Professional Documents
Culture Documents
SEMICONDUCTOR
SCIENCE AND
TECHNOLOGY
This page intentionally left blank
COMPREHENSIVE
SEMICONDUCTOR
SCIENCE AND
TECHNOLOGY
Editors-in-Chief
Pallab Bhattacharya
Department of Electrical Engineering and Computer Science,
University of Michigan, Ann Arbor, MI, USA
Roberto Fornari
Leibniz Institute for Crystal Growth, Berlin, Germany
and
Institute of Physics, Humboldt University, Berlin, Germany
Hiroshi Kamimura
Research Institute for Science and Technology,
Tokyo University of Science, Tokyo, Japan
Volume 1
PHYSICS AND FUNDAMENTAL THEORY
Volume Editor
Hiroshi Kamimura
Research Institute for Science and Technology,
Tokyo University of Science, Tokyo, Japan
The following article is a US Government works in the public domain and is not subject to copyright:
CHAPTER 4.09 ATOMIC RESOLUTION CHARACTERIZATION OF SEMICONDUCTOR MATERIALS
BY ABERRATION-CORRECTED TRANSMISSION ELECTRON MICROSCOPY
Permissions may be sought directly from Elsevier’s Science & Technology Rights
Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333;
email: permissions@elsevier.com. Alternatively you can submit your request online by
visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting
Obtaining permission to use Elsevier material
Notice
No responsibility is assumed by the publisher for any injury and/or damage to persons
or property as a matter of products liability, negligence or otherwise, or from any use
or operation of any methods, products, instructions or ideas contained in the material
herein. Because of rapid advances in the medical sciences, in particular, independent
verification of diagnoses and drug dosages should be made
ISBN: 978-0-444-53143-8
11 12 13 14 10 9 8 7 6 5 4 3 2 1
v
vi Contents of Volume 1
1.10 Electronic States and Transport Properties of Carbon Crystalline: Graphene, Nanotube,
and Graphite 359
Y. Iye, University of Tokyo, Kashiwa, Chiba, Japan
1.11 Angle-Resolved Photoemission Spectroscopy of Graphene, Graphite, and Related Compounds 383
T. Sato and T. Takahashi, Tohoku University, Sendai, Japan
1.12 Theory of Superconductivity in Graphite Intercalation Compounds 410
Y. Takada, University of Tokyo, Kashiwa, Chiba, Japan
Contributors to Volume 1
H. Aoki Chapter 1.05 p. 175
University of Tokyo, Tokyo, Japan
J. R. Chelikowsky Chapter 1.01 p. 1
University of Texas, Austin, TX, USA
W. R. Clarke Chapter 1.08 p. 279
University of New South Wales, Sydney, NSW, Australia
J. Deslippe Chapter 1.02 p. 42
University of California at Berkeley and Lawrence Berkeley National Laboratory,
Berkeley, CA, USA
M. Eto Chapter 1.03 p. 77
Keio University, Yokohama, Japan
Y. Iye Chapter 1.10 p. 359
University of Tokyo, Kashiwa, Chiba, Japan
J. K. Jain Chapter 1.06 p. 210
The Pennsylvania State University, University Park, PA, USA
H. Kamimura Chapter 1.03 p. 77
Tokyo University of Science, Tokyo, Japan
Y. Kangawa Chapter 1.04 p. 113
Kyushu University, Fukuoka, Japan
C.-T. Liang Chapter 1.08 p. 279
National Taiwan University, Taipei, China
S. G. Louie Chapter 1.02 p. 42
University of California at Berkeley and Lawrence Berkeley National Laboratory,
Berkeley, CA, USA
S. Murakami Chapter 1.07 p. 222
Tokyo Institute of Technology, Tokyo, Japan
N. Nagaosa Chapter 1.07 p. 222
University of Tokyo, Tokyo, Japan
T. Nakayama Chapter 1.04 p. 113
Chiba University, Chiba, Japan
T. Sato Chapter 1.11 p. 383
Tohoku University, Sendai, Japan
K. Shiraishi Chapter 1.04 p. 113
University of Tsukuba, Ibaraki, Japan
vii
viii Contributors to Volume 1
ix
x Preface
astonishing facilities for controlling the crystallographic structure and purity. Progress has called for research
by scientists and engineers from many disciplines, and the results of their efforts are documented in thousands
of archival articles, patents, and licensed technologies. Those new to the subject blanch at the thought of
assimilating even a small fraction of this immense knowledge base, and turn to the many specialist books that
attempt to summarize specific aspects of the field. To date, there is no comprehensive work that covers the
complete spectrum from fundamental semiconductor physics to design and manufacture devices which the
physics predicted to be possible, and which are now actually fabricated by materials scientists. In this sense, a
comprehensive work is different from an encyclopedia and a handbook. Comprehensive Semiconductor Science and
Technology consists of much longer chapters, which contain extensive cross-references to other relevant
information within the work and references to the vast available literature beyond the work. Each chapter
may be partly tutorial so that graduate students can greatly benefit from its reading.
This comprehensive comprises six volumes, and has been broadly divided into three main sections:
Section 1: Physics/Fundamental Theory – Edited by Hiroshi Kamimura
Section 2: Chemistry/Preparation – Edited by Roberto Fornari
Section 3: Devices/Application – Edited by Pallab Bhattacharya
Each of these complementary sections will provide a complete description of one aspect of the whole, and
will fuse together to give a comprehensive picture of the semiconductor world. The first section, which covers
Volumes 1 and 2, is concerned with the fundamental physics of semiconductors, showing how the electronic
properties and lattice dynamics change drastically when systems vary from bulk to a low-dimensional structure
and further to an interface and to a nanometer size such as quantum dots. The section describes all the
important characteristics of transport and optical properties. Further, this section includes many new topics
such as spin effects, carbon crystalline systems such as graphene and nanotubes, ultrafast coherent optical
phenomena, quantum information processing, etc., since the field of semiconductors continues to expand.
Throughout Volumes 1 and 2, there is an emphasis on the full understanding of the underlying physics.
Section 2, which covers Volumes 3 and 4, generally deals with technology of semiconductors. This includes
a description of the main methods employed for the preparation of bulk single crystals and thin layers. It is
shown that large single crystals may be conveniently grown from their melt, from solutions, or from the vapor
phase, and that the thermodynamical properties of the semiconductors to be grown actually decide about the
most appropriate method. The reader will also find some specific examples in the chapters devoted to bulk
growth of silicon, wide-bandgap compounds, II–VI and III–V semiconducting compounds. In addition, Section
2 also contains reviews on thin film technology, in particular MBE and MOVPE. In particular, how a basic
technology can be suited to preparation of semiconducting layers with tailored properties is discussed.
Examples of advanced low-dimensional heterostructures and nanostructures are provided along with a specific
chapter on integration of dissimilar materials. Additional chapters highlight the development of technologies
for deposition of high-quality ferroelectric and high-K materials to be applied as memories and gate isolators,
respectively. In addition to growth technology, Volumes 3 and 4 also include contributions regarding proces-
sing and characterization of semiconductors. The reader will find valuable information about the formation of
shallow junctions in semiconductors, the fabrication of ohmic and Schottky contacts as well as several chapters
describing the sophisticated methods used nowadays for investigating the physical and structural characteristics
of substrates, films, and quantum structures.
Volumes 5 and 6, which correspond to Section 3, contain chapters on the physics, technology and
application of devices, and circuits realized with diverse materials and heterostructure systems. The materials
include Si/SiGe, GaAs-, and InP-based heterostructures, antimonides, GaN-, ZnO-, and SiC-based materials,
graphene, HgCdTe, and materials for molecular electronics. Also included are carbon nanotubes and nano-
structured materials for flexible and stretchable electronics. The chapters describe in detail the properties of
high-speed, high-frequency, and high-power bipolar and field-effect transistors made with a variety of
heterostructure systems, negative differential resistance devices, single electron transistors, and microelectro-
mechanical system (MEMS)-based sensors. Several chapters describe the principles and properties of a wide
variety of optoelectronic devices including photodetectors, solar cells, light emitting diodes, and lasers. Passive
Preface xi
photonic devices such as waveguides and filters are also included. The operating wavelength of these devices
ranges from very short (ultraviolet) to very long (terahertz). The active region of some of these devices includes
low-dimensional quantum-confined heterostructures such as quantum wells and quantum dots. Concepts such
as the manipulation of slow and fast light and disordering of quantum structures for optoelectronic device
integration are described in detail in a couple of chapters. The use of photonic crystals and microcavities in
optical devices is elucidated. Electronics with molecules is included with a good description of the underlying
physics. Finally, the physics, fabrication, and characteristics of spin-based electronic and optoelectronic devices,
more commonly known as semiconductor spintronic devices, are described in a few chapters. Each chapter and
topic is uniquely distinct, complete with the appropriate background material and references.
xiii
xiv Editors in Chief
Roberto Fornari studied Solid State Physics at the University of Parma, Italy.
He is presently director of the Leibniz Institute for Crystal Growth (IKZ) in
Berlin and holds the Chair of Crystal Growth at the Physics Dept. of the
Humboldt University Berlin (joint appointment). Before moving to Germany
in 2003, he worked over twenty years as a research scientist at the Institute for
Electronic and Magnetic Materials (MASPEC, later IMEM) of the Italian CNR
where he led different research projects on growth and thermal processing of
bulk III-V semiconductors, HVPE and MOVPE of Nitrides, characterization of
semiconductors by electrical and optical techniques. He has authored/co-
authored about 180 scientific papers, eight patents and different book chapters.
He has edited books and proceedings on crystal growth and semiconductors
physics and was subject editor of the Encyclopedia of Materials published by
Pergamon Press in 2001. He is presently a member of the editorial board of
the Journal of Crystal Growth, the Crystal Research and Technology, and the Journal of
Optoelectronics and Advanced Materials.
He has been Chairman of the IUCr Commission for Crystal Growth and
Characterization of Materials from 1999 to 2005 and then member till 2008.
From 2001 to 2007 he served on the Executive Committee of the International
Organization for Crystal Growth and is currently President of this organization.
Volume 2
xv
xvi Contents of All Volumes
Volume 3
Volume 4
Volume 5
5.08 Silicon Single Electron Transistors Operating at Room Temperature and Their Applications
5.09 Electronics with Molecules
5.10 Electronic and Optoelectronic Properties and Applications of Carbon Nanotubes
5.11 Micro- and Nanostructured Semiconductor Materials for Flexible and Stretchable Electronics
5.12 MEMS-Based Sensors
5.13 III–V Compound Avalanche Photodiodes
5.14 Disordering of Quantum Structures for Optoelectronic Device Integration
5.15 Quantum-Well Lasers and Their Applications
5.16 Quantum Cascade Lasers
5.17 Slow and Fast Light in Quantum-Well and Quantum-Dot Semiconductor
Optical Amplifiers
Volume 6
1.01.1 Introduction 1
1.01.1.1 Early History 1
1.01.1.2 The Phillips–Kleinman Cancellation Theorem 3
1.01.2 Defining Pseudopotentials 4
1.01.2.1 Model Potentials 4
1.01.3 Solving the Electronic Structure Problem 6
1.01.4 The Empirical Pseudopotential Method 6
1.01.4.1 Optical Properties of Semiconductors 6
1.01.4.2 The Structure and Form Factors of Semiconductors 8
1.01.4.3 The Role of Nonlocality in Pseudopotentials 10
1.01.4.4 The EPM Applied to Diamond Structure Semiconductors 11
1.01.4.4.1 The electronic structure of silicon 11
1.01.4.4.2 The electronic structure of germanium, gallium arsenide, and zinc selenide 15
1.01.5 The Ab Initio Pseudopotential Method 26
1.01.5.1 Constructing Pseudopotentials from Density Functional Theory 28
1.01.5.2 Structural Properties of Semiconductor Crystals 31
1.01.5.2.1 Total electronic energy from pseudopotential–density functional theory 31
1.01.5.2.2 Phase stability of crystals 32
1.01.5.2.3 Vibrational properties 34
1.01.6 Summary and Conclusions 37
References 39
1
2 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
‘knowing’ and ‘doing’ are two different issues. The electrons and nucleus of an atom are treated as an
equations in question were much too complicated to inert ion core; the valence electrons are treated as an
be solved when Dirac made this statement. In fact, electron sea moving against the periodic background
they are much too complicated to be solved today, of the ion cores. This pseudopotential picture sets a
even with state-of-the-art computational platforms. length and energy scale to the problem that is deter-
As Dirac anticipated, the key to progress in this mined by only the valence states. A profound
area is to develop ‘approximate practical methods’. consequence of this length and energy scale is that
Today, we have a number of such practical meth- all atoms of the periodic table can be treated on an
ods that allow us to access some properties in a equal footing. Consider the situation without the
manner Dirac suggested. The first successes pseudopotential approximation; in this case, an
appeared some 30 or 40 years after Dirac’s state- atom of hydrogen and an atom of cesium would
ment and were led by the development of the require all the electronic states to be placed on an
pseudopotential concept. The pseudopotential equal footing. The energy and length scales between
model of a solid provided a practical model to the core and valance states would be different by
address the quantum mechanical problem and orders of magnitude and would enormously compli-
advances in computers provided the means to over- cate the problem.
come the numerical hurdles. The pseudopotential idea is not new. Several key
A key attribute of a pseudopotential is that it elements were recognized after the advent of quan-
allows us to address directly the electronically or tum mechanics. Both Enrico Fermi and Hans
chemically active valence states of an atom and Hellmann contributed seminal papers in the mid-
remove from consideration the electronically inert 1930s (Fermi, 1934; Hellman, 1935). Interestingly,
core states. This decomposition between active and they recognized somewhat different, but important
inactive states is readily done for most atoms, for elements of pseudopotential theory. Fermi’s focus
example, in a silicon atom the 1s22s22p6 core states was on determining the phase shift in wave functions
are tightly bound compared to the 3s23p2 valence of high-lying alkali atoms subject to perturbations
states. In Figure 1, we illustrate the pseudopotential from foreign atoms. He correctly recognized that if
model for a crystalline solid. In this case, the core one were interested in the long-range behavior of the
wave function, it was not necessary to get the details
correct near the nucleus. It should be possible to
replace the true potential near the nucleus with a
simple potential, or pseudopotential, that yields a
similar value for the long-range part. Fermi also
established the importance of a scattering length
within this paper.
Hellmann’s work is probably closer to our con-
temporary picture (Schwarz et al., 1999a, 1999b).
Hellmann recognized that the valence orbitals for
many-electron atoms contained pronounced oscilla-
tions in the region where the atomic core electrons
remain. Determining an accurate description of the
oscillatory behavior near the core and the long-range
behavior in the chemically relevant region would
render the problem computationally complex.
Hellmann also recognized these core electrons did
not play an important role in determining the che-
Nucleus mical bond; after all this was clear from the periodic
Core electrons table. Elements such as Si and Ge have different ion-
Valence electrons core configurations, but the valence electron config-
uration is the same, and their chemical properties are
Figure 1 Pseudopotential concept of a solid. The ion
cores consisting of the core electrons and valence electrons similar.
are inert. The chemically active valence electrons move In today’s language, we would argue that the
within this array of ion cores. valence electrons experience an additional kinetic
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 3
The sum is over the core states, c, and pv represents
Figure 2 Schematic of an ion-core pseudopotential. The the pseudopotential wave function. This form of the
all-electron potential and the pseudopotential ion-core wave function recognizes that near the nucleus the
potential are similar outside of the core region. valence wave function should appear atomic like and
contain elements similar to the core states. Away
from the nucleus, the core states will have little
energy contribution in the core region because of the
amplitude and we expect: v ðrÞ pv ðrÞ:
orthogonality requirement, that is, the valence state
Moreover, we expect pv(r) to be smoothly varying
of an atom is orthogonal to a core state by nodal over all space. Outside of the core, the potential will
structure in the core region. In this 1935 paper, vary slowly. Within the core, we expect its contribu-
Hellmann proposed that this kinetic energy term tions will be small compared to the core-like
could be quantified using the Thomas–Fermi expres- functions. To determine the admixture of the core
sion for the kinetic energy of a free electron gas, with the pseudo-wave-function, we demand the fol-
which depends only on the electron density. When lowing condition:
this term was added to the attractive Coulomb part of Z
the valence electron interacting with the ion core, he c ðrÞv ðrÞd3 r ¼ hc j v i ¼ 0 ð3Þ
argued that the resulting potential was ‘weak and
constant’. An example of a pseudopotential similar where we use the ‘bra–ket’ notation. The valence
to those Hellmann suggested is schematically illu- state must be orthogonal to the core state, owing to
strated in Figure 2. the nature of the eigenvalue problem. The orthogon-
ality requirement yields
av;c ¼ – c j pv ð4Þ
1.01.1.2 The Phillips–Kleinman
Cancellation Theorem
This form of the wave function was first proposed by
While the work of Hellmann and Fermi established Conyers Herring (Herring, 1940) in a different con-
the elementary ideas associated with pseudopoten- text. He proposed describing electronic states in
tials, a rigorous transformation of an all-electron crystals using orthogonalized plane waves. In this
potential to a pseudopotential was lacking. One of model, the valence states in a solid were replicated
the earliest such transformation is based on the ideas using plane waves that were orthgonalized to the
of Phillips and Kleinman (Kleinman and Phillips, core levels.
1960a, 1960b; Phillips and Kleinman, 1959; Phillips If we apply the Hamiltonian in Equation (1) to the
and Kleinman, 1962). Although their conclusions are valence wave function, we obtain
more far reaching, we will focus on solving the elec-
X
tronic structure problem for an isolated atom. We H pv þ c j pv ðEv – Ec Þc ¼ Ev pv ð5Þ
will make the one-electron approximation, that is, c
4 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
We have recognized that Hc ¼ Ecc. We can define where G are reciprocal lattice vectors. S(G) is a
a pseudopotential Hamiltonian, Hp, as structure factor given by
Hp ¼ H þ V R ð6Þ 1 X
SðGÞ ¼ expðiG ? tÞ ð11Þ
Na t
where we define a potential operator, V R, as
X where Na is the number of atoms in the basis. Va(G) is
VR¼ hc j . . .iðEv – Ec Þc ð7Þ
c
an atomic form factor given by
Z
The empty ‘ket’ given by j. . . > requires an operation 1
Va ðGÞ ¼ Va ðr ÞexpðiG?rÞd3 r ð12Þ
wherein the wave function is projected on c and the a
functional dependence now corresponds to c. If we Here a is the volume per atom and Va is the atomic
define a pseudopotential as Vp ¼ V þ V R , we have potential that we take to be spherically symmetric.
now transformed the one-electron Schrödinger equa- The structure factor contains information on the
tion to crystal structure and the form factors contain infor-
mation on the electronic interactions. The reciprocal
– h2 r2
Hp pv ¼ þ Vp pv ¼ Ev pv ð8Þ lattice vectors are designed to have the property:
2m
exp(iG?R) ¼ 1 (Kittel, 2005). This insures that the
This description of the atom has a number of advan- potential is periodic:
tages. The repulsive part of the potential, V R, largely P
V ðr þ RÞ ¼ Va ðGÞSðGÞexpðiG?ðr þ RÞÞ
cancels the attractive part of the all-electron poten- G
tial in the core region. The pseudo-wave-function is P ð13Þ
¼ Va ðGÞSðGÞexpðiG?rÞ ¼ VðrÞ
smooth as the core oscillations have been removed G
and the eigen-value, Ev, is identical to the all-electron
While this plane-wave expansion of the potential
potential. While the wave function is now amenable to
reflects the translational symmetry, the expansion is
a simple basis, the pseudopotential within this
only helpful if the number of waves (or reciprocal
construction is more complex than the all-electron
lattice vectors) is manageable. This will not be the
potential. Although the potential is weak and only
case for the all-electron potential owing to the sin-
binds the valence state, this potential is energy depen-
gularity of the Coulomb potential at the nucleus. The
dent, state dependent and involves a nonlocal,
pseudopotential avoids this issue by removing this
non-Hermitian operator. As such, the Phillips–
singularity. What remains is to define a procedure to
Kleinman potential is rarely, if ever, used for calculat-
construct the potential.
ing pseudopotentials. However, the cancellation
theorem is useful in demonstrating the essential
features of a pseudopotential.
1.01.2.1 Model Potentials
A realistic and powerful approach to find the required
form factors is based on experiment. For example, sup-
1.01.2 Defining Pseudopotentials pose we consider a potential similar to what Hellmann
proposed. Let us consider the following model, which
One could attempt to form a potential for an ele- only involves one parameter; the core radius, rc:
mental crystal by writing (
X 0 r rc
V ðrÞ ¼ Va ðr – R – tÞ ð9Þ Vcp ðr Þ ¼ 2
ð14Þ
R;t – Zv e =r r > rc
The next step in defining the atomic form factor is interesting consequence of Thomas–Fermi screening
to employ Equation (12): is that the potential now has a well-defined value
when G ¼ 0:
– 4Zv e 2 – 4e 2 n 2
Vcp ðGÞ ¼ cosðGrc Þ ð15Þ Vap ð0Þ ¼ ¼ – EF ð19Þ
a G 2 Ks2 3
The long-range nature of the coulomb tail requires The G ¼ 0 term corresponds to setting a reference
special handling in executing this integral (Kittel, energy for the solid as it corresponds to the average
2005). The form factor ‘rings’, that is, the form factor of the potential. This limiting case from Thomas–
oscillates as a cosine function owing to the disconti- Fermi screening is not a very good approximation
nuity in the potential at rc. What remains is to screen for a semiconductor such as silicon and is more
the ion-core pseudopotential by including a dielec- appropriate for a metal. We illustrate a schematic
tric screening function that replicates the role of the pseudopotential in Figure 3.
valence electron–electron interactions. Often The Phillips–Kleinman cancellation theorem
Thomas–Fermi screening is used for this purpose yields a state-dependent pseudopotential; the poten-
(Kittel, 2005): tial depends on the nature of the core states. For
example, in carbon there is no 1p core state and the
Vcp ðGÞ valence 2p is not required to be orthogonal to the
Vcp ðGÞ ¼ ð16Þ
TF ðGÞ core. This is not true for silicon, in which both s and p
Thomas–Fermi screening is given by states exist in the core; both the s and p valence states
see a repulsive term. Historically, this difference has
Ks2 been used to explain why silicon and carbon chem-
TF ðqÞ ¼ 1 þ ð17Þ
q2 istry are different. The previous model potential can
where 1/Ks is the Thomas–Fermi screening length: be modified to reflect the state dependence:
(
Ks2 ¼ 6ne 2 =EF , where n is the valence electron den- Al r rc;l
p
sity (n ¼ Zv/a) and EF is the Fermi energy. The Vc;l ðr Þ ¼ ð20Þ
2
– Zv e =r r > rc;l
atomic form factor is now given by
– 4e 2 n This potential is not a simple function of position, but
Vap ðGÞ ¼ cosðGrc Þ ð18Þ rather acts on the l-component present in the wave
G 2 þ Ks2
function (Cohen and Chelikowsky, 1982, 1989;
A common practice is to set the potential to zero Chelikowsky and Cohen, 1992). This can be accom-
outside of the second node to remove the oscillatory plished by using a projection operator as will be
or ringing behavior of the cos(Grc) term. An discussed later. Setting this technical issue aside,
V(q)
V(r) ~ ∫V(q) eiqr dq
~ (½ BOND LENGTH)–1
q
– 2/3 EF
Screened ion limit
for metals
Figure 3 Model of an atomic pseudopotential. The required form factors can be extracted from this model. Note the limiting
case of G ¼ 0 for a metallic system.
6 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
(
there are now additional parameters to be deter- X h2 2
mined as the well depth and size must be ðk þ GÞ – En ðkÞ GG9
2m
established for each l-component.
G9 )
ð23Þ
A common procedure in the early days of pseu- þVap ðG9 – GÞS ðG9 – GÞ n ðk; G9Þ ¼ 0
dopotential involved fitting the well depths and sizes
(Al, rc,l) to replicate optical data of ionized atoms. For This corresponds to a set of linear equations, one for
example, to construct an ion-core pseudopotential each G vector in the set. For a nontrivial solution, we
for an Al atom, one needs to consider the optical require the following:
excitations for an Alþ2 ion. In some cases, this is
relatively easy, for example, treating the Na atom h2
2
does not involve an ionized atom. However, an O det ðk þ GÞ – En ðkÞ GG9
2m
atom would require examining a C5+ ion.
ð24Þ
Experimentally, it is a difficult task to multiply ionize p
þ Va ðG9 – GÞS ðG9 – GÞ ¼ 0
such an atom and measure the atomic levels. Often
the potentials were estimated by extrapolation from
lighter atoms to heavier ones. Another issue concerns The diagonal elements of this determinant contain
improved dielectric functions for screening the the kinetic energy contribution; the off-diagonal
potentials. Both issues were the subject of a number elements contain the form factors for the pseudopo-
of studies, but without a definitive resolution. By the tential and the structure factor. In this example, we
mid-1960s, an inventory of such model potentials consider only elemental crystals so that the form
existed (Animalu and Heine, 1965; Cohen and factors can be separated out from the structure factor.
Heine, 1970) with some success, especially for metals. This eigenvalue problem is a standard mathematical
problem, and is easily handled, save for highly complex
systems. Typically, for common semiconductors a few
hundred plane waves are required to obtain a con-
1.01.3 Solving the Electronic verged result. This operation can often be performed
Structure Problem on a laptop computer for crystalline semiconductors.
function, "(!) ¼ "1(!) þ i"2(!) (Cohen and jMvc ðkÞj2 ¼ jhuv ðkÞjrjuc ðkÞij2 ð31Þ
Chelikowsky, 1989). They are related to each other as
The functions, un(k), are the periodic part of the
follows:
Bloch wave functions:
N 2 ¼ "1 þ i"2
X
"1 ¼ n2 – k2 ð25Þ un ðkÞ ¼ n ðk; GÞexpðiG?rÞ ð32Þ
G
"2 ¼ 2nk
where n is the band index. The periodic part obeys
The normal incident reflectivity of a semiconductor un ðk; r þ RÞ ¼ un ðk; rÞ. Implicit in this expression is
is given by the response functions: the existence of a bandgap that delineates the
valence and conduction bands. This form also
N – 1 2 ðn – 1Þ2 þ k2
R¼ ¼ ð26Þ assumes cubic symmetry. The physical content of
N þ 1 ðn þ 1Þ2 þ k2
the dielectric function is clear. The delta function
Hence, a knowledge of the dielectric function, "(!), insures energy conservation and the dipole matrix
is sufficient to yield the complex index of refraction element accounts for symmetry, that is, we assume
and the reflectivity. In fact, only the real (or imagin- dipole transitions. Another implicit factor is that the
ary) part of the dielectric function is required owing transitions from filled bands to empty bands are
to the Kramers–Kronig relation: direct; the wavevector of the initial and final states
Z 1
are equal. This is appropriate for optical transitions
2 !9"2 ð!9Þ where the wavelength of light is orders of magni-
"1 ð!Þ ¼ 1 þ P d!9 ð27Þ
0 !92 – !2 tude larger than the atomic length scale. Also, the
The principal part of the integral is required as is a Ehrenreich–Cohen form does not include excitonic-
knowledge of "2 over all frequencies, although, in (electron–hole) interactions, local field corrections, or
practice, this is not a stringent requirement. self-energy effects, which can be included (Rohlfing
There are useful sum rules that can be applied to and Louie, 2000) albeit at some considerable compu-
test the accuracy of response functions. One obvious tational cost. Moreover, the qualitative features of the
example is to use the ! ! 0 of the Kramers–Kronig optical spectra are correctly obtained from the
relationship: Ehenreich–Cohen expression.
Z 1
Once the imaginary part of the dielectric function
2 "2 ð!Þ
"1 ð0Þ ¼ 1 þ d! ð28Þ is known, it is straightforward using the Kramers–
0 !
Kronig transform to find the real part of the dielectric
Another sum rule is function and the reflectivity, which can be directly
Z 1 compared to experiment. To implement the
p
! ¼ !"2 ð!Þd! ð29Þ Ehenreich–Cohen expression, we need to know the
2 2 0
energy bands, En(k) and the corresponding wave
where !p is the plasma frequency given by functions n,k. This can be obtained once the pseu-
!p 2 ¼ 4ne 2 =m. dopotential is defined.
The connection between the macroscopic response Usually, the dipole matrix elements in Equation
function and the atomistic world can be made by a (31) are smoothly varying with k and can be assumed
computation of the imaginary part of the dielectric constant, except at high symmetry points. In this
function. A realistic, but somewhat simplified, expres- situation, it is useful to remove the matrix elements
sion comes from Ehrenreich and Cohen (EC59). The and isolate that part of "2(!) that produces the major
dielectric function involves transitions between the structure. The resulting function is called the ‘joint
valence band (v) to the conduction band (c): density of states’:
Z
4e 2 h X 2 2
"2 ð!Þ ¼ Jvc ð!Þ ¼ ½!vc ðkÞ – !d3 k ð33Þ
3m2 !2 v;c ð2Þ3 ð2Þ3 BZ
Z ð30Þ
ð!vc ðkÞ – !ÞjMvc ðkÞj2 d3 k This integral can be recast as
BZ
Z
The integral is over all states in the Brillouin zone 2 ds
Jvc ð!Þ ¼ ð34Þ
where the dipole matrix element is given by ð2Þ3 !¼!vc jrk !vc ðkÞj
8 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
where ds is a surface element in wavevector space where the lattice basis vectors are given by
defined by ! ¼ !vc(k). This is similar to the density of
â1 ¼ aðŷ þ ẑÞ=2; â2 ¼ aðx̂ þ ẑÞ=2; â3 ¼ aðx̂ þ ŷÞ=2
states, save here we replace the band energy by the
transition frequency between the valence and con- The diamond and zinc–blende crystals possess a pri-
duction bands. The structure in this joint density of mitive cell given by these vectors, with the cell
states occurs at critical points, just as in the density of volume given by
states (Cohen and Chelikowsky, 1989). These critical
points occur when c ¼ ja1 ?a2 a3 j ¼ a 3 =4 ð37Þ
There are two atoms in the basis located at
rk !vc kcp ¼ 0 ð35Þ
t ¼ ð1;1;1Þa=8, so the atomic volume, a is given
Physically, this occurs when the group velocity of the by a3/8.
hole in the valence band equals that of the electron in The reciprocal lattice vectors are given by
the conduction band. Four topologically distinct cri-
tical points occur: M0, M1, M2, and M3, which Gm1 ;m2 ;m3 ¼ m1 bˆ1 þ m2 bˆ2 þ m3 bˆ3 ð38Þ
represent a local minimum, two saddle points, and a where the reciprocal lattice basis vectors are given by
local maximum. The structure associated with these
critical points is presented in Figure 4. Structure in
either the imaginary part of the dielectric function or bˆ1 ¼ ð2=aÞð – x̂ þ ŷ þ ẑÞ=2; bˆ2 ¼ ð2=aÞðx̂ – ŷ þ ẑÞ=2
the density states can be identified with these critical bˆ3 ¼ ð2=aÞðx̂ þ ŷ – ẑÞ=2
points (Cohen and Chelikowsky, 1989).
This set of vectors has the desired property that
1.01.4.2 The Structure and Form Factors G ? R is an integral multiple of 2. In some disci-
of Semiconductors plines such as crystallography, the factor of 2 is
omitted from the definition of reciprocal lattice
This section focuses on cubic semiconductors that vectors and G ? R is an integer.
occur in either the zinc–blende or diamond struc- For diamond, the structure factor is given by
tures. These two structures are isostructural; they
have the same atomic arrangement, but the con- SðGÞ ¼ cosðG ? tÞ ð39Þ
stituent atoms can be different. Semiconductors
such as Si, Ge, GaAs, InSb, ZnSe, and CdTe
The structure factor for the zinc–blende crystal can-
occur in these structures. Figure 5 illustrates the
not be separated from the potential terms. For a zinc–
diamond structure. The diamond structure has
blende crystal, A B , where A is the cation and B is the
cubic symmetry and can be conceptualized by
anion, it is easy to show that the product of the form
considering two interpenetrating face-centered
factors and structure factor in Equations (23) and (24)
cubic crystals.
can be replaced by
The lattice vectors for diamond are given by
Rn1 ;n2 ;n3 ¼ n1 â1 þ n2 â2 þ n3 â3 ð36Þ VA ðGÞexpðiG?tÞ þ VB ðGÞexpð – iG?tÞ=2 ð40Þ
M2
M1
J(ω)
M0
M3
ω0 ω1 ω2 ω3
ω
Figure 4 Structure associated with the four types of critical points.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 9
0 1 0 VS serves as a reference
energy
pffiffiffi pffiffiffi
a 3 1= 2 1= 2 VS, VA both required
4 0 1 VA required
8 1 0 VS required
pffiffiffi pffiffiffi
11 1= 2 1= 2 VS, VA both required
Table 2 Pseudopotential form factors (Ry) and lattice constants for selected diamond and zinc–blende semiconductors
From Cohen ML and Bergstresser TK (1965) Band structures and pseudopotential form factors for fourteen semiconductors of the
diamond and zinc-blende structures. Physical Review 141: 789.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 11
P
1
1989). Nonlocal components for tetrahedral semi-
P yl Vl;c ðr ÞP l ¼ P ys Vs;c
p
Vcp ðrÞ ¼ p
ðr ÞP s
l¼0 ð42Þ conductors are available in the literature
þP yp Vp;c ðr ÞP p þ P yd Vd ;c ðr ÞP d þ
6
L3 Γ2´
Γ2´
4
Γ15
Γ15
2
L1
X1 Γ25´
Γ25´
0
L3´
–2
Energy (eV)
X4
–4
–6
Si X1
L1
–8
L2´
–10
Γ Γ1
–12
L Λ Γ Δ X U,K Σ Γ
Wavevector, k
Figure 7 Band structure for silicon.
Figure 9. The measured dielectric function is nearly variation of the field within the unit cell.
a factor of 2 larger near 3.4 eV than the calculated Unfortunately, the corrections for local fields do little
value for the energy region in question. This discre- to improve the dielectric function in this region as
pancy is also observed in the real part of the indicated in Figure 9. The role of the electron–hole
dielectric function, which is displayed in Figure 10. interactions in semiconductors was addressed by a
While the Ehrenreich–Cohen formalism number of groups. One of the first realistic attempts is
(Ehrenreich and Cohen, 1959) replicates the essential from Hanke and Sham (1974). They examined the
features of the optical constants, it omits local fields optical properties of diamond, including both local
and electron–hole interactions. Local-field correc- fields and exchange interactions. More recent work
tions were addressed by Adler (1962) and Wiser by Louie and collaborators (Hybertsen and Louie,
(1963). Local fields allow for the microscopic 1986; Rohlfing and Louie, 1998, 2000) has unequi-
vocally demonstrated the role of excitonic
contributions to the optical response functions.
In Figure 11, we exhibit the calculated and mea-
sured reflectivity spectrum for silicon. As for most
Si
40 solid-state spectra, the spectrum is not highly struc-
Experiment tured, and on first inspection appears to contain little
With local-field corrections
information. Such an assessment would be incorrect.
Without local-field
30 corrections
Details of the spectrum can be enhanced by numeri-
cally differentiating the spectrum. In Figure 12, we
ε2 (ω)
50
0.8
Si
40 Theory 0.7 Si
Experiment Theory
0.6 Experiment
30
0.5
20
R (E )
ε1 (E )
0.4
10 0.3
0 0.2
0.1
–10
0
0 1 2 3 4 5 6 7
–20 Energy (eV)
0 1 2 3 4 5 6 7
Energy (eV)
Figure 11 Measure and calculated normal incident
Figure 10 Real part of the dielectric function for Si. reflectivity for Si.
14 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
+1.0
(a) E1 + E′0 E2 E1′
+0.5
Si
–0.5
(eV–1)
Experiment
–1.0
1 dR
R dE
(b)
+0.5
–0.5
Theory
–1.0
2 3 4 5 6
Energy (eV)
Figure 12 Measured and calculated normal incident modulated reflectivity for Si.
the same energy region, that is, the lowest energy multiplicity of critical points within this energy
direct transitions at L and at have nearly the same region. This conclusion has been drawn by several
energy. In fact, a good deal of work has been authors (Welkowsky and Braunstein, 1972). The E2
peformed to determine precisely where in the structure dominates the reflectivity spectrum and
Brillouin zone this transition takes place. occurs at about 4.5 eV. This structure arises from
Piezoelectric experiments (Gobeli and Kane, 1965), large regions of the Brillouin zone.
chemical shifts in Ge–Si alloys (Tauc and Abraham, While optical properties served as a centerpiece
1961), electroreflectance (Pollak and Cardona, 1968), for the first EPM calculations, photoemission mea-
and some wavelength-modulation techniques (Koo surements provide information inaccessible to
et al., 1971) have suggested that the first reflectivity interband transitions. In particular, photoemission
peak arises near the zone center, perhaps along the can provide the absolute position of bands relative
direction. However, other work (Grover and to the vacuum. The valence band density of states for
Handler, 1974) suggested that this transition occurs silicon calculated using pseudopotentials is com-
along the direction. Unfortunately, the pseudopo- pared with experiment in Figure 13.
tential work has not resolved this issue since the Under certain conditions, it is possible to compare
reflectivity structure in question arises from contri- directly the theoretical density of states with photo-
butions near and along both the and emission measurements. The density of states, D(E),
directions. The dominant transition in the calcula- for crystalline matter, can be defined as
tions appears to be along the direction, but no firm Z
conclusion can be drawn. What is clear is that the 2 X
DðEÞ ¼ d3 kðE – En ðkÞÞ ð47Þ
complexity of this structure does suggest a ð2Þ3 n
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 15
tal gap.
L4 + L5
4 Γ8 Γ8
L6 0.6
Γ6 Γ6
2 Γ7
L6
Γ7 X5 Ge Γ8 0.4
0 Γ8
R (E )
L4 + L5
Ge
–2 L6 Γ7 Γ7
Energy (eV)
Theory
0.2 Experiment
–4 X5
–6
0
0 1 2 3 4 5 6 7
–8 L6 Energy (eV)
X5
Figure 15 Theoretical and experimental reflectivity for
–10 L
6 crystalline germanium. The experimental data is from
Philipp and Ehrenreich, (1963).
–12 Γ6 Γ6
L Λ Γ Δ x U, K Σ
Wavevector, k In Figure 16, the logarithmic derivative reflectiv-
ity spectrum is displayed. The derivative spectrum
Figure 14 Band structure for crystalline germanium. The
top of the valence band is considered as the energy zero. vividly illustrates the rich reflectivity structure avail-
able using modulated techniques. Perhaps the most
interesting feature of the germanium reflectivity
In Figure 14, we present the band structure of spectrum is the E2 peak at 4.5 eV. Initially, the origin
germanium as determined from empirical pseudopo- of this spectral feature appeared to be located at the
tentials. This calculation includes spin–orbit X-point and to involve transitions between the high-
interactions, which are most significant for heavier est valence band and the lowest conduction band.
elements (Cohen and Chelikowsky, 1989). Unlike The energy of this transition is near the observed
the case of silicon, most band structures for germa- E2 peak and the dipole matrix elements for these
nium are in reasonably good agreement. The band transitions are large. However, the phase space asso-
structure of germanium differs from silicon most sig- ciated with this critical point is too small to yield
nificantly in the conduction band arrangement. The such a prominent structure. The observation of a
15 and 29 bands, using the notation in Figure 7, well-defined interband reduced mass for the E2
reverse their ordering. Moreover, while germanium structure from electroreflectance (Aspnes, 1973) sug-
and silicon are both indirect semiconductors, the con- gests a relatively well-localized critical-point origin
duction band minimum in germanium occurs at the for this spectral feature. Ruling out transitions near X,
L-point as opposed to a point near X in silicon. As the critical point must reside elsewhere within the
noted earlier, the valence band configuration for dia- Brillouin zone. Empirical pseudopotential calcula-
mond semiconductors is insensitivie to the tions reinforce this interpretation; these calculations
constituents: silicon, germanium, and tin have almost (Chelikowsky and Cohen, 1973) yield a well-defined
identical valence band configurations, but they have critical point near the special k point (Chadi and
quite different conduction band configurations. Cohen, 1973).
In Figure 15, we present the reflectivity spectrum The density of states for germanium is illustrated
of germanium as calculated from pseudopotential in Figure 17 and compared to photoemission and
band theory and as determined by experiment inverse photoemission. Photoemission methods such
(Grover and Handler, 1974). Compared to silicon, as X-ray photoelectron spectroscopy (XPS) and
germanium exhibits a richer spectrum. The addi- ultraviolet photoemission spectroscopy (UPS) mea-
tional structure arises from the larger spin–orbit sure only properties of occupied states. Inverse
coupling occurring in germanium. For example, in photoemission spectroscopy allows one to detect
the E1 region near 2.2 eV, a distinct doublet structure the properties of empty states. Electrons with a
occurs. This doublet arises from the spin–orbit split- well-defined energy are directed at a sample. These
ting in the L39 level. electrons decay to the lowest unoccupied states,
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 17
2
Ge
Theory
1 Experiment
1 ΔR (eV–1)
0
R ΔE
–1
–2
0 1 2 3 4 5 6 7
Energy (eV)
Figure 16 Theoretical and experimental modulated reflectivity for crystalline germanium. The experimental data is from
Zucca and Shen (1970).
with some of the decay transitions being radiative. We expect that many properties of semiconduc-
Photons emitted in this process are detected and a tors, for example, optical gaps, cohesive energies
photon counts versus incident electron energy spec- dielectric constants, bond charges, etc., might be
trum is measured. Like photoemission, the inverse determined simply with an ionicity factor based on
photoemission can be related to the empty state VA(G) or a similar measure. Actually, this is the case,
density of states, ignoring any matrix element effects. and although VA(G) is not explicitly used, a more
One can obtain a direct measure of the ‘quasi-parti- sophisticated measure, based on similar concepts, of
cle’ gap, that is, the energy required to obtain a ionic versus covalent contributions to crystal binding
noninteracting electron–hole pair, by a combination in diamond and zinc–blende semiconductors has
of photoemission and inverse photoemission spectra. been devised by Phillips (1973).
In Figure 18, we illustrate the calculated band
structure for gallium arsenide using the EPM. The
1.01.4.4.2(ii) Gallium arsenide We can approach nonlocal pseudopotential EPM approach has yielded
the electronic structure of gallium aresenide from the one of the most accurate band structures for GaAs to
perspective of perturbing the germanium potential. date. In fact, as a result of the extensive experimental
Intuitively, we expect studies in this compound, this band structure may be
the most accurate band structure over a large energy
VS ðGÞ ¼ ½VGa ðGÞ þ VAs ðGÞ=2 VGe ðGÞ ð48Þ
range available for any material. By compiling the
That is, the symmetric part of the potential should most recent experimental data, we estimate an error
resemble the form factors of germanium. The anti- of about 0.1 eV over an energy span of nearly 20 eV.
symmetric form factor, Thus, we have an accuracy of roughly 1% or better;
this is almost an order of magnitude better than other
VA ðGÞ ¼ ½VGa ðGÞ – VAs ðGÞ=2 ð49Þ band structures.
which is a measure of the difference between the As we introduce antisymmetric form factors for
gallium and arsenic potentials, should contain the the gallium arsenide crystal potential, several signifi-
essential information which distinguishes germanium cant alterations from a germanium-like band
from gallium arsenide. Formally, it is the deviation of structure may be observed. First, the bottom valence
VA(G) from zero that distinguishes zinc–blende from band, which now has atomic s-like character and is
diamond semiconductors. In this sense, we expect a localized on the anion, splits off from the the rest of
smooth transition from a purely covalent semicon- the valence band to form the antisymmetric gap. This
ductor VA(G) ¼ 0, to a highly ionic semiconductor gap grows with increasing ionicity of the crystal or
where VA(G) deviates notably from zero. charge transfer between the cation and anion.
18 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
6 L4, 5
Γ8
L6 Γ8
4 Γ7 Γ7
X7
2 X6
L6 Γ6
Γ6 GaAs
Γ8 Γ8
0
L4, 5 Γ7 Γ7
L6
Energy (eV)
–2
X7
X6
–4
–6
L6 X6
–8
–10 X6
L6
–12 Γ6 Γ6
L Λ Γ Δ X K Σ Γ
Wavevector, k
Figure 18 Energy bands for crystalline gallium arsenide. The top of the valence band is taken to be the zero energy
reference.
oscillator with a resonant frequency at about 5 eV. electron–hole ‘excitonic’ interactions. Recent work
We can think of this resonant frequency as a funda- based on a many-body formulation of the Bethe–
mental property of GaAs that represents the average Salpeter equation confirms this interpretation as
bonding–antibonding energy level separation. indicted in Figure 22. The Bethe–Salpeter equation
Phillips divides up this average bonding–antibonding strongly enhances the magnitude of the imaginary
gap into a part that is ionic and a part that is covalent. part of the dielectric function, but does not shift the
One can extract from the dielectric function relevant critical points.
parameters for formulating ionicity scales (Phillips, The EPM-calculated reflectivity is shown in
1973). Such scales are useful for making predictions Figure 23. As for the silicon and germanium semi-
of a variety of semiconductor properties. With a conductors, we may divide up the zinc–blende
knowledge of the real and imaginary parts of the reflectivity structure into five distinct regions. For
dielectric function, we can derive the reflectivity GaAs, the E0 region extends from 1 to 2 eV, the E1
spectrum and reflectivity derivative spectrum. region 2-4 eV, the E90 region 4-5 eV, the E2 region 5-
While the structure of the calculated imaginary 6 eV, and the E92 region 6-7 eV. The lowest energy
part of the dielectric function is similar, the magni- region E0 is dominated by structure originating from
tude in the 2–4 eV region is lower by almost a the fundamental gap at . Spin–orbit interactions
factor of 2. This discrepancy was initially attributed split the upper valence bands of gallium arsenide by
to a failure of the Ehrenreich–Cohen dielectric func- about 0.3 eV; for heavier metal constituents, the split-
tion (Ehrenreich and Cohen, 1959) to include tings can be quite large, for example, 1 eV or more.
20 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
L
Lc6
E vs. k
Λ (Lowest C.B.)
Γ
Γc6
k
Δ
X
Xc6
ER spectrum
(Ga 3d Core)
2
–2
104 ΔR/R
–4
L X L X
–6
20 21 22
Energy (eV)
Figure 19 Fine structure in the electroreflectance spectrum for GaAs (Aspnes, 1976). Assignment to the conduction band
minima is indicated.
30
Ga As
25 Theory
Experiment 30
20 GaAs
25 Theory
15
Experiment
10 20
ε1 (E )
ε2 (E )
5 15
0
10
–5
5
–10
–15 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Energy (eV) Energy (eV)
Figure 20 Real part of the dielectric function for Figure 21 Imaginary part of the dielectric function for
gallium arsenide compared to experiment (Philipp and gallium arsenide compared to experiment (Philipp and
Ehrenreich, 1963). Ehrenreich, 1963).
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 21
GaAs
Theoretical
Experimental
1 dR (eV–1)
A
0
C
R dE
–2
0 1 2 3 4 5 6
Energy (eV)
Figure 24 Reflectivity derivative spectrum of gallium arsenide. See the text for a discussion of the structures: A, B, and C.
Experiment is from Zucca and Shen (1970).
about 0.1 eV higher in energy, it is nearly degenerate density of states extends from the onset of the third
with the E90 structure from and at 4.5 eV. In the valence band (at about 4 eV below the valence band
calculated derivative spectrum, this structure is masked maximum) to the valence band maximum. This
by the stronger M1 critical points, and this may be the region encompasses the top two valence bands and
case in the electroreflectance measurements. is predominantly p-like and is associated with anion
In Figure 26, we illustrate the calculated valence states.
band density of states for GaAs, and the results of
XPS measurements and inverse photoemission
(Chelikowsky et al., 1989). As for the diamond struc- 1.01.4.4.2(iii) Zinc selenide The prototypical II–
ture semiconductors, we may divide the density of VI semiconductor is zinc selenide. However, unlike
states into three general regions. The first region is the prototypical III–V semiconductor (GaAs) or our
the most tightly bound energy band. Electron states prototypical diamond semiconductor (Si or Ge), we
corresponding to this band are strongly localized on do not have as detailed a picture for this class of
the anion and are descendants of the atomic As 4s semiconductors. There are several reasons for this
states. The next region of note is a peak arising from situation. First, extensive experimental information
the onset of the second valence band. This band is lacking for ZnSe as compared to GaAs. The larger
shows almost no energy variation along the X–U bandgap in ZnSe requires higher photon energies
symmetry direction; in fact, it is very flat over the than GaAs for reflectivity measurements. Photon
entire square face of the Brillouin zone. This energy sources at these higher energies have not been rou-
band configuration results in a sharp onset of states tinely available. Also, some of the powerful optical
above the antisymmetric gap. The character of states techniques such as Schottky barrier electroreflec-
associated with the second valence band changes tance cannot be easily applied to ZnSe owing to
from predominantly cation s-like states at the band fabrication problems. Second, the Zn 3d-level resides
edge to predominantly anion p-like states at the close to the valence band. Traditionally, empirical
band maximum. The third region of interest in the pseudopotential band calculations do not explicitly
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 23
Λ6
Δ5
Δ5
Λ4,5
Γ8
GaAs
Λ6 Γ7
4
Δ5
Δ5
Energy (eV)
Λ6
Γ6
Γ8
0
Λ4,5
Δ5
Λ6 Γ7 Δ5
Λ6 Δ5
(0.3,0.3,0.3) Γ (0.5,0.0,0.0)
Λ Δ
Figure 25 Band structure of gallium arsenide near showing the critical point locations for the E90 structure. The indicated
transitions give rise to the structure labeled A, B, and C in Figure 24.
include this level in ZnSe. The 3d cation level in In Figure 28, we display the calculated and
gallium arsenide is significantly below the valence measured reflectivity spectrum for ZnSe
band and in Ge the 3d level is more than 30 eV (Freeouf, 1973; Walter et al., 1970). Some work
below the top of the valence band. using derivative spectroscopy is available (Theis,
In Figure 27, we illustrate the band structure of 1977). Owing to the increased ionic component of
ZnSe. Most features of this band structure can be the bonding in ZnSe, the reflectivity spectrum is
accounted for by extrapolating from Ge to GaAs. shifted to higher energy as compared to Ge or
Compared to Ge and GaAs, the bandgaps in ZnSe, GaAs. It also exhibits much more structure than
that is, the optical gap and the antisymmetric gap, Ge or GaAs because of the larger separation
have increased considerably in size. Moreover, the between bands.
valence bands have narrowed and show less For the most part, the reflectivity spectrum can be
dispersion. analyzed by analogy with GaAs. The E0 reflectivity
24 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
XPS BIS
p-GaAs(110) 1 × 1
X(2) X(11,12)
L(9)
L(3,4) L(8)
X(1)
X(15)
Γ(10,11)
X(7,8)
Γ(1) L(6,7)
X(3,4)
Γ(5) Γ(6–8)
L(1)
X(6)
Γ
–10 –5 0 5 10 15 20
Energy relative to Ev (eV)
Figure 26 Experimental X-ray photoemission spectroscopy (XPS) and Bremsstrahlung isochromat spectroscopy (BIS) or
inverse photoemission spectra for crystalline gallium arsenide (Chelikowsky et al., 1989). The critical point features of the
theoretical density of states are also identified with band structure features.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 25
L4, 5 Γ8
8 Γ8
L6
Γ7 Γ7
6
X7
4 X6
L6
Γ6 Zn Se Γ6
2
Γ8 Γ8
0
L4, 5
Γ7 Γ7
Energy (eV)
L6
X7
–2
X6
–4
L6 X6
–6
–8
–10
L6
X6
Γ6 Γ6
–12
L Γ X k Γ
Wavevector, k
Figure 27 Energy bands for crystalline zinc selenide. The top of the valence band is taken to be the zero energy reference.
structure lies between 2 and 4 eV and corresponds to in the reflectivity might arise from transitions invol-
the minimum energy gap at . The E1 structure at ving the Zn 3d-states.
4–5 eV arises from transitions near to or at the It should not be surprising that the density of
L point. The structure corresponding to the E2 peak states of ZnSe may also be interpreted in terms of
arises from a localized region near the special point an extrapolation from the Ge and GaAs band struc-
(2/a)(3/4, 1/4, 1/4). The E90 structure at 7–8 eV tures. In Figure 29, we compare the calculated
may be an exception having no analogy in GaAs. It density of states for ZnSe with photoemission mea-
lies above the E2 structure; thus, the origin of struc- surements and inverse photoemission (Chelikowsky
ture is reversed in ZnSe as compared to GaAs. et al., 1989). Some of the early pseudopotential band
However, the origin of the E90 structure appears to structures, which had potentials based on optical data
be the same, that from near and along the alone, were in mediocre agreement with the photo-
direction. The highest energy reflectivity structure, emission results. Specifically, local pseudopotentials
E91, occurs at 8–9 eV and corresponds to transitions chosen to reproduce optical gaps tend to overesti-
near the L point. mate the ionicity of the II–VI semiconductors. As a
With the exception of the line shape near the E2 consequence, the valence bands become quite narrow
peak, the pseudopotential results are consistent with compared to experiment. Nonlocal pseudopotentials
experiment. It is possible that the line-shape discre- do not suffer from this malady; they can be used to fit
pancy is an artifact of the sampling scheme used. We optical gaps with no corresponding increase in
note that at energies above 10 eV or so, the structure ionicity.
26 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
0.6
Experiment 1 ZnSe
0.5 Experiment 2
Theory
0.4
R 0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10
Energy (eV)
Figure 28 Reflectivity spectrum for zinc selenide. Two experimental spectra are presented: the dashed line is from Walter
et al. (1970) and the dotted line from Freeouf (1973).
1.01.5 The Ab Initio Pseudopotential interaction and the valence electron–valence elec-
Method tron interaction. A fundamental postulate of the
pseudopotential method is that the ion-core pseudo-
A deficiency of the empirical pseudopotential potential is not dependent on the chemical
method is that the potential is ‘biased’ by the experi- environment. We assume that this part of the poten-
mental data to which it was fit. For example, suppose tial can be transferred with no loss in accuracy. The
we fit the optical properties of GaSb and extract a Ga key problem is to determine the total pseudopoten-
potential, we must be careful in trying to use this fit tial by determining the potentials from the valence
Ga potential in another crystal. This potential might electrons interacting among themselves.
work fine in GaAs, which has a similar bonding A common procedure is to construct a self-
configuration when compared to GaSb, but the consistent potential. The wave functions obtained
potential may fail to describe GaN, which is consid- in an electronic structure calculation can be used
erably more ionic than GaSb. A related problem is to construct a new screening potential, which in
that the form factors are fit only to a subset of reci- turn can be used with the ion-core potential to
procal lattice vectors. If the crystal volume or compute a new total potential and new wave func-
structure changes, one must extrapolate to a new tions. When no changes occur in a feedback loop
subset. Another problem can occur at a surface. In of this kind, the solution is considered to be self-
crystalline GaAs, each Ga atom is surrounded by four consistent.
As atoms. For the cleavage plane the (110) surface, Let us assume a one-electron Hamiltonian, which
the Ga atom is bonded to three As atoms. Given the can be based on density functional theory
coordination change, the surface Ga atom cannot be (Hohenberg and Kohn, 1964; Kohn and Sham, 1965):
expected to retain a ‘bulk-like’ screening potential.
– h2 r2 p
There is no reason to be confident that the Ga þ Vion ðrÞ þ VH ðrÞ þ Vxc ðrÞ n ðrÞ ¼ En n ðrÞ
2m
pseudopotential extracted from the crystalline envir-
onment will be very accurate at the surface. Of ð50Þ
course, this problem is made worse if one wants to
examine a liquid containing Ga atoms where the The ion-core pseudopotential, Vpion, can be taken
coordination may continuously change. as a linear superposition of ion-core atomic poten-
We can consider the form factors to be composed tials. Determining the ionic potential can be
of two interactions: the ion core–valence electron accomplished by resorting to atomic structure
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 27
÷8
= –4 πeρ
Δ 2V
Solve: H
p
Form: VT = Vion +VH +Vxc
n-Zn Se (100) c 2 × 2
2
Δ 2
– p
X(1) L (3,4) Solve: Vion +VH +Vxc Ψn = EnΨn
2m
X(3,4)
Γ(6–8) Γ(15)
X(10)
X(2) L (8)
X(6) 2
Form: ρ = e Ψn
n, occup
addressed according to the system of interest. The This form assures us that the pseudo-wave-function
immediate issue here is how to define pseudo-wave- is nodeless and by taking even powers there is no
functions which can be used to define the corre- cusp associated with the pseudo-wave-function.
sponding pseudopotential. The parameters, c2n, are fixed by the following
Suppose we insist that the pseudo-wave-function criteria: (1) The all-electron and pseudo-wave-
be identical to the all-electron wave function outside functions have the same valence eigenvalue. (2)
of the core region. For example, let us consider the 3s The pseudo-wave-function is nodeless and is iden-
state for a silicon atom. We want the pseudo-wave- tical to the all-electron wave function for r > rc. (3)
function to be identical to the all-electron state out- The pseudo-wave-function must be continuous as
side the core region: well as the first four derivatives of the wave func-
p tion at rc. (4) The pseudopotential has zero
3s ðr Þ ¼ 3s ðr Þ r > rc ð52Þ
curvature at the origin. This construction is easy
where p3s is a pseudo-wave-function and rc defines to implement and extend to include other con-
the core size. straints. An example of an atomic pseudo-wave-
This assignment will guarantee that the pseudo- function for Si is given in Figure 31 where it is
wave-function will possess properties identical to the compared to an all-electron wave function. Unlike
all-electron wave function, 3s in the region away the 3s all-electron wave function, the pseudo-wave-
from the ion core. function has no nodes. The pseudo-wave-function
For r < rc, we alter the all-electron wave function. is much easier to express as a Fourier transform or
We are free to do this as we do not expect the valence a combination of Gaussian orbitals than the all-
wave function within the core region to affect the electron wave function.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 29
1.0
ZR
d 2 ln
– 2 ðr Þ2 ¼ 4 2 r 2 dr ¼ QðRÞ ð56Þ
3s Radial wavefunction of Si dE dr R
0
1 2 1 2 1 2 1 2
–5 –5 –5 –5
r (a0 ) r (a0 ) r (a0 ) r (a0 )
Vl (r) (Ry)
8 8 8 8
q 2Vl (q) (Ry/a02)
0 0 0 0
6 12 6 12 6 12 6 12
–8 –8 –8 –8
Figure 32 Ion-core pseudopotentials for carbon generated by four different methods in real and reciprocal space; (a)
Troullier and Martins (1991), (b) Kerker (1980), (c) Hamann et al. (1979), and (d) Vanderbilt (1985). The dotted and solid lines
correspond to the s and p pseudopotentials, respectively.
size of the core is determined by rc. As such, the element form has a great advantage as the integrals
upper limit of the real space integral is slightly are separable in that they are solely a function of
more than rc in practice. k + G or a function of k + G9. The individual inte-
A major strength of ab initio pseudopotentials is grals can be stored and retrieved as required in
that they can be employed for systems with many setting up the matrix.
atoms of low symmetry, for example, clusters, liquids, It is also possible to solve the Kohn–Sham problem
and surfaces, without having to fit any form factors. directly in real space (Chelikowsky et al., 1994). In this
However, a plane wave basis will require a large case, the Kleinman–Bylander form can be cast as
cutoff for such systems. In this case, the matrix ele- P p
ments in Equation (46) can be difficult to evaluate as VlkB ðx;y;zÞ p ðx;y;zÞ¼ Glm ulm ðx;y;zÞVl ðx;y;zÞ
lm
elements depend on both k + G and k + G9. Z
Kleinman and Bylander (1982) suggested an alternate p
ulm Vl p dxdydz ð58Þ
form for the required nonlocal matrix elements: Glm ¼Z
p p
P ulm Vl ulm dxdydz
þ GÞYlm ðkd
Ylm ðkd þ G9Þ
m
VlKB ðk þ G; k þ G9Þ ¼ Z where uplm are the reference atomic pseudo-wave-
p p
4a l ðr ÞVl ðr Þl ðr Þdr functions. The nonlocal nature of the pseudopoten-
Z
p tial is apparent from the definition of Glm; the value of
l ðr ÞVl ðr Þjl jk þ G9jr Þr 2 dr
Z these coefficients are dependent on the pseudo-
p
l ðr ÞVl ðr Þjl jk þ Gjr Þr 2 dr ð57Þ wave-function, p, acted on by the operator Vl.
This is very similar in spirit to the pseudopotential
where pl is the atomic reference pseudo-wave-func- defined by Phillips and Kleinman (1959). While we
tion and Ylm is a spherical harmonic with the angles have focused on a simple plane wave basis, there are
determined by the unit vector: k d þ G. This matrix other bases that can be employed, for example, one
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 31
can combine pseudopotentials with Gaussians (Chan Chelikowsky, 1989); however, to understand struc-
et al., 1986) or a uniform grid (Chelikowsky et al., tural properties, we need to employ a different
1994). These real space methods can be easily imple- approach. Since the form factors in the empirical
mented with real space ab initio pseudopotentials pseudopotential have been fit to optical transitions,
(Kronik et al., 2006). there is no reason to believe that they will be very
While our focus has been on norm conserving pseu- accurate for structural properties such as the phase
dopotentials, it is possible to generalize the stability, the equilibrium bond length, the bulk mod-
pseudopotential method to include systems that are ulus, and the phonon spectrum of a crystal. Ab initio
not based on such potentials. It is possible to make the pseudopotentials can be used for this purpose and, in
pseudopotentials even weaker by relaxing this condi- this section, we illustrate their application to the phase
tion. Vanderbilt (1990) proposed such a method for stability of tetrahedral semiconductors such as silicon.
constructing ultrasoft pseudopotentials, which are con- To evaluate the total energy of a crystal can be a
structed as a generalized eigenvalue problem. The difficult task. The total energy of the system contains
norm conservation constraint is relaxed, and the charge terms that individually diverge, for example, the
density within the ion core is not explicitly considered repulsive Coulomb terms between the ion cores and
as part of the pseudo-wave-function. The relaxation of the electron–electron interactions. These terms can
the norm conservation constraint allows one to consider be handled in Fourier or momentum space using a
a much larger core radius and a much softer potential. plane wave basis (Ihm et al., 1979). The total energy
One issue which is relevant for pseudopotential of the system can be written as
constructions, regardless of whether the potential is
intended for use with a plane wave basis or not, ET ¼ Ekin þ Eec þ EH þ Exc þ Ecc ð59Þ
concerns the issue of unbound, or weakly bound,
atomic states. If an atom does not bind a state of ET is the total electronic energy, Ekin represents the
interest, then the atomic wave function correspond- electronic kinetic energy, the electron–ion-core
ing to this state is clearly not normalizable. interaction energy is given by Eec, the electrostatic
Nonetheless, the pseudopotential corresponding to Coulomb energy is given by the Hartree energy, EH,
this state might be of some interest, for example, in the nonclassical exchange-correlation energy is given
a crystal such diverging wave functions are captured by Exc, and the ion core–ion-core classical term is
by the potentials of neighboring atoms. For example, given by Ecc. The momentum-space expressions for
Ba has a strong f-component resonance. However, the electronic energy terms are as follows:
these f-states are not bound for neutral Ba atom. In 1 X p 2 h 2 j k þ Gj 2
order to bind such states, one must consider highly Ekin ¼ n ðk þ GÞ ð60Þ
N n;k;G 2m
ionized atomic states, which result in very strong
pseudopotentials. Sometimes these potentials are so 1 X p
Eec ¼ ðk þ GÞ pn ðk þ G9Þ
strong as to be useless for a plane wave basis, or so far N n;k;G n
removed from the chemical environment of interest
Z ð61Þ
that their transferability may be suspect. Hamann p 1 Ze 2 3
Vc ðk þ G; k þ G9Þ þ G;G9 d r
(1989) has suggested a method for handling such c r
cases by integrating out the Kohn–Sham equation c X 4e 2 2
to a large distance and at that point terminating the EH ¼ 2
ðGÞ ð62Þ
2 G;G6¼0 G
pseudo-wave-function. The corresponding termi-
nated wave function is then used to generate a c X
Exc ¼ ðGÞ"xc ðGÞ ð63Þ
pseudopotential for the component of interest. 2 G
( "
e2 X 4 X 1
1.01.5.2 Structural Properties of Ecc ¼ Zv;s Zv;s9 cosðG?ðtS – tS9 ÞÞ
2 S;S9 c G;G6¼0 G 2
Semiconductor Crystals
2 #
1.01.5.2.1 Total electronic energy from –G
exp –
pseudopotential–density functional theory 4
2 c
2
) ð64Þ
Empirical pseudopotentials have been one of the most X erfcð
jR – tS þ tS9 jÞ 2
Each term yields the energy per cell; N is the total The first application involved the silicon crystal
number of cells, c is the cell volume, R is a lattice with a number of different phases: face-centered
vector, G is a reciprocal lattice vector, Z is the total cubic, body-centered cubic, hexagonal close-packed,
core charge, which is a sum of the individual core hexagonal diamond, cubic diamond, white tin and
charges Zv, t is the basis vector, k is the crystal simple cubic. For each structure, the energy was
momentum, and pn is the wave function for state n. minimized for a given volume by optimizing the
ðGÞVcp ðk þ G; k þ G9Þ, and pn ðk þ GÞ represent internal coordinates, for example, for a hexagonal
the Fourier components of the charge, ion-core structure the c/a ratio was optimized. Depending on
potential and the pseudo-wave-function. The sums the system studied the range of volumes considered
involving the pseudo-wave-functions run only over is varied. In the case of silicon. Cohen, et al. con-
the occupied states,
is a parameter that controls the sidered a volume of one-half the ambient volume of
convergence of the Ewald summation in real space the known diamond phase of silicon to an upper
versus momentum space (Ihm et al., 1979). limit of about 20% larger than the ambient volume.
In practice, it is easier to make use of the eigen- For such a diverse set of crystal structures, it is
value explicitly and subtract off the double counting important to make certain the convergence is similar
terms, that is, one can write the total energy as to all phases. For structures which yield metallic
1X systems, the k-point sampling is sometimes critical
ET ¼ En ðkÞ – EH þ Exc þ Ecc ð65Þ
N n;k because the variations in the Fermi surface make it
difficult to assure that occupied states are being
The sum is over all occupied states. The eigen- sampled. Once the total energy is computed for a
value term contains the exchange-correlation term,
finite number of volumes, the energy versus volume
whereas the total energy should contain the
points are fit to an equation of state such as the
exchange-correlation energy. This term can be
Birch (1952) equation or the Murnaghan (1944)
written as
equation.
X
Exc ¼ c ðGÞ½"xc ðGÞ – Vxc ðGÞ ð66Þ !
G B0 V ðV0 =V ÞB90 B0 V 0
ET ðV Þ ¼ ET ðV0 Þ þ þ1 – ð69Þ
B90 B90 – 1 B90 – 1
where "xc is related to the exchange correlation via a
functional derivative: where B0 and B90 are the bulk modulus and its pres-
sure derivative at the equilibrium volume V0. Using
ð "xc ½Þ
Vxc ¼ ð67Þ Equation (69), the fit of the calculated points yields
the equilibrium energy ET(V0), the equilibrium
where, in the local density approximation, volume, and the bulk modulus and its pressure deri-
Z vative at equilibrium. Representative calculated and
Exc ½ ¼ ðrÞ"xc ½ðrÞd 3 r ð68Þ measured values for the lattice constant and bulk
modulus results are given for Si in Table 1. The
In the simplest formalism, "xc is extracted from a results are impressive when one considers that this
homogeneous electron gas (Kohn and Sham, 1965). is an ab initio calculation requiring only the atomic
number and crystal structure as input. Generally, the
pseudpotential–density functional method yields lat-
1.01.5.2.2 Phase stability of crystals tice constants and bulk moduli to an accuracy of
The total energy, ET, of a crystal structure can be about 1% and 5%, respectively. The cohesive energy
calculated using Equation (65) as a function of the can be calculated by comparing the total energy at
lattice constant and any internal structural para- the equilibrium lattice constant with the energy of
meters. One of the first applications of this work the isolated atoms. Spin polarization effects (von
was carried out by Ihm and Cohen (1980), and Yin Barth and Hedin 1972; Gunnarsson et al., 1974) and
and Cohen (1980, 1982c, 1982b). This work repre- zero point vibrational energy need to be considered.
sents a seminal point in condensed matter physics Early results computed within the local density
(Chelikowsky, 2000). It demonstrated a workable approximation gave reasonable estimates of the
scheme for examining the structural properties of cohesive energies; however, in general the general-
solids and its extension led to molecular dynamics ized gradient approximation yields more accurate
using quantum forces (Car and Parrinello, 1985). cohesive energies (Becke, 1992).
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 33
1.01.5.2.2(i) Silicon and germanium phases The Car, 1995), most phase stability studies are done at
total energy versus volume curves for seven crystal zero temperature. In this cases, G ¼ U þ PV, where
phases of silicon are shown in Figures 33 and 34. It is the internal energy is given by the total electronic
encouraging that the lowest energy state corresponds energy and the pressure is given by P ¼ dET/dV.
to the diamond structure, as this structure is observed
experimentally under ambient conditions. (As with
–7.82
any such calculation, there is no guarantee that there
is an untried structure with a lower energy state.)
Ge
Experimental data for the structural properties of
silicon and germanium are listed in Table 3.
For small volumes other structural phases are
–7.84
lower in energy than the diamond phase; hence,
pressure-induced solid–solid structural phase transi-
tions are predicted. Structural transitions occur
Estructure (Ry/atom)
fcc
between phases when the Gibbs free energy hcp
bcc
G ¼ U þ PV – TS ð70Þ –7.86 4
sc
for the phases in question are equal. U is the internal β-Tin
energy, P is the pressure, and S is the entropy. 3
Although some attempts have been made to consider
phase transitions at finite temperature (Sugino and
–7.88
Hexagonal
2 diamond
–7.83 Diamond 1
Si
–7.90
0.6 0.7 0.8 0.9 1.0 1.1
Volume
–7.85
Figure 34 Total electronic energy for seven phases of
crystalline silicon as function of the atomic volume. The
Energy (Ry/atom)
For T ¼ 0, the pressure-induced phase transforma- the transition pressures of the Al compounds are
tion can be found by a common tangent line between underestimated.
the ET(V) curves of the two phases considered. The
negative of the slope of the tangent line is the transi-
tion pressure. Examples of common tangent 1.01.5.2.3 Vibrational properties
constructions are given in Figures 33 and 34 for An important application of the ab initio pseudopo-
the transitions from the diamond structure to the - tential approach is calculating the vibrational
tin phase for Si and Ge. As hydrostatic pressure is properties of solids. These calculations are based on
applied the path indicated in the figures, 1 ! 2 ! 3 the change in the energy of atom as it moves from an
! 4, illustrates the change from diamond at 1 to equilibrium position. Once we have established that
-tin at 4, with the transformation occurring along the energy of atom can be accurately calculated as a
the 2 ! 3 segment of the path where both structures function of position, phonon and lattice vibrational
can coexist. The predicted phase transition pressure modes can be calculated. The input needed for such
for the diamond ! -tin phase is 99 kbar for Si and calculations is minimal. In contrast to phenomenolo-
96 kbar for Ge; the measured values are 125 and 100 gic force constant models, which often require as
kbar, respectively (Yin and Cohen, 1982c). many as 15 parameters to achieve reasonable fits to
The successful prediction of high-pressure phonon dispersion curves, the total energy approach
structural phases such as hexagonal forms of Si and uses only the masses and atomic numbers of the
Ge and face-centered cubic Si are one of the impress- constituent atoms. To calculate the phonon disper-
ive results of the pseudopotential-density functional sion curve, !(q), along a specific direction, the
theory calculations (Chang and Cohen, 1984; Liu frozen-phonon approach is easy to implement
et al., 1988; Vohra et al., 1986). Moreover, these (Chadi and Martin, 1976; Heine and Weaire, 1970;
high-pressure phases are metallic and predicted to Wendel and Martin, 1979; Yin and Cohen, 1980). In
be superconductors (Chang et al., 1985). Two of the this approach, a supercell is chosen, and the atomic
predictions of superconductivity have been verified cores are displaced to simulate a particular phonon
while the others remain untested. mode. The change in the total energy arising from
The calculations for the group IV semiconductors the distortion depends on the type of distortion
serve as prototypes for the application of the total- assumed and the amplitude. Since a specific phonon
energy pseudopotential method to study structural wavevector, q, is chosen, this method is limited to
properties of solids. calculating !(q) at specific points in the Brillouin
zone. The phonon frequencies can also be evaluated
by computing the Hellman–Feynman forces on the
1.01.5.2.2(ii) III–V phases The first extensions to displaced atoms. Another approach that allows the
other solids were made for the III–V semiconductors: full phonon dispersion to be computed is based on
GaAs, AlAs, GaP, AlP, AlSb, GaSb, InP, InAs, and linear response theory (Baroni et al., 2001). However,
InSb (Froyen and Cohen, 1983a, 1983b; Zhang and the frozen-phonon approach is the simplest method
Cohen, 1987). Total electronic energy calculations for computing !(q) at symmetry points. Calculations
for these III–V semiconductors are displayed in at nonsymmetry points are possible but require larger
Figures 35 and 36. unit cells. Standard calculations usually limit the
Although these studies for III–V semiconductors computations to three or four q-points along a
generally give good results for the lattice constants symmetry direction.
and bulk moduli, the agreement between the calcu- As an example of a frozen phonon supercell
lated and measured properties of the high-pressure calculation, we briefly examine the calculations of
phases and for phase transition pressures and the phonon properties for the and X points of the
volumes is not always satisfactory. As an example of Brillouin zone for Si and Ge. The phonon polariza-
the latter, the calculated transition pressures for tions for these points can be easily determined using
the Al compounds are consistently lower by group theory as illustrated in Figure 37, and the
around 50% than the measured values (Zhang and primitive cell for the distorted lattice contains two
Cohen; 1987). In contrast, the calculated lattice atoms for the LTO() mode and four atoms for the
constants are within 0.4, 0.3, and 0.3% of the phonon modes at X. The phonon-distorted lattice
measured values for AlP, AlAs, and AlSb, respec- has inversion symmetry, which simplifies the
tively (Zhang and Cohen, 1987). It is unclear why calculation.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 35
150
AIAs AIP
CsCI
100 +
Rocksalt
β-Sn
Energy - mRy
Rocksalt
β-Sn
50
NiAs
Zinc–blende Zinc–blende
150
GaAs GaP
CsCI
100
Rocksalt
NiAs β-Sn
Energy - mRy
β-Sn Rocksalt
50
Zinc–blende Zinc–blende
0
0.6 0.7 0.8 0.9 1.0 1.1 0.6 0.7 0.8 0.9 1.0 1.1
V/V0 V/V0
Figure 35 Total electronic energy for some representative III–V structures as function of the atomic volume. The
calculations are from Froyen and Cohen (1983a, 1983b).
For a given amplitude u, the change in the total where M is the atomic mass. The calculation of the
energy arising from the distortion is ET(u). The total energies or forces is done typically for five
force constant k can be obtained from a second deri- different amplitudes ranging from 0.01 to 0.1 Å.
vative of the ET(u) or a first derivative of the force The total energies can fit by a quadratic function
F(u): to about 1%. For some phonon mode calculations
such as the LTO() modes in Si and Ge, higher-
@ 2 ET 2ET ðuÞ order fits such as the third-order term are needed.
k¼ ffi ð71Þ
@u2 u¼0 u2 These fits allow the evaluation of anharmonic terms.
or Table 4 illustrates some typical results. An impress-
ive achievement of these calculations is the good
@F F ðuÞ agreement with experiment obtained for the TA(X)
k¼ ffi – ð72Þ
@u u¼0 u modes in Si and Ge. For empirical force constant
model calculations, many-neighbor parameters are
The phonon frequency is given by needed to obtain the correct values for these modes
pffiffiffiffiffiffiffiffiffi because of the important role played by long-ranged
! ¼ 2f ¼ k=M ð73Þ forces.
36 Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories
Rocksalt 120 A5
Energy (mRy/molecule)
Energy (mRy/molecule)
Energy (mRy/molecule)
Rocksalt
60 60 Rocksalt
90
A5
60
30 30
A5
SH
Zinc–blende 30 Zinc–blende
0 Zinc–blende 0
0
InAs InSb
120
90
Rocksalt
Energy (mRy/molecule)
Energy (mRy/molecule)
90
A5
60
60 A5
Rocksalt
30
30
Zinc–blende
Zinc–blende
0
0
–30 –30
0.5 0.6 0.7 0.8 0.9 1.0 1.1 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Normalized volume Normalized volume
Figure 36 Total electronic energy for some representative III–V structures as function of the atomic volume. The
calculations are from Froyen and Cohen (1983a, 1983b).
The use of the total electronic energy or the earlier. This gives the value of the sound velocity.
forces via the Hellmann–Feynman theorem For example, the q ! 0 TA velocity vTA [110]
involves different approaches to calculate the propagating along the [110] direction with [110]
phonon frequencies, but the results are essentially polarization is associated with the shear modulus
the same as indicated in Table 4. Additional C11 C12:
information can be obtained from the force
calculation. For example, for the LOA(X) mode, C11 – C12 1=2
vTA ½110 ¼ ð74Þ
the forces on atoms 2 and 3 differ for finite 2M
amplitude (see Figure 37). This arises from the where M is the mass density. The calculated
effects of anharmonicity and illustrates that it is results for the velocity and shear modulus are in
more difficult to compress a bond than to stretch good agreement with experiment as shown in
it. For the phonon calculation, the average of Table 5.
the two force constants is used, and this quantity When interlayer force constants are calculated,
does not vary by more than 0.5% with the the resulting phonon spectrum is in good agreement
amplitude. with experiment. For example, for Si the to X
If the interlayer force constants are computed, phonon dispersion curve was computed by Yin
then the full dispersion can be calculated as discussed and Cohen (1982a) using the calculated interlayer
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 37
Si
f (energy) 15.16 ( 2%) 12.16 ( 1%) 13.48 ( 3%) 4.45 ( 1%)
f (force) 15.14 ( 3%) 11.98 ( 3%) 13.51 ( 3%) 4.37 ( 3%)
f (expt) 15.33 12.32 13.90 4.49
Ge
f (energy) 8.90 ( 2%) 7.01 ( 3%) 7.75 ( 6%) 2.44 (2%)
f (force) 8.89 ( 3%) 6.96 ( 3%) 7.78 ( 6%) 2.45 (2%)
f (expt) 9.12 7.21 8.26 2.40
Table 5 Comparison of calculated values (Yin and and whether it could be applied to solid and spe-
Cohen, 1982b) of C11–C12 and the TA velocity along the cifically semiconductors such as silicon. Some
polarization) for Si and Ge with experiment
[110] (with [110]
(McSkimin and Andreatch Jr., 1963).
workers thought that the electron–hole interactions
for a optical excitation would be so strong as to
C11 C12 (Mbar) vTA [110] (105 cm s 1) obscure any attempt at understanding such excita-
Si
tions using energy band theory. Given that the
Theory 1.07 4.79 EPM allowed adjustments to the one-electron
Expt. 1.03 4.69 potential, if the EPM had failed to work with the
Ge flexibility of choosing a potential, one would have
Theory 0.74 2.64 questioned whether it was possible to construct a
Expt. 0.82 2.77 meaningful one-electron potential and correspond-
ing energy band structure. The consequences of
this failure would have been very damaging to
prospects for using energy bands to interpret opti-
cal, dielectric, photoemission, and transport
16
properties of semiconductors.
LO Fortunately, this was not the case; the EPM pro-
vided a simple and effective way to couple band
structure theory to experimental work and its impact
TO
was immediate. While much of this effort took place
12 in the 1960s and 1970s, the EPM still serves as an
Phonon frequency (THz)
Heine V and Weaire D (1970) Pseudopotential theory of Phillips JC and Pandey KC (1973) Nonlocal pseudopotential for
cohesion and structure. Solid State Physics 24: 249. Ge. Physical Review Letters 30: 787.
Hellman H (1935) A new approximation method in the problem Pollak FH and Cardona M (1968) Piezo-electroreflectance in
of many electrons. Journal of Chemical Physics 3: 61. Ge. GaAs, and Si. Physical Review 172(3): 816.
Herring C (1940) A new method for calculating wave functions in Rehn V and Kyser D (1972) Symmetry of the 4.5 eV optical
crystals. Physical Review 57: 1169. interband threshold in gallium arsenide. Physical Review
Hohenberg P and Kohn W (1964) Inhomogeneous electron gas. Letters 28: 494.
Physical Review 136: B864. Rohlfing M and Louie S (2000) Electron-hole excitations and
Hybertsen MS and Louie SG (1986) Electron correlation in optical spectra from first principles. Physical Review B
semiconductors and insulators: Band gaps and 62: 4927.
quasiparticle energies. Physical Review B 34: 5390. Rohlfing M and Louie SG (1998) Electron-hole excitations in
Ihm J and Cohen ML (1980) Calculation of structurally related semiconductors and insulators. Physical Review Letters
properties of bulk and surface Si. Physical Review B 21: 1527. 81: 2312.
Ihm J, Zunger A, and Cohen ML (1979) Momentum-space Schwarz W, Andraea D, Arnold S, et al. (1999a) Hans G.A.
formalism for the total energy of solid. Journal of Physics C Hellmann (1903–1938): Part I. A pioneer of quantum
12: 4409. chemistry. Bunsen - Magazin 1: 10.
Kerker GP (1980) Nonsingular atomic pseudopotentials for solid Schwarz W, Andraea D, Arnold S, et al. (1999b) Hans G.A.
state applications. Journal of Physics C 13: L189. Hellmann (1903–1938): Part II. A German pioneer of quantum
Kittel C (2005) Introduction to Solid State Physics 8th edn., New chemistry in Moscow. Bunsen - Magazin 2: 60.
York: Wiley. Sinha SK (1973) Phonons in semiconductors. CRC Critical
Kleinman L and Bylander DM (1982) Efficacious form for model Reviews in Solid State and Materials Sciences 3: 273.
pseudopotentials. Physical Review Letters 48: 1425. Sugino O and Car R (1995) Ab-initio molecular-dynamics study
Kleinman L and Phillips JC (1960a) Crystal potential and energy of first-order phase-transitions – melting of silicon. Physical
bands of semiconductors. II. Self-consistent calculations for Review Letters 74: 1823.
cubic boron nitride. Physical Review 117: 460. Tauc J and Abraham A (1961) Optical investigation of the band
Kleinman L and Phillips JC (1960b) Crystal potential and energy structure of Ge-Si alloys. Journal of Physics and Chemistry of
bands of semiconductors. III. Self-consistent calculations for Solids 20: 190.
silicon. Physical Review 118: 1153. Theis D (1977) Wavelength-modulated reflectivity spectra of ZnSe
Kline JS, Pollak FH, and Cardona M (1968) Electroreflectance in and ZnS from 2.5 to 8 eV. Physica Status Solidi (b) 79: 125.
Ge-Si alloys. Helvetica Physica Acta 41: 968. Troullier N and Martins J (1991) Efficient pseudopotentials for
Kohn W and Sham LJ (1965) Self-consistent equations including planewave calculations. Physical Review B 43: 1993.
exchange and correlation effects. Physical Review 140: A1133. Vanderbilt D (1985) Optimally smooth norm-conserving
Koo J, Shen YR, and Zucca RRL (1971) Effects of uniaxial stress pseudopotentials. Physical Review B 32: 8412.
on E90-peak of silicon. Solid State Communications 9: 2229. Vanderbilt D (1990) Soft self-consistent pseudopotentials in a
Kronik L, Makmal A, Tiago ML, et al. (2006) Parsec – the generalized eigenvalue formalism. Physical Review B 41: 7892.
pseudopotential algorithm for real-space electronic structure Vanderbilt D, Louie SG, and Cohen ML (1986) Calculation of an
calculations: Recent advances and novel applications to nano- harmonic phonon couplings in carbon, silicon, and
structures. Physica Status Solidi (b) 243: 1063. germanium. Physical Review B 33: 8740.
Kunz AB (1971) Energy bands and soft X-rays absorption in Si. Vohra YK, Brister KE, Desgreniers S, Ruoff AL, Chang KJ, and
Physical Review Letters 27: 567. Cohen ML (1986) Phase-transition studies of germanium to
Ley L, Kowalczyk S, Pollak R, and Shirley DA (1972) X-ray 1.25 Mbar. Physical Review Letters 56: 19444.
photoemission spectra of crystalline and amorphous Si and von Barth U and Hedin L (1972) Local exchange-correlation
Ge valence bands. Physical Review Letters 29: 1088. potential for the spin-polarized case. I. Journal of Physics C
Liu AY, Chang KJ, and Cohen ML (1988) Theory of electronic, 5: 1629.
vibrational, and superconducting properties of fcc silicon. Walter JP, Cohen ML, Petroff Y, and Balkanski M (1970)
Physical Review B 37: 6344. Calculated and measured reflectivity of zinc telluride and
McSkimin HJ and Andreatch P, Jr. (1963) Elastic moduli of zinc selenide. Physical Review B 1: 2661.
germanium versus hydrostatic pressure at 25.0 and Welkowsky M and Braunstein R (1972) Interband transitions and
195.8 . Journal of Applied Physics 34: 651. exciton effects in semiconductors. Physical Review B 5: 497.
Murnaghan FD (1944) The compressibility of media under Wendel H and Martin RM (1979) Theory of structural properties
extreme pressures. Proceedings of the National Academy of of covalent semiconductors. Physical Review B 19: 5251.
Sciences of the United States of America 30: 244. Wiser N (1963) Dielectric constant with local field effects
Nilsson G and Nelin G (1971) Phonon dispersion relations in included. Physical Review 129: 62.
germanium at 80 K. Physical Review B 3: 364. Yin MT and Cohen ML (1980) Microscopic theory of the phase
Nilsson G and Nelin G (1972) Study of the homology between transformation and lattice dynamics of Si. Physical Review
silicon and germanium by thermal-neutron spectrometry. Letters 45: 1004.
Physical Review B 6: 3777. Yin MT and Cohen ML (1982a) Ab initio calculation of the
Pauling L (1960) The Nature of the Chemical Bond. Ithaca, NY: phonon dispersion relation: Application to silicon. Physical
Cornell University Press. Review B 25: 4317.
Philipp HR and Ehrenreich H (1963) Optical properties of Yin MT and Cohen ML (1982b) Theory of lattice-dynamical
semiconductors. Physical Review 129: 1550. properties of solids: Application to Si and Ge. Physical
Phillips JC (1973) Bonds and Bands in Semiconductors. New Review B 26: 3259.
York: Academic Press. Yin MT and Cohen ML (1982c) Theory of static structural
Phillips JC and Kleinman L (1959) New method for calculating properties, crystal stability, and phase transformations.
wave functions in crystals and molecules. Physical Review Application to Si and Ge. Physical Review B 26: 5668.
116: 287. Zhang SB and Cohen ML (1987) High-pressure phases of III–V
Phillips JC and Kleinman L (1962) Crystal potential and energy zinc-blende semiconductors. Physical Review B 35: 7604.
bands of semiconductors. IV. Exchange and correlation. Zucca RRL and Shen YR (1970) Wavelength-modulation spectra
Physical Review 128: 2098. of some semiconductors. Physical Review B 1: 2668.
Electrons in Semiconductors: Empirical and ab initio Pseudopotential Theories 41
Further Reading Kittel C (2004) Introduction to Solid State Physics, 8th edn. New
York: Wiley.
Cohen ML and Chelikowsky JR (1989) Electronic Structure and Marder MP (2000) Condensed Matter Physics. New York:
Optical Properties of Semiconductors, 2nd edn. Berlin: Wiley.
Springer. Martin RM (2004) Electronic Structure: Basic Theory and
Karixas E (2003) Atomic and Electronic Structure of Solids. Practical Method. Cambridge: Cambridge University
Cambridge: Cambridge University Press. Press.
1.02 Ab initio Theories of the Structural, Electronic, and
Optical Properties of Semiconductors: Bulk Crystals to
Nanostructures
J Deslippe and S G Louie, University of California at Berkeley and Lawrence Berkeley National Laboratory,
Berkeley, CA, USA
ª 2011 Elsevier B.V. All rights reserved.
1.02.1 Introduction 42
1.02.2 Ground-State Properties and DFT 44
1.02.3 Ab Initio pseudopotentials 47
1.02.4 Selected Applications of DFT to Ground-State Properties 49
1.02.5 Excited States and Spectroscopic Properties 54
1.02.6 Quasiparticle Properties and the Single-Particle Green’s Function 55
1.02.7 The GW Approximation 57
1.02.8 The GW Formalism and Applications 59
1.02.9 Two-Particle Excitations and the Bethe–Salpeter Equation 63
1.02.10 The GW-BSE Formalism and Optical Response 65
1.02.11 Interactions in Nanostructured Semiconducting Materials 71
1.02.12 Summary 73
References 74
42
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 43
wave function, which, in general, would give little 1964; Kohn and Sham, 1965). Such approaches have
or no physical insight. The evaluation of physical been very successful in the ab initio calculation of many
quantities of interest with appropriate approxima- ground-state properties for a wide variety of materials.
tions is then required and desirable. Moreover, the Spectroscopic properties, such as those probed in
majority of the properties of a condensed matter photoemission, tunneling or transport experiments,
system can be determined by investigating just the however, involve creating a specific excited particle
interactions of the valence (outer) electrons with the (electron or hole) above the ground state. Owing to
relatively slower-moving atomic ionic cores (nuclei many-particle interactions, the excited particle (or
plus core electrons) and the interactions of quasiparticle) acquires a self-energy, resulting in a
the valence electrons among themselves. In the modified dispersion and finite lifetime. Calculating
Born–Oppenheimer, or adiabatic, approximation for such quasiparticle properties is an N þ 1 particle pro-
electronic properties, the atomic positions are blem in many-body theory, where N is the number of
assumed to be a set of parameters due to the fact particles in the ground-state problem, that, in princi-
that they are slowly varying compared to motion of ple, requires a different theoretical treatment from
the electrons. This approximation allows the elim- those (such as the standard DFT-based approaches)
ination of two of the terms in Htot, leaving just a used for the ground-state problem (Hedin, 1965;
Hamiltonian for the electron wave function: Hedin and Lundqvist, 1969; Hybertsen and Louie,
1985, 1986). Understanding the optical properties
X p2 X e2 X – Zj e 2 from first principles is an N þ 2 particle (photo-excita-
i
Hel ¼ þ þ : ð3Þ
i
2m jr – ri9 j
i<i9 i i;j
jRj – ri j tion of an electron and hole pair) problem (Rohlfing
and Louie, 1998a, 1998b; Albretch et al., 1998; Benedict
Mechanical, structural, and dynamic quantities asso- et al., 1998). To study optical properties, therefore, it is
ciated with the motion of the atoms (such as phonon necessary to include the electron–hole interaction,
frequencies) can be calculated by evaluating ground- which can have significant effects on the optical
state total energies at different atomic positions. As response in many systems from bulk crystals to nanos-
we will discuss later in this chapter, the explicit tructures where such effects often qualitatively change
dependences of the core electrons can also be the calculated spectra. Spectroscopic properties are
removed through the pseudopotential approach, best discussed and computed in terms of a Green’s
leaving an equation for the valence electrons alone. functions approach (Kadanoff and Baym, 1999),
These interactions determine the electronic struc- where the Green’s functions describe the evolution of
ture of the material which, in turn, determines the the amplitude of a many-body system with the addi-
behavior of the material in the ground state and tion of one or more particles at certain time and
under external perturbations. We therefore focus positions.
the discussion in this chapter on the state-of-the-art Because of the reduced dimensionality of nano-
theoretical and computational methods for calculat- systems, many-electron interaction effects have been
ing these properties from first principles to further shown to be extraordinarily important in the spectro-
our understanding of the structural, electronic, and scopic properties of these systems including band-
optical properties of semiconducting materials from gap renormalization, quasiparticle spectral functions
bulk crystals to nanostructures. such as those measured in photoemission, and optical
It is useful to distinguish the ground-state proper- properties such as those probed in absorption
ties of a semiconductor from its electronic excited- and resonant photon scattering. For example, dom-
state or spectroscopic properties. Properties such as inantly strong excitonic effects (resulting from
structure, cohesive energies, charge densities, phase correlations between the photoexcited quasi-electron
stability, and vibrational properties are ground-state and the quasi-hole) in the optical response of semi-
properties because they are given collectively by all conducting and metallic nanotubes have been
the electrons in the many-body ground state. In gen- predicted by first-principles theory (Spataru et al.,
eral, these properties may be determined from 2004a; Deslippe et al., 2007) and have subsequently
knowing the ground-state total energy and related been confirmed by experiment (Wang et al., 2005,
quantities of the system as a function of the atomic 2007). Experimental advances have also allowed the
positions. Practically, for semiconductors, this is done measurement and characterization of excitation fea-
by applying methods such as those based on the den- tures in the optical response of other nanostructures.
sity functional theory (DFT) (Hohenberg and Kohn, In this chapter, we present some of the fundamental
44 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
ground-state properties (such as the structural and the Hartree energy described above, and Exc is a uni-
vibrational properties) of an interacting many- versal functional, termed the exchange–correlation
electron system. DFT does not approximate the functional, of the density encapsulating everything
ground-state energy by putting restrictions on a trial not included by the first three terms on the right side
many-body wave function; instead, DFT rigorously of Equation (9). The exchange–correlation functional,
reformulates the total energy in terms of a functional Exc[], includes all the effects of many-electron
of the charge density of the system. Within the exchange and correlations and the corrections to the
Kohn–Sham scheme, the charge density can be kinetic energy of the interacting system not captured
expressed exactly in terms of sums over single- by that of the noninteracting system. This formalism is
particle orbitals. As we will see, in practical calcula- a tremendous conceptual and technical simplification
tions, the basic approximation lies in the construction since the unwieldy 3N-dimensional many-electron
of the energy functional itself – no explicit many- wave function is, in principle, eliminated from the
body wave function enters the theory. Today, static problem entirely without any approximation to the
DFT within the Kohn–Sham formulation is arguably many-body total energy or charge density. The major
the most popular approach for first-principles inves- drawback of this approach is that although the exis-
tigations of the ground-state properties of condensed tence of the universal functional (Exc[]) was shown, its
matter systems from bulk semiconductors and metals exact form is not known. A number of highly successful
to nanostructures to complex biological molecules. approximations to Exc[] have been developed over the
For an interacting many-electron system in an years (Dreizler and Gross, 1995; Parr and Yang, 1989),
external static potential V(r), Hohenberg and Kohn including the local density approximation (LDA) in
(1964) demonstrated that, instead of the electronic which the exchange–correlation energy density is
ground-state energy being a functional of the com- taken as that of the homogeneous electron gas at the
plicated ground-state many-body wave function local density or the generalized gradient approxima-
(r2,r2,. . .,rN), it can be reformulated as a functional tions (GGAs) in which the effects of both the local
of the electronic ground-state charge density, , density and its gradients are included. These useful
alone. This functional obeys a variational principle, approximations to Exc[] have allowed practical and
that is, it is a minimal value at the physical electron quantitatively accurate calculations for ground-state
density. Moreover, Kohn and Sham introduced an properties to be made using the DFT approach.
explicit, practical formulation for the total energy Kohn and Sham (1965) showed that the electron
(atomic units are used): charge density, and hence the total energy, may be
Z R obtained by solving an associated system of noninter-
1 ðrÞðr9Þ
Ev ½ ¼ V ðrÞðrÞdr þ Ts ½ þ dr dr9 acting electrons with the same charge density as the
2 jr – r0 j
þ Exc ½
interacting system. In the Kohn–Sham approach,
properties of the original many-body material system
ð9Þ
of interest with interacting electrons are associated
Here, V(r) is the ionic potential, Ts is the kinetic with a fictitious system of noninteracting electrons
energy of a noninteracting electron system with the under the influence of an effective single-particle
equivalent ground-state density (r), the third term is potential Veff (see Figure 1). The equations for
Figure 1 Schematic of Kohn–Sham DFT. The real interacting system density is obtained in terms of single-particle
amplitudes of a fictitious non-interacting system. The energy functional is minimized by varying the effective potential.
46 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
solving for the physical electron density take on the The ground-state total energy as a function of
form of those of a self-consistent field problem. The atomic coordinates yields a large number of other
mapping makes use of the fact that the electron properties such as structural and vibrational para-
density of the noninteracting system is given by the meters. However, as mentioned above, Exc is
single-particle orbitals. The electron density of the unknown, and approximations must be made. A com-
noninteracting system (corresponding to a particular mon approach is to assume that the functional can be
Veff) is then used as an input to evaluate the total written in terms of expectation value of a local
energy functional of the real system, making use of operator
the variational principle that the physical density Z
minimizes the total energy functional given by Exc ¼ ðrÞ"xc ðrÞdr ð13Þ
Equation (9). By varying the noninteracting system
until the functional is minimized, we arrive at the
where "xc, an exchange–correlation energy density, is
physical density and energy of the real fully inter-
assumed to be a function of the density (r) alone in
acting electron system. This variational approach
the local density functional approximation (LDA) or
gives rise to a set of Euler–Lagrange equations
a function of (r) and its derivativesr(r) in GGA.
(the Kohn–Sham equations) that govern the single-
These approximations use data from the interacting
particle orbitals, ji(r), with Lagrange multipliers "i to
homogeneous electron gas problem, for which accu-
ensure orthogonal normality, for the fictitious non- rate numerical results have been obtained and
interacting system: approximate analytic expressions derived. Such
approximations have allowed the very accurate
r2 ab initio computation of many ground-state material
– þ Vion þ VH þ Vxc jnk ¼ "nk jnk ð10Þ
2 properties over the past several decades, in particu-
with lar, those of semiconductors. However, there are
well-known cases, such as systems where van der
X
occ
Waals interactions or strong electron correlations
ðrÞ ¼ jji ðrÞj ð11Þ
i
are important, that remain outside the scope of such
approximate functionals – though, in principle, not
and outside of an exact DFT ground-state formalism.
Exc It is important to note, however, the eigenvalues "i
Vxc ðrÞ ¼ : ð12Þ and eigenfunctions, ji(r), of the Kohn–Sham equa-
ðrÞ
tions (Equations (10)–(12)) do not correspond
The effective, self-consistent potential Veff for the rigorously to the excitation (or quasiparticle) energies
noninteracting system has two terms in addition to and amplitudes of the electrons/holes. Formally, only
the external potential: a Hartree term, VH, which is the electron charge density and total energy are mean-
the same as described above in relation to HF the- ingful quantities within the Kohn–Sham formalism.
ory, and an exchange–correlation term, Vxc, which is The Kohn–Sham eigenvalues are only Lagrange mul-
given by the functional derivative of Exc with tipliers in the Kohn–Sham variational construct, and
respect to the electron charge density. In principle, in general, they are not equal to the electron excitation
if the exact form of Exc were known, a self-consistent (or quasiparticle) energies of a system, even if we
solution to the Kohn–Sham equations would give knew the exact exchange–correlation functional. As
the exact electron density and, therefore, the exact an example, let us consider the case of the interacting
ground-state energy of the fully interacting system homogeneous electron gas. Because the density is
as a function of the atomic coordinates. Here, it is to homogenous, both the Hartree potential plus the
be noted that unlike the HF approximation, Vxc is external potential and the exact exchange–correlation
simply an energy- or state-independent local opera- potential in the Kohn–Sham equation do not
tor within the Kohn–Sham framework. However, depend on the coordinates of the electron, regard-
one should remember that it can have a highly less of the exact form of Vxc. Equation (10) then
nontrivial dependence as a functional of the whole reduces to
density, and that it contains the many-electron
interaction contributions to the kinetic energy that p2
þ const: ji ¼ "i ji ð14Þ
are not captured by Ts[]. 2m
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 47
The solutions to Equation (14) are always plane The use of pseudopotentials dates back to Fermi’s
waves or free-electron-like states. The dispersion work in the 1930s (Fermi, 1934). However, they
relation of the Kohn–Sham eigenvalues is then iden- are now best viewed conceptually in terms of the
tical to that of the free (noninteracting) electron Phillips–Kleinman cancellation theorem (Phillips
gas system. Therefore, if we were to interpret the and Kleinman, 1959). In this view, the valence orbi-
Kohn–Sham eigenvalues as electron quasiparticle tals (e.g., the Kohn–Sham orbitals in Equation (8))
excitation energies, then, the band structure and being orthogonal to the core orbitals may be written
therefore the effective mass, m, and occupied band- in the form of
width would be identical to those of free electrons, X
completely independent of the magnitude of the jji ðrÞi ¼ ji ðrÞi þ jfc ihfc ji i ð15Þ
c
interaction. In addition, the lifetime of a quasi-
electron or quasi-hole produced in such a system where ji ðr Þi is a smooth function (pseudo-
would be infinite. This is clearly incorrect. It was wavefunction) and jfc i are core electronic states
a similar misuse and misinterpretation of the or properly normalized sums of these states. The
Kohn–Sham eigenvalues (i.e., using the Kohn–Sham single-particle Schrodinger equation for the valence
eigenvalue gap as the quasiparticle excitation electrons can then be constructed within one of the
gap) that led to the famous band-gap problem in theories described in the previous section as
semiconductors. Despite this caveat, however, the
Kohn–Sham band structure nevertheless does often p2
– þ V ji ¼ "i ji ð16Þ
provide a reasonable starting point for the qualitative 2m
understanding of the electronic structure of semicon- where V is the standard independent-particle poten-
ductor materials. tial from the Kohn–Sham formalism, for example.
Solving this equation is then equivalent to solving
the following equation for the pseudo-wave-function
ji ðrÞi:
1.02.3 Ab Initio pseudopotentials
p2
– þ V þ VR i ¼ "i i : ð17Þ
Since the valence electrons of the constituent atoms 2m
determine most of the properties of a material and
the core electrons are often inert from their envir- The additional term, VR, given by
onment, it is possible to greatly simplify Equation X
VR ¼ ð" – "c Þjfc ðrÞihfc ðr9Þj ð18Þ
(3) by describing the interaction between these c
outer electrons with the combined nucleus and
is, in principle, a nonlocal, energy-dependent opera-
core electrons near an atom in terms of pseudo-
tor. However, since the core state energies are, in
potentials. The concept of pseudopotentials
general, significantly lower than those of the states
explains the apparent weak interaction between
of interest, VR is effectively a repulsive potential
the active electrons and the ionic cores in solids,
with negligible energy dependence. Thus, VR effec-
explaining the many successes of the nearly-free-
tively cancels the strong electron-ion interaction
electron model in describing simple metals.
potential, V, near the core region, resulting in a
Further, the use of pseudopotentials greatly
net potential seen by the valence electrons
improves the efficiency in electronic structure
that is weaker than the original potential. This
calculations of the properties of real materials.
weak net potential is what is termed the
The use of pseudopotentials removes the core
pseudopotential:
electrons from the calculations, reducing the com-
plexity of the problem. A weaker pseudopotential and Vp ¼ V þ VR : ð19Þ
nodeless pseudo-wave-functions result in smooth
eigenfunctions, greatly reducing the size of the basis The solutions to Equation (17) have the same
set required to study the valence electrons. The latter eigenvalues as the original Equation (16), and have
simplification results from the ability to remove the wave functions that are similar to those of the all-
part of the all-electron valence electron wave function electron wave functions outside of the core region
that oscillates rapidly in order to remain orthogonal to but significantly smoother near the core. The
the core states (Phillips and Kleinman, 1959). pseudopotential concept and methodology justify
48 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
why the interaction between the valence electrons p2
– þ Vpi i ¼ "i i ð20Þ
and the crystal potential appears to very weak in 2m
many systems, and provides a number of computa-
to arrive at the corresponding pseudopotential Vp
tional advantages such as the removal of the core
for the ith state of this particular element. The
electronic degrees of freedom in the calculation and
procedure is usually done for a neutral atom in a
a great reduction in the size and complexity of
configuration that is appropriate in the condensed
the necessary basis sets to expand the pseudo- state. An ab initio ionic pseudopotential describing
wave-function, since these are smooth functions. the intrinsic interaction of an electron with an
The pseudopotentials described above are not atom stripped of the other outer valence
unique since a different choice for the second term electrons is obtained by unscreening the neutral
on the right-hand side of Equation (15) would lead to pseudopotential, that is, removing the screening
equally valid but significantly different-looking Hartree and exchange–correlation potential
pseudopotentials. This ambiguity has led to many resulting from the valence electrons themselves.
different construction methods in order to optimize (The interactions with the other valence electrons will
the accuracy and computational efficiency of be treated explicitly in subsequent calculations on
pseudopotentials (Pickett, 1989). In general, modern real materials since these effects are environment
practical construction of ab initio pseudopotentials is dependent.) Various other constraints (such as
done by first solving the interacting all-electron pro- norm conservation; Hamann et al. (1979)) to ensure
blem for a given atom of interest within one of the accuracy and transferability can be implemented in
self-consistent theories described in the previous sec- this procedure for the construction of pseudo-
tion, such as DFT in the LDA. The resulting potentials (Pickett, 1989). The resulting ab initio
all-electron valence orbitals ji are used to form a ionic pseudopotentials are, in general, highly accu-
set of pseudo-wave-functions, i, using Equation (15) rate and transferable to different environments
implicitly, by joining j to a properly chosen smooth, where the atoms are not isolated (i.e., where the
nodeless function for small distances from the valence electron distribution differs greatly from
nucleus (called the cutoff radius), as illustrated in the isolated atoms) and have been successfully
Figure 2 for the case of the 3s and 3p states of used in first-principles calculations on materials
sodium. One then inverts the Schrödinger equation systems for many different properties in different
(for the constructed i and the true "i), environments.
1
Na s-Potential
3s
0.4
0
Wave functions
Potential (Ry)
p-Potential
3p –1
0.2
–2
d-Potential
–3
0
–4 All electron
–0.2 –5
0 1 2 3 4 5 0 1 2 3 4
r (a.u.) r (a.u.)
Figure 2 (Left) Constuction of ab initio pseudopotentials. The all-electron wave functions for the 3s and 3p states (dashed
lines) oscillate rapidly near the origin because of the orthogonality requirement to the core states. The pseudo-wave-functions
are constructed by replacing this oscillatory part of the wave function with a smooth, nodeless function. By inverting
Equation 20, one finds the corresponding pseudopotentials shown in the (right) panel. Figure courtesy of J.R. Chelikowsky.
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 49
1.02.4 Selected Applications of DFT the atomic wave functions. From the charge density, the
to Ground-State Properties Hartree and exchange–correlation potentials are con-
structed, and the Kohn–Sham eigenvector/eigenvalue
The use of the ab initio pseudopotential approach problem is then solved (typically through an iterative
within the Kohn–Sham formalism of DFT has diagonalization method) to get a set of new Kohn–Sham
proven to be a powerful approach for calculating eigenvalues and orbitals. The process is repeated
the ground-state properties of many materials until the change in orbitals or eigenvalues between
(Cohen, 1982) from bulk semiconductors and metals iterations is lower than a predefined convergence
to nanosystems such as molecules, wires, and sheets. criteria (i.e., when self-consistency is reached). The
There are now a number of DFT computational final products are the self-consistent Kohn–Sham
packages available that vary in the types of eigenstates of the associated noninteracting system
exchange–correlation functionals, basis sets, pseudo- from which the final electron density is obtained so
potentials, and numerical algorithms used. These that the electron ground-state energy of the physical
packages share the general computational flow struc- system may be evaluated using Equation (7) within a
ture of a typical Kohn–Sham calculation laid out in given approximation to the exchange–correlation
Figure 3. The input parameters of such a calculation functional used (e.g., the LDA).
are typically the atom types, initial structural In this section, we give several selected examples
arrangement, and methods for construction of of DFT calculations of ground-state properties.
pseudopotentials. Convergence parameters for the Figure 4 is taken from the classic work of Yin
basis set such as the plane-wave energy cutoff and Cohen (1980), who first showed that structural
(maximum energy plane wave to use in the bases) or energies (as a function atomic positions) may be
basis set components in the case of localized basis orbi- accurately determined using the ab initio pseudopo-
tals are specified from which an initial set of Kohn– tential density functional approach. The total energy
Sham orbitals and charge density are constructed using of the system is given by
–7.84 –7.82
Si Ge
–7.86 –7.84
tcc
hcp
Estructure (Ry/atom)
Estructure (Ry/atom)
bcc
tcc
hcp
4 bcc
–7.88 4 sc –7.86
β-TIN sc
β-TIN
3 3
–7.92 –7.90
0.6 0.7 0.8 0.9 1.0 1.1 0.6 0.7 0.8 0.9 1.0 1.1
Volume Volume
Figure 4 Phases of Si and Ge under pressure. The common tangent line between the E vs. V curves show the point where
the Gibb’s free energy at zero temperature (EþPV) of the two phases is the same, since P ¼ -dE/dV. The slope of the line gives
the transition pressure and the change in volume across the phase transition can be determined from the volume of the two
phases at the transition pressure. Modified from Yin MT and Cohen ML (1980) Microscopic theory of the phase-transformation
and lattice-dynamics of Si. Physical Review Letters 45: 1004.
including liquid phases have been done using the frozen phonon approach (Yin and Cohen, 1980),
ab initio MD approach. Below, we shall consider the one calculates the energy difference in the presence
simpler case of pressure-induced phase transition of a finite atomic distortion of the form
obtainable through the application of the standard ui ðkÞ ¼ u0 cosðk?i þ k Þ frozen into the lowest
DFT approach. energy structure. k here is an overall phase shift of
It is straightforward, utilizing the equation of state the distortion. By computing the change in the total
for the various structures, to determine the behavior energy of the system, E(uo), as a function of the
of structural phase transitions under pressure by distortion amplitude uo, one obtains the phonon
evaluating the Gibbs’ free energy (Kittel and properties via a power series expansion
Kroemer, 1980):
E ¼ K2 uo2 þ K3 uo3 þ K4 uo4 þ ð26Þ
G ¼ E þ PV – TS ð25Þ
where the harmonic contribution to the phonon
In particular, at low temperature, the critical transi- frequency is given by the coefficient K2 and the
tion pressure between two adjacent structures (as higher-order coefficients give rise to anharmonic
depicted in Figure 4) is given by the negative of contributions to the phonon frequency and to pro-
the slope of the common tangent (pressure) of the cesses such as phonon lifetime and phonon–phonon
two equations of state, yielding the point where interactions. This ability to easily compute the
the Gibb’s free energy of the two phases is the anharmonic contributions is a major advantage of
same. This leads to the determination of the critical the frozen phonon methodology. However, the size
pressure and the volume discontinuity at the phase of the supercell required to contain the distortion
transition and provides an ab initio understanding depends on the wavelength of the distortion; only a
and prediction of the high-pressure phases of matter finite set of discrete phonons (those of relatively
such as those of Si and Ge as shown in Figure 4 small wavelength, which are commensurate with
(Cohen, 1982). the lattice) can be calculated. Points away from
The preferred structure of a material system such symmetry points or high symmetry directions,
as that at a surface or near a defect can be found by which lead to a very large supercell, may not be
minimizing the total energy or the forces on the studied in practice.
atoms. This approach has been used with success in Another approach to calculating phonons involves
determining the structural parameters of complex the use of the density functional perturbation theory
materials, surfaces, defects, clusters, molecules, and formalism to create the dynamical matrix. In this
nanostructured systems of interest under imposed scheme, the atomic force constants,
boundary conditions. For nonperiodic systems, the
@2E
standard technique is to carry out the electronic Cij ¼ ð27Þ
@Ri @Rj
structure calculations employing a supercell scheme
(Cohen et al., 1975) in which the surface, defect, or are evaluated using linear response theory (Baroni
molecule is repeated periodically with sufficient et al., 2001; Yin and Cohen, 1982). The resulting
separation between them in order to mimic the iso- dynamical matrix can be diagonalized to obtain the
lated structure. As we will discuss more below, for phonon frequencies and eigenmodes. This approach
studying isolated nanostructures, a way to minimize allows the direct computation of phonon properties
the vacuum size needed is to replace the Coulomb at arbitrary wave vector, k, in the Brillouin zone, but
interaction by a new interaction that is the same as is limited to the harmonic approximation due to the
the Coulomb interaction within a cutoff radius but use of linear response theory. Figure 5 shows
zero outside of the radius. the results of such a linear-response calculation for
In addition, from information on the total energy the phonon dispersions and density of states for
of the system as a function of the atomic positions as silicon and germanium. In practice, both frozen pho-
well as the forces, we can study the lattice dyna- non and linear response approaches have given very
mics within the Born–Oppenheimer approximation. accurate phonon results for a variety of materials
There are two common schemes in determining ranging from metals to semiconductors to nanosys-
the phonon eigenmodes and dispersions from tems and molecules. This shows again the power of
first principles: (1) the frozen phonon method or (2) the ab initio pseudopotential DFT Kohn–Sham
the linear response method. Conceptually, in the formalism for calculating ground-state properties.
52 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
600
Frequency (cm–1)
400 Si
200
0
Γ K X Γ L X W L DOS
400
300
Frequency (cm–1)
Ge
200
100
0
Γ K X Γ L X W L DOS
Figure 5 Phonons dispersion relations and density of states (DOS) in Si and Ge calculated within the linear response theory.
After Giannozzi P, Degironcoli S, Pavone P, and Baroni S (1991) Ab-initio calculations of phonon dispersions in semiconductors.
Physical Review B 43: 7231.
In the study of nanostructures, the predictive importance of DFT over lower-level theories such
power of the ab initio pseudopotential density func- as a simple -orbital-only tight-binding model is
tional method has led to many important predictions. illustrated in the structural dependence of the band
We give one illustrative example here – the graphene gap. The qualitative differences between the two sets
nanoribbons. Graphene, a single atomic layer of gra- of results arise from changes at the boundaries of
phite with a honeycomb structure, has generated the structure that are not captured in tight-binding
considerable excitement since its isolation and study models based on pristine graphene. In carbon nano-
in 2005 (Novoselov et al., 2005; Zhang et al., 2005). tubes (i.e., graphene nanoribbons rolled into tubes),
This 2D system exhibits many highly unusual proper- somewhat similar corrections to the tight-binding
ties and is a promising material for future electronics results from relaxing the coordinates with respect
and other applications (Geim and Novoselov, 2007; to the total energy are found and termed cur-
Castro Neto et al., 2009). If cut into strips of nanometer vature effects (Blasé et al., 1994; Charlier et al., 1994,
widths with the dangling -bonds passivated with 1995; Louie, 1994). As we will see in the following
adatoms, the resulting homogeneous edge graphene sections, the DFT band structure, however, still needs
nanoribbons (e.g., with either zigzag shaped or to be corrected to represent the quasiparticle
armchair-shaped edges) are predicted to be all semi- dispersion.
conductors (Son et al., 2006a), unlike the case of Another interesting result from DFT calculations
carbon nanotubes (Saito et al., 1998). The semicon- is that the zigzag graphene nanoribbons have a semi-
ducting nature of the graphene nanoribbons has been conducting antiferromagnetic ground state that can
confirmed by recent experiments (Li et al., 2008). be made into a half metal by an applied external
Figure 6 shows the calculated gaps as a function of electric field (Son et al., 2006b). Figure 7 shows the
ribbon width for the armchair nanoribbons. The predicted ground-state spin-charge density for a
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 53
Δa(eV)
1.5 1.5 Δ3p
Δ3p + 1
1.0 1.0
Δ3p + 2
0.5 0.5
0.0 0.0
0 10 20 30 40 50 0 10 20 30 40 50
Figure 6 The Kohn–Sham electronic gap () of armchair graphene nanoribbons as a function of ribbon width between tight-
binding (left) and DFT LDA (right) calculations. The difference stems from the structural relaxation at the edges and a more
accurate treatment of electron–electron interaction effects. N is the number of rows of C-C dimers forming the width of the
ribbon. Armchair nanoribbons can be broken into three families with N¼3p, 3pþ1, or 3pþ2 for integer p. Note that the LDA
predicts a finite gap for the first family which is qualitatively different from the tight-binding result which predicts zero gap.
Modified from Son Y-W, Cohen ML, and Louie SG (2006a) Energy gaps in graphene nanoribbons. Physical Review Letters 97:
216803.
(a) 2 (b)
1
E–EF (eV)
0
–1
Electric field
–2 1.4
–3 0.0
0 π
–1.4
k(1/a)
(c) 2
1
E–EF (eV)
16-zigzag nanoribbon (16-ZGNR), which has 16 zig- one band localized on one edge and the other band
zag chains forming the width (Son et al., 2006b; on the opposite edge. The edge-state electrons are
Yang et al., 2007b). The DFT calculations show ferromagnetically coupled along one edge and anti-
that, for this system, the bands near the gap are ferromagnetically coupled across the ribbon. By
twofold degenerate spin-polarized edge states with applying a transverse electric field, thus creating a
54 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
LIF
spin orientation will be pushed into the band gap and
edge states with spin of opposite orientation will be
Fluorite
pushed away from the gap. In this a way, it is possible
to create a half metal with a 100% spin-polarized
LiCl
strength that would close the gap or by doping or Quasiparticle theory
gating the system with carriers in the presence of a
c-BN
Diamond
weaker applied transverse field.
w-AIN
w-BN
SiC, Fluorite, GaP, AIP
5
1.02.5 Excited States and
w-GaN
Znlo, ZnSe
InS
CdSe, CdTe, BP
c-GaN
CdS, AlSb
Spectroscopic Properties Many-body
corrections
AlAs
InP GaAs
Go GaSb
While DFT has proven to be an extremely useful and
Si
LDA
InAs
InSb
accurate method in calculating the ground-state
0
properties of semiconductors and many other con-
densed matter systems, the Kohn–Sham formalism 0 5 10 15
within static DFT does not generally give accurate Experimental band gap (eV)
excited-state properties such as band gaps, quasipar- Figure 8 Comparison of calculated energy band gaps
ticle energies, and optical spectra. Extensions of DFT with those measured by experiments. The diagonal line
to time-dependent density functional theory corresponds to a perfect agreement between experiment
(TDDFT), in principle, would be able to yield cer- and theory. Kohn–Sham band gaps calculated within LDA
(crosses) consistently underestimate the experimental
tain spectroscopic properties such as the optical
bandgap, whereas the GW quasiparticle gaps (diamonds)
absorption spectrum (Reining et al., 2002). In practice, consistently give good agreement with experiment.
however, approximations to date to the time-
dependent exchange–correlation functional have
restricted these studies to small finite systems such
as atoms or molecules. For bulk semiconductors and correlation effects systematically reduces the band
extended semiconducting nanostructures, current gaps since correlations lower the energy required to
implementations of TDDFT (such as those based create an excited electron or hole. This band-gap
on the adiabatic approximation) yield results, which problem and other similar problems that exist in
converge to those of independent particle transitions comparing calculated Kohn–Sham eigenvalues to
for the optical response and thus do not provide any experimental excitation energies do not stem from
improvements. This is due to an inadequacy in the the approximation of the exchange–correlation
description of both quasiparticle self-energy and functional in DFT, but arise from a fundamental
electron–hole interaction effects within current and conceptual difference between the theoretical
approximations, although considerable efforts are and measured quantities. As discussed in previous
being expended to make further progress (Reining sections and illustrated by the example of the homo-
et al., 2002; Botti et al., 2004). geneous interacting electron gas, the Kohn–Sham
Figure 8 shows a comparison of the calculated eigenvalues are merely Lagrange multipliers within
electronic band gap with experimental quasiparticle a variational derivation of the conditions for mini-
band gaps for a range of semiconductors and mizing the total energy – that is, they do not
insulators. The LDA Kohn–Sham eigenvalues physically correspond to the true quasiparticle ener-
underestimate the gaps by as much as 50% or gies in general.
more. For germanium, even the band topology is For an interacting system, it is important to dis-
qualitatively incorrect; LDA predicts Ge to be a tinguish the different kinds of excitations probed
metal instead of a semiconductor with a gap of in various experiments. Measurements employing
0.7 eV. HF theory, which often does well for small photoemission, transport, or tunneling techniques
molecules and atoms where screening is unimpor- yield, in general, information on individual particle-
tant, consistently overestimates the band gaps of like excitations or quasiparticles (excited electrons
semiconductors and insulators – the addition of and holes) of a system. For example, photoemission
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 55
(or inverse photoemission) probes a system with a properties of condensed matter systems. It is now
single additional hole (or electron) of a certain wave possible to accurately calculate the quasiparticle
vector and energy. Due to many-electron interac- and optical spectra of many systems from first prin-
tions, the excited particle acquires a self-energy ciples, in particular, those with moderately correlated
which changes its dispersion relation from that of electrons. The approach involves solving for the
the independent-particle or mean-field picture, and single-particle and two-particle Green’s function of
the excitation acquires a finite lifetime. In an optical the interacting-electron system. The advantage of
experiment, on the other hand, one creates neutral the approach is that one can obtain the spectroscopic
correlated electron–hole pair excitations. The properties including the relevant self-energy and
electron–hole interaction (excitonic) effects become electron–hole interaction effects without any empiri-
important and can in fact be dominant in many cal parameters and yet still be applicable to real
systems, particularly nanostructures. A treatment of materials. Applications of this approach have
excitonic effects requires an effective two-particle explained and predicted the spectroscopic properties
approach on top of the quasiparticle picture. Self- of a number of semiconducting systems from bulk
energy, excitonic, and other many-electron interac- crystals to nanostructures.
tion effects exist in all semiconductor systems to
various degrees. They are, however, particularly
relevant in understanding the spectroscopic proper- 1.02.6 Quasiparticle Properties and
ties of many systems of contemporary interest, the Single-Particle Green’s Function
including clusters, surfaces, polymers, defects, nano-
tubes, and other lower-dimensional materials since To predict the spectroscopic properties of semicon-
the Coulomb interaction is enhanced in reduced ductors and other systems, it is convenient to
dimensions. formulate the problem in terms of quasiparticles.
The behaviors of the single-particle-like excita- A direct way to calculate quasiparticle properties is
tions in semiconductor materials are best described through a Green’s function formalism of the many-
in terms of the quasiparticle concept (Hedin, 1965; body problem. The single-particle Green’s function
Hybertsen and Louie, 1985; Kadanoff and Baym, describes the propagation of the many-body system
1999). It is mostly transitions between the quasi- after a single particle is added at time t ¼ 0 and
particle states that are probed in spectroscopic location r. For an interacting many-electron system,
measurements. Because of electron–electron and it is defined as
other interactions in a solid, an excited particle of
wave vector k is dressed with a screening cloud of Gðr; r9;t Þ ¼ – ih0jT fcðr; t Þcy ðr9;0Þgj0i ð28Þ
electrons and other excitations (e.g., phonons or spin
waves) resulting in a different energy E, effective where the c’s are electron field operators, T the time
mass m, and a finite lifetime. In this chapter, we order operator, and j0i is the many-electron ground-
focus on the effects of electron–electron interaction state wave function. A useful alternate form is the
since they give rise to eV-scale modifications in the diagonal elements of G in an orbital basis
electronic excitations in semiconductors. An accu- representation
rate treatment of the exchange and dynamical
correlation effects, arising from the Pauli exclusion Gðp; t Þ ¼ – ih0jT fCp ðt ÞCpy ð0Þgj0i ð29Þ
principle and Coulomb repulsion, seen by an elec-
tron is essential in calculating the quasiparticle where p denotes the quantum numbers of a single-
excitation energies and properties. The understand- particle state, for example, the band index n and wave
ing of optical properties further involves the vector k. Physically, the Green’s function is related to
interaction between the excited quasi-electron and the probability of finding a particle with quantum
the hole that is left behind, which is mediated number p at time t if one was injected into the system
through all the other electrons in the system at time t ¼ 0. While the many-body ground-state and
(Strinati, 1988; Rohlfing and Louie, 1998a, 1998b; excited-state wave functions are extremely complex
Albretch et al., 1998; Benedict et al., 1998). and impractical to express explicitly in terms of the
As in the case of ground-state properties, tremen- 3N coordinates of the N electrons in a system, the
dous progress has been made in the past two decades single-particle Green’s function depends on only two
toward calculating the electronic excited-state space–time coordinates and is, thus, substantially
56 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
where Enk and cnk(r) are the complex solutions to the This physical description of the single-particle
quasiparticle Dyson equation excitations of an interacting electron system relies
Z on the lifetime of the quasiparticles being suffi-
Enk – H0 ðrÞ – VH ðrÞ cnk ðrÞ – ðr; r9; Enk Þcnk ðr9Þ ciently long on the time scale of the relevant
experimental probes that is, the widths of the
dr9 ¼ 0: peaks in the Green’s function should be sufficiently
ð36Þ small so that there are well-defined peaks in
the measured spectra. Within Fermi liquid theory,
Here, n and k are again the band and wave vector for excitation energies near the Fermi energy of a
quantum numbers describing the quasiparticle exci- metal or the fundamental energy gap region of a
tations in a crystal. The positions of the poles of the semiconductor or insulator, the excitations are
Green’s function G(!), hence, may be identified with long-lived and the quasiparticles are well-defined,
the complex solutions of Equation (36) and, as dis- allowing us to pursue this description (Mahan, 1981;
cussed above, are connected to the excitation energy Figure 9).
and lifetime of the quasiparticles. The quasiparticle
band structure and properties, thus, may be obtained
by solving Equation (36).
The quasiparticle equation, Equation (36), is 1.02.7 The GW Approximation
similar in form to the Schrödinger equation in one-
electron or self-consistent-field theories (such as the As discussed above, the Dyson’s equation, Equation
Kohn–Sham or HF formalism). However, in the (36), may be solved to yield the quasiparticle
Dyson’s equation, ðr; r9; !Þ is a nonlocal, non- energies, lifetimes, and amplitudes. To make the
Hermitian and energy-dependent operator, leading approach practical for real materials, we need an
to complex eigenvalues, Enk. As seen in Equation accurate and computationally tractable method to
(32), the real part of Enk corresponds to the quasipar- construct the nonlocal, non-Hermitian self-energy
ticle energy and the imaginary part gives the lifetime. operator ðr; r9;!Þ. It was shown by Hedin that
We see that this corresponds to a Green’s function can be systematically expanded in a series in
with poles off the real axis, giving rise to peaks of terms of the dressed Green’s function G and the
finite width in the functions G(k, !) and A(p, !) along screened Coulomb interaction W (Hedin, 1965;
the real ! axis. In practice, since Enk is often com- Hybertsen and Louie, 1985):
puted within perturbation theory, it is useful to write Z
the complex quasiparticle energy as a sum of an W ðr; r9; !Þ ¼ " – 1 ðr; r0; !ÞVc ðr0; r9Þdr0 ð38Þ
unperturbed non-interacting energy "nk plus a com-
plex self-energy nk containing the many-electron where "ðr; r9; !Þ is the full r; r9-dependent time-
(exchange–correlation) effects: ordered dielectric matrix of the system
(see Figure 10). The advantage of this expansion in
Enk ¼ "nk þ nk ð37Þ terms of the screened Coulomb interaction over an
Σ = 2 1 + 2 1
+ 2 1 + 2 1 + 2 1
+ 2 1 + 2 1 + 2 1...
Figure 10 Diagramatic expansion of the self-energy in terms of the screened Coulomb interation.
58 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
expansion in terms of the bare Coulomb interaction in determining the many-electron effects on the
Vc is that W is, in general, much weaker than Vc due quasiparticle properties, within the GW approxima-
to the dielectric screening. For example, the static tion, resides in the dielectric response function or
dielectric constant of a typical semiconductor is of matrix "ðr;r9; !Þ, which is a two-point function in
the order 10. Thus, a series expansion in W should the spatial coordinates and depends on frequency.
converge with fewer terms than one in Vc. An alter- It has been shown that both the nonlocality and
native view is that, since W and the dressed Green’s frequency-dependence of the screening matrix play
function themselves can be expressed as series expan- central roles in obtaining accurate quasiparticle
sions in the bare quantities, each diagram in properties in semiconductors (Hybertsen and
Figure 10 is a partial summation over terms in a Louie, 1985).
conventional expansion, leading to more accurate The construction of the dielectric response func-
physical results with only low-order terms. tion "ðr; r9; !Þ and the single-particle Green’s
In practice, the Dyson’s equation (Equation (36)) function G require the greatest effort in a typical
is typically solved within a perturbation theory GW calculation. Both have to be treated adequately
approach. In perturbation theory, it is greatly bene- to obtain quantitative results that may be compared
ficial to start from a mean-field scenario that is as with experiment. As mentioned above, if W is
close as possible to the physical system (Hybertsen replaced by the bare Coulomb interaction in
and Louie, 1985). The mean-field Hamiltonian that Equation (39), then Equation (36) reduces to the
is typically used in quasiparticle studies for semi- HF equations. From this point of view, the HF
conductors is the Kohn–Sham DFT Hamiltonian eigenvalues may be considered as a lowest-order
HKS within either the LDA or GGA, which already approximation to the quasiparticle energies, consis-
contains the effects of exchange and correlation in tent with Koopman’s theorem. The dielectric
an averaged way. Perturbation theory calculation is function "ðr; r9; !Þ provides the dynamical and spa-
then performed on H ¼ H – HKS ¼ - Vxc. It is tial-screening response of the electrons that gives rise
important to point out, however, that the quasipar- to correlation effects going beyond bare exchange.
ticle approach does not depend upon DFT, but Similarly, the eigenvalues from the Kohn–Sham
only utilizes it as a convenient starting point for equations in the density functional formalism may
many systems. For some systems, such as materials be viewed as another set of very approximate quasi-
with more correlated electronic states, other start- particle energies, with the exchange–correlation
ing mean-field Hamiltonians may be used such as potential Vxc(r) approximating the nonlocal, energy-
those from LDAþU or the hybrid functionals dependent self-energy operator in Equation (36).
(Miyake et al., 2006); or one may use some self- However, we must emphasize that, as discussed
consistent scheme to determine the best mean-field above, in principle, both the HF and Kohn–Sham
(van Schilfgaarde et al., 2006). In the studies of real eigenvalues are only Euler–Lagrange parameters in
materials, it is found that, for most cases, only minimizing the total energy in ground-state theories,
the first term of the series in Figure 10 for the and as illustrated in Figure 8, they are usually
self-energy operator is necessary, yielding what is not accurate enough as approximations to the quasi-
called the GW approximation (Hedin, 1965), since particle energies. It is only in the extent that
when expressed to first order in G and W is the exchange–correlation operators within self-
given by consistent field theories actually do approximate
Z that these theories approximate the solution of the
d! – i!
ðr; r9; EÞ ¼ i e Gðr; r9; E – !ÞW ðr; r9; !Þ ð39Þ Dyson’s equation. In fact, different formulations of
2
the DFT within the Kohn–Sham schedule can lead to
where ¼ 0þ. Within the GW approximation, one significantly different Kohn–Sham eigenvalues
computes using Equation (39) and solves the (Seidl et al., 1996).
quasiparticle Dyson’s equation for the complex The GW approximation was formulated for the
energy Enk and wave functions cnk ðrÞ, which yield electron gas in the 1960s by Hedin and others.
the quasiparticle properties of the system. If, instead However, the first-principles approach to make com-
of expanding in terms of G and W, we replace puting quasiparticle excitations practical in real
W with the unscreened (or bare) Coulomb interac- materials was not developed until over 20 years
tion Vc, Equation (39) would reduce to the HF later by Hybertsen and Louie (1985, 1986). This
exchange operator. Therefore, much of the physics approach has since been employed with success in
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 59
ab initio studies of the quasiparticle properties of semi- gaps are even negative. However, the GW quasipar-
conductors, insulators, surfaces, nanostructures, and ticle energies (which provide an appropriate
other material systems. A crucial step in the develop- description of particle-like excitations in an interact-
ment was the realization of the importance of the ing system) result in band gaps that are in excellent
off-diagonal elements of the dielectric matrix (called agreement with experiments for a range of materials –
local-field screening effects) in the self-energy as well from the small gap semiconductors such as InSb, to
as efficient ways to calculate and apply the equally moderate gap materials such as GaN and solid C60
important frequency dependence of the dielectric and to the large gap insulators such as LiF. Similarly
response. The latter is typically done by calculat- accurate results have been obtained for surface-state
ing explicitly the static dielectric matrix within the band gaps and for nanostructures.
random-phase approximation and extending it to
finite frequencies using schemes such as the gen-
eralized plasmon pole (GPP) model (Hybertsen
1.02.8 The GW Formalism and
and Louie, 1985). The frequency dependence of
Applications
the dielectric response has also been computed
from analytic continuation from a few points on
The procedure for an ab initio GW calculation of the
the imaginary frequency axis or explicitly on the
quasiparticle energies and wave functions is depicted
real frequency axis in recent applications (Rojas
in Figure 11. In principle, the GW formalism is not a
et al., 1995).
density-functional-based formalism – nor does it
Hybertsen and Louie (1985) showed that, with the
depend on a particular choice of basis set or input
inclusion of the full dielectric matrix, one can obtain
wave functions so long as the basis is complete and
highly accurate quasiparticle energies. Similar to
the Dyson’s equation is solved in full (i.e., diagona-
first-principles ground-state studies, the only inputs
lized within the basis). However, as discussed above,
to the calculations are the atomic numbers of the
a good mean-field starting solution is desirable. In
constituent elements and the geometric structure of
practice, GW calculations often take as input the
the system, which can be determined separately from
Kohn–Sham wave functions and eigenvalues. For
a total energy calculation. Since the screened
the case of most bulk semiconductors, it is further
Coulomb interaction, W, incorporates the dynamical
found that the Dyson’s equation as a matrix equation
many-body effects of the electrons, the dielectric
is nearly diagonal if it is expressed in terms of the
response function "ðr; r9; !Þ is key in determining
basis of the LDA Kohn–Sham eigenstates. This is not
the electron self-energy. Owing to the spatial distri-
bution of atomic positions and electron density
inhomogeneity, "ðr; r9; !Þ of a solid is a function of r
and r9 separately. In a k-space formulation, the crys- Procedure for ab initio GW calculations
talline dielectric function is a matrix "G; G9 ðq; !Þ in Mean Field Solution {ϕ nk (r ), εnk }
the reciprocal lattice vectors G; the off-diagonal
elements of this matrix describe the local field effects
due to the charge inhomogeneity in real materials Construct ε (q, G, G′, ω )
(Hybertsen and Louie, 1987). Inclusion of both local
field and dynamic effects is essential for an accurate Construct W= ε–1Vc
description of the quasiparticle properties of a real
material system.
Let us return to Figure 8, which compares both Construct Σ=GW
the Kohn–Sham gaps and the GW band gaps of a
number of insulating materials against the experi- p2
Diagonalize H = + VH + Σ{ϕ nk (r), Enk }
mentally measured quasiparticle band gaps. A 2m
perfect agreement between theory and experiment
would place the data points on the diagonal line. As
Figure 11 Procedure for carrying out GW calculations as
we discussed before, the Kohn–Sham gaps in the
developed by Hybertsen and Louie. From Hybertsen MS
LDA (or GGA) significantly underestimate the and Louie SG (1985) First-principles theory of
experimental values, giving rise to the well-known quasiparticles: Calculation of band gaps in semiconductors
band-gap problem. Some of the LDA Kohn–Sham and insulators. Physical Review Letters 55: 1418.
60 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
to say that LDA does an accurate job representing the cancel those of vertex corrections to (Hedin and
quasiparticle dispersion in the system; it only says Lundqvist, 1969; Holm and von Barth, 1998;
that the Schrödinger-like equation with either the Aryasetiawan and Gunnarsson, 1998; Aulbur et al.,
LDA Vxc or ¼iGW often has similar eigenfunctions. 2000). However, as in the case of the effect of
They may have significantly different eigenvalues as higher-order terms or vertex corrections to , only
illustrated in the band-gap problem discussed above. an a posteriori experience truly justifies the approx-
In the case of silicon, the LDA wave functions differ imations used.
from the fully diagonalized GW wave functions by A typical GW calculation using the procedure in
less than 0.1% (Hybertsen and Louie, 1985). Figure 11 has a few technical issues worthy of some
However, there are important cases where expressing discussion. The key ingredient to the calculation
as a matrix within the LDA orbital basis does not (that is not found in HF, for example) is the full
yield a diagonal representation and a more preferred two-point dynamic dielectric function within RPA.
basis set can be used. We discuss such cases below. The calculation of this function is based on the
In standard GW studies, the frequency-dependent polarizability and has been traditionally computed
polarizability and dielectric function are calculated as a sum over Kohn–Sham empty orbitals. Both the
within the random-phase approximation (RPA) using calculations of the empty orbitals and the matrix
quantities from the mean-field reference system. W is elements are considerably more demanding than
constructed from Equation (38) and G is constructed that of a typical ab initio pseudopotential DFT calcu-
from Equation (35). In principle, we then diagonalize lation for the ground state. Even requiring a larger
Equation (36) for the quasiparticle wave functions and number of empty orbitals is the expression for
eigenvalues with the approximation that ¼ iGW. itself. can be broken into two terms: COH þ
However, since in many cases, as discussed above, SEX. SEX, called the screened-exchange term, is
only the diagonal elements are sizable within the simply the exchange, or Fock, term from Hartee–
mean-field orbital basis, in applications to real mate- Fock with Vc replaced by W. COH is an entirely
rials, the effects of can be treated within first-order new term absent in HF theory describing the inter-
perturbation theory. In this approach, the idea is to action of an electron with the Coulomb-hole created
consider in the form ¼ Vxc þ ( Vxc), where around the charge quasiparticle. In the static limit,
Vxc is some appropriately chosen independent parti- COH is a local operator but in the dynamic limit, the
cle mean-field approximation to the exchange– COH term, like the polarizability, involves a sum
correlation, and then to calculate within perturbation over the empty orbitals. This sum converges very
theory the effect of the additional interaction slowly with respect to the number of empty orbitals.
( Vxc) within the GW approximation. For mod- In practice, the dependence of COH on empty orbi-
erately correlated electron systems, the best- tals significantly increases the CPU time needed to
available mean-field Hamiltonian may be taken to generate the large number of initial Kohn–Sham
be the Kohn–Sham Hamiltonian. Thus, for most orbitals. However, although each quasiparticle state’s
studies in the literature, the Green’s function is typi- COH converges quite slowly, often needing thou-
cally constructed using the Kohn–Sham wave sands of empty orbitals, differences between
functions and eigenvalues. In principle, the process quasiparticle energies converge significantly faster.
of correcting the eigenfunctions and eigenvalues In bulk systems, where absolute energies are unim-
(which determine W and G) could be repeated until portant, it is sufficient to converge the calculation to
self-consistency is reached or diagonalized in full; this lower number of conduction bands, whereas in
however, in practice, it is found that an adequate nanosystems such as molecules, it becomes essential
solution is often obtained within first-order perturba- to often use huge number of empty orbitals, up to an
tion theory on the Dyson’s equation for a given . energy cutoff of 10s of Rydbergs. Recent research has
Comparison of calculated energies with experiment focused on approximating this sum with various
shows that this level of approximation is very accu- schemes, such as calculating the high-energy state
rate for semiconductors and insulators and for most contribution within the static limit or reformulating
conventional metals. Also, there is only very limited the problem so that no empty orbitals are involved
experience on the importance of including the (Tiago and Chelikowsky, 2006; Giustino et al., 2009;
detailed structures of the spectral function self- Wilson et al., 2009; Umari et al., 2009a,b; Kohn
consistently for the quasiparticle energies; and it and Sham, 1965; Bruneval and Gonze, 2008;
seems that the effects of self-consistency tend to Reining et al., 1997).
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 61
From the above examples, it is clear that the Here avþ and bcþ are the second-quantization crea-
correct formalism, including many-electron effects, tion operators for holes and electrons, respectively.
is extremely important in describing the electronic For simplicity of notation, we use v and c to represent
properties of semiconductors, particularly nanostruc- both the band index and wave vector of the hole
tured systems, in connection with quasiparticle and electron state, respectively, and we use
excitations such as those measured in photoemission, jvc i ¼ avþ bcþ jN ;0i to denote a configuration in
tunneling, and other transport experiments. In many which a hole with wave vector k is created in the
cases, neglecting these effects can lead to errors on valence band v and a quasi-electron is created in
the order of magnitude of the quantities being the conduction band c with wave vector k þ q, as
measured or even give qualitatively incorrect beha- illustrated in Figure 15, where the q 0 limit is
vior –for example, germanium is predicted to be a taken.
metal within DFT–LDA while the GW approach Electron–electron interactions mix electron–hole
correctly captures its semiconducting character and configurations of the same center-of-mass q to
yields quantitatively the value of the band gap. form the excited (excitonic) state jN ;S i. The
electron–hole amplitude or exciton wave function
in real space, which describes the spatial correlation
of the electron and hole, is then given by
1.02.9 Two-Particle Excitations and X
the Bethe–Salpeter Equation S ðr; r9Þ ¼ AScv cc ðrÞcv ðr9Þ; ð41Þ
cv
We now discuss the formalism for the ab initio calcu- with c(r) being the quasiparticle amplitudes (or
lation of optical properties. For an interacting wave functions), for example, those obtained from
electron system, such as a semiconductor, we must a GW calculation. As in the case of the Dyson’s
include the interaction between the quasi-electron equation for quasiparticles, we use the two-particle
and the quasi-hole created in the optical excitation Green’s function’s equation of motion to arrive at
process. Thus, the study of the optical response an effective equation for AS and S. These quanti-
requires knowledge of the correlated electron–hole ties can be shown to satisfy a Bethe–Salpeter
excitations, jN ;0i ! jN ;S i, that do not change the equation of the form (Strinati, 1988; Rohlfing and
total number of electrons N in a system. (Here jN ;S i Louie, 2000)
denotes the Sth neutral excited state). That is, we X
consider excitations for which no particles are added EckQP – EvkQP ASvck þ vckK eh v9c9k9 ASv9c9k9
or subtracted from the system. k9v9c9
¼ S ASvck : ð42Þ
Just as the quasiparticle excitations are given by
the one-particle Green’s function G, the electron-
Here the EQPs are the quasiparticle energies and Keh
hole excitations may be obtained by investigating
is the electron–hole kernel describing the interaction
the interacting two-particle Green’s function G2
between the excited electron and hole. Solving this
(which is a two-particle generalization of Equation
Bethe–Salpeter equation yields the excitation ener-
(24)) and solving its equation of motion. In the
gies and the excited-state wave functions from which
above discussion on quasiparticles, we assume the
one can compute the optical absorption spectrum,
single-particle excitations are long-lived; the theory
exciton binding energies and wave functions, and
on optical excitations here also assumes that the
other related optical quantities.
electron–hole excitations are long-lived and that
the Tamm–Dancoff approximation (Kadanoff and (a) c c (b)
Baym, 1999) is valid. The excited state jN ;S i with c
energy S (referenced to the ground-state energy)
can then be expressed as a linear combination of
noninteracting electron–hole configurations plus
υ υ υ
correction terms that are not important in the eva-
luation of the optical oscillator strength: Figure 15 (a) Schematic of a photo-excited exciton state
as a superposition of independent electron–hole vertical
hole X
X elec transitions. (b) Feynman diagrams for the BSE consisting of
j N ; Si ¼ ASvc avþ bcþ j N ; 0i þ ð40Þ a repulsive exchange term and an attractive direct
v c interaction.
64 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
The interaction kernel, Keh, is an operator that Equation (46) is a coherent sum of the interband
describes the scattering of an electron–hole pair going transition matrix elements of the individual contri-
from one configuration to another ðjvc i ! jv9c9iÞ buting electron–hole pair transitions where the
due to the Coulomb interaction among the electrons. amplitudes and phases of the AS often lead to inter-
Using the notation 1¼(r1, 1, t1), K is given by the func- esting interference effects. It is absolutely crucial
tional derivative (Rohlfing and Louie, 1998a; Strinati, therefore to keep the relative phases of A intact dur-
1988) ing the calculation and to consider a sufficiently
complete basis sets of single-particle transitions.
½VH ð1Þð1; 3Þ þ ð1; 3Þ
K ð12; 34Þ ¼ : ð43Þ The Bethe–Salpeter formalism for two-particle
Gð4; 2Þ
excitations is therefore a natural extension of the
where the self-energy operator, , is typically treated GW formalism for quasiparticle excitations. In first-
within the GW theory in practical implementations. principles implementations, calculations within both
With the additional assumption that the functional formalisms use the same approximations for the self-
derivative of W with respect to G can be neglected energy and Green’s Function G. Figure 16 illus-
(Rohlfing and Louie, 2000), K is simplified to trates the importance of the BSE framework in
calculating the imaginary part of the dielectric func-
K ð12; 34Þ ¼ – ið1; 3Þð2 – ; 4ÞV ð1; 4Þ tion in silicon (Rohlfing and Louie, 1998a; Albretch
et al., 1998; Benedict et al., 1998). Only after including
þ ið1; 4Þð3; 2ÞW ð1þ ; 3Þ ð44Þ
excitonic effects is quantitative agreement with
¼ Kx þ Kd : experiment in the first peak height and also the
position of the second peak reached, resolving a
Equation (44) has two terms, as illustrated in long-standing puzzle in the field. This redistribution
Figure 15: a bare repulsive exchange interaction Kx of oscillator strength does not derive from the exis-
and a screened attractive interaction Kd. The meth- tence of bound excitons at the peak position but of
ods for the evaluation of the matrix elements of the the mixing interband excitations from a large k and
kernel K and the solution to the Bethe–Salpeter energy range giving a large enhancement in the first
equation have been discussed in detail in the litera- absorption peak. Such rearrangements of optical
ture (e.g., Rohlfing and Louie, 2000). When spin is strength also give rise to an apparent shift of nearly
considered, the exchange term only affects the singlet
excitons. Dynamical (i.e., the frequency dependent)
screening can further be neglected in Kd if the excita- 50 Expt
tion energies S are close to the energy of the Interacting
Nonint.
noninteracting pairs – in other words, the exciton- 40
binding energy is small compared to the optical
excitation energy itself. From the solution of the
30
Bethe–Salpeter equation (Equation (42)), the optical
ε2
0.5 eV for the second peak in Figure 16 (Rohlfing energy of the correlated exciton states relative to the
and Louie, 1998a). Beyond the oscillator strength interband transition continuum edge. With the exci-
redistribution, as in a simple hydrogenic model for tation energies and amplitudes of the electron–hole
excitons in semiconductors, a series of bound exciton pairs, A, one can then calculate the macroscopic
states is found below the onset of absorption edge, dielectric function for various light polarizations
though their oscillator strength is weak and binding using Equation (45), and also obtain quantities such
energy is only of the order of 10 meV for Si. As seen as electron–hole correlation functions in real space
below, the GW–BSE approach is capable of obtain- and higher-order optical effects such as multi-photon
ing highly accurate optical spectra from first absorption and phonon-assisted absorption spectra.
principles and has helped elucidate the optical When applying this method to isolated nanosystems
response of a wide range of systems ranging from in supercell calculations, it is important, as it is in the
nanostructures to bulk semiconductors to surfaces GW calculation, to replace W with the appropriate
and defects to 1D systems such as polymers and screened truncated interaction. As will be illustrated
nanotubes. in the final section, even with well-separated systems
that are considered reasonable in DFT calculations,
the nature of the interaction between the electron
1.02.10 The GW-BSE Formalism and and hole in an untruncated interaction calculation
Optical Response can be very different from the isolated case owing
to the unwanted influence of neighboring replicas.
Figure 17 illustrates the procedure for carrying out Because of the generally reduced screening and con-
an ab initio GW–BSE calculation to obtain quasipar- finement effects, one expects stronger excitonic
ticle and optical properties. The BSE calculations effects in reduced dimensional systems, which as
take as input the quasiparticle energies and wave seen below, is indeed the case.
functions from a GW calculation. In addition, the Figure 18 depicts the optical spectrum of bulk
BSE calculation takes the full dielectric matrix cal- GaAs. As seen from this figure and the results for Si
culated in the GW step, which is used to screen the in Figure 16, only after the inclusion of electron–
attractive direct electron–hole interaction. The hole interaction, there is good agreement between
direct and exchange part of the electron–hole kernel theory and experiment. As already discussed above
are calculated, and the electron–hole kernel plus the for Si, there is a redistribution of optical oscillator
kinetic energy term as a matrix in the quasiparticle strength to the lower energy peak, enhancing it by
electron–hole pair basis is then diagonalized yielding nearly a factor of 2. In addition, the electron–hole
the exciton wave functions and excitation energies. interaction shifts and heightens the second main
Exciton binding energies can be inferred from the peak, yielding agreement with experiment. This
30
Procedure for GW-BSE calculations 25
GaAs
Mean Field Solution {ϕnk (r ), εnk } 20
ε2
15
GW solution {ϕQPnk (r ), εQP nk } 10
5
Construct Bethe−Salpeter K (k,c,v,k′,c′,v ′)
0
0 2 4 6 8 10
Energy (eV)
Diagonalize BSE Hamiltonian A S c,v,k,E S
Figure 18 Calculated optical absorption of GaAs with
(solid line) and without (dashed line) electron–hole
Figure 17 Procedure for a GW-BSE calculation. The end interaction, compared to experimental data [Aspnes,
result of such a procedure is the ab initio calculation of the Lautenschlager]. Modified from Rohlfing M and Louie SG
quasiparticle and exciton states of a material system as well (1998a) Excitonic effects and the optical absorption
as a description of the optical properties through the spectrum of hydrogenated Si clusters. Physical Review
expression of the oscillator strength, Equation (45). Letters 80: 3320.
66 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
directly from the BSE calculation without the use With e–h interaction
4 (// polarization)
of any effective mass or other approximations. The
calculated binding energy of the lowest energy exci- 2 A1
ton states in GaAs (Rohlfing and Louie, 1998a) is
4.0 meV as compared to experimental value of 0
4.2 meV (Michaelis et al., 1996; Sell, 1972). The the- 2 With e–h interaction
ory has reproduced essentially all the observed (⊥ polarization)
bound excitonic structures to a good level of accu-
0
racy in GaAs. Thus, it is possible to obtain the 2 4 6 8
binding energy of bound excitons very accurately
Energy (eV)
from Equation (42), even if the binding energy is
only of the order of few meVs. To achieve such a Figure 20 Absorption vs. photon energy for the silicon
nanowire shown in Figure 19 for light polarized (top two
good description of the optical response of GaAs, a
panels) parallel to the wire axis and (bottom panel)
very fine sampling of the Brillouin zone and the perpendicular to the wire axis. The arrows marked by A and
inclusion of spin–orbit interaction are required, so A1 indicate the fundamental gap and the lowest-energy
that the quasiparticle energy bands are calculated exciton locations, respectively. The top panel shows the
with high accuracy. spectrum without electron–hole interaction effects
included, whereas the middle panel contains the full
Electron–hole effects are even more important solution to the BSE. The bottom panel illustrates that light
when applied to silicon nanowires (Yang et al., polarized perpendicular to the wire axis is absorbed very
2007c). Silicon nanowires are quasi-1D solid wires weakly due to depolarization effects. Modified from Yang L,
of silicon terminated by passivating adsorbates on the Spataru CD, Louie SG, and Chou MY (2007c) Enhanced
electron–hole interaction and optical absorption in a silicon
wire surface. Such nanowires represent a potential
nanowire. Physical Review B 75: 201304(R).
building block for nanoscale devices and have
recently been synthesized (Cui and Lieber, 2001;
Duan et al., 2001; Morales and Lieber, 1998; Holmes In Figure 20, we show the calculated optical
et al., 2000). Si nanowires with diameter as small as spectra of the Si nanowire shown in Figure 19
1.3 nm have been fabricated (Ma et al., 2003). employing the GW-BSE approach, with and without
Figure 19 depicts a model of a Si nanowire passi- electron–hole interaction effects included. The
vated with hydrogen of a diameter of 1.2 nm, with the absorption for light polarized parallel to the wire
wire axis along the [110] direction. axis exhibits two new peaks with inclusion of
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 67
electron–hole interaction, giving rise to an appear- (and quasiparticle self-energy) effects (Spataru et al.,
ance of a shift of the spectrum by nearly 2 eV 2004a, 2004b; Ando, 1997) are shown to be domi-
(comparison of top and middle panel in Figure 20). nantly important in these systems even at room
The dominant peaks in the spectrum (middle panel) temperature. Figure 21 compares the calculated
are due to resonant excitons. In addition, there is a absorption spectrum of a (8,0) semiconducting
bound exciton state although with very weak optical SWCNT between the cases with and without
strength, labeled A1, deep inside the quasiparticle electron–hole interactions. As for the Si nanowires,
band gap at 2.1 eV with a binding energy of 1 eV. because of depolarization effects (Ajiki and Ando,
This binding energy is two orders of magnitude 1994), the optical response of a tube-like object is
larger than the exciton-binding energies in bulk Si. the strongest when the polarization of the light is
This enhanced binding energy is a result of the along the tube axis – the polarization used in the
reduced dimensionality of the system, and of the calculations. The two resulting spectra are radically
unique nature of screening in lower dimensions as different. When electron–hole interaction effects are
discussed in the next section. included, the spectrum is dominated by bound and
When the GW-BSE method is applied to semi- resonant exciton states. Each of the structures
conducting SWCNTs, the results are extraordinarily. derived from a van Hove singularity in the noninter-
Because of the 1D nature of the nanotubes, excitonic acting joint density of states gives rise to a series of
(a) 250
A′1 (8,0)
200
With e−h interaction (b)
Without e−h interaction
150
ε2 (a.u.)
B′1
100 C′1
Exciton A’1
A′2
50 A
B′2 B C
A′2
C′2
0
1 1.5 2 2.5 3 3.5 4
Photon energy (eV)
75
200
|φ|2
|φ|2
50
100
25
0 0
−50 −25 0 25 50 −100 −50 0 50 100
Z (Å) Z (Å)
Figure 21 Excitons in semiconducting (8,0) SWCNTs. (a) Absorption spectra plotted with and without the electron–hole
interaction included. The oscillator strength from the continuum is almost entirely moved to the sharp bound exciton
transitions. (b) Exciton amplitude square plotted with the hole position fixed (dot in the center). (c,d) Exciton amplitude square
plotted along the tube axis with hole position fixed at z¼0.
68 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
exciton states, and these states rob all the optical nanotube. Again, the electron–hole interaction leads
transition strength from the continuum transitions. to a series of sharp lines in the absorption spectrum
Panels (b), (c) and (d) in Figure 21 give the spatial due to strongly bound excitonic states, and the optical
correlation between the excited electron and hole for strength as in the case of carbon nanotubes is virtually
two of the exciton states, one bound (A’1) and one completely transferred to these states. For the (8, 0) BN
resonant state (C’1). The extent of the exciton wave nanotube, which has the same diameter as the (8, 0)
function is about 2 nm for both of these states. For the SWCNT, the exciton binding energy of the lowest
(8,0) nanotube, the lowest-energy bound exciton has a energy exciton is over 2 eV (Park et al., 2006). This is
binding energy of nearly 1 eV. Note that the exciton- consistent with the fact that BN nanotubes are wide-
binding energy for bulk semiconductors of similar size gap insulators. Unlike the carbon case, however, the
band gap is in general only of the order of tens of meVs, lowest-energy exciton is composed of almost equal
illustrating again the dominance of many-electron weight from four sets of nearby interband transitions,
Coulomb effects in reduced dimensional systems. which lead to a considerably more localized electron–
Similar results have been obtained for other semicon- hole correlation in real space along the circumference
ducting carbon nanotubes (Spataru et al., 2004a, 2004b). direction.
These results predicted extraordinarily large exciton A surprising prediction was further made when
binding energies, although first met with skepticism, GW-BSE calculations were carried out for the metal-
have been verified by recent experiments (Wang et al., lic SWCNTs – bound/resonant exciton states exist
2005; Ma et al., 2005). Owing to the unique electronic despite metallic screening from the carriers (Spataru
structure of the carbon nanotubes, in addition to the et al., 2004a; Deslippe et al., 2007). To our knowledge,
optically active (or bright) excitons shown in these are the first bound excitons known to exist in
Figure 21, there exist also a number of optically inac- metallic systems, and they have subsequently been
tive (or dark) excitons associated with each of the bright observed by measurements (Wang et al., 2007). The
ones. These dark excitons can be observed experimen- existence of excitons in metallic SWCNTs is a result
tally with an applied magnetic field or light with of the increased efficiency of an attractive interaction
different polarization. These dark excitons also play in forming bound states in one dimension. This can
an important role in the optical properties of the nano- be seen as a result of the fact that, in one dimension,
tubes; for example, they strongly affect the radiative (1) electronic screening is greatly reduced and(2) any
lifetime of the optically created excitons in semicon- Rpotential (other than V ¼ 0) satisfying the constraint
ducting carbon nanotubes (Spataru et al., 2005). V ðx Þdx 0 (i.e., is negative on average) is guaran-
For the boron nitride (BN) nanotubes, which are teed to have at least one bound state (Kocher, 1977;
wide band-gap materials, it is found that excitonic Simon, 1976).
effects are even stronger than in SWCNTs. Figure 22 Figure 23 shows the LDA band structure for a
shows the calculated optical spectra of an (8, 0) BN (10, 10) and a (12, 0) metallic SWCNTs. For incident
E QP (eV)
20
16 2
12 0
8 K I′1
I2 I3 (x4) –2
4 II1 I4
5 6 7 8 9 Γ X
Photon energy (eV)
Figure 22 Absorption spectra and band structure of a (8,0) boron-nitride nanotube. (a) The spectrum is plotted with (solid)
and without (dashed) electron–hole interaction included. (b) Schematic of the four pairs of bands from which mixing of these
interband transitions forms the lowest bound exciton. Modified from Park CH, Spataru CD and Louie SG (2006) Excitons and
many-electron effects in the optical response of single-walled boron nitride nanotubes. Physical Review Letters 96: 126105.
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 69
2
2 1
Energy (eV)
0 (12,0)
0 (10,10)
−1
−2
−2
−3
0 0.2 0.4 0 0.2 0.4
k(2π/a)
Figure 23 Band structure of metallic nanotubes showing allowed optical transitions (solid arrow) and forbidden optical
transitions (dashed arrow).
60 15
40
10
20
5
0
0
1.5 2 2.5 1.5 2 2.5
Energy (eV)
Figure 25 Spectral shape of absorption spectra in the (10,10) metallic nanotube plotted with (solid) and without (dashed)
the electron–hole interaction and with Lorentzian broadening of 20 meV (left panel) and 80 meV (right panel). The right panel
plot has the noninteracting spectra shifted and scaled to match the position and height of the interacting spectra for
comparison.
(10,10) SWCNT. In the left panel of Figure 25, we for the (21,21) metallic SWCNT was 50 meV (Wang
see that the optical peak arises almost entirely from et al., 2007), in good agreement with values derived
the single exciton state, showing a profile that is from the ab initio calculations (Deslippe et al., 2009).
significantly different from the interband transition The large excitonic effects described here are not
case. As a comparison, the right panel in Figure 25 unique to the systems studied but are characteristic of
shows the noninteracting spectrum scaled and shifted 1D (as well as 0D and 2D) systems in general.
to match the peak height in the interacting case. We Graphene nanoribbons (see Figure 26), 2D graphene
see that the two lineshapes are distinctly different. and bilayer graphene all show similar large excitonic
This predicted change in lineshape was recently effects (Yang et al., 2007a, 2008, 2009). This is largely
confirmed by experiment using optical absorption because of the unique nature of screening in reduced
techniques on isolated metallic SWCNTs (Wang dimensional systems that will be discussed in the next
et al., 2007). The measured exciton-binding energy section.
(a) (b)
E11
1 E11
1
W/o e−h interaction W/o e−h interaction
With e−h interaction With e−h interaction
Optical absorption (a.u.)
11-AGNR 12-AGNR
E11
E11
0 2 4 0 2 4
Energy (eV)
Figure 26 Absorption spectra of armchair graphene nanoribbons with 11 carbon dimer rows (left panel) and 12 carbon
dimer rows (right panel) forming the width of the ribbon. Oscillator strength from the continuum is moved almost entirely into
the sharp bound exciton peaks. Modified from Yang L, Cohen ML, and Louie SG (2007a) Excitonic effects in the optical
spectra of graphene nanoribbons. Nano Letters 7(10): 3112–3115.
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 71
Z
1.02.11 Interactions in V ðzÞ ¼ dx9dy9dr2 W ðr9 þ r2 ; r2 Þjcc ðr9 þ r2 Þj2 jcv ðr2 Þj2 :
Nanostructured Semiconducting
Materials ð47Þ
The application of the GW-BSE method to semicon- The divergence at the origin is removed due to
ducting nanostructured materials has illuminated and the spatial extent to which the particles can still be
predicted several exciting new physical behaviors of separated when the z separation is zero. In an
these systems. As discussed above, many-body effects SWNT, this spatial extent is the tube diameter.
are crucial for even a qualitative understanding of
While not containing a singularity, the potential
their excited state properties. This can be seen as
still leads to exciton binding energies much greater
consequence of two general principles: first, the
than those found in higher-dimensional systems due
Coulomb interaction is more effective in lower
to the fact that the phase space over which the inter-
dimensions, and, second, screening has a unique nat-
action is large does not vanish.
ure in two- or less-dimensional systems.
In order to truly appreciate all the physics con-
The first point is illustrated in Figure 27. In
tained in the electron–hole kernel, Equation (43), it is
the case of three (or two) dimensions, a particle
useful to model the dielectric function, which deter-
under the influence of a Coulomb potential can
mines the screened interaction W. We can express an
avoid the singularity by traveling around the ori-
effective 1D dielectric function in the case of a nano-
gin of the potential. The expectation value of the
tube as "(q)¼1 – (q)Vbare (q) and the total screened
3D potential can be approximately expressed as
R 3 ðrÞ potential generated by a single ring of charge on a
d r jr j . If is reasonably well behaved, as it is tube is then (Deslippe et al., 2009; Leonard and
for the hydrogen model, than the 3D integral Tersoff, 2002)
about the origin integrates to a finite value with
R Vbare ðqÞ
no divergence. However, in the 1D case, dz jðzzjÞ V ðqÞ ¼
1 – ðqÞVbare ðqÞ
: ð48Þ
logarithmically diverges if the origin is included in
the integral. In the 1D Coulomb interaction, the For the case of a semiconducting tube with band
singularity is unavoidable for an oscillating parti- gap, Eg, and diameter, d, the susceptibility, , can be
cle, leading to large binding energies. In real obtained from a 1D generalization of the Penn model
quasi-1D nanosystems, it is the finite extent of (Penn, 1962).
the physical system in the other directions that The interesting behavior is that the dielectric
prevents the interaction between charge particles function in momentum space approaches unity at
in the system from diverging. For example, one both large and small q (Leonard and Tersoff, 2002;
can define an effective screened electron–electron van den Brink and Sawatzky, 2000). In a typical bulk
or electron–hole interaction in a quasi-1D system semiconductor on the other hand, the dielectric func-
using an averaging in the nonperiodic directions as tion approaches the bulk dielectric constant at small
(Deslippe et al., 2009) q. This behavior is a direct result of the fact that,
unlike in 3D semiconductors where the induced
charge is of the opposite sign as the added charge
3D/2D 1D and is a finite fraction (1 – 1/"0) of the added charge,
e
in 1D the total induced charge seen far away from the
e added charge particle is zero with regions where
charge density of the opposite sign as the external
h
charge is induced, as well as regions where charge
h density of the same sign is induced. A qualitative
explanation for the phenomenon is shown in
Figure 28 where the polarizable charge distribution
Figure 27 3D/2D vs. 1D electron–hole interaction. In 1D, of a semiconducting system is modeled by a simple
the singularity is unavoidable, leading to greatly enhanced ball-and-spring dielectric medium (Deslippe et al.,
exciton binding energies and other electron–electron
interaction effects. For example, the 1D bare attractive
2009). In three dimensions, because the surface area
Coulomb interaction has a bound state of infinite binding of a spherical shell is proportional to the radius
energy compared to the value of 13.6 eV of the 3D case. squared and the force generated by a charge at the
72 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
V(z) (eV)
–0.5 (8,0)
(8,0) Vbare
h
h
–1
0 5 10 15 20
z (nm)
Figure 29 Screened (solid) and bare (dashed) Coulomb
interaction in semiconducting (8,0) SWCNTs. Modified from
Figure 28 Screening in a polarizable medium in three
Deslippe J, Dipoppa M, Prendergast D, Moutinho MVO,
dimensions (left) and one dimension (right). In three
Capaz RB, and Louie SG (2009) Electron–hole interaction in
dimensions, the amount of charge induced inside a sphere
carbon nanotubes: Novel screening and exciton excitation
of radius r is constant as a function of increasing radius,
spectra. Nano Letters 9(4): 1330–1334.
whereas in one dimension the amount of charge induced
inside a box of length z goes to zero for large z.
origin on a spring at distance r is proportional to 1/r2, As discussed above, because essentially every
the total induced charge in a shell of radius r is k-point in a quasi-1D system is on a high symmetry
constant with respect to r. So, at large distances direction, the optical absorption to exciton states
from the external charge, there is a net induced obey strict symmetry selection rules across entire
charge of the opposite sign observed surrounding bands. In the semiconducting SWCNTs, the first
the external charge. In quasi-one-dimension, how- allowed optical excitation is just one of many (four
ever, the surface area perpendicular to the tube axis in the case of the zigzag tubes mentioned below)
of a box does not change as the box length changes in nearly degenerate exciton states due to the band or
the z-direction. The total induced charge in larger valley degeneracy of the quasiparticle states. These
and larger size boxes drops to zero; so at large length exciton states, and their relative oscillator strengths,
scales, there is effectively no screening (Deslippe come out directly from the GW-BSE calculation and
et al., 2009). are shown in Figure 30 for the (10, 0) SWCNT.
What is even more unusual is that at some inter- These states can be investigated through symmetry
mediate distance, the effect of screening in 1D breaking probes such as by applying a magnetic field
systems, such in SWCNTs, is to actually enhance or through careful analysis of photoluminescence
rather than weaken the Coulomb interaction. That is, spectra (Shaver et al., 2007).
there are ranges of separations over which two par-
ticles see an anti-screening effect. This is
demonstrated for the case of a (8, 0) SWNCT using
1.02
Equation (48) in Figure 29. Because the screened (10,0)
Coulomb interaction is very different in 1D, a very Dark –
0E6
1 Bright –
distinct excitation spectrum is created for the higher- 0A0
δ2
energy exciton states form from a given pair of inter- δ1
hΩ(0) (ev)
0.98
band transitions. The actual spectra are qualitatively - Dark 0B0
-
0B0
and quantitatively different from those of excitons in 0.96 -
0E6
3D or in 1D models that neglect the anti-screening
A -
0 0
effect. The higher states in the excitonic series are 0.94
considerably more bound due to the presence of the Triplet Singlet
anti-screening region; evidences consistent with 0.92
this prediction have been observed in recent mea- Figure 30 The first optical transition oA
0 in zigzag (10,0)
surements on semiconducting SWCNT (Lefebvre SWCNT is accompanied by several dark states due to the
and Finnie, 2007, 2008). well-defined symmetry in SWCNTs.
Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors 73
Kittel C and Kroemer H (1980) Thermal Physics, 2nd edn., Reining L, Onida G, and Godby RW (1997) Elimination of
Freeman. unoccupied-state summations in ab initio self-energy
Kocher CA (1977) Criteria for bound-state solutions in quantum calculations for large supercells. Physical Review B 56: R4301.
mechanics. American Journal of Physics 45: 71. Rohlfing M and Louie SG (1998) Excitonic effects and the
Kohn W and Sham L (1965) Self-consistent equations including optical absorption spectrum of hydrogenated Si clusters.
exchange and correlation effects. Physical Review A Physical Review Letters 80: 3320.
140: 1133. Rohlfng M and Louie SG (1998) Electron–hole excitations in
Lautenschlager P, Garriga M, Logothetidis S, and Cardona M semiconductors and insulators. Physical Review Letters
(1987) Interband critical points of GaAs and their 81: 2312.
temperature dependence. Physical Review B 35: 9174. Rohlfing M and Louie SG (2000) Electron–hole excitations and
Lefebvre J and Finnie P (2007) Polarized photoluminescence optical spectra from first principles. Physical Review B
excitation spectroscopy of single-walled carbon nanotubes. 62: 4927.
Physical Review Letters 98: 167406. Rojas HN, Godby RW, and Needs RJ (1995) Space–time
Lefebvre J and Finnie P (2008) Excited excitonic states in method for ab initio calculations of self-energies and
single-walled carbon nanotubes. Nano Letters 8(7): 1890. dielectric response functions of solids. Physical Review
Leonard F and Tersoff J (2002) Dielectric response of Letters 74: 1827.
semiconducting carbon nanotubes. Applied Physics Letters Saito R, Dresselhaus G, and Dresselhaus MS (1998) Physical
81(25): 4835. Properties of Carbon Nanotubes. London: Imperial College
Li X, Wang X, Zhang L, Lee S, and Dai H (2008) Chemically Press.
derived, ultrasmooth graphene nanoribbon semiconductors. Seidl A, Goerling A, Vogl P, Majewski JA, and Levy M (1996)
Science 319(5867): 1229–1232. Generalized Kohn–Sham schemes and the band-gap
Louie SG (1994) Electronic structure of C60 fullerites and problem. Physical Review B 53: 3764.
nanatubes. In: Kuzmany H, Fink J, Mehring M, and Sell D (1972) Resolved free-exciton transitions in the optical-
Roth S (eds.) Progress in Fullerene Research: International absorption spectrum of GaAs. Physical Review B 6: 3750.
Winter School on Electronic Properties of Novel Materials, Shaver J, Kono J, Portugall O, et al. (2007) Magnetic brightening
303. Singapore: World Scientific. of carbon nanotube photoluminescence through symmetry
Ma DDD, Lee CS, Au FCK, Tong SY, and Lee ST (2003) Small- breaking. Nano Letters 7(7): 1851–1855.
diameter silicon nanowire surfaces. Science Siegel DA, Park CH, Hwang C, et al. (in press) Many Body
299(5614): 1874–1877. Interactions in Quasi-Freestanding Graphene.
Ma YZ, Valkunas L, Bachilo SM, and Fleming GR (2005) Silvestrelli PL and Parrinello M (1999) Water dipole moment in the
Exciton binding energy in semiconducting single-walled gas and the liquid phase. Physical Review Letters 82: 3308.
carbon nanotubes. Journal of Physical Chemistry B Simon B (1976) The bound state of weakly coupled Schrodinger
109: 15671. operators in one and two dimensions. Annals of Physics
Mahan GD (1981) Many-Particle Physics. New York: Plenum. 97: 279.
Maultzsch J, Pomraenke R, Reich S, et al. (2005) Exciton Son Y-W, Cohen ML, and Louie SG (2006) Energy gaps in
binding energies in carbon nanotubes from two-photon graphene nanoribbons. Physical Review Letters 97: 216803.
photoluminescence. Physical Review B 72: 241402. Son Y-W, Cohen ML, and Louie SG (2006) Half-metallic
Michaelis JS, Unterrainer K, Gornik E, and Bauser E (1996) graphene nanoribbons. Nature 444(7117): 347–349.
Electric and magnetic dipole two-photon absorption in Spataru CD, Ismail-Beigi S, Benedict LX, and Louie SG (2004)
semiconductors. Physical Review B 54: 7917. Excitonic effects and optical spectra of single-walled carbon
Miyake T, Zhang P, Cohen ML, and Louie SG (2006) nanotubes. Physical Review Letters 92: 077402.
Quasiparticle energy of semicore d electrons in ZnS: Spataru CD, Ismail-Beigi S, Benedict LX, and Louie SG (2004)
Combined LDA þ U and GW approach. Physical Review B Quasiparticle energies, excitonic effects and optical
74: 245213. absorption spectra of small-diameter single-walled carbon
Morales AM and Lieber CM (1998) A laser ablation method for nanotubes. Applied Physics A 78: 1129.
the synthesis of crystalline semiconductor nanowires. Spataru CD, Ismail-Beigi S, Capaz RB, and Louie SG (2005)
Science 279: 208. Theory and ab initio calculation of radiative lifetime of
Novoselov KS, Geim AK, Morozov SV, et al. (2005) Two- excitons in semiconducting carbon nanotubes. Physical
dimensional gas of massless Dirac fermions in graphene. Review Letters 95: 247402.
Nature 438(7065): 197–200. Strinati G (1988) Application of the Green’s-function method to
Ortega JE and Himpsel FJ (1993) Inverse-photoemission study the study of the optical-properties of semiconductors.
of Ge(100), Si(100), and GaAs(100) – bulk bands and surface- Rivista del Nuovo Cimento 11(12): 1.
states. Physical Review B 47: 2130. Tassaing T (1998) The partial pair correlation functions of dense
Park CH, Spataru CD, and Louie SG (2006) Excitons and supercritical water. Europhysics Letters 42: 265.
many-electron effects in the optical response of single- Tiago ML and Chelikowsky JR (2006) Optical excitations in
walled boron nitride nanotube. Physical Review Letters organic molecules, clusters, and defects studied by first-
96: 126105. principles Green’s function methods. Physical Review B
Parr RG and Yang W (1989) Density-Functional Theory of Atoms 73: 205334.
and Molecules. Oxford University Press. Umari P, Stenuit G, and Baroni S (2009a) Optimal
Penn DR (1962) Wave-number-dependent dielectric function of representation of the polarization propagator for large-scale
semiconductors. Physical Review 128(5): 2093. GW calculations. Physical Review B 79: 201104(R).
Phillips JC and Kleinman L (1959) New method for calculating Umari P, Stenuit G, and Baroni S (2009b) GW quasiparticle spectra
wave functions in crystals and molecules. Physical Review from occupied states only. Physical Review B 81: 115104.
116: 287. van den Brink J and Sawatzky GA (2000) Non-conventional
Pickett WE (1989) Pseudopotential methods in condensed screening of the Coulomb interaction in low-dimensional and
matter applications. Computer Physics Reports 9: 115. finite-size systems. Europhysics Journal 50(4): 447.
Reining L, Olevano V, Rubio A, and Onida G (2002) Excitonic van Schilfgaarde M, Kotani T, and Faleev S (2006) All-electron self-
effects in solids described by time-dependent density- consistent GW approximation: Application to Si, MnO, and
functional theory. Physical Review Letters 88: 066404. NiO. Physical Review Letters 96: 226402.
76 Ab initio Theories of the Structural, Electronic, and Optical Properties of Semiconductors
1.03.1 Introduction 78
1.03.1.1 Isolated Donor Atom in Semiconductors 79
1.03.1.2 Quantum Computer Using Si:P 80
1.03.2 Localization Theories in Impurity Bands 81
1.03.2.1 Impurity Conduction 81
1.03.2.2 Mott Transition 82
1.03.2.3 Effects of Localization and Anderson Transition 82
1.03.2.4 Computer Studies of Anderson Model 84
1.03.2.5 Scaling Theory of Localization 84
1.03.3 Electron–Electron Interactions in Impurity Bands 86
1.03.3.1 Theoretical Formulation 87
1.03.3.1.1 Transfer diagonal representation 87
1.03.3.1.2 Anderson-localized states 87
1.03.3.1.3 Interaction terms 88
1.03.3.2 Theory of Intrastate Interaction 88
1.03.3.2.1 Spin susceptibility and specific heat 89
1.03.3.2.2 Comparison with experiment on Si:P 90
1.03.3.3 Interstate Interactions 91
1.03.3.3.1 Direct and kinetic-type exchanges 91
1.03.3.3.2 Spin-pair model 92
1.03.3.3.3 Specific heat anomaly and spin susceptibility 93
1.03.3.4 Numerical Simulation for Interacting Donor Electrons in Si:P System 94
1.03.3.4.1 Gaussian model 95
1.03.3.4.2 Electron configurations and many-electron states 96
1.03.3.4.3 Specific heat and spin susceptibility 98
1.03.3.4.4 Comparison with experiment 99
1.03.4 Hopping Conduction and Related Phenomena 101
1.03.4.1 VRH and Mott’s Law 101
1.03.4.1.1 Formulation of hopping conduction 101
1.03.4.1.2 Mott’s law 102
1.03.4.2 Coulomb Gap 102
1.03.4.3 Formula for Magnetoresistance in the Variable-Range Regime 104
1.03.4.3.1 Appearance of spin-dependent mechanisms 104
1.03.4.3.2 Calculated results 105
1.03.4.3.3 Comparison with experiments 106
1.03.4.4 Magnetocapacitance in Intermediate-Concentration Regime 107
1.03.4.5 Spin-Dependent Behavior of Magnetoresistance in Other Systems 109
1.03.5 Conclusions 109
References 110
77
78 Impurity Bands in Group-IV Semiconductors
nanofabrication technique on phosphorus-doped sili- specific heat at low temperatures and of spin suscept-
con (Si:P) (Kane, 1998). The qubits are nuclear spins ibility are reviewed, and it is shown how the observed
I¼1/2 of 31P atoms, which can be controlled by strange behaviors of specific heat and spin suscept-
nuclear magnetic resonance (NMR). The nuclear ibility can be explained by the theory described in
spins interact with electron spins in the phosphorus this subsection. In Section 1.03.3.4, numerical simula-
atoms by the hyperfine interaction. Any qubit opera- tion based on a cluster model is reviewed. Then, it is
tion is achieved by tuning the resonant frequency of shown how the physics of the localized states with
NMR and electrical control of electronic states using interstate as well as intrastate interactions is verified
small metallic gates attached on the silicon surface. by the numerical study for real Si:P systems.
One of the key elements in Kane’s quantum compu- The hopping conduction is described in Section
tation is the control of an antiferromagnetic coupling 1.03.4. In the impurity band the variable-range
between electron spins at adjacent donors (Section hopping plays an important role, as was proposed
1.03.1.2). This coupling was studied by one of the by Mott (1968). The Mott’s law, resistivity
authors of this chapter with his collaborators in the (T) _ exp[(T0/T)1/4] as a function of temperature
1970s for the impurity band of Si:P with intermediate T, is described in Section 1.03.4.1. Then it is
concentration regime, to explain experimental shown how the Mott’s law is modified in the low-
results of the anomalous behavior of specific heat at concentration regime of compensated samples. In
low temperatures. The spin coupling can be ferro- connection with this topic, the so-called Coulomb
magnetic as well as antiferromagnetic in the case of gap, which is a gap in the density of states at the
impurity band (Section 1.03.3.3). Therefore, our Fermi level caused by the long-range Coulomb
study of electronic states in the impurity band should interaction, is discussed in Section 1.03.4.2. In
be noteworthy from the modern point of view. Section 1.03.4.3, we describe the hopping conductiv-
The organization of this chapter is as follows. In ity in the intermediate concentration regime using
the rest of the present section, we briefly explain the the theoretical formalism given in Section 1.03.3. In
hydrogen 1s-like electronic state in an isolated particular, theoretical work on the magnetoresistance
impurity doped in semiconductor (Section 1.03.1.1) developed by Kurobe and Kamimura (1982) is
and the basic idea of Kane’s quantum computer using reviewed in Section 1.03.4.3, and that on magnetoca-
Si:P (Section 1.03.1.2). In Section 1.03.2, we overview pacitance in Section 1.03.4.4. It should be mentioned
the transport phenomena in the impurity band and that their spin-dependent theory has been applied
the localization theories of electronic states. The not only to impurity bands in doped semiconductors
historical development of localization theory and but also to a wide variety of materials and of systems
MI transition in doped semiconductors are also such as sandwiched semiconductors, amorphous
surveyed. semiconductors, doped polymers, carbon fibers, etc.,
Sections 1.03.3 and 1.03.4 are the main parts in this as briefly mentioned in Section 1.03.4.5.
chapter. These sections are devoted to a theoretical In reviewing theories for impurity bands in
investigation of the electronic states and hopping doped semiconductors, we should first mention
conduction in the insulating regimes of doped semi- that we are greatly indebted to the textbooks by
conductors. Section 1.03.3 begins with the theoretical Mott (1990), Kamimura (1986, 1985), Kamimura
formulation to examine the interplay between disor- and Aoki (1989), and Shklovskii and Efros (1984).
der and electron–electron interaction (Section
1.03.3.1). In Section 1.03.3.2 the important role of
1.03.1.1 Isolated Donor Atom in
the intrastate interaction in localized states, the lar-
Semiconductors
gest component of the interaction, is described in
detail. The description clarifies that the intrastate We briefly explain the electronic state in an isolated
interaction results in the coexistence of doubly occu- donor atom doped in silicon (Si) or germanium (Ge).
pied, singly occupied, and unoccupied states. The These elements possess a diamond crystal structure
experimental result of specific heat is compared in which (ns)2(np)2 valence electrons form sp3 cova-
with the theoretical result. In Section 1.03.3.3, we lent bonds. When an atom of group-V element is
describe the effects of the interstate interactions, doped into the materials, four valence electrons in
particularly antiferromagnetic or ferromagnetic cou- the atom participate in forming the sp3 covalent
pling between electron spins in the singly occupied bonds, whereas a remaining electron is bound to the
states. The theoretical and experimental results of nucleus of the atom by the Coulomb potential:
80 Impurity Bands in Group-IV Semiconductors
Here, "1 is the energy required to eject an electron in a regular lattice (Hubbard, 1963, 1964a, 1964b;
into the conduction band. Mott, 1976):
As regards the origin of the concentration- X y
X
dependent activation energy, "2, it was first ascribed H¼ tij ci cj þ U ni" ni# ð12Þ
ij i
to the energy required to remove an electron from
y
one donor and transfer it on another donor that where ci and ci are creation and annihilation opera-
already has an electron. The final state is called tors of an electron at site i with spin , respectively.
D ion, that is, a negative donor ion. Figure 2 The transfer integral tij is a constant V for the nearest
shows that, with increasing donor concentration, neighbors and zero otherwise. Further, U is the
both the donor ground (1s) state and D state Coulomb energy when two electrons occupy the
broaden to form a band of energy levels. They are same site, or it corresponds to the energy of D
called D band and D band, respectively. In the state. Let us consider the half-filling case where the
uncompensated case, the D band is completely filled number of electrons is the same as the number of
by electrons and the conduction occurs only by lattice sites (donors), corresponding to the uncom-
exciting electrons from the D band to the D band. pensated case in doped semiconductors. When the
Thus, the activation energy, "2, corresponds to the distance between the donors is so large that the
energy gap between these bands. It is seen in transfer integral V is much smaller than interaction
Figure 2 that this gap energy, "2, decreases as the energy U, it is favorable that each electron stays at
donor concentration, nD, increases and finally it one donor owing to the repulsive interaction energy
vanishes at a certain concentration, nc, at which between two electrons. This is the case of insulating
the MI transition takes place. Since the appearance phase (Mott insulator). With increasing donor con-
of the D band is due to the electron-correlation centration, nD, the donor–donor distance decreases
effect in a donor atom, this MI transition was first and in consequence the transfer integral, V, increases:
considered as the Mott transition, which was the D and D bands in Figure 2, which are called
induced by electron–electron interaction. lower and upper Hubbard bands, respectively, in
terms of the Hubbard model, are broadened with V.
When V U, the electrons move around to gain the
1.03.2.2 Mott Transition kinetic energy, which corresponds to the metallic
phase. At a critical value of the concentration, nc,
Mott pointed out an important role of electron– the MI transition takes place. This MI transition
electron interaction in the electronic states, originally due to the electron–electron interaction is called
to explain the insulating character of NiO (Mott, Mott transition.
1990). If the disorder effect is irrelevant in the Nowadays, it is known that the disorder effect
impurity band, the simplest model to consider the is relevant to the MI transition in the impurity
electron–electron interaction is the Hubbard model band (Anderson transition) rather than the elec-
tron–electron interaction (Mott transition). On the
other hand, the Hubbard model has been intensively
studied in the context to investigate the strongly
D − state D − band correlated systems, including copper-oxide materi-
ε2 als of high-Tc superconductivity (see a review
article by Imada et al. (1998)).
D state D band
ψ(r)
from each other by a critical energy Ec:
ðEÞ ¼ 0 for E < Ec r
ð13Þ
ðEÞ 6¼ 0 for E > Ec
where (E) represents the mobility at energy E. This
(b)
is why Ec is called as the mobility edge.
In the late 1960s, Mott developed an idea of exp(−|r − r0| /ξ)
ψ(r)
localization by disorder and suggested that the effect
of disorder plays an important role in the impurity
r
band (Mott, 1990; Mott and Davis, 1979). According r0
to his idea, at low donor concentration all states in
the D band are localized in the Anderson sense,
which he called Anderson-localized states. As the
donor concentration nD increases, D and D bands Figure 4 Wave function of an electron in (a) extended
states and (b) Anderson-localized states.
broaden and eventually overlap with each other. In
this case the states at the upper part of the D band
become extended, as in Figure 3, and a mobility edge
Ec separates extended states from localized states. The existence of localized states around EF for
This is the situation in the intermediate concentra- nD < nc was supported by experimental results of
tion regime, where the current is carried by electrons variable-range hopping conduction in Ge:Sb at very
excited above the mobility edge Ec. The activation low temperatures in the intermediate concentration
energy, "2, in Equation (10) is thus given by regime by Allien and Adkins (1972). Prior to their
experiment, the conduction was predicted by Mott
" 2 ¼ Ec – EF ð14Þ (1968), in which the conductivity varies with tem-
where EF is the Fermi energy. perature T as
As nD further increases, Ec EF diminishes and even-
hop ¼ 4 expð – A=T 1=4 Þ ð15Þ
tually disappears at nD ¼ nc. For nD > nc, the system
shows a metallic behavior. In this case, the mechan- This hopping conduction is treated in detail in
ism of the MI transition is of one-electron nature, Section 1.03.4.
namely, at nc the wave function at EF changes from a In addition to the concept of the variable-range
localized nature illustrated in Figure 4(b) to an hopping conduction, Mott (1970, 1972) also pro-
extended one in Figure 4(a). This MI transition is posed another important concept, minimum
called where Ef is the Fermi energy Anderson transi- metallic conductivity. According to his idea, the
tion. It was suggested that the MI transition at nc in zero-temperature conductivity (T ¼ 0) jumps at
doped semiconductors is of Anderson type rather nc from zero in the insulating phase to a minimum
than Mott type. metallic value given by
using the Kubo–Greenwood relationship he showed scaling idea of Thouless (1974) to the systems of
that 2 in Equation (10) is equal to min. larger size. In the scaling arguments, Thouless
Although the scaling theory does not favor the related the degree of localization to the sensitivity
presence of min, as we see later, the concepts of of the energy levels to the boundary condition. For
minimum metallic conductivity and variable-range samples of size L, a dimensionless conductance is
hopping stimulated many researchers in the field of defined by
disordered systems.
gðLÞ ¼ "ðLÞ="ðLÞ ð18Þ
where "(L) is the mean shift of the energy levels by
changing the boundary condition from periodic to
1.03.2.4 Computer Studies of Anderson
antiperiodic and "(L) is the mean spacing of energy
Model
levels. If the wave functions at energy E are localized
Through a number of attempts to check the validity with an exponential envelop, like
of Mott’s concepts theoretically and experimentally,
much progress was made in the research of disor- j ðrÞj exp½ – ðEÞjr – r0 j ð19Þ
dered systems in the 1970s. As one such attempt, with (E) being the inverse localization length, then
various theoretical groups tried to calculate the elec- "(L) and hence g(L) will behave as
tronic structures and metallic conductivity of
disordered systems by computer simulation. The gðLÞ ¼ exp½ – 2ðEÞL ð20Þ
computer simulation is based on the Hamiltonian If the wave functions are extended, on the other
proposed by Anderson (1958), called the Anderson hand, "(L) should be basically independent of L.
model. The Anderson model is essentially a tight- Then, g(L) yields the conductance in units of e2/h.
binding model in a regular lattice, with on-site ener- The quantity g is often called the Thouless number.
gies distributed randomly. In the notation of second Licciardello and Thouless found that, with suffi-
quantization, the Anderson model is expressed by the ciently large sample sizes, the states identified as
Hamiltonian extended previously exhibit the Thouless number
X y
X y that decreases with increasing sample size. They
H¼ "i ci ci þ tij ci cj ð17Þ
i ij
suggested, therefore, that these states are actually
localized states with long localization lengths. This
where i, j represent the sites in a regular lattice. The also meant that their earlier results, which were con-
one-electron energy, "i, which takes a random value, sistent with the existence of minimum metallic
is uniformly distributed over the range from W/2 conductivity, might not be conclusive.
to W/2. The transfer integral is tij ¼ V for the nearest This conclusion affected the research on disor-
neighbors and zero otherwise. The computer simula- dered systems since most researchers had assumed
tion can be carried out more easily for two- the existence of the minimum metallic conductivity.
dimensional systems than three-dimensional systems. Since the computer simulation was limited to sys-
Edwards and Thouless (1972) were probably the tems of finite size, it was necessary to derive real
pioneers for such a calculation. Yoshino and conductance in a system of infinite size. For this
Okazaki (1976, 1977) performed elaborate calcula- purpose, the scaling theory was most appropriate.
tions on a square lattice of size 100 100 and This was the dawn of the scaling theory of
obtained the eigenstates in this system. They found localization.
that for W/V ¼ 8 the wave functions at the center of
the band are strongly localized and decay exponen-
tially with a localization length of about 5 atomic
1.03.2.5 Scaling Theory of Localization
spacings. Further, they inferred that the MI transition
might take place at W/V 6.5. Abrahams et al. (1979) developed a scaling theory
Although the existence of exponentially localized based on the arguments of Thouless (1974). The
states in two dimensions was established from their problem concerns the behavior of the conductance g
results, there remained a problem concerning the in Equation (18) when the linear-dimension L of the
existence of extended states. Licciardello and system is changed. To change the system size, cubes
Thouless (1978) extended their earlier calculations of size L in d dimensions are put together to make a
(Licciardello and Thouless, 1975a,b) based on the larger cube of size bL. Abrahams et al. (1979, 1980)
Impurity Bands in Group-IV Semiconductors 85
asserted that g(L) is the relevant dimensionless ratio 2. For d ¼ 2, there is no true metallic conduction. All
that determines the change of energy levels when the the states are localized in the limit of infinite
hypercubes are fitted together. Thus, when a new g system size. The conductance as a function of L
for larger blocks is computed, the only relevant quan- always decreases with increasing L in two dimen-
tity determining the new g(bL) is the old one, namely sions, with a logarithmic dependence for small g
g(L). Hence, and an exponential dependence for larger g.
gðbLÞ ¼ f ½b; gðLÞ ð21Þ
The scaling theory gave great impetus to the
where f is a universal function depending only on the problem of the impurity conduction. In order to
dimensionality of the system. Taking an infinitesimal check experimentally whether or not the minimum
increase in the scale size, b ¼ 1 þ ", Abrahams et al. metallic conductivity exists at the MI transition in
obtained the following scaling equation for g(L): doped semiconductors, Rosenbaum et al. (1980) mea-
sured the conductivity of uncompensated Si:P
d ln gðLÞ 1 qf ðb; gÞ
¼ X ðgÞ ð22Þ samples with donor concentration very close to the
d ln L gðLÞ qb b¼1 MI transition. Their result indicates that the transi-
where (g) is called scaling function. tion is very sharp but the conductivity extrapolated
The asymptotic forms of (g) are given for g ! 0 to T ¼ 0 goes below the value of the minimum metal-
and g ! 1 in the following: (1) in the strongly loca- lic conductivity, min, predicted by Mott. Paalanen
lized regime (g ! 0), we have g(L) _ exp(2L/
) et al. (1982) tried to confirm the above-mentioned
with
being the localization length, so that (g) ! experimental results by making a high-resolution,
ln g þ constant; and (2) in the opposite limit of g !1, zero-temperature study of the MI transition, apply-
the conductance is given in terms of the conductivity ing uniaxial stress to one sample of slightly insulating
, as g(L) _ Ld 2 in d space dimensions, which concentration. Figure 6 shows their experimental
yields (g) ! d 2. Assuming that (g) is monotonic results, which can be fitted by the form
in g, Abrahams et al. obtained the behavior of the
ð0Þ ¼ c ½ðn=nc Þ – 1 & ð23Þ
scaling function as a function of g for one (d ¼ 1),
two (d ¼ 2), and three (d ¼ 3) dimensions, as illu- with n being the electron concentration (n ¼ nD in
strated in Figure 5. Important conclusions deduced the uncompensated samples), where c ¼ 260 20
from the figure are:
1. For dimensionality d > 2, there exists a mobility
edge at which conductivity continuously tends to 103
zero. This implied that, first, there is no minimum Insulator Metal
metallic conductivity, and, second, d ¼ 2 is the
lower critical dimensionality, above which both
σ (T = 0) (Ω−1cm−1)
σmin
1
10
gc d=3
β(g)
0 g
d=2
−1 1
0 2 4 6
d=1
nD (1018 cm−3)
( cm)1 and ¼ 0.55 0.10. This observed criti- determine the critical exponent (Milligan et al.,
cal behavior is not consistent with the scaling theory 1985; Löhneysen, 2003). Recently, Itoh et al. (1996,
by Abrahams et al. that predicted ¼ 1. The incon- 2004) performed a detailed study for the critical
sistency between the scaling theory and experimental behavior of conductivity on the metallic side and
result in Figure 6 may suggest an important role of that of dielectric susceptibility on the insulator side
the electron correlation at the transition in Si:P sys- in homogeneously doped p-type Ge samples. To
tems, which is not taken into account in the scaling examine the vicinity of the MI transition, the pre-
theory. paration of homogeneously doped samples is
The experimental result for another required. For this purpose, they employed the neu-
three-dimensional disordered system, Si-doped Alx tron transmutation doping (NTD) technique to a
Ga1xAs, is shown in Figure 7 (Katsumoto et al., chemically pure and isotropically enriched 70Ge
1987). In this system, carriers are created when elec- crystal. When the samples are irradiated with ther-
trons trapped in deep donor states are photoexcited to mal neutrons, some of the 70Ge atoms capture a
conductive states, resulting in the persistent photo- neutron and change to 71Ga acceptors. This NTD
conductivity. By this method, Katsumoto et al. tuned technique enables homogeneous doping to the
the electron concentration, n, to a resolution of 0.2% atomic level and fine-tuning of the carrier concen-
around the critical concentration. Figure 7 clearly tration. They concluded that 1 for compensated
shows that the conductivity changes as _ n/nc 1 samples and 0.5 for uncompensated samples (Itoh
in the metallic region (n > nc), indicating ¼ 1 in et al., 1996, 2004). Even in uncompensated samples,
accordance with the scaling theory. the critical behavior with 1 was observed in the
Regarding the puzzle of a critical exponent, that very vicinity of MI transition. However, this should
is, about a question of whether the value of is be attributable to unexpected compensation effect.
explained by the scaling theory or not, there have Their experimental result implies that the scaling
been a significant amount of experimental work to theory can describe the MI transition for the com-
pensated samples but cannot for the uncompensated
samples, possibly due to strong correlation effect in
n (1016 cm−3)
the latter. In this context, the understanding of the
2.4 2.8 MI transition requires further theoretical studies,
taking into account the electron–electron interaction
(Kawabata, 1985; Altshuler and Aronov, 1985;
Fukuyama, 1985; Kaveh, 1985; Di Castro, 1988;
0.4 Belitz and Kirkpatrick, 1994).
σ (S cm−1)
2 0.2
1.03.3 Electron–Electron
Interactions in Impurity Bands
0
0 0.02
There are two main streams of theoretical approach
n/nc−1
to investigate the MI transition in doped semicon-
ductors, which are complementary to each other.
The first stream of approach starts from the free-
0 electron picture in the metallic phase and investi-
520 600 gates how the localized states begin to be formed by
Total exposure time (a.u.) the scattering of free electrons with impurities. These
Figure 7 Experimental results for the conductivity as a studies were prevalent with use of the diagrammatic
function of the concentration of photoexcited electrons in Si- expansion and renormalization technique for the
doped AlxGa1xAs. The carrier concentration n is tuned by so-called weakly localized regime of doped semicon-
changing the total exposure time of light. The conductivity
ductors (Lee and Ramakrishnan, 1985). The second
changes as _ nD/nc 1 in the metallic region (n > nc) in the
vicinity of critical concentration nc (inset). Modified from stream of approach covers the insulating phase con-
Katsumoto S, Komori F, Sano N, and Kobayashi S (1987) sisting of the Anderson-localized states and
Journal of the Physical Society of Japan 56: 2259. investigates the Anderson localization itself.
Impurity Bands in Group-IV Semiconductors 87
In this section, we describe the theoretical work both random variables, while the second part
along the latter stream (Kurobe and Kamimura, 1982, represents the electron–electron Coulomb interac-
1983; Kamimura, 1986, 1985, 1978b, 1978a, 1980, tion U, where hij jU jkl i is its matrix element. If we
1982; Kamimura and Yamaguchi, 1978; Yamaguchi adopt the eigenstates of H0 as a set of basis func-
et al., 1979, 1980; Kamimura et al., 1982, 1983; tions, called transfer-diagonal representation, the
Takemori and Kamimura, 1982, 1983a, 1983b; Hamiltonian can be rewritten as
Kurobe et al., 1984). In the preceding section, we X 1X
have already seen that both the electron correlation H¼ " n̂ þ U n̂ n̂ –
2
and disorder play important roles in the MI transi- 1 X9 X y y
tions in doped semiconductors. Here, we would like þ U c c 9 c9 c ð25Þ
2 9
to point out remarkable effects of electron–electron
interaction in the intermediate concentration regime where ji represents an eigenstate of H0 with eigen-
of doped semiconductors. In this regime, T-linear energy ". It is expressed as
specific heat was observed by Kobayashi et al. X
(1977). This fact means that the density of states at ji ¼ vi jii ð26Þ
i
the Fermi level D(EF) is finite in spite of the insulat-
ing phase. Nevertheless, the spin susceptibility does with a unitary matrix vi. Here, the one-electron
not obey the Pauli law. Ue and Maekawa (1971) and energy " is also a random variable. We have also
Quirt and Marko (1972a,b) observed Curie-type sus- introduced the notation of
ceptibility in the intermediate concentration regime
y
in electron spin resonance (ESR) experiments. The n̂ X c c ; U X h jU ji; U X U
occurrence of the Curie-type susceptibility should be In Equation (25) we have separated the electron–
ascribable to the effect of electron correlation, as electron interactions into those within a state ji
pointed out by various authors (Kamimura, 1978a, and the remaining off-diagonal interactions,
1978b; Kapian et al., 1971; Kobayashi et al., 1978). denoted by the summation 9 excluding the case
Therefore, we have to investigate the electronic of ¼ ¼ ¼ . The former are called intrastate
states in the intermediate concentration regime, tak- interactions and the latter interstate interactions.
ing into account both effect of the electron–electron The interstate interactions are further classified
interaction and disorder on an equal footing. into direct Coulomb U , direct exchange
U , and the remaining interactions as we shall
see later.
1.03.3.1 Theoretical Formulation
1.03.3.1.1 Transfer diagonal
representation 1.03.3.1.2 Anderson-localized states
Following Kamimura (1985, 1982), the Hamiltonian in If there is stronger disorder, the transfer-diagonal
the presence of both disorder and electron–electron representation bases ji are localized in the
interaction is written generally in the tight-binding Anderson sense although there exist also extended
scheme as states above the mobility edge. In the strongly loca-
lized regime the amplitude of the wave functions
H ¼ H0 þ H1 decreases with the distance apart from the localiza-
P y P y tion center, on the average, in an exponential
H0 ¼ "i ci ci þ tij ci cj
i ij ð24Þ envelope function (Figure 3(b)). Thus, the coeffi-
1XX y y cients in the linear combination of impurity wave
H1 ¼ hij jU jklici cj 9 cl9 ck
2 ijkl 9
functions in Equation (26) can be written as
where y
ci and ci are the creation and annihilation vi ¼ v~i expð – jri – r j=
Þ ð27Þ
operators of an electron at the ith site with spin , where v~i represents a random phase factor,
the
respectively, ji i is a Wannier function at the ith localization length, and r the center of the localized
site with one-electron energy "i, and tij is the state ji. From the normalization condition, hji ¼ 1,
transfer integral between ith and jth sites. In this we obtain
Hamiltonian, the first part represents the single-
particle Hamiltonian with "i and tij, which are jvei j2 ¼ 4ðR0 =
Þ3 =3 ð28Þ
88 Impurity Bands in Group-IV Semiconductors
χ (μ 2B n/W )
intrastate interaction energy:
10 20 40 80
Z W 1
1
U ¼ U ðEÞ dE ð46Þ
W 0
W/U = 5
The numerical results of spin susceptibility and spe-
cific heat are presented in Figure 9 for several values
of W=U . 0.1
As seen in Figure 9(a), the spin susceptibility 0.1 1 10
shows Curie-type behavior at low temperatures so kB T/U
that kB T U . This is due to the localized spins of (b) 0.6
SO states and is expressed as
2B nS =kB T ð47Þ
20
where ns is the number of SO states per unit volume. It
should be noted that the appearance of this Curie-type 10
0.4
susceptibility is the result of the intrastate interaction,
because in the absence of this interaction the Pauli-
C/kB n
20 18 −3
n =1.2 × 10
18 −3
cm 0.7 × 10 cm
100 D
C (μJK −1 mol−1)
10
χ−1 (mole/e.m.u.)
18 −3
0.1 × 10 cm
50
0
150
18 −3
n =3.7 × 10 cm
100 D
0 2 4 50
T (K)
Figure 10 Electronic specific heat of Si:P for 0 2 4
nD ¼ 1.7 1018 cm3 as a function of temperature. The full
T (K)
curve represents the theoretical result by Yamaguchi et al.
(1979) with W ¼ 19 and U ¼ 3.4 meV, whereas the circles Figure 11 Curie–Weiss plots of the molar donor
represent the experimental result reported by Kobayashi susceptibility for several concentrations of phosphorus
et al. (1977). The enhanced region of the specific heat above reported by Andres et al. (1981). The solid line shows
the T-linear line is shown as the hatched area. free-spin behavior.
92 Impurity Bands in Group-IV Semiconductors
(a)
9.7
(a)
6
10 4.9
4
5 2
H = 0 kG
χ (μ 2B /K)
ΔC/kB
0 0
(b) (b)
10
6 H = 0 kG
9.7
4.9
5 4
2
0 0.2 0.4 0.6 0.8
T (K)
0 0.2 0.4 0.6 0.8
T (K)
Figure 13 Spin susceptibility and anomaly in electronic specific heat, C per pair of singly occupied states, as functions
of temperature T, calculated in the spin-pair model for (a) spin-triplet pairs and (b) spin-singlet pairs. Modified from Takemori T
and Kamimura H (1982) Solid State Communications 41: 885.
the physical picture described in the preceding sub- specific heat, and spin susceptibility do not depend
sections is confirmed. Furthermore, the direct on the special choice of the one-electron
calculations of the specific heat and spin susceptibil- Hamiltonian.
ity reproduce the characteristic features observed in As one choice, the one-electron Hamiltonian H0 is
Si:P. taken in such a way as
hijH0 jii ¼ h j jH0 jj i ¼ hijK jii þ hijVi jii ð66Þ
1.03.3.4.1 Gaussian model
for the diagonal elements and
The Hamiltonian is composed of the kinetic energy
of electrons K, Coulomb potential due to donor hijH0 jj i ¼ hijK j j i þ hijVi j j i þ hijVj j j i ð67Þ
impurities V, and electron–electron interaction U:
for the transfer energy. Note that the Coulomb
H ¼K þV þU ð61Þ potentials due to impurities other than i (i and j) do
where not contribute to the ith diagonal element (transfer
energy between i and j). This choice of one-electron
XNe
1 2 Hamiltonian effectively introduces the long-range
K ¼ p ð62Þ screening effect into the one-electron Hamiltonian.
i¼1
2m i
The one-electron Hamiltonian H0 is diagonalized,
X
Ne
and the eigenstates and eigenenergies are adopted
V ¼ Vi ð63Þ
i¼1
as the orthonormal basis functions and corresponding
state energies for the transfer-diagonal representa-
X e2
U¼ ð64Þ tion. Using this basis set, all the matrix elements of
i<j
jri – rj j the Hamiltonian are calculated.
Next, we calculate the matrix elements of the
Here, ri and pi are the space coordinate and momen- Hamiltonian in the multielectron configuration
tum of the ith electron, m is the effective mass, and space. The total number of the configurations in
Vi the Coulomb potential of the i th donor. The which six electrons occupy the six transfer-diagonal
number of electrons, Ne, is identical to that of donor !
12
impurities in uncompensated samples considered states is ¼ 924 if the spin degrees of freedom
here. No valley degeneracy of the conduction band 6
in Si is assumed. are considered. Although the Hamiltonian has no
The donor centers are distributed randomly in a symmetry in the coordinate space, it is invariant
sphere of volume Ne/nD, where nD is the donor against the rotation in the spin space. Each many-
concentration. A Gaussian wave function body eigenstate of the Hamiltonian is always an eigen-
state of total spin S and its z-component Sz and thus
i ðr Þ ¼ expð – jr – Ri j2 =2 Þ ð65Þ
either a spin-degenerate multiplet or a nondegenerate
is attached to each donor center Ri and the electronic singlet. It is, therefore, convenient to confine ourselves
wave functions are represented by their superposi- to the subspace with Sz ¼ 0. This reduces the number
tions. The spatial extent in Equation (65) is of configurations to 400, while we still retain the
determined to be 1.88 times the effective Bohr radius whole energy spectrum. The energy levels and
so as to give the largest binding energy (0.82 Ry) for many-electron wave functions in a cluster are then
an isolated donor. The Gaussian function is obtained by diagonalizing the obtained 400 400
employed instead of the usual exponential 1s func- Hamiltonian matrix in the configuration space.
tion, which enables us to obtain the analytical Since the Hamiltonian is diagonalized exactly, the
expression for all the matrix elements for the elec- calculated results do not depend on the choice of
tron–electron interaction (Boys, 1950). basis set for one-electron states. Advanced methods
First, the one-electron Hamiltonian is diagona- developed in quantum chemistry yield a more suita-
lized to obtain the orthonormal base for the ble basis set to elucidate many-body states. We
transfer-diagonal representation. On doing so, an adopted the multi-configuration self-consistent field
appropriate choice of the one-electron Hamiltonian (MCSCF) method to discuss the Anderson-localized
is required in order to facilitate the interpretation states in the uncompensated samples of Si:P (Eto and
of the results although the final calculated results Kamimura, 1988, 1989) and in the compensated
of physical quantities such as energy levels, samples (Eto and Kamimura, 1989).
96 Impurity Bands in Group-IV Semiconductors
jgrdij1"2"3";1#2#4#i – j1"2"4";1#2#3#i
1.0
The mixing coefficients have the opposite signs for
the two configurations, indicating that spins are par-
allel in states 3 and 4 in the present notation so that
the ground state is triply degenerate with total spin
0.5 S ¼ 1.
As regards the first excited state, the average
occupation numbers of transfer-diagonal states are
εo almost the same as in the ground state, as seen in
Table 1. The largest contributions come also
0
–1.6 –0.8 0.0 0.8 1.6 from the same two configurations as those in the
E (Ry*) ground state, but with the mixing coefficients of
Figure 15 Distribution of transfer-diagonal state energies the same sign. This means that electron spins are
of the six-donor clusters corresponding to the donor antiparallel in states 3 and 4 and hence the first
concentration nD ¼ 1.7 1018 cm3 of Si:P. excited state is a spin singlet (nondegenerate) with
Impurity Bands in Group-IV Semiconductors 97
(a)
(b)
Table 1 The state energies of transfer-diagonal states in a total spin S ¼ 0. The next largest components are
cluster shown in Figure 16
the configurations in which the transfer-diagonal
Average occupation state 3 or 4 is doubly occupied. This is due to the
number kinetic-exchange interaction between the two
Ground First excited
transfer-diagonal states, discussed in Section
State State energy " (Ry) state state 1.03.3.3.1. The effective exchange interaction J is
given by the energy difference E between the
1 1.474 1.947 1.958
ground and the first excited states, that is,
2 1.104 1.742 1.755
3 0.834 1.000 1.096 J ¼ E ¼ 0.0044 Ry.
4 0.582 1.001 0.916 Besides the spin-triplet clusters, there are a
5 0.424 0.289 0.276 large number of clusters in which the transfer-
6 1.202 0.018 0.001 diagonal states 3 and 4 are singly occupied with
The average occupation numbers are also tabulated in the ground antiparallel spins in the spin-singlet ground state
and first excited states. (S ¼ 0) and parallel spins in the first excited states
98 Impurity Bands in Group-IV Semiconductors
(S ¼ 1). In such cases, the energy difference can be magnetic field, reflecting the excitation among the
regarded as the antiferromagnetic coupling J Zeeman-split states. The specific heat increases again
between the spins in states 3 and 4. This result at higher temperatures, due to the excitations that
indicates the coexistence of ferromagnetic and change the occupation numbers of transfer-diagonal
antiferromagnetic interactions between the spins states.
in SO states. For a spin-singlet cluster (Figure 18(b)), a
Schottky-type hump appears in the specific heat,
but it is insensitive to the magnetic field. The peak
1.03.3.4.3 Specific heat and spin
of C is situated at kBT 102Ry, which reflects the
susceptibility
excitation energy to the spin-triplet state, E ¼ J.
Since all the eigenstates and eigenenergies for each clus-
The spin susceptibility is quite small compared
ter are obtained by the numerical study, we can readily
with the spin-triplet clusters. The tends to zero at
calculate the spin susceptibility and specific heat. In the
low temperatures, whereas increases with tempera-
presence of magnetic field H, the energy level En of each
tures up to kBT E because the excited state is a
spin multiplet splits into the Zeeman terms:
spin triplet.
Eðn; Sz Þ ¼ En þ 2B Sz H ð68Þ In order to simulate the experimental observation;
the specific heat and spin susceptibility are averaged
if we neglect the higher-order effects.
over all the clusters for the same donor concentra-
The specific heat C and spin susceptibility are
tion. The calculated results of specific heat for
then calculated by the following equations:
various magnetic fields are plotted against tempera-
E 2 E 2 ture in Figure 19. At high temperatures, such as
C=kB ¼ – kBT > 6 103 Ry, the curves for H ¼ 0 are almost
kB T kB T
2 T-linear but do not extrapolate to zero at T ¼ 0, in
P Eðn; Sz Þ exp½ – Eðn; Sz Þ=ðkB T Þ
¼ contrast to the experimental results. As discussed in
n;Sz kB T Z
Section 1.03.3.3.3, the T-linear specific heat is caused
" #2 by continuous distribution of state energy ". For the
X Eðn; Sz Þ exp½ – Eðn; Sz Þ=ðkB T Þ
– ð69Þ clusters of a finite size, the state energies " are
kB T Z
n;Sz discrete and the calculated specific heat retains the
and structure of the discrete excitations. Therefore, the
weak downward bending of the calculated curves of
X S 2 exp½ – Eðn; Sz Þ=ðkB T Þ
Sz2 z
C should be attributable to the limited size of the
¼ ¼ cluster. On the other hand, there appears a hump in
42B kB T H ¼0 k T
n;Sz B
Z
H ¼0
the specific heat for H ¼ 0 that grows and shifts to
ð70Þ
higher temperatures with increasing magnetic field.
where the partition function Z is given by As seen above, this is due to the ferromagnetic
X exchange interaction between the spins of SO states.
Z¼ exp½ – Eðn; Sz Þ=ðkB T Þ ð71Þ The hump is smaller relative to the T-linear part for
n;Sz
higher donor concentration. This indicates that
and the summation over Sz is taken from S to S the ratio of the number of SO states to that of donors,
when the energy level n has the total spin S. ns/nD, decreases with increasing nD, in agreement
Figure 18(a) shows the calculated results of the with the experimental data on Si:P (Kobayashi, 1979).
specific heat and spin susceptibility of a spin-triplet Figure 20 shows the Curie–Weiss plot of the
cluster. The spin susceptibility is of Curie type with a ensemble-averaged spin susceptibility for several
slight deviation from the Curie law owing to the values of donor concentration in Si:P. The curves
presence of spin-singlet excited states. The are almost T-linear (Curie type) with a downward
Schottky-type hump in the specific heat, which is bending at low temperatures. This characteristic
located at kBT 103 Ry in the absence of magnetic behavior is in good agreement with the experiment
field, is caused by the ferromagnetic exchange inter- on Si:P by Andres et al. (1981). The Curie–Weiss plot
action J between the SO states. The lowest state is a for six free spins is also plotted in the figure for
spin triplet and this state splits into three components comparison. The ratio of the number of SO states is
of Sz ¼ 1, 0, 1 in magnetic fields. The hump of C estimated from the susceptibility at
shifts to higher temperatures with increasing kBT ¼ 1.4 102 Ry,
Impurity Bands in Group-IV Semiconductors 99
(a) (b)
C/kB 0.8 2.0
C/kB
0.4 H = 3 × 10−3 (Ry*/μ B) 1.0
H=0
2 1 × 10−3 (Ry*/μ B)
1 2
3
0
0 0
12
6
χ/μ 2B (103 Ry*–1)
4 2
0 0
0.8 1.6 8 16
kBT (10−2 Ry*) kBT (10−2 Ry*)
Figure 18 Electronic specific heat C for various values of magnetic field and spin susceptibility , as functions of
temperature T, of (a) a spin-triplet cluster and (b) a spin-singlet cluster consisting of six donors.
nS =nD ¼ cluster =free spins ¼ 28:3% ð72Þ experiment on Si:P shows that the binding energy of
18 3 the orbitally nondegenerate one-electron state with
for nD ¼ 1.7 10 cm . The molar donor suscept-
A1 symmetry is larger than that for the other states
ibility decreases with increasing donor concentration
by 12 meV (Aggarwal and Ramdas, 1965). This
nD, which is also in accordance with the observation
in Si:P. This also indicates a decrease in the number implies that, for T < 100 K, the electronic state for
of SO states with increasing nD. the first electron is expressed as a superposition only
of A1 states of donors with no appreciable valley
effects.
1.03.3.4.4 Comparison with experiment When two electrons occupy a donor, the electro-
In order to compare the calculated results with nic configuration is specified by assigning the first
experimental data on Si:P, it is necessary to take electron to the orbital of A1 symmetry and the second
into account multivalley effects in the conduction electron to that of A1, E, or T2 symmetry with appro-
band of Si, because no valley degeneracy is assumed priate spin configuration. Thus, the two-electron
for the numerical calculations. There are six valleys energy levels are split due to the valley–orbit inter-
in the conduction band of Si (see Section 1.03.1.1). actions and by the Coulomb interaction between the
The crystal field of tetrahedral symmetry around an two electrons. The energy splitting can be estimated
impurity atom splits the one-electron energy levels from the binding energy of the D state, which was
of sixfold degeneracy into a nondegenerate state of evaluated to be 1.01 meV (11.7 K) for Si:P (Natori and
A1 symmetry, doubly degenerate state of E symme- Kamimura, 1977; Kamimura 1979). The splitting is
try, and triply degenerate state of T2 symmetry. expected to be smaller for Anderson-localized states,
This is called valley–orbit splitting. The effect is which are more extended than the isolated-donor
large for the first electron to occupy the state, but state, and we can consider that all the six valleys
small for the second electron because the crystal are available for the second electrons in the tempera-
field is screened by the first electron. The optical ture range larger than 1 K.
100 Impurity Bands in Group-IV Semiconductors
1.2 60
nD = 2.4 × 1018 cm–3
1/χ (10–2 Ry*/μ 2B )
0.8 1.7
40
1.0
0.4
20
Free spins
0 0.8 1.6
kBT (10−2 Ry*) 0 1.0 2.0
T (K)
Figure 20 Curie–Weiss plot of the ensemble-averaged spin
susceptibility for the cluster model of six donors corresponding Figure 21 Specific heat calculated for Si:P with impurity
to various donor concentrations in Si:P. The susceptibility for concentration nD ¼ 1.7 1018 cm3, as a function of
six free spins is also shown (broken line) for comparison. temperature, for H = 0 (full line) and 20 kOe (broken line). The
Modified from Takemori T and Kamimura H (1983a) Journal multivalley effect has been taken into account. Experimental
of Physics C: Solid State Physics 16: 5167; Takemori T and data taken by Kobayashi et al. (1976) are also shown for
Kamimura H (1983b) Advances in Physics 32: 715. H = 0 (triangles) and H = 19.1 kOe (circles).
Impurity Bands in Group-IV Semiconductors 101
1.03.4 Hopping Conduction and in the case of phonon emission with E < 0. Here,
Related Phenomena r is the distance between the localization centers
and
In this section, the hopping conductivity in doped 1 1
semiconductors at low temperatures T is discussed. Nph ðEÞ ¼ ; and f ð"Þ ¼ "=ðk T Þ
e E=ðkB T Þ – 1 e B þ1
When kBT is much smaller than the width of the
impurity band, the variable-range hopping (VRH) are boson and fermion distribution functions, respec-
tively. The state energies are measured from the
ðT Þ ¼ 0 exp½ðT0 =T Þ1=4 ð74Þ Fermi level. The localization length,
, is assumed
to be independent of state energy, so that the transfer
is observed, instead of activation-type conductivity,
integral between the two states gives the factor of
Equation (11), as predicted by Mott (1968).
The Mott’s theory of VRH is briefly reviewed e – 2r =
. The pre-exponential factor,
ph, is related to
in Section 1.03.4.1. In the low-concentration regime the product of the square of electron–phonon matrix
of compensated samples, the VRH shows a different element and the density of states of phonons.
T dependence because of the disappearance of the (Although
ph depends on r , its dependence is
density of states at the Fermi level caused by the much weaker than that of the exponential part.
long-range Coulomb interaction. The theory of Hence,
ph is treated as a constant.) The electron
the so-called Coulomb gap is described in Section spins are disregarded in Sections 1.03.4.1 and 1.03.4.2.
1.03.4.2. In Sections 1.03.4.3 and 1.03.4.4, VRH and At sufficiently low temperatures, it is usually
related phenomena are studied for the intermediate expected that kBT j" " j, j"j, and j" j. Then,
concentration regime, based on the transfer-diagonal the hopping rate can be written in a unified form of
representation formalism presented in the preced- ¼
ph exp½ – 2r =
– =kB T ð77Þ
ing section. The theory successfully explains
experimental results of magnetoresistance and where
magnetocapacitance. 1
" ¼ ½j" – " j þ j" j þ j" j ð78Þ
2
Here is equal to j" " j when the state energies
1.03.4.1 VRH and Mott’s Law
lie on opposite sides of the Fermi energy and max (j"j,
Let us begin with the hopping conduction between j" j) otherwise. The resistance of the hopping is
Anderson-localized states, neglecting the electron– given by
electron interaction. The hopping conductivity is
kB T kB T
usually evaluated using a model of random resis- R ¼ 2
¼ 2 expð2r =
þ " =kB T Þ ð79Þ
tance network proposed by Miller and Abrahams e e
ph
(1960) (Section 1.03.4.1.1). Based on the model, the (Shklovskii and Efros, 1984). Miller and Abrahams
Mott’s law of the VRH is derived in Section (1960) proposed a random resistance network in
1.03.4.1.2. which resistance R connects a pair of vertices, as
shown in Figure 22(a). The conductivity of the
1.03.4.1.1 Formulation of hopping whole system is determined by a set of {R }. An
conduction important feature of the network is its extremely
Consider an elementary hopping process from ji to wide spectrum of resistance, R , due to the expo-
j i when the state energies " and " are different. nential dependence on r and .
The hopping of an electron is accompanied by the When the temperature is not too low, the r -
phonon absorption or emission for the energy con- dependent part in Equation (79) dominantly deter-
servation. The number of electrons making the mines the conductivity in the system. Then electrons
transition per unit time is given by hop from a localized state to its nearest-neighbor
¼
ph e – 2r =
Nph ðEÞf ð" Þ½1 – f ð" Þ ð75Þ state with smallest r , as illustrated by process (i)
in Figure 22(b). The temperature dependence in
in the case of phonon absorption with E ¼ " Equation (79) results in an activation-type
" > 0, and conductivity:
¼
ph e – 2r =
½Nph ð – EÞ þ 1 f ð" Þ½1 – f ð" Þ ð76Þ ¼ 3 expð – "3 =kB T Þ
102 Impurity Bands in Group-IV Semiconductors
r¼ ð80Þ
8Dð0ÞkB T
and average hopping energy
α
Rαβ β kB
¼ ðT0 T 3 Þ1=4 ð81Þ
4
(b) (ii)
(i) This gives the optimized hopping rate
EF
¼
ph exp½ – ðT0 =T Þ1=4 ð82Þ
Optimized r (T ) 3
with kBT0 ¼ 18.1/[D(0)
].
Figure 22 (a) Random resistance network proposed by The conductivity is expressed as ¼ e 2 DDð0Þ
Miller and Abrahams (1960) to evaluate the hopping from the Einstein relation, where the diffusion con-
conductivity in the Anderson-localized regime. The
stant D is written as D ¼ r 2 . Thus, we arrive at
resistance R between sites and is given by Equation
(79). (b) Schematic drawing for (i) the nearest-neighbor
¼ 0 exp½ – ðT0 =T Þ1=4 ð83Þ
hopping and (ii) variable-range hopping conduction. In the
latter, the hopping length increases with decreasing 2 2
with 0 ¼ e D(0)r
ph, where r is given by Equation
temperature.
(80). Generally, in d-dimensional systems, we have
thus negatively charged, resulting in acceptors. Some whereas the Coulomb energy from donor i is counted
of the donor levels are occupied by an electron and in "j because donor i is occupied in the ground state.
the others are empty. Here, the former are neutral The condition (88) must hold for any pair of donors
while the latter are positively charged. The electro- with "i < 0 and "j > 0.
nic states in the system are identified by the number Consider donors whose energies fall in a narrow
of electrons ni (¼1 or 0) at donor site i if the electron band [ – "~=2; "~=2] around the Fermi level. From
spins are neglected. This situation is described by the Equation (88), any two donors in this band must be
Hamiltonian separated from each other by a distance rij larger
" than e2/"~ when the energies are on the opposite
P
d on e2 1 X don X
don
ð1 – ni Þð1 – nj Þ sides of the Fermi level, "i"j < 0. As a result,
H ¼ "D n i þ
i 2 i j 6¼i
ri j the donor concentration in this band n("~) cannot
#
don X
X acc
1 – ni 1 Xacc Xacc
1 exceed ð3=4Þð"~=e 2 Þ3 . Thus, the density of states
– þ ð86Þ Dð"~Þ ¼ dnð "~Þ=d "~ tends to zero when "~! 0, at least
i k
rik 2 k l6¼k rkl
as fast as "~2 . (D("~) cannot tend to zero faster than "2
where "D is the energy level of the isolated donors, because the decrease in the density of states stems
don and acc mean the summations over all donor from the strong interaction between the low-energy
and acceptor sites, respectively. The second term in states. Otherwise, the separation between the donors is
the Hamiltonian represents the long-range Coulomb too large and hence the interaction is too weak to be
interaction between ionized donors, the third term responsible for lowering the density of states so much.)
the Coulomb interaction between ionized donors and We conclude that
acceptors, and the last term the Coulomb interaction
3 "~2
between acceptors. The set of occupation numbers Dð"~Þ ¼ 3 ð89Þ
{ni} in the ground state is determined by minimizing e6
H with a fixed total number of electrons, or by where 3 is a numerical coefficient. The reduction in
minimizing the density of states in the vicinity of the Fermi level
X
don is called Coulomb gap by Efros and Shklovskii
H~ ¼ H – ni (1975). The width of the Coulomb gap is estimated
i
from D() D0(0), where D0(0) is the density of
The chemical potential is chosen to be zero. states at the Fermi level in the absence of Coulomb
As a result, the energy level at donor site i is given interaction. Thus, e3D0(0)1/2/3/2.
by Similar arguments in the two-dimensional case
" # lead to the law
qH e2 X
don
1 – nj Xacc
1
"i ¼ ¼ "D þ – þ ð87Þ 2 j"~j
qni rij rik Dð"~Þ ¼ 2 ð90Þ
j 6¼i k e4
The numerical coefficients were estimated to be
The determined values of "i’s become random due to
3 ¼ 3/ and 2 ¼ 2/ (Baranovskii et al., 1978).
the long-range Coulomb interaction from the other
The width of the Coulomb gap is e4D0(0)/2 in
impurities. Hence, the width of the impurity band is
the two-dimensional case.
r Þ, where r¼ ð4nD =3Þ – 1=3 is the
of an order of e 2 =ð
The Coulomb gap in the density of states signifi-
average distance between donors.
cantly affects the T dependence of the VRH
For an electron transfer from a filled donor i to an
conduction at temperature T . The hopping
empty donor j in the ground state, the total energy
conductivity is derived in the same way as that in
must be increased:
the preceding subsection. Let us find the optimized
e2 hopping distance r which maximizes in Equation
ij ¼ "j – "i – >0 ð88Þ
rij (77). From the above-mentioned argument, the hop-
The increment ij in energy is derived by the fol- ping energy is evaluated to be " ¼ 9e2/r for a given
lowing argument. First, an electron is transferred hopping distance r, with 9 being a numerical factor.
from donor i to infinity, which requires the energy Hence, we minimize the hopping rate of
"i. Second, the electron is placed on donor j, which is " 0
#
2r e2
empty in the ground state. The energy cost in the ¼
ph exp – –
kB T r
second stage is "j e2/rij since donor j is empty,
104 Impurity Bands in Group-IV Semiconductors
in Shklovskii and Efros (1984). where n is the number operator in state ji and
spin .
Regarding functional forms of the intrinsic hop-
1.03.4.3 Formula for Magnetoresistance in ðlÞ
ping rates , we assume that they are similar to
the Variable-Range Regime
those given in Equation (75) and (76):
As discussed in Section 1.03.3, intrastate and inter- ðlÞ
ðlÞ
state interactions play important roles in the ¼
ph e – 2r =
Nph ðE ðlÞ Þ
Anderson-localized states. In this subsection, we 0 exp½ – 2r =
ðlÞ – E ðlÞ =kB T ð97Þ
show that these interactions are important for the
magnetoresistance in the VRH conduction in the
intermediate concentration regime of uncompen- εβ εβ
EF
sated samples. We assume that the long-range part εα εα
U U
of the Coulomb interaction is screened out by U
εβ
extended states over several impurity sites. Kurobe εα εβ
εα
and Kamimura (1982) calculated the resistivity due
to the VRH between the Anderson-localized states (1) S U (2) S S (3) D U (4) D S
by the percolation method (Pollak, 1972) taking into Figure 23 Four kinds of hopping processes in the
account the intrastate interaction. presence of intrastate interaction U.
Impurity Bands in Group-IV Semiconductors 105
E
T = 0.2 K
Ec
3
Ec
0.5 K
Δρ /ρ
2 EF
1.0 K EF
1
3.0 K
2μ BH
0 10 20 30 40
D–(E ) D+(E )
H (kOe)
Figure 25 Schematic illustration of the negative
Figure 24 Calculated result for the positive component of
contribution to the magnetoresistance due to the state-
the magnetoresistance in the variable-hopping conduction
energy dependence of the localization length.
regime, as a function of magnetic field for various
temperatures, in three dimensions. The parameters used
are U ¼ 3.4 meV, D1(EF) ¼ 9.5 1019/cm3 eV,
1 ¼ 49 Å, and
2 ¼ 98 Å. Modified from Kurobe A and Kamimura H (1982) parallel to the magnetic field, this effect gives rise
Journal of the Physical Society of Japan 51: 1904. to a negative magnetoresistance. Combining this
effect with the opposite effect discussed above, the
total magnetoresistance is written conventionally in
temperature T decreases. The saturation value also the form
increases when the density of states D1(EF) decreases,
or when the localization lengths decrease with the ratio ðH Þ ðH Þ – ð0Þ
¼ ¼ P ðH Þ þ N ðH Þ ð105Þ
of
1/
2 being fixed. The analytical expression for the ð0Þ ð0Þ
saturated value is given by where P(H) and N(H) represent positive and
8 d d 1=ðdþ1Þ negative parts of the magnetoresistance, respectively.
< c ðdÞ
1=ðd þ1Þ
1 þ
2
ðH Þ
ð0Þ sat
j
¼exp
: D1 ðEF ÞkB T
2d
1 þ
2
2d For comparison with experimental results, we
discuss the magnetoresistance in a transition metal
#) ð104Þ dichalcogenide system, 1T-TaS2, which has a layer-
!
d d 1=ðd þ1Þ type sandwich structure with a commensurate
1 þ3
2 –1
–
ð
1d þ
2d Þ2 þ4
22d charge density wave (CDW) phase below
T ¼ 200 K (Williams et al., 1974, 1975; Wilson et al.,
in d dimensions (Kurobe and Kamimura, 1982; 1974, 1975). In this phase, the material is an insula-
Kurobe, 1986). tor with an energy gap which is interpreted as
arising from the CDW modulation. DiSalvo and
1.03.4.3.3 Comparison with experiments Graebner (1977) found a T1/3 law of conduction in
As an effect of the magnetic field on the transition this phase, which should be ascribed to the VRH
rate, we have so far considered the change in occupa- conduction among the Anderson-localized states in
tion probabilities of UO, SO with spin ¼ , and the energy gap in the quasi-two dimensional system.
DO states. In addition, we have to take into account The magnetoresistance was later measured in this
an effect of the magnetic field on the localization system by Kobayashi and Muto (1979), Onuki et al.
length. Since the localization length is a function of (1980), and also by Tanuma (1983). Figure 26 shows
state energy, it is changed through the Zeeman shift, their experimental results of the magnetoresistance
as schematically shown in Figure 25. This mechan- as a function of magnetic field for various
ism was first proposed by Fukuyama and Yosida temperatures.
(1979). The localization length increases to Let us apply the present theory to the above-
(" þ BH) for spin-up electrons while it decreases mentioned experimental results. We assume that
to
(" BH) for spin-down electrons if
(") is an
1(") ¼
2(")/2 and that the localization lengths
increasing function of state energy " (we assume depend on the state energy, ", in the following way:
that EF is located below the mobility edge Ec). Since
the majority carriers have the magnetic moments
1 ð" Þ ¼
2 ð" Þ=2 _ ðEc – " Þ – 1 ð106Þ
Impurity Bands in Group-IV Semiconductors 107
0.2 T = 0.4 K
2
T = 1.42 K
0
1 0.6 K
Δρ /ρ
0.39 K
–0.2 1.0 K
Δρ /ρ
0
–0.4
0.1 K
–1
–0.6
0 50 100
H (kG)
0.05 K
–0.8 Figure 27 Calculated result for the magnetoresistance of a
two-dimensional 1T-TaS2 system, in the presence of
intrastate interactions. Both positive and negative
0 20 40 60 80 components are taken into account. The parameters are
(EcEF) ¼ 1.0 meV, U ¼ 1.7 meV, D1 (EF) ¼ 5.3 1014 cm2eV
H (kOe)
and
1(EF) ¼ 100 Å. Modified from Kamimura H (1982)
Figure 26 Experimental data on magnetoresistance of Progress of Theoretical Physics Supplement 72: 206.
1T-TaS2 in the temperature range from 1.42 to 0.05 K.
Modified from Kobayashi N and Muto Y (1979) Solid State
Communications 30: 337. P(H) disappears in Equation (105). When the tem-
perature is lower than a certain temperature
corresponding to the order of average interstate inter-
where Ec is the mobility edge although there is action, the magnetoresistance becomes negative even
no mobility edge in two-dimensional systems of in low fields. The observation of an anomaly in the
infinite size (Abrahams et al., 1979, 1980). Choosing specific heat of this material (Nishio et al., 1978) sug-
the values of the parameters to be Ec " ¼ 1.0 meV, gests the existence of spin-dependent interstate
U ¼ 1.7 meV, D1(EF) ¼ 5.3 1014 cm2eV, and interactions, as discussed in Section 1.03.3.3, where
1(EF) ¼ 100 Å, the magnetoresistance was calcu- the magnitude of the interaction is estimated to be
lated. The calculated result is shown in Figure 27. about 1 K. If the average interstate interaction is taken
The result shows that the magnetoresistance is posi- as 0.5 K, the calculated result of magnetoresistance is
tive in low magnetic fields and then changes to in good agreement with the experimental results
negative in high magnetic fields. The magnitude of (Kamimura et al., 1982, 1983).
the positive part decreases with increasing tempera-
ture. These features are consistent with the
1.03.4.4 Magnetocapacitance in
experimental results. However, experimental data
Intermediate-Concentration Regime
show only negative magnetoresistance below about
0.1 K (see Figure 26). This behavior does not agree In this subsection, we consider the magnetic-field
with the theoretical result. This discrepancy between dependence of the electrical polarizability for
the theoretical result and experiments can be Si:P in the intermediate-concentration regime. The
explained by the spin-dependent interstate interac- magnetocapacitance reflects elementary processes of
tions between Anderson-localized states, which is the hopping between the Anderson-localized states
given by Equation (53). although it is not a transport property.
When the interstate interaction is antiferromag- New et al. (1982) reported an experimental result
netic, the spins in the SO states retain spin-singlet that the electrical polarizability depends drastically
states and hardly become parallel to the magnetic on the magnetic field. Figure 28(a) shows their
field if the magnetic field is weak. Thus, the suppres- experimental result of the magnetocapacitance
sion of hopping processes of types 2 and 3 in (MC) of Si:P, where the MC is plotted as a function
Figure 23 may not occur, and in consequence of magnetic field H. The MC is defined as
108 Impurity Bands in Group-IV Semiconductors
(a) (b)
10–1
10–1
T=1K
10–2
ln Reff (H)
ln Reff (H)
10–2
T = 1.3 K
2
10–3 3
10–3 4
1.9
5
2.9
4.2
1 10 0.1 1 10
B (T) B (T)
Figure 28 (a) Experimental result for logarithmic of magnetocapacitance, Reff(H), plotted against magnetic field for Si:P at
various temperatures. The donor concentration is nD ¼ 1.0 1018 cm3. (b) Calculated result for Reff(H) for Si:P with donor
concentration nD ¼ 1.7 1018 cm3 with U ¼ 3.4 meV and W ¼ 19 meV. (a) Modified from New D, Lee NK, Tan HS, and Castner
TG (1982) Physical Review Letters 48: 1208. (b) Modified from Kurobe A, Takemori T, and Kamimura H (1984) Physical Review
Letters 52: 1457.
4nD D ¼ 1 þ 2 þ 3 þ 4 ð109Þ
– 0 ¼ ð108Þ
1 – 4nD D =ð30 Þ 2e 2 X hn ð1 – n – Þihð1 – n Þð1 – n – ÞijhjX j ij2
1 ¼
V " – "
Here 0 is the dielectric constant of the host material. ð110Þ
The electrical polarizability is obtained from the
relation of ¼ 0 þ 4 . This experimental result 2e 2 X hn ð1 – n – Þihð1 – n Þn – ijhjX j ij2
2 ¼
shows that ln Reff(H) _ H2 for low fields and the V ð" þ U Þ – "
dependence becomes weaker for higher fields, imply- ð111Þ
ing the behavior of spin-dependent electrical
polarizability. 2e 2 X hn n – ihð1 – n Þð1 – n – ÞijhjX j ij2
3 ¼
New et al. (1982) ascribed their experimental result V " – ð" þ U Þ
to the Zeeman splitting of excited spin-triplet states in
ð112Þ
spin-singlet paris of donors. Their model may be
suitable for the low-concentration regime of the 2e 2 X hn n – ihð1 – n Þn – ijhjX j ij2
donors. For the present case, however, we have to 4 ¼ ð113Þ
V ð" þ U Þ – ð" þ U Þ
consider the Anderson-localized states in the inter-
mediate-concentration regime. Kurobe et al. (1984) where h i represents the thermal average and h jX j i
adopted the theory in the preceding subsection to is the matrix element of the x-coordinate which is
calculate in the intermediate-concentration regime. proportional to the electric dipole moment. In general,
Although there are various excitations, the domi- there is also contribution to from excitations into
nant contribution to electrical polarizability comes states of antibonding characters, which reduces in
from the one-electron excitations among the the low-concentration regime, as discussed by New
Impurity Bands in Group-IV Semiconductors 109
et al. (1982). In the intermediate concentration, however, such as metallic or doped polyacetylene (Kaneko
the antibonding states have much higher energies. et al., 1994; Ishiguro et al., 1995; Choi et al., 1999),
Besides, the states are more extended and hence the metallic polypyrrole (Singh et al., 2008), entangled
matrix elements in Equations (110–113) are small. carbon-nanotube network (Kim et al., 1998), carbon
Hence, the above-mentioned excitations among the nanomaterials (Demishev et al., 2008), disordered
Anderson-localized states dominantly contribute to . nanographite network (Takahara et al. 2007), etc.
The calculated result is shown in Figure 28(b) on
simplifying assumptions that the matrix element
h jX j i does not depend on the kind of process 1.03.5 Conclusions
and that the localization length,
, and intrastate
interaction, U, are independent of the state energies. In this chapter, we have described how the impurity
For low magnetic fields, we have band in group-IV semiconductors manifests a rich
diversified physics. In Section 1.03.2, we have surveyed
ln½Reff ðH Þ _ ðB H =kB T Þ2 ð114Þ the transport phenomena in the impurity band, MI
Physically, spins of SO states tend to be aligned with transition, and the historical development of localiza-
increasing H. Since an electron cannot be excited to a tion theory. In Section 1.03.3, we have described the
state which is already occupied by another electron theoretical formulation to treat the interplay of disor-
of the same spin, the contribution from an SO state to der and electron–electron interactions in the strongly
another SO state in Equation (111) (2) decreases Anderson-localized regime just in the insulating side
with increasing magnetic field. Therefore, at high of the MI transition, developed by Kamimura and
fields (H > 5 T), where this type of excitation is co-workers. Their theory successfully explains the
almost suppressed, MC becomes less dependent on anomaly of specific heat and spin susceptibility at low
magnetic field. All these characteristics of MC are in temperatures. The theoretical formulation has been
good agreement with experimental results. confirmed by numerical simulation for the electronic
states in the Anderson-localized regime. Section 1.03.4
has been devoted to the description of the hopping
conduction in the Anderson-localized regime. We
1.03.4.5 Spin-Dependent Behavior of
have reviewed the Mott’s law for the VRH and
Magnetoresistance in Other Systems
Coulomb gap theory for the compensated samples.
As we have described in the previous subsections, the Then, we have described the theory by Kurobe and
theory of magnetoresistance and MC developed by Kamimura on the magnetoresistance and MC in the
Kurobe and Kamimura (Kurobe and Kamimura, 1982; VRH regime, considering spin-dependent effects due
Kamimura et al., 1982, 1983; Kurobe, 1986) has played to electron–electron interactions. Their theory has
an important role in clarifying the spin-dependent now been applied to magnetoresistance phenomena
effects on the transport properties due to the elec- observed in a wide variety of systems such as polymers,
tron–electron interactions in doped semiconductors carbon-nanotubes, and other exotic materials.
(New et al., 1982; Agrinskaya and Kozub, 1998; Recent development of the nanofabrication tech-
Agrinskaya et al., 2000; Kozub, 2002; Meir, 2001). nology on semiconductors is opening new research
Recently, their theory has been widely applied to fields in solid-state physics. As we have mentioned in
other systems. For example, two-dimensional hole Section 1.03.1.2, for example, it is an interesting sub-
system in the GaAs/AlGaAs heterostructure in mag- ject to control single impurity states in doped
netic fields (Yoon et al., 2000; Noh et al., 2004), silicon semiconductors for the quantum information proces-
MOSFETs in magnetic fields (Mertes et al., 2001), sing. It is very impressive that Kamimura and
amorphous indium oxide films (Frydman and co-workers predicated the spin-dependent effects
Ovadyahu, 1995; Lee et al., 2002; Orlyanchik and on various physical phenomena in the strongly
Ovadyahu, 2005), amorphous Bi films (Hernandez Anderson-localized regime two decades ago before
et al., 2003), Mn–Fe–Ge antiferromagnets (Du et al., extensive researches on solid-state quantum compu-
2007), and icosahedral Al-Pd-Re materials (Srinivas ters have begun. We believe that the present chapter
et al., 2001). In particular, it is interesting to see that will be useful for graduate students and young
the theory of Kurobe and Kamimura has been used researchers who will work on the physics and
for elucidating magnetoresistance phenomena in technology of semiconductors. (See Chapters 1.05,
one-dimensional and quasi-one-dimensional systems 1.06, 2.02 and 2.04).
110 Impurity Bands in Group-IV Semiconductors
(eds.) The Metal Non-Metal Transition in Disordered Kobayashi N, Ikehata S, Kobayashi S, and Sasaki W (1977)
Systems, p. 327. Edinburgh: SUSSP Publications. Specific heat study of heavily P doped Si. Solid State
Kamimura H (1979) Electronic states of a negative donor ion in Communications 24: 67.
silicon and germanium. Festschrift in honour of Sir Nevili Kobayashi N, Ikehata S, Kobayashi S, and Sasaki W (1979)
Mott. In: Gaskell PH and Mackenzie JD (eds). Journal of Magnetic field dependence of the specific heat of heavily
Non-Crystalline Solids 32: p. 187. phosphorus doped silicon. Solid State Communications
Kamimura H (1980) Diagonal and off-diagonal correlation in 32: 1147.
Anderson localized states. (Special Issue on Impurity Bands Kobayashi N and Muto Y (1979) Anomalously large negative
in Semiconductors.). Philosophical Magazine 42: 763. magnetoresistance of 1T-TaS2 at ultra low temperatures.
Kamimura H (1982) Theoretical model on the interplay of Solid State Communications 30: 337.
disorder and electron correlations. Progress of Theoretical Kozub VI (2002) A possible role of the D-band in hopping
Physics Supplement 72: 206. conductivity and metal-insulator transition in 2D structures.
Kamimura H (1985) Electron–electron interactions in the Physica Status Solidi (b) 230: 163.
Anderson-localised regime near the metal-insulator Kurobe A (1986) Theory of magnetoresistance in the Anderson-
transition. In: Efros AL and Pollak M (eds.) Electron–Electron localised regime of semiconductors. Journal of Physics C:
Interaction in Disordered Systems, ch. 7. Amsterdam: North- Solid State Physics 19: 2201.
Holland Publishing Company. Kohn W and Luttinger JM (1955) Theory of donor states in
Kamimura H (1986) Impurity Bands. In: Butcher PN, March NH, silicon. Physical Review 98: 21.
and Tosi MP (eds.) Crystalline Semiconducting Materials and Kurobe A and Kamimura H (1982) Correlation effects on variable
Devices, ch. 8, New York and London: Plenum Press. range hopping conduction and the magnetoresistance.
Kamimura H and Aoki H (1989) The Physics of Interacting Journal of the Physical Society of Japan 51: 1904.
Electrons in Disordered Systems. Oxford: Oxford University Kurobe A and Kamimura H (1983) Theory of magnetoresistance
Press. in amorphous semiconductors. Journal of Non-Crystalline
Kamimura H and Mott NF (1976) The variable range hopping Solids 59/60: 41.
induced by electron spin resonance in n-type silicon and Kurobe A, Takemori T, and Kamimura H (1984) Origin of
germanium. Journal of the Physical Society of Japan magnetocapacitance in the Anderson-localized regime.
40: 1351. Physical Review Letters 52: 1457.
Kamimura H, Kurobe A, and Takemori T (1983) Lee YJ, Kim YS, and Shin HK (2002) Magnetoresistance of
Magnetoresistance in Anderson-localized systems. Physica amorphous indium oxide films at the region of weak-strong
BþC 117–118: 652. localization crossover. Journal of Physics: Condensed
Kamimura H, Takemori T, and Kurobe A (1982) Interplay of Matter 14: 483.
Disorder and Coulomb Interactions in Anderson-Localized Lee PA and Ramakrishnan TV (1985) Disordered electronic
States. In: Nagaoka Y and Fukuyama H (eds.) Andemon systems. Reviews of Modern Physics 57: 287.
Localization, (Springer Series in Solid State Sciences 39.), Licciardello DC and Thouless DJ (1978) Conductivity and
p. 156. Berlin: Springer-Vedag. mobility edges in disordered systems. II. Further calculations
Kamimura H and Yamaguchi E (1978) Electron correlation for the square and diamond lattices. Journal of Physics C:
effects on Anderson localized states. Solid State Solid State Physics 11: 925.
Communications 28: 127. Licciardello DC and Thouless DJ (1975a) Constancy of
Kane BE (1998) A silicon-based nuclear spin quantum minimum metallic conductivity in two dimensions. Physical
computer. Nature 393: 133. Review Letters 35: 1475.
Kaneko H, Ishiguro T, Tsukamoto J, and Takahashi A (1994) Licciardello DC and Thouless DJ (1975b) Conductivity and
Low-temperature magnetoresistance of heavily mobility edges for two-dimensional disordered systems.
FeCl3-doped polyacetylene: Hopping conduction among Journal of Physics C: Solid State Physics 8: 4157.
localized states with Coulomb correlation. Solid State Löhneysen Hv (2003) Heavily doped semiconductors: Magnetic
Commications 90: 83. moments, electron–electron interactions and the Metal-
Kapian TA, Mahanti SD, and Hartmann WM (1971) Electron insulator transition. In: Hewson AC and Zlatić V (eds.) Concepts
correlation effects on the low-temperature thermodynamics in Electron Correlation, pp. 155–168. Amsterdam: Kluwer.
of amorphous semiconductors. Physical Review Letters Marko JR, Harrison JP, and Quirt JD (1974) Specific-heat
27: 1796. studies of heavily doped Si:P. Physical Review B 10: 2448.
Katsumoto S, Komori F, Sano N, and Kobayashi S (1987) Fine Mein Y (2001) Percolation-type description of the metal–
tuning of metal-Insulator transition in Al0.3Ga0.7As using insulator transition in two dimensions. Physica A 302: 391.
persistent photoconductivity. Journal of the Physical Society Mertes KM, Zheng H, Vitkalov SA, Sarachik MP, and
of Japan 56: 2259. Klapwijk TM (2001) Temperature dependence of the
Kaveh M (1985) Effect of electron-electron interactions on resistivity of a dilute two-dimensional electron system in high
critical exponents near the metal-insulator transition. parallel magnetic field. Physical Review B 63: 041101.
Philosophical Magazine 52: L1. Miller A and Abrahams E (1960) Impurity conduction at low
Kawabata A (1985) Progress of Theoretical Physics Supplement concentrations. Physical Review 120: 745.
84: 16. Milligan RF, Rosenbaum TF, Bhatt RN, and Thomas GA (1985) A
Kim GT, Choi ES, Kim DC, et al. (1998) Magnetoresistance of an review of the metal-insulator transition in doped
entangled single-wall carbon-nanotube network. Physical semiconductors. In: Efros AL and Pollak M (eds.) Electron–
Review B 58: 16064. Electron Interaction in Disordered Systems, Ch. 3,
KirkPatrick S (1973) Hopping conduction: Experiment versus Amsterdam: North-Holland.
theory. In: Stuke J and Brenig W, (eds.), Proceedings of the Mott NF (1967) Electrons in disordered structures. Advances in
5th International Conference on Amorphous and Liquid Physics 16: 49.
Semiconductors, Garmisch, p. 183. London: Taylor and Mott NF (1968) Introductory talk; conduction in non-crystalline
Francis. materials. Journal of Non-Crystalline Solids 1: 1.
Kobayashi S, Fukagawa Y, Ikhata S, and Sasaki W (1978) NMR Mott NF (1970) Conduction in non-crystalline systems IV.
study on electronic states in phosphorus doped silicon. Anderson localization in a disordered lattice. Philosophical
Journal of the Physical Society of Japan 45: 1276. Magazine 22: 7.
112 Impurity Bands in Group-IV Semiconductors
Mott NF (1972) Conduction in non-crystalline systems IX. The Srinivas V, Rodmar M, Poon SJ, and Rapp Ö. (2001) Evidence
minimum metallic conductivity. Philosophical Magazine for an insulating ground state in high-resistivity icosahedral
26: 1015. AlPdRe from the magnetoresistance. Physical Review B
Mott NF (1976) The metal-insulator transition in an impurity 63: 172202.
band. Journal of Physics Colloques 37: C4–301. Takahara K, Takai K, Enoki T, and Sugihara K (2007) Effect of
Mott NF and Davis EA (1979) Electronic Processes in Non- oxygen adsorption on the magnetoresistance of a
Crystalline Materials 2nd edn.; Oxford: Clarendon Press. disordered nanographite network. Physical Review B
Natori A and Kamimura H (1977) The electronic structure of a 76: 035442.
D- centre in many-valley semiconductors. Journal of the Takano Y (2008) Superconductivity in Diamond.
Physical Society of Japan 43: 1270. In: Sussmann RS (ed.) CVD Diamond for Electronic Devices
New D, Lee NK, Tan HS, and Castner TG (1982) New and Sensors. Wiley, Ch. 19.
magnetoelectric effect: Evidence for spin-dependent Takemori T and Kamimura H (1982) Effect of interstate
electrical polarizabilities from magnetocapacitance interactions on specific heat and magnetic susceptibility in
measurements on n-type silicon. Physical Review Letters Anderson-localized system. Solid State Communications
48: 1208. 41: 885.
Nielsen MA and Chuang IL (2000) Quantum Computation and Takemori T and Kamimura H (1983a) Computer simulation of
Quantum Information. Cambridge: Cambridge University interacting donor electrons in the Anderson-localised regime
Press. of semiconductors. Journal of Physics C: Solid State Physics
Mott NF (1990) Metal-Insulator Transitions, 2nd edn.; London: 16: 5167.
Taylor and Francis. Takemori T and Kamimura H (1983b) Effect of interstate
Nishio Y, Kobayashi N, and Muto Y (1978) Specific heat of interactions on specific heat and magnetic susceptibility
1T-Ta0.93Ti0.07S2 in the Anderson localized states. Solid in Anderson localized systems. Advances in Physics
State Communications 50: 781. 32: 715.
Noh H, Yoon J, Tsui DC, and Shayegan M (2004) Insulating Tanuma S (1983) Electrical properties of layered materials in
behavior of dilute two-dimensional holes in GaAs under an high magnetic field. In: Landwehr G (ed.) Lecture Notes in
in-plane magnetic field. Physical Review B 70: 241306. International Conference on the Application of High
Onuki Y, Inada R, and Tanuma S (1980) Effect of impurities of Magnetic Fields in Semiconductor Physics, p. 147. Berlin:
selenium and iron on the Anderson localization of 1T-TaS2. Springer.
Physica B 99: 177. Thouless DJ (1974) Electrons in disordered systems and the
Orlyanchik V and Ovadyahu Z (2005) Conductance noise in theory of localization. Physics Report 13C: 93.
interacting Anderson insulators driven far from equilibrium. Ue H and Maekawa S (1971) Electron-spin-resonance studies
Physical Review B 72: 024211. of heavily phosphorus-doped silicon. Physical Review B
Paalanen MA, Rosenbaum TF, Thomas GA, and Bhatt RN 3: 4232.
(1982) Stress tuning of the metal-insulator transition at Williams PM, Parry GS, and Scruby CB (1974) Diffraction
millikelvin temperatures. Physical Review Letters 48: 1284. evidence for the Kohn anomaly in 1T TaS2. Philosophical
Pollak M (1972) A percolation treatment of dc hopping Magazine 29: 695.
conduction. Journal of Non-Crystalline Solids 11: 1. Wilson JA, DiSalvo FJ, and Mahajan S (1974) Charge-density
Quirt JD and Marko JR (1972) Absolute spin susceptibilities and waves in metallic, layered, transition-metal dichalcogenides.
other ESR parameters of heavily doped n-type silicon. Physical Review Letters 32: 882.
I. Metallic samples. Physical Review B 5: 1716. Wilson JA, DiSalvo FJ, and Mahajan S (1975) Charge-density
Quirt JD and Marko JR (1972) Absolute spin susceptibilities and waves and superlattices in the metallic layered transition
other ESR parameters of heavily doped n-type silicon. II. A metal dichalcogenides. Advances in Physics 24: 117.
unified treatment. Physical Review B 7: 3842. Yamaguchi E, Aoki H, and Kamimura H (1979) Intra- and
Rosenbaum TF, Andres K, Thomas GA, and Bhatt RN (1980) inter-state interactions in Anderson-localised states. Journal
Sharp metal-insulator transition in a random solid. Physical of Physics C: Solid State Physics 12: 4801.
Review Letters 45: 1723. Yamaguchi E, Aoki H, and Kamimura H (1980) Intra- and
Sasaki W and Kinoshita J (1968) Piezoresistance and magnetic interstate interactions in Anderson-localised states. Journal
susceptibility in heavily doped n-type silicon. Journal of the of Non-Crystalline Solids 35/36: 47.
Physical Society of Japan 25: 1622. Yoshino S and Okazaki M (1976) Numerical study of localization
Scruby CB, Williams PM, and Parry GS (1975) The role of in Anderson model for disordered systems. Solid State
charge density waves in structural transformations of 1T Communications 20: 81.
TaS2. Philosophical Magazine 31: 255. Yoshino S and Okazaki M (1977) Numerical study of electron
Shklovskii BI and Efros AL (1984) Electronic Properties of localization in Anderson model for disordered systems:
Doped Semiconductors (Springer Series in Solid-State Spatial extension of wavefunction. Journal of the Physical
Science 45). Society of Japan 43: 415.
Singh KJ, Clark WG, Ramesh KP, and Menon R (2008) 1H-NMR Yoon J, Li CC, Shahar D, Tsui DC, and Shayegan M (2000)
and charge transport in metallic polypyrrole at ultra-low Parallel magnetic field induced transition in transport in the
temperatures and high magnetic fields. Journal of Physics: dilute two-dimensional hole system in GaAs. Physical
Condensed Matter 20: 465208. Review Letters 84: 4421.
Skal AS and Shklovskii BI (1971) Mott equation for low Yokoya T, Nakamura T, Matsushita T, et al. (2005) Origin of the
temperature edge conductivity. Fizika Tverdogo Tela metallic properties of heavily boron-doped superconducting
16: 1820 (Soviet Physics–Solid State 16: 1190). diamond. Nature 438: 647.
1.04 Atomic Structures and Electronic Properties of
Semiconductor Interfaces
T Nakayama, Chiba University, Chiba, Japan
Y Kangawa, Kyushu University, Fukuoka, Japan
K Shiraishi, University of Tsukuba, Ibaraki, Japan
ª 2011 Elsevier B.V. All rights reserved.
113
114 Atomic Structures and Electronic Properties of Semiconductor Interfaces
1.04.2 Interface Formation and The chemical potential for the ideal gas is
Stability given by (Fowler, 1936; Tolman, 1938; Fowler and
Guggenheim, 1939; Mayer and Mayer, 1940;
1.04.2.1 Semiconductor Epitaxy
Rodebush, 1931; Fowler and Sterne, 1932; Kassel,
1.04.2.1.1 Epitaxial growth diagram 1936):
Fabrication of semiconductor devices such as high-
¼ – kB T lnðgkB T =p trans rot vibr Þ ð1Þ
electron-mobility transistors (HEMTs) and light-
emitting diodes (LEDs) are carried out using vapor 3=2
trans ¼ 2mkB T =h 2 ð2Þ
phase epitaxy (VPE) techniques, for example,
n on=2
molecular beam epitaxy (MBE) and metal organic rot ¼ ð1=Þ 83 ðIA IB :::Þ1=n kB T =h 2 ð3Þ
vapor phase epitaxy (MOVPE). The adsorption–
desorption behavior of the constituent atoms on vibr ¼ Pi3N – 3 – n f1 – expð – hvi =kB T Þg – 1 ð4Þ
semiconductor surfaces is of fundamental impor-
tance in these growth techniques. In particular, the where trans, rot, and vibr are the partition functions
for the translational motion, the rotational motion,
behavior of constituent atoms must be known to
and the vibrational motion, respectively. Here, kB is
understand surface and thin film growth-related
Boltzmann’s constant, T the temperature, g the
phenomena such as the change in surface recon-
degree of degeneracy of the electron energy level, p
structions and growth processes. In the present
the BEP of the particle, m the mass of one particle, h
section, we discuss the relationships between
the Planck’s constant, the symmetric factor, II the
adsorption–desorption behavior and growth condi-
moment of inertia, n the degree of freedom of the
tions such as beam equivalent pressure (BEP),
rotation, N the number of atoms in the particle, i the
p, and temperature, T.
degree of freedom for the vibration, and the fre-
We can consider adsorption–desorption behavior
quency. II is written as
from a viewpoint of vapor–solid phase equilibrium.
If the free energy of an atom/molecule in the II ¼ mI r 2 ð5Þ
vapor phase is smaller than that on the surface, an
where mI is the reduced mass and r the radius of
impinging atom/molecule is easily desorbed from
gyration.
the surface and stabilized in vapor phase. On the
Practical applications of this theory are as follows.
other hand, an atom/molecule is absorbed if the
During the typical GaAs MBE process, monatomic
free energy of the atom/molecule in vapor phase is Ga and As2 molecule are supplied on the substrate
larger than that on the surface. Here, the free energy surface under an ultrahigh vacuum (UHV) condition.
of the atom/molecule, which is called the chemical In the UHV condition, the ideal-gas assumption is
potential (), can be computed by using quantum applicable because impingement of each atom/mole-
statistical mechanics. The energy gain of an adsorbed cule is negligible. In the case of monatomic Ga, the
atom/molecule on the surface (i.e., the adsorption value of N and I are 1 and 0, respectively. Therefore,
energy (Ead)) is estimated, for example, by ab initio all that is required to compute the chemical potential
calculations. By comparing with Ead, we can discuss of monatomic molecule is considering the partition
the adsorption–desorption behavior as shown in function of translational motion. The value of para-
Figure 1. meter g for monatomic Ga in Equation (1) is 2. The
values of g for typical elements are summarized in
Adsorption E Desorption Table 1 (Glasstone et al., 1964). In the case of As2
molecule, we should consider the partition functions
for rotational and vibrational motions. Here, the value
μ of the symmetry number in (3) is 2 for As2. The
values of and r for As2 in Equations (4) and (5) can
Ead Ead be estimated by ab initio molecular orbital calculations.
μ The values of and r for As2 estimated by the
Gaussian 98 program (Frisch et al., 1998) are
Figure 1 Schematics of the magnitude correlation
446 cm1 and 1.0621 Å (Kangawa et al., 2002), respec-
between chemical potential and adsorption energy Ead and tively. By using these values, we can compute the
its relationship with the adsorption–desorption behavior. chemical potentials of monatomic Ga and As2
116 Atomic Structures and Electronic Properties of Semiconductor Interfaces
[110]
VI O, S, Se, Te, Po 3 1 2
VII F, Cl, Br, I 2 [110]
0 He, Ne, Ar, Kr, Xe, Rn 1
[001]
molecule as a function of growth parameters p and T. [110]
On the other hand, the adsorption energy Ead of Ga or Figure 2 Schematic of GaAs(001)-(4 2)2 surface.
As2 on reconstructed surfaces can be obtained by ab Adsorption sites are indicated by the letters A–E.
initio calculations. For the results in this section, the Reproduced with permission from Kangawa Y, Ito T,
first-principles pseudopotential method based on the Taguchi A, Shiraishi K, and Ohachi T (2001) A new
theoretical approach to adsorption–desorption behavior of
local-density functional formalism (Hamman et al.,
Ga on GaAs surfaces. Surface Science 493: 178.
1979) is used, adopting the Kleinman–Bylanders
separable pseudopotentials, careful treatment of the
cut-off value of local potential (Ohno, 1993), the con- calculated as the difference between the total energy
ventional repeated-slab geometry, and termination of when the Ga adatom is located at the most stable site
bottom-layer Si by fictitious H atoms. Here, we show (E-site) and when the Ga atom is in the vacuum
the results of the calculated adsorption energy of Ga region. In the calculation, we gradually pulled the
on GaAs(001)-(4 2)2 surface and desorption Ga apart from the E site to the vacuum region and
energy (Ead) of As2 from c(4 4) surface. confirmed the convergence of the total energy
First, we discuss about adsorption–desorption difference when the distance between the position
behavior of Ga on GaAs(001)-(4 2)2 surface. of Ga atom and the E site is larger than 4 Å.
Figure 2 illustrates the GaAs(001)-(4 2)2 surface. Consequently, the calculated adsorption energy Ead
The calculated results show that the Ga adatom is for Ga at the E site on the Ga-rich surface is esti-
most stable at the site ‘E’ on the surface where two mated to be 3.3 eV. This implies that the Ga
Ga–Ga bonds (E-1 and E-2 bonds) and a Ga–As bond adsorbed structure is formed when the chemical
(E-3 bond) are formed around the adatom. The most potential for Ga atom is larger than the calculated
favorable site for adatom on the semiconductor sur- adsorption energy (3.3 eV). On the other hand, it is
face is frequently discussed using an electron thought that the GaAs(001)-(4 2)2 surface with-
counting model (Farrel et al., 1987; Pashley, 1989). out Ga adatom is stable when the chemical potential
This is a model for discussing the structural stability for Ga is less than the adsorption energy (3.3 eV).
of reconstructed surface structure of compound Figure 3 shows the chemical potential for Ga atom as
semiconductors. The lowest-energy structure is a function of temperature. Considering the typical
obtained by filling all dangling bonds on the anion BEP value (pGa ¼ 1.0 105 torr), the calculated line
atom and keeping empty those on the cation atom. crosses the line of Ga ¼ 3.3 eV at 700 C; there-
This is because energy levels of dangling bonds on fore, the critical temperature for Ga adsorption is
anion atom are generally located around the valence 700 C at this BEP value. By similar calculations
bands and those on cation atom are around the con- to obtain crossing points at other BEPs, a p–T dia-
duction bands. The stability of reconstructed gram for the adsorption–desorption transition was
structure and the adsorption site of constituent obtained as Figure 4. This result agrees well with
atoms on GaAs surface is well discussed by Ito experiments, that is, GaAs(001)-(4 2) surface
(1995). The adsorption energies of Ead can be becomes unstable and Ga droplets are formed at
Atomic Structures and Electronic Properties of Semiconductor Interfaces 117
–2.0
700 Desorption
1.0 ×10–4 torr
Temperature (°C)
c(4 × 4) unstable
650
–2.5
1.0 ×10–5 torr
Chemical potential (eV)
600
600 C during the conventional MBE growth under 1.04.2.1.2 Competition of growth direction
Ga-rich condition (Yamada et al., 1989), whereas Ga Control of the interface/surface shape is a crucial factor
desorption proceeds and reconstructed surface for fabricating semiconductor devices with better-
appears at 700 C after turning off the Ga flux defined epi-layers. In this section, we discuss, for exam-
(Kojima et al., 1985; Gibson et al., 1990). This suggests ple, how to control the surface shape of cubic GaN
that the theoretical approach is feasible for predicting (c-GaN) during growth. If we can control the surface
the dependence of the temperature and BEP on shape, we can also control the interface shape between
adsorption–desorption behavior. the layer and the post-growth layer. In case of c-GaN
Next, we discuss the stability conditions of growth, control of surface shape is indispensable to
GaAs(001)-c(4 4) surface. The conditions where fabricate a better-defined epi-layer as well as to control
the -c(4 4) appears are predicted by comparing the the grown phase.The following information has been
adsorption energy of As-dimer on the surface obtained about c-GaN growth: {111} facet formation
118 Atomic Structures and Electronic Properties of Semiconductor Interfaces
during growth causes hexagonal GaN (h-GaN) mixing, (a) Top view
that is, in the regions grown toward the <111> direc- 2nd layer N
1st layer Ga
tions, the stacking sequence of . . .ABCABC. . . easily
collapses and changes to . . .ABABAB. . . (Kuwano et al.,
A B C
1994). This implies that two-dimensional (2D) growth
D E
without facet formation is important to grow single-
phase c-GaN. Here, we apply the theoretical approach Side view
described in Section 1.04.2.1.1 on the adsorption process A B C
D E
of monatomic Ga and N on GaN(001) and (111) facet
planes in order to investigate the 2D growth condition
range. The shape of a growing surface can be investi- 1st layer Ga
gated by comparing the adsorption energy of the 2nd layer N
constituent atoms on each surface; top surface and (b) Ga N
facet plane.
0
–2.1 eV
–6.4 eV
(Neugebauer et al., 1998; Al-Brithen et al., 2005). –3
Therefore, we assumed the (4 1) reconstructed –4
surface as the initial substrate and investigated the –5
relationship between adsorption–desorption behavior –6
of monatomic Ga and N on the surface and growth
A B C D E A B C D E
conditions such as BEP and temperature. Here, we
Site (–) Site (–)
consider monatomic N, because the source gas is
supplied on the substrate surface in case of radiofre- Figure 6 (a) Schematic drawing of GaN(001)-(41)
quency (rf) MBE. Figure 6 shows a schematic drawing surface. (b) Adsorption energies of Ga and N on sites A–E
shown in (a). Reproduced with permission from Kangawa Y,
of the GaN(001)-(4 1) surface and the adsorption
Matsuo Y, Akiyama T, Ito T, Shiraishi K, and Kakimoto K
energy on the sites denoted by the capital letters (2007) Theoretical approach to initial growth kinetics of GaN
(A–E), respectively. In Figure 6, one can see that the on GaN(001). Journal of Crystal Growth 300: 62.(Kangawa
most favorable sites for adsorption are ‘E’ site for Ga et al. 2007)
and ‘A’ site for N. In the case of Ga adsorption, the
adsorption energy is 2.1 eV. This value is larger than In the same way, we can discuss the adsorption of
the formation energy of a Ga droplet (2.8 eV; a second atom on the N-adsorbed GaN(001)-(4 1)
Honing and Karmer, 1969). Therefore, the formation surface. Figures 8(a) and 8(b) show a schematic
of a Ga droplet is easier than that of a Ga-adsorbed drawing of the N-adsorbed GaN(001)-(4 1) surface
structure in this case. In the case of N adsorption, the and the adsorption energy on each site, respectively.
adsorption energy is 6.4 eV. Using these values of In Figure 8(a), it can be seen that the most favorable
adsorption energy, we can obtain phase diagrams sites are ‘D’ for Ga adsorption and ‘B’ for N adsorp-
related to the surface structures (Figure 7). In tion. However, if a N atom is adsorbed near the N
Figure 7 shows that no Ga-adsorbed structure atoms on the surface, that is, the case of adsorption on
would be formed, though Ga droplets seem to ‘A’ and ‘D’ sites, N atoms would desorb as a N2
be formed on the surface at high Ga BEP and in the molecule instead of N-dimer formation because a
low-temperature region. Here, the boundary line N-dimer is unstable on the surface. Phase diagrams
satisfies the conditions of Ga ¼ EGadroplet where were made using the calculated adsorption energies,
Ga is the chemical potential of monatomic Ga and that is, 3.7 eV for Ga and 6.0 eV for N, as shown in
EGa-droplet the formation energy of a Ga droplet Figure 9. The results imply that the stable region of
(2.8 eV). On the other hand, the N-adsorbed the Ga-adsorbed structure appears in the phase
structure is stable at typical growth conditions as diagram after the first N adsorption. That is, a
shown in the phase diagram. These results suggest Ga-adsorbing site appears after adsorption of the
that a N-adsorbed structure appears instead of a first N adatom. On the other hand, it is found that
Ga-adsorbed structure in the case of adsorption of N adsorption might occur under the condition shown
the first atom. in the diagram; however, N adatoms seem to desorb
Atomic Structures and Electronic Properties of Semiconductor Interfaces 119
(a) (b)
Temperature (°C)
800 Ga desorption
Ead-Ga > μGa
700 N adsorption
Ead-N > μN
600 Ga droplet
10–6 10–5 10–4 10–3 10–2 10–6 10–5 10–4 10–3 10–2
Ga BEP (torr) N BEP (torr)
Figure 7 p–T dependence of adsorption–desorption behavior of (a) Ga and (b) N on the initial surface. Reproduced with
permission from Kangawa Y, Matsuo Y, Akiyama T, Ito T, Shiraishi K, and Kakimoto K (2007) Theoretical approach to initial
growth kinetics of GaN on GaN(001). Journal of Crystal Growth 300: 62.
Temperature (°C)
800
A B C Ga adsorption
D E Ead-Ga < μGa
700 N adsorption
Side view Ead-N> μN
D A 600
Ga droplet
B C
E 10–6 10–5
10–4 10–3 10–2 10–6 10–5 10–4 10–3 10–2
Ga BEP (torr) N BEP (torr)
–3
Subsequently, Ga adsorption on the N adatom
–4
occurs. Here, the N adatom is stabilized by being
–5 covered with a Ga adatom. This is different than
–6 the GaAs case, that is, group-III atoms are incorpo-
N2 N2
rated in the crystal by being covered with group-V
A B C D E A B C D E
Site (–)
atoms in the case of GaAs. Though more detailed
Site (–)
examinations are required, Ga adsorption might be
Figure 8 (a) Schematic drawing of N-adsorbed GaN(0 0 the rate-limiting process during c-GaN growth and
1)-(41) surface. (b) Adsorption energies of Ga and N on
sites A–E shown in (a). Reproduced with permission from
the adsorption energy is 3.7 eV in this case.
Kangawa Y, Matsuo Y, Akiyama T, Ito T, Shiraishi K, and We have already noted that the vapor–solid equi-
Kakimoto K (2007) Theoretical approach to initial growth librium condition changes during the growth process.
kinetics of GaN on GaN(001). Journal of Crystal Growth That is, change in coverage of adatom/molecule
300: 62.
leads to a change in adsorption site for the subsequent
adatom/molecule and adsorption energy. Therefore,
as a N2 molecule if a N adatom migrates to the sites the adsorption–desorption behavior changes regu-
on another N adatom. These results suggest that larly during the growth process. In a similar
c-GaN growth on the GaN(001)-(4 1) surface pro- manner, we have to consider the influence of the
ceeds via the following process during the initial spatial dependence of the adsorption–desorption
growth stage. First, a N-adsorbed structure is formed behavior if there are more than one plane on the
instead of a Ga-adsorbed structure. surface. In the following paragraph, we discuss the
120 Atomic Structures and Electronic Properties of Semiconductor Interfaces
(a) Top view (b) Top view Table 2 Desorption energies of Ga at T1 and B
1st layer Ga sites shown in Figure 10 on Ga-monolayer and
1st layer Ga
2nd layer Ga 2nd layer Ga bilayer surfaces
Monolayer Bilayer
Site T1 B T1 B
B
B T1 Desorption energy (eV) 4.1 3.8 3.7 3.4
T1
Reproduced with permission from Kangawa Y,
Akiyama T, Ito T, Shiraishi K, and Kakimoto K (2009)
Theoretical approach to structural stability of GaN:
How to grow cubic GaN. Journal of Crystal Growth
Side view Side view 311: 3106.
B T1
B T1 Ga monolayer is stable compared with Ga bilayer
surface because desorption energy of Ga on the
monolayer surface (3.8 eV) is larger than that on
the bilayer surface (3.4 eV).
Here, the adsorption energy is equal to the nega-
Figure 10 Schematic of (a) Ga monolayer and (b) Ga
tive of the desorption energy if there is no activation
bilayer structure on (111) surface. Reproduced with energy during the desorption process. That is, the
permission from Kangawa Y, Akiyama T, Ito T, Shiraishi K, adsorption energy of Ga at B site on the monolayer
and Kakimoto K (2009) Theoretical approach to structural and bilayer surface is 3.8 eV and 3.4 eV, respec-
stability of GaN: How to grow cubic GaN. Journal of Crystal tively. Using these values and the adsorption energy
Growth 311: 3106.
of Ga on (001)-(4 1) surface (3.7 eV), we can
classify the growth conditions into three regions:
adsorption–desorption behavior on the faceted region (I)–(III), as shown in Figure 11. In regions
surface and the change in growth form. (I) and (II), the chemical potential, Ga , is less than
To compare the adsorption energies of constitu- 3.4 eV, which is the adsorption energy of Ga on
ent atoms on GaN(001)-(4 1) surface with those (111) Ga bilayer surface. Therefore, Ga monolayer
on {111} facet plane, we studied the adsorption– surface appears on the faceted surface and the
desorption behavior of monatomic Ga and N on adsorption energy on the surface is 3.8 eV. In case
{111} facet plane. Figures 10(a) and 10(b) show of region (I), Ga is less than 3.7 eV, which is the
schematics of Ga monolayer and bilayer surfaces, adsorption energy of Ga on GaN(001)-(4 1). In this
respectively, which were observed on GaN(0001) region, impinging Ga adsorbs only on the (111) facet
(Adeimann et al., 2002; Northrup et al., 2000). These plane. In the case of region (II), impinging Ga adsorbs
structure also appear on GaN(111) because the on the facet plane, and then adsorbs on the (001)
atomic arrangement of the topmost and second surface because Ead-Ga(111) < Ead-Ga(001) < Ga , where
monolayers of (111) is the same as that of (0001). Ead-Ga(hkl) is the Ga adsorption energy on (hkl) plane.
Therefore, we discuss the adsorption–desorption In the case of region (III), Ga is larger than 3.4 eV,
behavior on these surfaces henceforth. In Figures which is the Ga adsorption energy on Ga bilayer
10(a) and 10(b), one can see that there are two surface, and, therefore, the bilayer surface appears
kinds of atomic sites, that is, T1 site which is on the on the facet plane under the growth conditions. In
Ga in underling layer and B site which is on the the region, impinging Ga atoms adsorb on the (001),
bridging site between Ga atoms in the underling and then adsorb on the facetted surface because Ead-
layer. Table 2 shows the desorption energies of Ga(001) < Ead-Ga(111) < Ga . These results imply that
Ga occupying the T1 and B sites on Ga mono- 2D growth occurs when the growth condition is
layer and bilayer surfaces. In Table 2, it is found suitable in regions (I) and (II), whereas facet forma-
that Ga atoms at the B sites are unstable com- tion occurs when the growth condition is in region
pared with that at T1 sites. This implies that (III). That is, we can control the surface shape and
decomposition of each layer starts by desorption suppress h-GaN mixing by choosing proper growth
of Ga at B sites. Moreover, the results suggest that conditions.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 121
Monolayer
600
Ga droplet
(III) –3.4 < μ Ga < –2.8 (eV)
10–6 10–5 10–4 10–3 10–2
–3.7 –3.8
Ga partial pressure (torr)
Bilayer
Figure 11 (a) Phase diagram of growth form and (b) schematic of growth mode in regions (I)–(III). Reproduced with
permission from Kangawa Y, Akiyama T, Ito T, Shiraishi K, and Kakimoto K (2009) Theoretical approach to structural stability
of GaN: How to grow cubic GaN. Journal of Crystal Growth 311: 3106.
1.04.2.2 Growth Modes in Heteroepitaxy formation. In the case of the VW mode, three-dimen-
sional growth proceeds without forming a two-
1.04.2.2.1 Growth modes
dimensional wetting layer. It is noted that the FM
Heteroepitaxial growth is one of the most important
mode originally means the growth mode in which a
techniques for the fabrication of semiconductor
two-dimensional growth layer appears after the for-
heterostructures which is the basic building block
mation of misfit dislocations at the interface between
of semiconductor optical and electronic devices.
a substrate and a growth layer, as described in the
Therefore, an understanding of heteroepitaxial
original paper of van der Merwe (1991). Nowadays,
growth mechanism is crucial for both scientific and
all two-dimensional growths are often categorized
technological viewpoints (Venables et al., 1984).
into the FM mode, contrary to the original definition.
Among them, it is well known that several growth
It is well known that growth modes of a semicon-
modes appear in a heteroepitaxial growth with lat- ductor are quite complex and the physical
tice-mismatched systems. In Figure 12, three typical mechanisms of various growth modes have intensively
examples of Frank–van der Merwe (FM) mode, studied so far (van der Merwe, 1963a,b, 1991;
Stranski–Krastanov (SK) mode, and Volmer–Weber Matthews and Blakeslee, 1975, 1976; Tersoff, 1988,
(VW) mode are schematically illustrated. As shown, a 1998; Nakajima, 1999). However, it is difficult to sys-
two-dimensional flat epitaxial layer is observed tematically explain the various and complex behaviors
during growth in the FM mode. In the SK mode, of the growth modes observed experimentally. This is
however, three-dimensional islands (called SK because semiconductor crystal growth involves
islands) appear after two-dimensional wetting-layer various and complex phenomena: for example, the
formation of misfit dislocations at growth interfaces
and of three-dimensional growth islands, as well as the
(a) (b) (c) inter-diffusion of atoms at the growth interfaces.
In the case of lattice-mismatched epitaxial
growths, the most crucial question is about how the
strain relaxation occurs. There have been a lot of
reports that discuss the relationship between growth
Figure 12 Schematic illustration of three typical growth processes and each individual strain relaxation
modes: (a) Frank–van der Merwe (FM) mode; (b) Stranski– mechanism. For example, van der Merwe (1963a,b,
Krastanov (SK) mode; and the (c) Volmer–Weber (VM) mode. 1991) and Matthews and Blakeslee (1975, 1976)
122 Atomic Structures and Electronic Properties of Semiconductor Interfaces
investigated the relationship between the generation heteroepitaxial growth can be qualitatively described
of misfit dislocations and growth processes from the on the basis of energetics.
viewpoint of the energy competition between the First, we define the ‘optimal growth mode’, which
strain in growth layers and the generation of misfit is key to the discussion in lattice-mismatched hetero-
dislocations. However, their approaches focused on epitaxy. The optimal growth mode has the minimum
the misfit dislocations and three-dimensional island free energy at a given film layer thickness h. Here, we
formations were not taken into account. On the other determine the optimal growth mode by considering
hand, Tersoff (1988, 1998) intensively studied the the energetics, instead of the system free energies, for
relationship between the three-dimensional SK simplicity. To determine the growth mode, we
island formation and growth mode, although he did should consider the possible growth modes. If two
not discuss the generation of misfit dislocations. mechanisms of misfit dislocation and SK island for-
Nakajima (1999) systematically studied the competi- mations are considered, possible growth modes can
tion between strain energies, surface energies, and be categorized as follows:
interfaced energies, and discussed the growth
modes. However, he did not include the effect of 1. mode with neither SK islands nor misfit disloca-
misfit dislocations. tions (two-dimensional coherent growth),
In this subsection, we show the simple but sys- 2. mode with only SK island formation (SK coherent
tematic theoretical studies that discuss the growth),
competition between several strain relaxation 3. mode with only misfit dislocation formation (two-
mechanisms, such as the generation of misfit dislo- dimensional growth with misfit dislocation), and
cations and three-dimensional island formations 4. mode with both misfit dislocations and SK islands
(Okajima et al., 2000). This theory is not only easy (SK growth with misfit dislocation).
to understand but also can naturally include various These growth modes are schematically illustrated in
mechanisms such as misfit dislocation and Figures 13(a)–13(d). Accordingly, the optimal growth
three-dimensional island formations. Moreover, we mode is one with the lowest energy among the above
comment on the importance of atomistic level infor- four growth modes at a given film thickness h.
mation for systematic understanding of various Therefore, by considering the change in the optimal
macroscopic growth processes. growth mode with the increase of h, we can system-
atically investigate the epitaxial growth process.
1.04.2.2.2 Theory based on macroscopic Next, we describe the system energies of four
energetics types of growth modes. Here, we define the physical
In this subsection, we discuss the growth mode of parameters that will commonly be used in this sub-
lattice-mismatched epitaxy when two kinds of strain section as follows:
relaxation mechanisms of misfit dislocation and
three-dimensional island formations are taken into
account (Okajima et al., 2000). First, we discuss a
(a) (b)
theory that includes misfit dislocation and SK island
formations. This theory includes several macroscopic
parameters, for example, the elastic constants. In the
following, we will discuss a procedure that deter-
mines these macroscopic parameters based on the
atomistic simulations (e.g., first-principles calcula-
(c) (d)
tions) showing an example that initial
heteroepitaxial growth process of InAs/GaAs(110)
can quantitatively be reproduced by the discussion
using first-principles calculations.
It should be noted that dynamics as well as the
energetics play a crucial role in crystal growth. In the Figure 13 Schematic illustrations of four growth modes
that appear when misfit dislocation and SK island formation
present discussion, however, we only take into
are competed: (a) two-dimensional coherent growth, (b) SK
account energetics in order to simplify the discus- coherent growth, (c) two-dimensional growth with misfit
sions. In the following discussion, we will show that dislocation formation, and (d) SK growth with misfit
various and complex behaviors of lattice-mismatched dislocation formation.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 123
–1
: surface energy of a growth layer, i: Increase in the opt h 2D – MD
l2D – MD ðh Þ ¼ l0 1 – c ð10Þ
surface area by SK island formation, h
SK Ed Ed2
2D 1 Eopt MD ðh Þ ¼ þ – ð15Þ
Ecoh ðh Þ ¼ þ M"02 h ð6Þ l0 2M ð1 –
Þ"02 l02 h
2
In the case of SK coherent growth, an increase in By comparing the system energies of four possible
surface area and strain relaxation by the formation of growth modes given in Equations (6), (7), (14), and
SK islands should be added and given as follows: (15) at a given film thickness of h, we can determine
the optimal growth mode as a function of h. Further
1 more, we can discuss the various growth processes by
SK
Ecoh ðh Þ ¼ ð1 þ Þ þ M ð1 –
Þ"02 h ð7Þ
2 considering the h dependence of the optimal growth
In the case of two-dimensional growth with misfit mode.
dislocations, the system energy is a function of an Here, we briefly mention the treatment of VW
average distance in misfit dislocations l as well as a growth in which a wetting layer does not appear. To
film thickness h as follows: describe VW growth, we should set the parameter
(surface energy of a growth layer) as negative. By this
treatment, we can describe various and complex
2D 1 l0 2 Ed
EMD ðh; l Þ ¼ þ M"02 1 – hþ ð8Þ growth processes, such as the three-dimensional
2 l l
island formation, without wetting layer formation,
Similarly, the system energy of SK growth with subsequent formation of misfit dislocations, and
misfit dislocations can also be described as h and l: coalescence of three-dimensional islands.
SK 1 2 l0 Ed
EMD ðh; l Þ ¼ ð1 þ Þ þ M ð1 –
Þ"0 1 – hþ ð9Þ 1.04.2.2.3 Mode transition during epitaxy
2 l l
Tracing the trajectory of the optimal growth mode,
In real epitaxial growth, an average distance of misfit while varying the layer thickness h, should reveal the
dislocations l should be the value lopt(h) at which the change in growth mode during heteroepitaxy. By
system energy becomes minimum. In addition, lopt(h) varying Ed, two types of typical growth behavior are
can be described as follows in two-dimensional demonstrated when other parameters are appropri-
growth with misfit dislocations: ately chosen, as shown in Figure 14. In the case of
124 Atomic Structures and Electronic Properties of Semiconductor Interfaces
2D-coh SK-coh
SK-MD
Ed = 2.0 2D-MD
P
T
2D-MD
Ed = 1.0
2D
C1
2D SK
hC1 hT hC1 hP
Layer thickness
2D
2D C1 : FM-mode
hC1 < hT
2D-coh 2D-MD
SK
2D T C1
hC1 > hT
2D-coh SK-coh
P : SK-mode
SK-MD 2D-MD
Figure 14 Change in the free energy with layer thickness h. By tracing the trajectory of the optimal growth mode, the
change in the growth mode can be described. Two typical growth behaviors (FM and SK) are illustrated. Typical FM and SK
modes can well be reproduced by changing the formation energy of misfit dislocation (Ed ¼ 1 and 2 eV Å1 for FM and SK
modes). Reproduced with permission from Shiraishi K, Oyama N, Okajima L, et al. (2002) First principles and macroscopic
theories of semiconductor epitaxial growth. Journal of Crystal Growth, 237–239: 206–211.
the small formation energy of a misfit dislocation in actual systems. Now, we briefly mention the effect
(Ed ¼ 1.0 eV Å1), the initial two-dimensional coher- of dynamics. For example, it becomes difficult to
ent growth mode changes into the two-dimensional generate misfit dislocations at the interface when
growth mode with misfit dislocations, when the layer the epi-layer thickness is quite large. In this case,
thickness h exceeds hc 2D-MD . However, no change even though the free energy of the growth mode
occurs after h ¼ hc 2D-MD . This growth process corre- with misfit dislocation is at its lowest, misfit disloca-
sponds to the typical FM growth that is observed in tions are rarely generatd. This is related to the fact
InAs/GaAs(110) and InAs/GaAs(111) heteroepitax- that the experimental critical thicknesses of many
ial systems (Belk et al., 1997; Yamaguchi et al., 1997). lattice-mismatched systems are much larger than
On the other hand, the two-dimensional coherent the theoretical values.
growth mode first changes into the SK-coherent
one at h ¼ hT, when Ed is comparatively large 1.04.2.2.4 Theory based on atomistic
(Ed ¼ 2.0 eV Å1). By increasing the growth layer energetics
thickness, this SK-coherent growth mode changes We have noted in the above that the formation ener-
to the SK growth mode with misfit dislocations gies of misfit dislocations play a crucial role in
when h exceeds hc SK-MD . Further increase in the determining epitaxial growth processes. These ener-
growth layer thickness, however, induces the change gies of misfit dislocations are often discussed based
of SK growth mode with misfit dislocations into two- on the elastic theory. However, the above approach
dimensional growth mode with misfit dislocations. cannot include the individual and atomistic charac-
This is a typical SK growth process that is observed teristics of materials. To investigate the formation
in many lattice-mismatched systems. energies of misfit dislocations, quantum mechanical
It is well known that dynamics and energetics play and atomistic approaches are desirable, although
an essential role in determining the growth processes these involve generally need large-scale calculations.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 125
Here, we show an example of a first-principles inves- surface was depressed just above the misfit dislocation
tigation of misfit dislocations formed at the line due to the strain field caused by the misfit disloca-
heteroepitaxial interface of InAs/GaAs(110) systems tions. The vertical displacement of the top surface As
(Oyama et al., 1999). Moreover, we discuss the growth atoms is about 0.54 Å when the InAs epi-layer thickness
processes based on the first-principles results. is 4 ML. This vertical displacement is in good agree-
It has been reported by scanning tunneling micro- ment with the reported value (Oyama et al., 1999).
scopy (STM) experiments that misfit dislocations The above first-principles results also gave the
with average distance of 60 Å appear at the very total energy difference between coherent and dis-
initial stage of InAs heteroepitaxial growth on a located growth modes. The calculated energy
GaAs(110) surface after 2.8 ML (‘monolayer units’) difference are DE(2 ML) ¼ 3.15 eV per cell and
InAs growth (Belk et al., 1997). In addition, it is also DE(4 ML) ¼ 0.85 eV per cell, respectively. From
observed that two-dimensional growth continues these values, we can estimate the critical thickness
without forming SK islands. The existence of misfit as well as the phenomenological parameters. The
dislocations can be observed based on STM experi- calculated phenomenological parameters (Ed and
ments by estimating the vertical displacement of M) and the critical thickness hc 2DMD are
surface atoms with 0.5–0.7 Å. 0.96 eV A1, 1.22 1011 N m2, and 2.35 ML,
The first-principles calculations with norm-conser- respectively, for the InAs/GaAs(110) system. The
ving pseudopotentials (Hamman et al., 1979; Ihm et al., value of elastic constant M is of the same order
1979) were carried out for model systems in which two as the reported bulk modulus of InAs
and four InAs monolayers are coherently grown on (0.58 1011 N m2). On the other hand, the value
GaAs(110) substrates, and are grown with misfit dis- of the formation energy of a misfit dislocation is
locations, respectively. One InAs epi-layer unit was much smaller than the typical value of misfit dis-
forcibly removed for every 15 (GaAs) units to generate locations in crystals (1.86.9 eV Å1). This is
misfit dislocations. After the optimization, we obtained because this dislocation core structure only con-
the final structures. The so-obtained structures of four tains stable In–As bonds without relatively unstable
InAs epi-layers are shown in Figure 15. The 90 In–In or As–As bonds (Kobayashi and Nakayama,
perfect-misfit dislocations were generated in the 2008). Moreover, the critical thickness is obtained
[001] direction. The most noticeable points in this as 2.35 ML. Thus, the first isolated misfit disloca-
structure are the fivefold coordinated In atoms that tion is expected to appear when the epi-layer
have covalent bonds with the nearest five As atoms thickness exceeds 3 ML, in fairly good agreement
along the dislocation line. Although this fivefold coor- with the experiments (Yamaguchi et al., 1997). The
dinated structure was not confirmed experimentally, it above consideration indicates that the atomistic
is expected that further experimental developments information of misfit dislocations is crucial for
will observe such brand-new dislocation core struc- understanding the macroscopic growth modes
tures. Due to the dislocation formation, the epi-layer appearing in the heteroepitaxial growth.
2.52 2.52
2.57 2.57
:Ga
2.64 2.63
:In
3.12 2.86
2.70 2.60 :As
2.39 2.36
2.28 2.30 [110]
2.34 2.34
Figure 15 The dislocation core structure obtained for InAs/GaAs(110) heterointerface with an InAs layer is 4 ML. Light, dark,
and black circles indicate In, Ga, and As atoms, respectively. The numbers in the figure represent bond lengths (Å).
Reproduced with permission from Shiraishi K, Oyama N, Okajima L, et al. (2002) First principles and macroscopic theories of
semiconductor epitaxial growth. Journal of Crystal Growth, 237–239: 206–211.
126 Atomic Structures and Electronic Properties of Semiconductor Interfaces
X
1.04.2.3 Defect Formation in Heteroepitaxy H¼ VAB
A;BX
1.04.2.3.1 Heterovalency-induced defects þ k0 minðjQAþ j; jQA – jÞ
In the previous section, we explained how the A
elastic strain originating from lattice mismatch pro- 3 X
X
þ kn minðjQA j; jQB jÞ ð16Þ
duces nanostructures such as quantum-dots in n¼1 A;B
heteroepitaxy. In both hetero- and homoepitaxy
of semiconductor films, another kind of nanostruc- In this model, all atoms are assumed to be located at
ture is often generated near the interface zinc-blende tetrahedral site. The first term represents
originating from other origins. In this section, we the bonding energies between the first-nearest neigh-
concentrate on the stacking-fault tetrahedron boring atoms, A and B, and between the second-
defects (SFTs) (Finch et al., 1963; Mendelson, nearest neighboring surface dimer-bond atoms in
1964; Kanisawa et al., 2001; Nakayama and the same epitaxy layer. The values of VAB are eval-
Kobayashi, 2005) and show that they are generated uated from the first-principles calculations of
by (i) the heterovalency, that is, the valency mis- cohesive energy for zinc-blende AB crystal and sur-
match of two semiconductors, and (ii) the existence face energies of various configurations. The second
of impurity atoms on the growth surface, by term in the equation represents the energy gains due
to the electron transfer between four tetrahedral
demonstrating the cases of ZnSe/GaAs interface
bonds around an atom, A, where electrons are trans-
formation and Si homoepitaxy.
ferred from the donor bonds such as As–Se to the
ZnSe and GaAs have the same zinc-blende crystal
acceptor bonds like As–Zn. Since this transfer
structure and similar lattice constants of 5.65 and
changes the electron occupancy from the antibond-
5.64 Å, respectively. Therefore, there is no elastic
ing states to the bonding states, the gain energy is
strain at ZnSe/GaAs interfaces. However, the het-
proportional to the sum of excess or deficit electron
eroepitaxy of ZnSe on a (001) GaAs substrate often
number, QAþ or QA around the A atom. After this
produces various defects such as vacancies and inter-
intra-site electron transfer, there still exist excess or
stitials at the interface, followed by macroscopic
deficit electrons, QA ¼ QAþ þ QA, around the atom
defects like dislocations, partials, and SFTs (Kuo
A. We can assume that such electrons are transferred
et al., 1996a, 1996b; Yasuda et al., 1996). What is the within the first (n ¼ 1) to third-nearest (n ¼ 3) neigh-
origin of such extensive defect formation at ZnSe/ boring sites, hence the third term being introduced.
GaAs interfaces? The parameters, k0, k1, k2, and k3, are taken to be 2.00,
As shown in Section 1.04.2.1.1, most semicon- 0.32, 0.08, and 0.02 eV, respectively, which are again
ductor surface structures satisfy the electron obtained by fitting the GaAs and ZnSe surface
counting condition, where the surfaces stabilize energies in the first-principles calculations.
to become semiconducting by decreasing the dan- The growth simulation is performed using the
gling-bond electrons, that is, by producing cation/ conventional stochastic Monte Carlo method (Ito
anion dimer bonds on the surface and transferring and Shiraishi, 1998a, 1998b), where the absorption,
the excess electrons of surface-cation dangling evaporation, surface-atom diffusion, dimerization,
bonds into the partially occupied dangling bonds and atom exchange between the surface and substrate
of surface anion atoms. In the case of ZnSe/GaAs are considered as fundamental processes. The occur-
heteroepitaxy, however, the electron-deficit rence probability of these processes is described by
Zn–As acceptor and electron-excess Ga–Se the Boltzman factor, exp(E), where E is the surface
donor bonds inevitably appear at the interface energy change between before and after the process
and the electron transfer occurs among these and ¼ 1/kBT. The growth simulation starts on
bonds in addition to the surface dangling bonds GaAs (001) (2 4)2 surfaces with an area of
(Nakayama, 1992a; Oda and Nakayama, 1992; (12 12) and (20 20).
Murayama et al., 1998; Nakayama and Three typical cases are shown here for the initial
Murayama, 2000a). growth of ZnSe on GaAs: (i) the case when Zn and Se
The simplest atomistic model that includes these are simultaneously supplied from the beginning, and
charge-transfer effects has the total energy form as the case when Zn and Se are simultaneously supplied
(Nakayama and Sano, 2001) after some amount of (ii) Zn or (iii) Se exposure.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 127
(001)
Zn
3
Se Ga Ga Ga Ga
Ga
(–110)
As 2
(b)
(001)
Vacancy
1 z = –1 Zn Sc
V As As
z = –2 As As
0
A z = –3
Ga Ga Ga Ga
–1
(c)
Se
–2 As As As
As
–3 )
(11 10 Ga Ga Ga Ga
0) (–1
Figure 17 Atom structures of trench sites of GaAs (001)
Figure 16 Calculated ZnSe atom configurations (a) at (2 4) 2 surfaces viewed from the (110) direction: (a)
the initial stage of growth and (b) after 4 ML ZnSe before ZnSe adsorption, (b) after Zn adsorption, and (c) after
adsorption on GaAs (001) (2 4)2 substrate. Zn and Se Se adsorption. In (b), the second adsorption occurs at the
atoms are simultaneously supplied. Reproduced with cation site by Se atom, while, in (c), As evaporation occurs
permission Nakayama T and Sano K (2001) Monte Carlo after the Se adsorption. Reproduced with permission
simulation of defect formation in ZnSe/GaAs heterovalent Nakayama T and Sano K (2001) Monte Carlo simulation of
epitaxy. Journal of Crystal Growth 227/228: 665. defect formation in ZnSe/GaAs heterovalent epitaxy.
Journal of Crystal Growth 227/228: 665.
processes. However, once a certain number of anti- the anion sites which had ever been occupied by As
sites are produced nearby, they do not disappear atoms. Moreover, one can see that a number of sub-
through the diffusion and atom-exchange processes strate As atoms evaporate and disappear on the
between different layers, and, as seen in surface and even in the inner z ¼ 2 layer. This As
Figure 17(b), the vacant sites appear over these anti- evaporation is induced by the Se adsorption as fol-
sites. Moreover, it should be noticed here that, as seen lows: in case (iii), the supplied Se atom first occupies
in Figure 16(b), the boundary between adsorbed and the cation trench site as shown Figure 17(c).
vacant regions is occupied by antisite atoms of Zn and Although the Se desorption from this configuration
Se. This indicates the displacement tendency of is a major process in the next step, the second-nearest
adsorbed atoms and might induce large defects such neighboring As atom sometimes evaporates. This is
as dislocations and SFTs observed in experiments. because there are many electrons around the
We now consider the effects of initial treatment of adsorbed Se due to the Se–As donor bonds, and
the substrate. Figure 18(a) shows the grown atom when the neighboring As atom evaporates the large
configuration after 4 ML of ZnSe are adsorbed in energy gain is obtained by transferring excess
case (ii), while Figure 18(b) shows the surface struc- electrons into the dangling bond of the nearest As
ture after Se atoms are supplied on GaAs substrate in generated by As evaporation.
case (iii). In Figure 18(a), all trench sites in the Moreover, since there is no As supply, vacant sites
z ¼ 1 layer are occupied by Zn atoms, and there is are never occupied by As atoms once As atoms eva-
no antisite and vacancy, hence no sign of defect porate. Since the resulting surface is so bumpy, there
production even when further growth proceeds. On appear more number of defects than those shown in
the other hand, in Figure 18(b), Se atoms are Figure 16(b) when ZnSe is grown on this surface.
adsorbed not only on the GaAs surface but also at All the above-mentioned results are qualitatively
in agreement with experiments; the defect starts just
(a) (b)
at the interface and the Zn/Se exposure before ZnSe
z=4
supply decreases/increases the defect density in
grown layers. These results clearly demonstrate that
(001)
–3
–3 )
10
Si(111)
(11
0) (–1
Face film
Figure 18 Calculated ZnSe atom configurations after wn
4 ML ZnSe adsorption on GaAs (001) (2 4) 2 surfaces. Ridge G ro t e
Apex tra
Zn and Se are simultaneously supplied after some amount bs
of (a) Zn and (b) Se exposure. The symbol notations are the Su
same as in Figure 16. Reproduced with permission from
Nakayama T and Sano K (2001) Monte Carlo simulation of
defect formation in ZnSe/GaAs heterovalent epitaxy. Figure 19 Schematic picture of stacking-fault
Journal of Crystal Growth 227/228: 665. tetrahedron (SFT) in Si(111) grown film.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 129
(a) (b)
(c) (d)
Ridge SF
Figure 21 Side views of surface structures during Si growth. (a) Surface after the growth of two Si bilayers in the case when
two Cl atoms are left on the substrate; (b) surface after a few Si atoms are supplied on the surface shown in (a). The dome
structure is produced over the Cl atoms. (c) Surface after the growth of three bilayers with one Cl atom left on the substrate. (d)
Surface after the growth of three bilayers with two Cl atoms left on the substrate. Dotted lines, labeled SF and Ridge,
respectively, show the stacking fault face and the ridge line of SFT. Reproduced with permission from Kobayashi R and
Nakayama T (2004) Theoretical study on generation and atomic structures of stacking- fault tetrahedra in Si film growth. Thin
Solid Films 464/465: 90.
region and have a diamond structure. Moreover, we fact, a Si/SiO2 interface can provide us a nearly
can see the stacking-fault faces and the ridge line ideal two-dimensional electron gas (2DEG) that
around this region, indicating the growth of the leads to the discovery of quantum Hall effects, etc.
SFT-like structure. From these results, we can con- (von Klitzing et al., 1980). This is due to the sharp and
clude that the SFTs can be produced when more nearly defect-free characteristics of the Si/SiO2
than one Cl atoms are present nearby on the sub- interface. The typical interface defect density is of
strate, and the dome structure shown here is one of the order of 1010 cm2, and the TEM observation
the candidates for the apex structure of SFTs. This really indicates the sharp interface structure. Such
conclusion is consistent with the observation that the sharp interface comes from the layer-by-layer oxida-
SFT density is proportional to the square of the Cl tion of Si surfaces and interfaces, which has been
coverage on the Si substrate (Mazumdar et al., 1995; observed on Si(111) (Ohishi and Hattori, 1994) and
Takakuwa et al., 1997). In fact, after producing such Si(100) surfaces (Watanabe et al., 1998).
an apex, there appear three partial dislocations as
ridge lines (Kobayashi and Nakayama, 2008). 1.04.2.4.1 Phenomenological theory
The Deal and Grove model of Si oxidationhas been
widely accepted for a long time. First, we introduce
1.04.2.4 Interface Formed by Oxidation
the conventional understanding of Si thermal oxida-
The Si/SiO2 interface is a building block of Si metal- tion by Deal and Grove (1965). The suggested steps
oxide field-effect transistor (MOSFET) in large- consist of an oxidant diffusion process followed by an
scale integration (LSI) circuits. The realization of interfacial reaction process. When a Si substrate is
MOSFET in modern LSI circuits stems from the exposed to an oxygen (O2) atmosphere at around
high-quality Si/SiO2 interfaces (Kahng, 1960). In 700–1100 C, the substrate surface changes into Si
Atomic Structures and Electronic Properties of Semiconductor Interfaces 131
oxide. This is the thermal oxidation process. While that the transport of Si is also important during the
this process can be described by a simple chemical thermal oxidation, though the Deal–Grove scheme
reaction formula, Si þ O2 > SiO2, its microscopic considers only the transport of O. Thus, the move-
picture is much complicated. During the oxidation, ment of both O and Si should be clarified
O2 molecules from the atmosphere must meet the microscopically in detail to obtain a precise scheme
substrate Si to react, but the surface of the substrate is of the thermal oxidation process.
covered by Si oxide. Therefore, O2 molecules must
move through the Si oxide covering the surface. So,
1.04.2.4.2 Mechanism of oxidation
Deal and Grove assumed that the O2 molecules move
Although the Deal and Grove model can describe the
(the diffusion process) and arrive at the interface
important characteristics of thermal oxidation, it is
between the Si oxide and substrate Si and react
agreed that it fails to explain the oxidation process,
with the latter (the reaction process). By combining
especially in the thin-oxide case. Here, we introduce
these two processes, they successfully explained the
a more realistic scheme for the mechanism of the Si
kinetics of oxide thickness growth. Experiments with
thermal oxidation process based on the knowledge
O isotopes basically supported this scheme (Han and
obtained by the computational sciences and related
Helms, 1988; Lu et al., 1995): isotope-labeled O atoms
experiments (Kageshima and Shiraishi, 1998;
are located at Si/SiO2 interfaces, which indicates that
Kageshima et al., 1999; Uematsu et al., 2000, 2001,
O2 molecules diffuse through SiO2 layer and react at
2004; Fukatsu et al., 2003; Watanabe et al., 2006;
the Si/SiO2 interface. Moreover, their model could
Ming et al., 2006).
reproduce the experimental oxidation rate as follows.
The time evolution of oxide thickness can be written
as X 2 þ AX ¼ B ðt þ Þ, where X and t are oxide
• Oxidation direction Kageshima and Shiraishi
clearly show by the first-principles calculations
thickness and oxide time, respectively, and A and B that the preferential growth direction of the oxide
are, respectively, the constants related to oxidation nucleus on the surfaces is vertical to the substrate,
reaction and O2 diffusion through SiO2. The charac- whereas, at the interfaces, it is lateral. Moreover,
teristic nature of their model is that there exist two they have shown that Si atoms are inevitably
regions called the reaction- and diffusion-limited emitted from the interface to release the stress
oxidation regions that correspond to thin and thick induced during Si oxide growth (Kageshima and
oxide thickness, respectively. In the case of reaction- Shiraishi, 1998)
limited oxidation, oxide thickness X is proportional First, the growth directions of an oxide nucleus on
to the oxidation time, t. On the other hand, pffiffi in the Si surfaces and at Si-oxide/Si interfaces are dis-
diffusion limited case, X is proportional to t . cussed. The surface growth direction is investigated
However, as Deal and Grove also pointed out in by using the Si(100) surface with buckled dimmers as
their original paper, this picture cannot explain sev- the initial surface. For such a surface, the most stable
eral issues related to the thermal oxidation process adsorption site of the initial O atom is the back-bond
(Deal and Grove, 1965; Deal, 1988). In particular, it of the lower dimer atom (Kato et al., 1998). The first
cannot explain the so-called initial enhanced oxida- O atom is placed at that site and another is added
tion, the rapid oxide growth when the oxide between the SiSi bonds neighboring the first
thickness is thinner than a few tens of nanometers. SiOSi bond. Second, the most stable adsorption
One important factor, which is not taken into account site for the second O atom is determined by the first-
in the Deal–Grove scheme, is that the volume of the principles results. The possible sites are shown in
newly formed oxide is more than 2 times larger than Figure 22(a). The calculated total energies are 0.10,
that of the reacted substrate Si. Since the reaction 0.02, 0.33, and 0.61 eV per unit cell, for the sites B,
occurs at the interface surrounded by the substrate C, D, and E, respectively, relative to the total energy
and the surface oxide, the excess volume should of the site A. Using the dihydride Si(100) surface
cause a large compressive strain on the newly formed model (Northrup, 1991) as the initial surface,
oxide. Such accumulated strain could be released by Kageshima and Shiraishi also investigated the growth
viscoelastic deformation of the Si oxide. The Si den- direction for a H-terminated surface. The most stable
sity drastically decreases after Si thermal oxidation adsorption site for the initial O atom is the outermost
due to the volume expansion, indicating that Si spe- SiSi bond. The first O atom is placed at that site
cies should move from interface region to release and another one is added between the SiSi bonds
accumulated strain. Therefore, it is natural to think neighboring the first SiOSi bond. Then, the most
132 Atomic Structures and Electronic Properties of Semiconductor Interfaces
C D
C
E C
B
A B
D A
B A
Figure 22 Atomic structures for studying the oxide nucleus growth. (a) Top view for the clean surface, (b) top view for the
dihydride surface, and (c) side view for the oxide/Si interface with the less-stressed quartz-like oxide. The filled circles are O
atoms, the empty circles are Si atoms, and the small hatched circles are H atoms. Reproduced with permission from
Kageshima H and Shiraishi K (1998) First-principles study of oxide growth on Si(100) surfaces and at SiO2/Si(100) interfaces.
Physical Review Letters 81: 5936.
stable adsorption site of the second O atom can be by the difference between ‘on the surface’ and ‘at
determined. The calculated total energies are 0.12, the interface.’ Since the vertical oxide growth on
0.22, and 0.59 eV per unit cell, for the sites B, C, the surfaces is independent of the surface reconstruc-
and D, respectively, relative to the total energy of the tion, the stress (rather than the bonding nature or
site A (Figure 22(b)). These calculations indicate the charge transfer) seems to control the growth
that the oxide nucleus on the (100) surface preferen- direction. Actually, it is easy for SiOSi bonds on
tially grows vertically into the substrate, being the surfaces to expand vertically because the surface
independent of the surface reconstruction. atoms in the Si region can move upwards with almost
Next, we discuss the oxide growth direction at Si/ total freedom, while it is not easy for the bonds to
SiO2 interfaces. For the investigation of the growth expand laterally. Thus, the initial oxide nucleus on
direction for interfaces, the quartz/Si(100) interface the surface should grow vertically in order to mini-
model is used as the initial interface (Figure 22(c)) mize the stress. In the case of the interface, the
(Kageshima and Shiraishi, 1998). While the real vertical expansion of SiOSi bonds is not easy
oxide layer formed by oxidation is amorphous, for because their movement is restricted by the covered
simplicity, the oxide is modeled by a crystal of SiO2. oxide layer. Therefore, the energy gain due to the
However, this crystal model certainly has a perfect stress release by vertical growth is quite restricted.
bond network without large stress, which is an On the other hand, to minimize the interface energy,
important feature of the real amorphous oxide inter- the initial oxide nucleus at the interface should grow
face. The first O atom is added to the interface and laterally. These results have been confirmed by
the stable structure is determined. Next, the second examining the stress distribution of the calculated
and third O atoms are introduced to the interface, atomic structures estimated from the shortening of
assuming that all of the SiOSi bonds formed are the Si–Si bond lengths. These findings show the
connected. The calculations show that the structure importance of the stress in determining the growth
in which the second O atom is inserted into the site A direction. A simple phenomenological model that
is energetically more stable (by 0.29 eV per unit cell) includes oxidation induced compressive strain also
than the structure in which the second one is inserted reproduces the lateral oxide growth (Shiraishi et al.,
into the site B. Moreover, the structure in which the 2000). Recently, it has been pointed out that a simple
second and third O atoms are inserted into the sites A Monte Carlo simulation that includes the diffusion of
and C is more stable (by 0.05 eV per unit cell) than O species through SiO2 region can reproduce layer-
the structure in which the second and third ones are by-layer growth qualitatively (Watanabe et al., 2004).
inserted into the sites A and B. These results indicate The above considerations agree with the experi-
that the oxide nucleus at the Si-oxide/Si(100) inter- mental results fairly well. A previous STM-based
face preferentially grows laterally, parallel to the measurement of the oxide growth on a clean
interface. Si(111) surface (Ono et al., 1993) showed that oxide
The preferential growth direction of the oxide islands are formed in the initial stage at 600 C. The
nucleus for the (100) substrate is thus governed depth of the islands reaches several atomic layers at
Atomic Structures and Electronic Properties of Semiconductor Interfaces 133
the very initial stage. Furthermore, many experi- lower temperature could result in the formation of a
ments have clearly shown that the oxide grows well-controlled atomically thin uniform oxide layer.
atomically layer by layer at the Si-oxide/Si(100) The efficiency of these oxidation processes is sup-
and (111) interfaces (Gibson and Lanzerotti, 1989; ported by the experiments (Ohishi et al., 1994;
Komeda et al., 1998; Ohishi and Hattori, 1994; Watanabe et al., 1998).
Watanabe et al., 1998). These are consistent with the
findings based on the computational results. • Stress release We now turn to the mechanism for
the release of the accumulated stress during oxida-
The results discussed above indicate that a uni-
tion. Stress release was first investigated using a
form oxide layer can be obtained with any thickness
by thermal oxidation, once a uniform surface oxide dihydrided Si(100) surface as the initial surface. We
layer is formed. Therefore, the preparation of the sequentially insert O atoms between SiSi bonds of
initial surface oxide is crucial for obtaining a uniform the surface, assuming atomical layer-by-layer oxide
oxide layer with atomically controlled thickness, growth. This assumption simplifies the analysis of the
which gives an important guiding principle for mod- accumulated stress, as will be shown below. When
ern Si nanotechnologies. Although first-principles eight O atoms per unit cell are introduced
results also indicate that the initial growth direction (Figure 23(a)), the formed oxide has a SiOSi
of the oxide nucleus on the surfaces is vertical into network similar to that of the cristobalite of crystal
the substrate, this is true only when the thermody- SiO2 (Wyckoff, 1963). However, the structure is
namics govern the oxidation process. Actually, the highly compressed compared to that of the cristoba-
STM measurement has shown that oxidation does lite. The a and b axes of the obtained oxidized region,
not form islands, but instead forms an atomically thin which are parallel to the interface, are 23% shorter
surface oxide layer from the very first stage at room than the corresponding axes of the a-cristobalite.
temperature, where the oxidant cannot diffuse into Despite the elastic theory, the c axis, which is per-
the substrate easily (Ono et al., 1993). It has been pendicular to the interface, is only 20% longer than
reported that the O2 adsorption in the second layer the corresponding axis of the a-cristobalite. Thus, the
of the clean Si(100) surface has a nonzero barrier of structure is largely compressed to about three-
about 0.3 eV, while the adsorption in the outermost fourths of the volume of the a-cristobalite. This sug-
layer is barrierless (Watanabe et al., 1998). Therefore, gests the existence of strain release mechanism
thermal oxidation at a lower oxidant pressure and during the oxide growth. One possibility is the
Figure 23 Side views of the atomic structures for studying the accumulation and the release of the stress. (a) The structure
after sequential oxidation by two Si atomic layers; (b, c) the structures before and after the Si emission on the dihydride
surface; (d) the structure with the Si emission after oxidation by two Si atomic layers; (e, f ) the structures before and after the
emission at the oxide/Si interface with the less-stressed quartz-like oxide. The broken circles indicate the position where the
Si atom is emitted. Reproduced with permission from Kageshima H and Shiraishi K (1998) First-principles study of oxide
growth on Si(100) surfaces and at SiO2/Si(100) interfaces. Physical Review Letters 81: 5936.
134 Atomic Structures and Electronic Properties of Semiconductor Interfaces
breaking, deformation, and rebonding of the formed Moreover, when six O atoms per unit cell are intro-
SiOSi network, which would correspond to a duced to the emitting structure (Figure 23(d)), the
viscous flow of oxide. However, bond breaking and resulting bond network resembles the quartz struc-
deformation after oxide formation require a lot of ture of SiO2. The a axis of the obtained oxidized
energy. Therefore, there must be some other region is only 8% longer than the corresponding
mechanisms that work to release the stress before axis of the b-quartz. The b and c axes are only 1%
the compressed oxide is formed. and 0.2% shorter, respectively, than the correspond-
ing axes of the b-quartz. Thus, the accumulated strain
It is found that the atomic structure, when three O
is successfully released by removing Si atoms during
atoms per unit cell are introduced, is the key to the
Si oxidation. In the real experimental situation, it is
stress release (Figure 23(b)) (Kageshima and
expected that the remaining stress in the formed
Shiraishi, 1998). In this structure, an O atom is quite
oxide would be completely released after the Si
close to a surface Si atom, which has only one SiO
emission during oxide growth.
bond. Thus, these two atoms can form a bond by
Silicon emission also occurs at the Si-oxide/Si
breaking the bonds with the second-layer Si atom.
interfaces. Si emission from the interfaces has been
Moreover, the second-layer Si atom, whose two
considered using the quartz/Si(100) interface
bonds were broken, could be emitted from the sur-
model (Kageshima and Shiraishi, 1998). The total
face because of laterally compressed stress on it
energy of the emitting structure (Figure 23(f)) is
(Figure 23(c)). The calculated total energy of such more stable (by 0.41 eV per unit cell) than that of
a Si-emitting structure indicates it to be only 0.04 eV the nonemitting structure (Figure 23(e)), although
per unit cell higher than that of the nonemitting two Si dangling bonds are formed after Si emission.
structure, though there remain two dangling bonds. This means that, even at the oxide/Si interfaces, Si
This structure resembles the well-known A center (or atoms are preferentially emitted during oxide
the VO center) in bulk Si crystal (Pajot, 1994; Chadi, growth. Moreover, although layer-by-layer oxida-
1996). In addition, when we sequentially insert O tion is assumed as mentioned above, further
atoms, the total energies for all of the emitting struc- calculations show that the Si emission is indepen-
tures are more stable than those for the dent of the oxide growth mode. Even after the
corresponding nonemitting structures (Figure 24). initial vertical oxide growth on the surfaces, the
The energy advantage is up to 2 eV per unit cell. emission can occur again when the oxide islands
This is because the two remaining dangling bonds connect with each other. Since stress accumulation
first form a weak bond by laterally compressed stress, is inevitable in the Si oxidation process, the release
and are finally terminated by forming a SiOSi of this stress by Si emission should be essential and
bond. This also indicates that the Si emission scarcely universal.
results in the creation of the interfacial gap states.
• Si emission As discussed above, the emitted Si
atoms should play an important role in the oxidation
process. Since the energy advantage of the
3 Si emission (up to 2 eV) is smaller than the
Energy advantage
2 Emission preferential
(eV \ unit cell)
1019
Si
B concentration (cm–3)
Si species B 5 ×1013 cm–2
1018
1200 °C 24 h
Figure 25 Schematic illustration of Si oxidation
processes. Oxygen molecules should diffuse through a SiO2 28SiO
2 thickness
layer before reacting with Si at Si/SiO2 interfaces.
200 nm
1017
300 nm
from the Si/SiO2 interface. As a result, the oxidation 650 nm
process contains diffusion of both O and Si species,
As-impla
as illustrated in Figure 25.
Next, we show the experimental findings of Si 1016
0 50 100 150 200
species backflow from Si/SiO2 interface to SiO2 Depth (nm)
region during oxidation. First experimental example
Figure 27 Diffusion profiles of B in SiO2 with various
is that the B and Si self-diffusion near the Si/SiO2 thicknesses. Samples were implanted with B to a dose of
interface is remarkably enhanced (Uematsu et al. 5 1013 cm2 and annealed at 1200 C for 24 h. The nearer
2004; Fukatsu et al., 2003). This is thought to be the the Si/SiO2 interface, the broader the B profiles becomes.
effect of emitted Si species from the Si/SiO2 inter- Reproduced with permission from Uematsu M, Kageshima
H, Takahashi Y, Fukatsu S, Itoh KM, and Shiraishi K (2004)
faces. To investigate the B diffusion in SiO2 by
Correlated diffusion of silicon and boron in thermally grown
secondary-ion mass spectroscopy (SIMS) analysis, SiO2. Applied Physics Letters 85: 221.
the samples are prepared as follows. The isotopically
enriched 28Si epi-layer was thermally oxidized in dry
O2 to form 28SiO2 with thickness 200, 300, and Figure 27 shows the depth profiles of 11B before
650 nm. The samples were implanted with 30Si at and after annealing at 1250 C for 6 h. As shown in
50 keV to a dose of 2 1015 cm2 and capped with Figure 27, the profiles of 11B become broader with
a 30-nm-thick silicon nitride layer. Subsequently, decreasing thickness of the 28SiO2 layer, that is, B
the samples were implanted with 11B at 25 keV to a diffusivity increases with decreasing distance from
dose of 3 1015 cm2. The final structure is shown in the Si/SiO2 interface. If B diffusion is governed by a
Figure 26. The samples were pre-annealed at single process, the B diffusivity should depend on
1000 C for 30 min to eliminate implantation the distance. However, it is physically unnatural.
damages, and were annealed in the resistively heated The distance dependence of diffusivity is also
annealing furnace at various temperatures in the observed in the Si self-diffusion in SiO2 (Fukatsu
range of 1100–1250 C. The diffusion profiles of 11B et al., 2003). In the case of Si self-diffusion, SiO
and 30Si were measured by SIMS. molecules generated at the interface and diffusing
into the oxide enhance Si self-diffusion (Fukatsu
et al., 2003). These results indicate that SiO mole-
Si3N4 30 nm cules also enhance B diffusion, because B diffusivity
is higher near the interface, where the SiO
concentration is high. Moreover, first-principles
200–650 nm calculations show that interstitial BO complexes
can diffuse through SiO2 layers with relatively low
28SiO
2 activation barrier (Otani et al., 2003). Considering
28Si that this interstitial BO complex is stoichiometri-
Implanted
30Si
cally equivalent to the BSiSiO complex, these
first-principles results may indicate that the
Implanted 11B existence of SiO species also enhances B diffusion
Figure 26 The sample structure employed for
(Uematsu et al., 2006).
considering the enhanced B diffusion near Si/SiO2 Taking into account the effect of SiO, coupled
interfaces. diffusion equations that include normal thermal B
136 Atomic Structures and Electronic Properties of Semiconductor Interfaces
diffusion and SiO-assisted B diffusion can readily be species which segregate at the surface. Ming et al.
constructed (Uematsu et al., 2004). As shown in (2006) performed the high-resolution Rutherford
Figure 27, a numerical simulation that includes backscattering (HRBS) measurement and confirm
the mechanism that SiO species generated from the existence of surface Si component around
Si/SiO2 interface enhance B diffusion well repro- 361 keV in addition to the Si peak at the Si/SiO2
duces the distance dependence of B diffusivity, interface near 350 keV, only when interfacial SiO2
although constant diffusivities are assumed. growth occurs, as clearly shown in Figure 28. This
Moreover, the time dependence of B diffusivities experiment clearly confirmed that Si emission during
has also been reported. It is expected that SiO con- Si oxidation, which was predicted by the first-prin-
centration generated from Si/SiO2 interface ciples calculations (Kageshima and Shiraishi, 1998) is
increases with longer annealing time. Considering surely observed.
the above discussions that SiO species enhance the Here, we introduce the physical origin of initial
B diffusion, B diffusivity is expected to be increased enhanced oxidation, which has been a mystery of Si
with longer annealing time. Actually, the clear oxidation for a long time. First, we introduce the Si
enhancement in B diffusivity has been confirmed; emission model (Kageshima et al., 1999). If the effect
1.5 1016 and 3.0 1016 cm2 s1 B diffusivities of such SiO interstitials as discussed previously under
are obtained after 8 and 24 h 1200 C anneals, the so-called Si emission model is considered, the
respectively. This time dependence also supports initial enhanced oxidation can systematically be
the fact that B diffusion is assisted by SiO reproduced (Kageshima et al., 1999; Uematsu et al.,
(Uematsu et al., 2006). 2000). As per Deal and Grove, the oxide growth rate
Next, we show much direct proof of Si species equation can be derived from the reaction–diffusion
emission during Si oxidation (Ming et al., 2006). equation, while newly considered SiO flow from the
Experiments use the characteristic material proper- interface to the surface should be included besides
ties of SiO2 in thin HfO2. It is known that SiO2 and the O2 flow from the surface to the interface
HfO2 reveal phase separation when HfO2 is thin (Figure 29).
enough. Thus, when the HfO2/SiO2/Si stacked sam- Due to the SiO flow, the interfacial reaction rate
ple is oxidized and Si substrate oxidation occurs, it is of O2 must be modified. SiO should be much easily
expected that emitted Si species diffuse through oxidized on the surface than in the oxide because the
HfO2 and segregate at the surface. Thus, surface- oxidation of SiO should be incorporated with the
sensitive observations can detect the emitted Si volume expansion. Then, in the thin-oxide limit,
15
400 keV He+ HfO2 /SiO2 /Si [111] channeling
As-grown
900 °C × 2 min
10
Counts (a.u.)
×5 Hf
5 Si
0
320 340 360 380 400
Energy (keV)
Figure 28 High-resolution Rutherford backscattering spectra of as-grown and annealed HfO2/SiO2/Si. Reproduced with
permission from Ming Z, Nakajima K, Suzuki M, et al. (2006) Si emission from the SiO2/Si interface during the growth of SiO2 in
the HfO2/SiO2/Si structure. Applied Physics Letters 88: 153516.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 137
Gas Oxydant
Point (2) k′
S
k: rapid reaction C Si C SO
x Point (1)
k : slow reaction
DSi Interfacial reaction rate
k C Si(x) DO k = k0 (1–C ISi /C 0Si)
Oxide
x
C ISi C IO
k
Silicon
V
Emitted Si
Figure 29 Schematic view of the reaction-diffusion kinetics in the Si emission model. From Kageshima H, Uematsu M,
Akiyama T, and Ito T (2007) Microscopic mechanism of silicon thermal oxidation process. ECS Transactions 6: 449.
vacuum environment. On the other hand, the reflec- layers are isotropic, while the surface/interface layers
tance difference spectroscopy (RDS) is a powerful (ILs) are anisotropic. Thus, since the polarization ani-
in situ optical measurement to observe the electronic sotropy originates from the anisotropy of surface/
structures of semiconductor surfaces/interfaces and interface electronic states, the RDS can teach us the
their time evolution (Aspnes and Studna, 1985; nature of surface/interface electronic structures. In
Hingerl et al., 1993; Uwai and Kobayashi, 1994; particular, the spectral shape reflects the localized nat-
Nakayama, 1997; Murayama et al., 1998). Since the ure of the interface states. The simplest picture of
electromagnetic light penetrates deep into the Si polarization around the semiconductor interface is
substrate, that is, an order of 1000 Å, the RDS can obtained by the bond polarization model (Nakayama
detect, in principle, the layer-by-layer thermal oxi- and Murayama, 1999). The bond polarization in
dation. Nakayama and Murayama proposed the use the nth semiconductor layer is represented by
of RDS to control the layer-by-layer thermal oxida- n ¼ ð – 1Þn b ð!Þ þ sn ð!Þ, where b ð!Þ is the bulk
tion, by theoretically investigating the variation of polarization corresponding to the light with h! energy,
RDS spectra of buried SiO2/Si interfaces (Nakayama while sn ð!Þis the deviation of polarization from the
and Murayama, 2000). bulk values. The (-1) prefactor denotes that the RDS
At first, we briefly explain the RDS. The RDS measures the reflectance difference because the odd
measures the difference in reflectance of surface/inter- and even layers contribute to the spectra with alter-
face between two perpendicular light-polarization native signs. By summing up the contributions from all
directions as shown in Figure 31(a). The inner bulk layers, the spectra become
X X
ð!Þ ¼ n ?e – 2na ¼ b ð!Þ=2 þ sn ð!Þ ð18Þ
(a)
where – 1 is the decay length of the light in semi-
(001) RDS spectra ∝ conductor. Figure 31(b) shows the typical spectral
ε2,(−110)− ε2,(110) shapes of the RDS. When there exist interface states
(−110) strongly localized at the interface, the sn ð!Þ term
becomes extremely large because the optical transi-
(010)
tions between localized states are large. Thus, the
RDS shows the peak-like shape at the transition
energy, as shown by A in Figure 31(b). When there
are no interface states but the electrons in bulk have
localized nature, the first term, b ð!Þ=2, produces
spectra having the shape of bulk dielectric function,
as shown by B in Figure 31(b). When the electronic
Bulk is isotropic Surface is anisotropic
states are extended from the interface, we can rewrite
the sn ð!Þ term as qb ð!Þ=q!?" because the bond
(b) polarization gradually changes the excitation energy
from the surface for extended electronic states. As a
result, the RDS spectra have the energy-derivative
shape for the dielectric function, as shown by C in
RDS spectra
(a) (b)
[001]
O O
Si Si
SiO2
Si
(c) (d)
O
Si
Si O Si
[–110]
0 O
2 1) [110]
(00
3
Figure 32 Schematic diagrams of SiO2/Si (001) interfaces. (a) Flat interface A with the desorption of interface Si atoms;
(b) flat interface B after 1 ML oxidation of A interface; and (c) flat interface with crystalline SiO2 layers around the interface.
(d) Bird’s eyeview of layer-by-layer oxidation of Si(001) surface as noted from the surface along [001]. Reproduced with
permission from Nakayama T and Murayama M (2000b) Atom-scale optical determination of Si-oxide layer thickness during
layer-by-layer oxidation: Theoretical study. Applied Physics Letters 77: 4286.
edge SiSi bonds terminated at SiO2/Si interface interface A, and vice versa. Namely, these two inter-
alternatively change the direction between [110] faces alternatively appear during the layer-by-layer
and [110]. Therefore, since the RDS measures the oxidation. The detailed atomic positions at these
reflectance difference between two perpendicular interfaces are taken from the first-principles calcula-
directions, that is, the anisotropy of interface polar- tions (Kageshima and Shiraishi, 1998). Here, we
ization originating from the interface electronic concentrate on two features in Figures 33: (i) we
structure, one can optically detect the change of first notice that the spectra of the interfaces, A and
interface-bond direction and know the advance of B, have similar shapes and opposite signs. This result
monolayer oxidation. indicates that when the A and B interfaces appear
Figures 33(a) and 33(b) show the calculated RDS alternatively during oxidation, the RDS signals
spectra of two SiO2/Si interfaces, A and B, which are oscillate between positive and negative values,
displayed in Figures 32(a) and 32(b), respectively which is similar to the reflectance high-energy elec-
(Nakayama and Murayama, 2000). It is noted that the tron diffraction (RHEED) oscillation in the epitaxial
interface B corresponds to the interface when the layer-by-layer growth. Especially, such an observa-
monolayer oxidation is completed starting from the tion is apparently effective around 3.5/4.5 eV, where
140 Atomic Structures and Electronic Properties of Semiconductor Interfaces
the RDS signal is large. In this case, the period of discussed above, that the anisotropic RDS signals
oscillation corresponds to the bilayer oxidation and appear due to the modulation of bulk electronic
one can know the number of oxidized layers by structures that are extended around the interface
counting the oscillation; (ii) the other feature (Nakayama and Murayama, 1999). In fact, such a
observed in Figure 33 is the spectral shape of the modulation can be examined by analyzing the layer
large peaks around 3.5 and 4.5 eV, which, res- contributions to the RDS, which are also shown in
pectively, correspond to the E1 and E2 van Hove Figure 33(c).
singularity energies of bulk Si shown in This theoretical prediction was confirmed by
Figure 33(d). These peaks have shapes similar to the RDS experiments by Yasuda et al. (2001).
those of the energy derivatives of "2, which indicates, Figure 34(a) shows the spectral oscillation mea-
based on the general theory of spectral shapes sured at around 3.5 eV, while the observed spectra
after 2 ML oxidation are shown in Figure 33(d).
The spectral oscillation is clearly seen as a func-
(a) tion of oxidization time, being in good agreement
with the theoretical prediction. We note that the
20 Å
with many kinds of metals, such as Au, Ag, Cu, Ni, Al,
etc. Such intermixing typically ranges from nano- to
micrometers and provides serious damages to nanos-
t2 t3 cale devices such as circuit shortening. Thus, it is
O2 in
important to understand what the motive force of
intermixing is and how the intermixing proceeds.
On the other hand, in the case of metalSi
101 102 103 combinations that show the intermixing at inter-
Elapsed time (s) faces, it is well known that most transition metals
and rare-earth metals, typically Ni metal, produce
(b) metal silicides by intermixing, while sp-orbital
The 3rd layer
(one-step ox.) metals like Al and a few transition metals such as
Au and Ag have no silicide phases. In addition, in
103 spite of various bulk silicide phases of NixSiy, only
the silicides with specific stoichiometry of
(x,y) ¼ (1,2), (1,1), (2,1) are allowed to grow on Si
t3 (S)
(a) (b)
Au/Si(111) Al/Si(111)
Au Al
Al
Au
Si(1) Si(1) Si(1)
Au Si(1)
Al
Si(2) Si(2) Si(2)
Si(2)
Si(3) Si(3)
Si Si(1)
Si Si
Figure 36 Calculated electron-density contour map around (a) Au/Si and (b) Al/Si interfaces. Three ML Au and Al are
deposited on Si(111) surface and innermost metal atoms are moved into Si layers by about 1.8 Å (z ¼ 1.8 Å in Figure 35(a)).
From Murayama et al. (2001); Nakayama et al. (2002b, 2008). Reproduced with permission from Nakayama T, Itaya S, and
Murayama D (2006) Nano-scale view of atom intermixing at metal/semiconductor interfaces. Journal of Physics: Conference
Series 38: 216.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 143
atoms and spread their valence-electron density bonding energy gain. In this way, one can know
widely over the Si substrate. As a result, the electron that the small atomic radius and the larger/smaller
density increases around Si atoms, and such an electronegativity of metal atoms are relevant fac-
increase screens up and decreases the charge density tors to realize intermixing.
of Si–Si bond (Hiraki, 1980). Once such screening
weakens the interface SiSi bonds, the interface Si
atoms can diffuse again into Al layers similar to the 1.04.2.5.2 Silicidation
case of Au/Si interface. In fact, the calculated barrier Next, we consider what factors distinguish the
of Si diffusion in the case of Al/Si is also small, silicide formation between various metals. When
around 0.2 eV. Therefore, one can roughly say that MSi2 silicide (M ¼ metal atom) is selected as a
the screening due to the charge extension is the representative phase among a variety of stoichio-
motive force of intermixing for Al/Si interfaces. metric phases, the calculated formation energies
In this way, the charge redistribution during the of TiSi2, NiSi2, AuSi2, and AlSi2 become 0.4,
diffusion and the motive force of intermixing 0.1, þ0.3, and þ0.4 eV, respectively (Nakayama
are somewhat different between Au/Si and Al/Si et al., 2008, 2009). This result indicates that Ti
interfaces. From this result, we can expect the and Ni silicides are stable, while Au and Al sili-
difference of diffusion and intermixing patterns of cides are unstable, being in good agreement with
metal atoms between Au/Si and Al/Si interfaces. the observed bulk phase diagrams (Madelung,
Since the production of AuSi bonds is the origin 1997). The reason for such difference in formation
of Au atom diffusion, the Au diffusion has direc- energies is clearly understand by noting the band
tional tendency and the penetration of Au into Si is structures. Figures 37(a) and 37(b) show the cal-
expected to proceed in a needle-like line pattern. culated band structures of Ni and Au bulks
On the other hand, since the charge extension of Al (Nakayama et al., 2008, 2009). In the case of Ni,
toward Si atoms in various directions is the origin one can see five flat bands between 4 and
of Al-Si intermixing, the Al diffusion has no direc- þ0.5 eV, which are made of Ni d orbitals. The
tivity and the intermixing of Al and Si is expected band that ranges from 9 to 7 eV and crosses
to proceed with a two-dimensional face-like front these d bands is made of Ni s orbitals. Similar d
interface. and s bands are seen for bulk Au between 7.5
It should be noted here that, in contrast to the and 2 eV and from 11 to 4 eV, respectively.
cases shown above, no mixing is observed at most The most important difference between Ni and
of the metal/(wide-gap-semiconductor) interfaces, Au is the electron occupation in d bands; all d
such as Au/GaN, although Au shows intermixing bands are occupied by electrons for Au, while
at interfaces with several semiconductors. There about half of the electrons are unoccupied in the
are two factors to prevent intermixing. One is the top d band of Ni.
elastic energy loss; since the atomic radius of Au, On the other hand, Figures 37(c) and 37(d)
1.44 Å, is much larger than the average atomic radii show the band structures of NiSi2 and AuSi2 sili-
of Ga and N, 1.0 Å, the diffusion of Au pushes the cides. One can see the deformation of metal-
GaN substrate to induce a simple compression originated bands, which is caused by the produc-
strain of GaN layers and produces the elastic tion of covalent-like bonds between metal and Si
energy loss. For example, Pb metal layers were atoms due to the hybridization of s þ d orbitals of
observed showing no intermixing on Si substrate, metal atom and s þ p orbitals of Si, as seen in the
because the atomic radius of Pb, 1.8 Å, is much case of Au/Si interface shown in Figure 36(a).
larger than that of Si, 1.1 Å. The other factor to However, we can roughly identify that the rela-
prevent the mixing is the chemical-bonding energy tively flat d-like bands, from 9 to 3 eV for
loss; since the electronegativities of Ga, Au, and N NiSi2 and from 9 to 6 eV for AuSi2, are located
atoms are 1.6, 2.4, and 3.0, respectively, the Ga far below the Fermi energy in the case of silicides,
atoms prefer locating among N atoms to Au compared to the case of bulks. This result indicates
atoms. When the Au atom is located on GaN sur- that the d bands of Ni, which are partially unoc-
face, the top-layer Ga is surrounded by Au and N. cupied in bulk, become fully occupied in NiSi2.,
On the other hand, when the Au atom moves thus indicating the electron charge transfer from Si
between Ga and N, the top-layer Ga is surrounded sp orbitals to Ni d orbitals. However, in the case of
only by Au atoms, thus losing the Ga-N high ionic AuSi2, the d bands are already fully occupied by
144 Atomic Structures and Electronic Properties of Semiconductor Interfaces
(a) (b)
25 Ni Au
15
20
10
15
Energy (eV)
Energy (eV)
10 5
5
EF0
EF0
–5
–5
–10
–10
G XK G L KW X G XK G L KW X
(c) (d)
NiSi2 AuSi2
5
5
EF0
Energy (eV)
–5 –5
–10 –10
–15
G XK G L KW X G XK G L KW X
Figure 37 Calculated band structures of (a) Ni, (b) Au, (c) NiSi2, and (d) AuSi2. Reproduced with permission from Nakayama
T, Shinji S, and Sotome S (2008) Why and how atom intermixing proceeds at metal/Si interfaces: Silicide formation vs. random
mixing. ECS Transactions 16(10): 787–795 and Nakayama T, Sotome S, and Shinji S (2009) Stability and Schottky barrier of
silicides: First-principles study. Microelectronic Engineering 86: 1718.
electrons in bulk Au, and such a transfer never regions to grow in thermal equilibrium. It is clearly
occurs. From these results, one can conclude that seen that all NiSi2, NiSi, Ni2Si, and Ni3Si phases are
the stability of metal silicides is realized by the realized in the bulk growth, depending on the growth
electron transfer from Si to unoccupied d-orbital condition, while the Ni3Si phase becomes difficult to
states of transition-metal atoms. This stabilization grow on the Si substrate. Such changes in stability are
scenario of metal silicides well explains why tran- caused by two factors. One is the elastic strain in
sition metals, except Au and Ag, produce stable NixSiy on the Si substrate. As the proportion of Ni
silicides at metal/Si interfaces and why Al does increases, the lattice constant of NixSiy increases and
not produce silicides. the compressed strain in it increases. The other factor
Next, we consider why only the NixSiy silicides is the energy loss of the interface bonding. With
with specific stoichiometry, (x,y) ¼ (1,2), (1,1), (2,1), increasing Ni, the number of stable Si–Si covalent
are allowed to grow on Si substrate among various bonds decreases, which also promotes the instability
silicide phases. Figures 38(a) and 38(b) show the of Ni3Si.
calculated phase diagrams of NixSiy as a function of Finally, we point out that the above stabilization
chemical potentials of Ni and Si, in the case of mechanism is closely related to the work function at
the bulk growth and the growth on the Si substrate, silicide/Si interfaces (Nakayama et al., 2009). Since
respectively (Nakayama et al., 2008, 2009). Chemical the p-orbital valence energy states of Si are located
potentials represent the supply ratios of Ni and Si above those of Ni, the charge transfer occurs from Si
and the gray regions correspond to the allowed to Ni in NixSiy and lowers the positions of Fermi
Atomic Structures and Electronic Properties of Semiconductor Interfaces 145
Ni2Si
interface. Figure 40 schematically shows the relation
–4.3 of the complex band structures of A and B bulks to
–41.6 –41.4 –41.2
electronic states at the interface. In A and B bulks,
(b) μ of Ni [Ht] because of the crystal periodicity, the electronic
–3.8 states are characterized by the extended Bloch wave
functions, n;k ðrÞ, where n is the band index and k
NiSi2 the Bloch wave-number real vectors. Reflecting the
periodicity, the energy spectra, "n,k, show the band
–4
structure having band gaps. However, since the
Ni3Si
NiSi
the z-direction, that is, perpendicular to the interface,
–4.2 the electronic states with the energy within the band
gaps of bulks and complex wave-number kz are
allowed to exist. Because Im kz 6¼ 0, such states
–41.6 –41.4 –41.2
show amplitude decay along the z-direction and are
μ of Ni [Ht]
sometimes called as the evanescent wave states. For
Figure 38 Calculated phase diagrams of NixSiy as functions example, the electronic state having energy "1 in
of Ni and Si chemical potentials, in cases of (a) bulk growth and Figure 40 is constructed by the connection of two
(b) growth on the Si substrate. Gray regions are regions where propagating Bloch states in A and four evanescent-
NixSiy is allowed to grow in thermal equilibrium. Reproduced
with permission from Nakayama T, Shinji S, and Sotome S wave states in B, which all are described by the
(2008) Why and how atom intermixing proceeds at metal/Si intersection points of the band structures and the
interfaces: Silicide formation vs. random mixing. ECS energy plane. Since this state propagates along the
Transactions 16(10): 787–795 and Nakayama T, Sotome S, z-direction to z ¼ 1in A (z < 0) but decays sharply
and Shinji S (2009) Stability and Schottky barrier of silicides:
in B (z > 0), we can call it the state originating from
First-principles study. Microelectronic Engineering 86: 1718.
E E
(a) (b)
ψε1 (r)
Si conduction band Si conduction band
ε1
NiSi2 0.55eV TiSi2 0.69 eV
0.42
0.27 TiSi Complex band
Ni2Si NiSi 0.48 Ti2Si 0.64 ψε2 (r)
Ti3Si
Ni3Si 0.21 0.40
ε2
Valence band Si Valence band
Real band
Figure 39 Calculated Fermi-energy positions of NixSiy
and TixSiy. Reproduced with permission from Nakayama T,
Sotome S, and Shinji S (2009) Stability and Schottky barrier
of silicides: First-principles study. Microelectronic Re kz
Re kz
Engineering 86: 1718.
Im kz Im kz
Material A Material B
energy originating from Si electronic states for Ni-
rich silicides. Thus, as seen in Figure 39, the work Figure 40 Schematic diagram of complex band
function of silicides increases as the proportion of Si structures of A and B materials and connection of wave
decreases in NixSiy. function at A/B interface.
146 Atomic Structures and Electronic Properties of Semiconductor Interfaces
material A. On the other hand, the electronic state origins for interfaces owing to the breakdown of
having energy "2 is made of evanescent-wave states translational symmetry at the interface and the pro-
of both A and B materials, and is localized near the duction of new types of atomic bonds between A and
interface, thus called the interface state. It is apparent B at the interface. Since, at present, there are no
that the state that has energy within the band gaps of definitive experimental tools, such as STM, for the
both A and B bulks becomes the type-(ii) interface surface to observe the interface atomic structures, it
state, while the state that has energy within the band is sometimes difficult to separate these origins.
spectrum of either A or B becomes the type-(i) state The details of the intrinsic origins are considered
that propagates deep into either A or B. These are in the next two subsections. Here, we briefly illus-
necessary conditions of energy for judging the types trate examples of the extrinsic origins in Figure 42.
of interface states. It should be noted here that the New atomic environment shown in the number
number of complex bands that produce the interface region ‘2’ and the existence of impurity atoms
states is in general infinite, though the finite ones are shown in regions ‘6’ and ‘7’ often produce the inter-
displayed in Figure 40. However, their contribution face states in the band gap. The step structures of the
becomes considerably small when they are located substrate shown in ‘3’ and ‘4’, which are left in the
far from the real band structures. heteroepitaxy, and the appearance of misfit disloca-
For the type-(i) bulk-like interface states, the tions shown in ‘5’, which are produced by the lattice
existence of interface is sometimes renormalized to mismatch between A and B, also become the struc-
be described as some potential, combined with the tural origins of interface states.
effective-mass approximation. This treatment is
often called the quantum-well-picture representa- 1.04.3.1.2 Semiconductor/semiconductor
tion. For example, as shown in Figure 41 for the interface states
case of GaAs/AlAs superlattice, AlAs layers are trea- We first consider the Ge/GaAs (110) nonpolar
ted as potential barriers and GaAs ones as quantum- interface as an example. Figure 43 shows the
wells. By adopting this picture, many interesting atomic structure near the interface (Pickett and
electronic phenomena are analyzed. In such cases, Cohen, 1978). Because Ge and GaAs have similar
the key quantity that governs the electronic states is lattice constants, the interface has few dislocations
the band offset, which is the energy difference of and becomes remarkably sharp. The atomic bonds
bands between the two semiconductors (represented represented by broken lines, GeGa and GeAs,
by Ec and EV in Figure 41). The details of this are not present in tetrahedral bulk semiconductors.
band offset are discussed in the next subsection. GeGa and GeAs bonds have 1.75 and 2.25
The physical origins of the existence of type-(ii) electrons, as opposed to the ordinary semiconduc-
interface states are also categorized into two groups: tors, which have 2.0. Thus, these bonds are often
(a) extrinsic origins such as dislocations and impurity called acceptor and donor bonds, respectively.
atoms at interfaces, which are sometimes produced To judge the existence of interface states, it is
and incorporated in the case of interface formation, as convenient to compare the two-dimensional band
shown in the previous section; and (b) intrinsic structure of an interface with the projected spectra
E E
Ve (z)
LB Lw
Conduction
Ec (AlAs) ΔEc
z band
Ec (GaAs)
Band gap
Ev (GaAs)
Ev (AlAs) z ΔEv
Valence
Vh(z) band
kz
kz
AlAs GaAs
AlA GaAs AlAs GaAs AlAs
Figure 41 Quantum-well diagram and band offset, in the case of GaAs/GaAs superlattice.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 147
(a)
6
5
4 7
1
2 3
(b)
Figure 42 Schematic examples of interface structures that produce extrinsic interface states.
0 B2 P2
Interface layer
–2
Energy (eV)
P1
–4
1st Ge layer S2 B1
–6
z –8
x
–10
y
–12 S1
Figure 43 Atomic structure of Ge/GaAs (110) interface.
Reproduced with permission from Pickett WE, Louie SG, and
Cohen ML (1978) Self-consistent calculations of interface Γ X M X′ Γ
states and electronic structure of the (110) interfaces of k
Ge–GaAs and AlAs–GaAs. Physical Review B 17: 815.
Figure 44 Band structure of Ge/GaAs (110) interface.
Reproduced with permission from Pickett WE, Louie SG, and
of bulk band structures into two-dimensional Cohen ML (1978) Self-consistent calculations of interface
Brillouin zone. Figure 44 shows such a comparison states and electronic structure of the (110) interfaces of
for Ge/GaAs interface, where band structures are Ge–GaAs and AlAs–GaAs. Physical Review B 17: 815.
calculated using super unit cells and self-consistent
pseudopotential method (Pickett et al., 1978). GeAs bond, while S2 and B2 ((c) and (a)) are
Shaded regions correspond to bulk bands, thus similar states localized at the GeGa bond. In this
pointing to the existence of only the bulk-like way, the interface states at nonpolar sharp inter-
interface states even at the interface. The electro- faces appear to originate from new interface bonds.
nic states, S1B2, appearing outside the shaded However, there appear no interface states within
regions (gap, pockets), are states intrinsic to inter- the fundamental band-gap region, 0.01.0 eV in
face and are localized at the interface as shown in Figure 44. This is because all dangling bonds
Figures 45(a)–45(d). S1 and B1 states ((d) and (b)) that exist on the Ge and GaAs surfaces are termi-
are made of s and p orbitals and localized at the nated at the interface. As a result, the physical
148 Atomic Structures and Electronic Properties of Semiconductor Interfaces
Bond charge
Reproduced with permission from Pickett WE and Cohen ML (1978) Theoretical trends in
the abrupt (110) AlAs–GaAs, Ge–GaAs, and Ge–ZnSe interfaces. Journal of Vacuum
Science and Technology 15: 1437.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 149
0.5
0
1.5
Region VI
EF
1.0
0.5
0
−14 −12 −10 −8 −6 −4 −2 0 2 4
Energy (eV)
Figure 47 Valence charge distribution around the Al/Si Figure 48 Local DOS around the Al/Si (111) interface.
(111) interface: (a) contour map display and (b) distribution Regions I–VI are displayed in Figure 47. Reproduced with
averaged along the interface. Reproduced with permission permission from Louie SG and Cohen ML (1976) Electronic
from Louie SG and Cohen ML (1976) Electronic structure of a structure of a metal–semiconductor interface. Physical
metal–semiconductor interface. Physical Review B 13: 2461. Review B 13: 2461.
150 Atomic Structures and Electronic Properties of Semiconductor Interfaces
(a) Al-Si interface semiconductors. One can see that the number of
states with E = 0 to 1.2 eV
MIGSs becomes small as the ionicity of semiconductor
1.8 1.5 1.8 1.5 2.1
increases. This is because a semiconductor of large
1.8
1.7
0.6 1.0 0.0
ionicity has larger band gap and thus the DOS is largely
0.3
0.3 decreased because the complex band leaves the real
1.8
1.5 1.8 1.5 band, has a large imaginary part of wave number, ImKz,
1.8
0.9 2.7
(b) and the MIGS is strongly localized at the interface.
2.0
Al-Si interface P (z)
States with E = 0 to 1.2 eV
1.04.3.2 Band Alignment
1.0
and proposed that
CNL is determined only by the When the interface states appear due to the
electronic structure of the bulk semiconductor. As extrinsic origins such as structural disorders, we
explained above, the MIGS is the eigenstate in com- have to consider other methods for calculating
plex band structures having energies within the band
CNL . Based on their systematic XPS
gap of the bulk system. Since the complex bands experiments,Spicer et al. (1980, 1989) were the first
connect with real bands at the band edge of real to point out the importance of defect-induced inter-
bands, the MIGSs with lower energies possess a face electronic states to determine the Schottky
valence-band character, while those having higher barrier. Drummond (1999) extended this concept
energies carry a conduction-band character (see and categorized observations into various defect
Figure 51(a)). The effective Fermi energy,
CNL , is types such as vacancies and antisite atoms.
defined as the boundary between these valence-band- It is often observed that the DOS of interface
and conduction-band-like states, and are thus states shows a U-like shape in the band gap for a
approximately given by the sign-changed boundary variety of disordered metal/semiconductor inter-
energy of the cell-averaged propagating Green faces, as shown in Figure 52(a). Hasegawa and
function: Ohno called these states as disorder-induced gap
states (DIGS). They found that the Fermi energies
Z XZ jnk ðr Þjnk ðr þ RÞ of metals match the energy position of the lowest
G ðR;
CNL Þ ¼ dr d3 k ¼0
n
CNL – "nk DOS value of DIGS; thus, such energy state acts as a
unitcell
CNL-like Fermi energy of semiconductors
ð20Þ
(Hasegawa and Ohno, 1986). Figure 52(b) shows
where jnk ðrÞ and "nk are eigen wave function and the basic concept they used for calculating
CNL for
energy of the electronic states of bulk semiconductor, disordered interfaces. They considered that the
respectively, with the band index n, and the Bloch DIGS appears due to the disorder of semiconductor
wave number k. Here, R is the real lattice vector bonds at the metal/semiconductor interfaces and
perpendicular to the interface. In a one-dimensional proposed that the CNL-like energy that separates
system,
CNL coincides with the branching point of the bonding and antibonding states is obtained as an
the complex band. Using
CNL calculated by this average of sp3-orbital energies of semiconductors.
formula, Tersoff succeeded in reproducing SBHs of Here, we would like to point out two important
various metal/semiconductor interfaces. Table 4 features in these kinds of
CNL theories. At first, due
shows the calculated SBH values, together with the to some ambiguity in determining the boundary
observed values in experiments. The agreement is between valance- and conduction-band-like MIGS
quite good, especially for semiconductors with or DIGS, there exist several versions of the definition
small band-gap energies. This is because the MIGS of
CNL . For example, using a Penn-like model,
picture, which assumes the metallic electronic states Cardona and Christensen (1987) proposed to adopt
in semiconductor layers near the interface, is well the dielectric mid-gap energy as
CNL . However,
applicable to such systems because of the large den- most of these theories present similar values of
sity of MIGSs.
CNL for many semiconductors. This occurs because
f e (eV)
From Tersoff J (1984a) Schottky barrier heights and the continuum of gap states.
Physical Review Letters 52: 465, Tersoff J (1984b) Theory of semiconductor
heterojunctions: The role of quantum dipoles. Physical Review B 30: 4874 and Tersoff J
and Harrison WA (1987) Transition-metal impurities in semiconductors – their connection
with band lineups and Schottky barriers. Physical Review Letters 58: 2367.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 153
–04
x = 0.53 x = 0.25 the simple average of orbital energies of semiconduc-
–08
tors, thus again depending on only the bulk
x=0
properties.
–12 Al O /native oxide
2 3
Ev 1.04.3.2.3 Band bending
1011 1012 1013 1014
(cm–2 eV–1) Most semiconductors contain impurity atoms that
are introduced unintentionally in the growth process
Disordered
(b) semiconductor E
or intentionally by doping. Here, we consider an n-
layer type semiconductor having movable electron carriers
Ec
as an example. Figures 53(a) and 53(b) show the
Anti-bonding band alignments before and after the contact. Before
Semi- III II I contact, in a semiconductor, some of electrons origi-
conductor EHO nating from donor atoms transfer into interface states
Bonding and the positive-impurity sites appear around the
interface. Due to this positive charge distribution,
Er
Insulator or metal Nss the potential (therefore, the conduction and valence
Figure 52 (a) DOS of interface states at disordered
bands) bends as shown in Figure 53(a), which is
interfaces. (b) Schematic diagram explaining the appearance called the band bending. When metal and semicon-
of disorder-induced gap states (DIGSs). Reproduced with ductor are in contact with each other, electron
permission from Hasegawa H and Ohno H (1986) Unified transfer occurs from the interface states of semicon-
disorder induced gap state model for insulator– ductor and inner donor states to the electronic states
semiconductor and metal–semiconductor interfaces. Journal
of Vacuum Science and Technology B 4: 1130.
in metal in case of the alignment in Figure 53(a), to
equalize the Fermi energies of metal and semicon-
ductor, M and S. The alignment after the contact is
they commonly assume the existence of interfaces
states and the charge transfer between metal states (a) (b)
Depletion layer
and interface states. Since such charge transfer is
conceptually similar to that between metallic materi- μS φC
μM μM μS
als, we can say that the full contact of electronic states
is assumed between metals and semiconductors in
these conventional
CNL theories, details of which
will be discussed in the following section. Metal n-Type
The second important feature is that
CNL is semiconductor
defined as intrinsic to not the interface but the bulk Figure 53 Schematic view of band bending at metal/
properties of semiconductors, which is equivalent to semiconductor interface: (a) before and (b) after contact.
154 Atomic Structures and Electronic Properties of Semiconductor Interfaces
shown in Figure 53(b). A more precise band-bending EV ¼ EC þ EGB þ EGA . On the other hand, since
feature is obtained by solving the Poisson equation in the valence-band structures of zinc-blende semicon-
the electromagnetism, considering the charge trans- ductors are well described by the nearest-neighbor
fer self-consistently. The region around the interface tight-binding approximation using sp3 orbital sets,
with no electrons is often called the depletion layer, Harrison and Tersoff (1986) and Monch (1996) pro-
with typical width around 100 Å. posed that the valence-band offset, EV, is obtained
Here, we briefly comment on the experimental as the energy difference of the valence-band top
methods to measure SBHs. Figure 54 shows schema- between A and B; they calculated various EV values
tically the principal views of three methods. In the by adopting empirical parameters. These results are
current–voltage (I–V) method, one applies a voltage shown in Table 5 for various combinations of semi-
perpendicular to the interface and measures the ther- conductors. It is seen that the agreement with
mal electrons that pass the Schottky barrier potential experiments is largely improved by the latter theory.
from semiconductor to metal. Because the current is This is because the affinity is sensitive to the atomic
proportional to expð –
B =kB T Þ, one can determine structures of material surfaces and the atomic and
B from the temperature dependence. In the capaci- electronic structures are generally different for sur-
tance–voltage (C–V) method, one applies the faces and interfaces. On the other hand, though we do
alternating voltage V and measures the capacitance C
not consider the charge transfer at the interface in
of inversion layer.pSince the width of inversion layer is
ffiffiffiffiffiffiffiffiffiffiffiffiffi these theories, the agreement is not so poor, indicat-
proportional to
B – V , the capacitance becomes
ing the quantity intrinsic to bulk semiconductor is an
proportional to 1=
B – V , from which
B is deter-
important factor to determine the offset.
mined. Because the capacitance also includes effects
In these theories, we use the vacuum level as the
of tunneling currents, resistance in semiconductors,
reference energy for comparison. Recently, Van der
and distribution of impurity atoms, the analysis is
Walle and Neugenbauer (2003) proposed the band
sometimes complicated. In the method using internal
alignment assuming that the hydrogen impurity level
photoelectric effect, one produces hot electrons by
has the same energy position in most of the semi-
photoelectric excitation and measures the inverse
conductors. This proposal can be conceptually
current flowing over the barrier. These methods
have an advantage in observing the deeply buried similar to the above-mentioned two theories.
interface. We can also measure the barriers by using To advance a more reliable calculation of the band
XPS and STM when metal layers are thin enough. offset, one has to consider the charge transfer at the
interface (Nakayama, 1993). In this case, the scenario
based on the
CNL theory at metal/semiconductor
1.04.3.2.4 Band offset
interfaces applies straightforwardly to semiconductor/
Next, we consider how the band offset appears at a
semiconductor interfaces. Figure 55 illustrates how the
semiconductor/semiconductor interface. Anderson
band offset is determined at the semiconductor/semi-
(1962) proposed that the conduction band offset at
conductor interface. At the semiconductor interfaces
A/B interface is given by the electron affinity differ-
ence between A and B materials as Ec ¼ B – A .
Using the observed band-gap energies, one can Table 5 Calculated valence-band offset at various
also derive the valence-band offset as semiconductor/semiconductor (110) nonpolar interfaces,
in eV
A B
Semiconductors The results in Table 5 demonstrate that not only
A B
the electronic structures of bulk semiconductors but
εBF also the charge transfer at the interface are important
εAF εBF
to determine the band offset at the interface. It should
εAF
be noted here that the charge transfer across the inter-
face is often realized by the orbital hybridization
between the constituent interface atoms of semicon-
Figure 55 Schematic view to explain the formation of ductors, thus producing the dipole at the interface. To
band offset. examine whether such a dipole is really produced at the
interface by charge transfer, we had better introduce
A and B, there exist MIGS-like interface states and another freedom at the interface and study the variation
these states are occupied by interface electrons from of charge transfer. Figure 56 shows the calculated
the bottom to the effective Fermi energies (left figure). valence-band offset at ZnSe/ZnTe (001) interface as a
When the interface is realized, the charge transfer function of the strain (Nakayama, 1992). The strain can
occurs between these states to equalize such Fermi be varied by changing the lattice constant, that is, the
energies (middle figure). As a result, one obtained the composition, of the substrate for ZnSe/ZnTe system.
band alignment shown in the right. Because by increasing the Te composition, the differ-
Harrison and Tersoff (1988) were the first to ence of bulk energy, ", decreases, it reflects the
adopt this picture and calculate the band offsets for deformation potentials of the bulk system. On the
various combinations of semiconductors; their results other hand, there is an increase in the charge transfer
are shown in Table 5. It is seen that the
CNL theory from ZnTe to ZnSe, , and this increases the dipole
also succeeded in predicting valence- and conduc- potential, V. As a result of compensation between "
tion-band offsets of a number of semiconductor/ and V, the valence-band offset gradually decreases. It
semiconductor interfaces. should be noted that such a correlation between "
: III–V compounds
ΔV (eV)
ΔV Si
: III–VI compounds
Δ & ΔEv,hh (eV)
0.2
1.6 –1 Δρ
SiC
Δ
Charge transfer
Dipole potential
0 2 4 6
0 Ionicity, ε cp – ε ap (eV)
0 0.5 1
LznTe
X=
LZnSe + LZnTe
Figure 56 (a) Strain dependence of valence-band offset at ZnSe/ZnTe (001) interface. Strain changes corresponding
to the composition of the substrate. (b) Band offset at wurtzite/zinc-blende interface for the heavy-hole states of
various semiconductors as a function of ionicity. (a) Reproduced with permission from Nakayama T (1992b) Valence
band offset and electronic structures of zinc-compound strained superlattices. Journal of the Physical Society of Japan
61: 2434. (b) Reproduced with permission from Murayama M and Nakayama T (1994) Chemical trend of band offsets
at wurtzite/zinc-blende heterocrystalline semiconductor interfaces. Physical Review B 49: 4710.
156 Atomic Structures and Electronic Properties of Semiconductor Interfaces
and V indicates the charge transfer expected by the foreign-family atoms at the interface, one can some-
CNL theory really occurs. times control the modulation of the band offset.
At the heterovalent interfaces such as ZnSe/GaAs Lambrecht and Segall (1990) developed the simple
(001), the charge transfer governs the band offset tight-binding model to include such effects in the
itself. Figure 57(a) shows the measured valence- case of covalent bonding systems.
band offset as a function of R ¼ Zn/Se ratio at the More precise values of band offsets are evaluated
interface, which is realized by changing the growth using the first-principles calculations. The results
treatment (Nicolini et al., 1994). Zn–As acceptor and using the local density-functional calculations are
Se–Ga donor bonds exist at this interface. The charge also shown in Table 5. When the quasiparticle GW
transfer of about 101e occurs between these bonds, calculation is applied, one can obtain the most pre-
which produces a dipole potential V. Depending on cise values, for example, 0.5 eV for GaAs/AlAs
the growth treatment, the configuration of these interface (Zhang et al., 1989). These first-principles
bonds changes as shown in Figures 57(b) and 57(c) calculations can consider the effects of any atomic
and the direction of charge transfer is opposite in structures of interface and apply to new systems that
these two cases. As a result, as seen in Figure 57(a), have not ever been produced in experiments. For
the offset varies by about 0.7 eV. example, the heterocrystalline interfaces, which is
The above result for ZnSe/GaAs interface the interface between the same semiconductors
demonstrated that the atomic structures at the inter- with different crystal structures, are proposed and
face are also important to determine the band offset produced in experiments (Murayama and
in some of the interfaces. In fact, by producing the Nakayama, 1993, 1994; Hibino and Ogino, 2000).
heterovalent bonds by intentionally inserting The calculated band offsets at hexagonal/cubic
(a)
1.6
ZnSe/GaAs(001)
band offset
1.2
R = Zn/Se
0.8
Substrate types
0.4 p+ n+
p n
pc (4 × 4) nc (4 × 4)
p 3×1 n 3×1
0 No doping R
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
ΔE (eV)
(b) (c)
Charge transfer
Se
Zn
direction
Ga
As
Small R Large R
Figure 57 (a) Valence band offset, E, at heterovalent ZnSe/GaAs (001) interface, as a function of the composition ratio of
R ¼ Zn/Se. (b,c) Charge transfer at anion and cation mixed interfaces. Reproduced with permission from Nicolini R, Vanzetti L,
Mula G, et al. (1994) Local interface composition and band discontinuities in heterovalent heterostructures. Physical Review
Letters 72: 294.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 157
(wurtzite/zinc-blende) homomaterial twin interfaces biding energies of Au, and found that the binding
are shown in Figure 56(b). This result indicates that energy increases as the size of dots and thus the
not only the heteroatoms but also the topology of interface area increases, as shown in Figure 58(b).
bond connection are the origins of offsets. This result clearly indicates that at Au/HfO2 inter-
face the electrons transfer from Au to HfO2, which
promotes the increase of Au effective WF. Similarly,
1.04.3.3 New Features of Schottky Barrier by using the CV techniques, WF shifts were observed
on HfAlOx substrate for Al metal with a decrease of
1.04.3.3.1 Breakdown of Schottky–Mott
0.36 eV, and for Ni and Au metals with increases of
limit
0.20 and 0.22 eV, respectively.
Koyama et al. (2004) deposited Au, Pt, and Al metals
The change of WFs from in-vacuum to on-high-k-
on HfSiO substrate and measured their effective
dielectrics is schematically summarized by arrows in
WFs, which results are displayed in Figure 58(a).
Figure 58(c), for various metals. The Fermi energies
They found that effective WFs of Au and Pt become
of p-metals such as Ni and Au, which have larger
large compared to WFs in vacuum, opposite to what
WFs, are shifted toward the valence band of high-k
happens with Al This result indicates that the S
dielectrics, whereas those of n-metals, such as Al,
parameter becomes larger than 1, that is, beyond
having smaller WFs move toward the conduction
the conventional limit. A similar anomaly was
band. These WF changes are quite different from
observed in the XPS experiment by the Miyazaki
expectations deduced by the conventional
CNL the-
group (Shiraishi et al., 2005). They deposited Au
nanoscale dots on HfO2 substrate, measured the ories. The conventional theories state that there is a
single
CNL level in the band gap of a high-k dielectric
and the metal Fermi energies are aligned to this
CNL ,
(a) 6 as shown in Figure 51(c). Thus, the variation in
Figure 58(c) was beyond our ordinary understanding.
Effective WF on HisiD (eV)
Ps
Au
5.5 In order to clarify what happens at metal/high-k
interfaces, Shiraishi et al. (2005) investigated the
5 electronic structures using the first-principles cal-
culations, and found that two important
4.5
preconditions implicitly assumed in conventional
Al S > 1 !! theories are broken at these interfaces. The first is
concerned with the penetration length of MIGSs
4
4 4.5 5 5.5 6 from the interface into insulating materials.
Vacuum WF (eV) Figures 59(a) and 59(b) show the charge densities
(b) (c)
CB Interface
Au4f 2 nm (a)
3 nm Al
Intens. (a.u.)
W
Ni
Ru
4 nm Metal Si
Au
EF
HfO2 VB Metal (jellium) HfO2
88 86 84 82 (b)
Binding energy (eV) O
Hf
Figure 58 (a) Effective work functions of Al, Au, and Pt on
HfSiON as a function of vacuum work function, reported by
Koyama et al. (2004). (b) Observed binding energies of Au
nanodots on HfO2 by XPS measurement (Shiraishi et al.,
Interface
2005). The sizes of dots are described in the figure. (c)
Schematic diagram of the observed Fermi-energy Figure 59 Contour plots of MIGS wave functions at (a)
movement of metals on Hf-related high-k materials. metal/Si and (b) metal/HfO2 interfaces (Shiraishi et al., 2005).
158 Atomic Structures and Electronic Properties of Semiconductor Interfaces
Energy (eV)
metal/Si, the MIGS penetrates deep into Si, about 15 EF
five atomic layers. Thus, the Si side of the interface 10
has movable electrons and looks like a metal. The 5
band alignment is realized by moving electrons in 0
these states. Note here that, since the MIGS pene- Γ XK Γ L KWX
trates deep and touches a number of Si atoms
around the interface, the detailed interface atomic (b) Au
structure does not govern the electronic structure of 40
the interface. In other words, one can say that the 35
Energy (eV)
full contact of electronic states is realized at this 30
metal/Si interface. On the other hand, in case of 25
20 DOS
metal/HfO2, the penetration of MIGS into HfO2 is EF
15
seen to be at most one to two atomic layers, thus the 10
full contact of electronic states not being realized at 5
metal/HfO2. This occurs because of the high ioni- Γ XK Γ L KWX
city and large band gap of HfO2. The present result Figure 60 Band structures of (a) Al and (b) Au bulk metals,
indicates that the electronic structure at metal/ calculated by the first-principles method. Schematic
high-k interface is very sensitive to the interface diagrams of DOS are displayed on the right. From Shiraishi
K, Akasaka Y, Miyazaki S, et al. (2005) Universal theory of
atomic structures, that is, to which atom contact is
work functions at metal Hf-based high-k dielectrics
realized, and we had better start with the bonding interfaces – guiding principles for gate metal selection. In:
picture of atoms to understand the electronic Technical Digest of IEEE International Electron Devices
structure at this interface. Meeting, p. 29. Washington, DC, USA, December.
The other precondition is related to the indivi-
duality of metals. Figures 60(a) and 60(b) show
band structures of Al and Au metals around the in the above, it is clear that these conditions are not
respective Fermi energies, EF. The schematic pic- satisfied at metal/high-k interfaces.
tures of DOS are also shown on the right. As the
extended s- and p-orbital electrons are valence elec- 1.04.3.3.2 Generalized charge neutrality
trons, Al has featureless DOS around EF. On the level
other hand, due to the localized d-orbital electrons, In order to simulate the electronic structures of
Au has small-dispersion bands below EF. Thus, DOS metal/high-k dielectrics interfaces, we have to take
is extremely large below the Fermi energy, whereas into account the interface atomic structures and char-
it is comparable to the Al case above the Fermi acteristics of metals. To realize such interfaces,
energy. As shown in the following, this kind of Shiraishi et al. (2005) adopted the effective-four-
metal character is not considered in the conven- level tight-binding model of an interface. The sche-
tional theory. matic diagram of this model is described in
The full contact of electronic states at the inter- Figure 61(a). They characterize a metal with the
face and the featureless metal DOS are essential Fermi energy, EF, and the effective local DOS at
conditions in conventional theories. This is apparent the interface below and above EF, Docc and Dunocc.
because
CNL is defined using quantities intrinsic to The high-k dielectric such as HfO2 is represented by
bulk materials, such as EVB, EG, DVB, and DCB, and the energies of valence-band top and conduction-
does not include the interface information, such as band bottom (EVB and ECB, respectively) and the
atomic structures and characteristics of metals. The effective local DOS of valence and conduction
same conditions are also assumed in the case of the bands at the interface (respectively, DVB and DCB).
theory of DIGSs (Hasegawa and Ohno, 1986); for Since HfO2 is an ionic material, the conduction bands
example, the interface is assumed to have random are mainly made of Hf d-orbitals, while the valence
amorphous-like structures, and some sort of averages bands are made of O p-orbitals. tM-Hf is a transfer
that realize the full contact of electronic states are energy between the occupied metal states and con-
implicitly taken to deduce
CNL . However, as shown duction-band states of HfO2, while tM-O is that
Atomic Structures and Electronic Properties of Semiconductor Interfaces 159
φGCNL
EVB By solving this equation, the generalized charge neu-
s. p. d Docc trality level of HfO2 that should match the Fermi
tM–O DVB
O 2p energy of metal is obtained by
GCNL ¼ EF – V as
ð23Þ
s. p Hf 5d Hf 5d
s. p where EG ¼ ECB EVB is the band-gap energy of
φGCNL φGCNL HfO2.
EF
d EF In the case of ordinary metal/semiconductor inter-
s. p O 2p O 2p faces, we can expect the full contact of electronic
states, tM-Hf ¼ tM-O, and the DOS of the metal is fea-
Al HfO2 Au, Ni HfO2 tureless, Docc ¼ Dunocc. Thus, we can reproduce the
conventional
CNL as EVB þ EG DVB =ðDVB þ DCB )
Figure 61 (a) Schematic diagrams of interface
hybridization models to derive the generalized neutrality (Cardona and Christensen, 1987), which is the
levels at metal/nonmetal interfaces. EF, Doccu., and quantity intrinsic to a bulk semiconductor. This is
Dunoccu are Fermi energy and DOS variables for the metal why we call the present
G CNL
the generalized charge
below and above EF, while ECB, EVB, DCB, and DVB are neutrality level.
the conduction- and valence-band edges and their DOS
Next, we explain how the new
G theory
for the nonmetal. tM-Hf and tM-O are orbital hybridization CNL
energies between electron occupied and unoccupied explains the unusual behavior of work functions at
states, which promote charge transfer across the metal/high-k interfaces. As representative interfaces
interface. (b,c) Hybridization-induced charge transfer at made of n and p metals, the Al/HfO2(110) and Ni/
Au/HfO2 and Al/HfO2 interfaces. Bold arrows denote HfO2(110) interfaces were investigated by the first-
major charge transfer.
principles calculations. Figures 62(a) and 62(b),
respectively, display the most stable adsorption posi-
tions of Al and Ni atoms on the HfO2 substrate, while
between the unoccupied metal states and valence- Figure 62(c) shows the adsorption energies of Al and
band states of HfO2. It should be noted here that the Au on HfO2 as a function of the adsorption position
charge transfer between a metal and HfO2 is gener- (Nakayama et al., 2006b). Adsorption positions, 1–7,
ally realized only by such orbital hybridization are denoted on the right. It is clearly seen that Al
between unoccupied and occupied states. Transfer atoms prefer to locate on oxygen atoms and produce
energies between occupied states and those between the connection only with oxygen atoms. This occurs
unoccupied states never induce the charge transfer because the reactivity of Al with oxygen is high. As a
between the metal and HfO2, thus being not relevant result, one can reasonably approximate tM-Hf << tM-O.
to determine the band alignment and being excluded Meanwhile, the variation of Al DOS is monotonously
in the present model. continuous and featureless as shown in Figure 61(b),
By applying the second-order perturbation theory indicating that Docc ¼ Dunocc. Therefore, the charge
of quantum mechanics, the charge transfer from a transfer occurs from O to Al, as shown in
metal to HfO2 is written as (Nakayama, 1993; Figure 61(b). As a result,
G approaches the bottom
CNL
Nakayama et al., 2006b) of HfO2 conduction bands, ECB, and the effective WF
of Al decreases as shown in Figure 58(c). With
jtM-Hf j2 Docc DCB jtM-O j2 Dunocc DVB
_ – ð21Þ respect to Au/HfO2, the Au atom appears to have
ECB – EF EF – EVB
no preference for the adsorption position and is
This charge transfer produces the dipole potential at expected to locate between Hf and O atoms on aver-
the interface and increases the energies of EVB and age on the HfO2 surface, thus indicating tM-Hf ¼ tM-O.
ECB in HfO2. Since the charge transfer should be However, as seen in Figure 60(b), the Au DOS
completed by inducing a final dipole potential of below the Fermi energy is much larger than that
V, the self-consistent equation of V becomes as above the Fermi energy. Thus, we can approximate
160 Atomic Structures and Electronic Properties of Semiconductor Interfaces
(a) (b) 38
3.56Å 1.89Å 1.95Å
2.64Å 1.97Å
36
CB
Energy (eV)
2.84Å
34
Al
32 Au
Al-adsorbed structure Ni-adsorbed structure VB
30
(c) HfO2
1 2 3 4 5 6 7 ∞
Adsorption energy [eV]
the interface, which all are defined as in Figure 61(a). insulator (band gap is about 5.6 eV), SBH behavior is
This equation represents the balance of two charge expected to be similar to the Schottky limit according
neutrality levels (Nakayama, 1993). to the conventional CNL concepts. However, the
Finally, we remark on the universality of the new observed behavior rather resembles the Bardeen limit.
GCNL
concept.
GCNL
is constructed by using the repre- The physical origin of this unusual behavior is in the
sentative physical quantities of the interface. The mechanism of the SBH formation being much different
energies EF, EVB, and ECB and DOS values Docc, from the conventional mechanism. This is governed by
Dunocc, DVB, and DCB represent properties intrinsic the thermodynamics of the interface reaction.
to bulk materials originating from bulk band struc- The physical mechanism is as follows. The rela-
tures, while the transfer energies, tM-Hf and tM-O, are tively higher energy level of oxygen vacancy (Vo) in
introduced to describe the microscopic atomic struc- HfO2 causes a notable thermodynamic behavior of
tures of interfaces, that is, the interface bonds. This interfaces, when HfO2 is in contact with Si. Recent
fact indicates that both band and bond pictures, experiments indicate that the Vo level is located
which, respectively, correspond to itinerant and loca- about 0.4 eV above the bottom of Si conduction
lized characters of electrons, are necessary to band (about 1.2 eV below the bottom of HfO2 con-
describe electronic structures at the interface. In duction band) (Takeuchi et al., 2004).
this way, the first-principles theoretical approaches It is well known that Hf atoms can bind much
have played key roles in creating the new science more strongly than Si atoms to O atoms. Actually, the
concepts such as
G in recent Si nanotechnology, formation enthalpy of HfO2 is larger than that of
CNL
and they are further leading not only the Si nano- SiO2 by about 2.2 eV. Further more, recent first-
technology but also the frontier fields of nanoscience principles calculations show that the Vo formation
such as organic semiconductors and nano-bio devices energy in bulk HfO2 is larger than that in bulk SiO2
(Oda and Nakayama, 2008). by about 1.2 eV (Scopel et al., 2004). At first glance,
the fact that Hf–O bonds are stronger than SiO
bonds indicates that the partial oxidation of poly-Si
1.04.3.3.3 Interface reaction and gates by pulling an O atom out of the HfO2 dielec-
Fermi-level pinning trics is an endothermic reaction with 1.2 eV energy
We first introduce an example where the thermody- loss, and this reaction occurs with difficulty.
namics of interface reaction determine the SBH. An However, the situation changes completely if we
unexpected SBH behavior at the interface between take into account the electron behavior.
heavily B doped Si (pþpoly-Si) and HfO2. has been FLP in Hf-related high-k gate stacks with pþpoly Si
reported. Hobbs et al. (2003) reported that the differ- gates can naturally be explained by taking into account
ence in SBH between nþpoly-Si and pþpoly-Si the electron behavior as follows. In Figures 65(a)–
becomes only 0.2 eV, although their intrinsic Fermi- 65(c), the mechanism of Vo formation in HfO2 and
level difference amount to 1 eV reflecting the Si band subsequent electron transfer across the poly-Si/HfO2
gap (Figure 64). This is called ‘Fermi level pinning’ interface is schematically illustrated. First, let us assume
(FLP) in LSI jargon. Since HfO2 is a typical wide-gap that the poly-Si is partially oxidized by the formation of
SiOSi bonds by O atoms being pulled out of HfO2.
Si conduction
As a result, SiOSi bonds in the poly-Si and Vos in
band HfO2 are formed (Figure 65(a)). An O atom in HfO2
0.2 eV
Ferml level 0.2 eV
takes an O2 ion form, but an O atom in a SiOSi
of n + poly Si bond is neutral. Accordingly, two additional electrons
Ferml level are generated after one O atom is pulled out, and if
0.6 eV
of p + poly Si these electrons remain inside HfO2, they occupy the Vo
Si valence level in HfO2 (Figure 65(b)). The assumption that two
band additional electrons remain in HfO2 corresponds to the
On SiO2 On HfO2 same situation as the bulk calculations that give a
1.2 eV energy loss. However, since HfO2 is in contact
Figure 64 Schematic illustration of the unusual behaviors
with the poly-Si gate, electrons have to transfer into the
of the relative Schottky barrier heights (SBHs) at the pþpoly-
Si/HfO2 interface compared with that at the nþpoly-Si/HfO2 gate. This is because the Vo level is located above the
interface. For comparison, the well-known SBH behavior at poly-Si Fermi level (Figure 65(c)). Now, we estimate
the poly-Si/SiO2 interfaces is also shown on the left. roughly the two ultimate cases of energy loss (gain)
162 Atomic Structures and Electronic Properties of Semiconductor Interfaces
when HfO2 is in contact with the poly-Si gate: one is completely corroborate the above discussion based on
with an nþpoly-Si gate and the other is with a pþpoly- the ‘oxygen vacancy model’.
Si gate. For an nþpoly-Si gate, the two-electron trans- Now, we move on the Vfb shift originating from
fer results in an energy gain of 0.8 eV (2 0.4 eV), and the formation of Vo in HfO2. For an nþpoly-Si gate,
the total energy loss is reduced from 1.2 eV (bulk value) the interface reaction that induces the electron trans-
to 0.4 eV. Despite the energy reduction due to the fer occurs with difficulty, since it is endothermic. For
electron transfer, the reaction for an nþpoly-Si gate is a pþpoly-Si gate, however, the interface reaction
still endothermic. For a pþpoly-Si gate, on the other accompanied by the formation of Vo in HfO2 occurs
hand, the situation is quite different. The Fermi level easily, since this reaction has a large energy gain of
position of a pþpoly-Si gate is located about 1.5 eV 1.8 eV. At the same time, electrons transfer from
below the Vo level in HfO2. As a result, the two- HfO2 into the poly Si occurs. As a result, an interface
electron transfer results in a total energy gain of dipole is formed as illustrated in Figure 67. This
1.8 eV (2 1.5 eV 1.2 eV). Surprisingly, the interface dipole formation raises the position of the Fermi
reaction, accompanied by Vo formation and subsequent level of the pþpoly-Si gate and flat band voltage
electron transfer, becomes exothermic with an energy (Vfb) decreases. It is noticeable that the energy gain
gain of 1.8 eV, when the HfO2 is in contact with a of the interface reaction decreases when the Fermi
pþpoly-Si gate. Actually, TEM observations show level is elevated. A simple consideration indicates
that some interfacial reaction layers are observed in a that the position at which the Fermi level is pinned
pþpoly-Si gate HfAlOx MISFET, as shown in corresponds to the energy level that makes the
Figure 66, and such reaction layers have not been energy gain of the interfacial reaction zero. The
observed in nþpoly-Si gate MISFETs. These results final Fermi level position satisfies the equation 1.2
(0.4 þ x) ¼ 0, where x is the final pinning position
measured from the nþpoly-Si Fermi level. The
Fermi level position of a pþpoly-Si gate obtained
p + poly -Si HfAIOx from this is about 0.2 eV below the nþpoly-Si gate
Fermi level, which is in fairly good agreement with
the experimental results (Hobbs et al., 2003). As
discussed above, ‘oxygen vacancy model’ can quanti-
tatively reproduce the Fermi level pining position of
IL pþpoly-Si gate/HfO2 interfaces.
Si substrate
20 nm Next, we comment on the effect of inserting cap
insulator between pþpoly-Si and HfO2 gate dielec-
Figure 66 Cross section of replacemnt pþgate HfAlOx trics. Recent experiments show that the FLP of
MISFET observed by TEM. Reproduced with permission
pþpoly-Si gates cannot essentially be improved by
from Shiraishi K, Yamada K, Torii K, et al. (2006) Oxygen-
vacancy-induced threshold voltage shifts in Hf-related high- inserting SiO2 or SiN cap layers between pþpoly-Si
k gate stacks. Thin Solid Films 508: 305. gates and high-k Hf-related dielectrics (Cartier et al.,
Atomic Structures and Electronic Properties of Semiconductor Interfaces 163
(a) (b)
Si CB
1.1 eV Vo EC
0.4 eV 0.4 eV
Si VB
(EF) x eV
0.2 eV – –
EV
1.1–x eV
Interface Pinning level
dipole
Poly-Si gate HfO2
Cap Vo
Poly-Si layer
gate Vo HfO2 +
SiO2
+
SiO2
Figure 67 (a) Schematic illustration of interface dipole formation and subsequent Fermi-level elevation toward the pinning level.
(b) Schematic illustration of the cap layer effect for Fermi-level pinning (FLP). Reproduced with permission from Shiraishi K, Yamada
K, Torii K, et al. (2006) Oxygen-vacancy-induced threshold voltage shifts in Hf-related high-k gate stacks. Thin Solid Films 508: 305.
2004). The oxygen vacancy model naturally explains IL, as illustrated in Figure 68(c). Their experimental
the results of these experiments. The schematic illus- finding indicates that interaction between HfO2 and a
trations are shown in Figure 67. As shown in this Si substrate is very important for FLP as well as that
figure, the pinning position is governed only by the between poly-Si gate and HfO2. In other words, FLP
energy position that balances the Vo formation disappears only when the interaction between Si and
energy loss and the electron transfer energy gain. HfO2 is weak enough. Further, it is expected that
The final pinning position measured from the indirect interaction between Si and HfO2 can induce
nþpoly-Si Fermi level (x) satisfies the equation 1.2̃ FLP. In short, the thermal equilibrium between Vo
(0.4 þ x) ¼ 0, which is the same equation without a formation and annihilation reaction at the Si/HfO2
cap layer. Accordingly, if O atoms can penetrate interface given by the following equation determines
through the cap layer until the system reaches ther- the SBH (FLP position) at pþpoly-Si/HfO2 inter-
mal equilibrium, the final pinning position remains face, as shown in Figure 69:
the same, regardless of the existence of a cap layer.
ðHfO2 Þ þ Si ! SiO2 þ HfO2 þ Vo 2þ þ 2e ð25Þ
Systematic experiments have been reported by
Kamimuta et al. (2005). They examined three geo-
metries of poly-Si/HfO2 gate stacks. The schematic
illustrations are shown in Figure 68. The usual gate 1.04.3.3.4 Correlation between interfaces
stack structures in which FLP is observed in pþpoly- It has been reported that SBHs of p-metals decrease
Si gates are structures with no barrier layer and thin remarkably, revealing FLP behavior when IL is thin,
IL as shown in Figure 68(a). At first, the effect of after high-temperature annealing. However, FLP does
boron segregated near the Si/HfO2 interfaces not appear when IL is thick enough (Lee et al., 2006).
(Takayanagi et al., 2003) or interfacial Hf–Si bonds This is called ‘Vfb roll-off’. The noticeable fact is that
(Hobbs et al., 2003) were thought to be the cause of the energy position of FLP is similar to the pinning
FLP of p-poly-Si gates. However, FLP cannot be position of pþpoly-Si gates (Hobbs et al., 2003), as
avoided even if a barrier layer is inserted between a mentioned in previous section. This experimental
pþpoly-Si gate and HfO2, as described in fact indicates that FLP of p-metal is also governed
Figure 68(b), indicating that neither B effect nor by a mechanism similar to FLP of pþpoly-Si gates
Hf–Si bonds can be the cause of FLP at pþpoly-Si/ mentioned above (Akasaka et al., 2006).
HfO2 interfaces. They have found that FLP relaxa- Now, we consider the mechanism of FLP of
tion can be achieved only when the gate stack p-metal gates subjected to a high-temperature
structure contains both thick barrier layer and thick treatment. Since usual p-metals are nonreactive
164 Atomic Structures and Electronic Properties of Semiconductor Interfaces
e– transfer O transfer
(a)
Occur Occur
FLP 2+ Poly-Si
Vo
Occur
Very small
p+ poly-Si2 HfO Si sub.
(b)
Not Occur
occur Poly-Si
FLP Vo2+ with
barrier layer
Occur
Very small
p+ poly Barrier HfO2 Si sub.
(c)
Not Not Poly-Si
occur occur with barrier
FLP
relaxation Vo2+ layer and
thick IL
Occur
Very small
p+ poly Barrier HfO2 Si sub.
Figure 68 Schematic illustration of three poly-Si/high-k gate stack structures examined by Kamimuta et al. (2005). (a) Usual
pþpoly-Si gate stack structure. (b) Gate stack structure with barrier layer between a pþpoly-Si gate and a high-k dielectric. (c)
Gate stack structure in which barrier layers are inserted both between a pþpoly-Si gate and a high-k dielectric and between a Si
substrate and a high-k dielectric. Reproduced with permission from Akasaka Y, Nakamura G, Shiraishi K, et al. (2006) Modified
oxygen vacancy induced fermi level pinning model extendable to P-metal pinning. Japanese Journal of Applied Physics 45: L1289.
(a) (b)
Electron Interface
transfer dipole
EC EC
E(Vo) E(Vo)
EV + + E(p+) EV E(p+)
p+poly- HfO2 p+poly- HfO2
Si-gate Si-gate
Interface
dipole Thermal Electron
SiO2 equilibrium SiO2 transfer
Vo Vo annihilation
O transfer – – Vo
EV(HfO2) – –
materials, the situation of p-metal/high-k gate dielectrics is inhibited when IL is thick enough
stack can be schematically illustrated as in (Figure 70(b)). As discussed in Figure 68, the
Figure 70. O transfer hardly occurs from high-k interface reaction between high-k dielectric and
dielectrics to p-metals due to low reactivity of Si substrate still occurs when IL is thin. Further-
p-metals. However, O transfer is still possible if IL more, reaction with the Si substrate induces
is thin (Figure 70(a)). Transfer of O from high-k electron transfer from Vos to p-metal gates. A
Atomic Structures and Electronic Properties of Semiconductor Interfaces 165
Not
Occur p-Metal
occur
FLP Vo 2+ with thin
IL
Occur
Very small
p-metal HfO2 Si sub.
(b)
Not Occur
FLP occur p-Metal
relaxation Vo2+ with thick
Occur IL
Very small
p-metal HfO2 Si sub.
Figure 70 Schematic illustration of two typical p-metal/high-k gate stack structures. (a) A p-metal/high-k gate stack
structure with thin IL. (b) A p-metal/high-k gate stack structure with thick IL. Reproduced with permission from Akasaka Y,
Nakamura G, Shiraishi K, et al. (2006) Modified oxygen vacancy induced fermi level pinning model extendable to P-metal
pinning. Japanese Journal of Applied Physics 45: L1289.
schematic illustration of this situation is given in Figure 71, the pinning position corresponds to the
Figure 71. It is noticeable that the net reaction energy at which the energy loss (G1) and energy
between high-k dielectrics and the Si substrates is gain (G2) by electron transfer from Vo in high-k to a
the same as that between poly-Si gates and high-k gate metal are canceled by each other (i.e., G1
dielectrics, although electron transfer and O trans- G2 ¼ 0). This means that the FLP positions of
fer directions are opposite to each other. In fact, p-metal gates are the same as that of the pþpoly-
the reaction equation with the Si substrate can be Si gate, irrespective of the metal species. In fact,
described as our observed EWFs of p-metals are almost inde-
pendent of metal species, as shown by the C–V
ðHfO2 Þ þ Si ! SiO2 þ HfO2 þ Vo 2þ þ 2e ð26Þ
curves in Figure 72.
which is the same as in the case of pþpoly-Si gates It is better to note that the Vo-related mechanism
described in the previous subsection. According to is not the only cause of Vfb shift. The Vo-related
(a)
Vo
EF Semiconductor
HfO2 (Si)
Metal
Reaction with Si sub. – G1
O SiO2
(b)
Vo
EF elevation Energy gain by Semiconductor
EF
electron transfer G2 (Si)
Metal
Reaction with Si sub. – G1
HfO2
SiO2
Figure 71 Schematic illustration of interface reaction with the Si substrate and subsequent electron transfer from Vo to gate
metals in p-metal/high-k gate stacks. (a) O transfer into the Si substrate through thin IL. (b) Subsequent electron transfer from
a Vo level to a p-metal gate which induces gate Fermi-level elevation.
166 Atomic Structures and Electronic Properties of Semiconductor Interfaces
WF(Ru) = 4.7 eV
FLP can be avoided if the reaction between high-k
WF(TiN) = 4.72 eV
WF(Ir) = 4.63 eV dielectrics and the Si substrate is suppressed. In order
2×10–6 to suppress the corresponding reaction, there are two
possibilities. One is insertion of thick IL, and the other
C (F cm–2)
(a) Vo
Semiconductor
EF elevation Energy gain by
EF (Si)
electron transfer G2
Metal
Reaction with Si sub. – G1
HfSiON
(b)
Vo
EF elevation O injection
EF HfSiON by ozone at RT
Metal
Figure 73 Schematic illustration of the experimental procedure that confirms our Vo model. (a) FLP occurs after a high-
temperature treatment. (b) After the removal of Si substrates, ozone is injected into Hf-based high-k dielectrics at room temperature.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 167
Oxygen
chemical Reducing process Oxidizing process
potential
CeO2 Ce2O3 CeO2 Ce2O3
CeO2 Ce2O3 CeO2 Ce2O3
Release Absorb
Vo2+ Vo2+ O
O O 2– O 2–
High-k i High-k i
μo
Stable flat and voltage μo is kept constant
with Ce-oxide capping
o –2e → O
I2– –
V2–
o +O
Figure 75 Schematic illustrations of the recipe for fixing the chemical potential of oxygen during reduction and oxidation
processes by using a capping layer of multivalent oxides. In the reduction process, CeO2 supplies O atoms into the oxide
layer. On the other hand, Ce2O3 absorbs O atoms from the oxide layer in oxidizing ambient. From Kouda M, Umezawa N,
Kakushima K, et al. (2009) Charged defects reduction in gate insulator with multivalent materials. In: Digest of Technical, 2009
Symposium of VLSI Technology, p. 200. Kyoto, Japan, 18 June 2009.
40
limited number of atomic bonds. The molecule is
sometimes a zero- or one-dimensional system and
20
has a small number of freedoms, while the electrode
0 is a two- or three-dimensional system and has a
120
larger number of freedoms. Therefore, the contact
e (K)
140
160
between molecule and electrodes is also a typical
eratur
Electron number
experiments. At each temperature, the gate voltage was
swept with a 10 mV step between 4 and 4 V, 10 v = 0.2
corresponding to the effective gate voltage between 2 and Electrodes
6 V, at a sweep rate of 63 mV s1. Reproduced with 0.1
permission from Sakurai Y, Iwata J, Muraguchi M, et al.
(2010) Temperature dependence of electron tunneling 0
between two dimensional electron gas and Si quantum
dots. Japanese Journal of Applied Physics 49(1): 014001.
v = 0.2
assumption that sufficiently wave-packet-like wave 0.1
functions, which satisfy ‘geometrical matching’ 0
between different-dimensional systems, can contri- 0 20 40 60
bute to electron tunneling of these systems. Time, t (a.u.)
Other interest is concerned with the dynamics at Figure 77 Transient current behavior at nanocontact
the interface made of two different-dimensional sys- molecular bridge (quantum point contact) system.
tems because these systems have different degree of Reproduced with permission from Tomita Y, Ishii H, and
Nakayama T (2009) Transient current behavior through
freedom. As discussed in other chapters in this pub-
molecular bridge systems: Effects of intra-molecule current
lication, electronic transport through nanoscale on quantum relaxation and oscillation. e-Journal of Surface
systems, such as molecular bridges, semiconductor Science and Nanotechnology 7: 606.
170 Atomic Structures and Electronic Properties of Semiconductor Interfaces
systems. This occurs because the energy and entropy Bell GR, Belk JG, McConville CF, and Jones TS (1999) Species
intermixing and phase transitions on the reconstructed (001)
have the general tendency to move from small-free- surfaces of GaAs and InAs. Physical Review B 59: 2947.
dom system to large-freedom system, which Caldeira AO and Leggett AJ (1985) Influence of damping on
phenomena are often called the friction and the loss quantum interference: An exactly soluble model. Physical
Review A 31: 1059.
of information in physics. In this way, such interfaces Car R, Kelly PJ, Oshiyama A, and Pantelides ST (1984)
provide a new stage for studying quantum friction Microscopic theory of atomic diffusion mechanism in silicon.
(Ishii et al., 2008; Tomita et al., 2009), in addition to Physical Review Letters 52: 1814.
Cardona M and Christensen NE (1987) Acoustic deformation
the conventional representative friction systems potentials and heterostructure band offsets in
(Caldeira and Leggett, 1985). On the other hand, we semiconductors. Physical Review B 35: 6182.
can also see the oscillation of current in Figure 77. Cartier E, Narayanan V, Gusev EP, et al. (2004) Systematic
study of pFET Vt with Hf-based gate stacks with poly-Si and
This occurs due to the quantum motion of electrons FUSI gates. In: Technical Digest of 2004 Symposium on VLSI
between the molecule and electrode, the period of Technologies, pp. 44–45. Honolulu, Hawaii, USA, 15–17
which reflects the difference of Fermi energy in the June.
Chadi DJ (1996) Core structure of thermal donors in silicon.
electrode and the energy level in the molecule. In Physical Review Letters 77: 861.
other words, we can observe the electronic properties Cowley AM and Sze SZ (1965) Surface states and barrier height
of the electrode in the current through the molecule. of metal–semiconductor systems. Journal of Applied
Physics 36: 3212.
In this way, not only the electronic properties of the Deal BE (1988) The Physics and Chemistry of SiO2 and the
molecule (or QPC or QD) but also the interaction Si–SiO2 Interface (Helms CR and Deal BE, eds.), p. 5.
between the molecule and the electrode through the New York: Plenum.
Deal BE and Grove AS (1965) General relationship for the
interface becomes important in the dynamics of thermal oxidation of silicon. Journal of Applied Physics
nanoscale objects. Another example of such quantum 36: 3770.
friction is seen in the attenuation behavior of micro- Drummond TJ (1999) Schottky barriers on GaAs: Screened
pinning at defect levels. Physical Review B 59: 8182.
scopic molecular vibration (Shigeno and Nakayama, Farrel HH, Harbison JP, and Peterson LD (1987) Molecular-
2007). (See Chapter 1.01). beam epitaxy growth mechanisms on GaAs(100) surfaces.
Journal of Vacuum Science and Technology B 5: 1482.
Finch RH, Qucissor HJ, Thomas G, and Washburn J (1963)
Structure and origin of stacking faults in epitaxial silicon.
References Journal of Applied Physics 34: 406.
Fowler RH (1936) Statistical Mechanics. New York: Cambridge
Adeimann C, Braut J, Jalabert D, et al. (2002) Dynamically University Press (chapter 3).
stable gallium surface coverages during plasma-assisted Fowler RH and Guggenheim EA (1939) Statistical
molecular-beam epitaxy of (0001) GaN. Journal of Applied Thermodynamics. New York: Cambridge University Press
Physics 91: 9638. (chapter 3).
Akagi K, Tsuneyuki S, Yamashita Y, Hamaguchi K, and Fowler RH and Sterne TE (1932) Statical mechanics with
Yoshinobu J, (2004) Structural and chemical property of particular reference to the vapor pressures and entropies of
unsaturated cyclic-hydrocarbon molecules regularly crystals. Reviews of Modern Physics 4: 635.
chemisorbed on Si(001) surface. Applied Surface Science Frisch MJ, Trucks GW, Schlegel HB, et al. (1998) Gaussian 98,
234: 162. Revision A. 9. Pittsburgh, PA: Gaussian.
Akasaka Y, Nakamura G, Shiraishi K, et al. (2006) Modified Fukatsu S, Takahashi T, Itoh KM, et al. (2003) Effect of the Si/
oxygen vacancy induced Fermi level pinning model SiO2 interface on self-diffusion of Si insemiconductor-grade
extendable to P-metal pinning. Japanese Journal of Applied SiO2. Applied Physics Letters 83: 3897.
Physics 45: L1289. Gibson EM, Foxon CT, Zhang J, and Joyce BA (1990) Gallium
Al-Brithen HAH, Yang R, Haider RB, et al. (2005) Scanning desorption from GaAs and (Al,Ga)As during molecular beam
tunneling microscopy and surface simulation of zinc-blende epitaxy growth at high temperatures. Applied Physics Letters
GaN(001) intrinsic 4 reconstruction: Linear gallium 57: 1203.
tetramers? Physical Review Letters 95: 146102. Gibson JM and Lanzerotti MY (1989) Observation of interfacial
Anderson RL (1962) Experiments on Ge–GaAs heterojunctions. atomic steps during silicon oxidation. Nature (London) 340: 128.
Solid State Electronics 5: 341. Glasstone S, Laidler KJ, and Eyring HE (1964) The Theory of
Aspnes DE and Studna AA (1985) Anisotropies in the above- Rate Processes. McGraw-Hill.
band-gap optical spectra of cubic semiconductors. Physical Hamman DR, Schluter M, and Chiang C (1979) Norm-conserving
Review Letters 54: 1956. pseudopotentials. Physical Review Letters 43: 1494.
Austing DG, Honda T, and Tarucha S (1996) Single electron Han CJ and Helms CR (1988) 18O tracer study of Si oxidation in
charging and single electron tunneling in sub-micron dry O2 using SIMS. Journal of the Electrochemical Society
AlGaAs/GaAs double barrier transistor structures. Solid- 135: 1824.
State Electronics 40: 237. Harrison WA and Tersoff J (1986) Tight-binding theory of
Bardeen WJ (1947) Surface states and rectification at a metal heterojunction band lineups and interface dipoles. Journal of
semi conductor contact. Physical Review 71: 717. Vacuum Science and Technology B 4: 1068.
Belk JG, Sudijono JL, Zhang XM, Neave JH, Jones TS, and Hasegawa H and Ohno H (1986) Unified disorder induced gap
Joyce BA (1997) Surface contrast in two dimensionally state model for insulator–semiconductor and metal–
nucleated misfit dislocations in InAs/GaAs(110) semiconductor interfaces. Journal of Vacuum Science and
heteroepitaxy. Physical Review Letters 78: 475. Technology B 4: 1130.
Atomic Structures and Electronic Properties of Semiconductor Interfaces 171
Heine V (1965) Theory of surface states. Physical Review A Kageshima H, Uematsu M, Akiyama T, and Ito T (2007)
138: 1689. Microscopic mechanism of silicon thermal oxidation
Hibino H and Ogino T (2000) Si twinning superlattice: Growth Of process. ECS Transactions 6: 449.
new single crystal Si. Surface Review Letters 7: 631. Kahng D and Atalla MM (1960) Silicon-silicon dioxide field
Hingerl K, Aspnes DE, Kamiya I, and Florez LT (1993) induced surface devices, presented at the IRE-AIEE Solid-
Relationship among reflectance-difference spectroscopy, State Device Research Conference. Pittsburgh: Carnegie
surface photoabsorption, and spectroellipsometry. Applied Institute of Technology.
Physics Letters 63: 885. Kajita S, Minato T, Kato HS, Kawai M, and Nakayama T (2007)
Hiraki A (1980) A model on the mechanism of room temperature First-principles calculations of hydrogen diffusion on rutile
interfacial intermixing reaction in various metal– TiO2(110) surfaces. Journal of Chemical Physics 127: 104709.
semiconductor couples: What triggers the reaction? Journal Kajita S, Nakayama T, and Kawai M (2007) Ab-initio calculation
of the Electrochemical Society 127: 2662. method of electronic structures of charged surfaces using
Hobbs C, Fonseca L, Dhandapani D, et al. (2003) Fermi level repeated slab and density-variable charge sheets. Journal of
pinning at the polySi/metal oxide interface. In: Technical the Physical Society of Japan 76: 044701.
Digest of 2003 Symposium on VLSI Technology, p. 9. Kyoto, Kamimuta Y, Koyama M, Ino T, et al. (2005) Extended Abstracts
Japan, June. of the 2005 Conference on Solid State Device and Materials,
Honing RE and Karmer DA (1969) Vapour pressure data for the p. 24. Kobe, Japan, September.
solid and liquid elements. RCA Reviews 30: 285. Kangawa Y, Akiyama T, Ito T, Shiraishi K, and Kakimoto K
Hu SM (1975) Anomalous temperature effect of oxidation (2009) Theoretical approach to structural stability of GaN:
stacking faults in silicon. Applied Physics Letters 27: 165. How to grow cubic GaN. Journal of Crystal Growth
Ihm J, Zunger A, and Cohen ML (1979) Momentum-space 311: 3106.
formalism for the total energy of solids. Journal of Physics C: Kangawa Y, Ito T, Hiraoka YS, Taguchi A, Shiraishi A, and
Solid State Physics 12: 4409. Ohachi T (2002) Theoretical approach to influence of As2
Ikarashi N, Watanabe K, and Miyamoto Y (2000) High-resolution pressure on GaAs growth kinetics. Surface Science 507–
transmission electron microscopy of an atomic structure at a 510: 285.
Si(001) oxidation front. Physical Review B 62: 15989. Kangawa Y, Ito T, Taguchi A, Shiraishi K, and Ohachi T (2001) A
Ikeda M, Kresse G, Kadoshima M, Nabatame T, Satake H, and new theoretical approach to adsorption–desorption
Toriumi A (2006) Extended Abstracts of the 2006 Conference behavior of Ga on GaAs surfaces. Surface Science 493: 178.
on Solid State Device and Materials, p. 222. Yokohama, Kangawa Y, Matsuo Y, Akiyama T, Ito T, Shiraishi K, and
Japan, September. Kakimoto K (2007) Theoretical approach to initial growth
Inoue M, et al. (2005) Fluorine incorporation into HfSiON kinetics of GaN on GaN(001). Journal of Crystal Growth
dielectric for Vth control and its impact on reliability for poly- 300: 62.
Si gate pFET. In: Technical Digest of IEEE International Kanisawa K, Butcher MJ, Tokura Y, Yamaguchi H, and
Electron Devices Meeting, p. 413. Washington, DC, USA, Hirayama Y (2001) Local density of states in zero-
December. dimensional semiconductor structures. Physical Review
Ishii H, Sugiyama K, Ito E, and Seki K (1999) Energy level Letters 87: 196804.
alignment and interfacial electronic structures at organic/metal Kassel LS (1936) The calculation of thermodynamic functions
and organic/organic interfaces. Advanced Materials 11: 605. from spectroscopic data. Chemical Reviews 18: 277.
Ishii H, Tomita Y, Shigeno Y, and Nakayama T (2008) Relaxation Kato K, Uda T, and Terakura K (1998) Backbond oxidation of the
process of transient current through nanoscale systems: Si(001) surface: Narrow channel of barrier less oxidation.
Density matrix calculations. e-Journal of Surface Science Physical Review Letters 80: 2000.
and Nanotechnology 6: 213. Kita T, Wada O, Nakayama T, and Murayama M (2002) Optical
Itoh M, Bell GR, Avery AR, Jones TS, Joyce BA, and reflectance study of the wetting layers in (In,Ga) As self-
Vvedensky DD (1998) Island nucleation and growth on assembled quantum dot growth on GaAs(001). Physical
reconstructed GaAs(001) surfaces. Physical Review Letters Review B 66: 195312.
81: 633. Kobayashi R and Nakayama T (2004) Theoretical study on
Ito T (1995) Recent progress in computer-aided materials generation and atomic structures of stacking- fault
design for compound semiconductors. Journal of Applied tetrahedra in Si film growth. Thin Solid Films 464/465: 90.
Physics 77: 4845. Kobayashi R and Nakayama T (2008) Atomic and electronic
Ito T and Shiraishi K (1998a) Theoretical investigations of structures of stair-rod dislocations in Si and GaAs. Japanese
adsorption behavior on GaAs(001) surfaces. Japanese Journal of Applied Physics 47: 4417.
Journal of Applied Physics 37: L262. Kojima T, Kawai NJ, Nakagawa T, Ohta K, Sakamoto T, and
Ito T and Shiraishi K (1998b) Theoretical investigations of Kawashima M (1985) Layer-by-layer sublimation observed
adsorption behavior on GaAs(001) surfaces. Japanese by reflection high-energy electron diffraction intensity
Journal of Applied Physics 37: 4234–4243. oscillation in a molecular beam epitaxy system. Applied
Ito T, Tsutsumida K, Nakamura K, et al. (2004) Systematic Physics Letters 47: 286.
theoretical investigations of adsorption behavior on the Komeda T, Namba K, and Nishioka Y (1998) Simultaneous
GaAs(001)-c(4 4) surfaces. Applied Surface Science observation of SiO2 Surface and SiO2/Si interface using self-
237: 194. assembled-monolayer island. Japanese Journal of Applied
Iwamoto K, Ito Y, Kamimuta Y, et al. (2007) Re-examination of Physics 37: L214.
flat-band voltage shift for high-k MOS devices. In: Technical Konc K and Martin RM (1981) Atomic structure and properties
Digest of 2007 Symposium on VLSI Technologies, of polar Ge–GaAs(100) interfaces. Physical Review B
pp. 70–71. Kyoto, Japan, 12–14 June. 24: 3445.
Kageshima H and Shiraishi K (1998) First-principles study of Kouda M, Umezawa N, Kakushima K, et al. (2009) Charged
oxide growth on Si(100) surfaces and at SiO2/Si(100) defects reduction in gate insulator with multivalent materials.
interfaces. Physical Review Letters 81: 5936. In: Digest of Technical, 2009 Symposium of VLSI
Kageshima H, Shiraishi K, and Uematsu M (1999) Universal Technology, p. 200. Kyoto, Japan, 18 June 2009.
theory of Si oxidation rate and importance of interfacial Si Koyama M, Kamimuta Y, Ino T, et al. (2004) Careful examination
emission. Japanese Journal of Applied Physics 38: L971. on the asymmetric Vfb shift problem for poly-Si/HfSiON gate
172 Atomic Structures and Electronic Properties of Semiconductor Interfaces
stack and its solution by the Hf concentration control in the Murayama M and Nakayama T (1993) Electronic structures of
dielectric near the poly-Si interface with small EOT expense. hetero-crystalline semiconductor superlattices. Journal of
In: Technical Digest of IEEE International Electron Devices the Physical Society of Japan 61: 2419.
Meeting, p. 499. San Francisco, CA, USA, December. Murayama M and Nakayama T (1994) Chemical trend of band
Kuo LH, Kimura K, Miwa S, Yasuda T, Ohtake A, and Yao T offsets at wurtzite/zinc-blende heterocrystalline
(1996a) Dependence of generation and structures of semiconductor interfaces. Physical Review B 49: 4710.
stacking faults on growing surface stoichiometries in ZnSe/ Murayama M, Nakayama T, and Natori A (2001) Au–Si Bonding
GaAs. Applied Physics Letters 69: 1408. on Si(111) Surfaces. Japanese Journal of Applied Physics
Kuo LH, Kimura K, Yasuda T, et al. (1996b) Effects of interfacial 40: 6976.
chemistry on the formation of interfacial layers and faulted Murayama M, Shiraishi K, and Nakayama T (1998) Reflectance
defects in Zn Se/GaAs. Applied Physics Letters 68: 2413. difference spectra calculations of GaAs(001) As- and Ga-rich
Kuwano N, Nagatomo Y, Kobayashi K, et al. (1994) reconstruction surface structure. Japanese Journal of
Transmission electron microscope observation of cubic GaN Applied Physics 37: 4109.
grown by metal organic vapor phase epitaxy with Nakajima K (1999) Equilibrium phase diagrams for Stranski–
demithylhydrazine on (001) GaAs. Japanese Journal of Krastanov structure mode of III–V ternary quantum dots.
Applied Physics 33: 18–22. Japanese Journal of Applied Physics 38: 1875.
Lambrecht WR and Segall B (1990) Interface-bond-polarity Nakayama T (1992a) Electronic structures of heterovalent (001)
model for semiconductor heterojunction band offsets. semiconductor superlattices: GaP/ZnS and GaAs/ZnSe.
Physical Review B 41: 2832. Journal of the Physical Society of Japan 61: 2458.
Lee BH, Oh J, Tseng HH, Jammy R, and Huff H (2006) Gate Nakayama T (1992b) Valence band offset and electronic
stack technology for nanoscale devices. Materials Today structures of zinc-compound strained superlattices. Journal
9: 32. of the Physical Society of Japan 61: 2434.
Louie SG, Chelikowsky JR, and Cohen ML (1977) Ionicity Nakayama T (1993) Band offsets: The charge transfer effect.
and the theory of Schottky barriers. Physical Review B Physica B 191: 16.
15: 2154. Nakayama T (1997) Reflectance difference spectra of
Louie SG and Cohen ML (1976) Electronic structure of a metal– semiconductor surfaces and interfaces. Physica Status
semiconductor interface. Physical Review B 13: 2461. Solidi B 202: 741.
Lu HC, Gustafsson T, Gusev EP, and Garfunkel E (1995) An Nakayama T, Itaya S, and Murayama D (2006a) Nano-scale
isotopic labeling study of the growth of thin oxide films on view of atom intermixing at metal/semiconductor interfaces.
Si(100). Applied Physics Letters 67: 1742. Journal of Physics: Conference Series 38: 216.
Madelung O (ed.) (1997) Landort–Boernstein New Series, Group Nakayama T and Kobayashi R (2005) Electronic structures of
IV Physical Chemistry, vols. 5A–5J. Berlin: Springer. natural quantum-dot system: Si stacking-fault tetrahedron.
Massoud HZ, Plummer JP, and Irene EA (1985) Thermal Physica Status Solidi (c) 2: 3125.
oxidation of silicon in dry oxygen growth-rate enhancement Nakayama T, Kobayashi R, Sano K, and Murayama M (2002a)
in the thin regime. Journal of the Electrochemical Society Defect formation in heterovalent ZnSe/GaAs epitaxy:
132: 2685. Theoretical study. Physica Status Solidi (b) 229: 311.
Matsudo T, Ohta T, Yasuda T, et al. (2002) Observation of Nakayama T and Murayama M (1999) Tight-binding-calculation
oscillating behavior in the reflectance difference spectra of method and physical origin of reflectance difference spectra.
oxidized Si(001) surfaces. Journal of Applied Physics Japanese Journal of Applied Physics 38: 3497.
91: 3637. Nakayama T and Murayama M (2000a) Electronic structures of
Matthews JW and Blakeslee AE (1975) Defects in epitaxial hexagonal ZnO/GaN interfaces. Journal of Crystal Growth
multilayers: II. Dislocation pile-ups, threading dislocations, 214/215: 299.
slip lines and cracks. Journal of Crystal Growth 29: 273. Nakayama T and Murayama M (2000b) Atom-scale optical
Matthews JW and Blakeslee AE (1976) Defects in epitaxial determination of Si-oxide layer thickness during layer-by-
multilayers: III. Preparation of almost perfect multilayers. layer oxidation: Theoretical study. Applied Physics Letters
Journal of Crystal Growth 32: 265. 77: 4286.
Mayer JE and Mayer MG (1940) Statistical Mechanics. Wiley. Nakayama T, Murayama M, and Murayama D (2002b) Au–Si
Mazumdar MK, Mashiko Y, Koyama MH, Takakuwa Y, and intermixing mechanism at Au/Si metal/ semiconductor
Miyamoto N (1995) Generation kinetics of pyramidal hillock interfaces. In: Proceedings of the 26th International
and crystallographic defect on Si(111) vicinal surfaces grown Conference on the Physics of Semiconductors, Edinburgh,
with SiH2Cl2. Journal of Crystal Growth 155: 183. UK, 29 July–2 August 2002.
Meirav U, Kastner KA, and Wind SJ (1990) Single-electron Nakayama T and Sano K (2001) Monte Carlo simulation of
charging and periodic conductance resonances in GaAs defect formation in ZnSe/GaAs heterovalent epitaxy. Journal
nanostructures. Physical Review Letters 65: 771. of Crystal Growth 227/228: 665.
Mendelson S (1964) Stacking fault nucleation in epitaxial silicon Nakayama T, Shinji S, and Sotome S (2008) Why and how atom
on variously oriented silicon substrates. Journal of Applied intermixing proceeds at metal/Si interfaces: Silicide
Physics 35: 1570. formation vs. random mixing. ECS Transactions
Ming Z, Nakajima K, Suzuki M, et al. (2006) Si emission from the 16(10): 787–795.
SiO2/Si interface during the growth of SiO2 in the HfO2/SiO2/ Nakayama T, Shiraishi K, Miyazaki S, et al. (2006b) Physics of
Si structure. Applied Physics Letters 88: 153516. metal/high-k interfaces. ECS Transactions 3: 129–140.
Mizuo S and Higuchi H (1982) Retardation of Sb diffusion in Si Nakayama T, Sotome S, and Shinji S (2009) Stability and
during thermal oxidation. Japanese Journal of Applied Schottky barrier of silicides: First-principles study.
Physics 20: 739. Microelectronic Engineering 86: 1718.
Monch W (1996) Empirical tight-binding calculation of the Naser B, Ferry DK, Heeren J, Reno JL, and Bird JP (2006) Large
branch-point energy of the continuum of interface-induced capacitance in the nanosecond-scale transient response of
gap states. Journal of Applied Physics 80: 5076. quantum point contacts. Applied Physics Letters
Mott NF (1938) Note on the contact between a metal and an 89: 083103.
insulator or semi-conductor. Proceedings of the Cambridge Neugebauer J, Zywietz T, Scheffler M, Northrup JE, and Van de
Philosophical Society 34: 568. Walle CG (1998) Clean and As-covered zinc-blende
Atomic Structures and Electronic Properties of Semiconductor Interfaces 173
GaN(001) surfaces: Novel surface structures and surfactant Sakurai S and Nakayama T (2002) Electronic structures and
behavior. Physical Review Letters 80: 3097. etching processes of chlorinated Si(111) surfaces. Japanese
Nicolini R, Vanzetti L, Mula G, et al. (1994) Local interface Journal of Applied Physics 41: 2171.
composition and band discontinuities in heterovalent Sakurai Y, Iwata J, Muraguchi M, et al. (2010) Temperature
heterostructures. Physical Review Letters 72: 294. dependence of electron tunneling between two dimensional
Northrup JE (1991) Structure of Si(100)H: Dependence on the H electron gas and Si quantum dots. Japanese Journal of
chemical potential. Physical Review B 44: 1419. Applied Physics 49(1): 014001.
Northrup JE, Neugebauer J, Feenstra RM, and Smith AR (2000) Sasaki S, De Franceschi S, Elzerman JM, et al. (2000) Kondo
Structure of GaN(0001): The laterally contracted Ga bilayer effect in an integer-spin quantum dot. Nature 405: 764.
model. Physical Review B 61: 9932. Schottky W (1938) Halbleitertheorie der Sperrschicht.
Oda K and Nakayama T (1992) Effects of interface atomic Naturwissenshcaften 26: 843.
configurations on electronic structures of semiconductor Scopel WL, da Silva AJR, Orellana W, and Fazzio A (2004)
superlattices. Japanese Journal of Applied Physics 31: 2359. Comparative study of defect energetics in HfO2 and SiO2.
Oda M and Nakayama T (2008) Energy-level alignment, Applied Physics Letters 84: 1492.
ionization, and stability of bio-amino acids at amino-acid/Si Shigeno Y and Nakayama T (2007) Current-induced quantum
junctions. Japanese Journal of Applied Physics 47: 3712. friction of nano-linked molecular vibration. Journal of
Ohishi K and Hattori T (1994) Periodic changes in SiO2/Si(111) Physics: Conference Series 89: 012023.
interface structures with progress of thermal oxidation. Shiraishi K, Akasaka Y, Miyazaki S, et al. (2005) Universal theory
Japanese Journal of Applied Physics 33: L675–L678. of work functions at metal Hf-based high-k dielectrics
Ohno T (1993) Energetics of As dimmers on GaAs(001) As-rich interfaces – guiding principles for gate metal selection. In:
surfaces. Physical Review Letters 70: 631. Technical Digest of IEEE International Electron Devices
Ohta A, Miyazaki S, Akasaka Y, et al. (2006) Extended Abstracts Meeting, p. 29. Washington, DC, USA, December.
of 2006 International Workshop on Dielectric Thin Films for Shiraishi K and Ito T (1998) Ga-adatom-induced As
Future ULSI Devices – Science and Technology, p. 61. rearrangement during GaAs epitaxial growth: Self-surfactant
Kawasaki, Japan, November. effect. Physical Review B 57: 6301.
Ohtake A, Yasuda T, Hanada T, and Yao T (1999) Real-time Shiraishi K, Kageshima H, and Uematsu M (2000)
analysis of adsorption processes of Zn on the GaAs(001)- Phenomenological theory on Si layer-by-layer oxidation with
(2 4) surface. Physical Review B 60: 8713. small interfacial islands. Japanese Journal of Applied
Okajima K, Takeda K, Oyama N, Ohta E, Shiraishi K, and Ohno T Physics 39: L1263.
(2000) Phenomenological theory of semiconductor epitaxial Shiraishi K, Oyama N, Okajima L, et al. (2002) First principles
growth with misfit dislocations. Japanese Journal of Applied and macroscopic theories of semiconductor epitaxial
Physics 39: L917. growth. Journal of Crystal Growth 237–239: 206–211.
Ono Y, Tabe M, and Kageshima H (1993) Scanning-tunneling- Shiraishi K, Yamada K, Torii K, et al. (2006) Oxygen-vacancy-
microscopy observation of thermal oxide growth on Si(111) induced threshold voltage shifts in Hf-related high-k gate
7 7 surfaces. Physical Review B 48: 14291. stacks. Thin Solid Films 508: 305.
Otani M, Shiraishi K, and Oshiyama A (2003) First-principles Spicer WE, Cao R, Miyano K, et al. (1989) From synchrotron
calculations of boron-related defects in SiO2. Physical radiation to I–V measurements of GaAs Schottky barrier
Review B 68: 184112. formation. Applied Surface Science 41/42: 1.
Otani M and Sugino O (2006) First-principles calculations of Spicer WE, Lindau I, Skeath P, and Su CY (1980) Unified defect
charged surfaces and interfaces: A plane-wave nonrepeated model and beyond. Journal of Vacuum Science and
slab approach. Physical Review B 73: 115407. Technology 17: 1019.
Oyama N, Ohta E, Takeda K, Shiraishi K, and Yamaguchi H Takahashi Y, Namatsu H, Kurihara K, Iwadate K, Nagase K, and
(1999) First-principles calculations on atomic and electronic Murase K (1996) Size dependence of the characteristics of Si
structures of misfit dislocations in InAs/GaAs(110) and single-electron transistors on SIMOX substrates. IEEE
GaAs/InAs(110) heteroepitaxies. Journal of Crystal Growth Transactions on Electron Devices 43: 1213.
201/202: 256. Takakuwa Y, Mazumder MK, and Miyamoto N (1997) Surface
Pajot B (1994) Some atomic configurations of oxygen. adsorbate related nucleation of crystallographic defect in
In: Shimura F (ed.) Oxygen in Silicon, Semiconductors and chemical vapor deposition of silicon with dichlorosilane.
Semimetals Series, vol. 42 p. 191. San Diego, CA: Academic Applied Surface Science 117/118: 88.
Press. Takayanagi M, Watanabe T, Iijima R, et al. (2003) Extended
Pashley MD (1989) Electron counting model and its application Abstracts of International Workshop on Gate Insulator, p.
to island structures on molecular-beam epitaxy grown 174. Tokyo, Japan, November.
GaAs(001) and ZnSe(001). Physical Review B 40: 10481. Takeuchi H, Ha D, and King TJ (2004) Observation of bulk HfO2
Pickett WE and Cohen ML (1978) Theoretical trends in the defects by spectroscopic ellipsometry. Journal of Vacuum
abrupt (110) AlAs–GaAs, Ge–GaAs, and Ge–ZnSe Science and Technology A 22: 1337.
interfaces. Journal of Vacuum Science and Technology Tan TY and Gösele U (1985) Point defects, diffusion
15: 1437. processes, and swirl defect formation in silicon. Applied
Pickett WE, Louie SG, and Cohen ML (1978) Self-consistent Physics A 37: 1.
calculations of interface states and electronic structure of Tersoff J (1984a) Schottky barrier heights and the continuum of
the (110) interfaces of Ge–GaAs and AlAs–GaAs. Physical gap states. Physical Review Letters 52: 465.
Review B 17: 815. Tersoff J (1984b) Theory of semiconductor heterojunctions: The
Pollmann J and Pantelides ST (1980) Electronic structure of the role of quantum dipoles. Physical Review B 30: 4874.
Ge–GaAs and Ge–ZnSe(100) interfaces. Physical Review B Tersoff J (1988) New empirical approach for the structure and
21: 709. energy of covalent systems. Physical Review B 37: 6991.
Ravi KV and Varker CJ (1974) Oxidation-induced stacking faults Tersoff J (1998) Enhanced nucleation and enrichment of
in silicon. I. Nucleation phenomenon. Journal of Applied strained-alloy quantum dots. Physical Review Letters
Physics 45: 263. 81: 3183.
Rodebush WH (1931) The calculation of chemical equilibrium Thomas DJD (1963) Surface damage and copper precipitation
from spectroscopic data. Chemical Reviews 9: 319. in silicon. Physica Status Solidi B 3: 2261.
174 Atomic Structures and Electronic Properties of Semiconductor Interfaces
Tolman RC (1938) The Principles of Statistical Mechanics. Yamaguchi H, Belk JG, Zhang XM, et al. (1997) Atomic-scale
Oxford: University Press. imaging of strain relaxation via misfit dislocations in highly
Tomita Y, Ishii H, and Nakayama T (2009) Transient current mismatched semiconductor heteroepitaxy: InAs/
behavior through molecular bridge systems: Effects of GaAs(111)A. Physical Review B 55: 1337.
intra-molecule current on quantum relaxation and oscillation. Yasuda T, Kuo LH, Kimura K, et al. (1996) In situ
e-Journal of Surface Science and Nanotechnology 7: 606. Characterization of ZuSe/GaAs(100) interfaces by
Torii K, Shiraishi K, Miyazaki S, et al. (2004) Technical Digest of reflectance difference spectroscopy. Journal of Vacuum
2004 IEEE International Electron Device Meeting, p. 129. San Science and Technology B 14: 3052.
Francisco, CA, 13–15 December. Yasuda T, Kumagai N, Nishizawa M, Yamasaki S, Oheda H, and
Tu Y and Tersoff J (2002) Microscopic dynamics of silicon Yamabe K (2003) Layer-resolved kinetics of Si oxidation
oxidation. Physical Review Letters 89: 086102. investigated using the reflectance difference oscillation
Uematsu M, Kageshima H, and Shiraishi K (2000) Unified method. Physical Review B 67: 195338.
simulation of silicon oxidation based on the interfacial silicon Yasuda T, Yamasaki S, Nishizawa M, et al. (2001) Optical
emission model. Japanese Journal of Applied Physics anisotropy of oxidized Si(001) surfaces and its oscillation in
39: L699. the layer-by-layer oxidation process. Physical Review
Uematsu M, Kageshima H, and Shiraishi K (2001) Simulation of Letters 87: 037403.
wet oxidation of silicon based on the interfacial silicon Zhang SB, Hybertsen MS, Cohen ML, Louie SG, and
emission model and comparison with dry oxidation. Journal Tomanek D (1989) Quasiparticle band gaps for ultrathin
of Applied Physics 89: 1948. GaAs/AlAs(001) superlattices. Physical Review Letters
Uematsu M, Kageshima H, Fukatsu S, et al. (2006) Enhanced Si 63: 1495.
and B diffusion in semiconductor-grade SiO2 and the effect
of strain on diffusion. Thin Solid Films 508: 270.
Uematsu M, Kageshima H, Takahashi Y, Fukatsu S, Itoh KM, and
Shiraishi K (2004) Correlated diffusion of silicon and boron in Further Reading
thermally grown SiO2. Applied Physics Letters 85: 221.
Uwai K and Kobayashi N (1994) Dielectric response of As- Breuer H-P and Petruccione F (2002) The Theory of Open
stabilized GaAs surfaces observed by surface photo- Quantum Systems. Oxford University Press.
absorption. Applied Physics Letters 65: 150. Cappaso F and Marrgaritondo G (1987) Henetrojunction
Venables JA, Spiller GTD, and Hanbucken M (1984) Nucleation Band Discontinuities: Physics and Device Applications.
and growth of thin films. Reports on Progress in Physics North-Holland.
47: 399. Edward TY, Mccaldin JO, and Mcgill TC (1992) Band offset an in
Van de Walle CG and Neugebauer J (2003) Universal alignment semiconductor heterojunctions. In: Ehrenreich H and
of hydrogen levels in semiconductors, insulators and Turnbull D (eds.) Solid State Physics, vol. 46, pp. 2–147. San
solutions. Nature 423: 626. Diego, CA: Academic Press.
van der Merwe JH (1991) Misfit dislocation generation in Gao D and Wei SH (eds.) (1993) II–VI Semiconductor
epitaxial layers. Critical Reviews in Solid State and Materials Compounds. World Scientific.
Sciences 17: 187. Glastone S, Laidler KJ, and Eyring H (1964) The Theory of Rate
van der Merwe JH (1963a) Crystal interfaces. Part I. Semi- Processes. New York: McGraw-Hill.
infinite crystals. Journal of Applied Physics 34: 117. Kamimura H and Toyozawa Y (eds.) (1983) Recent Topics in
van der Merwe JH (1963b) Crystel interfaces. Part II. Finite Semiconductor Physics. Singapore: World Scientific.
overgrowths. Journal of Applied Physics 34: 123. Kangawa Y, Ito T, Taguchi A, Shiraishi K, Irisawa T, and
von Klitzing K, Dorda G, and Pepper M (1980) New method for Ohachi T (2002) Applied Surface Science 190: 517–520.
high-accuracy determination of the fine-structure constant Moench W (2001) Semiconductor Surfaces and Interfaces.
based on quantized Hall resistance. Physical Review Letters Berlin: Springer.
45: 494–497. Ogawa T and Kanemitsu Y (eds.) (1995) Optical Properties of
Watanabe H, Kato K, Uda T, et al. (1998) Kinetics of initial layer- Low-Dimensional Materials. Singapore: World Scientific.
by-layer oxidation of Si(001) surfaces. Physical Review Salaneck WR, Seki K, Kahn A, and Pireaux J-J (eds.) (2002)
Letters 80: 345. Conjugated Polymer and Molecular Interfaces. Marcel
Watanabe T and Ohdomari I (2007) A kinetic equation for Dekker.
thermal oxidation of silicon replacing the deal-grove Shiraishi K, Yamada K, Torii K, et al. (2004) Oxygen vacancy
equation. Journal of the Electrochemical Society 154: G270. induced substantial threshold voltage shifts in the Hf-based
Watanabe T, Tatsumura K, and Ohdomari I (2004) SiO2/Si high-K MISFET with pþpoly-Si gates – a theoretical
interface structure and its formation studied by large-scale approach. Japanese Journal of Applied Physics 43: L1413.
molecular dynamics simulation. Applied Surface Science Sze SM and Kwok KNg (2007) Physics of Semiconductor
237: 125. Devices. Wiley.
Watanabe T, Tatsumura K, and Ohdomari I (2006) New linear- Tersoff J and Harrison WA (1987) Transition-metal impurities in
parabolic rate equation for thermal oxidation of silicon. semiconductors – their connection with band lineups and
Physical Review Letters 96: 196102. Schottky barriers. Physical Review Letters 58: 2367.
Wyckoff RWG (1963) Crystal Structure. NewYork: Interscience. Weiss U (2008) Quantum Dissipative Systems. World Scientific.
Yamada K, Inoue N, Osaka J, and Wada K, (1989) In situ You JH and Johnson HT (2009) Effect of dislocations on
observation of molecular beam epitaxy of GaAs and AlGaAs electrical and optical properties in GaAs and GaN.
under deficient As4 flux by scanning reflection electron In: Ehrenreich H and Spaepen F (eds.) Solid State Physics,
microscopy. Applied Physics Letters 55: 622. vol. 61, pp. 144–261. New York: Academic Press.
1.05 Integer Quantum Hall Effect
H Aoki, University of Tokyo, Tokyo, Japan
ª 2011 Elsevier B.V. All rights reserved.
175
176 Integer Quantum Hall Effect
Energy
fully filled, then we can regard the system as a
v.b.
giant closed shell.
2. Transport properties. The closed shell is not an
ordinary one, since the quantized Hall conductiv-
ity (integer times e2/h) is given entirely in terms Distance from interface(z)
of physical constants (e : elementary charge,
h : Planck’s constant). (b)
Electrode
1.05.1.1 2D Electron Gas
As a background we should start with describing GaAs
the two-dimensional electron gas (2DEG). The AlGaAs c.b.
usual electron gas is a system of electrons that c.b.
move more or less freely in a continuous space, EF
as typically realized in simple metals. In
semiconductor physics, we can realize 2DEG in
metal-oxide-semiconductor field-effect transistors
~100Å
(MOSFETs; Figure 1(a)) mainly before c. 1970s,
and subsequently QHE is observed primarily in Figure 1 Structures of MOSFET (a) and semiconductor
semiconductor heterostructures (Figure 1(b)) heterostructure (b). In each frame, the upper panel shows
the sample structure, while the lower panel the electronic
(Ando et al., 1982). band structure around the interface. Typical wave functions
In these structures, electrons are confined to a 2D are plotted in black against the direction perpendicular to
plane, due to a Schottky barrier between the metal the interface.
and the oxide in an MOSFET, or between different
semiconductors in a heterostructure. There, an elec-
tron moves in the 2D plane, with the electronic
structure of the constituent materials entering only the band structure comprises 2D electronic bands
through the effective mass in the effective-mass associated with the quantized levels in the normal
approximation. The wave function has a finite thick- direction, which are called subbands. When only the
ness in the direction perpendicular to the plane, but lowest subband is occupied by electrons (or, more
the motion along this direction is quantized, so that precisely, if the transitions between adjacent subbands
Integer Quantum Hall Effect 177
can be neglected), we can regard the motion genuinly part. The resulting drift of the center coordinate
2D. The Schrödinger equation reads accounts for the classical Hall effect.
When the system is disordered, due to, for exam-
1 2
H ¼ p þ U ðzÞ ¼E ð1Þ ple, impurities, then an electron is scattered, and
2m
the drift velocity acquires a component along E
where m is the effective mass of the electron, p is the (Figure 2(b)). If we define the conductivity, the
momentum in the 2D plane (x, y), and U(z) is the quantity is a tensor in the presence of a magnetic
confining (Schottky) potential. If we ignore disorder field, where the current, j, and E are related as
(interface roughness, impurities, etc.), the wave func- 0 1 0 10 1
jx xx xy Ex
tion is expressed as @ A¼@ A@ A
jy yx yy Ey
¼ exp iðkx x þ ky yÞ fn ðzÞ ð2Þ
0 1 0 10 1 ð3Þ
Ex xx xy jx
up to a normalization constant, where the 2D motion @ A¼@ A@ A
reduces to plane waves and fn is the nth quantized Ey yx yy jy
wave function along z.
In semiconductor heterostructures, typically where m is the conductivity tensor and m the
GaAs/AlGaAs grown with the molecular beam resistivity tensor. They are inverse matrices with
epitaxy (MBE), the Fermi energy is EF 10 meV, each other, so that we have, with xx = yy, yx = -xy,
so that the electron system is a degenerate Fermi 0 1 0 1–1 0 1
xx xy xx xy xx – xy
gas at liquid He temperature (0.4 meV). @ A¼@ A ¼ 1 @ A
yx yy – xy xx 2xx þ 2xy
xy xx
0 1 0 1
xx xy xx – xy
1.05.1.2 2DEG in Strong Magnetic @ A¼ 1 @ A
Fields – Classical Mechanics 2 þ 2
yx yy xx xy xy xx
If we apply a magnetic field, B, normal to a 2DEG, ð4Þ
classically an electron undergoes a circular motion
If we introduce a phenomenological relaxation
(called Larmor’s motion) due to the Lorentz force,
time, 0, in zero magnetic field to describe the
ev B (e: elementary charge, v: velocity of the
scattering in a classical transport theory with the
electron). When there is an external electric field, E,
equation of motion given by m(d/dt þ 1/
the classical orbit becomes a trochoid (Figure 2(a)),
0)v ¼ e(E þ v B), then we have
where the center of the circular motion drifts in a
direction perpendicular to E with a drift velocity 0
xx ¼
cE B/B 2 (c: speed of light). 1 þ !2c 02
ð5Þ
This is because the electron is accelerated (decel- 0 !c 0 nec xx
xy ¼ ¼ – þ
erated) when it moves along (against) E, while the 1 þ !2c 02 B !c 0
Lorentz force (_ v) is always balanced with the where 0 ¼ ne2 0/m is the conductivity in the
centrifugal force (_ the radius of the circular absence of magnetic fields, and
motion), so that the trajectory is elongated in the
accelerated part while contracted in the decelerated !c ¼ eB=m c
(a) (b)
z y z y
B B
(X, Y )
x x
Ex Ex
Figure 2 Classical orbits for a charged particle in a magnetic field (k z) in an applied electric field (k x) for a clean system (a)
and in a disordered system with scatterers (crosses) (b).
178 Integer Quantum Hall Effect
2D
EG D
S
D y
D
x
Figure 3 (a) Sample geometry for measuring the QHE. (b) Top views of a Hall bar and a Corbino sample, with source (S) and
drain (D) electrodes. (c) Equipotential lines in the QHE condition.
Then the one-particle Hamiltonian, H 0 for the where 0 X ch/e ¼ 4 107G cm2 is the flux quantum,
clean system, which is quadratic in p, can be so that n amounts to the number of flux quanta pene-
expressed as trating the unit area. Alternatively, we can say that the
total degeneracy, S2/2,2, is of the order of the number
1
H 0 ¼ h!c a y a þ ð12Þ of cyclotron orbits that cover the sample area S.
2
To see this, we can look at the wave functions. We
in terms of the operator a only, which is natural since first note that the operator b is related
with the orbit
b involving (X, Y) should not appear in a translation- 1
center as X 2 þ Y 2 ¼ 2,2 b y b þ . The formula-
ally invariant system. Since the Hamiltonian has the 2
same form as a linear harmonic oscillator, the energy tion so far does not depend on the gauge (i.e., how
eigenvalues are we fix the vector potential A which has an ambiguity
related with the gauge transformation). The wave
1 function does depend on the gauge. Let us adopt
EN ¼ h!c N þ ; N ¼ 0; 1; 2; . . . ð13Þ
2 1
the symmetric gauge, A ¼ B r. This amounts to
where N is called the Landau index. 2
So we have here a truly abnormal situation where taking, out of the degenerate wave functions, the set
the energy spectrum, despite the system being a bulk, that diagonalizes, simultaneously with the energy, the
is completely discrete (Figure 4(a)). In 3D systems angular momentum, L ¼ r p, which is along z when
we do have Landau’s quantization, but the extra the motion is within the (x,y) plane. We can show that
motion along B makes the density of states a con- L̂z ¼ ½r ðp – ðe=cÞAz ¼ a y a – b y b ð16Þ
tinuum (Figure 4(a)). In 2D each level, called the
Landau level as labeled by Landau index, has then a with h = 1. For the N ¼ 0 Landau level we have a
macroscopic degeneracy. The degeneracy can be harmonic-oscillator form for Lˆz ¼ byb, so that the
estimated by noting that the density of states of a eigenfunction having the eigenvalue of Lz ¼ m is
2DEG, which is a constant, DðE Þ ¼ m =2h2 per unit given by
y m
area, integrated over an interval h!c should corre- b
spond to the degeneracy per unit area: jmi ¼ pffiffiffi j0i; m ¼ 0;1;2; . . .
m!
where j0i is the vacuum of boson b. If we go to the
n ¼ h!c DðEÞ ¼ 1=ð2,2 Þ ð14Þ
first-quantized form the wave function is expressed,
This number can be expressed as in polar coordinates (r,
), as
jmj
n ¼ B= 0 ð15Þ Nm ðrÞ_ exp – im
– r 2 =4,2 r m LN ðr 2 =2,2 Þ ð17Þ
(a) (b)
3D
2
Density of states
0
0
2D
1
N=0 1 2
hωc
0 N=0
0 1 3 5 7 9
2
hωc 2
hωc 2
hωc 2
hωc 2
hωc
E
Figure 4 (a) Density of states for a clean system in the absence (dashed lines) and in the presence (solid lines) of a magnetic
field in two-dimensional (lower panel) or in three-dimensional (upper) systems. (b) Wave functions for a 2D system for various
values of the Landau index N.
180 Integer Quantum Hall Effect
ðmÞ
where LN ðzÞ is associated Laguerre polynomial. mechanical hopping between cyclotron orbits at dif-
They are depicted in Figure 4(b). By restricting the ferent positions. The hopping matrix element
radius R of these wave functions within a radius of a hNXjVjN9X9i becomes large when the random poten-
disk, we recover the degeneracy of a Landau level. tial varies rapidly in space (on the magnetic length
In terms of the degeneracy, we can now define the scale, Equation (8), since each function jNXi has a
Landau level filling factor, that is, the fraction of the spatial extension ,).
occupied states for a given Landau level. If we denote In the presence of the hopping, each Landau level
the density of electrons by ne, the Landau level filling is broadened from the line spectrum. Electronic
factor is structure and transport properties of the disordered,
Landau-quantized systems in 2D were theoretically
X ne =n ¼ 2,2 ne ð18Þ studied with the self-consistent Born approximation
In other words, 1/ is the number of flux quanta in the 1970s by Uemura, Ando and co-workers
per electron. This is the essence of Landau’s quantiza- (Ando et al., 1974, 1974b, 1982). In this approxima-
tion, formulated by Landau in 1930, and the discovery tion, the lifetime of an electronic state due to disorder
of the QHE in 1980 coincided with its half centenary. is taken into account in a self-consistent way, which
When only the lowest (or lowest few) Landau is imperative, since the unperturbed system has
level(s) are occupied (i.e., . 1) for strong enough anomalous, delta-function spectra. Let us assume
magnetic fields, the situation is called the quantum that the random potential V(r) ¼ i 2,2V0(r ri)
limit. For GaAs with a typical density of electrons is expressed as a sum of the contributions from short-
n 1011 cm2, we have range (delta-function like) impurity potentials at
position ri, with the average number of impurities
4
¼ n 1011 cm – 2 per unit area ni. For dense impurities (to be more
B ðTÞ precise, for the dimensionless concentration ci X 2
where B is in units of tesla and ne in 1011 cm2, so that ,2ni 1), we can adopt the Born approximation.
the quantum limit is realized for B & 4 T. There, the self-energy due to the impurity scattering
So far we have considered a clean system. In is ðEÞ ¼ ci V02 GðEÞ, where the self-consistency
the presence of disorder, such as a random potential demands that Green’s function, G(E), contains the
V(r) arising from impurities and interface roughness, effect of (E). Then each Landau level is broadened
pffiffiffi
cyclotron orbits that have different center coordinates with a width, ¼ 2 ci V0 :
(X, Y) are no longer degenerate, but are subject to In other words, the ratio between the Landau-
scattering. Then the equation of motion for (X, Y) level broadening and the cyclotron energy is
may be obtained from its commutator with H as
=ðh!c Þ 1=ð!c 0 Þ1=2 ð21Þ
y
tities, we have
H ¼ N þ h!c cNX cNX
NM X
2 h!c ¼ 0:12 B ðTÞðm0 =m ÞmeV
y pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ hNX jV jN 9X 9icNX cN 9X 9 ð20Þ
NXN 9X 9
0:15 B ðTÞ=
where cyNXcreates an electron in the Landau’s wave where B is measured in units of tesla, the effective
function jNXi (here in the Landau gauge (A ¼ (0, Bx)) mass m ¼ 0.067m0 (m0: bare mass of an electron) for
rather than in the symmetric gauge), and the second GaAs and ¼ e 0/m is the carrier mobility here
term on the right-hand side represents the quantum measured in units of 104 cm2 V1 s1.
Integer Quantum Hall Effect 181
1.05.2 Integer QHE – Experiments the Hall resistance Rxy are quantized very accurately
into integer multiples of h/e2 ¼ 25 813, or when
The Landau quantization in 2DEG described above translated to the Hall conductivity as
is reflected in various properties, especially in trans-
e2
port properties. There are prehistories for the the xy ¼ – N ; N ¼ integer ð22Þ
h
QHE physics that are precursors of the discovery.
As early as in the 1960s, Shubnikov de Haas effects Astonishing, because (1) the quantized value only
were observed in Si-MOSFETs by Landwehr’s contains the fundamental physical constants, the
group, and by groups in US and in Japan. elementary charge e and Planck’s constant h (whereas
Shubnikov de Haas effect is an oscillatory (in B) ordinarily transport properties are naturally affected
transport phenomenon that generally occurs in mag- by various material parameters, etc.), and (2) this
netic fields, but the effect in 2DEG is peculiar, since occurs in disordered systems. This has become
the oscillations in xx and xy reflect the line-like known as the quantum Hall effect. A few years after
spectrum Landau levels. In the 1970s more elaborate this discovery the fractional QHE was discovered, so
experimental studies did exhibit oscillatory beha- the original effect is sometimes called the integer
viors indicative of Landau levels in 2D. Figure 5 quantum Hall effect (IQHE). Remarkably, the accu-
racy of the quantization is experimentally confirmed
shows a typical example (Igarashi et al., 1975;
to be better than 107. Hall effect was discovered
Kawaji and Wakabayashi, 1981; Kawaji, 2008), along
by Edwin Hall in 1879, so it was almost exactly a
with a theoretical result. It was also recognized that
century later when the remarkable Hall effect was
there are regions of vanishing xx and flat xy
discovered. Subsequently, experimental data were
between Landau levels, which were called plateaux.
refined, where the quantum Hall steps are almost a
In 1980, von Klitzing, Dorda, and Pepper found an
series of step functions, as typically shown in
astonishing behavior in the Hall conductivity as
Figure 7, which were then interpreted in terms of
shown in Figure 6 (von Klitzing et al., 1980). In a
the localization, as we shall see in the section on
Si-MOSFET in strong magnetic fields, the heights of
localization.
nec –σxy
10
H
1.05.2.1 Materials and Sample Geometry
Experiment Historically, the IQHE was discovered in Si-
Theory MOSFETs, but subsequently studied for 2DEGs in
8
semiconductor heterostructures, typically GaAs/
AlGaAs (Figure 1). Integer QHE has also been
σxx , –σxy (10–4Ω –1)
Rxx /Ω
400
200
0
23.0 23.5 24.0 24.5
UH/mV Uxx/mV VG/V
RH /Ω
6500
25 2.5
6400
B = 13.0 T
T = 1.8 K
6300
20 2.0
6200
23.0 23.5 24.0 24.5
VG/V
15 1.5
Uxx
10 1.0
5 0.5
UH
0
0 5 10 15 20 25
N= 0 N= 1 N= 2
VG/V
Figure 6 The original QHE result. For the Hall (UH) and longitudinal (Uxx) voltages against the gate voltage VG. Inset shows the
detail around the fully occupied N = 0 Landau level. Reprinted figure with permission from K. von Klitzing, G. Dorda and M. Pepper,
Physical Review Letters 45, 494 (1980). Copyright 2010 by the American Physical Society. For information, see: http://link.aps.org/
14 000
12 000 GaAs-AlxGa1–xAs
T = 50 mK
10 000
8000
ρxy (Ω)
6000
4000
2000
300
5
200 6 3
ρxx (Ω)
7 4 2
100
N=1
2 1
0
0 20 40 60 80
B (kG)
Figure 7 A typical result for IQHE in GaAs-AlxGa1 xAs. Reprinted figure with permission from M. A. Paalanen et al., Physical
Review B 25, 5566 (1982). Copyright 2010 by the American Physical Society. For information, see: http://link.aps.org/
Integer Quantum Hall Effect 183
Energy
exchange interactions), so that every Landau level
is almost twofold spin-degenerated. Later, the
effects of the spin splitting on QHE have been ela-
borated with various experimental techniques. In the
case of Si-MOSFET, a valley degeneracy also exists,
since the bulk Si has multiple valleys in the band Inelastic light
structure. scattering
~ Luminescence
Filling factor
14 12 10 8
200
Capacitance (pF)
180
50
40 –Sxx
Thermopower (μVK–1)
30
20
10
0
–10
–Sxy
–20
–30
–40
0.0 1.0 2.0 3.0 4.0
Chemical potential
Figure 11 A typical theoretical result for the diagonal (Sxx) and off-diagonal (Sxy) components of the thermopower tensor
against chemical potential. Reprinted figure with permission from M. Jonson and S. M. Girvin, Physical Review B 29, 1939
(1984). Copyright 2010 by the American Physical Society. For information, see: http://link.aps.org/
electron systems in the presence of disorder, and was equipotential contour as we have shown above.
subsequently culminated as the scaling theory of loca- Quantum mechanically, however, an electron tunnel
lization in 1979. In a clean crystal, every wave function between these orbits even for a slowly varying poten-
is a Bloch state that extends over the entire sample. In a tial, while the quantum mechanical hopping becomes
disordered system, wave functions can be spatially more frequent for rapidly varying (i.e., short-ranged)
localized, whose spatial extension is characterized by random potentials (see Figure 13). So the localization
the localization length that depends on energy. In the is a quantum mechanical effect. Then plateaux in xy
presence of magnetic fields, the Anderson localization and regions of vanishing xx have been suggested to
was suggested to occur as well for wave functions in come from the localization. A peculiarity in the loca-
the Landau levels in the 1970s (Figure 12). For ran- lization in the QHE system is that the disorder has a
dom potentials slowly varying in space, an electron dual role: disorder causes the localization on the one
follows, semiclassically, an orbit that is an hand, but it gives rise to quantum mechanical hopping
(a)
ψ
| |2 y
x
(b)
Localization length
(c)
D(E )
(d)
E
a bc d e
(e)
Figure 12 Typical wave functions in a disordered QHE system obtained in a computer simulation for various eigenenergies
as indicated on the right panel, which shows the density of states and the localization length against energy. From H. Aoki,
Journal of Physics C 10, 2583 (1977).
186 Integer Quantum Hall Effect
(a) (b)
l (c)
Figure 13 A random potential rapidly varying as compared with the magnetic length l (left panel) and a slowly varying one
(rigth) are schematically shown. (c) A typical contour plot with shaded areas indicating negative energies and thick lines E ¼ 0
contour. From H. Aoki, Reports on Progress in Physics 50, 655 (1987).
Z Z
of cyclotron guiding centers to contribute to transport 1 1
at the same time, as pointed out by Aoki and Here hhAB ii X dt expð – t Þ dhAð – ihÞ
L2 0 0
Kamimura (1989). Bðt Þi; where hi is the canonical ensemble average
plus the average over disorder, ¼ 1/kBT, a
positive infinitesimal, and A(t) the Heisenberg repre-
1.05.3.2 Linear-Response Theory sentation of A.
The most standard method to treat conductivities xy is expressed as a combination
microscopically is the linear-response theory, or hhY_ X_ iihhX_ Y_ ii for the following reason. For the
the Kubo formula. Aoki and Ando (1981, 1993) have conductivity tensor in a magnetic field B, Onsager’s
applied this to the QHE problem. While the plateaux reciprocal theorem dictates that
in xy and vanishing regions in xx are explained by ðBÞ ¼ ð – B Þ
the localization, the puzzle remains as to why the
value of the plateaux in disordered systems is quan- so that we can decompose into the symmetric part
tized into the universal e2/h, which is originally the s and the antisymmetric part a. Then the current j
value for the clean 2DEG with just an integer is expressed as
X
Landau-level filling. In other words, if a disorder j ¼ s E þ axy ðE êz Þ ð25Þ
makes xx zero, why does the disorder allow xy to
stick to the quantized value. where the part of the current (i.e., the Hall current)
In the linear-response theory, the conductivity is that is induced by the magnetic field and is perpen-
given by a current–current correlation function. In dicular to the applied electric field E is related to a
the QHE system, the current has to do with the (Landau et al., 1984).
dynamics of the cyclotron guiding center (X, Y), The linear-response formula may be written in
which is subject to quantum mechanical hopping. terms of Green’s function as
The conductivities, both longitudinal and Hall, are Z 1
e 2 h
given as correlation functions of X_ and Y_ as xy ¼ dEf ðEÞ
iL2 –1
* " #+
xx ¼ e 2 X_ X_ q
Tr X_ ReGðEÞY_ ImGðEÞ – X_ $ Y_
ne ð23Þ qE
xy ¼ – þ xy
B ð26Þ
–1
e _ _ _ _
2
where G ðE Þ ¼ ðE – H þ iÞ is the Green’s
xy ¼ YX – XY ð24Þ
2 function, f(E) ¼ [e(E m) þ 1]1 is the Fermi’s
Integer Quantum Hall Effect 187
distribution function, and $ denotes the term with X_ implies that, (ii) at least in the limit of strong magnetic
and Y_ exchanged. In this expression, the Hall con- fields all the states cannot be localized, otherwise xy
ductivity xy is contributed by all the states below EF, would be zero over the whole region. This is impor-
unlike xx that is related to an energy-dissipating tant, since, according to the scaling theory of
process around EF. In terms of the eigenstates of the localization all the states are localized in two dimen-
Hamiltonian, the Hall conductivity is expressed as sions. The only way to go around this universal
! argument is to go to another universality class, either
ihe 2 X f ð" Þ – f ð"
Þ jx jy – jy jx
to the unitary class (e.g., systems in magnetic fields
xy ð!Þ ¼ 2 ð27Þ
L
6¼ " – "
" – "
þ i where the time-reversal symmetry is broken) or to the
symplectic class (e.g., systems that have spin–orbit
where "
is
-th eigenenergy, and j
the current interactions). QHE system belongs to the unitary
matrix elements between the eigenstates. class, which is why delocalized states are allowed to
From this formalism, we can show, in the presence exist, which in turn makes the QHE to be realized.
of localization arising from disorder, that (i) the Hall The theoretical picture explained here for broadend
conductivity xy should be rigorously flat (at T ¼ 0) Landau levels, xx and xy are summarized in
as a function of the density n of electrons in the Figure 15.
region where the states are localized. This accounts As we shall see below, the quantization of xy into
for the flatness of plateaux in the QHE, as depicted in e2/h can be shown for finite magnetic fields, where the
Figure 14. required condition is EF being in a gap in the density
We can also show that the quantized xy ¼ Ne2/h
of states or in a localized regime (mobility gap). Even
in a plateau should hold in the limit of strong magnetic
in the treatment above, however, different Landau
fields where adjacent Landau levels are well separated
levels are not treated as being completely indepen-
and when EF is in the gap between them. This in turn
dent. As seen from the fact that the Landau’s
quantization is formally equivalent to a 1D harmonic
oscillator, the current operator has matrix elements
Density of states
Inverse loc
length
0
σxx
σxx
0 0
3
e 2 = ntotalec
–σxy(e 2/h)
h H nec 2
H
–σxy
0
0 1 2 3
0
0 1 E/hωc
2πl 2n
Figure 15 Theoretical density of states (with shaded
Figure 14 Theoretical density of states D(E) (with shaded areas indicating the localized states) along with the inverse
areas indicating the localized states), the longitudinal localization length, xx, and xy are schematically plotted
conductivity xx, and the Hall conductivity xy are against energy for a series of Landau levels. Each white
schematically plotted against the Landau-level filling 2,2n. region representing the delocalized states actually has zero
The horizontal dashed line indicates how the quantized width in the thermodynamic limit at T ¼ 0. From H. Aoki and
value is achieved despite the presence of localization. T. Ando, Solid State Communications 38, 1079 (1981).
188 Integer Quantum Hall Effect
between N and N 1 levels, so the mixing between in which a QHE system is wound into a cylinder (but
them is implicitly included (in fact, the treatment in the magnetic field is still applied perpendicular to
terms of the guiding center (X, Y) corresponds to every point on the cylinder), as in Figure 16. To this
taking this mixing to the leading order). we add a magnetic flux (due to, say, a solenoid) that
It may first seem counterintuitive that xy attains pierces the cylinder. Through its vector potential,
the quantized value despite the presence of localized A
¼ /2r (in cylindrical coordinates), the flux exerts
states. Physically, we can say the following. From the an Aharonov–Bohm effect on the electrons. We can
equation of motion for the guiding center the Hall eliminate the vector potential with a gauge transforma-
current jx in an electric field Ey is given as tion at the cost of the boundary condition twisted to
(
) ! exp[i(/ 0)
], on the wave function , where
X
delocalized 0 ¼ eh/e is the flux quantum.
jx ¼ – ð1=B Þ h
9jqV =qy j
9i þ eEy
If we increase with time, we recover the
9
P original periodic boundary condition every time
There is a kind of sum rule, all
9 h
9jV =qy j
9i ¼ 0; increases by 0 in a time interval t. When
which means that delocalized states move faster to the Fermi energy EF is in a gap in the density
just compensate the localized states, which has been of states or in the energy region for localized states
confirmed numerically and field theoretically as well. (as described in the previous section), the occupa-
tion of electrons must be the same when we make
! þ 0. The only change, then, should be a
1.05.3.3 Středa–Widom Formula
transfer of an integer (M) number of electrons from
As another formalism for the Hall conductivity, one electrode to another, which we assume to be
Středa’s formula (Středa, 1982) is also often evoked. attached to either edge of the cylinder as Hall
He showed that the Kubo formula for the Hall con- probes. From the Maxwell equation, an electric
ductivity can be decomposed into a form, field LEy ¼ (1/c)q/qt is induced along the cylin-
xy ¼ Ixy þ IIxy ; where Ixy is the Drude-like part der of circumference L, which should result in a
that tends to Ixy ! – !c 0 xx in the relaxation time Hall current, jx ¼ (e)M/(Lt) ¼ (Me2/h)Ey,
picture. The second term is expressed as and we have xy ¼ Me2/h. So the derivation
applies arbitrary strength of the magnetic field as
qN ðEÞ
IIxy ¼ ec ð28Þ far as the Fermi energy EF is in a gap in the density
qB E¼EF of states, or in the energy region for localized
where N(E) is the integrated density of states. When states.
the Fermi energy EF is in an energy gap, Ixy
vanishes, while we can show that IIxy ¼ – ðe 2=hÞN
when EF is in a gap between the Nth and (N þ 1)th
Landau levels. Thermodynamically, we have the
electric field, E ¼ r(m/e) with being the electro-
chemical potential, so that the Hall current
jH ¼ cr M with M being the magnetization can
be expressed as jH ¼ ecE (qM/q). If we use,
following Widom, the thermodynamic Maxwell’s Φ
relation, qM/q ¼ qN/qB, which comes straight B
from M ¼ q/qB with the grand potential , we
Hall
recover the formula for IIxy (Widom, 1982). Note current
that the Středa–Widom formula is applicable to the
situation where the spectrum has an energy gap.
localization (which is a nonperturbative problem) have 1.05.4.2 Quantum Criticality and xx xy
been extensively studied with numerical methods and Diagram
field theoretical methods. Figure 12 shows typical
One way to display the critical behavior is the xx–xy
behaviour, obtained numerically, of the wave function
diagram. Namely, while at T ¼ 0 xy is a step function
along with the localization length versus energy, which
and xx nonzero at discrete points, at finite tempera-
shows that the localization is strong toward the edges of
tures not only the Fermi distribution is smeared by
a Landau level and the weakest at the center.
kBT, but we have also a finite inelastic scattering
Classically, this corresponds to the fact that contours
length L. This can be summarized as the (xx, xy)
of a random potential, along which the cyclotron guid-
flow diagram when T is varied for various values of
ing center drifts in the semiclassical picture, are valleys
EF, as originally evoked in the nonlinear model
(hills) for energies well below (above) the level center,
(Figure 17(a)). Namely, we can map the QHE
while the contours tend to percolating paths for ener-
system onto a field-theoretic model called the non-
gies close to the center. However, the localization is
linear model, for which the renormalization into
quantum mechanical phenomenon related with inter-
larger sample sizes can be discussed (Huckestein,
ference of wave functions.
1995).
If we look at the behavior of the localization length
We can also vary the sample size for numerical
in a numerical method (finite-size scaling for the
works on the QHE system to look at the renormali-
Thouless number), the localization length, although a
zation flow. Alternatively, we can change the
smooth function of E, exhibits a rather singular beha-
temperature, which changes the inelastic scattering
vior, in which the inverse localization length becomes
length, hence effectively changes the sample size
zero only at the level center (i.e., extended states only
(Figure 17(b)). This implies that experimental (xx,
occur there), as schematically shown in Figure 15.
xy) flow lines can be obtained by varying the tem-
The delocalized states, marginally allowed to appear
perature (Figure 17(c)).
in the presence of magnetic fields in two dimensions,
We can also discuss the so-called plateau-
thus coalesce into a single point E0 (the center of each
to-plateau transition, which is the energy interval over
Landau level) on the energy axis. The localization
which one plateau crosses over to another. In terms of
length, denoted here as , behaves around this point as
the localization, the states that satisfy L" < (E) are
1=jE – E0 js ð31Þ effectively delocalized, where L" is the inelastic scatter-
ing length and the localization length (Figure 18(a)).
where s (numerically shown to be & 2) is the localiza- If we assume the inelastic scattering time behaves like
tion critical exponent (Aoki and Ando, 1985; Ando "(Tp) with p ¼ 2 in the diffusive regime, the energy
and Aoki, 1985). This value may be compared with region whose width is Tp/2s behaves as delocalized,
the exponent s ¼ 7/3 for the percolation problem. since the localization length has a critical behavior
The effect of quantum mechanical tunneling has jE E0js while L" _ 1/Tp/2. If the magnetic
also been extensively studied in terms of a network field is varied instead, the plateau transition width
model due to Chalker and Coddington, who have against B has the width B Tp/2s with the same
modeled the quantum mechanical tunneling across exponent. Figure 18(b) displays an experimental result.
the paths (see Figure 13(c)), (Chalker and One important question is how the QHE plateaux
Coddington, 1988; Kramer et al., 2005). vanish as the degree of disorder is increased, which
The critical exponent s has also been shown to should take place since for a large-enough disorder the
depend on the Landau level (e.g., larger for N ¼ 1 Landau quantization, along with QHE, should go away.
than for N ¼ 0). A mapping onto the nonlinear
Kivelson et al. (1992) have considered the problem
model gives an estimate , exp 2SCBA , with in terms of what is called the global phase diagram on
SCBA (N þ 1/2)e2/h being xx in the self-consistent a xyxx plane. In this diagram xx is regarded as
Born approximation (Ando and Uemura, 1974; Ando a measure of the strength of disorder, while xy
1974a, 1974b; Ando et al., 1975) and , the magnetic the Landau-level filling. As shown in Figure 19(a),
length. So the localization length increases with N, but the midpoints, xy ¼ (N þ 1/2)e2/h, between the pla-
so does the critical exponent. The localization length teaux in the Hall conductivity are bifurcation points in
also depends strongly on whether the randomness is the xxxy flow lines, so that we can take the points,
rapidly varying in real space or slowly varying as
compared with the magnetic length. xy ¼ 1= 2xx þ 2xy ¼ ðN þ 1=2Þe 2=h, as boundaries
Integer Quantum Hall Effect 191
3
1.5K
T = 0.32K
1.78K
10.0K 0.35K
2 n=1 (1 –)
σxx /σSCBA
2.0
(0 –)
0.35K
n=0
1
T = 1.5K
0 0
0 1 1 3 2 5 0 1 2 3.0 3.5 5.0 5.5 6.0
2 2 2 –σxy / (e 2/h)
Figure 17 Results for the xxxy diagram: (a) Renormalization flow lines obtained in the nonlinear model [Reprinted
figure with permission from A. M. M. Pruisken, Physical Review Letters 61, 1297 (1988) Copyright 2010 by the American
Physical Society. For information, see: http://link.aps.org/]. (b) A numerical result [From H. Aoki and T. Ando. Surface
Science 170, 249 (1986)]. (c) An experimental result [From M. Yamane et al., Journal of the Physical Society of Japan 58, 1899
(1989)].
(a) (b)
Inverse localization length ∝ E s 10 μm
1 18 μm
32 μm
64 μm
ΔB (T)
T p/2s 0.3
1
~ T p/2
Lε
E0
E
0.1
30 100 300 1000
T (mK)
Figure 18 (a) In a plot for the inverse localization length against energy, the effectively extended states (white region) is
shown for the inelastic scattering length L" at a given temperature T [after H. Aoki and T. Ando, Surface Science 170, 249
(1986)]. (b) An experimental result for the plateau-to-plateau transition width on the B-axis plotted against T for various values
of the sample size. Reprinted figure with permission from S. Koch et al., Physical Review Letters 67, 883 (1991). Copyright
2010 by the American Physical Society. For information, see: http://link.aps.org/
up to which the QHE effect survives when the strength B ! 0, so that the fate of the delocalized states is
of disorder is increased (Figure 19(b)). The global quite nontrivial. It has been suggested that the ener-
phase diagram has also been experimentally examined gies corresponding to the delocalized states go up in
for IQHE systems. energy, which is called the floating.
It is also important to consider how the situation Another question of heuristic interest is what is
crosses over to the weak magnetic field case, since all the classical limit (i.e., Planck’s constant h ! 0). In
the states should become localized in the limit of this limit, the Landau-level filling 2,2n ¼ (h/eB)n
192 Integer Quantum Hall Effect
(a)
σ 2 σxx
σxy
1
σxx
N=0
n 1 3
2 2
σxy /(e 2/h)
(b)
ρxx0
σxy /(e 2/h) = 1
2
4 3
0 1 2
ρxy 0/(h/e 2)
Figure 19 (a) How a xxxy diagram is obtained from their n-dependence is schematically shown, with the flow lines
bifurcating at xy /(e2/h) ¼ half integer indicated. (b) The global phase diagram.
vanishes for a fixed n. The measure of the Landau- larger sample sizes than in Figure 12, show that the
level mixing, on the other hand, is the Landau-level wave function is very scattered both in amplitude and
broadening divided by the cyclotron energy, spatial extent.
pffiffiffiffiffiffiffiffiffi pffiffiffi
=h!c 1= !c 0 . Since =h!c diverges like 1= h To be more precise, the self-similarity in critical
for h ! 0 with fixed B and 0, the classical limit wave functions extends beyond a single scale trans-
amounts to a regime of dilute (_ h) in weak mag- formation, so the idea has been subsequently
netic fields. developed into the multifractal analysis, with which
we can analyze the delocalized states around the
Landau level center (Huckestein, 1995). As described
1.05.4.3 Fractal Wave Functions and in the section below, real-space imaging such as
Dynamical Scaling
STM begins to visualize the fractal states.
It is an intriguing question to ask whether a transition Among the physical quantities that are affected
between localized and extended states, in general, is by the fractality of the wave functions is the
similar to phase transitions in statistical mechanics. A anomalous diffusion in transport. Namely, the
suggestion came in 1983 by Aoki to the effect that the dc conductivity obeys, in ordinary systems,
wave function at the Anderson transition, which has Einstein’s relation, ¼ limq;!!0 ¼ e 2 DðEF ÞD 0 ,
no characteristic length scale with the diverging loca- where D 0 is the diffusion constant. At a critical
lization length, should be self-similar (fractal) (Aoki, point, the conductivity becomes q, !-dependent. In
1983), just as the critical point in a phase transition is the dynamical scaling ansatz, the ac conductivity
characterized by scale-invariant states with a diver- should depend on !, where the relaxation rate
ging correlation length. goes to zero like 1/ 1/z around the transition.
For the QHE system in particular, the transition Here, z is the dynamical critical exponent, which is
point occurs at the center of each Landau level, so usually 2 in noninteracting systems. If we combine
that the delocalized states at the center of a Landau this with the sample size scaling in L/, q, !-depen-
level are just not the usual extended states, but fractal dence should take a form
(sometimes called critical) wave functions (r) that
have a fractal dimensionality, d < 2. In other words, ðq; !Þe 2 DðEF ÞD q=!1=z ð32Þ
the density autocorrelation decays with a power law,
hj (r) (r þ R)j2i 1/R2 d . Typical wave functions At the criticality,
pffiffiffi the
fractality
p offfiffiffithe
states is shown
in Figure 20, which are obtained numerically for to lead to D q= ! D 0 = q= ! with ¼ 2 d.
Integer Quantum Hall Effect 193
(a)
N=0
d=0
(c) N=1
d = 0.71l
(b)
N=1
d=0
Figure 20 Typical wave functions (represented as their squared amplitude) around the center of each Landau level
obtained in a computer simulation for a disordered QHE system with a system size of 300l 300l for (a) N ¼ 0 and (b) N ¼ 1
Landau levels with short-range scatteres, and for (c) N ¼ 1 with long-range (d 6¼ 0) scatteres. Reprinted figure with permission
from T. Terao et al, Physical Review B 54, 10350 (1996). Copyright 2010 by the American Physical Society. For information,
see: http://link.aps.org/
1.05.5 QHE Edge States and Edge of boundary states in field theories. In a finite QHE
Transport sample, the edge states exist, which appear on the
energy spectrum as the edge modes that cross from
The role of sample edges in the QHE transport has one Landau level to another in a Landau gap. These
been an issue of interest from an early stage when modes have to exist, and sometimes called topologi-
the problem was raised by Halperin in the 1980s cally protected in QHE systems. More importantly, the
(Halperin, 1982). Classically, there are edge currents topological nature of the QHE edge states manifests
that correspond to cyclotron motions skipping itself, as shown by Hatsugai, in the properties that
along the edges (Figure 21(b)). Quantum mechani- (1) the edge Hall conductivity is also expressed as a
cally, edge states also exist (Figure 21(a)). The topological invariant (Chern number), and (2) this
energy diagram plotted against the real-space quantity exactly coincides with the topological Chern
position across the sample width will look like number for the bulk Hall conductivity (Hatsugai,
Figure 22(a). Edge states are rather ubiquitous in 1997). Alternatively, we can visualize, in
quantum mechanics, but the peculiarity in the QHE Figure 22(b), that we can continuously change the
is that the edge currents flow, with no backscatter- situation from the edge-picture limit (where currents
ing, in a definite direction dictated by the direction exist only at edges) to the bulk-picture (where a bulk
of the applied magnetic field. When one takes the potential gradient exists) by deforming the current
electron–electron interaction into account, the edge distribution, but there is always the total current pre-
states may be regarded as special, incompressible servation, I bulk ¼ I left edge I right edge, which implies
electronic states called the chiral Tomonaga– bulk edge
xy ¼ xy (Koshino et al., 2001, 2002a, 2002b).
Luttinger liquid, usually discussed in the context One way to describe the edge transport is
of the fractional QHE. Büttiker’s formula (Büttiker, 1992) which is an exten-
In terms of the topological nature of the QHE, the sion of Landauer’s formula for transport processes in
appearance of edge states is a prime example of a much terms of an S matrix for transmission and reflection
wider concept of the bulk-edge correspondence, channels in ballistic transports. Büttiker’s formula can
that is, the nature of edge states are dictated by the explain some of the experimental results, including
nature of bulk states, which corresponds to the problem the quantization of the Hall conduction. However,
194 Integer Quantum Hall Effect
Edge
(a) (b)
Figure 21 Typical edge states in a QHE system in quantum mechanics (a) and classical mechanics (b) From H. Aoki, in
G. Landwehr (ed.): Application of High Magnetic Fields in Semiconductor Physics (Springer, 1983), p. 11.
(a) (b)
μ0
μ0 eV
Bulk picture
μ2
μ2 Bulk Edge
μ1
E
μ1
Edge picture
Figure 22 (a) Energy spectrum against the position along the width is schematically shown for a sample with edges with the
chemical potential at the left edge (1) and at the right (2). (b) How the edge picture (in which all the currents are carried by
edge states) continuously crosses over to the bulk picture (where a bulk contribution exists) is schematically shown. From M.
Koshino et al., Physical Review B 66, 081301(R) (2002).
this does not mean that all the QHE currents are dimension of the phase coherence length L . The
carried by edges. In general, there exist both bulk transport is described by the Kubo formula xx, xy
and edge Hall currents. Bulk and edge states can even if the sample size is much greater than L where the
be hybridized. The details have been studied both Hall field and Hall current distributed over the sam-
experimentally and theoretically, which largely indi- ple, while the edge transport in the manner of
cate that there are contributions from both bulk and Büttiker starts to dominate when the size is compar-
edges, whose proportions depend on the width of the able with L .
sample, T-dependent xx, etc. Real-space imaging One of the characteristic features in the edge
results also indicate that there are bulk potential transport is that highly nonlocal phenomena can
gradients in QHE samples. arise that extend over length scales far exceeding
Physically, a macroscopic system may be regarded the bulk mean free path. This can be probed in
as comprising subsystems each having a linear Hall-bar samples by selectively injecting currents
Integer Quantum Hall Effect 195
from the contacts affects the conduction, where the difference between different polarizations probing
conduction can strongly depend on which terminals the potential) to show that an electric field exists
to employ in multiterminal measurements. over the whole sample, although the field becomes
The width of edge states has experimentally been stronger toward the edges. An example of the imaging
measured with various methods, including transport is shown in Figure 23. This is followed by various
(e.g., QHE breakdown), magnetocapacitance, and other methods, including scanning capacitive, force,
magnetoplasmon measurements. One recent method and polarization optical microscopies (Morgenstern,
utilizes the nuclear spins. Usually, the nuclear spins 2007).
are irrelevant to electron systems, since the nuclear One such probing, which is particularly suited to
Zeeman energy is as small as 1/2000 of the electronic examine nonequilibrium carriers around the hot spots,
counterpart. When there are almost degenerate elec- is the imaging of cyclotron emission due to Komiyama
tronic states, however, nuclear spins can affect the et al. (2004). In a Hall bar sample in the QHE condition
electron system via the hyperfine interaction. This (with xx ¼ 0, xy 6¼ 0), the electric lines of force are
occurs not only in the fractional QHE systems close forced to be distorted to make the rectangular sample
to a spin-polarization transition, but also in the inte- geometry compatible with the Hall angle of 90 ,
ger QHE where left and right edge states are resulting in singularities which have to appear at two
degenerate in energy, so that the nuclear spin is positions around the electrodes (Figure 24(a)). These
probed via the scattering between the left and right are called the hot spots. In the imaging of cyclotron
edge channels. emission (Figure 24(b)), the QHE detector, which is
In semiconductor superlattices, which realize a itself an IQHE system having by nature sensitive
stack of 2DEGs, an application of a strong magnetic photoresponses at the cyclotron resonance frequency,
field along the growth direction makes the system a is used to scan the QHE system. In the result the hot
stack of IQHE systems. There, the edge states are also spots are clearly seen, with edge channel also visible
stacked along the edge surfaces, which is called sheath for the Landau-level filling 6¼ integer. The imaging of
currents and have been experimentally observed. cyclotron emission has also been used to examine the
breakdown of QHE (see section 1.05.8).
Recently, STM and scanning tunneling spectro-
scopy (STS) begin to be performed for a special kind
1.05.6 Real-Space Imaging of 2DEG, that is, the adsorbate-induce 2DEG, where
for example, Cs atoms are deposited on cleaved
Experimentally, there have been a body of studies for n-InSb. The STS imaging is also used to directly
real-space imaging of the QHE systems, including the observe the quantum Hall transition from one loca-
Hall field. Fontein et al. have studied a potential profile lized regime to another via the delocalized states that
imaging with the electro-optical effect (which utilizes have a fractal character as shown in Figure 25
the birefringence (Pockels effect) with the phase (Hashimoto et al., 2008).
Figure 23 A typical real-space image for the potential distribution obtained with the electro-optic imaging for B ¼ 10.8 T
corresponding to the center of a plateau. From R. Knott et al., Semiconductor Science and Technology 10, 117 (1995).
196 Integer Quantum Hall Effect
(a)
S D
Condenser lens
1400 14.6mm
1000
QHE
1000 2000 3000 4000 etector
X (μm)
Figure 24 (a) The electric lines of force are schematically shown for the QHE condition, with S: source electrode and
D: drain, (b) A real-space imaging obtained with the cyclotron emission detected by the QHE device (inset). Reprinted figure
with permission from K. Ikushima et al., Physical Review Letters 93, 146804 (2004). Copyright 2010 by the American Physical
Society. For information, see: http://link.aps.org/
c
dl/dV (a.u.)
dl/dV (a.u.)
b
d
a
Low
–120 –110 –100 –90 –80
Sample voltage (mV)
Figure 25 A real-space image obtained by STS for 0.01 monolayer of Cs on a cleaved n-InSb(110) at T ¼ 0.3 K for various value
of the Landau-level filling, (a)-(d), as indicated in inset. Reprinted figure with permission from K. Hashimoto et al., Physical Review
Letters 101, 256802 (2008). Copyright 2010 by the American Physical Society. For information, see: http://link.aps.org/
1.05.7 QHE Resistance Standard and et Mesures (CIPM), that the QHE be adopted as the
the Fine-Structure Constant electrical resistance standard, with which ohm is
defined via the QHE resistance,
After the recognition that the accuracy of the QHE
quantization is experimentally better than 106 RK ¼ h=e 2 ¼ 25 812:807 ð33Þ
(nowadays around 107 or even better), it was
decided, in 1990 by Comité International des Poids where RK is called the von Klitzing constant.
Integer Quantum Hall Effect 197
(a) (b)
H 10–8α 10–6h
RK NPL-88
RK LNE-01
KJ PTB-91
RK NMI-97
KJ NMI-89
RK NIST-97
KJ2RK NPL-90
KJ2RK NIST-98
KJ2RK NIST-07
ae U Washington-87
ae Harvard-06 CODATA-02
CODATA-06
CODATA-02
CODATA-06 5 6 7 8 9 10
597 598 599 600 601 602 603 604 [h/(10–34Js) – 6.6260] × 105
Figure 26 (a) Values of the fine-structure constant
obtained with various methods. Those marked with RK are from the QHE
measurements at various institutes, ae from the electron magnetic moment anomaly, (b) A similar plot for the Planck constant h.
CODATA-recommended values are also indicated. From P. J. Mohr et al., Reviews of Modern Physics 80, 633 (2008).
The constant, now adopted as a resistance stan- this phenomenon, known from an early stage of the
dard, has a profound significance related with a QHE studies, is called the breakdown of QHE. The
fundamental physical constant. Namely, the breakdown is important from both applicational
fine-structure constant aspects (since it affects the accuracy of the QHE) and
fundamental aspects since this is an interesting
e2
¼ .1=137:036 ð34Þ nonequilibrium phenomena. Several factors can be
hc involved, among which are tunneling between differ-
which is the coupling constant in the quantum elec- ent Landau orbits or Landau levels, and/or electron
trodynamics (QED) (Kinoshita, 2007), one of the heating effect. Also relevant is how the breakdown is
most basic physical constants, is directly related related with the current profile in real space in the
with RK via
¼ 2/RKc. Here, c ¼ 2.99 792. . . 108 Hall bar. Although there are some prevailing experi-
m s1, the speed of light in vacuum, is, in SI, a mental and theoretical results, we seem to be some
defined value. Figure 26(a) shows the values with way from a unified picture (Nachtwei, 1999).
error bars of
obtained by various methods includ- Sometimes metastable or bistable states are observed
ing that from RK obtained by various institutes. If we around the breakdown regime, which is also of interest.
combine RK with the Josephson constant, KJ ¼ 2e/h, As for nonequilibrium phenomena, we can men-
we can deduce the value of Planck’s constant via tion that the microwave-induced magnetoresistance
h ¼ 4/K2JRK (Figure 26(b)). For a recent review, see oscillation is one remarkable nonequilibrium phe-
Mohr et al. (2008). nomenon, although this occurs in a weak magnetic
field regime rather than in the QHE regime.
Ey (Vcm–1)
–80 –40 0 40 80
ρxx
4
Ex (Vcm–1) 2
0 5
B(T)
–2
–4
–6
–0.8 –0.4 0 0.4 0.8
jx (Am–1)
Figure 27 A typical current–voltage characteristics at ¼ 2 (the arrow in the inset). From Nachtwei, Physica E 4, 79 (1999).
generally fabricated by applying a lateral confining Quantum-dots are also investigated in strong
potential to a 2DEG system. This can be realized by magnetic fields. In a magnetic field, B = rot A,
an electrode that exerts an electrostatic confining applied perpendicular to the dot plane the
potential, or alternatively we can mesa-etch the sys- Hamiltonian, in the one-body problem here, reads
tem to have a finite system (Figure 28(a)). The
confining potential is usually cylindrically parabolic 1 2 1
H ¼
pi þ ðe=c ÞAðrÞ þ m !20 r 2 ð35Þ
to a good approximation. 2m 2
(b) 50
40
(a) A
30
E (meV)
Drain
Side
AlGaAs Gate 20
InGaAs
AlGaAs
10
Source
0
0 2 4 6 8 10
B (T)
Figure 28 (a) A typical mesa-etched quantum-dot structure. From P. A. Maksym et al, Physical Review B 79, 115314 (2009).
(b) Energy levels of the Fock–Darwin states against magnetic field for a harmonic confinment. From P. A. Maksym et al,
Journal of Physics: Condensed Matter 12, R299 (2000).
Integer Quantum Hall Effect 199
where h!0 is the confinement energy and we have and Landau’s quantization interfere with each other,
ignored the Zeeman energy. since the application of a magnetic field, B ¼ rot A,
The exact eigenstates of this Hamiltonian are gives rise to, semiclassically, the Peierls phase,
Z
B
1/4
3/8
1/2
–σxy /(e2/h)
3
0
–1
–2
–3
–4 –2 0 2 4
E
Figure 29 Hofstadter butterfly (energy spectrum against magnetic field) for a square lattice (upper panel), and the
corresponding QHE at the value of magnetic field for which the magnetic flux penetrating a unit cell of the periodic system is
3/8 in units of the flux quantum [after H. Aoki, in G. Landwehr (ed.): Application of High Magnetic Fields in Semiconductor
Physics III (Springer, 1991), p.17].
1,
1
E/tx
1,
1
0
3 3 2 –2 –3
0,4
0,3
4,0
1,
1,02,–20,2 2,0 0,1 3,0
2 –1 1 1,–2
–1
–1,3 –2,2
3,–1 ,1 –2,1
–2 2 1,–3 2, –1 –3,1
0
2
–3 3
1 1,–4
0,
–4,1
1,–1
E/tx
2 2 –3,4 4,–3 ,1
–1
0 –2,3 3,–2
2,–1
2 2 0,4
0,3 0,1 –1,2 1,0 2,
0
4,0
–1 3,0
0,2
3 –3
2 1 –1 –2
–2 3,–3 3,–3
–1,2 2,–1
2,–2 –2,2
–3 –3 –2 2 3 0,0 1,–1 –1,1 0,0
0
0 0.2 0.4 0.6 0.8 1 0 30 60 90
φ θ (degree)
j ¼ s E where s = (yz, zx, xy) (Figure 30). 1.05.10 Integer versus Fractional
Experimental realization is yet to come. QHEs
A kind of Landau quantization has been known to
occur in 3D systems, typically in a quasi-2D organic The integer QHE has subsequently been developed
conductor (TMTSF in the Bechgaard salt family). into the fractional QHE (FQHE) as is described in
This is the field-induced spin-density wave (SDW) the present volume (see Chapter 1.06). In FQHE the
in strong magnetic fields with a many-body origin. Hall conductivity is quantized into xy ¼ e2/h for
Namely, the Landau quantization takes place within fractional Landau-level fillings ¼ 1/3, 2/3, 3/5, . . .
the pockets formed by incompletely-nested Fermi as opposed to the IQHE for ¼ integer. Let us com-
surfaces, which gives rise to a series of gaps around pare the IQHE with FQEH in this section.
the main SDW gap. Physically, the integer QHE is primarily understood
Integer Quantum Hall Effect 201
in terms of a one-body problem as described here, Although the FQHE as a new class of many-body
while FQHE is inherently a many-body effect, and states is associated with fractional Landau-level
this is a customary way to distinguish the IQHE and fillings, the electron–electron interaction exists for
FQHE. It is then natural that the IQHE was origin- integer fillings as well. In fact, an integer is a kind
ally discovered in Si-MOSFET, while the FQHE of fraction in that Laughlin’s wave function for the
effect in GaAs–AlGaAs heterostructures, where the FQHE liquid for clean systems, allowed for odd
latter system is atomically much cleaner with typical fractions ( ¼ 1/m with m an odd integer), also
mobility exceeding 106cm2 V1 ? s1 against the for- accommodates m ¼ 1. So the real question is the
mer’s 104cm2 V1 ? s1. The FQHE has in fact been nature of the energy gap: in the integer QHE the
observed in clean-enough Si-MOSFETs. excitation gap is primarily the one-body gap
One clear way to realize that the FQHE, a many- between the adjacent Landau levels (or between
body effect, emerges as the degree of disorder is the mobility edges in disordered systems), while
lowered is to look at the historical developments the excitation gap in the FQHE has a many-body
from IQHE to FQHE in Figure 31, which shows origin.
how the fractional effect appears as the sample qual- The relative magnitude of the one-body and
ity (as characterized by the carrier mobility) becomes many-body gaps depends on the degree of disorder
higher. In the clean limit, the system is indeed in the (and the g-factor in the case in which the adjacent
limit of strong electron correlation in that the kinetic Landau levels are Zeeman-split ones). To quantify
energy is quenched due to Landau’s quantization so this, we can examine various energy scales.
that the ratio of the interaction energy to the kinetic Figure 32 plots
energy is infinite (although more rigorously we have
to consider the inter-Landau-level matrix elements). 1. h!C (the cyclotron energy),
2. e2/", (typical size of the electron–electron
Coulomb interaction),
where e is the elemental charge, "(¼ 13 for GaAs)
the dielectric constant of the material and , the
magnetic length, and
3. gBB (the Zeeman energy),
2 1
3 3 where g is Landé’s g-factor, B ¼ he=ð2mc Þ the
Bohr magneton. We can see that for B few tesla
the Zeeman energy is relatively negligible, while the
cyclotron energy and the Coulomb energy are com-
1982 parable. For disordered systems we have to compare
ρxx
0
these with the Landau-level broadening , which is
roughly estimated in the self-consistent Born approx-
imation as =ðh!c Þ1=ð!c 0 Þ1=2 , where 0 is the
1984
scattering relaxation time in zero magnetic field. So
the Landau-level broadening is comparable with the
cyclotron energy for !c 0 1.
0
3 2 While the two-dimensionality in IQHE appears in
2 3 5
3 5
6 4
7 the fact that the Hall conductivity represents a topo-
45 9
79 5 11
11 8
logical quantum number (Chern number) in terms of
7 13
13
the Berry’s curvature, in FQHE the spatial dimension-
1987 ality of two is essential in allowing the composite
4 5 fermion picture (i.e., a Chern–Simons gauge field the-
5 7
0 oretic treatment) of the many-body quantum liquid.
1 1/2 1/3
Landau level filling This is the reason why the FQHE accommodates such
novel concepts as anyon quasi-particles.
Figure 31 The developments of the fractional structure in
samples of progressively higher quality From D. C. Tsui et al.,
The relevant chapter should be referred to for the
Physical Review Letters 48, 1559 (1982), A. M. Chang et al., FQHE, so suffice it to mention that FQHE is
ibid 53, 997 (1984), R. Willett et al., ibid 59, 1776 (1987). regarded as an IQHE of composite fermions in the
202 Integer Quantum Hall Effect
35
30 N=1
hωc hωc
25 N=0 Zeeman splitting
Energy (meV)
20
15
e2/εl
10
5
gμBB
0
0 5 10 15 20
B (T)
Figure 32 Various energy scales (cyclotron, Coulomb, and Zeeman) against magnetic field, here plotted for GaAs. Inset
schematically shows the Landau levels with Zeeman splitting.
4
0.6 6
1.05.11 Recent Developments and
ρxx (kΩ/ )
ρxy (kΩ)
5
Related Phenomena
6
0.4 4
There are a multitude of recent developments in the
integer QHE and closely related physics. Extensions
are being made both for various classes of materials, 0.2 2
and various novel phenomena. Let us here briefly
mention them.
0.0 0
0 5 10 15
1.05.11.1 QHE in Oxides B (T)
Figure 33 QHE in an oxide heterostructure ZnO/
A notable breakthrough in the attempts to widen the
MgxZn1 xO. Reprinted figure with permission from A.
classes of materials is the realization of QHE in oxide Tsukazaki et al., Physical Review B 78, 233308 (2008).
heterostructures. The system is zinc oxide (ZnO), Copyright 2010 by the American Physical Society. For
which is an insulator, or a wide-gap semiconductor. information, see: http://link.aps.org/
Integer Quantum Hall Effect 203
(a) (b)
E
K K′
ky
kx
Figure 34 Crystal structure (a) and the band dispersion (b) of graphene.
204 Integer Quantum Hall Effect
length as usually defined (Equation (8)), and N ¼ 1, Soon after the fabrication of graphene samples, an
2,. . . (N ¼ 1, 2,. . .) correspond to pelectron
ffiffiffiffi (hole) anomalous IQHE was observed by Geim’s group, and
Landau levels. The Landau levels _ N is not uni- by Kim’s group (Novoselov et al., 2005; Geim and
formly spaced in sharp contrast to the usual, Novoselof, 2007; Castro Neto et al., 2009). As
uniformly spaced Landau levels, and, in particular, Figure 36 shows, the IQHE in graphene has
there is the N ¼ 0 Landau level right pffiffiffiat the Dirac e2
point E ¼ 0. The cyclotron energy _ B is also unu- xy ¼ – 2 ð2N þ 1Þ ð39Þ
h
sual. The N ¼ 0 Landau level is quite peculiar, which 2
e
is brought home by noting that the level is comple- as contrasted with the usual xy ¼ – 2 N when the
h
tely outside the Onsager’s semiclassical quantization carriers are filled up to the Nth Landau level. The
scheme in magnetic fields (because the Landau tube, prefactor of 2 in these equations is for the spin
which is the set of cylinders of varying radii for the degeneracy. The peculiarity of the Dirac cone
semiclassical quantization, cannot be defined around appears as the factor (2N þ 1) replacing the ordinary
the Dirac point). In fact, this level is an outcome of N as shown by Zheng and Ando, 2002; Gusynin and
a topological property of the massless Dirac cone, Sharapov, 2005. There are K and K9 points that
so that its presence is topologically protected. If we contribute equally to the Hall conduction, so that if
go back to the original honeycomb lattice in mag- we take out the factor of 2 along with the spin
netic fields, the problem becomes a Hofstadter e2
degeneracy, we can write xy ¼ – 4 ðN þ 1=2Þ,
butterfly for the honeycomb lattice (Figure 35), h
where, remarkably, a fraction (1/2) appears.
which was obtained by Rammal in 1985.
The situation around the N ¼ 0 Landau level is
Incidentally, it is heuristic to look at the Landau’s
indeed unusual. Let us compare the situation with the
quantization in the (massive) relativistic particles.
ordinary one where we have a (massive) conduction
From the Dirac equation for a Dirac particle
band and a (massive) valence band in Figure 37. In
in a magnetic field, the energy levels are
the latter case, we have an ordinary IQHE sequence,
EN ¼ [m2c4 þ mc2(h!c(2N þ 1 1))]1/2 (MacDonald,
0 ! 1 ! 2 ! . . ., for electrons (in the conduction band)
1983), (which contains the nonrelativistic limit,
and another, 0 ! 1 ! 2 ! . . ., for holes (in the
EN ¼ mc2 þ h!c(N þ 1 1/2)), as the leading term in
valence band). For a Dirac cone, the IQHE step
h!c/mc2 expansion). In the massless limit, m ! 0, we across N ¼ 0 at the Dirac point (i.e., an electron–
have to note that !c ¼ eH/mc also contains the mass hole symmetric point) cannot be 0 ! þ 1 nor
to recover the graphene Landau levels. 1 ! 0, which would be incompatible with the elec-
tron–hole symmetry. Instead, the step changes from
1/2 to þ1/2 at N ¼ 0. Another peculiarity is that
the quantum Hall step remains robust even at room
temperature (Figure 38), which is quite unusual.
We have mentioned that the quantum Hall num-
ber in usual systems is identified as a topological
quantum number according to Thouless et al. How
E is the topological number identified in the graphene
QHE? Hatsugai et al. (2006) have shown that we can
0 indeed algebraically calculate the Chern number in
the manner of Thouless et al. extended to the QHE
on the honeycomb lattice, which precisely gives
e2
xy ¼ – 2 ð2N þ 1Þ. The topological analysis also
h
leads to another topological character, namely edge
states (Figure 39) for a finite graphene in storng
mangetic fields appear (which should not be confused
with the graphene edge states in zero magnetic field,
B an interesting issue in its own right). We have men-
Figure 35 Energy spectrum against magnetic field for a tioned above that the bulk and edge QHE topological
honeycomb lattice. numbers are equal, that is, edge bulk
xy ¼ xy , which can
Integer Quantum Hall Effect 205
B = 14 T and T = 4K 7/2
5/2
10
3/2
σxy = (2N + 1) (e 2/h)
σxy (4e2/h)
ρxx (kΩ) 1/2
0
–1/2
5 ρxx
–3/2
–5/2
–7/2
0
–4 –2 0 2 4
n (1012 cm–2)
Figure 36 QHE observed in graphene. From K. S. Novoselov et al., Nature 438, 197 (2005).
σxy /(–e2/h)
2
σxy /(–e2/h)
3/2
(a) 1 (b)
1/2
0
–1
–1/2
–2
–3/2
vb cb
k k
E
E
–1 N = 0 1
Figure 37 Band dispersion and Landau levels (lower panels) and the QHE steps (upper) are compared between the
ordinary 2D system with valence and conduction bands (a) and graphene with k-linear, zero-gap dispersion (b).
20
ρxx (kΩ)
(a) (b)
E E
N=1
N=0
N=1
N=0
x x
Figure 40 Energy spectrum against the real-space position along the width for a finite sample is schematically compared
between the ordinary QHE system (a) and graphene QHE system (b).
E
B M
+q –q +q –q
Figure 41 Quantum Hall effect, anomalous Hall effect, and spin Hall effect are schematically shown.
Integer Quantum Hall Effect 207
from system’s magnetic structure, band structure and have an extra term that couples the real spin and
interactions, so they occur in 3D systems as well. the pseudospin, and this gives rise to an energy gap
The anomalous Hall effect, in which the Hall (a mass gap in the language of the Dirac theory).
resistivity is Since the gapped state cannot be reached by an
adiabatic change of the system, the insulator is called
RH ¼ Rnormal B þ Ranomalous M ð40Þ a topological insulator. Associated with the gap in the
bulk are the edge states, which are also dictated to
with Rnormal being the normal Hall resistivity,
exist and called topological edge states. In this
Ranomalous the anomalous Hall resistivity, and M
the magnetization of the material, was discovered respect, the QSHE is distinct from the spin Hall
in metallic ferromagnets in the nineteenth century, effect, and the whole situation is rather similar to
where the Hall current flows even in zero external the IQHE, where an essential difference is that the
magnetic fields. Theoretical analysis was initiated in edge states carry charges in IQHE while edge states
the 1950s by Karplus and Luttinger as a (multi-) carry spins in the QSHE. The energy dispersion
band effect in the presence of spin–orbit interaction. (Figure 42) of the edge states in a QSHE system,
To be more precise, two mechanisms have been which may be thought of as arising from Kramers’
identified, one (called extrinsic) is the impurity doublets, is gapless with the two dispersion crossing
scattering in the presence of spin–orbit interaction at a point in the Brillouin zone. The conductivity
and magnetization (skew scattering þ side jump) and calculated with the linear-response theory for each of
the other (called intrinsic) is Berry’s curvature con- the spin-up and spin-down electrons gives a quan-
tribution to the Hall conductivity as in the QHE, tized (Chern number) conductivity, where the
which is included in the linear-response formula. directions of the current are opposite between the
Experimentally, the anomalous Hall effect has up and down spins, with a quantized spin Hall con-
been observed in various materials, typically ductivity shown to be spin
xy ¼ e=2.
Nd2Mo2O7 with noncollinear spin configurations While the original proposal, due to Kane and
and (Sr, Ca)RuO compounds. Mele (2005) was made for graphene, the system has
In the spin Hall effect, which also occurs in zero turned out to have too small a spin–orbit interaction.
magnetic field, " spins and # spins flow in the oppo- Subsequently, the effect was shown to be realized in
site directions in an electric field (as opposed to the quantum-wells with narrow-gap semiconductors,
ordinary Hall effect in which opposite charges flow HgTe/CdTe. While both of HgTe and CdTe have
in the opposite directions in a magnetic field). This zincblende crystal structure, their band structures are
effect, which also comes from the spin–orbit interac- affected by the spin–orbit interactions which are sig-
tion, is another topological effect. The effect, nificant in these materials. Specifically, the bulk
predicted in the 1970s, is recently being experimen-
tally observed with, for example, Kerr rotation
microscopy in both n- and p-type GaAs-based semi- 1
conductor heterostructures.
QHE, anomalous Hall effect and spin Hall effect E/t
are related in that they are manifestations of the
phase of the wave functions in magneto-transport 0 I
properties. In a wide context, electrical polarization
is also formulated recently in terms of Berry’s
phase.
An intimate relation between QHE and the spin –1
Hall effect has been brought home by a more recent 0 π/a kx 2π/a
quantum spin Hall effect (QSHE) (König et al., 2008; Figure 42 Energy spectrum for a system that exhibits
Murakami, 2008) which was predicted originally for quantum spin Hall effect. Bunch of lines represent bulk
graphene. The idea starts as follows. Graphene is states, while a pair of isolated lines the edge states. Inset
described, as mentioned above, by a Dirac equation shows how the spin-up and spin-down electrons flow in real
space. Reprinted figure with permission from C. L. Kane and
for two spatial dimensions, where K and K9 points in E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). Copyright
the Brillouin zone are distinguished by a pseudospin 2010 by the American Physical Society. For information,
z. In the presence of a spin–orbit interaction, we see: http://link.aps.org/
208 Integer Quantum Hall Effect
0.4
ν=4 0.25
ρxy (h /e2)
ρxx (kΩ)
0.6
ν=5 0.2
ν=6
0.15
0.4 ν=7
0.1
0.2
0.05
0.0 0.0
0 1 2 3 4 5 6 7 8
B (T)
Figure 43 QHE in a HgTe quantum-well [after A. Pfeuffer-Jeschke et al., Physica B 256–258, 486 (1998).]
HgTe has a band structure called inverted, since a Ando T, Matsumoto Y, and Uemura Y (1975) Journal of the
Physical Society of Japan 39: 279.
band that usually forms the valence band lies above Ando T and Uemura Y (1974b) Journal of the Physical Society of
another band that usually forms the conduction band, Japan 36: 959.
where the former consists of two (heavy- and light- Aoki H (1983) Journal of Physics C 16: L205.
Aoki H (1986) Applied Physics Letters 48: 559.
mass) bands that touch with each other at a point in Aoki H and Ando T (1981) Solid State Communications 38: 1079
the Brillouin zone with opposite curvatures (i.e., the (reprinted in Aoki H and Ando T (1993) Solid State
bulk system is a semimetal). While IQHE itself Communications 88: 951).
Aoki H and Ando T (1985) Physical Review Letters 54: 831.
(Figure 43) became observable in HgTe/CdTe Aoki H and Ando T (1986) Physical Review Letters 57: 3093.
quantum-wells and superlattices grown by MBE, a Büttiker M (1992) In: Reed M (ed.) Nanostructured Systems,
recent breakthrough is the observation of the QSHE. 192. New York: Academic Press.
Castro Neto AH, et al. (2009) Reviews of Modern Physics, 81: 109.
The band structure behind this is, when HgTe is Chalker JT and Coddington PD (1988) Journal of Physics C
sandwiched between CdTe to make a quantum- 21: 2665.
well, the bands in HgTe well remain inverted for Fletcher R (1999) Semiconductor Science and Technology
14: R1.
wide wells, while the bands become normal for thin Geim AK and Novoselof KS (2007) Nature Materials 6: 183.
enough wells. So the well thickness can act to control Geisler MC, et al. (2004) Physical Review Letters 92: 256801.
the presence or otherwise of a mass gap, and the Gusynin VP and Sharapov S (2005) Physical Review Letters
95: 146801.
system should be a quantum spin Hall insulator Halperin BI (1982) Physical Review B 25: 2185.
above a critical thickness. With the spin Hall effect, Hashimoto K, et al. (2008) Physical Review Letters 101: 256802.
we can electronically manipulate the spin degrees of Hatsugai Y (1997) Journal of Physics: Condensed Matter
9: 2507.
freedom, so that applications to the spintronics is Hatsugai Y, et al. (2006) Physical Review B 74: 205414.
anticipated. (See Chapters 1.03, 1.06 and 1.07). Hofstadter DR (1976) Physical Review B 14: 2239.
Huckestein B (1995) Reviews of Modern Physics, 67: 357.
Igarashi T, Wakabayashi J, and Kawaji S (1975) Journal of the
Physical Society of Japan 38: 1549.
References Ikebe Y, et al. (2010) Physical Review Letters 104: 256802.
Ikushima K, et al. (2004) Physical Review Letters 93: 146804.
Ando T (1974) Journal of the Physical Society of Japan Kamimura H and Aoki H (1989) Physics of Interacting
36: 1521. Electrons in Disordered Systems. Oxford: Oxford University
Ando T (1974) Journal of the Physical Society of Japan 37: 1233. Press.
Ando T and Aoki H (1985) Journal of the Physical Society of Kane CL and Mele EJ (2005) Physical Review Letters
Japan 54: 2238. 95: 226801.
Ando T, Fowler AB, and Stern F (1982) Reviews of Modern Kawaji S (2008) Proceedings of the Japan Academy, Series B
Physics, 54: 437. 84: 199.
Integer Quantum Hall Effect 209
Kawaji S and Wakabayashi J (1981) In: Physics in High Magnetic Pruisken AMM (1990) In: Prange RE and Girvin SM (eds.)
Fields, Chikazumi S and Miura N (eds.) p. 284. Springer. The Quantum Hall Effect, 2nd edn., 117. New York:
Kinoshita T (2007) Progress of Theoretical Physics Supplement Springer.
167: 62. Středa P (1982) Journal of Physical: Condensed Matter
Kivelson S, et al. (1992) Physical Review B 46: 2223. 15: L718.
König M, Buhmann H, Molenkamp LW, et al. (2008) Journal of Thouless DJ, Kohmoto M, Nightingale P, and den Nijs M (1982)
the Physical Society of Japan 77: 031007. Physical Review Letters 49: 405.
Koshino M, et al. (2001) Physical Review Letters 86: 1062. Tsukazaki A, et al. (2008) Physical Review B 78: 233308.
Koshino M, et al. (2002) Physical Review B 65: 045310. von Klitzing K, Dorda G, and Pepper M (1980) Physical Review
Kramer B, Ohtsuki T, and Kettemann S (2005) Physics Reports Letters 45: 494.
417: 211. Widom A (1982) Physics Letters A 90: 474.
Landau LD, Lifshitz EM, and Pitaevskii LP (1984) Zheng Y and Ando T (2002) Physical Review B 65: 245420.
Electrodynamics of Continuous Media, 2nd edn., Pergamon.
Laughlin RB (1981) Physical Review B 23: 5632.
MacDonald AH (1983) Physical Review B 28: 2235.
Maksym PA, et al. (2000) Journal of Physics: Condensed Matter Further Reading
12: R299.
Mohr PJ, Taylor BN, and Newell DB (2008) Reviews of Modern Aoki H (1987) Reports on Progress in Physics 50: 655.
Physics 80: 633. Chakraborty T and Pietiläinen P (1988) The Fractional Quantum
Morgenstern M (2007) In: Scanning Probe Microscopy, Hall Effect. Springer.
pp. 349–371. New York: Springer. Das Sarma S and Pinczuk A (eds.) (1997) Perspective in
Morimoto T, et al. (2008) Physical Review B 78: 073406. Quantum Hall Effects. New York: Wiley.
Murakami S (2008) Progress of Theoretical Physics Supplement Landwehr G (1986) Festkörperprobleme 26: 17.
176: 279. Prange RE and Girvin SM (eds.) (1990) The Quantum Hall Effect,
Nachtwei G (1999) Physica E 4: 79. 2nd edn. New York: Springer.
Novoselov KS, et al. (2005) Nature 438: 197. Yoshioka D (2002) The Quantum Hall Effect. Springer.
1.06 Fractional Quantum Hall Effect and Composite
Fermions
J K Jain, The Pennsylvania State University, University Park, PA, USA
ª 2011 Elsevier B.V. All rights reserved.
Glossary
Composite fermions Bound states of electrons Hall effect Observation of a voltage transverse to
and quantized vortices. the direction of the current flow in the presence of a
Degeneracy Multiplicity of an energy level. magnetic field.
Fermi sea The state in which fermions occupy all Integral quantum Hall effect Observation of
states below certain energy. plateaus in the Hall resistance at quantized values
Filling factor Number of filled Landau levels. characterized by integers.
Fractional quantum Hall effect Observation of Landau levels Quantized kinetic energy levels of
plateaus in the Hall resistance at quantized values electrons in a magnetic field.
characterized by simple fractions.
210
Fractional Quantum Hall Effect and Composite Fermions 211
1.06.1 Introduction
B
Collective quantum behavior is of fundamental
interest in condensed matter physics, where a collec- I
tion of interacting particles behaves in a surprising
manner that would be difficult to guess from the I
knowledge of the properties of single particles.
Superconductivity and superfluidity are two such VL
VH
well-known examples. During the last three decades,
a new collective quantum fluid state of matter has
been discovered and studied, which occurs when
electrons are confined to two dimensions, cooled to Figure 1 Schematics of magnetotransport measurements.
near absolute zero temperature, and subjected to a I, VL, and VH are the current, longitudinal voltage, and the Hall
strong magnetic field. The most dramatic, and unex- voltage, respectively. The longitudinal and Hall resistances
pected, manifestation of this phase is the ‘fractional are defined as RL X VL/I and RH X VH/I.
quantum Hall effect’ (Tsui et al., 1982). The fractional
quantum Hall effect (FQHE) and many other
The phenomenon has a classical origin. The
properties of this state result from the formation of
Lorentz equation of electrodynamics
a new class of topological particles, called ‘composite
fermions’, which are bound states of electrons 1
F¼q Eþ vB ð3Þ
and quantized microscopic vortices. The composite- c
fermion quantum fluid provides a new paradigm for
gives the force on a particle of charge q, moving with
collective behavior. This chapter describes the essen-
a velocity v, in the presence of electric and magnetic
tial phenomenology of the fractional quantum Hall
fields. A consequence of this equation is that for
effect, how the composite fermion theory resolves it,
crossed electric and magnetic fields, say E ¼ Eŷ and
and other consequences of the existence of composite
B ¼ Bẑ, the charged particle drifts in a direction
fermions.
perpendicular to the plane containing the two fields,
with a velocity v ¼ c(E/B)x̂.
Current density is given by J ¼ qv, where is
the (three-dimensional) density of particles. That
1.06.2 Quantum Hall Effect produces the Hall resistivity
Phenomenology
Ey B
1.06.2.1 The Hall Effect H ¼ ¼ ð4Þ
Jx qc
The Ohm’s law is given by
The Hall measurement is used routinely to measure
J ¼ E ð1Þ the density of the mobile charges, as well as the sign
of the charge carriers (i.e., whether they are electrons
where is the conductivity, and J ¼ qv is the cur- or holes).
rent density for particles of charge q and density
moving with a velocity v. The more familiar I ¼ V/R
can be obtained from this local relation. In the pre- 1.06.2.2 Integral Quantum Hall Effect
sence of a magnetic field, the electron current flows
When the Hall experiments are performed in a sys-
in a direction perpendicular to the plane containing
tem in which electrons are confined to two
the electric and the magnetic fields. Alternatively,
dimensions (such confinement is achieved, for exam-
passage of a current induces a voltage perpendicular
ple, by constructing AlGaAs–GaAs quantum-wells to
to the direction of the current flow, called the Hall
confined motion in one dimension), the behavior
voltage, VH, as seen in Figure 1. In addition to the
changes in a qualitative manner, as shown in
usual (longitudinal) resistance, R ¼ VL/I, a new resis-
Figure 2, taken from Stormer (1999). Instead of
tance, called the Hall resistance, is defined as
being proportional to B, now Hall resistance shows
VH plateaus. Most prominent are plateaus on which the
RH ¼ ð2Þ
I Hall resistance is precisely quantized at
212 Fractional Quantum Hall Effect and Composite Fermions
43 2
h
1 RH ¼ ð8Þ
fe 2
1
1 eA 2 H ¼ ay a þ ð22Þ
H¼ pþ ð9Þ 2
2mb c
This is the familiar Hamiltonian of a simple harmo-
Here, e is defined to be a positive quantity, the
nic oscillator.
electron’s charge being e, and mb is the band mass
The angular momentum operator is defined as
of the electron. The canonical momentum is defined
as p ¼ ihÑ, and A satisfies
L ¼ – h b y b – ay a ð23Þ
Ñ A ¼ B ẑ ð10Þ
Exploiting the property [H,L] ¼ 0, the eigenfunctions
The Schrödinger equation are chosen to diagonalize H and L simultaneously.
The analogy to the Harmonic oscillator problem
H ¼ E ð11Þ
immediately gives the solution
can be solved in several gauges, but the most con-
venient for the fractional quantum Hall effect is the H jn; mi ¼ En jn; mi ð24Þ
symmetric gauge:
1
En ¼ n þ ð25Þ
Br B 2
A¼ ¼ ð – y; x; 0Þ ð12Þ
2 2
ðb y Þmþn ðay Þn
Choosing the unit of length as the magnetic length jn; mi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffi j0; 0i ð26Þ
ðm þ nÞ! n !
1=2
hc where n is called the Landau level index, and
,¼ ð13Þ
eB m ¼ n, n þ 1,. . . is the angular momentum quan-
and the unit of energy as the cyclotron energy tum number. The single-particle orbital at the
bottom of the two ladders defined by the two sets of
eB raising and lowering operators is
h!c ¼ h ð14Þ
mb c
1 1
the Hamiltonian can be expressed as hrj0; 0i X 0;0 ðrÞ ¼ pffiffiffiffiffi e – 4zz ð27Þ
2
which satisfies
1 q2 1 q q
H¼ –4 z – z þ z
þ z ð15Þ
2 qzqz 4 qz q
z aj0; 0i ¼ b j0; 0i ¼ 0 ð28Þ
Here we have defined new variables z and z: The single-particle states are especially simple in
the lowest Landau level (n ¼ 0):
z ¼ x – iy ¼ r e – i ; z ¼ x þ iy ¼ r ei ð16Þ
1
We further define the following sets of ladder ðb y Þm zm e – 4z z
0;m ¼ hrj0; mi ¼ pffiffiffiffiffi 0;0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð29Þ
operators: m! 22m m !
1 z q where we have used Equations (18) and (27). Aside
b ¼ pffiffiffi þ2 ð17Þ from the ubiquitous Gaussian factor, a general state
2 2 qz
in the lowest Landau level is simply given by a
1 z q polynomial of z. In other words, apart from the
b y ¼ pffiffiffi –2 ð18Þ
2 2 qz Gaussian factor, the lowest Landau level wave func-
tions are analytic functions of z.
y 1 z q
a ¼ pffiffiffi –2 ð19Þ
2 2 qz
1 z q
a ¼ pffiffiffi þ2 ð20Þ 1.06.3.1 Landau-Level Degeneracy
2 2 q
z
which have the property that States with a given n but different m are degener-
ate. To obtain the degeneracy of a Landau level,
a; ay ¼ 1; b; b y ¼ 1 ð21Þ consider a disk of radius R centered at the origin,
and ask how many states lie inside it. The degen-
and all the other commutators are zero. In terms of eracy can be shown to be independent of the
these operators, the Hamiltonian can be written as Landau level. Taking, for simplicity, the lowest
214 Fractional Quantum Hall Effect and Composite Fermions
of electrons at an integral filling factor is very action is 10710 . Further, there is no small parameter
simple: it contains a unique ground state contain- in the theory, which prevents a meaningful perturba-
ing an integral number of filled Landau levels, tion theory.
separated from excitations by the cyclotron or
the Zeeman energy gap. (In other words, the
state is incompressible, because to compress the 1.06.6 Composite Fermions: Basic
ground state creates finite energy excitations.) Foundations
It should be noted that the detailed explanation
of the existence of the plateaus also requires a With the observation of many fractions, an analogy
consideration of disorder-induced Anderson loca- between the FQHE and the IQHE could be identi-
lization of some states. fied, which led to the postulation of composite
Fractional Quantum Hall Effect and Composite Fermions 215
fermions, and an explanation of the FQHE as the argued that these correlations survive projection into
IQHE of composite fermions (Jain, 1989). The most the lowest Landau level. Q
2p
important accomplishment of the composite fermion The Jastrow factor j < k(zj zk) attaches 2p
(CF) theory is the identification of the particles of the quantized vortices to each electron in . The
FQHE which can be considered as weakly interact- bound state of an electron and 2p vortices is inter-
ing to a good approximation. The CF theory can be preted as a particle called the composite fermion. As
motivated by a Chern–Simons mean field theory composite fermions move about, the vortices bound
(Jain, 1989; Lopez and Fradkin, 1991; Halperin et al., to them produce Berry phases, which cancel part of
1993), but its final outcome can be stated succinctly. the Aharonov Bohm phases originating from the
The eigenfunctions and eigenenergies for the ground external magnetic field. The effective magnetic
and (low-energy) excited states of strongly interact- field can be determined in a Berry-phase calculation.
ing electrons at an arbitrary lowest Landau level When a composite fermion, that is an electron along
filling are expressed in terms of the known solu- with its vortices, is taken in a closed loop enclosing an
tions of the noninteracting electron problem at the area A, it acquires a Berry phase
Landau level filling as follows:
Y
2p BA
¼ – 2 þ 22pNenc ð38Þ
¼ P LLL zj – zk ð34Þ 0
j <k
where Nenc is the number of composite fermions inside
and the loop. The first term is the familiar Aharonov Bohm
D P E
(AB) phase due to a charge going around in a loop.
j <k r1
E ¼
jk
þ Vel – bg þ Vbg – bg ð35Þ The second is the Berry phase due to the 2p vortices
h j i going around Nenc particles, with each particle con-
where tributing a phase of 2. For uniform density states, we
replace Nenc in Equation (38) by its average value A
and equate the entire phase to the AB phase due to an
¼ ð36Þ
2p 1 effective magnetic field, B, to obtain Equation (37).
Composite fermions thus sense a magnetic field B
B ¼ B – 2p0 ð37Þ
that is much smaller than the applied magnetic field,
Here, are the eigenfunctions of noninteracting and can even be zero or negative. This property of
electrons at , P LLL projects the wave function to its composite fermions distinguishes them from electrons,
right into the lowest Landau level, p is an integer, and and lies at the root of most of the phenomenology.
B is an effective magnetic field. These equations Composite fermions form Landau-like levels in the
define a one-to-one correspondence between the reduced magnetic field, which are called levels, and
ground and excited states at filling factor (or mag- occupy of them.
netic field B) and (magnetic field B). The problem Composite fermions represent a new class of par-
is often much simpler at , and for ¼ n (where n is ticles realized in nature. A vortex is a topological
an integer), we know the exact solution for the object, because the quantum mechanical phase asso-
ground and low energy states. The wave function ciated with a closed loop around a vortex is exactly 2,
for the ground state at ¼ 1/(2p þ 1) was known independent of the shape and the size of the loop.
previously (Laughlin, 1983). (The topological character of vortices is implicit in
Electrons capture vortices to turn into composite the fact that we count them.) As a result, composite
fermions because this is the most effective way for fermions are collective, topological particles. All fluids
them to stay away from one another. A typical wave of composite fermions are thus topological quantum
function satisfying the Pauli principle vanishes as fluids. The most direct consequence of the topological
r when two particles approach one another, r being quantization of the vorticity of composite fermions is
the distance separatingQ them, but the unprojected the effective magnetic field, which is responsible for
wave functions j < k(zj zk)2p vanish as r2p þ 1, the FQHE and several other phenomena.
with the Jastrow factor contributing 2p to the expo- The Chern–Simons field theoretic formulation of
nent and contributing the rest. The Jastrow factor composite fermions (Jain, 1989; Lopez and Fradkin,
is very effective in keeping particles apart from one 1991; Halperin et al., 1993) proceeds through a sin-
another, producing favorable correlations. It can be gular gauge transformation defined by
216 Fractional Quantum Hall Effect and Composite Fermions
!2p
Y z –z 1.06.7 Consequences
¼ j k 9 ð39Þ
z j – z k 1.06.7.1 Theory of Fractional Quantum Hall
j <k
Effect
under which the eigenvalue problem transforms
into The FQHE of electrons is a manifestation of the
integral QHE for composite fermions. The latter
H 99 ¼ E9 ð40Þ occurs because a gap opens when composite fermions
" # fill an integral number of levels, that is, when
1 X e e
2
H9 ¼ pi þ Aðri Þ – aðri Þ þV ð41Þ ¼ n. These fillings correspond to electron filling
2mb i c c
factors given by the sequences:
2p X 9 n
aðri Þ ¼ 0 Ñi ij ð42Þ ¼ ð47Þ
2 j
2pn 1
The above transformation thus amounts to attaching These fractions can be obtained by defining the
a point flux of strength 2p0 to each electron, which original problem in terms of holes – rather than
is how the composite fermion is modeled in this electrons – in the lowest Landau level, and making
approach. composite fermions by binding vortices to holes.
To proceed further, one makes a mean-field The composite fermion (CF) theory provides
approximation, which amounts to spreading the a natural explanation of many experimental facts.
point flux on each electron into a uniform magnetic The fractions f given by Equations (48) and (49)
field. Formally, one writes are precisely the prominently observed fractions.
Furthermore, the fractions appear in sequences
A – a X A þ
A ð44Þ because they are all derived from the integer sequence
of the IQHE. Some of the experimental sequences are:
Ñ A ¼ B ẑ ð45Þ
f ¼ 1=3; 2=5; 3=7; 4=9; . . . ; 10=21 ð50Þ
The transformed Hamiltonian then becomes
f ¼ 2=3; 3=5; 4=7; . . . ; 10=19 ð51Þ
1 X e
2
f ¼ 1=5; 2=9; . . . ; 6=25 ð52Þ
H9 ¼ pi þ A ðri Þ þV þ V 9 ¼ H0 þ V þ V 9
2mb i c
The fractions have odd denominators because the
ð46Þ
vortex quantum number 2p is an even integer.
where V is the Coulomb interaction and V 9 denotes Different flavors of composite fermions (i.e., com-
the terms containing
A. The solution to H0 posite fermions carrying different numbers of
describes free composite fermions in an effective vortices) occur in different filling factor regions.
magnetic field B , and V þ V 9 is the effective inter- The filling factor range 1/2 1/3 is described
action between them, which is to be treated in terms of composite fermions carrying two vortices;
perturbatively. When one transforms this problem the range 1/3 > 1/5 in terms of composite fer-
into the field theoretical Langrangian, it is seen to mions carrying four vortices (with B antiparallel to
be equivalent to the familiar Chern–Simons theory. B for 1/3 > > 1/4); and so on. The region 1 > 1/2
This approach is believed to capture the topological of electrons maps into 0 < 1/2 of holes as a result
properties of composite fermions, but does not lend of particle–hole symmetry in the lowest Landau level,
itself to a systematic perturbative treatment because and can be understood in terms of composite fermions
of the lack of a small parameter. made of holes.
Fractional Quantum Hall Effect and Composite Fermions 217
1.06.7.2 Quantitative Tests against Exact and 3/7 (the first, second, and third rows, respec-
Results tively) are shown in Figure 3 (dashes). This figure
also shows the predicted energies from the CF the-
Exact results can be obtained for the Hamiltonian in
ory, without any adjustable parameters, as dots. The
Equation (32) for a small number of particles by a
ground state corresponds to one, two, or three filled
brute force diagonalization, because the dimension of
the Fock space in the lowest Landau level is finite for levels of composite fermions, the the lowest energy
a finite system. Typically, depending on the filling branch of excitations is a particle–hole pair of com-
factor, 10–16 electrons can thus be studied. The exact posite fermions. The predicted energies agree to
solution gives the eigenenergies and eigenfunctions within 0.1%, and the overlaps between the exact
for all eigenstates. The CF explanation of the FQHE and the CF wave functions are greater than 99%
is fully confirmed by comparison to exact results for the numerical systems. Furthermore, the states
(Dev and Jain, 1992; Wu et al., 1993; Jain and in between the special fractions n/(2pn 1) are
Kamilla, 1998) well described in terms of composite fermions at a
A convenient geometry is the spherical geometry, nonintegral filling factor.
in which electrons move on the surface of a sphere, Monte Carlo methods allow determination of the
with a radial magnetic field produced by a magnetic thermodynamic limits of various experimentally
monopole of strength Q at the center, which produces measurable quantities, such as excitation gaps.
a total magnetic flux of 2Q0. The eigenstates are These are often in 20–50% agreement with experi-
conveniently labeled by the total orbital angular ment, with the discrepancy caused primarily by
momentum L. The exact spectra at ¼ 1/3, 2/5, disorder. An important quantity is collective modes,
101
111
168
175
227
230
277
31
42
–0.41
–0.42
E (e2/εl)
–0.43
–0.45
179
212
304
328
416
52
83
21
22
35
33
8
8
–0.44
–0.45
E (e2/εl)
–0.46
–0.47
–0.49
1182
127
263
493
621
952
744
21
22
35
33
45
8
8
–0.46
–0.47
E (e2/εl)
–0.48
–0.49
–0.51
0 3 6 9 12 0 3 6 9 12
L L
Figure 3 Comparison of spectra obtained from exact diagonalization (dashes) and CF theory (dots). The spectra in the three
rows are for ¼ 1/3, 2/5, and 3/7, respectively. The x-axis label L is the total orbital angular momentum of the state. From Jain
JK and Kamilla RK (1998) Composite fermions: Particles of the lowest Landan level. In: Heinonen O (ed.) Composite Fermions,
ch. 1. New York: World Scientific.
218 Fractional Quantum Hall Effect and Composite Fermions
which are understood as excitons of composite fer- found between the low energy spectra of the exact
mions; the experimental measurements of the solutions of interacting electrons at B (from numer-
energies and the dispersions (Pinczuk et al., 1993; ical diagonalization) and the exact solutions of
Kukushkin et al., 2009) at various fractions such as noninteracting electrons at B. In addition, several
1/3, 2/5, 3/7, and 4/9 are in excellent agreement experiments have measured the radius of the cyclo-
with the predictions of the composite fermion theory tron orbit of the current carrying entities in the
(Scarola et al., 2000). vicinity of ¼ 1/2, and confirmed that it is deter-
mined by B rather than the applied magnetic field
1.06.7.3 Composite Fermion Fermi Sea (Willet et al., 1993; Kang et al., 1993; Goldman et al.,
1994; Smet et al., 1996).
No general principle excludes FQHE at even-
denominator fractions. Such FQHE has been
observed, for example, at f ¼ 5/2. The CF theory
provides a natural explanation for why even denomi- 1.06.7.5 Spin Physics
nator FQHE is rare: the model of noninteracting The spin degree of freedom is frozen when the
composite fermions produces only odd-denominator Zeeman splitting is large compared to the interaction
fractions; any even-denominator fraction must neces- energy. In that limit, all FQHE states in the lowest
sarily owe its existence to weak residual interactions Landau level are fully spin polarized. One might
between composite fermions, and, therefore, can be expect, by application of the Hund’s maximum spin
expected to be much weaker. rule to electrons in the lowest Landau level, that the
The nature of state at ¼ 1/2, the simplest frac- state would be fully polarized at all Zeeman energies.
tion, has been of interest. It is obtained as the n ! 1 That is not the case, however. The actual state is
limit of the sequence f ¼ n/(2n þ 1). If the model of determined by application of the Hund’s rule to
noninteracting composite fermions continues to be composite fermions.
valid in this limit, their effective magnetic field B In a model that assumes that composite fermions
vanishes and they form a Fermi sea, called the CF can be taken as nearly independent particles with an
Fermi sea (Halperin et al., 1993; Kalmeyer and Zhang, effective mass, their physics is straightforward. The
1992). The lack of FQHE follows because the Fermi IQHE of composite fermions occurs at ¼ n ¼ n" þ
sea has no gap to excitations. Several experiments n#, where n" is the number of occupied spin-up
have confirmed the formation of composite fermions
levels and n# is the number of occupied spin-down
in the vicinity of ¼ 1/2 and of the CF Fermi sea at
levels. This fraction corresponds to electron fillings
¼ 1/2. These include. Shubnikov de Haas oscilla-
tions of composite fermions (Du et al., 1994), linear n n" þ n#
opening of the CF gap (Du et al., 1993); measurement ¼ ¼ ð53Þ
2pn 1 2pðn" þ n# Þ 1
of the cyclotron resonance of composite fermions
(Kukushkin et al., 2007); and measurements of the The spin polarization of the state is given by
CF Fermi wave vector through the semiclassical n" – n#
e ¼ ð54Þ
cyclotron orbit of composite fermions (Willet et al., n" þ n#
1993; Kang et al., 1993; Goldman et al., 1994; Smet
and wave function for the FQHE state is
et al., 1996)
Y
2p
2pnþ1
n ¼P LLL n" ;n# zj – zk ð55Þ
j <k
1.06.7.4 Effective Magnetic Field
or
The reduced effective magnetic field, which is a
Y
2p
direct evidence of binding of vortices to electrons 2pnþ1 ¼P
zj – zk ð56Þ
n LLL n" ;n#
and the formation of composite fermions, has been j <k
confirmed directly by numerous means. The appear-
ance of fractional sequences that correspond to the where
integer sequence of the noninteracting fermions and n" ;n# ¼ n" n# ð57Þ
the formation of a Fermi sea at 1/2 filled Landau
level are experimental proofs of the effective mag- are wave functions of IQHE states with n" spin-up
netic field, as is the one-to-one correspondence and n# spin-down Landau levels occupied.
Fractional Quantum Hall Effect and Composite Fermions 219
where Pf stands for Pfaffian. The Pfaffian of an anti- (Laughlin, 1983), and has been measured in shot
symmetric matrix M is defined, apart from an overall noise experiments (de Picciotto et al., 1997).
normalization factor, as
Pf Mij ¼ AðM12 M34 MN – 1;N Þ ð64Þ
1.06.8 Open Issues
where A is the antisymmetrization operator. The
usual Bardeen–Cooper–Schrieffer wave function for The FQHE and other phenomena in the lowest
fully polarized electrons can be written as Landau level are well described by the CF theory.
BCS ¼ A½0 ðr1 ; r2 Þ0 ðr3 ; r4 Þ 0 ðrN – 1 ; rN Þ
ð65Þ However, the CF theory in its simplest form is not
fully satisfactory for the second Landau-level
which is a Pfaffian. (The fully symmetric spin part is physics. One difference between the two Landau
1 levels is indicated by the observation that while
not shown explicitly.) Analogously, Pf ¼
zi – zj there is no FQHE at ¼ 1/2, an FQHE occurs at
describes a p-wave pairing of electrons (p-wave ¼ 5/2. Further, far fewer fractions are observed in
because the system is fully spin-polarized), and the second Landau level, and they are much more
Pf
1=2 is interpreted as the p-wave paired state of delicate than those in the lowest Landau level. While
composite fermions carrying two vortices. The other fractions in the second Landau level also have
Pfaffian wave function was originally motivated in a odd denominators (such as 2 þ 1/3 and 2 þ 2/5),
conformal field theory approach, which also suggests their wave functions are rather different than those
that its quasiparticles obey so-called nonabelian of the corresponding fractions in the lowest Landau
braiding statistics (Moore and Read, 1991). level. Several imaginative approaches are currently
being pursued, which have excitations with exotic
character (e.g., nonabelian braiding properties).
(See Chapters 1.03, 1.05 and 1.07).
1.06.7.7 Fractional Charge
The excitations of an FQHE state are excited com-
posite fermions. A sole composite fermion in an References
otherwise empty level is often called a CF quasi-
Chang C-C and Jain JK (2004) Microscopic origin of the next-
particle, and a missing composite fermion from an generation fractional quantum Hall effect. Physical Review
otherwise filled level is called a CF quasihole. This Letters 92: 196806.
description has been confirmed extensively in exact de Picciotto R, Rezhnikov M, Heiblum M, Umansky V, Bunin G,
and Mahalu D (1997) Direct observation of a fractional
diagonalization studies. In a localized representation, charge. Nature 389: 162.
these represent a localized excess or deficiency of Dev G and Jain JK (1992) Band structure of the fractional
charge relative to the uniform ground state. The quantum Hall effect. Physical Review Letters 69: 2843.
Du RR, Stormer HL, Tsui DC, Pfeiffer LN, and West KW (1993)
charge excess associated with a CF quasiparticle is Experimental evidence for new particles in the fractional
the sum of the charge of an electron (e) and the quantum Hall effect. Physical Review Letters 70: 2944.
charge of 2p vortices: Du RR, Stormer HL, Tsui DC, Pfeiffer LN, and West KW (1994)
Shubnikov –de Haas oscillations around ¼ 1/2 Landau level
filling factor. Solid State Communications 90: 71.
– e ¼ – e þ 2pev ð66Þ
Du RR, Yeh AS, Stormer HL, Tsui DC, Pfeiffer LN, and West KW
(1995) Fractional quantum Hall effect around ¼ 3/2:
where ev is the charge of a single vortex. The charge
Composite fermions with a spin. Physical Review Letters
of the vortex can be shown to be 75: 3926.
Goldman VJ, Su B, and Jain JK (1994) Detection of composite
ev ¼ ve ð67Þ fermions by magnetic focusing. Physical Review Letters
72: 2065.
The local charge of a CF-quasiparticle at Halperin BI, Lee PA, and Read N (1993) Theory of the half-filled
¼ n/(2pn þ 1) therefore has the fractional value Landau level. Physical Review B 47: 7312.
Jain JK (1989) Composite-fermion approach for the fractional
e quantum Hall effect. Physical Review Letters 63: 199.
– e ¼ – ð68Þ Jain JK and Kamilla RK (1998) Composite fermions: Particles of
2pn þ 1
the lowest Landan level. In: Heinonen O (ed.) Composite
The topological quantization of the vorticity implies Fermions, ch. 1, pp. 1–90. New York: World Scientific.
Kalmeyer V and Zhang SC (1992) Metallic phase of the quantum
that this charge is precisely quantized. The fractional Hall system at even-denominator filling fractions. Physical
charge was originally predicted theoretically Review B 46: 9889.
Fractional Quantum Hall Effect and Composite Fermions 221
Kang W, Stormer HL, Pfeiffer LN, Baldwin KW, and West KW Stormer HL (1999) Nobel lecture: The fractional quantum Hall
(1993) How real are composite fermions? Physical Review effect. Reviews of Modern Physics 71: 875–889.
Letters 71: 3850. Tsui DC, Stormer HL, and Gossard AC (1982) Two-dimensional
Kukushkin, IV, Smet JH, Scarola VW, Umansky V, and von magnetotransport in the extreme quantum limit. Physical
Klitzing K (2009) Dispersion of the excitations of fractional Review Letters 48: 1559.
quantum Hall states. Science 324: 1044. von Klitzing K, Dorda G, and Pepper M (1980) New method for
Kukushkin IV, Smet JH, Schuh D, Wegscheider W, and von high-accuracy determination of the fine-structure constant
Klitzing K (2007) Dispersion of the composite-fermion cyclotron based on quantized Hall resistance. Physical Review Letters
resonance mode. Physical Review Letters 98: 066403. 45: 494.
Kukushkin, IV, von Klitzing K, and Eberl K (1999) Spin Willett RL, Eiscnstein JP, Stormer HL, Tsui DC, Gossard AC,
polarization of composite fermions: Measurements of the and English JH (1987) Observation of an even-denominator
fermi energy. Physical Review Letters 82: 3665. quantum number in the fractional quantum Hall effect.
Laughlin RB (1981) Quantized Hall conductivity in two Physical Review Letters 59: 1776.
dimensions. Physical Review B 23: 5632. Willett RL, Ruel RR, West KW, and Pfeiffer LN (1993)
Laughlin RB (1983) Anomalous quantum Hall effect: An Experimental demonstration of a Fermi surface at one-half
incompressible quantum fluid with fractionally charged filling of the lowest Landau level. Physical Review Letters
excitations. Physical Review Letters 50: 1395. 71: 3846.
Lopez A and Fradkin E (1991) Fractional quantum Hall effect and Wójs A, Yi K-S, and Quinn JJ (2004) Fractional quantum Hall
Chern–Simons gauge theories. Physical Review B 44: 5246. states of clustered composite fermions. Physical Review B
Moore G and Read N (1991) Nonabelions in the fractional 69: 205322.
quantum Hall effect. Nuclear Physics B 360: 362. Wu XG, Dev G, and Jain JK (1993) Mixed-spin incompressible
Pan W, Stormer HL, Tsui DC, Pfeiffer LN, Baldwin KW, and states in the fractional quantum Hall effect. Physical Review
West KW (2003) Fractional quantum Hall effect of composite Letters 71: 153.
fermions. Physical Review Letters 90: 016801.
Park K and Jain JK (1998) Phase diagram of the spin
polarization of composite fermions and a new effective
mass. Physical Review Letters 80: 4237.
Pinczuk A, Dennis BS, Pfeiffer LN, and West KW (1993)
Observation of collective excitations in the fractional Further Reading
quantum Hall effect. Physical Review Letters 70: 3983.
Scarola VW, Park K, and Jain JK (2000) Rotons of composite Das Sarma S and Pinczuk A (eds.) (1996) Perspectives in
fermions: Comparison between theory and experiment. Quantum Hall Effects. New York: Wiley.
Physical Review B 61: 13064. Heinonen O (ed.) (1998) Composite Fermions. New York: World
Smet JH, Weiss D, Blick RH, et al. (1996) Magnetic focusing of Scientific.
composite fermions through arrays of cavities. Physical Jain JK (2007) Composite Fermions. London: Cambridge
Review Letters 77: 2272. University Press.
1.07 Spin Hall Effect
S Murakami, Tokyo Institute of Technology, Tokyo, Japan
N Nagaosa, University of Tokyo, Tokyo, Japan
ª 2011 Elsevier B.V. All rights reserved.
1.07.1 Introduction – Early History and Background of Spin Hall Effect (SHE) 224
1.07.2 Intrinsic and Extrinsic Mechanism of SHE 227
1.07.2.1 Intrinsic SHE in p-Type Semiconductors 228
1.07.2.2 Rashba Model for n-Type Semiconductors 230
1.07.2.3 Impurity Scattering Effect 231
1.07.2.3.1 Spin-independent impurities 231
1.07.2.3.2 SO coupling modified by impurities 235
1.07.2.3.3 SHE in the hopping regime 236
1.07.2.4 Mesoscopic SHE 237
1.07.2.5 Experimental Observation of SHE in Semiconductors 239
1.07.2.5.1 Measurement of SHE in semiconductors by KR 239
1.07.2.5.2 Measurement of SHE in semiconductors by spin LED 241
1.07.2.6 Various Issues on the SHE 243
1.07.2.6.1 Definition of the spin current 243
1.07.2.6.2 Ab initio calculations 245
1.07.3 SHE in Metals 245
1.07.3.1 Experiments of SHE in Metals 245
1.07.3.2 Theories of SHE in Metals 251
1.07.4 Quantum Spin Hall Effect 253
1.07.4.1 Inverted Band Structure and SHI 253
1.07.4.2 QSHE and Z2 Topological Invariants in 2D and 3D 254
1.07.4.3 Helical Edge Modes in QSH systems 257
1.07.4.4 Models and Candidate Materials for QSHE 258
1.07.4.5 Experimental Observation of QSHE 259
1.07.4.6 Further Developments in the QSHE 261
1.07.4.6.1 Spin-charge separation and charge fractionalization 261
1.07.4.6.2 Effective theory and quantized magneto-electric effect 262
1.07.4.6.3 Superconducting QSHSs 262
1.07.4.6.4 Localization problem related to the QSHS 263
1.07.4.6.5 SHE in strongly correlated systems 264
1.07.5 Summary and Perspectives on SHE 264
1.07.5.1 Comparison between AHE and SHE 264
1.07.5.2 Comparison between Integer QHE and QSHE 265
1.07.5.3 Generalization of SHE and Future Directions 266
1.07.5.3.1 Optical SHE 266
1.07.5.3.2 Multiferroics 267
1.07.5.3.3 Future directions 268
References 269
Glossary
Aharonov–Bohm effect The wave function of a potential along the path, leading to the change in the
charged quantum mechanical particle acquires a interferenceofthedifferencepathsduetothemagnetic
phase given by the contour integral of the vector flux. This effect is called Aharonov–Bohm effect.
222
Spin Hall Effect 223
Aharonov–Casher effect Analogously to the Hanle effect In 1924, Hanle showed the variation
Aharonov–Bohm effect, the wave function of a of the polarization of the resonant fluorescent light
quantum mechanical particle with spin acquires a in atoms, which is called the Hanle effect. This
phase factor which is given by the contour integral effect is now used to measure the lifetime and
of the vector product of the electric field and the g-factor of atoms and molecules.
spin. This effect is called Aharanov–Casher effect. Keldysh formalism L. Keldysh developed a
Angle-resolved photoemission spectroscopy Green’s function method which is applicable even
(ARPES) The momentum- and energy-resolved to states far away from the equilibrium. This
spectroscopy of the emitted electrons from the method is called the Keldysh formalism, and is now
sample kicked by the high-energy incident light. applied in various transport problems.
This can detect the spectral function of the Green’s Kondo effect When a magnetic impurity is
function for occupied states. inserted into a metallic host, its magnetic moment
Berry phase In 1984, M.V. Berry formulated the is screened by the spins of the conduction
quantum phase associated with the adiabatic electrons to form a spin singlet. This phenomenon
change of the Hamiltonian. Through the adiabatic is called the Kondo effect.
change, the wave function is confined within the Kramers theorem In the system with the odd
subspace of the Hilbert space, which leads to the number electrons, it is proved that the electronic
gauge structure and is represented by this Berry states are energetically (at least doubly) degenerate
phase. when the Hamiltonian has the time-reversal
BW phase One of the p-wave pairing states for symmetry. The doublet related by the time-reversal
superfluid with the isotropic energy gap. It is now symmetry operation is called Kramers doublet.
established that the B-phase of 3He is described by Kubo formula In 1957, R. Kubo formulated the
this state. linear response theory based on the density matrix
Chiral edge channel In quantum Hall system, formalism. This provides the most fundamental
there appear edge channels at the edges of the formula of the response functions, and basis to
sample, which have the definite propagation study the physical properties of systems.
direction and are called chiral edge channels. The Light-emitting diode (LED) The light is emitted
backward scattering is forbidden in the chiral edge through the recombination of the electrons and
channels, which explains the quantized Hall holes when the current is applied into the p–n
conductance in quantum Hall system. junction.
Coulomb blockade In the tunnel junction with Magnetic Kerr rotation The rotation of the
small capacitance, the Coulomb charging energy polarization occurs when the light is reflected from
can be larger than the thermal energy, and the the magnetic system due to the relativistic
tunneling process is blocked. This Coulomb spin–orbit interaction.
blockade depends on the charge q at the junction Magneto-electric effect The phenomenon where
periodically, leading to the characteristic oscillation the magnetization is induced by the electric field or
of the tunneling current as a function of gate the electric polarization is induced by the magnetic
voltage. field.
Dirac fermion P.A.M. Dirac introduced the Majorana fermion In 1937, E. Majorana proposed
relativistic equation of quantum particles with the real fermions satisfying the relation f† = f called
four-component spinor wave function starting from the Majorana fermion (f† (f) is the creation
a few fundamental requirements. This is called (annihilation) operator of the fermion). Roughly
Dirac equation, and the fermions obeying Dirac speaking, Majorana fermion is half of the usual
equation are the Dirac fermions. In condensed fermion. Majorana fermion appears in the exact
matter systems, it often occurs that the band solution to two-dimensional Ising model, and also
structure is described by four- or two-component in the theory of pairing state of spinless fermions.
Dirac equations locally in momentum space. Mott scattering N.F. Mott derived the formula for
eg and t2g orbitals Under the crystal field of Oh the cross section of the Coulomb scattering
symmetry, the five d-orbitals are split into two between the two identical fermions with spin 1/2.
groups, that is, eg-orbitals (x2 y2, 3z2 r2) and By using this formula, the Mott detector to measure
t2g-orbitals (xy,yz,zx). the spin state of electrons has been developed.
224 Spin Hall Effect
Multiferroics Materials showing multiple Seebeck effect The effect where the voltage drop
symmetry breakings. Usually, it means those is induced by the temperature gradient.
showing the magnetic order and the ferroelectricity Soliton In field theory, the order parameter field
simultaneously. can have the spatial dependence creating a
Noether’s theorem The theorem connecting the localized kink structure. This localized structure is
invariance with respect to some transformation and called soliton.
the conservation law in field theory. By using this Umklapp scattering In crystals, the conservation
theorem, one can define the current from the of the momentum is relaxed by the reciprocal
invariance of the Lagrangian. lattice vectors G’s, and scattering process using
Noncentrosymmetric this G’s is called Umklapp scattering.
superconductor Superconductors without the Virtual bound state The d-orbitals in metals are
inversion symmetry. In the presence of the almost localized near the atomic nucleus. However,
spin–orbit interaction, novel properties such as the due to the hybridization with the conduction
mixture of singlet and triplet pairing emerge. electrons, the d-levels have the finite lifetime and
Nonlinear sigma model A model in field theory energy broadening, and are called the virtual bound
describing the field with constraint. This is used to states.
analyze the localization problem in disordered Z2 group The group whose elements are (1,1)
systems. only.
1.07.1 Introduction – Early History Lorentz force acting on the orbital motion of the
and Background of Spin Hall Effect electrons under the magnetic field; hence, the sign
(SHE) of Hall coefficient is determined by the sign of the
charge carriers while its magnitude by the carrier
Hall effect, which is one of the most important effects density. A few years after. Hall himself discovered
in condensed matter physics, has been discovered by the anomalous Hall effect (AHE) (Hurd, 1973), in
Hall in 1893 (Hurd, 1973). When the charge current j which the transverse Hall voltage originates from the
flows with an applied magnetic field perpendicular to spontaneous magnetization even without the exter-
it, a voltage drop occurs in the direction transverse nal magnetic field in metallic ferromagnets. This
both to the current and magnetic field. This voltage effect is driven by the relativistic spin–orbit (SO)
drop VH between the two edges of the sample, called interaction, while there are both intrinsic (Karplus
Hall voltage, divided by the current j defines the Hall and Luttinger, 1954) and extrinsic (Smit, 1955, 1958;
resistance. This Hall effect originates from the Berger, 1970) mechanisms. The intrinsic mechanism
Spin Hall Effect 225
is due to the anomalous velocity of the Bloch wave In these works, the SHE has been considered as an
function, and is purely the band structure effect. extrinsic effect, due to impurities in the presence of
Recently, this intrinsic contribution is reexamined SO coupling. Nevertheless, there were no detailed
from the modern topological viewpoint, that is, the discussions on the size of the effect, and this extrinsic
Berry curvature in the momentum space (Nagaosa, effect is not easily controllable because it is caused by
2006). The extrinsic ones, on the other hand, are due impurities. On the other hand, two groups indepen-
to the impurity scattering influenced by the SO dently proposed an intrinsic SHE in different
interaction and are classified into (1) the skew scat- systems. Murakami et al. (2003) proposed the SHE
tering (SS) mechanism (Smit, 1955, 1958), where the in p-type semiconductors such as p-GaAs. Sinova
scattering probability Pk ! k9 from k to k9 is different et al. (2004) independently proposed the SHE in
from Pk9 ! k, and (2) the side jump SJ mechanism n-type semiconductors in two-dimensional (2D) het-
(Berger, 1970), where the wavepacket experiences erostructures. These proposals are called the intrinsic
the transverse shift at the scattering. SHE, because they do not rely on impurities but they
In nonmagnetic metals, one can imagine that each of stem from the band structure itself, as we discuss in
the up- and down-spin electrons show the AHE in the this review. These two proposals triggered extensive
opposite directions due to the SO interaction. Then, research, both theoretical and experimental. After
one expects the spin current flow or the spin voltage 5 years from the theoretical predictions of the intrin-
drop in the transverse direction both in the direction of sic SHE, more than 500 papers have been published.
the charge current and the spin-polarization direction. In p-type semiconductors, for example, the SHE
This effect, called the spin Hall effect (SHE), is the in the linear response is expressed as
subject of this chapter, which constitutes a rapidly
growing field in the spintronics. The SHE was first jji ¼ s "ijk Ek ð1Þ
proposed by D’yakonov and Perel (1971a,b). They where jij is the current of the ith component of the
considered the extrinsic SS as the origin of the SHE spin along the direction j and "ijk is the totally anti-
predicting the spin accumulation at the edge of the symmetric tensor in 3D (see Figure 1(a)). The spin
sample due to the spin current. Here, one remark is current jijroughly corresponds to a product of velocity
that the spin current is time-reversal even, in sharp vj and the spin Si. It describes a spin current with
contrast to the spin accumulation or charge current spins along +iˆdirection going to the +jˆdirection, and
which are time-reversal odd. This means that the spins along iˆ direction going to the jˆ direction. In
conversion of the spin current into the spin accu- Figure 1(a), the electric field is applied along the
mulation requires the irreversibility due to the spin z-direction (k = z), and it produces the spin currents
relaxation. Later Hirsch (1999) proposed an experi- jy x (S xvy) and jx y (S yvx) with jy x = jx y . For
mental method to detect the SHE by the example, the Sx-spin current flows in the y-direction,
measurement of the voltage. With the charge cur- and S y-spin current flows in the x-direction. As an
rent flowing in the x-direction in the spin Hall example, we consider a thin semiconductor slab and
system, the spin Hall voltage is produced along the apply an electric field as shown in Figure 1(b). The
y-direction. When the two edges at y = L/2, with L electrons with opposite spins are deflected toward
being the width of the sample, are connected to each opposite directions, resulting in a spin current per-
other with a conductor, the spin current circulates pendicular to the applied field.
along the y-direction. This circulating spin current In the context of SHE, we often restrict ourselves to
induces the voltage at the conductor along the a situation with time-reversal symmetry, which means
x-direction. In Hirsch (1999), the voltage signal the absence of magnetism and magnetic field. We can
has been estimated to be of the order of 60 mV for consider the SHE in systems without the time-reversal
the sample size l ¼ 100 mm for a usual metal. Zhang symmetry. In such case the charge Hall effect arises (or
(2000) also discussed the SHE with a semiclassical in ferromagnet AHE arises), and the Hall current is a
Boltzmann equation taking into account the spin mixture of charge and spin current. Only when the
diffusion, since SO interaction is the origin for system respects the time-reversal symmetry, the Hall
both the SHE and the spin relaxation through the current is a pure spin current. The spin current will
spin-flip processes. As mentioned above, this spin induce a spin accumulation near the sample boundary.
relaxation is also indispensable for the spin accumu- The amount of accumulated spins is roughly estimated
lation at the edges of the sample. as a product of spin current and spin relaxation time Ts
226 Spin Hall Effect
E p-GaAs
Current Down-spin
(Murakami et al., 2003). The spin accumulation also Wunderlich et al. (2005) observed a circularly polar-
affects the spin current itself near the boundaries ized light emitted from a spin light-emitting diode
(Ma et al., 2004; Hu et al., 2004). Because spin current (LED) structure, using spin current from the p-type
is even under time reversal, it can be induced even semiconductor. This circularly polarized light is
when the time-reversal symmetry is preserved. It also attributed to the SHE in the p-type semiconductor.
implies that the spin Hall current is dissipationless. In After those works, there followed a number of experi-
doped semiconductors, however, the longitudinal mental papers. In particular, the SHE has been
conductivity is finite and the system undergoes observed in various metals such as Al, Pt, and Au.
Joule heating. Nevertheless, there exist spin Hall As the Hall effect has a quantized variant, that is,
insulators (SHIs), which are band insulators with the quantum Hall (QH) effect, the SHE also has an
nonzero SHE; in such systems the longitudinal con- analogous effect: quantum spin Hall effect QSHE.
ductivity is zero, and the SHE accompanies no (Note that the spin Hall conductance is not quan-
dissipation (Murakami et al., 2004a). tized, so it cannot be called quantized SHE.) The
Because the predicted effect in p-type semiconduc- QSH effect is a spin analog of the QH effect, and
tors and in n-type heterostructures is large enough to they share many fundamental aspects in common.
be measured, this intrinsic SHE attracted much atten- Yet, the QSH effect is realized in zero magnetic
tion, and one of the important issues was the disorder field, which is in strong contrast with the QH effect.
effect. Among many works on disorder effect, Inoue This effect has been experimentally observed
et al. (2003, 2004) considered dilutely distributed recently (König et al., 2007).
impurities with short-ranged potentials in the n-type Although there have been many works concerning
semiconductor in heterostructure, and calculated the the SHE in these few years, there are many issues to
SHE, incorporating the vertex correction in the ladder be resolved. In this review, we also introduce these
approximation. Remarkably, the resulting spin hall remaining issues and open questions. There are
conductivity (SHC) is exactly zero in the clean limit. already several articles on the recent developments
After this work, many people thought the SHE to be of this field (Alderson and Hurd, 1971; Awschalom
fragile to impurities; namely, only a small amount of and Flatte, 2007; Bratkovsky, 2008; Chazalviel, 1975;
impurities might completely kill the intrinsic SHE. Fabian et al., 2007; Fertig, 2003; Nagaosa, 2008; Wolf
However, this turned out to be a special case, and is et al., 2001). One may be referred also to other intro-
not true in general. In fact, the SHC is, in general, ductory articles on the recent developments and
nonzero even in the presence of disorder, as we see perspectives of the SHE (Bauer, 2004; Day, 2005;
later. Inoue and Ohno, 2005). Readers can also be referred
In such circumstances, two seminal experiments on to earlier short reviews (Murakami, 2005), and more
the SHE have been done. Kato et al. (2004) observed recent ones (Schliemann, 2006; Engel et al., 2006).
spin accumulation in n-type GaAs by means of Kerr This chapter is organized as follows. In Section
rotation (KR). The observed spin accumulation at the 1.07.2, we introduce a basic framework and discuss
two edges was interpreted as that caused by a bulk spin the mechanism of the SHE in semiconductors. In
current. Both unstrained and strained samples, with Section 1.07.3, we discuss the SHE in metals.
various orientations of the sample, are used to measure Section 1.07.4 is devoted to the QSH effect. We
the SHE, and the resulting spin accumulation agreed give concluding remarks and discuss miscellaneous
well with that expected from the extrinsic SHE. topics in Section 1.07.5.
Spin Hall Effect 227
1.07.2 Intrinsic and Extrinsic Here An(k) is a gauge field, and is subject to a gauge
Mechanism of SHE transformation accompanied by a change of phase for
the wave function of unk: u9nk = ei(k)unk. The corre-
The intrinsic SHE can be described from various sponding field strength Bn(k) = rk An(k) remains
viewpoints. In this section, we start with the Berry unchanged, that is, gauge invariant. Because a
phase theory of the intrinsic SHE. According to this contour integral of A(k) along a loop in k space is
theory, the SHE is driven by the ‘‘Berry phase cur- the Berry phase, An(k) and Bn(k) are called Berry
vature in momentum-space.’’ This occurs because of connection and Berry curvature, respectively.
the wave nature of the electrons. The Berry phase We consider a wavepacket as a superposition of
(Berry, 1984; Shapere and Wilczek, 1989; Bohm et al., the Bloch states. It is localized both in the real space
2003) is a change of a phase of a quantum state and in the momentum space. The width x of the
caused by an adiabatic cyclic change of some para- wavepacket in the real space and the width k in
meters characterizing the Hamiltonian. As this phase momentum space is restricted by the uncertainty
is determined only by a path in the parameter space, principle: xk & 2. We consider k is much
it is sometimes referred to as a geometric phase. smaller than the size of the Brillouin zone
Various physical phenomena involve this Berry k 2=a; where a is the lattice constant, and
phase, depending on which parameters are regarded consequently we have x a. This wavepacket
to change adiabatically, and are studied and acquires a phase when the wave number k is changed.
reviewed in Shapere and Wilczek (1989), Bohm The motion of a wavepacket made of the Bloch states
et al. (2003), Bérard and Mohrbach (2004, 2006), then undergoes a shift because of the interference of
Bliokh (2005), Bliokh and Bliokh (2005), Culcer Bloch states.
et al. (2004), Culcer and Winkler (2007a,b, Culcer By incorporating the effect of Bn(k), the
et al. 2005a), Dai and Zhang (2007), Erlingsson et al. Boltzmann-type semiclassical equation of motion
(2005), Hu et al. (2003a), Imura and Shindou (2005), (EOM) acquires an additional term due to the
Raimondi et al. (2007, 2006), Shindou and Imura Berry phase (Adams and Blount, 1959; Blount, 1962;
(2005), Sinova et al. (2006), Sun et al. (2004), Lu Sundaram and Niu, 1999):
et al. (2006), and Středa (2006). As for the electronic 1 qEn ðkÞ _
states, this Berry phase is known to appear in various x_ ¼ þ k Bn ðkÞ ð4Þ
h qk
kinds of Hall effects, such as the QH effect (Thouless
et al., 1982; Kohmoto, 1985; Sundaram and Niu, hk_ ¼ – e ðE þ x_ BðxÞÞ ð5Þ
1999), the AHE, or the SHE. To demonstrate the The term kB_ n ðxÞ represents the effect of Berry
effect of the Berry phase in k space, we here consider phase, and is called an anomalous velocity.
a clean system with crystal translational symmetry, When we apply an electric field, the wave number
where the Bloch wave number k is a good quantum will change in time. The second term on the
number. We regard k as parameters which are adia- right-hand side of Equation (4) then gives rise to an
batically changed. Then the notion of the Berry additional velocity perpendicular to the electric field,
phase can be expressed in a gauge field in k-space namely the Hall effect. This theory has already been
(Berry, 1984). It plays an important role in AHE proposed in the 1950s in a discussion of AHE (Karplus
(Matl et al., 1998; Ye et al., 1999; Chun et al., 2000; and Luttinger, 1954; Nagaosa, 2006), and has renewed
Ohgushi et al., 2000; Taguchi et al., 2001; Jungwirth interest from the QHE (Thouless et al., 1982) and the
et al., 2002; Fang et al., 2003) and in magnetic super- Berry phase (Berry, 1984). The SHE is among the
conductors (Murakami and Nagaosa, 2003). phenomena due to this anomalous velocity. This
We define the U(1) gauge field An(k) and corre- intrinsic Hall conductivity (7) was first recognized by
sponding field strength Bn(k) as Karplus and Luttinger (1954). This Berry phase in
Z momentum space has been studied in the recent
q y qunk
Ani ðkÞ ¼ – i nk nk X – i unk dx ð2Þ works on AHE (Ye et al. 1999; Ohgushi et al. 2000;
qki unit cell qki Taguchi et al. 2001; Jungwirth et al., 2002; Fang et al.
Bn ðkÞ ¼ rk An ðkÞ ð3Þ 2003; Yao et al. 2004), as well as those on the SHE.
These equations were originally proposed by Adams
where unk(x) is the periodic part of the Bloch wave and Blount (1959) and Blount (1962). They are re-
function nk(x) = eik ? xunk(x) and n is the band index. derived in Chang and Niu (1995, 1996) and Sundaram
228 Spin Hall Effect
and Niu (1999) using time-dependent variational et al., 2000; Taguchi et al., 2001; Jungwirth et al., 2002)
principle and investigated in the context of the Berry including first-principles calculations for SrRuO3
phase. We note that when there is a band degeneracy (Fang et al., 2003), thin film Sr1 xCaxRuO3 (Mathieu
for every wavevector k, for example, as a result of et al., 2004), bcc Fe (Yao et al., 2004), thin film Mn5Ge3
symmetry such as Kramers degeneracy, the above (Zeng et al., 2006), and CuCr2Se4 xBrx (Yao et al.,
formalism is modified accordingly (Murakami et al., 2007). In some cases, a quantitative agreement has
2003; Culcer et al., 2005b). The wavepacket consists of been reached as a function of parameters such as
the degenerate wave functions composition x, the magnetization M, or the tempera-
Z ture T. Meanwhile, apart from such intrinsic terms
X calculated from the first-principles calculation, an
jw i ¼ ddkaðk;t Þ i ðk;t Þji i ð6Þ
i extrinsic term coming from impurities always exists,
and it depends on whether the intrinsic terms
where jii (i = 1,. . .,N) denote the degenerate Bloch
dominate.
wave functions and N is the degeneracy of the con-
To see what the field Bn(k) is like and how we can
sidered states. The Berry connection and curvature
enhance this field, we rewrite the formula as follows:
should then be replaced from U(1) to SU(N). The
resulting EOMs for x and k are similar to
X nkqH mk mkqH nk
qk qk
Equations (4) and (5), with Bn(k) replaced by the Bn ðkÞ ¼ i ð8Þ
mð6¼nÞ ðEn ðkÞ – Em ðkÞÞ2
matrix element of the SU(N) Berry curvature sand-
wiched by the vector i. In the course of propagation, where we assumed that the energy spectrum has no
the components i can change, and this propagation degeneracy. This formula shows that the Berry phase is
also involves the Berry connection. a multiband effect, and it vanishes when there is only
These EOMs, Equations (4) and (5), can be used one band. We note that it is straightforward to extend
for calculating the Hall conductivity. For 2D systems, this formula to a spectrum with degeneracy; the gauge
for example, the Hall conductivity xy in a clean field becomes non-Abelian (Wilczek and Zee, 1984).
system is calculated from Equations (4) and (5) as Because this formula has a square of energy difference
Z in its denominator, Bn(k) becomes large, when two
e2 X
xy ¼ – d2 k nF ð"n ðkÞÞBnz ðkÞ ð7Þ energy spectra come close. In particular, when two
2h n BZ
bands touch (i.e., are degenerate) at a certain wavevec-
where nF("n(k)) is the Fermi distribution function for tor k, B(k) can diverge. We note that degeneracy does
the nth band, and the integral is over the entire not necessarily mean divergence of Bn(k). For example,
Brillouin zone. This formula agrees with the result when the HamiltonianPnear the degeneracy point is
from the Kubo formula in a bare vertex diagram. In expressed as H ðkÞ ¼ a¼x;y;z a ka , like the massless
this sense, this formula represents the intrinsic contri- Dirac fermions, Bn(k) diverges and has a monopole or
bution, which does not rely upon impurity scattering. antimonopole, while it does not when P only two com-
In reality, there are also extrinsic contributions coming ponents are involved as in H ðkÞ ¼ a¼x;y a ka .
from impurity scattering, which should be evaluated
separately. Given the Hamiltonian, the vector field
1.07.2.1 Intrinsic SHE in p-Type
Bn(k) is calculable, and we can get the intrinsic Hall
Semiconductors
conductivity, as in the ab initio calculation of the AHE
(Fang et al., 2003; Yao et al., 2004). This Hall effect In cubic p-type semiconductors with diamond struc-
comes not only from the states on the Fermi level but ture (e.g., Si, Ge) or zincblende structure (e.g., GaAs,
also from all the occupied states. We note that this InSb), the valence band consists of two doubly-
formula can be recast into a sum of a quantized part degenerate bands called heavy-hole (HH) and
and a part which comes from the states on the Fermi light-hole (LH) bands (Kittel, 1987). These two
energy (Haldane, 2004). When the Fermi energy lies in bands are degenerate at the -point (k = 0). They
a gap, the extrinsic contribution vanishes and the consist of the three p-orbitals, px, py, pz, with spin up
intrinsic one (7) can be shown to be quantized in a and down. In the presence of the relativistic SO
unit of e2/h, which explains the integer QHE (Thouless coupling, these six states are split into fourfold
et al., 1982; Kohmoto, 1985). This formula was also degenerate S = 3/2 states and twofold degenerate
applied to the AHE in ferromagnetic metals (Matl S = 1/2 states. Here, S denotes the total angular
et al., 1998; Ye et al., 1999; Chun et al., 2000; Ohgushi momentum of the atomic orbital, obtained through
Spin Hall Effect 229
h2 5
H ðkÞ ¼ 1 þ 2 k2 – 2 2 ðk?SÞ2 ð10Þ
2m 2
1.5
CB
This approximation is called spherical approxima- Helicity:
tion, because it restores spherical symmetry, while a 1.4
calculation without it is also possible (Bernevig et al., k FL kH
F λ = ± 3/2
0 HH
2004). This Hamiltonian is theoretically easy to –EF
k?S
Energy, E (eV)
E(p)
2006; Li and Shen, 2007; Shen, 2004; Shen et al., 2005,
py
2004; Sinitsyn et al., 2004; Wang et al., 2006a; Wang and
Vasilopoulos, 2007; Winkler, 2003; Zhou et al., 2008; Li
et al., 2005; Lipparini and Barranco, 2007; Lucignano
et al., 2008; Miah, 2006; Rashba, 2004a, 2005; Schulz
px
and Trimper, 2008; Sensharma and Mandal, 2006;
py Shen, 2005; Song et al., 2006b; Wang et al., 2008,
2006d; Zhang and Shen, 2008; Dimitrova, 2005).
px
Figure 4 Dispersion of the Rashba model and the spin 1.07.2.3 Impurity Scattering Effect
directions of the eigenstates Sinova et al. (2004).
In this section, we explain various studies on the
effects of impurity scatterings on SHE. The extrinsic
defined the spin current jzy to be a symmetrized SHE has been studied in several methods, partially in
1 qH order to examine to what extent the experimental
product of the spin sz and the velocity vy ¼ ,
h qky reports can be accounted for by the extrinsic
namely mechanism. Before going into details, we note that
the impurity scattering effect on the SHE is similar to
1 h h2 ky z that on the AHE (Crépieux and Bruno, 2001). What
jyz ¼ v y ; z ¼ ð21Þ
2 2 2m has been established in the AHE is as follows. The
To calculate the SHC, we calculate the correlation impurity effects are classified into the skew scattering
function between the spin current jzy and the charge (SS) and the side jump (SJ). First, the SO coupling at
current jx defined as the impurity gives rise to a left–right asymmetry at
the scattering, as is known as Mott scattering. This
e qH hkx
R y mechanism of the AHE is called as the SS
jx ¼ – evx ¼ – ¼ –e – ð22Þ
h qkx m h (Figure 6(a)). Second, the impurity potential in the
where the electron charge is taken to be e = jej. By Hamiltonian gives rise to an additional anomalous
using the Kubo formula for the bare bubble diagram in term to the velocity. It gives an SJ of the electron
Figure 5 representing the intrinsic SHE, the resulting trajectory at the scattering event, without changing
is zyx = zxy = e/8, which is independent of the the direction of the particle away from the scattering
Rashba coupling
. The Rashba term in Equation center (Berger, 1970; Figure 6(b)). These two con-
(16) can be regarded as a k-dependent effective tributions, SS and SJ, sum up to give an extrinsic
Zeeman field Beff =
(ẑ k). In equilibrium the spins AHE. In the Kubo formula, the SS and SJ correspond
are pointing either parallel or antiparallel to Beff for to Figures 7(d) and 7(c), respectively, whereas the
the lower and upper bands, respectively. An external self-energy correction (Figure 7(b)) may be
electric field Ejjx̂ changes the wavevectors k of Bloch regarded as the correction to the intrinsic SHE
wave functions, and Beff also changes accordingly. The (Figure 7(a)). The situation for the SHE well resem-
spins will then precess around Beff, and tilt to the z- bles that for the AHE, and we use the same
direction, depending on the sign of ky. This appears as terminology as in the AHE. The readers are also
the SHE, and the SHC is calculated to be e/8, in referred to the papers focusing on some new aspects
agreement with the Kubo formula. We also note that of disorder effects (Hankiewicz and Vignale, 2006;
the same result was obtained by the functional integral Tserkovnyak et al., 2007).
method (Rebei and Heinonen, 2006). In addition, var-
ious aspects of the SHE in Rashba-type models have 1.07.2.3.1 Spin-independent impurities
been studied (Bryksin and Kleinert, 2006; Bernevig, There are many works on the calculation of the SHC
2005; Chang, 2005; Cheche, 2006; Cheche and Barna, in the presence of impurities. The most studied case
is the system with randomly distributed impurities
i
with -function potential:
jj jk
X
V ðxÞ ¼ V ðx – xi Þ ð23Þ
Figure 5 Diagram for calculating the intrinsic SHC. i
232 Spin Hall Effect
Spin
jji jk jj i jk jji jk
Figure 7 The calculation of the SHC. (a) bare vertex, (b) self-energy corrections, (c) vertex corrections (side-jump), and
(d) skew scattering.
One can include an effect of the self-energy In Krotkov and Das Sarma (2006), the Rashba
broadening 1/ by disorder (Schliemann and Loss, model with an arbitrary form of (spin independent)
2004, 2005) (Figure 7(b)). In both Rashba and kinetic term is considered, and it is found that the
Luttinger models, the intrinsic SHE is gradually cancelation of the SHC is specific to the quadratic
2 2
reduced as is decreased from 1. In the clean h k
limit, the SHC reproduces its intrinsic value. In the form of the kinetic term. In Dimitrova
2m
Rashba model, the characteristic energy scales are (2004b) and Chalaev and Loss (2005), in particular,
the Rashba energy "R ¼ m
R =h2 and the Fermi it is argued that because the conventional definition
energy " ; the SHC is reduced when h= exceeds
pffiffiffiffiffiffiffiffiffiffi F of spin current (21) is proportional to S_ y , its expecta-
"R "F . In the Luttinger model, the SHC is reduced tion value is physically expected to vanish in the
when h= exceeds "SO, which is the energy splitting steady state: hjyz i hS_ y i ¼ 0: In an explicit form,
between the LH and HH bands at the Fermi energy. the spin current in the Rashba model is written as
Subsequently, Inoue et al. (2003, 2004) found an
important result. They assumed dilutely distributed hky z 1 1 y
jyz ¼ ¼ ½H ;y ¼ – s_ ð24Þ
impurities with a -function potential and took the 2m 4im
R 2m
clean limit. In a clean limit, they obtained a vertex- which is roughly because the SO term is linear in k
correction contribution (Figure 7(c)) e/8 to the (Dimitrova, 2004b; Chalaev and Loss, 2005). The
SHC, exactly canceling the intrinsic value e/8. In vanishing of the SHC can be shown rigorously
addition, subsequent papers found that the SHC using the Keldysh formalism (Sugimoto et al., 2006).
vanishes even for finite and for long-range impurity Using the Keldysh formalism, one can show that the
potential (Mishchenko et al., 2004; Dimitrova, 2004b; time derivative O_ of an arbitrary operator O, which is
Chalaev and Loss, 2005; Khaetskii, 2006; Raimondi and independent of k, x, time t, and frequency !, has a
Schwab, 2005; Rashba, 2004b; Mal’shukov and Chao, vanishing expectation value in the steady state,
2005; Liu and Lei, 2004, 2005c,a,b; Sugimoto et al., 2006; _ = 0 even with a general form of the impurity
hOi
Krotkov and Das Sarma, 2006). In Raimondi and (Sugimoto et al., 2006). We note that a system with
Schwab (2005), the SHC is calculated for weak disorder Dresselhaus and Rashba terms also satisfies the simi-
h=
R pF "F , with general form of impurity lar relationship
potential. For weak disorder, we can evaluate the SHC
within the Born approximation and ladder summation 1 d
jyz ¼ ð
R s y –
D s x Þ ð25Þ
for vertex corrections. It is found that the SHC is zero. 2mð
2D –
2R Þ dt
Spin Hall Effect 233
and its SHC vanishes for general impurity potential Long-range impurity potential will give an extrinsic
for finite (Liu and Lei, 2005c). term, which has an opposite sign and a bit smaller
In the calculation of linear response such as the compared with the intrinsic one; thus, is the extrinsic
SHC, the Keldysh formalism is equivalent to the piece due to long-range impurity potential makes the
Kobo formula calculation. Nevertheless, the SHC smaller (Liu and Lei, 2005a,b). The Keldysh
Keldysh formalism can sometimes give us a clearer formalism is used for more generalized cases in the
physical picture for transport properties. In addition, Rashba model and related models (Liu and Lei, 2004,
the Keldysh formalism is suitable for calculating 2005a,b,c; Sugimoto et al., 2006).
time- and space-dependent properties. For example, One may wonder whether the SHE vanishes in
in Mishchenko et al. (2004), the Keldysh formalism is other systems, and there remain some controversies
used to derive diffusion equation for the spin and in this respect. After debates on the vanishing of
charge in the two-probe geometry. The impurity the SHC for finite , has now been reached consensus
potential is assumed to be of the -function form. that the SHE vanishes in the Rashba model with
The resulting diffusion equation describes the SHE, -function impurities in the clean limit (Inoue et al.,
the spin relaxation, and diffusion in a unified way. It 2003, 2004) and even for finite (Mishchenko et al.,
is found that the spin Hall current appears only near 2004; Dimitrova, 2004b; Chalaev and Loss, 2005;
the electrodes, whereas the bulk spin Hall current Khaetskii, 2006; Raimondi and Schwab, 2005; Liu
vanishes irrespective of the lifetime (Mishchenko and Lei, 2004; Sugimoto et al., 2006; Nomura et al.,
et al., 2004), in accordance with the vanishing SHC by 2005a; Sheng et al., 2005a) or for finite-ranged impu-
the Kubo formula. In the calculation, some approx- rities (Inoue et al., 2004; Khaetskii, 2006; Raimondi
imation is used based on an assumption that EF is and Schwab, 2005; Liu and Lei, 2004). Here, we note
much larger than the SO coupling and the self- that the Rashba model is exceptional, in that the SHE
energy broadening 1: EF , EF 1.
vanishes rather accidentally, namely because the spin
To analyze the origin of the vanishing SHC in the
current operator jzy is proportional to s_ y ¼ i½H ;s y
Rashba model, several works have been done without
(Shekhter et al., 2005; Dimitrova, 2004b; Chalaev
such approximation. In the Keldysh formalism, the
and Loss, 2005; Rashba, 2004b; Liu and Lei, 2005c).
density matrix is calculated using the quantum
In fact, one can check that the SHE does not vanish
Boltzmann equation in the electric field E. The
for other models in general (Murakami, 2004;
resulting density matrix contains several terms linear
Bernevig and Zhang, 2004; Liu and Lei, 2005a,b;
in the electric field E. Some of such linear terms are
Krotkov and Das Sarma, 2006); for example, when
proportional to the Fermi distribution function nF("),
the Rashba model is generalized, to have a higher-
while the others are proportional to dnF/d". The
former terms stem from interband matrix elements order term in k in the Rashba coupling (Murakami,
of the (spin) current operators, and correspond to 2004) or to deviate the band dispersion from para-
intrinsic contributions, while the latter terms repre- bolic to a more general one (Krotkov and Das Sarma,
sent intraband matrix elements, leading to extrinsic 2006), the SHC no longer vanishes. In addition, there
contributions. In the study by Liu and Lei (2004), the are some models where the vertex correction does
range of the impurity potential is finite, and the self- not cancel the intrinsic value (Liu and Lei, 2005a,b),
energy broadening is ignored, and the vanishing or even vanishes by symmetry (Murakami, 2004;
SHC has been concluded. Later, this result is gener- Bernevig and Zhang, 2004). Magnetic impurities
alized by considering also the self-energy broadening also make the SHC to be nonvanishing, where the
(Sugimoto et al., 2006; Liu et al., 2006a). In Liu and exchange coupling between magnetic impurities and
Lei (2005c) the system with Dresselhaus and Rashba itinerant electrons can be isotropic (Inoue et al., 2006)
terms is analyzed on the same assumptions, and again or anisotropic (Wang et al., 2007). The SHC depends
zero SHC is concluded. On the other hand, the cubic on the anisotropy of the magnetic interaction (Wang
Rashba model is also investigated on the same et al., 2007) or on the amount of magnetic impurities
assumptions, which gives nonvanishing SHC (Liu when both magnetic and nonmagnetic impurities
and Lei, 2005a,b). The extrinsic term has the same coexist (Inoue et al., 2006). We note that in this case
sign and its magnitude is somewhat smaller than the the SHC does not vanish even in the clean limit. In
intrinsic value 9e/8. In the case of the Luttinger the 2D hole gas (2DHG) models such as the cubic
model, on the other hand, the -function impurities Rashba model, the vertex correction vanishes for
will have no effect on the SHC (Murakami, 2004). -function impurities (Bernevig and Zhang, 2005).
234 Spin Hall Effect
P i
A Boltzmann-type equation can also be used to be short ranged: V ðrÞ ¼ V Ni¼1 ðx – xi Þ, where Ni
study impurity scattering with arbitrary angular is the number of impurities. For each disorder con-
dependence (Shytov et al., 2006). We consider a 2D figuration, they calculate the eigenstates
system numerically and put them into the Kubo formula.
The treatment of the small imaginary part of the
1
H ¼ "k – bðkÞ? þ V ðxÞ ð26Þ frequency in the Kubo formula requires care. In
2
the linear response theory of the Kubo formula, the
where k = (kx, ky) and b(k) represents the SO controls the speed of adiabatic switching of the
coupling. We note that b(k) = b(k) required by external electric field. In a calculation of longitudi-
time-reversal symmetry. It is assumed that nal charge conductivity, in order to absorb energy
from the external electromagnetic field, the level
bz ¼ 0; bx þ iby ¼ b0 ðkÞeiN ð27Þ
spacing of the system, , should be much smaller
i
where kx + iky = ke . Since the result is also affected than (Imry, 1997): . We can expect that
by the behavior of b0(k) and the energy dispersion "k similar argument applies to the SHE. In Nomura
near the Fermi energy, we write et al. (2005b), the SHC is calculated for various
system sizes, where the relationship is
q" 1þ
jb0 j _ kN̂ ; vk ¼ _k ð28Þ retained. Bulk quantity is obtained by extrapolating
qk
to ! + 0. For the k-linear Rashba model, the
near the Fermi energy. These three exponents, N, Ñ, extrapolation seems to suggest s 0 within numer-
and
, characterize the intrinsic SHE. For 2D elec- ical error. On the other hand, for the cubic Rashba
tron system on a (100) surface of a III–V model and the modified Rashba model, s in the
semiconductor, we have
= 0 and Ñ = 1. bulk is extrapolated to be nonzero. It is also found
Furthermore, N = 1 corresponds to the pure that the variation of the s by changing disorder
Rashba term, and N = 1 to the pure (linear-k) configuration is much larger than the average
Dresselhaus term. The 2D hole system with the value of s.
cubic Rashba term is represented as N = Ñ = 3 and The tight-binding counterpart of the Rashba
= 0. Here, the direct SO term
? (k rV) is model is numerically studied (Sheng et al., 2005a).
neglected because the constants
and are, in The model is represented by a tight-binding
principle, independent. This gives vanishing result Hamiltonian with Rashba SO coupling on a square
for the pure Rashba system or the pure Dresselhaus lattice:
system (N = 1). For 2D hole system, for example, X y
the SHC is different from that for the isotropic H ¼ H TB þ "i ci ci ð30Þ
i;
scattering potential (Shytov et al., 2006). The order
of magnitude for the SHC observed in the spin- where HTB represents the nearest-neighbor tight-
LED experiment by Wunderlich et al. (2005) seems binding Hamiltonian with the Rashba coupling
consistent with the value predicted here. (Ando model (Ando, 1989)), and "i is set to be a
Numerical studies based on the Kubo formula random variable within [W/2, W/2] when disorder
have also been carried out. An earlier numerical effect is considered. In Sheng et al. (2005a), the
calculation for the Rashba model (Nomura et al., authors apply the Laughlin’s gauge argument
2005a) suggested that the SHC is finite for weak (Laughlin, 1981) to relate the SHC to the carried
disorder for finite systems. Nevertheless, it has been spins during insertion of a flux quantum into a hole
superseded by a later paper (Nomura et al., 2005b, of the system, which is periodic only along one direc-
2006). In the latter paper (Nomura et al., 2005b) tion (see also Figure 27). This flux insertion
three models are studied – the (k-linear) Rashba corresponds to the twisting of the phase in the
model, the k-cubic Rashba model, and a modified boundary condition. The energy spectrum, shown
Rashba model: as a function of boundary phases, shows anticrossings
between energy levels, from which the authors con-
h2 k2
H¼ þ
k2 ðx ky – y kx Þ ð29Þ clude that there is no spin transfer across the sample
2m
during flux insertion, resulting in the vanishing SHC.
The calculation is based on the Kubo formula with Nevertheless, it is not clear whether the gapless
ji ¼ ðe=hÞqH =qki ; jyz ¼ h– 1 qH =qky ;sz =2, where system like the Rashba model can be treated within
sz ¼ ðh=2Þz . The disorder potential is assumed to the Laughlin’s gauge argument to calculate the SHC.
Spin Hall Effect 235
The diagrams contributing to the SHE are shown in is studied with the Kubo formula. By including ver-
Figure 7. These diagrams are classified into the SJ and tex corrections, the SS vanishes and the SJ
the SS in the similar way as in the AHE (Crépieux and contribution is reduced to
Bruno, 2001). The spin current operator has a term
involving V(x), which gives rise to the SJ (Figure 7(c)). SJ
s ¼ e
n=2 ð40Þ
The other contribution comes from the impurity term which is a half of the value (37) where the SO
V(x) itself in the Hamiltonian, corresponding to the SS coupling of the impurities is neglected. This remains
(Figure 7(d)). The resulting formula is true when
R ! 0, meaning that the resulting SHC
for
R ! 0 is a half of the value at
R = 0. Thus, the
SJ
s ¼ e
n ð37Þ
SHC is nonanalytic at
R = 0 (Tse and Sarma, 2006).
for the SJ and The cubic Rashba model is treated in the same way.
( In accordance with the previous results on the intrin-
1=3 :3D sic SHE, the vertex correction vanishes, and the
SS
s ¼ – 4e
"F =h ð38Þ
1=2 :2D result is additive in the lowest order for the intrinsic
and extrinsic parts (Tse and Sarma, 2006). This ana-
for the SS. The ratio between the two contributions is
lysis shows a rather peculiar feature of the (linear)
SJ =SS ðh=Þ="F which is smaller for cleaner sam- Rashba model.
ples. For 1013–1012s and "F 1–10 meV, for We note that the same problem is studied in Hu
example, they are of the same order. et al. (2006) with the Boltzmann transport theory. The
For 3D n-type semiconductors, the resulting value of results look somewhat different from the Kubo for-
1 1 1 1
the SHC is SJs = 0.75 m andpffiffiffiSJ
s 6 m mula results (Tse and Sarma 2006). It is not usually
with nominal values of parameters: 2
¼ 4:7Å, n = 3 easy to trace the reason for the difference in the
1016 cm3, and xx = 3 103 1m 1. The sum is results with Kubo formula and Boltzmann transport
s = 5.2 1 m1. It reproduces the order of magnitude theory. Nevertheless, one of the reasons may lie in
for the experimental value 0.7 1m 1, obtained in an the fact that in Hu et al. (2006) the distribution func-
experiment in Kato et al. (2004), although we took the tion is considered to be diagonal in the chirality basis
simplest model and picked up only the lowest-order jk
i, while it may not be always true when the
diagrams. For 2D p-type semiconductors with the para- external field is present, as studied in Shytov et al.
meters p = 2 1012 cm2 and xx 1.09 1013 1, (2006). The readers are referred also to Hankiewiez
as in the experiments by Wunderlich et al. (2005), and Vignale (2006, 2008), Tse et al. (2005), Cheng and
8 1
the SHC is evaluated as SJ s = 3.10 10 Wu (2008), Chudnovsky (2007), Grimaldi et al.
SS 6 1
and s = 2.72 10 . The total SHC is (2006a), and Sherman et al. (2006) for related subjects.
2.68 106 1. In this case the ratio jSJ SS
s /s j is as
2
small as 2 10 , because the sample is relatively clean.
The results of Engel et al. (2005) and those in Tse and 1.07.2.3.3 SHE in the hopping regime
Das Sarma (2006) qualitatively agree with each other. The SHE in strongly disordered systems is governed
When both types of the SO coupling, namely that by hopping conduction. In Beckmann et al. (2005),
of the bulk and that at the impurities, coexist, the Damker et al. (2004), and Entin-Wohlman et al.
problem becomes much more complicated. This (2005), the SO coupling is included in the hopping
problem is studied in Tse and Sarma (2006) for amplitudes, and interference effect between hopping
Rashba and cubic Rashba models, and it was paths is studied. The effect of the SO coupling
revealed that the Rashba model shows a peculiar appears in the third order in hopping amplitudes;
interplay between the intrinsic and the extrinsic for three sites i, l, and m, the interference occurs
parts, whereas in the cubic Rashba model extrinsic between the direct path i ! l and the indirect path
and intrinsic parts are additive in the lowest order. i ! m ! l. Because of this interference, when an
The Rashba Hamiltonian with SO-coupled external electric field is applied, the transverse spin
impurities current calculated as a time derivative of the spin
polarization becomes nonzero. In Beckmann et al.
(2005) and Damker et al. (2004), only the Rashba SO
h2 k2 coupling is studied in terms of the rate equations in
H¼ þ
R ðx ky – y kx Þ þ V ðxÞ –
ð rV ðxÞÞ?k
2m the Markovian limit. In Entin-Wohlman et al. (2005),
ð39Þ the Rashba and Dresselhaus SO coupling are
Spin Hall Effect 237
included in the hopping amplitudes, and this SHE is Rashba model for a four-terminal geometry was first
proportional to !2, where ! is the frequency of the numerically studied by Bulgakov et al. (1999), prior to
external electric field (Entin-Wohlman et al., 2005). the theoretical proposals of the intrinsic SHE
We note that the way to calculate the spin current as (Murakami et al., 2003; Sinova et al., 2004). In
a time derivative of the spin polarization was adopted Bulgakov et al. (1999), the scattering-wave formalism
in several works (Entin-Wohlman et al., 2005; Shi is applied to the four-terminal (continuum) Rashba
et al., 2006) as discussed in Section 1.07.2.6.1 model without disorder. For example, without the SO
coupling whose eigenstates are plane waves are
attached to the system. One can calculate the transition
1.07.2.4 Mesoscopic SHE
probabilities that an incoming wave goes into an out-
In this subsection, we discuss various calculations on going wave in other leads. It was numerically obtained
SHE in small systems. This includes (1) the calcula- that a left–right asymmetry occurs in the transmission
tions based on Kubo formula and (2) the calculations probability into the transverse channels, because of the
by the Landauer–Büttiker formalism. We discuss Rashba SO coupling. When the incoming electrons are
these two types of calculations separately. not polarized, the outgoing electrons in the transverse
The Kubo formula calculation for small-size sys- channel are spin polarized, which we can call the SHE.
tems (<100 sites) of 2DEG has been done (Sheng On the other hand, polarized incoming electron gives
et al., 2005a; Moca and Marinescu, 2007a). The result- rise to a transverse voltage, associated as an inverse
ing SHC strongly fluctuates as a function of the SHE (ISHE) (Bulgakov et al., 1999).
Fermi energy EF. Nevertheless, by averaging over This problem was then studied in the four-terminal
the twisting angle in the boundary conditions, the Landauer–Büttiker formalism (Nikolić et al., 2005b;
SHC becomes a smooth function of EF. We note that Sheng et al., 2005b). In this formalism, the authors
this trick has been adopted for numerical calculation used a tight-binding Hamiltonian with Rashba SO
of Hall conductivity in the QHE. As a function of the coupling on a square lattice. The random on-site
system size, the SHC is s 0.48(e/4) for a clean energy, "i, is set to be a random variable within
system, while s for disordered systems exponen- [W/2, W/2] when disorder effect is considered.
tially decays as a function of the system size When we consider a square sample with the size
hsi _ exp(L/s), with a characteristic length s L L, and attach the leads with no SO coupling at
(Sheng et al., 2005a; Moca and Marinescu, 2007a). the four edges, we can apply the four-terminal
This is consistent with the above-mentioned result Landauer–Büttiker formalism. The current is driven
that the SHC in disordered Rashba models is zero. through the lead 0 to 1 in the inset of Figure 8. Nikolić
In Chen et al. (2005a), the 2DHG is investigated et al. (2005b) and Sheng et al. (2005b) studied systems
numerically with the Kubo formula. The model is a with size up to 100 100. The SHE without any
discretized version of the Luttinger–Rashba model. disorder is studied by changing the system size and the
The SHC is calculated with the Kubo formula in a SO coupling. The SHC with spin along the z-axis, Gz,
finite-size sample, with the imaginary part set to is of the order of e/4, but not exactly equal to the
zero. In a clean system, the SHC is of the order of universal value e/8 calculated from the Kubo for-
11e/8, as long as the EF is near the band bottom. mula in the clean system. This SHC Gz depends on
When the system size is not large (around L = 20–30), the system size L or the SO interaction VRSO. As the SO
zxy largely fluctuates by changing disorder config- coupling VRSO increases, the SHC increases (Sheng et al.,
uration. This fluctuation is strongly suppressed when 2005b). The SHC with spins along the x-direction,
an average over the twisting angle in the boundary Gx, is also nonzero, which is not the case for the
condition is performed, as in the electron systems. Kubo formula calculation in a system with infinite
The resulting value agrees quite well with a result in size. The spin Hall current for both edges of the
a large system L = 1000; hence, one can regard that sample has the opposite sign for the x- and z-direction.
this averaging over the boundary conditions removes When the size of disorder W is increased, the SHC is
a finite-size effect efficiently. As one introduces the gradually suppressed, and when W 5t the SHC
disorder gradually by changing W, the SHC is mono- becomes vanishingly small. With the system size
tonically suppressed. The SHC almost vanishes L 100, the size dependence of Gx and Gz depends
when W 5t. on the SO interaction. There also arises a spin com-
The Landauer–Büttiker formalism has been applied ponent along the y-direction, and it has the same sign
to many types of systems. In fact, the SHE in the for both edges of the sample. It corresponds to the
238 Spin Hall Effect
1.2
0.0050
W=0 0.0025 G xsH
0.8 W = 1t 0.0000
W = 2t –0.0025 tSO = 0.01to
–0.0050
0.4
0.2
GsH (e/4π)
0.1 60 × 60
current-induced spin polarization (CISP) studied the- larger for the 2DHG than that for the 2DEG in
oretically (Edelstein, 1990; Aronov et al., 1991; Inoue general. Nonetheless, they are of the order e/8 and
et al., 2004) and experimentally. A similar system with depend unmonotonically on
and the system size L.
the Rashba coupling also in the leads is studied by An effect of impurity SO coupling is studied by
Wang and Chan (2005). Qualitatively similar results Pareek (2004) in a Y-shaped geometry. Here a charge
with the previous cases of leads without SO coupling current is passed between two terminals, while the
are obtained, whereas the existence of the SO coupling third terminal works as a voltage probe without any
in the leads makes the definition of the spin current charge current. A pure spin current is then produced
somewhat ambiguous (Figure 9). in the third probe, and it can be regarded as a result of
The reciprocal effect of the SHE can be studied an extrinsic SHE.
by the Landauer–Büttiker formalism (Hankiewicz Nonequilibrium spin accumulation has been stu-
et al., 2005a). In the Landauer–Büttiker formalism, died in a two-terminal geometry, namely in a
the lead has no SO coupling, and the spin current is quantum wire geometry (Nikolić et al., 2005a; Wang
uniquely defined, thanks to the spin conservation. et al., 2006b). The Keldysh nonequilibrium Green’s
This conservation of spin automatically ensures the function is combined with the Landauer formalism to
Onsager relation study numerically the spin accumulation. The tight-
xy yx
binding model studied by Ando (1989) is used for the
GSC ¼ – GCS ð41Þ calculation. The spins accumulate at both the edges,
This relationship holds for disorder-averaged con- with their directions along the z-axis and the x-axis
ductances, because its proof uses the inversion opposite for the two edges, in accordance with the
symmetry on the x–y plane (Hankiewicz et al., calculation of the spin current by the Landauer–
2005a). This relationship can be directly checked Büttiker formalism in the previous section (Nikolić
numerically. The 2DEG as well as the 2DHG are et al., 2005b; Sheng et al., 2005b). In Nikolić et al.
studied. The 2DEG is described by the Rashba (2005a), the leads are without the SO coupling,
model, and the 2DHG by a cubic Rashba model whereas in Wang et al. (2006b) the leads include the
p2 i
3 þ 3 – SO coupling in the leads as well, which changes the
H¼ þ ðp – pþ Þ, where = x iy. spin-dependent scattering between the leads and the
2m 2 –
The resulting SHC (which is equal to the GyxCS) is sample. In Wang et al. (2006b), the spin accumulation
Spin Hall Effect 239
0.01
–6
0.001
0.0001
–8 0 20 40 60 80
λ (me V nm)
0 20 40 60 80
λ (meV nm)
Figure 11 Calculated voltage drop between terminals 3 and 4 in the H-shaped device. From Hankiewicz EM, Molenkamp
LW, Jungwirth T, and Sinova J (2004) Manifestation of the spin Hall effect through charge-transport in the mesoscopic regime.
Physical Review B 70: 241301.
polarization, focused to 1 mm size, is incident onto the experimental data. It can also account for the absence
sample, and they measured the rotation angle of the of dependence on crystal orientation. Thus, the
polarization plane upon reflection. The rotation quantitative explanation of the observed SHE is still
angle is proportional to the spin polarization perpen- to be resolved.
dicular to the plane. When a static magnetic field is Further experiments using the spin mapping by the
applied, the magnetization will precess around the KR have been done by the same group. Sih et al. (2005)
magnetic field, and decays as A?ð1=½ð!L s Þ2 þ 1Þ by investigated the SHE and CISP in (110) AlGaAs QWs
the Hanle effect, where !L ¼ gB B=h is the Larmor with various direction of electric field. In the (110)
frequency and s is the spin coherence time. By QW, the Dresselhaus field is perpendicular to the
measuring the field dependence of the angle, one plane, while the Rashba field is in-plane; thereby, the
can determine s. separation between the two fields is possible. For
Kato et al. (2004) observed the SHE in n-type Ejj[001], the CISP due to the Dresselhaus field
semiconductors by measuring spin accumulation at vanishes, and the total CISP is in-plane. Thus, the
the edges of the sample by KR (Bauer, 2004). The CISP is absent in the KR signal, because the KR
spin accumulation is uniformly distributed along both detects the spin component perpendicular to the
the edges, while away from the edges there is no spin plane. Therefore, the out-of-plane spin is accumulated
accumulation, as can be seen from Figure 12. From only at the two edges of the sample. On the other
experimental data they evaluated the amount of spin hand, for other orientations of the electric field such as
accumulation and spin lifetime as a function of an ½110 and ½112, the CISP has an out-of-plane compo-
external magnetic field. The measured SHC is nent, and appears in the whole sample, in addition to
5 103 1 cm1. They concluded the observed the SHE signal at the edges.
SHE to be extrinsic for the following reasons: (1) Sih et al. (2006) also investigated the device with
spin splitting is negligibly small in the sample and (2) side arms, in order to see whether the observed spin
the effect has no dependence on crystal orientation. accumulation at the edges is due to a spin current, or
Nevertheless, Bernevig and Zhang (2004) argued due to other edge effect. The measurement was done
that the observed SHE can be intrinsic, coming from for 3D n-GaAs (2 mm thick) with a Si doping with
the dresselhaus term representing bulk inversion- n = 3 1016 cm3 at T = 30 K. The observed profile
symmetry breaking. They showed that even if the of spin distribution is well fitted by assuming that the
spin splitting due to the Dresselhaus term is negligi- spins are generated through the SHE in the main
bly small, the SHE can be as large as the channel, and flow into the side arm.
Spin Hall Effect 241
B (T)
B (T)
0 0
θK (μrad)
–4 8
–0.2 –0.2
0 10 20 0 10 20
t (ns) t (ns)
Figure 13 Experimental setup for the time-resolved measurement of SHE. From Stern NP, Steuerman DW, Mack S,
Gossard AC, and Awschalom DD (2008) Time-resolved dynamics of the spin Hall effect. Nature Physics 4: 843.
obtained SHE is mostly intrinsic. More refined argu- strong SOs (Figure 15). They found it is indepen-
ment, by showing vanishing vertex correction, also dent of the strength of disorder scattering suggesting
supports this conclusion (Bernevig and Zhang, 2005). the intrinsic nature of the effect, and obtained the
Nomura et al. (2005c) studied numerically the SHE- similar phenomenological aspects as compared with
induced edge-spin accumulation in a 2DHG with the experiment (Wunderlich et al., 2005).
(a) Ip
LED 1
+Ip
1
LED 1
0
p n
y
1.5-mm x
n
CP (%)
channel z –1
–Ip
LED 2 (b)
9 +Ip 1
LED 1
2.5
(d)
1.5 0
(meV)
0.5 LED 2
(c) –1
1
10 20 30 1.505 1.510 1.515 1.520
P2D (1011 cm–2) E (eV)
Figure 14 (a) The device setup for the spin LED, (b) circular polarizations measured for the opposite directions of current,
(c) circular polarizations measured at LED1 and LED2, and (d) fit with the theory. From Wunderlich J, Kaestner B, Sinova J,
and Jungwirth T (2005) Experimental observation of the spin-Hall effect in a two-dimensional spin–orbit coupled
semiconductor system. Physical Review Letters 94: 047204.
Spin Hall Effect 243
8
(a) Spin-density profile (c) Edge-spin density
Szedge/E (e /4πvF)
10
Sz (x)/E (e /4πvF)
4
0
5
–4
–8 0
8 (b) Spin-current profile (b) Bulk-spin current
10
jxzbulk/E (e /8π)
jxz(x)/E (e /8π)
4
5
2
0 0
0 10 20 30 40 50 0 2 4 6 8 10 12 14
X (kF–1) EF (h/τ)
Figure 15 Spatial dependence of the spin accumulation and spin current in a model for the 2D hole gas. From Nomura K,
Wunderlich J, Sinova J, Kaestner B, MacDonald AH, and Jungwirth T (2005c) Edge-spin accumulation in semiconductor two-
dimensional hole gases. Physical Review B 72: 245330.
A theoretical study on the spin accumulation by the One can adopt the symmetrized product
electric field can also be found in Kleinert 1
ðvi s j þ s j vi Þ between the velocity v and the spin s
et al. (2005). 2
as a definition of the spin current as in Sinova et al.
(2004). The result given by the Kubo formula with
1.07.2.6 Various Issues on the SHE this definition is in general different from that by the
semiclassical theory described above (Murakami
1.07.2.6.1 Definition of the spin current
et al., 2004b). This difference comes from noncom-
In the presence of the SO coupling, the total spin is
mutability between the spin S and the velocity v. In
not conserved. Hence, there is no unique way to
other words, this comes from the nonuniqueness of
define a spin current. Naively we expect that the
the definition of spin. One can modify the semiclas-
spin current js should satisfy the equation of continu-
sical theory to give the same result as the Kubo
qs i
ity þ r?Jsi ¼ 0; this relationship requires the formula by adding three contributions: spin dipole,
qt torque moment, and change of wavepacket spins due
conservation of total spin, namely
to electric field (Culcer et al., 2004).
Z Z
1.07.2.6.2 Ab initio calculations characterized by, for example, k ? p theory, and the
On the other hand, the intrinsic SHE is calculated for theory can be rather well controlled. However, it has
Si, GaAs, W, and Au for various values of the Fermi the following disadvantages. (1) due to the small
energy (Guo et al., 2005). For p-type GaAs the pre- number of carriers, the conductivity itself is small,
dicted value is maximally 300 1 cm1, which is of and correspondingly the SHC is even smaller; (2)
the same order of magnitude as that from the the interface with magnetic materials, which are
Luttinger model (Zhang and Yang, 2005). The mostly metals, suffers from the impedance mismatch
p-type Si has somewhat smaller value, 50 1 cm1, (Schmidt et al., 2000); and (3) the systems are rather
which is due to smaller SO coupling than GaAs. For fragile against disorder and thermal agitation. These
n-type GaAs, the size of the SHC is 50 1 cm1 disadvantages are overcome by the metallic systems.
and undergoes a sign change as a function of carrier For example, (1) the SHC is much larger than that of
density, and n-type Si is found to have a negative doped semiconductors typically by the factor of 102–
SHC ( 50 1 cm1). It is found that even without 105; (2) the junction with ferromagnetic metals is
doping, GaAs and Si show a small but finite SHE (43 simple and techniques in metallic spintronics such as
and 7 1 cm1, respectively). It is due to a small the spin injection (Johnson, 1993; Johnson and Silsbee,
hybridization. In this sense, these undoped semicon- 1985), spin diffusion (Shchelushkin and Brataas,
ductors are SHI. In the metallic systems such as W 2005a,b), and nonlocal effect can be used; and (3)
and Au, the SHC is calculated to be 1390 and the quantum coherence is much more robust against
731 1 cm1, respectively, and is rather robust disorder and thermal agitation due to the large Fermi
against disorder (Yao and Fang, 2005). These ab initio energy, and hence the operation at room temperature
calculations are done by calculating only the bare is possible. These advantages make the metallic
vertex diagram (Figure 5) in the Kubo formula. systems more promising candidates for the real appli-
Nevertheless, analytic calculations revealed that the cations of the SHE in spintronies.
vertex correction can be of the same order as the In semiconductors, the optical detection of the
intrinsic SHE, and an effect of vertex correction on spin accumulation such as the KR has been used to
such ab initio calculations remains to be unsettled. detect SHE, since the spin diffusion length is much
For applications to spintronies (Žutić et al., 2004), larger than the spot size of the laser beam of the order
the SHE may potentially be used as an effective of 1 mm. In metals, on the other hand, the spin diffu-
means to inject spins into semiconductor spintronics sion length is much shorter, and the electrical
devices. There have been many proposals for semi- detection is used (Huh et al., 2008). There are two
conductor spintronics devices such as Datta–Das spin ways to study the SHE in metals. One is to use the
transistor (Datta and Das, 1990). Nevertheless, an nonlocal spin diffusion due to the injection of the
effective spin injection into semiconductors is one spin-polarized current from the ferromagnet
of the elusive issues, because spin injection from a (Takahashi and Maekawa, 2003, 2007, 2008a,b). An
ferromagnetic metal suffers from conductance mis- early experiment on the SHE in Al (Valenzuela and
match (Schmidt et al., 2000). There have been many Tinkham, 2006) uses CoFe as the ferromagnet. A thin
kinds of attempts to overcome this difficulty, and the Al Hall cross is oxidized and contacted with the
SHE might be one of the way out toward effective ferromagnetic electrodes as shown in Figure 16.
spin injection into semiconductors. In particular, one When the current flows from the ferromagnetic elec-
of the merits of the SHE is that it does not require trode to Al Hall bar, the spin diffusion occurs to the
magnetic field or magnetism. The main obstacle for other side of the bar, where the Hall voltage is mea-
the SHE toward application is its smallness. Ab initio sured. This is the reciprocal effect to the SHE in
calculations would be important for searching sys- which the charge current induces the spin current.
tems with large enough SHC. This is called the ISHE, and is now widely used to
measure the SHC in metals. As shown in Figure 16
the spin transistor configuration with two ferromag-
1.07.3 SHE in Metals netic electrodes FM1 and FM2 has been used to
characterize the spin diffusion length
sf, the spin
1.07.3.1 Experiments of SHE in Metals
polarization P, and the angle between the magne-
Up to now we have focused mainly on the semicon- tization m and the electrode axis. The chemical
ductor systems with small carrier concentration. The potential difference between up- and down-spins
local band structure in momentum bfk-space is well decays as the distance x from the distance from the
246 Spin Hall Effect
(a) (b)
FM2
FM1 λsf = 705 nm
V– 10
ΔR (mΩ)
I– 1 λsf = 455 nm
I– 0.1 tAI = 12 nm
tAI = 25 nm
I+ V+
1 2 3 4
LFM (μm)
(c) (d)
LFM = 2 μm 1
2 B⊥ (T)
• sin θ
–0.4 –0.2 0.0 0.2 0.4
V/I (mΩ)
sin θ
0 0
2 B⊥
V/I (mΩ)
0 M
–2 θ
–2
–1
0 1 2 3 –3 0 3
B⊥ (T) B⊥ (T)
Figure 17 (a–d) Experimental data for the SHE in Al. From Valenzuela SO and Tinkham M (2006) Direct electronic
measurement of the spin Hall effect. Nature 442: 176.
y Pt Cu z
y
Cu
x x Pt
Cu 1 2 3
Cu 1 2 3 Py
Py
(a) Cu
Is
(b) (c)
Ie s
Is
Is
Py Cu Pt
1 2 3 Ie
Is
(e)
(d)
Is
Is Ie
Py Cu Pt
1 2 3 s Ie
Figure 18 (a–e) Experimental setup for the measurement of SHE and ISHE in Pt. From Kimura T, Otani Y, Sato T, Takahashi
S, and Maekawa S (2007) Room-temperature reversible spin Hall effect. Physical Review Letters 98: 156601.
(a) z (b)
F
y
200 nm 150 nm
FePt injector x
d
M
C
B Au E
V
Au hall cross D
Fept
A
0.5
200 +
–31.5
V
ΔV –
–19.5
0 100
0 0.5 e–
RISHE (mΩ)
RLHE (mΩ)
–32.0
V (μV)
I(mA) V+
ISHE I = –0.5 mA V+
0
LHE
20.0 M
–32.5
+
–100
e– V
ΔV
19.5 –
–33.0
–200
V–
–10 –5 0 5 10 15 –8 –4 0 4 8
H (k0e) H (k0e)
Figure 19 (a–d) Experimental setup and results for the measurement of ISHE in Au. From Seki T, Hasegawa Y, Mitani S, et al. (2008) Giant spin Hall effect in perpendicularly
spin-polarized FePt/Au devices. Nature Materials 7: 125.
Spin Hall Effect 249
around the magnetic field with the Gilbert where is the relaxation rate. The second term
damping. More explicitly, the EOM for the mag- represents the contribution from the AHE in NiFe
netization reads system. Taking the derivative dV(H)/dH, the ISHE
contribution is odd with respect to H HFMR while
dm dm
¼ m H eff þ m ð58Þ AHE contribution is even. Therefore, it is clear
dt dt
from the data in Figure 21 that the ISHE contri-
where the last term on the right-hand side represents bution is dominant. Another test of ISHE is the
the Gilbert damping, and is typically 103–101. angel dependence. From (58), the signal is
This Gilbert damping drives the magnetization pre- expected to be proportional to sin ( is the
cession amplitude smaller and smaller and the angle between the external magnetic field and the
magnetization eventually turns the direction of the charge current), which is also consistent with the
effective magnetic field Heff. From the viewpoint of data. These experimental results provide rather
the conservation of the angular momentum, the spin firm evidence for the ISHE in Pt metal, but the
angular momentum dissipates to the environment magnitude of the SHC could not be derived
from the ferromagnet, that is, the flow of the spin (Saitoh et al., 2006). Recently, an extension of this
current is produced from the ferromagnet to the experiment has been done by the same group using
reservoir. In the present configuration, the reservoir similar devices (Ando et al., 2008a,b). In this experi-
is the nonmagnetic metal. To summarize, the spin ment, the magnetization relaxation in NiFe is
current with the polarization direction parallel to controlled by the spin current injection from Pt
the averaged magnetization is injected to the non- by the SHE. They have carefully studied the
magnetic metal from the ferromagnet by the change in the ferromagnetic resonance signal,
ferromagnetic resonance. This injected spin current dI(H)/dH, due to the charge current Jc in Pt
will produce the charge current jc in the nonmagnetic layer. The idea here is just opposite to that in
metal through the ISHE as expressed by the above experiment. For example, the injected
jc ¼ DISHE js ð59Þ
spin current to NiFe will modify the Gilbert
damping, which depends on the direction of the
where DISHE is a coefficient for ISHE efficiency. In spin current polarization. They studied the reso-
the original experiment by Saitoh et al. (2006) and nance lineshape for different values of the current
Inoue et al. (2007), Ni81Fe19/Pt sample with the in Pt. Note that the change in the spectra is
250 Spin Hall Effect
d/(H)/dH (a.u.)
Ni81Fe19/Pt
10–4V/T
dV (H )/dH
0 0
Ni81Fe19
Ni81Fe19/Pt
100 150 100 150
H (mT) H (mT)
(c)
ISHE (d) Ni81Fe19 (e) Ni Fe /Pt
81 19
dV/dH HFMR H 10–5V/T 10–4V/T
calc.
dV (H)/dH
exp.
AHE
0
dV/dH H
0 = 90°
HFMR
100 150 120 140
H (mT) H (mT)
Figure 21 (a–e) Experimental data for the ISHE in Pt. From Saitoh E, Ueda M, Miyajima H, and Tatara G (2006)
Conversion of spin current into charge current at room temperature: Inverse spin-Hall effect. Applied Physics Letters
88: 182509.
symmetric for positive and negative direction of drop due to ESHE is measured as a function of the
the current for = 0 ( is the angle between the temperature difference, T, between the two ends
external magnetic field and the charge current) of the sample, the applied magnetic field H, and
suggesting that the heating effect is the dominant also the position xp of the attached Pt metal mea-
contribution. When = 90 , on the other hand, the sured from the center of the sample. Results
spectral change is clearly different for the different showed that r is almost constant along the x-
direction of the charge current. By examining the direction, which suggests that the temperature gra-
asymmetric part, that is, dI(H,jc)/dH dI(H,jc)/ dient term, that is, the second term, gives
dH, they could estimate the change of the Gilbert the dominant contribution to r in the expres-
damping coefficient due to the charge current sion r = (qc/qn)rn + (qc/qT)rT er
jc through the SHE. Considering the values of spin (c: spin-dependent chemical potential for spin ,
diffusion lengths
Pt = 7 nm,
Py = 3 nm, and : scalar potential). It has been argued that the
1
the conductivities Pt c = 0.064(m cm) and contribution from rn decays quickly within the
Py 1
c = 0.065(mcm) , they obtained the spin Hall length scale of the spin diffusion length (5 nm)
angle for Pt as 0.08, which is much larger than much shorter than the size of the sample (6 mm),
0.0037 obtained by Kimura et al. (2007). and does not contribute to the bulk of the sample.
The ISHE has been recently applied to the From this measurement, the estimated spin
experimental observation of the spin Seebeck effect Seebeck coefficient for Ni81Fe19 is
(Uchida et al., 2008). With the magnetic field S spin > 2 nV K1 at T = 300 K, which is orders of
applied also along the x-direction, the spin current magnitude smaller than the usual Seebeck coeffi-
is produced by the difference of the chemical cient S > 20 mV K1. In addition to the studies
potentials " and # for up- and down-spins, mentioned above, the increasing number of publi-
respectively. In the attached Pt metal, where the cations address the metallic SHE (Fan and Eom,
decay of the chemical potential difference, = 2008; Harii et al., 2008; Kent, 2006; Ma et al., 2008;
" #, occurs along the z-direction, the ISHE Otani and Kimura, 2008; Song et al., 2006a; Stern
occurs and results in the electric field ESHE or the et al., 2007, 2008; Valenzuela and Tinkham, 2007;
voltage drop along the y-direction. This voltage Vila et al., 2007).
Spin Hall Effect 251
1.07.3.2 Theories of SHE in Metals extension of the all-electron linear muffin-tin orbital
method based on the density functional theory with
Theoretically, both the intrinsic and extrinsic
local density approximation. Figure 22 shows the
mechanisms can contribute to the SHE in metals as
relativistic band structure of Pt, and also the SHC
in the case of semiconductors. From the studies on the
s, as a function of EF. The 6s bands are extending
AHE, the microscopic mechanism depends strongly
broadly from 10 eV to higher energy, while the 5d
on the degree of disorder characterized by the long-
itudinal resistivity or conductivity c (Onoda et al., bands are within the range of 7 eV < " < 1 eV.
2006b; Miyasato et al., 2007). For very clean system Therefore, the electronic states near EF = 0 are the
where c is larger than 106 1 cm1, the extrinsic mixture of the 6s and 5d bands. The situation is
SS is expected to be dominant where the spin Hall different for Au, where one extra electron is added
angle is the proper quantity. In the intermediate compared with Pt, and the Fermi energy is shifted
region where c lies between 104 and 106 1 upward. The electronic states near the Fermi energy
cm1, the intrinsic contribution is the dominant one, are mostly 6s and 6p bands in this case. Therefore, we
where s is of the order of 103 1 cm1. This value need to look at the electronic structure in more detail
of 103 1 cm1 corresponds to e2/ha with a > 5 Å to understand the mechanism of SHE. As shown in
as the lattice constant, representing the QH conduc- Figure 22 the SHC s peaks at the true Fermi level
tivity in 3D. Therefore, it is reasonable to assume the (0 eV), with a large value of 2200 1 cm1. This
intrinsic mechanism for Pt metal since the conductiv- gigantic value of the SHC is orders of magnitude
ity of Pt at room temperature is of the order of larger than the corresponding value in p-type semi-
105 1 cm1. The first-principles band structure cal- conductors such as Si, Ge, GaAs, and AlAs.
culation for the intrinsic contribution has been done Furthermore, the calculated SHC in simple metal Al
(Guo et al., 2008). Note that the SHE does not simply is only 17 1 cm1, being two orders of magnitude
scale with the SO interaction strength, which suggests smaller than that of Pt.
that the perturbative treatment of SO interaction is One important point is the degenerate band struc-
not valid. Pt shows particularly large SHE surviving ture at L- and X-points near EF in the scalar-
even up to room temperature among the 5d elements relativistic band structure (i.e., without the SO
with the similar SO interaction strength. The band coupling). These double degeneracies are lifted by
structure of Pt is calculated using a fully relativistic the SO coupling, with large SO splittings (0.66 and
4 4
3 3
SOC
2 noSOC 2
1 1
0 0
–1 –1
–2 –2
Energy (eV)
–3 –3
–4 –4
–5 –5
–6 fcc Pt –6
–7 –7
–8 –8
–9 –9
(a) (b)
–10 –10
–11 –11
W L Γ X W Γ –20 0 20
σxyz (102 Ω–1cm–1)
Figure 22 Band structure and SHC as a function of Fermi energy for Pt. From Guo GY, Murakami S, Chen TW, and
Nagaosa N (2008) Intrinsic spin Hall effect in platinum: First-principles calculations. Physical Review Letters 100: 096401.
252 Spin Hall Effect
0.93 eV, respectively). As has been discussed for and hence the SHC shows the maximum when EF
p-type GaAs, the magnetic monopoles contribute lies within the gap. One can also argue from this
resonantly to the SHC, and this structure is realized effective Hamiltonian that the intrinsic SHE is
at L-and X-points. This is confirmed by the momen- robust against impurity scattering since the vertex
tum-resolved contribution to the SHC as shown in corrections for the SHC from the short-ranged
Figure 23 where the contribution from the region impurity scattering vanish in the clean limit due to
near L- and X-points are dominant. It has been also the symmetry H(k) = H(k). Therefore, the intrin-
found that the SHC decreases monotonically as the sic contribution in Pt is expected to be robust
temperature T is raised, as shown in the inset of against the disorder potential, and the observed
Figure 23. This rather strong temperature depen- SHE is most probably of the intrinsic origin.
dence is also due to the near degeneracies since the A related work has been done by Kontani et al.
small energy scale is relevant to the SHC. (2008) using the tight-binding Hamiltonian for the
Nevertheless, the SHC s > 240 1 cm1 at t2g orbitals. Sr2MO4 (M = Ru,Rh,Mo) has been
T = 300K is still large and is close to the measured explicitly considered, and the large SHC of the
value (>200 1 cm1). The calculated SHC for Al order of 700 1 cm1 was obtained for Sr2RuO4
at T = 4 and 300 K is 17 and 6 1 cm1, respec- (Kontani et al., 2007, 2008). They also calculated
tively. The former value is similar to the the orbital Hall conductivity corresponding to the
experimental values (27, 34) at 4.2 K. orbital current. Tight-binding model is convenient
To understand the microscopic mechanism of to draw an intuitive picture in the real space. The
the SHE, the effective Hamiltonians H(k) for the transfer integrals among the t2g orbitals become
two doubly degenerate bands at X- and L-points complex in the presence of the SO interaction as
have been constructed. This 4 4 matrix can be has been discussed in the context of the AHE by
written as the linear combination of matrices, Onoda and Nagaosa (2002). This phase of the trans-
which is similar to the Luttinger model. The pre- fer integral can be regarded as the Peierls phase for
vious analysis for the p-type semiconductors can be the fictitious magnetic field, and induces the
equally applied to the present case. An important Aharonov–Bohm effect in the crystal. A related
issue is whether the contributions from conduction paper by Tanaka et al. (2008) considered the SHE
band and valence bands cancel or not. For example, and the orbital Hall effect in 3D 4d and 5d transition
the contribution to the SHC becomes zero or finite metals. The vertex correction relevant to the SS
when the Fermi energy is within the gap between contribution has not been considered in this work.
the two bands. This issue has been discussed in the Therefore, the obtained results are for the intrinsic
context of SHI, and the inverted band structure as in mechanism only.
the case of HgTe has the finite SHC even if the As for the giant SHE observed in Au by Seki et al.
Fermi energy is within the gap, while it vanishes in (2008), the role of local electron correlation has been
the usual band structure as in the case of GaAs. The considered by Guo et al. (2009). They considered the
present case of Pt corresponds to the former one, vacancy, Fe, and Pt as the possible defects in Au and
2
σ (103 Ω-cm–1)
300
Ωz(k) (atomic units)
1
200
0
100 0 150 300
T (K)
0
W L Γ X W Γ
Figure 23 Momentum-resolved contribution to the SHC. Inset: The temperature dependence of the SHC. From Guo GY,
Murakami S, Chen TW, and Nagaosa N (2008) Intrinsic spin Hall effect in platinum: First-principles calculations. Physical Review
Letters 100: 096401.
Spin Hall Effect 253
calculated the local density of states for each case. 1.07.4 Quantum Spin Hall Effect
Only in the case of Fe in Au, there appears the
resonant electronic states at the Fermi energy which The QSHE and the topological insulator have been
can contribute to the enhanced SS. Fe in Au has been theoretically proposed and experimentally con-
known as a representative Kondo magnetic impurity firmed. In the present section, we review these
with a low Kondo temperature TK = 0.4 K. Therefore, recent developments.
naively, it is not relevant to the giant SHE observed at
room temperature. However, the detailed first-princi-
ples calculation has shown that the Kondo effect in Fe
1.07.4.1 Inverted Band Structure and SHI
is orbital-dependent in nature. Previously, the crystal
field splitting between eg and t2g orbitals is around From the systems discussed so far, we can extend
0.1 eV, which is much smaller than the hybridization our discussion to insulators. We can see that the
energy > 1 eV, and has been neglected. On the SHE can be nonzero in some classes of insulators,
other hand, the LDA + U calculation shows that the which we call SHIs in Murakami et al. (2004a).
electron correlation induces the large orbital polariza- The reason why the SHI arises can be understood
tion and the energy splitting between eg and t2g orbitals in terms of band structure. The SHE is caused by
is of the order of 2 eV. This means that the spins in eg the SO coupling. In terms of the band structure,
orbitals are in the Kondo limit, while the electrons in the respective bands are classified into multiplets
t2g orbitals are in the mixed-valence region. The for- of the angular momentum. SHE appears if the
mer ones lead to the Kondo effect at low temperature, fillings of the bands within the same multiplet
while the latter can contribute to the large SHE since are different. Instead, the bands within the same
the SO interaction is active in the t2g orbitals while it is multiplet are all empty or all occupied, and their
quenched in eg orbitals. Another scenario for the contribution to the SHE cancels each other. In the
Kondo effect of Fe in Au has been recently proposed Luttinger model representing the valence bands of
by Costi et al. (2009). cubic semiconductors, the J = 3/2 multiplet con-
Actually, taking into account the SO interaction in tains the LH and HH bands; the SHE appears
the first-principles calculation, the energy splitting because the fillings of the HH band and the LH
between the effective m = 1 and m = 1 states occurs band are different.
(m: in the z-component of the orbital angular In the semiconductors such as -Sn, HgTe, HgSe,
momentum). According to Engel et al., (2005), the and -HgS, the band structure is inverted from the
spin Hall angle S = s/xx due to the SS is given cubic semiconductors such as GaAs (Figure 24(a)).
by the formula In the inverted band structure (Figure 24(b1)), the
Z order of the bands is inverted from LH–HH–CB to
dI ðÞSðÞsin CB–HH–LH. The mathematical structure of the
S ¼ Z ð61Þ Hamiltonian is essentially the same with the cubic
dI ðÞð1 – cosÞ semiconductors. The only change is that the energy
difference at the point between 6 (CB) and 8
where is the angle between the incident and scat- (LH and HH) bands has a different sign. Thus, the
tered waves, is the solid angle, S() is the skewness original LH band and the CB become a CB and the
function, and I() is the strength of the scattering. S() valence band in the inverted semiconductors, respec-
and I() are given by the spin-dependent phase shifts tively. Hence, in the hole picture the LH band is fully
of the scattering, and the large energy splitting occupied and the HH band is empty, and the result-
between m = 1 states suggests that S can be as ing SHC is nonzero even for the band insulators. In
large as 0.1. The reason why S cannot be larger these semiconductors with inverted band structure,
than 0.1 is that the phase shift 1 for the p-wave the bandgap is zero, which comes from the degener-
scattering is needed as first noted by Fert and Jaoul acy at the point due to the cubic symmetry. When
(1972). The SHE due to the SS by Ce and Yb impu- one breaks the cubic symmetry by uniaxial strain, for
rities has also been considered recently (Tanaka and example, there opens a finite gap, while the SHC
Kontani, 2008), and it is concluded that the spin Hall remains nonzero (Figure 24(b2)). This nonzero
angle S is given by 8 sin(2/7), expecting the SHC is a result of the nature of the gap; the gap is
giant SHE. opened by the SO coupling.
254 Spin Hall Effect
(a) Conventional (b1) Zero-gap (b2) Zero-gap + Uniaxial strain (c) Narrow-gap
(GaAs) (HgTe) (HgTe) (PbTe)
CB LH
LH
EF EF EF EF
HH
HH
HH
LH
CB CB
π (111)
k= a
k=0 k k=0 k k=0 k k
Figure 24 Band structures of (a) Cubic semiconductors such as GaAs, zero-gap semiconductors (b1) without strain and
(b2) with uniaxial strain along z-axis, and (c) narrow-gap semiconductors such as PbTe. The LH, HH, and CB in (a) stand for
the light-hole, heavy-hole, and conduction bands. The LH, HH, and CB correspond to the LH, HH, and CB bands in (a).
Narrow-gap semiconductors such as PbTe, PbSe, "xy = e2/h. Meanwhile, for the 2D down-spin subsys-
and PbS are also SHI. The bandgap in these narrow- tem, the magnetic field is opposite, and #xy = e2/h.
gap semiconductors is also caused by the SO They have gapless chiral edge states having opposite
coupling, and the SHC is nonzero. The crystal struc- sense of rotations. Superposition of these two subsys-
ture is that of the rocksalt. The direct gap of the size tems then results in two gapless edge states which have
0.15–0.3 eV is at the four L-points (Figure 24(c)). opposite spins and flow directions, thus carrying a spin
The SHC for these compounds is around 0.04e/a, current. This phase realizes topological order, as is
where a is the lattice spacing (Murakami et al., 2004a). similar to the QH phase. This QSH phase is character-
In the SHI described above, as in contrast with the ized by the Z2 topological number, as we discuss below.
QSH system discussed later, there are no gapless The whole system preserves the time-reversal
edge channels. It is then a question of how the spin symmetry. This state requires a field, which acts as
current flows in the SHI. The nature of the spin +ẑ magnetic field for up-spin and ẑ magnetic field
current in the SHI was numerically studied using for down-spin. It is not achievable by an external
the Keldysh formalism (Onoda and Nagaosa, magnetic field, but instead it is effectively achieved
2005b). It is revealed that the spin current flows by the SO coupling, which preserves time-reversal
along the edge only when an electrode is attached symmetry. In some materials the SO coupling is
along the edge. This may imply that by providing the strong enough to realize this topological phase, as
edge channel from the electrode, the SHI can carry we discuss subsequently. There is one important
the spin current. aspect of the QSH system QSHS, which is absent in
the QH system; the above example with two QH
subsystems conserves the spin sz, while the SO cou-
1.07.4.2 QSHE and Z2 Topological
pling does not conserve sz in general, hybridizing the
Invariants in 2D and 3D
two subsystems together. The question is whether
The QSHE is theoretically proposed as the spin there is still some interesting physics even in the
analog of the QHE (Kane and Mele, 2005a,b; absence of sz conservation. Remarkably, the answer
Bernevig and Zhang, 2006; Fu et al., 2007a; Fukui is positive; the edge states are robust against pertur-
and Hatsugai, 2007a; Hatsugai et al., 2006; Sheng bations which do not break the time-reversal
et al., 2005c). The QSH phase is possible both in 2D symmetry, such as nonmagnetic impurities or inter-
and 3D. We first consider the 2D QSH phase. In the action (Wu et al., 2006; Xu and Moore, 2006). This
phase, the bulk is gapped while the edge is gapless shows the peculiarity of the edge states.
carrying a spin current. The key concept of the QSHS is the time-reversal
The simplest version of the 2D QSH can be con- symmetry, which gives rise to the Kramers degener-
structed as a superposition of two QH subsystems with acy between k and k. In the band theory, the wave
opposite spins (see Figure 25). For the 2D up-spin numbers satisfying k X k (mod G) play an important
subsystem, we apply an external perpendicular role. Such momenta are called the time-reversal-
magnetic field (jj + ẑ ), realizing the QH state with invariant momenta (TRIM), and are expressed in 2D
Spin Hall Effect 255
B B
Spin
1
as k = i where i¼ðn1 n2 Þ ¼ ðn1 b1 þ n2 b2 Þ with n1, for I-asymmetric systems (Fu and Kane, 2006).
2
n2 = 0, 1 and b1, b2 are reciprocal lattice vectors Here, w depends on the U(1) phase of the Bloch
(Kane and Mele, 2005a; Fu and Kane, 2006; Fu wave function; the phase should be chosen to be
et al., 2007a). The degeneracy is different depending smooth over the whole Brillouin zone. On the
on the presence or absence of the spatial inversion other hand, for I-symmetric systems, due to double
symmetry I. In I-asymmetric systems the Kramers the- degeneracy of each state, the calculation of i
orem implies double degeneracy only at the TRIM, involves the SU(2) gauge choice of juk,mi. In fact,
and at other wave numbers k the eigenenergies are free however, for I-symmetric systems the formula for
from degeneracies. On the other hand, in I-symmetric the index i is even simpler (Fu et al., 2007a):
system the eigenenergies are degenerate at all k. Y
N
The QSH phase is characterized by the Z2 topo- i X 2m ði Þ ð64Þ
logical number taking only two values = 0 and m¼1
= 1. In some literature they are called as = even where 2m(i)(=1) is the parity eigenvalue of the
and = odd, respectively, or X (1) = 1. When Kramers pairs at each of these points for I-symmetric
= odd (or = 1), the system is in the QSH phase, systems. This shows a crucial role of the I-symmetry
and when = even (or = 0), the system is an ordin- in the theory of the QSHSs. This is because the I
ary insulator. It is worth noting that is calculated symmetry relates between k and k, as does the
from the Bloch wave function in the bulk as time-reversal symmetry. The Z2 topological number
explained in the next section. in 2D is defined as
The Z2 topological number in noninteracting Y
clean systems is defined in the following way (Fu and ð – 1Þ ¼ i ð65Þ
i
Kane, 2006; Fu et al., 2007a). Let N denote the num-
ber of Kramers pairs below EF. First, we define a where the product over i runs over the four TRIMs in
(2N) (2N) matrix w, defined as the 2D Brillouin zone. The Z2 topological number
allows various equivalent expressions. For example, in
wmn ðkÞ ¼ u – k;m h j U juk;n ð62Þ the I-asymmetric systems, the Z2 topological number
can define how the wave functions are glued together at
where U is the time-reversal operator represented as
the TRIMs (Fu and Kane, 2006; Moore and Balents,
U = iyK, with K being complex conjugation. uk,m is
2007; Roy, 2006a). We classify the wave functions
the periodic part of the mth Bloch wave function
below EF into two classes juk,2j 1i (j = 1, 2, . . ., N),
(m = 1, 2, . . ., 2N). This matrix w(k) is unitary at any
and juk,2ji (j = 1, 2, . . ., N). When the Z2 topological
k, and is also antisymmetric at k = i. Then, for each
number is trivial ( = 0), by a judicious unitary trans-
TRIM we define the index i as
formation, one can make these states to satisfy
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
det wði Þ
i X ð63Þ uk;2j ¼ Uuk;2j – 1 ; uk;2j – 1 ¼ – Uuk;2j
Pf wði Þ ð66Þ
256 Spin Hall Effect
We note that U2 = 1. When = 1, in contrast, it is is gapped for both cases and looks similar. The topo-
not possible for (66) to hold, but there should neces- logical order, which is not evident in the bulk
sarily appear additional phase factors. Notice that bandstructure, appears as the existence of the robust
the wave functions are degenerate only at the edge states. The correspondence between the bulk
TRIMs. The nontrivial Z2 topological number and the edge is as follows:
works as an obstruction for the wave functions to
be glued at the TRIMs in the simple way given in • bulk: Z2 topological number is = 1 ( = 0) and
(66) (Fu and Kane, 2006; Moore and Balents, 2007; • edge: there are odd (even) numbers of Kramers
pairs of gapless edge states.
Roy, 2006b). because the topological order survives
perturbations, such as nonmagnetic disorder and Because an odd number cannot be zero, the system
interactions, the definition of the Z2 topological with = 1 should have at least one Kramers pair of
number can be further extended (Qi and Zhang, gapless edge states. This bulk–edge correspondence
2008) to systems with nonmagnetic disorder (Kane can be shown by the argument similar to the well-
and Mele, 2005b) and those with interactions (Lee known Laughlin’s gedanken experiment (Fu and
and Ryu, 2008). Kane, 2006). Suppose that we consider the system
To illustrate the relationship between Z2 topolo- on a ribbon, with two opposite ends attached
gical number and the edge states, we calculate the (Figure 27). In one direction the system is periodic,
bandstructure for geometries with edges in the while in the other there are edges. Then, when we
Kane–Mele model (Kane and Mele, 2005a,b), which increase the flux penetrating the hole from zero to
is the 2D tight-binding model showing the QSH and half the flux quantum, the change of the system
insulator phases by changing some parameters (for characterized by the time-reversal polarization (Fu
details, see Section 1.07.4.4). Let us consider a ribbon and Kane, 2006) depends on the Z2 topological num-
geometry, which is finite in one direction and is ber, and it is also related with the number of Kramers
infinite in the other. The phase diagram is in the pair of edge states.
inset of Figure 26, where
R,
SO, and
v are The analogous phase is also possible in 3D (Moore
model parameters. The QSH and I in the phase and Balents, 2007; Fu et al., 2007a; Roy, 2006b). In this
diagram represent the QSH and insulator phases, phase in 3D, the system is an insulator in the bulk and
respectively. In the QSH phase (Figure 26(a)), supports gapless surface states carrying spin currents.
there exists gapless edge states, irrespective of the In 3D, the Z2 topological numbers are defined
geometry. In contrast, for the insulator phase in the following way. The TRIMs are
(Figure 26(b)) there are no gapless edge states. In i¼ðn1 n2 n3 Þ ¼ 1=2ðn1 b1 þ n2 b2 þ n3 b3 Þ with n1, n2,
fact, there are edge states, but they do not go across n3 = 0, 1 in 3D. There are four Z2 topological numbers
the gap. In some cases, in general, these edge states in j (j = 0,1,2,3) (Moore and Balents, 2007; Fu et al.,
the insulator state may cross the Fermi energy; 2007a; Teo et al., 2008), given by
nevertheless, even if they cross the Fermi energy, it
is not an intrinsic property, and the crossing can Y
8 Y
ð – 1Þ0 ¼ i ; ð – 1Þk ¼ i¼ðn1 n2 n3 Þ ð67Þ
disappear by perturbations preserving time-reversal i¼1 nk ¼1;nj 6¼k ¼0;1
symmetry. On the other hand, the bulk bandstructure
1
E/t
5 λ R / λ so I
QSH
0 0
–5 λ V / λ so
(a) –5 0 5 (b)
–1
0 π ka 2π 0 π ka 2π
Figure 26 Band structures of the Kane–Mele model on a ribbon geometry, for parameters in (a) the QSH phase and (b) the
insulator phase. Inset: phase diagram of the Kane–Mele model. From Kane CL and Mele EJ (2005) Z2 topological order and
the quantum spin Hall effect. Physical Review Letters 95: 146802.
Spin Hall Effect 257
(a) (b)
Gapless
Inversion Inversion
symmetric symmetric
I QSH I QSH
δ δ
Gapless
m m
Figure 28 Universal phase diagram between the QSH and the insulator (I) phases in (a) 3D and (b) 2D (Murakami, 2007;
Murakami et al., 2007; Murakami and Kuga, 2008).
V I 0
Figure 29 Schematic of the helical edge states. From Kane CL and Mele EJ (2005a) Quantum spin Hall effect in graphene.
Physical Review Letters 95: 226801.
open a gap. It is demonstrated experimentally in neighbor bond from i to j. ij takes the value 1
CdTe/HgTe/CdTe QW that the two-terminal con- depending on whether the next-nearest-neighbor
ductance decreases rapidly when the magnetic field hopping from j to i is clockwise or counterclockwise.
is increased (König et al., 2007, 2008). A fractional i takes the value 1 for the A and B sublattices,
charge appears when the time-reversal symmetry is respectively. The phase diagram is as shown in the
broken by magnetic domains put on the edge, as inset of Figure 26. We note that there are other
discussed in Section 1.07.4.6.1. It is proposed theore- tight-binding models showing the QSH phase
tically that the proximity effect to the ferromagnets (Onoda and Nagaosa, 2005b; Qi et al., 2006a,b). This
with the helical edge channels introduces the mag- model shows the QSH and the insulator phases, by
netic domain wall and a fractional charge e/2, changing the model parameters. The phase diagram
which can be detected as the shift of the Coulomb is shown in the inset of Figure 28.
blockade periodicity (Qi et al., 2008b). A theoretical proposal of the 2D QSH on the
cubic semiconductor with strain gradient has been
made (Bernevig and Zhang, 2006). In this setup, the
1.07.4.4 Models and Candidate Materials strain gradient plays a role of a spin-dependent
for QSHE effective magnetic field through the SO coupling,
The QSHE was first discovered in a model for gra- and gives rise to the spin-dependent QH effect, that
phene by Kane and Mele (2005a,b). This model is is, the QSH effect. In this proposal, the Landau
given as levels are formed, and in the presence of interaction,
X X there might be a chance to have exotic states similar
HKM ¼ t ciy cj þ i
SO ij ciy sz cj to the fractional QH states.
ði;j Þ hhi;j ii
X X The 3D QSH phase is realized in the model
þ i
R ciy ðs b
kij Þz cj þ
v i ciy ci ð69Þ proposed by Fu et al. (2007a). It is the tight-binding
hi;j i i
model on a diamond lattice:
where hi,ji runs through the nearest-neighbor pairs X X
(i,j), and hhi,jii runs through the next nearest-neighbor H ¼t ciy cj þ ið8
SO =a2 Þ ciy s?ðd1ij d2ij Þcj ð70Þ
pairs (i,j). bkij is the unit vector along the nearest- hij i ðhij iÞ
Spin Hall Effect 259
touch at k = 0. Thereby the parity of the valence alloy is proposed to be the 3D QSH phase, that is, the
band and the Z2 topological number changes have STI (Fu and Kane, 2007; Fu et al., 2007a). The angle-
been predicted theoretically in advance (Bernevig resolved photoemission spectroscopy (ARPES) is used
et al., 2006). In the experiments (König et al., 2007, for this alloy, and the Fermi surface of the surface state
and –
2008) the charge conductance, G, is measured in a crosses the M line five times (see Figure 33).
two-terminal setup, and found to be G 2e2/h when From the bandstructure calculation, the indices i at
and –
d > dc and G 0 when d < dc. An effect of the external M are opposite in sign as predicted theoretically
magnetic field has also been studied. As the QSHE is (Fu and Kane, 2007). This means that there should be
and –
fragile to perturbations which breaks time-reversal odd number of Fermi points between M . Thus,
–
and indicate that it is
symmetry, the charge conductance G shows a rapid the five crossings between M
decrease when the magnetic field is applied. This indeed the 3D QSH phase (Hsieh et al., 2008). Further
means that in the presence of magnetic field, there relevant references in this regard are Dai et al. (2008),
appears a gap in the helical edge channel to suppress Day (2008b), Hirahara et al. (2006), König et al. (2008),
the conductance (Figures 31 and 32). and Koroteev et al. (2008).
The 3D QSH phase has also been observed experi- Another experiment on the 3D QSH is carried out
mentally in the alloy Bi0.9Sb0.1 (Hsieh et al., 2008). This on Bi2Se3 (Xia et al., 2008). The ARPES measurement
16
14
R = h/(2e2)
12
10
Rxx / kΩ
8
6
4
2 1 μm × 1 μm, 1.8 K
1 μm × 0.5 μm, 1.8 K
0 1 μm × 1 μm, 4.2 K
–1 0 1 2
(Vg – Vth) / V
Figure 31 Conductance as a function of gate voltage. From König M, Buhmann H, Molenkamp LW, et al. (2008) The
quantum spin Hall effect: Theory and experiment. Journal of the Physical Society of Japan 77: 031007.
0.14 0°
0°
(B|| x)
0.12
15°
30°
0.10 45°
60°
G (e 2/h)
0.08 75°
90°
0.06 (B|| z)
0.04
x
0.02 I
y z
0.00 90°
–0.10 –0.05 0.00 0.05 0.10
B (T)
Figure 32 Conductance as a function of external magnetic field. From König M, Buhmann H, Molenkamp LW, et al. (2008)
The quantum spin Hall effect: Theory and experiment. Journal of the Physical Society of Japan 77: 031007.
Spin Hall Effect 261
0.1
Topological Hall insulator
1 2 3 4, 5
0.0
EB (eV)
–0.1
Figure 33 ARPES data for Bi0.9Sb0.1. From Hsieh D, Qian D, Wray L, et al. (2008) A topological Dirac insulator in a
quantum spin Hall phase. Nature 452: 970.
showed that in Bi2Se3 the surface states traverse the half-filled. In the ground state, dj = d is uniform,
bulk gap. The dispersion of the surface states forms a which opens the gap. In the real space, this corre-
single Dirac fermion at the point, meaning that sponds to the periodic array of dimers. For each
Bi2Se3 is the STI if it is insulating in the bulk. dimer, there are bonding and antibonding orbitals
Nevertheless, in reality the material is n-doped in and an electron occupies the bonding orbital.
nature, and it is not even insulating in the bulk. If the Therefore, the electron number per atom is 1/2.
carriers are compensated and the material becomes There is an energy gap corresponding to the energy
insulating in the bulk, it will become the STI. First- splitting between the bonding and antibonding
principles calculations have been done on this and states. Now, we consider an isolated atom separated
related materials (Zhang et al., 2009), and it was pre- from the dimers. This can be regarded as the kink or
dicted that Bi2Se3 is really the STI. In addition, domain wall between the two degenerate dimeriza-
Sb2Te3 and Bi2Te3 are also predicted to be STI, tion patterns, and called a soliton. Because the
while they have a smaller gap than Bi2Se3. isolated atom has one orbital decoupled from
the other atoms, it has the energy between those of
the bonding and antibonding orbitals in the dimers.
1.07.4.6 Further Developments in the QSHE Let us define here the charge of the soliton. An
important remark here is that the charge is measured
QSHS and topological insulator are the subject of
from that of the ground state. For example, we count
intensive studies, and constitute a growing field. We
the charge of the soliton from that of the ground-state
mention here some of the interesting recent devel-
configuration. As noted above, there is half electron
opments and future directions.
per atom in the ground state, that is, e/2 charge.
Therefore, the charge of the soliton q is defined as
1.07.4.6.1 Spin-charge separation and
q = (e) (e/2) = e/2 if an electron occupies the
charge fractionalization
soliton site, while q = 0 (e/2) = e/2 if the soliton
The simplest example of the spin-charge separation
site is empty. This means that the soliton charge is a
is the soliton excitations in the conjugate polymer
fraction (half) of the unit charge e. This corresponds to
model such as polyacetylene. Suppose we have the
the case of jdjj = t, but the physics remains qualita-
tight-binding model (Niemi and Semenoff, 1986)
tively the same also in the limit of jdjj << t. In this case,
X
H¼ – ðt þ ð – 1Þj dj Þðcjy cj þ1 þ h:c:Þ ð71Þ the field theoretical methods can be applied (Niemi
j and Semenoff, 1986). In the continuum approxima-
where there is one orbital at each atom j, coupled to tion, the Hamiltonian reads
the atom at j + 1 by the transfer integral t + (1)jdj, Z
with dj being the dimerization. We assume that the H¼ dx y
ðxÞ½z qx þ x mðxÞ ðxÞ ð72Þ
average electron number is 1/2 per atom, that is,
262 Spin Hall Effect
where (x) = t( 1(x), 2(x)) is the two-component spinfull electron model discussed above. The only
field operator for the right-going and left-going elec- difference is that the system is 2D and the soliton is
trons, and m(x) is the mass term. With the sign change replaced by the vortex with the flux . The phase
of the mass m(x) between the two limits x ! 1, a winding around this vortex guarantees the existence
mid-gap state is topologically guaranteed (Niemi and of the mid-gap state in the QSHS, and the occupancy
Semenoff, 1986). This is the simplest example of the of that mid-gap state determines the charge and spin
fractionalization but can be generalized to many quantum numbers quite analogous to that in the spin
situations in a straightforward way. and charge solitons (Ran et al., 2008a,b; Qi et al.,
Now, we consider the spinfull electrons. Each 2008b; Moore, 2008). Therefore, QSHS offers the
atom can accommodate two electrons, that is, up- first explicit model showing the spin-charge separa-
spin and down-spin electrons. Then, there is one tion in 2D, which has been examined for many years
electron per atom in the ground state. There are in the context of the theory of high temperature
four possibilities of the soliton in this case: (1) superconductors (Lee et al., 2006).
vacancy, (2) double occupancy, (3) up-spin electron,
and (4) down-spin electron. Let us count the charge q
of the soliton for each of (1)–(4) states. It reads (1): 1.07.4.6.2 Effective theory and quantized
q = 0 (e) = e, (2): q = (2e) (e) = e, (3): magneto-electric effect
q = (e) (e) = 0, and (4): q = (e) (e) = 0. In the case of QH systems, the effective theory, that
Therefore, the soliton with (1) or (2) configuration is, the Chern–Simons theory, describes the low-
is positively and negatively charged, respectively, energy phenomena. Therefore, it is desirable to
and is called charge soliton. On the other hand, the develop an effective field theory for the QSHS also.
soliton with (3) or (4) configuration is neutral. When Recently, an interesting work has been done in this
we consider the spin quantum number S, it is 1/2 for direction (Qi and Zhang, 2008). The idea is to define
the configuration (3) and (4), while (1) and (2) and the fundamental time-reversal invariant (TRI) insu-
also the ground state are the spin singlet. Therefore, lator in (4 + 1)-dimensions, described by the (4 + 1)-
the (3) and (4) forms the spin-doublet pair and is dimensional Chern–Simons theory and characterized
called the spin soliton. Summarizing the above con- by the second Chern number. Due to the time-rever-
sideration, the charge and spin are carried by the sal symmetry, the coefficient of the Chern–Simons
different kinds of soliton, and this phenomenon can term is restricted to be 0 or , which corresponds to
be called spin-charge separation. We have shown that the Z2 topological number discussed in Section
even though the constituent particle have both unit 1.07.4.2. By dimensional reduction, this state is
charge and spin, the excitation in the many-body related to the QSH system in (2 + 1)-dimensions
state could have fractional quantum numbers and and the topological insulator in (3 + 1)-dimensions.
also the spin-charge separation. As an explicit application of this effective theory, the
This idea has been applied to the QSHS. As quantized magneto-electric effect was predicted,
already noted in Section 1.07.4.3, the helical edge where an electric field produces the magnetization
modes in QSHS separate the right-going and left- in the same direction, with a universal constant of
going branches of the dispersion at the two edges of proportionality quantized in odd multiples of the
the sample for up- and down-spin electrons, respec- fine-structure constant ¼ e 2 =hc.
tively. Therefore, the degrees of freedom are reduced
to half at each edge, leading to the fractionalized
charge. The effective Hamiltonian for the helical 1.07.4.6.3 Superconducting QSHSs
edge modes at one edge can be written as (72) with The topological concept related to the time-reversal
1(x) ( 2(x)) corresponding to the right-going up- symmetry has been generalized to the superconduc-
spin electron (left-going down-spin electron), while tors. In particular, the superconducting proximity
the mass m(x) corresponds to the magnetic order effect between the surface of the (3 + 1)-dimensional
parameter. Therefore, at the magnetic domain wall STI and the usual s-wave superconductor has been
where m(x) changes sign, one expects the fractional discussed (Fu and Kane, 2008). The Hamiltonian
charge e/2 as discussed in Section 1.07.4.3 (Qi et al., reads
2008b). Z
In the bulk of the QSHS, there are both up- and H¼ dx½ y
ð – iv?r – Þ þ y y
þ h:c: ð73Þ
" #
down-spin states, and the situation is similar to the
Spin Hall Effect 263
where = t( " #) is the two-component spinor cor- number can still be defined, and the helical edge
responding to the surface Dirac fermion dictated by modes as the Andreev bound states appear when
the topology of the bulk state, while is the super- the order parameter p for the p-wave pairing is
conducting order parameter induced by the larger than that for the s-wave pairing s. An inter-
proximity effect. This Hamiltonian can be diagona- esting spin-transport property is found in the
lized to result in a superconducting state resembling Andreev reflection at the junction of the normal
the chiral p + ip spinless pairing state. Therefore, a metal and the noncentrosymmetric superconductor.
novel vortex satisfying the non-Abelian statistics Another recent development of this field is the math-
(Read and Green, 2000; Ivanov, 2001), and the chiral ematical classification of the possible topological
edge modes (Buchholtz and Zwicknagl, 1981; classes including both the insulators and supercon-
Matsumoto and Sigrist, 1999; Read and Green, ductors. It is concluded that there are five classes
2000) carrying Majorana fermions are expected. Fu according to the time-reversal symmetry and chiral
and Kane (2008) considered the geometry of the two symmetry. The readers are referred to the recent
superconductors with the order parameter 0 and papers (Roy, 2006a,b; Schnyder et al., 2008).
0ei attached to the topological insulator with the
separation W. Then, the energy dispersion of the 1.07.4.6.4 Localization problem related to
Andreev bound state is given by the QSHS
E(q) = [v2q2 + 0 2 cos2(/2)]1/2, with q being The effect of disorder on the electronic states is an
the wave number along the edge. The fermion opera- important issue in condensed matter physics. In par-
tor for this bound state constitutes the Majorana wire, ticular, it is known that the dimensionality,
and the circuit of this wire can be used to create, symmetry, and topological properties of the system
manipulate, and fuse Majorana bound state (Fu and determine the localization/delocalization behavior
Kane, 2008). of the wave functions (Lee and Ramakrishnan,
On the other hand, one can consider the (p ip)- 1985). There are basically three universality classes,
pairing state as the superconducting analog of the that is, (1) orthogonal, (2) symplectic, and (3) unitary
QSH state in parallel to the spinless chiral (p + ip)- classes, which corresponds to the (1) spinless time-
pairing state which can be regarded as the super- reversal system, (2) time-reversal system with SO
conducting QH system (Qi et al., 2008a,c; Goryo interaction, and (3) system with broken time-reversal
et al., 2008; Grover and Senthil, 2008; Roy, 2006c; symmetry. It is remarkable that all the states are
Sengupta et al., 2006). In the (p ip)-pairing state, the localized in 2D however weak the disorder strength
helical edge modes appear as the Andreev bound is in (1) and (3), while the metal–insulator transition
states. However, because the negative and positive occurs in (2). The integer QH system, which corre-
energy states are redundant in the Bogoliubov–de sponds to the case (3) but also is characterized by the
Gennes Hamiltonian, only the half of the fermion topological number called Chern number related to
degrees of freedom should be taken. This means that the quantized Hall conductance, has distinct locali-
the Bogoliubov quasi-particles at the helical edge zation properties. For example, there remain
modes are Majorana fermions. In the case of a vortex, extended states at the center of each Landau level
a pair of the Majorana fermion states at the core of (Girvin and Prange, 1987). Therefore, it is an intri-
the vortex appears. On this low-energy state, the guing problem if the similar modification of the
time-reversal symmetry operation acts as the super- localization properties due to the Z2 topological
symmmetry operation changing the even/odd of the numner occurs also in the QSHS, which belongs to
fermion number. This leads to a nontrivial nonlocal the symplectic class (2). Concerning this problem,
correlation of some order parameter. They also pro- Onoda et al. (2007) numerically studied the localiza-
pose the BW-phase of 3He as the (3 + 1)-dimensional tion length and its scaling behavior of the
superfluid analog of the topological insulator Kane–Mele model (Kane and Mele, 2005a,b). The
(Qi et al., 2008c). exponent describes the divergence of the localiza-
Recently Tanaka and Kontani (2008) have pro- tion length as a function of energy difference from
posed that the noncentrosymmetric superconductors the mobility edge, _ jE Ecj . By numerical cal-
can be a candidate for the superconducting QSHS. In culation, it is found to be 1.6, which is different
the presence of the SO interaction, the singlet and from the symplectic value 2.7 (Onoda et al.,
triplet pairing states are mixed, and one cannot 2007). On the other hand, the QSH phase is studied
define the spin Chern number. However, the Z2 by extending the network model (Obuse et al., 2007).
264 Spin Hall Effect
The result is consistent with the symplectic value the honeycomb lattice structure concludes that the
(Obuse et al., 2008a,b). system is a QSHS at room temperature (Shitade et al.,
For the 3D QSHS, the localization of the (2 + 1)- 2009). Due to the electron correlation, the novel
dimensional Dirac fermion at the surface is an inter- antiferromagnetic state is expected at lower tempera-
esting issue. A model of a single 2D Dirac fermion is ture. Further studies on the interplay between the
studied by constructing an effective theory includ- electron correlation and SHE are extremely impor-
ing a nontrivial topological term, which is tant especially in correlated oxides, which have many
responsible for the protection of the extended states advantageous features such as the systematic control
(Ostrovsky et al., 2007; Ryu et al., 2007). Therefore, of the band width, carrier doping, and the high
the localization is prohibited by the topology in the energy scales.
case of the surface states of the STI. From the view-
point of the nonlinear sigma model approach, the
topological terms are classified by the homotopy 1.07.5 Summary and Perspectives on
consideration of the target space, which is deter- SHE
mined by the symmetry of the problem, and are
1.07.5.1 Comparison between AHE and
restricted to the two cases: (1) there is no topological
SHE
term and (2) the topological sectors are classified by
Z2 number. The former corresponds to the usual SHE is closely related to the AHE in metallic ferro-
symplectic class and also the 2D topological insula- magnets as discussed in Section 1.07.1. Both are
tor, while the latter to the surface Dirac fermion of driven by the relativistic SO interaction, and intrinsic
the 3D STI. The readers are referred to some and extrinsic mechanisms are considered. More
further references on this issue (Sheng et al., 2006; explicitly, one may regard the SHE as the two copies
Essin and Moore, 2007; Obuse et al., 2008a,b; of AHE for up- and down-spins. This analogy is
Shindou and Murakami, 2009). correct in the zeroth order approximation, but there
are several differences as described below.
1.07.4.6.5 SHE in strongly correlated One important difference is that the charge is a
systems conserved quantity and the charge current in AHE is
One of the most important issues to be studied in the well defined as the Noether current associated with
future is the role of electron correlation in SHE the U(1) gauge symmetry, while the spin is not con-
(Raghu et al., 2008). Concerning this issue, an inter- served in the presence of the SO interaction and
esting material has been reported, that is, Sr2IrO4 correspondingly the continuity equation for the
(Kim et al., 2008). As the atomic number increases spin and spin current cannot be derived in generic
from 3d, 4d to 5d, SO interaction increases while the situations. Therefore, the definition of the spin cur-
electron correlation decreases since the extent of the rent is somewhat arbitrary, and the direct observation
d-orbitals gets larger and larger. In the case of 5d of the spin current is much more difficult than that of
orbitals, the SO interaction and the correlation charge current, as discussed earlier. This is why the
energy U are of the order of 0.5 eV, both of which experimental observation of the SHE has been rather
play important roles, while the crystal field splitting indirect, by measuring the consequent spin accumu-
between eg and t2g orbitals is much larger. In Sr2IrO4, lation near the edge of the sample, for example. In
Ir4+ ion has d5 configuration, and one hole occupies that case, the analysis of the spin diffusion equation
the t2g orbitals. Because the orbital angular moment is including the decay of the spin density due to the SO
active in t2g orbitals, the strong SO interaction mixes interaction, which is phenomenologically intro-
the spin and orbital states, leading to the effective duced, should be performed.
Jeff = 1/2 states and Jeff = 3/2 states. The single hole More direct detection of the spin current is in
occupies one of the doubly degenerate Jeff = 1/2 terms of the voltage drop perpendicular to it. This
states, and the Coulomb interaction U leads to the effect is the so-called Aharonov–Casher effect in the
Mott insulating state (Kim et al., 2008). This experi- vacuum, but is very small because the magnitude of
ment opens a promising route of the SHE in oxides the effect in the vacuum contains the rest mass of the
because the effective Hubbard model for Jeff = 1/2 electron mc2 in the denominator. In solids, the effec-
states contains the spin-dependent complex transfer tive SO interaction can be tremendously enhanced
integrals due to the SO interaction. Actually, a tight- by the factor of mc2/E with E being the bandgap.
binding analysis on a related material, Na2IrO3, with This factor can be as large as 106, and hence the
Spin Hall Effect 265
detection of the spin current in solids is much easier. and 1 is the phase shift for the p-wave scattering
This idea has been already pursued in the ISHE as (Fert and Jaoul, 1972). This can be of the order of
discussed in Section 1.07.3 for metallic SHE. 102 since
/ > 0.1 and 1 > 0.1 are possible but
In the case of AHE, the theoretical studies on the not larger. Therefore, the giant SHE observed in Au
cross-over between the extrinsic and intrinsic domi- (Seki et al., 2008) suggests an essential difference
nated regions have been developed, taking into between the AHE and SHE. For example, the SHE
account the band crossing near the Fermi energy is not a simple two copies of AHE for up- and down-
which resonantly enhances the AHE (Miyasato spins. This is natural since the x and y components of
et al., 2007; Onodaet al., 2006b). The representative the spin operator play some role in the quantum
model system is the Rashba Hamiltonian with the fluctuation, which leads to the singlet formation in
spin polarization. For example, the cross-over from the Kondo effect as discussed in Section 1.07.3.2 (Guo
the very clean metal, where the SS is dominant, to et al., 2009; Tanaka et al., 2008). This can be the
the usual metallic region, where the intrinsic contri- mechanism of the enhanced SHE compared with
bution determines H, occurs at h=> where is AHE. In any case, the role of the electron correlation
the transport lifetime, and is the SO energy. At this and the quantum fluctuation of the spins will be an
cross-over, the longitudinal conductivity xx is much important issue in the future.
larger than e2/h, that is, xx > (e2/h)("F/). If one
increases the disorder strength furthermore, the sec-
1.07.5.2 Comparison between Integer QHE
ond cross-over occurs at h=ð"F Þ 0:1 to the region
and QSHE
where a new scaling law H _ (xx) with the expo-
nent > 1.6 holds. This theoretical prediction The QHE and the QSHE have many aspects in
(Onoda et al., 2006b), when translated to the 3D common. They come from topological orders in
systems by replacing e2/h by e2/(ha) > 103 (Ohm gapped systems. The QH system is represented
cm)1 (where a is the lattice constant assumed to be by the topological number (the Chern number)
around 4 Å), is in agreement fairly well with the taking any integers: = . . ., 2, 1, 0, 1, 2,. . . . This
recent experimental studies over the many decades is because the underlying symmetry is the contin-
of the disorder strength (Miyasato et al., 2007). uous U(1) symmetry associated with the charge
Therefore, h=ð Þ and hence the absolute value of conservation. This Chern number is equal to the
the longitudinal conductivity is a key parameter to number of chiral edge states, and also appears as
control the behavior of the AHE, which resolves the the quantized Hall conductance xy = e2/h. In the
long-standing controversy. 2DEG in a strong magnetic field, this number is
From the viewpoint of the above results, the SHE equal to the number of Landau levels below the
in semiconductors is all in the strongly disordered Fermi energy. On the other hand, QSHE is related
region where the 1.6-power law is expected, but to the discrete symmetry, that is, the time-reversal
detailed study of the xx-dependence of s has symmetry, which is not related to the conservation
never been done. In the case of metals, on the other of a physical observable. Correspondingly, QSHS is
hand, the vital role of the band crossings is common characterized by the Z2 topological number taking
which leads to the enhanced intrinsic SHE in the only two values, = even and = odd, that is,
usual metallic systems as discussed in Section (1) = 1. As stated earlier, the simplest example
1.07.3.2. If the disorder is further reduced, the extrin- of the QSHS is realized as a superposition of two
sic SS contribution is dominant, and the SHC s is QH systems of opposite effective magnetic fields
proportional to the diagonal charge conductivity xx, for opposite spins. Only for such kinds of the
and the spin Hall angle S = s/xx characterizes the QSHSs, the corresponding topological number is
spin Hall response. The anomalous Hall angle is also the spin Chern number: s = " # = . . ., 4,
defined in a similar way, that is, = s/xx. The 2, 0, 2, 4, . . . . In this criterion, Z2 even (odd)
typical value of is of the order of 103, correspond- corresponds to s = 0(2)(mod 4). Nevertheless, gen-
ing to the ratio of the SOI and the Fermi energy. eric QSHSs allow terms which break spin
When the resonant scattering by the virtual bound conservation preserving the time-reversal symme-
state of d-orbitals is active, is of the order of try. With such spin-nonconserving terms, the set of
(
/)1 where
is the SOI energy, is the hybri- topological numbers is reduced to only two classes,
dization energy between the d-orbitals and the odd and even. This is the Z2 topological number,
s-bands, that is, the width of the virtual bound state, which characterizes the QSHS.
266 Spin Hall Effect
The edge states are also similar, whereas the 2007; Wang and Li, 2007; Yao and Niu, 2008; Zhang
spin-nonconserving terms make the QSH different et al., 2007). In this case the spin of light is nothing but
from a mere superposition of two QH systems. the circular polarized states of light. One can derive
The QH system has chiral edge states whose num- the EOM for the center of mass position rc and
ber is equal to the Chern number. They are momentum kc of the wavepacket of light including
immune to backscattering from impurities, since the lowest-order correction due to the finite wave-
the modes with different chiralities are separated length. It reads as
by the sample width. The QSHS has helical edge
states, which consist of pairs of Kramers-degener- dxc kc
¼ vðxc Þ þ k̇c ðzc jkc jzc Þ ð74Þ
ate edge states. When the spin, sz, is conserved, the dt kc
number of edge states is equal to the spin Chern dk c
¼ – ½rvc ðxc Þkc ð75Þ
number of the original QH system. When the spin, dt
sz, is no longer conserved (but time-reversal sym-
metry is preserved), some edge states may open a d jzc Þ
¼ – k̇c ?kc jzc Þ ð76Þ
gap; nevertheless, if the Z2 topological number is dt
odd, at least one Kramers pair of edge states where v(x) = 1/n(x) (n(x): refractive index) is the
remains gapless. velocity of light, and jz) = [z+, z] represents the
An important difference between the QH and polarization state of light. kc and kc are the SU(2)
QSHSs is that the QHE usually requires a strong connection and curvature matrices, which corre-
magnetic field, while the QSHE is realized in the spond respectively to An(k) and Bn(k) in the
absence of the magnetic field. In fact, the QHE Berry-phase theory of electron transport. The sec-
requires breaking of the time-reversal symmetry, ond term on the right-hand side of (74) is the
but not the uniform magnetic field; the QHE can anomalous velocity due to the Berry curvature.
arise from external staggered magnetic field This term will induce the shift of the trajectory of
(Haldane, 1988). However, in real systems thus far, light once the force, that is, the spatial gradient of
QHE arises only in external uniform magnetic field. the refractive index rn(x) is applied. As shown in
In 2D systems it can be realized because of the Figure 34 this leads to the Hall effect of light at the
cyclotron motion in the plane perpendicular to the refraction at the spatial change of n (Onoda et al.,
magnetic field. Due to the Landau level formation by 2004, 2006a). If we make the spatial change of n to be
the strong magnetic field, the motion perpendicular abrupt, it reduces to the case of interface refraction/
to the magnetic field opens a gap. In 3D, on the other reflection, which is described by the famous Snell’s
hand, the motion along the magnetic field usually law. The new finding is that once the light beam is
remains gapless; this means that the system cannot injected to the interface, the shift of the refracted
be the QH phase in most 3D systems. On the con- and reflected beams occurs perpendicular to the
trary, the QSHS is realized in the absence of the plane of incident light typically of the order of a
magnetic field; there is no fixed direction for the fraction of the wavelength. This phenomenon has
magnetic field, and the QSH phase can be realized
in 3D as well as in 2D. More deeply, the QSHS or the
topological insulator in 3D has no correspondence to Anomalous velocity
the QH system. k (z c | Ω kc|z c )
Small n
1.07.5.3 Generalization of SHE and Future Berry curvature
(z c | Ω kc|z c )
Directions
k
1.07.5.3.1 Optical SHE
The concept of the SHE can be generalized in many
directions. One example is the optical SHE, that is, Large n
the SHE of light (Fedorov, 1955; Pancharatnam,
1956; Imbert, 1972; Chiao and Wu, 1986; Tomita
Figure 34 Schematic of the optical SHE at refraction. The
and Chiao, 1986; Berry, 1987a,b; Onoda et al., 2004, polarization is left circular. The gradient of the refractive
2006a; Kavokin et al., 2005; Bliokh and Bliokh, 2006; index induces a transverse anomalous velocity due to Berry
Day, 2008a; Duval et al., 2006; Glazov and Kavokin, curvature.
Spin Hall Effect 267
been known as the Imbert shift (Fedorov, 1955; due to the spatial variation of the spin directions
Imbert, 1972) since long ago, whereas it is surprising (Meier and Loss, 2003). This is more promising
that it shares the similar mechanism with the SHE in when one wants to suppress the dissipation, com-
electrons. This transverse shift has been reconsid- pared with the metals and semiconductors. The
ered by Bliokh in the case of Gaussian beam (Bliokh design of the QSHE has been developed along this
and Bliokh, 2006). Because the semiclassical EOMs line as discussed in Section 1.07.1 but Mott insulat-
(74)–(76) assumes slow spatial variation of n, in a ing magnetic systems are more common and usual.
strict sense they cannot be applied for the abrupt In the presence of the SO interaction, one expects
interface; as a result, in the abrupt interface the shift the coupling between the spin current and the
depends on the detail of the beam, for example, electric polarization/electric field as a natural gen-
whether it is a Gaussian beam (Bliokh and Bliokh, eralization of SHE.
2006; Onoda et al., 2006a). Let us start with the commutation relationship
Recently, this transverse shift has been observed among the components of the spin operator, that is,
experimentally by using the weak measurement
(Hosten and Kwiat, 2008). With the setup shown in ½s a ; s b ¼ ih"abc s c ð77Þ
Figure 35 the transverse shift at the interface refrac- z
which can be translated into that of s and the angle
tion is measured to the accuracy of 1 nm, and the within the xy-plane (s+ = sx + isy ei) as
result shows good agreement with the theory, as
shown in Figure 36. The readers are referred to the ½; s z ¼ ih ð78Þ
original papers for more details. Optical SHE in exci-
ton-polariton system due to the precession of the spin This relation is analogous to the commutation rela-
state by the k-dependent Zeeman field has been dis- tion between the phase j and the particle number n
cussed and observed experimentally (Kavokin et al., of the bosons.
2005; Leyder et al., 2007). This mechanism is distinct In this language, the magnetic ordering can be
from that due to the Berry curvature discussed above. regarded as the superfluidity of the spin current,
Note that the SHE of excitons due to the Berry and once difference of the phases i and j occurs
curvature has been proposed by Yao and Niu (2008) between the neighboring sites i and j, the Josephson
and Kuga et al. (2008) recently. super spin current is induced, which then produces
the electric polarization. In this scenario, the electric
polarization P produced by the two spins si and sj is
1.07.5.3.2 Multiferroics given by (Katsura et al., 2005)
Another generalization of SHE is to the magnetic
P ¼ eij ðsi sj Þ ð79Þ
insulators. The charge transport is forbidden in the
insulator, but the spin current can be nonzero once with being a constant proportional to the SO inter-
the magnetism is there. For example, the spin cur- action. Here, the spin current is related to the vector
rent is associated with the spin wave propagation spin chirality si sj.
VAP zI
θw PS
xI P2 L2
y
θI
θT
HWP P1 VAP
L1
Figure 35 Experimental setup for the measurement of optical SHE. From Hosten O and Kwiat P (2008) Observation of the
spin Hall effect of light via weak measurements. Science 319: 787.
268 Spin Hall Effect
46
70
44 θI = 64°
42
60
40
38 δH
y-Displacement (nm)
50
36
34
40
0 20 40 60 80
γI(degrees)
30
δV
20
10
λ = 633 nm
0 20 40 60 80
θI(degrees)
Figure 36 Experimental results of optical SHE. From Hosten O and Kwiat P (2008) Observation of the spin Hall effect of light
via weak measurements. Science 319: 787.
According to this consideration, one expects the gradient, and the voltage drop are the external forces
ferroelectric state for the cycloidal helimagnets, to drive the heat, spin, and charge currents, respec-
that is, when the spin rotation plane includes the tively, and one can consider various kinds of off-
direction of the helical wavevector. Experimentally, diagonal responses among these. In particular, the
the multiferroic behavior of RMnO3 has been Hall responses have the topological aspects as
discovered earlier by Kimura et al. (2003) indepen- described above, which needs to be explored in the
dently of the theoretical works mentioned above, future. Orbital current, on the other hand, is rather
and the magnetic structure in the ferroelectric subtle since the information of the shape of the wave
phase was shown to be the incommensurate cycloid function is lost once the current is taken out of the
later (Kenzelmann et al., 2005; Arima et al., 2006) sample to the leads. However, one can develop var-
supporting the theoretical proposal. The origin of ious analogies between the spin current and orbital
the ferroelectricity in several other multiferroic current. Similar ideas have been discussed in the
materials is now shown to be the spin current physics of graphene recently, where the valley
mechanism (Katsura et al., 2005). In any case, the degrees of freedom is analogous to those of spin,
generalization of the spin Hall idea to the insulating and valleytronics might be an interesting subject
magnetic insulating materials offers an interesting (Rycerz et al., 2007).
route to the dissipationless spintronics. Another direction is to consider the electron–
electron and electron–phonon interactions in SHE.
1.07.5.3.3 Future directions As we discussed above, the d-electrons can be a
The future of the studies on SHE is briefly discussed promising candidate for future spintronics, where
here. One can imagine two directions for the future the electron correlation plays a crucial role. The
developments. One direction is to consider the var- enhanced SHE by local correlation by Fe impurity
ious kinds of current instead of the spin current such has been discussed in Section 1.07.3. SHE with 5d
as the heat current and orbital current. In particular, electrons was mentioned in Section 1.07.6.5. The
the heat current is an important issue also from the many-body effects combined with the nontrivial
viewpoint of applications, and the interplay between topology is a vast field of research as represented by
this current and the spin current is an interesting fractional QHE. Whether the analogous fractiona-
problem. Temperature gradient, magnetic field lized state exists or not for QSHE is an open
Spin Hall Effect 269
question, although the quenched kinetic energy by Awschalom DD and Flatte ME (2007) Challenges for
semiconductor spintronics. Nature Physics 3: 153.
Landau level formation is missing in the case of QSHE. Bardarson JH, Adagideli I, and Jacquod P (2007) Mesoscopic
From the viewpoint of the materials, it started spin Hall effect. Physical Review Letters 98: 196601.
with the semiconductors and now extends to metals, Bauer GEW (2004) Applied physics–mesmerizing
semiconductors. Science 306: 1898.
band insulators, Mott insulators, and even supercon- Beckmann U, Damker T, and Böttger H (2005) Spin transport in
ductors. Accordingly, the concept of the spin current disordered two-dimensional hopping systems with Rashba
and SHE is generalized. The condition is that the SO spin–orbit interaction. Physical Review B 71: 205311.
Bellucci S and Onorato P (2007a) Spin Hall accumulation in
interaction is effective, which generally requires ballistic nanojunctions. European Physical Journal B 59: 35.
heavy elements. It is very important to consider the Bellucci S and Onorato P (2007b) Spin Hall effect and spin–orbit
possible mechanisms to enhance the SO interaction. coupling in ballistic nanojunctions. Physical Review B
75: 235326.
An interesting proposal has been made (Huertas- Bellucci S and Onorato P (2008) Spin filtering and spin Hall
Hernando et al., 2006): the curvature in graphene, accumulation in an interferometric ballistic nanojunction
with Rashba spin–orbit interaction. Physical Review B
fullerenes, nanotubes, and nanotube caps induces 77: 075303.
the new term of SO interaction. This kind of theore- Bérard A and Mohrbach H (2006) Spin Hall effect and Berry
tical design will be useful to extend the horizon of phase of spinning particles. Physics Letters A 352: 190.
Bérard A and Mohrbach He (2004) Monopole and Berry phase
the SHE. Various other issues such as the role in momentum space in noncommutative quantum
of correlation (Hu et al., 2003b; Kou et al., 2005; mechanics. Physical Review D 69: 127701.
Dimitrova, 2004a), electron–phonon coupling Berger L (1970) Side-jump mechanism for the Hall effect of
ferromagnets. Physical Review B 2: 4559.
(Grimaldi et al., 2006b), the enhancement by bilayer Bernevig BA (2005) On the nature of spin currents in Rashba
effect (Jin and Li, 2007), and SHE in atoms (Liu et al., and Luttinger-type systems. Physical Review B 71: 073201.
2007c; Zhu et al., 2006) are left for future studies. Bernevig BA, Hu J, Mukamel E, and Zhang S-C (2004)
Dissipationless spin current in anisotropic p-doped
(See Chapters 1.01, 1.02, 1.05 and 1.06). semiconductors. Physical Review B 70: 113301.
Bernevig BA, Hughes TL, and Zhang S-C (2006) Quantum spin
Hall effect and topological phase transition in HgTe quantum
wells. Science 314: 1757.
Bernevig BA and Zhang S-C (2004) Intrinsic spin-Hall effect in
References n-doped bulk GaAs. cond-mat/0412550.
Bernevig BA and Zhang S-C (2005) Intrinsic spin Hall effect in the
Abanin DA, Lee PA, and Levitov LS (2006) Spin-filtered edge two-dimensional hole gas. Physical Review Letters 95: 016801.
states and quantum Hall effect in graphene. Physical Review Bernevig BA and Zhang S-C (2006) Quantum spin Hall effect.
Letters 96: 176803. Physical Review Letters 96: 106802.
Adagideli r and Bauer GEW (2005) Intrinsic spin Hall edges. Berry MV (1984) Quantal phase factors accompanying
Physical Review Letters 95: 256602. adiabatic changes. Proceedings of the Royal Society of
Adams EN and Blount EI (1959) Energy bands in the presence of London, Series A 392: 45.
an external force field–II: Anomalous velocities. Journal of Berry MV (1987a) The adiabatic phase and Pancharatnam’s
Physics and Chemistry of Solids 10: 286. phase for polarized light. Journal of Modern Optics 34: 1401.
Alderson JE and Hurd CM (1971) Spin component of Hall effect Berry MV (1987b) Interpreting the anholonomy of coiled light.
and transverse magnetoresistance in dilute alloys containing Nature 326: 277.
magnetic solutes. Journal of Physics and Chemistry of Solids Bleibaum O (2005) On the spin-Hall effect in a dirty Rashba
32: 2075. semiconductor cond-mat/0503471.
Ando K, Kajiwara Y, Takahashi S, et al. (2008a) Angular Bleibaum O (2006) Spin diffusion equations for systems with
dependence of inverse spin-Hall effect induced by spin Rashba spin–orbit interaction in an electric field. Physical
pumping investigated in a Ni81Fe19/Pt thin film. Physical Review B 73: 035322.
Review B 78: 014413. Bliokh KY (2005) Topological spin transport of a relativistic
Ando K, Takahashi S, Harii K, et al. (2008b) Electric electron. Europhysics Letters 72: 7.
manipulation of spin relaxation using the spin Hall effect. Bliokh KY and Bliokh YP (2005) Spin gauge fields: From Berry
Physical Review Letters 101: 036601. phase to topological spin transport and Hall effects. Annals
Ando T (1989) Numerical study of symmetry effects on of Physics 319: 13.
localization in two dimensions. Physical Review B 40: 5325. Bliokh KY and Bliokh YP (2006) Conservation of angular
Arii K, Koshino M, and Ando T (2007) Dynamical spin Hall momentum, transverse shift, and spin Hall effect in reflection
conductivity in a disordered two-dimensional system and refraction of an electromagnetic wave packet. Physical
calculated in a self-consistent Born approximation. Physical Review Letters 96: 073903.
Review B 76: 045311. Blount EI (1962) Formalisms of band theory. In: Seitz F and
Arima T, Tokunaga A, Goto T, Kimura H, Noda Y, and Tokura Y Turnbult D (eds.) Solid State Physics. p. 305. New York:
(2006) Collinear to spiral spin transformation without changing Academic Press.
the modulation wavelength upon ferroelectric transition in Bohm A, Mostafazadeh A, Koizumi H, Niu Q, and Zwanziger J
Tb1 x DyxMnO3. Physical Review Letters 96: 097202. (2003) The Geometric Phase in Quantum Systems. Berlin:
Aronov AG, Lyanda-Geller Y, and Pikus FG (1991) Spin Springer.
polarization of electrons by an electric current. Soviet Physics, Bratkovsky AM (2008) Spintronic effects in metallic,
Journal of Experimental and Theoretical Physics 73: 537. semiconductor, metal–oxide and metal–semiconductor
270 Spin Hall Effect
heterostructures. Reports on Progress in Physics Culcer D, Sinova J, Sinitsyn NA, Jungwirth T, MacDonald AH,
71: 026502. and Niu Q (2004) Semiclassical spin transport in spin–
Brusheim P and Xu HQ (2006) Spin Hall effect and Zitterbewegung orbit-coupled bands. Physical Review Letters 93: 046602.
in an electron waveguide. Physical Review B 74: 205307. Culcer D and Winkler R (2007a) Generation of spin currents and
Brusheim P and Xu HQ (2007) Spin transport and spin Hall spin densities in systems with reduced symmetry. Physical
effect in an electron waveguide in the presence of an in- Review Letters 99: 226601.
plane magnetic field and spin–orbit interaction. Physical Culcer D and Winkler R (2007b) Steady states of spin
Review B 75: 195333. distributions in the presence of spin–orbit interactions.
Bryksin VV and Kleinert P (2006) Theory of electric-field-induced Physical Review B 76: 245322.
spin accumulation and spin current in the two-dimensional Culcer D, Yao Y, MacDonald AH, and Niu Q (2005a) Electrical
Rashba model. Physical Review B 73: 165313. generation of spin in crystals with reduced symmetry.
Buchholtz LJ and Zwicknagl G (1981) Identification of p-wave Physical Review B 72: 045215.
superconductors. Physical Review B 23: 5788. Culcer D, Yao Y, and Niu Q (2005b) Coherent wave-packet
Bulgakov EN, Pichugin KN, Sadreev AF, Streda P, and Seba P evolution in coupled bands. Physical Review B
(1999) Hall-like effect induced by spin–orbit interaction. 72: 085110.
Physical Review Letters 83: 376. Dai X, Fang Z, Yao Y-G, and Zhang F-C (2006) Resonant
Bychkov YA and Rashba EI (1984) Oscillatory effects and the intrinsic spin Hall effect in p-type GaAs quantum well
magnetic susceptibility of carriers in inversion layers. Journal structure. Physical Review Letters 96: 086802.
of Physics C: Solid State Physics 17: 6039. Dai X, Hughes TL, Qi XL, Fang Z, and Zhang SC (2008) Helical
Chalaev O and Loss D (2005) Spin-Hall conductivity due to edge and surface states in HgTe quantum wells and bulk
Rashba spin–orbit interaction in disordered systems. insulators. Physical Review B 77: 125319.
Physical Review B 71: 245318. Dai X and Zhang FC (2007) Light-induced Hall effect in
Chang M-C (2005) Effect of in-plane magnetic field on the spin semiconductors with spin–orbit coupling. Physical Review B
Hall effect in a Rashba–Dresselhaus system. Physical 76: 085343.
Review B 71: 085315. Damker T, Böttger H, and Bryksin VV (2004) Spin Hall-effect in
Chang M-C and Niu Q (1995) Berry phase, hyperorbits, and the two-dimensional hopping systems. Physical Review B
Hofstadter spectrum. Physical Review Letters 75: 1348. 69: 205327.
Chang M-C and Niu Q (1996) Berry phase, hyperorbits, and the Datta S and Das B (1990) Electronic analog of the electro-optic
Hofstadter spectrum: Semiclassical dynamics in magnetic modulator. Applied Physics Letters 56: 665.
Bloch bands. Physical Review B 53: 7010. Day C (2005) Two groups observe the spin Hall effect in
Chazalviel JN (1975) Spin-dependent Hall-effect in semiconductors. Physics Today 58: 17.
semiconductors. Physical Review B 11: 3918. Day C (2008a) Light exhibits a spin Hall effect. Physics Today
Cheche TO (2006) Strain effect on the intrinsic spin Hall effect. 61: 18.
Physical Review B 73: 113301. Day C (2008b) Quantum spin Hall effect shows up in a quantum
Cheche TO and Barna E (2006) Control of the spin Hall current in well insulator, just as predicted. Physics Today 61: 19.
two dimensional electronic gas. Applied Physics Letters Dimitrova OV (2004a) Spin-Hall conductivity and Pauli
89: 042116. susceptibility in the presence of electron–electron
Chen SH, Liu MH, Chen KW, and Chang CR (2007) Broken spin- interactions. cond-mat/0407612.
Hall accumulation symmetry by magnetic field and coexisted Dimitrova OV (2004b) Vanishing spin-Hall conductivity in 2D
Rashba and Dresselhaus interactions. Journal of Applied disordered Rashba electron gas. cond-mat/0405339.
Physics 101: 09D–513. Dimitrova OV (2005) Spin-Hall conductivity in a two-
Chen T-W, Huang C-M, and Guo GY (2006) Conserved spin and dimensional Rashba electron gas. Physical Review B
orbital angular momentum Hall current in a two-dimensional 71: 245327.
electron system with Rashba and Dresselhaus spin–orbit Dresselhaus G (1955) Spin–orbit coupling effects in zincblende
coupling. Physical Review B 73: 235309. structures. Physical Review 100: 580.
Chen WQ, Weng ZY, and Sheng DN (2005a) Numerical study of Duval C, Horváth Z, and Horváthy PA (2006) Fermat principle for
spin Hall transport in a two-dimensional hole–gas system. spinning light. Physical Review D 74: 021701(R).
Physical Review B 72: 235315. D’yakonov MI and Perel’ VI (1971a) Current-induced spin
Chen WQ, Weng ZY, and Sheng DN (2005b) Numerical study of orientation of electrons in semiconductors. Physics Letters A
the spin-Hall conductance in the Luttinger model. Physical 35: 459.
Review Letters 95: 086605. D’yakonov MI and Perel’ VI (1971b) Possibility of orienting
Cheng JL and Wu MW (2008) Kinetic investigation of the electron spins with current. JETP Letters 13: 467. (ZhETF
extrinsic spin Hall effect induced by skew scattering. Journal P is. Red. 13,657 (1971)).
of Physics – Condensed Matter 20: 085209. Edelstein VM (1990) Spin polarization of conduction
Chiao RY and Wu Y-S (1986) Manifestations of Berry’s topological electrons induced by electric current in two-dimensional
phase for the photon. Physical Review Letters 57: 933. asymmetric electron systems. Solid State
Chudnovsky EM (2007) Theory of spin hall effect: Extension of Communications 73: 233.
the Drude model. Physical Review Letters 99: 206601. Engel H-A, Halperin BI, and Rashba EI (2005) Theory of spin Hall
Chun SH, Salamon MB, Lyanda-Geller Y, Goldbart PM, and conductivity in n-doped GaAs. Physical Review Letters
Han PD (2000) Magnetotransport in manganites and the role 95: 166605.
of quantal phases: Theory and experiment. Physical Review Engel H-A, Rashba EI, and Halperin BI (eds.) (2006) Theory of
Letters 84: 757. Spin Hall Effects. Handbook of Magnetism and Advanced
Costi TA, Bergqvist L, Weichselbaum A, et al. (2009) Kondo Magnetic Materials vol. 5. Newyork: Wiley.
decoherence: Finding the right spin model for iron Engel HA, Rashba EI, and Halperin BI (2007) Out-of-plane spin
impurities in gold and silver. Physical Review Letters polarization from in-plane electric and magnetic fields.
102: 056802. Physical Review Letters 98: 036602.
Crépieux A and Bruno P (2001) Theory of the anomalous Hall Entin-Wohlman O, Aharony A, Galperin YM, Kozub VI, and
effect from the Kubo formula and the Dirac equation. Vinokur V (2005) Orbital ac spin-Hall effect in the hopping
Physical Review B 64: 014416. regime. Physical Review Letters 95: 086603.
Spin Hall Effect 271
Erlingsson SI and Loss D (2005) Determining the spin Hall Grimaldi C, Cappelluti E, and Marsiglio F (2006b) Spin-Hall
conductance via charge transport. Physical Review B conductivity in electron–phonon coupled systems. Physical
72: 121310. Review Letters 97: 066601.
Erlingsson SI and Loss D (2006) Determining the spin Hall Grover T and Senthil T (2008) Topological spin Hall states,
conductance via charge current and noise. Physica E 34: 401. charged skyrmions, and superconductivity in two
Erlingsson SI, Schliemann J, and Loss D (2005) Spin dimensions. Physical Review Letters 100: 156804.
susceptibilities, spin densities, and their connection to spin Guo G-Y, Maekawa S, and Nagaosa N (2009) Enhanced spin Hall
currents. Physical Review B 71: 035319. effect by resonant skew scattering in the orbital-dependent
Essin AM and Moore JE (2007) Topological insulators beyond Kondo effect. Physical Review Letters 102: 036401.
the Brillouin zone via Chern parity. Physical Review B Guo GY, Murakami S, Chen TW, and Nagaosa N (2008) Intrinsic
76: 165307. spin Hall effect in platinum: First-principles calculations.
Fabian J, Matos-Abiague A, Ertler C, Stano P, and Žutić I Physical Review Letters 100: 096401.
(2007) Semiconductor spintronics. Acta Physica Slovaca Guo GY, Yao Y, and Niu Q (2005) Ab initio calculation of the
57: 565. intrinsic spin Hall effect in semiconductors. Physical Review
Fan J and Eom J (2008) Direct electrical observation of spin Hall Letters 94: 226601.
effect in Bi film. Applied Physics Letters 92: 142101. Haldane FDM (1988) Model for a quantum Hall effect without
Fang Z, Nagaosa N, Takahashi KS, et al. (2003) The anomalous Landau levels: Condensed-matter realization of the ‘‘parity
Hall effect and magnetic monopoles in momentum space. anomaly. Physical Review Letters 61: 2015.
Science 302: 92. Haldane FDM (2004) Berry curvature on the Fermi surface:
Fedorov FI (1955) K teorii polnogo otrazheniya. Dokl. Akad. Anomalous Hall effect as a topological Fermi-liquid property.
Nauk SSSR 105: 465. Physical Review Letters 93: 206602.
Fert A and Jaoul O (1972) Left–right asymmetry in the scattering Hankiewicz EM, Li J, Jungwirth T, Niu Q, Shen S-Q, and Sinova J
of electrons by magnetic impurities, and a Hall effect. (2005a) Charge Hall effect driven by spin-dependent chemical
Physical Review Letters 28: 303. potential gradients and Onsager relations in mesoscopic
Fertig HA (2003) Spinning holes in semiconductors. Science systems. Physical Review B 72: 155305.
301: 1335. Hankiewicz EM, Molenkamp LW, Jungwirth T, and Sinova J (2004)
Fu L and Kane CL (2006) Time reversal polarization and a Z2 Manifestation of the spin Hall effect through charge-transport in
adiabatic spin pump. Physical Review B 74: 195312. the mesoscopic regime. Physical Review B 70: 241301.
Fu L and Kane CL (2007) Topological insulators with inversion Hankiewicz EM, Sinitsyn NA, and Sinova J (2005b) Spin
symmetry. Physical Review B 76: 045302. currents and intrinsic spin-Hall effect in low dimensional
Fu L and Kane CL (2008) Superconducting proximity effect and systems. Journal of Superconductivity 18: 151.
majorana fermions at the surface of a topological insulator. Hankiewicz EM and Vignale G (2006) Coulomb corrections to
Physical Review Letters 100: 096407. the extrinsic spin-Hall effect of a two-dimensional electron
Fu L, Kane CL, and Mele EJ (2007a) Topological insulators in gas. Physical Review B 73: 115339.
three dimensions. Physical Review Letters 98: 106803. Hankiewicz EM and Vignale G (2008) Phase diagram of the spin
Fu YP, Wang D, Huang FJ, Li YD, and Liu WM (2007b) The Hall effect. Physical Review Letters 100: 026602.
contribution of spin torque to the spin Hall coefficient and Harii K, Ando K, Inoue HY, Sasage K, and Saitoh E (2008)
spin motive force in a spin–orbit coupling system. Journal of Inverse spin-Hall effect and spin pumping in metallic films
Physics – Condensed Matter 19: 496220. (invited). Journal of Applied Physics 103: 07F.
Fukui T and Hatsugai Y (2007a) Quantum spin Hall effect in Hatsugai Y, Fukui T, and Suzuki H (2006) Topological
three dimensional materials: Lattice computation of Z2 description of (spin) Hall conductances on Brillouin zone
topological invariants and its application to Bi and Sb. lattices: Quantum phase transitions and topological
Journal of the Physical Society of Japan 76: 053702. changes. Physica E 34: 336.
Fukui T and Hatsugai Y (2007b) Topological aspects of the Hattori K and Okamoto H (2006) Spin separation and spin Hall
quantum spin-Hall effect in graphene: Z2 topological order effect in quantum wires due to lateral-confinement-induced
and spin Chern number. Physical Review B 75: 121403. spin–orbit-coupling. Physical Review B 74: 155321.
Fukuyama H (2006) Interband effects of magnetic field on Hirahara T, Nagao T, Matsuda I, et al. (2006) Role of spin–orbit
orbital susceptibility and Hall conductivity – case of bismuth. coupling and hybridization effects in the electronic structure
Annalen Der Physik 15: 520. of ultrathin Bi films. Physical Review Letters 97: 146803.
Galitski VM, Burkov AA, and Das Sarma S (2006) Boundary Hirsch JE (1999) Spin Hall effect. Physical Review Letters 83: 1834.
conditions for spin diffusion in disordered systems. Physical Hosten O and Kwiat P (2008) Observation of the spin Hall effect
Review B 74: 115331. of light via weak measurements. Science 319: 787.
Ganichev SD, Bel’kov VV, Golub LE, et al. (2004) Experimental Hsieh D, Qian D, Wray L, et al. (2008) A topological Dirac
separation of Rashba and Dresselhaus spin splittings in insulator in a quantum spin Hall phase. Nature 452: 970.
semiconductor quantum wells. Physical Review Letters Hu JP, Bernevig BA, and Wu CJ (2003a) Spin current in spin–
92: 256601. orbit coupling systems. International Journal of Modern
Ganichev SD, Ivchenko EL, Bel’kov VV, et al. (2002) Spin- Physics B 17: 5991.
galvanic effect. Nature 417: 153. Hu L, Gao J, and Shen S-Q (2004) Effects of spin imbalance
Girvin SM and Prange RE (1987) The Quantum Hall Effect. on the electric-field-driven quantum dissipationless spin
Berlin: Springer-Verlag. current in p-doped semiconductors. Physical Review B
Glazov M and Kavokin A (2007) Spin Hall effect for electrons 70: 235323.
and excitons. Journal of Luminescence 125: 118. Hu L, Huang Z, and Hu S (2006) Spin Hall effect in a diffusive two-
Goryo J, Kohmoto M, and Wu YS (2008) Quantized spin Hall dimensional electron gas in the presence of both extrinsic and
effect in He-3-A and other p-wave paired Fermi systems. intrinsic spin–orbit coupling. Physical Review B 73: 235314.
Physical Review B 77: 144504. Hu LB, Gao J, and Shen SQ (2003b) Coulomb interaction in the
Grimaldi C, Cappelluti E, and Marsiglio F (2006a) Off-Fermi spin Hall effect. Physical Review B 68: 153303.
surface cancellation effects in spin-Hall conductivity of a Hu LB, Gao J, and Shen SQ (2003c) Influence of spin transfer
two-dimensional Rashba electron gas. Physical Review B and contact resistance on measurement of the spin Hall
73: 081303. effect. Physical Review B 68: 115302.
272 Spin Hall Effect
Hu LB and Huang Z (2006) Why does the intrinsic spin Hall spin-Hall effect in a spin–orbit coupled two-dimensional hole
effect in a Rashba two-dimensional electron gas disappear gas. Physica E 34: 47.
in the dc limit?. Physics Letters A 352: 250. Kane CL (2007) Graphene and the quantum spin Hall effect.
Huertas-Hernando D, Guinea F, and Brataas A (2006) Spin–orbit International Journal of Modern Physics B 21: 1155.
coupling in curved graphene, fullerenes, nanotubes, and Kane CL and Mele EJ (2005a) Quantum spin Hall effect in
nanotube caps. Physical Review B 74: 155426. graphene. Physical Review Letters 95: 226801.
Huh SG, Eom J, Koo HC, and Han SH (2008) Electrical detection Kane CL and Mele EJ (2005b) Z2 topological order and the
of the spin Hall effect in a two-dimensional electron gas. quantum spin Hall effect. Physical Review Letters 95: 146802.
Journal of the Korean Physical Society 52: 79. Kane CL and Mele EJ (2006) Physics: A new spin on the
Huh SG, Koo HC, Eom J, Yi H, Chang J, and Han SH (2007) insulating state. Science 314: 1692.
Unbalanced spin accumulation induced by spin Hall effect. Karplus R and Luttinger JM (1954) Hall effect in ferromagnetics.
Journal of Magnetism and Magnetic Materials 310: E705. Physical Review 95: 1154.
Hurd CM (1973) The Hall Effect in Metals and Alloys. New York: Kato YK and Awschalom DD (2008) Electrical manipulation of
Plenum. spins in nonmagnetic semiconductors. Journal of the
Imbert C (1972) Calculation and experimental proof of the Physical Society of Japan 77: 031006.
transverse shift induced by total internal reflection of a Kato YK, Myers RC, Gossard AC, and Awschalom DD (2004)
circularly polarized light beam. Physical Review D 5: 787. Observation of the spin Hall effect in semiconductors.
Imry Y (1997) Introduction to Mesoscopic Physics. Oxford: Science 306: 1910.
Oxford University Press. Katsura H, Nagaosa N, and Balatsky AV (2005) Spin current and
Imura KI and Shindou R (2005) Wave-packet dynamics of Bloch magnetoelectric effect in noncollinear magnets. Physical
electrons role of Berry phase. Physica E 29: 637. Review Letters 95: 057205.
Inoue HY, Harii K, Ando K, Sasage K, and Saitoh E (2007) Kavokin A, Malpuech G, and Glazov M (2005) Optical spin Hall
Detection of pure inverse spin-hall effect induced by spin effect. Physical Review Letters 95: 136601.
pumping at various excitation. Journal of Applied Physics Kent AD (2006) Quantum physics: New spin on the Hall effect.
102: 083915. Nature 442: 143.
Inoue J, Bauer GEW, and Molenkamp LW (2003) Diffuse Kenzelmann M, Harris AB, Jonas S, et al. (2005) Magnetic
transport and spin accumulation in a Rashba two- inversion symmetry breaking and ferroelectricity in TbMnO3.
dimensional electron gas. Physical Review B 67: 033104. Physical Review Letters 95: 087206.
Inoue J, Bauer GEW, and Molenkamp LW (2004) Suppression of Khaetskii A (2006) Nonexistence of intrinsic spin currents.
the persistent spin Hall current by defect scattering. Physical Physical Review Letters 96: 056602.
Review B 70: 041303. Kim BJ, Jin H, Moon SJ, et al. (2008) Novel jeff = 1/2 Mott state
Inoue J and Ohno H (2005) Taking the Hall effect for a spin. induced by relativistic spin–orbit coupling in Sr2IrO4.
Science 309: 2004. Physical Review Letters 101: 076402.
Inoue J-i, Kato T, Ishikawa Y, Itoh H, Bauer GEW, and Kimura T, Goto T, Shintani H, Ishizaka K, Arima T, and Tokura Y
Molenkamp LW (2006) Vertex corrections to the anomalous (2003) Magnetic control of ferroelectric polarization. Nature
Hall effect in spin-polarized two-dimensional electron gases 426: 55.
with a Rashba spin–orbit interaction. Physical Review Letters Kimura T, Otani Y, Sato T, Takahashi S, and Maekawa S (2007)
97: 046604. Room-temperature reversible spin Hall effect. Physical
Ivanov DA (2001) Non-abelian statistics of half-quantum Review Letters 98: 156601.
vortices in p-wave superconductors. Physical Review Kittel C (1987) Quantum Theory of Solids. New York: Wiley.
Letters 86: 268. Kleinert P and Bryksin VV (2006) Theory of spin-Hall transport of
Jiang YJ and Hu LB (2007) Symmetry properties of spin heavy holes in semiconductor quantum wells. Journal of
currents and spin polarizations in multiterminal Physics – Condensed Matter 18: 7497.
mesoscopic spin–orbit-coupled systems. Physical Review Kleinert P, Bryksin VV, and Bleibaum O (2005) Spin
B 75: 195343. accumulation in lateral semiconductor superlattices
Jiang ZF, Li RD, Zhang S-C, and Liu WM (2005) Semiclassical induced by a constant electric field. Physical Review B
time evolution of the holes from Luttinger Hamiltonian. 72: 195311.
Physical Review B 72: 045201. Knap W, Skierbiszewski C, Zduniak A, et al. (1996) Weak
Jin P-Q and Li Y-Q (2006) Generalized Kubo formula for spin antilocalization and spin precession in quantum wells.
transport: A theory of linear response to non-Abelian fields. Physical Review B 53: 3912.
Physical Review B 74: 085315. Kohmoto M (1985) Topological invariant and the quantization of
Jin PQ and Li YQ (2007) Magnification of the spin Hall effect in a the Hall conductance. Annals of Physics 160: 343.
bilayer electron gas. Physical Review B 76: 235311. König M, Buhmann H, Molenkamp LW, et al. (2008) The
Jin PQ and Li YQ (2008) Understanding spin transport from quantum spin Hall effect: Theory and experiment. Journal of
motion in SU(2) U(1) fields. Physical Review B the Physical Society of Japan 77: 031007.
77: 155304. König M, Wiedmann S, Brune C, et al. (2007) Quantum spin Hall
Jin P-Q, Li Y-Q, and Zhang F-C (2006) SU(2) U(1) unified insulator state in HgTe quantum wells. Science 318: 766.
theory for charge, orbit and spin currents. Journal of Physics Kontani H, Naito M, Hirashima DS, Yamada K, and Inoce JI
A: Mathematical and General 39: 7115. (2007) Study of intrinsic spin and orbital Hall effects in Pt
Johnson M (1993) Spin accumulation in gold films. Physical based on a (6s, 6p, 5d) tight-binding model. Journal of the
Review Letters 70: 2142. Physical Society of Japan 76: 103702.
Johnson M and Silsbee RH (1985) Interfacial charge-spin Kontani H, Tanaka T, Hirashima DS, Yamada K, and Inoue J
coupling: Injection and detection of spin magnetization in (2008) Giant intrinsic spin and orbital hall effects in
metals. Physical Review Letters 55: 1790. Sr2MO4 (M = Ru, Rh, Mo). Physical Review Letters
Jungwirth T, Niu Q, and MacDonald AH (2002) Anomalous Hall 100: 096601.
effect in ferromagnetic semiconductors. Physical Review Koroteev YM, Bihlmayer G, Chulkov EV, and Blugel S (2008)
Letters 88: 207208. First-principles investigation of structural and electronic
Kaestner B, Wunderlich J, Jungwirth T, Sinova J, Nomura K, properties of ultrathin Bi films. Physical Review B
and MacDonald AH (2006) Experimental observation of the 77: 045428.
Spin Hall Effect 273
Kou S-P, Qi X-L, and Weng Z-Y (2005) Spin Hall effect in a systems: A helicity-basis nonequilibrium Green’s function
doped Mott insulator. Physical Review B 72: 165114. approach. Physical Review B 73: 205327.
Kravchenko VY and Tsoi VS (2007) Spin hall effect in Liu SY, Lei XL, and Horing NJM (2006b) Vanishing spin-Hall
nonmagnetic conductors under the classical hall conditions. current in a diffusive Rashba two-dimensional electron
JETP Letters 86: 548. system: A quantum Boltzmann equation approach. Physical
Kravchenko VY and Tsoi VS (2008) Spin hall effect without spin– Review B 73: 035323.
orbit coupling in a homogeneous ferromagnet. Journal of Liu X-J, Liu X, Kwek LC, and Oh CH (2007c) Optically induced
Nanoelectronics and Optoelectronics 3: 86. spin-Hall effect in atoms. Physical Review Letters
Krotkov PL and Das Sarma S (2006) Intrinsic spin Hall 98: 026602.
conductivity in a generalized Rashba model. Physical Lou P and Xiang T (2005) Spin Hall current and two-dimensional
Review B 73: 195307. magnetic monopole in a Corbino disk. cond-mat/0501307.
Kuga S-i, Murakami S, and Nagaosa N (2008) Spin Hall effect of Lu JQ, Zhang XG, and Pantelides ST (2006) Tunable spin Hall
excitons. Physical Review B 78: 205201. effect by Stern–Gerlach diffraction. Physical Review B
Laughlin RB (1981) Quantized Hall conductivity in two 74: 245319.
dimensions. Physical Review B 23: 5632. Lucignano P, Raimondi R, and Tagliacozzo A (2008) Spin Hall
Lee PA, Nagaosa N, and Wen X-G (2006) Doping a Mott effect in a two-dimensional electron gas in the presence of a
insulator: Physics of high-temperature superconductivity. magnetic field. Physical Review B 78: 035336.
Reviews of Modern Physics 78: 17. Luttinger JM (1956) Quantum theory of cyclotron resonance
Lee PA and Ramakrishnan TV (1985) Disordered electronic in semiconductors: General theory. Physical Review
systems. Reviews of Modern Physics 57: 287. 102: 1030.
Lee SS and Ryu S (2008) Many-body generalization of the Z2 Ma JS, Koo HC, Chang J, et al. (2008) Spin Hall effect induced
topological invariant for the quantum spin Hall effect. by a Pd/CoFe multilayer in a semiconductor channel. Journal
Physical Review Letters 100: 186807. of the Korean Physical Society 53: 1357.
Leurs BWA, Nazario Z, Santiago DI, and Zaanen J (2008) Non- Ma T and Liu Q (2006a) Sign changes and resonance of intrinsic
abelian hydrodynamics and the flow of spin in spin–orbit spin Hall effect in two-dimensional hole gas. Applied Physics
coupled substances. Annals of Physics 323: 907. Letters 89: 112102.
Levitov LS, Nazarov YV, and Eliashberg GM (1985) Ma T and Liu Q (2006b) Spin Hall conductance of the two-
Magnetoelectric effects in conductors with mirror isomer dimensional hole gas in a perpendicular magnetic field.
symmetry. Zh. Eksp. Teor. Fiz. 88: 229. Physical Review B 73: 245315.
Leyder C, Romanelli M, Karr JP, et al. (2007) Observation of the Ma X, Hu L, Tao R, and Shen S-Q (2004) Influences of spin
optical spin Hall effect. Nature Physics 3: 628. accumulation on the intrinsic spin Hall effect in two-
Li J, Hu L, and Shen S-Q (2005) Spin-resolved Hall effect driven dimensional electron gases with Rashba spin–orbit coupling.
by spin–orbit coupling. Physical Review B 71: 241305. Physical Review B 70: 195343.
Li J and Shen SQ (2007) Spin-current-induced charge Mal’shukov AG and Chao KA (2005) Spin Hall conductivity of a
accumulation and electric current in semiconductor disordered two-dimensional electron gas with Dresselhaus
nanostructures with Rashba spin–orbit coupling. Physical spin–orbit interaction. Physical Review B 71: 121308.
Review B 76: 153302. Mal’shukov AG and Chu CS (2006) Spin cloud induced around
Li Y and Tao R (2007) Current in a spin–orbit-coupling system. an elastic scatterer by the intrinsic spin Hall effect. Physical
Physical Review B 75: 075319. Review Letters 97: 076601.
Lin Q, Liu SY, and Lei XL (2006) Spin-Hall effect in a diffusive Mal’shukov AG and Chu CS (2007) Spin orientation and spin-
two-dimensional electron system with spin–orbit coupling Hall effect induced by tunneling electrons. Physical Review B
under an in-plane magnetic field. Applied Physics Letters 76: 245326.
88: 122105. Mal’shukov AG, Tang CS, Chu CS, and Chao KA (2005a) Strain-
Lipparini E and Barranco M (2007) Zeeman and spin–orbit induced coupling of spin current to nanomechanical
effects on the spin-Hall conductance. Physica E 36: 190. oscillations. Physical Review Letters 95: 107203.
Liu CX, Hughes TL, Qi XL, Wang K, and Zhang SC (2008) Mal’shukov AG, Wang LY, and Chu CS (2007) Spin-Hall
Quantum spin Hall effect in inverted type-II semiconductors. interface resistance in terms of Landauer-type spin dipoles.
Physical Review Letters100. Physical Review B 75: 085315.
Liu MH, Bihlmayer G, Blügel S, and Chang CR (2007a) Intrinsic Mal’shukov AG, Wang LY, Chu CS, and Chao KA (2005b) Spin
spin-Hall accumulation in honeycomb lattices: Band Hall effect on edge magnetization and electric conductance
structure effects. Physical Review B 76: 121301. of a 2D semiconductor strip. Physical Review Letters
Liu SY, Horing NJM, and Lei XL (2006a) Long-range scattering 95: 146601.
effects on spin Hall current in p-type bulk semiconductors. Mathieu R, Asamitsu A, Yamada H, et al. (2004) Scaling of the
Physical Review B 73: 205207. anomalous Hall effect in Sr1 xCaxRuO3. Physical Review
Liu SY, Horing NJM, and Lei XL (2007b) Inverse spin Hall effect Letters 93: 016602.
by spin injection. Applied Physics Letters 91: 122508. Matl P, Ong NP, Yan YF, et al. (1998) Hall effect of the colossal
Liu SY and Lei XL (2004) Spin Hall effect in a diffusive Rashba magnetoresistance manganite La1 xCaxMnO3. Physical
two-dimensional electron gas. cond-mat/0411629. Review B 57: 10248.
Liu SY and Lei XL (2005a) Disorder effects on the spin-Hall Matsumoto M and Sigrist M (1999) Quasiparticle states near the
current in a diffusive Rashba two-dimensional heavy-hole surface and the domain wall in a px ipy-wave
system. Physical Review B 72: 155314. superconductor. Journal of the Physical Society of Japan
Liu SY and Lei XL (2005b) (erratum) disorder effects on the spin- 68: 994.
Hall current in a diffusive Rashba two-dimensional heavy- Maytorena JA, Lopez-Bastidas C, and Mireles F, (2006a)
hole system. Physical Review B 72: 249901(E). Dynamic spin and charge Hall conductivities in two-
Liu SY and Lei XL (2005c) Vanishing of the dissipationless spin dimensional electron gases with Rashba and Dresselhaus
Hall effect in a diffusive two-dimensional electron gas with coupling. cond-mat/0603722.
spin–orbit coupling. cond-mat/0502392. Maytorena JA, Lopez-Bastidas C, and Mireles F (2006b) Spin
Liu SY and Lei XL (2006) Spin Hall effect in infinitely large and and charge optical conductivities in spin–orbit coupled
finite-size diffusive Rashba two-dimensional electron systems. Physical Review B 74: 235313.
274 Spin Hall Effect
Meier F and Loss D (2003) Magnetization transport and Niemi AJ and Semenoff GW (1986) Fermion number
quantized spin conductance. Physical Review Letters fractionization in quantum field theory. Physics Reports
90: 167204. 135: 99.
Miah MD (2006) Injection and detection of spin current and spin Nikolić BK, Souma S, Za rbo LP, and Sinova J (2005a)
Hall effect in gaas. Materials Letters 60: 2863. Nonequilibrium spin Hall accumulation in ballistic
Min H, Hill JE, Sinitsyn NA, Sahu BR, Kleinman L, and semiconductor nanostructures. Physical Review Letters
MacDonald AH (2006) Intrinsic and Rashba spin–orbit 95: 046601.
interactions in graphene sheets. Physical Review B 74: 165310. Nikolić BK and Zârbo LP (2006) Extrinsically vs. intrinsically
Mishchenko EG, Shytov AV, and Halperin BI (2004) Spin current driven spin Hall effect in disordered mesoscopic
and polarization in impure two-dimensional electron systems multiterminal bars. Europhysics Letters 77: 47004.
with spin–orbit coupling. Physical Review Letters Nikolić BK, Zârbo LP, and Souma S (2005b) Mesoscopic spin
93: 226602. Hall effect in multiprobe ballistic spin–orbit-coupled
Miyasato T, Abe N, Fujii T, et al. (2007) Crossover behavior semiconductor bridges. Physical Review B 72: 075361.
of the anomalous Hall effect and anomalous Nernst effect Nikolić BK, Zârbo LP, and Souma S (2006) Imaging mesoscopic
in itinerant ferromagnets. Physical Review Letters spin Hall flow: Spatial distribution of local spin currents and
99: 086602. spin densities in and out of multiterminal spin–orbit coupled
Moca CP and Marinescu DC (2005) Longitudinal and spin Hall semiconductor nanostructures. Physical Review B
conductance of a one-dimensional Aharonov–Bohm ring. 73: 075303.
Journal of Physics – Condensed Matter 18: 127. Nikolić BK, Zârbo LP, and Welack S (2005c) Transverse spin–
Moca CP and Marinescu DC (2007a) Finite-size effects in a two- orbit force in the spin Hall effect in ballistic semiconductor
dimensional electron gas with Rashba spin–orbit interaction. wires. Physical Review B 72: 075335.
Physical Review B 75: 035325. Nitta J, Akazaki T, Takayanagi H, and Enoki T (1997) Gate
Moca CP and Marinescu DC (2007b) Spin-Hall conductivity of a control of spin–orbit interaction in an inverted
spin-polarized two-dimensional electron gas with Rashba In0.53Ga0.47As/In0.52Al0.48As heterostructure. Physical
spin–orbit interaction and magnetic impurities. New Journal Review Letters 78: 1335.
of Physics 9: 343. Nomura K, Sinova J, Jungwirth T, Niu Q, and MacDonald AH
Moca CP, Marinescu DC, and Filip S (2008) Spin Hall effect in a (2005a) Nonvanishing spin Hall currents in disordered spin–
symmetric quantum well by a random Rashba field. Physical orbit coupling systems. Physical Review B 71: 041304.
Review B 77: 193302. Nomura K, Sinova J, Jungwirth T, Niu Q, and MacDonald AH
Moore J (2008) Topological order – how spin splits the electron. (2006) Erratum: Nonvanishing spin Hall currents in
Nature Physics 4: 270. disordered spin–orbit coupling systems [Physical Reviews B
Moore JE and Balents L (2007) Topological invariants of time- 71, 041304(R) (2005)]. Physical Review B 73: 199901.
reversal-invariant band structures. Physical Review B Nomura K, Sinova J, Sinitsyn NA, and MacDonald AH (2005b)
75: 121306. Dependence of the intrinsic spin-Hall effect on spin–orbit
Murakami S (2004) Absence of vertex correction for the spin interaction character. Physical Review B 72: 165316.
Hall effect in p-type semiconductors. Physical Review B Nomura K, Wunderlich J, Sinova J, Kaestner B, MacDonald AH,
69: 241202. and Jungwirth T (2005c) Edge-spin accumulation in
Murakami S (2005) Intrinsic spin Hall effect. In: Kramer B (ed.) semiconductor two-dimensional hole gases. Physical
Advances in Solid State Physics. p. 197, vol. 45, Berlin: Review B 72: 245330.
Springer. Obuse H, Furusaki A, Ryu S, and Mudry C (2007) Two-
Murakami S (2006) Quantum spin Hall effect and enhanced dimensional spin-filtered chiral network model for the Z2
magnetic response by spin–orbit coupling. Physical Review quantum spin-Hall effect. Physical Review B 76: 075301.
Letters 97: 236805. Obuse H, Furusaki A, Ryu S, and Mudry C (2008a) Boundary
Murakami S (2007) Phase transition between the quantum spin criticality at the Anderson transition between a metal and a
Hall and insulator phases in 3D: Emergence of a topological quantum spin Hall insulator in two dimensions. Physical
gapless phase. New Journal of Physics 9: 356. Review B 78: 115301.
Murakami S, Iso S, Avishai Y, Onoda M, and Nagaosa N Obuse H, Subramaniam AR, Furusaki A, Gruzberg IA, and
(2007) Tuning phase transition between quantum spin Ludwig AWW (2008b) Boundary multifractality at the integer
Hall and ordinary insulating phases. Physical Review B quantum Hall plateau transition: Implications for the critical
76: 205304. theory. Physical Review Letters 101: 116802.
Murakami S and Kuga S-i (2008) Universal phase diagrams for Ohe J-i, Takeuchi A, Tatara G, and Kramer B (2008) Inverse spin
the quantum spin Hall systems. Physical Review B Hall effect in the Rashba spin–orbit system. Physica E
78: 165313. 40: 1554.
Murakami S and Nagaosa N (2003) Berry phase in magnetic Ohgushi K, Murakami S, and Nagaosa N (2000) Spin
superconductors. Physical Review Letters 90: 057002. anisotropy and quantum Hall effect in the kagome lattice:
Murakami S, Nagaosa N, and Zhang S-C (2003) Dissipationless Chiral spin state based on a ferromagnet. Physical
quantum spin current at room temperature. Science 301: 1348. Review B 62: R6065.
Murakami S, Nagaosa N, and Zhang S-C (2004a) Spin-Hall Onoda M, Avishai Y, and Nagaosa N (2007) Localization in a
insulator. Physical Review Letters 93: 156804. quantum spin Hall system. Physical Review Letters
Murakami S, Nagaosa N, and Zhang S-C (2004b) SU(2) non- 98: 076802.
Abelian holonomy and dissipationless spin current in Onoda M, Murakami S, and Nagaosa N (2004) Hall effect of
semiconductors. Physical Review B 69: 235206. light. Physical Review Letters 93: 083901.
Nagaosa N (2006) Anomalous Hall effect – a new perspective. Onoda M, Murakami S, and Nagaosa N (2006a) Geometrical
Journal of the Physical Society of Japan 75: 042001. aspects in optical wave-packet dynamics. Physical Review E
Nagaosa N (2007) A new state of quantum matter. Science 74: 066610.
318: 758. Onoda M and Nagaosa N (2002) Topological nature of
Nagaosa N (2008) Spin currents in semiconductors, metals, and anomalous Hall effect in ferromagnets. Journal of the
insulators. Journal of the Physical Society of Japan 77: 031010. Physical Society of Japan 71: 19.
Spin Hall Effect 275
Onoda M and Nagaosa N (2005a) Role of relaxation in the spin Ran Y, Zhang Y, and Vishwanath A, (2008b) Helical metal inside
Hall effect. Physical Review B 72: 081301. a topological band insulator. arXiv:0810.5121.
Onoda M and Nagaosa N (2005b) Spin current and Rashba EI (1960) Properties of semiconductors with an
accumulation generated by the spin Hall insulator. Physical extremum loop .1. Cyclotron and combinational resonance
Review Letters 95: 106601. in a magnetic field perpendicular to the plane of the loop.
Onoda S, Sugimoto N, and Nagaosa N (2006b) Intrinsic versus Soviet Physics: Solid State 2: 1109 (Fiz. Tverd. Tela
extrinsic anomalous Hall effect in ferromagnets. Physical (Leningrad) 2, 1224 (1960)).
Review Letters 97: 126602. Rashba EI (2004a) Spin currents, spin populations, and
Ostrovsky PM, Gornyi, IV, and Mirlin AD (2007) Quantum dielectric function of noncentrosymmetric semiconductors.
criticality and minimal conductivity in graphene with long- Physical Review B 70: 161201.
range disorder. Physical Review Letters 98: 256801. Rashba EI (2004b) Sum rules for spin Hall conductivity
Otani Y and Kimura T (2008) Spin current and spin Hall effect in cancellation. Physical Review B 70: 201309.
metallic nanostructures. IEEE Transactions on Magnetics Rashba EI (2005) Spin dynamics and spin transport. Journal of
44: 1911. Superconductivity 18: 137.
Pancharatnam S (1956) On the phenomenological theory of Read N and Green D (2000) Paired states of fermions in two
light propagation in optically active crystals. Proceedings of dimensions with breaking of parity and time-reversal
the Indian Academy of Sciences XLIV(5): 247. symmetries and the fractional quantum Hall effect. Physical
Pareek TP (2004) Pure spin currents and the associated Review B 61: 10267.
electrical voltage. Physical Review Letters 92: 076601. Rebei A and Heinonen O (2006) Spin currents in the Rashba
Pareek TP (2007) Measuring spin Hall conductance and Rashba model in the presence of nonuniform fields. Physical Review
spin–orbit interaction via electrical measurement in Y shape B 73: 153306.
conductor. International Journal of Modern Physics B 21: 4575. Ren W, Qiao ZH, Wang J, Sun QF, and Guo H (2006) Universal
Pershin YV and Di Ventra M (2008) A voltage probe of the spin spin-Hall conductance fluctuations in two dimensions.
Hall effect. Journal of Physics – Condensed Matter Physical Review Letters 97: 066603.
20: 025204. Reynoso A, Usaj G, and Balseiro CA (2006) Spin Hall effect in
Pfeffer P and Zawadzki W (1999) Spin splitting of conduction clean two-dimensional electron gases with Rashba spin–
subbands in III–V heterostructures due to inversion orbit coupling. Physical Review B 73: 115342.
asymmetry. Physical Review B 59: R5312. Roy R (2006a) On the Z2 classification of quantum spin Hall
Qi X-L, Hughes TL, Raghu S, and Zhang S-C, (2008a) Topological models. cond-mat/0604211.
superconductivity and superfluidity. arXiv:0803.3614. Roy R (2006b) Three dimensional topological invariants for time
Qi XL, Hughes TL, and Zhang SC (2008b) Fractional charge and reversal invariant Hamiltonians and the three dimensional
quantized current in the quantum spin Hall state. Nature quantum spin Hall effect. cond-mat/0607531.
Physics 4: 273. Roy R (2006c) Topological invariants of time reversal invariant
Qi X-L, Hughes TL, and Zhang S-C (2008c) Topological field superconductors. cond-mat/0608064.
theory of time-reversal invariant insulators. Physical Review Rycerz A, Tworzydlo J, and Beenakker CWJ (2007) Valley filter
B 78: 195424. and valley valve in graphene. Nature Physics 3: 172.
Qi X-L, Wu Y-S, and Zhang S-C (2006a) A general theorem Ryu S, Mudry C, Obuse H, and Furusaki A (2007) Z2 topological
relating the bulk topological number to edge states in two- term, the global anomaly, and the two-dimensional
dimensional insulators. Physical Review B 74: 045124. symplectic symmetry class of anderson localization.
Qi X-L, Wu Y-S, and Zhang S-C (2006b) Topological quantization Physical Review Letters 99: 116601.
of the spin Hall effect in two-dimensional paramagnetic Saitoh E, Ueda M, Miyajima H, and Tatara G (2006) Conversion of
semiconductors. Physical Review B 74: 085308. spin current into charge current at room temperature: Inverse
Qi X-L and Zhang S-C (2008) Spin-charge separation in the spin-Hall effect. Applied Physics Letters 88: 182509.
quantum spin Hall state. Physical Review Letters Schliemann J (2006) Spin Hall effect. International Journal of
101: 086802. Modern Physics B 20: 1015.
Qiao Z, Ren W, Wang J, and Guo H (2007) Low-field phase Schliemann J (2007) Ballistic side-jump motion of electrons and
diagram of the spin Hall effect in the mesoscopic regime. holes in semiconductor quantum wells. Physical Review B
Physical Review Letters 98: 196402. 75: 045304.
Qiao ZH, Wang J, Wei YD, and Guo H (2008) Universal Schliemann J and Loss D (2004) Dissipation effects in spin-Hall
quantized spin-Hall conductance fluctuation in graphene. transport of electrons and holes. Physical Review B
Physical Review Letters 101: 016804. 69: 165315.
Raghu S, Qi XL, Honerkamp C, and Zhang SC (2008) Topological Schliemann J and Loss D (2005) Spin-Hall transport of heavy
Mott insulators. Physical Review Letters 100: 156401. holes in III–V semiconductor quantum wells. Physical Review
Raichev OE (2007) Intrinsic spin-Hall effect: Topological B 71: 085308.
transitions in two-dimensional systems. Physical Review Schmidt G, Ferrand D, Molenkamp LW, Filip AT, and van
Letters 99: 236804. Wees BJ (2000) Fundamental obstacle for electrical spin
Raimondi R, Gorini C, Dzierzawa M, and Schwab P (2007) injection from a ferromagnetic metal into a diffusive
Current-induced spin polarization and the spin Hall effect: A semiconductor. Physical Review B 62: R4790.
quasiclassical approach. Solid State Communications Schnyder AP, Ryu S, Furusaki A, and Ludwig AWW (2008)
144: 524. Classification of topological insulators and superconductors
Raimondi R, Gorini C, Schwab P, and Dzierzawa M (2006) in three spatial dimensions. Physical Review B 78: 195125.
Quasiclassical approach to the spin-Hall effect in the two- Schulz M and Trimper S (2008) Thermally assisted spin Hall
dimensional electron gas. Physical Review B 74: 035340. effect. Physics Letters A 372: 5905.
Raimondi R and Schwab P (2005) Spin-Hall effect in a Seki T, Hasegawa Y, Mitani S, et al. (2008) Giant spin Hall effect
disordered two-dimensional electron system. Physical in perpendicularly spin-polarized FePt/Au devices. Nature
Review B 71: 033311. Materials 7: 125.
Ran Y, Vishwanath A, and Lee D-H (2008a) Spin-charge Sengupta K, Roy R, and Maiti M (2006) Spin Hall effect in triplet
separated solitons in a topological band insulator. Physical chiral superconductors and graphene. Physical Review B
Review Letters 101: 086801. 74: 094505.
276 Spin Hall Effect
Sensharma A and Mandal SS (2006) Spin–spin Hall effect in semiconductors with the spin Hall effect. Physical Review
two-dimensional electron systems with spin–orbit Letters 97: 096605.
interaction. Journal of Physics: Condensed Matter 18: 7349. Sih V, Myers RC, Kato YK, Lau WH, Gossard AC, and
Shapere A and Wilczek F (eds.) (1989) Geometric Phases in Awschalom DD (2005) Spatial imaging of the spin Hall effect
Physics. Singapore: World Scientific. and current-induced polarization in two-dimensional
Shchelushkin RV and Brataas A (2005a) Spin Hall effect, Hall electron gases. Nature Physics 1: 31.
effect, and spin precession in diffusive normal metals. Sinitsyn NA, Hankiewicz EM, Teizer W, and Sinova J (2004) Spin
Physical Review B 72: 073110. Hall and spin-diagonal conductivity in the presence of
Shchelushkin RV and Brataas A (2005b) Spin Hall effects in Rashba and Dresselhaus spin–orbit coupling. Physical
diffusive normal metals. Physical Review B 71: 045123. Review B 70: 081312.
Shekhter A, Khodas M, and Finkel’stein AM (2005) Chiral spin Sinitsyn NA, Hill JE, Min H, Sinova J, and MacDonald AH (2006)
resonance and spin-Hall conductivity in the presence of the Charge and spin Hall conductivity in metallic graphene.
electron–electron interactions. Physical Review B 71: 165329. Physical Review Letters 97: 106804.
Shen R, Chen Y, Wang ZD, and Xing DY (2006) Conservation of Sinova J, Culcer D, Niu Q, Sinitsyn NA, Jungwirth T, and
spin currents in spin–orbit-coupled systems. Physical MacDonald AH (2004) Universal intrinsic spin Hall effect.
Review B 74: 125313. Physical Review Letters 92: 126603.
Shen S-Q (2004) Spin Hall effect and Berry phase in two- Sinova J, Murakami S, Shen SQ, and Choi MS (2006) Spin-Hall
dimensional electron gas. Physical Review B 70: 081311. effect: Back to the beginning at a higher level. Solid State
Shen S-Q (2005) Spin transverse force on spin current in an Communications 138: 214.
electric field. Physical Review Letters 95: 187203. Smit J (1955) The spontaneous Hall effect in ferromagnetics-i.
Shen S-Q, Bao Y-J, Ma M, Xie XC, and Zhang FC (2005) Physica 21: 877.
Resonant spin Hall conductance in quantum Hall systems Smit J (1958) The spontaneous hall effect in ferromagnetics-ii.
lacking bulk and structural inversion symmetry. Physical Physica 24: 39.
Review B 71: 155316. Song F, Beckmann H, and Bergmann G (2006a) Geometrical
Shen S-Q, Ma M, Xie XC, and Zhang FC (2004) Resonant spin decay of the spin Hall effect measured in Fe(CsAu),
Hall conductance in two-dimensional electron systems with multilayers. Physical Review B 74: 035335.
a Rashba interaction in a perpendicular magnetic field. Song HZ, Zhang P, Duan SQ, and Zhao XG (2006b) Intrinsic Hall
Physical Review Letters 92: 256603. effect and separation of Rashba and Dresselhaus spin
Sheng DN, Sheng L, Weng ZY, and Haldane FDM (2005a) Spin splittings in semiconductor quantum wells. Chinese Physics
Hall effect and spin transfer in a disordered Rashba model. 15: 3019.
Physical Review B 72: 153307. Souma S and Nikolić BK (2005) Spin Hall current driven by
Sheng DN, Weng ZY, Sheng L, and Haldane FDM (2006) quantum interferences in mesoscopic Rashba rings.
Quantum spin Hall effect and topologically invariant Chern Physical Review Letters 94: 106602.
numbers. Physical Review Letters 97: 036808. Stern NP, Ghosh S, Xiang G, Zhu M, Samarth N, and
Sheng L, Sheng DN, and Ting CS (2005b) Spin-Hall effect in Awschalom DD (2006) Current-induced polarization and the
two-dimensional electron systems with Rashba spin–orbit spin Hall effect at room temperature. Physical Review Letters
coupling and disorder. Physical Review Letters 94: 016602. 97: 126603.
Sheng L, Sheng DN, Ting CS, and Haldane FDM (2005c) Stern NP, Steuerman DW, Mack S, Gossard AC, and
Nondissipative spin Hall effect via quantized edge transport. Awschalom DD (2007) Drift and diffusion of spins
Physical Review Letters 95: 136602. generated by the spin Hall effect. Applied Physics Letters
Sheng L and Ting CS (2006a) Asymptotic behavior of spin Hall 91: 062109.
conductance in mesoscopic two-dimensional electron Stern NP, Steuerman DW, Mack S, Gossard AC, and
systems. Physica E 33: 216. Awschalom DD (2008) Time-resolved dynamics of the spin
Sheng L and Ting CS (2006b) Intrinsic spin Hall effect in Hall effect. Nature Physics 4: 843.
mesoscopic systems. International Journal of Modern Středa P (2006) Normal and spin Hall effects as a response to
Physics B 20: 2339. the gradient of the chemical potential. Physical Review B
Sherman EY, Najmaie A, van Driel HM, Smirl AL, and Sipe JE 73: 113316.
(2006) Ultrafast extrinsic spin-Hall currents. Solid State Sugimoto N, Onoda S, Murakami S, and Nagaosa N (2006) Spin
Communications 139: 439. Hall effect of a conserved current: Conditions for a nonzero
Shi J, Zhang P, Xiao D, and Niu Q (2006) Proper definition of spin Hall current. Physical Review B 73: 113305.
spin current in spin–orbit coupled systems. Physical Review Sun Q-f, Guo H, and Wang J (2004) Spin-current-induced
Letters 96: 076604. electric field. Physical Review B 69: 054409.
Shindou R and Imura KI (2005) Noncommutative geometry and Sun Q-f, Wang J, and Guo H (2005) Quantum transport theory
non-abelian Berry phase in the wave-packet dynamics of for nanostructures with Rashba spin–orbital interaction.
Bloch electrons. Nuclear Physics B 720: 399. Physical Review B 71: 165310.
Shindou R and Murakami S (2009) Disorder effect on 3-dimensional Sun Q-f and Xie XC (2005) Definition of the spin current: The
Z2 quantum spin Hall systems. Physical Review B 79: 045321. angular spin current and its physical consequences. Physical
Shitade A, Katsura H, Qi X-L, Zhang S-C, and Nagaosa N, Review B 72: 245305.
(2009) Topological states in a transition metal oxide Na2IrO3. Sundaram G and Niu Q (1999) Wave-packet dynamics in slowly
Physical Review Letters 102: 256403. perturbed crystals: Gradient corrections and Berry-phase
Shytov AV, Mishchenko EG, Engel HA, and Halperin BI (2006) effects. Physical Review B 59: 14915.
Small-angle impurity scattering and the spin Hall Taguchi Y, Oohara Y, Yoshizawa H, Nagaosa N, and Tokura Y
conductivity in two-dimensional semiconductor systems. (2001) Spin chirality, Berry phase, and anomalous Hall effect
Physical Review B 73: 075316. in a frustrated ferromagnet. Science 291: 2573.
Sih V and Awschalom DD (2007) Electrical manipulation of spin– Takahashi S and Maekawa S (2003) Spin injection and detection
orbit coupling in semiconductor heterostructures. Journal of in magnetic nanostructures. Physical Review B 67: 052409.
Applied Physics 101: 081710. Takahashi S and Maekawa S (2007) Nonlocal spin Hall effect
Sih V, Lau WH, Myers RC, Horowitz VR, Gossard AC, and and spin–orbit interaction in nonmagnetic metals. Journal of
Awschalom DD (2006) Generating spin currents in Magnetism and Magnetic Materials 310: 2067.
Spin Hall Effect 277
Takahashi S and Maekawa S (2008a) Spin current in metals and Wang P and Li YQ (2008) The influence of inelastic relaxation
superconductors. Journal of the Physical Society of Japan time on intrinsic spin Hall effects in a disordered two-
77: 031009. dimensional electron gas. Journal of Physics – Condensed
Takahashi S and Maekawa S (2008b) Spin current, spin Matter 20: 215206.
accumulation and spin Hall effect. Science and Technology Wang P, Li Y-Q, and Zhao X (2007) Nonvanishing spin Hall
of Advanced Materials 9: 014105. currents in the presence of magnetic impurities. Physical
Tanaka T and Kontani H (2008) Giant extrinsic spin Hall effect Review B 75: 075326.
due to rare-earth impurities. New Journal of Physics Wang XD and Zhang XG (2005) Spin symmetry and spin
11: 013023. current of helicity eigenstates of the Luttinger
Tanaka T, Kontani H, Naito M, et al. (2008) Intrinsic spin Hall Hamiltonian. Journal of Magnetism and Magnetic
effect and orbital Hall effect in 4d and 5d transition metals. Materials 288: 297.
Physical Review B 77: 165117. Wang XF and Vasilopoulos P (2007) Intrinsic spin-Hall
Teo JCY, Fu L, and Kane CL (2008) Surface states and conductivity of a periodically modulated two-dimensional
topological invariants in three-dimensional topological electron gas in a perpendicular magnetic field: Interplay
insulators: Application to bi(sub 1–x)sb(sub x). Physical between the Zeeman term, spin–orbit interaction, and
Review B 78: 045426. modulation potential. Physical Review B 75: 075331.
Thouless DJ, Kohmoto M, Nightingale MP, and den Nijs M Wang ZG and Zhang P (2007) Conversed spin Hall conductance
(1982) Quantized Hall conductance in a two-dimensional in a two-dimensional electron gas in a perpendicular
periodic potential. Physical Review Letters 49: 405. magnetic field. Physical Review B 75: 233306.
Tomita A and Chiao RY (1986) Observation of Berry’s Wilczek F and Zee A (1984) Appearance of gauge structure in
topological phase by use of an optical fiber. Physical Review simple dynamical systems. Physical Review. Letters
Letters 57: 937. 52: 2111.
Tse W-K and Das Sarma S (2006) Spin Hall effect in doped Winkler R (2003) Spin–Orbit Coupling Effects in Two-
semiconductor structures. Physical Review Letters 96: 056601. Dimensional Electron and Hole Systems. Berlin: Springer.
Tse W-K, Fabian J, Ẑutić I, and Das Sarma S (2005) Spin Wolf SA, Awschalom DD, Buhrman RA, et al. (2001) Spintronics:
accumulation in the extrinsic spin Hall effect. Physical A spin-based electronics vision for the future. Science
Review B 72: 241303. 294: 1488.
Tse W-K and Sarma SD (2006) Intrinsic spin Hall effect in the Wong A, Maytorena JA, Lopez-Bastidas C, and Mireles F (2008)
presence of extrinsic spin–orbit scattering. Physical Review Spin-torque contribution to the AC spin Hall conductivity.
B 74: 245309. Physical Review B 77: 035304.
Tserkovnyak Y, Halperin BI, Kovalev AA, and Brataas A (2007) Wu C, Bernevig BA, and Zhang S-C (2006) Helical liquid and the
Boundary spin Hall effect in a two-dimensional edge of quantum spin Hall systems. Physical Review Letters
semiconductor system with Rashba spin–orbit coupling. 96: 106401.
Physical Review B 76: 085319. Wu MW and Zhou J (2005) Spin-Hall effect in two-dimensional
Uchida K, Takahashi S, Harii K, et al. (2008) Observation of the mesoscopic hole systems. Physical Review B 72: 115333.
spin Seebeck effect. Nature 455: 778. Wunderlich J, Kaestner B, Sinova J, and Jungwirth T (2005)
Valenzuela SO and Tinkham M (2006) Direct electronic Experimental observation of the spin-Hall effect in a two-
measurement of the spin Hall effect. Nature 442: 176. dimensional spin–orbit coupled semiconductor system.
Valenzuela SO and Tinkham M (2007) Electrical detection of Physical Review Letters 94: 047204.
spin currents: The spin-current induced Hall effect (invited). Xia Y, Wray L, Qian D, et al. (2008) Electrons on the surface of
Journal of Applied Physics 101: 09B103. bi2se3 form a topologically ordered two dimensional gas with
Vila L, Kimura T, and Otani Y (2007) Evolution of the spin hall a non-trivial berry’s phase. arXiv:0812.2078v1.
effect in Pt nanowires: Size and temperature effects. Xing Y, Sun Q-f, Tang L, and Hu J (2006a) Accumulation of
Physical Review Letters 99: 226604. opposite spins on the transverse edges of a two-dimensional
Vurgaftman I, Meyer JR, and Ram-Mohan LR (2001) Band electron gas in a longitudinal electric field. Physical Review B
parameters for III–V compound semiconductors and their 74: 155313.
alloys. Journal of Applied Physics 89: 5815. Xing Y, Sun Q-f, and Wang J (2006b) Nature of spin Hall effect in
Wang CM, Lei XL, and Liu SY (2006a) Nonvanishing spin Hall a finite ballistic two-dimensional system with Rashba and
conductivity in two-dimensional Rashba electron systems Dresselhaus spin–orbit interaction. Physical Review B
with nonparabolic Kane dispersion. Physical Review B 73: 205339.
73: 113314. Xing Y, Sun Q-f, and Wang J (2007) Symmetry and transport
Wang CM, Lin SY, and Lei XL (2006b) Rashba spin–orbit- property of spin current induced spin-Hall effect. Physical
coupled two-dimensional electron gas under an intense Review B 75: 075324.
terahertz field: Theoretical calculations. Physical Review B Xing YX, Sun QF, and Wang J (2008) Influence of dephasing on
73: 035333. the quantum Hall effect and the spin Hall effect. Physical
Wang J and Chan KS (2005) Numerical study of the spin Hall Review B 77: 115346.
current in a quantum wire. Physical Review B 72: 045331. Xiong JW, Hu LB, and Zhang ZX (2006) Suppression of direct
Wang J, Wang B, Ren W, and Guo H (2006c) Conservation of spin Hall currents in two-dimensional electronic systems
spin current: Model including self-consistent spin–spin with both Rashba and Dresselhaus spin–orbit couplings.
interaction. Physical Review B 74: 155307. Chinese Physics Letters 23: 1278.
Wang J, Zhu BF, and Liu RB (2008) Proposal for direct Xu C and Moore JE (2006) Stability of the quantum spin Hall
measurement of a pure spin current by a polarized light effect: Effects of interactions, disorder, and Z2 topology.
beam. Physical Review Letters 100: 086603. Physical Review B 73: 045322.
Wang JW and Li SS (2007) Spin Hall effect of excitons with Yang M-F and Chang M-C (2006) Streda-like formula in the spin
spin–orbit coupling. Applied Physics Letters 91: 052104. Hall effect. Physical Review B 73: 073304.
Wang LY, Tang CS, and Chu CS (2006d) DC spin current Yang W, Chang K, and Zhang SC (2008) Intrinsic spin Hall effect
generation in a Rashba-type quantum channel. Physical induced by quantum phase transition in HgCdTe quantum
Review B 73: 085304. wells. Physical Review Letters 100: 056602.
278 Spin Hall Effect
Yao W and Niu Q (2008) Berry phase effect on the exciton Zhang S and Yang Z (2005) Intrinsic spin and orbital
transport and on the exciton Bose-Einstein condensate. angular momentum Hall effect. Physical Review Letters
Physical Review Letters 101: 106401. 94: 066602.
Yao Y and Fang Z (2005) Sign changes of intrinsic spin Hall Zhang ZY (2006) Spin accumulation on a one-dimensional
effect in semiconductors and simple metals: First-principles mesoscopic Rashba ring. Journal of Physics – Condensed
calculations. Physical Review Letters 95: 156601. Matter 18: 4101.
Yao Y, Kleinman L, MacDonald AH, et al. (2004) First-principles Zhou B, Liu CX, and Shen SQ (2007) Topological quantum phase
calculation of anomalous Hall conductivity in ferromagnetic transition and the Berry phase near the Fermi surface in hole-
bcc Fe. Physical Review Letters 92: 037204. doped quantum wells. Europhysics Letters 79: 47010.
Yao Y, Liang Y, Xiao D, et al. (2007) Theoretical evidence of the Zhou L, Ma ZS, and Zhang C (2008) Temperature dependence
berry-phase mechanism in anomalous Hall transport: First- of the intrinsic spin Hall effect in Rashba spin–orbit coupled
principles studies of CuCr2Se4 xBrx. Physical Review B systems. Europhysics Letters 82: 67003.
75: 020401. Zhu S-L, Fu H, Wu CJ, Zhang SC, and Duan LM (2006) Spin Hall
Ye J, Kim YB, Millis AJ, et al. (1999) Berry phase theory of the effects for cold atoms in a light-induced gauge potential.
anomalous Hall effect: Application to colossal Physical Review Letters 97: 240401.
magnetoresistance manganites. Physical Review Letters Žutić I, Fabian J, and Das Sarma S (2004) Spintronics:
83: 3737. Fundamentals and applications. Reviews of Modern Physics
Zarea M and Ulloa SE (2006) Spin Hall effect in two-dimensional 76: 323.
p-type semiconductors in a magnetic field. Physical Review Zyuzin VA, Silvestrov PG, and Mishchenko EG (2007) Spin Hall
B 73: 165306. edge spin polarization in a ballistic 2D electron system.
Zeng C, Yao Y, Niu Q, and Weitering HH (2006) Linear Physical Review Letters 99: 106601.
magnetization dependence of the intrinsic anomalous Hall
effect. Physical Review Letters 96: 037204.
Zhang FC and Shen SQ (2008) Theory of resonant spin
Hall effect. International Journal of Modern Physics B Further Reading
22: 94.
Zhang H, Liu C-X, Qi X-L, Dai X, Fang Z, and Zhang S-C, (2009). Engel H-A, Rashba EI, and Halperin BI (eds.) (2006) Theory of
Topological insulators at room temperature. Nature Physics Spin Hall Effects. Handbook of Magnetism and Advanced
5: 438. Magnetic Materials, Vol. 5, Wiley.
Zhang L, Liu FQ, and Liu C (2007) The influence of external König M, Buhmann H, Molenkamp LW, et al. (2008)
electromagnetic field on an exciton spin current. Journal of The quantum spin Hall effect: Theory and
Physics – Condensed Matter 19: 346222. experiment. Journal of the Physical Society of Japan
Zhang P and Niu Q (2004). Charge-Hall effect driven by spin 77: 031007.
force: Reciprocal of the spin-Hall effect. cond-mat/ Murakami S (2005) Intrinsic Spin Hall Effect. Vol. 45 of Advances
0406436. in Solid State Physics. Berlin: Springer.
Zhang P, Wang ZG, Shi JR, Xiao D, and Niu Q (2008b) Theory of Nagaosa N (2008) Spin currents in semiconductors, metals, and
conserved spin current and its application to a two- insulators. Journal of the Physical Society of Japan
dimensional hole gas. Physical Review B 77: 075304. 77: 031010.
Zhang S (2000) Spin Hall effect in the presence of spin diffusion. Schliemann J (2006) Spin Hall effect. International Journal of
Physical Review Letters 85: 393. Modern Physics B 20: 1015.
1.08 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
W R Clarke and M Y Simmons, University of New South Wales, Sydney, NSW, Australia
C-T Liang, National Taiwan University, Taipei, China
ª 2011 Elsevier B.V. All rights reserved.
279
280 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
Glossary
Band-gap engineering The process of controlling Four-terminal configuration A measurement
or altering the band gap of a material by controlling configuration for measuring the resistance of a
the growth and composition of layers of different device in which the current at the source and drain
semiconductors, such as GaAs, AlGaAs, InGaAs, and the voltage at two points along the device are
and InAlAs. measured simultaneously. With this information, it
Chemical potential The chemical potential of a is possible to determine the resistance of the
thermodynamic system is the amount by which device between the two voltage probes and
the energy of the system would change if an eliminate contact resistance.
additional particle was introduced, with the g-Factor A dimensionless quantity which
entropy and volume held fixed. If the chemical characterizes the magnetic moment and
potential of a source lead of a quantum wire is gyromagnetic ratio of a particle. The g-factor
greater than the chemical potential of the drain determines how strongly the particle will couple to
lead, then the electrons will flow from source to an external magnetic field.
drain. g-Factor anisotropy The extent to which the
Density of states The number of states at each measured g-factor of carriers changes as the
energy level of a system. direction of the magnetic field changes.
Drude conductivity The conductivity of an Group velocity The velocity of the envelope of a
electronic system at zero temperature predicted by group of interfering waves having slightly different
the classical Drude model. frequencies and phase velocities.
Effective mass A value of the mass assigned to an Heterostructure A single crystal made up of
electron traveling in a periodic lattice that allows the layers of different crystalline materials such as
electron to be treated as a particle traveling in free GaAs, AlGaAs, and AlAs.
space. Inelastic scattering A form of scattering event in
Einstein’s relation Also known as Einstein– which the energy of the incident particles is not
Smoluchowski relation, it states that the mobility of conserved.
charges in a semiconductor is proportional to a Low-dimensional system Systems of electrons
diffusion coefficient D and inversely proportional to or holes that are confined to approximately one
the product of the Boltzmann constant and the Fermi wavelength in one or more directions.
absolute temperature. Many-body phenomena Phenomena that can
Elastic scattering A form of scattering event in only be described by including the interactions
which the energy of the incident particles is between particles and cannot be described by a
conserved, only their direction of propagation is single-particle model.
modified. Mean free path The average distance an electron
Fermi energy The energy in a solid below which all travels before being scattered.
energy states are occupied at zero temperature Mesoscopic Pertaining to a size regime,
and above which all states are unoccupied at zero intermediate between the microscopic and the
temperature. macroscopic, that is characteristic of a region
Fermi velocity The velocity of particles having where a large number of particles can interact in a
kinetic energy equal to the Fermi energy. quantum-mechanically correlated fashion.
Fermi wavelength The de Broglie wavelength of a Mobility The average drift velocity of carriers per
quantum wave packet with kinetic energy equal to unit electric field in a homogeneous
the Fermi energy. semiconductor. Mobility is determined by the
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 281
number of scattering events and therefore is often Weak localization A quantum mechanical
quoted as a measure of purity of the system. property of 2D systems characterized by an
Modulation doping A technique used in molecular enhanced resistance at zero magnetic field that
beam epitaxy in which successive heterostructure rapidly falls away with the application of a
layers are grown with different types and amounts perpendicular field. The enhanced resistance at
of dopants. zero field arises due to time-reversal symmetry.
Momentum relaxation time The characteristic For each particle with a probability to travel a
time for the momentum distribution of electrons in a closed loop, there is an equal probability that the
solid to approach or relax to equilibrium after an particle will travel the same closed loop in the
external influence is removed. opposite direction. These two probability
Phonon The quantum of acoustic or vibrational pathways constructively interfere, increasing the
energy, considered a discrete particle and used probability that the particle will be trapped in this
especially in models to calculate thermal and closed loop and unable to take part in conduction.
vibrational properties of solids. The size of these localizing pathways is dependent
Quantum point contact A quasi-1D system in on the phase coherence length and therefore
which the length of the channel approaches zero. weak localization is most apparent at low
Scattering time The time between scattering temperature where the phase coherence is
events of an electron. enhanced. Applying a magnetic field breaks the
Spin–orbit coupling In an atom, it is the interaction time-reversal symmetry and thereby causes a
between a particle’s spin and its orbital angular decrease in the resistance.
momentum. In a semiconductor, the spin–orbit Zeeman splitting The breaking of spin
coupling refers to the coupling of the spins to the degeneracy by the application of an external
orbital angular momentum of the bands in which magnetic field.
they travel. The orbital angular momentum of the Zero-bias anomaly In 1D systems, the zero-bias
band derives from the atomic orbitals from which it is anomaly refers to an enhanced conductance at
formed. In GaAs, the conduction band derives from zero source–drain bias that falls away rapidly as the
s-orbitals and therefore has l ¼ 0. The valence band source–drain bias is increased. The effect draws its
derives from p-orbitals and therefore l ¼ 1. name from an analogous effect seen in quantum-
Transconductance The derivative of the current dots due to the Kondo effect in 0D.
through a device as a function of source–drain voltage.
conventional diffusive transport, where the electrons In Section 1.08.3, we mirror the discussion on electron
are scattered many times by impurities as they make systems by considering ballistic transport in 1D hole
their way through the device. Interestingly, funda- systems. In particular, we highlight the important role
mental research in these high-purity GaAs-based that spin–orbit coupling and the large effective
devices has been incorporated back into the semi- mass have on both the single-particle properties and
conductor industry, which now uses them in high- many-body effects in ballistic hole channels.
speed applications such as monolithic microwave
integrated circuits (MMICs).
In this chapter, we study the unique properties of 1.08.1.1 Important Length Scales and
ballistic transport in GaAs devices. In particular, we Mesoscopic Systems
examine the intriguing effects that appear in ballistic,
1D quantum wires. The truly quantum nature of As semiconductor devices scale down in size, there
ballistic 1D wires was first demonstrated by two reaches a limit where device behavior no longer
groups independently. In 1988, van Wees et al. scales with dimension but novel effects start to
(1988) and Wharam et al. (1988) both showed that occur. These effects come into play when the dimen-
the conductance of a ballistic 1D channel was quan- sions of the device become comparable to an
tized in units of G ¼ 2e2/h. This seminal discovery important length scale, such as the mean free path
could be explained by a simple single-particle model. or the Fermi wavelength. At this point, the system
However, subsequent experiments revealed the enters what is called the mesoscopic regime, where
appearance of new features with no single-particle the wave character of the charge particle comes into
analog. Most notable has been the discovery of a play and the kinetic energy becomes quantized. This
small conductance plateau at the oddly unquantized quantization comes about purely from the spatial
value of G ¼ 0.7(2e2/h). This so-called 0.7 structure confinement of the charge carriers. This confinement
reveals a plethora of many-body effects, the origin of can occur in just one spatial dimension to form a 2D
which continues to be a source of contention in the film or in two spatial dimensions to form a 1D wire or
field of mesoscopic physics. in all three spatial dimensions to form a quantum-dot
The chapter will proceed with a brief introduction or an artificial atom. In each case, the dimensionality
to transport in mesoscopic systems with important of charge motion is the number of spatial directions
length scales in Section 1.08.1.1. Since all 1D devices in which the electron eigenstates are free to evolve
are derived from 2D systems, we will present a brief and thereby to transport charge. Understanding
introduction to 2D systems in Section 1.08.1.2. Then, mesoscopic systems requires an understanding of
in Section 1.08.1.3, we introduce the fundamental con- several key length scales. These are summarized as
cepts of conduction in 1D, which includes an follows.
introduction to the density of states and to the
Landauer formula for transport in 1D systems, quan- 1.08.1.1.1 Important length scales
tum point contacts, and the saddle point potential. In The three most important length scales in meso-
Section 1.08.2, we provide a historical account of the scopic systems are: the Fermi wavelength, the mean
discovery of conductance quantization in n-type GaAs free path, and the phase coherence length.
heterostructures. This includes an introduction to the Fermi wavelength, F: The Fermi wavelength for 2D
split-gate technique (Section 1.08.2.2) and several other systems, F, is the de Broglie wavelength of electrons
methods to fabricate 1D systems (Section 1.08.2.3), at the Fermi energy and is given by
including shallow etching techniques, cleaved-edge rffiffiffiffiffi
2 2 h
overgrowth, V-groove quantum wires, and gated, F ¼ ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi
kF ns 2m EF
undoped heterostructures. We then review the out-
comes for the quantized conductance plateau in both where kF is the Fermi wave number, ns is the sheet
magnetic field and as a function of source–drain bias in carrier concentration, m is the effective mass, and EF
Section 1.08.2.4. This allows us to measure energy-level is the Fermi energy. At low temperatures, current in
subband spacings and the g-factor in 1D systems. In devices is passed by carriers having an energy close
Section 1.08.2.5, we review the appearance of a feature to the Fermi energy, so that the Fermi wavelength
at 0.7(2e2/h) seen in high-quality electron systems, becomes the relevant wavelength. Carriers with less
before finishing with a discussion of the Kondo energy have longer wavelengths, but they do not
model and zero-bias anomaly (Section 1.08.2.6). contribute to the conduction. The lowest energy
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 283
mode in a quantum wire is half the Fermi wave- Phase coherence length, l : When electrons are con-
length; therefore, a wire must be at least this wide fined spatially, we need to consider the electron as a
in order to see conductance quantization. In GaAs/ wave packet with a phase coherence length. Electron
AlGaAs heterostructures F 400 Å for an electron motion can no longer be described by a single
concentration of ns 3 1011 cm2. For 1D systems Schrödinger equation but can interact with many
kF is given by ns. other degrees of freedom and exchange their energy.
Mean free path, lmfp: A simple explanation of the The phase-coherence length is essentially the dis-
mean free path is the average distance a carrier tance the charge carrier can travel before its phase
travels before it gets scattered into a different wave- becomes uncorrelated with its original value. The
vector direction. If we consider an electron moving phase-coherence length can only be destroyed by
through a perfect crystalline lattice, the electron inelastic scattering events – ones that cause phase
moves as if it were in vacuum, but with a different randomizing collisions. Typically, this is caused by
mass. However, if there is any deviation from perfect fluctuating scatterers. Some examples of this include
crystallinity either caused by defects or impurities in electron–electron interactions, where the mutual
the lattice or, from thermal vibrations, then the elec- Coulomb repulsion of electrons causes them to con-
tron’s path is scattered and it changes its momentum. tinually move and act as scattering sites. Phonons or
The mean free path lmfp is given by lattice vibrations also cause dephasing, but these
typically reduce considerably at low temperatures.
lmfp ¼ vF Indeed, both electron–electron and electron–phonon
where vF is the Fermi velocity and is the momen- scattering decrease at low temperatures. Impurity
tum relaxation time. The Fermi velocity is given by scattering can also be phase-randomizing if the
impurity has an internal degree of freedom that can
hkF fluctuate with time, such as magnetic impurities
VF ¼
m where the spin fluctuates with time. The weak
The mean free path dictates the length a quantum localization correction is only significant when
wire must be before ballistic transport is observed. l > lmfp.
Ideally, the length of the wire should be less than the
mean free path. Typically, for high-mobility, modu- 1.08.1.1.2 Mesoscopic transport
lation-doped GaAs/AlGaAs heterostructures, with Transport in mesoscopic systems can be divided into
sheet carrier densities of 3 1011 cm2, and mobi- three distinct regimes depending on the relative
lities of 5 106 cm2 V1 s1, the mean free path can magnitude of the size of the system, L, mean free
be as long as 100 mm at low temperatures. At room path, lmfp, and the phase coherence length, l, (see
temperature, this value can decrease considerably Table 1). In the diffusive regime, the sample size is
due to the large amount of electron–phonon scatter- much larger than the mean free path and the trans-
ing. The consequence of such long mean free paths at port is essentially independent of the form of the
low temperatures is that it is easy to create devices system. In the quantum coherent regime, the sample
where the device size has comparable dimensions to size is similar to the mean free path; therefore classi-
the mean free path and the system becomes ballistic. cal transport begins to break down as edge and
Table 1 Trends in g-factor with increasing ID confinement for quantum wires oriented along different crystallographic axes
in parallel and perpendicular magnetic fields.
Important
Regime length scales Impact on electrical conduction
Diffusive regime l < L Quantum corrections to the conductivity can exist due to phase coherence within
regions of size lø
Quantum lmfp < L < l Classical theories of conduction breakdown and the device becomes a more
coherent complex quantum system
regime
Ballistic regime L< lmfp Electrons move through the device like billiard balls on a table and the system
behaves in a more simple purely quantum mechanical way.
boundary effects start to dominate. However, in the constant a is very small, with a ¼ 5.6533 þ 0.0078x
ballistic regime, the electron makes a ballistic motion (Adachi, 1985), minimizing strain and scattering at
in the system and the system boundary plays the role GaAs/AlGaAs interfaces. This is in stark contrast
of the scatterer instead of impurities. Metallic sys- to the crystalline-amorphous Si–SiO2 interfaces in
tems are typically diffusive since their mean free Si-MOSFETs.
paths are of the order of 100 Å and Fermi wave- Confining the carriers to a 2D plane is made
lengths of 1–2 Å, such that quantization of the possible through band-gap engineering. While the
electron energy levels is not so important. lattice spacing in GaAs (5.653 Å) is very similar to
However, for semiconductor systems where mean AlAs (5.660 Å), the band gap between them is very
free paths can be 50 mm long and the Fermi wave- different. If we replace a fraction, x, of the Ga atoms
lengths are large, 500 Å, both the boundaries and with Al atoms in AlxGa1xAs, we can span a whole
quantization of the system play important roles. range of band gaps from Eg ¼ 1.424 þ 1.247x eV at
Before introducing conduction in ballistic 1D sys- room temperature. With the use of techniques such
tems, it is important to briefly consider the 2D as molecular beam epitaxy (MBE), it is possible to
systems from which they are typically formed. engineer the band gap by growing a series of layers
with different doping and mole fractions x to form
heterojunction devices.
When a layer of AlGaAs is grown on top of a layer
1.08.1.2 2D Systems
of GaAs, discontinuities in the conduction and
Two-dimensional electron gases (2DEGs) are valence bands form. It is possible to trap carriers
important since they represent the conducting and thereby form a 2D sheet of carriers at the
layer that is subsequently confined laterally to cre- AlGaAs/GaAs interface. However, by a technique
ate 1D systems. A 2DEG occurs when electrons are called modulation doping, it is possible to position
confined to an interface between two different donors in an AlGaAs layer some distance away from
materials, such as the interface between a thin sili- the interface where the 2DEG is formed. The elec-
con and silicon dioxide insulating layer or the trons from these donors tunnel to the lower energy
interface between GaAs and AlGaAs. Here, the ver- state at the interface leaving behind ionized impurity
tical confinement of the electron gas means that the states. Since the electron gas formed is spatially
spacing of all the energetically accessible particle- removed from these ionized donors, the scattering
in-a-box energy levels for modes in the z-direction from these charged impurities is reduced and the
is greater than the level broadenings and kBT. The mean free path increases.
density of states in the z-direction is measurably When the doped layer is n-type, the Fermi energy
discrete and the z-component of the wave function in this layer is shifted toward the conduction band in
has a standing wave form. Therefore, the electron analogy to a positively biased surface gate. This
cannot move classically in the z-direction and the forces the bands in the intrinsic layers to bend
transport is 2D. upward in order to balance the internal electric
The dimension at which this vertical confinement field and creates a triangular quantum-well in the
occurs is system specific but is usually of the order of conduction band at the GaAs/AlGaAs heterointer-
the Fermi wavelength, F, and in a metal-oxide face. This is shown schematically in Figure 1(a).
semiconductor field effect transistor (MOSFET) or Conversely, if the doped layer is p-type, the bands
high-electron-mobility transistor (HEMT) is typi- in the intrinsic layers bend down, forming a
cally hundreds of angstroms at low temperatures. quantum-well in the valence band in which holes
The most authoritative text on 2D systems is that accumulate to form a 2D hole system (2DHS) as
by Ando et al. (1992). However, a brief summary as to shown in Figure 1(b).
how they are formed will be given here before we A big advantage of GaAs systems over silicon-
consider 1D systems. based devices is that it is possible to produce sam-
A 2DEG can be formed in GaAs with the impor- ples where the electron mean free path is much
tant advantage that it is formed at a crystalline GaAs/ larger than that of bulk materials with the same
AlGaAs heterointerface. GaAs can be considered as a carrier concentration. This makes the ballistic
special case of the more general group of ternary regime readily accessible in GaAs devices. Lower-
compounds AlxGa1xAs, where x represents the Al dimensional systems can then be created by using
mole fraction. As a result, the variation in lattice etching or surface gates to confine the 2D system to
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 285
(a) (b)
n-AIGaAs AIGaAs GaAs p-AIGaAs AIGaAs GaAs
EF
2DEG
2DHG
EF
Figure 1 a) Band diagram of a GaAs/AlGaAs heterostructure with an n-type doping layer and intrinsic AlGaAs spacer layer
forming a two-dimensional election gas (2DEG) at the GaAs/AlGaAs interface. b) The analogous band diagram for p-type
doping to form a two-dimensional hole gas (2DHG).
lϕ
1.08.1.3 1D Systems
(b)
When the width of the 2D system is confined later- Ballistic
ally to a length scale that is comparable to the Fermi
wavelength, then we form a quasi-1D system. Again, W
we can experience diffusive or ballistic transport
depending on the comparison between the length of
L << lmfp
the wire and the mean free path. In the case where
the length of the wire is longer than the mean free
(c) Quantum point contact
path, the electrons suffer many elastic scattering
events and the wire is diffusive (see Figure 2(a)). If
however the length of the wire becomes comparable W
to the mean free path, the wire becomes ballistic (see
Figure 2(b)). There is also a situation where the W ≈ L « lmfp
width is comparable to the length and is much less
than the mean free path, and this corresponds to a Figure 2 Electron trajectories in the (a) diffusive and
quantum point contact (QPC) see Figure 2(c). (b) ballistic regimes and in (c) a quantum point contact. Here
W is the width of the constriction, L is the length l is the phase
In ballistic wires or QPCs, the constriction in the coherence length, and lmfp is the mean free path. Modified
2DEG is narrow enough that the electron wave from van Wees BJ, van Houten H, Beenakker CWJ, et al.
functions form 1D subbands. In an ideal case, the (1988) Quantized conductance of point contacts in a two-
longitudinal momentum is conserved and no dissi- dimensional electron gas. Physical Review Letters 60: 848.
pation takes place in the semiconductor. The
conductance of such a 1D conductor does not 1.08.1.3.1 Quantum transport in 1D
depend on the length of the channel but only systems
depends on the number of 1D modes in the channel. The Schrödinger equation for a quasi-1D system is
Here, we can no longer define a local conductivity given by
as in the diffusive case and we can no longer directly
P2
use Einstein’s relation between the conductivity and þ V ðyÞ ¼ E
2m
the diffusion constant. Instead, however, we must
use the Landauer formula, which connects the con- where V(y) is the lateral confinement potential. The
ductance of a ballistic conductor with a Fermi-level electron wave function within the quasi-1D conduc-
property of such a system. tor can be written as
286 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
n=3 0.8
ID DOS (a.u.)
0.6
n=2
0.4
n=1
0.2
n=0
0.0
y 0 1 2 3 4 5 6
E (meV)
Figure 3 One-dimensional (1D) subbands: (a) eigenfunctions of a simple harmonic oscillator and (b) density of states for a
1D system with a square well potential.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 287
4.0
10 3.0
G /e 2/h
0z
2.0
–10
–4 0 1.0
–2 4 ωy
2
0 2 ωx
x 0 4
2 0.0
–2 y –2 0 2 4 6 8
4 –4 (E –Vo)/ hωx
Figure 5 (a) Schematic diagram of a symmetric saddle point potential. (b) Calculated conductance of a quantum point contact
(QPC) as a function of energy E relative to the energy of the longitudinal harmonic oscillator energy, h!x/2 plotted for differently
shaped saddle potentials parametrized by the ratio !x/!y (Büttiker, 1990). Here, spin degeneracy has not been included and as a
result the conductance quantization is in units of e2/h. Reproduced with permission from Büttiker M (1990) Quantised
transmission of a saddle point constriction. Physical Review B 41: 7906. Copyright (1990) by the American Physical Society.
wire will depend on a transmission probably matrix temperature, and effective mass and are summarized
Tn,m so that briefly here for clarity.
e2 Disorder plays a pivotal role by determining all
G¼ Tn;m the important length scales. It can be directly corre-
h
lated to the mobility of the system and therefore the
where Tn,m depends exponentially on the energy of
mean free path. Disorder also sets the phase coher-
the mode ("n) and is given by
ence length as a larger number of scattering events
Tn;m ¼ n;m 1=ð1 þ e – "n Þ increases the probability of a carrier losing its phase
memory. However, disorder also influences the
The total conductance is therefore given by Fermi wavelength. This is because disorder deter-
e2 X e2 mines the minimum density at which a mesoscopic
G¼N Tm;n ¼ N T system can operate and hence the maximum Fermi
h h
wavelength. As a result, the disorder of a mesoscopic
As a result, the shape of the saddle point potential plays system sets the upper limits of both the length and
an important role in determining the shape of the width of a quantum wire.
conductance quantization plateaus. For very short Confinement potential is also a crucial factor. As we
quantum wires (!y/!x < 1), there is very little evidence have seen, the relative curvature of the confinement
of conductance quantization. Where the confinement is potential parallel and perpendicular to the wire will
symmetric (!y/!x ¼ 1), plateaus develop in the con- influence the shape of the conductance plateaus – the
ductance and as this ratio increases, the plateaus length of the steps and the sharpness of the risers. The
become flatter and longer (Büttiker, 1990). shape of the confinement potential (i.e., parabolic or
square well) determines the relative spacing between
energy levels while the strength of the confinement
1.08.1.4 Observing Conductance
potential will determine the absolute energy-level spa-
Quantization
cing. The energy-level spacing is also affected by the
From the above discussion, it is evident that many effective mass of the carriers. Having a small effective
factors influence whether or not conductance mass increases the energy-level spacing.
quantization can be observed in 1D systems. Maximizing the energy-level spacing is particu-
These include disorder, confinement potential, larly important for the observation of conductance
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 289
quantization to mitigate the effects of thermal broad- the dimensionality from 3 to 2 dimensions. Fowler used
ening, which smears out the plateaus. It follows that an n-type silicon MOSFET, with two p-type implanted
lower temperatures are also preferable for the obser- regions patterned either side of a 1 mm channel covered
vation of conductance quantization. by a metal top gate. The p-type implanted regions fully
Today, conductance quantization is readily obser- depleted the lightly doped n-type channel. However, by
vable in GaAs due to the development of high-purity applying a positive bias to the top gate, it was possible to
heterostructures, nanolithography techniques, and induce a 1D conducting channel. The devices, however,
milli-Kelvin measurements. In the next section we were highly disordered and the authors were restricted
review the development of some of these techniques to looking at the transition from 2D to 1D variable
and the insight into 1D ballistic transport that has range hopping. Since that time, there have been many
been gained as a result. investigations of how to controllably confine the width
of the conducting channel.
The first variable width system was demonstrated
1.08.2 Ballistic Conduction in n-type in GaAs-based heterostructures by Thornton et al.
GaAs-Based 1D Systems (1986). Here, they were able to look at the transport
1.08.2.1 Introduction characteristics of a 2DEG in a narrow channel formed
by the split gate of a GaAs–AlGaAs heterojunction
Ballistic transport was first observed in 1D devices field effect transistor, as shown in Figure 6. A split
based on GaAs/AlGaAs n-type heterostructures. One gate (SG) is literally a gate that has a gap in it so that
of the reasons for this is that GaAs-based heterojunc- the two sides can be biased independently. By applying
tions have long mean free paths and are generally a negative voltage to these gates, they first deplete
easier to fabricate. In addition, the smaller effective the carriers underneath them and then start to deplete
mass of electrons increases the subband spacing, the carriers laterally.
thereby enhancing quantization effects. Today, advan- In this way, the width of the 1D channel could be
cing growth and fabrication techniques have made it varied from 0.4 mm down to 0 mm. Thornton et al.
possible to readily achieve long mean free paths and investigated the conductance in the channel as a
hence obtain 1D ballistic conductance quantization. function of magnetic field for different temperatures,
Moreover, high-purity 1D systems have become a as shown in Figure 7. They were able to show that, at
fertile playground for the study of unique many- small sample widths, the magnetic field behavior did
body phenomena indicative of spontaneous spin polar- not fit standard models of 2D conductance but did fit
ization at zero magnetic field and Kondo-like physics. 1D models. As such, they were able to demonstrate a
In Section 1.08.2.2, we describe the developmental transition from 2D to 1D conductance as they
steps that lead to the first observations of ballistic reduced the width of the conducting region.
conductance plateaus. In Section 1.08.2.3, we survey However, since the dimensions of their device were
several of the most commonly used techniques for still relatively large (0.6 mm wide and 16 mm long)
fabricating ballistic 1D systems, while the effect of a compared with the mean free path, they did not
source–drain bias and in-plane magnetic field will be observe ballistic transport through their 1D system.
reviewed in Section 1.08.2.4. Many-body physics in
1D systems will be addressed in the Sections 1.08.2.5 Split-gate electrodes
and 1.08.2.6 is where we take a detailed look at the
0.7 structure and zero-bias anomaly and review the 2DEG
theories that have emerged to explain their origins.
0.41 K
0.46 K
4.0 × 10–7
0.56 K
0.6 × 15μ Gap
3.0
0.6 K
2.0
1.0
0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Magnetic field (T)
Figure 7 Conductance as a function of magnetic field indicated by crosses. The solid lines indicate the best fit to
one-dimensional (1D) weak localization theory at different temperatures. The inset shows the gate defining the narrow channel
in the underlying heterojunction. From Thornton TJ, Pepper M, Ahmed H, Andrews D, and Davis GJ (1986) One dimensional
conduction in the two dimensional electron gas of a GaAs-AlGaAs heterojunction. Physical Review Letters 56: 1198, Fig. 1.
Following these initial results, two groups pur- of 50 nm. At low temperatures, the longitudinal
sued this technique to make 1D systems in materials resistance showed quantized conductance steps as
in which the elastic mean free path was much longer the voltage on the SG was made more negative (see
than both the length and the width of the 1D chan- Figure 8):
nel. As a result, the first evidence for ballistic
h 2e 2 N
conductance quantization was independently R¼ ; G¼
2e 2 N h
observed by both the Delft–Philips group led by
van Wees et al. (1988) and the Semiconductor where N is the number of 1D subbands in the constric-
Physics group, Cambridge University led by tion. As the split-gate voltage was made increasingly
Pepper and co-workers (Wharam et al., 1988). Both negative, the plot showed an initial rise of the resis-
of these results will be discussed here for tance at Vg ¼ 0.4 V due to the depletion of the
completeness. electrons under the gate to create the narrow channel.
In the Wharam paper (Wharam et al., 1988), the SGs This was then followed by the continual depopulation
defined a channel that was 0.4 mm long and 0.5 mm wide of the 1D channels giving steps in the resistance on the
and 0.7 mm below the surface. Due to the spreading order of h/2e2. The measurements were taken in a
resistance of the electric fields below the gates and the quasi-four-terminal configuration with the contacts
negative voltage applied to the SG, it was possible to several microns away. As a result, the resistance of
reduce the width of the channel and therefore depopu- the 2DEG regions at either end contributed a series
late the 1D subbands. The mobility of the 2DEG was resistance to the overall resistance of the sample. When
between 2.5 105 and 1 106 cm2 V1s1 for carrier the series resistance was subtracted, the plateaus were
concentrations between 2 and 5 1011 cm2. At low found to be quantized to within 1%.
temperatures, the elastic mean free path is given by Generally, the advantage of using a four-terminal
pffiffiffiffiffiffiffiffi resistance measurement is that one can minimize the
h 2n resistance of the 2DEG region, which is in series with
lmfp ¼
e the 1D channel. It is worth mentioning, however, that
So, for these samples, this corresponded to a mean at present quantized conductance steps cannot be
free path in excess of 2 mm with a Fermi wavelength used as a conductance standard. One of the key
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 291
1/3
R (h/2e 2 ) 4
R (kΩ)
1/4
3
1/5
1/6
1/7 2
1/8
1/9
1/10
1
0
–3 –2 –1
Vg(V)
Figure 8 The channel resistance as a function of split-gate voltage for two different carrier concentrations in a high-mobility two-
dimensional electron gas (2DEG), showing quantization of the resistance in units of h/2e2. The dimensions of the SG were 0.5 mm
wide and 0.4 mm long and measurements were taken at T 0.1 K. Different carrier concentrations were achieved by illumination.
The inset shows a schematic diagram of the device used in this work. From Wharam DA, Thornton TJ, Newbury R, et al. (1988)
one-dimensional transport and the quantisation of the ballistic resistance. Journal of Physics C: Solid State Physics 21: L209, Fig. 1.
overgrowth wires, V-groove quantum wires, and Interestingly, the confinement strength of these
electrostatically induced 1D systems, and are briefly shallow-etched samples is much larger than that of
reviewed here. a typical SG device, resulting in large 1D subband
spacings of 20 meV. In these devices, therefore,
1.08.2.3.1 Shallow etching techniques the 1D conductance plateaus were observed to
Another method for the fabrication of 1DEGs persist up to a relatively high temperature of 30 K.
was developed by Scherer et al. (1987). Here, they These shallow-etched QPCs form a very good sys-
laterally patterned a 2DEG with low-energy tem for studying both quantized 1D conductance
ion-beam-assisted etching to form channels as low plateau and the 0.7 structure at elevated tempera-
as 75 nm wide. They noted that there was no loss in tures (see Section 1.08.2.5).
electron mobility by the creation of the wires and
that this technique would permit investigation of 1.08.2.3.2 Cleaved-edge overgrowth wire
devices where the wave nature of electron transport Yacoby and co-workers (Yacoby et al., 1996) pio-
would be dominant. Kristensen and co-workers then neered a new technique to fabricate 1D wires
went on to use shallow-etching techniques to form using cleaved-edge overgrowth by MBE and a
three different types of quantum point contacts (see novel in situ contacting scheme. They first grow a
Figure 10). In their type-I sample, the 1D constric- modulation-doped GaAs quantum-well embedded
tion was formed by shallow wet etching as shown in between two thick AlGaAs layers with a doped
Figure 10(a). The etched constriction was then cov- GaAs top layer as shown in Figure 11(a) . In this
ered by a 10-mm-wide, 100-nm-thick Ti/Au top gate. case, the resulting 2DEG is 500 nm below the top
Negatively biasing the top gate simultaneously surface. A long and narrow tungsten stripe is then
reduced the density in the wire and laterally confined evaporated over the top of the heterostructure that
the wire. In their type-II and type-III devices is ultimately used to gate the wire. The quantum
(Figure 10(b)), etched trenches define the quantum wire itself is prepared by cleaving the sample in
point contact within two large areas of 2DEG. In this ultrahigh vacuum as shown in Figure 11(a) and
case, the 2DEG region can be used as side gates to then overgrowing the smooth cleavage plane
alter the width of the conducting channel. In type II with a second modulation doping scheme (see
(III) devices, the trenches were etched 60 nm (90 nm) Figure 11(b)). As shown in Figure 11(b), electrons
to remove the donor layer. are introduced at the edge of the quantum-well,
(a) (b)
500 nm
AlGaAs Si donors
AlGaAs Ti/Au top gate
Source
QWR
50 nm Drain
QWR Gate2: Sidewalls with QWR
Figure 12 (a) Cross-sectional transmission electron microscopy (TEM) images of the quantum wire ridge region, charge
distribution due to doping is schematically shown. Taken from Fig. 1 in Electron transport in AlGaAs/AlGaAs V-groove
quantum wires. (b) A tilted scanning electron microscopy (SEM) image of the S-shaped device. Black arrows indicate the
trajectory of the current flow. Two metallic gates of 0.25 mm width each are deposited, one on the sidewall area and the other
on the full sidewall–wire–sidewall path. Taken from Fig. 1 (b) in Conductance quantization in V-groove quantum wires,
(a) From Dwir B, Kaufamn D, Berk Y, Rudra A, Paleski A, and Kapon E (1999) Electron transport in GaAs/AlGaAs V-groove
quantum wires. Physica B 259–261: 1025. (b) From Kaufman D, Berk Y, Dwir B, Rudra A, Palevski A, and Kapon E (1999)
Conductance quantization in V-groove quantum wires. Physical Review B 59: R10433.
that can act as an in situ top gate. Positively biasing essentially eliminated and the epitaxial GaAs–
the gate induces an electron gas at the intrinsic AlGaAs interface minimizes interface roughness
GaAs–AlGaAs interface. A quantum wire can be scattering.
formed by patterning the in situ top gate into three Several groups have used the undoped hetero-
distinct regions as shown in Figure 13. This is structure to understand quantum transport in low-
usually done by a combination of electron-beam disorder quantum wires. Reilly et al. (2001) used
lithography and wet-etching techniques. In this induced quantum wires to study many-body effects
way, the central region of the top gate can be used as a function of wire length. Typically, they noted
to induce a 1D electron system while the outer that for these induced architectures, a change in
regions can be biased negatively to control the density always coincided with a change in the shape
lateral confinement of the wire. of the quantum-well and it was difficult to disentan-
The key advantage of induced, undoped quantum gle the contributions from each effect. More recently,
wires is that they have ultra-low levels of disorder. Sarkozy et al. (2009) have also used induced hetero-
The fact that there is no modulation doping means structures to study many-body effects in low-
that scattering from remote-ionized impurities is disorder quantum wires. In particular, they looked
VT > 0
+ + +
Barrier
Wire
Figure 13 (a) Schematic cross section of an semiconductor–insulator–semiconductor field effect transistor (SISFET)
quantum wire device. The metallically doped cap is patterned into three regions that act as independent surface gates.
Positively biasing the middle gate induces a one-dimensional interacting election system (1DES) directly beneath the gate at
the GaAs/AlGaAs interface while a negative bias on the outer gates controls the lateral confinement potential. (b) Scanning
electron micrograph of an induced quantum wire device fabricated by electron beam lithography and wet etching. From Kane
BE, Facer GR, Dzurak AS, et al. (1998) Quantized conductance in quantum wires with gate-controlled width and electron
density. Applied Physics Letters 72: 3506.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 295
at the zero-bias anomaly, which is believed to result energy crossing the spin-split subbands. By measur-
from Kondo-like physics (see Section 1.08.2.6). ing the transconductance, it is possible to get a more
accurate position of the conductance steps. If we
consider the lowest trace for Figure 15(a), we see
1.08.2.4 Energy-Level Splitting and the 1D
the transconductance of the 1D channel versus gate
g-Factor
voltage for B ¼ 0. The position of the n ¼ 1 and n ¼ 2
1.08.2.4.1 Zeeman splitting in 1D electron conductance plateaus occur when dG/dVg is zero. As
systems a parallel magnetic field is applied to the current
In Section 1.08.2.2, we introduced the first observa- direction, a linear splitting of the transconductance
tion of ballistic conductance in 1D systems. At zero peaks becomes apparent.
magnetic field, the conductance steps are quantized
in units of 2e2/h, where the factor 2 comes from 1.08.2.4.2 Source–drain bias
electron spin degeneracy. When a large magnetic spectroscopy
field is applied in the plane of the 1D channel, the The quantized conductance plateaus also split lin-
electron spin degeneracy is lifted and the conduc- early with the application of a large bias between
tance plateaus become quantized in units of e2/h. source and drain. The effects of a source–drain bias
This effect is called Zeeman splitting in 1D electron on 1D subbands were also investigated by Patel
systems. The first Zeeman splitting in a 1D electron et al. (1991a, 1991b), as shown in Figure 15(b).
system was observed by Wharam et al. (1988), as seen They measured a linear splitting of the transcon-
in Figure 14. Here a magnetic field of 13.6 T was ductance peaks both with increasing magnetic field
applied in the plane and parallel to the 1D subbands and for increasing source–drain bias. Increasing
of a high-mobility 1 106 cm2 V1 s1 2DEG at the source–drain voltage lifts the momentum
T 0.1 K. The magnetic field lifted the spin degen- degeneracy and splits each conductance plateau
eracy, doubling the number of plateaus observed (in the differential conductance vs. gate voltage
in the quantized resistance. These steps occur at trace) into two. Again, a linear splitting in the
R ¼ h/2(N þ 1/2)e2. transconductance peaks is observed, this time as
Patel et al. (1991a) extended the parallel field work Vsd increases.
of Wharam et al. (1988) and measured the transcon- Source–drain bias spectroscopy is based on a
ductance dG/dVg of a 1D channel as a function of SG model developed by Glazman and Khaetskii (1989)
voltage at various magnetic fields and at various and is now a standard method for measuring subband
source–drain biases, as shown in Figure 15. The spacings in 1D systems . The method adds a direct
positions of the peaks correspond to the Fermi current (DC) bias to the alternating current (AC)
excitation voltage in order to change the relative
chemical potentials, , at the source and the drain
1500 of the quantum wire. When no DC bias is applied, the
1/9 differential conductance follows the standard quan-
1400
Resistance (Ω)
1/10 1300
1200 XN
1/11 dI e2
G¼ ¼ 2n
1100 dV h
n¼0
1000
where N is the number of subbands occupied.
–1.5 –1.0 –0.5 However, when a DC bias equivalent to the subband
Vg(V) energy spacing is applied, the differential conduc-
tance becomes
Figure 14 The channel resistance of a 1D electron system
as a function of applied gate voltage when a magnetic field dI XN
e2
of 13.6 T is applied parallel to the channel. The G¼ ¼ ð2n þ 1Þ
dV n¼0
h
measurement temperature T is 0.1 K. The spin splitting
gives rise to additional quantized plateaus at R ¼ h/(21e2), h/
Applying a source–drain bias is known to select dif-
(23e2), and an incipient plateau at h/(19e2). From Wharam
DA, Thornton TJ, Newbury R, et al. (1988) One-dimensional ferent numbers of channels from the source and
transport and the quantisation of the ballistic resistance. drain, and is expected to give rise to a half-integer
Journal of Physics C: Solid State Physics 21: L209, Fig. 1. plateau in zero magnetic field. In other words, the
296 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
(a) 12 (b) 10
10 B = 15T B = 0T
8 Vsd (mV)
dG /dVg (a.u.)
dG /dVg (a.u.)
8
B = 10T 6 1.5
6
B = 5T 4 1.0
4
2 2 0.5
B = 0T
n=1 n=2 n=1 n=2
0 0 0.0
–3.6 –3.4 –3.2 –3 –2.8 –2.6 –2.4 –3.6 –3.4 –3.2 –3 –2.8 –2.6 –2.4
Vg (V) Vg (V)
Figure 15 (a) Transconductance as function of gate voltage at different magnetic fields. (b) Transconductance as a function of
gate voltage at different source drain biases for B ¼ 0 T. From Patel NK, Nicholls JT, Martin-Moreno L, et al. (1991a) Properties of
a ballistic quasi-one-dimensional constriction in a parallel high magnetic field. Physical Review B 44: R10973, Fig. 2.
conductance is quantized in odd integer multiples of magnetic fields with those in the presence of
e2/h. Therefore, by measuring at what DC bias the source–drain biases, Patel et al. found that for the
conductance plateau becomes quantized in odd inte- second and third subband edge, the 1D g-factor was
ger multiples of e2/h, it is possible to determine the 1.1 – significantly enhanced over the bulk g-factor
subband energy spacing for each subband. By tuning 0.4. This large discrepancy was attributed to an
the source–drain voltage, one can then probe the 1D enhancement of the g-factor due to exchange and
subband spectrum. It is important to note that correlation mechanisms. This pioneering experi-
these half-integer plateaus at high Vsd are still spin ment did not require a perpendicular field as is
degenerate. Further application of a magnetic field normally required for the determination of the
results in the formation of quarter-integer plateaus g-factor for 2DEGs.
at 1.25(2e2/h), 1.75(2e2/h), 2.25(2e2/h), 2.75(2e2/h), Thomas et al. (1996) further extended the work of
3.25(2e2/h), . . . . Patel and co-workers and were able to measure the
Thomas et al. (1998b) went on to show that in g-factor when the current is either perpendicular or
finite source–drain bias experiments where well- parallel to the applied in-plane magnetic field.
resolved integer and half-integer plateaus were Their experimental results are reproduced in
observed quarter-integerplateaus were also seen for Figure 16. For Vg < 4 V, additional spin-split pla-
every subband even at zero magnetic field. teau are observed interleaved between those at zero
field. For Vg > 4 V, the Zeeman energy is compar-
able to the subband spacings and both sets of spin-
1.08.2.4.3 Determination of the g-factor split plateaus cannot be easily resolved. From the
in 1D inset (b), we can see gk and g? for all 26 subbands,
The extent to which the spins couple to an external and it is evident from these results that the g-factor
magnetic field is determined by the g-factor, which is almost isotropic. It is worth pointing out that
can be measured by finding the source–drain voltage when the number of occupied 1D subbands was
at which the splitting of the transconductance peaks high, the absolute value of the g-factor was found
is equal to the Zeeman splitting in an in-plane mag- to be 0.4, that is, close to the bulk value of GaAs.
netic field. At this point, the g-factor is given by the For the last few occupied subbands of the channel,
relation the anisotropy of the confinement potential can be
described by a saddle point (Martin-Moreno et al.,
eVsd ¼ 2gB SB
1992) and little anisotropy of the in-plane g-factor is
where B is the magnetic field, B is the Bohr magne- observed. However, as the number of 1D subbands
ton, and S ¼1=2. decreases, the g-factor was observed to increase
As shown in Figure 15, a linear splitting of the gradually. It was suggested that the enhanced
transconductance peaks was observed for both g-factor (1) over its bulk value was due to elec-
increasing magnetic field and for increasing tron–electron interaction effects as the constriction
source–drain bias. By comparing the data at high became narrow.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 297
20
1.5 (b)
g⊥
• The conductance tends toward 0.5(2e2/h) in a par-
allel magnetic field (Thomas et al., 1996).
g-factor
1 • The conductance has been observed to tend
toward 0.5 2e2/h with both increasing (Reilly
15
0.5 et al., 2001) and decreasing density (Thomas
G (in units of 2e 2/h)
I
II et al., 1998a, 2000; Pyshkin et al., 2000).
0
10
0 10 20
Subband Index
• The feature is activated with temperature.
Typically, it strengthens with temperature and
Sample A evolves downward from the 2e2/h plateau as
T = 60 mK G (2e2/h) 1 (a) temperature increases (Thomas et al., 1998a).
5 0.5
T = 600 mK
• It is often accompanied by a peak in the conductance
at zero source–drain bias (Cronenwett et al., 2002).
0
–5.8 –5.6
Vg (V) 1.08.2.5.1 Identifying the 0.7 structure
0 Although the 0.7 structure was observed in some of
–6 –5 –4 –3
the earliest experiments of conductance quantiza-
Gate voltage Vg (V)
tion in 1D, it was not appreciated as a phenomenon
Figure 16 (I) Gate voltage characteristics at B ¼ 0. (II) The in its own right until Thomas et al. (1996) performed
gate characteristics (offset by 0.3 V for clarity) in a magnetic a detailed study of the anomalous plateau. The
field of 11 T. Insets: (a) detail of the structure at 0.7 2e2/h; (b)
initial oversight is likely to be due to the fact that
the in-plane g-factors as a function of subband index, as
obtained from the Zeeman splitting at 8.2 T. From Thomas KJ, disorder within the channel can cause resonant
Nicholls JT, Simmons MY, Pepper M, Mace DR, and Ritchie backscattering and produce similar features at arbi-
DA (1996) Possible spin polarization in a one-dimensional trary conductances. It is therefore necessary to first
electron gas. Physical Review Letters 77: 135, Fig. 1. distinguish the 0.7 structure from these resonances.
This is possible because, unlike the 0.7 structure,
resonances will change on successive cool-downs,
1.08.2.5 Many-Body Physics in 1D – The 0.7
get weaker with temperature, appear in the risers at
Structure
higher subband indices with increasing amplitude,
In a very clean 1D channel, a clear plateau-like and will disappear when the channel is shifted lat-
structure close to (0.7 2e2/h) has been observed at erally. Therefore, it is important when investigating
zero magnetic field, which can be seen in the inset the 0.7 structure to measure not only the tempera-
(a) to Figure 16 (Thomas et al., 1996). This feature is ture and magnetic field dependence but also the
now well known as the 0.7 structure or 0.7 anomaly, ability to laterally shift the channel on the same
whose conductance value is placed between the sample.
spin-degenerate conductance plateau at 2e2/h and Liang et al. (2000) performed such a study using a
the spin-split conductance plateau at e2/h, and can- multi-layered gated 1D structure as shown in
not be explained within a single–particle picture. Figure 17(a). Here, a cross-linked PMMA layer sits
The 0.7 structure has been observed in numerous between the SG labeled in black and three finger
1D systems with different sample designs establish- gates (F1, F2, and F3) labeled in gray and acts as an
ing that it is a universal effect. Its origin remains the insulating layer, so that by biasing the finger gate
subject of intense experimental and theoretical voltages, it is possible to control the strength of
debate and has focused on zero-field spin polariza- lateral confinement of the 1D channel.
tion (Thomas et al., 1996; Spivak and Zhou, 2000; As shown in Figure 17(b), with decreasing nega-
Bruus et al., 2001; Starikov et al., 2003), spin-density tive finger gate voltage, VF2, the 1D conductance steps
wave formation (Reimann et al., 1999), pairing of become less pronounced and finally disappear due to
electrons (Flambaum and Kuchiev, 2000), singlet– weakening of the lateral confinement strength. The
triplet formation (Rejec et al., 2000), Kondo-like energy separation of the 1D subbands was determined
interactions (Cronenwett et al., 2002; Meir et al., from source–drain bias experiments at various VF2 and
2002; Lindelof, 2001), and electron–phonon effects demonstrated a good linear fit in Figure 17(c). As the
(Seelig and Matveev, 2003). finger gate voltage, VF2 was made more negative the
There are some general features of the 0.7 struc- energy spacing between the first and second subband,
ture that make it unique and intriguing: E1,2 (VSG) decreased, giving rise to the reduction in
298 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
(b) Device A
First cooldown
4
T = 1.2 K
1
0.05 μm VF2 = –1.8 V
0
–1.2 –1 –0.8 –0.6 –0.4
VSG (V)
0.3 μm 0.6 μm
(c)
2.4
2.1
ΔE12 (mV)
F1 F2 F3 1.8
SG
0.8 μm 1.5
Figure 17 (a) A schematic diagram showing a multi-layer-gated structure. The black areas represent the split-gate (SG). The
gray lines correspond to three overlaying finger gates F1, F2, and F3. There is a layer of crosslinked polymethylmethacrylate
(PMMA) in between SG and F1, F2, and F3 to act as a gate dielectric. (b) Conductance measurements as a function of SG
voltage VSG at various finger gate voltages VF2. From left to right VF2¼ 0 to 1.8 V in 0.3 V steps. (c) Measured subband
spacing at various VF2. (a) From Liang, (1999), Fig. 1(a). (b) From Liang (1999), Fig. 2(a). (c) From Liang, (1999), Fig. 2(b).
flatness of the conductance plateaus. Using the saddle et al., 1991a). In Figure 15, Patel et al. observed an
point potential model (Buttiker, 1990), they estimated additional shoulder-like structure at around
the change in lateral confinement strength by a factor Vg¼3.4 V in the transconductance of a high-mobility
of 2 over this range. Therefore, the 0.7 structure per- 1D channel. With increasing applied source drain bias,
sists despite a change in lateral confinement strength the shoulder-like structure appeared to split into two.
by a factor of 2, while the 1D ballistic conductance While the structure was observed, its origin could not
plateau disappears. This result showed compelling be explained. It was not until 1996 when Thomas and
evidence that the 0.7 structure is an intrinsic property co-workers performed a systematic study of an ultra-
of a clean 1D channel and persists over a wide range of clean 1D channel in an in-plane magnetic field that
lateral confinement potentials. the 0.7 structure was investigated thoroughly.
In this work, Thomas et al. used a high-quality
1.08.2.5.2 In-plane magnetic field 2DEG with a mobility of 4.5 106 cm2 V1 s1 at a
behavior of the 0.7 feature carrier density of 1.8 1011 cm2 showing nearly 30
Historically, the 0.7 structure was observable in some quantized plateaus and a well-defined structure at
of the earliest experiments (van Wees et al., 1988) and 0.7(2e2/h). The 0.7 structure was observed to be repro-
first commented on by Patel and co-workers (Patel ducible on thermal cycling, ruling out the possibility
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 299
T = 100 mK (a)
3 α2
Sample A
α1 β
G (in units of 2e 2/h)
B
2
10
B (T)
B
1
N+1
Q
Energy R
N P
N–1
Magnetic Field α0
0
–1.4 –1.2 –1
Gate voltage Vg (V)
Figure 19 Differential conductance G(Vg) traces for –1.25 –1.2 –1.15
different parallel magnetic fields from 0 to 15.6 T in steps of Gate voltage Vg (V)
0.6 T. The traces are offset for clarity. Inset: Schematic
energy diagram for a linear Zeeman splitting of one- Figure 20 Gray-scale plots of the transconductance,
dimensional (1D) subbands and subsequent crossings. dG/dVg,as a function of Vg and Bk The labels 1, 2, and
From Graham AC, Thomas KJ, Pepper M, Cooper NR, show the 0.7 analogs at the Zeeman crossing between
Simmons MY, and Ritchie DA (2003) Interaction effects at spin levels 1" and 2#, 1" and 3#, and 2" and 3#,
crossings of spin-polarized one-dimensional subbands. respectively. From Sfigakis F, Graham AC, Thomas KJ,
Physical Review Letters 91: 136404, Fig. 1. Pepper M, Ford CJB, and Ritchie DA (2008b) Spin Effects in
One Dimensional Systems. Journal of Physics C 20: 164213.
Fig. 6.
until the point where the Zeeman energy was equal
to the subband spacing and the integer plateaus dis-
appeared. This happened at a crossing point, such as to the 0.7 feature is visible. As Bk was increased, this
Q in the inset. Finally, the half-integer plateaus wea- plateau evolved into the 0.5(2e2/h) and the white
kened again as Bk increases further and the integer region broadened. By Bk ¼ 11 T, the first (N ¼ 1")
plateaus reappeared after this crossing (see R in the and second subbands (N ¼ 2#) crossed. After the
inset). They considered that the appearance of the crossing, the N ¼ 1" showed a discontinuous shift
1.5(2e2/h) plateau, which weakened and then evolved from the crossing point, marked 1. This discontinu-
into the 2e2/h plateau before reappearing for Bk > 8 T, ity corresponds to the 0.7 analog. This discontinuous
resembled the evolution of the 0.7 structure to shift was also observed at the crossing of the N ¼ 2"
0.5(2e2/h) with increasing Bk. As a consequence, they and the N ¼ 3# lines marked by , and the N ¼ 1"
have named the 1.5 (2e2/h) plateau, the 0.7 analog. and the N ¼ 3# lines marked by 2. Interestingly, at
The reappearing 2e2/h plateau carried the opposite these points, there was no anti-crossing of the sub-
spin than before the crossing, while the lowest sub- bands, but a gap formed abruptly after the crossing.
band, which does not encounter a crossing, does not Graham et al. argued that this 0.7 analog is a conse-
change its spin. quence of the strong exchange interactions and
The evolution of these conductance characteris- where there is a lifting of the zero-field spin degen-
tics is more clearly seen in gray-scale plots of the eracy in the 0.7 structure, there is also a lifting of the
transconductance as a function of Bk and Vg, as shown degeneracy of the crossing point for the 0.7 analog
in Figure 20. White regions represent the plateaus in (Graham et al., 2003; Graham et al., 2004). In further
conductance and the dark regions correspond to the works using DC bias spectroscopy, they showed that
transitions between plateaus. As Bk was increased the 0.7 structure and its analogs were caused by the
from zero, a splitting of the 1D subbands of opposite highest energy spin-up subband pinning to the che-
spin was observed, as shown in Figure 21 for point P. mical potential, as predicted by Kristensen et al.
At Bk ¼ 0, the first white plateau, 0, corresponding (2000), together with an abrupt rearranging of
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 301
TA (K)
–1.2 10
Vgs = 0.309 V
–1.6 TA = 1.11 K 5
–2.0 0
0.0 0.5 1.0 1.5 2.0 2.5 0.30 0.31 0.32 0.33
1/T (K–1) Vgs (V)
Figure 22 (a) Temperature dependence of the conductance suppression G0-G(T) at fixed gate voltages Vg ¼ 0.305V and
0.309 V, measured on device A. The data show Arrhenius behavior, G(T)/G0 ¼ 1-Cexp(TA/T),with an activation temperature TA.
(b) Measured activation energy as a function of gate voltage. (a) From Kristensen A, Bruus H, Hensen AE, et al. (2000) Bias and
temperature dependence of the 0.7 conductance anomaly in quantum point contacts. Physical Review B 62: 10950, Fig 8(a).
(b) from Kristensen, (2000), Fig 8(b).
G ¼ 2e2/h plateau to form the anomalous plateau at 1.3 1011 to 3 1010 cm–2 shifted the 0.7 structure to
G ¼ 0.7(2e2/h). 0.5(2e2/h) (Thomas et al., 2000), indicating a complete
Cronenwett et al. (2002) noted that the tempera- spin polarization (Wang and Berggren, 1996; Gold
ture dependence of the 0.7 structure is reminiscent of and Calmels, 1996). However, subsequent work on
the Kondo effect in quantum-dots with one unpaired induced electron gases showed that the conductance
electron. Extending this analogy, they performed of the 0.7 structure tended toward 0.5(2e2/h) at both
temperature dependence and source–drain bias stu- the highest and lowest densities, with further reports
dies and found several other conspicuous similarities. in an induced 1D wire of length 2 mm showing that
These included a conductance peak at zero source– the 0.7 structure evolved continuously to 0.5(2e2/h)
drain bias. In addition, plots of the conductance along with increasing 1D density (Reilly et al., 2001). Reilly
the 0.7 plateau as a function of T collapse onto a (2005) showed that the position and strength of the
single function of empirical Kondo-like form using a 0.7 structure did not depend on the absolute value of
single scaling parameter designated the Kondo tem- the carrier density but on the relative difference
perature, TK. Since then, several theories have been between the potentials in the 1D and 2D regions.
developed based on a Kondo explanation for the 0.7 Wirtz et al. (2002) performed interesting studies
structure, which explain many of the experimental on the hydrostatic effect on the 0.7 structure. Using
observations (Meir et al., 2002; Hirose et al., 2003; both pressure and illumination, the electron density
Schmeltzer et al., 2005) (see Section 1.08.2.6). in the 2DEG was reduced from 2.14 1011 to
However, Kondo models require that an unpaired 0.61 1011 cm2, and a shift in the conductance of
electron can become localized in the 1D region to the 0.7 structure toward the spin-split value of e2/h
act as a form of magnetic impurity. While it has been was observed (see Figure 23). The electron density
shown that such a state is theoretically possible under below the SG is lower than in the 2DEG, so it is
certain assumptions (Meir et al., 2002), other density possible that the exchange energy could be signifi-
functional (Starikov et al., 2003) and Hartree–Fock cantly enhanced with the Hartree term being
(Sushkov, 2003) calculations did not confirm a loca- completely collapsed inside the constriction.
lized spin state in real QPC devices. Theoretically, such an enhancement could lead to a
spin polarization. Theoretical results by Bruus et al.
1.08.2.5.4 Density dependence of the 0.7 (2001) suggested that an increase of the spin gap
feature between the up and down states should result in a
Studies of the carrier density dependence of the 0.7 shift of the 0.7 structure toward e2/h.
structure are not straightforward. Initial experiments Using a cleaved-edge overgrown 1D wire, de
showed that lowering the carrier concentration from Picciotto et al. (2004) observed a 0.7 structure in the
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 303
appearance of these quarter-integer plateaus did not results in a potential difference that can be measured.
change with increasing magnetic field, further sup- The authors showed that, in a ballistic 1D system, the
porting the concept of spin polarization at zero steps in the conductance between the quantized
magnetic field. More recently, plateaus at 0.25 (2e2/h) values gave rise to peaks in the measured thermo-
and 0.75 (2e2/h), nearly identical to those observed in power. By measuring the height of the thermopower
GaAs quantum wires, have also been observed in peaks, it was possible to determine the temperature
nonequilibrium transport studies in In0.75Ga0.25As of the hot electrons in the heating channel.
quantum wires (Simmonds et al., 2008). Their appear- Appleyard et al. (2000) also went on to show the
ance has also been attributed to zero-field presence of a small shoulder in the thermopower
spin-polarized quarter-integer plateaus. near the 0.7 structure. Application of a magnetic
Sfigakis et al. (2008a) considered that in a single field converted the 0.7(2e2/h) to the spin polarized
particle picture, the quarter-integer plateau for a 0.5(2e2/h) value, accompanied by a decline in the
given subband should appear at the same time when thermopower signal to zero. The observed thermo-
sweeping the source–drain bias. For the high index power signal at 0.7(2e2/h) is consistent with the first
subbands, this appeared to be the case. However, as subband being split, with the spin# subband being
the subband index decreased, the (N þ 0.25) 2e2/h transmitted, and the other pinned near the chemical
plateau appeared at increasingly higher bias than potential. Since a fully transmitted channel leads to a
the (N þ 0.75) 2e2/h. The late formation in source– zero thermopower, the 0.7 structure could be
drain bias voltage of the 0.25(2e2/h) plateau with explained by partial transmission of the minority
respect to the formation of the 0.75(2e2/h) plateau spin" subband. While the Mott formulation does
was shown to be consistent with the predictions from not predict a peak in G at 0.7(2e2/h), it is likely that
a spin-gap model. Here, self-consistent calculations in the regime of strongly interacting electrons an
with a saddle point model and a density-dependent additional mechanism is required to explain the finite
energy gap between spin subbands were shown to measure thermopower coincident with 0.7 structure.
exhibit the same characteristics (Sfigakis, 2005). The thermal conductance,
, and electrical con-
ductance, G, of noninteracting electrons in quantum
wires are known to be related by the Wiedemann–
1.08.2.5.6 Thermopower, thermal Franz equation
conductance, and shot noise dependence
of the 0.7 feature 2 kB2 T
K ¼ – G
In addition to conventional transport measurements, 3e 2
the 0.7 structure has also been observed in studies of As a consequence, if the conductance is quantized in
the thermopower, thermal conductance, and shot units of 2e2/h, then the thermal conductance will be
noise experiments. While not comprehensive, the quantized in units of 2kB2 T =3h. Below 2e2/h, the
following section briefly reviews these experiments. Wiedemann–Franz equation no longer holds, but at
The thermopower of a 1D system, S, is related to gate voltages near the 0.7 structure, a half plateau in
the conductance, G: the thermal conductance was observed by Chiatta
et al. (2006).
V 2 kB2 qð1nGÞ
S¼ ¼ – ðTe þ Tl Þ Finally, shot noise experiments in a QPC in the
Te – Tl I ¼0 3e qm
regime of the 0.7 structure were also performed. Shot
by the Mott formula (Mott, 1963), where is the noise is the temporal fluctuation of current resulting
chemical potential of the contacts relative to the 1D from the quantization of charge. DiCarlo et al. (2006)
subbands, Te is the electron temperature, Tl is the reported simultaneous measurements of the shot
lattice temperature, and V is the voltage measured noise and DC transport in a QPC. In particular,
across the constriction. Initial thermopower experi- they focused on the 0.7 structure and its evolution
ments used a 1D constriction to measure the electron in a parallel magnetic field. They observed a suppres-
temperature (Appleyard et al., 1998). Here, electrons sion of the noise, compared with that expected for
were heated on one side of a 1D constriction by the spin degenerate transport, near 0.7(2e2/h) in zero
application of an electric current. However, the hot magnetic field in agreement with earlier results by
electrons are prevented from passing through to the Roche et al. (2004). With increasing Bk, this suppres-
other side by the 1D constriction and a temperature sion was observed to evolve into the signature
difference is established across the channel. This of a spin-resolved transmission. They found
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 305
quantitatively good agreement between their noise less efficiently screened, and the conductance is
data and the phenomenological model for density- reduced toward its high-temperature limit of 0.5(2e2/
dependent level splitting (Reilly, 2005). h). This gives rise to the 0.7 structure and provides an
Since the first reported observation of the 0.7 struc- explanation for why the plateau at 0.7(2e2/h) is
ture, its origin has been intensely debated. stabilized by temperature.
Throughout the deliberations, both Coulomb and A critical assumption of the Kondo model is that
exchange energies have been shown to be very impor- an open 1D system is able to support a bound state.
tant in quasi-1D systems since they can cause a This assumption appears to be robust according to
spontaneous spin polarization. Indeed, the concept of recent spin-density functional theory calculations
a spin gap has been fairly successful in explaining most (Meir et al., 2002; Hirose et al., 2003; Rejec and
of the experimental results of the 0.7 structure Meir, 2006) and tunneling spectroscopy measure-
(including the temperature dependence, magnetic ments between parallel quantum wires (Auslaender
field, and bias dependence) (Wang and Berggren, et al., 2005; Yoon et al., 2007). From this assumption it
1996; Spivak and Zhou, 2000; Kristensen et al., 2000; follows that (Meir, 2008):
Reilly et al., 2002; Berggren and Yakimenko, 2002;
Graham et al., 2005; Lassl et al., 2007; Graham et al., 1. The bound state will blockade the current, result-
2007). Central to this interpretation is the concept that ing in a reduced conductance between 0.5(2e2/h)
the spin gap must depend on carrier density (gate and 2e2/h.
voltage), otherwise a quantized plateau at 0.5(2e2/h) 2. As the Fermi level is increased, the Coulomb
would be observed. To date, however, none of the blockade energy is overcome and the universal
SG models have addressed the appearance of a peak value of 2e2/h is recovered.
in the differential conductance of the quantum wire 3. In the presence of a magnetic field, the energy of
for G < 2e2/h at very low temperatures (Cronenwett the localized state splits and the conductance is
et al., 2002). This is discussed in the next section. reduced to 0.5(2e2/h).
4. There is a reduction in shot noise around the 0.7
plateau due to charge being carried by two spin
1.08.2.6 The Kondo Model and Zero-Bias channels, one of perfect transmission and the other
Anomaly with reduced transmission.
1.08.2.6.1 Overview These predictions are consistent with experimental
Although a spontaneous spin-splitting model explains observations.
many of the salient features of the 0.7 structure, there However, the greatest strength of the Kondo
sometimes exists a peak in the conductance around model is arguably its ability to explain the zero-bias
Vsd ¼ 0 that cannot be explained within the spin-split anomaly. The zero-bias anomaly was originally stu-
paradigm. This conduction peak has been termed the died by Cronenwett et al. (2002), who showed that the
zero-bias anomaly, in analogy to the zero-bias anomaly 1D zero-bias anomaly shared many of the features of
observed in quantum-dots. The zero-bias anomaly has
its 0D counterpart including:
led to a Kondo-like description of the 0.7 structure that
also describes many of its properties – in particular, its 1. the collapse of G(T) curves onto a single function
unusual temperature dependence. with a single scaling parameter designated the
The Kondo model predicts that near pinch-off Kondo temperature, TK;
(Meir, 2008), a 1D system contains a bound state that 2. an exponential dependence of TK on VG;
acts as a magnetic impurity. Without electron–elec- 3. a correlation between zero-bias peak width and
tron interactions, the bound state would block carriers TK; and
of the same spin from passing through the 1D channel 4. splitting of the zero-bias peak in a parallel mag-
– limiting transport to a single spin channel and redu- netic field.
cing the conductivity to 0.5(2e2/h). However,
electron–electron interactions cause the bound state Subsequent experiments have confirmed the exis-
to polarize electrons in the leads and thereby screen its tence of the zero-bias anomaly as a fundamental
effect. At T ¼ 0, the magnetic impurity is screened property of quantum wires but have found less cor-
perfectly by the carriers in the leads and the conduc- respondence with the 0D zero-bias anomaly (Sfigakis
tivity reaches the unitary limit, 2e2/h. As the et al., 2008a; Sarkozy et al., 2009). In particular, the 1D
temperature is increased, the magnetic impurity is zero-bias anomaly does not always split with parallel
306 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
magnetic field and the observed splitting is often less 2.5(2e2/h) that trend upward as Vsd is shifted away
than predicted (Chen et al., 2009). from zero (Glazman and Raikh, 1988; Ng and Lee,
Today, there is a growing consensus that bound 1988; Patel et al., 1991b).
states can and do exist in quantum wires resulting in Below 2e2/h, however, they observed a lot more
Kondo-like physics. However, whether or not this is structure in the nonlinear conductance. The data
the origin of the 0.7 structure is still a point of con- showed a clear zero-bias peak – the so-called zero-
tention as there is mounting evidence to suggest that bias anomaly. Importantly, there was no bunching of
the 0.7 structure and Kondo-like phenomena co-exist lines along Vsd ¼ 0 and hence, no plateau at 0.7(2e2/h)
as separate and distinct effects (Sfigakis et al., 2008a). was observed in the linear conductance. At this low
temperature, the zero-bias anomaly appeared to
1.08.2.6.2 Experimental and theoretical enhance the conductance at Vsd ¼ 0 and wash out
evidence for a Kondo model the 0.7 structure. Nonetheless, as Vsd was moved
The detailed studies of the zero-bias anomaly per- away from zero, two distinct plateaus formed at
formed by Cronenwett et al. (2002) preceded a 0.8(2e2/h) and 0.2(2e2/h), which are believed to result
concrete theory of Kondo-like physics in 1D and from the splitting of spin bands.
spin-density functional theory calculations demon- As the temperature was increased to 600 mK,
strating a stable bound state in QPCs (Meir et al., the zero-bias peak was suppressed as shown in
2002; Hirose et al., 2003; Rejec and Meir, 2006; Meir, Figure 25(b). Now, a clear bunching of lines at
2008). Cronenwett et al. used surface gate depletion 0.7(2e2/h) around Vsd ¼ 0 is observed. This is a
to define a 1D channel in a 2DES and performed strong indication that the formation of the 0.7
source–drain bias measurements on the first three structure coincides with the destruction of the
subbands. Figure 25(a) shows the nonlinear differ- zero-bias anomaly with temperature. The strong
ential conductance as a function of source–drain bias correlation between the zero-bias anomaly and 0.7
for a range of surface gate voltages at base tempera- structure was further exemplified at high-parallel
ture and zero magnetic field. Bunching of the lines magnetic field. Figure 25(c) shows source–drain
along Vsd ¼ 0 indicates the familiar plateaus in the bias measurements performed at base temperature
linear conductance at 2e2/h and 2(2e2/h). These pla- and Bk ¼ 8 T. In this case, the zero-bias peak was
teaus split to form half-plateaus at 1.5(2e2/h) and completely destroyed and the spin-split plateau
(a) T = 80 mK, Bll = 0 T (b) T = 600 mK, Bll = 0 T (c) T = 80 mK, Bll = 8 T
3
5/2
5/2
2 2 2
G (2e2/h)
3/2
3/2
1 1 1 1
0.8
0.7
1/2
1/2
0
–1 0 1 –1 0 1 –1 0 1
Vsd (mV) Vsd (mV) Vsd (mV)
Figure 25 Nonlinear differential conductance as a function of source–drain bias at various surface gate voltages for the first
three subbands of a quantum wire. (a) At base temperature and zero parallel magnetic field, there is a strong zero-bias peak in
the nonlinear conductance below (2e2/h) – the zero-bias anomaly. (b) Raising the temperature suppresses the zero-bias
anomaly and reveals a bunching of lines near 0.7(2e2/h) due to the 0.7 structure. (c) At base temperature and B|| ¼ 8 T, the
zero-bias anomaly has vanished and there is a strong spin-split plateau at 0.5(2e2/h). These results demonstrate the intimate
relationship between the 0.7 structure and the zero-bias anomaly. Modified with permission from Cronenwett SM, Lynch HJ,
Goldhaber-Gordon D, et al. (2002) Low-temperature fate of the 0.7 structure in a point contact: A Kondo-like correlated state
in an open system. Physical Review Letters 88: 226805. Copyright (2002) by the American Physical Society.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 307
at 0.5(2e2/h) is clearly visible. Again, the appear- TK scaled exponentially with gate voltage as shown
ance of the spin-split subband coincides with in Figure 26(c). This is also consistent with TK in
the destruction of the zero-bias anomaly with quantum-dots, which depends exponentially on the
magnetic field. energy of the bound spin ("0) relative to the Fermi
Cronenwett et al. went on to study the tempera- energy in the leads ("0 þ U) according to
ture dependence of the zero-bias anomaly in detail
"0 ð"0 þ U Þ
and showed important similarities between the 1D TK exp
GU
zero-bias anomaly and Kondo physics observed in
quantum-dots (Goldhaber-Gordon et al., 1998; where G is the energy broadening (Haldane, 1978).
Cronenwett et al., 1998; van der Wiel et al., 2000). Moreover, Cronenwett et al. found that the zero-
Figure 26(a) shows the temperature dependence of bias peak splits with magnetic field and that between
zero-bias peaks for five different gate voltages. 0.7(2e2/h) and 2e2/h, the peak widths increase like
Plotting the peak height as a function of tempera- 2kTK/e. Both of these observations are consistent
ture over a range of gate voltages produced the with Kondo physics in quantum-dots. However,
curves shown in the inset of Figure 26(b). more recent experiments have shown that the zero-
Cronenwett et al. found that these curves could bias peak does not always split with magnetic field
and that any observed splitting is significantly less
be scaled onto one function using a single scaling
than predicted (Chen et al., 2009). Sarkozy et al. (2009)
parameter TK as shown in the main panel of
also performed a detailed study of the zero-bias
Figure 26(b). The functional form of the scaled
anomaly as a function of magnetic field and tempera-
curve agrees with a modified expression of the
ture and found further discrepancies between the
Kondo conductance:
behavior of the zero-bias anomaly and 1D Kondo
physics model, namely:
2e 2 1 1
G¼ þ
h 2f ðT =TK Þ 2 1. a nonmonotonic increase of TK with surface gate
voltage;
where f(T/TK) is a universal function for the Kondo 2. the FWHM of the zero-bias peak did not scale
conductance. This provided compelling evidence for with Gmax of the peak; and
a Kondo explanation of the zero-bias anomaly and 3. a linear peak splitting with gate voltage at fixed
hence the 0.7 structure. Moreover, it was found that magnetic field.
G(2e2/h)
0.8
TK (K)
0.5 –466 mV
1.0 1
0.7
G (2e2/h)
0.6
–488 mV
0.5
0.1 1 T (K)
0.5 0 0.1
0
–1 0 1 0.01 0.1 1 10 –492 –462
Vsd (mV) T/Tk Vg
Figure 26 Temperature dependence data of the zero-bias anomaly by (Cronenwett et al. 2002). (a) Temperature
dependence of the zero-bias anomaly at five different surface-gate voltages. As the temperature is increased from 80 to 670
mK, the zero-bias anomaly is destroyed. (b) Zero-bias anomaly peak conductance as a function of temperature for various
gate voltages and scaled onto a single functional form using the scaling parameter TK. The inset shows the unscaled data. (c)
TK as a function of surface gate voltage, Vg, showing an exponential correspondence between TK and Vg. The plot also shows
the linear conductance of the first subband as a function of Vg for 80 mK (solid), 210 mK (dotted), 560 mK (dashed), and 1.6 K
(dot-dashed) showing the emergance of the 0.7 structure as the zero-bias anomaly is suppressed.
308 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
While these discrepancies may suggest that the 1D they provided a good qualitative description of
Kondo model is incomplete, the model is still more many of the important features of the 0.7 structure.
effective at describing the zero-bias anomaly com- Figure 27(a) shows the conductance of the first
pared to a spin-split model, which cannot account for subband as a function of the Fermi energy ("F)
the occurrence of the zero-bias anomaly. relative to the energy of the bound state ("0) calcu-
Guided by the strong experimental evidence, lated at four different temperatures (Meir et al.,
Meir and co-workers developed a comprehensive 2002). As the temperature increases, a distinct pla-
model for the 0.7 structure based on Kondo physics teau forms at 0.7(2e2/h), in excellent agreement with
(Meir et al., 2002; Hirose et al., 2003; Rejec and the experimental data shown in the inset of
Meir, 2006; Meir, 2008). A key prerequisite of the Figure 27(a).
theory is the formation of a quasi-bound state in The calculations also show the formation of a
the QPC. This localized state is generated by strong plateau with increasing parallel magnetic
multiple reflections between the edges of a poten- field as shown in Figure 27(b). While this is in
tial barrier. In the case of a square potential qualitative agreement with experimental data
barrier, a particle with higher energy can be shown in the inset, the position of the calculated
reflected as it enters and exits the barrier. plateau is significantly lower than that observed
Resonant reflections between these two interfaces experimentally. However, Meir et al. (2002) attribute
lead to a quasi-bound state localized within the this quantitative disagreement to the perturbative
barrier region. Interestingly, spin-density functional approach used in their calculations rather than a
theory calculations indicate that this remains true fundamental difference in the underlying physical
even for the smooth potential of a QPC, thereby mechanism.
producing a narrow quasi-bound state that may act The Kondo model also provides a very good
as a single magnetic impurity. description of the evolution of the zero-bias anomaly
Although the calculations by Meir et al. were as a function of temperature and magnetic field as
perturbative and therefore only semi-quantitative, shown in Figure 28(a) and Figure 28(b),
1.0
(a)
Data
1.0
0.5
Conductance (2e2/h)
0.5
VG
1.0
(b)
Data
1.0
0.5
0.5
VG
0.0
0.0 0.2 0.4 0.6 0.8
εF/lε0l
Figure 27 Linear conductance of the first subband as a function of the Fermi energy "F relative to the energy of the
quasi-bound state "0. The calculations by Meir et al. are based on spin-density functional theory calculations of a Kondo
model (Meir, 2002). (a) The calculated conductance at 50, 100, 200 and 600 mK. The results agree both qualitatively and
quantitatively with the experimental observations of Cronenwett et al. shown in the inset (Cronenwett 2002). (b) The calculated
conductance of the first subband in parallel magnetic fields of 0 T, 0.07, 0.12, and 0.4 T (from top to bottom, respectively).
The calculations agree qualitatively with experimental observations shown in the inset, and quantitative discrepancies are
attributed to the perturbative nature of the calculations. Reproduced with permission from Meir Y, Hinose K, and Wingreen
NS (2002) Kondo model for the ‘‘0.7 anomaly’’ in transport through a quantum point. Physical Review Letters 89: 196802
Copyright (2002) by the American Physical Society.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 309
1.0 1.0
(a) (b)
Data
Conductance (2e2/h)
Conductance (2e2/h)
1.0
0.5
V V
0.0 0.0
−0.5 0.0 0.5 −0.5 0.0 0.5 0
Bias voltage V/lε0l
Bias voltage V/lε0l
Figure 28 Calculations of the nonlinear conductance and zero-bias anomaly based on spin-density functional theory
calculation of a Kondo model (Meir, 2002). (a) Calculated zero-bias anomaly at for Fermi energies "F ¼ 0.1, 0.03, and 0.01 from
top to bottom respectively at four different temperatures T ¼ 60, 100, 200, and 400 mK. The calculations agree both
quantitatively and qualitatively by the experimental data taken by Cronenwett, shown in the inset (Cronenwett, 2002).
(b) Zero-bias anomaly near the 0.7 structure for different parallel magnetic fields corresponding to a Zeeman splitting of ¼ 0,
0.04, 0.07, and 0.1, where ¼ 0.04. Again, the data agree with the splitting of the zero-bias anomaly peak observed
experimentally shown in the inset (Cronenwett, 2002).
respectively. Through this work, combined with that perhaps two separate and distinct phenomena
experimental studies, there is growing consensus exist near 0.7(2e2/h) concomitantly. In the next
that 1D systems (1.) can support a quasi-bound section, we review the experiments performed in
state, (2.) do display Kondo-like physics, and p-type 1D systems. The unique properties of the
(3.) the zero-bias anomaly and 0.7 structure are valence band provide a different system to probe
intimately related phenomena based on spin. the 0.7 structure.
However, it does not necessarily follow that the 0.7
structure has its origins in Kondo-like physics. There is
evidence to suggest that while the zero-bias anomaly 1.08.3 Ballistic Transport in p-Type
and 0.7 structure coexist, their underlying mechanisms GaAs-Based 1D Systems
are separate and distinct. Using etched channels
Sfigakis et al. (2008a) formed open quantum-dot struc- 1.08.3.1 Introduction
tures that displayed enhanced Kondo behavior while Studying ballistic transport in p-type 1D systems
retaining many of the features of the 0.7 structure . brings with it the advantages of a larger effective
These 0D–1D hybrid structures allowed Sfigakis et al. mass, strong spin–orbit coupling, and spin 3/2 com-
to study the temperature dependence of the 0.7 struc- pared with spin 1/2 for electrons. The large effective
ture and zero-bias anomaly independently. They hole mass m ¼ 0.2 – 0.4me leads to enhanced many-
pffiffiffiffi
found that the two features appeared at different body interactions as parametrized by rs _ m = ns .
energy scales with the 0.7 structure appearing after While 2D electron systems are limited to rs < 5,
the zero-bias anomaly had been destroyed by increas- hole systems can readily achieve rs >> 10. This lar-
ing temperature. ger value of rs allows a study of the importance of
In summary the origin of the 0.7 structure interaction effects in 1D.
remains one of the most intriguing and challenging The strong spin–orbit coupling makes it possible
puzzles in the field of 1D ballistic transport. Many of to manipulate the hole spin with an applied electric
its properties indicate the presence of spontaneous field. In addition, the total angular momentum j ¼ 3/2
spin polarization – such as its behavior in magnetic gives low-dimensional hole systems a much richer
field and source–drain bias and density depen- spin physics than their electron counterparts
dences. However, other properties such as the (Winkler, 2003). The strong spin–orbit coupling
zero-bias conductance peak and temperature depen- and anisotropic Zeeman splitting in these systems
dence are more commensurate with a Kondo-like also give important insights into the origin of the
model. These conflicting properties suggest that the 0.7 structure and zero-bias anomaly. The study of
physics is more complicated than first thought and hole quantum wires in GaAs is still relatively new
310 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
compared to the broad spectrum of literature on 1D discuss the 0.7 structure and the zero-bias anomaly
electron systems. However, their complex spin prop- in the context of p-type systems.
erties and ability to be tuned with external electric
and magnetic fields allow unique insights into the
spin properties of 1D systems. 1.08.3.2 The Hole Band Structure
Progress in the study of 1D GaAs hole systems has
The structure of the valence band in bulk GaAs
been slower for two main reasons. First, the higher
systems is more complex than the simple parabolic
effective mass in p-type systems results in smaller
conduction band for electrons (see Figure 29). In
subband spacings that require lower temperatures to the conduction band, transport is described well by
resolve clear conductance quantization. Second, and the effective mass approximation due to the fact that
most important, fabricating stable ballistic p-type the conduction band is, to a good approximation,
hole wires has proved to be significantly more chal- parabolic and therefore d2E(k)/dk2 is a constant. In
lenging than is the case for n-type electron wires contrast, the definition of m in the valence band is
(Zailer et al., 1994; Daneshvar et al., 1997; Rokhinson complicated by the fact that the bands are nonpara-
et al., 2006). However, recent developments in device bolic and anisotropic around k ¼ 0 (Winkler, 2003).
fabrication technology and high-purity heterostruc- Furthermore, the curvature of the valence band is
tures have allowed the fabrication of 1D hole systems negative, resulting a negative m. This concept is
that demonstrate clear conductance quantization. nonphysical and so we treat these carriers as
This has led to a rapid development in our under- positively charged particles with a positive mass.
standing of both single-particle and many-body Conduction band electrons originate from s-like
phenomena in ballistic 1D channels. l ¼ 0 states with spin 1/2, with a quantized projection
In this section, we review the importance of the in any direction of ms ¼ h/2. In contrast, there are
hole band structure before going on to discuss how three valence band states. In a tight-binding picture,
different fabrication techniques have allowed pro- these band states are derived from the three
gressively more stable and reliable 1D hole systems p-orbitals px, py, and pz, and therefore have an orbital
in Section 1.08.3.2. In Section 1.08.3.3 we review angular momentum l ¼ 1. The spin–orbit interaction
the observation of ballistic transport in 1D hole lifts the sixfold degeneracy of the valence band states
wires, including measurements of the subband spa- at k ¼ 0 leading to a splitting of the valence band
cing, Zeeman splitting, and measurements of the states to give four degenerate states of total angular
g-factor anisotropy. Finally, in Section 1.08.3.4 we momentum j ¼ 3/2, separated from two degenerate
E E
(a) (b)
Conduction
band Conduction
band
Eg K Eg K
Valance band Valance band
HH HH
Δ0 ±3/2 Δlh-hh ±3/2
j = 3/2
LH
±1/2 LH
±1/2
SO } j = 1/2 SO
Figure 29 (a) Schematic band diagram for bulk GaAs, showing the split-off (SO), light-hole (LH), heavy-hole (HH), and
conduction bands. (b) Schematic band diagram for a two-dimensional (2D) hole system (2D HS) where the HH–LH
degeneracy is lifted at k ¼ 0 due to the 2D confinement.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 311
states of lower energy with j ¼ 1/2. The doubly (Zülicke, 2006). We can consider two limiting case. In
degenerate states are known as the spilt-off band the case of very narrow wires, the mj ¼ 1/2 light-hole
and have m ¼ 0.17 (Kittel, 1996). The fourfold band becomes higher in energy than the mj ¼ 3/2
degenerate states form two band masses, heavy heavy-hole band. The other limit is where the 2D
holes (m ¼ 0.5me) with z-component of the angular confinement from the quantum-well is stronger than
momentum jz ¼ 3/2 and light holes (m ¼ 0.087me) the lateral quantum wire confinement. In this case,
with jz ¼ 1/2. the ground state remains predominantly mj ¼ 3/2
Confining the GaAs system to 2D lifts the fourfold HH-like. This has important implications for how
degeneracy of the heavy hole (HH) and light hole the holes couple with an external magnetic field as
(LH) bands, as depicted Figure 29(b), so that the HH the confinement strength varies.
(mJ ¼ 3/2) band becomes the highest energy hole Figure 30 shows the calculated change in g-factor
band. Two ladders of energy states are formed as a during the 2D-to-1D cross-over for a square wire
result of the different effective masses of the bands parallel to the [233]crystallographic axis. In this fig-
(Winkler and Nesvizhskii, 1996). However, much of ure, Wz/Wx is the ratio of the height of the quantum
this substructure is hidden from magneto-transport wire (Wz) relative to the width Wx. In the 2D limit,
measurements of 2DHSs, which are described the g-factor for the low-symmetry [311] GaAs is also
primarily by transport through only the highest highly anisotropic. The holes are strongly coupled to
energy band – typically HH1. In these systems, the an external magnetic field perpendicular to the [311]
Fermi energy is small (EF 1 meV), significantly plane with gz 7. In contrast, holes coupled to a
less than the light-hole heavy-hole splitting magnetic field in the plane and parallel to the wire
(lhhh 10 meV), so that only the heavy-hole have gy 0.6. The holes are even more weakly
band is occupied. coupled to a magnetic field in the plane and perpen-
Eisenstein et al. (1984) showed that, in 2D asym- dicular to the wire, with gx 0.2.
metric quantum-wells, a lifting of the spin
degeneracy of the HH1 band results in two effective
masses, which appear as two distinct frequencies in 2D limit 1D limit
Shubnikov de Haas oscillations. In contrast, the states 7
g∗z
in symmetric, square quantum-wells remain doubly 6 z
degenerate and only one effective mass is present. 5
Even so, assigning an effective mass to 2D GaAs hole x g∗x
g ∗x,y,z 4
systems is nontrivial due to the fact that m is depen- 3 y
dent on both the symmetry of the device (Eisenstein 2 g∗y
et al., 1984; Iye et al., 1986; Pan et al., 2003) and the
1
density (Pan et al., 2003; Noh, 2005).
0
Historically, transport measurements of 2DHSs 0 0.2 0.4 0.6 0.8 1
have been performed on silicon-doped GaAs- Wz / Wx
AlGaAs heterostructures grown on a (311)A-oriented
GaAs substrate. The hole mobility in these structures Figure 30 Calculated g-factors for a rectangular hole
quantum wire oriented along [233] as a function of one-
depends on the direction of the current path and is at
dimensional (1D) confinement.¯ The g-factors gx and gy
a maximum for currents parallel to the [233] direc- correspond to in-plane magnetic fields orthogonal and
tion (the fast direction). Between this direction and parallel to the wire, respectively, while gz corresponds to
the [011] direction (the slow direction), the mobility magnetic fields perpendicular to the plane. The 1D
may change as much as 20% (Davies, 1991; confinement is characterized by the ratio of the quantum-
well height (Wz) to the width of the wire (Wx). When Wz/Wx ¼
Heremans and Santos, 1994). This anisotropy must
0, the system is in the 2D limit but tends towards the 1D limit
also be taken into account when calculating transport as Wz/Wx ! 1. The calculations predict that increasing the
properties. Recent developments in MBE have 1D confinement will reduce the out-of-plane g-factor while
allowed the growth of high-quality p-type hetero- simultaneously increasing the inplane g-factors, gx and gy.
structures on (100)-oriented GaAs substrates. The 1D confinement also affects the g-factor anisotropy.
Below Wz/Wx ¼ 0.4 spins couple most strongly to fields
However, their 1D transport properties are largely
parallel to the wire. However, when Wz/Wx > 0.4, fields
unexplored. orthogonal to the wire couple more strongly. From Zülicke U
Further confining the valence band to just one (2006) Electronic and spin properties of hole point contacts.
dimension causes a mixing of the HH and LH states Physica Status Solidi (c) 3: 4354.
312 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
The data in Figure 30 show that the g-factors and 1.08.3.3.1 Etched quantum wires
the g-factor anisotropy change as the hole system The first observation of conductance quantization in
approaches the 1D limit. This occurs because the a 1D hole system was made by Zailer et al. (1994).
confinement reduces the separation between the Original data from this work and a schematic of the
HH and LH subbands and allows mixing of these device structure are shown in Figure 31. In these
two bands. Most importantly, we see that when experiments, Zailer et al. used wet-etching to define
Wx > 2Wz, there is a cross-over of the in-plane aniso- two parallel wires in a 2DHS of peak mobility
tropy, that is, gx > gy. This cross-over is a result of 320 000 cm2 V1 s1 at 50 mK. Using electron beam
the LH subband becoming the highest-energy sub- lithography, Zailer et al. were able to etch two 300-
band and therefore the subband that dominates the nm-wide trenches separated by a 1100-nm gap. A
transport properties. It is worth noting that here 300-nm dot was then etched between the two
the traditional nomenclature becomes confused as trenches to divide the gap into two, 400-nm wide
the holes occupying the light-hole subband (charac- 1D channels. By depositing metal into the etched
terized by jz ¼ 1/2), in fact, have the greater regions, Zailer et al. formed self-aligned depletion
effective mass in this regime. gates that could be used to tune the width of each
wire independently. By using the depletion gates to
completely pinch off one channel and vary the width
1.08.3.3 Fabricating Stable 1D Hole of the other channel, Zailer et al. were able to explore
Systems the current quantization through a single channel.
The wet-etch technique has been used success-
The enhanced m in the valence band implies that fully to fabricate n-type quantum wires with very
resolving conductance plateaus in p-type systems is large subband spacings and high stability. However,
intrinsically more difficult than for n-type systems. Zailer et al. found that the current through their
The problem is twofold. First, increasing m reduces p-type devices suffered from random telegraph
the subband energy spacing. As a result, p-type sys-
tems must be measured at much lower temperatures
to remove the effect of thermal broadening, which
8
smears out the conductance plateaus. Second, increas-
ing m reduces the Fermi velocity and, hence, the
B = 0T
mobility of 2DHSs. This makes it more difficult to 6
achieve conduction at low carrier densities, which, in
G (e2/h)
G (e2/h)
used the surface gate-depletion method to define
6
the 1D channel – a method that has also been used B=0
with great success in n-type systems. A schematic of
the device structure and original data from
Daneshvar et al. is shown in Figure 32. Three metal 4
gates were patterned on the surface of the GaAs
heterostructure – two SGs separated by a small dis-
tance and a third gate that runs between them. 2
Applying a positive bias to the SGs locally depletes B⊥ = 0.7T
the 2DHS directly beneath them and forms a 1D
channel. Increasing the bias on the SG constricts
0
and eventually pinches off the 1D channel. The mid- 2.5 2.75
dle gate can be used to vary the density in the 1D Side gate voltage (V)
channel without affecting the width of the channel.
Figure 32 Original data from Daneshvar et al. showing
Similar techniques are used routinely in n-type quantized inflections in the conductance up to 5(2e2/h) in a
systems to define 1D electron systems with high surface gate-depletion quantum wire (Daneshvar, 1997). A
stability and clear conductance quantization (cf. van small magnetic field enhances the ballistic transport
Wees et al., 1988). In contrast, Daneshvar et al. were through the device to produce well-defined and well-
only able to resolve inflections in the conductance up quantized conductance plateau. (Inset) Schematic of the
surface gate-depletion device architecture used by
to 5(2e2/h). The ballistic transport was facilitated by Daneshvar et al. on a single-layer hole system. The two side
the application of an in-plane magnetic field, which gates define and confine the one-dimensional (1D) channel
lengthened the inflections to form well-defined pla- while the middle gate controls the density in the wire.
teaus and provided a good estimate of the 1D g-factor Reproduced with permission from Daneshvar AJ, Ford CJB,
as a function of subband index. Hamilton AR, simnons MY, Pepper M, and Ritchie DA (1997)
Enhanced g factor of a one-dimensional hole gas with
quantized conductance. Physical Review B 55: R 13409.
1.08.3.3.3 Local anodic oxidation Copyright (1997) by the American Physical Society.
In the five years that followed, there was very little
work done in this field. However, Rokhinson et al.
(2002) fabricated a 1D channel that showed clear region formed the source and drain that taper
but irregular conductance plateaus up to 9(2e2/h). A down to the 1D constriction. The 2DHSs either
schematic of the device structure and data from this side of the constriction then acts as in-plane gates.
work are shown in Figure 33. In this device, local Applying a positive bias to these 2DHSs allows the
anodic oxidation was used to form a 1D constriction 1D channel to be confined and even completely
in a 2DHS situated just 35 nm from surface. The pinched off.
local anodic oxidation technique uses a biased However, the 2D system must be kept close to the
atomic force microscope (AFM) tip to locally surface because the oxidation does not penetrate
induce an oxidation reaction between the GaAs deeply into the heterostructure. Due to the close
and native water film. In this way, the AFM tip proximity to the surface, the mobility of the 2DHS
can be used to oxidize lines in the GaAs substrate ( ¼ 500 000 cm2 V1 s1 at 0.3 K) was less than half
down to 10 nm resolution. Rokhinson et al. used the that of Daneshvar et al. Nonetheless, Rokhinson et al.
local anodic oxidation technique to pattern their (2004, 2006) used this technique to fabricate 1D
2DHS into three separate regions. The central hole systems that show well-defined and stable
314 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
10
10
8
G (2e2/h)
8 In-plane
gate
1 DHS
In-plane 6
gate
G (2e2/h)
6 4
4 2
0
2 2.8 3.2 3.6 4.0
Split gate bias (V)
0 Figure 34 Conductance quantization observed in a
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 0.6
bilayer one-dimensional hole system (1DHS) by Danneau
Vg1 = Vg2 (V)
et al. (Inset) Schematic of the surface gate-depletion
Figure 33 Original data from Rokhinson et al. showing architecture used by Danneau et al. to define the quantum
irregular conductance quantization plateaus up to 9(2e2/h). wires in the bilayer 2DHS. This device was fabricated on a
(Inset) Schematic of the device architecture used by metallically doped substrate that could be used as an in situ
Rokhinson et al. Using local anodic oxidation, a shallow back-gate to vary the density in the lower layer. From
two-dimensional hole system (2DHS) is pattern into three Hamilton AR, Danneau R, Klochan O, et al. (2008) The 0.7
separate regions. The middle region forms the one- anomaly in one-dimensional hole quantum wires. Journal of
dimensional (1D) channel while the 2DHSs either side of the Physics: Condensed Matter 20: 164205.
channel act as in-plane gates to control the current through
the channel. From Rokhinson LP, Tsui DC, Pfeiffer LN, and
West KW (2002) AFM local oxidation nanopatterning of a gates, as shown in Figure 34. A combination of
high-mobility shallow 2D hole gas. Superlattices and
Microstructures 32: 99. front- and back-gate biases allows the transport
properties of the top or bottom quantum wire to
be measured independently.
conductance plateaus at 2e2/h and 2(2e2/h) and have A positive bias applied to the two side gates
even used these devices to perform systematic stu- defines the 1D channels in the two quantum-wells.
dies of the 0.7 structure and the zero-bias anomaly. The central top-gate is used to control the density
They have also extended the technique to develop in the upper quantum wire, and the back-gate con-
spin filters, which have revealed important informa- trols the density in the lower quantum wire. Thus,
tion about the origins of the 0.7 structure and are the top quantum wire can be measured in isolation
discussed in Section 1.08.3.4.3. by depleting the lower wire with the back-gate. In
addition, a forth surface gate (not shown) could be
used to stop the current flow in the top quantum
1.08.3.3.4 Bilayer quantum wires wire without it being depleted. This allows the
In 2006, Danneau et al. applied the surface gate- lower quantum wire to be measured without having
depletion method to form two hole wires in a to deplete the upper wire, and can be used to
modulation-doped double quantum-well. In these measure the compressibility of 1D systems
heterostructures, two sheets of high-mobility holes (Danneau et al., 2006a, 2006b).
are formed in 20–nm-wide GaAs quantum-wells Using this bilayer heterostructure, Danneau
separated by a 30-nm AlGaAs barrier. The barrier et al. were able to form highly stable and repro-
is sufficiently thick so that there is no tunneling ducible conductance quantization for the first 11
between the two 2D hole layers. The as-grown subbands and a strong 0.7 structure. Figure 34
hole densities in the two wells were shows the conductance of the quantum wire in
1 1011 cm2
, with peak mobilities in excess of the upper quantum-well as a function of the
106 cm2 V1 s1. Ohmic contacts were made to side-gate bias without any of the resonances that
both 2DHSs in parallel, and an overall back-gate hampered previous studies of lower-mobility hole
was used to control the hole density in the lower devices.
quantum-well. Two parallel quantum wires are The bilayer device design has been particularly
defined along the [233] direction by three surface successful. Danneau et al. subsequently performed
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 315
G (2e2/h)
lit
Sp
the 0.7 structure and the associated zero-bias anom-
4
aly (Danneau et al., 2008).
However, it is still unclear why the bilayer
device design has been so successful compared to 2
single-layer devices that have also used surface
gate depletion to define the 1D channel. Indeed, 0
0.0 0.4 0.8 1.2
the peak mobility of each of the two layers of the Side gate bias (V)
bilayer devices is significantly lower than the
Figure 35 Conductance quantization observed in an
mobility of the single-layer 2DHS of Daneshvar
electrostatically induced 1DHS by Klochan et al. (Inset)
et al. One possibility is that remote ionized dopants Schematic of the one-dimensional (1D) semiconductor–
cause instabilities in the 1D channel. In the devices insulator–semiconductor field effect transistor (SISFET)
of Daneshvar et al., the quantum-well was symme- device architecture. Electron beam lithography was used to
trically doped – with the 2DHS sandwiched pattern the metallically doped capping layer into three
independent surface gates. The middle gate is negatively
between two layers of remote ionized dopants. biased to form the 1D channel while the side gates are used
The bilayer system is also symmetrically doped; to control the 1D confinement. From Hamilton AR, Danneau
however, the top 2DHSs screens the lower R, Klochan O, et al. (2008) The 0.7 anomaly in one-
2DHSs from the upper band of remote ionized dimensional hole quantum wires. Journal of Physics:
dopants and vice versa. This effectively halves the Condensed Matter 20: 164205.
impact that the remote ionized dopants have on
each layer.
1D quantum wire and the 2D reservoirs to the left
It is not possible to rule out that geometric prop-
erties of the wire (such as the dimensions of the 1D and right of it are controlled with a negative bias
channel and the 2D-to-1D transition) may have con- applied to the central top-gate. At the same time, a
tributed to the enhancement of the channel stability. positive bias applied to the two side gates controls
However, recent experiments by Klochan et al. (2006) the effective width of the 1D wire formed 200 nm
have confirmed that by completely removing remote below the gate.
ionized dopants, the stability of the 1D channel is As a result, they were able to fabricate a 1D hole
improved dramatically. system with remarkable stability – even very close
to pinch-off. The device shows seven clean con-
1.08.3.3.5 Electrostatically induced ductance steps, with additional structure below the
quantum wires first plateau. These devices were found to be
Klochan et al. removed the need for dopants by exceptionally stable, without the hysteresis or
electrostatically inducing a 1D hole system in a instabilities that occur in most modulation-doped
GaAs SISFET (Clarke et al., 2006; Klochan et al., devices. Remarkably, Klochan et al. demonstrated
2006) (see Figure 35). This is simply an induced that the current through their 1D channel
single heterojunction in which the holes are intro- remained constant – even for conductances as low
duced by a negative bias applied to a degenerately as 0.3(2e2/h).
doped pþ-GaAs gate electrode, instead of by
modulation doping the AlGaAs layer. This
approach avoids unwanted scattering from remote 1.08.3.3.6 Cleaved-edge overgrowth hole
ionized impurities. The holes are confined in a wires
triangular potential well, with densities from To date, almost all experiments on ballistic 1D hole
1.6 1010 to 1.9 1011 cm2 and mobilities up to systems have been performed on QPCs fabricated in
700 000 cm2 V1 s1. To define a quantum wire, the heterostructures grown on the (311)A GaAs. A
pþ-GaAs layer is divided into three electrically notable exception to this trend are the p-type wires
separate gates using electron beam lithography produced by Pfeiffer et al. (2005), who used cleaved-
and a shallow wet etch. The hole density in the edge overgrowth to fabricate 2-mm-long, p-type
316 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
ce
4 Finger gate urfa
)S
10
(1
Conductance (2e2/h)
QW
3 gexp
2
2 gexp
gexp
0
1.5 2 2.5
Gate voltage (V)
Figure 36 Conductance quantization observed in a 2-mm-long quantum hole wire fabricated using cleaved-edge over-
growth by Pfeiffer et al. In this case the conductance is quantized in units of 0.77(2e2/h) as a result of 1D–2D wave function
mismatch (Peiffer, 2005). (Inset) Schematic of the cleaved-edge overgrowth quantum wire. A quantum-well heterostructure
grown in [100] GaAs is cleaved along the [110] crystallographyic plane. Additional layers of modulation-doped AlGaAs are
grown on the newly cleaved surface, creating an enhancement in the carrier density in the quantum-well along the cleaved
edge. Surface gates are then biased such that the two-dimensional hole system (2DHS) in the quantum-well is depleted but a
2-mm-long 1D channel remains along the cleaved edge. Reproduced with permission from Pfeiffer LN, de Picciotto R, West
KW, Baldwin KW, and Omony CHL (2005) Ballistic hole transport in a quantum wire. Applied Physics Letters 87: 073111, of
American Institute of Physics.
quantum wires from a carbon-doped heterostructure subbands. However, the plateaus were quantized in
grown on a (100) GaAs substrate. units of 0.77(2e2/h) as opposed to 2e2/h. de Picciotto
A schematic of the device and the conductance et al., (2000, 2001) explained this discrepancy as
plateau observed is shown in Figure 36. A high- a competition between 1D–2D wave function
mobility ( ¼ 1 500 000 cm2 V1 s1) 2DHS was mismatch and the inevitable residual disorder along
grown on a (100) GaAs substrate. A series of 13 the ungated regions.
parallel metallic finger gates were then patterned
on the surface of the heterostructure – each gate
being 2 mm wide and separated from each other by 1.08.3.4 Ballistic Transport in Hole
2 mm. The heterostructure was then cleaved through Quantum Wires
the finger gates such that a clean [011] surface was In recent years, the marked improvement in the
formed along the cleaved edge. A second MBE stability of p-type quantum wires and QPCs has
growth step was then used to grow a modulation- allowed single-particle and many-body phenomena
doped AlGaAs heterostructure on the [011] surface – in 1D hole systems to be explored in much greater
perpendicular to the original 2DHS quantum-well. detail.
The additional doping along this edge increases the
electron density in the quantum-well near the 1.08.3.4.1 Source–drain bias spectroscopy
cleaved-edge interface. As a result, applying a posi- One of the defining factors regarding the clarity of
tive bias to a finger gate-depletes the 2DHS directly the conductance plateau is the energy spacing
beneath it, but leaves a 2 -mm-long, 1D conducting between the 1D subbands, which is related to the
channel along the cleaved-edge interface. Increasing strength of the confinement potential. We have seen
the bias on the finger gate constricts the wire and that experiments on n-type wires defined by wet
ultimately pinches off the channel. etching tend to have a very strong confinement
These devices showed well-defined and reprodu- potential (Kristensen et al., 1998a, 2000; Skaberna
cible conductance plateaus for the three lowest et al., 2000; Ramvall et al., 1997; Regul et al., 2002).
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 317
This results in large subband spacings, up to 20 mV. with the observation of very well-defined conduc-
As a result, conductance quantization can be tance plateaus in these bilayer systems.
observed in these devices at temperatures up to tens
of kelvin (Kristensen et al., 1998a, 2000; Skaberna 1.08.3.4.2 Zeeman splitting and the
et al., 2000; Ramvall et al., 1997) before thermal broad- g-factor anisotropy
ening destroys the conductance plateaus. For hole quantum wires, the Zeeman splitting of the
On the other hand, 1D wires defined by surface 1D subbands is highly dependent on the orientation
gate depletion exhibit subband spacings of the order of the magnetic field. For quantum wires fabricated
of 1 or 2 mV and so require much lower measure- on (311)A heterostructures, the interplay of the con-
ment temperatures to resolve the conductance finement potential and the crystal anisotropy implies
plateaus (Daneshvar et al., 1997; Thomas et al., 1995; that the Zeeman splitting is much larger when the
Pyshkin et al., 2000; Cronenwett et al., 2002). Another magnetic field is applied parallel to the quantum wire
distinguishing feature is that the subband spacings in (along [233]), than when it applied perpendicular to
etched QPCs show little variation between energy the quantum wire (along [011]). This is quite differ-
levels (Kristensen et al., 2000; Regul et al., 2002), ent to 1D electrons, where the Zeeman splitting has
whereas the subband spacings in surface-gated sys- been shown to be isotropic and independent of the
tems tend to decrease rapidly for higher energy levels direction of the in-plane field.
(Daneshvar et al., 1997; Thomas et al., 1995; Pyshkin The way in which 1D hole systems couple with an
et al., 2000; Cronenwett et al., 2002). in-plane magnetic field is dependent on both the
To date, there have been only two studies to mea- direction of the magnetic field and the orientation
sure the subband energy spacing in 1D hole systems of the wire relative to the [233] and [011] crystal-
using source–drain bias spectroscopy. One study con- lographic axes. This means that there are four
sidered a single hole wire and the other a bilayer hole possible g-factors that must be considered:
system. Daneshvar et al. (1997) applied the source–
1. Wire k [233]and B k [233];
drain bias spectroscopy technique to their single 1D
2. Wire k [233]and B k [011];
hole system defined by surface gate depletion. The
3. Wire k [011]and B k [233]; and
subband energy spacings measured were observed to
4. Wire k [011]and B k [011].
decrease from 0.27 mV to 0.09 mV for the first five
subbands. These spacings are significantly smaller than As early as 1997, it was shown that the g-factor in a
the subband energy spacings measured in n-type 1D hole systems increases with 1D confinement
devices defined by the same technique. However, this (Daneshvar et al., 1997). However, it was not until
is consistent with the fact that the effective mass of very recently that the g-factor anisotropy was studied
holes is approximately 6 times larger than that of in detail along the two major orthogonal directions.
electrons and therefore we expect the subband energy Danneau et al. (2006a) studied the g-factor aniso-
spacing to be reduced by a factor of 6 for p-type tropy in a quantum wire oriented parallel to the [233]
systems. axis and found that increasing the 1D confinement
Danneau et al. (2006a, 2006b) also measured the increased the g-factor for in-plane magnetic fields
subband spacings of a p-type bilayer QPC using oriented both parallel and perpendicular to the
source–drain bias spectroscopy. They found that in wire. For magnetic fields parallel to the wire, there
the bilayer device, the subband spacing ranged from was a dramatic 40% increase in g-factor as the wire
E1,2 ¼ 0.41 mV to E5,6 ¼ 0.28 mV for the first five was confined to a single subband. For fields perpen-
subbands, with very little variation in the three high- dicular to the wire, the g-factor showed a more
est subband spacings. The larger subband spacings, modest 15–20% increase. Danneau et al. suggested
and the small variation between energy levels in the that the enhanced sensitivity to in-plane magnetic
bilayer devices is more consistent with that observed fields was due to the the 1D confinement causing
in etched QPCs than in surface-gated devices. This the total angular momentum quantization axis to
may indicate a stronger confinement potential set up rotate into the plane and parallel to the wire. As a
by the two QPC gates, the midline gate and the in situ result, the spins couple more strongly to in-plane
back gate. Despite the fact that both devices used magnetic fields and in particular to fields parallel to
surface gate depletion, the source–drain bias spectro- the wire.
scopy revealed that the bilayer system had a much However, experiments by Koduvayur et al. (2008)
stronger confinement potential, which is consistent suggested that the total angular moment quantization
318 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
Bk[233]
Bk[011]
Wire k [233] Nonmonotonic increase Little to no increase (g 0)
Wire k [011] 100% increase Little to no increase in g-factor
axis always rotates toward [233], regardless of the transport measurements, accessing a different subband
orientation of the wire. In these experiments, they requires a change in the confinement potential.
compared the g-factor anisotropy for all four combi- Despite a series of investigations, there is still
nations of wire and magnetic field orientations. Their much to be understood about the g-factor in 1D
results are summarized in Table 2. hole systems, how it changes with a magnetic field,
For a quantum wire oriented along the [233] direc- and how it can be tuned with 1D confinement and
tion Koduvayur et al. observed that for fields parallel to spin–orbit coupling.
the wire, the g-factor increased rapidly as the number
of subbands in the wire decreased from 6 to 4 but then
dropped abruptly for the third subband. Conversely, 1.08.3.5 The 0.7 Structure in 1D Hole
for fields perpendicular to the wire, Koduvayur et al. Systems
recorded g-factors very close to zero that showed no
One of the most direct benefits of improved device
significant change with increasing 1D confinement.
stability in 1D hole systems has been the ability to
This result disagrees with recent theoretical results
resolve and study the 0.7 structure in 1D holes – one
(Zülicke, 2006) that predict that the 2D g-factor
of the most intensely studied many-body phenomena in
along the [011] direction (which has a finite value of n-type 1D systems. One might expect to see a significant
g 0.3) will only increase with 1D confinement. qualitative or quantitative difference between the n-
For a quantum wire oriented along the [011] and p-type 0.7 structures given that holes have much
direction, the observations are consistent with greater effective mass, particle–particle interaction
experiment. For magnetic fields oriented along the strength, spin–orbit coupling, and effective spin-3/2.
[233] direction, the g-factor increases monotonically However, to date, experiments have shown that the
as the number of subbands decreases. In addition, for 0.7 structure in 1D hole systems behaves almost identi-
fields parallel to the [011] direction, Koduvayur et al. cally to the analogous structure in n-type systems.
observed a constant but finite value g ¼ 0.3 – very In particular, the anomalous 0.7 plateau:
close to the g-factor expected in the 2D limit. These
results indicate that even when the wire is oriented 1. appears at a conductance value of 0.7(2e2/h);
along the [011] axis, 1D confinement causes the 2. shows activated behavior, becoming stronger with
quantization axis to align itself along the [233] direc- increasing temperature;
tion, leading Koduvayur et al. to conclude that ‘‘. . .g- 3. typically coincides with a zero-bias anomaly; and
factor anisotropy is primarily determined by the 4. moves toward 0.5(2e2/h) in the presence of an in-
crystalline anisotropy of spin–orbit interactions.’’ plane magnetic field.
This rational may also explain the unusual non- Interestingly, the behavior of the plateau in a mag-
monotonic behavior of the g-factor in the wires netic field shows the same anisotropy as the
oriented along the [233] direction. However, it is not conductance plateaus at higher subbands, providing
clear why this random fluctuation in g-factors has not further evidence that the origin of the 0.7 structure is
been observed previously. Although, it has been pre- fundamentally spin-based. Furthermore, it was found
dicted by Csontos and Zülicke (2007) that spin–orbit that the zero-bias anomaly mirrors the 0.7 structure
coupling may introduce large and apparently random anisotropy – disappearing in an in-plane magnetic
variations in the g-factor for different subbands. These field at the same time the 0.7 structure is reduced to
theoretical predictions of Csontos et al. are based on 0.5(2e2/h). This highlights a close interdependence of
perfectly cylindrical or square-shaped wires as these two phenomena. Recent experiments have used
opposed to the square well and saddle potentials in spin–orbit coupling to probe the level of spin-polar-
practical devices. Moreover, the theory calculates the ization around 0.7(2e2/h). The results from these
g-factors for wires of constant width, while for experiments indicate that the anomalous plateau at
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 319
G (2e2/h)
T = 650 mK
1.0
1.08.3.5.1 Characterizing the 0.7 structure
One of the main differences between electron and 0.5
G (2e2/h)
structure observed in bilayer 1DHSs. These studies
showed that despite rs in the electron and hole 0.5
samples differing by a factor of 5 or more, many of
the zero-field properties of the 0.7 structure were T = 20 mK
T = 320 mK
B=0T
identical and, therefore, were independent of carrier 0.0
–0.5 0.0 0.5 –0.5 0.0 0.5
type. This suggests that the 0.7 structure may be Vsd (mV) Vsd (mV)
more sensitive to the potential landscape and/or
Figure 37 (a) Evolution of the 0.7 structure in a bilayer
the disorder environment than to the strength of the quantum wire as a function of temperature. The data show
interactions. that the 0.7 structure has the same activated behavior as
The temperature dependence of the 0.7 structure observed in n-type systems – becoming more evident as the
was observed to be very similar to that in n-type temperature increases. Conversely, conductance plateaus
for higher subbands become less well defined as a result of
quantum wires. Danneau et al. showed an enhance-
thermal broadening. b) Source–drain bias spectroscopy at
ment of the 0.7(2e2/h) plateau with increasing T ¼ 0 showing the presence of a strong zero-bias anomaly
temperature up to 650 mK (see Figure 37(a)). In below the first subband. c) Increasing the temperature
contrast to the single particle plateaus, the 0.7 struc- destroys the zero-bias peak. From Danneau R, Klochan O,
ture became more prominent as the temperature Clarke WR, et al. (2008) 0.7 structure and zero-bias anomaly
in ballistic hole quantum wires. Physical Review Letters 100:
was increased, and moved closer to 0.5(2e2/h). As 016403.
the temperature was increased further, the strength
of the state weakened once more, but only after all
the single-particle plateaus had been thermally
smeared. 1.08.3.5.2 g-Factor anisotropy of the 0.7
This activated behavior was studied in detail in structure and zero-bias anomaly
n-type systems where it was shown by Cronenwett The similarities between the 0.7 structure observed
et al. (2002) to have similarities to the 0D Kondo in n- and p-type ballistic wires diverge with the
effect. The Kondo model was also used to explain application of an in-plane magnetic field. It is
another peculiar feature of the 0.7 structure – a through this difference that we have gained valuable
peak in the conductance around zero source–drain information about the origin of the 0.7 structure.
bias, also known as the zero-bias anomaly. Earlier, we showed that the Zeeman splitting of the
Danneau et al. confirmed the existence of the 1D subbands in hole quantum wires is highly depen-
zero-bias anomaly in 1D hole systems, as shown dent on the orientation of the magnetic field. This
in Figure 37(b). Moreover, it was shown that in extreme anisotropy of the spin splitting in 1D holes
analogy to n-type systems, the p-type zero-bias can be used to probe the 0.7 structure: if it is related
anomaly was also destroyed with increasing tem- to spin, then it should show an anisotropic response
perature and correlated well with the simultaneous to an in-plane magnetic field as well.
enhancement of the 0.7(2e2/h) conductance plateau. Danneau et al. (2008) used in-plane magnetic
This is evident from the missing zero-bias peak in fields to demonstrate that the 0.7 structure shared
Figure 37(c). the same g-factor anisotropy as that observed in
320 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
higher 1D subbands. The fact that the 0.7 feature However, for in-plane magnetic fields oriented
always moves toward 0.5(2e2/h) at the same rate that perpendicular to the wire (Figures 38(c) and 38(d))
degeneracy of the higher subbands is lifted suggests the case was very different. In this case, the 0.7
that the underlying mechanism is the same for both – structure showed very little change in its absolute
providing strong evidence that the 0.7 structure is a position – even at 3.6 T. Likewise, the higher sub-
spontaneous spin-split level. In these experiments, bands showed no signs of Zeeman splitting at this
Danneau et al. studied the magnetic field dependence field. The fact that the 0.7 structure shares the same
of the 0.7 structure formed in a bilayer 1DHS g-factor anisotropy is an indication that the anoma-
oriented along the [233] crystallographic axis (see lous plateau at 0.7(2e2/h) is the result of same form of
Figures 38(a)–38 (d)). Danneau et al. found that for spontaneous spin polarization.
magnetic fields oriented parallel to the wire, the spins In electron systems, it has been shown that the 0.7
coupled strongly to the external magnetic field. This structure is accompanied by an enhanced conduc-
was evidenced by the fact that the 0.7 structure tance at zero source–drain bias, which falls away
reacted rapidly to the applied field, reaching rapidly as the magnitude of the source–drain bias is
0.5(2e2/h) at approximately 3.6 T. At the same time, increased. The resulting conductance peak at Vsd ¼ 0
the higher subbands also experienced strong Zeeman is a characteristic signature of an anomaly in the
splitting, becoming fully spin split at 3.6 T. density of states at the Fermi energy, and shares
2
BII =
3.6 T
1.0
g (2e2/h)
g (2e2/h)
1
BII = 0.5
0T T = 20 mK
B = 3.6 T
II
0 0.0
3.4 3.6 3.8 4.0 4.2 4.4 4.6 0 1 2 3 4 –0.5 0.0 0.5
Vsg (V) BII (T) Vsd (mV)
(c) (d) (f)
2 B T
=4T
1.0
g (2e2/h)
g (2e2/h)
1
B T 0.5
=0T T = 20 mK
B = 3.6 T
T
0 T
0.0
4.0 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 0 1 2 3 4 –0.5 0.0 0.5
Vsg (V) B (T)
T Vsd (mV)
Figure 38 Conductance data from a bilayer quantum wire oriented along the [2 –33] crystallographic axis. (a) Conductance
quantization and the 0.7 structure as a function of an in-plane magnetic field parallel to the wire B||. (b) Gray scale of the
transconductance in the wire as a function of B||. The dark regions correspond to plateaus in the conductance. These data
show that the higher sub–bands couple strongly to B|| and that the 0.7 structure is also strongly coupled – moving to 0.5G0 at
3.6 T. (c) Conductance quantization and 0.7 structure as a function of in-plane magnetic field orthogonal to the wire, B?. (d)
Transconductance in the wire as a function of B? showing that B? has very little effect on the higher subbands and similarly
has no effect on the position of the 0.7 structure up to 3.6 T. (e) Source–drain bias spectroscopy at B|| ¼ 3.6 T showing that the
zero-bias anomaly has not been destroyed by the in-plane magnetic field. (f) Source–drain bias spectroscopy at B? ¼ 3.6 T
showing that the zero-bias anomaly has not been destroyed by the in-plane magnetic field. Reproduced with permission from
Danneau R, Klochan O, Clarke WR, et al. (2008) 0.7 structure and zero-bias anomaly in ballistic hole quantum wires. Physical
Review Letters 100: 016403. Copyright (2008) by the American Physical Society.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 321
Magnetoresistance (kΩ)
Danneau et al. found that for fields parallel to the focusing
wire, the zero-bias anomaly collapsed by Bk ¼ 3.6 T. 3 4
This field lifts the spin degeneracy of the 1D sub-
bands, and suppresses the zero-bias anomaly. As
shown in Figure 38(e), the destruction of the zero- 2
bias anomaly coincided with the point at which the
0.7 structure reached 0.5(2e2/h). However, if the same 1
magnetic field was applied perpendicular to the
quantum wire, the Zeeman splitting of the integer
1D subbands was much weaker as shown in 0
Figure 38(f), and the zero-bias anomaly was still
present. In summary, the zero magnetic field data
–1
for electron and hole quantum wires are strikingly –1.0 –0.5 0.0 0.5 1.0
similar, with a clear 0.7 structure that becomes stron- B (T)
T
ger with increasing temperature. However in
contrast to electron systems the results for p-type Figure 39 (Inset) AFM micrograph of a p-type spin filter
composed of two p-type QPCs separated by 0.8 mm. The
systems strongly suggest that the destruction of the injector QPCs, with contacts 1 and 2, passes a current of 1
zero-bias anomaly is spin related, and that the zero- nA while the voltage is measured across the detector QPC
bias anomaly and 0.7 structure are intimately related. using contacts 3 and 4. The main panel shows the magne-
toresistance measured across the detector QPC as a
1.08.3.5.3 Spin polarization near the 0.7 function of magnetic field perpendicular to the plane B?. For
B? < 0 holes passing through the injector QPC are focused
structure away from the detector QPC and as a result the magne-
The fact that the 0.7 structure moves toward toresistance is determined primarily by Subnikov-de-Haas
G ¼ 0.5(2e2/h) in a parallel magnetic field strongly oscillations in the two-dimensional hole system (2DHS).
suggests that the origin of the anomalous plateau is However, for B? > 0, holes are focused into the detector
a spin-split state due to a spontaneous spin polariza- QPC, resulting in peaks in the voltage across the QPC. The
first and sharpest peak is comprised of a doublet of peaks
tion at B ¼ 0. However as discussed for 1D electron separated by 36 mT – showing that the different spins in the
wires in Section 1.08.2.5, the origin of the 0.7 struc- subbands are indeed resolved and that these devices can
ture remains highly debated. be used as spin filters. Reproduced with permission from
A further experiment in 1D hole systems that has Rokhinson LP, Larkina V, Lyanda YB, Pfeiffer LN, and West
provided important information on the 0.7 structure KW (2004) Spin separating in cyclotron motion. Physical
Review Letters 93: 146601. Copyright (2004) by the
was performed by Rokhinson et al. (2004). In this American Physical Society.
work two quantum point contacts were separated
by a small distance, d, as show in the inset of
Figure 39. A magnetic field was applied perpendi- analogous to turning on spin–orbit coupling in
cular to the plane and used to focus the holes injected 2DHSs by asymmetrizing the quantum-well confine-
through the first QPC into the second QPC. When ment potential.
the magnetic radius of the holes was d/n2 (where Rokhinson et al. found that as the conductance
n ¼ 1,2,3, . . .) there was a peak in the voltage mea- through the injector QPC approached 0.7(2e2/h),
sured across the second QPC. However, as shown in the height of the second peak collapsed as shown
Figure 39, the first peak is in fact a doublet, corre- in Figure 40. This result indicates that the injector
sponding to spin states that have been split by spin– current is predominately spin polarized near 0.7(2e2/h)
orbit coupling. The doublet is most prominent (Rokhinson et al., 2004). Interestingly, in 1D hole sam-
when both QPCs are gated to pass exactly one ples that did not show a well-defined 0.7 structure, the
spin-degenerate mode. The level of splitting can doublet remained intact below 0.3(2e2/h).
also be tuned by changing the asymmetry of the In summary we have reviewed the 0.7 structure
injecting QPC’s confinement potential. This is and zero-bias anomaly in high-quality hole quantum
322 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
V (μV)
0.96 heterostructures. Since the first realization of ballistic
0.97
0.92 transport in GaAs quantum wires by van Wees et al.
0.5 0.73
–2 0.66
and Wharam et al., the field has blossomed over the
0.68 past two decades as new experiments, and in parti-
0.43
0.26 cular, the ability to create purer and smaller systems
0.0 has exceeded new length scales, allowing new device
0.10 0.15 0.20 0.1 0.2 0.3
architectures to be probed.
Vg (v) B (T)
T
Nonetheless, there remains important questions to
Figure 40 (a) Conductance through the injector quantum be answered regarding the nature of transport
point contact (QPC) below the first subband as a function of through these devices, in particular the role of spin
gate voltage showing a well-defined 0.7 structure. The
vertical lines indicate the points at which spin-polarized
and interactions at very low densities. Most notably is
measurements were performed. (b) Voltage drop across the the 0.7 structure, whose origin is still a point of
detector QPC as a function of magnetic field for different contention. Although many experiments suggest
injector conductances. The data show that when the that the anomalous plateau originates from a sponta-
conductance through the injector is equal to (2e2/h), the
neous spin splitting at zero magnetic field, others
voltage across the detector QPC registers two voltage
peaks indicating two closely spaced spin states whose suggest that Kondo-like physics may govern the
degeneracy has been lifted by spin–orbit coupling. underlying mechanism. It appears that at the date of
However, when the conductance in the injector QPC is submission neither mechanism can explain all the
reduced to 0.7(2e2/h), the detector QPC registers only a experimental observations. With the ability to fabri-
single voltage peak indicating a single spin state and hence
a spin-polarized current. The upper solid and dashed lines
cate both ballistic electron and hole systems with
compare these two cases directly. Reproduced with different spin states and phonon coupling, there is
permission from Rokhinson LP, Pfeiffer LN, and West KW still a lot more to learn from 1D systems.
(2006) Spontaneous spin polarization in quantum point Perhaps most importantly, the techniques used to
contacts. Physical Review Letters 96: 156602. Copyright
study ballistic transport in 1D have provided a solid
(2006) by the American Physical Society.
foundation for understanding other systems. For
example, the work on 1D systems in GaAs has
wires. Despite the fact that interaction effects should opened the door to studies of 0D structures allowing
be much stronger in these systems, because of the the fabrication of artificial atoms and coupled
enhanced hole effective mass, the zero magnetic quantum-dots. These fields are now expanding to
field data for electron and hole quantum wires are include the study of quantum-dots as a possible
strikingly similar. However, unlike 1D electron sys- route to a solid-state quantum computer. The work
tems the Zeeman splitting caused by an in-plane has also contributed to an increased understanding of
magnetic field is highly anisotropic for 1D holes, due quantum transport in other material systems such as
to the strong spin–orbit coupling. Both the 0.7 struc- nanowires and carbon nanotubes. (See Chapter 2.03).
ture and the zero-bias anomaly share this anisotropic
response to an in-plane magnetic field demonstrating
that they are linked, and spin related. Finally, ballistic References
hole spin filters were used to study the spin polariza-
tion around G ¼ 0.7(2e2/h). These devices were able to Adachi S (1985) GaAs, AlAs, and AlxGa1 xAs – material
parameters for use in research and device applications.
resolve two spin states in the 1D subbands. However, Journal of Applied Physics 58: R1.
near the 0.7(2e2/h) plateau, only one spin state was Ando T, Fowler AB, and Stern F (1992) Electronic properties of
resolved – indicating that the lowest subband is in two-dimensional systems. Reviews of Modern Physics 54: 437.
Appleyard NJ, Nicholls JT, Pepper M, Tribe WR, Simmons MY,
fact spin polarized at B ¼ 0. More work is needed to and Ritchie DA (2000) Direction-resolved transport and
understand the electronic and spin properties of holes possible many-body effects in one-dimensional
confined to 1D systems, but already it is clear that 1D thermopower. Physical Review B 62: R16275.
Appleyard NJ, Nicholls JT, Simmons MY, Tribe WR, and
holes have unusual spin properties with no counter- Pepper M (1998) Thermometer for the 2D electron gas using
part in spin-1/2 electron systems. 1D thermopower. Physical Review Letters 81: 3491.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 323
Auslaender OM, Steinberg H, Yacoby A, et al. (2005) Spin- de Picciotto R, Stormer HL, Yacoby A, Pfeiffer LN, Baldwin KW,
charge separation and localization in one dimension. and West KW (2000) 2D–1D coupling in cleaved edge
Science 308: 88. overgrowth. Physical Review Letters 85: 1730.
Bagraev NT, Buravlev AD, Klyachkin LE, et al. (2002) Quantized DiCarlo L, Zhang Y, McClure DT, et al. (2006) Shot-noise
conductance in silicon quantum wires. Semiconductors signatures of 0.7 structure and spin in a quantum point
36: 439. contact. Physical Review Letters 97: 036810.
Berggren KF and Yakimenko II (2002) Effects of exchange and Dwir B, Kaufamn D, Berk Y, Rudra A, Paleski A, and Kapon E
electron correlation on conductance and nanomagnetism in (1999) Electron transport in GaAs/AlGaAs V-groove quantum
ballistic semiconductor quantum point contacts. Physical wires. Physica B 259–261: 1025.
Review B 66: 085323. Eisenstein JP, Störmer HL, Narayanamurti V, Gossard AC, and
Bruus H, Cheianov VV, and Flensberg K (2001) The anomalous Wiegmann W (1984) Effect of inversion symmetry on the
0.5 and 0.7 conductance plateaus in quantum point band structure of semiconductor hetero-structures. Physical
contacts. Physica E 10: 97. Review Letters 53: 2579.
Büttiker M (1990) Quantised transmission of a saddle point Flambaum VV and Kuchiev MY (2000) Possible mechanism of
constriction. Physical Review B 41: 7906. the fractional-conductance quantization in a one-
Chen T-M, Graham AC, Pepper M, Farrer I, and Ritchie DA dimensional constriction. Physical Review B 61: R7869.
(2009) Non-Kondo zero-bias anomaly in quantum wires. Fowler AB, Harstein A, and Webb RA (1982) Conductance in
Physical Review B 79: 153303. restricted dimensionality accumulation layers. Physical
Chiatta O, Nicholls JT, Proskuryakov Y, Lumpkin N, Farrer I, and Review Letters 48: 196.
Ritchie DA (2006) Quantum thermal conductance of Gold A and Calmels L (1996) Valley- and spin-occupancy
electrons in a one-dimensional wire. Physical Review Letters instability in the quasi-one-dimensional electron gas.
97: 056601. Philosophical Magazine Letters 74: 33.
Chou HY, Luscher S, Goldhaber-Gordon D, et al. (2005) High- Glazman LI and Khaetskii AV (1989) Non-linear quantum
quality quantum point contacts in GaN/AlGaN hetero- conductance of a lateral microconstraint in a hetero-
structures. Applied Physics Letters 86: 073108. structure. Europhysics Letters 9: 263.
Clarke WR, Micolich AP, Hamilton AR, Simmons MY, Muraki K, Glazman LI and Raikh ME (1988) Resonant Kondo transparency
and Hirayama Y (2006) Fabrication of induced two- of a barrier with quasilocal impurity states. JETP Letters
dimensional hole systems on (311)A GaAs. Journal of 47: 452.
Applied Physics 99: 023707. Goldhaber-Gordon D, Shtrikman H, Mahalu D, Abusch-
Cronenwett SM, Lynch HJ, Goldhaber-Gordon D, et al. (2002) Magder D, Meirav U, and Kastner MA (1998) Kondo effect in
Low-temperature fate of the 0.7 structure in a point contact: a single electron transistor. Nature 391: 156.
A Kondo-like correlated state in an open system. Physical Graham AC, Pepper M, Simmons MY, and Ritchie DA (2005)
Review Letters 88: 226805. Anomalous spin-dependent behavior of one-dimensional
Cronenwett SM, Oosterkamp TH, and Kouwenhoven LP subbands. Physical Review B 72: 193305.
(1998) A tunable Kondo effect in quantum dots. Science Graham AC, Pepper M, Simmons MY, and Ritchie DA (2006)
281: 540. New interaction effects in quantum point contacts at high
Csontos D and Zülicke U (2007) Large variations in the hole spin magnetic fields. Physica E 34: 588.
splitting of quantum-wire subband edges. Physical Review B Graham AC, Sawkey DL, Pepper M, Simmons MY, and
76: 073313. Ritchie DA (2007) Energy level pinning and the 0.7 spin state
Daneshvar AJ, Ford CJB, Hamilton AR, Simmons MY, in one dimension: GaAs quantum wires studied using finite
Pepper M, and Ritchie DA (1997) Enhanced g factor of a one- bias spectroscopy. Physical Review B 75: 035331.
dimensional hole gas with quantized conductance. Physical Graham AC, Thomas KJ, Pepper M, et al. (2004) 0.7
Review B 55: R13409. Analogue structures and exchange interactions in
Danneau R, Clarke WR, Klochan O, et al. (2006a) Zeeman quantum wires. Solid State Communications 131: 591.
splitting in ballistic hole quantum wires. Physical Review Graham AC, Thomas KJ, Pepper M, Cooper NR, Simmons MY,
Letters 97: 026403. and Ritchie DA (2003) Interaction effects at crossings of
Danneau R, Clarke WR, Klochan O, et al. (2006b) Conductance spin-polarized one-dimensional subbands. Physical Review
quantization and the 0.7 2e2/h conductance anomaly in Letters 91: 136404.
one-dimensional hole systems. Applied Physics Letters Haldane FDM (1978) Scaling theory of the asymmetric
88: 012107. Anderson model. Physical Review Letters 40: 416.
Danneau R, Klochan O, Clarke WR, et al. (2008) 0.7 structure Hamilton AR, Danneau R, Klochan O, et al. (2008) The 0.7
and zero-bias anomaly in ballistic hole quantum wires. anomaly in one-dimensional hole quantum wires. Journal of
Physical Review Letters 100: 016403. Physics: Condensed Matter 20: 164205.
Davies AG (1991) The Fractional Quantum Hall effect in High Heremans JJ and Santos MB (1994) Transverse magnetic
Mobilitity Two-dimensional Hole Gases. PhD Thesis, focusing and the dispersion of GaAs 2D holes at (311)A.
University of Cambridge. Surface Science 305: 349.
de Picciotto R, Pfeiffer LN, Baldwin KW, and West KW (2004) Hirose K, Meir Y, and Wingreen NS (2003) Local moment
Nonlinear response of a clean one-dimensional wire. formation in quantum point contacts. Physical Review
Physical Review Letters 92: 036801. Letters 90: 026804.
de Picciotto R, Pfeiffer LN, Baldwin KW, and West KW (2005) Imry J (1997) Introduction to Mesoscopic Physics, p. 94.
Temperature-dependent 0.7 structure in the conductance of Oxford: Oxford University Press.
cleaved-edge-overgrowth one-dimensional wires. Physical Iye Y, Mendez EE, Wang WI, and Esaki L (1986)
Review B 72: 033319. Magnetotransport properties and subband structure of the
de Picciotto R, Pfeiffer LN, Baldwin KW, and West KW (2008) two-dimensional hole gas in GaAs–Ga1xAlxAs hetero-
The 0.7 structure in cleaved-edge-overgrowth wires. Journal structures. Physical Review B 33: 5854.
of Physics C 20: 164204. Kane BE, Facer GR, Dzurak AS, et al. (1998) Quantized
de Picciotto R, Stormer HL, Pfeiffer LN, Baldwin KW, and conductance in quantum wires with gate-controlled
West KW (2001) Four-terminal resistance of a ballistic width and electron density. Applied Physics Letters
quantum wire. Nature 411: 51. 72: 3506.
324 Ballistic Transport in 1D GaAs/AlGaAs Heterostructures
Kaufman D, Berk Y, Dwir B, Rudra A, Palevski A, and Kapon E Pepper M (1978) Magnetic localisation in silicon inversion
(1999) Conductance quantization in V-groove quantum layers. Philosophical Magazine B 37: 187.
wires. Physical Review B 59: R10433. Pfeiffer LN, de Picciotto R, West KW, Baldwin KW, and
Kittel C (1996) Introduction to Solid State Physics. New York, Quay CHL (2005) Ballistic hole transport in a quantum wire.
NY: Wiley. Applied Physics Letters 87: 073111.
Klochan O, Clarke WR, Danneau R, et al. (2006) Ballistic Pyshkin KS, Ford CJB, Harrell RH, Pepper M, Linfield EH, and
transport in induced one-dimensional hole systems. Applied Ritchie DA (2000) Spin splitting of one-dimensional
Physics Letters 89: 092105. subbands in high quality quantum wires at zero magnetic
Koduvayur SP, Rokhinson LP, Tsui DC, Pfeiffer LN, and field. Physical Review B 62: 15842.
West KW (2008) Anisotropic modification of the effective Ramsak A and Jefferson JH (2005) Shot noise reduction in
hole g-factor by electrostatic confinement. Physical Review quantum wires with the 0.7 structure. Physical Review B
Letters 100: 126401. 16: 161311.
Kothari H, Ramamoorthy A, Akis R, et al. (2008) Linear and Ramvall P, Carlsson N, Maximov I, et al. (1997) Quantized
non-linear conductance of ballistic quantum wires with conductance in a heterostructurally defined GaAs0.25In0.75As/
hybrid confinement. Journal of Applied Physics InP quantum wire. Applied Physics Letters 71: 918.
103: 013701. Regul J, Keyser UF, Paesler M, et al. (2002) Fabrication of
Kristensen A, Bruus H, Hensen AE, et al. (2000) Bias and quantum point contacts by engraving GaAs/AlGaAs
temperature dependence of the 0.7 conductance anomaly in hetero-structures with a diamond tip. Applied Physics
quantum point contacts. Physical Review B 62: 10950. Letters 81: 2023.
Kristensen A, Jensen JB, Zaffalon M, et al. (1998a) Reilly DJ (2005) Phenomenological model for the 0.7
Conductance quantization above 30 K in GaAlAs shallow conductance feature in quantum wires. Physical Review B
quantum point contacts smoothly joined to the background 72: 033309.
2DEG. Journal of Applied Physics 83: 607. Reilly DJ, Buehler TM, O’Brian JL, et al. (2002) Density
Kristensen A, Jensen JB, Zaffalon M, et al. (1998b) Temperature dependent spin-polarization in ultra-low-disorder quantum
dependence of the ‘‘0.7’’ 2e2/h quasi-plateau in strongly wires. Physical Review Letters 89: 246801.
confined quantum point contacts. Physica B 251: 180. Reilly DJ, Facer GR, Dzurak AS, et al. (2001) Many-body spin
Landauer R (1957) Spatial variation of current and fields due to related phenomena in ultra-low disorder quantum wires.
localized scatterers in metallic conduction. IBM Journal of Physical Review B 63: 121311.
Research and Development 1: 223. Reimann SM, Koskinen M, and Manninen M (1999) End states
Lassl A, Schlagneck P, and Richter K (2007) Effects of short- due to spin-Peierls transition in quantum wires. Physical
range interactions on transport through quantum point Review B 59: 1613.
contacts: A numerical approach. Physical Review B Rejec T and Meir Y (2006) Magnetic impurity formation in
75: 0453456. quantum point contacts. Nature 442: 900.
Liang CT, Simmons MY, Smith CG, Kim GH, Ritchie DA, and Rejec T, Ramsak A, and Jefferson JA (2000) Conductance
Pepper M (1999) Spin-dependent transport in a clean anomalies in quantum wires. Journal of Physics C 12: L233.
one-dimensional channel. Physical Review B 60: 10687. Roche P, Segala J, Glattli DC, et al. (2004) Fano factor reduction
Liang C-T, Simmons MY, Smith CG, Kim GH, Ritchie DA, and on the 0.7 conductance structure of a ballistic one-
Pepper M (2000) Spin-dependent transport in clean one- dimensional wire. Physical Review Letters 93: 116602.
dimensional channel. Physical Review B 60: 10687. Rokhinson LP, Larkina V, Lyanda YB, Pfeiffer LN, and West KW
Lieb E and Mattis D (1962) Theory of ferromagnetism and (2004) Spin separating in cyclotron motion. Physical Review
ordering of electronic energy levels. Physical Review Letters 93: 146601.
125: 164. Rokhinson LP, Pfeiffer LN, and West KW (2006) Spontaneous
Lindelof PE (2001) Effect on conductance of an isomer state in a spin polarization in quantum point contacts. Physical Review
quantum point contact. Optical Organic and Inorganic Letters 96: 156602.
Materials 4415: 77. Rokhinson LP, Tsui DC, Pfeiffer LN, and West KW (2002)
Martin-Moreno L, Nicholls JT, Patel NK, and Pepper M (1992) AFM local oxidation nanopatterning of a high mobility
Nonlinear conductance of a saddle-point constriction. shallow 2D hole gas. Superlattices and Microstructures
Journal of Physics – Condensed Matter 4: 1323. 32: 99.
Meir Y (2008) The theory of the ‘0.7 anomaly’ in quantum point Sarkozy S, Sfigakis F, Das Gupta K, et al. (2009) Zero-bias
contacts. Journal of Physics: Condensed Matter 20: 164208. anomaly in quantum wires. Physical Review B
Meir Y, Hirose K, and Wingreen NS (2002) Kondo model for the 79: 161307(R).
‘‘0.7 anomaly’’ in transport through a quantum point contact. Scherer A, Roukes ML, Craighead HG, Ruthen RM, Beebe ED,
Physical Review Letters 89: 196802. and Harbison JP (1987) Ultra-narrow conducting channels
Mott NF (1963) The Theory of the Properties of Metals and formed in AlGaAs/GaAs hetero-structures by low energy.
Alloys. Oxford: Clarendon. Applied Physics Letters 51: 233.
Ng TK and Lee PA (1988) On-site repulsion and resonant Schmeltzer D, Saxena A, Bishop AR, and Smith DL (2005)
tunnelling. Physical Review Letters 61: 1768. Electron transmission through a short interacting wire: 0.7
Noh H (2005) Effective mass of dilute two-dimensional holes in a conductance anomaly. Physical Review B 71: 045429.
GaAs hetero-structure. Journal of the Korean Physical Seelig G and Matveev KA (2003) Electron-phonon scattering in
Society 47: 272. quantum point contacts. Physical Review Letters
Pan W, Lai K, Bayrakci SP, et al. (2003) Cyclotron resonance at 90: 176804.
microwave frequencies in two-dimensional hole systems in Sfigakis F (2005) Electron Transport in Etched GaAs/AlGaAs
AlGaAs/GaAs quantum wells. Applied Physics Letters 83: 3519. quantum Wires. PhD Thesis, University of Cambridge.
Patel NK, Nicholls JT, Martin-Moreno L, et al. (1991a) Properties Sfigakis F, Ford CJB, Pepper M, Kataoka M, Ritchie DA, and
of a ballistic quasi-one-dimensional constriction in a parallel Simmons MY (2008a) Kondo effect from a tunable bound state
high magnetic field. Physical Review B 44: R10973. within a quantum wire. Physical Review Letters 100: 026807.
Patel NK, Nicholls JT, Martin-Moreno L, et al. (1991b) Evolution of Sfigakis F, Graham AC, Thomas KJ, Pepper M, Ford CJB, and
half plateaus as a function of electric field in a ballistic quasi- Ritchie DA (2008b) Spin effects in one dimensional systems.
one-dimensional constriction. Physical Review B 44: 13549. Journal of Physics C 20: 164213.
Ballistic Transport in 1D GaAs/AlGaAs Heterostructures 325
Simmonds PJ, Sfigakis F, Beere HE, et al. (2008) Quantum Yoon Y, Mourokh L, Morimoto T, et al. (2007) Probing the
transport in In0.75Ga0.25As quantum wires. Applied Physics microscopic structure of bound states in quantum point
Letters 92: 152108. contacts. Physical Review Letters 99: 136805.
Skaberna S, Versen M, Klehn B, Kunze U, Reuter D, and Zailer I, Frost JEF, Ford CJB, et al. (1994) Phase coherence,
Wieck AD (2000) Fabrication of a quantum point contact by interference and conductance quantization in a confined
dynamic plowing technique and wet chemical etching. two-dimensional hole gas. Physical Review B 49: 5101.
Ultramicroscopy 82: 153. Zülicke U (2006) Electronic and spin properties of hole point
Spivak B and Zhou F (2000) Ferromagnetic correlations in contacts. Physica Status Solidi (c) 3: 4354.
quasi-one-dimensional conducting channels. Physical
Review B 61: 16730.
Starikov AA, Yakimenko II, and Berggren KF (2003) Scenario for
the 0.7-conductance anomaly in quantum point contacts.
Physical Review B 67: 235319. Further Reading
Sushkov OP (2003) Restricted and unrestricted Hartree–Fock
calculations of conductance for a quantum point contact. Ashcroft NW and Mermin ND (1976) Solid State Physics, p. 319.
Physical Review B 67: 195318. London: Thomson Learning.
Thomas KJ, Nicholls JT, Appleyard NJ, et al. (1998a) Interaction Bird JP and Ochiai Y (2004) Electron spin polarization in
effects in a one dimensional constriction. Physical Review B nanoscale constrictions. Science 303: 1621.
58: 4846. Casey HC and Panish MB (1978) Heterojunction Lasers, p. 191.
Thomas KJ, Nicholls JT, Pepper M, Tribe WR, Simmons MY, New York, NY: Academic Press.
and Ritchie DA (2000) Spin properties of low density Davies AG (2000) Quantum electronics: The physics and
one-dimensional wires. Physical Review B 61: R13365. technology of low-dimensional electronic systems into the
Thomas KJ, Nicholls JT, Simmons MY, Pepper M, Mace DR, and new millennium. Philosophical Transactions of the Royal
Ritchie DA (1996) Possible spin polarization in a Society A 358 151–172.
one-dimensional electron gas. Physical Review Letters 77: 135. Davies JH (1998) The Physics of Low Dimensional
Thomas KJ, Nicholls JT, Simmons MY, Pepper M, Mace DR, and Semiconductors. New York, NY: Cambridge University
Ritchie DA (1998b) Non-linear transport in single mode one Press.
dimensional electron gas. Philosophical Magazine B 77: 1213. Datta S (1995) Electronic Transport in Mesoscopic Systems.
Thomas KJ, Simmons MY, Nicholls JT, Mace DR, Pepper M, Cambridge: Cambridge University Press.
and Ritchie DA (1995) Ballistic transport in one-dimensional Ferry DK and Goodnick SM (1997) Transport in Nanostructures.
constrictions formed in deep two-dimensional electron Cambridge: Cambridge University Press.
gases. Applied Physics Letters 67: 109. Heinzel T (2003) Mesoscopic Electronics in solid state
Thornton TJ, Pepper M, Ahmed H, Andrews D, and Davis GJ nanostructures. Weinheim: Wiley-VCH.
(1986) One dimensional conduction in the two dimensional Ihn T (2004) Electronic Quantum Transport in Mesoscopic
electron gas of a GaAs–AlGaAs heterojunction. Physical Semiconductor Structures. New York, NY: Springer.
Review Letters 56: 1198. Imry Y (1997) Introduction to Mesoscopic Physics. New York,
van der Wiel WG, de Franceschi S, Fujisawa T, Elzerman JM, NY: Oxford University Press.
Tarucha S, and Kouwenhoven LP (2000) Kondo effect in the Kelly MK (1995) Low dimensional physics. In: Low Dimensional
unitary limit. Science 289: 2105. Semiconductors, ch. 4, pp. 76–101. New York, NY: Oxford
van Wees BJ, van Houten H, Beenakker CWJ, et al. (1988) Science Publications.
Quantized conductance of point contacts in a two- Morimoto T, Iwase Y, Aoki N, et al. (2003) Nonlocal resonant
dimensional electron gas. Physical Review Letters 60: 848. interaction between coupled quantum wires. Applied
Wang CK and Berggren KF (1996) Spin splitting of subbands in Physics Letters 82: 2952.
quasi-one-dimensional electron quantum channels. Physical Puller VI, Mourokh LG, Shailos A, and Bird JP (2004) Detection
Review B 54: 14257. of local-moment formation using the resonant interaction
Wharam DA, Thornton TJ, Newbury R, et al. (1988) One- between coupled quantum wires. Physical Review Letters
dimensional transport and the quantisation of the ballistic 92: 096802.
resistance. Journal of Physics C: Solid State Physics 21: L209. Sfigakis F, Graham AC, Thomas KJ, Pepper M, Ford CJB, and
Winkler R (2003) Spin–Orbit Coupling Effects in Two- Ritchie DA (2008) Spin Effects in One Dimensional Systems.
Dimensional Electron and Hole Systems. Berlin: Springer. Journal of Physics C 20: 164213.
Winkler R and Nesvizhskii AI (1996) Anisotropic hole subband Smith CG (1996) Low dimensional quantum devices. Reports in
states and interband optical absorption in [mmn]-oriented the Progress of Physics 59: 235–282.
quantum wells. Physical Review B 53: 9984. Thomas KJ (1997) Transport Properties of High Mobility One
Wirtz R, Newbury R, Nicholls JT, Tribe WR, Simmons MY, and Dimensional Electron Gases. PhD Thesis, Cambridge.
Pepper M (2002) Tuning the electron transport properties of Wang CK and Berggren KF (1998) Local spin polarization in
a one-dimensional constriction using hydrostatic pressure. ballistic quantum point contacts. Physical Review B
Physical Review B 65: 233316. 57: 4552.
Yacoby A, Stormer HL, Wingreen NS, Pfeiffer LN, Baldwin KW, Yacoby A, Auslaender OM, Steinberg H, et al. (2006) Tunneling
and West KW (1996) Nonuniversal conductance quantization spectroscopy of quantum wires: Spin-charge separation
in quantum wires. Physical Review Letters 77: 4612. and localization. Physica Status Solidi (b) 243: 3593.
1.09 Thermal Conductivity and Thermoelectric Power
of Semiconductors
I Terasaki, Nagoya University, Nagoya, Japan
ª 2011 Elsevier B.V. All rights reserved.
326
Thermal Conductivity and Thermoelectric Power of Semiconductors 327
the thermal current and the electrical current. These of the gradient of the chemical potential r and the
are known as the thermoelectric phenomena. By gradient of the inverse temperature Ñð1=T Þ:
making full use of the thermoelectric phenomena in
1 1
semiconductors, we are able to convert electric j ¼ L11 Ñ þ L12 Ñ ð1Þ
T T
power into heat, and vice versa. Such a technology
is called thermoelectrics. 1 1
q ¼ L21 Ñ þ L22 Ñ ð2Þ
In this chapter, we briefly review the thermal T T
conductivity and thermoelectricity of semiconduc- where Lij’ s are called the transport coefficients (the
tors. This subject has more than a 100-year history Onsager coefficients). The chemical potential con-
since Drude established his kinetic theory of electron sists of an electrostatic part e ¼ eV and a chemical
gas in 1900. Partly because of the space constraint, we part c, wher e (<0) is the electron charge. Then, the
focus on recent developments and show only electric field E is given by
minimal set of old data (selected with the author’s
1
biased view). Obviously, the data shown here are far E ¼ – ÑV ¼ – Ñð – c Þ ð3Þ
e
from complete, particularly the old ones. Regretably,
it is almost impossible to cite all the preceding studies However, Ñc cannot be observed separately in
in order from an enormous numbers of papers. Those real experiments, and is considered to be included
who wish to know this area more in detail can refer to in the observed E hereafter (Ashcroft and Mermin,
many textbooks and reviews available. A good 1976).
summary of Drude’s theory is given by Ashcroft Boltzmann has given a formula to calculate the
and Mermin (1976), from which one can grasp gross distribution function f(k, r) of a particle subject to
features of the electron transport. Callen (1985) has external fields. f(k, r) should change from its
explained a basic treatment of nonequilibrium equilibrium value f0, but the deviation will be
thermodynamics in his unique framework of thermal small enough, if the external fields are weak.
physics. A standard theory of the transport phenom- In this weak limit, f f0 can be expanded in
ena in solids is concisely described by Ziman (1976). terms of external fields. To include the non-
Thermal conductivity of new complex materials is equilibrium state, Boltzmann has introduced the
extensively reviewed by Tritt (2004), and the current scattering term describing the change of f(k, r)
status of the thermoelectrics is thoroughly compiled from f0 through a process that a particle with
by Rowe (2006). momentum k is scattered to acquire a momentum
The chapter is organized as follows: In Section of k9.
1.09.2, we discuss the Boltzmann equations for Then, the deviation from the equilibrium should
electrons and phonons, and derive explicit expres- be due to diffusion, external field, and scattering, as
sions for thermal conductivity. Then, we take a quick expressed by
look at the thermal conductivities of various
e @f
semiconductors in Section 1.09.3. We also discuss vk ?Ñf þ E ? Ñk f ¼ ð4Þ
h @t scattering
the thermoelectric effects in Section 1.09.4, and the
thermoelectrics in Section 1.09.5. In Section 1.09.6, where Ñk refers to the gradient by the momentum k.
the thermal and thermoelectric properties in The first and second terms on the left-hand side
nanostructured materials are briefly introduced. We correspond to the diffusion and the acceleration,
show that they raise difficult questions to our respectively. For the sake of simplicity, we omitted
understanding of thermal conduction. Finally, we the contribution from magnetic fields in Equation (4).
briefly summarize this chapter in Section 1.09.7. To find the expressions for electrons, we assume
the Fermi–Dirac distribution function for f0, that is,
–1
1.09.2 The Boltzmann Equation ð" k – Þ
f0 ¼ exp þ1 ð5Þ
kB T
1.09.2.1 Equations for Conduction
Electrons
where "k is the eigenenergy for the momentum k.
In the framework of nonequilibrium thermodynamics The scattering term is, in fact, very difficult to
(see Callen, 1985), the electrical current density j and handle strictly, so that the relaxation time approxi-
the thermal current density q are written as functions mation is often used, in which the scattering term can
328 Thermal Conductivity and Thermoelectric Power of Semiconductors
be written using the deviation from the equilibrium Note that Kˆn is a second-rank tensor through
distribution g ¼ f f0 as 0 1
vkx vkx vkx vky vkx vkz
B C
@f 1 1 B C
¼ – g ¼ – ðf – f0 Þ ð6Þ vk vk ¼ B C
@t scattering B v ky vkx v ky vky v ky vkz C ð14Þ
@ A
where is the scattering time which characterizes the vkz vkx vkz vky vkz vkz
timescale of the dissipation of the system. It is origin-
We also note that Equations (11) and (12) are iden-
ally defined in the form of the inverse number 1/ as
tical to Equations (1) and (2), and Onsager’s relation,
the probability per unit time that a particle with a
L12 ¼ L21, is readily satisfied in Equations (11) and
momentum k and an energy ! is scattered to one
(12). All the parameters are reduced to scalar quan-
with a momentum k9 and an energy !9.
tities for a cubic crystal, where the conductivity is
We further assume that f depends on r only
given by
through the space variation of temperature T ðrÞ.
Making use of the following expression of the ¼ e 2 K0 ð15Þ
group velocity
The rest of the parameters are similarly given by
1
vk ¼ Ñk"k ð7Þ 1 K1
h S¼ ð16Þ
eT K0
we obtain
K2
@f0 h "k – i 9 ¼ ð17Þ
g¼ – vk e Eþ ð – ÑT Þ ð8Þ T
@" "¼"k T
where S is the Seebeck coefficient (thermoelectric
and calculate the electrical current density as power or thermopower) and 9 the thermal conduc-
Z tivity in the absence of electric field. The electron
1 1 thermal conductivity is always measured for the
j¼ evk gd3 k ¼ evk f d3 k ð9Þ
43 43 open-circuit condition of j =0; thus, by eliminating
In deriving Equation (9), we used the fact that the E from Equation (12), we get
above integral gives zero for f ¼ f0, because the
q ¼ S 2 T ÑT – 9ÑT
electron moves in random directions. The prefactor
of 1/43 ¼ 2/(2)3 represents the density of states for S2 ð18Þ
¼ 9 1 – ð – ÑT Þ
the momentum space times the spin degrees of 9
freedom. and the thermal conductivity of conduction electron
Next, we find the expression of the thermal in real experiments is given by
current density in the same framework. Recalling
that the chemical potential and the eigenenergy are S 2
¼ 9 1 – ð19Þ
the free and internal energies per electron, respec- 9
tively, we can define the heat per electron as "k – ,
The second term is usually small, and we may often
and the thermal current density as
Z
neglect the difference between and 9 unless we
1 treat thermoelectric materials (see Section 1.09.5).
q ¼ ð"k – Þvk gd3 k ð10Þ
43 Using , S, and 9, we eventually simplify
Substituting Equation (8) in Equations (9) and (10), Equations (11) and (12) as
we obtain
j ¼ Eþ Sð – ÑT Þ ð20Þ
e
j ¼ e 2 Kˆ0 E þ Kˆ1 ð – ÑT Þ ð11Þ q ¼ ST Eþ 9ð – ÑT Þ ð21Þ
T
1 ˆ
q ¼ e Kˆ1 E þ K2 ð – ÑT Þ ð12Þ where all the parameters , S, and 9 can be deter-
T mined by experiments.
where From Equations (15) and (17), one can find that
Z the conductivity and thermal conductivity are
1 @f0 calculated from a similar integral for conventional
Kˆn ¼ 3 – vk vk ð"k – Þn d3 k ð13Þ
4 @" "¼"k metals, in which the Fermi energy EF is much larger
Thermal Conductivity and Thermoelectric Power of Semiconductors 329
than the thermal energy kBT. By expanding @f0/@" in Then, we finally obtain the lattice thermal
terms of kBT/EF (the Sommerfeld expansion), one conductivity as
can derive K2 in terms of K1 or as Z
1 @N0
ˆ ph ¼
h!k vk vk – d3 k ð27Þ
2 kB2 T 2 kB T 2 83 @" "¼h!
K2 ¼ ðE F Þ þ O ð22Þ
3 e2 EF
Let us consider the physical meaning of Equation
and obtain the Wiedemann–Franz law given by (27). For the sake of simplicity, we assume cubic
symmetry, in which the second-rank tensor of
T 2 kB2 Equation (14) is reduced to the scalar quantity of
¼ ¼ L0 ð23Þ
el 3 e2 vk 2 . If we further assume that the velocity and scat-
tering time are independent of k, we may rewrite
where we write el, instead of 9, to emphasize that the
Equation (27) as
thermal conductivity discussed here is due to conduc-
" Z #
tion electrons, not due to phonons. The numerical value 1 1 @N0 3
of L0 is 2.4 108 V2 K2 (155 mV/K)2 is known as ph ¼ v,
3 83
h!k –
@" "¼h!
dk ð28Þ
the Lorentz number. This relation states that electrons
in solids can carry heat as well as electricity, or where we take the thermal average of the velocity as
equivalently, that a metal is a good conductor of
electricity and heat. Note that the Lorentz number vx2 ¼ vy2 ¼ vz2 ¼ v2 =3 ð29Þ
L0 is given only by the universal constants. This and introduce the mean free path , ¼ v. Noting that
justifies the Wiedemann–Franz law that the ratio of the quantity in the brace is the specific heat of phonons,
the thermal to the electrical conductivity at fixed one can approximate the thermal conductivity as the
temperatures is independent of material parameters. product of the sound velocity, the mean free path, and
This is known as one of the most remarkable successes the specific heat.
of Drude’s theory. First, we show the electron thermal conductivity
in the Drude model (Ashcroft and Mermin, 1976).
Drude regarded electrons in a metal as ideal gas
1.09.2.2 Equations for Phonons molecules, and developed a kinetic theory of such
gases. Suppose a metal rod is subject to a temperature
The Boltzmann equation can also be applied to the gradient (one-dimensional (1D) model). The elec-
lattice thermal conductivity, the contribution of pho- trons coming from the hot side have a larger
nons to the whole thermal conductivity. Since the
thermal energy and equivalently, a larger kinetic
phonon does not carry charge, we only consider the
energy of ", so that the net heat flow is given as
energy (heat) flow, in which the distribution function
at thermal equilibrium is given by the Bose–Einstein nvx
q¼ ½"ðT ½x – vx Þ – "ðT ½x þ vx Þ ð30Þ
distribution as 2
–1 d"
h!k ¼ nvx ,ðT1 – T2 Þ ð31Þ
N0 ¼ exp –1 ð24Þ dT
kB T
where n is the carrier density and v(T) is the thermal
where !k is the phonon frequency for the wavevector velocity at temperature T. The factor of 1/2 repre-
k (note that the chemical potential is zero for sents that the half of electrons are moving toward the
phonons). Similarly in the case of electrons, we position x. We take the thermal average of the
write the thermal current density of phonons as velocity of Equation (29) and obtain the expression as
Z 1 1
1
q ¼ h!k N vk d3 k ð25Þ el ¼ Cel v2 ¼ Cel v,el ð32Þ
83 3 3
where N is the distribution function for phonons in where Cel ¼ nd"/dT and , are the electronic specific
the presence of temperature gradient. We take the heat per volume and the electron mean free path,
relaxation time approximation given by respectively. One may notice that this is a special
form of Equation (28). The interpretation of this
@N0 equation is straightforward. A good conductor of
N – N0 ¼ – vk ð – ÑT Þ ð26Þ
@" heat should satisfy that (1) the number of the heat
330 Thermal Conductivity and Thermoelectric Power of Semiconductors
carrier is large (large Cel), (2) the speed for the heat Steigmeier and Abeles (1964) calculated the para-
carrier is high (large v), and (3) the heat carrier meter A in a solid solution system. According to
travels far (large or ,). In fact, Equation (32) is their calculation, A is given by
valid for phonons as well: V X m – mi 2
A¼ 3
fi ð37Þ
1 1 4v i
m
ph ¼ Cph vs2 ¼ Cph vs ,ph ð33Þ
3 3
where V is the volume per atom, fi the volume fraction
in which Cel is replaced by the phonon-specific heat of the atom i, mi the mass of the atom i, and m the average
Cph; vs and are the sound velocity and the phonon mass. Thus, the parameter A is large, when the mass
scattering time, respectively. – mi is large. The third term is the phonon–
difference m
To step further, let us go back to Equation (27), phonon (normal) scattering. Since the lattice in real
and discuss the relaxation time . By definition, is a materials deviates from the ideal harmonic oscillator,
function of the momenta of a phonon scattered from the restoring force can be nonlinear. In terms of pho-
k to k9. While the conduction electrons in metals nons, it corresponds to the phonon–phonon scattering.
have nearly the same energy as EF, the phonons (Recall that the anharmonicity corresponds to products
may have different energy ranging from zero to of four phonon operators like bky bk9y bq9 bq .) This scattering
kBT. Thus, it is more reasonable to assume that is is phenomenologically given as
dependent on energy, or equivalently, frequency !
N – 1 ¼ B!a T b ð38Þ
through k k9. Eventually, the mean free path of
phonons may strongly depend on !, implying that where a and b are some exponents of the order of
Equation (33) is often oversimplified. unity. These parameters are determined by fitting
It has long been discussed how the scattering time the thermal conductivity measured in real materials.
of phonons depends on ! (see Yang, 2004). Usually, The last scattering is the phonon–phonon Umklapp
we can assume that the scattering events from differ- scattering. At high temperatures where the momen-
ent origins are additive in the form of scattering rates tum change k – k9 is large enough to exceed a
given as reciprocal lattice vector, then phonons are scattered
X from one Brillouin zone to another. Such scattering is
–1
total ¼ i – 1 ð34Þ called Umklapp scattering. Slack and Galginaitis
i (1964) proposed the following formula
This is known as Matthiessen’s rule in electric resis-
h 2 D
tivity, in which the total resistivity is given by the U – 1 ¼ !2
T exp – ð39Þ
Mv2 D 3T
sum of the resistivity due to impurity scattering and
that due to electron–phonon scattering (Ziman, where D is the Debye temperature and the Grüneisen
1976). parameter. Equations (38) and (39) have similar
In the case of phonons, we usually consider three temperature dependence except for the exponential
to four kinds of scattering, as pointed out by term in the latter, and thus the former is sometimes
Callaway (Callaway, 1959; Callaway and v Baeyer, ignored for the data fitting. At first glance, one may
1960). The first one is the boundary scattering corre- wonder if the thermal conductivity is uniquely fit with
sponding to the elastic scattering. This term is the above four scattering times. Fortunately, they affect
independent of energy, expressed in terms of a the curve differently, and the scattering times are
parameter L as reasonably determined by fitting the experimental
data. We see how experimental data are explained
B – 1 ¼ v=L ð35Þ with these scattering times in the following section.
L corresponds to a characteristic length of the
system, which can be the dimensions of the sample,
the size of the grains, etc. The second one is the 1.09.3 Thermal Conductivity
point-defect scattering corresponding to the impur- 1.09.3.1 Single-Crystalline
ity scattering of electrons. Klemens (1960) pointed Semiconductors
out that this scattering time PD is proportional
to !4, so that Figure 1 shows the thermal conductivity of pure and
doped silicon, silicon carbide, and diamond (Slack,
PD – 1 ¼ A!4 ð36Þ 1964). Below around 20 K, the thermal conductivity
Thermal Conductivity and Thermoelectric Power of Semiconductors 331
104 104
Thermal conductivity (W m–1 K–1)
Diamond 70
100 Doped Si
T3
–1
10 102
100 101 102 103 T –1
Temperature (K)
1 10 100
Figure 1 Thermal conductivity of pure and doped silicon, Temperature (K)
silicon carbide, and diamond. From Slack GA (1964)
Thermal conductivity of pure and impure silicon, silicon Figure 2 Thermal conductivity of isotope-pure and
carbide, and diamond. Journal of Applied Physics 35: natural Ge. From Asen-Palmer M, Bartkowski K, Gmelin E,
3460–3466. et al. (1997) Thermal conductivity of germanium crystals
with different isotopic compositions. Physical Review B 56:
9431–9447.
increases approximately in proportion to Tp
(3 > p > 2), takes a broad maximum around 20–80 K,
and decreases with temperature approximately The thermal conductivity of diamond is much higher,
inversely proportional to T above 100 K. owing to higher Debye temperature and higher sound
The temperature dependence is qualitatively velocity. Because of this property, diamond is
understood as follows: according to Equation (33), expected to be used as a semiconductor wafer of
the thermal conductivity is given by the product of next generation.
the specific heat, the phonon mean free path, and the As discussed in the previous section, the point-
sound velocity. In many cases, the sound velocity defect scattering given by Equations (36) and (37)
does not change with temperature very much, and depends on the mass difference. This means that the
the other two predominantly determine the tempera- isotope atom can be a scattering center of phonons.
ture dependence of the thermal conductivity. As is Geballe and Hull (1958) found that the isotopically
well known, the specific heat is expected to scale with enriched 74Ge (96%) has three times larger thermal
T3 for T D, and to be 3NkB for T D. conductivity than natural Ge. Asen-Palmer et al.
In contrast, the phonon mean free path is dominated (1997) successfully measured the thermal conductivity
by impurity scattering at low temperatures, and of isotope-pure 70Ge (99.99%) as shown in Figure 2.
expected to be weakly dependent on temperature. The observed thermal conductivity is eight times
With increasing temperature, the phonon begins to larger at maximum than that of natural Ge crystal.
be scattered by other phonons and/or electrons. Figure 3 shows thermal conductivity of various
In particular, Equation (39) indicates that the semiconductors with particularly low thermal con-
phonon–phonon scattering rate is roughly propor- ductivity (Kurosaki et al., 2005). They are used for
tional to T for T D, so that the mean free path thermoelectric materials, which are discussed in
is inversely proportional to T at high temperatures. Section 1.09.5. All the materials include heavy ions
Thus, the product of the specific heat and the phonon such as Ag, Tl, Te, and Bi, and consequently have
mean free path is expected to be proportional to Tp low sound velocities. Other complex materials with
(2 < p < 3) for low temperatures and to 1/T for high low thermal conductivity are summarized in a review
temperatures (see also Figure 2). The thermal by Snyder and Toberer (2008). By comparing data
conductivity peak corresponds to the crossover from Figures 1–3, we can say that the magnitude of
between the two temperature regions. the thermal conductivity ranges from 101 to
The magnitude of the thermal conductivity is 104 W m1 K1. This dynamic range is significant,
80 W m1 K1 at room temperature for doped silicon. but much smaller than that of electrical conductivity,
Since a typical value of thermal conductivity of metals in which it ranges from 1020 to 106 S cm1. This
is 10–102 W m1 K1 (Ashcroft and Mermin, 1976), implies that neither ideal heat insulator nor heat
this value is reasonably high for phonon conduction. conductor is available in real materials.
332 Thermal Conductivity and Thermoelectric Power of Semiconductors
150
Thermal conductivity (W m–1 K–1)
50
0
0 50 100 150 200 250 300
Temperature (K)
Figure 4 Lattice thermal conductivity of skutterudite antimonide. The contributions of three scattering mechanisms are
separately drawn. From Yang J (2004) Theory of thermal conductivity in crystalline semiconductors. In: Trit TM (ed.) Thermal
Conductivity. Kluwer.
Thermal Conductivity and Thermoelectric Power of Semiconductors 333
of CoSb3 and CeCoFe3Sb12, in which the huge mass decreases as T3. The temperature dependence is quali-
difference results in a large A. The third term dom- tatively understood as follows: in the present situation,
inates the decreasing thermal conductivity at high the mean free path is as short as the atomic distance, and
temperatures. For T D, all the phonon modes consequently independent of temperature. Although the
are thermally excited, and the specific heat is sound velocity cannot be defined in disordered mate-
saturated as a classical value of 3NkB. In such condi- rials, it is expected to be similar to that in crystals. Thus
tions, the contribution from !4 does not cause strong the last term, the specific heat, is the origin of the
temperature dependence; the phonon–phonon temperature dependence of the thermal conductivity.
scattering rate is proportional to T exp(D/T) T. Another feature to be pointed out is that all the curves
are reasonably close to one another, implying somewhat
universal behavior.
1.09.3.2 Amorphous Solids and Glasses
Cahill et al. (1992) proposed a theoretical expres-
The amorphous solids and glasses are materials in sion for the thermal conductivity of disordered solids,
which atoms have a short-range order, but no trans- by assuming that the thermal energy of each atom is
lational symmetry. As a result, no phonons are well lost during half the period of an oscillation of the
defined, or all the phonon modes are localized at each atom. Based on the Debye model, the minimum
atoms. Cahill et al. (1992) measured and analyzed the thermal conductivity min is given by
thermal conductivity of various disordered crystals, 2 1=3 X 2 Z i =T 3 x
and showed minimum thermal conductivity, as was nL T x e dx
min ¼ kB vi ð41Þ
originally proposed by Slack (1979). 6 i
i 0 ðex – 1Þ2
Figure 5 shows the thermal conductivity of amorph-
where vi corresponds to three sound modes (two for
ous solids ( -SiO2 and CdGeAs2) and disordered solids
transverse and one for longitudinal). The cutoff
(Or1Ab33An66 feldspar and Ba0.67La0.33F2.33). All the
frequency i is given in temperature unit as
curves have low magnitudes at room temperature (less
than 102 W cm1 K1 ¼ 1 W m K1), and weakly h
i ¼ vi ð62 nL Þ1=3 ð42Þ
dependent on temperature above 10 K. With decreasing kB
temperature below 10 K, the thermal conductivity
where nL is the density of atoms. Note that Equation
(41) contains no free parameters, because vi and nL
10–1 can be determined by different experiments.
Or1 Ab33 An66 Figure 6 shows the calculated min plotted against
thermal conductivity measured in various disordered
Thermal conductivity (W cm−1 K−1)
100
Cu
b O
Sr
κs
a Chain
c
b
a c T –1
10
10 100
Temperature (K)
Figure 7 Anisotropic thermal conductivity of the 1D magnetic insulator Sr2CuO3 along the a-, b-, and c-axis directions.
From Sologubenko AV, Felder E and Giannò K, et al. (2000a) Thermal conductivity and specific heat of the linear chain cuprate
Sr2CuO3: Evidence for thermal transport via spinons. Physical Review B 62: R6108–R6111.
Thermal Conductivity and Thermoelectric Power of Semiconductors 335
for this high thermal conductivity. Other 1D spin conductivity is seen in the spin-ladder copper oxide
chains such as Sr14Cu24O41 and SrCuO2 show Ca9La5Cu24O41, in which the a-axis thermal conduc-
similarly large thermal conductivity along the chain tivity is 20 times larger than the c-axis one at 170 K
direction. (Hess et al., 2001; Kudo et al., 2001; (Hess et al., 2001).
Sologubenko et al., 2000b). If the magnetic excitation dominates the heat con-
Figure 8 shows the mean free path of the duction in magnetic materials, one may easily
magnetic excitation evaluated from the Drude-like imagine that external fields will change the thermal
expression: conductivity. Such thermomagnetic effects are
widely seen in magnetic materials such as CuGeO3
1
mag ¼ Cmag vmag ,mag ð43Þ (Ando et al., 1998), R2CuO4 (Sales et al., 2004), and
3
(La,Ca)MnO3 (Chen et al., 1997). Figure 9 shows the
As pointed out in the previous section, the Drude- thermal conductivity of the layered vanadium oxide
like expression for is oversimplified. However, in K2V3O8 (Sales et al., 2002). This particular oxide is an
the case of magnetic excitation, we do not know antiferromagnetic insulator with the local moment
anything about the scattering process, and are unable on V ions aligned along the a-axis. This antiferro-
to calculate scattering times. The evaluated mean magnetic order exhibits a spin-flop transition, in
free path is as large as 100 nm at low temperatures, which the critical field is less than 1 T. The thermal
which strongly suggests that magnetic excitations in conductivity suddenly increases above the critical
this material travel over 250 unit cells without scat- field, suggesting that the spin orientation determines
tering. Note that the lattice thermal conductivity is the heat conduction to some extent.
less anisotropic; the anisotropy is basically deter- Like spin-valve/giant magnetoresistance effects in
mined by the anisotropy of the sound velocity multilayered ferromagnetic films (Zutic et al., 2004),
(Crommie and Zettl, 1991). Even in chain-like mate- the magnetothermal effect observed here may be
rials such as Sr2CuO3, the chemical bonds applied to a thermal spin valve (Sales et al., 2004). At
perpendicular to the chain are similar to those in the moment, it is remarkable only at low temperatures
the chain. As a result, the restoring force will be at high fields (larger than 1 T). Even if a large response
more or less isotropic. Even in the presence of van in the spin thermal conductivity were discovered at
der Waals coupling in one direction, the anisotropic room temperature, the lattice thermal conductivity
ratio of remains around 5–7 (Terasaki et al., 2004). would remain unchanged, and overlap the magnetic
In this respect, the 300-K anisotropy between a- and contribution. Thus, the magnetothermal conductivity
b-axis directions is remarkable, which may be used as could not be large at room temperature, in comparison
a heat guide in solids. More anisotropic thermal with magnetoconductivity.
6.5
103 +gμH
H//a, Q//a
6
κa (W m−1 K−1)
5.5 − gμH
Is (Å)
5 H //c, Q//a
H = 0.55 T
4.5
4
102 T = 3.7 K
H = 0.95 T
3.5
20 40 60 80 100 200 0 0.5 1 1.5 2 2.5 3 3.5 4
Temperature (K) H (T)
Figure 8 Mean free path of magnetic excitation in Figure 9 Thermal spin valve in K2V3O8. From Sales BC,
Sr2CuO3. From Sologubenko AV, Felder E and Giannò K Lumsden MD, Nagler SE, Mandrus D and Jin R (2002)
et al. (2000a) Thermal conductivity and specific heat of the Magnetic field enhancement of heat transport in the 2D
linear chain cuprate Sr2CuO3: Evidence for thermal Heisenberg antiferromagnet K2V3O8. Physical Review
transport via spinons. Physical Review B 62: R6108–R6111. Letters 88: 095901.
336 Thermal Conductivity and Thermoelectric Power of Semiconductors
600
104
2.8 × 1016 cm–3
Thermoelectric power (μV k–1)
500
103
–S (μV K–1)
400
102
300 1.7 × 1019 cm–3
200 101
n-type-doped SI
100 1 3 5 7 9
2 4 6 8 100
0 50 100 150 200 250 300
0 Temperature (K)
180 200 220 240 260 280 300
Temperature (K) Figure 14 The Seebeck coefficients of lightly doped
silicon. From Weber L and Gmelin E (1991) Transport
Figure 13 The Seebeck coefficient of pentacene. properties of silicon. Applied Physics A 53: 136–140.
Reproduced from Mühlenen Av, Errien N, Schaer M, Bussac
M-N and Zuppiroli L (2007) Thermopower measurements on
pentacene transistors. Physical Review B 75: 115338. Gmelin, 1991). The Seebeck coefficient of the lightly
doped silicon is well described by Equation (62), in
in the figure represent different samples deposited on which the Seebeck coefficient is linear in 1/T. The
different substrates. The Seebeck coefficient is magnitude is much larger than kB/jej ¼ 86 mV K1,
essentially independent of temperature, and the and the activation energy Eg evaluated with Equation
magnitude is well understood from Equation (56). (62) is semiquantitatively consistent with the optical
The Seebeck coefficient of organic conductors is band gap.
reviewed by Kaiser (2001), in which lightly doped In the case of degenerate semiconductors, the
organic conductors show temperature-independent Seebeck coefficient is essentially the same as that of
thermopower explained by the Heikes formula. The a noninteracting Fermi gas given by Equation (51).
Heikes formula also explains the Seebeck coefficient The Seebeck coefficient of the heavily doped silicon
of transition-metal oxide semiconductors such as shown in Figure 14 exhibits T-linear dependence
La1xSrxCrO3 (Karim and Aldred, 1979) and like those of metals. At high temperatures, minority
La1xSrMnO3 (Palstra et al., 1997). carriers are thermally excited to decrease the magni-
Let us consider the Seebeck coefficient in lightly tude of the Seebeck coefficient, where the Seebeck
doped semiconductors. For a lightly doped semi- coefficient roughly obeys Equation (62). Figure 15
conductor, the chemical potential is pinned near the shows the Seebeck coefficient of the skutterudite
center of the band gap, and has a gap of 2Eg to the
bottom (top) of conduction (valence) band. 300
Substituting Eg for in Equation (49), we get
Seebeck coefficient (μV–1 K)
250 1212OB22
10OB22
1 K1 2NB13
S¼ 200 2NB9
eT K0
Z
150
@f0 2
– v ð"k Þ þ Eg Þd3 k ð61Þ
1 @" k
¼ Z 100
eT @f0 2 3
– v d k
@" k 50
p-type CoSb3
Eg 0
ð62Þ
eT 300 400 500 600 700 800 900
Temperature (K)
In other words, each thermally excited electron will
acquire the activation energy, and carry large entropy. Figure 15 The Seebeck coefficient of the skutterudite
antimonide CoSb3. Reproduced from Caillat T,
Hence, the Seebeck coefficient is large compared with Borshchevsky A and Fleurial JP (1996) Properties of single
those of metals. Figure 14 shows the Seebeck coeffi- crystalline semiconducting CoSb3. Journal of Applied
cient of lightly and heavily doped silicon (Weber and Physics 80: 4442–4449.
340 Thermal Conductivity and Thermoelectric Power of Semiconductors
antimonide CoSb3 with various doping concentra- The Harman method is known as a unique
tions (Caillat et al., 1996). At low temperatures, technique to measure thermal conductivity using
the Seebeck coefficients decrease with decreasing the Peltier effect (Harman, 1958). Suppose that a
temperature, suggesting that the electrical conduc- sample is hung by electrical leads in vacuum at a
tion is dominated by degenerate carriers. In contrast, temperature T. Upon a steady current I0, the Peltier
the Seebeck coefficients have peaks and decrease heat I0 is generated at one current contact to the
with increasing temperature at high temperatures, sample, and is absorbed at the other. As the sample is
showing that the minority carriers begin to be excited in vacuum, that is, thermally isolated, the Peltier heat
thermally. The temperature at which the Seebeck is finally balanced by a back flow of thermal current
coefficients peak corresponds to a crossover from through the sample, and causes a temperature
the degenerate to intrinsic regions. difference T between the contacts to satisfy
K T ¼ I0 ð66Þ
1.09.4.2 The Peltier and Thomson Effects where K is the thermal conductance of the sample.
If the Seebeck coefficient is measured in advance, the
The Peltier effect is the reverse phenomenon of the Peltier coefficient will be given by ¼ ST. Thus, K
Seebeck effect; the electrical current flowing through can be obtained from the observed T. Figure 16
the junction connecting two materials will emit or shows an example of T observed in the Harman
absorb heat per unit time at the junction to balance method (Satake et al., 2004). The temperatures at the
the difference in the chemical potential of the two left side (TL) and those at the right side (TR) change
materials. Thanks to this effect, an electronic refrigera- with external currents. As shown in the inset, the
tor can be made, which is known as the Peltier cooler. antisymmetric part of the data is plotted to remove
The Peltier cooler has been applied to niche areas such the contribution from the Joule heating (which is
as infrared detectors, CPU coolers, wine cellars, etc., proportional to I0 2 ), which is linear in external
because the cooling power is lower than that of com- current, as expected from the Peltier effect.
pressor-based refrigerators. This technique is called The third thermoelectric effect is the Thomson
thermoelectrics, and is discussed in the next section. effect. In the presence of temperature gradient and
Similarly to the Seebeck coefficient, the Peltier external current, thermal current generates in pro-
coefficient is defined as the coefficient of the ther- portion to the temperature gradient and external
mal current Q to the electrical current I, given by current given as
Q ¼ I ð63Þ @T
qx ¼ – T jx ð67Þ
@x
or equivalently, is defined as the coefficient of the
thermal current density to the electrical current
density
ΔT(l 0)−ΔT(–l 0) (K)
4
q ¼ j ¼ ð – ÑT Þ ð64Þ 2
3
2 300 K
where the current density is further expressed by the
ΔT = TR − TL (K)
electrical conductivity. 1 1
The Peltier coefficient is related to the Seebeck 0
0 5 10 15
coefficient. From Equations (11) and (12), we may l 0 (mA)
0 100 K
associate with S as
¼ ST ð65Þ
−1
This relationship is not specific to Equations (11) and
(12), but can be demonstrated from the first principle, −10 0 10
being known as Onsager’s reciprocal theorem. Current l0 (mA)
Thanks to this relationship, the Peltier coefficient
Figure 16 The temperature difference by the Peltier effect
has been rarely measured. The amount of heat is
in the sample hung in vacuum. From Satake A, Tanaka H,
difficult to measure precisely than the temperature Ohkawa T, Fujii T and Terasaki I (2004) Thermal conductivity
difference, so that the Seebeck coefficient is much of the thermoelectric layered cobalt oxides measured by the
easier to measure. Harman method. Journal of Applied Physics 96: 931–933.
Thermal Conductivity and Thermoelectric Power of Semiconductors 341
where T is the Thomson coefficient. This effect is Recalling that the entropy goes to zero at 0 K (the
understood through the temperature dependence of third law of thermodynamics), we can assume that
the Seebeck coefficient. Suppose that the tempera- the Seebeck coefficient also goes to zero at 0 K, and
ture of the left half of a rod-like sample is T1, and that calculate the absolute Seebeck coefficient from the
of the right half is T2. This is equivalent to two rods measured Thomson coefficient.
with different Peltier coefficients of T1S(T1) and
T2S(T2) welded at the edge. Then, the welded
junction will have the Peltier heat with respect 1.09.4.3 The Galvanomagnetic Effects
to external current. Eventually, the Thomson A conduction electron is coupled with external mag-
coefficient is written as netic fields through the Lorentz force, and is
dS consequently deflected along the direction perpen-
T ¼ T ð68Þ dicular to both the current and magnetic field. The
dT
deflected electrons are accumulated to one side of
The measurement of the Thomson coefficient is the sample to cause the transverse voltage, and cancel
important, because the absolute Seebeck coefficient the Lorentz force in the steady state. This is known as
can be obtained from T. Note that the thermoelec- the Hall effect, and the transverse voltage is called
tric voltage V is experimentally observed in the the Hall voltage. Similarly to this, the electron in the
open-circuit condition as presence of temperature gradient will be deflected to
V ¼ Slead ðT0 – T1 Þ þ Ssample ðT1 – T2 Þ the direction perpendicular to both the temperature
ð69Þ gradient and the magnetic field, and eventually cause
þSlead ðT2 – T0 Þ the transverse voltage. This is known as the Nernst
¼ ðSsample – Slead ÞðT1 – T2 Þ ð70Þ effect, a typical galvanomagnetic effect. The Nernst
coefficient N is defined as
where T0 is the ambient temperature, T1 and T2 are
the temperatures at the edges of the sample. Ssample @T
Ey ¼ NBz ð71Þ
and Slead are the Seebeck coefficients of the sample @x
and the voltage leads, respectively. This indicates that where Bz is the external magnetic field applied along
the Seebeck coefficient is always measured as the the z-direction, and the temperature gradient is along
difference from the Seebeck coefficient of the leads. the x-direction.
Figure 17 shows the Thomson coefficient of The Nernst effect is exceptionally large in semi-
various metals (Lander, 1948). As discussed above, metals (Li and Rabson, 1970). This is a similar
the Seebeck coefficient is an entropy per carrier. situation of the classical magnetoresistance which
has a large magnitude in semimetals (Clayhold,
1996). In particular, the high-mobility semimetals
20
have the Nernst voltage as large as the Seebeck
Thomson coefficient (micro V K–1)
Ag
voltage in magnetic fields of several teslas.
Cu +
Figure 18 shows the Nernst voltage of Bi1xSbx in
0 Au
W – various magnetic fields (Li and Rabson, 1970). For
Mo small magnetic fields less than 1 T, the Nernst
−20 Pt voltage exceeds 1 mV K1 at low temperatures.
Another class of materials exhibiting large Nernst
coefficients is a superconductor in the mixed state.
−40 Needless to say, a superconductor is a material that
Pd
exhibits zero resistivity below the superconducting
transition temperature Tc. External magnetic field
−60 above the lower critical field Bc1 can penetrate a
300 900 1500 2100
Temperature (K)
type II superconductor as quantized magnetic fluxes.
This state is called the mixed state, in which the
Figure 17 The Thomson coefficient of metals.
quantized fluxes form a triangular lattice. The flux
Reproduced from Lander JJ (1948) Measurements of
Thomson coefficients for metals at high temperatures and lattice is squeezed by external current through the
of Peltier coefficients for solid–liquid interfaces of metals. Lorenz force, and begins to flow above a critical
Physical Review 74: 479–488. current density. Similarly to this, the flux can flow
342 Thermal Conductivity and Thermoelectric Power of Semiconductors
1.5 114 K
ascribed to the phase fluctuation of the supercon-
188 K ducting order parameter far above Tc. A similar
1
large Nernst signal is observed in organic conductors
(Nam et al., 2007).
0.5
Te-doped Bi88Sb12 300 K
0 1.09.5 Thermoelectrics
0 0.2 0.4 0.6 0.8 1
B (T) 1.09.5.1 Thermoelectric Devices
Figure 18 The Nernst voltage of the high-mobility
semimetal Bi1xSbx. From Li SS and Rabson TA (1970) The
As discussed in the previous section, the Seebeck
Nernst and the Seebeck effects in Te-doped Bisingle bond and Peltier effects are cross-correlations between
Sb alloys. Solid State Electronics 13: 153–160. electricity and heat carried by an electron in
solids. When the Seebeck coefficient and the
in the presence of temperature gradient (Huebener, electrical conductivity are large in a material, such
1995) from high- to low-temperature side. According a material can convert electric energy into heat and
to Faraday’s electromagnetic induction, the voltage vice versa. This is schematically illustrated in
will be generated in the direction of B ð – ÑTÞ, Figure 20. A rod of a material subject to the
which is nothing but the Nernst voltage. temperature gradient causes the thermoelectric
Figure 19 shows the Nernst voltage NB in the voltage V ¼ ST across the sample, where T is
mixed state of the high-temperature superconductor the temperature difference between the edges of
YBa2Cu3O7 (Huebener, 1995). With increasing tem- the sample, and S is the Seebeck coefficient. If the
perature, the signal in a fixed field sharply rises above voltage V is sufficiently high, the rod works as a
battery, and generates electricity at an external
a certain temperature, reaches a maximum, and then
load connected to the rod. In this situation,
decreases slowly. The external field enhances the
the thermoelectric voltage corresponds to the open-
range in which the Nernst signal is observable. This
circuit voltage of the battery, and the resistance of the
is qualitatively explained as follows: the quantized
rod corresponds to the internal resistance of the
fluxes are frozen at sufficiently low temperatures,
and cannot exist far above Tc. Consequently, the
3
H=1T
H=2T
Current
H=3T
Ey/∇xT (μV K–1)
2 H=4T
H=6T
H=8T
H = 10 T
H = 12 T
1
Sample
0
70 80 90 100 110
Temperature (K)
Figure 19 The Nernst voltage of the high-temperature
superconductor YBa2Cu3O7 in the mixed state. Figure 20 A schematic picture of thermoelectric power
Reproduced from Huebener RP (1995) Superconductors generation. Reprinted from Narlikar AV (ed.) (2005) Layered
in a temperature gradient. Superconductor Science and Cobalt oxides as a thermoelectric material in Frontiers in
Technology 8: 189–198. magnetic Materials. pp 327–346. Springer: Berlin: Figure 1.
Thermal Conductivity and Thermoelectric Power of Semiconductors 343
battery. Such a material is called a thermoelectric coefficient, and the net thermal conductance of the
material, and the device made from is called a ther- device, respectively. For the sake of simplicity, we
moelectric device. One can find a good review on assume that all the parameters of the device are
thermoelectrics given by Mahan (1997). independent of temperature. At the junction
Actually, Figure 20 is an oversimplified picture (cold side), the pumped heat per unit time QC is
for the thermoelectric device. Since the current lead given by
is usually a good conductor of heat, the heat applied
1
at the left edge will flow through the lead, and will QC ¼ STC I – RI 2 – K T ð72Þ
2
not generate sufficient temperature difference and,
in the worst case, may damage the load. To avoid where the second term is the Joule heat in the sample
this, one should make a pair of rods as shown in (we assume that a half of the heat goes to each side)
Figure 21. This structure is nothing but a thermo- and the third term is the back flow of the thermal
couple, and the heat applied at the junction now current by the temperature difference. Similarly, at
flows through the pair of thermoelectric legs. In this the hot side, the emitted heat per unit time QH is
given by
respect, the thermoelectric device is a thermocouple
with an ability to generate electricity, as shown in 1
QH ¼ STH I þ RI 2 – K T ð73Þ
Figure 21(b). Obviously, the pair should be a pair 2
of n- and p-type materials to maximize the Thus, the total work per unit time is given by
thermoelectric voltage. Another requirement is that
the thermal conductivity should be as low as possible W ¼ QH – QC ¼ ðST þ IRÞI ð74Þ
to maximize the temperature difference across Equation (74) means that the work to refrigerate is
the legs. equal to the voltage drop of the device in addition to
The thermoelectric device does not only generate the thermoelectric voltage.
electric power from heat, but also converts electric The heat balance at the hot and cold sides in the
power into heat through the Peltier effect, that is, it thermoelectric power generator shown in
cools the junction with external current, as shown Figure 21(b) is similarly given by
in Figure 21(a). This is known as thermoelectric
refrigeration or Peltier cooling. Thanks to the 1
QH ¼ STH I – RI 2 þ K T ð75Þ
Onsager relation of Equation (65), the Peltier 2
coefficient is equal to the Seebeck coefficient 1
QC ¼ STC I þ RI 2 þ K T ð76Þ
multiplied by absolute temperature. Thus, materials 2
with the large Seebeck coefficient near or below By connecting an external load Rext ¼ xR, we find the
room temperature can be used for thermoelectric current I ¼ ST/(1 þ x)R. Then, the output power P
refrigeration. is equal to
We begin with the heat balance of the thermo-
electric cooling device shown in Figure 21(a). Let R, ðST Þ2 x
P ¼ IV ¼ ð77Þ
S, and K be the net resistance, the net Seebeck R ð1 þ xÞ2
(a) (b)
p-type
TC TH TH TC
xR
QC QH QH QC
n-type
Figure 21 A schematic picture of thermoelectric devices: (a) thermoelectric refrigerator and (b) thermoelectric power
generator. Reprinted from Sorell CC, Sugihara S, and Nowotny J (eds.) (2005) Materials for Energy Conversion Devices.
Cambridge, UK: Woodhead Publishing: Figure 13.2.
344 Thermal Conductivity and Thermoelectric Power of Semiconductors
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
which takes a maximum Pmax ¼ (ST)2/4R at x ¼ 1, TC 1 þ ZT – TH
max ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð86Þ
that is, Rext ¼ R. Pmax is determined by the following T ð 1 þ ZT þ 1Þ
material parameter:
S2 after some calculations. For the power generation,
¼ S2 ð78Þ
the efficiency is given as
which is called the (thermoelectric) power factor. W
Here, we evaluate the maximum heat absorption ¼ ð87Þ
QH
of the cooling device under the conditions of
constant TH and TC. Then, a necessary condition of VI
¼ ð88Þ
dQC/dI ¼ 0 gives the optimum current I0 ¼ STC/R. STH I – RI 2 =2 þ K T
By substituting I0 into Equation (72), we have xT
¼ ð89Þ
2 ð1 þ xÞT þ ð1 þ xÞ2 =Z þ xT =2
ðSTC Þ
QCmax ¼ – K T ð79Þ
2R
By taking d/dx ¼ 0, we find that the maximum
ðSTC Þ2 efficiency is
¼K – T ð80Þ
2RK pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
T ð 1 þ ZT – 1Þ
Then, we introduce the figure of merit Z defined by max ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð90Þ
TH 1 þ ZT þ TC
S2 S2 S2
Z¼ ¼ ¼ ð81Þ
RK
According to Equations (86) and (90), material
properties are associated with the conversion
and rewrite QCmax as
efficiency through ZT. This is reasonable, because
ZTC 2 the conversion efficiency and ZT are both dimen-
QCmax ¼K – T ð82Þ
2 sionless quantities. In this respect ZT is the most
Thus, the maximum heat absorption is directly important parameter for thermoelectrics, and is
proportional to Z (or the power factor) for T ¼ 0. called the dimensionless figure of merit.
Next, we evaluate the lowest achievable tempera- We can take some notes on the above results.
ture TCmin under the conditions of constant QC and First, max given by Equation (86) and max given
TH. A necessary condition of dTC/dI ¼ 0 gives the by Equation (90) are reduced to the Carnot efficiency
optimum current I1 ¼ STCmin =R. By substituting I1 as ZT ! 1. This is reasonable, because thermoelec-
into Equation (72), we have tric energy conversion is a conversion through the
electron transport, which is an irreversible process
ðSTCmin Þ2 QC accompanying the Joule heat. Second, as shown in
T ¼ – ð83Þ
2KR K Figure 22, the efficiency is larger for larger ZT and
The maximum temperature difference (i.e., lowest T. Considering that the conversion efficiency of a
achievable temperature) is again directly propor- solar battery is 10–15%, we think that a similar is
tional to Z for QC ¼ 0. expected for practical use. For ZT̄ < 1, Equation (90)
In addition, we evaluate the maximum efficiency. can be expanded in terms of ZT̄. In the lowest order,
The energy conversion efficiency for a cooling max is
device is characterized by the coefficient of perfor-
mance (COP): max ZðTH – TC Þ=4 ð91Þ
QC STC I – RI 2 =2 – K T
¼ ¼ ð84Þ Thus, Z(TH TC) ¼ 0.4 is needed for 10%
W ðST þ IRÞI
efficiency. For a larger ZT, max tends to saturate,
Taking d/dI ¼ 0, we obtain the optimized current I2: and 10% corresponds to Z > 3 103 K1 for
T ¼ 300 K in Figure 22, which corresponds to
ST
I2 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð85Þ ZT ¼ 1.8 at 600 K. Third COP of a commercial
R 1 þ ZT – 1
refrigerator is 1.2–1.3, which corresponds to
where T ¼ ðTH þ TC Þ=2. By substituting I2 into , ZT ¼ 3–4. Thus, much improvement in ZT is needed
we find to replace a freon-gas refrigerator.
Thermal Conductivity and Thermoelectric Power of Semiconductors 345
20 3 × 10–3 K–1 metals have very small Seebeck coefficient, while con-
4 × 10–3 K–1 ventional semiconductors have very low conductivity.
Hence, an optimum carrier concentration is evaluated
to be 1019–1020 cm3 that is a typical carrier concen-
10 tration of degenerate semiconductors. Once the carrier
concentration is set to be the optimum value, the only
way to maximize the conductivity is to maximize the
mobility. In fact, the state-of-the-art thermoelectric
0 materials are high-mobility semiconductors.
0 100 200 300 400 500 600
In the case of degenerate semiconductors, the
ΔT (K)
lattice thermal conductivity is predominant to the
Figure 22 The conversion efficiency plotted as a function electron thermal conductivity. The electron thermal
of temperature difference. The cold temperature is set to conductivity is easily evaluated from the conductivity
300 K. Reprinted from Sorell CC, Sugihara S, and Nowotny J
through the Wiedemann–Franz law given by
(eds.) (2005) Materials for Energy Conversion Devices.
Cambridge, UK: Woodhead Publishing: Figure 13.2. Equation (23). Typical conductivity of thermoelectric
materials is 500–1000 S cm1 at room temperature,
1.09.5.2 Thermoelectric Materials which corresponds to el ¼ 0.4–0.8 W mK1. Good
thermoelectric materials show a low thermal conduc-
Thermoelectric materials are characterized by ZT, tivity of 2–3 W mK1, so that it has been a central
that is, they show large Seebeck coefficient, high con- issue how to reduce the lattice thermal conductivity.
ductivity, and low thermal conductivity at the same Figure 24 shows ZT for various thermoelectric
time. Such requirements are difficult to be satisfied, materials (Mahan, 1997). Thermoelectric materials
because the three parameters are functions of carrier so far used for practical applications are Bi2Te3,
concentration, which cannot be tuned independently. PbTe, and Si1xGex. They show ZT 1 at optimum
Figure 23 schematically shows how the three para- temperatures. n-Type BiSb is superior at low
meters depend on carrier concentration (Snyder and temperatures, but has no p-type counterpart. Bi2Te3
Toberer, 2008). The Seebeck coefficient S decreases shows the highest performance near room tempera-
with carrier concentration n, whereas the conductivity ture, and is used for Peltier coolers commercially
available. PbTe shows the highest performance near
1 700–800 K, and has been used for thermoelectric
S σ κ power generators operating at moderately high
ZT temperatures. Si and Ge are poor thermoelectric
materials at room temperature because of the high
Parameters (a. u.)
0.5 at maximum.
An experimental manifestation of PGEC is seen in
the filled skutterudite CexFe3CoSb12 (Sales et al.,
p-type 1997). This is the first unambiguous example of
0 ZT >1, which is a prototypical success in the material
(b) BiSb Si1–x Gex search in the 1990s. The unit cell of the unfilled
1 Bi2Te3
PbTe skutterudite CoSb3 consists of the eight subcells
whose corners are occupied by Co atoms. Six subcells
out of the eight are filled with Sb plackets, forming the
ZT
Ba8Ni6Ge40
Ba8Zn8Ge38 Type 1
2.0
Ba8Cu6Ge40 Ba8Al16Si30 clathrates
κ L (150 K) (W K−1m−1)
Ba8Ga16Ge30 Cs8Zn4Sn42
Cs8Sn44
1.5 Ba8Ga16Si30
Rb8Zn4Sn42
Rb8Ga16Sn38
Sr8Ga16Si30
Sr8Ga16Ge30
1.0
Ba8In16Ge30
n1 Ba8In16Ge28Sn2
β-Ba8Ga16Sn30 Eu8Ga16Ge30
p1
0.5
Figure 25 The crystal structure of the clathrate compound 1.2 1.3 1.4 1.5 1.6 1.7
A6Ga16Ge30. The large dots correspond to A ions, and the Guest free space (Å)
Ga and Ge ions form the face-shared polyhedral cages.
Figure 27 The thermal conductivity as a function of spacing
Reproduced from Suekuni K, Avila MA, Umeo K, et al. (2008)
in the rattling site. From Suekuni K, Avila MA, Umeo K, et al.
Simultaneous structure and carrier tuning of dimorphic
(2008) Simultaneous structure and carrier tuning of dimorphic
clathrate Ba8Ga16Sn30. Physical Review B 77: 235119:
clathrate Ba8Ga16Sn30. Physical Review B 77: 235119.
Figure 1(b).
x=0
1.09.5.3 Oxide Thermoelectrics
κ L (W K–1 m–1)
Ca3Co4O9 (Funahashi et al., 2000), (Bi,Pb)2Sr2Co2O8 As an origin of the large Seebeck coefficient,
(Itoh and Terasaki, 2000), TlSr2Co2Oy (Hébert et al., Koshibae et al. (2000) proposed an extended Heikes
2001), and (Pb,Co)Sr2Co2Oy (Maignan et al., 2002) formula for transition-metal oxides:
have been found to show good thermoelectric
performance. Some single crystals show even kB gA p
S¼ ln ð92Þ
ZT >1 at 1000 K. As shown in Figure 28, the Q gB 1 – p
CdI2-type hexagonal CoO2 layer is common to where gA and gB are the degeneracy of the electron
these cobalt oxides, which should be a key ingredient configuration of A and B ions, Q is the charge differ-
for the unusually high thermoelectric performance of ence between A and B ions, and p is the atomic
the layered cobalt oxides. content of the A ion. Since
Figure 29 shows the thermoelectric parameters of
the three kinds of the layered cobalt oxides. In the gA p
kB ln ð93Þ
upper panel the resistivity of the three cobalt oxides gB 1 – p
along the in-plane direction is shown. The most- is equal to the entropy per carrier, Equation (92) is a
conducting compound is NaxCoO2, where a low
special case of Equation (56).
of 200 m cm at 300 K decreases in a metallic-like Let us apply the above formula to NaCo2O4.
fashion with decreasing temperature down to 1.5 K. Assuming that Na and O exist as Naþ and O2 in
This is as conductive as the superconducting Cu NaCo2O4, we expect that Co ions exist as Co3þ and
oxides, meaning that the layered cobalt oxide is one Co4þ with a ratio of Co3þ:Co4þ¼1:1. Then, p for
of the most conductive layered oxides. The other two NaCo2O4 is equal to 0.5, and S for p 0.5 is simply
compounds show fairly large
of 1–10 m cm at 300 K reduced to
with an upturn below about 50 K. The Seebeck coeffi-
cient S of the three cobalt oxides along the in-plane kB gA
S¼ ln ð94Þ
direction is shown in the middle panel. The magnitude Q gB
of S is as large as 100–150 mV K1 at 300 K, being
comparable with that of conventional thermoelectric Magnetic measurements reveal that the Co4þ and
semiconductors. The thermal conductivity of the three Co3þ ions are in the low spin state in NaCo2O4.
cobalt oxides along the in-plane direction is shown in The configuration of the low-spin-state Co3þ is
the lower panel. Clearly, a more complicated block (t2g)6, whose entropy is zero. On the other hand,
layer shows lower , implying that is predominatly the low-spin-state Co4þ has a hole in the t2g states,
determined by the block layer. High was thus which is sixfold degenerate (two from spin and
far expected for oxides because of the light mass of three from t2g orbitals) to carry large entropy of
oxygen, which is not always true in the layered kB ln6. Suppose that electric conduction occurs by
cobalt oxides. exchanging Co3þ and Co4þ. Then, a hole on Co4þ
Co
Co SrO
BiO
CaO
CaO BiO
Co Co
Figure 28 Crystal structures of the layered cobalt oxides. Reprinted from Sorell CC, Sugihara S, and Nowotny J (eds.) (2005)
Materials for Energy Conversion Devices. Cambridge, UK: Woodhead Publishing: Figure 13.8, and Narlikar AV (ed.) (2005)
Layered Cobalt oxides as a thermoelectric material in Frontiers in magnetic Materials. pp 327–346. Springer: Berlin: Figure 5.
Thermal Conductivity and Thermoelectric Power of Semiconductors 349
100
observed photoemission spectra quantitatively
reproduces the experimental results. Ishida et al.
50 NaxCoO2
(2007) measured the temperature dependence of
the chemical potential from the photoemission
0
spectra, and found that the electronic states gradu-
8 (c) ally change from a high-temperature localized
NaxCoO2
picture to a low-temperature itinerant picture.
6
κ (W mK–1)
S2 S2
ZT ðxÞ ¼ ¼ ð95Þ
=xðxel þ ph Þ
ðel þ ph =xÞ
6.0
This corresponds to the fact that the lattice
thermal conductivity is effectively reduced by a
factor of x. 4.0
Z20T
Imfp (Å) ( , , )
BiSbTe3 alloy
KL (W/m-K) ( ,
400 300 K kz
–S (μV K–1)
300
2 ML
200 kb
8 ML 16 ML
100 ky k2
4 ML
k1
0
0 1 2 3 4 5 6 7
Figure 33 Corresponding wave vectors in the k space, k1,
Sr(Ti0.8Nb0.2)O3 layer thickness (nm) k2, and kb correspond to cross-sectional planes and kf is the
radius of the Fermi sphere. V1 is the volume of the electrons
Figure 32 The Seebeck coefficient of Sr(Ti0.8Nb0.2)O3/ that participate in thermionic emission above the barrier if
SrTiO3 superlattice films as a function of thickness of the lateral momentum is conserved. V2 is that volume if the
Sr(Ti0.8Nb0.2)O3 layer at 300 K. From Ohta H, Kim S, Mune Y, lateral momentum is not conserved. Reproduced from
et al. (2007) Giant thermoelectric Seebeck coefficient of a Vashaee D and Shakouri A (2004) Electronic and
two-dimensional electron gas in SrTiO3. Nature Materials 6: thermoelectric transport in semiconductor and metallic
129–134. superlattices. Journal of Applied Physics 95: 1233–1245.
352 Thermal Conductivity and Thermoelectric Power of Semiconductors
neighboring layer to get a lateral momentum k2. is constant everywhere in the device. Under such
Then, the momentum space to conserve total conditions, all the wells are in thermal equilibrium,
momentum is limited in V1, while the space to and the electron transfer is dissipationless.
allow the violation of the momentum conservation
is enlarged to V2. As a result, such momentum-
nonconserving filters will remain conductivity as 1.09.6.2 Nanowires
high as the bulk value, and enhance the Seebeck
Hicks and Dresselhaus (1993b) performed the same
coefficient significantly.
calculation for 1D wires as they did for 2D films and
Zide et al. (2006) experimentally demonstrated
showed that ZT is enhanced with decreasing
the enhancement of the Seebeck coefficient in
diameter of the wire. They have further found
the superlattice. Figure 34 shows the in-plane and
that the Bi wires change their ground state from
cross-plane Seebeck coefficients in InGaAs/
semimetal to semiconductor, because the quantum
InAlGaAs superlattices. Normally, Seebeck
confinement lifts the bottom of the conduction band
coefficients are weakly anisotropic because the above the top of the valence band. Figure 35 shows
anisotropic factor is cancelled in the calculation (see the electronic band diagram of Bi as a function of
Equation (49)). Contrary to this, the observed values wire diameter (Lin et al., 2000). For a diameter as
are unconventionally anisotropic: the cross-plane data large as 200 nm, the band of the holes at the T point is
are three times larger than the in-plane ones, which is above electrons at the L point. The energy shifts to
essentially independent of doping levels. higher values with respect to electron doping, and
Humphrey and Linke (2005) proposed a new electrons and holes coexist up to 40 meV. The energy
design of thermoelectric superlattice, and showed levels of the L electrons increase roughly in inverse
that the conversion efficiency of this device reaches proportion to the wire diameter, while the energy
the Carnot efficiency, which corresponds to ZT ! 1. levels of the T holes rapidly decrease below 50 nm.
Their device consists of many quantum-wells, and a As a result, a transition from semimetal to semicon-
small temperature gradient is applied in accordance ductor occurs below 49 nm. Heremans et al. (2000)
with implemented variation of chemical potential in have fabricated Bi nanowire bundles and clarified
order that the electron distribution function that the electronic states of Bi nanowire can be
regarded as single-band semiconductor.
–1 As discussed in Section 1.09.3, silicon is a good
" – ðrÞ conductor of heat, and hence the thermoelectric
f0 ¼ exp þ1 ð98Þ
kB T ðrÞ figure of merit ZT remains a low value of 0.005 at
600 100
80
500
Seebeck coefficient (μV K–1)
Le− (A)
Cross-plane 60 Le− (B, C) T holes
Energy (meV)
400
40
300 20 −Δ0 = 38
49.0 nm
200 In-plane 0
L holes EgL = 15
−20
100
−40
0 0 50 100 150 200
0 2 4 6 8 10 Wire diameter (nm)
Carrier concentration (1018cm–3)
Figure 35 Calculated energy band diagram of Bi wire
Figure 34 The in-plane and cross-plane Seebeck plotted as a function of wire diameter. Below 49 nm, the
coefficient in the InGaAs/InGaAlAs superlattice films. From energy level of the holes at T point goes down below the
Zide JMO, Vashaee D, Bian ZX, et al. (2006) Demonstration energy level of the electrons at L point. Reproduced from Lin
of electron filtering to increase the Seebeck coefficient in Y-M, Sun X, and Dresselhaus MS (2000) Theoretical
In0.53Ga0.47As/In0.53Ga0.28Al0.19 As superlattices. Physical investigation of thermoelectric transport properties of
Review B 74: 205335. cylindrical Bi nanowires. Physical Review B 62: 4610–4623.
Thermal Conductivity and Thermoelectric Power of Semiconductors 353
10–7
10–8
3000
κ (T) (W/m K)
2000
10–9 1000
0
100 200 300
T (K)
10 100
Temperature (K)
Figure 37 Themal conductivity of a single carbon nanotube. The sample mount is shown in the inset. The lower inset shows
the thermal conductivity of carbon nanotubes with different diameters. The solid, dashed, and dotted curves correspond to
diameters of 14.80, and 200 nm, respectively. Reproduced from Kim P, Shi L, Majumdar A, and McEuen PL (2001) Thermal
transport measurements of individual multiwalled nanotubes. Physical Review Letters 87: 215502.
Keppens V, Mandrus D, Sales BC, et al. (1998) Localized Murata N, Haruyama J, Reppert J, et al. (2008)
vibrational modes in metallic solids. Nature 395: 876–878. Superconductivity in thin films of boron-doped carbon
Kim P, Shi L, Majumdar A, and McEuen PL (2001) Thermal nanotubes. Physical Review Letters 101: 027002.
transport measurements of individual multiwalled Nagao Y, Terasaki I, and Nakano T (2007) Dielectric constant
nanotubes. Physical Review Letters 87: 215502. and acconductivity in the layered cobalt oxide Bi2Sr2CoO6þd.
Kim W, Zide J, Gossard A, Klenov D, Stemmer S, Shakouri A, Physical Review B 76: 144203.
and Majumdar A (2006) Thermal conductivity reduction and Narlikar AV (ed.) (2005) Layered Cobalt oxides as a
thermoelectric figure of merit increase by embedding thermoelectric material in Frontiers in magnetic Materials.
nanoparticles in crystalline semiconductors. Physical Review pp 327–346. Spirnger: Berlin.
Letters 96: 045901. Nam M-S, Ardavan A, Blundell SJ, and Schlueter JA (2007)
Klemens PG (1960) Thermal resistance due to point defects at Fluctuating superconductivity in organic molecular metals
high temperatures. Physical Review 119: 507–509. close to the Mott transition. Nature 449: 584–587.
Kobayashi W, Satake A, and Terasaki I (2002) Thermoelectric Niemeijera Th and Vianen HAW (1971) A note on the thermal
properties of the Brownmillerite Ca1yLayCo2xAlxO5. conductivity of linear magnetic chains. Physics Letters A
Japanese Journal of Applied Physics 41: 3025–3028. 34: 401–402.
Kobayashi W and Terasaki I (2006) Thermoelectric properties of Nolas GS, Cohn JL, Slack GA, and Schujman SB (1998)
Pb–Sr–Co–O single crystals. Applied Physics Letters Semiconducting Ge clathrates: Promising candidates for
89: 072109. thermoelectric applications. Applied Physics Letters
Koshibae W, Tsutsui K, and Mackawa S (2000) Thermopower in 73: 178–180.
cobalt oxides. Physical Review B 62: 6869–6872. Ohta S, Nomura T, Ohta H, and Koumoto K (2005) High-
Kuroki K and Arita R (2007) Pudding Mold band drives large temperature carrier transport and thermoelectric properties
thermopower in NaxCoO2. Journal of the Physical Society of of heavily La- or Nb-doped SrTiO3 single crystals. Journal of
Japan 76: 083707. Applied Physics 97: 034106.
Koumoto K, Terasaki I, and Funahashi R (2006) Complex oxide Ohta H, Kim S, Mune Y, et al. (2007) Giant thermoelectric
materials for potential thermoelectric applications. MRS Seebeck coefficient of a two-dimensional electron gas in
Bulletin 31: 206–210. SrTiO3. Nature Materials 6: 129–134.
Kramer B and MacKinnon A (1993) Localization: Theory Ohtaki M, Tsubota T, Eguchi K, and Arai H (1996) High-
and experiment. Reports on Progress in Physics temperature thermoelectric properties of (ZnAl)O. Journal of
56: 1469–1564. Applied Physics 79: 1816–1818.
Kudo K, Ishikawa S, Noji T, et al. (2001) Spin gap and hole Ohtaki M, Hayashi R, and Araki K (2007) Thermoelectric
pairing in the spin-ladder cuprate Sr14xAxCu24O41 properties of sintered ZnO incorporating nanovoid structure:
(A¼Ca and La) studied by the thermal conductivity. Influence of the size and number density of nanovoids.
Journal of the Physical Society of Japan 70: 437–444. Proceedings of the 26th International Conference on
Kurosaki K, Kosuga A, Muta H, Uno M, and Yamanaka S (2005) Thermoelectrics, 3–7 June, 2007, pp.112–116, Jeju.
Ag9TITe5: A high-performance thermoelectric bulk material Okada S and Terasaki I (2005) Physical properties of Bi-based
with extremely low thermal conductivity. Applied Physics rhodium oxides with RhO2 hexagonal layers. Japanese
Letters 87: 061919. Journal of Applied Physics 44: 1834–1837.
Lander JJ (1948) Measurements of Thomson coefficients for Okada S, Terasaki I, Okabe H, and Matoba M (2005) Transport
metals at high temperatures and of Peltier coefficients for properties and electronic states of the layered rhodium oxide
solid–liquid interfaces of metals. Physical Review 74: 479–488. (Bi1 xPbx)1.8Ba2Rh1.9Oy. Journal of the Physical Society of
Li SS and Rabson TA (1970) The Nernst and the Seebeck Japan 74: 1525–1528.
effects in Te-doped Bisingle bond Sb alloys. Solid State Okuda T, Nakanishi K, Miyasaka S, and Tokura Y (2001) Large
Electronics 13: 153–160. thermoelectric response of metallic perovskites:
Lin Y-M, Sun X, and Dresselhaus MS (2000) Theoretical Sr1 xLaxTiO3 (0 < x < 0.1). Physical Review B 63: 113104.
investigation of thermoelectric transport properties of Palstra TTM, Ramirez AP, Cheong S-W, Zegarski BR,
cylindrical Bi nanowires. Physical Review B 62: 4610–4623. Schiffer P, and Zaanen J (1997) Transport mechanisms in
Mahan GD (1997) Good thermoelectrics. Solid State Physics doped LaMnO3: Evidence for polaron formation. Physical
51: 81–157. Review B 56: 5104–5107.
Mahan GD, Sofo JO, and Bartkowiak M (1998) Multilayer Poudel B, Hao Q, Ma Y, et al. (2008) High-thermoelectric
thermionic refrigerator and generator. Journal of Applied performance of nanostructured bismuth antimony telluride
Physics 83: 4683–4689. Bbulk alloys. Science 320: 634–638.
Maignan A, Hébert S, Pelloquin D, Michel C, and Hejtmanek J Rosenbaum TF, Andres K, Thomas GA, and Bhatt RN (1980)
(2002) Thermopower enhancement in misfit cobaltites. Sharp metal-insulator transition in a random solid. Physical
Journal of Applied Physics 92: 1964–1967. Review Letters 45: 1723–1726.
Meisner GP, Morelli DT, Hu S, Yang J, and Uher C (1998) Rowe DM (2006) Thermoelectrics Handbook: Macro to Nano.
Structure and lattice thermal conductivity of fractionally filled Boca Raton, FL: CRC Press.
skutterudites: Solid solutions of fully filled and unfilled end Sales BC, Mandrus D, Chakoumakos BC, Keppens V, and
members. Physical Review Letters 80: 3551–3554. Thompson JR (1997) Filled skutterudite antimonides:
Motoyama N, Eisaki H, and Uchida S (1995) Magnetic Electron crystals and phonon glasses. Physical Review B
susceptibility of ideal spin 1/2 Heisenberg antiferromagnetic 56: 15081–15089.
chain systems, Sr2CuO3 and SrCuO2. Physical Review Sales BC, Lumsden MD, Nagler SE, Mandrus D, and Jin R
Letters 76: 3212–3215. (2002) Magnetic field enhancement of heat transport in the
Mott NF (1972) Conduction in non-crystalline systems IX: The 2D Heisenberg antiferromagnet K2V3O8. Physical Review
minimum metallic conductivity. Philosophical Magazine Letters 88: 095901.
26: 1015–1026. Sales BC, Jin R, and Mandrus D (2004) Thermal spin valve.
Mühlenen Av, Errien N, Schaer M, Bussac M-N, and Zuppiroli L Proceedings of the 27th International Thermal Conductivity
(2007) Thermopower measurements on pentacene Conference and the 15th International Thermal Expansion
transistors. Physical Review B 75: 115338. Symposium. DEStech Publications Lancaster, PA.
358 Thermal Conductivity and Thermoelectric Power of Semiconductors
Satake A, Tanaka H, Ohkawa T, Fujii T, and Terasaki I (2004) Takeuchi T, Kondo T, Takami T, et al. (2004) Contribution of
Thermal conductivity of the thermoelectric layered cobalt electronic structure to the large thermoelectric power in
oxides measured by the Harman method. Journal of Applied layered cobalt oxides. Physical Review B 69: 125410.
Physics 96: 931–933. Terasaki I, Sasago Y, and Uchinokura K (1997) Large
Sharp JW, Poon SJ, and Goldsmid HJ (2001) Boundary thermoelectric power in NaCo2O4 single crystals. Physical
scattering and the thermoelectric figure of merit. Physica Review B 56: R12685–R12687.
Status Solidi (a) 187: 507–516. Terasaki I and Nonaka T (1999) Thermoelectric properties of
Singh DJ (2000) Electronic structure of NaCo2O4. Physical Sr1xLaxPbO3 (x < 0.02). Journal of Physics: Condensed
Review B 621: 13397–13402. Matter 11: 5577–5582.
Singh DJ and Pickett WE (1994) Skutterudite antimonides: Terasaki I, Tanaka H, Satake A, Okada S, and Fujii T (2004)
Quasilinear bands and unusual transport. Physical Review B Out-of-plane thermal conductivity of the layered
50: 11235–11238. thermoelectric oxide Bi2xPbxSr2Co2Oy. Physical Review B
Slack GA (1964) Thermal conductivity of pure and impure 70: 214106.
silicon, silicon carbide, and diamond. Journal of Applied Terasaki I, Kobayashi W, and Ishiwata S (2006) Nano-block
Physics 35: 3460–3466. integration in the misfit-layered oxides. Proceedings of the
Slack GA (1979) The thermal conductivity of nonmetallic 25th International Conference on Thermoelectrics, 6–10
crystals. Solid State Physics 34: 1–71. August 2006, pp. 283–286, Vienna.
Slack GA (1995) New materials and performance limits for Tritt TM (ed.) (2004) Thermal Conductivity. Kluwer.
thermoelectric cooling. In: Rowe DM (ed.) CRC Handbook of Vashaee D and Shakouri A (2004) Electronic and thermoelectric
Thermoelectrics pp. 407–440, Boca Raton: CRC–Press. transport in semiconductor and metallic superlattices.
Slack GA and Galginaitis S (1964) Thermal conductivity and Journal of Applied Physics 95: 1233–1245.
phonon scattering by magnetic impurities in CdTe. Physical Venkatasubramanian R (2000) Lattice thermal conductivity
Review 133: A253–A268. reduction and phonon localization like behavior in
Snyder GJ and Toberer ES (2008) Complex thermoelectric superlattice structures. Physical Review B 61: 3091–3097.
materials. Nature Materials 7: 105–114. Venkatasubramanian R, Siivola E, Colpitts T, and OQuinn B
Sologubenko AV, Felder E, Giannò K, et al. (2000a) Thermal (2001) Thin-film thermoelectric devices with high room-
conductivity and specific heat of the linear chain cuprate temperature figures of merit. Nature 413: 597–602.
Sr2CuO3: Evidence for thermal transport via spinons. Viennois R, Girard L, Ravot D, et al. (2004) Inelastic neutron
Physical Review B 62: R6108–R6111. scattering experiments on antimony-based filled
Sologubenko AV, Giannó K, Ott HR, Ammerahl U, and skutterudites. Physica B 350: 403–405.
Revcolevschi A (2000b) Thermal conductivity of the Xu ZA, Ong NP, Wang Y, Kakeshita T, and Uchida S (2000)
hole-doped spin ladder system Sr14 xCaxCu24O41. Physical Vortex-like excitations and the onset of superconducting
Review Letters 84: 2714–2717. phase fluctuation in underdoped La2 xSrxCuO4. Nature
Sorell CC, Sugihara S, and Nowotny J (eds.) (2005) Materials 406: 486–488.
for Energy Conversion Devices. Cambridge, UK: Woodhead Yang J (2004) Theory of thermal conductivity in crystalline
Publishing. semiconductors. In: Trit TM (ed.) Thermal Conductivity.
Steigmeier EF and Abeles B (1964) Scattering of phonons by Kluwer.
electrons in Germanium–Silicon alloys. Physical Review Weber L and Gmelin E (1991) Transport properties of silicon.
136: A1149–A1155. Applied Physics A 53: 136–140.
Suekuni K, Avila MA, Umeo K, and Takabatake T (2007) Zerec I, Keppens V, McGuire MA, Mandrus D, Sales BC, and
Cage-size control of guest vibration and thermal Thalmeier (2004) Four-well tunneling states and elastic
conductivity in Sr8Ga16Si30xGex. Physical Review B response of clathrates. Physical Review Letters 92: 185502.
75: 195210. Zide JMO, Vashaee D, Bian ZX, et al. (2006) Demonstration of
Suekuni K, Avila MA, Umeo K, et al. (2008) Simultaneous electron filtering to increase the Seebeck coefficient in
structure and carrier tuning of dimorphic clathrate In0.53Ga0.47As/In0.53Ga0.28Al0.19 As superlattices. Physical
Ba8Ga16Sn30. Physical Review B 77: 235119. Review B 74: 205335.
Takahata K, Iguchi Y, Tanaka D, Itoh T, and Terasaki I (2000) Ziman JM (1976) Chapter 7 in Principles of the Theory of Solids,
Low thermal conductivity of the layered oxide (Na, 2nd ed. Cambridge: Cambridge University Press.
Ca)Co2O4: Another example of a phonon glass and an Zotos X, Naef F, and Prelovsek P (1997) Transport and
electron crystal. Physical Review B 61: 12551–12555. conservation laws. Physical Review B 55: 11029–11032.
Takahata K and Terasaki I (2002) Thermal conductivity of Zutic I, Fabian J, and Sarma SD (2004) Spintronics:
AxBO2-type layered oxides Na0.77MnO2 and LiCoO2. Journal Fundamentals and applications. Reviews of Modern Physics
of the Physical Society of Japan 41: 763–764. 76: 323–410.
1.10 Electronic States and Transport Properties of Carbon
Crystalline: Graphene, Nanotube, and Graphite
Y Iye, University of Tokyo, Kashiwa, Chiba, Japan
ª 2011 Elsevier B.V. All rights reserved.
359
360 Electronic States and Transport Properties of Carbon Crystalline
production of fullerene molecules by an arc dis- structures. Under appropriate conditions, fullerenes
charge method (Dresselhaus et al., 1996). can form dimers, trimers, and various polymers. One
The arc discharge method yielded another, no less of the most beautiful allotropes is carbon nanopea-
important, family of allotropes, that is, carbon nanotubes. pods, that is, carbon nanotubes enclosing C60
Iijima (1991), while investigating under an electron molecules (Smith et al., 1998).
microscope the soot deposits on the cathode of his arc Exploring further beyond the boundary of pure
discharge apparatus, discovered thin tubes made of gra- carbon territory, one finds a vast landscape of mate-
phitic layers. Although one can spot in retrospect a few rials diversity. By using a metal-containing carbon
earlier reports on similar substances (Oberlin et al., 1976), rod in the arc discharge process, metallofullerenes,
it was the report by Iijima that triggered the subsequent that is, fullerene molecules encapsulating metal
research activities on carbon nanotubes (Saito et al., 1998; ion(s), can be produced. Nanotubes can accommo-
Dresselhaus et al., 2001). Whereas the initial discovery date various foreign species (including fullerenes as
was on multi-walled nanotubes, single-walled nano- mentioned above). Extensive studies have been made
tubes were later prepared successfully (Iijima and on graphite intercalation compounds (GICs)
Ichihashi, 1993; Bethune et al., 1993). (Dresselhaus and Dresselhaus, 1981). A wide variety
A further surprise was brought about in 2004 by the of guest species (intercalant) can be inserted in the
Manchester University group led by Geim (Novoselov interlayer spacing of graphite in a well-ordered fash-
et al., 2004; Geim and MacDonald, 2007; Geim and ion. The resulting GICs are good examples of
Novoselov, 2007). By an astonishingly simple peel- synthetic metals.
and-stick technique, they prepared samples of a sin-
gle-layer sheet of graphite on an oxidized surface of
silicon wafer. Although cleavage was routinely done by 1.10.1.2 Bond and Local Coordination
those who handled graphite samples, it came as a total
surprise that samples of single-layer graphite of semi- Carbon atom has four valence electrons in the 2s and
macroscopic size can be obtained by such a method. 2p atomic orbitals. The carbon–carbon (C–C) bond
The discovery of these novel species of carbon can be either sp3- or sp2-hybridized bond. The rich-
allotropes has vastly broadened the horizon of con- ness of carbon allotropes stems from various
densed matter physics. The exotic electronic combinations of the sp3- and sp2-hybridized bonds.
properties of nanotubes and graphenes and their In diamond crystals, each carbon atom is tetra-
potential applications to future electronics are cur- hedrally coordinated. Strong C–C bond is formed
rently under vigorous study worldwide. with each of the neighboring four atoms by sp3-
The carbon materials listed above and depicted in hybridization. The C–C bond length in diamond is
Figure 1 are representative ones that have been most 0.154 nm. The crystal structure is cubic with lattice
extensively studied. There are many more exotic allo- constant 0.357 nm. The diamond lattice can be
tropes, even limiting ourselves to those purely made of viewed as consisting of two face-centered cubic
carbon atoms. In addition to C60 and C70, there are so- 111
(f.c.c.) sublattices of which one is shifted by
called higher-order fullerenes such as C84. The crys- 444
tals made of the fullerene molecules can take different relative to the other.
Figure 1 Carbon allotropes: diamond, graphite, C60 molecule, and carbon nanotube.
Electronic States and Transport Properties of Carbon Crystalline 361
Carbon atoms in a graphitic sheet form a honey- single electron tunneling (SET) transistor, in which
comb network so that each atom is trigonally tunneling of individual electrons between the
coordinated. Three of the four valence electrons are quantum-dot and the leads can be controlled by a
used for the sp2-hybridized -bonds with the three gate voltage. A single fullerene molecule trapped in
neighbors. The remaining one is in the -orbital and the nano-sized gap between two electrodes can be
can be easily delocalized. The C–C bond length in the an ultimate quantum-dot. Thus, graphene, nano-
graphitic sheet is 0.142 nm. The three-dimensional tube, and fullerene are regarded as wonderful
(3D) graphite is made of a stack of graphitic sheets. playgrounds for nanoscience and as precious parts
The ordinary graphite (Bernal graphite) exhibits the of nanotechnology.
stacking order ABAB. . .. A thermodynamically In the subsequent sections, some of the basics
unstable variant called rhombohedral graphite has a about the carbon-based electronic systems are
stacking sequence ABCABC. . .. explained, and the excitement brought by the
In the case of amorphous carbon, the C–C bonds research efforts since the late 1950s is conveyed.
consist of mixture of sp3- and sp2-hybridized bonds. This chapter mainly focuses on the transport
Actually, one important parameter that characterizes properties.
the amorphous carbon is the ratio of sp3- and
sp2-hybridized bonds. Materials that are high in
sp3-hybridized bonds are referred to as tetrahedral 1.10.2 Electronic Structures
amorphous carbon or as diamond-like carbon owing
1.10.2.1 Diamond
to the similarity of many physical properties to those
of diamond. Raman spectroscopy provides a useful Since this chapter is principally concerned with the
means of characterization. The Raman spectrum of graphitic systems, we only briefly touch upon the
diamond shows a sharp peak at ! ¼ 1332 cm1, while electronic structure of diamond. Diamond is a wide
the frequency of the Raman-active in-plane mode of gap semiconductor. Figure 2(a) shows the band
graphite is ! ¼1582 cm1. The Raman spectra of structure of diamond (Saslow et al., 1966; Saravia and
amorphous carbon generally exhibit broadened mix- Brust, 1968), which is similar to that of silicon. The
ture of these two peaks. conduction-band minima are located along the –X
lines of the Brillouin zone. The surfaces of equal
energy are ellipsoids with m1 ¼ 1.4 m0 and mt ¼ 0.36 m0.
1.10.1.3 Dimensionality
The valence band top is located at the -point. The
The line-up of carbon allotoropes introduced in the effective masses of heavy and light holes
previous section provides a set of model systems for are mhh ¼ 2.12 m0 and m1h ¼ 0.7 m0. The split-off band
low-dimensional physics; that is, diamond (3D), gra- is 6 meV below the heavy and light hole bands
phite (quasi-2D), graphene (2D), nanotube (1D), and and the effective mass is m1h ¼ 1.06 m0. The indirect
fullerene (0D). Dimensionality is a key ingredient gap is 5.47 eV and the direct gap at the -point
that profoundly affects transport and other electro- is 7.3 eV.
nic properties. Indeed, phenomena intimately linked
to the dimensionality have been observed in these
1.10.2.2 2D Graphite (Graphene)
systems. 2D electron system created at the semicon-
ductor heterointerface is an experimental stage of Calculations of the electronic band of 2D version
many intriguing physics, including quantum Hall of graphite (i.e., graphene in today‘s terminology)
effect (QHE) and electron localization. Up until have been made by many authors for over 50
the advent of graphene, the experimental systems years (Wallace, 1947; Coulson, 1947; Bassani and
were virtually limited to GaAs/AlGaAs and Si- Parravicini, 1967; Painter and Ellis, 1970; Zunger,
MOS. Graphene provides not just another 2D sys- 1978). Figure 2(b) shows the calculated band
tem but a very unique one, as explained in later structures of 2D graphite taken from Zunger
sections. Carbon nanotubes are regarded as nearly (1978). The Brillouin zone edges denoted by P
ideal 1D quantum wire. Some of the properties of and Q in the figure correspond to K and M in the
carbon nanotubes seem to corroborate the today’s conventional notation. The s-orbital and
Tomonaga–Luttinger liquid (TLL) behavior theo- two in-plane p-orbitals form the sp2-hybridized
retically developed for interacting 1D electron -bonds which make up the strong honeycomb
systems. Nanotube is also used in a device called framework of graphene. The bonding -bands
362 Electronic States and Transport Properties of Carbon Crystalline
(a)
Δ1
Λ1 Δ2′
k2 20
Γ12′
Λ3 Δ5
l
Λ Λ3 Γ2
Γ k2 L3 Δ2′
10 Γ15
Δ Σ X1
k2 L1 Λ1
K
XZ W Δ1
Γ25′
0
E (eV)
L3′ Λ3 Δ5
X4
–10
Λ1 Δ2′
L1 C
X1
–20
L2′ Δ1
Λ1
Γ1
–30
(b) +
10 (a)
Q2v (b) (c)
+
Γ1v
–
P3+ –
Γ2g Γ2g
–
Γ2g Q2u
+
0 + +
P2 +
Qlu +
Γ3u Qlu
+
Γ3u Γ3u
+
–
Energy (eV)
Q2g
– Q2g +
Q2g
– P3–
–10
P3 P3– +
+
Γ3g – + Q2u
– Γ3g Q2u
–
Q2u Γ3g
+ +
– P1+ Γ2u Q2g
+ Γ2u +
Q2g Q2g
–
Γ2u P1+ +
–20 P1+ Qlu
+ +
+
P3+ + P3 Γlg +
P3+ Qlu Qlu Qlg
+ +
+ Qlg Qlg
Γlg +
Γlg
–30
P Γ Q P Γ Q P Γ Q
Figure 2 The Brillouin zone and electronic band of (a) diamond and (b) two-dimensional graphite. (a) Reprinted with permission
from Saslow W, Bergstresser TK, and Cohen ML (1966) Band structure and optical properties of diamond. Physical Review Letters
16: 354. Copyright (1966) by the American Physical Society. (b) Reprinted with permission from Zunger A (1978) Self-consistent
LCAO calculation of the electronic properties of graphite. I. The regular graphite lattice. Physical Review B 17: 626. Copyright
(1978) by the American Physical Society.
and antibonding 9-bands are shown by three As it constitutes the basis for the electronic struc-
solid curves in the lower part and another three ture of all graphitic systems, let us take a closer look
in the upper part of the band diagrams in at the graphene -band structure. The electronic
Figure 2(b). The dashed curves represent the structure of the graphene can be reasonably well
bonding -band and antibonding 9-band formed described by a simple tight-binding Hamiltonian for
by the remaining out-of-plane p-orbital. The the - and 9-bands. The notations here follow those
bonding -band and antibonding 9-band touch by Ando (2005).
with each other at the K-point, where the Fermi The honeycomb lattice of graphene consists of two
level of the undoped graphene resides. interpenetrating triangular sublatteices (referred to as
Electronic States and Transport Properties of Carbon Crystalline 363
(a) (b)
4
z K
M
y
3
x Γ
2
γ0 EF
0
Monolayer Bilayer
−1
(c)
ε ε −2
−3
κy κy K Γ M K
Wave vector
K K
κx κx
Monolayer Bilayer
Figure 3 (a) Left: Honeycomb lattice of graphene. Right: Stacking of bilayer graphene. (b) The - and 9-bands of graphene.
(c) Characteristic band structures at the K point for monolayer (left) and bilayer (right) graphene.
X X
A and B sublattices hereafter) as depicted in Figure 3(a) ðrÞ ¼ A ðrÞðr RA Þ þ B ðrÞðr – RB Þ ð1Þ
with black and white atoms. Each atom of one sublattice RA RB
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffi at the corners of the Brillouin zone as shown on the
kx a 3ky a kx a
" ðkÞ ¼ 0 1 þ 4cos cos þ 4cos2 ð4Þ left of Figure 3(c). The corresponding density of
2 2 2
states reads
which is shown in Figure 3(b). Note that
" ðkÞ ¼ 0 at the K- and K9-points. In the Dð"Þ ¼
j"j
ð6Þ
vicinity of the K-points, the dispersion is linear; 2h2 vF2
that is,
This linear dispersion in the vicinity of the K-
" ðkÞ ¼ hvF jkjj ð5Þ and K9-points is what makes the graphitic
systems (graphene and nanotube) unique in
for a << 1, where k X k – K is the wave
pffiffiffi vector mea- their electronic properties, as will be seen in the
sured from the K-point and vF ¼ 30 a=2h is the following.
group velocity of the electrons (vF ¼ 1106 m s – 1 Within the effective mass approximation, the
for graphene). The same relation holds for the Schrödinger equation for graphene -band electrons
K9-point. This can be represented as cone structure can be written as
Here, FAK ðrÞ, etc. are the slowly varying envelope which operates in the pseudo-spin space that describes
functions. The above matrix equation can be rewrit- the amplitude of electron waves on the A and B sub-
ten in the following more compact form: lattices. The eigenvalues of this Hamiltonian are given
( by Equation (5). This form is convenient for highlight-
hvF ðs?k̂ ÞFK ðrÞ ¼"FK ðrÞ ing the difference from the bilayer case to be discussed
ð8Þ
hvF ðs ?k̂ ÞFK9 ðrÞ ¼ "FK9 ðrÞ in the next subsection.
Although the concept of massless Dirac fermion in
Here, s ¼ x ;y is the pseudo-spin Pauli matrices solid state is not new but has been invoked in discussing
(the pseudo-spin corresponding to the A and B sub- certain aspects of the electronic properties of graphite
lattice), and k̂ ¼ – ir is the wave number operator. (McClure, 1956) and other systems such as bismuth
Equation (8) is isomorphic to a relativistic Dirac equa- (Fukuyama and Kubo, 1970) and HgTe, demonstration
tion for a particle with vanishing rest mass, known as of its genuine occurrence with all subtleties has been
Wyle‘s equation. In this sense, electrons and holes in first achieved in monolayer graphene. However, occur-
monolayer graphene are often referred to as massless rence of genuine massless Dirac fermions in solid state
Dirac fermions. Here, vF ¼ 1 106 m s – 1 plays the is not exclusive to graphene. Another example recently
role of the speed of light. elucidated is the zero-gap state in a molecular
The quasi-particles in monolayer graphene can be conductor -(bis(ethylenedithia)tetrathiafulvalene
described by the effective Hamiltonian (BEDT-TTF))2I3 under pressure (Tajima et al., 2000),
! whose electronic structure can be described by a tilted
0 ˆ x – i ˆ y
Ĥmono ¼ hvF ð9Þ Weyl equation with anisotropic Fermi velocity
ˆ x þ i ˆ y 0 (Katayama et al., 2006; Kobayashi et al., 2009).
Electronic States and Transport Properties of Carbon Crystalline 365
1.10.2.3 Bilayer Graphene the vertically aligned atoms gives rise to interlayer
coupling that changes the band structure of bilayer
In a graphene bilayer with Bernal (AB) stacking shown
graphene from the monolayer case.
on the right of Figure 3(a), one of the two sublattices
The Schrödinger equation for bilayer graphene
in the top layer comes right above a sublattice in the
(to be contrasted with Equation (7) for the monolayer
bottom layer, while the other sublattice does not find
case) can be written as
such a counterpart. The transfer integral 1 between
K
Ĥbi FK ðrÞ ¼ "FK ðrÞ
FK ðrÞ ¼ ðF1B ðrÞ; F1A ðrÞ; F2A ðrÞ; F2B ðrÞÞ
0 1
0 vF ˆ x þ iˆ y 0 0
B C
B C ð10Þ
B vF ˆ x iˆ y 0 0 C
K B 1 C
Ĥbi ¼ B
B
C
C
B 0 0 v ˆ
þ i ˆ
C
B 1 F x y C
@ A
0 0 vF x iy
ˆ ˆ 0
ð10Þ
together with a similar expression for the K9 valley.
This results in four valley-degenerate bands as estimated as m 0.034m. In contrast to the monolayer
shown on the right of Figure 3(c). A symmetric case, electrons and holes in bilayer graphene act as massive
pair of the higher penergy Dirac fermions as shown on the right of Figure 3(c).
ffiffiffiffiffiffiffi bands are separated by a
gap of magnitude 21 at ¼ 0, which originates
from the interlayer coupling 1 . The effective
Hamiltonian for the lower energy band pair relevant
1.10.2.4 Single-Wall Carbon Nanotubes
to the transport and other low energy properties can
be written as Single-wall nanotubes (SWNTs) are formed by roll-
0 1 ing up a sheet of graphene into a nanometer-sized
vF2 2 cylinder. Electronic motion along the circumference
B 0 ˆ
x – iˆ y C
B 1 C direction is then subjected to a periodic boundary
Ĥbi ¼ B C ð11Þ
@ v2 2 A condition that depends on how the sheet is rolled up.
F
ˆ x þ iˆ y 0 The way a graphene sheet is rolled up to make
1
SWNT is specified by a lattice translation vector,
which is to be contrasted with Equation (9) for the called chiral vector,
monolayer case. The energy eigenvalues of Equation
(11) are pffiffiffi
1 3
L ¼ n1 a1 þ n2 a2 ¼ n1 ð1; 0Þ þ n2 – ;
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! 2 2
1 2 1 pffiffiffi ð13Þ
"ðkÞ ¼ þ ðhvF Þ – 1 2 1 3
4 1 2 ¼ n1 – n2 ; n2
ð12Þ 2 2
ðhvF Þ2 2
as shown in Figure 4. The hexagons separated by L
12
in the original graphene sheet are rolled onto
Thus, the low energy bands of bilayer graphene are one another in the SWNT. The length of L is
quadratic and meets at k ¼ 0. This dispersion relation expressed as
can be written in the form of a massive
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
electron "ðkÞ ¼ h2 2 =2m , with the effective mass jLj ¼ jn1 a1 þ n2 a2 j ¼ n21 ja1 j2 þn22 ja2 j2 þ2n1 n2 a1 ? a2
m ¼ 1 =2vF2 , as opposed to the linear dispersion in the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
monolayercase.Usingavalue 1 0.39 eVtakenfrombulk ¼ a n21 þ n22 – n1 n2 ,a1 ?a2 ¼ – a2 =2
graphite and vF 1106ms1, the effective mass is ð14Þ
366 Electronic States and Transport Properties of Carbon Crystalline
Figure 4 Construction of a carbon nanotube from a where X ðn1 þ n2 Þðmod3Þ is an integer that takes a
graphene sheet. Chiral vector. Reproduced with permission value 0 or 1. If ¼ 0, the SWNT is metallic
from Ando T (2005) Theory of electronic states and transport because it has two bands crossing at the K- and
in carbon nanotubes. Journal of the Physical Society of
K9-points without gap. If ¼ 1, on the other
Japan 74: 777.
hand, there is a non-zero-gap so that the SWNT is
semiconducting.
An SWNT with a nonspecial L possesses a
helical (chiral) structure. There are two special
Since the chiral vector L becomes the circumference, categories of nonchiral SWNT: zigzag nanotubes
the diameter of the SWNT is with ðn1 ; n2 Þ ¼ ðm; 0Þ and armchair nanotubes
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi with ðn1 ; n2 Þ ¼ ð2m; mÞ. According to the above
L a n12 þ n22 – n1 n2 criterion for metallicity, zigzag nanotubes are
d¼ ¼ ð15Þ
metallic when m is devisible by 3 and semicon-
Thus, every SWNT is specified by a set of two ducting otherwise. By contrast, all armchair
integers ðn1 ;n2 Þ. The primitive translation vector T nanotubes are metallic because n1 þ n2 ¼ 3m.
along the length of nanotube is written as The periodic boundary conditions for the envel-
ope function are
T ¼ m1 a1 þ m2 a2 ð16Þ
The condition L ? T ¼ 0 reads 2
FK ðr þ LÞ ¼ e – iK?L FK ðrÞ ¼ exp – i FK ðrÞ
3
ð20Þ
m1 ð2n1 – n2 Þ þ m2 ð2n2 – n1 Þ – 0 ð17Þ 2
K9 – iK9?L K K9
F ðr þ LÞ ¼ e F ðrÞ ¼ exp i F ðrÞ
3
which can be solved as
The extra phase appearing here as an electron travels
n1 – 2n2 2n1 – n2
m1 ¼ and m2 ¼ ð18Þ around the cylinder corresponds to a fictitious
P P
magnetic flux of magnitude ¼ – ð =3Þ0 for the
where p is the greatest common divisor of n1 – 2n2 K-point and ¼ ð =3Þ0 for the K9-point (0 ¼ h=e
and 2n1 – n2 . The unit cell of an ðn1 ;n2 Þ nanotube is being the flux quantum) piercing the tube.
the rectangular region defined by T and L. The envelope wave function is given by a plane
The electronic bands of an SWNT is obtained wave FK ðrÞ_ exp – ikx x þ iky y . The kx -component
by imposing a periodic boundary condition is quantized to
Electronic States and Transport Properties of Carbon Crystalline 367
example, an armchair ribbon becomes metallic or - and 9-bands meet at the K- and K9-points.
semiconducting according to whether the width N Interlayer coupling gives rise to a finite overlap of
of the ribbon is N ¼ 3m – 1 ðm ¼ 1;2;3; Þ or the - and 9-bands, so that 3D graphite is a semi-metal.
otherwise. One notable difference of nanoribbons The first Brillouin zone of graphite is a hexagonal
from nanotubes lies in the existence of edge states. prism shown in Figure 6. The -band structure of
In particular, the band calculations predict that a graphite in the vicinity of "F, that is, near the H–K–H
zigzag ribbon has a flat (dispersionless) band at zone edge, is described by the Slonczewski–Weiss–
" ¼ 0, so that the density of states has a sharp peak McClure (SWMcC) model (McClure, 1957;
at the Fermi level (Fujita et al., 1996). This is in Slonczewski and Weiss, 1958):
marked contrast with the bulk graphene whose den-
sity of states is zero at " ¼ 0. This flat band is due to 0
1
E1 0 H13 H13
the edge states constructed from the nonbonding B C
orbitals of the atoms along the zigzag edge. Since B0 E2 C
H23 – H23
B C
H ¼B C ð25Þ
the weight of the edge states relative to the bulk B H13
H23 E3 H33 C
@ A
states increases with decreasing ribbon width, the
H13 – H23 H33 E3
former plays an important role in nanometer-sized
ribbons. For example, the large density of states at the where
Fermi level would lead to a large Pauli paramagnetic
contribution to the magnetic suseptibility and may E1 ¼ þ 21 cosðkz c Þ þ 25 cos2 ðkz c Þ
even give rise to the so-called flat-band E2 ¼ 21 cosðkz c Þ þ 25 cos2 ðkz c Þ
ferromagnetism.
E3 ¼ 22 cos2 ðkz c Þ
pffiffiffi
6 ð26Þ
H13 ¼ ð – 0 þ 24 cosðkz c ÞÞeij a
4
1.10.2.6 Bulk Graphite pffiffiffi
6
The global electronic structure of graphite is H23 ¼ ð0 þ 24 cosðkz c ÞÞeij a
4
given by the 2D graphite model described in pffiffiffi
H23 ¼ 33 cosðkz c Þeij a
Section 1.10.2.2. Graphite, however, is a 3D system
consisting of ABAB stacking of graphene sheets with
the interlayer spacing c ¼ 0.335 nm. For the detailed If the parameter 3 which describes the trigonal
band structure near the Fermi level, interlayer coupling warping is neglected, the SWMcC model yields
plays an important role. In the case of 2D graphite, the the following four eigenvalues:
(a) (b)
E
A H
H′
H′ H K E1
Γ
K′ M K′
K
A H
H′ H′
H L
0 E3
EF
2γ2
E2
K K H H
Figure 6 The electron and hole Fermi surfaces of graphite located along the H–K–H (and H9–K9–H9) edges of the
Brillouin zone.
Electronic States and Transport Properties of Carbon Crystalline 369
8 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> 1 1 3 in semiconductor heterointerfaces. Graphene samples
> ð
<2 1 3E þE Þ ðE1 –E3 Þ2 þ ð0 –24 cosðkz c ÞÞ2 ðaÞ2
4 4 used in transport experiments are typically prepared
E¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi by pressing mechanically cleaved pieces of graphite
>
>1
: ðE2 þE3 Þ 1 ðE2 –E3 Þ2 þ 3 ð0 þ24 cosðkz c ÞÞ2 ða Þ2
>
onto an oxidized surface of Si substrate (Novoselov
2 4 4
et al., 2005b). With enough skills and persistence, a few
ð27Þ
micron-sized ultrathin graphite pieces can be identi-
The -band near the H–K–H edge is illustrated in fied under an optical microscope. Electrical contacts
Figure 6. There occur electron pockets centered at are made by standard lithographic techniques and
the K- and K9-points and hole pockets at the H- and metal evaporation. With the underlying Si substrate
H9-points. The bandwidth along the H–K–H edge is used as a back gate, the system constitutes a field-
4 1 1.56 eV and the band overlap that makes the effect-transistor (FET) structure. Figure 7 shows the
3D graphite a semimetal is 22 39meV. sheet resistance of a monolayer graphene sample as a
Reflecting the quasi-2D structure of graphite, its function of the back gate bias Vg . Positive (negative)
Fermi surfaces are highly elongated along the Vg induces electrons (holes) of density n ¼ Cg Vg =e in
kz -direction. The effective mass for motion within the graphene sheet. For a typical thickness 300 nm of
the basal plane is me ¼ 0:057m for electrons and the insulating SiO2 layer, the coefficient is
mh ¼ 0:039mh for holes (Soule et al., 1964). (In the Cg =e 8 1014 m – 2 V – 1 . Thus, the system functions
early stage of the electronic structure study of gra- as an ambipolar FET with carrier mobilities of the
phite, the assignment of the electron and hole pockets order of 0:1 – 1m2 V – 1 s – 1 .
was opposite to what was later established.) The effec- Another method to prepare graphene relies on gra-
tive mass for the out-of-plane motion is estimated to phitization of the surface layer of a silicon carbide
be on the order of m? ¼ ð5 – 15Þm for both electrons crystal (Berger et al., 2006). The (0001) surface of a
and holes. These values are inferred from the angular 4H-SiC wafer is vacuum heat treated to form an epi-
dependence of the cyclotron mass mc fitted to the taxial layer of graphene. This technique has a potential
following formula assuming ellipsoidal Fermi surfaces: advantage of preparing a large area sample of graphene.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 cos2
sin2
1.10.3.2 Transport at Zero Magnetic Field
¼ þ ð28Þ
mc ð
Þ m2k mk m? The unconventional nature of the electronic state in
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi graphene is reflected in various transport properties
The anisotropy ratio m? =mk is 17 for the electron
Fermi surface and 12 for the hole Fermi surface (Soule (Geim and Novoselov, 2007; Castro Neto et al., 2009).
et al., 1964). Direct measurement of the effective masses
together with correct assignment of carrier pockets were
achieved by cyclotron resonance using circularly polar- 6
ized microwave (Suematsu and Tanuma, 1972).
Detailed studies of magneto-quantum oscilla-
tions suggest existence of additional small pieces
of Fermi surface. These are so-called minority 4
ρ (kΩ)
The electronic dispersion in monolayer graphene is 20m2 V – 1 s – 1 at helium temperature are cur-
characterized by the Dirac cones, that is, massless rently reported. This means that carrier transport is
chiral fermions. Indeed, graphene offers a unique ballistic on submicron scale even at room tempera-
opportunity of exploring an analog of quantum elec- ture. Carrier mobilities currently achieved are
trodynamics (QED) in solid state. Charge carriers in presumably limited by impurities, by structural dis-
bilayer graphene provide massive chiral fermions order (corrugation of graphene sheet) and by
that are no less exotic. coupling with the substrate, and leave room for
Massless Dirac fermion described by the improvement.
Hamiltonian, Equation (9), has a unique feature As seen in Figure 7, sheet resistance of gra-
that the electron state in the conduction band and phene increases as Vg ! 0, but remains on the
the hole state in the valence band are intercon- order of several kilo-ohm even right at the charge
nected, exhibiting properties analogous to charge- neutrality point, where one naively expects the
conjugation symmetry in QED. For the case of system to become nonconducting owing to vanish-
graphene, this symmetry arises from the honeycomb ing carrier density. Conductivity on the order of
lattice structure with A and B sublattices. The quasi- e 2=h per channel at the charge neutrality point has
particles are described by two-component wave been theoretically predicted (Fradkin, 1986; Shon
functions which is analogous to spinor wave func- and Ando, 1998). This conduction without carriers
tion in QED, with pseudo-spin specifying the is yet another counterintuitive property of gra-
sublattice playing the role of real spin in the latter. phene. The reason for the existing discrepancy
The symmetry is such that an electron with energy between experimentally observed values of the
E propagating in one direction belongs to the same minimum conductivity that scatters around 4e 2=h
branch of the electronic spectrum as the hole with and the theoretical value 4e 2=h is not clear at the
energy –E propagating in the opposite direction. moment. Theoretically, proper treatment of strong
Electrons and holes belonging to the same branch electron–electron interaction may be crucial. On
have pseudo-spin pointing in the same direction the other hand, the fact that a real graphene
(parallel to the momentum for electrons and anti- sample at their charge neutrality point most likely
parallel for holes). Chirality, which is defined as consists of puddles of electrons and holes may
projection of pseudo-spin on the direction of significantly affect the experimental results
motion, is positive for electrons and negative for (Adam et al., 2007).
Theoretical prediction for the high-frequency con-
holes.
ductivity of Dirac fremion in graphene is ¼ e 2=4h,
One of the most counterintuitive properties of
as opposed to 4e 2 =h for the DC conductivity. Optical
graphene is the so-called Klein paradox. It has been
transmittance given by T X ð1 þ 2=c Þ – 2 becomes
theoretically predicted that a relativistic particle can
freely penetrate a potential barrier by transforming – 2 – 2
e2 1
itself to its antiparticle. In graphene, this is mani- T¼ 1 þ 2 ¼ 1 þ 1 – ð29Þ
4hc 2
fested as reflectionless tunneling of charge carriers
through a potential barrier by unimpeded transfor-
mation between electron-like and hole-like quasi- where X e 2 =hc 1=137 is the fine structure
particles. Namely, an electrostatic potential cannot constant. The opacity of monolayer graphene is
backscatter a massless Dirac fermion. Another way to given by ð1 – T Þ ¼ 0:023. Visible light trans-
understand the suppression of backscattering in gra- mission measurements on suspended graphene
phene is to invoke destructive interference between membrane have proved that monolayer graphene
absorbs about 2.3% of the incident white light
one backscattering process and its time-revered
(Nair et al., 2008).
counterpart on account of Berry9s phase associated
with the rotation of pseudo-spin. This is analogous to
the anti-localization in the presence of a strong spin–
orbit interaction. 1.10.3.3 Quantum Hall Effect
Charge carriers in graphene can indeed propagate Application of magnetic field normal to the gra-
without scattering over large distances of the order of phene plane causes quantum oscillations of
micrometers. Carrier mobilities as high as resistance (Shubnikov–de Haas effect) which evolve
1:5m2 V – 1 s – 1 are at room temperature and to QHE under higher fields (Novoselov et al., 2005a;
Electronic States and Transport Properties of Carbon Crystalline 371
Zhang et al., 2005). The QHE in graphene is one of N ¼ 0 and 1 LLs is E ¼ 400K BðT Þ, which, in
the most conspicuous phenomena that reveal the conjunction with high mobility, makes it possible to
exotic nature of the system. Figure 8 shows the observe QHE even at room temperature. In moder-
characteristic QHE behavior in monolayer and ate magnetic fields, each LL has fourfold degeneracy.
bilayer graphene (Novoselov et al., 2005a, 2006). The spin degeneracy of LL is lifted by a strong
Here the Hall conductance xy and the longitudinal magnetic field with g-factor somewhat smaller
resistance xx are plotted as a function of the charge than 2. For the N ¼ 0 LL, even the valley degener-
density controlled by the gate bias. In both cases, the acy seems to be lifted under high magnetic fields
step of the Hall conductance xy is 4e 2=h. The factor whose mechanism is yet to be elucidated (Zhang
4 reflects the twofold spin degeneracy and the two- et al., 2006).
fold valley degeneracy. For monolayer graphene, Bilayer graphene has parabolic conduction and
the sequence of Landau levels is shifted by 1/2, valence bands that are degenerate at the charge neu-
compared with the conventional QHE, trality point. The LLs of massive chiral quasi-
so that the
Hall plateau values are xy ¼ 4 e 2=h ðjN j þ 1=2Þ, particles are given by
N being the Landau level (LL) index . The xx peak heB pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EN ¼ N ðN – 1Þ ð31Þ
at n ¼ 0 (Vg ¼ 0) implies that the N ¼ 0 LL which m
stays at the charge neutrality point, E ¼ 0 is shared The QH plateau sequence is xy ¼ 4e 2=h int
equally by electrons and holes. This is a signature of which is the same as the conventional 2DEG with
the massless Dirac fermion described by the hamil- the same spin and valley degeneracy, except that the
tonian, Equation (9). The shift by 1/2 is lowest LLs of the valence and conduction bands
alternatively viewed as arising from an additional (N ¼ 0 and 1) are merged so that the plateau at
phase , known as Berry’s phase, gained by a quasi- n ¼ 0 is missing.
particle as it completes cyclotron motion with fixed The gapless electronic band of bilayer graphene is
chirality. modified by applying an electrostatic bias between
The LL spectrum of a massless Dirac fermion the layers. The resulting energy gap of the order of
(linear dispersion) is 0.1eV can be tuned by the field effect. In the presence
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi of nonzero gap, the above-mentioned anomaly in
EN ¼ vF 2ehjN jB ð30Þ
the QHE due to the degeneracy of the N ¼ 0 and 1
to be contrasted with the ordinary LLs disappears. To observe this splitting at finite
EN ¼ ðheB=mÞðN þ 1=2Þ for a massive particle (para- interlayer bias, the charge neutrality point has
bolic dispersion). Equation (29) implies large to be shifted by additional chemical doping
separation of low-lying LLs. The gap between the (Castro et al., 2007).
(a) (b)
+7/2
Monolayer 4 Bilayer
+5/2 3
6
10 +3/2 2
σxy (4e 2/h)
σxy (4e 2/h)
1
ρxx (kΩ)
ρxx (kΩ)
+1/2
0 4
−1/2
−1
5
−3/2 −2
2
−5/2 −3
−7/2 −4
0 0
−4 −2 0 2 4 −4 −2 0 2 4
Charge density (1012 cm−2) Charge density (1012 cm−2)
Figure 8 Quantum Hall effect in (a) monolayer and (b) bilayer graphene. Reprinted with permission from Geim AK and
MacDonald AH (2007) Graphene: Exploring carbon flatland. Physics Today 60: 35. Copyright (2007) American Institute of
Physics.
372 Electronic States and Transport Properties of Carbon Crystalline
1.10.4 Transport Properties of subbands, and nonzero (zero) density of states at the
Carbon Nanotubes Fermi level for the metallic (semiconducting) SWNT.
As shown by Equation (23), the electronic structure
1.10.4.1 Electronic State in SWNTs
of an SWNT is profoundly affected by an AB flux
As mentioned in Chapter 3.02, the electronic structure threading through the tube. Full conversion of a semi-
of SWNT is determined by the chiral vector, that is, conducting nanotube to a metallic one (or vice versa)
how a hypothetical graphene sheet is rolled to make is beyond experimental reach, because it will require
the SWNT. An armchair SWNT which has an elec- an extremely strong magnetic field in excess of 1000 T
tronic structure similar to monolayer graphene is for a typical nanotube diameter of 1 nm. Still, a pre-
metallic. Zigzag nanotubes are, by contrast, either cursory effect of AB flux on the band structure of
quasi-metallic or semiconducting depending on the SWNT has been verified by observation of mag-
chiral vector. The difference in electronic structures netic-field-induced splitting of the optical spectra
and their correlation with the chirality of nanotube has under magnetic fields up to 45 T (Zaric et al., 2004).
been beautifully verified by spectroscopic measure-
ments using a scanning tunneling microscope (STM)
1.10.4.2 Electron Transport in SWNTs
tip (Issi et al., 1995; Wildoer et al., 1998). Figure 9 shows
the calculated and measured spectra for metallic and For electrical transport experiments, it is necessary to
semiconducting SWNTs (Dekker, 1999). The spectra make electrical contacts. A usual way to achieve this
show the characteristic van Hove singularities of 1D is to disperse or CVD-grow onto a substrate on which
Metallic Semiconducting
nanotube nanotube
(9,9) (11,7)
0.1 0.1
Density of states
Density of states
0.0 0.0
−1 0 1 0 1
−1
Energy (eV)
Energy (eV)
3
Differential conductance
Differential conductance
3
0
(normalized)
(normalized)
2 2
1 1
0 0
−1 0 1 −1 0 1
Voltage (V) Voltage (V)
Figure 9 Calculated (top panels) and measured (bottom panels) tunnel spectra of a metallic nanotube (left) and a
semiconducting nanotube (right). Reproduced with permission from Dekker C (1999) Carbon nanotubes as molecular
quantum wires. Physics Today 52: 21.
Electronic States and Transport Properties of Carbon Crystalline 373
a large array of fine metallic electrodes has been Figure 10 shows the conductance of an SWNT
pre-patterned. After depositing the nanotubes, the sample measured with highly transparent (low-
sample is scanned with an atomic force microscope resistance) contacts achieved with titanium electro-
(AFM) to find nanotubes that by coincidence bridge des (Kong et al., 2001). The value of conductance at
two or more electrodes. Another more controlled low temperatures is close to 4e 2 =h. The conductance
method is to sprinkle nanotubes on an insulating shows conspicuous fluctuations as a function of gate
substrate, take a scanning image of the deposited bias which is attributed to Fabry–Perot-type inter-
area with an AFM, and use that information to ference of ballistic electrons repeatedly reflected at
make a suitable electrode pattern by electron beam the both ends (contacts) of the SWNT sample.
lithography. A problem here is that it is difficult to It is remarkable that metallic SWNT can remain
prejudge whether the target nanotube is metallic or metallic at all, if one is reminded of intrinsic fragility
semiconducting. The contact resistance varies widely of 1D electron systems associated with Peierls
depending on the kind of metal used for the elec- instability and strong tendency of localization. The
trode. Titanium and nickel are often used to achieve tubular structure of SWNT renders robustness
low-resistance contacts, while gold and aluminum against Peierls distortion. The local defect potential
tend to result in high contact resistance. The trend is effectively weakened because it is seen by electrons
seems to be correlated with the carbide formation as averaged over the circumference (White and
tendency of the respective metals. Todorov, 1998). Suppression of backscattering men-
The intrinsic conductance of armchair nanotube is tioned earlier in the graphene context also plays a
expected to be 4e 2 =h. Here, the factor 4 refers to the role here (Ando, 2005).
spin and valley degeneracy. That a metallic SWNT Once electron–electron interaction is switched
acts as a coherent quantum wire was first demon- on, a 1D electron system is predicted to behave as a
strated by Tans et al. (1997) and by Bockrath et al. so-called TLL rather than an ordinary Fermi liquid.
(1997). They observed and analyzed the Coulomb Signatures of TLL behavior are power-law depen-
blockade phenomenon at low temperatures, and dence of the conductance on temperature and bias
found that successively added single electrons are voltage, which are experimentally observed
carried by one additional molecular orbital, which by Bockrath et al. (1999). The TLL nature of
implied that electrons in SWNT were delocalized SWNTs is also corroborated by the photoemission
over an appreciable length (a few mm). spectroscopy (Ishii et al., 2003).
2.0 I II III
25 K
20 K
1.5 50 K
Conductance (G0)
100 K
1.0 290 K
0.5
0.0
−20 −10 0 10 20 30
Vg (V)
Figure 10 Conductance of a nanotube sample as a function of gate bias at different temperatures. The conductance at low
temperatures is close to 2 G0 ¼ 4e2/h in value and exhibits random but reproducible fluctuations. Reprinted with permission
from Kong J, Yenilmez E, Tombler TW, et al. (2001) Quantum interference and ballistic transmission in nanotube electron
waveguides. Physical Review Letters 87: 106801. Copyright (2001) by the American Physical Society.
374 Electronic States and Transport Properties of Carbon Crystalline
(a) (b)
θ
20 0°
19.8°
T = 1.22°K 29.8°
18
16
Magnetoresistance
–105 Δ:p 14 39.7°
50.1°
12
V,(mV)
–6 7.40 58.3°
10
n = 1 electron
6.60 63.5°
–4 8
68.2°
3.643
n = 1 hole 6
3.412
–2 2.952
2.936 n = 2 electron 73.3°
4
1.90 77.0°
0 80.0°
Magnetic field, T 2
0 90.0°
4 6 8 10 12 14 16 18 20 22 24
H (kg)
Figure 11 (a) Traces of longitudinal magnetoresitance xx (B) in a natural single crystal of graphite at T ¼ 1.1K showing the
Shubnikov–de Haas oscillations due to electrons and holes. (b) The magnetoresistance data at different field angles from the
c-axis indicate highly elongated electron and hole Fermi surfaces. (a) Reprinted with permission from Woolam JA (1970) Spin
splitting, Fermi energy changes, and anomalous g shifts in single-crystal and pyrolytic graphite. Physical Review Letters
25: 810. Copyright (1970) by the American Physical Society. (b) Reprinted with permission from Soule DE, McClure JW, and
Smith LB (1964) Study of Shubnikov–de Haas effet. Determination of the Fermi surfaces in graphite. Physical Review
134: A453. Copyright (1964) by the American Physical Society.
Hall voltage to counteract the Lorentz force, a very quadratic B-dependence for compensated metals, a
large positive magnetoresistance appears. In high- B-linear behavior over a wide range of magnetic
quality graphite crystals, xx ðBÞ=xx ð0Þ typically field (McClure and Spry, 1968; Woolam, 1970).
reaches 103 at B 1 T and 104 at 10 T. Since The linear increase of xx(B) tends to saturate at
xx xy , the diagonal magneto-conductivity is higher magnetic fields (Brandt et al., 1974; Lowrey
and Spain, 1977). For some samples, xx(B) even
xx ¼ xx = 2xx þ 2xy 1=xx at all magnetic
starts to decrease at still higher fields (Iye, 1985;
fields. Yaguchi and Singleton, 1998).
In a semi-classical transport model, the Hall The characteristic B-linear behavior has been
conductivity of a semi-metal in the high field interpreted as due to B-dependent screening of
limit is ideally zero. In actual samples, it takes a ionized impurity potential which is thought to be
nonzero value that reflects a small imbalance the dominant scattering mechanism at low tempera-
between the electron and hole densities, tures (McClure and Spry, 1968). The saturation of
xy ðBÞ – ðne – nh Þje j=B. This imbalance is typi- xx ðBÞ at higher field may be attributed to the so-
cally jne – nh j=ðne þ nh Þ 0:3 – 0:03%, and is due called magnetic freeze-out effect, that is, localization
to the presence of ionized impurities. For the major- of the excess carriers to the impurity sites, which is
ity of graphite crystals, the imbalance goes toward known to occur for example in InSb (von Ortenberg,
the electron excess side, that is, more donors than 1973). In the case of graphite, magnetic freeze-out
acceptors. does not affect the total carrier density so much but
Figure 12(a) shows traces of magnetoresistance changes the ionized impurity scattering centers to
in single-crystal (kish) graphite at low temperatures neutral ones. The two sets of data in Figure 12
(Iye et al., 1985). The classical part of the diagonal represent two types of kish graphite samples in
resistivity xx(B) exhibits, instead of the standard terms of ionized impurity concentration. Type B
376 Electronic States and Transport Properties of Carbon Crystalline
Graphite
ρxx 150 mk
192
Type A 255
330
390
0.01 Ω cm 485
600
0 175 mk
257
335
ρxx Type B 390
465
0.01 Ω cm
620
0
0
0 50 100 150 200 250
B (kG)
Figure 12 Traces of longitudinal magneto-resitance xx (B) in single-crystal (kish) graphite at different temperatures. The
two set of data are obtained from samples with different ionized impurity concentrations (Iye, 1985).
sample contains an order of magnitude higher con- oscillations: one due to the electron Fermi surface
centration of ionized impurities than type A. That and the other due to the hole Fermi surface
the resistivity saturation is more conspicuous in type (Woolam, 1970). Due to the high carrier mobilities,
A sample seems to corroborate this picture. However, the SdH oscillations can be seen in graphite from
genuine freeze-out that should diminish xy B to zero magnetic fields as low as 0.1 T. The SdH effect
is not observed in this field range, so that the resis- appears as a series of sharp dips, whose shape
tivity saturation may be interpreted as precursor of reflects that of the 1D density of states associated
the freeze-out effect. with the motion parallel to the field. Figure 11(b)
shows magnetoresistance traces for different values
of field angle
from the c-axis (Soule et al., 1964).
1.10.5.3 Magneto-Quantum Oscillations The shift of the SdH oscillation pattern approxi-
Since the early days of graphite research, the mately obeys the cos
dependence indicating
LL structure of graphite has been calculated (Nakao, highly elongated electron and hole Fermi surfaces.
1976; Dresselhaus, 1974), and investigated by various In magnetic fields higher than 3 T, the SdH dips
magneto-quantum oscillation measurements (Soule split into pairs, reflecting the Zeeman splitting
et al., 1964; Woolam, 1970, 1971a) and magneto-optical (Woolam, 1970). Minority hole pockets are also
spectroscopy (Schroeder et al., 1971; Doezema et al., identified (Woolam, 1971b).
1979). Figure 13(a) shows a calculated LL structure More recently, direct probing of LL density of
along the K–H edge of the Brillouin zone for states is made possible by the use of a low-temperature
B ¼ 1 T. scanning tunneling spectroscope (LT-STM) (Matsui
The trace shown in Figure 11(a) contains two et al., 2005). Figure 14(a) shows the tunneling spec-
series of Shubnikov-de Haas (SdH) quantum trum of graphite under perpendicular magnetic fields.
Electronic States and Transport Properties of Carbon Crystalline 377
Energy (eV)
3
b c n = –1 σ=+
a
2 σ=–
b c c 0
1
–20 a
b 0 EF
c a EF
–1 1 2 3
a 45 –0.02
b 6 b
c 7
a n = –1
b c 8
n = –2 σ = +
a c –0.04
–40 9
b c
a 10 a b σ=–
a b c 11 b
a b c 12 –0.06
a b c c
c n=0
–60 13 a
–0.08 σ=–
b n=1
c
a σ=+
–0.10
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
(K) ξ (H) ξ = cokz /2π
Figure 13 (a) Landau levels of graphite in a magnetic field of 1T. (b) Landau levels of graphite in a magnetic field of 25 T.
Only the lowest spin-split Landau levels of electron and hole bands cross the Fermi level. (a) Reproduced with permission
from Nakao (1976) Landau level structure and magnetic breakthrough in graphite. Journal of the Physical Society of Japan
40: 761. (b) Reprinted with permission from Iye Y, Tedrow PM, Timp G, et al. (1982) High-magnetic field electronic phase
transition in graphite observed by magnetoresistance anomaly. Physical Review B 25:5478. Copyright (1982) by the American
Physical Society.
The LLs are manifest as a series of peaks in the spectra For those electronic properties of graphite asso-
indicated by triangles. The observed spectra show ciated with energy scale larger than the electron–
good agreement with the calculated ones with an hole band overlap, j2 j 0:02eV, the Dirac-like
appropriate choice of the surface potential s , as spectrum is obviously relevant so that interpretation
shown in Figure 14(b). in such terms is possible, as discussed, for example, in
The advent of graphene has brought resurgence of LL spectroscopy (Orlita et al., 2009) and photoemis-
interest in graphite. The electronic properties of sion spectroscopy (Gruneis et al., 2008). For low-
graphite are now scrutinized under new light. energy-scale properties such as transport, however,
Luk’yanchuk and Kopelevich (2004, 2006) recently the 3D nature of the graphite band is quite essential.
made a new analysis of the SdH and dHvA oscilla-
tions in bulk graphite. Based on the values of residual
1.10.5.4 Magnetic-Field-Induced
phase, they claim that the spectra indicate coexis-
Electronic Phase Transition
tence of normal (massive) electrons and Dirac-like
(massless) holes. Schneider et al. (2009) made a simi- In magnetic fields higher than 7.4 T, both electrons
lar but more precise experiment at 10 mK and and holes are only in their lowest LLs, that is, the
measured SdH oscillations up to N 90 quantum system is in the extreme quantum limit. The magneto-
numbers. They have shown that the data precisely fit resistance traces in Figure 12 show distinct anomalies
the prediction of the SWMcC model, so that there is in the high-field quantum limit regime. The abrupt
no ground for invoking massless Dirac fermions in increase of resistance is indicative of a phase transition
3D graphite. to a new magnetic-field-induced electronic state. As
378 Electronic States and Transport Properties of Carbon Crystalline
(a)
4 B = 6T
5T
3 4T
dI/dV (nA/V) 3T
2T
2
0T
Kish graphite
T = 55 mK
0
(b) 30
φs = 17 meV
= 0 meV
dI/dV (nA/V)
20
10
Kish graphite
B=6T
T = 55 mK
0
−200 −100 0 100 200
V (mev)
Figure 14 (a) Tunneling spectrum of graphite in several magnetic fields obtained by a low-temperature scanning tunneling
microscope (LT-STM). (b) Comparison of the spectrum with the calculated local density of states curves for two different
values of surface potential s. Reprinted with permission from Matsui T, Kambara H, Niimi, Y, Tagami K, Tsukada M, and
Fukuyama H (2005) STS observation of Landau levels at graphite surfaces. Physical Review Letters 94: 226403. Copyright
(2005) by the American Physical Society.
seen in the figure, the onset magnetic field decreases along the c-axis (parallel to B). Yoshioka and
with decreasing temperature. Figure 15(a) shows the Fukuyama considered the possibility of Peierls-type
phase boundary of the field-induced phase plotted instability of such magnetic-field-induced 1D system
with T in a logarithmic scale versus 1/B. The data and proposed that the observed anomaly could be
points for the type A sample (crystals with lower attributed to occurrence of a charge-density-wave
ionized impurity concentration) lie on a straight line, instability (Yoshioka and Fukuyama, 1981; Sugihara,
which means that the phase boundary of the field- 1984; Takada and Goto, 1998). To be precise, two
induced phase can be empirically expressed as charge density waves associated with two valleys
(along the H–K–H and H9–K9–H9 edges) occur out-
Tc ðBÞ ¼ T expb – B =B c ð32Þ of-phase with each other to reduce the Hartree term of
the Coulomb interaction, so that a valley density wave
for this field range. may be a better term for the proposed state. The critical
The Landau subband structure in the quantum limit temperature given by Yoshioka and Fukuyama reads
regime is shown in Figure 13(b) for B ¼ 25 T (Iye et al.,
1982). In the quantum limit, the electronic motion in
Kish graphite β′
2000 B || c-axis
α β1
β2
1000 α′
1.1 K
0
2.2 K
0
Type B 3.2 K
200 0
4.2 K
0
100 8.0 K
0 10 K
50 0
4 5 6 0 10 20 30 40 50 60
1/B (×10–3 kG–1) Magnetic field (T)
(c)
10
α
8 α′
Temperature (K)
YF theory
6
0
20 30 40 50 60
Magnetic field (T)
Figure 15 (a) The phase boundary of the magnetic-field-induced phase (Iye, 1985). (b) Magnetoresistance traces up to 55 T
at different temperatures. (c) The phase boundary in a wider range of temperature and magnetic field, showing a re-entrant
behavior at higher magnetic fields. (c) Reprinted with permission from Yaguchi H and Singleton J (1998) Destruction of the
field-induced density-wave state in graphite by large magnetic fields. Physical Review Letters 81: 5193. Copyright (1998) by
the American Physical Society.
where N0" ð"F Þ is the Fermi-level density of states 55 T measured in a pulse magnet (Yaguchi and
of the lowest spin-up electron subband, and ũ is Singleton, 1998). At the highest field, the resistance
a parameter representing the effective electron– comes back to the trace extrapolated from the low-
electron interaction strength. This is an analog field region, indicating a reentrant behavior.
of a Bardeen-Cooper-Schrieffer (BCS) -type formula Figure 15(c) is a phase diagram over a wider
for pairing instability, Tc "F exp½ – ð1=N ð"F ÞV Þ
. temperature and magnetic field range. The dashed
Note that the empirical formula, Equation (31), reflects curve represents the prediction of Yoshioka–
the fact that N0" ð"F Þ_ B. The lower critical tempera- Fukuyama theory, Equation (32). Thus, the valley-
ture for the type-B samples is attributed to the pair density-wave phase exists only at temperatures
breaking effect of ionized impurity (Iye et al., 1984; Iye, below 10 K. The disappearance of the density-
1985). Studies of neutron-irradiated graphite also cor- wave phase at higher field is naturally understood
roborate this conclusion (Yaguchi et al., 1999). since the ð0; " Þ subband relevant to the instability
Figure 15(b) shows magnetoresistance traces up to becomes unoccupied.
380 Electronic States and Transport Properties of Carbon Crystalline
Fradkin E (1986) Critical behavior of disordered degenerate Kroto HW, Heath JR, O’Brien SC, Curl RF, and Smalley RE
semiconductors II. Spectrum and transport properties in (1985) C60: Buckminsterfullerene. Nature 318: 162.
mean-field theory. Physical Review B 33: 3263. Lowrey WH and Spain IL (1977) High field
Fujita M, Wakabayashi K, Nakada K, and Kusakabe K (1996) galvanomagnetic properties of graphite. Solid State
Peculiar localized state at zigzag graphite edge. Journal of Communications 22: 615.
the Physical Society of Japan 65: 1920. Luk’yanchuk IA and Kopelevich Y (2004) Phase analysis of
Fukuyama H and Kubo R (1970) Interband effects on magnetic quantum oscillations in graphite. Physical Review Letters
susceptibility. II. Diamagnetism of bismuth. Journal of the 93: 166402.
Physical Society of Japan 28: 570. Luk’yanchuk IA and Kopelevich Y (2006) Dirac and normal
Geim AK and MacDonald AH (2007) Graphene: Exploring fermions in graphite and graphene: Implications of the
carbon flatland. Physics Today 60: 35. quantum Hall effect. Physical Review Letters
Geim AK and Novoselov KS (2007) The rise of graphene. Nature 97: 256801.
Materials 6: 183. Matsubara K, Sugiura K, and Tsuzuku T (1990) Electrical
Gruneis A, Attaccalite C, Pichler T, et al. (2008) Electron– resistance in the c direction of graphite. Physical Review B
electron correlation in graphite: A combined angle-resolved 41: 969. [Erratum: (1992) Physical Review B 49: 1943.]
photoemission and first-principles study. Physical Review Matsui T, Kambara H, Niimi Y, Tagami K, Tsukada M, and
Letters 100: 037601. Fukuyama H (2005) STS observation of Landau levels at
Ichida M, Mizuno S, Saito Y, Kataura H, Achiba Y, and graphite surfaces. Physical Review Letters 94: 226403.
Nakamura A (2002) Coulomb effect on the fundamental McClure JW (1956) Diamagnetism of graphite. Physical Review
optical transitions in semiconductin single-walled carbon 104: 666.
nanotubes: Divergent behavior in the small-diameter limit. McClure JW (1957) Band structure of graphite and de Haas–van
Physical Review B 65: 241407. Alphen effect. Physical Review 108: 612.
Iijima S (1991) Helical nanotubule of graphitic carbon. Nature McClure JW and Spry WJ (1968) Linear magnetoresistance in
354: 56. the quantum limit in graphite. Physical Review 165: 809.
Iijima S and Ichihashi T (1993) Single-shell carbon nanotubes of Morgan GJ and Uher C (1981) The c-axis resistivity of highly-
1-nm diameter. Nature 363: 603. oriented pyrolytic graphite. Philosophical Magazine B44: 427.
Ishii H, Kataura H, Shiozawa H, et al. (2003) Direct observation Nair RR, Blake P, Grigorenko AN, et al. (2008) Fine structure
of Tomonaga–Luttinger-liquid state in carbon nanotubes at constant defines visual transparency of graphene. Science
low temperatures. Nature 426: 540. 320: 1308.
Issi J-P, Langer L, Heremans J, and Olk CH (1995) Electronic Nakao K (1976) Landau level structure and magnetic
properties of carbon nanotubes: Experimental results. breakthrough in graphite. Journal of the Physical Society of
Carbon 33: 941. Japan 40: 761.
Iye Y and Dresselhaus G (1985) Non-ohmic transport in the Novoselof KS, Geim AK, Morozov SV, et al. (2004) Electric field
magnetic-field-induced charge-density-wave phase of effect in atomically thin carbon films. Science 306: 666.
graphite. Physical Review Letters 54: 1182. Novoselov KS, Geim AK, Morozov SV, et al. (2005a) Two-
Iye Y, McNeil LE, and Dresselhaus G (1984) Effect of impurities dimensional gas of massless Dirac fermions in graphene.
on the electronic phase transition in graphite in the magnetic Nature 438: 197.
quantum limit. Physical Review B 30: 7009. Novoselov KS, Jiang D, Schedin F, et al. (2005b) Two-dimensional
Iye Y, McNeil LE, Dresselhaus G, Boebinger G, and Berglund atomic crystals. Proceedings of the National Academy of
PM (1985) The electronic phase transition in graphite under Sciences of the United States of America 102: 10451.
strong magnetic field. In: Chadi JD and Harrison WA (eds.) Novoselov KS, McCann E, Morozov SV, et al. (2006)
Proceedings of the 17th International Conference on the Unconventional quantum Hall effect and Berry’s phase of 2
Physics of Semiconductors, p. 981. San Francisco, CA, USA, in bilayer graphene. Nature Physics 2: 177.
1984. New York: Springer. Novoselov KS, Morozov SV, Mohinddin TMG, et al. (2007)
Iye Y, Murayama C, Mori N, Yomo S, Nicholls JT, and Electronic properties of graphene. Physica Status Solidi (b)
Dresselhaus G (1990) Effect of pressure on the high- 244: 4106.
magnetic-field electronic phase transition in graphite. Oberlin A, Endo M, and Koyama T (1976) Journal of Crystal
Physical Review B 41: 3249. Growth 32: 335.
Iye Y, Tedrow PM, Timp G, et al. (1982) High-magnetic- Orlita M, Faugeras C, Schneider JM, Martinez G, Maude DK,
field electronic phase transition in graphite observed and Potemski M (2009) Graphite from the viewpoint of
by magnetoresistance anomaly. Physical Review B 25: 5478. Landau level spectroscopy: An effective graphene bilayer
Katayama S, Kobayashi A, and Suzumura Y (2006) Pressure- and monolayer. Physical Review Letters 102: 166401.
induced zero-gap semiconducting state in organic Painter GP and Ellis DE (1970) Electronic band structure and
conductor -(BEDT-TTF)2I3 salt. Journal of the Physical optical properties of graphite from a variational approach.
Society of Japan 75: 054705. Physical Review B1: 4747.
Kataura H, Kumazawa Y, Maniwa Y, et al. (1999) Optical properties Saito R, Dresslhaus D, and Dresselhaus MS (1998) Physical
of single-wall carbon nanotubes. Synthetic Metals 103: 2555. Properties of Carbon Nanotubes. London: Imperial College
Kempa H, Esquinazi P, and Kopelevich Y (2002) Field-induced Press.
metal–insulator transition in the c-axis resistivity of graphite. Saravia LR and Brust D (1968) Band structure and interband
Physical Review B 65: 241101. optical absorption in diamond. Physical Review 170: 683.
Kobayashi A, Katayama S, and Suzuura Y (2009) Theoretical study Saslow W, Bergstresser TK, and Cohen ML (1966) Band
of the zero-gap organic conductor -(BEDT-TTF)2I3.. Science structure and optical properties of diamond. Physical
and Technology of Advanced Materials 10: 02409. Review Letters 16: 354. [Erratum: (1968) Physical Review
Kong J, Yenilmez E, Tombler TW, et al. (2001) Quantum Letters 21: 715.]
interference and ballistic transmission in nanotube electron Schneider JM, Orlita M, Potemski M, and Maude DK (2009) A
waveguides. Physical Review Letters 87: 106801. consistent interpretation of the low temperature magneto-
Kopelevich Y, Esquinazi P, Torres JHS, da Silva RR, and transport in graphite using the Slonczewski–Weiss–McClure
Kempa H (2003) Graphite as a highly correlated electron 3D band structure calculations. Physical Review Letters
liquid. Advances in Solid State Physics 43: 207. 102: 166403.
382 Electronic States and Transport Properties of Carbon Crystalline
Schroeder PR, Dresselhaus MS, and Javan A (1971) High- Woolam JA (1970) Spin splitting, Fermi energy changes, and
resolution magnetospectroscopy of graphite. In: Carter DL anomalous g shifts in single-crystal and pyrolytic graphite.
and Bate RT (eds.) Proceedings of the International Physical Review Letters 25: 810.
Conference on the Physics of Semimetals and Narrow- Woolam JA (1971a) Graphite carrier locations and quantum
Bandgap Semiconductors, p. 139. Dallas, TX, USA, 1970, transport to 10 T (100 kG): Physical Review B 3: 1148.
London: Pergamon. Woolam JA (1971b) Minority carriers in graphite. Physical
Shon NH and Ando T (1998) Quantum transport in two- Review B 4: 3393.
dimensional graphite system. Journal of the Physical Society Yaguchi H, Iye Y, Takamasu T, Miura N, and Iwata T (1999)
of Japan 67: 2421. Neutron-irradiation effects on the magnetic-field-induced
Slonczewski JC and Weiss PR (1958) Band structure of electronic phase transitions in graphite. Journal of the
graphite. Physical Review 109: 272. Physical Society of Japan 68: 1300.
Smith BW, Monthioux M, and Luzzi DE (1998) Encapsulated Yaguchi H and Singleton J (1998) Destruction of the field-
C60 in carbon nanotubes. Nature 396: 323. induced density-wave state in graphite by large magnetic
Soule DE, McClure JW, and Smith LB (1964) Study of fields. Physical Review Letters 81: 5193.
Shubnikov–de Haas effet. Determination of the Fermi Yoshioka D and Fukuyama H (1981) Electronic phase transition
surfaces in graphite. Physical Review 134: A453. of graphite in a strong magnetic field. Journal of the Physical
Suematsu H and Tanuma S (1972) Cyclotron resonances in Society of Japan 50: 725.
graphite by using circularly polarized radiation. Journal of the Zaric S, Ostojic GN, Kono J, et al. (2004) Optical signatures of
Physical Society of Japan 33: 1619. Aharonov–Bohm phase in single-walled carbon nanotubes.
Sugihara K (1984) Charge-density wave and magnetoresistance Science 304: 1129.
anomaly in graphite. Physical Review B 29: 6722. Zhang Y, Jiang Z, Purewal JP, et al. (2006) Landau-level splitting
Tajima N, Tamura M, Nishio Y, Kajita K, and Iye Y (2000) in graphene in high magnetic fields. Physical Review Letters
Transport property of an organic conductor -(BEDT-TTF)2I3 96: 136806.
under high pressure. Journal of the Physical Society of Japan Zhang Y, Tan Y-W, Stormer H, and Kim P (2005) Experimental
69: 543. observation of the quantum Hall effect and Berry’s phase in
Takada Y and Goto H (1998) Exchange and correlation effects graphene. Nature 438: 201.
in the three-dimensional electron gas in strong magnetic Zunger A (1978) Self-consistent LCAO calculation of the
fields and application to graphite. Journal of Physics: electronic properties of graphite. I. The regular graphite
Condensed Matter 10: 11315. lattice. Physical Review B17: 626.
Tans SJ, Devoret MH, Dai H, et al. (1997) Individual single-
wall carbon nanotubes as quantum wires. Nature
386: 474.
von Ortenberg M (1973) On the problem of magnetic freezeout
in indium antimonide. Journal of Physics and Chemistry of
Solids 34: 396. Further Reading
Wallace PR (1947) The band theory of graphite. Physical Review
71: 622. Ando T, Zheng Y, and Suzuura H (2002) Dynamical conductivity
White CT and Todorov TN (1998) Carbon nanotubes as long and zero-mode anomaly in honeycomb lattices. Journal of
ballistic conductors. Nature 393: 240. the Physical Society of Japan 71: 1318.
Wildoer JWG, Venema LC, Rinzler AG, Smalley RE, and Castro Neto AH, Guinea F, Peres NMR, Novoselov KS, and
Dekker C (1998) Electronic structure of atomically resolved Geim AK (2009) The electronic properties of graphene.
carbon nanotubes. Nature 391: 59. Reviews of Modern Physics 81: 109.
1.11 Angle-Resolved Photoemission Spectroscopy of
Graphene, Graphite, and Related Compounds
T Sato and T Takahashi, Tohoku University, Sendai, Japan
ª 2011 Elsevier B.V. All rights reserved.
383
384 Angle-Resolved Photoemission Spectroscopy
Crystal surface
x y
Figure 2 Energy-conservation diagram of the
Figure 1 Schematic view of ARPES. photoemission process.
Angle-Resolved Photoemission Spectroscopy 385
photoelectron emitted into the vacuum (outside the of electrons in a crystal as a function of the momen-
crystal) are described as follows: tum parallel to the surface (kjj). The schematic
pffiffiffiffiffiffiffiffiffiffiffi diagram is shown in Figure 4. This method appears
Kjj ¼
2mEK sin ð4Þ
to be very powerful when the crystal under consid-
pffiffiffiffiffiffiffiffiffiffiffi eration is a highly two-dimensional layered material.
K? ¼ 2mEK cos ð5Þ
In this case, the energy dispersion perpendicular to
where is the polar angle of photoelectron as shown the layer is negligibly small and therefore the
in Figure 1. obtained band dispersions as a function of kjj
5. When the photoelectron is emitted from the represent the band structure of the material.
crystal into the vacuum through the surface, the Figure 5 shows a set of ARPES spectra of layered
momentum parallel to the surface is conserved compound 1T-VSe2 and the experimental band dis-
because the transverse symmetry parallel to the sur- persions obtained from these spectra (Terashima
face is hold (Figure 3). This is the most important et al., 2003). Several characteristic bands from Se 4p
assumption in ARPES:
and V 3d orbitals are clearly seen in the experi-
kjj ¼ Kjj ð6Þ mental band structure. We also find that the V 3d
band reaches the Fermi level at midway between
Using Equations (1)–(6), we obtain the relationship and M points in the Brillouin zone, producing a
between the energy and the momentum of the initial metallic Fermi surface. The band structure calculation
state in the crystal: is also shown for comparison in Figure 5. The good
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi agreement between the experiment and the calcula-
hkjj ¼ 2mðEi þ h! – W Þsin ð7Þ
tion indicates that ARPES is able to map out the band
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
structure of layered materials with high precision.
hk? ¼ 2mððEi þ h! – W Þcos2 þ V0 Þ ð8Þ
The concept of Fermi surface is one of important
where V0 is defined as the sum of E0 and W (V0 ¼ E0 þ W) physical concepts based on which we discuss the
and is called the inner potential. By applying these electronic properties of materials. By measuring
formulas to ARPES experimental results, we are able many ARPES spectra all over the Brillouin zone
to map out experimentally the relationship between the and plotting the spectral intensity at the Fermi level
initial-state energy (Ei) and the momentum (kjj and k?), as a function of two-dimensional momentum (wave
namely the band dispersions of electrons in the crystal. vector), we are able to map out experimentally the
two-dimensional Fermi surface. Figure 6 shows an
1.11.2.2 Determination of Band example applied for 1T-VSe2, where the characte-
Dispersions ristic two-dimensional cylindrical Fermi surface is
1.11.2.2.1 Layered materials
clearly mapped out by ARPES (Terashima et al.,
As described above, when we measure the polar 2003).
angle () and the kinetic energy (EK) of photoelec-
trons, we are able to map out the band dispersions (Ei)
Energy band
zII
θ0
EF
Photoelectron intensity
K θ1
θ θ θ2
KII
Energy
e– θ3
Crystal surface
k θ4 √2mEK
k = sinθ
h
θ5
kII
e– EF k0 k1 k2 k3 k4 k5
Binding energy Wave vector
Figure 3 Momentum parallel to the surface of
photoelectrons is conserved during the photoemission Figure 4 Schematic diagram to show how to determine
process. the E–k relation (band structure) by ARPES.
386 Angle-Resolved Photoemission Spectroscopy
1T-VSe2 EF V 3dz 2
M(L)
1 Se 4pz*
Intensity (a.u.)
Se 4px,y
3
4
Γ(A)
5
6
Γ M
7 6 5 4 3 2 1 EF A L
Binding energy (eV) Wave vector
Figure 5 Experimentally determined band structure of quasi-1D layered compound 1T-VSe2 (Terashima et al., 2003).
ky
High
kx
Γ(A)
Intensity
Low
1T-VSe2
First Brillouin zone
Ef (see Figure 2), the escape depth becomes much done along the (010) direction at the (001) cleaved
shorter (one-half to one-third) because the wave func- surface. So, kjj and k? are defined along the (010) and
tion of the photoelectrons outside the crystal becomes (001) directions, respectively. When we compare the
an evanescent wave inside the crystal and cannot enter experimental band dispersions with the band struc-
deep inside the crystal. This situation enhances the ture calculated along several high-symmetry lines,
broadening of momentum perpendicular to the we find that the band dispersions along X and
surface, and finally a large portion of k? is simulta- XWX high-symmetry lines simultaneously appear
neously seen in ARPES measurements. In other in the experimentally determined band structure.
words, when we conduct ARPES measurements at a This indicates that two different k? points are meas-
certain polar angle (), the kjj component is uniquely ured simultaneously at a fixed kjj due to the k?
determined from Equation (7), while the k? compo- integration effect. As shown above, the band structure
nent is automatically integrated in all (or a large of three-dimensional materials can be determined by
portion of) Brillouin zone perpendicular to the crystal ARPES in many cases.
surface because of the short escape depth of photo-
electrons. In general, the DOS is relatively larger on
the high-symmetry line in the Brillouin zone in com- 1.11.2.3 Experimental Apparatus
parison with other portions between high-symmetry Figure 8 shows a schematic diagram of a high-
lines. When we integrate the DOS along k? at a fixed resolution ARPES spectrometer. The apparatus
kjj, the high-symmetry line produces a peak in the consists of mainly four parts: (1) a large electrostatic
ARPES spectrum. This means that the peak position hemispherical electron energy analyzer with the aver-
in the ARPES spectrum of three-dimensional mate- age diameter of about 40 cm; (2) a microwave-driven
rials traces the band dispersion on the high-symmetry discharging lamp to produce high-intensity VUV light;
line in the Brillouin zone. This indicates that we can (3) a sample preparation vacuum chamber where
compare the band dispersions obtained by ARPES for samples are prepared by cleaving, sputtering, evapora-
three-dimensional materials with the band structure tion, etc.; and (4) an ultrahigh vacuum analyzer
calculated along the high-symmetry lines. chamber where the sample is irradiated by VUV light
Figure 7 shows the band structure of LaSb deter- from the discharging lamp. In addition, the apparatus is
mined by ARPES, compared with the band structure equipped with a liquid-helium cryostat to cool down
calculation (Kumigashira et al., 1998). LaSb is a typi- the sample, a sample transfer system, many vacuum
cal three-dimensional material with face-centered pumps, etc. We can use synchrotron radiation light to
cubic crystal structure. ARPES measurements were excite photoelectrons.
e1 e3
EF
e2
X
z(001)
h2 k
a
1.0 h1
Binding energy (eV)
Γ X
W
c b j
X X
K
Σ
i
Δ W 2.0
d
2nd Brillouin zone g
X Γ Δ X y(010)
X f
3.0
X
1st Brillouin zone
x(100)
4.0
X W X
Γ X
Hemispherical
electron analyzer
CCD
camera
Installation TMP
chamber
Main
He-flow chamber He discharge
cryostat lamp
Monochromator
Manipulator Cryostat
Sample
Preparation
chamber TMP
Radiation
shield
The recent progress in the energy resolution in schematic diagram to explain how the energy and
ARPES experiments owes mainly to the improve- the polar angle of photoelectrons are measured
ment of electron energy analyzer. Figure 9 shows the with the hemispherical analyzer. By using the
Hemispherical analyzer
Slit E1 E2 E3 Energy
e θ1
gl
An MCP
θ2
Electron lens
Screen
Photoelectron
hω
CCD camera
θ1
θ2
Sample
Figure 9 Two-dimensional detection hemispherical electron analyzer.
Angle-Resolved Photoemission Spectroscopy 389
Graphene ribbon
Graphene
Graphite
Nanotube
GIC
Figure 11 Crystal structures of graphene, graphite, and related materials.
390 Angle-Resolved Photoemission Spectroscopy
base of the sp2-hybridized carbon materials is In all the carbon-based materials, the key to
graphene, single-layered graphite, which is purely understand the interesting physical properties
two dimensional. As discussed later, this graphene lies in the carbon 2p and orbitals whose atomic
behaves as a zero-gap semiconductor where the wave functions are schematically shown in
mass of carriers is regarded as zero (a Dirac fermion). Figure 12(a). The orbital is elongated parallel to
When the graphene stacks along the out-of-plane the graphite plane while the orbital perpendicular
direction, it becomes three-dimensional graphite. to the plane. The band structure for high-symmetry
In contrast to the zero-gap nature of graphene, gra- lines in the Brillouin zone derived from these orbitals
phite is a semimetal with a small overlap of C 2p are shown in Figure 12(b). Because two carbon
bands due to the finite interaction between graphene atoms are included in the hexagonal unit cell in the
sheets in the crystal. Various atoms or molecules can case of the AB stacking, we find two bonding
be inserted between graphene sheets of graphite. bands and four bonding bands below EF. The
These compounds are called GICs and are usually corresponding antibonding counterparts are also
metallic and sometimes superconducting. It has been seen above EF ( and bands). The bandwidth
theoretically predicted that by cutting the side of along the momentum perpendicular to the plane
graphite (graphite ribbon), anomalous electronic (kz) is very small compared to that for the in-plane,
states called edge-localized states appear in reflecting the quasi-two-dimensional nature of
the vicinity of EF (Fujita et al., 1996; Kobayashi, graphite. As one immediately notices from
1993; Nakada et al., 1996; Wakabayashi et al., 1999). Figure 12(b), both and bands cross EF at H or
Moreover, carbon nanotube, created by smoothly K point and produce small hole and electron Fermi
connecting both sides of graphene sheets, shows a surfaces. These small Fermi surfaces are responsible
variety of interesting properties owing to its chirality. for the electric conduction and hence the semimetal-
The carbon nanotube would be technologically lic nature of graphite.
useful for transistors, power fuel cells, displays, Figure 13 shows valence-band ARPES spectra of
and so on. kish graphite (artificially grown single-crystal
(a) (c) kz
Carbon atom
A
R Δ
π orbital L Q Γ
σ orbital S
ε
kx M P ∧
T
K
ky
(b)
Energy (Ry) Energy (eV)
0.0 σ* 0.0
–4.0
–0.4 π*
–8.0
–0.8 π
–12.0
σ
–1.2 –16.0
–20.0
–1.6
–24.0
κ ∧ Γ Σ M T KPH Q A R L S H
Figure 12 (a) Schematic view of the wave function of and orbitals. (b, c) Calculated band structure (Tatar and Rabii,
1982) and Brillouin zone of graphite.
Angle-Resolved Photoemission Spectroscopy 391
Γ(A) Γ(A)
K(H) K(H)
(a) M(L) (b) M(L)
M(L)
Intensity (a.u.)
Intensity (a.u.)
K(H) M(L)
Γ(A)
Γ(A)
16 12 8 4 EF 16 12 8 4 EF
Binding energy (eV) Binding energy (eV)
Figure 13 Valence-band ARPES spectra of graphite measured along two high-symmetry lines: (a) KM (AHL) and
(b) M (AL).
graphite) measured at 20 K along (a) KM (AHL) oriented pyrolytic graphite (HOPG). This suggests
and (b) M (AL) directions. We clearly find highly that ARPES with a high-quality single crystal is
dispersive bands along both directions. For example, essential to establish the intrinsic electronic structure
along the KM (AHL) direction (Figure 13(a)), one of graphite.
band has the top of dispersion at the (A) point at To visualize more clearly the dispersive features
4 eV and disperses toward the higher binding energy in ARPES spectra, we have mapped out the band
on approaching the K(H) point. This band has the structure and show the result in Figure 14. The
bottom at about 12 eV at the K(H) point and then experimental band structure was obtained by taking
disperses back toward the lower binding energy on the second derivative of ARPES spectra and plotting
approaching the M(L) point. We also find another the intensity by gradual shading as a function of wave
prominent peak located at about 8 eV at the (A) vector and binding energy. Bright areas correspond
point. This band rapidly approaches EF on moving to experimental bands. Figure 14 also shows the
toward the K(H) point, and the spectrum shows a first-principles band calculation of graphite (Tatar
clear Fermi-edge cutoff indicative of a tiny hole and Rabii, 1982) using the Johnson and Dresselhaus
pocket centered at the K(H) point. On the other (1973) model (the JD model) with the AB stacking
hand, along the M (AL) direction (Figure 13(b)), sequence. It is clear that the experimental band
the band does not cross EF but stays at 3 eV around structure and the bulk band calculation show a
the M(L) point, indicative of the absence of Fermi good agreement along both directions. However, we
surface along this direction. These features are com- also notice quantitative differences. For example, the
monly observed in previous ARPES results band due to the C2p bonding states is located at
(Takahashi et al., 1985; Santoni et al., 1991). It is also about 4 eV at the (A) point in both the experiment
noted that we are able to observe separately the and the calculation, while the dispersive feature from
electronic structure along the KM (AHL) and M the (A) point to the K(H) point shows a quantitative
(AL) directions due to the single-crystal nature of deviation. The lower-lying branch of the band
kish graphite, which is not possible with the highly (2 band) along the KM (AHL) direction is located
392 Angle-Resolved Photoemission Spectroscopy
(a) (b)
EF EF
4 4
Binding energy (eV)
High
12 12
16 16 Low
Γ K M Γ M
A H L A L
Wave vector Wave vector
Figure 14 Valence-band structure of graphite along two high-symmetry lines KM (AHL) and M (AL) determined by
ARPES, compared with the band-structure calculation (Tatar and Rabii, 1982).
at slightly higher binding energy in the experiment While the ARPES technique determines the
than in the calculation. A similar trend is also seen in occupied electronic states, the band structure in
the M (AL) direction. The band is located at a the unoccupied side is also elucidated by the
slightly higher binding energy in the experiment angle-resolved inverse photoemission spectroscopy
than in the calculation at the (A) point, whereas (ARIPES) (Schäter et al., 1986; Ohsawa et al., 1987)
both perfectly coincide with each other near EF at the and angle-resolved secondary-electron emission
K(H) point in the energy scale of Figure 14. The spectroscopy (ARSEES) (Maeda et al., 1988). In the
finite deviation of the energy position of bands ARIPES, electrons are impinged into the sample sur-
between the experiment and the calculation may be face and emitted photons are detected. Figure 15(a)
due to many-body effects and/or final-state effects shows ARIPES spectra of HOPG as a function of
(Strocov et al., 2001; Heske et al., 1999). polar angle. The ARIPES spectra show clear
(a) (b)
Graphite (HOPG)
ARIPES
69°
65° 20
Intensity (a.u)
60°
55°
50°
45°
40°
35°
10
30°
25°
20°
15° Ev
10°
5°
0° EF
0 5 10 15 20 25 K Γ M
Energy relative to EF
Figure 15 (a) Angle-resolved inverse photoemission spectra of graphite. (b) Comparison of ARIPES-derived unoccupied
band structure with the band calculation.
Angle-Resolved Photoemission Spectroscopy 393
structures denoted by markers, whose energies and an expansion around the K(H) point. As clearly seen
intensities are very sensitive to the change of polar in Figure 16(b), the spectral intensity at EF is
angle. Figure 15(b) shows a comparison of experi- sharply peaked at the K(H) point, demonstrating
mentally determined unoccupied band structure the existence of an extremely small Fermi surface
with the band calculation (Tatar and Rabii, 1982). at the K(H) point. The volume of the observed small
An overall agreement between the ARIPES experi- hole-like Fermi surface can be estimated by the
mental result and the calculation is found as in the following method. We first assume that the hole
case of the occupied band structure, in both the pocket has an ellipsoidal shape elongated along
and bands. A similar result has been obtained by the kz (KH) axis with a circular cross section in the
comparison with a refined band calculation (Fretigny kx–ky plane. The longer axis (2b) of the ellipsoid is
et al., 1989). All these experimental results demon- supposed to be just half of the unit cell along the kz
strate that the band-structure calculation serves as a axis, namely the distance between the K and H
good starting point to discuss the electronic structure points, being 0.942 Å1, as predicted from the
of graphite. band calculation (Charlier et al., 1991). We
determined the shorter axis (2a) of the ellipsoid as
1.11.3.1.1 Fermi surface 0.034 0.004 Å1 by extrapolating the band disper-
Since the graphite is a low-carrier semimetal, we sion near the K(H) point toward EF with a linear
expect the presence of tiny Fermi surface(s). This function. The estimated volume (4a2b/3) of the
Fermi surface is successfully determined by the hole-like Fermi surface is 5.7 0.7 104 Å3,
high-resolution ARPES. Figure 16(a) displays the which corresponds to the carrier number of
ARPES intensity at EF of kish graphite plotted as 9.2 1.1 1018 cm3. This value is apparently
a function of two-dimensional wave vector. The larger than the hole carrier number (2.0 1018)
spectral weight is integrated within 10 meV with estimated from the measurement of the Hall
respect to EF and the intensity is plotted by assum- coefficient (Brandt et al., 1988), possibly due to
ing the hexagonal symmetry. The bright area the finite energy and momentum resolution in the
corresponds to the high intensity. Figure 16(b) is ARPES measurement.
kx (Å–1)
–1.0 0.0 1.0
(a)
M(L)
K(H)
1.0
ky (Å–1)
0.0 Γ(A)
–1.0
–1.6
(b)
ky (Å–1)
–1.7
–1.8
–0.2 –0.1 0.0 0.1 0.2
kx (Å–1)
Figure 16 (a) Fermi surface of graphite determined by ARPES. (b) Expansion near the K(H) point.
394 Angle-Resolved Photoemission Spectroscopy
0.10
(a) Zigzag ribbon (b) Armchair ribbon
1A 1A 2A 1B
1B
Intensity (a.u.)
1
2B
1
2 Γ(A)
2A 3
2B 2
3 4
5 0.05
K(H)
M(L)
N–1
N A N–1
N N
NB NA N B
(c) (d) 0.0
–3 –3
–0.05
0 0
E
–0.10
0.6 0.4 0.2 EF
3 3 Binding energy (eV)
–π 0 π –π 0 π
Figure 18 Near-EF ARPES spectra of graphite showing
Figure 17 (a, b) Zigzag and armchair ribbons. (c, d) the edge-localized state. Inset shows the measured cut in
Calculated energy dispersion for zigzag and armchair the Brillouin zone (red line near the K(H) point). Red arrows
ribbons. indicate prominent structures around kx ¼ 0 Å1.
Angle-Resolved Photoemission Spectroscopy 395
matrix-element effect in the photoelectron exciting anomalous band is simply denied since such a back-
process because the polarization of incident light is ground peak in ARPES spectrum would show no
different between the two cases. We have measured dispersion, contrary to the small but finite energy
ARPES spectra for several cuts near the K(H) point dispersion of the band. The edge-localized state pre-
and found that this anomalous feature appears only dicted from the calculation for graphite ribbons
around the K(H) point. (Fujita et al., 1996; Nakada et al., 1996; Kobayashi,
In Figure 19, we plot the ARPES spectral inten- 1993; Wakabayashi et al., 1999) may account for the
sity as a function of binding energy and wave vector, observed anomalous band. According to the calcula-
together with the projection of bulk band on the (001) tions, a zigzag edge of graphite ribbon possesses an
plane (dots) (McClure, 1957). As clearly seen in the almost-flat band in the close vicinity of EF around the
figure, the projection of the band shows a good K point, in good agreement with the present observa-
agreement with the hole-like band in the experi- tion. We speculate that cleaved surface of kish
ment. The anomalous feature is observed well graphite contains similar zigzag edges at the steps.
outside the projection of the band, indicating that In fact, a recent STS/STM study (Matsui et al., 2005)
the dispersion along kz of the band is not respon- reported the existence of zigzag steps and the resul-
sible for the anomalous feature. The antibonding tant edge-localized states on the graphite surface.
electron-like band ( band) might be the origin It is also revealed that the zigzag step affects the
of the anomalous feature. However, as seen in electronic structure substantially away from the
Figure 19, the band is situated well above EF in step because the edge-localized state is observed on
this momentum region and cannot account for the the flat surface about 35 Å away from the step.
anomalous flat band below EF. The coupling of Probably this long-distance effect of the zigzag step
electrons with a certain bosonic excitation as enables the observation of the edge-localized states
observed in cuprate superconductors (Campuzano by ARPES, which probes the electronic structure
et al., 1999; Lanzara et al., 2001; Sato et al., 2003) averaged over a wide area of the surface. The possi-
cannot explain this observation. Because the band bility of dangling bonds are not ruled out to explain
produced by such an excitation should follow or the observed anomalous electronic states (Jensen and
mimic the original hole-like dispersion (Norman Blase, 2004; Hahn and Kang, 1999). It has been theo-
retically suggested (Jensen and Blase, 2004) that the
et al., 1997) with a considerable renormalization of
character of the graphite surface is dramatically
the bandwidth near EF. However, the observed
affected by the presence of dangling bonds. These
anomalous band shows an electron-like character in
dangling bonds are associated to undercoordinated
contrast to the hole-like band. The angle-
carbon atoms at the point or extended defects at the
integrated-type background as the origin of the
step edges. According to the STM measurement by
Hahn and Kang (1999) such dangling bonds, which
are not passivated by adsorbates contribute to the
EF DOS near EF. We speculate that dangling atoms or
dimers at armchair or zigzag edges, which are not
Binding energy (eV)
0.6
1.11.3.3 Many-Body Interactions
Low
–0.10 –0.05 0.0 0.05 0.10 ARPES is a strong experimental technique to deter-
kx (Å–1) mine the band structure of solids. By qualitatively
analyzing the energy dispersion and the energy
Figure 19 ARPES-intensity plot near EF of graphite as a
function of wave vector and binding energy measured near width of the peak in the ARPES spectrum, we are
the K(H) point (see Figure 18), compared with the projection able to obtain an insight into the many-body interac-
of calculated bands (McClure, 1957). tions responsible for the anomalous physical
396 Angle-Resolved Photoemission Spectroscopy
properties. For instance, as shown in Figure 20(a), a signature of the strong electron-mode coupling
when electrons are strongly coupled to some kind of in kish graphite (Sugawara et al., 2007). In
collective excitations or bosonic modes, a kink Figure 20(b), we plot ARPES spectra of kish
appears in the energy dispersion near EF. The energy graphite near EF measured at 7 K along a cut that
scale of the kink basically reflects the characteristic includes the K(H) point (line at K(H) in the inset to
energy of the mode, and the effective mass of Figure 20(c)). Figure 20(c) shows the corresponding
the quasi-particle band is highly renormalized in intensity plot as a function of wave vector and
the vicinity of EF due to the interaction. The disper- binding energy. We clearly identify in Figure 20(b)
sion kink has been studied recently in various a couple of hole-like bonding bands separated by
materials, such as high-Tc cuprate superconductors 0.6 eV, which disperse toward EF when approaching
(Campuzano et al., 2004; Damascelli et al., 2003), the K(H) point. The splitting of bands is created
charge-density-wave materials (Valla et al., 2000; by the AB stacking sequence of kish graphite as
Schäfer et al., 2003), and surface systems (Valla et al., seen in the band calculations (Tatar and Rabii,
1999; Hengsberger et al., 1999). As for the origin of 1982; Charlier, 1991; McClure, 1957) (see also
the mode, mainly phonons and magnetic excitations Figure 12). An anomalously sharp quasi-particle
were proposed so far. peak appears in the upper band at a narrow angle
By obtaining a flat clean surface with a negligible region centered at the K(H) point, while the peak
contribution from the edge-localized states, we find becomes significantly broad and does not show
prominent energy dependence when the position of
peak exceeds 0.2 eV. We also find a tiny peak
(a) (b) slightly away from the K(H) point in the close
kF vicinity of EF, which shows a remarkable resem-
EF k
blance to the spectral line shape of the surface band
Mode near EF in Be surface (Hengsberger et al., 1999). The
energy Kink With marked sharpening of the spectral line shape as well
Intensity (a.u)
150
ReΣ (meV)
λ = 0.70 ± 0.08 60 are directly connected to each other. To elucidate
100 40 the character of the mode quantitatively, we apply the
Debye model to reproduce the obtained self-energy on
20 an assumption of a linear scattering rate higher than
50
0 the Debye energy !D to account for the energy depen-
Intensity (a.u.)
kz (a)
(a) (b) ∼
C8 K ΓΚ Ζ He I
C8K A Z
B
kx K Λ T
Σ ΓΔ
Y
ky
70°
65°
Intensity (a.u.)
Graphene layer K Y K Γ
kx 60°
Unit cell 1st layer α (K) (Γ)
2nd layer β Y 55°
3rd layer γ
Γ 50°
4th layer δ
5th layer α (M) 45°
ky 40°
35°
Figure 22 (a) Crystal structure and (b) Brillouin zone 30°
of C8K. 25°
20°
15°
10°
studied by ARPES, especially in the 1980s, in C6Li 5°
0°
(Eberhardt et al., 1980; McGovern et al., 1980), C8K
10 5 EF
(Takahashi et al., 1986), C8Rb (Eberhardt et al., 1980; Binding energy (eV)
(b)
McGovern et al., 1980), and C8Cs (Gunasekara
et al., 1987). The crystal structure and Brillouin EF
zone for C8K are shown in Figure 22. The graphite
Binding energy (eV)
1.11.4.1.2 Interlayer band 11.5 K) are significantly higher than those of other
As seen in Figure 23(b), some discrepancies are also GICs. To determine its pairing symmetry, several
found between the ARPES experiment and the cal- experimental results have been reported. Although
culation: (1) the so-called K 4s band (interlayer band) the isotropic s-wave superconducting order para-
in the calculation shows a nearly-free-electron-like meter within the conventional BCS theory has been
dispersion around the point while we find only a inferred from the thermodynamic (Sutherland et al.,
nondispersive band at EF in the ARPES experiment; 2007) and magnetic properties (Lamura et al., 2006),
(2) the calculated highest energy level of the band substantial deviations and anomalies beyond the
at the point is located at about 7 eV while it is conventional framework have also been suggested
experimentally observed at about 6 eV. As for the by the tunneling spectroscopy (Bergeal et al., 2006;
interlayer band, it has a large energy dispersion Kurter et al., 2007) and the electron spin resonance
along the interlayer direction (Z) as shown in experiment (Muranyi et al., 2008), proposing the
Figure 23(b). This interlayer dispersion may distort strong-coupling superconductivity, the anisotropic
the present ARPES spectra, causing a discrepancy gap, and/or the multicomponent gaps. The discovery
between the experiment and the calculation. of high-Tc GICs has also stimulated several theo-
Although we do not find a clear signature for the retical proposals for the superconducting mechanism.
interlayer band, the present experimental data sug- An earlier band calculation (Csanyi et al., 2005) has
gest the existence of partially filled band around the proposed an unconventional pairing mechanism via
point. This is derived from the following two the interlayer band with intermediate bosons such as
experimental facts. First, the experimentally low-energy acoustic plasmons (Csanyi et al., 2005)
observed band observed at the K~ point shows a and excitions (Allender et al., 1973). Pairing mechan-
good agreement with that of the ONK model, which ism by the resonating valence bond has also been
predicts the presence of the interlayer band below EF discussed (Black-Schaffer and Doniach, 2007).
(Ohno et al., 1979), while the band does not show Recent first-principles calculations (Boeri et al.,
a good agreement with the model predicting 2007; Calandra and Mauri, 2005; Mazin, 2005;
the absence of the interlayer band below EF Mazin et al., 2007; Sanna et al., 2007) have suggested
(DiVincenzo and Rabii, 1982). The size of electron that the electron coupling to the out-of-plane carbon
pocket in the experiment is similar to that in the and intercalant phonons is responsible for achieving
ONK model, which suggests an electron charge higher Tc’s, while a clear demonstration to experi-
transfer of 0.6e from potassium atoms (e is the unit mentally distinguish various models has not yet been
charge). In order to accommodate the remaining made. This key issue as well as the discrepancies in
electronic charge of about 0.4e, another band is neces- the nature of superconducting gap among experi-
sary somewhere below EF in the experimental band ments request further precise investigations on the
structure. This extra band is supposed to be located momentum and Fermi-surface dependence of the
around the center of the Brillouin zone according to superconducting gap. However, such an experiment
the theoretical interpretations for C8K (Ohno et al., had hardly been done because of the small super-
1979; DiVincenzo and Rabii, 1982). Second, the non- conducting gap as well as the lack of high-quality
dispersive feature observed at EF shows the largest single crystals.
peak intensity at 0 , and when we increase the Figure 24 shows (a) crystal structure and
polar angle, the peak intensity is again enhanced (b) Brillouin zone of C6Ca. Since the Ca atom stacks
around the point in the next Brillouin zone. This with the rhombohedral coordination with an
strongly suggests the existence of a band below EF AAA sequence, the Brillouin zone of C6Ca is
around the point, which would likely be the inter- different from that of C8A. The projected surface
layer band. It is noted that this interlayer band is Brillouin zone has a hexagonal shape but is rotated
more clearly observed in C6Ca as we describe in by 30 with respect to the Brillouin zone of C8A
the next section. (Figure 22(b)).
C6Ca single crystals were usually grown by
the chemical vapor transport method (Weller et al.,
1.11.4.2 C6Ca
2005), while it is also produced by the alloy method
The discovery of superconductivity in C6Yb and (Emery et al., 2005). The inset in Figure 25 shows a
C6Ca (Weller et al., 2005) in 2005 has revived a picture of C6Ca single-crystal grown by the alloy
considerable interest since their Tc values (6.5 and method by heating kish graphite mixed with molten
400 Angle-Resolved Photoemission Spectroscopy
(a) (b)
Γ
M
: Ca
K
:C
L
X
T
Figure 24 (a) Crystal structure and (b) the Brillouin zone of C6Ca.
K(H)
Intensity (a.u.)
Intensity (a.u.)
M(L)
Γ(A)
Γ(A)
16 12 8 4 EF 16 12 8 4 EF
Binding energy (eV) Binding energy (eV)
Figure 26 Valence-band ARPES spectra of C6Ca measured along (a) the K and (b) M directions in the graphite Brillouin
zone (Sugawara et al., 2008).
C6Ca C6Ca
B′
EF
M
1 B
1.0 K
2 A
ky (Å–1)
0.0 M Γ A′
EF
Binding energy (eV)
K
High
–1.0
4
12
of band due to the superstructure of Ca interca-
lants, because the shape and size of Fermi surface
16 are distinctly different from those of the Fermi
M K Γ M K Low surface. On the other hand, the band structure
Wave vector calculations (Boeri et al., 2007; Calandra and Mauri,
Figure 27 Band structure of C6Ca determined by ARPES. 2005; Mazin, 2005; Mazin et al., 2007; Sanna et al., 2007)
The top panel shows an expansion of the near-EF region. have predicted an additional band at the point in
C6Ca (named the interlayer band), which has a sizable
and Rabii, 1982). In contrast, the small Fermi surface contribution from the intercalant states and produces a
observed at the point is not explained by the small free-electron-like spherical Fermi surface at the
simple rigid shift of graphite bands or the folding point. Considering a good agreement in the location
402 Angle-Resolved Photoemission Spectroscopy
and the shape, we assign the small Fermi surface shows dominant peaks at 50–75 and 160–180 meV
observed at the point to the interlayer band originating in the out-of-plane and the in-plane
(Ohno et al., 1979) in C6Ca. In fact, the observed vibrations of carbon atoms, respectively, in good
circular Fermi surface is not ascribed to the calculated agreement with the anomalies observed in the
two Fermi surfaces in C6Ca, since one calculated ARPES spectra. This suggests that electrons on the
Fermi surface has a hexagonal shape and the other circular Fermi surface at the point are strongly
is not centered at the point (Sanna et al., 2007; coupled to the out-of-plane vibrations while those
Mazin et al., 2007). on the triangular Fermi surface at the K point are
coupled dominantly with the in-plane phonons.
This experimentally demonstrates that the interlayer
1.11.4.2.3 Superconducting gap electrons strongly interact with the surrounding lat-
Figure 29 shows ARPES spectra near EF of C6Ca at tice unlike in the free-electron picture, as predicted
6 K measured at two representative Fermi vectors from the first-principles calculations (Sanna et al.,
(kF’s) on the circular Fermi surface at the point 2007; Boeri et al., 2007). The reason why electrons
and the triangular Fermi surface at the K point, at the K point are only weakly coupled to the out-of-
respectively, as indicated by circles in Figure 28. plane phonons may be the antisymmetry of the
Note that the spectral intensity shows a sudden wave function with respect to the graphene layer
drop at 80 meV on the circular Fermi surface (Calandra and Mauri, 2005). On the other hand, the
(point A) while a small but clear dip is observed at interlayer band forming the circular Fermi surface
170 meV in the spectrum measured on the triangu- at the point is expected to be strongly coupled to
lar Fermi surface (point B). We find that the energy the out-of-plane phonons due to the symmetry
position of dip does not vary in the momentum (Boeri et al., 2007). We note that a spectral dip
region around the kF point. This is not explained related to the coupling with the Ca vibrations is
by the presence of another band located around the not clearly observed in Figure 29, possibly because
dip energy, but is reminiscent of the strong many- the dip is broad (20–30 meV) compared to the Ca
body effects observed in high-Tc cuprates phonon energy (15 meV).
(Campuzano et al., 2004; Damascelli et al., 2003) Figure 30(a) shows ultrahigh-resolution ARPES
and Ba0.67K0.33BiO3 (Chainani et al., 2001). In fact, spectra near EF of C6Ca measured at two represent-
as seen in Figure 29, the Eliashberg function 2F(!) ative k points A and B as shown in Figure 28 at
calculated for C6Ca (Calandra and Mauri, 2005) temperatures below/above Tc (6 and 15 K). At point
A, the midpoint of the leading edge in the spectrum
at 6 K is shifted toward high binding energy by
C6Ca approximately 0.8 meV with respect to EF while the
Tc = 11.5 K
spectrum at 15 K has the midpoint at EF. This clearly
indicates the opening/closing of a superconducting
gap as a function of temperature. On the other hand,
at point B, the midpoint of the leading edge is located
at around EF even at 6 K, and the spectrum looks to
Intensity (a.u.)
Ca
0.5 (xy) B
the gap, we also plot the 6-K spectra measured at
points A9 and B9 on the same Fermi surfaces but on
different high-symmetry lines (Figure 28), since the
0.4 0.3 0.2 0.1 EF anisotropy, if it exists, should be largest between
Binding energy (eV) points A and A9 (B and B9). As seen from
Figure 29 ARPES spectra near EF of C6Ca at 6 K Figure 30(b), the leading edge at point A9 (B9)
measured at points A and B (see Figure 28), compared to almost coincides with that of points A (B). This
the Elieshberg function 2F(!) (Calandra and Mauri, 2005). suggests that the anisotropy of superconducting gap
Angle-Resolved Photoemission Spectroscopy 403
(a) (b)
C6Ca T=6K
Tc = 11.5 K A
Intensity (a.u.)
Intensity (a.u.)
A'
B
A B'
T=
6K ∆ = 1.8 meV
B Γ = 0.5 meV A
15 K
10 5 EF –5 10 5 EF –5
Binding energy (meV) Binding energy (meV)
Figure 30 (a) Superconducting gap of C6Ca measured at representative kF points A and B in Figure 28. (b) Top: direct
comparison of ARPES spectrum at T ¼ 6 K measured at four kF points in Fig. 28. Bottom: Numerical simulation of ARPES
spectrum at point A.
1970) that the and bands contact each other only and the temperature gradient on the surface. Here
at K point of the Brillouin zone in graphene and the we focus on the ARPES results of graphene film
energy at this contacting point is called the Dirac prepared by the third method.
energy (ED). It is obvious that ED is equal to EF in Graphene thin films were prepared by annealing
pristine (undoped) graphene. To experimentally an n-type Si-rich 6H-SiC(0001) single crystal with
investigate the electronic structure of graphene by resistive heating in an ultrahigh vacuum of 1 1010
ARPES, a clean surface of graphene has to be pre- Torr (Bostwick et al., 2007; Zhou et al., 2007, 2008;
pared under ultrahigh vacuum. There are several Rotenberg et al., 2008). By controlling the tempera-
methods to fabricate the graphene sample. First is ture and the heating time, it is possible to selectively
to peel off the bulk graphite many times known as a grow single-layer graphene and multilayer graphene
micro cleaving (Novoselov et al., 2005). Second is to with more than 10 sheets (Kusunoki et al., 2000).
evaporate/dissociate organic materials like ethylene Graphene films are characterized by the low-energy
on a clean metal surfaces such as Ni(111) (Nagashima electron diffraction (LEED) and the energy position
et al., 1994a), TaC(111) (Nagashima et al., 1994b), and of and bands observed by ARPES.
Pt(111) (Fujita et al., 2005). Third is to segregate a Figure 32(a) shows the LEED pattern of gra-
graphene sheet on SiC(0001) crystal surface by phene grown on 6H-SiC(0001) measured with
heating the SiC crystal (Vanbommel et al., 1975). primary electron energy of 107 eV. The LEED
First method has a difficulty in obtaining a large pattern of graphene on SiC(0001) consists of three
surface area suitable for ARPES measurements. In different spots: (1) the (1 1) spots from the bulk SiC
p p
the second method, we expect a strong interaction substrate (SiC (1 1)); (2) the weak 6 3 6 3
between the substrate and the graphene film, so that spots which originate in the bulk SiC layer bonded
near-EF energy bands of graphene are strongly with the interface graphene layer (buffer layer);
modulated by the hybridization by the energy and (iii) the C (1 1) spots due to the honeycomb
bands of the metallic substrate. The third method atomic pattern of graphene (Ni et al., 2008).
seems useful for ARPES measurement because the Figure 32(b) shows the valence-band ARPES
interaction between substrate and the graphene film spectra of graphene measured along the KM direc-
is much weaker than the second case, and a clean tion. Several highly dispersive bands are clearly
surface with the large area (few mm2) is easily observed, and the overall band dispersion is basically
obtained by controlling the annealing temperature similar to that of bulk graphite (see Figure 14(a)).
Graphene/SiC(0001)
(a) (b)
Graphene (1 x 1)
6 3x6 3
M
Si (1 x 1)
Intensity (a.u.)
(c) EF
π K
σ
Binding energy (eV)
12 Γ
16
16 12 8 4 EF
Γ K M
Wave vector Binding energy (eV)
Figure 32 (a) LEED pattern of graphene grown on 6H-SiC(0001). (b) Valence-band ARPES spectra of graphene.
(c) Experimentally determined band structure compared with the band calculation of graphene.
Angle-Resolved Photoemission Spectroscopy 405
We find a single band which rapidly disperses For example, the bottom of band at point in
toward EF on approaching the K point. Figure 32(c) graphene is shifted by about 0.7 eV toward the high
shows the ARPES-derived band structure of binding energy compared to that in graphite. As for the
graphene/SiC(0001) compared with the calculated band, the top (bottom) of dispersion at the (K) point
bands. The experimental band structure was obtained in graphene is located at 0.5 eV higher (0.7 eV lower)
by taking the second derivative of ARPES spectra and binding energy than that of graphite. These experi-
plotting the intensity as a function of wave vector and mental results indicate that the band width of both
binding energy. As expected, the band structure of and bands is larger in graphene than in graphite.
graphene consists of the highly dispersive and This quantitative difference in the band width may be
bands. The overall band structure is well reproduced ascribed to the difference in the lattice constant.
by the calculation, while some quantitative differences Actually, the wave vector of the K point estimated
are found in the energy position of the top of band from the present ARPES results is slightly different
and the bottom of band at the point. between graphene and graphite; 1.72 0.01 Å1 for
graphene and 1.70 0.01 Å1 for graphite. The
corresponding in-plane lattice vectors () are
1.11.5.2 Comparison with Graphite 2.44 0.01 and 2.46 0.01 Å, respectively, showing a
Figure 33 shows the comparison of experimental good agreement with the reported value (2.46 Å)
band structure between (a) graphene/SiC(0001) and (Lukesh and Pauling, 1950) for the case of graphite.
(b) graphite/SiC(0001). It is noted that the multilayer The lattice contraction in graphene may be due to a
graphene sample in this study shows the almost slight mismatch of the lattice constant between
identical band structure to that of bulk kish graphite graphite and SiC as observed by Raman spectroscopy
(Sugawara et al., 2006, 2007). Hence we refer to this (Ni et al., 2008). In addition to the and bands, we find
multilayer graphene as graphite in this section. As a weakly dispersing band around 3–4 eV binding
clearly visible in Figure 33, the band structures of energy in graphene, which is not predicted in the
graphene and graphite look very similar to each band calculation (Painter and Ellis, 1970) and may be
p p
other, consisting of the highly dispersive and due to the 6 3 6 3 structure of the buffer layer
bands. The band has the bottom of dispersion at (Mattausch and Pankratov, 2007; Emtsev et al., 2008).
around 8–9 eV at the point and the top at EF at the Figure 34 shows the plot of ARPES intensity
K point, while the band has the top at around around the K point as a function of the wave vector
4–5 eV at point and shows a strong downward and the binding energy for (a) graphene and (b)
dispersion toward the K point. However, detailed graphite. To remove the effect from the FD function,
comparison reveals several quantitative differences. the ARPES intensity is divided by the FD function
σ σ
4
Binding energy (eV)
8 8
K(H) M(L)
12 12
Γ(A)
16 16
Γ K M Γ(A) K(H) M(L)
Wave vector Wave vector
Figure 33 Comparison of valence-band structure between (a) graphene/6H-SiC(0001) and (b) graphite/6H-SiC(0001).
406 Angle-Resolved Photoemission Spectroscopy
High
0.6 0.6
0.8 kx = 0.8 kx =
0.0 Å–1 0.0Å–1
1.0 1.0
–0.04 0.0 0.04 Intensity (a.u.) –0.04 0.0 0.04 Intensity (a.u.) Low
kx (Å–1) kx (Å–1)
Figure 34 Comparison of near-EF band structure between (a) graphene/6H-SiC(0001) and (b) graphite/6H-SiC(0001),
together with the corresponding energy distribution curve at the K point.
convoluted with the Gaussian which reflects the in graphene estimated from the interval between the
instrumental resolution. The corresponding ARPES two peaks in the ARPES spectrum is about 280 meV.
spectra at the K point are also shown for comparison. This band gap has also been observed (Zhou et al.,
It is remarked in Figure 34(a) that the Dirac energy 2007, 2008) and discussed in terms of the sublattice-
ED of graphene is not at EF, but is about 0.4 eV below symmetry breaking due to the graphene–substrate
it, suggesting that the graphene sheet on SiC is doped interaction. On the other hand, it has also been
with electrons, as also evident from the appearance of reported that there is no gap opening at ED and the
the band at EF. It is speculated that electrons are observed small anomaly of band dispersion around
transferred from the n-type SiC substrate and/or the ED is explained in terms of the electron–plasmon
buffer layer (Bostwick et al., 2007; Rotenberg et al., coupling (Bostwick et al., 2007; Rotenberg et al.,
2008; Zhou et al., 2007; Zhou et al., 2008). On the other 2008). The reason for this difference is unclear at
hand, ED of graphite is at or very close to EF, as seen present, while it has also been discussed that the
in Figure 34(b). The band structure of graphite quantitative value of the band gap may depend on
grown on SiC shows a close resemblance to that of the sample preparation process (Zhou et al., 2008).
bulk kish graphite (see Figure 14) (Sugawara et al., As for the origin of band gap in graphene observed
2007). The energy splitting of band at the K point in the present ARPES study, it would be most likely
due to the interlayer coupling between graphene related to the interaction between the graphene sheet
sheets is clearly observed and the energy interval is and the SiC substrate. The band-gap size decreases
almost identical to that of kish graphite. A sharp on increasing the number of layers (Zhou et al., 2007),
quasi-particle peak appears at EF, and the Dirac and finally vanishes in multilayer graphite (graphite)
energy ED is at or very close to EF as in kish graphite. as shown in Figure 34. As described above, the pre-
These experimental results suggest that the graphite sent ARPES results have revealed that the lattice of
sample grown on SiC in the present study consists of graphene grown on SiC(0001) slightly contracts
a sufficiently large number of graphene sheets and compared with that of bulk graphite, as evident
hardly suffers the charge-transfer effect from the from the larger wave vector of the K point and the
substrate (Zhou et al., 2007; Ohta et al., 2007). In resultant larger bandwidth of the and bands in
contrast, the graphene sheet grown on SiC substan- graphene. It is also noted that a graphene sheet grown
tially suffers the charge-transfer effect as evident on TaC(111), which has a relatively large lattice
from the large chemical potential shift as seen in mismatch to the substrate, exhibits a substantially
Figure 34(a). Further, very interestingly, there large (1.3 eV) energy gap between the and
seems to be an energy gap between the and the bands (Oshima and Nagashima, 1997). All these
bands in graphene on SiC, while no or a negligibly experimental results suggest that the energy gap is
small gap is seen in graphite. The size of energy gap caused by the lattice strain due to the lattice
Angle-Resolved Photoemission Spectroscopy 407
mismatch between the graphene sheet and the sub- particular, the direct ARPES observation of the inter-
strate, and the gap gradually vanishes as the strain is layer state and its relationship to the superconductivity
gradually released at the topmost layers of multilayer in C6Ca has definitely resolved the long-standing
graphene. This indicates that the effect from the problem in the mechanism of superconductivity in
substrate should be taken into account when gra- GICs. It is expected that high-resolution ARPES will
phene grown on a SiC crystal is investigated, and be applied to well-characterized single- or multilayer
also suggests that the magnitude of the band gap can graphene as well as well-fabricated graphene ribbons
be controlled by tuning the lattice parameters of the to elucidate their novel electronic structures.
base substrate and resultant hybridization strength (See Chapters 1.02, 1.10, 1.12 and 2.05).
between the graphene sheet and the substrate. We
note that it is important to grow a graphene film
free from the lattice mismatch/strain to investigate References
the genuine Dirac-fermion-like behavior.
Allender D, Bray J, and Bardeen J (1973) Model for an exciton
mechanism of superconductivity. Physical Review B
7: 1020–1029.
1.11.6 Concluding Remarks and Bena C and Kivelson SA (2005) Quasiparticle scattering and
Summary local density of states in graphite. Physical Review B
72: 125432.
Bergeal N, Dubost V, Noat Y, et al. (2006) Scanning tunneling
The recent remarkable improvement of the energy spectroscopy on the novel superconductor CaC6. Physical
and momentum resolutions in ARPES enables us to Review Letters 97: 077003.
Black-Schaffer AM and Doniach S (2007) Resonating valence
determine experimentally the band structure and bonds and mean-field d-wave superconductivity in graphite.
the Fermi surface of various novel materials with Physical Review B 75: 134512.
great accuracy. In fact, the Fermi surface and the Boeri L, Bachelet GB, Giantomassi M, and Andersen OK (2007)
Electron–phonon interaction in graphite intercalation
superconducting gap of high-temperature cuprate compounds. Physical Review B 76: 064510.
superconductors have been precisely revealed by Bostwick A, Ohta T, Seyller T, Horn K, and Rotenberg E (2007)
ARPES. Further, electrons dressed with interactions, Quasiparticle dynamics in graphene. Nature Physics
3: 36–40.
namely quasi-particles, are directly observed by Brandt NB, Chudinov SM, and Ponomarev YG (1988) Graphite
high-resolution ARPES as a small anomaly (kink) in and Its Compounds. Amsterdam: North-Holland.
the energy dispersion near EF and have been inten- Calandra M and Mauri F (2005) Theoretical explanation of
superconductivity in C6Ca. Physical Review Letters
sively discussed in relation to the mechanism and the 95: 237002.
origin of the novel properties such as the high-Tc Campuzano JC, Ding H, Norman MR, et al. (1999) Electronic
superconductivity. Theoretical calculations, such as spectra and their relation to the (,) collective mode in
high-Tc superconductors. Physical Review Letters
that of the band structure, are directly compared with 83: 3709–3712.
the experimental results from ARPES experiments to Campuzano JC, Randeria M, and Norman MR (2004)
examine the validity of the theory. Thus, ARPES is Photoemission in the high temperature superconductors.
In: Bennemann KH and Ketterson JB, (eds.) The Physics of
now regarded as one of the essential experimental Superconductors, vol. 2, pp. 167–265. Berlin: Springer.
techniques in solid-state physics and materials Chainani A, Yokoya T, Kiss T, Shin S, Nishio T, and Uwe H
science. On the other hand, the discovery and/or (2001) Electron–phonon coupling induced pseudogap and
the superconducting transition in Ba0.67K0.33BiO3. Physical
the synthesis of new functional materials have been Review B 64: 180509.
successively achieved in recent years. In particular, Charlier JC, Michenaud JP, Gonze X, and Vigneron JP (1991)
carbon-based new materials such as fullerenes and Tight-binding model for the electronic properties of simple
hexagonal graphite. Physical Review B 44: 13237–13249.
carbon nanotubes have revived/stimulated the study Csanyi G, Littlewood PB, Nevidomskyy AH, Pickard CJ, and
of the mother material (graphite) and produced Simons BD (2005) The role of the interlayer state in the
additional new functional carbon-based material, electronic structure of superconducting graphite
intercalated compounds. Nature Physics 1: 42–45.
grapheme, and high-Tc GICs. Damascelli A, Hussain Z, and Shen ZX (2003) Angle-resolved
In this chapter, we have reviewed high-resolution photoemission studies of the cuprate superconductors.
ARPES studies of graphite, GICs, and graphene. Reviews of Modern Physics 75: 473–541.
DiVincenzo DP and Rabii S (1982) Theoretical investigation of
The high-resolution ARPES has revealed that this the electronic properties of potassium graphite. Physical
simple sp2 material exhibits a variety of interesting Review B 25: 4110–4125.
properties such as the interlayer state, the edge-localized Dresselhaus MS and Dresselhaus G (2002) Intercalation
compounds of graphite. Advances in Physics 51: 1–186.
state, the Dirac-fermion, and the Fermi-surface- Dynes RC, Narayanamurti V, and Garno JP (1978) Direct
dependent high-temperature superconductivity. In measurement of quasiparticle-lifetime broadening in a
408 Angle-Resolved Photoemission Spectroscopy
strong-coupled superconductor. Physical Review Letters Kurter C, Ozyuzer L, Mazur D, et al. (2007) Large energy gaps in
41: 1509–1512. CaC6 from tunneling spectroscopy: Possible evidence of
Eberhardt W, McGovern IT, Plummer EW, and Fisher JE (1980) strong-coupling superconductivity. Physical Review B 76:
Charge-transfer and non-rigid-band effects in the graphite 220502-4.
compound LiC6. Physical Review Letters 44: 200–204. Kusakabe K and Maruyama M (2003) Magnetic nanographite.
Emery N, Herold C, d’Astuto M, et al. (2005) Superconductivity Physical Review B 67: 092406.
of bulk CaC6. Physical Review Letters 95: 087003. Kusunoki M, Suzuki T, Hirayama T, Shibata N, and Kaneko K
Emtsev KV, Speck F, Seyller T, Ley L, and Riley JD (2008) (2000) A formation mechanism of carbon nanotube films on
Interaction, growth, and ordering of epitaxial graphene on SiC(0001). Applied Physics Letters 77: 531–533.
SiC{0001} surfaces: A comparative photoelectron Lamura G, Aurino M, Cifariello G, et al. (2006) Experimental
spectroscopy study. Physical Review B 77: 155303-10. evidence of s-wave superconductivity in bulk CaC6. Physical
Fretigny C, Saito R, and Kamimura H (1989) Electronic Review Letters 96: 107008.
structures of un-occupied bands in graphite. Journal of the Lanzara A, Bogdanov PV, Zhou XJ, et al. (2001) Evidence for
Physical Society of Japan 58: 2098–2108. ubiquitous strong electron–phonon coupling in high-
Fujita M, Wakabayashi K, Nakada K, and Kusakabe K (1996) temperature superconductors. Nature 412: 510–514.
Peculiar localized state at zigzag graphite edge. Journal of Lukesh JS and Pauling L (1950) The problem of the graphite
the Physical Society of Japan 65: 1920–1923. structure. American Mineralogist 35: 125.
Fujita T, Kobayashi W, and Oshima C (2005) Novel structures of Maeda F, Takahashi T, Ohsawa H, Suzuki S, and Suematsu H
carbon layers on a Pt(111) surface. Surface and Interface (1988) Unoccupied electronic band structure of graphite
Analysis 37: 120–123. studied by angle-resolved secondary-electron emission and
Gunasekara N, Takahashi T, Maeda F, Sagawa T, and inverse photoemission. Physical Review B 37: 4482–4488.
Suematsu H (1987) Electronic band-structure of C8Cs Maruyama M and Kusakabe K (2004) Theoretical prediction of
studied by highly angle-resolved ultraviolet photoelectron synthesis methods to create magnetic nanographite. Journal
spectroscopy. Journal of the Physical Society of Japan of the Physical Society of Japan 73: 656–663.
56: 2581–2589. Matsui T, Kambara H, Niimi Y, Tagami K, Tsukada M, and
Hahn JR and Kang H (1999) Vacancy and interstitial defects at Fukuyama H (2005) STS observations of landau levels at
graphite surfaces: Scanning tunneling microscopic study of graphite surfaces. Physical Review Letters 94: 226403.
the structure, electronic property, and yield for ion-induced Mattausch A and Pankratov O (2007) Ab initio study of
defect creation. Physical Review B 60: 6007–6017. graphene on SiC. Physical Review Letters 99: 076802.
Hannay NB, Geballe TH, Matthias BT, Andres K, Schmidt P, and Mazin II (2005) Intercalant-driven superconductivity in YbC6 and
MacNair D (1965) Superconductivity in graphitic CaC6. Physical Review Letters 95: 227001.
compounds. Physical Review Letters 14: 225–226. Mazin II, Boeri L, Dolgov OV, Golubov AA, Bachelet GB,
Heersche HB, Jarillo-Herrero P, Oostinga JB, Giantomassi M, and Andersen OK (2007) Unresolved
Vandersypen LMK, and Morpurgo AF (2007) Bipolar problems in superconductivity of CaC6. Physica C
supercurrent in graphene. Nature 446: 56–59. 460: 116–120.
Hengsberger M, Purdie D, Segovia P, Garnier M, and Baer Y McChesney JL, Bostwick A, Ohta T, Emtsev KV, Seyller T,
(1999) Photoemission study of a strongly coupled electron– Horn K, and Rotenberg E (2007) Massive enhancement of
phonon system. Physical Review Letters 83: 592–595. electron–phonon coupling in doped graphene by an
Heske C, Treusch R, Himpsel FJ, et al. (1999) Band widening in electronic singularity, ArXiv: 0705.3264.
graphite. Physical Review B 59: 4680–4684. McClure JW (1957) Band structure of graphite and de Haas–van
Iijima S (1991) Helical microtubules of graphitic carbon. Nature Alphen effect. Physical Review 108: 612–618.
354: 56–58. McGovern IT, Eberhardt W, Plummer EW and Fischer JE (1980)
Jensen P and Blase X (2004) Step barrier for gold adatoms and Band structures of graphite and graphite-intercalation
small clusters diffusing on graphite: An ab initio study. compounds as determined by angle resolved photoemission
Physical Review B 70: 165402. using synchrotron radiation. Physica B and C 99: 415–419.
Johnson LG and Dresselh G (1973) Optical properties of Miyamoto Y, Nakada K, and Fujita M (1999) First-principles
graphite. Physical Review B 7: 2275–2284. study of edge states of H-terminated graphitic ribbons.
Kaminski A, Mesot J, Fretwell H, et al. (2000) Quasiparticles in Physical Review B 59: 9858–9861.
the superconducting state of Bi2Sr2CaCu2O8þ
. Physical Molodtsov SL, Schiller F, Danzenbächer S, et al. (2003) Folded
Review Letters 84: 1788–1791. bands in photoemission spectra of La-graphite intercalation
Kawashima Y and Katagiri G (1995) Fundamentals, overtones, compounds. Physical Review B 67: 115105.
and combinations in the Raman spectrum of graphite. Muranyi F, Urbanik G, Kataev V, and Buchner B (2008) Electron
Physical Review B 52: 10053–10059. spin dynamics of the superconductor CaC6 probed by ESR.
Kim JS, Kremer RK, Boeri L, and Razavi FS (2006) Specific heat Physical Review B 77: 024507-5.
of the Ca-intercalated graphite superconductor CaC6. Nagashima A, Itoh H, Ichinokawa T, Oshima C, and Otani S
Physical Review Letters 96: 217002. (1994a) Change in the electronic states of graphite
Kittel C (1953) Introduction to Solid State Physics. New York: overlayers depending on thickness. Physical Review B
Wiley. 50: 4756–4763.
Kobayashi K (1993) Electronic structure of a stepped graphite Nagashima A, Tejima N, and Oshima C (1994a) Electronic
surface. Physical Review B 48: 1757–1760. states of the pristine and alkali-metal-intercalated monolayer
Koike Y, Suematsu H, Higuchi K, and Tanuma S (1978) graphite/Ni(111) systems. Physical Review B
Superconductivity in graphite-potassium intercalation 50: 17487–17495.
compound C8K. Solid State Communications Nakada K, Fujita M, Dresselhaus G, and Dresselhaus MS (1996)
27: 623–627. Edge state in graphene ribbons: Nanometer size effect and
Kroto HW, Heath JR, Obrien SC, Curl RF, and Smalley RE edge shape dependence. Physical Review B
(1985) C60 – Buckminsterfullerene. Nature 318: 162–163. 54: 17954–17961.
Kumigashira H, Kim H-D, Ito T, et al. (1998) High-resolution Ni ZH, Chen W, Fan XF, Kuo JL, Yu T, Wee ATS, and Shen ZX
angle-resolved photoemission study of LaSb. Physical (2008) Raman spectroscopy of epitaxial graphene on a SiC
Review B 58: 7675–7680. substrate. Physical Review B 77: 115416-6.
Angle-Resolved Photoemission Spectroscopy 409
Norman MR, Ding H, Campuzano JC, et al. (1997) Unusual Sugawara K, Sato T, Souma S, Takahashi T, and Suematsu H
dispersion and line shape of the superconducting state (2006) Fermi surface and edge-localized states in graphite
spectra of Bi2Sr2CaCu2O8þ
. Physical Review Letters studied by high-resolution angle-resolved photoemission
79: 3506–3509. spectroscopy. Physical Review B 73: 045124.
Novoselov KS, Geim AK, Morozov SV, et al. (2005) Two- Sugawara K, Sato T, Souma S, Takahashi T, and Suematsu H
dimensional gas of massless Dirac fermions in graphene. (2007) Anomalous quasiparticle lifetime and strong electron–
Nature 438: 197–200. phonon coupling in graphite. Physical Review Letters
Ohishi M, Shiraishi M, Nouchi R, Nozaki T, Shinjo T, and 98: 036801.
Suzuki Y (2007) Spin injection into a graphene thin film at Sugawara K, Sato T, and Takahashi T (2008) Fermi-surface-
room temperature. Japanese Journal of Applied Physics Part dependent superconducting gap in C6Ca. Nature Physics
2 – Letters and Express Letters 46: L605–L607. 5: 40–43.
Ohno T, Nakao K, and Kamimura H (1979) Self-consistent Sutherland M, Doiron-Leyraud N, Taillefer L, Weller T,
calculation of the band structure of C8K including the Ellerby M, and Saxena SS (2007) Bulk evidence for single-
charge-transfer effect. Journal of the Physical Society of gap s-wave superconductivity in the intercalated graphite
Japan 47: 1125–1133. superconductor C6Yb. Physical Review Letters
Ohsawa H, Takahashi T, Kinoshita T, Enta Y, Ishii H, and 98: 067003.
Sagawa T (1987) Unoccupied electronic band structure of Takahashi T, Gunasekara N, Sagawa T, and Suematsu H (1986)
graphite studied by angle-resolved inverse photoemission. Electronic band-structure of C8K studied by angle-resolved
Solid State Communications 61: 347–350. ultraviolet photoelectron spectroscopy. Journal of the
Ohta T, Bostwick A, McChesney JL, Seyller T, Horn K, and Physical Society of Japan 55: 3498–3502.
Rotenberg E (2007) Interlayer interaction and electronic Takahashi T, Tokailin H, and Sagawa T (1985) Angle-resolved
screening in multilayer graphene investigated with angle- ultraviolet photoelectron spectroscopy of the unoccupied
resolved photoemission spectroscopy. Physical Review band structure of graphite. Physical Review B
Letters 98: 206802. 32: 8317–8324.
Oshima C and Nagashima A (1997) Ultra-thin epitaxial films of Tatar RC and Rabii S (1982) Electronic properties of graphite
graphite and hexagonal boron nitride on solid surfaces. – a unified theoretical study. Physical Review B
Journal of Physics: Condensed Matter 9: 1–20. 25: 4126–4141.
Painter GS and Ellis DE (1970) Electronic band structure and Terashima K, Sato T, Komatsu H, Takahashi T, Maeda N, and
optical properties of graphite from a variational approach. Hayashi K (2003) Charge-density wave transition of 1T-VSe2
Physical Review B 1: 4747–4752. studied by angle-resolved photoemission spectroscopy.
Rotenberg E, Bostwick A, Ohta T, McChesney JL, Seyller T, and Physical Review B 68: 155108.
Horn K (2008) Origin of the energy bandgap in epitaxial Tombros N, Jozsa C, Popinciuc M, Jonkman HT, and van
graphene. Nature Materials 7: 258–259. Wees BJ (2007) Electronic spin transport and spin
Sanna A, Profeta G, Floris A, Marini A, Gross EKU, and precession in single graphene layers at room temperature.
Massidda S (2007) Anisotropic gap of superconducting Nature 448: 571–U4.
CaC6: A first-principles density functional calculation. Valla T, Fedorov AV, Johnson PD, and Hulbert SL (1999) Many-
Physical Review B 75: 020511. body effects in angle-resolved photoemission: Quasiparticle
Santoni A, Terminello LJ, Himpsel FJ, and Takahashi T (1991) energy and lifetime of a Mo(110) surface state. Physical
Mapping the Fermi surface of graphite with a display-type Review Letters 83: 2085–2088.
photoelectron spectrometer. Applied Physics A – Materials Valla T, Fedorov AV, Johnson PD, Xue J, Smith KE, and
Science and Processing 52: 299–301. DiSalvo FJ (2000) Charge-density-wave induced
Sato T, Matsui H, Takahashi T, et al. (2003) Observation of band modifications to the quasiparticle self-energy in 2H-TaSe2.
renormalization effects in hole-doped high-Tc Physical Review Letters 85: 4759–4762.
superconductors. Physical Review Letters 91: 157003. Vanbommel AJ, Crombeen JE, and Vantooren A (1975) LEED
Schäter I, Schlüter M, and Skibowski M (1986) Conduction- and auger electron observations of SiC(0001) surface.
band structure of graphite studied by combined angle- Surface Science 48: 463–472.
resolved inverse photoemission spectroscopy and target Vitali L, Schneider MA, Kern K, Wirtz L, and Rubio A (2004)
current spectroscopy. Physical Review B 35: 7663–7670. Phonon and plasmon excitation in inelastic electron
Schäfer J, Sing M, Claessen R, et al. (2003) Unusual spectral tunneling spectroscopy of graphite. Physical Review B
behavior of charge-density waves with imperfect nesting in a 69: 121414.
quasi-one-dimensional metal. Physical Review Letters Wakabayashi K, Fujita M, Ajiki H, and Sigrist M (1999) Electronic
91: 066401. and magnetic properties of nanographite ribbons. Physical
Schafhäl C (1840) Ueber die Verbingungen des Kohlenstoffes Review B 59: 8271–8282.
mit Silicium, Eisen und anderen Metallen, welche die Weller TE, Ellerby M, Saxena SS, Smith RP, and Skipper NT
verschiedenen Gallungen von Roheisen, Stahl und (2005) Superconductivity in the intercalated graphite
Schmiedeeisen bilden. J. Prakt. Chem. 21: 129–157. compounds C6Yb and C6Ca. Nature Physics 1: 39–41.
Siebentritt S, Pues R, Rieder KH, and Shikin AM (1997) Surface Zhou SY, Gweon GH, Fedorov AV, et al. (2007) Substrate-
phonon dispersion in graphite and in a lanthanum graphite induced bandgap opening in epitaxial graphene. Nature
intercalation compound. Physical Review B 55: 7927–7934. Materials 6: 770–775.
Strocov VN, Charrier A, Themlin JM, et al. (2001) Photoemission Zhou SY, Siegel DA, Fedorov AV, et al. (2008) Origin of the
from graphite: Intrinsic and self-energy effects. Physical energy bandgap in epitaxial graphene – Reply. Nature
Review B 64: 075105. Materials 7: 259–260.
1.12 Theory of Superconductivity in Graphite Intercalation
Compounds
Y Takada, University of Tokyo, Kashiwa, Chiba, Japan
ª 2011 Elsevier B.V. All rights reserved.
Glossary
Allen-dynes formula In 1975, Allen and Dynes Density functional theory (DFT) A theory to
modified the McMillan’s formula for Tc, producing a describe an interacting many-body system in terms
rather large change in Tc for the electron–phonon of its density distribution in three dimensions.
nondimensional coupling constant larger than 2. Eliashberg function In the phonon mechanism,
Coulomb pseudopotential A phenomenological the phonon-mediated attractive interaction is
parameter to estimate the effect of the Coulomb characterized by both the electron–phonon
repulsion on the Cooper pairing by taking into coupling (!) and the phonon density of states F(!),
account both retardation and screening effects. but in calculating Tc only their combination
410
Theory of Superconductivity in Graphite Intercalation Compounds 411
(!)2F(!), which is called the Eliashberg function, is based on the numerical solution of the Eliashberg
important. equation for various systems.
First-principles calculation A calculation is said Phonon mechanism If the Cooper pairs are
to be from first principles or ab initio if it starts created by the attractive interaction induced by
directly at the level of established laws of physics virtual exchange of phonons between electrons,
which contain no fitting or phenomenological the superconductivity is considered to be brought
parameters. about by the phonon mechanism.
Graphite intercalation compounds Plasmon mechanism In the low-density electron
(GICs) Complex materials in which the graphite gas, plasmons can play a similar role to phonons in
layers remain largely intact, and the guest element superconductivity. We indicate this role as the
or molecule is inserted between the layers. plasmon mechanism.
GW approximation In 1965, Hedin proposed an Spin-fluctuation mechanism In a system with
approximation scheme to calculate the electron strong spin fluctuations at either long wavelengths
self-energy in terms of the one-electron Green’s (related to a ferromagnetic instability) or short
function G and the effective electron–electron wavelengths (related to an antiferromagnetic
interaction W without any vertex corrections. If the instability), the fluctuations can play a role similar to
same scheme is applied to superconductivity, it is phonons in superconductivity. We indicate this role
nothing but the Eliashberg theory for as the spin-fluctuation mechanism.
superconductivity. Stage In a GIC, not every layer is necessarily
Isotope effect If Tc depends directly or indirectly intercalated by guest elements or molecules. In the
on the mass M of the ions building up the lattice, we so-called stage 1 GICs like KC8, graphite and
call the dependence the isotope effect. In the intercalated layers alternate, while in stage 2 GICs
standard Bardeen–Cooper–Schrieffer (BCS) such as KC24, two graphite layers with no guest
theory, Tc changes in proportion to M with the material in between alternate with an intercalated
isotope coefficient ¼ 0.5. layer. Similarly, we can define stage n with n > 2.
McMillan’s formula In 1968, McMillan proposed a Vertex function A factor to describe the change in
simple formula for Tc in the phonon mechanism the electron charge during scattering processes
composed of an average phonon energy !0, the due to the modification of the many-body wave
electron–phonon nondimensional coupling function. Its deviation from unity (or the bare
constant , and the Coulomb pseudopotential , coupling) is called vertex corrections.
(a) (b)
CaC6
101 YbC6
LiC2 SrC6
d
Tc (K)
100
KC8
RbC8 C5C8
10–1
10–2
4.0 4.5 5.0 5.5 6.0
d (Å)
: M : C
Figure 1 (a) Crystal structure of MCx (x ¼ 2, 6, 8). The case of x ¼ 6 is illustrated here, in which the metal atoms, M’s, are
arranged in a rhombohedral structure with the stacking sequence, implying that M occupies the same spot in the
framework of a graphene lattice at every three layers (or at the distance of 3d with d the distance between the adjacent
graphite layers). (b) Superconducting transition temperature Tc observed in the first-stage alkali- and alkaline-earth-
intercalated graphites plotted as a function of d.
1982; Koike et al., 1978; Kobayashi and Tsujikawa, 1.12.1.3 Central Issues
1979; Koike et al., 1980), but only a limited success
In spite of all those efforts and the existence of such a
was achieved at that time; the highest attained Tc was
generally accepted picture, there remain several very
around 2–5 K in the last century. For example, it is
important and fundamental questions:
1.9 K in LiC2 (Belash et al., 1989).
A breakthrough occurred in 2005 when Tc went up 1. Can we understand the mechanism of supercon-
to 11.5 K in CaC6 (Weller et al., 2005; Emery et al., ductivity in both alkali GICs with Tc in the range
2005) (and even to 15.4 K under pressures up to 0.01–1.0 K and alkaline-earth GICs with Tc in the
7.5 GPa (Gauzzi et al., 2007)). In other alkaline-earth range 1–10 K from a unified point of view? In
GICs, the values of Tc are 6.5 and 1.65 K for YbC6 other words, is there any standard model for
(Weller et al., 2005) and SrC6 (Kim et al., 2006), respec- superconductivity in GICs with Tc ranging over
tively, as indicated in Figure 1(b). Since then, very three orders of magnitude?
intensive experimental studies have been made in 2. What is the actual reason why Tc is enhanced so
those and related compounds (Emery et al., 2005; abruptly (or by about a hundred times) by just
Kim et al., 2007, 2006; Kurter et al., 2007). Theoretical substituting K with Ca, the atomic mass of which
studies have also been performed mainly by making is almost the same as that of K? In terms of the
state-of-the-art first-principles calculations of the standard model, what are the key controlling phy-
electron–phonon coupling constant to account for sical parameters to bring about this huge
the observed value of Tc for each individual super- enhancement of Tc ? This change of Tc from
conductor (Mazin, 2005; Csanyi et al., 2005; Calandra KC8 to CaC6 is probably the most important
and Mauri, 2005; Sanna et al., 2007). Those experi- issue in exploring superconductivity across the
mental/theoretical works have elucidated that, entire family of GICs.
although there are some anisotropic features in the 3. Is there any possibility to make a further enhance-
superconducting gap, the conventional phonon-driven ment of Tc in GICs ? If possible, what is the
mechanism to bring about s-wave superconductivity optimum value of Tc expected in the standard
applies to those compounds. This picture of super- model and what kind of atoms should be interca-
conductivity is confirmed by, for example, the lated to realize the optimum Tc in actual GICs?
observation of the Ca isotope effect with its exponent
¼ 0.50, the typical Bardeen–Cooper–Schrieffer In order to provide reliable answers to the
(BCS) value (Hinks et al., 2007). above questions, it is indispensable to make a
Theory of Superconductivity in Graphite Intercalation Compounds 413
first-principles calculation of Tc with sufficient accu- competition between the attractive and the repulsive
racy and predictive power. Recently, such a interactions between electrons. Accumulation of such
calculation has been completed by the present author, information might pave the way to the synthesis of a
and based on the calculation, some interesting predic- room-temperature superconductor, a big dream in
tions have been proposed (Takada, 2009a; 2009b). materials science. From this perspective, a continu-
The present chapter not only reports some details of ous effort has been made for a long time to develop a
this work on the superconducting mechanism in GICs good theory for first-principles calculations of Tc,
but also makes a brief summary of the current status starting with a microscopic Hamiltonian.
of the theories for first-principles calculations of Tc.
1.12.2.2 McMiland’s and Allen-Dynes’
1.12.1.4 Organization of This Chapter Formulas for Tc
This chapter is organized as follows: In Sections In the phonon mechanism, for example, there has
1.12.2–1.12.4, a critical review of the theories for been a rather successful framework for this purpose,
quantitative calculations of Tc is given. More speci- known as the McMillan’s formula (McMillan, 1968)
fically, we make comments on the theories based on or its revised version (the Allen-Dynes’ formula)
the McMillan’s or the Allen-Dynes’ formula employ- (Allen and Dynes, 1975; Allen and Mitrović, 1982;
ing the concept of the Coulomb pseudopotential Carbotte, 1990), both of which are derived from the
in Section 1.12.2. In Section 1.12.3 we explain the Eliashberg theory of superconductivity (Eliashberg,
theory on the level of the so-called G0W0 approxima- 1960). In this framework, the task of a microscopic
tion, where Tc can be obtained without using , a calculation of Tc is reduced to the evaluation of the
very important advantage. The same advantage can so-called Eliashberg function 2F(!) from first prin-
be enjoyed in the density functional theory for super- ciples; the function 2F(!) enables us to obtain both
conductors, which will be addressed in Section 1.12.4. the electron–phonon coupling constant and the
In Sections 1.12.5–1.12.8, a review on superconduc- average phonon energy !0, through which we can
tivity in GICs is given; starting with a summary of make a first-principles prediction of Tc by
the experimental works in Section 1.12.5, a standard additionally introducing a phenomenological
model for considering the mechanism of supercon- parameter (the Coulomb pseudopotential (Morel
ductivity in GICs is introduced in Section 1.12.6. In and Anderson, 1962)) in order to roughly estimate
Section 1.12.7, the calculated results of Tc are given the effect of the Coulomb repulsion between
for the alkaline-earth GICs and they are compared electrons on Tc.
with the experimental results. The prediction of the At present this framework is usually regarded as
optimum Tc is given in Section 1.12.8. Finally in the canonical one for making a first-principles predic-
Section 1.12.9, the conclusion of this chapter is tion of Tc. In fact, the superconducting mechanism of
given, together with some perspectives on the many (so-called weakly correlated) superconductors is
researches in this and related fields in the future. believed to be clarified by using this framework,
whereby the key phonon modes to bring about super-
conductivity are identified. We can point out that
1.12.2 First-Principles Calculation of Tc superconductivity in MgB2 with Tc ¼ 39 K is a good
recent example (Bohnen et al., 2001; Choi et al., 2002a;
1.12.2.1 Goal of the Problem
2002b; Kong et al., 2001) to illustrate the power of this
It would be one of the ultimate goals in the enterprise framework. The case of CaC6 has also been investi-
of condensed matter theory to make a reliable pre- gated along this line of theoretical studies (Mazin,
diction of Tc only through the information on 2005; Calandra and Mauri, 2005).
constituent elements of a superconductor in consid-
eration. A less ambitious yet very important goal is to
1.12.2.3 Coulomb Pseudopotential
make an accurate evaluation of Tc directly from a
microscopic (model) Hamiltonian pertinent to the Nevertheless, this framework is not considered to be
superconductor. If we could find the dependence of very satisfactory, primarily because a phenomenolo-
Tc on the parameters specifying the model gical parameter is included in the theory.
Hamiltonian, we could obtain a deep insight into Actually, it cannot be regarded as the method of
the mechanism of superconductivity or the predicting Tc in the true sense of the word, if the
414 Theory of Superconductivity in Graphite Intercalation Compounds
corresponds to the GW approximation with respect to where the gap function p and the pairing interac-
the phonon-mediated attractive interaction between tion K 9p;p9 are, respectively, defined by
electrons. Z
1
d!
Let us explain this G0W0 scheme here (we employ p X 2"p ImF R ðp; !Þ ð5Þ
0
units in which h ¼ kB ¼ 1). For simplicity, imagine
the three-dimensional (3D) electron gas in which an and
electron is specified by momentum p and spin . If Z
1
2 "p þ "p9
we write the electron annihilation operator by cp, we Kp;p9 ¼ d V ðp – p9; iÞ ð6Þ
can define the abnormal thermal Green’s function 0 þ ð"p þ "p9 Þ2
2
with !p being the fermion Matsubara frequency. At 1.12.3.2 Comments on the Formulation
T ¼ Tc where the second-order phase transition Five comments are in order on this framework:
occurs, this function satisfies the following exact
gap equation: 1. Based on Equations (4) and (6), we can obtain Tc
directly from the microscopic one-electron dis-
F ðp; i!p Þ ¼ – Gðp;X ÞGð – p; – i!p Þ
i!pX persion relation "p and the effective electron–
Tc I˜ðp; p9; i!p ; i!p9 ÞF ðp9; i!p9 Þ electron interaction V(q, i) with no need to
!p9 p9 separate the phonon-mediated attraction from
ð2Þ the Coulomb repulsion.
2. In spite of the similarity of Equation (4) to the
where G(p, i!p) is the normal Green’s function and
˜ p9; i!p ; i!p9 Þ is the irreducible electron–electron BCS gap equation, artificial cutoffs involved in
Iðp;
constructing the BCS model are avoided in the
effective interaction. present scheme; natural cutoffs are automatically
Now, in the spirit of the G0W0 approximation, we introduced by the calculation of Kp;p9 defined in
will replace G(p, i!p) by the bare one G0(p, i!p) X Equation (6).
(i!p "p)1 in Equation (2), where "p(¼ p2/2m ) 3. Except for the spin-singlet pairing, no assumption
is the bare one-electron dispersion relation with m is made on the dependence of the gap function on
the band mass and the chemical potential. We will angular valuables in deriving Equation (4), so that
also consider the case in which Iðp; ˜ p9; i!p ; i!p9 Þ is
this gap equation can treat any kind of anisotropy
well approximated as a function of only the variables in the gap function, indicating that it can be
(p p9, i!p i!p9) to write applied to s-wave, d-wave, etc. and even their
˜ p9; i!p ; i!p9 Þ ¼ V ðp – p9; i!p – i!p9 Þ
Iðp; ð3Þ mixture like (sþd)-wave superconductors.
4. As can be seen by its definition, the gap function
just like the effective interaction in the random- p in Equation (5) does not correspond to the
phase approximation (RPA), though we do not physical energy gap except in the weak-coupling
intend to confine ourselves to the RPA at this region. Similarly, Kp;p9 is not a physical entity.
point. By substituting Equation (3) into Equation Both quantities are introduced for the mathema-
(2) and making an analytic continuation on the ! tical convenience so as to make Tc invariant in
plane to transform F(p, i!p) to the retarded function transforming Equation (2) into Equation (4). The
FR(p, !) on the real-! axis, we get a gap equation key point here is that we need not solve the full
for FR(p, !). Then, by taking the imaginary parts in gap equation (2) but a much simpler one (4) in
both sides of the gap equation and integrating order to obtain Tc in Equation (2). Of course, if we
over the ! variable, we finally obtain an equation want to know the physical gap function rather
depending only on the momentum variable p. than p to compare with experiment, we need to
Concretely, the equation can be cast into the solve the full gap equation, Equation (2), with Tc
following BCS-type gap equation: determined by Equation (4).
5. Historically, Cohen was the first to evaluate Tc in
X p9 "p9
p ¼ – tanh Kp;p9 ð4Þ degenerate semiconductors on the level of the
p9
2"p9 2Tc G0W0 approximation (Cohen, 1964; 1969).
416 Theory of Superconductivity in Graphite Intercalation Compounds
Unfortunately, the pairing interaction is not cor- Taking these situations into account, we have
rectly derived in his theory, as explicitly pointed assumed that the superconductivity is brought about
out by the present author (Takada, 1980) who, by the polar-coupling phonons associated with the
instead, has succeeded in obtaining the correct stress-induced ferroelectric phase transition. Then
pairing interaction (Takada, 1978) by consulting we have calculated the effective electron–electron
the pertinent work of Kirzhnits et al. (1973). interaction V(q, i) in the RPA, in which the long-
range attraction induced by the virtual exchange of
polar-coupled phonons is included with the long-
1.12.3.3 Assessment: Application to SrTiO3 range Coulomb repulsion on the same footing. By
In order to assess the quality of this basic framework substituting this V(q, i) into Equation (6), we have
of calculating Tc from first principles, we have obtained Tc directly from a microscopic model and
applied it to SrTiO3 and compared the results with the results of Tc are in surprisingly good quantitative
experiments (Takada, 1980). This material is an insu- agreement with experiment, as shown in Figure 2.
lator and exhibits ferroelectricity under a uniaxial This success indicates that the present basic frame-
stress of about 1.6 kbar along the [100] direction, work including the adoption of the RPA is very useful
but it turns into an n-type semiconductor by either at least in the polar-coupled phonon mechanism.
Nb doping or oxygen deficiency, whereby the con-
duction electrons are introduced in the 3d band of Ti
around the point with the band mass of m 1.8me
(me: the mass of a free electron). At low temperatures, 1.12.4 Density Functional Theory for
superconductivity appears and the observed Tc Superconductors
shows interesting features; Tc depends strongly on
1.12.4.1 Hohenberg–Kohn–Sham Theorem
the electron concentration n and it is optimized with
Tc 0.3 K at n 1020 cm3. Its dependence on the Recently, much attention has been paid to an exten-
pressure is unsual; Tc decreases rather rapidly with sion of the density functional theory (DFT) to treat
hydrostatic pressures, but it increases with the [100] superconductivity, mainly because it provides
uniaxial stress. another scheme for first-principles calculations of
0.04
0.1
ΔTc (K)
0
–0.04
0.05
–0.08
–0.12 n = 6.3 × 10 cm
19 –3
30
m* = 1.8me : [100]
: [110]
ω t (cm–1)
0.01 20 : H.P.
0.05 0.1 0.5 1 5 10
n (1020 cm–3) 10
level with its energy eigenvalue "i in the KS equation where i is the gap function for ith KS level. In just
have no physical relevance; they are merely intro- the same way as its energy eigenvalue "i (which is
duced for the mathematical convenience so as to measured from the chemical potential), i is not the
obtain the exact n(r) in connecting the noninteracting quantity to be observed experimentally but just
reference system with the real many-electron system. introduced for the mathematical convenience so as
The Hohenberg–Kohn theorem can be applied to to obtain the exact Tc by solving this BCS-type
the ordered ground state as well on the understand- equation, Equation (7). Similarly, the pair interaction
ing that the order parameter itself is regarded as a Kij, defined as the second-functional derivative of
functional of n(r). In providing some approximate Fxc[n, ] with respect to and , does not have
functional form for Fxc[n], however, it would be any direct physical meaning, either.
more convenient to treat the order parameter as an We note here the very impressive fact that the
additional independent variable. For example, in final forms for the two gap equations, Equations (4)
considering the system with some magnetic order, and (7), are exactly the same, in spite of the fact that
we usually employ the spin-dependent scheme in they are derived from quite different foundations and
which the fundamental variable is not n(r) but the reasoning. We also note that because of this similar-
spin-decomposed density n(r), leading to the spin- ity, we may judge that, as long as Kij is properly
polarized exchange-correlation energy functional chosen, the physics described by is also included
Fxc[n], based on which the spin-dependent KS in the framework of DFT for superconductors, at
equation is formulated. least to the extent that it is included in the G0W0
scheme explained in Section 1.12.3.
1.12.4.2 Gap Equation
Similarly, in treating the superconducting state in the
framework of DFT, it is better to construct the energy
1.12.4.3 Applications
functional by employing both n(r) and the electron-
pair density (r,r9)(X h"(r)#(r9)i) as basic vari- In 2005, this DFT framework was extended to expli-
ables, leading to the introduction of the exchange- citly take care of the phonon-mediated attractive
correlation energy functional Fxc[n(r), (r,r9)], where interaction (Lüders et al., 2005) and it has been
a(r) is the electron annihilation operator (Oliveira applied to many superconductors (Sanna et al., 2007;
et al., 1988; Kurth et al., 1999). In accordance with this Marques et al., 2005; Floris et al., 2005, 2007; Profeta
addition of the order parameter as a fundamental et al., 2006; Sanna et al., 2006; Floris et al., 2007). In
variable to DFT, not only the exchange-correlation order to perform these calculations for actual super-
potential Vxc(r) but also the exchange-correlation pair conductors, it is necessary to provide a concrete form
418 Theory of Superconductivity in Graphite Intercalation Compounds
for Fxc[n, ]. In the judgment of the present author, issue extends more than four decades. In 1965, the
the presently available form for Fxc[n, ] contains the first report of superconductivity was made for KC8,
information equivalent to that included in the RbC8, and CsC8 (Hannay et al., 1965) in which Tc was
Eliashberg theory for the part of the phonon- not reliably determined; it depended very much on
mediated attraction, indicating that no vertex correc- samples. Subsequent works (Alexander et al., 1981;
tions are considered in this treatment, while for the Chaiken et al., 1990; Iye and Tanuma, 1982; Koike
part of the Coulomb repulsion, it contains only very et al., 1978, 1980; Kobayashi et al., 1979; Kobayashi
crude physics; the screening effect is treated in the and Tsujikawa, 1981; Pendrys et al., 1981) confirmed
Thomas–Fermi static-screening approximation, the occurrence of superconductivity in KC8 with
which is nothing but the result of the RPA only in Tc ¼ 0.15 K, but superconductivity did not appear
the static and the long-wavelength limit, forgetting in RbC8 and CsC8 when Tc was down to 0.09 and
the detailed dynamical nature of the screening effect. 0.06 K, respectively. Later works have found that Tc
Mainly for this reason, Tc in the present form of is actually 26 mK for RbC8, but no superconductivity
Fxc[n, ] is not expected to be very accurate, even is found in either LiC6 or the second- or higher-stage
though the calculated results for Tc seem to be in alkali GICs, though the calculation of Tc based on
good agreement with experiment. the McMillan’s formula (McMillan, 1968) predicted
an observable value of Tc even for KC24 (Kamimura
et al., 1980; Inoshita and Kamimura, 1981). It seems
1.12.4.4 Basic Problems
that the usual first-principles calculation of 2F(!)
In relation to the above point, it would be appropriate tends to provide an unrealistically large contribution
to give the following comment: in the calculations of from the intralayer high-energy carbon oscillations
the normal-state properties in the local-density approx- to . This unfavorable tendency in the calculation of
imation (LDA) and generalized gradient approximation seems to prevail even in CaC6 (Calandra and
(GGA) (Perdew et al., 1997) to DFT, we usually antici- Mauri, 2005).
pate that errors in the calculated results are of the order The anisotropy of the critical magnetic field Hc2
of 1 and 0.3 eV for LDA and GGA, respectively. These was also a matter of interest, drawing attention of
errors are much larger than those expected in the both experimentalists (Chaiken et al., 1990;
calculation of quantum chemistry (0.05 eV). In DFT Dresselhaus et al., 1989; Iye and Tanuma, 1982;
for superconductors, calculations of Tc (which is of the Koike et al., 1980; Roth et al., 1985) and theorists
order of 0.001 eV in general) are done simultaneously (Jishi et al., 1991, 1992). Note that the gap function
with those for the normal-state properties. This implies p defined in Equation (4) has nothing to do with the
that the errors anticipated for Tc would be very large anisotropic behavior of Hc2, though in developing a
compared to Tc itself. phenomenological theory (Jishi et al., 1991, 1992)
We should also point out that the present form for some critical comments were made on the results of
Fxc[n, ] is not useful in the discussion of the electronic p (Takada, 1982) with the assumption that the
mechanisms like the plasmon and the spin-fluctuation anisotropy in Hc2 should reflect on p.
ones, prompting us to improve on the approximate In search of higher Tc, many attempts have
form for Fxc[n, ]. In addition, there are several pro- been made to synthesize new GIC superconductors
blems in the fundamental theory; for example, it is by such as NaC2 (Tc ¼ 5K) (Belash et al., 1987), LiC2
no means clear whether the second-functional deriva- (Tc ¼ 1.9K) (Belash et al., 1989), and alkali-metal
tive of Fxc[n, ] is a well-defined quantity, in just the amalgams such as KHgC4 (Tc ¼ 0.73 K) and KHgC8
same way as we have already experienced in the (Tc ¼ 1.90 K) (Alexander et al., 1980; Tanuma, 1981;
energy-gap problem (Perdew et al., 1997, 1983; Sham Koike and Tanuma, 1981; Pendrys et al., 1981;
and Schlüter, 1983) in semiconductors and insulators. Alexander et al., 1981; Iye and Tanuma, 1982), but
a larger enhancement of Tc was not achieved until
CaC6 was found in 2005 with Tc ¼ 11.5 K
1.12.5 Experiment on (Weller et al., 2005). Subsequently, many works
Superconductivity in GICs have been done on alkaline-earth GIC superconduc-
tors (Emery et al., 2005; Hinks et al., 2007; Kadowaki
From this section, let us get back to the review on et al., 2007; Kim et al., 2006, 2007; Kurter et al., 2007;
superconductivity in GICs. As briefly mentioned in Lamura et al., 2006; Sugawara et al., 2009;
Section 1.12.1, the history of the researches on this Valla et al., 2009), but none ever succeeded in
Theory of Superconductivity in Graphite Intercalation Compounds 419
synthesizing a new GIC with Tc larger than 15.4 K 2. The valence electrons released from M will trans-
which was observed in CaC6 under pressures (Gauzzi fer either to the graphite bands or to the 3D
et al., 2007). Thus, some new idea seems to be needed band composed of the intercalant orbitals and the
to further enhance Tc. The present author hopes that graphite interlayer states (Posternak et al., 1984;
the suggestions given in Section 1.12.8 help experi- Halzwarth et al., 1984; Koma et al., 1986). We shall
mentalists synthesize a new GIC superconductor define the factor f as the branching ratio between
with Tc much higher than 10 K. these two kinds of bands. For example, Zf and
Z(1 f) electrons will go to the and the 3D
bands, respectively.
1.12.6 Standard Model for 3. The electrons in the graphite bands are char-
Superconductivity in GICs acterized by the 2D motion with a linear
1.12.6.1 Characteristic Features of the dispersion relation (known as a Dirac cone in the
System case of graphene) on the graphite layer.
4. The dispersion relation of the graphite interlayer
Basically because GICs are not recognized as strongly band is very similar to that of the 3D free-electron
correlated systems, the usual ab initio self-consistent gas, folded into the Brillouin zone of the graphite
band structure calculation is very useful in elucidating (Csanyi et al., 2005). Thus, its energy level is very
the important features of the electronic structures of high above the Fermi level in the graphite,
GICs in the normal state. According to such calcula- because the amplitude of the wave function for
tions, it is found that there is no essential qualitative this band is small on the carbon atoms. In MCx, on
difference between alkali and alkaline-earth GICs (see the other hand, the cation MZþ is located in the
Figure 3). The main common features among these
interlayer position where the amplitude of the
GICs may be summarized in the following way:
wave functions is large, lowering the energy
1. In MCx, each intercalant metal atom acts as a level of the interlayer band below the Fermi
donor and changes from a neutral atom M to an level. The dispersion of the interlayer band is
ion MZþ with valence Z. modified from that of the free-electron gas because
(a) (b)
3D band
EF
–1eV
KF3D
D
F2
K
2D band
KB
Z
(η, η, η,) (η, –η, 0) (η, η, 0) (η, 0, 0)
L Γ χ X Γ T KF3D Kz
Figure 3 (a) Band structure of CaC6 (Calandra and Mauri, 2005). (b) Fermi surface of KC8 (Wang et al., 1991). Both materials
are characterized by the common feature that the electronic system is composed of the 2D bands of graphite and the 3D
interlayer band.
420 Theory of Superconductivity in Graphite Intercalation Compounds
>
band mass m and the Fermi energy EF. Here the f
2D π-electrons
d v M Z+
M Z+
M Z+ e–
>
e–
3D itinerant
hybridization occurs with s-orbitals, allowing us to 1–f electrons: m* M Z+
M Z+ M Z+
energies of LO- and LA-phonons, respectively, and with experiment. Note that smaller values of Tc are
V0(q) is the bare Coulomb interaction 4 e2/q2. (The obtained for heavier alkali atoms because of the
subscript ‘0’ indicates that it is the bare interaction to be smaller couplings as characterized by both !p and
screened by both 2D and 3D mobile electrons.) p . This success indicates that the present simple
!
Owing to the coupling with valence electrons, model applies well at least to alkali GIC
both !LO(q) and !LA(q) depend on f, but the superconductors.
f-dependence is not important, if we write the
phonon-mediated interaction in terms of the corre-
sponding transverse phonon energies, !TO(q) and
1.12.7 Superconductivity in Alkaline-
!TA(q). Thus, we specify the phonon energies in
Earth GICs
terms of !TO(q) and !TA(q). In actual calculations,
we assume that !TO(q) ¼ !t(¼ constant) and 1.12.7.1 CaC6
!TA(q) ¼ ctjqj, with !t of the order of 150 K and ct
Now let us consider alkaline-earth GIC supercon-
of the order of 105 cm s1 for the oscillation perpen-
ductors. We shall investigate them by adopting the
dicular to the graphite plane.
same simple model and by using exactly the same
calculation code developed in 1982 in order to see
1.12.6.3 Calculation of Tc for Alkali-Doped whether the model and therefore the picture on the
GICs mechanism of superconductivity successfully applied
to alkali GIC superconductors can also be relevant to
By combining this polar-phonon-mediated attrac- these newly synthesized superconductors or not
tive interaction W0(q, !) with the bare Coulomb (Takada, 2009a; 2009b). The parameters specifying
interaction between electrons V0(q) on the same the model will be changed in the following way, if
footing and considering the polarization effects of CaC6 is considered instead of KC8:
both 2D and 3D electrons, we faithfully calculate
V(q, !) the effective interaction between 3D elec- 1. Because the valence Z changes from monovalence
trons in the RPA (see Figure 4(b)). The obtained to divalence, the attractive interaction W0, which
V(q, !) is put into the kernel, Equation (6), of the is in proportion to Z2, increases by 4 times.
gap equation (4) to obtain Tc from first principles. 2. The interlayer distance d decreases from 5.42 to
The calculated results for Tc in alkali GICs are 4.524 Å, so that the 3D electron density
plotted as a function of f in Figure 5 to find that n increases.
the overall magnitude of Tc is in the range of 3. The factor f to determine the branching ratio
0.1–0.01 K for f 0.5, which is in good agreement decreases from about 0.6 to 0.16.
4. The effective band mass for the 3D interlayer
band m increases from me to about 3me.
LiC2 Z=1
1 KC8 m* = me 5. The atomic number of the ion A hardly changes
from 39.1 to 40.1.
0.1 o
Experiment Considering the changes in the parameters, we have
Tc (K)
0.0001
1. Overall, Tc becomes higher for smaller f. This can
CsC8 be understood by the fact that the screening effect
0.00001 due to the 2D electrons, which makes the polar-
0 0.2 0.4 0.6 0.8 1.0 coupled interaction weak, becomes smaller with
f
the decrease of f.
Figure 5 Calculated results for Tc as a function of the 2. The enhancement of Tc by about one order is
branching ratio f for alkali GIC superconductors in which brought about by doubling Z, if m is kept to be
Z ¼ 1 and m ¼ me. Reproduced with permission from
the same value.
Takada Y (2009) Unified model for superconductivity in
graphite intercalation compounds: Prediction of optimum Tc 3. The enhancement of Tc by about one order is also
and suggestion for its realization. Journal of the Physical brought about by tripling m from me to 3me, if Z is
Society of Japan 78(1): 013703:1–4. taken as Z ¼ 2.
422 Theory of Superconductivity in Graphite Intercalation Compounds
20 20
CaC6 Experiment on CaC6 Z=2
(CaC6: under pressure)
Z=2 CaC6: m* = 3.3me
15 15 o
CaC6: m* = 2.8me
o Experiment on YbC6
Tc (K)
m* = 4me o
Tc (K)
10
Experiment 10
m* = 3me
5 o YbC6: m* = 3.0me
m* = me 5
m* = 2me Experiment on SrC6
SrC6: m* = 1.5me
o
0 0.2 0.4 0.6 0.8 1.0
f 0 0.2 0.4 0.6 0.8 1.0
Figure 6 Calculated Tc as a function of f for m in the f
range of me–4me with other parameters suitably chosen for Figure 7 Calculated Tc as a function of f for alkaline-earth
CaC6. The experimental result is reproduced well, if we GICs with m determined so as to reproduce EF provided by
choose m 3me. Reproduced with permission from the bandstructure calculation. Reproduced with permission
Takada Y (2009) Unified model for superconductivity in from Takada Y (2009) Unified model for superconductivity in
graphite intercalation compounds: Prediction of optimum Tc graphite intercalation compounds: Prediction of optimum Tc
and suggestion for its realization. Journal of the Physical and suggestion for its realization. Journal of the Physical
Society of Japan 78(1): 013703:1–4. Society of Japan 78(1): 013703:1–4.
CaC6
50 superconductivity in metal GICs can be understood by
YbC6
40
LiC2(Z = 1)
the picture that the 3D electrons in the interlayer band
30 supplied by the ionization of metals experience the
20 SrC6
attractive interaction induced by the virtual exchange
of the polar-coupled phonons of the metal ions. We
10 BaC6
have also predicted a further enhancement of Tc well
0 10 20 30 40 0 10 20 30 40 50 beyond 10 K, giving some suggestions to realize such
m*/ me m*/ me superconductors in the family of GICs.
Figure 8 Prediction of Tc as a function of m for various
By first principles we usually mean the calcula-
GICs in pursuit of optimum Tc: (a) Z dependence for light- tions based on not the model but the first-principles
atom intercalant and (b) dependence on intercalant atom. Hamiltonian. Thus, it might be considered as inap-
We assume the fractional factor f ¼ 0. Reproduced with propriate to call the present G0W0 scheme first
permission from Takada Y (2009) Unified model for principles, but it is not an easy task to specify the
superconductivity in graphite intercalation compounds:
Prediction of optimum Tc and suggestion for its realization.
key parameters to control Tc by just implementing
Journal of the Physical Society of Japan 78(1): 013703:1–4. the calculations based on the first-principles
Hamiltonian. We can identify the importance of the
parameters, m and Z, only through the calculations
irrespective of any choice of other parameters, and based on the model Hamiltonian, leading to a better
Tc is optimized at m near 15me. The optimized Tc and unambiguous understanding of the mechanism of
depends rather strongly on the parameters to control superconductivity without probing too much into the
the polar-coupling strength such as Z and the atomic very details of each system which sometimes obscure
mass A; if we choose a trivalent light atom such as the essence in first-principles approaches. Besides,
boron to make !t large, the optimum Tc is about because of the errors involved in the numerical cal-
100 K, but the problem about the light atoms is that culations of normal-state properties as mentioned in
m will never become heavy due to the absence of Section 1.12.4, more accurate results of Tc will be
either d- or f-electrons. Therefore, we do not expect obtained by way of a suitable model Hamiltonian
that Tc would become much larger than 10 K, even if rather than directly from the first-principles one.
BeC2 or BC2 were synthesized. From this perspec- As a project in the future, it would be important
tive, it will be much better to intercalate Ti or V, to construct a more powerful scheme for the first-
rather than Be or B. Taking all these points into principles calculation of Tc by combining the schemes
account, we suggest synthesizing three-element in Sections 1.12.3 and 1.12.4, based on which we may
GICs providing a heavy 3D electron system by the make more detailed suggestions to synthesize GIC
introduction of heavy atoms into a light-atom polar- superconductors with Tc much larger than 10 K.
crystal environment. (See Chapters 1.10 and 1.11).
Allen PB and Dynes RC (1975) Transition temperature of strong- Floris A, Sanna A, Massidda S, and Gross EKU (2007) Two-
coupled superconductors reanalyzed. Physical Review B band superconductivity in Pb from ab initio calculations.
12: 905–922. Physical Review B 75: 054508:1–6.
Allen PB and Mitrović B (1982) Theory of superconducting Tc. Gauzzi A, Takashima S, Takeshita N, et al. (2007) Enhancement
In: Ehrenreich H, Seitz F, and Turnbull (eds.) Solid State of superconductivity and evidence of structural instability in
Physics, vol. 37, pp. 1–92. New York: Academic. intercalated graphite CaC6 under high pressure. Physical
Belash IT, Bronnikov AD, Zharikov OV, and Palnichenko AV Review Letters 98: 067002:1–4.
(1987) On the superconductivity of graphite intercalation Gunnarsson O (1997) Superconductivity in fullerides. Reviews
compounds with sodium. Solid State Communications of Modern Physics 69: 575–606.
64: 1445–1447. Hannay NB, Geballe TH, Matthias BT, Andres K, Schmidt P, and
Belash IT, Bronnikov AD, Zharikov OV, and Pal’nichenko AV MacNair D (1995) Superconductivity in graphitic
(1989) Superconductivity of graphite intercalation compounds. Physical Review Letters 14: 225–226.
compound with lithium C2Li. Solid State Communications Hebard AF, Rosseinsky MJ, Haddon RC, et al. (1991)
69: 921–923. Superconductivity at 18 K in potassium-doped C60. Nature
Belash IT, Bronnikov AD, Zharikov OV, and Pal’nichenko AV 350: 600–601.
(1990) Effect of the metal concentration on the Hinks DG, Rosenmann D, Claus H, Bailey MS, and
superconducting properties of lithium-, sodium- and Jorgensen JD (2007) Large Ca isotope effect in the CaC6
potassium-containing graphite intercalation compounds. superconductor. Physical Review B 75: 014509:1–6.
Synthetic Metals 36: 283–302. Hohenberg P and Kohn W (1964) Inhomogeneous electron gas.
Belash IT, Zharikov OV, and Pal’nichenko AV (1989) Physical Review 136: B864–B871.
Superconductivity of GIC with Li, Na and K. Synthetic Metals Holzwarth NAW, Louie SG, and Rabii S (1984) Interlayer states
34: 455–460. in graphite and in alkali-metal-graphite intercalation
Bohnen K-P, Heid R, and Renker B (2001) Phonon dispersion compounds. Physical Review B 30: 2219–2222.
and electron-phonon coupling in MgB2 and AlB2. Physical Inoshita T and Kamimura H (1981) Theory of phonon-limited
Review Letters 86: 5771:1–4. resistivity and electron-phonon interaction in higher-stage
Calandra M and Mauri F (2005) Theoretical explanation of graphite intercalation compounds. Synthetic Metals
superconductivity in C6Ca. Physical Review Letters 3: 223–232.
95: 237002:1–4. Iye Y and Tanuma S (1982) Superconductivity of graphite
Carbotte JP (1990) Properties of boson-exchange intercalation compounds with alkali-metal amalgams.
superconductors. Reviews of Modern Physics Physical Review B 25: 4583–4592.
62: 1027–1157. Jishi RA and Dresselhaus MS (1992) Superconductivity in
Chaiken A, Dresselhaus MS, Orlando TP, et al. (1990) graphite intercalation compounds. Physical Review B
Anisotropic superconductivity in C4KHg. Physical Review B 45: 12465–12469.
41: 71–81. Jishi RA, Dresselhaus MS, and Chaiken A (1991) Theory of the
Choi HJ, Roundy D, Sun H, Cohen ML, and Louie SG (2002a) upper critical field in graphite intercalation compounds.
The origin of the anomalous superconducting properties of Physical Review B 44: 10248–10255.
MgB2. Nature 418: 758–760. Kadowaki K, Nabemoto T, and Yamamoto T (2007)
Choi HJ, Roundy D, Sun H, Cohen ML, and Louie SG (2002b) Synthesis and superconducting properties of graphite
First-principles calculation of the superconducting transition compounds intercalated with Ca:C6Ca. Physica C
in MgB2 within the anisotropic Eliashberg formalism. 460–462: 152–153.
Physical Review B 66: 020513(R):1–4. Kamimura H (1987) Graphite intercalation compounds. Physics
Cohen ML (1964) Superconductivity in many-valley Today 40(12): 64–71.
semiconductors and in semimetals. Physical Review Kamimura H, Nakao K, Ohno T, and Inoshita T (1980) Electronic
134: A511–A521. structure and properties of alkali-graphite intercalation
Cohen ML (1969) Superconductivity in low-carrier-density compounds. Physica B 99: 401–405.
systems: Degenerate semiconductors. In: Parks RD (ed.) Kim JS, Boeri L, O’Brien JR, Razavi FS, and Kremer RK (2007)
Superconductivity, vol. 1, chap. 12, pp. 615–664. New York: Superconductivity in heavy alkaline-earth intercalated
Marcel Dekker. graphites. Physical Review Letters 99: 027001:1–4.
Csanyi G, Littlewood PB, Nevidomskyy AH, Pickard CJ, and Kim JS, Kremer RK, Boeri L, and Razavi FS (2006) Specific heat
Simons BD (2005) The role of the interlayer state in the of the Ca-intercalated graphite superconductor CaC6.
electronic structure of superconducting graphite Physical Review Letters 96: 217002:1–4.
intercalated compounds. Nature Physics 1: 42–45. Kirzhnits DA, Maksimov EG, and Khomskii DI (1973) The
de Gennes PG (1966) Superconductivity of Metals and Alloys. description of superconductivity in terms of dielectric
New York: Benjamin. response function. Journal of Low Temperature Physics
Dresselhaus MS, Chaiken A, and Dresselhaus G (1989) 10: 79–93.
Anisotropic superconductivity in C4KHg-GICs. Synthetic Kobayashi M and Tsujikawa I (1979) Susceptibility fluctuation in
Metals 34: 449–454. superconducting first stage potassium graphite intercalation
Dresselhaus MS and Dresselhaus G (2002) Intercalation compound near transition temperature. Journal of the
compounds of graphite. Advances in Physics 51: 1–186. Physical Society of Japan 46: 1945–1946.
Eliashberg GM (1960) Interactions between electrons and Kobayashi M and Tsujikawa I (1981) Superconducting transition
lattice vibrations in a superconductor. Soviet Physics – JETP in the first stage potassium graphite intercalation
11: 696–702. compounds. Physica B 105: 439–443.
Emery N, Heérold C, d’Astuto M, et al. (2005) Superconductivity Kohn W and Sham LJ (1965) Self-consistent equations
of bulk CaC6. Physical Review Letters 95: 087003:1–4. including exchange and correlation effects. Physical Review
Fischer JE and Thompson TE (1978) Graphite intercalation 140: A1133–A1138.
compounds. Physics Today 31(7): 36–45. Koike Y, Suematsu H, Higuchi K, and Tanuma S (1978)
Floris A, Profeta G, and Lathiotakis NN (2005) Superconducting Superconductivity in graphite-potassium intercalation
properties of MgB2 from first principles. Physical Review compounds C8K. Solid State Communications
Letters 94: 037004:1–4. 27: 623–627.
Theory of Superconductivity in Graphite Intercalation Compounds 425
Koike Y, Suematsu H, Higuchi K, and Tanuma S (1980) Roth G, Chaiken A, Enoki T, Yeh NC, Dresselhaus G, and
Superconductivity in graphite-alkali metal intercalation Tedrow PM (1985) Enhanced superconductivity in
compounds. Physica B 99: 503–508. hydrogenated potassium-mercury–graphite intercalation
Koike Y and Tanuma S (1981) Anisotropic superconductivity in compounds. Physical Review B 32: 533–536.
the graphite potassium-amalgam intercalation compound Sanna A, Franchini C, Floris A, et al. (2006) Ab initio prediction of
C8KHg. Journal of the Physical Society of Japan pressure-induced superconductivity in potassium. Physical
50: 1964–1969. Review B 73: 144512:1–7.
Koma A, Miki K, Suematsu H, Ohno T, and Kamimura H Sanna A, Profeta G, Floris A, Marini A, Gross EKU, and
(1986) Density-of-states investigation of C8K and Massidda S (2007) Anisotropic gap of superconducting
occurrence of the interlayer band. Physical Review B CaC6: A first-principles density functional calculation.
34: 2434–2438. Physical Review B 75: 020511(R):1–4.
Kong Y, Dolgov OV, Jepsen O, and Andersen OK (2001) Sham LJ and Schlüter M (1983) Density-functional theory of the
Electron-phonon interaction in the normal and energy gap. Physical Review Letters 51: 1888–1891.
superconducting states of MgB2. Physical Review B Sham LJ and Schlüter M (1985) Density-functional theory of the
64: 020501(R):1–4. energy gap. Physical Review B 32: 3883–3889.
Kurter C, Ozyuzer L, Mazur D, et al. (2007) Large energy gaps in Sugawara K, Sato T, and Takahashi T (2009) Fermi-surface-
CaC6 from tunneling spectroscopy: Possible evidence of dependent superconducting gap in C6Ca. Nature Physics
strong-coupling superconductivity. Physical Review B 5: 40–43.
76: 220502(R):1–4. Takabayashi Y, Ganin AY, Jeglič P, et al. (2009) The disorder-
Kurth S, Marques M, Lüders M, and Gross EKU (1999) Local free non-BCS superconductor Cs3C60 emerges from an
density approximation for superconductors. Physical Review antiferromagnetic insulator parent state. Science
Letters 83: 2628–2631. 323: 1585–1590.
Lamura G, Aurino M, Cifariello G, et al. (2006) Experimental Takada Y (1978) Plasmon mechanism of superconductivity in
evidence of s-wave superconductivity in bulk CaC6. Physical two- and three-dimensional electron systems. Journal of the
Review Letters 96: 107008:1–4. Physical Society of Japan 45: 786–794.
Lüders M, Marques MAL, Lathiotakis NN, et al. (2005) Ab initio Takada Y (1980) Theory of superconductivity in polar
theory of superconductivity. I. Density functional formalism semiconductors and its application to n-type
and approximate functional. Physical Review B semiconducting SrTiO3. Journal of the Physical Society of
72: 024545:1–17. Japan 49: 1267–1275.
Marques MAL, Lüders M, Lathiotakis NN, et al. (2005) Ab initio Takada Y (1982) Mechanism of superconductivity in graphite-
theory of superconductivity. II. Application to elemental alkali metal intercalation compounds. Journal of the Physical
metals. Physical Review B 72: 024546:1–13. Society of Japan 51: 63–72.
Mazin, II (2005) Intercalant-driven superconductivity in YbC6 Takada Y (1993) s- and p-wave pairings in the dilute electron
and CaC6. Physical Review Letters 95: 1–4 227001. gas: Superconductivity mediated by the Coulomb hole in the
McMillan WL (1968) Transition temperature of strong-coupled vicinity of the Wigner-crystal phase. Physical Review B
superconductors. Physical Review 167: 331–344. 47: 5202–5211.
Mizutani U, Kondow T, and Massalski TB (1978) Low- Takada Y (1993) Consideration of the mechanism of
temperature specific heats of graphite intercalation superconductivity in fullerenes: Beyond Migdal’s theorem
compounds with potassium and cesium. Physical Review B and the dynamic Coulomb effect. Journal of Physics and
17: 3165–3173. Chemistry of Solids 54: 1779–1788.
Morel P and Anderson PW (1962) Calculation of the Takada Y (2009) Unified model for superconductivity in graphite
superconducting state parameters with retarded electron- intercalation compounds: Prediction of optimum Tc and
phonon interaction. Physical Review 125: 1263–1271. suggestion for its realization. Journal of the Physical Society
Ohno T, Nakao K, and Kamimura H (1979) Self-consistent of Japan 78(1): 013703:1–4.
calculation of the band structure of C8K including the charge Takada Y (2009) Mechanism of superconductivity in graphite
transfer effect. Journal of the Physical Society of Japan intercalation compounds including CaC6. Journal of
47: 1125–1133. Superconductivity and Novel Magnetism 22: 89–92.
Oliveira LN, Gross EKU, and Kohn W (1988) Density-functional Takada Y and Hotta T (1998) Superconductivity in the alkali-
theory for superconductors. Physical Review Letters doped fullerides: Competition of phonon-mediated
60: 2430–2433. attractions with Coulomb repulsions in polaron pairing.
Pendrys LA, Wachnik R, Vogel FL, et al. (1981) International Journal of Modern Physics B 12: 3042–3051.
Superconductivity of the graphite intercalation compounds Tanuma S (1981) Dimensionality of electronic structure and
KH8C8 and RbHg8C8. Solid State Communications superconductivity of graphite intercalation compounds.
38: 677–681. Physica B 105: 486–490.
Perdew JP, Burke K, and Ernzerhof M (1996) Generalized Valla T, Camacho J, Pan Z-H, et al. (2009) Anisotropic electron-
gradient approximation made simple. Physical Review phonon coupling and dynamical nesting on the graphene
Letters 77: 3865–3868, 78:1396(E). sheets in superconducting CaC6 using angle-resolved
Perdew JP and Levy M (1983) Physical content of the exact photoemission spectroscopy. Physical Review Letters
Kohn-Sham orbital energies: Band gaps and derivative 102: 107007:1–4.
discontinuities. Physical Review Letters 51: 1884–1887. Wang G, Datars WR, and Ummat PK (1991) Fermi surface of the
Posternak M, Baldereschi A, Freeman AJ, and Wimmer E (1984) stage-1 potassium graphite intercalation compound.
Prediction of electronic surface states in layered materials: Physical Review B 44: 8294–8300.
Graphite. Physical Review Letters 52: 863–866. Weller TE, Ellerby M, Saxena AS, Smith RP, and Skipper NT
Profeta G, Franchini C, Lathiotakis NN, et al. (2006) (2005) Superconductivity in the intercalated graphite
Superconductivity in lithium, potassium, and aluminum compounds C6Yb and C6Ca. Nature Physics 1: 39–41.
under extreme pressure: A first-principles study. Physical Zabel H and Solin SA (eds.) (1992) Graphite Intercalation
Review Letters 96: 047003:1–4. Compounds II. New York: Springer.
426 Theory of Superconductivity in Graphite Intercalation Compounds