Defects in

Materials
and Devices
Microelectronic
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
CRC Press is an imprint of the
Taylor & Francis Group, an informa business
Boca Raton London New York
Defects in
Materials
and Devices
Microelectronic
Edited by
Daniel M. Fleetwood • Sokrates T. Pantelides
Ronald D. Schrimpf
ß 2008 by Taylor & Francis Group, LLC.
Figure on cover: Z-contrast image of a silicon-silicon dioxide-hafnium dioxide structure showing an isolated Hf atom in
the SiO
2
interlayer (image courtesy of K. Van Benthem and S. J. Pennycook). The expanded image is an electron density
plot for this structure (courtesy of A. G. Marinopoulos and S. T. Pantelides).
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2009 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed in the United States of America on acid-free paper
10 9 8 7 6 5 4 3 2 1
International Standard Book Number-13: 978-1-4200-4376-1 (Hardcover)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For orga-
nizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Fleetwood, Daniel.
Defects in microelectronic materials and devices / Daniel Fleetwood, Sokrates T. Pantelides, and
Ronald D. Schrimpf.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-4200-4376-1 (alk. paper)
1. Microelectronics--Materials--Testing. 2. Metal oxide semiconductor field-effect
transistors--Testing. 3. Integrated circuits--Defects. I. Pantelides, Sokrates T. II. Schrimpf, Ronald
Donald. III. Title.
TK7871.F5485 2008
621.381--dc22 2008018722
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
ß 2008 by Taylor & Francis Group, LLC.
Contents
Prefac e
Editors
Contr ibutors
1. D efects in U lt ra-Shallow Junctions
Mark E. Law, Renata Camillo-Castillo, Lance Robertson, and Kevin S. Jones
2. H ydrogen-Relat ed Defect s in S ilicon, Germanium,
and Si licon– Germanium Alloys
A.R. Peaker, V.P. Markevich, and L. Dobaczewski
3. Defects in S traine d-Si MOSFETs
Yongke Sun and Scott E. Thompson
4. The Ef fect o f D efec ts on Ele ctron Transport in Nanome ter-Sca le
Electronic D evices: Im purities and Interf ace R oughness
M.V. Fischetti and S. Jin
5. Electrical Char acterizat ion of Defects in Gate Di electrics
Dieter K. Schroder
6. Dominating Defects in the MO S System: P
b
and E
0
Centers
Patrick M. Lenahan
7. Oxide Traps, Border Traps, and Interf ace T raps in SiO
2
Daniel M. Fleetwood, Sokrates T. Pantelides, and Ronald D. Schrimpf
8. From 3D Imaging of Atoms to Macroscopic D evi ce Pr operties
S.J. Pennycook, M.F. Chisholm, K. van Benthem, A.G. Marinopoulos,
and Sokrates T. Pantelides
9. D efec t Energy Leve ls in Hf O
2
and R ela ted Hi gh-K Gate Oxides
J. Robertson, K. Xiong, and K. Tse
10 . S pe ctro sco pic S tu di es of E lec trically Acti ve Defects i n High- K
Gate Dielectrics
Gerald Lucovsky
11. Def ects in CMOS G ate Dielectr ics
Eric Garfunkel, Jacob Gavartin, and Gennadi Bersuker
ß 2008 by Taylor & Francis Group, LLC.
12. Negati ve B ias T emperatur e Instabilities in Hi gh-k Gate Dielectrics
M. Houssa, M. Aoulaiche, S. De Gendt, G. Groeseneken, and M.M. Heyns
13. Defect F ormation and Anni hilation in E lectronic Devices
and t he Role of Hydrogen
Leonidas Tsetseris, Daniel M. Fleetwood, Ronald D. Schrimpf, and Sokrates T. Pantelides
14 . T owa rd E ngineeri ng M od eling o f Negative Bias
Temperatur e Instability
Tibor Grasser, Wolfgang Goes, and Ben Kaczer
15 . We ar-Out and Time -Depende nt Die lect ric Breakdown
in Sil icon Oxides
John S. Suehle
16 . De fects As sociate d with Diel ectri c Bre akdown
in SiO
2
-Based Gate D ielec trics
Jordi Suñé and Ernest Y. Wu
17. Defects i n Thin and Ultrathi n S ilicon Dioxides
Giorgio Cellere, Simone Gerardin, and Alessandro Paccagnella
18. S tructural Def ects in Si O
2
– Si Caused by Ion B omba rdment
Antoine D. Touboul, Aminata Carvalho, Mathias Marinoni, Frederic Saigne,
Jacques Bonnet, and Jean Gasiot
19 . Impac t of R adiation- Induced Defec ts on Bi polar Device Oper ation
Ronald D. Schrimpf, Daniel M. Fleetwood, Ronald L. Pease, Leonidas Tsetseris,
and Sokrates T. Pantelides
20. S ilicon Di oxide– Sil icon C arbide Interfaces: Current St atus
and R e cent A dvances
S. Dhar, Sokrates T. Pantelides, J.R. Williams, and L.C. Feldman
21 . De fects i n SiC
E. Janzén, A. Gali, A. Henry, I.G. Ivanov, B. Magnusson, and N.T. Son
22. Defects i n Gal lium Arsenide
J.C. Bourgoin and H.J. von Bardeleben
Appendix A: Selected High-Im pac t Journal Articl es on Defe cts
in Micr oel ectronic Materials and Devices
ß 2008 by Taylor & Francis Group, LLC.
Preface
Defects in microelectronic materials can profoundly affect the yield, performance, long-
term reliability, and radiation response of microelectronic devices and integrated circuits
(ICs). This book provides a comprehensive survey of defects in silicon-based metal oxide
semiconductor (MOS) field-effect transistor technologies, which dominate the worldwide
microelectronics marketplace. This book also discusses the defects in linear bipolar tech-
nologies, silicon carbide based devices, and gallium arsenide materials and devices. An
appendix is provided with supplemental material on highly cited papers on defects in
these and other materials systems (e.g., GaN, ZnO, C) that are being investigated
for present and future microelectronics technologies. The chapters described below
summarize decades of experience in characterizing defect properties and their impact on
microelectronic devices, and also look forward to the challenges that will have to be
overcome as new materials (e.g., high-K gate dielectrics and high-mobility substrate
materials) are incorporated into increasingly more highly scaled devices and ICs.
Chapters 1 through 4 focus on yield- and performance-limiting defects and impurities in
the device silicon layer, and=or at the critical Si=SiO
2
interface. In Chapter 1, Law et al.
discuss yield-limiting defects that are important to control in highly scaled technologies
with ultra-shallow junctions. The impact of defects on diffusion and activation processes is
emphasized. Defects in Si, Ge, and SiGe alloys associated with hydrogen are discussed by
Peaker et al. in Chapter 2. The incorporation of hydrogen and its complexes in these
materials are reviewed in detail, with examples provided from a broad range of experi-
mental and theoretical work. In Chapter 3, Sun and Thompson describe dislocations in
transistors that have been designed with highly strained layers to enhance carrier mobi-
lities. In Chapter 4, Fischetti and Jin discuss the effects of ionized dopant atoms and
interface roughness on electron transport in very highly scaled devices and ICs. An
extensive review of mobility-limiting scattering processes is presented.
Chapters 5 through 8 describe electrical, analytical, spectroscopic, and state-of-the-art
microscopic methods to characterize defects in MOS gate oxides. In Chapter 5, Schroder
provides a comprehensive review of electrical and analytical techniques to estimate dens-
ities and energy distributions of defects in MOS gate dielectrics and=or at the Si=SiO
2
interface. A wide variety of examples are provided. These enable a comparison of the
relative advantages and limitations of different characterization methods. In Chapter 6,
Lenahan provides an extensive review of electron spin resonance studies that have pro-
vided significant insight into the microstructure of the dominant SiO
2
hole trap. This is the
E
0
defect, a trivalent Si center in SiO
2
associated with an O vacancy. The dominant defect at
the Si=SiO
2
interface is also characterized extensively; this is the P
b
defect, a Si dangling
bond that frequently is passivated by hydrogen during device processing, but when
depassivated after electrical stress or ionizing radiation exposure can function as an
interface trap. In Chapter 7, the properties of oxide and interface traps are reviewed by
Fleetwood et al., with particular emphasis on separating the effects of ‘‘true’’ oxide and
interface traps from the effects of near-interfacial oxide traps that exchange charge with the
Si on the timescale of typical device measurements (border traps). Thermally stimulated
current and low frequency noise techniques are described to estimate oxide and border
trap densities and energy distributions. In Chapter 8, Pennycook et al. present the results of
ß 2008 by Taylor & Francis Group, LLC.
aberration-corrected scanning transmission electron microscopy on MOS structures; these
techniques provide information on defects and impurities with sub-angstrom resolution.
Chapters 9 through 11 focus on defects in high-dielectric constant (high-K) materials that
are under intense development to replace SiO
2
(more precisely, nitrided SiO
2
) as the
preferred gate dielectric material for Si-based ICs at or beyond the 45 nm technology
node. In Chapter 9, Robertson et al. describe theoretical calculations of defect energy levels
in high-K dielectrics, where it is found that the O vacancy is the main electrically active
defect in HfO
2
-based high-K gate dielectrics. In Chapter 10, Lucovsky surveys an extensive
amount of spectroscopic data on transition metal oxides. These materials are compared
and contrasted with SiO
2
, and a critical review of the suitability of these materials is
provided for incorporation into device manufacturing. In Chapter 11, Garfunkel et al.
review experimental results and computational calculations of defects in HfO
2
-based
gate stacks. The effects of these defects on device electrical properties are emphasized.
In Chapter 12, Houssa et al. survey negative bias temperature instabilities (NBTIs) for
high-K materials, and they separate the effects of interface and bulk traps on device
electrical response. In Chapter 13, Tsetseris et al. discuss the role of hydrogen in NBTI in
SiO
2
, and summarize first-principles calculations of defect formation, defect dynamics, and
defect annihilation in the Si=SiO
2
system. A comprehensive engineering model of NBTI is
developed by Grasser et al. in Chapter 14. The effects of poststress relaxation are shown to
be important to a complete understanding of the underlying mechanisms, as well as in
developing techniques to predict device lifetimes from accelerated measurements.
Chapters 15 through 18 discuss defects in ultrathin oxides (SiO
2
and silicon oxynitride).
In Chapter 15, Suehle describes the role of defects and hydrogen in SiO
2
on the time-
dependent dielectric breakdown of MOS devices and ICs. The present understanding of
defect generation processes is discussed, as are statistical models for the resulting device
failure distributions. Defects associated with dielectric breakdown and conduction in
oxides degraded by long-term, high-field stress are discussed in detail by Suñé and Wu
in Chapter 16. The percolation model of breakdown is described, and the relative roles of
holes and hydrogen in the breakdown process are evaluated experimentally and analytic-
ally. The effects of radiation and stress-induced defects in thin oxides are described by
Cellere et al. in Chapter 17 and Touboul et al. in Chapter 18. Cellere et al. emphasize
processes that lead to radiation and stress-induced leakage currents, which are especially
important for nonvolatile memory cells that can be discharged via extremely low currents,
and Touboul et al. discuss the physical damage caused by a high energy ion when it passes
through a dielectric layer. The latter is of particular concern for the reliability of space
systems.
The effects of radiation-induced defects on linear bipolar devices are discussed by
Schrimpf et al. in Chapter 19. Even though similar defects are formed in the base oxides
of linear bipolar transistors as in MOS structures (primarily oxide and interface-trap
charge), the effects of these charges on device operation can be quite different for linear
bipolar transistors than for MOS devices. One example of this is enhanced low-dose-rate
sensitivity, which is the excess buildup of (primarily) interface traps during low-electric-
field irradiations of typical base oxides that can be affected by trap densities and hydrogen
concentrations of the structures.
In Chapter 20, Dhar et al. present an overview of defects in oxides on SiC wafers. The
higher growth temperatures and the differing interfacial layers for SiC, relative to Si, lead
to significant differences in defect properties. Most notably, dangling bond defects cannot
be passivated as easily at the SiC=SiO
2
interface with hydrogen treatment as for the Si=SiO
2
interface. A comprehensive survey of defects in SiC is provided by Janzén et al. in Chapter
21. An extensive array of experimental data and theoretical calculations are tabulated and
discussed in detail. Defects in GaAs are reviewed by Bourgoin and von Bardeleben in
ß 2008 by Taylor & Francis Group, LLC.
Chapter 22. Particular emphasis is placed on As antisite defects associated with the EL2.
Finally, the appendix lists a large number of highly cited journal articles related to defects
in the Si=SiO
2
system, high-K dielectrics, GaAs, GaN, and ZnO. Brief synopses are pro-
vided of more than 450 highly cited articles; the interested reader can use these reference
lists as a starting point to obtain more information on the nature and effects of defects and
impurities in these or other semiconductor-based material systems.
This book was encouraged by the program managers of two multidisciplinary university
research initiatives (MURIs) sponsored by the Air Force Office of Scientific Research,
Gerald Witt and Kitt Reinhardt, on the effects of ionizing radiation on microelectronic
materials and devices. We appreciate their support during these programs, as well as the
efforts of all of our MURI collaborators—you will see many of their contributions in this
book, as well as contributions from many other experts in the field who graciously agreed
to provide chapters in their fields of specialty. We also wish to thank all our valued
professional colleagues, research collaborators, and sponsors who have contributed so
much to these efforts, as well as Jill Jurgensen, Allison Shatkin, and Shelley Kronzek at
Taylor & Francis for their interest in and assistance with this book. We finally wish to thank
Arun Kumar, Jennifer Smith, and the entire production team for their strong and timely
support.
Daniel M. Fleetwood
Sokrates T. Pantelides
Ronald D. Schrimpf
Nashville, Tennessee
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
Editors
Daniel M. Fleetwood received BS, MS, and PhD degrees from Purdue University in 1980,
1981, and 1984, respectively. He joined Sandia National Laboratories as a member of the
technical staff in 1984. In 1990, he was named a distinguished member of the technical staff
in the Radiation Technology and Assurance Department at Sandia. Dan accepted a pos-
ition as professor of electrical engineering at Vanderbilt University in 1999. Between 2001
and 2003, he served as an associate dean for research in the School of Engineering. In 2003,
he was named the chairman of the Electrical Engineering and Computer Science Depart-
ment at Vanderbilt University. He has authored or coauthored more than 300 papers on
radiation effects and low frequency noise. He served as a guest editor of the December 1990
and April 1996 issues of the IEEE Transactions on Nuclear Science, and presently is the
executive vice-chairman of the IEEE Nuclear and Plasma Sciences Society (NPSS) Radi-
ation Effects Steering Group. Dan has received seven outstanding paper awards for the
IEEE NSREC and three for the HEART Conference, as well as several meritorious confer-
ence paper awards. Dan is a fellow of the IEEE and the American Physical Society.
Sokrates T. Pantelides is the William A. and Nancy F. McMinn professor of physics at
Vanderbilt University, Nashville, Tennessee. He holds a secondary appointment as a
distinguished visiting scientist at Oak Ridge National Laboratory, Oak Ridge, Tennessee.
He received his PhD in physics from the University of Illinois at Urbana-Champaign in
1973. After a postdoctoral appointment at Stanford University, he joined the IBM
T. J. Watson Research Center in Yorktown Heights, New York, in 1975 where he carried
out theoretical research in semiconductors and served as manager, senior manager, and
program director. He joined Vanderbilt University in 1994. He has authored or coauthored
more than 300 research articles and edited eight books. He is a fellow of the American
Physical Society and the American Association for the Advancement of Science. His
research focuses on the structure, defect dynamics, and electronic properties of electronic
materials, radiation effects, transport in molecules and thin films, and catalysis.
Ronald D. Schrimpf received BEE, MSEE, and PhD degrees from the University of
Minnesota in 1981, 1984, and 1986, respectively. He joined the University of Arizona in
1986, where he served as an assistant professor (1986), an associate professor (1991), and a
professor (1996) of electrical and computer engineering. He joined Vanderbilt University in
1996 and was an invited professor at the University of Montpellier II, Montpellier, France,
in 2000. Ron is currently a professor of electrical engineering at Vanderbilt University,
where his research activities focus on microelectronics and semiconductor devices. In
particular, he has a very active research program dealing with the effects of radiation on
semiconductor devices and integrated circuits. Ron is also the director of the Institute for
Space and Defense Electronics (ISDE) at Vanderbilt University. The engineering staff of
ISDE performs design, analysis, and modeling work for a variety of space and defense-
oriented organizations, as well as commercial semiconductor companies. Ron is a fellow of
the IEEE and past chairman of the IEEE NPSS Radiation Effects Steering Group.
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
Contributors
M. Aoulaiche
IMEC
Leuven, Belgium
and
Department of Electrical Engineering
University of Leuven
Leuven, Belgium
H.J. von Bardeleben
Institut de NanoSciences de Paris
Université Paris 6
Paris, France
K. van Benthem
Materials Science and Technology Division
Oak Ridge National Laboratory
Oak Ridge, Tennessee
and
Center for Nanophase Materials Sciences
Oak Ridge National Laboratory
Oak Ridge, Tennessee
Gennadi Bersuker
Sematech International
Austin, Texas
Jacques Bonnet
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
J.C. Bourgoin
Genie des Semiconducteurs (GESEC)
Research, Inc.
Paris, France
Renata Camillo-Castillo
IBM
Burlington, Vermont
Aminata Carvalho
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
Giorgio Cellere
Department of Information
Engineering
University of Padova
Padova, Italy
M.F. Chisholm
Materials Science and Technology Division
Oak Ridge National Laboratory
Oak Ridge, Tennessee
S. Dhar
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
and
Cree Incorporated
Durham, North Columbia
L. Dobaczewski
Institute of Physics
Polish Academy of Sciences
Warsaw, Poland
L.C. Feldman
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
and
Institute for Advanced Materials, Devices
and Nanotechnology
Rutgers University
New Brunswick, New Jersey
ß 2008 by Taylor & Francis Group, LLC.
M.V. Fischetti
Department of Electrical and Computer
Engineering
University of Massachusetts
Amherst, Massachusetts
Daniel M. Fleetwood
Department of Electrical Engineering
and Computer Science
Vanderbilt University
Nashville, Tennessee
A. Gali
Department of Atomic Physics
Budapest University of Technology
and Economics
Budapest, Hungary
Eric Garfunkel
Departments of Chemistry
Rutgers University
Piscataway, New Jersey
Jacob Gavartin
Accelrys
Cambridge, United Kingdom
S. De Gendt
IMEC
Leuven, Belgium
and
Department of Chemistry
University of Leuven
Belgium
Jean Gasiot
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
Simone Gerardin
Department of Information Engineering
University of Padova
Padova, Italy
Wolfgang Goes
Christian Doppler Laboratory for TCAD
in Microelectronics
Institute for Microelectronics
Wien, Austria
Tibor Grasser
Christian Doppler Laboratory for TCAD
in Microelectronics
Institute for Microelectronics
Wien, Austria
G. Groeseneken
IMEC
Leuven, Belgium
and
Department of Electrical Engineering
University of Leuven
Leuven, Belgium
A. Henry
Department of Physics, Chemistry
and Biology
Linköping University
Linköping, Sweden
M.M. Heyns
IMEC
Leuven, Belgium
and
Department of Electrical Engineering
University of Leuven
Leuven, Belgium
M. Houssa
IMEC
Leuven, Belgium
and
Department of Electrical Engineering
and Physics
University of Leuven
Leuven, Belgium
I.G. Ivanov
Department of Physics, Chemistry
and Biology
Linköping University
Linköping, Sweden
ß 2008 by Taylor & Francis Group, LLC.
E. Janzén
Department of Physics, Chemistry
and Biology
Linköping University
Linköping, Sweden
S. Jin
Department of Electrical and Computer
Engineering
University of Massachusetts
Amherst, Massachusetts
Kevin S. Jones
Department of Materials Science
and Engineering
University of Florida
Gainesville, Florida
Ben Kaczer
IMEC
Leuven, Belgium
Mark E. Law
Department of Electrical
and Computer Engineering
University of Florida
Gainesville, Florida
Patrick M. Lenahan
Department of Engineering Science
and Mechanics
The Pennsylvania State University
University Park, Pennsylvania
Gerald Lucovsky
Department of Physics
North Carolina State University
Raleigh, North Carolina
B. Magnusson
Department of Physics, Chemistry and
Biology
Linköping University
Linköping, Sweden
and
Norstel AB
Norrköping, Sweden
Mathias Marinoni
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
A.G. Marinopoulos
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
V.P. Markevich
School of Electrical and Electronic
Engineering
University of Manchester
Manchester, United Kingdom
Alessandro Paccagnella
Department of Information Engineering
University of Padova
Padova, Italy
Sokrates T. Pantelides
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
A.R. Peaker
School of Electrical and Electronic
Engineering
University of Manchester
Manchester, United Kingdom
Ronald L. Pease
RLP Research
Los Lunas, New Mexico
S.J. Pennycook
Materials Science and Technology Division
Oak Ridge National Laboratory
Oak Ridge, Tennessee
and
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
J. Robertson
Department of Engineering
Cambridge University
Cambridge, United Kingdom
ß 2008 by Taylor & Francis Group, LLC.
Lance Robertson
DRS Technologies
Dallas, Texas
Frederic Saigne
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
Ronald D. Schrimpf
Department of Electrical Engineering
and Computer Science
Vanderbilt University
Nashville, Tennessee
Dieter K. Schroder
Department of Electrical Engineering
Center for Solid State Electronics
Research
Arizona State University
Tempe, Arizona
N.T. Son
Department of Physics, Chemistry
and Biology
Linköping University
Linköping, Sweden
John S. Suehle
Semiconductor Electronics Division
National Institute of Standards
and Technology
Gaithersburg, Maryland
Yongke Sun
Department of Electrical and Computer
Engineering
University of Florida
Gainesville, Florida
Jordi Suñé
Departament d’Enginyeria Electrònica
Universitat Autònoma de Barcelona
Barcelona, Spain
Scott E. Thompson
Department of Electrical and Computer
Engineering
University of Florida
Gainesville, Florida
Antoine D. Touboul
Institut d’Electronique du Sud
UMR-CNRS 5214
University of Montpellier II
Montpellier, France
K. Tse
Department of Engineering
Cambridge University
Cambridge, United Kingdom
Leonidas Tsetseris
Department of Physics and Astronomy
Vanderbilt University
Nashville, Tennessee
and
Department of Physics
Aristotle University of Thessalonki
Thessalonki, Greece
J.R. Williams
Department of Physics
Auburn University
Auburn, Alabama
Ernest Y. Wu
IBM Microelectronics Division
Burlington, Vermont
K. Xiong
Department of Engineering
Cambridge University
Cambridge, United Kingdom
and
University of Texas at Dallas
Dallas, Texas
ß 2008 by Taylor & Francis Group, LLC.
1
Defects in Ultra-Shallow Junctions
Mark E. Law, Renata Camillo-Castillo, Lance Robertson, and Kevin S. Jones
CONTENTS
1.1 Introduction............................................................................................................................ 1
1.2 Shallow Junction Processing Primer ................................................................................... 2
1.3 Implant Process: Nuclear and Electronic Stopping .......................................................... 2
1.4 Annealing................................................................................................................................ 3
1.5 Defect Evolutionary Processes ............................................................................................. 5
1.5.1 Submicroscopic Defects.............................................................................................. 6
1.5.2 Dot Defects................................................................................................................... 8
1.5.3 311 Defects ................................................................................................................... 8
1.5.4 Loops .......................................................................................................................... 12
1.6 Defects in Ultra-Shallow Processes ................................................................................... 13
1.6.1 Surface Effects............................................................................................................ 13
1.6.2 Millisecond Annealing ............................................................................................. 14
1.6.3 Regrowth-Related Defects ....................................................................................... 20
1.7 Conclusions........................................................................................................................... 22
References...................................................................................................................................... 23
1.1 Introduction
The widely quoted International Technology Roadmap for Semiconductors (ITRS) [1]
enumerates critical device dimensions (physical gate lengths, oxide thickness, junction
depths, etc.) needed to meet performance goals of the future. The ITRS bases its predictions
on a scaled planar silicon technology, and is an invaluable aid for defining process tool
capabilities and providing a sense of scale for modeling efforts needed in the near future.
Many of the techniques and structures for these devices are not yet known and much of
the roadmap has been coded ‘‘research required.’’ The device at the far end of the road
might be a traditional bulk device [2], a dual gate silicon-on-insulator (SOI) device [3],
a ‘‘FinFET’’ [4], or even a vertical device [5].
The shallow junction specification of the roadmap is quite daunting. High conductivity,
extremely shallow junctions have to be formed. It is likely that future devices will have
metastable dopant concentrations in the source and drain to attempt to control parasitic
elements. Lattice defects have an enormous impact on the final electrical characteristics of
the junction. As we will see, damage from implantation controls the junction depth and
ß 2008 by Taylor & Francis Group, LLC.
final activation. Understanding the evolutionary path of the defects becomes critical in
understanding the electrical properties of doped layers in silicon.
1.2 Shallow Junction Processing Primer
Implantation has been used for over 20 years for device fabrication to introduce dopants
into the substrate silicon, and annealing has been used to repair the damage from the
implant. Throughout the years, implant energy and temperature cycles have been signifi-
cantly reduced. Typical times of annealing today are several seconds and energies less than
1 keV are not uncommon. There are two major challenges to understanding the junction
formation process. The first is prediction of the active, substitutional concentration of the
dopant atoms, which is critical since only the active concentration contributes to the
electrical properties. The other challenge is to understand the diffusion profile accurately,
which determines in most cases the junction depth. The junction depth and steepness also
determine electrical properties of the device.
At a first glance, the implant and anneal cycle does not seem to be a difficult system to
understand. Ionic stopping in solids is fairly well understood. Solid-state diffusion is also a
well-characterized problem. However, the common dopant atoms in silicon are tightly
bound when substitutional and do not diffuse by themselves. They require an interaction
with a lattice point defect, either a vacancy or an interstitial to move. The dopants pair with
the defects to forma mobile species. Consequently, the apparent diffusivity of the dopant is
proportional to the local point defect concentration in silicon. However, the crystal damage
created by the implantation process, initially in the form of vacancies and interstitials,
dramatically enhance the diffusivity of the dopant atoms as they evolve during annealing.
This process is called transient-enhanced diffusion (TED). During annealing, point defects
also react with the dopant atoms to form clusters, which deactivate the dopant atoms.
Predicting the damage interaction with the dopant atoms becomes very challenging
indeed, particularly since there usually are no good ways to measure the point defect
concentrations directly in the silicon. The junction depth and activation of the dopant layer
both depend on the point defects, which are produced in large concentrations from the
implant process itself.
1.3 Implant Process: Nuclear and Electronic Stopping
Understanding the implant process requires an understanding of how ions come to rest in
the silicon substrate. The ions lose energy in the crystal in two ways. The ions can
experience nuclear collisions with the lattice ions. Typically, this displaces the lattice ion
creating a vacancy and an interstitial. The lattice atom recoils with an energy that is
frequently sufficient to create other recoils. The implanted ion can also lose energy to the
electronic cloud present in the crystal. The ions positive charge tends to accelerate the
electrons in the vicinity, which creates a drag force on the ion. Electronic stopping does not
create crystal damage. Since the collisions have a randomnature, the profile ends up with a
statistical distribution.
Nuclear stopping is the more interesting, complicated component and has a much larger
impact on the final electrical properties of the structure. Point defects are generated in
abundance fromboth the primary ion and the generated secondaries. At high doses and for
implants with mass greater than silicon, it is not uncommon to amorphize the surface of the
ß 2008 by Taylor & Francis Group, LLC.
wafer. Implants are typically characterized as being either below or above the amorphiza-
tion threshold.
Implantation creates a large total population of point defects and smaller net defect
population following initial recombination. The total population includes all of the gener-
ated Frenkel pairs from primary and secondary collisions. These Frenkel pairs typically
anneal out, leaving behind no damage. However, we have also added atoms to the crystal
in numbers equal to the dose. These excess atoms can drive defect precipitation. Since
most dopants are very soluble, on average most of the excess atoms are lattice silicon
interstitials. This simple picture of the results of annealing is frequently referred to as the
‘‘þ1’’ model of diffusion [6].
Implants below the amorphous threshold can generate enough point defects to result in
observable extended defect structures. The extended defects are precipitation of the excess
interstitials produced from the implantation. The extended defects tend to form at the
projected range of the implant, since this is the position of the maximum of the excess
interstitial population. Implants above the amorphization threshold amorphize the top
surface. There is an excess interstitial population beyond the amorphous interface fromthe
extra implanted primary ions and atoms that are recoiled from the amorphous layer.
During annealing at short times and low temperatures, the amorphous layer regrows.
This is called solid-phase epitaxial regrowth (SPER). At higher temperatures, extended
defects formin the region at or near the original amorphous interface. The defects are often
referred to as end-of-range (EOR) defects.
Current ion implantation trends are focused on delivering the highest beam current
(for high throughput) at the lowest energy (for shallowjunctions). For process technologies
below 0.25 mm, ion implanters in production are usually operated in deceleration
mode. The dopants are extracted at a higher energy and the decelerated to a lower
energy in the main beam-line. Though there are concerns over possible impurity issues
and energy contamination problems, these are the current production workhorses for
shallow junctions.
Plasma implantation is another possibility. A plasma with a dopant carrying gas is
created and an electric field generated by a DC bias accelerates ions to the surface of the
wafer. The systems are similar to design as a reactive ion etcher, but they operate with
higher delivered ion energy to implant the dopants. Beamcurrent and processing through-
put in these systems are impressive. However, it is not a mass selected system and
contamination remains a concern.
Another ion implantation approach is to use clustered ion beams. In these systems a
molecule is ionized and implanted. The kinetic energy of the molecule is imparted to each
atomic species as a ratio of the atomic mass to the total molecular mass. The resultant
kinetic energy of each implanted species is much lower than the total kinetic energy of the
ion cluster. The implanter can consequently operate at a higher acceleration voltage, and
the effective dose rate of dopant ions delivered to the wafer can be much higher. There is
still debate on whether the cluster implant will also cluster dopants together and create
problems during annealing, so these are not yet in production but could be adopted soon.
1.4 Annealing
The damage created by implantation must be annealed to increase the mobility of the
carriers and to fully activate the dopant. Amorphous layers regrow at fairly low temper-
atures (4008C–6008C), but this does not remove all of the crystalline damage from the EOR
damage region. Higher temperatures are required to eliminate the crystalline damage from
ß 2008 by Taylor & Francis Group, LLC.
the projected range region in nonamorphizing implants and the EOR damage region of
amorphizing implants.
During annealing, the dopants diffuse. This diffusion for most dopants is accomplished
on the microscopic scale by interaction with point defects. As a simple picture, the
diffusivity is proportional to the excess point defect population. This leads to the damage
from the implant, increasing the diffusivity until the damage is fully annealed, known in
the literature as TED. This process is reasonably well understood. The implant-induced
damage condenses into various extended defect structures, which are discussed in more
detail subsequently. As these extended defects evolve, they release interstitials that drive
the diffusion enhancement. These interstitials are thought to recombine at surfaces. The
extended defect dissolution and evolution controls the diffusion enhancement duration
and magnitude. This has driven the study of extended defect behavior in silicon. As
mentioned earlier, interaction with defects can also control dopant activation processes.
Complete understanding of the dopant, point defect, and extended defect system is still
being sought.
Because of transient effects brought on by damage, the trend in annealing has been to
reduce the time of the anneal. Furnace annealing in silicon is now mostly a thing of the
past. Current annealing technologies are done in rapid thermal annealing (RTA) for times
in the duration of seconds and temperatures between 9008C and 12008C. In current
operation, anneals are called spike anneals, which have no dwell time at temperature
whatsoever. The entire anneal is done as a ramp up to peak temperature and an immediate
cool down. These short time anneals present difficulty in understanding the dopant and
defect system, as many of the ‘‘pseudoequilibrium’’ approximations break down. For
example, regrowth processes of the amorphous layer become more important as time at
temperature becomes shorter and no longer can be neglected.
In recent years millisecond annealing has come to the forefront in the semiconductor
industry in its attempt to continue to drive Moore’s law by reducing the total thermal
budget imparted to the wafer. Commercially known as flash-assist rapid thermal process
(fRTP), this process is designed to operate within the time gap between spike, rapid, and
laser thermal processing techniques. The process offers three main advantages over con-
ventional RTA systems that stem from the differences in the heating technology. Tungsten
filament lamps are utilized in conventional RTAs compared to the water-walled arc lamps
in flash systems. The arc lamps, which are very high-quality optical sources, deliver greater
power; have a faster response time; and deliver short wavelength radiation that is more
effective in heating the silicon substrate. A high-pressure argon plasma in an arc lamp,
when heated to 12,000 K, produces radiation power of 1 Â10
6
W, enabling very high ramp
rates that are four orders of magnitude higher than a conventional RTA, which has a
resultant power of 1 Â10
3
W. The smaller thermal mass of the argon in the arc lamps also
enables the arc lamps to respond approximately 10 times faster than tungsten lamps [7].
Since the transition from heating to cooling is also a function of the response time of the
heat source, a wafer heated by flash techniques will transition much faster from heating to
cooling. Finally, over 95% of the arc radiation is below the 1.2 mm band gap absorption of
silicon compared to 40%for radiation generated by the tungsten lamps [7], hence it is more
effective in heating the wafer.
The process uses a continuous arc lamp to heat the bulk of the wafer to an intermediate
RTP (iRTP) temperature, where the dwell time is essentially zero. This heating is slower
than the thermal conduction rate through the wafer, thus the entire wafer remains at
approximately the same temperature [8]. The iRTP serves as the initial temperature
of the flash anneal. Subsequently, a capacitor bank is discharged through an arc lamp,
which adds additional power to the device side of the wafer at a rate much faster than the
thermal conduction rate. Short time pulses allow for heating of the surface of the wafer
ß 2008 by Taylor & Francis Group, LLC.
to the peak flash temperatures (fRTP), while the substrate never attains these high tem-
peratures. This is possible since the time constant of the flash, which is on the order 1 ms, is
much shorter than the thermal time constant of the wafer ($10–20 ms). Therefore, a thin
slice of the device side of the wafer is heated and cooled rapidly at rates on the order of
1 Â10
6
8C s
À1
. The fast cooling is achieved since the bulk of the wafer acts as a heat sink,
removing heat from the top layer via conduction much more efficiently and faster than
can be accomplished in bulk cooling. The high absorbance of the reactor chamber also
complements the cooling rates. However, as the flash time pulse approaches the time
constant of silicon, the flash elevates the bulk temperature considerably, therefore only
allowing for cooling by radiative and convective methods, resulting in much slower cool-
ing rates. Heating rates up to the intermediate temperature are similar to conventional
spike annealing (508C s
À1
–3008C s
À1
), as are cooling rates during the bulk radiative cooling
(up to $1508C s
À1
).
1.5 Defect Evolutionary Processes
In high concentrations found after implantation, free interstitials cluster to reduce free
energy. The interstitial cluster configurations believed to occur in ion-implanted silicon
include the di-interstitial, interstitial chain, {311} rod defect, and dislocation loops. Trans-
mission electron micrograph (TEM) images of {311} defects and dislocation loops are
shown in Figu res 1.1 and 1.2, respec tively. The free inter stitial has strai n energy asso ciated
with it because it is larger than any interstitial site. It also has a free energy of 1 eV from
each unbonded orbital. Adi-interstitial represents a more stable configuration compared to
the free interstitial since it reduces the number of unbonded orbitals. Theoretically, by
FIGURE 1.1
{311} defects imaged in plan-view, weak-beamdark-field TEM. The left picture is after a preanneal of 2 h at 7508C.
The right picture is of an in situ anneal of the sample at left after 15 additional minutes at 7708C. (FromLaw, M.E.
and Jones, K.S., International Electron Device Meeting, 2000, p. 511. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
forming an interstitial chain in which interstitials are bonded both to the lattice as well as to
each other in a linear fashion, the number of dangling bonds can be further reduced. This is
supported by results of Cowern [9] obtained by modeling interstitial supersaturation
measurements, which suggest that interstitial clusters have stable configurations
(e.g., n ¼8) below the size of a {311} defect. The interstitial chain configuration has been
used in many models for the formation of extended defects in silicon [10–12]. Formation of
such an interstitial chain elongated in the <011> direction is the foundation for modeling
{311} defects by adding several <011> chains in the <233> direction, forming an extrinsic
stacking fault on the {311} habit plane. Tan [10] has shown that such an extended defect
further reduces the free energy of the excess interstitials since the {311} defect has no
dangling bonds along the sides of the defect (strained reconstructed bonds exist at the
ends of the rod). Of the aforementioned defects, only {311} defects and dislocation loops
have been directly observed by electron microscopy. The {311} defect is a metastable defect,
having a stability intermediate between self-interstitial clusters and dislocation loops. Both
{311} defects and dislocation loops represent an increase in the free energy of the silicon
lattice compared to its equilibrium state. {311} defects have a lower energy of formation
than loops, but loops lead to a configuration with a lower chemical potential per intersti-
tial, and are therefore a more stable configuration than {311} defects. Dislocation loops are
categorized according to structure as either faulted or unfaulted. Faulted loops are known
as Frank loops, whereas perfect (unfaulted) loops are known as Shockley loops. The
Burger’s vector for Frank loops is a=3 <111> and for Shockley loops it is a=2 <110>,
where a denotes the lattice constant. Each of these types will be discussed in turn.
1.5.1 Submicroscopic Defects
Weak-beam dark-field imaging conditions generally yield the highest resolution images of
extended defects when imaging using amplitude or diffraction contrast. Under these
FIGURE 1.2
Dislocation loops imaged in plan-view, weak-beam dark-field TEM.
ß 2008 by Taylor & Francis Group, LLC.
cond itions the sm allest cluste rs (d islocation loo ps or precipitat es) that can be observed
usually com prise over 100 atoms . If excess inter stitials are introduce d into the crys tal, then
as the y precipitat e out they can form submi crosc opic inter stitial clus ters (SMICs ). Some -
times the se clus ters gr ow into bare ly observabl e dot defects. Figu re 1.3 show s an example
of a TEM of dot defe cts. Dot defe cts and SMICs eve ntually ma y evolve int o {311} defe cts
and dislocat ion loops, or may dis solve before they get large r. If dopant s are abov e solid
solubili ty in the crys tal then submi croscopi c dopant clusters may also fo rm. These sub-
micro scopic defe cts are imp ortant becaus e they can dr amatical ly affect dopant diffusion
and activati on upon formatio n and dis solution.
The m ost well-u nderstood exa mples of submi crosc opic dopant clu sters are associ ated
with boron. It is wi dely accepte d that bor on cluste rs into imm obile boron inter stitial
comple xes [13,14] . Most of these clu sters are bel ieved to be made of few er than 10 atoms
and thus are subm icroscop ic. At short tim es, boron forms interst itial-rich cluste rs that alter
the damag e formatio n. At longe r tim es, these cluste rs become boron rich and are more
stable. The breakup of the cluste rs tend s to release inter stitials and can be modi fied by
the interstitial conte nt — they dis solve slower in an oxidizin g env ironmen t, which inje cts
inter stitials [15]. Figu re 1.4 shows the boron clus tered in a damag e region and held
imm obile [16].
Arseni c is less we ll unde rstood than boron. At low dose s, arsen ic absorbs interst itials [17].
At intermedi ate dose level s, the arsenic forms cluste rs cen tered aroun d vacanci es [18,19] .
In this dose range , the proces s inje cts int erstitial s that can form extended defects [20]. At
higher doses, arsenic forms larger precipitates, and the defect character seems to change
[21,22]. The kinetic rates for both processes need to be fully understood to be able to
maximize the dopant activation. The characterization of these behaviors is a critical near
termchallenge for junction profiling.
FIGURE 1.3
TEMof dot defects. These defects are too small to classify as a specific type and typically consist of approximately
100 interstitials.
ß 2008 by Taylor & Francis Group, LLC.
Many com mon implants are also acti ve in trap ping defe cts. Carbon is thou ght to clus ter
with inter stitials [23]. Nitrogen [24] and fluori ne [25] beha ve simi larly, in that they trap a
vacancy during regrowth process es, and then act as an inter stitial trap during annealin g. It
is likely that we do not yet understan d all of the impurity int eraction s with defe cts in small
clus ters. It wou ld appe ar mo st impuritie s a re likely to be point defect acti ve and form
clus ters of some type below observabl e TEM detectio n.
1.5.2 Dot Defects
As SMI Cs coars en abov e 100 atoms, they become visi ble. Below a few 100 ato ms, it is often
not possible to determine the exact natu re of these defe cts, e.g., are the {311} ’ s loo ps or
volu me precipitat es? These are terme d dot defects. These dot defec ts can be extre mely
num erous and are more commo n as the implant energy and annea ling cycles decrease .
They are often observed after flash or nonme lt lase r anne aling of ultra low- energy
impl ants. They a lso are observe d to form in the EOR region after solid-p hase epitaxial
rec rystalliza tion (SPE R) of an amorpho us laye r. SP ER (typ ically less than 700 8C) is being
expl ored as an alternati ve to high -temp erature anneal ing for junct ion formation. Figu re 1.3
shows an image of wha t are commonl y observe d as dot-lik e defects.
1.5.3 311 Defec ts
The {311} defe ct is a long thi n plate of excess inter stitials lyi ng on a {311} plane [11,26,2 7].
An example of the defe ct st ructure is sh own in Figure 1.1. They have been reporte d
to typically remain 40–80 Å wide, independent of the length of the {311} defect [28]. {311}
defects are observed in ion-implanted silicon above a certain threshold dose and energy.
10
19
10
18
10
17
10
16
0 2000 4000
Furnace anneal at 800ЊC
None
35 min
65 min
118 min
180 min
Boron implant
2 ϫ10
14

cm
−2
at 60 keV
Depth (Å)
C
o
n
c
e
n
t
r
a
t
i
o
n

(
c
m

3
)
6000
FIGURE 1.4
Boron profile as a function of depth and anneal time. The peak region of the boron that is immobile during the
anneal has formed boron-interstitial clusters. (After Michel, A.E., Rausch, W., and Ronsheim, P.A., Appl. Phys.
Lett., 51, 487, 1987. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
For silicon implanted at 40 keV and doses above 7 Â10
12
cm
À2
, {311} defects are observed
after annealing at temperatures between 7008C and 8508C for several minutes [28,29].
{311} defects are observed both in the projected range of nonamorphizing implants
as well as in the EOR in amorphizing implants. Technological interest in the {311} defect
arose after studies of implants into doping superlattices showed that the dissolu-
tion kinetics of these defects matched the time scale of the effect known as TED [30],
which describes the enhanced diffusion of implanted dopants during annealing of ion-
implanted layers. The possible evolutionary pathways observed for {311} defects in
ion-implanted silicon are dissolution, conservative unfaulting, and congruent dissolution
and unfaulting.
Pan has shown that {311} defects are metastable defects compared to dislocation loops
[31]. Though {311} defects have negligible strain energy due to bonds along the elongated
sides of the defect and relatively small stacking fault energy due to the small dislocation
Burger’s vector, a=21 [116] [32,33], {311} defects become unstable at temperatures above
$7008C. Dissolution curves for {311} defects in the EOR damage region created by
120 keV 1 Â10
15
cm
À2
Si
þ
implants at 7508C and 8008C are shown in Figure 1.5. Modeling
the dissolution of interstitials from {311} defects with a decaying exponential function,
the characteristic decay time constants are 50 and 10 min for 7508C and 8008C, respect-
ively. With annealing temperatures above 9008C, either the {311} defect nucleation and
dissolution process time scales are much shorter than typical annealing times, or the {311}
defect does not exist. At higher temperatures, when the supersaturation of interstitial
concentration above intrinsic levels falls below a critical level, circular dislocation loops
of sufficient size become a thermodynamically favorable defect configuration with respect
to {311} defects. Therefore, once the activation barrier for the formation of a loop is
overcome, loop growth is preferred over {311} growth. This explains why, at sufficient
doses (>1 Â10
14
cm
À2
) [34] and longer annealing times and higher annealing tempera-
tures, {311} defects are not observed; only dislocation loops. Extension of the same
reasoning explains how it is possible that interstitials from {311} defects, when provided
with the necessary energy to leave the defect, could then be accommodated by a disloca-
tion loop and lower the energy of the system. In addition, Li and Jones [35] have
shown that a {311} defect can form an elliptical dislocation loop by what appears to be
dislocation unfaulting.
0 50 100
3ϫ10
13
2ϫ10
13
1ϫ10
13
I
n
t
e
r
s
t
i
t
i
a
l
s

i
n

{
3
1
1
}

d
e
f
e
c
t
s

(
a
t
o
m
s
/
c
m
2
)
0
Time (min)
150
t
750
=50 min
t
800
=10 min
800ЊC
750ЊC
200 250
FIGURE 1.5
Density of silicon interstitials in {311}
defects as a function of annealing time at
7508C and 8008C for a 120 keV 1 Â10
15
cm
À2
Si
þ
implant.
ß 2008 by Taylor & Francis Group, LLC.
Kim [12], Takeda [11], and Tan [10] have modeled the fo rmation of the {311}
defect. These model s and emp irical evidence [32,33] result in a con sensus of a=21
< 116 > for the Burger ’ s vector of the {311} defect. Few have observe d or model ed
the transfor mation of a {311} defect int o a dis location loo p. Eagle sham [28] has observe d
that {311} defe cts can unfault to form Frank loops. The defect reaction proposed by
Eagle sham
a=21 < 116 > þ a=21 < 661 > ¼ a=3 <111 >
gives a plau sible unfau lting rea ction. Eagle sham then postu lated that Frank loops may
unfau lt to form Shockl ey dislocat ion loops throu gh the commo n unf aulting reaction ,
a=3 < 111 > þ a=6 < 112 > ¼ a=2 < 110 > ,
which has been observe d in many cubic lattices .
Li and Jo nes [35] have observe d similar transf ormati ons of the {311} defe ct into
dislocat ion loops via in situ TEM measure ments. This transf ormati on may occ ur while
cons erving the number of interstitial s bound in the defect. If such con servation is main-
tained, then the defe ct transf ormatio n itself should not affect the supersat ura tion of the
inter stitial concen tration in the surrou nding crys tal.
Disso lution of inter stitials from {311} defe cts is anothe r possibl e evol utionar y pat h for
the inter stitial s, as the defect become s thermod ynami cally unstab le. Eagl esham [30]
propo sed that the dissolu tion of inter stitials from {311} defects is the source of int erstitial s
that induce TED , since a correlat ion exists betw een both time constant and energy barri er
for bot h TED and {311} dis solution. Suc h dis solution is expe cted since recom bination,
surfa ce or bulk, represen ts a lower free energy of the system. At high er tem perature s,
{311} defects are relative ly unstable and dissolve upon anneal ing. This occurs after only
3 min at 815 8 C [30].
In situ anneal ing in the TEM allows indivi dual defect beha vior to be observe d and
mo nitored so that the dissolu tion proce ss can be obse rved [36]. A 100 keV Si 10
14
cm
À 2
impl ant was use d to damage a silicon wafer. Th ese sam ples were then prean neale d at
750 8 C for 2 h in a con vention al furnace . The sam ples were the n anne aled in situ in the
TEM at a vari ety of temperat ures . Figu re 1.1 shows the evol ution at 0 and 15 min. In thi s
tim e period, no te that the defect in the lowe r left has comple tely dis solved. Th e longer
defects a re not more stab le energetical ly than smaller defects. This work clearly sh ows
that longer defe cts can dissol ve muc h faste r than shor ter defects . Figure 1.6 shows the
dissol ution of nine differe nt defe cts from the {311} ense mble. These dissolu tion curves
are fi t better by line ar decay rates than exponenti al decay rates. Linear decay fi ts are
shown in Figu re 1.6, and the decay rates as a function of initi al defect size are extracte d.
Captu re and release of int erstitial s on the {311} defects occ urs on ly at the end of the
defects , and therefo re is proportio nal to the num ber of defe cts, D
311
. This provi des
two dis tinct resu lts. First, indivi dual defe cts dis solve at a nea rly constant rate , since the
dissol ution is proporti onal on ly to the end size. The length of the defect does not
determi ne the dissolu tion rate.
{311} defects may also unfault to form dislocation loops, while simultaneously releasing
inter stitials int o the surroundi ng crystal. Figu re 1.7 show s as an exa mple a {311} defect in
a 100 keV, 2 Â10
14
cm
À2
Si
þ
ion-implanted silicon wafer which had been preannealed in a
furnace at 8008C for 5 min. After the furnace preanneal, this defect contains 6400 inter-
stitials. The sample is then annealed in situ in a TEM. Over the 40 min in situ anneal, the
{311} defect unfaults to form a dislocation loop. Over this same interval the number of
interstitials bound by the defect decreases from 6400 to 1500.
ß 2008 by Taylor & Francis Group, LLC.
0
50
100
150
200
250
300
350
0 20 40 60 80 100 120
S
i
z
e

(
n
m
)
Time (min)
FIGURE 1.6
Size of individual defects as a function of in situ anneal time. Best fits are linear decays. The slowest decaying
defect is about average size at time zero. (From Law, M.E. and Jones, K.S., International Electron Device Meeting,
2000, p. 511. With permission.)
FIGURE 1.7
In situ annealing showing a {311} defect unfaulting into a dislocation loop.
ß 2008 by Taylor & Francis Group, LLC.
1.5.4 Loops
The formation threshold for dislocation loops [34] (around 1 Â10
14
cm
À2
for implants
below 100 keV) is considerably higher than the reported threshold dose for {311} defects of
only 7 Â10
12
cm
À2
for 40 keV Si
þ
implants [28]. Thus, it appears that the nucleation
barrier for theformation of a dislocation loop is higher than that of a {311} defect. For
higher-energy implants (380 keV–1 MeV), the threshold dose for loops can drop as low
as 4 Â10
13
cm
À2
[37]. The decrease in the threshold dose with increasing energy is thought to
be due either to the increase in damage deposition [37] in the crystal or to the increased
separation of the Frenkel pairs [38–40] that reduces the I–V recombination efficiency. As
mentioned previously, it has been proposed that loops may simply evolve from unfaulting
of {311} defects [28]. Li and Jones [35] showed that, for nonamorphizing implants, all of the
dislocation loops that were observed to formcame from{311} defects. It has not been proven
that the same process occurs at higher temperatures, 9008C–12008C. Once the nucleation stage
for dislocation loops has been completed, loops either remain stable, coarsen, or dissolve.
Dislocation loops are much more stable than {311} defects, requiring temperatures of
10008C–11008Cto dissolve [34,41,42]. Figure 1.8 shows that, at lower annealing temperatures,
dislocation loops continue in the growth stage kinetically for hours. This time scale is well
beyond the {311} defect growth time scale, and occurs as {311} defects are dissolving. Often
oxidation-induced interstitial injection prevents complete dissolution of dislocation loops,
instead resulting in the growth of the loops into large stacking faults. Dislocation loop
dissolution can provide interstitials for some diffusion enhancement [43], but because the
temperature is so high, the relative enhancement C
I
=C
I
* (C
I
* is the intrinsic concentration of
interstitials) is not as large as the effect from{311} dissolution at lower temperatures.
The presence of dislocation loops in the silicon crystal represents an increase in the
free energy of the lattice when compared to the equilibrium lattice. During annealing,
the minimum size for a stable dislocation loop depends both on the temperature
2ϫ10
14
1ϫ10
14
0
0 50 100 150
Time (min)
I
n
t
e
r
s
t
i
t
i
a
l
s

i
n

d
i
s
l
o
c
a
t
i
o
n

l
o
o
p
s

(
a
t
o
m
s
/
c
m
2
)
200 250
800ЊC
750ЊC
FIGURE 1.8
Density of silicon interstitials in dislocation loops as a function of annealing time at 7508C and 8008C for 120 keV,
1 Â10
15
cm
À2
Si
þ
implant.
ß 2008 by Taylor & Francis Group, LLC.
and the time-de pendent supersat uration of interst itials. Due to the fi nite tim e requi red
for the nucleatio n st age, there is a dis tribution of dislocat ion loo p sizes at a give n tim e. As
thermal annealin g conti nues, the dislocat ion loops move from a nucleati on stage to a growth
stage. Du ring the gr owth stage all of the dis location loops app ear to inc rease in size , so the
total numb er of inter stitial s bound by dislocat ion loo ps increa ses. Aft er the gr owth stage,
the loops then enter a coarsen ing stag e where the num ber of dislocat ion loo ps decre ases
and the aver age loop size increa ses, whil e the con centratio n of bound inter stitials in
the dislocat ion loops rem ains relative ly unc hanged [44]. The coa rsening pha se has been
model ed both by Liu [45] and Laa nab [46]. In the coars ening phase the sup ersatura tion of
inter stitials aroun d the loo ps drops signi fi cantly [47]. Some autho rs [47 –49] have termed this
coars ening stage of dislocat ion loops ‘‘Ostwa ld ripeni ng, ’’ since its beha vior mim ics that of
second phase precipitat es in many met als systems , where larger preci pitates grow at the
expens e of sm aller preci pitates to redu ce the surfa ce free energy of the precipitat es.
Typi cally loops enter a coarsen ing pha se wi thin 30 – 60 min at 800 8 C [50]. For room
temperat ure impl ants, if the imp lant conditio ns are such that the defects fo rm at dep ths
grea ter than 10 00 Å and the impla nt spe cies is below its solid solubi lity, then it has been
shown that high a nnealing tempe ratures ( >1000 8 C for 24 h) are require d to dis solve EOR
defects comple tely [34,41] . These tem perature s are , in general , well beyond the thermal
budge ts of mo dern integr ated circ uit (IC) proces sing. Howe ver, as the imp lant energi es
decreas e, the implant temperat ure control inc reases and the dose rate increa ses the EOR
defects become less st able [51 –53]. This is attribu ted to a decreas e in the net excess
inter stitials coming to rest in the EOR regi on, as opp osed to a tru e surfa ce prox imity effect
[54]. A review of the effect of impl ant and anne al cond itions on the stabilit y of the se defe cts
has been presen ted elsew here [40,55] .
Theore tically, no dislocat ion loo p is thermo dynami cally stable in a single- crystal silicon,
since it represen ts an increa se in the free energy of the crystal. Ki netic limit ations limit thi s
theo retical the rmody namic outlook on the stabilit y of dislocat ion loo ps in ion-imp lanted
silicon. As a result of the se limitati ons, many dislocat ion loops in ion-imp lanted silicon are ,
for all pra ctical purp oses, st able during postimplan tatio n anneal ing. Dislocatio n loops that
remain in submicro n silicon transi stors after proce ssing can be detri mental to the device.
Disloca tion loo ps in silicon tend to get met al impuritie s, which dis rupt the electro n
transpo rt in the vicini ty of the loop. Buck [56] and Land i [57] have shown that, if disloca-
tion loops exis t in the space charge regio n of a junct ion, they can cause high leakage
currents. For these reasons, the electrical junction of a silicon transistor must form deep
enough so that dislocation loops are never in the space charge region of the device.
1.6 Defects in Ultra-Shallow Processes
1.6.1 Surface Effects
In an effort to investigate the role of the surface on interstitial recombination, a series of
experiments was conducted using preamorphized Si. Czochralski (CZ) grown (100) Si wafers
were implanted with 1 Â10
15
cm
À2
germanium ions at 5, 10, and 30 keV. This produced
amorphous layers of 120, 220, and 480 Å below the surface, respectively. After annealing
at 7508C for varying times, TEM images were taken and the defects quantified. Figure 1.9
shows that quantification of the defects showed that, as the energy decreased to 5 keV,
the dissolutionkinetics rapidly accelerated. The question arises: Is this a result of the proximity
to the surface, or the decrease in interstitial population in the end of range region?
To study this further, lapping experiments were performed. A sample was implanted
ß 2008 by Taylor & Francis Group, LLC.
with a 10 keV Ge 1 Â 10
15
cm
À 3
implant. Next, pieces of the wafer were chemically, mechan-
ically polished to reduce the thickness for the amorphous layer from 180 Å down to 20 Å.
The aforementioned 5 keV sample had an amorphous thickness of $ 100 Å. Thus, if the surface
is the reason for the rapid dissolution, then thinning the sample should have a dramatic effect
on defect dissolution rates. After thinning, the samples were annealed at 7508C for two
different times, and plan-view transmission electron microscopy (PTEM) was used to quantify
the trapped interstitials in the defects. Figure 1.10 shows that, as the amorphous layer was
thinned, no surface effect was observed until the amorphous layer was less than 60 Å.
This means that the surface proximity did not have a measurable effect on the defect evolution
between 60 and 180 Å. This is counter to most of the models, which consider the surface as
an infinite sink for interstitials. Hence, there should exist a significant gradient of interstitials
toward the surface. However, these results imply that the surface is not as strong a sink as
previously thought.
1.6.2 Milli second Annea ling
Dama ge anneal ing in the millis econd time regime presents a unique oppor tunity to
inv estigate the ear ly st ages of the damage evol ution proces s, which we re no t previousl y
poss ible with forme r availabl e anne aling proces ses such as RTA . The high tem perature s
attainab le by the flash annealin g techni que, coupled with the extrem ely short anneal times,
has enabled the defect evolut ion to be ma pped from the early stag es in the evolut ion to
matu re defect struc tures. These have been exten sively charac terized in the past, so thi s
process allows for a more complete picture of the defect evolutionary processes. Such
knowledge is crucial to an understanding of the interstitial concentrations, and hence the
10
15
10
14
10
13
TEM detection limit
10
12
10
11
T
r
a
p
p
e
d

i
n
t
e
r
s
t
i
t
i
a
l
s

(
#
/
c
m
2
)
10
10
10
9
1 10 100 1000
Time (s)
10
4
10
5
30 keV
10 keV
5 keV
FIGURE 1.9
Dissolution kinetics of defects as the energy of the implant is reduced. At a threshold of 5 keV, the defects rapidly
dissolve even though earlier kinetics is not significantly different.
ß 2008 by Taylor & Francis Group, LLC.
mechani sms that are largel y respons ible for dopant diffusion in the silico n latt ice duri ng
thermal proce ssing cycles.
We now conside r defect evol ution for amo rphizing germ anium impl ants in silico n,
during fl ash anneal ing. Tr aditional ly defe ct evolut ion studi es have bee n con ducted by
time-de pende nt studies usi ng iso thermal anneals [30,58 –60] in which speci fic defe ct
struc tures could be isolat ed. The nature of the fl ash anneal ing proces s does not pe rmit
for such investig ations, since the anneal time is set by the full wi dth at hal f maximu m
(FWHM ) of the radiat ion pulse, which is on the or der of millisec onds. Thus, to investigate
the evol ution, on e must reso rt to isochron al anneal s at different tem peratures . Studies
cond ucted on the eff ect of fl ash anneal ing on the evolut ion of the damage for amorp hizing
germ anium implants into silicon reveal the presence of defects at different stages in their
evolution, for different flash anneal temperatures. Such studies have been conducted on
30 keV, 1 Â 10
15
cm
À 3
germanium implants performed on (100) n-type CZ grown silicon
wafers [61]. Peak flash temperatures of 10008C, 11008C, 12008C, and 13008C are investigated,
for a constant iRTP anneal temperature of 7008C. Figure 1.11 illustrates the PTEM images of
the damage observed as the flash temperature is varied. The 7008C iRTP anneal results in the
formation of dot-like defect structures of very high density. Similar dot-like defects are present
in the microstructure after the 10008C and 11008C flash anneals, but appear to be larger and of
a lower density. As the flash temperature is increased to 12008C and 13008C, respectively,
{311}-type defects and dislocation loops are evident.
The damage presen t in the struc ture is cons istent with type II defects [34], com monly
known as EOR dam age, which occ urs beyond the amorp hous –crystal line int erface. That
different defect structures are identi fied a t different anneal ing temperat ures in this work is
not surp rising. Iso chronal anneals, altho ugh a valid experi mental approach to temperat ure-
depen dent studi es, do not yield EOR defects in the same pha se of their evol ution. At a given
isochron al a nnealing time, lowe r tem perature anneal s generat e EOR defec ts in their earlier
nucleati on, growth and coarsen ing stages, while high -temp erature anneals resu lt in defe cts
further along in their evolution, pos sibly in the dissol ution regime [62]. The evol ution of the
type II dam age observe d in this study at the high er anneal temperat ures concu r with
previous findings, as the EOR defects are observed to evolve from {311}-type defects into
dislocation loops with increasing flash anneal temperature.
The existence of {311}-type defects and dislocation loops at such high temperatures can
be explained by considering that the anneal times at these temperatures were extremely
1ϫ10
15
1ϫ10
14
1ϫ10
13
1ϫ10
12
100 50 0 150 200
1800 s
2400 s
T
r
a
p
p
e
d

i
n
t
e
r
s
t
i
t
i
a
l
s

(
#
/
c
m
2
)

Amorphous layer depth (Å)
FIGURE 1.10
Number of trapped interstitials as a function of
amorphous layer depth. Instead of varying
energy, the surface was thinned with polishing.
ß 2008 by Taylor & Francis Group, LLC.
short. Keys [62] demonstrates that it is possible to affect similar defect structures and
diffusion states at different anneal temperatures by determining an equivalent anneal
time. Knowledge of the time, t
1
, required for defect dissolution at a given anneal tempera-
ture, T
1
, allows one to calculate the anneal time, t
2
, required to affect similar defect
structures at another temperature, T
2
, from the ratio of the time constants for decay, t, in
accordance with the relation:
t
2
¼
t
1
t
1
t
2
(1:1)
in which t is related to the anneal temperature, T, by an Arrhenius relation given by
Equation 1.2 that includes an activation energy, E
a
, a preexponential factor, t
0
, and the
Boltzmann constant, k,
t ¼ t
0
exp
E
a
kT

(1:2)
The value of the E
a
used in the calculation is based on the dominant defect structure
present in the microstructure. Thus, in the case of the {311}-type defect, approximately 5 ms
at 12008C is required to affect total {311}-type defect dissolution based on an activation
energy of 3.7 eV [30,58]. This time exceeds the size of the radiation pulse produced by the
(a) (b)
(c) (d) (e)
66 nm
FIGURE 1.11
WBDF PTEM images of the EOR defects imaged under g
220
two beam conditions of a 30 keV, 1 Â10
15
cm
À2
Ge
amorphizing implant into Si (100): (a) 7008C iRTP; (b) 7008C iRTP, 10008C fRTP; (c) 7008C iRTP, 11008C fRTP;
(d) 7008C iRTP, 12008C fRTP; and (e) 7008C iRTP, 13008C fRTP. (FromCamillo-Castillo, R.A. et al., Appl. Phys. Lett.,
88, 232104, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
flash lamps by an order of m agnitude. He nce the presence of {311}- type defect in the
micro structur e afte r the 1200 8 C fRTP is plausible, since the time of the flash anneal was
insuf fic ient to affect dissolu tion based on the equiva lent anneal time calcul ation . A similar
argu ment holds for the occ urrence of dislocat ion loops upo n anneal ing at 1300 8 C. An
anneal tim e of 0.3 ms is requi red to acc ount fo r the diffe rences in the int erstitial densi ties
betwe en 1200 8C and 1300 8 C fla sh anneal s, which falls within the range of the FW HM of the
flash pulse at 1300 8C, con firmi ng that the anne al time is suf fi cient to affect {311}- type defect
dissolu tion, and expl aining the presence of on ly dislocat ion loops in the micro structure.
Ther e has bee n no previo us experi mental evid ence of the EOR defe cts evol ving from the
dot-lik e structur es to {311}- type defects. To con fi rm that these defe cts are in the early stages
of their evol ution, pos t flash thermal proce ssing, consisting of a spike rapid thermal anneal
at 95 08C, is perform ed on the ma terial con taining these dot-lik e structur es. Th is is based on
the notion that the additi onal the rmal budge t sh ould evol ve the damag e if it is in its
infancy. Ex aminati on of the microst ructure subsequ ent to the 950 8C spi ke RTA reveals
the presence of dislocat ion loops in the struc ture, which corrobo rates the the ory that the
dot-lik e defects evol ve into mo re stable defect struc tures wi th subs equent anneal ing [61].
The exa ct con figura tion of small inter stitial clus ters has been the center of a number
of inv estigati ons, yet very little is still known about them. Rece nt experim ental and
theo retical data [9,63,64 ] demons trate that preci se cluste r size s exh ibit enhanced stabilit y,
indicate d by the existe nce of minima and maxi ma in the cluste r binding energy curve.
Howev er, cons iderable debate rem ains over the exa ct sizes of the stable clu sters. Cowern
[9] found that interst itial clus ters that con sist of mo re than 20 atoms have a similar
different ial formati on energy to the {311}- type defect, sugge sting that the int erstitial clus-
ters unde rgo a transi tion to {311}- type defe cts a t a smaller clus ter size. Other investi gation s
[65,66] support this ide a and suggest that the transi tion from sm all int erstitial cluste rs to
{311}- type defects occ urs for inter stitial cluste rs con taining eight ato ms. He nce, the sm all
inter stitial cluste rs observe d at 700 8 C iRTP, and the 1 000 8C and 1100 8C flash may in fact
be {311}-typ e defects , since the total num ber of ato ms in these struc tures exceed s eight
atoms , and the sm allest defect that can be imaged by a conve ntional TEM is approxi mately
100 atoms [67]. Other st udies [60] of lower- energy germ anium amorp hizing implants
propose that small interstitial clusters may exhibit defect morphology very similar to
plate-like dislocation loops. These dislocation loops were shown to be very unstable,
dissolving with an activation energy of 1.13 Æ0.14 eV [60]. If the dot-like defects in these
experiments are analogous to small {311}-type defects alluded to by Cowern [9] or the
loops observed by King [60], then any additional thermal budget applied to them should
result in a defect dissolution behavior that adheres to the respective dissolution kinetics
of these defects.
Figure s 1.12 and 1.13, resp ectively, depict the defect densi ties and trapped interstitial
populations as a function of the peak anneal temperature [7,61]. It is evident that both of
these quantities decrease as the flash temperature is increased above the 7008C iRTP
anneal, indicating interstitial loss from the EOR damage as the defects evolve over the
course of the flash anneal. This loss of interstitials during the flash anneal process may be
attributed to interstitial recombination at the amorphous–crystalline interface during
regrowth of the amorphous layer, recombination at the surface on completion of the
regrowth, and interstitial loss to the bulk of the material. These simultaneous factors
suggest that the defects are in a coarsening regime in which a fraction of the interstitials
is not recaptured by evolving defect structures. Consequently, during the flash anneal, the
system is viewed as a ‘‘leaky box’’ from which interstitials are lost as the EOR defects
undergo coarsening.
Investigations of the kinetics of the defect decay are traditionally performed by fitting
the interstitial density with time over various temperatures to an exponential function of the
ß 2008 by Taylor & Francis Group, LLC.
form of Equation 1.1. In these cases isolated defect structures such as {311}-type defects
[30,58] and dislocation loops [59,60] were examined, enabling the kinetics for each defect
type to be extracted. Since the nature of the flash anneal process limits the investigations
10
13
10
12
10
11
10
10
10
9
10
8
10
7
0 500 1000 1500
Defect density
TEM detection limit
fRTP (ЊC)
D
e
f
e
c
t

d
e
n
s
i
t
y

(
c
m

2
)
FIGURE 1.12
Defect density as a function of fRTP anneal temperature for a 1 Â10
15
cm
À2
, 30 keV Ge amorphizing implant into
Si (100).
10
16
10
15
10
14
10
13
10
12
10
11
10
10
10
9
1000 500 0 1500
Trapped interstitial population (cm
−2
)
TEM detection limit
fRTP (ЊC)
T
r
a
p
p
e
d

i
n
t
e
r
s
t
i
t
i
a
l

d
e
n
s
i
t
y

(
c
m

2
)
FIGURE 1.13
Trapped interstitial density as a function of fRTP anneal temperature for a 1 Â10
15
cm
À2
, 30 keV Ge amorphizing
implant.
ß 2008 by Taylor & Francis Group, LLC.
to isochronal anneals, which affects EOR defects at different stages in their evolution,
conventional kineti c anal yses metho ds in which spe cifi c defect struc tures are isolat ed
cannot be empl oyed to determine the rate of inter stitial loss fo r single isolated defe ct
struc tures. Rather the inter stitial loss dat a fo r the durat ion of the flash anneal over differi ng
defect st ructures is obtained. This necess itates the developme nt of ano ther approac h to
analy zing the attai nable expe rimental dat a, so that mean ingful compariso ns can be made
to past studies [61]. The adop ted appro ach ent ails the selectio n of a referenc e anneal , from
which the decay in the trapped int erstitial dens ity can be tracke d with the flash anneal
temperat ure, allowing for the kinetics of the interst itial decay to be extracted . The interstitial
value cor respond ing to the 700 8 C iRT P anneal temperat ure can serve as such a ref erence for
the initi al trap ped int erstitial value. This is validate d by the obse rvatio n that all of the wafe rs
are hea ted to this tem perature prio r to app lication of the flash, and hen ce allo ws for the
decay in the trapped interst itials duri ng the fl ash portion of the thermal profi le, i.e., fl ash
anneal , to be extracted such that the effect of the flash on the defects coul d be isolated.
Figure 1.13 notic eably demo nstrat es that the trapped int erstitial decay follo ws an expo-
nenti al relati on as the flash temperat ure is increa sed. He nce, in accordance with kinetic rate
theo ry, the int erstitial decay can be app roximate d by the relatio n:
@ C
interstitials
@ t
¼ À
C
inter stitials
t
(1 : 3)
where
C
interstitials
is the concentrat ion of trap ped inter stitials (cm
À 2
)
t (s) is the anneal tim e
t (s) is the capt ured interst itial lifetime, which is relate d to the anne al temperat ure,
T (K), by an Arrhenius expre ssion that inc ludes an acti vation energy, E
a
t ¼ t
0
exp
E
a
kT

(1 : 4)
The proce ss simu lator FLOO PS [68] is utilize d to calculate the trapped inter stitial densi ty
for each flash anneal tem perature from an initial trap ped interst itial density (value for the
700 8 C iRTP anne al), by fitting the parame ters t
0
and E
a
. The tem perature –tim e vari ations
for each anneal need to be inc orporat ed int o the simulat ion to allo w fo r an accu rate
determi nation of t as time and temperat ure are altered. This approach facilitates the precise
integration of the trapped interstitials with time. The decay rates derived from the fits of
the experi mental trapped int erstitial pop ulatio ns are illus trated in Figure 1.14. The int er-
stitial decay rate is found to vary linearly with the inverse flash temperature, yielding an
activation energy, E
a
of 2.1 Æ0.05 eV and preexponential factor, K
0
, of 3.3 Â10
10
s
À1
. We
note that the interstitial decay rates varied over two orders of magnitude for the flash
temperatures investigated, from approximately 100 s
À1
at 10008C, compared to 2000 s
À1
at
12008C. Such high decay rates had never been observed.
Comparison of the interstitial decay rates for {311}-type defects [30] and dislocation
loops [59,60] demonstrates the much higher interstitial decay rates during the flash anneal
compared to former studies. The interstitial decay rates for the flash anneals range two
orders of magnitude between 1 Â10
2
and 1 Â10
3
s
À1
for anneal temperatures from 10008C
to 13008C, which are three orders of magnitude larger than the maximum rate previously
reported, obtained by Seidel et al. [59] for the interstitial decay from stable dislocation
loops. The interstitial decay rates for {311}-type defects and small dislocation loops were
much lower. The temperature range of Seidel’s experiments coincides with flash temper-
atures lower than 12008C. Yet, the interstitial decay rates are vastly dissimilar for this
ß 2008 by Taylor & Francis Group, LLC.
temperature regime, suggesting that the interstitial loss is most likely from a defect not
similar to the dislocation loop. This suggests the existence of a highly unstable defect
structure at these temperatures, and correlates with the dot defects observed at this
temperature during the flash anneal.
Examination of the differences in the activation energies for defect dissolution in the
literature provides further insight into the characteristics of the proposed highly unstable
defect. The extracted activation energies for interstitial loss during the flash anneal
are 2.7 and 1.6 eV smaller than the values obtained for dislocation loop and {311}-type
defect dissolution, respectively. That the activation energy determined for the interstitial
decay during the flash anneal is not similar to those previously extracted values
supports the theory that the interstitial loss is not from comparable defect structures.
It also clearly indicates that this defect is less stable than the {311} defect and disloca-
tion loops.
That dot-like defects exist in the structure for flash anneal temperatures of 11008C and
lower show that these defects only exist in the early stages of annealing, and either evolve
into {311}-type defects or dissolve. The decrease in the trapped interstitial concentration
between 11008C and 12008C flash anneal temperatures suggests that some of these defects
dissolved, losing interstitials to either the surface to the bulk of the material. The data
imply that those dot-like defects that did not dissolve must have therefore evolved into
the {311}-type defect detected after the 12008C anneal, since they are no longer observed
in the microstructure. This validates the supposition that the dot-like defect is a precursor
for the {311}-type defect. Consequently, the extracted kinetics for the decrease in trapped
interstitial density as a function of the temperature applies to the dissolution of this dot-like
defect structure.
1.6.3 Regrowth-Related Defects
Regrowth-related defects arise upon recrystallization of an implantation-induced amorph-
ous layer. These defects may arise if the amorphous–crystalline interface is rough. This can
occur for lighter ions, at higher implant energies, or if the temperature of the wafer rises
FIGURE 1.14
Arrhenius plot of the time constant derived for
defect decay extracted from the simulated
experimental data, indicating an activation
energy, E
a
, of 2.1 Æ0.05 eV for dissolution.
1
5.5 6 6.5 7
10
4
/T (eV)
7.5 8 8.5
10
100
1000
10
5
10
4
30 keV Ge flash-assist RTP
K
d
e
f
e
c
t

(
s

1
)
ß 2008 by Taylor & Francis Group, LLC.
during the implant. These defects may be in the form of hairpin dislocations [69] or
stacking faults and microtwins. Because the amorphous–crystalline interface is not planar,
extended defects can form upon solid-phase recrystallization as the interface begins
to propagate toward the surface. In addition, regrowth-related defects may arise in
patterned structures such as at a gate edge. These defects arise when the lateral and vertical
regrowth velocities are such that a pinch point forms where the two interfaces meet.
This also requires the depth of the implant-induced amorphous layer to typically be over
700–800 Å thick [70].
Figure 1.15 shows an image of the extended defects that can form near the mask edge
during regrowth. This defect is a half loop dislocation that propagates up to the surface.
Research has shown that this defect occurs because of the way the amorphous layer is
shaped during regrowth. Figure 1.16a shows a regrowth condition that leads to these
mask edge defects. Figure 1.16b clearly shows the bottom corners beginning to pinch
100 nm
FIGURE 1.15
Half loop dislocation.
50 nm
(b)
(a)
FIGURE 1.16
The amorphous layer in cross section during regrowth. Both had 40 keV 10
15
Si
þ
implants, and were annealed at
7508C for 13 min. On the left there is no nitride pad and the pinching of the regrowth in the corner is evident. The
right has a nitride pad and avoids pinching.
ß 2008 by Taylor & Francis Group, LLC.
off [70]. Th e defe ct forms along the pinch when the two crystal line surfa ces are not
perfec tly aligned atomicall y. Chan ging the st ress in the wafer can change thi s behavior.
Upo n the introduct ion of ten sile stress from either a thin film such as a nitride layer [71]
or throu gh bending the wafe r [72], the solid-p hase regr owth velocit ies can be altere d.
Figu re 1.16 shows the elim ination of the pinch point when the wafer has a nit ride implant
mas k during anneal . Figure 1.17 sh ows the top view of thi s proce ss. Defects lie around the
mas k and can provi de fau lts in subs equent circuits .
1. 7 Conclusions
Unde rstandi ng defect behavior is critica l fo r optimi zing shallo w junctions in highly sca led
com plemen tary metal- oxide semico nductor (CMOS ) manufact uring. Even though the
impl ant energie s have been signi ficantly reduce d, which reduce s the to tal dam age depos-
ited, extended defects still form. The surfa ce is not as strong a sink as it was once believe d
to be. Defects still form and control junct ion charac teristi cs.
Advanc es in anneal ing technol ogies that allow millis econd anne als were also thou ght to
rend er the defects unimp ortant. Altho ugh these junctions freq uently show little or no
diffus ion, the defects can control and determi ne the activati on and thus the resistance of
the junction. Mo dels for defect beha vior in silico n in thi s tim e regime are still fairly
primi tive and work rem ains to be done.
For scaled devices, extended defects c an be the size of the active area of the devi ce and
bridge the channel. Understan ding ho w defects fo rm around mask edge s is crit ical for
reli ability predic tion. Mask edge defe cts can bridge devi ces and provid e a leakag e path
shor ting the device.
There are many issues that remain to be resolved for shallow junctions in highly scaled
CMOS manufacturing technologies. Defects can still control diffusion and activation
processes; this does not change as the layers get shallow. These defects continue to
be important yield, performance, and reliability issues for microelectronic materials
and devices.
100 nm
FIGURE 1.17
Top view of the mask edge defects. On the left is a case with pinching, and on the right it is prevented with the
mask. (From Olson, C.R. et al., J. Vac. Sci. Technol. B, 24, 446, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
Re fer enc es
1. Semiconductor Industry Association, International Technology Roadmap for Semiconductors . Avail-
able at: http: ==www.itrs.net =reports.html, 2006 edition.
2. Thompson, S.E. et al., A logic nantotechnology featuring strained-silicon, IEEE Electron Dev. Lett.,
25, 191, 2004.
3. Doris, B. et al., Extreme scaling with ultra-thin silicon channel MOSFET’s (XFET), IEEE IEDM,
San Francisco, CA, 2002, p. 267.
4. Wong, H.S.P., Frank, D.J., and Solomon, P.M., Device design considerations for double-gate,
ground-plane, and single-gated ultra-thin SOI MOSFET’s at the 25 nmchannel length generation,
IEEE IEDM, San Francisco, CA, 1998, p. 407.
5. Hergenrother, J.M. et al., The vertical replacement-gate (VRG) MOSFET: A 50-nm vertical
MOSFET with lithography-independent gate length, IEEE IEDM, Washington, DC, 1999, p. 75.
6. Giles, M.D., Transient phosphorus diffusion below the amorphization threshold, J. Electrochem.
Soc., 138, 1160, 1991.
7. Camm, D.M. and Lojek, B., High power arc-lamp RTP system for high temperature annealing
application, RTP Proc., Raleigh-Durham, NC, 1994, p. 259.
8. Stuart, G.C. et al., Temperature diagnostics for a dual-arc FRTP tool, IEEE Intl. Conf. on ATP of
Semi., San Francisco, CA, 2002, p. 77.
9. Cowern, N.E.B. et al., Energetics of self-interstitial clusters in silicon, Phys. Rev. Lett., 82, 4460,
1999.
10. Tan, T.Y., Atomic modeling of homogeneous nucleation of dislocations from condensation of
point defects in silicon, Phil. Mag. A, 44, 101, 1981.
11. Takeda, S. et al., Interstitial defects on {113} in Si and Ge line defect configuration incorporated
with a self-interstitial atom chain, Phil. Mag. A, 70, 287, 1994.
12. Kim, J. et al., Extended Si {311} defects, Phys. Rev. B, 55, 16186, 1997.
13. Pelaz, L. et al., B diffusion and clustering in ion implanted Si: The role of B cluster precursors,
Appl. Phys. Lett., 70, 2285, 1997.
14. Lilak, A.D. et al., A physics based modeling approach for the simulation of anomalous boron
diffusion and clustering behaviors, IEEE IEDM, Washington, DC, 1997, p. 493.
15. Radic, L., Lilak, A.D., and Law, M.E., Dependence of boron cluster dissolution on the annealing
ambient, Appl. Phys. Lett., 81, 826, 2002.
16. Michel, A.E., Rausch, W., and Ronsheim, P.A., Implantation damage and the anomalous transient
diffusion of ion implanted boron, Appl. Phys. Lett., 51, 487, 1987.
17. Brindos, R., Keys, P., and Jones, K.S., Effect of arsenic doping on {311} defect dissolution in silicon,
Appl. Phys. Lett., 75, 229, 1999.
18. Fair, R.B. and Tsai, J.C.C., Profile parameters of implanted-diffused arsenic layers in silicon,
J. Electrochem. Soc., 123, 583, 1977.
19. Luning, S. et al., Kinetics of high concentration arsenic deactivation at moderate to low temper-
atures, IEEE Trans. Electron Dev., 17, 457, 1992.
20. Dokumaci, O. et al., Transmission electron microscopy analysis of heavily doped As-doped, laser,
and thermally annealed layers in silicon, J. Appl. Phys., 78, 828, 1995.
21. Nobili, D. et al., Precipitation, aggregation, and diffusion in heavily arsenic-doped silicon, Phys.
Rev. B, 49, 2477, 1994.
22. Nobili, D. et al., Precipitation as the phenomena responsible for the electrically inactive arsenic in
silicon, J. Electrochem. Soc., 130, 992, 1983.
23. Scholz, R. et al., Carbon-induced undersaturation of silicon and self-interstitials, Appl. Phys. Lett.,
72, 200, 1998.
24. Adam, L.S. et al., Experimental identification of nitrogen-vacancy complexes in nitrogen
implanted silicon, Appl. Phys. Lett., 79, 623, 2001.
25. Simpson, P.J. et al., Observation of fluorine-vacancy clusters in silicon, Appl. Phys. Lett., 85, 1538,
2004.
26. Bourret, A. et al., Defect structure in CZ silicon and germanium studied by high-resolution
electron microscopy, Microscopy of Semiconductor Materials, Oxford, UK, 1981, p. 210.
ß 2008 by Taylor & Francis Group, LLC.
27. Davidson, S.M., Study of radiation damage in silicon using scanning electron microscopy, Nature,
227, 487, 1970.
28. Eaglesham, D.J. et al., Implant damage and transient enhanced diffusion in Si, Nucl. Instrum.
Meth. B, 106, 191, 1995.
29. Liu, J. et al., The effect of boron implant energy on transient enhanced diffusion in silicon, J. Appl.
Phys., 81, 1656, 1997.
30. Eaglesham, D.J. et al., Implantation and transient B diffusion in Si: The source of the interstitials,
Appl. Phys. Lett., 65, 2305, 1994.
31. Pan, G.Z. et al., Microstructural evolution of {113} rod-like defects and {111} dislocation loops in
silicon-implanted silicon, Appl. Phys. Lett., 71, 659, 1997.
32. Ferreira Lima, C.A. and Howie, A., Defects in electron-irradiated germanium, Phil. Mag. A, 34,
1057, 1976.
33. Salisbury, I.G. and Loretto, M.H., {113} Loops in electron-irradiated silicon, Phil. Mag. A, 39, 317,
1979.
34. Jones, K.S. et al., A systematic analysis of defects in ion implanted silicon, Appl. Phys. A, 45, 1,
1988.
35. Li, J.-H. and Jones, K.S., {311} defects in silicon: The source of the loops, Appl. Phys. Lett., 73, 3748,
1998.
36. Law, M.E. and Jones, K.S., A new model for {311} defects based on in-situ measurements,
International Electron Device Meeting, San Francisco, CA, 2000, p. 511.
37. Schreutelkamp, R.J. et al., Pre-amorphization damage in ion-implanted silicon, Mater. Sci. Rep., 6,
275, 1991.
38. Laanab, L. et al., Variation of end of range density with ion beam energy and the predictions of
the excess interstitials model, Nucl. Instrum. Meth. B, 96, 236, 1995.
39. Cho, N.-H. et al., MeV ion induced damages and their annealing behavior in silicon, Mater. Res.
Soc. Symp., San Francisco, CA, 1996, p. 430.
40. Jones, K.S. and Gyulai, J., Annealing of implantation damage in silicon, Ion Implantation Science
and Technology, Ziegler, J.F. (ed.), Ion Implantation Technology Co., Yorktown, 1996.
41. Jones, K.S. et al., Enhanced elimination of implantation damage upon exceeding the solid
solubility, J. Appl. Phys., 62, 4114, 1987.
42. Fair, R.B., Damage removal=dopant diffusion tradeoffs in ultra-shallow implanted p
þ
–n junc-
tions, IEEE Trans. Electron Dev., 37, 2237, 1990.
43. Michel, A.E., Anomalous transient diffusion of ion implanted dopants: A phenomenological
model, Nucl. Instum. Meth. B, 37=38, 379, 1989.
44. Listebarger, J.K. et al., Use of type II (end of range) damage as detectors for quantifying interstitial
fluxes in ion-implanted silicon, J. Appl. Phys., 73, 4815, 1993.
45. Liu, J. et al., Evolution of dislocation loops in silicon in an inert ambient (I), Solid State Electron., 38,
1305, 1995.
46. Laanab, L. et al., A model to explain the variations of end-of-range defect densities with ion
implantation parameters, Mater. Res. Soc. Symp., 1993, p. 302.
47. Bonafos, C.B. et al., TED of boron in presence of EOR defects: The role of the evolution of
Si self-interstitial supersaturation between the loops, Intl. Conf. Ion Imp. Tech., Austin, TX,
1997, p. 1.
48. Claverie, A. et al., Interactions between dopants and end of range defects in silicon, Solid State
Phen., 47–48, 195, 1996.
49. Bonafos, C. et al., The effect of the boron doping level on the thermal behavior of end-of-range
defects in silicon, Appl. Phys. Lett., 71, 365, 1997.
50. Listebarger, J.K. et al., Study of end of range loop interactions with Bþ implant damage using a
boron doped diffusion layer, J. Appl. Phys., 78, 2298, 1995.
51. Ajmera, A.C. and Rozgonyi, G.A., Elimination of end-of-range and mask edge lateral damage in
Ge
þ
preamorphized, B
þ
implanted Si, Appl. Phys. Lett., 49, 1269, 1986.
52. Ajmera, A.C. et al., Point defect=dopant diffusion considerations following preamorphization of
silicon via Si
þ
and Ge
þ
implantation, Appl. Phys. Lett., 52, 813, 1988.
53. Prussin, S. and Jones, K.S., Role of ion mass, implant dose, and wafer temperature on end-
of-range defects, J. Electrochem. Soc., 137, 1912, 1990.
ß 2008 by Taylor & Francis Group, LLC.
54. Ganin, E. and Marwick, A., Is the end-of-range loops kinetics affected by surface proximity or ion
beam recoils distribution? Mater. Res. Soc. Symp., San Diego, CA, 1989, p. 147.
55. Jones, K.S. and Rozgonyi, G.A., Extended defects from ion implantation and annealing,
Rapid Thermal Processing: Science and Technology, Fair, R.B. (ed.), Academic Press, Orlando, FL,
1993, p. 123.
56. Buck, T.M. et al., Gettering rates of various fast-diffusing metal impurities at ion-damaged layers
on silicon, Appl. Phys. Lett., 21, 485, 1972.
57. Landi, E. and Solmi, S., Electrical characterization of pþ=n shallow junctions obtained by boron
implantation into preamorphized silicon, Solid. State Electron., 29, 1181, 1986.
58. Stolk, P.A. et al., Physical mechanisms of transient enhanced dopant diffusion in ion-implanted
silicon, J. Appl. Phys., 81, 6031, 1997.
59. Seidel, T.E. et al., A reviewof rapid thermal annealing (RTA) of B, BF
2
and As ions implanted into
silicon, Nucl. Instrum. Meth. B, 7=8, 251, 1985.
60. King, A.C. et al., Defect evolution of low energy, amorphizing germanium implants in silicon,
J. Appl. Phys., 93, 2449, 2003.
61. Camillo-Castillo, R.A. et al., Kinetics of the end-of-range damage dissolution in flash-assist rapid
thermal processing, Appl. Phys. Lett., 88, 232104, 2006.
62. Keys, P., Phosphorus-defect interactions during thermal annealing of ion implanted silicon, PhD
dissertation, University of Florida, Gainesville, FL, 2001.
63. Chichkine, M.P., De Souza, M.M., and Sankara Narayanan, E.M., Growth of precursors in silicon
using pseudopotential calculations, Phys. Rev. Lett., 88, 085501, 2002.
64. Gilmer, G.H. et al., Diffusion and interactions of point defects in silicon: Molecular dynamics
simulations, Nucl. Instrum. Meth. B, 102, 247, 1995.
65. Chichkine, M.P., De Souza, M.M., and Sankara Narayanan, E.M., Self-interstitial clusters in
silicon, Mater. Res. Soc. Symp., San Francisco, CA, 2000, p. 610.
66. Claverie, A. et al., Modeling of the Ostwald ripening of extrinsic defects and transient enhanced
diffusion in silicon, Nucl. Instrum. Meth. B, 186, 281, 2002.
67. Robertson, L.S. et al., Annealing kinetics of {311} defects and dislocation loops in the end-of-range
damage region of ion implanted silicon, J. Appl. Phys., 87, 2910, 2000.
68. Law, M.E. and Cea, S.M., Continuum based modeling of silicon integrated circuit processing: An
object oriented approach, Comput. Mater. Sci., 12, 289, 1998.
69. Sands, T. et al., Influence of the amorphous-crystalline interface morphology on dislocation
nucleation in pre-amorphized silicon, XIII Intl. Conf. on Defects in Semiconductors, The Metallur-
gical Society, Los Angeles, CA, 1984, p. 531.
70. Olson, C.R. et al., Effect of stress on the evolution of mask-edge defects in ion-implanted silicon,
J. Vac. Sci. Technol. B, 24, 446, 2006.
71. Ross, C. and Jones, K.S., The role of stress on the shape of the amorphous–crystalline interface and
mask-edge defect formation in ion-implanted silicon, Mater. Res. Soc. Symp., San Francisco, CA,
2004, p. 810.
72. Rudawski, N.G. et al., Effect of uniaxial stress on solid phase epitaxial regrowth and mask edge
defect formation in two-dimensional amorphized Si, Mater. Res. Soc. Symp., San Francisco CA,
2006, p. 912.
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
2
Hydrogen-Related Defects in Silicon, Germanium,
and Silicon–Germanium Alloys
A.R. Peaker, V.P. Markevich, and L. Dobaczewski
CONTENTS
2.1 Introduction........................................................................................................................ 27
2.2 Incorporation, Diffusion, and Solubility of Hydrogen................................................. 29
2.2.1 Incorporation ........................................................................................................... 29
2.2.2 Diffusion................................................................................................................... 30
2.2.3 Solubility .................................................................................................................. 33
2.3 Isolated Hydrogen............................................................................................................. 34
2.4 Passivation .......................................................................................................................... 37
2.5 Electrically Active Hydrogen Complexes ...................................................................... 38
2.5.1 Vacancy–Hydrogen Complexes............................................................................ 38
2.5.2 VOH.......................................................................................................................... 39
2.5.3 CH............................................................................................................................. 40
2.5.4 Transition Metal–Hydrogen Complexes ............................................................. 41
2.5.5 Au–H
n
....................................................................................................................... 41
2.5.6 Pt–H
2
......................................................................................................................... 42
2.6 Hydrogen Molecules ......................................................................................................... 43
2.7 Hydrogen Clusters and Platelets..................................................................................... 44
2.8 Hydrogen in Germanium................................................................................................. 44
2.8.1 Hydrogen Incorporation, Diffusivity, and Solubility in Ge ............................. 45
2.8.2 Isolated Hydrogen Species and Complexes with Intrinsic Defects in Ge...... 45
2.8.3 Hydrogen Passivation of Deep Centers and Activation
of Neutral Impurities in Ge................................................................................... 46
2.8.4 Device-Related Aspects of Hydrogen in Ge ....................................................... 46
2.9 Hydrogen in Silicon–Germanium Alloys....................................................................... 47
2.10 Summary............................................................................................................................. 49
References...................................................................................................................................... 49
2.1 Introduction
Hydrogen is a very common impurity in semiconductors and has been investigated in great
detail in silicon. There is also a small amount of work on hydrogen in silicon–germanium
and germanium. Czochralski-grown ingots of silicon and germanium contain almost no
ß 2008 by Taylor & Francis Group, LLC.
hydrogen. However, subsequent processing can introduce considerable concentrations in the
near surface regions and in some cases deep in the bulk of the materials. How hydrogen is
incorporated into silicon and germanium is discussed in detail in Section 2.2, so here it is
sufficient to say that processes such as chemical etching, polishing, and plasma processing
release considerable quantities of atomic hydrogen, which is free to diffuse and react with the
semiconductor. In some technological processes hydrogen is deliberately incorporated, a
classic case is where silicon metal-oxide semiconductor (MOS) devices are heated in forming
gas at about 4508C to passivate interface traps or surface states, a process that was patented
by Fowler in 1974 [1]. This passivation process is discussed in Chapter 7 of this book, but
quite obviously there is the potential for hydrogen to diffuse into the semiconductor itself.
The incorporat ion of hydrog en in silicon and germ anium is a rathe r com plex proces s
that depends on the form of the hydro gen and the dopin g and impurit y content of the
semi conduct or. Hydrogen mo lecule s diffu se in silicon, but more impo rtantly, so do atom ic
defects of hyd rogen . Th ese can exis t in pos itively or nega tively charge d state s or, excep-
tiona lly, in neutra l state s. Th e charge d single atomi c species are ext remely reactive and are
com monly observed to passiv ate shallo w donor s and acceptors as we ll as deep-le vel
defects . Muc h work has been devo ted to studyi ng the reaction prod ucts of various defect
spe cies with hydro gen. Th is rea ctivity makes the diffus ion beha vior of atomi c hydro gen at
lower temperat ures some what comple x; a key issue is its trapping at vari ous impu rity
sites . The diffus ion depends on the detai l of the silicon dopin g, the saturati on of captur e
sites , and the subsequ ent thermal releas e of the hyd rogen from these trapping sites. In
cons equenc e, the solubi lity and diffusion of hydrog en in high ly doped silico n is dramat-
ically different than that of its behavior in undop ed mate rial.
Iso lated hydro gen atoms are known to exist in silico n, silico n –germani um, and germ a-
nium at low tem peratures ; vari ous tech niques have bee n used to determi ne their precis e
posit ion in the semi conduct or lattice. Similarl y detailed work has been conduct ed on
hyd rogen pass ivation and on electri cally active hydro gen com plexes. Althou gh this chap-
ter is about hydro gen, it is impo rtant to say that deuter ium is of ten used inde pende ntly
or in combin ation in studies of hyd rogen-re lated defects . Th e rea sons for this are that
(1) seco ndary ion mass spectro metry (SIMS) has a muc h high er sens itivity to deu terium
than hydrog en and (2) that some techni ques such as local mode spec troscopy can
disti nguish betwe en hy drogen and deuter ium, there by makin g the ide ntities of com plexes
(espec ially those con taining mo re than one hydrogen atom) easi er to resol ve.
In general, the total concentrat ion of hydrog en in silicon determi ned by SIMS is much
high er than the concen tration of hyd rogen and hydrogen com plexes dete rmine d the ir
elect rical or optical measu remen ts. Comb inations of the oretical studies and optical work
have rev ealed that this is due to the pres ence of hyd rogen mo lecules. Th ese are neutr al and,
in their usu al siting withi n the silicon lattice , are largel y elect rically and optical ly inactiv e.
Howe ver, much work has now been done to resol ve the weak opti cal activit y of molecu lar
spe cies usi ng Raman studi es and optic al absorp tion tech niques. In rec ent year s, this work
has been given a particular stimulus by the developm ent of the Smar tCut process for
produ cing silicon on insulato r wafers [2]. In this techniq ue, lig ht ions (predomi nantly
hyd rogen) are implante d a t high concen trations to prod uce gas bubbl es jus t below the
surfa ce. Subsequ ent hea t treating can then be use d to defoliat e a thin uniform laye r. This
behavior is now used extensively to produce silicon on insulator wafers for high perform-
ance, extremely scaled complementary MOS (CMOS), and indeed for many other applica-
tions. The technique can be applied to silicon, silicon–germanium, and germanium, and is
now of considerable commercial importance. All these issues are discussed in detail in the
subsequent sections of this chapter, but attention should be drawn to a number of previous
reviews of hydrogen in silicon and related materials and also to conference proceedings
focused specifically to hydrogen in silicon [3–8].
ß 2008 by Taylor & Francis Group, LLC.
2.2 I ncorporation, Diffusion, and Solubili ty of Hydrogen
2.2.1 Incorporat ion
Hydro gen can be inc orporat ed into silicon in molecu lar or atom ic form. In the latter case, it
can exis t in the pos itive, nega tive, or met astable neutral charge state s. Hydrogen in its
atomi c form is very rea ctive a nd can combin e with impu rities in the silicon. As a conse-
quence , the beha vior of atomi c hyd rogen is strongly influence d, at leas t in the low-
temperat ure regime (less than $200 8 C), by the pres ence of impuritie s. Al l the abov e factors
affect the incor poration, sol ubility, and diffusiv ity at differe nt temperat ures. He nce, the
actual value s for inc orporat ion rate , diffusion , and sol ubility dep end on the expe rimental
cond itions used. So, a very wide sca tter exists in the experi mental value s in the literature .
Howev er, it is eviden t that hyd rogen has a very high intrin sic diffusiv ity and a low
equilibr ium solubility.
The mo st fun damental mo de of incorporat ion is diffu sion at high tem peratures
( $ 1000 8C) in an atmo sphere of hydro gen gas. Sever al publi cations base solubility and
diffusion studies on incorporat ion in this way. Howeve r, the detecti vity of hydrogen in
SIMS is rather low due to the high ba ckgroun d levels of hyd rogen in mo st SIM S machines .
Thus, deuterium has been use d as a hydro gen analog in several st udies. Th e improve ment
in detecti vity is machi ne speci fic, but more than two orders of magnitu de impro vement is
state d in some publicati ons result ing in densi ties [D] ! 5 Â 10
13
cm
À3
being quan tifiab le.
An importan t issue in relatio n to inc orporat ion into silico n from hyd rogen or deu terium
gas is that the proce ss does not dep end line arly on the par tial press ure p of the gas. The
experi ments in 1956 of Van Wieri ngen and Warmol tz [9] measure d the transpo rt of
hydro gen through a thin-w alled silico n cylind er and foun d the st eady-stat e permeati on
at 1 050 8C to be a function of p
0.5
, and at 1200 8 C, p
0.54
. If diffus ion is the rate-limiti ng step in
these expe riments , the mo st likely expl anatio n is that the hy drogen diffuses pred ominantl y
as atoms rather than molecu les. The dissociati on energy of a hydrogen mo lecule is $4 eV,
but catalyti c dissociati on of hydrogen at temperat ures 1050 8 C is well kno wn and prob-
ably can occur at the silico n surfa ce. Subsequ ent measure ments of rap id-quen ched sam ples
by Newman et al. foun d the appa rent sol ubility of bot h hyd rogen and deuter ium v aried as
p
0.54
at 1300 8 C [10]. The conclusion was also drawn that incor poration and diffusion occur
in ato mic form, alth ough molecu les were shown to be presen t after quench ing and
assume d to be due to rec ombina tion of the atomi c species .
Anoth er ver y widel y use d met hod of hydro gen and deuter ium incor poration is from a
discharge produced in the gas at a pressure of $2 Torr by either a DC current or radio
frequency plasma. Often the sample is remote from the discharge to minimize damage due
to energetic ions. Atomic species reach the sample by drift in the gas flow. The atomic
species often recombine before reaching the sample, a process enhanced by the proximity
of the surface of the discharge tube. As a result, the concentration of atomic hydrogen at the
sample surface is difficult to control and quantify. However, it is a very effective method
for low- temperat ure inc orporat ion, as sh own in Figu re 2.1, where the SIM S-dete rmined
deuterium profile is shown after exposure to a remote 70 W plasma for 30 min with the
sample held at 1508C [11]. After deuteration, the samples were rapidly cooled to room
temperature (RT). The figure shows high concentration ($10
20
cm
À3
) incorporation at
the surface, while deeper into the silicon the concentration of deuterium is similar to the
dopant concentration for the case of boron but not in the case of phosphorus. This rather
complex behavior will be discussed later in this chapter.
Ion-beam sources are also used to incorporate hydrogen as H
þ
, although of course
considerable damage is introduced into the silicon lattice by these protons. These tech-
niques have been reviewed in detail by Seager [12], and more recently have been the basis
ß 2008 by Taylor & Francis Group, LLC.
of the commercially important SmartCut process [2]. These are processes where hydrogen
is introduced intentionally; hydrogen incorporation into silicon also occurs as an uninten-
tional consequence of plasma etching and related processes.
It is also possible to introduce hydrogen into silicon electrochemically at lowtemperature.
This has been reported for hydrogen fluoride and oxidizing etches [13], and even water
[14]. Essentially any process that produces atomic hydrogen at the surface of the silicon
provides species that can diffuse rapidly into the semiconductor and react with dopant or
other impurity species. Among the many technologically important processes in which
hydrogen incorporation occurs are chemical mechanical polishing (CMP) of silicon and
the fabrication of microelectromechanical systems (MEMS) structures by chemical or plasma
etching.
2.2.2 Diffusion
Diffusion falls into two clearly distinct regimes. The first is high-temperature behavior
in which the hydrogen migrates predominantly as atomic species via an interstitial
mechanism. The second is a complex low-temperature behavior that again reflects the
motion of atomic species, but where hydrogen binds to impurities in the silicon, forming
stable or metastable defects. As a result, the hydrogen transports through the lattice via a
hopping mechanism. Under these circumstances the effective diffusivity depends not only
on the temperature, but also on the concentration of the impurities present, the concentra-
tion of the hydrogen, and in some cases the position of the Fermi level. The key issues are
presented below; for more detail, the reader is referred to a comprehensive review of the
0
10
15
10
16
10
17
10
18
10
19
10
20
10
21
0.80
[B] = 5ϫ10
18
cm
Ϫ3
[P] = 1ϫ10
17
cm
Ϫ3
[B] = 2ϫ10
17
cm
Ϫ3
1.20 1.80
Depth (mm)
D
e
u
t
e
r
i
u
m

c
o
n
c
e
n
t
r
a
t
i
o
n

(
a
t
o
m
s

c
m
Ϫ
3
)
2.40 3.00
FIGURE 2.1
Deuterium depth profile in Czochralski-grown Si after plasma treatment at 1508C for 30 min using three different
slices of material doped with (a) 5 Â10
18
boron, (b) 2 Â10
17
boron, and (c) 10
17
phosphorus. (From Johnson, N.M.
and Moyer, M.D., Appl. Phys. Lett., 46, 787, 1985. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
early work on diffusion and solubi lity of hydro gen in crystal line silicon [15], publis hed by
Stav ola in 1999.
Van Wieringe n and Wa rmoltz ’ s measu remen t [9] of the permeati on of hy drogen
throug h a thin-walle d silico n cylinde r provides a ver y reliab le direct measure ment of
diffusiv ity over the temperat ure range 1090 8 C –1200 8C. The relati onship they derive d is
D( T ) ¼ 9: 4 Â 10
À 3
exp ( À 0: 48 =kT ) [cm
2
s
À 1
]
where kT is express ed in eV .
These value s and the ir extrapo lation repre sent the up per limit of all publi shed expe ri-
mental data. This is understan dable in the conte xt of the experi mental con ditions in
which the temperat ure is suf fi ciently high to render trapping process es in the transp ort
insigni ficant.
The minimum energy posit ion for H
þ
is a bond-cen ter (BC) site locate d midw ay between
two firs t neighbor Si a toms. To diffus e, the hydrog en must jump betwe en (BC) sites .
Howev er, the energie s of other int erstitial sites are suf ficie ntly close so as to make it
unlik ely that there will be a uniqu e path for hydro gen diffus ion. In cons equen ce, static
total energy calcu lations could be unreli able. To overco me this lim itation, Bud a et al. [16]
have underta ken ab initio mo lecular dyn amic s simulation of high-temp eratur e diffusion of
H in Si (1200– 1800 K) and obt ained diffusiv ity values that are very similar to the expe ri-
mental values of Van Wie ringen and Warmol tz [9], as do Blo chl et al. using fi rst-pri nciples
calcul ations based on rate -theory formalisms [17]. More rec ently, Panzari ni and Colombo
have bee n ab le to extend the molecu lar dyna mics met hod to lowe r tem perature s. They
calcul ated the diffus ivity over the range 800 –1800 K [18], and obtain excellent ag reement
with Wieringe n and Warmol tz ’s expe rimen ts over the range cov ered by the expe rimen ts,
but predict lower diffu sivities than the ext rapolatio n would indica te becaus e of changes in
the jump length. It is important to point out that this is predicted to occur at temperatures
above that at which trapping=detrapping would be expected to be important. No experi-
mental data are available at the lower temperatures that these molecular dynamic calcu-
lations cover, but there is a trend toward the many experimental results that fall below the
extrapolation of the Wieringen and Warmoltz data.
The above trends can be seen in Figu re 2.2, which presents a com parison of diffus ivity
results from the literature. The line is an extrapolation of Wieringen and Warmoltz’s work
[9]. The Panzarini and Colombo results [18] can be seen as tending toward the main
grouping of experimental results at slightly lower temperatures. However, it is interesting
to note that two sets of low-temperature measurements [19,20] fit very well to the Wieringen
and Warmoltz extrapolation. The data of Gorelkinskii and Nevinnyi [19] were taken
over the range 126–143 K and measures the reorientation of a single jump of H
þ
, so that
the retrapping considerations do not enter into the process. Because of this, the mechan-
isms of the high-temperature measurement are mimicked in this low-temperature
electron paramagnetic resonance (EPR) work. A similar argument can be applied to the
data of Kamiura et al. [20], who released hydrogen from CH pairs and observed its
recapture at nearby P. They undertook this experiment for D as well as H and, as expected,
observed a slightly lower diffusivity for D. Other work mostly reports rather lower
diffusivities than the extrapolation at lower temperatures. Mogro-Campero et al. [21]
calculated the diffusivity from the penetration depth of hydrogen derived from passivation
studies under conditions that were likely to include both the effects of trapping and
detrapping. Newman [22] used the interaction of H with O to determine the diffusivity
by observing the relaxation of stress-induced dichroism of the 9 mm oxygen infrared (IR)
absorption band. The measurements enabled the hydrogen diffusion coefficient to be
estimated as 1.7 Â10
2
exp(À1.2 eV=kT)cm
2
s
À1
for temperatures in the range 2258C–3508C.
ß 2008 by Taylor & Francis Group, LLC.
At low tem perature s where trapping is importan t, muc h expe rimen tal and modeling
work has been done. Zund el and Weber [23] have st udied the in fluence of bor on con cen-
tratio n on the diffus ivity over the range 608 C –140 8C on the hydro gen profi le by obse rving
the passiv ation of the bor on. Th ey find that the eff ective hydrog en diffu sion coef ficient
shows no dep endence on the diffus ivity of free hydro gen and is entirely trap -limite d.
Figu re 2.3 show s an Arrhe nius plot of the produ ct of the effect ive diffusiv ity and acceptor
concen tration as a functi on of temperat ure, indicati ng that the effect ive diffus ivity
is redu ced as the bor on concen tration increases . The measu remen ts discuss ed above
probabl y relate to H
þ
, or poss ibly in the cas e of W ieringen and Warmol tz, H
0
(althou gh
a widel y held view is that H
0
is met astable due to the negativ e-U charac ter of H in Si).
Johnso n and He rring [24] have repo rted measure ments on D
À
at low tem perature s by
dissoci ating P –D comple xes; these are shown in Figure 2.2. These diffusiv ities are con sider-
ably lower than the Wieri ngen and Warmol tz extrapo lation (only a reducti on of
ffiffiffi
2
p
fo r D
com pared to H wou ld be expected from simp le diffusion theo ry), but are higher than mo st
H
þ
resu lts at tem peratures where trap ping and hoping diffusion would be expe cted to
dominat e. In selecti ng data for Figu re 2.2, we have omitt ed H
þ
result s that combin e the
effect s of dr ift with diffusion , i.e., result s in whic h the motion of H
þ
has been observe d under
the action of an el ectric field. At RT the eff ects of drift can be ver y subs tantial compar ed to
diffus ion alone .
1 2 3 4 5 6 7 8
1000/T (K
Ϫ1
)
10
Ϫ22
10
Ϫ20
10
Ϫ18
10
Ϫ16
10
Ϫ14
10
Ϫ12
10
Ϫ10
10
Ϫ8
10
Ϫ6
10
Ϫ4
10
Ϫ2
D
i
f
f
u
s
i
v
i
t
y

(
c
m
2

s
Ϫ
1
)
Experiment
Van Wieringen and Warmoltz [9]
Gorelkinskii and Nevinnyi [19]
Johnson and Herring [24]
Newman et al. [22]
Zundel and Weber [23]
Kamiura et al. [20]
Mogro-Campero et al. [21]
Theory
Panzarini and Colombo [18]
Buda et al. [16]
Van Wieringen extrapolation
Ϫ100°C 0°C 500°C 100°C
FIGURE 2.2
Compilation of diffusivity as a function of temperature from various publications. The solid line is an extrapol-
ation of the data of Van Wieringen and Warmoltz [9]. Theoretical calculations and experimental data are as
referenced in the figure.
ß 2008 by Taylor & Francis Group, LLC.
2.2.3 Solubility
The equilibrium solubility of hydrogen in silicon has been determined over the range
9008C–13008C. Four sets of experimental results determined by different methods are
shown in Figure 2.3. The early work of Van Wieringen and Warmoltz [9] described earlier
measured concentration of molecules and so is scaled by a factor of 2 in Figure 2.4 to enable
2.4
140 120
Temperature (°C)
100 80 60
10
0
10
1
10
2
10
3
10
4
2.5 2.6 2.7
10
3
/T [K
Ϫ1
]
D
e
f
f

N
A

(
c
m
Ϫ
1

s
Ϫ
1
)
2.8 2.9 3.0
N
A
= 3.8ϫ10
16
cm
Ϫ3
N
A
= 1.4ϫ10
15
cm
Ϫ3
FIGURE 2.3
Low-temperature diffusivity showing
dependence on acceptor concentra-
tion. (From Zundel, T. and Weber, J.,
Phys. Rev. B, 46, 2071, 1992. With
permission.)
6.5 7 7.5 8 8.5
10,000/T [K
Ϫ1
]
10
14
10
15
10
16
S
o
l
u
b
i
l
i
t
y

(
c
m
Ϫ
3
)
Van Wieringen and Warmoltz [9]
Ichimiya and Furuichi [25] (tritium)
Binns et al. [26] from SIMS (deuterium)
Binns et al. [26] from BH absorption
1200°C
1000°C
FIGURE 2.4
High-temperature solubility of hydrogen, deuterium, and tritium in silicon.
ß 2008 by Taylor & Francis Group, LLC.
direct compar ison with the othe r data that relate to atomic concentrat ions. Ichi miya and
Furui chi [25] measu red the concen tration of tri tium atoms usi ng the emitte d beta radiation
as a measu re of concen tration. Binns et al. [26] hea ted high -resisti vity silico n in D
2
and
quench ed to RT, measu ring the deuter ium concentrat ion by SIMS. Al so shown in Figu re
2.3 are data derive d by the same authors from optical absorptio n measu remen ts associate d
with the BH or BD compl ex. In thi s cas e, the deu terium SIMS data were used as a
calibrat ion and extended to hyd rogen. Experime ntal con ditions were establish ed such
that a ll the hyd rogen wou ld be expecte d to form BH pairs.
The hyd rogen and deuteriu m concen trations shown in Figu re 2.1 after 1508C pl asma
incorporation are many orders greater than the equilibrium values at high temperature
reported above. As will be discussed later the high surface concentration is attributed to
hydrogen molecules, while the saturated values at the plateaux equal to the boron concen-
tration are due to metastable boron complexes.
2.3 Isolated Hydrogen
The identification and lattice location of isolated hydrogen in silicon has been a much
bigger challenge for experimentalists than for theorists. Initially, a point of particular
interest has been the diffusion of hydrogen at the atomic level. As discussed above,
hydrogen migrates, according to the present theoretical consensus [6], as isolated ionic
species through interstitial sites of the silicon crystal. The potential-energy surfaces govern-
ing this migration have minima at the interstitial BC site, and at the interstitial tetrahedral
site (T), giving rise to a donor level and an acceptor level, respectively. The pathway of
migration is expected to depend on the position of these levels relative to the Fermi level.
Ab initio calculations [27] predicted that the neutral BC hydrogen, H
0
(BC), in an n-type
material represents a local minimum in free energy. However, this minimum energy
configuration is metastable inasmuch as either positive BC hydrogen, H
þ
(BC), or negative
T-site hydrogen, H
À
(T), represents the global energy minimum for any position of the
Fermi level. This implies that the donor level lies above the acceptor level, with the
consequence that monatomic hydrogen should behave as a negative-U defect in silicon
when the H
0
impurity can jump swiftly between BC and T sites.
The high reactivity and RT mobility of interstitial hydrogen makes it very difficult, or
even impossible, to observe the H
þ
(BC), H
0
(BC), and H
À
(T) configurations of hydrogen
directly. There is much experimental work where this fact has been ignored and, as a result,
the hydrogen–oxygen and hydrogen–carbon complexes have been erroneously interpreted
as the isolated hydrogen. One way to avoid this problem is to introduce hydrogen by
proton implantation at cryogenic temperatures where hydrogen and intrinsic defects are
immobile. In such an approach, the structure of isolated hydrogen in silicon may be
studied by in situ applications of standard spectroscopic techniques. The first report
where this approach has been used presented the results of the local vibrational mode
(LVM) absorption studies of the low-temperature proton-implanted silicon [28]. The
observed 1990 cm
À1
line has been attributed to the presence of hydrogen in the crystal;
however, it took almost 20 years to show unambiguously that this line (1998 cm
À1
precisely)
refers to the H(BC) configuration of isolated hydrogen [29,30]. The same procedure has
been used by Holm et al. [31] for n-type silicon where, after low-temperature proton
implantation using deep-level transient spectroscopy (DLTS), a E
c
ÀE
t
¼0.16 eV donor
level (E3
0
) has been observed. Finally, in p-type or intrinsic material where H
0
(BC) converts
into H
þ
(BC), Gorelkinskii and Nevinnyi [32] used band gap light to produce a steady-state
population of H
0
(BC) and observed the EPR signal, labeled AA9. The metastability of the
ß 2008 by Taylor & Francis Group, LLC.
E3
0
center was established by Bech Nielsen et al. [33], and its annealing behavior was found
to match that of the AA9 signal. On this basis, the DLTS signal E3
0
was finally assigned to
electron emission from H
0
(BC).
Further detailed studies performed by Bonde Nielsen et al. [34] showed that at least two
kinds of BC hydrogen donors exist. One, H
0
(BC), is associated with hydrogen at the BC
site; the other, H
0
(BC)–O
i
, is associated with BC hydrogen weakly bound to interstitial
oxygen. A key point in this interpretation is that H
þ
(BC)–O
i
converts to H
þ
(BC) in oxygen-
lean material, whereas in oxygen-rich material the reverse process occurs. Both centers
are abundant after low-temperature proton implantation giving rise to two signals,
denoted E
0
and E3
00
, which are observed when DLTS is applied in situ to as-implanted
samples. These signals are always present after implantation into the depletion layer of
a reverse-biased diode (no free electrons available in the implantation region). They are
indistinguishable as far as their electronic DLTS signals are concerned, but can be dis-
cerned on the basis of characteristic differences in their formation and annealing properties.
The tracking of charge accumulated in the depletion layer shows that both centers can
convert temporarily into singly negatively charged centers. These, so far hidden, centers
are thermally stable up to $250 K, where they both anneal.
As a consequence of the perturbation, the H(BC)–O
i
center should have a lower sym-
metry than the H(BC) center. This has been confirmed by the application of the uniaxial-
stress technique to the DLTS method [35]. Figure 2.5 depicts the splitting pattern of the E3
0
emission signal which reveals the expected trigonal symmetry of H(BC), whereas the
splitting under <100> stress of the E3
00
signal into two components indicates a lower
(monoclinic-I) symmetry as expected for H(BC)–O
i
.
The position of the T-site acceptor level has been much more difficult to identify. The
acceptor level has been defined as the energy difference between H
0
(BC) þe
c
and H
À
(T)
and placed slightly below midgap on the basis of a kinetic study [36,37]. As the first step in
this work an initial amount of H
þ
was released in a diode space-charge layer from
hydrogen-passivated shallow donors by reverse-bias illumination and converted to H
À
by the application of a filling pulse. Then the subsequent thermally induced removal of
H
À
was recorded as an exponential capacitance transient. This is ascribed to the first step
of the process H
À
(T) ! H
0
(BC) þe
c
! H
þ
(BC) þ2e
c
and the activation enthalpy of this
step obtained [36,37]. It was found, consistent with the anticipated negative-U property,
0.1 1 10
Zero stress
<100>
<111>
<110>
100
E3Ј
E3Љ<100>
0.4 GPa
100
Emission rate (s
Ϫ1
)
L
a
p
l
a
c
e

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y
1000
FIGURE 2.5
Laplace DLTS spectra revealing the trigonal sym-
metry of the center associated with the E3
0
signal,
and a lower symmetry of the center associated
with the E3
00
signal (inset). (Adapted from Bonde
Nielsen, K. et al., Physica B, 308, 134, 2001. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
that this enthalpy was larger than that described above [31] for the second step of the
process. The authors obtained the position of the acceptor level from detailed-balance
considerations. In view of the evidence that at least two kinds of BC hydrogen donors
exist [H(BC) and H
0
(BC)–O
i
] the obvious question arose whether a similar situation
applied to the T-site acceptor. If so, two different configurations of H
À
(T) with and without
an oxygen atom nearby, should exist.
The main idea behind the H
À
(T) (or H
À
(T)–O
i
) configuration detection was that this
configuration annealing is triggered by a thermal electron emission process. The emission
was then observed as a capacitance transient at temperatures slightly below 250 K, the
temperature at which it is known that the H
À
(T) centers anneal. As depicted in Figure 2.6,
the single-shot capacitance-transient measurements for long decay times reveal the emis-
sion [38]. The transient shown has been recorded with an oxygen-rich hydrogen-
implanted diode with about 55% of the implants initially present as H(BC) and the rest
as H(BC)–O
i
. Both centers were first converted into the negative hidden centers by bias
removal at 110 K. The capacitance transient was then recorded after reapplication of the
bias at 242 K. As can be seen, the transient consists of two exponential components. The
fast decay is ascribed to electron emission from H
À
(T) and labeled as AT
0
. The slow decay
is analogously ascribed to emission from H
À
(T)–O
i
and labeled AT
00
. These assignments
were based on detailed studies of the stabilities of both signals and on the analogy to the
stabilities of the H(BC) and H
0
(BC)–O
i
configurations. From the transients AT
0
and AT
00
measured at different temperatures, the Arrhenius plots are depicted in the inset to
Figure 2.6. The remarkable agreement over many decades between the AT
00
data and
those obtained in Ref. [36] can be noted. The linear-regression analysis yielded the
activation DH
00
¼0.79 Æ0.03 eV for the emission from H
À
(T)–O
i
, and DH
0
¼0.65 Æ0.10
eV for the emission from H
À
(T).
The electron emission enthalpies found for H
À
(T)–O
i
and H
À
(T) exceed those of the
corresponding BC donor states in accordance with the expected negative-U property.
However, to demonstrate this property more directly, the correlation between the amount
of charge bound as H
0
(BC)–O
i
and H
À
(T)–O
i
has been examined. For this purpose, a
reverse-biased sample was prepared containing, predominantly, H
þ
(BC)–O
i
. The
H
þ
(BC)–O
i
was then partially converted to H
À
(T)–O
i
by the application of a filling pulse
with the number of the converted centers controlled by the pulse duration. The remaining
FIGURE 2.6
Single-shot capacitance transient at 242 K
revealing two components AT
0
and AT
00
[38].
The inset shows the Arrhenius analysis of
AT
00
(r) and AT
0
(D). The data (.) of Ref. [36]
have been included. (Adapted from Bonde
Nielsen, K. et al., Physica B, 308, 134, 2001.
With permission.) Time (10
3
s)
0 5 10 15 20 25 30
Δ
C
(
t
)
/
Δ
C
0
0.1
1
ATЈ
ATЉ
1000/T (K
Ϫ1
)
3.5 4.0 4.5
l
n
(
e
n
/
T
2
)
Ϫ22
Ϫ20
Ϫ18
Ϫ16
Ϫ14
Ϫ12
Ϫ10
ß 2008 by Taylor & Francis Group, LLC.
H
þ
(BC)–O
i
gives rise to a capacitance transient (the E3
00
signal) following the filling pulse.
The amplitude of this transient (see Figure 2.7a) is a measure of the amount of charge
emitted in the process H
0
(BC)–O
i
!H
þ
(BC)–O
i
and is denoted as DC
BC
00 . An illumination
pulse applied immediately after the E3
00
transient has leveled off causes a shift in steady-
state capacitance, which is denoted as DC
T
00 . This capacitance shift measures the charge
change as the H
À
(T)–O
i
configuration converts back to the H
þ
(BC)–O
i
configuration. Then,
choosing different filling pulse lengths to vary the relative amounts of H
0
(BC)–O
i
and
H
À
(T)–O
i
, the linear correlation between DC
T
00 and DC
BC
00 depicted in Figure 2.7b is
obtained. The slope of this correlation line reveals that two electrons are emitted when
H
À
(T)–O
i
is converted optically to H
þ
(BC)–O
i
. This result establishes the one-to-one
correspondence E3
00
$ AT
00
, and hence supports the assignment of AT
00
to H(T)–O
i
and
evidences the negative-U behavior of the H(T)–O
i
defect.
The electrical measurements performed for the H
À
(T) and H
À
(T)–O
i
configurations of
hydrogen seem to be the only spectroscopic data where the properties of these species can
be observed. The negatively charged hydrogen is diamagnetic and does not produce a
signal suitable for the magnetic resonance technique. Despite the fact that the H
þ
(BC)
configuration can be observed in the far-IR LVM absorption measurements, the H
À
(T)
configuration is not observed by this method. The detailed studies on this problem
performed recently by Pereira et al. [39] showed that negatively charged hydrogen in the
antibonding position attached to one silicon atom can swiftly reconfigure between four
equivalent interstitial positions as the barrier separating these positions in silicon is low.
Therefore, the antibonding configuration of negatively charged hydrogen in silicon may be
short-lived due to rapid jumps of hydrogen among these equivalent antibonding sites,
which would broaden the LVM lines and hamper their detection.
2.4 Passivation
The term passivation is used generically to describe the removal of electrical activity.
Hydrogen is a remarkably effective passivator in its atomic form due to its high reactivity.
It migrates through the lattice and reacts with unsatisfied bonds of defects and impurities
rendering them electrically inactive. This is the basis of the widely used technique of
forming gas anneal to reduce the interface trap density in MOS devices. Passivation is an
ΔC
BCЉ
(pF)
0.0 0.5 1.0
Δ
C
T
Љ

(
p
F
)
0
1
2
(b) (a)
ΔC

Time (s)
0 10 20 30 40 50 60
C
a
p
a
c
i
t
a
n
c
e

(
p
F
)
0
1
2
3
ΔC
BCЉ
FIGURE 2.7
Illustration of the correlation between E3
00
signal and the reservoir for the AT
00
signal. (a) The pulsing sequence to
obtain DC
T
00 and DC
BC
00 as explained in the text. (b) The DC
T
00 versus DC
BC
00 dependence showing that the H
À
(T)–O
i
configuration binds twice more electrons than H
0
(BC)–O
i
. (FromBonde Nielsen, K. et al., Phys. Rev. B, 65, 075205-1,
2002. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
important process in improving the electrical properties of amorphous silicon, and in
recent years it has been used extensively in polycrystalline solar cell materials to reduce
recombination at grain boundaries.
In crystalline silicon the passivation of boron acceptors is the process that has been most
extensively studied [40,41]. Although there has been much debate about the mechanism of
passivation, it is now generally accepted that the hydrogen resides at the bond center
between the silicon and the boron; this has strong support from vibrational spectroscopy
[42], perturbed angular correlation experiments [43], channeling studies [44], and several
theoretical calculations [45]. However, the hydrogen is not strongly bound to the boron.
Indeed, this is true for all shallow dopants. The dissociation energies for the acceptors have
been determined by Zundel and Weber [46] and are given in Table 2.1.
The technique used was to study the dissociation of the acceptor hydrogen complexes in
the junction field of a Schottky barrier diode, which was annealed with an applied reverse
bias, to sweep the hydrogen from the near surface region of the junction when the dopant
hydrogen pairs dissociated. This overcame the problem of earlier experiments in which the
hydrogen dissociated from the dopant, only to recombine again, making analysis of the
experimental data extremely difficult. The work of Zundel and Weber [46] showed that
when the bias was applied the dissociation of the acceptor hydrogen complexes could be
accurately described by first-order kinetics. The technique has also been used to study the
dissociation kinetics of donor hydrogen complexes in silicon [47,48]; the dissociation
energies are given in Table 2.1.
It is not only shallow donors and acceptors that can be passivated by hydrogen; there is a
very substantial literature on the passivation of deep-level defects being passivated in
silicon after exposure to a hydrogen plasma [49–51]. A few specific cases and their
structural relation to electrically active complexes are discussed in the following sections.
2.5 Electrically Active Hydrogen Complexes
2.5.1 Vacancy–Hydrogen Complexes
Defects that possess an unsatisfied dangling-bond form in silicon a particular class of deep
levels with rather similar energy levels. The level position in this group is predominantly a
property of the dangling bondandhas only a weak dependence on the detailedstructure of a
specific defect. The origin of dangling-bond levels can be traced back to the formation and
decoration of lattice vacancies in the semiconductor crystal. For example, when Si is sub-
jected to ion implantation a rather dense concentration of vacancy clusters will be generated
with the consequence that several close-lying acceptor-type dangling-bond levels appear
rather deep in the upper half of the band gap. Even with the high-resolution Laplace DLTS
TABLE 2.1
Hydrogen–Dopant Complexes: Dissociation Energies (E) of Dopant–Hydrogen
Pairs and Values of the Attempt Frequencies (v)
Acceptors B Al Ga In
E
A
(eV) 1.28 Æ0.03 1.44 Æ0.02 1.40 Æ0.03 1.42 Æ0.05
v
0A
(s
À1
) 2.8 Â10
14
3.1 Â10
13
6.9 Â10
13
8.4 Â10
13
Donors P As Sb
E
D
(eV) 1.2 Æ0.05 1.12 Æ0.05 1.13
v
0D
(s
À1
) 6 Â10
13
4 Â10
13
3 Â10
13
ß 2008 by Taylor & Francis Group, LLC.
method these multilevel structures are hardly resolvable [52]. However, if only two or
possibly three close levels are present, the application of Laplace DLTS can be used to
resolve similar dangling-bond acceptor levels VH(À=0), V
2
H(À=0), and PV(À=0) [52]. The
similarity of the electronic structures of these centers has been established by EPR by Bech
Nielsen et al. [53] and by Stallinga et al. [54]. The neutral charge state of the three defects
(VH
0
, V
2
H
0
, and PV
0
) has very similar EPR spectral parameters, indicating that the wave
functions of the odd electron are indeed very similar. The acceptor level of these centers
originates fromthe emission of an electron fromthe dangling-bond orbital to the conduction
band, leaving the orbital singly occupied. Consequently, the level energies are also expected
to be similar. The assignment of these levels as VH(À=0) and V
2
H(À=0) are based on the
comparison of annealing properties of the DLTS signals with the EPR data of Refs. [53,54].
The similarity of the V
2
H(À=0) and VH(À=0) levels is easily rationalized. The divacancy
V
2
(À=0) level may be qualitatively understood as originating from two weakly interacting
elongated dangling bonds. In V
2
H
0
hydrogen terminates one of these bonds leaving the
other practically undisturbed. The presence of the V
2
(À=0) level in the same energy range
supports this qualitative picture. In VH
0
hydrogen breaks the unsaturated bond structure of
the monovacancy generating a structure consisting of one dangling bond, a Si–Si bridge, and
a hydrogen-terminated bond. The similarity of the PV(À=0) signal with VH(À=0) shows that
the Si–H fragment of VH
0
may be regarded as a pseudo group V impurity, as far as giving
rise to a very similar dangling-bond level when binding to a lattice vacancy.
While VH and V
2
H can be members of the pseudo group V impurity group, then basing
on similar arguments, the VH
2
complex can be considered as a pseudo vacancy–oxygen
(VO) complex. The electrical activity of VO comes from the prolonged reconstructed Si–Si
bond, while the other two silicon atoms surrounding the vacancy are bridged by oxygen.
This is equivalent to the structure of the VH
2
complex based on the LVM [55] and EPR
[56–58] measurements, where the two silicon atoms instead of being bonded to interstitial
oxygen have their bonds terminated with hydrogen. Similar to the case of VO, the electrical
activity of VH
2
comes from the prolonged Si–Si bond; however, in this case the bond
experiences a different stress from the surrounding atoms. As a result, one can expect that
VH
2
forms a similar VO acceptor level with energy close to E
c
À0.17 eV. There are some
suggestions [59] that the E
c
À0.2 eV level can be associated with the VH
2
complex but this
needs to be confirmed.
2.5.2 VOH
Low-temperature electron-irradiation of Si crystals introduces vacancies and self-intersti-
tials, neither of which is stable as an isolated defect at RT. They diffuse and are trapped by
impurities such as oxygen. The VO center is a well-known and well-understood defect that
is stable up to 3008C. The oxygen atom passivates two of the Si dangling bonds and the
electrical activity of the center comes from a weakly reconstructed Si–Si bond bridging the
remaining Si neighbors bordering the vacancy. This bond is associated with an acceptor level
at E
c
À0.17 eV. Hydrogen, often unintentionally introduced during growth or processing,
readily reacts with the reconstructed bond leading to the formation of VOH and VOH
2
defects. The latter is electrically inactive but has been detected in IR absorption studies [60].
The former defect is electrically active and possesses acceptor and donor levels at E
c
À0.31 eV
and E
v
þ0.28 eV, respectively [52,61–64]. When neutral, it has been observed in EPR
experiments [64], and it was established that it possessed monoclinic-I (C
1h
) symmetry
below $180 K, but above 240 K it exhibits an apparent orthorhombic-I (C
2v
) symmetry due
to the H atom jumping between the two Si dangling bonds with an activation energy of
0.18 eV. The same effect has been observed in the Laplace DLTS measurements under
uniaxial stress [65,66]. At the temperatures used for these measurements (around 160 K),
ß 2008 by Taylor & Francis Group, LLC.
the hydro gen atom can jump swif tly betwe en two equiva lent pos itions, thus the observe d
symm etry appe ars as or thorhom bic a s in the cas e of VO (see Figu re 2.8 and com pare the
peak spl itting pat tern [66]) . The defe ct anne als out at aro und 300 8 C [64,65] .
2.5.3 CH
Unli ke the cas e of the isolated hydro gen atom in silicon, where two electro ns bound by
hyd rogen have a nega tive corre lation energy U [38], it has been shown that hydrogen near
carbo n is a positive- U species wi th donor and acceptor levels at E
v
þ 0.33 eV and E
c
À 0.16 eV,
respec tively . The latt er is almost identical with the donor level of isolat ed H. Furt her,
H can be trapped as a met astable spe cies, sep arated fro m substitut ional carbo n, and pos-
sesses a donor level at E
c
À 0.22 eV [67]. This sta te (C –H)
I
is observe d after low- temperat ure
proton impl antation. Aft er zero- bias anneal at 225 K, it is irr eversib ly los t, con verting to the
mo re stable hydrogen –carbon com plex, (C –H)
II
, with the acceptor level at E
c
À 0.16 eV [67].
The elect ron emis sion rate from this stable comple x is simi lar to that of the acceptor level of
the VO comple x. The con version rati o betwe en (C –H)
I
and (C –H)
II
is pra ctically 1:1,
indica ting a simple atomi c recon figura tion. The (C –H)
II
con figura tion is sta ble to just
abov e RT, a nd is also obse rved in sam ples wet etched in the dark at RT. It is irrevers ibly
lost by white light illumi nation at RT, whereu pon no othe r levels are obse rved in the upper
hal f of the gap. Th e (C –H)
II
level is identi cal to the ca rbon –hydro gen defect studi ed by
Endr ös [68] and Kam iura et al. [69]. Th eoretical mo deling descri bed in Ref. [67] led to the
conclu sion that the (0=þ ) elect rical level of hyd rogen in the CH
BC2
con figura tion (see
Figu re 2.9) is in exc ellent ag reement wi th (C– H)
I
. This defect con figura tion has previousl y
been obse rved by IR absorpti on [70] in simi lar mate rial. Th e observe d acceptor level of
(C–H)
II
at E
c
À0.16 eVhas been identifiedwith the CH
BC1
configuration, with an energy level
coincidentally close to the isolated H
BC
donor. The E
c
À0.16 eVlevel splitting under uniaxial
stress measuredwith Laplace DLTS andthe (C–H)
II
complex stress-inducedalignment effect
agree with the calculated defect response to the stress for the CH
1BC
complex configuration,
lending further weight to the assignment.
FIGURE 2.8
Laplace DLTS peak stress-induced splitting
observed for (a) VO and (b) VOH complexes.
(Adapted from Dobaczewski, L. et al., Physica B,
340, 499, 2003. With permission.)
1 10 10 100
Emission rate (s
Ϫ1
)
L
a
p
l
a
c
e

D
L
T
S

a
m
p
l
i
t
u
d
e

(
a
r
b
.

u
n
i
t
s
)
(a) VO T=82 K
<111>
0.15 GPa
<110>
0.15 GPa
<100>
0.10 GPa
0 GPa
(b) VOH T=160 K
<111>
0.40 GPa
<110>
0.65 GPa
<100>
0.45 GPa
0 GPa
ß 2008 by Taylor & Francis Group, LLC.
2.5.4 Transition Metal–Hydrogen Complexes
The hydrogenation process of a silicon sample can make transition metal (TM) impurities
electrically inactive. On the other hand, there are many cases in which for such a sample
the hydrogenation introduces new electrically active centers as a result of the TM–hydro-
gen complexes formation. As a general case, the hydrogenation does not necessarily result
from an introduction of hydrogen from external sources. It can happen that during device
processing hydrogen is released from other trapping centers, in particular from shallow
acceptors and donors. Thus, in some cases hydrogenation replaces the isolated TM-related
electrically active centers with new electrically active defects. Theory predicted that, in
these complexes, up to four hydrogen atoms can be involved, and some of these complexes
can be observed in four different charge states (see Ref. [71]). This variety of complexes and
charge states makes their unambiguous identification difficult. In particular, it can be
difficult to determine their potential roles as generation or recombination centers in device
operation. In this respect, DLTS can be very effective, enabling many hydrogen complexes
to be characterized. Examples are silver [72], gold [51,73–75], copper [76], chromium [77],
cobalt [78], iron [79], nickel [80], palladium [81], titanium [82], vanadium [77,83], and
platinum [84–86]. A common feature of all these different complexes is that they are not,
in general, stable above 2008C, and cannot be formed above this temperature.
2.5.5 Au–H
n
Among many TM–hydrogen complexes, the ones formed with platinum and gold have
attracted the most attention, due to the technological importance of these metals and
complexes. The microscopic structures of hydrogen complexes with gold have been ana-
lyzed by theory [71,87,88], LVM spectroscopy [89], and magnetic resonance techniques
[90,91]. The LVM studies revealed a C
2n
symmetry of the AuH
2
complex based on
vibrational modes of the isotopic shifts for mixtures H and D; however, it was not possible
to resolve the issue of the hydrogen location in these complexes, with either direct TM–H
or Si–H bonding possible. The lack of Si isotopic fine structure in the vibrational modes
suggested the former case. The same complex has been analyzed in Refs. [90,91], where it
Td1a
Td1b Td2
AB1
AB2
X
BC1 BC2
BC3a
BC3b
(001)
(110)
FIGURE 2.9
Configurations of interstitial H (small dots) in Si (Ge). Host atoms are shown as white circles. A substitutional
X¼C, Ge (Si) atom is represented by the black circle. Hydrogen sites are labeled according to the usual notation,
where BC, AB, and Td stand for bond-centered, antibonded, and tetrahedral interstitial sites, respectively.
It should be noted that BC and AB structures produce large lattice relaxations that are not shown. (From Balsas, A.
et al., J. Phys. Condens. Matter, 17, S2155, 2005. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
has been concluded that hydrogen in the complex is directly bonded to silicon, presumably
in the antibonding position.
There is a great deal of data on the electrical studies of hydrogen passivation of gold in
silicon. Following the work of Pearton et al. [51], Sveinbjörnsson and Engström [73] studied
the complexes formed during hydrogenation using DLTS and reported that there are four
electrically active deep levels (referred to as G1, G2, G3, and G4) resulting from the
formation of AuH complexes. It is believed that G1, G2, and G4 are different charge states
of the same AuH. Recent ab initio calculations embracing this defect system support this
interpretation [87]. It is generally accepted that gold (without hydrogen) forms an acceptor
which acts as a majority-carrier trap in n-type silicon. The G4 level appears to be very close
in energy to the gold acceptor and has almost identical electron emission characteristics not
resolved by conventional DLTS [73]. Figure 2.10 shows a comparison of the Laplace and
conventional DLTS spectra obtained from the same Si:Au:H sample [74]. The Laplace
DLTS spectrum shown in the main part of the figure reveals that there are two separate
and quite distinct bound- to free-electron emission rates at the measurement temperature
of 260 K. This is consistent with the previous interpretation of deconvolved DLTS spectra,
and confirms unambiguously that the conventional DLTS peak at the Au-acceptor position
in hydrogenated silicon consists of two contributions.
2.5.6 Pt–H
2
Platinum-related deep centers in silicon are widely used to control the minority-carrier
lifetime in fast-switching semiconductor devices. However, unintentionally introduced
hydrogen interacts with platinum, leading to the formation of new electrically active
complexes. The existence of PtH
n
(where n ¼1–3) complexes have been detected by EPR
[90–93], LVM [84,94], and DLTS [84–86,95]. The platinum–dihydrogen complex (PtH
2
) is of
particular interest from the point of view of application of platinum for minority-carrier
lifetime reduction in power devices (increase the switching speed) because, among all
platinum–hydrogen complexes, it has the shallowest electronic level. This makes the
platinum–dihydrogen center an effective carrier recombination center. But as the electronic
FIGURE 2.10
DLTS and Laplace DLTS spectra of hydrogen-
ated silicon containing gold. The conventional
DLTS spectrum is shown as an inset at the top
right of the figure. The broad peak centered at
260 K is attributed to electron emission from the
gold acceptor and G4. The main spectrum uses
the Laplace technique and clearly separates the
gold-acceptor level and the gold-hydrogen
level G4. (Adapted from Deixler, P. et al., Appl.
Phys. Lett., 73, 3126, 1998. With permission.)
G4 (Si:Au,H)
Gold acceptor
200 240 280 320
T (K)
C
o
n
v
e
n
t
i
o
n
a
l

D
L
T
S

o
u
t
p
u
t
Rate window 50 s
−1
10
0
10
1
10
2
10
3
10
4
10
5
Emission rate (1 s
−1
) at 260 K
S
p
e
c
t
r
a
l

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n

ß 2008 by Taylor & Francis Group, LLC.
level is well away from the midgap position, the carrier generation via this defect is less
effective, so the device leakage is reduced. In spite of a number of different studies by both
theoretical [71,88] and experimental groups, there are some uncertainties relating to the
PtH
2
structure. It is well known that it possesses C
2n
symmetry, which is the result of a
static Jahn–Teller distortion combining tetragonal and trigonal components analogous
with the isolated negatively charged vacancy. This symmetry has been demonstrated
with use of EPR [90–93], LVM [94], and DLTS [85,86] methods. However, the position of
hydrogen atoms in platinum–hydrogen complexes is disputed. Some models assume that
hydrogen atoms are attached directly to the platinum; however, it is more generally
believed that two hydrogen atoms bond to two Si atoms neighboring the substitutional
Pt and pointing away from the platinum.
2.6 Hydrogen Molecules
In addition to interacting with impurities and defects in semiconductors, H can also
form H
2
molecules in the silicon lattice. Early theoretical calculations [96,97] predicted
that the H
2
molecule in Si would sit at a tetrahedral interstitial (T) site where it should be
electrically and optically inactive. It has also been suggested that H
2
is one of the most
stable forms of hydrogen in silicon and plays an important role during the in-diffusion of
hydrogen and in a variety of H-related reactions [3,4]. However, in spite of its proposed
importance, the H
2
molecule was not observed in semiconductors until recently.
H
2
in crystalline silicon was observed first by Raman spectroscopy in hydrogen-plasma-
treated Si samples [98]. A Raman line at 4158 cm
À1
, which was assigned in Ref. [98] to H
2
,
was proved later to be caused by molecules associated with H-saturated platelets [99].
Subsequently a vibrational line due to the isolated interstitial H
2
at the T-site in the Si lattice
was discovered in Raman [100] and IR absorption spectra [101]. The line associated with
the isolated H
2
was found to be at 3601 cm
À1
(Raman, RT) or 3618 cm
À1
(IR, 10 K).
Vibrational lines due to D
2
and HD were also observed. Furthermore, theoretical calcula-
tions found vibrational frequencies for isolated H
2
in Si that are in agreement with
experiment [102–105]. It appears that the interaction between H
2
and the host lattice
weakens the H–H bond, which results in a drop in its stretch frequency by about 550
cm
À1
relative to that of the free molecule ($4160 cm
À1
).
Properties of the isolated H
2
molecules in silicon have been studied in detail by two
groups with the use of IR absorption [106–109] and Raman spectroscopies [110–112]. It is
found that all of the experimental observations can be explained in terms of a nearly free
rotator, which interacts with host atoms surrounding a tetrahedral cage and exchanges
energy with the host crystal. In oxygen-rich Si crystals H
2
molecules can be trapped
at T-sites near interstitial oxygen (O
i
) atoms [113,114]. Careful annealing studies have
shown that the binding energy of H
2
to O
i
is 0.26 Æ0.02 eV, and the activation energy
for diffusion of H
2
among interstitial sites is 0.78 Æ0.02 eV [114]. The diffusion of the
molecule is expected to occur along the trigonal axis from the T to hexagonal and then to
another T-sites. Multivacancy voids in Si crystals have also been found to be effective
trapping sites for H
2
molecules [115,116].
The H
2
* defect, another hydrogen dimer with one hydrogen atomclose to a BCsite andthe
other one along the same trigonal axis at an antibonding site, was also predicted to be stable
in semiconductors [117]. The H
2
* defect in Si was identified later in an IR absorption
spectroscopy study combinedwith ab initio modeling [118]. Theory predicts that the relative
stability of interstitial H
2
as compared to the H
2
* increases with the host lattice constant
for the group IVsemiconductors [119]. In Si, interstitial H
2
is expected to be the ground state
ß 2008 by Taylor & Francis Group, LLC.
of hydrogen. However, molecular dynamics simulations have shown that the interaction of
the interstitial H
2
molecule with a vacancy and a self-interstitial can result in its dissociation
and transformation to the H
2
* defect [120].
2.7 Hydrogen Clusters and Platelets
Under certain circumstances hydrogen can be incorporated into silicon in such high
concentrations that it exists in a precipitated form. The formation of such defects is
well-documented after exposure to a hydrogen plasma or after ion implantation. The
latter process is of considerable commercial significance with the advent of the SmartCut
process [2].
Weldon et al. have investigated the precipitation of hydrogen and the formation of
microvoids and platelets after the implantation of hydrogen [121]. They have studied the
evolution of the internal defect structure as a function of implanted hydrogen concentration
and anneal temperature and found that the chemical action of hydrogen leaves the formation
of (100) and (111) internal surfaces above 4008C via an agglomeration of the initial defect
structure. In addition, between 2008C and 4008C molecular hydrogen is released. This is
subsequently trapped in the microvoids bounded by the internal surfaces resulting in the
buildup of internal pressure. This is the mechanism that leads to the blistering of silicon, and
is the basis of the process used for layer transfer in the SmartCut process.
First-principles calculations of such platelets are difficult as they involve studying large
defect structures using relatively small clusters. However, modeling work by Reboredo et al.
[122], based on calculations of configuration energies using density functional theory,
suggest that mechanisms for the nucleation and growth of aggregates of second nearest
neighbor hydrogenated vacancies are energetically favorable. One of the complexes
proposed is a second nearest neighbor pair of two VH
4
molecules. The bimolecular complex
is predicted to accumulate progressively, forming the platelets. Such complexes are expected
to have a preference for formation along {111} and {100} planes as is observed in experiments.
The formation and evolution of hydrogen platelets in silicon after exposure to remote
plasma has been investigated in detail by Lavrov and Weber [123]. The structure of such
platelets is rather complex, but electron microscopy studies suggest that they consist of {111}
platelets with a hydrogen-saturated internal surface in which each silicon–silicon bond in a
(111) plane is replaced by two Si–H bonds [124]. However, it does seem that other forms of
platelets can exist [125]. In this case H
2
* aggregates are involved in the formation of the {111}
platelets. The existence of two different structures has been confirmed by Lavrov and
Weber using polarized Raman scattering spectroscopy [123]; the relative concentrations of
the two species of platelet depend on the plasma conditions. They observe that the H
2
*
structure predominates under low-temperature formation conditions and transforms into
the [2Si–H]
n
þH
2
structure as the temperature rises above 1008C.
2.8 Hydrogen in Germanium
Whereas the properties of hydrogen in silicon have been studied extensively in the past
and are now reasonably well understood, understanding of hydrogen in germanium
requires further efforts of researchers. Taking into account the similar properties of crys-
talline Si and Ge and very similar chemistries of {Si, H} and {Ge, H} compounds, one would
ß 2008 by Taylor & Francis Group, LLC.
not expect qualitative differences in the behavior of H in these materials. Indeed, some
hydrogen-related phenomena like the passivation of deep centers and reactivation of
neutral centers have been found to be very similar in Ge and Si. However, some experi-
ments involving H, which were done on comparable Si and Ge samples, showed different
results. For example, in contrast to Si, in Ge only one successful attempt of passivation of
shallow acceptors has been reported [126], and passivation of shallow donors has never
been realized. A brief review on the properties of hydrogen in Ge is presented in the
following sections.
2.8.1 Hydrogen Incorporation, Diffusivity, and Solubility in Ge
The solubility and diffusivity of hydrogen near the melting point has been studied in the
1960s by permeation experiments [9,127]. According to Frank and Thomas [127], the
diffusion coefficient D and solubility S of atomic hydrogen in the temperature range
8008C–9108C can be described by the following equations:
D
H
¼ 2:7 Â10
À3
exp(À0:38=kT) [cm s
À1
] (2:1)
S
H
¼ 1:6 Â10
24
exp(À2:3=kT) [cm
À3
] (2:2)
Hydrogen diffusivity at elevated temperatures is high and typical for interstitial diffusion.
It indicates that trapping does not affect the movement of hydrogen in this temperature
range. Near the melting point the S
H
value is about 4 Â10
14
cm
À3
, but the extrapolation
of Equation 2.2 yields a negligible equilibrium concentration of H near RT; e.g., a value of
10
5
cm
À3
is obtained at 3508C. However, in practice, much higher values have been
observed in high-purity Ge crystals [128]. This indicates that at lower temperatures there
is a trapping enhanced increase of the hydrogen concentration in germanium. Further,
experimental values of hydrogen diffusivity at lower temperatures have been found to be
much lower than the values extracted from Equation 2.1, thus indicating that at lower
temperatures, as in the case of silicon, hydrogen diffusion is trap-limited [4,129].
Hydrogen incorporation during growth and following cooling to RT seems to be very
efficient in the case of Ge [128]. In contrast, hydrogen (deuterium) incorporation into
germanium from H(D) plasma is not very effective, compared to this process for the case
of Si. Thermal effusion (TE) measurements of D-plasma-treated Ge samples showed that
out-diffusion of hydrogen occurs at 2008C, a significantly lower temperature than in Si
(3508C–4008C) [130]. Further, the amount of D that penetrates into a Ge sample exposed to
a D plasma decreases with the ambient temperature, while the opposite holds for Si. At
2508C all hydrogen was found to be located on or near the sample surface, even for long
exposure times. It appears that the Ge surface constitutes an effective diffusion barrier or
binding center for atomic hydrogen, thus preventing significant concentrations of H inside
the bulk material. Ge–D and GeD
2
species (mono- and dideuterides) formed at the Ge
surface were suggested as sources of the high surface concentrations [130]. GeD and GeD
2
species desorb at 3008C and 2008C, respectively.
2.8.2 Isolated Hydrogen Species and Complexes with Intrinsic Defects in Ge
Theory predicts negative-U properties for isolated hydrogen atoms in all semiconductor
materials [131]. In Ge, the E(À=þ) level of the isolated H atom is calculated [131] to be close
to or within the valence band, leading to nonamphoteric behavior. For all Fermi-level
positions, H
À
should be the stable equilibrium configuration. Although one might expect
ß 2008 by Taylor & Francis Group, LLC.
that, as for the case of silicon, the isolated H atoms could be observed in nonequilibrium
metastable neutral and singly positively charge state in germanium. Isolated hydrogen
atoms in Ge crystals have been detected recently by IR absorption and DLTS measure-
ments after low-temperature hydrogen implantation [132,133]. In a work by Budde et al.
[132] two LVM lines at 1794 and 745 cm
À1
were observed and assigned to an A
2u
vibration
of H
BC
þ and a degenerate wagging mode of H
À
close to the T-site, respectively. It was
found from DLTS measurements that the donor level of bond-centered hydrogen in Ge is
at E
c
À0.11 eV [133]. A search for the complementary interstitial-site acceptor level failed,
indicating that this level could be very shallow or resonant with the valence band. The
experimental results obtained are consistent with the theoretical predictions [131].
The H
2
molecule is expected to be the ground state of H in Ge [120], but only recently
Raman signals due to the isolated hydrogen molecule at the tetrahedral interstitial site and
trapped in hydrogen-induced platelets have been observed in H(D)-plasma-treated Ge
samples [134,135]. Two LVM bands 3826 and 3834 cm
À1
were assigned to ortho and
para interstitial H
2
trapped at the T-site within the Ge host lattice [135]. The LVM band
at 4155 cm
À1
was shown to be related to H
2
trapped in platelets [134].
The H
2
* defect was also observed in Ge crystals by means of IR absorption measure-
ments after hydrogen implantation [136]. Several absorption lines were assigned to LVMs
of the H
2
* complex: stretch modes at 1774 and 1989 cm
À1
, along with a bend mode and its
first overtone at 765 and 1499 cm
À1
[136]. As in the case of silicon, hydrogen can terminate
dangling bonds of the native vacancy (V) and interstitial (I) defects. With the use of IR
absorption spectroscopy, LVMs due to VH
2
, VH
4
, V
2
H
6
, and IH
2
defects were identified in
Ge samples implanted with hydrogen [137,138].
2.8.3 Hydrogen Passivation of Deep Centers and Activation of Neutral Impurities in Ge
Hydrogen passivation of deep donor and acceptor states in n- and p-type Ge was reported
by Pearton and coworkers in a number of publications [139,140]. A review of this work was
given in Ref. [4]. It was found that, in a few cases, the interaction of hydrogen with deep
centers resulted in the creation of new hydrogen-related levels. A good example is the
interaction of hydrogen with the triple acceptor Cu
s
, having energy levels at E
v
þ0.04 eV
(Cu
s
0=À
), E
v
þ0.33 eV (Cu
s
À=2À
), and E
c
À0.26 eV (Cu
s
2À=3À
). It was shown that the CuH
and CuH
2
complexes are double and single acceptors, respectively [141–143]. A neutral
CuH
3
defect was suggested from a simple model that three hydrogen atoms could replace
the three holes in the triple acceptor.
High-purity Ge was the first semiconductor material where electrically active hydrogen-
related centers were identified unambiguously. Using high-resolution photothermal
ionization spectroscopy, it was revealed that the interaction of hydrogen with neutral
impurities, carbon, silicon, and oxygen resulted in the formation of complexes exhibiting
shallow acceptor, A(H, C) and A(H, Si), and shallow donor, D(H, O), characters [144–146].
The involvement of one hydrogen atom in these complexes was proved by observations of
isotope shifts after replacement of H by D and the resulting splitting of the spectral lines in
material grown in a mixed ambient [145].
2.8.4 Device-Related Aspects of Hydrogen in Ge
Hydrogen is a very important impurity in the production technology of sensitive Ge-based
gamma-radiation detectors. In fact, high-purity germanium with a net doping concentra-
tion lower than 10
11
cm
À3
, which is used for the production of the detectors, can only be
grown under a hydrogen ambient [128,129,144]. One of the reasons is that a high amount
ß 2008 by Taylor & Francis Group, LLC.
of hyd rogen can be inc orporat ed, which passiv ates part of undes irable electrica lly active
impu rities and contami nants such as Cu.
The renewe d inter est in using high -mobili ty Ge layers in mains tream micro electroni c
technol ogy bring s this mate rial back int o the focus of pres ent day research [147]. Becau se
bulk Ge wafers are rather expe nsive, there is a high deman d for germ anium- on-ins ulator
(GeOI) structur es. It has bee n shown that GeOI wafe rs can be produced using the
Smar tCut proce ss [148]. Proton impl antation is a basic step of the SmartCut techno logy
[149]. Typical implantat ion con ditions are 120 –160 keV energy and a fl uence of about
5 Â 10
16
H
2
þ
ions cm
À 2
[148]. Up on subsequ ent anneal ing, the imp lantatio n dam age and
hydro gen cause the formatio n of micr ocracks or blister s and then the exf oliation of a thin
layer of Ge, thereb y mak ing it poss ible to transf er the layer from the impl anted sample to
a substrat e to which it bon ds. It has been foun d that the formati on energy for micro crack
formatio n in the group IV semico nductors scales with the bon d energy. In the case of
germ anium, the a ctivation energy was found to be 1.0 eV, wi th a Ge ÀÀ Ge bon d energy
of 1.6 eV [148].
2.9 H ydrogen in S ilicon– Germanium A lloys
At pres ent, m uch attentio n is focused on silico n –germani um alloy s due to their appl ication
in devices [150,151]. Among the desirable properties are high carrier mobilities that are
important for ultrafast devices, and the use of strained SiGe materials in band gap
engineering for optoelectronic devices [152–155] as well as commercial Si-based techno-
logies (e.g., Chapter 3).
The main questions to be answered in relation to the behavior of hydrogen in dilute SiGe
alloys are whether (1) species of the least abundant element (the minority species) are
effective sinks for hydrogen atoms, and (2) the minority species can influence the dynamic
properties of hydrogen atoms significantly. These issues have been addressed recently in a
number of publications [39,156–160].
To understand hydrogen-related defects in SiGe it is necessary to consider some prop-
erties of the alloy material first. SiGe alloys form a fully miscible solid solution with a face-
centered cubic lattice structure. In dilute SiGe crystals the minority species produces a
randomly distributed internal strain that is absent in pure silicon and pure germanium. In
Si-rich SiGe, the substitutional Ge forms four Si–Ge bonds that are longer than the
averaged Si–Si bond [161]. This induces strain locally, which causes compression of
some Si–Si bonds, while others are elongated in the vicinity of the Ge atoms. In Ge-rich
SiGe, the situation will be very similar since perturbed Ge–Ge bonds will exist in the
vicinity of the silicon atoms.
The formation of H
BC
defect involves two opposing energy terms: (1) the lowering of the
energy that results from the formation of the three atom bond X
1
–H–X
2
and (2) the increase
in energy associated with the outward displacement of the two host atoms bonded to
hydrogen. The energy term in (2) will depend on the length of the Si–Si (or Ge–Ge)
bond before the hydrogen atom is accommodated and, henceforth, it is more energetically
favorable to add hydrogen to an elongated bond than to a compressed one. This implies that
the X
i
–H bond lengths of H
þ
BC
will depend on the stress conditions of the pertinent BC site;
e.g., compressed Si–Si (or Ge–Ge) bonds result in shorter X
i
–H bonds. As a result of
the different bond lengths, the vibrational frequencies associated with the stretching
of the X
i
–H bonds will differ from site to site, as well as the thermal stability of different
H
þ
BC
sites. Shortened X
i
–H bonds result in H
þ
BC
centers vibrating at higher frequencies
and with lower thermal stability. This means that the frequencies (and thermal stabilities)
ß 2008 by Taylor & Francis Group, LLC.
of different H
þ
BC
sites will correlate with the local strain conditions of the BC site that
accommodates the hydrogen atom. In this context, the H
þ
BC
in dilute SiGe alloys is especially
interesting, since the local strain imposed by the minority species may be probed by
measuring the strain-inducedLVMfrequency shifts on H
þ
BC
sites neighboredby the minority
species, with respect to the frequencies observed for isolated H
þ
BC
centers. Knowing the local
strain properties is essential to clarify the origins of the unique properties of SiGe alloys and
to develop the device potential of these materials.
Configurations and vibrational properties of elemental hydrogen centers in Si, Ge, and
dilute SiGe alloys have been modeled recently by an ab initio supercell method [156].
Poss ible sites for an iso lated hydro gen atom in dilute Si Ge alloy s are sh own in Figu re 2.9.
According to results of the modeling [156], all of the bond-centered configurations for an
isolated hydrogen atom are nearly degenerate in both the Si-rich and Ge-rich dilute SiGe
alloys; i.e., the binding energies of the hydrogen atom to the minority species are supposed
to be negligible. However, the calculated LVM frequencies have been found to be different
for a hydrogen atom at the different BC sites [156]. It has been found that H
À
ions may
be trapped at low temperatures in the form of H
À
AB
–Si–Ge units in dilute SiGe alloys. In a
Si-rich host, the presence of soft Si–Ge bonds, when compared to the abundant and stiff
Si–Si bonds, encourages the formation of Si–H bonds at the expense of the Si–Ge ones. On
the other hand, in Ge-rich alloys, H
À
AB
–Si–Ge defects are about 0.27 eV more stable than the
isolated H
À
AB
–Ge defect.
Different configurations of isolated hydrogen atoms have been observed in dilute Si-rich
and Ge-rich SiGe alloys by means of LVM spectroscopy after low-temperature (20 K)
proton (deuteron) implantation [39,157,158]. LVMs of X
s
H
þ
BCn
(n¼1, 2, 3) defects have
been detected. For details of the LVM frequencies for different X
s
H
þ
BCn
configurations,
readers are referred to Ref. [158]. In agreement with the results of ab initio modeling
[156], the analysis of intensities of LVMs bands associated with different H
þ
BC
-related
configurations has shown that there are no preferential sites for the H
þ
BC
species in the
vicinity of Ge atoms in Si-rich SiGe alloys [158]. Furthermore, isochronal annealing studies
have shown that thermal stabilities of different X
s
H
þ
BCn
defects are similar.
In Ge-rich SiGe alloys implanted with protons at low temperature, in addition to LVM
lines due to bond-centered hydrogen atoms, three LVM lines at 815.8, 1430.2, and 1630.7
cm
À1
have been observed and assigned to the negatively charged hydrogen atom located
at an antibonding site and attached to a substitutional silicon atom [39]. It has been found
that a conversion process takes place at temperatures in the range 20–125 K between
negatively charged hydrogen at a germanium antibonding site to a silicon antibonding
site. This provides direct evidence that silicon acts as an effective trap for negatively
charged hydrogen in SiGe alloys. Further, it has been reported in Ref. [159] that Si atoms
in Ge-rich SiGe are effective nucleation sites for the formation of a dominant H
2
*-like defect
at and above RT. This complex consists of a Ge–H
BC
Á Á ÁSi–H
AB
structure along a <111>
axis, where the H atoms have effectively broken a Si–Ge bond in the alloy. Several
absorption lines have been assigned to LVMs of the complex: a stretch mode at 1850.7
cm
À1
, along with a bend mode and its first overtone at 813.3 and 1592.6 cm
À1
.
The formation of Ge–H
BC
Á Á ÁSi–H
AB
complex strongly reduces the trapping of H by vacan-
cies and self-interstitials, which are major H-trapping centers in pure Ge [159].
In an in situ Laplace DLTS study of electrically active defects induced in Si-rich dilute
SiGe samples by proton implantation at 60 K, in addition to an E3
0
signal due to H
0
(BC) !
H
þ
(BC) transition, which is similar to that observed in pure silicon, another signal {E3
0
(Ge)}
has been observed and assigned to a defect with a hydrogen atom in the Ge
s
H
BC2
configuration [160]. The activation energy of the electron emission from the E3
0
(Ge)
donor has been found to be $158 meV when extrapolated to zero electric field. This is
slightly lower than the value of $175 meV for the E3
0
defect. It was deduced from the
ß 2008 by Taylor & Francis Group, LLC.
constructed configuration diagram of the Ge-strained site that alloying with $1% Ge does
not significantly influence the low-temperature migration of hydrogen as compared to
elemental Si [160].
In summary, the recent results of the properties of hydrogen-related defects in dilute
SiGe alloys [39,156–160] have shown that in Si-rich SiGe alloys the minority species (Ge
atoms) are not effective sinks for hydrogen atoms and do not influence the diffusivity of
hydrogen significantly. In Ge-rich SiGe alloys the binding energy of the H
À
AB
–Si–Ge center
has been found to be significantly higher than that for the isolated H
À
AB
–Ge defect. Further,
the formation of Ge–H
BC
Á Á ÁSi–H
AB
complex is very effective in Ge-rich SiGe. This causes
the suppression of the trapping of H atoms by other lattice defects, particularly by
vacancies and self-interstitials.
2.10 Summary
This chapter has summarized the behavior of hydrogen in silicon, germanium, and silicon–
germanium alloys in terms of its incorporation, diffusion, and defect reactions. It is evident
that the behavior is very complex as a consequence of the high diffusivity and reactivity of
the atomic species. Hydrogen can passivate electrically active defects or react with impur-
ities to form new complexes that have electronic states within the band gap. Much work
has been done in this field with the result that the literature contains detailed studies of
most of the commonly observed defects, especially those of technological importance in
silicon material and devices. In many cases a comprehensive understanding has been
established of the electronic properties of the defect and its stability. In an increasing
number of cases a combination of experimental and theoretical studies has resulted in an
appreciation of the detailed structure of the defect and its energetics. Far less work has been
done on germanium and silicon–germanium, but considerable effort is now being devoted
to hydrogen in these materials. It is anticipated that the next few years will see a maturity
of vision in relation to hydrogen-related defects in germanium and silicon–germanium
comparable with our present understanding of hydrogen in silicon.
References
1. Fowler, A.B., Process for the elimination of interface states in MIOS structures, US Patent
3,849,204, 1974.
2. Aspar, B. et al., Basic mechanisms involved in the SmartCut
1
process, Microelectron. Eng., 36,
233, 1997.
3. Pankove, J.I. and Johnson, N.M. (Eds.), Hydrogen in Semiconductors (Semiconductors and Semi-
metals, Vol. 34), Academic Press, Boston, 1991.
4. Pearton, S.J., Corbett, J.W., and Stavola, M., Hydrogen in Crystalline Semiconductors (Springer
Series in Materials Science, Vol. 16), Springer-Verlag, Heidelberg, 1992.
5. Newman, R.C., Hydrogen atoms and complexes in monocrystalline semiconductors,
Philos. Trans. R. Soc. A, 350, 215, 1995.
6. Estreicher, S.K., Hydrogen-related defects in crystalline semiconductors: A theorist’s perspec-
tive, Mater. Sci. Eng., R14, 314, 1995.
7. Willardson, R.K. and Beer, A.C. (Eds.), Hydrogen in Semiconductors (Semiconductors and Semi-
metals, Vol. 61), Academic Press, San Diego, 1999.
8. Van de Walle, C.G. and Neugebauer, J., Hydrogen in semiconductors, Annu. Rev. Mater. Res., 36,
179, 2006.
ß 2008 by Taylor & Francis Group, LLC.
9. Van Wieringen, A. and Warmoltz, N., On the permeation of hydrogen and helium in single
crystal silicon and germanium at elevated temperatures, Physica, 22, 849, 1956.
10. Newman, R.C. et al., Dipole moments of H
2
, D
2
, and HD molecules in Czochralski silicon,
Phys. Rev. B, 60, 12775, 1999.
11. Johnson, N.M. and Moyer, M.D., Absence of oxygen diffusion during hydrogen passivation of
shallow-acceptor impurities in single-crystal silicon, Appl. Phys. Lett., 46, 787, 1985.
12. Seager, C.H., Hydrogenation methods, in Hydrogen in Semiconductors (Semiconductors and
Semimetals, Vol. 34), Pankove, J.I. and Johnson, N.M. (Eds.), Academic Press, Boston, 1991,
Chap. 2, p. 17.
13. Seager, C.H., Anderson, R.A., and Panitz, J.K.G., The diffusion of hydrogen in silicon and
mechanisms for ‘‘unintentional’’ hydrogenation during ion beam processing, J. Mater. Res., 2,
96, 1987.
14. Tavendale, A.J., Williams, A.A., and Pearton, S.J., Hydrogen injection and neutralization of
boron acceptors in silicon boiled in water, Appl. Phys. Lett., 48, 590, 1986.
15. Stavola, M., Hydrogen diffusion and solubility in C-Si, in Properties of Crystalline Silicon, Hull, R.
(Ed.), Section 9.8, EMIS Data Review No. 20, INSPEC, IEE, London, 1999, p. 511.
16. Buda, F. et al., Proton diffusion in crystalline silicon, Phys. Rev. Lett., 63, 294, 1989.
17. Blochl, P.E., Van de Walle, C.G., and Pantelides, S.T., First-principles calculations of diffusion
coefficients: Hydrogen in silicon, Phys. Rev. Lett., 64, 1401, 1990.
18. Panzarini, G. and Colombo, L., Hydrogen diffusion in crystalline silicon: A tight-binding
molecular dynamics study, Phase Transitions, B52, 137, 1994.
19. Gorelkinskii, Yu.V. and Nevinnyi, N.N., EPR of interstitial hydrogen in silicon: Uniaxial stress
experiments, Mater. Sci. Eng. B, 36, 133, 1996.
20. Kamiura, Y., Yoneta, M., and Hashimoto, F., Hydrogen diffusivities below room temperature in
silicon evaluated from the photoinduced dissociation of hydrogen-carbon complexes, Appl. Phys.
Lett., 59, 3165, 1991.
21. Mogro-Campero, A., Love, R.P., and Schubert, R., Drastic changes in the electrical resistance of
gold-doped silicon produced by a hydrogen plasma, J. Electrochem. Soc., 132, 2006, 1985.
22. Newman, R.C. et al., Hydrogen diffusion and the catalysis of enhanced oxygen diffusion in
silicon at temperatures below 5008C, J. Appl. Phys., 70, 3061, 1991.
23. Zundel, T. and Weber, J., Trap-limited hydrogen diffusion in boron-doped silicon, Phys. Rev. B,
46, 2071, 1992.
24. Johnson, N.M. and Herring, C., Diffusion of negatively charged hydrogen in silicon, Phys. Rev. B,
46, 15554, 1992.
25. Ichimiya, T. and Furuichi, A., On the solubility and diffusion coefficient of tritium in single
crystals of silicon, Int. J. Appl. Radiat. Isot., 19, 573, 1968.
26. Binns, M.J. et al., Hydrogen solubility in silicon and hydrogen defects present after quenching,
Semicond. Sci. Technol., 8, 1908, 1993.
27. Van de Walle, C.G. et al., Theory of hydrogen diffusion and reactions in crystalline silicon,
Phys. Rev. B, 39, 10791, 1989.
28. Stein, H.J., Vacancies and the chemical trapping of hydrogen in silicon, Phys. Rev. Lett., 43, 1030,
1979.
29. Budde, M., Hydrogen-related defects in proton-implanted silicon and germanium, PhD thesis,
University of Aarhus, Denmark, 1998.
30. Budde, M. et al., Vibrational lifetime of bond-center hydrogen in crystalline silicon, Phys. Rev.
Lett., 85, 1452, 2000.
31. Holm, B., Bonde Nielsen, K., and Bech Nielsen, B., Deep state of hydrogen in crystalline silicon:
Evidence for metastability, Phys. Rev. Lett., 66, 2360, 1991.
32. Gorelkinskii, Yu.V. and Nevinnyi, N.N., Electron paramagnetic resonance of hydrogen in
silicon, Physica B, 170, 155, 1991.
33. Bech Nielsen, B., Bonde Nielsen, K., and Byberg, J.R., EPR experiments on hydrogen-
implanted silicon crystals: Annealing properties of bond center hydrogen, Mater. Sci. Forum,
143–147, 909, 1994.
34. Bonde Nielsen, K. et al., Bond-centered hydrogen in silicon studied by in situ deep-level transient
spectroscopy, Phys. Rev. B, 60, 1716, 1999.
ß 2008 by Taylor & Francis Group, LLC.
35. Bonde Nielsen, K. et al., Electronic levels of isolated and oxygen-perturbed hydrogen in silicon
and migration of hydrogen, Physica B, 308–310, 134, 2001.
36. Johnson, N.M., Herring, C., and Van de Walle, C.G., Inverted order of acceptor and donor levels
of monatomic hydrogen in silicon, Phys. Rev. Lett., 73, 130, 1994.
37. Johnson, N.M., Herring, C., and Van De Walle, C.G., Reply to comment on inverted order of
acceptor and donor levels of monatomic hydrogen in silicon. Phys. Rev. Lett., 74, 4566, 1995.
38. Bonde Nielsen, K. et al., Acceptor state of monoatomic hydrogen in silicon and the role of
oxygen, Phys. Rev. B, 65, 075205, 2002.
39. Pereira, R.N. et al., Antibonding configurations of hydrogen in silicon-germanium alloys,
Phys. Rev. B, 73, 085201, 2006.
40. Sah, C.T, Sun, Y.-C.J., and Tzou, J.-T.J., Deactivation of the boron acceptor in silicon by
hydrogen, Appl. Phys. Lett., 43, 204, 1983.
41. Pankove, J.I. et al., Neutralization of shallow acceptor levels in silicon by atomic hydrogen,
Phys. Rev. Lett., 51, 2224, 1983.
42. Stavola, M. et al., Vibrational characteristics of acceptor-hydrogen complexes in silicon,
Appl. Phys. Lett., 50, 1086, 1987.
43. Skudlik, H. et al., Influence of electronic parameters on the electric-field gradients induced by H
at the probe atom
111
In=
111
Cd in Si, Phys. Rev. B, 46, 2159, 1992.
44. Bech Nielsen, B., Andersen, J.U., and Pearton S.J., Lattice location of deuterium interacting with
the boron acceptor in silicon, Phys. Rev. Lett., 60, 321, 1988.
45. Chang, K.J. and Chadi, D.J., Theory of hydrogen passivation of shallow-level dopants in
crystalline silicon, Phys. Rev. Lett., 60, 1422, 1988.
46. Zundel, T. and Weber, J., Dissociation energies of shallow-acceptor–hydrogen pairs in silicon,
Phys. Rev. B, 39, 13549, 1989.
47. Zhu, J., Johnson, N.M., and Herring, C., Negative-charge state of hydrogen in silicon, Phys. Rev. B,
41, 12354, 1990.
48. Pearton S.J. and Lopata J., Dissociation of P-H, As-H, and Sb-H complexes in n-type Si,
Appl. Phys. Lett., 59, 2841, 1991.
49. Benton, J.L. et al., Hydrogen passivation of point defects in silicon, Appl. Phys. Lett., 36, 670, 1980.
50. Pearton, S.J. and Haller E.E., Palladium- and platinum-related levels in silicon: Effect of a
hydrogen plasma, J. Appl. Phys., 54, 3613, 1983.
51. Pearton, S.J. and Tavendale, A.J., Hydrogen passivation of gold-related deep levels in silicon,
Phys. Rev. B, 26, 7105, 1982.
52. Bonde Nielsen, K. et al., Deep levels of vacancy-hydrogen centers in silicon studied by Laplace
DLTS, Physica B, 273, 167, 1999.
53. Bech Nielsen, B. et al., Identification of the silicon vacancy containing a single hydrogen atom by
EPR, Phys. Rev. Lett., 79, 1507, 1997.
54. Stallinga, P. et al., Electron paramagnetic resonance study of hydrogen-vacancy defects in
crystalline silicon, Phys. Rev. B, 58, 3842, 1998.
55. Bech Nielsen, B. et al., H interacting with intrinsic defects in Si, Mater. Sci. Forum, 196–201, 933,
1995.
56. Chen, W.M. et al., Microscopic identification and electronic structure of a di-hydrogen-vacancy
complex in silicon by optical detection of magnetic resonance, Phys. Rev. Lett., 64, 3042, 1990.
57. Stallinga, P. and Bech Nielsen, B., Comment on microscopic identification and electronic struc-
ture of a di-hydrogen–vacancy complex in silicon by optical detection of magnetic resonance,
Phys. Rev. Lett., 80, 422, 1998.
58. Johannesen, P. et al., Silicon vacancy containing two hydrogen atoms studied with electron
paramagnetic resonance and infrared absorption spectroscopy, Phys. Rev. B, 66, 235201, 2002.
59. Markevich, V.P., unpublished data.
60. Markevich, V.P. et al., Observation and theory of the VOH
2
complex in silicon, Phys. Rev. B, 61,
12964, 2000.
61. Irmscher, K., Klose, H., and Maass, K., Hydrogen-related deep levels in proton-bombarded
silicon, J. Phys. C, 17, 6317, 1984.
62. Peaker, A.R. et al., Hydrogen reactions with electron irradiation damage in silicon, Physica B,
273–274, 243, 1999.
ß 2008 by Taylor & Francis Group, LLC.
63. Feklisova, O.V. et al., On the nature of hydrogen-related centers in p-type irradiated silicon,
Physica B, 308–310, 210, 2001.
64. Johannesen, P., Bech Nielsen, B., and Byberg, J.R., Identification of the oxygen-vacancy defect
containing a single hydrogen atom in crystalline silicon, Phys. Rev. B, 61, 4659, 2000.
65. Coutinho, J. et al., Effect of stress on the energy levels of the vacancy-oxygen-hydrogen complex
in Si, Phys. Rev. B, 68, 184106, 2003.
66. Dobaczewski, L. et al., Piezospectroscopic studies of the re-orientation process of defect com-
plexes, Physica B, 340–342, 499, 2003.
67. Andersen, O. et al., Electrical activity of carbon-hydrogencenters inSi, Phys. Rev. B, 66, 235205, 2002.
68. Endrös, A., Charge-state-dependent hydrogen-carbon-related deep donor in crystalline silicon,
Phys. Rev. Lett., 63, 70, 1989.
69. Kamiura, Y. et al., Electronic state, atomic configuration and local motion of hydrogen around
carbon in silicon, Defect Diffus. Forum, 183, 25, 2000.
70. Hoffmann, L. et al., Weakly bound carbon-hydrogen complex in silicon, Phys. Rev. B, 61, 16659,
2000.
71. Jones, R. et al., The interaction of hydrogen with deep level defects in silicon, Solid State Phen.,
71, 173, 2000.
72. Yarykin, N. et al., Silver-hydrogen interactions in crystalline silicon, Phys. Rev. B, 59, 5551, 1999.
73. Sveinbjörnsson, E.Ö. and Engström O., Reaction kinetics of hydrogen-gold complexes in silicon,
Phys. Rev. B, 52, 4884, 1995.
74. Deixler, P. et al., Laplace-transform deep-level transient spectroscopy studies of the G4 gold-
hydrogen complex in silicon, Appl. Phys. Lett., 73, 3126, 1998.
75. Rubaldo, L. et al., Gold-hydrogen complexes in silicon, Mater. Sci. Eng. B, 58, 126, 1999.
76. Knack, S. et al., Evolution of copper-hydrogen-related defects in silicon, Physica B, 308–310, 404,
2001.
77. Sadoh, T. et al., Deep levels of vanadium- and chromium-hydrogen complexes in silicon,
Mater. Sci. Forum, 143–147, 939, 1994.
78. Jost, W., Weber, J., and Lemke, H., Hydrogen-induced defects in cobalt-doped n-type silicon,
Semicond. Sci. Technol., 11, 22, 1996.
79. Feklisova, O.V. et al., Dissociation of iron-related centers in Si stimulated by hydrogen,
Mater. Sci. Eng. B, 71, 268, 2000.
80. Shiraishi, M. et al., DLTS analysis of nickel–hydrogen complex defects in silicon, Mater. Sci. Eng. B,
58, 130, 1999.
81. Sachse, J.-U., Weber, J., and Lemke, H., Deep-level transient spectroscopy of Pd-H complexes in
silicon, Phys. Rev. B, 61, 1924, 2000.
82. Jost, W. and Weber, J., Titanium-hydrogen defects in silicon, Phys. Rev. B, 54, R11038, 1996.
83. Sadoh, T., Nakashima, H., and Tsurushima, T., Deep levels of vanadium and vanadium-
hydrogen complex in silicon, J. Appl. Phys., 72, 520, 1992.
84. Weinstein, M.G. et al., Pt-H complexes in Si: complementary studies by vibrational and capaci-
tance spectroscopies, Phys. Rev. B, 65, 035206, 2001.
85. Kamiura, Y. et al., Effect of uniaxial stress on the electronic state of a platinum-dihydrogen
complex in silicon and charge-state-dependent motion of hydrogen during stress-induced
reorientation, Phys. Rev. B, 69, 045206, 2004.
86. Kolkovsky, V. et al., Electrical activity of the PtH
2
complex in silicon: High-resolution Laplace
deep-level transient spectroscopy and uniaxial-stress technique, Phys. Rev. B, 73, 195209, 2006.
87. Resende, A. et al., Theory of gold-hydrogen complexes in silicon, Mater. Sci. Forum, 258–263, 295,
1997.
88. Hourahine, B. et al., Platinum and gold dihydrides in silicon, Physica B, 340–342, 668, 2003.
89. Evans M.L. et al., Vibrational spectroscopy of defect complexes containing Au and H in Si,
Mater. Sci. Eng. B, 58, 118, 1999.
90. Huy, P.T. and Ammerlaan, C.A.J., Complexes of gold and platinum with hydrogen in silicon,
Physica B, 302–303, 233, 2001.
91. Huy, P.T. and Ammerlaan, C.A.J., Electronic and atomic structure of transition-metal-hydrogen
complexes in silicon, Physica B, 308–310, 408, 2001.
ß 2008 by Taylor & Francis Group, LLC.
92. Höhne, M. et al., EPR spectroscopy of platinum-hydrogen complexes in silicon, Phys. Rev. B, 49,
13423, 1994.
93. Williams, P.M. et al., Structure-sensitive spectroscopy of transition-metal-hydrogen complexes
in silicon, Phys. Rev. Lett., 70, 3816, 1993.
94. Uftring, S.J. et al., Microscopic structure and multiple charge states of a PtH
2
complex in Si,
Phys. Rev. B, 51, 9612, 1995.
95. Sachse, J.-U. et al., Electrical properties of platinum-hydrogen complexes in silicon, Phys. Rev. B,
55, 16176, 1997.
96. Mainwood, A. and Stoneham, A.M., Interstitial muons and hydrogen in crystalline silicon,
Physica, 116B, 101, 1983.
97. Corbett, J.W. et al., Atomic and molecular hydrogen in the Si lattice, Phys. Lett., 93A, 303, 1983.
98. Murakami, K. et al., Hydrogen molecules in crystalline silicon treated with atomic hydrogen,
Phys. Rev. Lett., 77, 3161, 1996.
99. Leitch, A.W.R., Alex, V., and Weber, J., H
2
molecules in c-Si after hydrogen plasma treatment,
Solid State Commun., 105, 215, 1998.
100. Leitch, A.W.R., Alex, V., and Weber, J., Raman spectroscopy of hydrogen molecules in crystal-
line silicon, Phys. Rev. Lett., 81, 421, 1998.
101. Pritchard, R.E. et al., Isolated interstitial hydrogen molecules in hydrogenated crystalline silicon,
Phys. Rev. B, 57, R15048, 1998.
102. Okamoto, Y., Sato, M., and Oshiyama, A., Comparative study of vibrational frequencies of H
2
molecules in Si and GaAs, Phys. Rev. B, 56, R10016, 1997.
103. Van de Walle, C.G., Energetics and vibrational frequencies of interstitial H
2
molecules in
semiconductors, Phys. Rev. Lett., 80, 2177, 1998.
104. Hourahine, B. et al., Hydrogen molecules in silicon located at interstitial sites and trapped in
voids, Phys. Rev. B, 57, R12666, 1998.
105. Estreicher, S.K. et al., Dynamics of interstitial hydrogen molecules in crystalline silicon, J. Phys.:
Condens. Matter, 13, 6271, 2001.
106. Chen, E.E. et al., Key to understanding interstitial H
2
in Si, Phys. Rev. Lett., 88, 105507, 2002.
107. Chen, E.E. et al., Rotation of molecular hydrogen in Si: unambiguous identification of ortho-H
2
and para-D
2
, Phys. Rev. Lett., 88, 245503, 2002.
108. Fowler, W.B., Walters, P., and Stavola, M., Dynamics of interstitial H
2
in crystalline silicon,
Phys. Rev. B, 66, 075216, 2002.
109. Shi, G.A. et al., Rotational-vibrational transitions of interstitial HD in Si, Phys. Rev. B, 72, 085207,
2005.
110. Lavrov, E.V. andWeber, J., Ortho andpara interstitial H
2
insilicon, Phys. Rev. Lett., 89, 215501, 2002.
111. Hiller, M., Lavrov, E.V., and Weber, J., Raman scattering study of H
2
in Si, Phys. Rev. B, 74,
235214, 2006.
112. Hiller, M., Lavrov, E.V., and Weber, J., Ortho–para conversion of interstitial H
2
in Si, Phys. Rev.
Lett., 98, 055504, 2007.
113. Pritchard, R.E. et al., Interactions of hydrogen molecules with bond-centered interstitial oxygen
and another defect center in silicon, Phys. Rev. B, 56, 13118, 1997.
114. Markevich, V.P. and Suezawa, M., Hydrogen-oxygen interaction in silicon at around 508C,
J. Appl. Phys., 83, 2988, 1998.
115. Ishioka, K. et al., Hydrogen molecules trapped by multivacancies in silicon, Phys. Rev. B, 60,
10852, 1999.
116. Weber, J. et al., Properties of hydrogen-induced voids in silicon, J. Phys.: Condens. Matter, 17,
S2303, 2005.
117. Chang, K.J. and Chadi, D.J., Diatomic-hydrogen-complex diffusion and self-trapping in crystal-
line silicon, Phys. Rev. Lett., 62, 937, 1989.
118. Holbech, J.D. et al., H
2
* defect in crystalline silicon, Phys. Rev. Lett., 71, 875, 1993.
119. Estreicher, S.K., Roberson, M.A., and Maric, Dj.M., Hydrogen and hydrogen dimers in c-C, Si,
Ge and a-Sn, Phys. Rev. B, 50, 17018, 1994.
120. Estreicher, S.K., Hastings, J.L., and Fedders, P.A., Radiation-induced formation of H
2
* in silicon,
Phys. Rev. Lett., 82, 815, 1999.
ß 2008 by Taylor & Francis Group, LLC.
121. Weldon, M.K. et al., On the mechanism of the hydrogen induced exfoliation of silicon, J. Vac. Sci.
Technol. B, 15, 1065, 1997.
122. Reboredo, F.A., Ferconi M., and Pantelidies S.T., Theory of the nucleation, growth and structure
of hydrogen-induced extended defects in silicon, Phys. Rev. Lett., 82, 4870, 1999.
123. Lavrov, E.V. and Weber, J., Evolution of hydrogen platelets in silicon determined by polarised
Raman spectroscopy, Phys. Rev. Lett., 87, 185502, 2001.
124. Muto, S., Takeda, S., and Hirata, M., Hydrogen-induced platelets in silicon studied by transmis-
sion electron microscopy, Philos. Mag., A72, 1057, 1995.
125. Nickel, N.H. et al., Nucleation of hydrogen-induced platelets in silicon, Phys. Rev. B, 62, 8012,
2000.
126. Pearton, S.J., Corbett, J.W., and Shi, T.S., Hydrogen in crystalline semiconductors, Appl. Phys. A,
43, 153, 1987.
127. Frank, R.C. and Thomas Jr. J.E., The diffusion of hydrogen in single-crystal germanium, J. Phys.
Chem. Solids, 16, 144, 1960.
128. Hansen, W.L., Haller, E.E., and Luke, P.N., Hydrogen concentration and distribution in high-
purity germanium crystals, IEEE Trans. Nucl. Sci., 29, 738, 1982.
129. Hall, R.N., HP Ge: purification, crystal growth, and annealing properties, IEEE Trans. Nucl. Sci.,
31, 320, 1984.
130. Stutzmann, M. et al., A comparison of hydrogen incorporation and effusion in doped crystalline
silicon, germanium, and gallium arsenide, Appl. Phys. A, 53, 47, 1991.
131. Van de Walle, C.G. and Neugebauer, J., Universal alignment of hydrogen levels in semicon-
ductors, insulators and solutions, Nature, 423, 626, 2003.
132. Budde, M. et al., Local vibrational modes of isolated hydrogen in germanium, Phys. Rev. Lett., 85,
2965, 2000.
133. Dobaczewski, L. et al., Donor level of bond center hydrogen in germanium, Phys. Rev. B, 69,
245207, 2004.
134. Hiller, M., Lavrov, E.V., and Weber, J., Hydrogen-induced platelets in Ge determined by Raman
scattering, Phys. Rev. B, 71, 045208, 2005.
135. Hiller, M. et al., Interstitial H
2
in germanium by Raman scattering and ab initio calculations,
Phys. Rev. B, 72, 153201, 2005.
136. Budde, M. et al., Local modes of the H
2
* dimer in germanium, Phys. Rev. B, 54, 5485, 1996.
137. Budde, M. et al., Vacancy-hydrogen complexes in group-IV semiconductors, Physica B, 273–274,
208, 1999.
138. Budde, M. et al., Identification of the hydrogen-saturated self-interstitials in silicon and germa-
nium, Phys. Rev. B, 57, 4397, 1998.
139. Pearton, S.J. and Tavendale, A.J., Hydrogen passivation of deep metal-related donor centers in
germanium, J. Appl. Phys., 54, 820, 1983.
140. Pearton, S.J. et al., Deuterium in germanium; interaction with point defects, J. Appl. Phys., 55,
1464, 1984.
141. Haller, E.E., Hubbard, G.S., and Hansen, W.L., Hydrogen-multivalent acceptor complexes in
high-purity germanium, IEEE Trans. Nucl. Sci., 24, 48, 1977.
142. Haller, E.E. et al., Deep level transient spectroscopy of high-purity germanium diodes=detectors,
IEEE Trans. Nucl. Sci., 26, 265, 1979.
143. Pearton, S.J., Hydrogen passivation of copper-related defects in germanium, Appl. Phys. Lett., 40,
253. 1982.
144. Haller, E.E., Hansen, W.L., and Goulding, F.S., Physics of ultrapure germanium, Adv. Phys., 30,
93, 1981.
145. Haller, E.E., Isotope shifts in the ground state of shallow, hydrogenic centers in pure germanium,
Phys. Rev. Lett., 40, 584, 1978.
146. Haller, E.E., Joos, B., and Falicov, L.M., Acceptor complexes in germanium: systems with
tunneling hydrogen, Phys. Rev. B, 21, 4729, 1980.
147. Claeys, C. et al. (Eds.), Proceedings of symposium T, E-MRS 2006 Spring Meeting on Germa-
nium based semiconductors from materials to devices, Mater. Sci. Semicond. Process., 9, 407, 2006.
148. Tong, Q.-Y. et al., Layer splitting process in hydrogen-implanted Si, Ge, SiC, and diamond
substrates, Appl. Phys. Lett., 70, 1390, 1997.
ß 2008 by Taylor & Francis Group, LLC.
149. Bruel, M., Silicon on insulator material technology, Electron. Lett., 31, 1201, 1995.
150. Stoneham, A.M. and Jain, S.C. (Eds.), GeSi Strained Layers and Their Applications, Institute of
Physics, Bristol, UK, 1995.
151. Grimmeiss, H.G., Silicon–germanium—a promise into the future, Semiconductors, 33, 939, 1999.
152. King, C.A. et al., Si=Si
1Àx
Ge
x
heterojunction bipolar transistors produced by limited reaction
processing, IEEE Electron Device Lett., 10, 52, 1989.
153. Konig, U. and Dambkes, H., SiGe HBTs and HFETs, Solid-State Electron., 38, 1595, 1995.
154. Shäffler, F., High mobility Si and Ge structures, Semicond. Sci. Technol., 12, 1515, 1997.
155. Pearsall, T.P. et al., Spectroscopy of band-to band optical transition in SiGe alloys and super-
lattices, Phys. Rev. B, 57, 9128, 1989.
156. Balsas, A. et al., Vibrational properties of elemental hydrogen centers in Si, Ge and dilute SiGe
alloys, J. Phys.: Condens. Matter, 17, S2155, 2005.
157. Pereira, R.N., Dobaczewski, L., and Bech Nielsen, B., Infrared absorption studies in proton- and
deuterium-implanted Si
1Àx
Ge
x
, Physica B, 340–342, 803, 2003.
158. Pereira, R.N. et al., Local modes of bond-centered hydrogen in Si:Ge and Ge:Si, Phys. Rev. B, 71,
195201, 2005.
159. Pereira, R.N. et al., Hydrogen self-trapping near silicon atoms in Ge-rich SiGe alloys, Appl. Phys.
Lett., 88, 142112, 2006.
160. Bonde Nielsen, K. et al., Bond-center hydrogen in dilute Si
1Àx
Ge
x
alloys: Laplace deep-level
transient spectroscopy, Phys. Rev. B, 68, 045204, 2003.
161. Aubry, J.C. et al., First-shell bond lengths in Si
x
Ge
1Àx
crystalline alloys, Phys. Rev. B, 59, 12872,
1999.
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
3
Defects in Strained-Si MOSFETs
Yongke Sun and Scott E. Thom pson
CONT ENTS
3.1 Intr oductio n ............... ............... ............... .. ............... ............... ............... ..... ............... .......... 57
3.2 Stat e of the Art: Stra ined-S i MOSFE Ts .. ............... ............... ............... ..... ............... .......... 57
3.3 Yie ld of Integ rated Circu its ............... ...... ............... ............... ............... ..... ............... .......... 58
3.4 De fects in Stra ined Layers ............... ........ ............... ............... ............... ..... ............... .......... 61
3.5 Stra ined-Lay er Critical Thickness and Stra in Relaxati on ............... ...... ............... .......... 64
3.6 Proce ss Flow ............... ............... ............... . ............... ............... ............... ..... ............... .......... 65
3.7 Stra in and Alternati ve Wafe r Orie ntation ............... ............... ............... .. ............... .......... 67
Refere nces ......... ............... ............... ............... ..... ............... ............... ............... ..... ............... .......... 67
3.1 I ntro du ction
Strain ed-Si channel metal-ox ide-semico nductor field-ef fect transis tors (MO SFETs) are use d
in nearly all 90 nm and smaller comme rcial log ic technol ogies. Stra ined chann els improve
the elect ron and ho le motilities by alteri ng the semi conduct or band structure. Stra ined
semico nductor layers are kno wn to produ ce dislocat ion defects and have been exten sively
studi ed on blan ket wafers on whi ch several ma terials science rev iew pape rs have rec ently
been publis hed [1 –4]. Howe ver, muc h less has bee n pub lished on strained- Si defe cts
relating to manufact uring comme rcial produ ction tech nologi es whi ch we address in this
chapte r. Th e chapte r is outlin ed as fo llows: Secti on 3 .2 brie fl y reviews strai ned-Si
MOSFE Ts in st ate-of-th e-art produ ction technol ogies, Secti on 3.3 cov ers the yie ld of inte-
grated circuits, Se ction 3.4 fo cuses on defects in st rained- Si layers, Se ction 3.5 dis cusses
strai n relaxa tion, Secti on 3. 6 descri bes proce ss flow thermal cy cles and manu facturi ng
trade offs, and Se ction 3.7 brie fly int roduces the potenti al bene fi ts and probl ems involvi ng
incorporating strain in alternative wafer orientations.
3.2 State of the Art: Strained-Si MOSFETs
In this section, we briefly describe techniques used in commercial 90 and 45 nm logic
technologies to introduce uniaxial stress into the Si channel; more process details are
published elsewhere [5,6]. The techniques in production include high stress tensile and
ß 2008 by Taylor & Francis Group, LLC.
com pressive SiN capping layers, and selective ep itaxial Si Ge deposite d in recessed =raise d
source s and drains (S=D). Examp les of the se prod uction techniq ues are shown in Figu re 3.1,
where the transmis sion electro n micr ograph s (TEM ) are obtain ed by Chip works [7].
Future generations wi ll bring the SiGe closer to the chann el, increa se the Ge concen tration
[8 –11], and inc rease the stress in the cappi ng layers , which is approach ing 3 GP a. Oth er
poss ible fut ure tech niques fo r proces s stress or mob ility enhance ment may includ e tensil e
shallo w trench oxide, or embed ded SiC [12] fo r nMO SFETs and hybri d orientat ed (110)
wafers [13]. Together, these technique s are expected to lead to chann el stresses of
$ 1–2 GPa, approach ing the st ress level of wafer- based biaxi al stress [14].
The common defects in strained layers are misfit dislocations induced by film strain
relaxation. To minimize defect density, the process-induced strain, unlike wafer-based strain,
is introduced late in the process flow. Epitaxial SiGe is introduced postisolation and gate
formation by etching the S=D and growing selective epitaxial SiGe [8,15,16]. The high stress
capping layers are introduced even later postsalicide formation. In some cases when the high
stress capping layers for stress memorization are introduced before S=D anneal [17,18], they
can cause drain shorting defects at S=D anneals at elevated temperatures.
3.3 Yield of Integrated Circuits
Before inv estigati ng the defe cts in strained- Si in Secti on 3.4, it is fi rst help ful to unde rstand
how these defects alter yield. Yield is one of the key factors in determining the success of a
production technology, or a technology option like strained-Si, since yield directly affects
the die cost of an integrated circuit. Yield considerations have played, and will continue to
130
250
500
750
1000
90 65 45 Time
Node (nm)
S
t
r
e
s
s

(
M
P
a
)
SiGe
AMD/Intel uP
IBM/Xbox-360
NMOS PMOS
Applied materials:
1.8–3.0 GPa films
TI/cell phone
60 nm
FIGURE 3.1
Various structures to introduced strain into the channel of MOSFETs, e.g., embedded SiGe in S=D, and stress liners.
(FromChipworks Report. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
play, a key role in determini ng which technol ogies are com merciall y adopte d. To under-
stand chi p yield, it is nec essary to understan d that, for a logic die to yield (i.e., be full y
functi onal and sellabl e), eve ry single logic transis tor, which typic ally numb ers in the
hund reds of millio n per chip, needs to work. This is unli ke mem ory chips where redundant
mem ory blocks are added and fuse s set to repair nonyi elding bits, ro ws, columns, or
even blocks . Figure 3.2 shows some typical die photos and transi stor devi ce cross
sectio ns that are fab ricated usi ng st rained- Si. Th e chips serve the three domi nant marke ts
of com puter, com munica tion, and consumer elect ronic s, and high light the wide succe ss of
this tech nology in not only high perform ance but also low- cost marke ts.
How st rained- Si defec ts affect yield dep ends on how suscepti ble yie ld is to the defe cts.
Yield is a function of defect density and amoun t of critica l chip are a sensi tive to the
defects. The critica l area is de fined as the chi p area in which, if a defect occ urs, it will
cause a no nfunctional device. Using the example of strained- Si, this importan t con cept of
critica l area will become clearer and will help to unde rstand the choice s m ade for the fi rst
three gene rations of strained- Si. For st rained- Si devices, the dominan t failure modes are
mis fit dis locations and stack ing fau lts. The strain fi eld aro und the dislocat ions and stackin g
faults captur es m etal impu rity atoms and increa ses the dopant diffusion by several or ders
of magni tude [19]. Stacking fau lts and dislocat ions that a re containe d entirel y wi thin the
neutr al space charge regio n of the S= D junct ions do not impact transis tor yield, while those
exten ding fro m source to dr ain cause high leakage current (see Figu res 3. 3 and 3.4) and
nonfunctional transistors. Some levels of dislocations and stacking faults are present and
can be found in commercial chips on the market, as seen in the Chipworks report on Intel’s
90 nm strai ned-Si tech nology. Figu re 3.5 shows a stackin g fault in the S =D on a comme rcial
chip that does not degrade the functionality.
Defects will have a more deleterious impact on the yield as the channel length of the
MOSFET is scaled. At shorter channel length, the dislocations or stacking faults are more
likely to penetrate through the source to the drain and cause a nonfunctional device. Due to
the deleterious nature of these defects, study of these defects is an important subject for
improving the quality of strained-Si in high-volume manufacturing.
90 nm 65 nm 45 nm
~400 M Transistors/chip ~400 M Transistor/chip ~100 M Transistor/chip
microprocessor
FIGURE 3.2
Examples of logic chips fabricated with strained-Si channel MOSFETs: Intel and AMD microprocessors.
ß 2008 by Taylor & Francis Group, LLC.
Island
Silicon island
Buried oxide
Substrate
Dislocations
100 nm
Gate
Stacking fault
FIGURE 3.3
Examples of stacking fault and dislocation defects between source and drain of a MOSFET. (The left panel of this
figure is reprinted with permission from Yang, J., Neudeck, G.W., and Denton, J.P., Appl. Phys. Lett., 77, 20, 2000.
The right panel of this figure is reprinted with permission from Sleight, J.W., Chuan Lin, and Gula, G.J., IEEE
Electron Devices Lett., 20, 5, 1999.)
0.0
10
Ϫ11
10
Ϫ10
10
Ϫ9
10
Ϫ8
10
Ϫ7
10
Ϫ6
10
Ϫ5
10
Ϫ4
10
Ϫ3
10
Ϫ2
0.5 1.0
Normal device
V
ds
= 50 mV
V
ds
= 2.0 V
Anomalous device V
sub
= Ϫ10 V
Anomalous device V
sub
= 0 V
Vg (V)
D
r
a
i
n

c
u
r
r
e
n
t

(
A
)
1.5 2.0
FIGURE 3.4
Current versus voltage curve for a normal and MOSFET with a source to drain defect. (Reprinted with permission
from Sleight, J.W., Chuan Lin, and Gula, G.J., IEEE Electron Device Lett., 20, 5, 1999.)
FIGURE 3.5
Examples of stacking faults in embedded
SiGe source and drain. (From Pentium 4
prescott Microprocessor report, www.chip-
works.com.)
Neutral junction SiGe
Active P+
junction
STI
Dislocation
ß 2008 by Taylor & Francis Group, LLC.
3.4 D ef ects in St rained Layers
Mis fit dislocat ions and stackin g faults in the S =D Si Ge epit axy can be generated duri ng
epitaxi al growth and thermal process es due to st rain relaxa tion. The mis fi t dislocat ions a re
one- dimension al thread ing defects, while the stackin g fau lts are two-dime nsional planar
defects. Th ey are differe nt yet c losely related to each ot her. The formati on of threadi ng
dislocat ions and stacking faults is strongly in fluenced by seve ral factors , includ ing st rain
level, latt ice temperat ure, and implantat ion dam age. Thread ing dis locations are fur ther
classi fied into two types , edge and screw dislocat ions, as sh own in Figure 3.6. A dislocat ion
can be visualize d by imagining cuttin g a crystal along a plane and slipp ing one half across
the othe r by a lattice vect or. The halves will fi t back tog ether withou t leavi ng a defect. But if
the cut only goes part way through the crys tal, the boundar y of the cut will leave a
dislocat ion. Th e boundar y of the cut is the dislocat ion line ; the vect or of the slip is the
Burgers vect or. When the Burgers vect or is perp endicu lar (paral lel) to the dislocat ion
line, the dis location is called edge (scre w) dislocat ion. As we will see in the followi ng
discuss ion, the dominan t dislocat ion ty pe in strained- Si is neither pure edge nor screw, but
is mixed.
A mix ed-type dis location is bes t visua lized by the concep t of an edg e dislocat ion. An
edge dis location is forme d by slipp ing one half of the crys tal in the dire ction perp endicular
to the cuttin g plane. Then the defe ct is create d as if one plane of atoms is inserte d to a
regul ar cry stal and ends at the defe ct line as show n in Figure 3.7 . He re, even thou gh
there exists one extra plane of atoms in half of the crystal , exc ept for the nea rby are a of the
dislocat ion, the atoms still maintai n the stackin g order of crys tal. Th us, the extra half plane
is not a pl anar defe ct, in con trast to the stacking fault- induced edge dislocat ion that will
be introduce d next. It is easy to see that a dislocat ion line cannot end inside the crystal .
Disloca tions either end at the surfa ce or form a ring insi de the crystal . Most dislocat ions in
strai ned-Si devices termin ate at bot h ends of the shallo w trench isolation (STI) that
surround s all devices. Likewi se, in the proces s of crys tal gro wth, the dislocat ions in the
substrat e ten d to propag ate into the newly gr own cry stal unle ss tech niques are used to
termin ate the defects on the sides [4,20].
The stacking fault, just as its name implies, is caused by wrong stacking order of atoms
in the process of crystal or film growth. A face-centered cubic (FCC) crystal lattice is
forme d by clos e packing of the atoms , which are shown in Figure 3. 8a and b, where the
close-packed plane is the (111) plane. If we take the base plane as the A-plane, the second
(a) (b)
FIGURE 3.6
Two types of dislocations: (a) edge and (b) screw dislocation.
ß 2008 by Taylor & Francis Group, LLC.
plane as the B-plane, then the atoms can take either A or C configuration in the third plane.
The ABCABC. . . sequence of atomarray forms the face-centered cubic (FCC) crystal lattice.
A stacking fault occurs when a whole or part of one (111) plane is absent. When this
happens, the stacking sequence becomes, for example, ABCABABABC. . . or ABCABA-
CABC. . . The former sequence is shown in Figure 3.8c. Normally, a stacking fault is formed
by part of the absent (111) plane, and is bordered by an edge dislocation. This dislocation is
different from the dislocations introduced earlier, and is a partial dislocation, since its
Burgers vector is not the translation vector of the crystal lattice. However, a perfect
FIGURE 3.7
A 608 dislocation in Si. The dislocation is created in
crystal structure along the line where the extra half
plane ends.
<111>
<110>
FIGURE 3.8
FCCcrystal latticeandtheformationof
a stacking fault. (a) and (b) show
the FCC stacking sequence, and
(c) shows howa stacking fault forms.
(a)
A
B
C
A
B
A
B
C
A
(c)
(b)
A A
A A A
A A
B
B
B
C
C
C
A
B
C
A
B
C
A
B
C
ß 2008 by Taylor & Francis Group, LLC.
dislocation may dissociate into partial bounding stacking faults under certain circumstan-
ces. Since a stacking fault is a planar defect with broken or distorted bonds extending
through a large thickness of the film, it is likely to be more deleterious to device operation
than a single threading dislocation.
The dislocations and stacking faults in SiGe=Si system exhibit quite complicated
characteristics. These defects normally can glide along a crystal plane. Understanding the
movement of defects during thermal cycles is important to diminish the impact of these
defects to yield. Because the dislocation takes place when the two halves of crystal glide over
each other on one crystal plane (shear stress exceeds the yield stress of the material), the
crystal structure determines in which plane the dislocations have the lowest energy to form.
Si has a diamond structure, and is formed by two FCC lattices by displacing each other one
quarter of the body diagonal of the cubic cell. All Si atoms have four nearest neighbors,
which are positioned at the vertices of a tetrahedron and connected to the center atom by
four bonds. Any two nearest atoms are along a <111> direction, and any two atoms on the
vertices are along a <110>direction. By plotting the atompositions in the Si crystal along the
<111>direction, we can see that the atoms in (111) plane are close packed in two layers and
connected with the other double layers by one bond per atom (see Figure 3.9). The inter-
actions within the double layer are strong and hard to break, but the interactions between
double layers are relatively weak. Thus, the lowest energy glide for diamond structures is in
the (111) plane. Therefore, the dislocation lines in Si are generally in the (111) plane. To
determine the direction of the dislocation line, we need to inspect the directions formed by
connecting atoms in the (111) plane. In any (111) plane, there exist three <110> directions.
They are the connecting lines between the three atoms at the vertices of a tetrahedron in
one (111) plane. Due to the two planes fitting with each other after the glide, the dislocation
lines are usually along the <110> directions, which form a 608 angle to the Burgers vector.
Hence, this type of dislocation is also called the 608 dislocation. The majority of observed
dislocations in the compressively stressed SiGe system are found to be of the 608 type. The
608 dislocation is a mixed type of dislocation that is characteristic of both edge and screw
<111>
A
C
B
FIGURE 3.9
Atom positions in Si crystal along the <111> direction. Note there are two close layers connected with the other
double layers by one bond per atom.
ß 2008 by Taylor & Francis Group, LLC.
dislocat ions. The formatio n of the 60 8 dis location is illustrat ed in Figure 3 .7. Anoth er typic al
dislocat ion line is along the < 112 > directi on in the (111) plane. The < 112 > directi ons are
line s that connect atoms in differe nt tetrahe drons. A < 112 > dis location is an edge disloca-
tion. An atom on a dis location line has one extra elect ron to form a full shell, and thus form a
dang ling bond (s ee Figure 3 .7).
A perfe ct a=2 < 110 > 60 8 dislocat ion can dissociate into two a=6 < 112 > par tial disloca-
tions, i.e., a 90 8 and a 30 8 partia l dis location. Of necess ity, a stackin g fau lt betwe en the two
par tial dis locations must also be generat ed. This dissociati on depen ds crit ically on strain
and lattice temperat ure. For (001) or ientated films unde r tension such as Si=SiGe , stackin g
fau lts may ext end throug h the st rained- Si fi lm thickne ss, whereas mo st studi es cons ider
that the partia ls are pushe d together at the inter face and the dislocat ions are perfect 60 8
dislocat ions for compr essivel y strained films such as Si Ge=Si.
3.5 Strained-Layer Critical Th ickness and Strain Relaxat ion
To induce a nd maintai n high stress in the chann el, the thickne ss of the SiGe epit axy in the
S =D usu ally reaches over 100 nm. This raises concer ns of strain relaxa tion, since for
the epit axial layers gro wn on a sub strate, there exists a critica l thi ckness h
c
, below which
the lattice mismatch can be entir ely a ccommod ated by strain, and beyo nd which, part
of the mismat ch m ust be acco mmod ated by dislocat ions. If st rain is relaxed in the Si Ge
epit axy and dislocat ions are generated in the edg e of the S =D region, they probabl y
pene trate into the channel and thu s cause large leakage . For control ling the defects, the
SiGe epilaye rs in the S=D regi on must be limit ed under the critica l thickness or treate d
with special thermal cycles.
The equilibr ium critica l thickness h
c
was studi ed by Ball and van der Merwe [21], and by
Matth ews and Blake slee [22], who found that fo r Si
1Àx
Ge
x
films gro wn on Si, the relati on
betwe en h
c
(nm) and the Ge atom fraction x
c
can be written as
x
c
¼
0:55
h
c
ln
4h
c
b

:
For 608 glide dislocations, the Burgers vector b ¼a
0
=2 <110> $0.4 nmfor Si. Obviously, the
critical thickness for SiGe filmdecreases with Ge content. For SiGe filmcontaining 20% Ge,
the critical thickness is about 14 nm.
The thickness for low-temperature molecular beam epitaxy (MBE) grown layers can
be much larger than the critical thickness without inducing any dislocations [23,24].
Figu re 3.10 shows a series of SiGe films grown at 494 8 C with differe nt Ge c ontent.
The s
exc
=m¼0 line indicates the Matthews and Blakeslee criterion for absolute stability,
where s
exc
is excess stress and m is the shear modulus. As we can see, films with much
larger thickness than the critical thickness are still fully strained. This is because normal
as-grown epilayers are far from the thermodynamic equilibrium due to the low substrate
temperature (e.g., T<0.5T
melt
). These layers are in metastable states. Strain relief needs
excess stress, which decreases with temperature [25]. For example, in Figure 3.10,
the excess stress needed at 4948C obeys s
exc
=m¼0.026. For thick SiGe epitaxy grown in
the S=D region, the S=D annealing in high temperatures such as 9008C–10008C can cause
strain relaxation and create dislocations [23,24]. For epitaxial layers with thickness less
than the critical thickness, annealing at high temperature such as 9008C for even 30 min
does not cause observable differences [23].
ß 2008 by Taylor & Francis Group, LLC.
Except for the stress that is induced intention ally in the chann el, the mech anical st ress
induce d by thermal cycles in the comple mentary met al-ox ide-semico nductor (CMOS )
proces s should be control led to min imize stress-i nduced defe cts. One impo rtant thermal
budge t is the formati on of STI, which is before the SiGe dep osition. It is well known that
stress is strongly concentrat ed at bot h the top and bot tom corners of the STI. Inherent ly
large mecha nical st ress and damag e in STI, comb ined with the subsequ ent process -induced
stress and defects, would generat e STI dislocat ions. If located withi n the junction depleti on
region, the STI dislocat ions can cause ab normally large leakage cur rent throu gh the
junct ion and transi stor. The dislocat ion generat ion is clos ely relate d to the S=D anne aling
[26,27] . During thermal anne aling, crystal defects from ion impl antat ion, esp ecially in
heavy -dose arsenic- implante d S=D regions [28], produ ce STI dis location nucleati on and
generat e dis locations , which glide to the regio n of maximu m stress unde r the comb ined
infl uence of mechani cal stress and high -temp erature process ing, and inter act with othe rs
to evolve a larger dislocat ion. Ike da et al. [26] propo sed two-step anne aling to bot h
activate the S =D dop ants and minimi ze the mech anical stress. Ishim aru et al. [29] and
Park et al. [30] propo sed the technol ogy of dimini shing the resi dual stress in STI
throug h high -tempera ture densi ficati on and the optimi zed com bination of tre nch fillin g
tetrae thoxysi lane (TEOS –O
3
) oxide, respectiv ely. Dami ano et al. [28] propo sed addi tional
oxidatio n, and Jeon et al. [31] change d the S=D anneal from rap id the rmal anne aling (RTA)
to tube anneal ; bot h claim ed they com pletely elim inated the trench dis locations .
3.6 P roc ess Flow
For the purp ose of sustai ning large enough strai n in the channel , st ress is int roduced in
the strained- Si proces s fl ow as late as poss ible to avoid high-temp eratur e anneal s and st rain
relaxa tion. A ty pical 130 to 45 nm front- end CMOS proces s fl ow is shown in Table 3.1. The
key the rmal cycles that will potenti ally cause devi ce defects are the STI, S =D formati on, and
gate oxide formation, each of which requires $30 s or longer thermal budgets at temperature
0.000
0.0
0.026 = s
exc
/m
10
2
10
3
10
4
0.005 0.010 0.015 0.020
Misfit
S
t
r
a
i
n
e
d

l
a
y
e
r

t
h
i
c
k
n
e
s
s

(
Å
)
Stable
Unrelieved
Partially or
fully relieved
Metastable Metastable
FIGURE 3.10
Thickness and misfits for a series of SiGe films
grown at 4948C. Fully strained and partially
strain-relieved films are represented by filled
and open circles, respectively. (Reprinted with
permission from Tsao, J.Y., Dodson, B.W.,
Picraux, S.T., and Cornelison, D.M., Phys.
Rev. Lett., 59, 21, 1987.)
ß 2008 by Taylor & Francis Group, LLC.
aroun d 1000 8C. Only sligh t modi fi cations to a stand ard CMO S logic technol ogy proces s
fl ow are need ed to insert the longi tudinal com pressive and tensil e stress into the p- and
nMO SFETs, respec tively. For strained- Si pMOS FETs, a Si recess etch is inserted after spacer
formati on and followed by selective epitaxi al che mical vapor dep osition (CVD) of SiGe in
the S =D region. The speci fic proces s flow st eps are show n in Figure 3.11 . Ger manium
concen tration of 17%–35% is use d, which cre ates a lattice spaci ng larger than Si . The
mismat ch in the SiGe to Si lattice cause s the sm aller latt ice cons tant Si channel to be under
com pressive stress, the magnitu de of which also depen ds on the spacing of the Si Ge epitaxy
to the chann el in addi tion to the Ge concentrat ion. A 30 nm recess ed SiGe layer at the S=D
locatio n result s in a 250 MPa st ress in the middl e of the chann el; whil e the sam e SiGe layer,
when present at the S=D extension location, stresses the channel to $900 MPa. As an
exa mple, the three-d imension al finite el ement analy sis in Figu re 3.12 sh ows the stress
distribution for a 45-nm gate length transistor after the SiGe deposition. Strain can also be
engineered at other process steps like STI and silicide [32]. Integrating the SiGe layer in the
S=D extension location also has a key advantage to reduce the effect of the SiGe deposition
thermal budget on S=D. Furthermore, by confining the SiGe to the S=D region and introdu-
cing it late in the process flow, integration challenges, such as misfit dislocations, yield,
and increased self-heating due to the low thermal conductivity of SiGe, are reduced.
TABLE 3.1
Process Modification of Strained-Si Technology to a Standard CMOS Process Flow
. Typical CMOS Process Flow Strained-Si
. Gate stack
. Gate patterning
. Offset spacers (optional)
. n-Type LDDþhalo/pocket implants
. p-Type LDDþhalo/pocket implants
. Spacer deposition and etch (STI)
. Recess and SiGe deposition (pMOSFETs)
. n-Type HDD implants
. p-Type HDD implants
. RTA
. Silicide protection deposition and etch
. Silicidation
. High stress capping layer (nMOSFETs)
FIGURE 3.11
Strained-Si process flow.
Gate
Gate
SiGe
channel stressor
Si
1Ϫx
Ge
x
channel stressor
NiSi
Gate
(a)
(b)
(c)
Si recess etch
STI
STI
STI
ß 2008 by Taylor & Francis Group, LLC.
Longitudinal uniaxial tensile stress is introduced into the nMOSFETs by engineering the
stress and thickness of the Si nitride-capping layer [32–35]. There are several techniques to
nearly completely neutralize the capping layer strain on p-type MOSFETs; one is the use of a
Ge implant and masking layer [33]. Another technique to relax the strain is to selectively
remove the capping from the p-type transistors. Wafer-based strained-Si, e.g., Si channels
grown on virtual SiGe substrates, also attracted a lot of attention in the last decade. However,
compared to process-induced strain, the strain induced in the channel is biaxial tensile strain.
This has been proven not to be as efficient as the longitudinal tensile strain for nMOSFETs [36],
and even degrades the performance of the pMOSFETs [37,38]. Second, thermal annealing can
easily cause strain relaxation in the strained-Si layer [39–41]. In contrast, a significant advan-
tage of the uniaxial stress Si process flow is that, on the same wafer, compressive stress is
introduced into the pMOSFETs and tensile stress in the nMOSFETs to improve both the
electron and hole mobility. Since the nitride capping layer is already present to support
unlanded contacts, only a few new process steps are introduced at less than a 2% wafer
cost increase.
3.7 Strain and Alternative Wafer Orientation
Critical thickness and strain relaxation also relate to the wafer orientation. Although there
are attempts to use wafers with different orientations, combined with epitaxial strain, to
potentially enhance carrier mobility, the critical thickness for dislocation introduction is
very low in non-(001) wafers, and kinetically suppressing dislocation nucleation is very
difficult [42]. Diamond cubic and zinc blende structures create undesirable partial disloca-
tion stacking in (111), (110), and (112), leading to relaxed layers with very high threading
dislocation densities [43]. Thus, for CMOS technology incorporated with strain, at present,
the best recommendation continues to be to use (001) wafers.
References
1. Mooney, P.M. and Chu, J.O., SiGe technology: Heteroepitaxy and high-speed microelectronics,
Annu. Rev. Mater. Sci., 30, 33, 2000.
2. Mooney, P.M., Strain relaxation and dislocations in SiGe=Si structures, Mater. Sci. Eng. Rep., R17,
105, 1996.
MPa
45 nm
1000 MPa 1000 MPa
Si
0.83
Ge
0.17
600 MPa
400 MPa
200
0.1
0.0
µm
0
200
400
600
800
1000
FIGURE 3.12
Stress distribution after SiGe deposition
obtained by 3D finite element simulation
for a 45 nm gate length transistor.
ß 2008 by Taylor & Francis Group, LLC.
3. Tsuya, H., Present status and prospect of Si wafers for ultra large scale integration, Jpn. J. Appl.
Phys., 43, 4055, 2004.
4. Taraschi, G., Pitera, A.J., and Fitzgerald, E.A., Strained-Si, SiGe, and Ge on-insulator: Review of
wafer bonding fabrication techniques, Solid-State Electron ., 48, 1297, 2004.
5. Arghavani, R., Method of inducing stresses in the channel region of a transistor, patent (US).
Available at: http: ==www.freshpatents.com =Method-of-inducing-stresses-in-the-channel-region-
of-a-transistor-dt20051117ptan20050255667.php.
6. Thompson, S.E. et al., A 90-nm logic technology featuring strained-silicon, IEEE Trans. Electron
Devices, 51, 1790, 2004.
7. Ch ipworks Report. Availabl e at: http: ==ww w. ch ip wo rks .com =seamark.aspx?sm ¼ s4%
3BD ated fl 14%3BDev ic eCate gory13%3BDigital þ Logic&ss ¼ no&tv ¼ strained þ Si&tn ¼
1ReportCode%2CTitle%2CDescription%2CWh at sInside%2C WhoSho uldBuy%2CWhyToBuy%
2CManufacturer%2CDeviceCategory&ns ¼ 1.
8. Chidambaram, P.R. et al., 35% drive current improvement from recessed-SiGe drain extensions
on 37 nm gate length PMOS, in VLSI Symp. Tech. Dig., Honolulu, HI, 2004, p. 48.
9. Bai, P. et al., A 65 nm logic technology featuring 35 nm gate lengths, enhanced channel strain,
8 Cu interconnect layers, low-k ILD and 0.57 mm
2
SRAM cell, in IEDM Tech. Dig., 2004, p. 657.
10. Ohta, H. et al., High performance 30 nmgate bulk CMOS for 45 nmnode with s-shaped SiGe-SD,
in IEDM Tech. Dig., 2005, p. 4.
11. Horstmann, M. et al., Integration and optimization of embedded-SiGe, compressive and tensile
stressed liner films, and stress memorization in advanced SOI CMOS technologies, in IEDM Tech.
Dig., 2005, p. 233.
12. Ang, K. et al., Enhanced performance in 50 nm N-MOSFETs with silicon-carbon source=drain
regions, in IEDM Tech. Dig., 2004, p. 1069.
13. Yang, M., Chan, V., and Ku, S.H., On the integration of CMOS with hybrid crystal orientations, in
VLSI Symp. Tech. Dig., 2004, p. 160.
14. Welser, J., Hoyt, J.L., and Gibbons, J.F., Electron mobility enhancement in strained Si N-type
metal-oxide-semiconductor field-effect-transistors, IEDM Tech. Dig., 1992, p. 1000.
15. Thompson, S.E. et al., A logic nanotechnology featuring strained-silicon, IEEE Electron Device
Lett., 25, 191, 2004.
16. Lee, W.H. et al., High performance 65 nm SOI technology with enhanced transistor strain and
advanced-low-k BEOL, in IEDM Tech. Dig., 2005, p. 61.
17. Chen, C. et al., Stress memorization technique (SMT) by selectively strained-nitride capping
for sub-65 nm high-performance strained-Si device application, in VLSI Symp. Tech. Dig.,
2004, p. 56.
18. Singh, D.V. et al., Stress memorization in high-performance FDSOI devices with ultra-thin silicon
channels and 25 nm gate lengths, in IEDM Tech. Dig., 2005, p. 505.
19. Braga, N. et al., Formation of cylindrical n=p junction diodes by arsenic enhanced diffusion along
interfacial misfit dislocations in p-type epitaxial Si=Si(Ge), Appl. Phys. Lett., 65, 1410, 1994.
20. Cohen, G.M. et al., Dislocation-free strained-silicon-on-silicon by in-place bonding, Appl. Phys.
Lett., 86, 251902, 2005.
21. Ball, C.A. and van der Merwe, J.H., The growth of dislocation-free layers, in Dislocations in Solids,
Nabarro, F.R.N., Eds., North-Holland Press, Amsterdam, 1983, Chapter 27.
22. Matthews, J.W. and Blakeslee, A.E., Defects in epitaxial layers: I. Misfit dislocations, J. Cryst.
Growth, 27, 118, 1974.
23. Houghton, D.C. et al., Equilibriumcritical thickness for Si
1Àx
Ge
x
strained layers on (100) Si, Appl.
Phys. Lett., 56, 460, 1990.
24. Timbrell, P.Y. et al., An annealing study of strain relaxation and dislocation generation in SiGe
heteroepitaxy, J. Appl. Phys., 67, 6292, 1990.
25. Tsao, J.Y. et al., Critical thicknesses for Si
x
Ge
1Àx
strained-layer plasticity, Phys. Rev. Lett., 59,
2455, 1987.
26. Ikeda, S. et al., The impact of mechanical stress control on VLSI fabrication process, in IEDM Tech.
Dig., 1997, p. 77.
27. Ha, D. et al., Anomalous junction leakage current induced by STI dislocations and its impact on
dynamic random access memory devices, IEEE Trans. Electron Devices, 46, 940, 1999.
ß 2008 by Taylor & Francis Group, LLC.
28. Damiano, J. et al., Characterization and elimination of trench dislocations, in VLSI Symp. Tech.,
Dig., 1998, p. 212.
29. Ishimaru, K. et al., Mechanical stress induced MOSFET punch-through and process optimization
for deep submicron TEOS-O
3
filled STI devices, in VLSI Tech. Dig., 997, p. 123.
30. Park, M.H. et al., Stress minimization in deep sub-micron full CMOS devices by using an
optimized combination of the trench filling CVD oxides, in IEDM Tech. Dig., 1997, p. 669.
31. Jeon, C. et al., Generation of trench dislocation in 0.25 mm logic technology and its elimination,
Proc. 6th Int. Conf. on VLSI and CAD (IVCC 1999), October, 26–27, 1999, p. 463.
32. Ito, S. et al., Mechanical stress effect of etch-stop nitride and its impact on deep submicrometer
transistor design, in IEDM Tech. Dig., 2000, p. 247.
33. Shimizu, A. et al., Local mechanical-stress control (LMC): A new technique for CMOS-
performance enhancement, in IEDM Tech. Dig., 2001, p. 433.
34. Ghani, T. et al., A 90-nmhigh volume manufacturing logic technology featuring novel 45-nmgate
length strained-silicon CMOS transistors, in IEDM Tech. Dig., 2003, p. 978.
35. Ge, C.H., Process-strained-Si (PSS) CMOS technology featuring 3-D strain engineering, in IEDM
Tech. Dig., 2003, p. 73.
36. Uchida, K. et al., Physical mechanisms of electron mobility enhancement in uniaxial stressed
MOSFETs and impact of uniaxial stress engineering in ballistic regime, in IEDM Tech. Dig., 2005,
p. 129.
37. Rim, K. et al., Fabrication and mobility characteristics of ultra-thin strained Si directly on insulator
(SSDOI) MOSFETs, in IEDM Tech. Dig., 2003, p. 49.
38. Sun, Y., Thompson, S.E., and Nishida, T., Physics of strain effects in semiconductors and metal-
oxide-semiconductor field-effect transistors, J. Appl. Phys., 101, 104503, 2007.
39. Fiorenza, J.G. et al., Film thickness constraints for manufacturable strained-silicon CMOS, Semi-
cond. Sci. Technol., 19, L4, 2004.
40. Hirashita, N., Moriyama, Y., and Sugiyama, N., Misfit strain relaxation in strained-Si layers on
silicon-germanium-on insulator substrates, Appl. Phys. Lett., 89, 091916, 2006.
41. Samavedam, S.B. et al., Relaxation of strained-Si layers grown on SiGe buffers, J. Vac. Sci. Technol.,
B, 17, 4, 1999.
42. Kuan, T.S. and lyer, S.S., Strain relaxation and ordering in SiGe layers grown on (100), (111), and
(110) Si surfaces by molecular-beam epitaxy, Appl. Phys. Lett., 59, 2242, 1991.
43. Lee, M.L., Antoniadis, D.A., and Fitzgerald, E.A., Challenges in epitaxial growth of SiGe buffers
on Si (111), (110), and (112), in Proc. ICSI4 Conference, Awaji, Japan, 2005.
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
4
The Effect of Defects on Electron Transport in
Nanometer-Scale Electronic Devices: Impurities
and Interface Roughness
M.V. Fischetti and S. Jin
CONTENTS
4.1 Introduction.......................................................................................................................... 71
4.2 Semiclassical Electron Transport ....................................................................................... 72
4.2.1 Transport in Bulk Semiconductors......................................................................... 72
4.2.2 Low-Field Mobility................................................................................................... 73
4.2.3 Transport in Confined, Low-Dimensionality Systems ........................................ 75
4.2.4 Collision Kernels ....................................................................................................... 76
4.3 Ionized Impurities ............................................................................................................... 78
4.3.1 Impurity Potential in Bulk Semiconductors.......................................................... 78
4.3.2 Dielectric Screening .................................................................................................. 80
4.3.3 Full-Band Expression for the Screened Impurity Scattering Rate ..................... 85
4.3.4 Additional Corrections to the Impurity Potential................................................ 88
4.3.5 2D Model.................................................................................................................... 90
4.3.6 Multi-Subband Linear Screening Approximations .............................................. 92
4.3.7 Ballistic Limit and the DOS Bottleneck: The Role of Impurity Scattering ....... 94
4.4 Interface Roughness ............................................................................................................ 98
4.4.1 Effective Potential in Thin-Body Si ........................................................................ 99
4.4.2 Roughness as a Perturbation................................................................................. 100
4.4.3 Direct Scattering at Atomic Steps......................................................................... 103
4.4.4 Scattering via Roughness-Induced Coulomb Terms ......................................... 105
4.4.5 Screening .................................................................................................................. 107
4.4.6 Mobility Calculation............................................................................................... 109
4.4.7 Simulation Results and Discussion ...................................................................... 109
4.5 Conclusions......................................................................................................................... 115
Acknowledgments ..................................................................................................................... 115
References.................................................................................................................................... 115
4.1 Introduction
Crystal defects have long ceased to be a major concern in devices of the present very large-
scale integration (VLSI) technology, thanks to the dramatic improvements we have
ß 2008 by Taylor & Francis Group, LLC.
recently witnessed regarding crystal growth, gettering, and annealing after implantation
and other radiation-rich processing steps (such as reactive-ion etching [RIE]). Ignoring
neutral defects such as those associated with dislocations and stacking faults—possibly
returning to the fore as strained channel materials are becoming more and more commonly
employed—more realistically, defects to be considered at present are intentionally induced
defects. If we define, from an electron transport perspective, a defect as any entity which
breaks the three-dimensional (3D) symmetry of the semiconductor lattice, the ionized
impurities introduced by doping and interfaces are the major defects to be considered.
Here, first we consider ionized impurities, discussing the models employed to describe
their associated potential, various approximations used to account for dielectric screening,
and two major competing approximations, the Conwell and Weisskopf (CW) [1] and the
Brooks–Herring (BH) [2] models, employed to obtain the associated scattering and
momentum relaxation rates. Then, we consider genuine defects, namely those interfacial
defects commonly labeled as interface roughness. As charge carriers are squeezed harder
against the interfaces by the increasingly large perpendicular electric fields present in thin
insulator devices, scattering with the roughness is a growing concern. We limit our
analysis to extensions to thin bodies and quantum wells of the continuum empirical
model originally proposed by Prange and Nee [3], Saitoh [4], Sakaki [5], and Ando et al. [6].
4.2 Semiclassical Electron Transport
Before moving to the core of this chapter, we recall some basic tools employed in dealing
with electronic transport in semiconductors.
4.2.1 Transport in Bulk Semiconductors
At the semiclassical level charge transport in semiconductors is described by the
Boltzmann transport equation (BTE). Its validity at the 10 nm scale is a question that
goes beyond the scope of this chapter, so we naively assume the BTE as valid.
The BTE describes the irreversible temporal evolution of the distribution function
f(r, k, m, t) (i.e., the density, f of the particles in band m at position r with wave vector
k at time t):
@f (k, r, m, t)
@t
¼ À
dk
dt
Á r
k
f (k, r, m, t) À
dr
dt
Á r
r
f (k, r, m, t) þ
@f (k, r, m, t)
@t

coll
: (4:1)
In words, this equation expresses the total time variation of the population of an infini-
tesimal volume in phase space as due to particles that leave the volume drifting away in
k-space under external forces (the term (dk=dt) Á r
k
f), by diffusing away in real space (the
term (dr=dt) Á r
r
f), and due to particles leaving and entering the volume via collisions
(the term (@f=@t)
coll
). The last term is the crucial and hard-to-treat collision term (often
called collision integral or collision operator):
@f (k,r,m,t)
@t

coll
¼
¸
n

dk
0
(2p)
3
S(k,m; k
0
,n; r)f (k
0
,r,n,t)[1Àf (k,r,m,t)]
À
¸
n

dk
0
(2p)
3
S(k
0
,n; k,m; r)f (k,r,m,t)[1Àf (k
0
,r,n,t)], (4:2)
ß 2008 by Taylor & Francis Group, LLC.
where S(k
0
, n; k, m; r) represents the probability per unit time that a particle will make a
transition from a state (hk
0
, m) in reciprocal space into the infinitesimal volume centered
around (hk
0
, n) because of a collision. These terms, aside from the less complicated band
structure effects, contain the entire physics of the problem. This is where the knowledge of
the defect enters when analyzing its effect on electronic transport.
Note that in Equation 4.1 the drift term (dk=dt) Á r
k
f is determined by the electric field
F¼Àr
r
w, the electrostatic potential w being determined by the Poisson equation
rÁ [«(r)rw(r, t)] ¼ Àr(r, t) Àr
d
(r), (4:3)
where
«(r) is the position-dependent static permittivity of the material(s) constituting the device
r
d
(r) is the charge density of the ionized donors and acceptors, eN
þ
D
(r) ÀeN
À
A
(r)
r(r, t) is the charge density of the free carriers related to the distribution function
f(k, r, t) via its separate hole (valence bands) and electron (conduction bands) contri-
bution:
r(r,t) ¼ 2e
¸
m
valence

dk
(2p)
3
f (k, r, m, t) À2e
¸
m
conduction

dk
(2p)
3
f (k, r, m, t), (4:4)
having assumed, here and in the following, that the distribution function is normalized to
the total local particle density:
2
¸
m

dk
(2p)
3
f (k, r, m, t) ¼ n(r,t):
Thus, Equations 4.1 and 4.3 are coupled since the equations of motion for a particle in phase
space depend on the electrostatic potential w(r, t) (assuming a homogeneous material),
dk
dt
¼ +
e
h
r
r
w(r, t), (4:5)
dr
dt
¼
1
h
r
k
E
m
(k), (4:6)
the ‘‘minus’’ sign in Equation 4.5 holding for holes, the ‘‘plus’’ for electrons, e is the
magnitude of the electron charge, h is the reduced Planck constant, and E
m
(k) is
the dispersion relation for carriers in the semiconductor.
Solving this integro-differential equation (possibly nonlinear since the collision kernel
S(k
0
, n; k, m; r) depends on f itself when accounting for dielectric screening and=or
interparticle collisions) is best done with Monte Carlo techniques [7]. Below we present
results using self-consistent Monte Carlo simulations employing 3D transport accounting
for the full band structure of the semiconductor, coupled to a 2D solution of the Poisson
equation in the electronic device [8].
4.2.2 Low-Field Mobility
Of particular interest—for both its simplicity and relatively straightforward comparison with
experimental information—is to consider the homogeneous, steady-state, weak-field limit of
the BTE. As stated above, thanks to the acceleration theorem [9], the driving term dk=dt in
ß 2008 by Taylor & Francis Group, LLC.
Equation 4.1 can be written as ÆeF=h, where e is the magnitude of the electron charge,
F¼Àr
r
w is the applied electric field, w is the electrostatic potential, and the plus=minus
sign applies to holes=electrons. By expressing the unknown distribution function f as its
expression at equilibrium, f
0
(given by a Fermi–Dirac distribution) plus a correction d f
proportional to the magnitude F of the field, one can show [10] that the current density,
J ¼ Æ2e

dk
8p
3
y(k) f (k) ¼ Æenhyi, (4:7)
(n being the carrier density and h . . . i the statistical average) can be written as the following
linear function of F:
J
i
¼ Æen
¸
j
m
ij
F
j
, (4:8)
where the proportionality constant m (here more generally represented as a tensor to
account for a possible anisotropy of the band structure) is the low-field mobility and is
given by
m
ij
¼ À
e
h
y
i
t
p,j
1
f
@f
@k
j

, (4:9)
where 1=t
p,i
is the ith component of the all-important momentum relaxation rate
1
t
p,i
(k)
¼

dk
0
(2p)
3
1 À
y
i
(k
0
)t
p,i
(k
0
)
y
i
(k)t
p,i
(k)
¸
S(k
0
, k): (4:10)
Note that this expression actually defines t
p,i
as the solution of an integral equation. For
elastic and isotropic scattering, Equation 4.10 simplifies to the well-known expression:
1
t
p,i
(k)
¼

dk
0
(2p)
3
(1Àcos u)S(k
0
, k), (4:11)
where u is the angle between k and the i-axis. The term 1 Àcos u intuitively represents the
fraction of the ith component of the momentum lost (or gained) in the collision. In general,
however, the evaluation of the term within square brackets in Equation 4.10 requires either
drastic approximations or laborious computations. In the simple case of an isotropic and
parabolic dispersion, the diagonal components of the mobility tensor coincide and are
given by the well-known Kubo–Greenwood expression [11,12] (obtained from Equation 4.9
by transforming the integration over k into an integration over angles and energy E):
m ¼ À
2e
3nm*

1
0
dEr(E)Et
p
(E)
@f
0
@E
¼
2e
3nm*k
B
T
0

1
0
dEr(E)Et
p
(E) f
0
(E), (4:12)
where
f
0
is the Fermi–Dirac equilibrium distribution function at the lattice temperature T
t
p
is given by Equation 4.10
r(E) is the density of state (DOS) at energy E
ß 2008 by Taylor & Francis Group, LLC.
Modifications to this expression are required when dealing with nonparabolic, anisotropic,
or more complicated band structures. These expressions can be found in Refs. [13–16] and
in Equation 4.151 for the 2D case. They add to the algebraic complexity, but not to the
physical meaning.
4.2.3 Transport in Confined, Low-Dimensionality Systems
Inparticular regions of the device, the electrostatic potential w, the presence of heterointerfaces,
or other geometrical constraints may confine the carriers into regions whose dimensions are
comparable to (or smaller than) the wavelength of the envelope wave function, causing
quantum confinement. The channel of metal-oxide-semiconductor field effect transistors
(MOSFETs), quantum wells, and ultrathin bodies are typical examples. In this case one may
usually be able to describe transport in semiclassical terms on the plane perpendicular to the
confinement direction. Some additional approximations reducethe complexityof the problem:
(1) the confining potential w varies slowly also along the confining direction z, although not as
slowlyas inthein-planedirection, sothat it canbe assumedroughlyconstant over aunit cell. (2)
The matrix elements between Bloch functions of different bands are negligible. Then, the
electronic wave function in band n can still be factored into a high-frequency periodic Bloch
function and a slowly varying envelope wave function, f
n
, which can be expanded over
components of crystal momentum K measured in the in-plane directions from the minimum
of the valley in which confinement occurs [17–19]:
f
n
(r) ¼
¸
k
a
Kn
f
Kn
(R, z),
where each envelope component f
Kn
(R, z) obeys the Luttinger–Kohn (effective mass)
equation
[E
n
(Àir) þV(R, z)]f
Kn
(R, z) ¼ E
n
(K)f
Kn
(R, z), (4:13)
where E
n
(Àir) is the (nonlocal) operator obtained by replacing the crystal momentum k
with the operator Àir, as if we were dealing with free particles. The total potential energy
V(R, z) results, in principle, from the sum of various contributions [6]:
V(R, z) ¼ Æew(R, z) þV
i
(R, z) þV
xc
(R, z), (4:14)
(the sign ‘‘plus’’ for holes, ‘‘minus’’ for electrons) expressing the potential energy due to the
ionized impurities in the depletion layer and to the induced charges in the space charge
layer, Æew, to the image charges at the semiconductor–insulator interface, V
i
, and to the
exchange and correlation effects, V
xc
. More generally and correctly, this last term is not a
simple local function of position, but is a functional of the electron density. The image and
many-body corrections have effects opposite and roughly equal in magnitude in a few
cases for which solutions have been found [6,20]. Thus, the approximation is made of
ignoring these corrections and of retaining only the term Æew accounting for the charge
density of the dopants and of the free carriers. Therefore, the variable R in w can be treated
approximately as a parameter, thanks to the slow variation of the potential in the in-plane
direction. Thus, the envelope wave function can be simplified even more, by factoring it
into in-plane plane waves and a subband envelope z
na
:
f
Kn
(R, z) % z
na
(K, z)e
iKÁR
: (4:15)
ß 2008 by Taylor & Francis Group, LLC.
Then, the electron charge density in Equation 4.4 is replaced by
r(r, t) ¼ r
(3D)
(r, t) þ2e
¸
m,a
h
m

dKf
(2D)
(K, R, m, a, t)jz
ma
(K, z)j
2
, (4:16)
where h
m
is þ1 for the valence bands (holes), À1 for the conduction bands (electrons),
r
(3D)
is the charge density given by Equation 4.4 due to particles which are not confined
(bulk or 3D particles), whose transport is described by the BTE (Equation 4.1), while the
second component is due to the contribution of confined (2D or quantized particles) whose
in-plane distribution function f
(2D)
obeys the semiclassical BTE:
@f
2D
(K, R, m, a, t)
@t
¼À
dK
dt
Á r
K
f
(2D)
(K, R, m, a, t) À
dR
dt
Á r
R
f
2D
(K, R, m, a, t)
þ
@f
(2D)
(K, R, m, a, t)
@t

coll
, (4:17)
where the collision terms takes a form similar to Equation 4.2, with the appropriate
scattering kernels reflecting the confined nature of the carriers via the low-dimensionality
matrix elements involving the subband wave functions z (see Equation 4.20).
4.2.4 Collision Kernels
The collision kernels S(k
0
n
0
; kn; r) in Equation 4.2 are obtained typically by employing
the lowest order coupling between the carriers and the scattering potentials (which
are assumed to conserve spin and not to depend on this variable). In order to evaluate
this lowest order coupling and evaluate the scattering rate from Fermi golden rule, we
need the matrix element of the scattering potential between the initial and the final states,
M
s
(k
0
n
0
, kn). Let us denote by V
s
(r) ¼ S
s,q
V
s,q
e
iqÁr
, the scattering potential of type s, where
s runs over phonon branches, Coulomb potential for scattering with ionized impurities or
other carriers, alloy potential, etc. Then, proceeding in a standard way:
M
s
(k
0
n
0
, kn) hk
0
n
0
jV
s
jkni ¼
1
N
c
V
c

dr e
Àik
0
Ár
u*
n
0
k
0 (r)V
s
(r)u
nk
(r)e
ikÁr
,
where
N
c
is the number of cells in the normalization volume
V
c
the volume of a cell
u
Ã
nk
(r) is the Bloch wave
We can now express the integral over the variable r as a sum over cells labeled by the index
a and equal integrals over a cell, so that r ! R
a
þr and
M
s
(k
0
n
0
, kn) ¼
1
N
c
¸
q
¸
a
e
i(kÀk
0
þq)ÁR
a
V
s,q
1
Vc

cell
dr e
i(kÀk
0
þq)Ár
u*
n
0
k
0 (r)u
nk
(r):
The sum over a yield a nonzero contribution (equal to N
c
) only when kÀk
0
þq¼G, for
some G-vector of the reciprocal lattice. Therefore,
M
s
(k
0
n
0
, kn) ¼
¸
G
V
s,kÀk
0
þG
I(kn, k
0
n
0
; G), (4:18)
ß 2008 by Taylor & Francis Group, LLC.
where I (kn, k
0
n
0
; G) is the overlap factor between Bloch waves,
I(kn, k
0
n
0
; G) ¼
1
V
c

cell
dr e
iGÁr
u*
n
0
k
0 (r)u
nk
(r):
For phonon scattering, the only G-vector contributing to the sum is the particular vector,
G
u
, which maps k
0
Àk to the first Brillouin zone. If G
u
¼ 0, we talk of Normal processes, of
Umklapp processes otherwise. For the other interactions we are particularly interested
here, such as Coulomb scattering (impurity, interparticle), in principle we should retain a
sum over all possible G-vectors. We indeed retain these terms as a matter of computational
convenience. In practice, the squared Coulomb matrix elements will decrease very fast
(as G
À4
) and only the first term will contribute. As matter of simplicity of notation, we drop
the sum over G-vectors for now. Now, the scattering kernel associated to this process can
be calculated from the Fermi golden rule (or first Born approximation):
S
s
(k
0
n
0
; kn) ¼
2p
h
jM
s
(k
0
n
0
, kn)j
2
d[E
n
(k) ÀE
n
0 (k
0
) Æhv
s
(k
0
Àk)], (4:19)
where hv
s
(q) is the energy of the elementary excitation associated to the scattering poten-
tial, and the DOS of the material is implicitly accounted for by the delta function expressing
energy conservation. The overall kernel S appearing in the collision integral is obtained by
summing over all scattering processes s.
Ignoring the degeneracy correction (1 Àf ), which is accounted for in the approximation
described in the following, the rate at which a carrier of crystal momentum k in band
n scatters with the scattering potentials of type s is obtained from:
1
t
s
(k,n)
¼
¸
k
0
,n
0
S
s
(k
0
n
0
; kn) ¼
2p
h
1
(2p)
3
¸
n
0

dk
0
jM
s
(k
0
n
0
, kn)j
2
d[E
n
(k) ÀE
n
0 (k
0
) Æhv
s
(k
0
Àk)]:
(4:20)
Scattering involving carriers in a low-dimensionality region is treated in a similar way, but
the matrix element (Equation 4.18) is replaced by the matrix element between envelope
wave functions in bands n
0
, n (which in all practical situations are both identical and equal
to the lowest conduction band, simplification which we assume in the following by
dropping the band index) and subbands a
0
, a:
M
(2D)
s
(K
0
a
0
, Ka) %
L
z
N
c
V
c

dRdz e
ÀiK
0
ÁR
z*
a
0 (K
0
, z)V
s
(R, z)z
a
(K, z)e
iKÁR
, (4:21)
where L
z
is the normalization length in the quantization direction z, the envelope wave
functions have been normalized so that

L
z
dz z*
a
0 (K
0
, z)z
a
(K, z) ¼ d
K
0
, K,
and having approximated the overlap factor with unity, since only low energy states all
lying in a small region of the Brillouin zone are involved. In the common case of a
ß 2008 by Taylor & Francis Group, LLC.
scattering potential dependent only on the magnitude R of the in-plane coordinate R, the
matrix element above takes the form:
M
(2D)
s
(K
0
a
0
, Ka) %
L
z
N
c
V
c

dz z*
a
0 (K
0
, z)V
s
(jK
0
ÀKj, z)z
a
(K, z) V
s,a
0
a
(jK
0
ÀKj), (4:22)
where V
s
(Q, z) is the Fourier–Bessel transform of the scattering potential
V
s
(Q, z) ¼ 2p

1
0
dR RJ
0
(QR)V
s
(R, z),
where J
0
is the Bessel function.
Degeneracy effects (represented by the (1 Àf)-terms in Equation 4.2) are important in
high-density regions. In these regions the electric fields are typically small and the distri-
bution function approaches its equilibrium value. Therefore, a quasiequilibrium approxi-
mation is employed, by approximating f with a heated Fermi–Dirac function at a density
and temperature corresponding to the local density and temperature.
4.3 Ionized Impurities
4.3.1 Impurity Potential in Bulk Semiconductors
At the simplest level, the problem of treating electronic scattering with ionized impurities
is deceptively simple. Consider the simple potential
V
(0)
imp
(r) ¼
e
2
Z
4p«
s
r
, (4:23)
due to the presence of a charge eZ in a homogeneous macroscopic medium characterized
by a dielectric constant «
s
. The Fourier transform V
q
entering the matrix element given by
Equation 4.18 will be simply ( . . . with some regularization tricks to handle a tricky integral):
V
(0)
q
¼
e
2
Z
«
s
q
2
, (4:24)
which can be inserted into Equation 4.20 to obtain the scattering rate. The problem with
this approach is well known. Given the infinite range of Coulomb interactions, the cross
section (and relative scattering rate) diverges. Equally important is the fact that seldom
(if ever) is the problem related to the presence of an isolated impurity in an intrinsic
semiconductor without any free carrier. These considerations have led to the formulation
of two different approaches to bypass the infinite-range problem. In the BH model [2],
the unscreened potential V
(0)
imp
is replaced by the Yukawa-type potential:
V
(BH)
imp
(r) ¼
e
2
Z
4p«
s
e
Àb
DHr
r
, (4:25)
where b
DH
is the screening wave vector we discuss shortly, which leads to the matrix
element (assuming here and in the following Z¼1 for singly ionized dopants):
M
(BH)
(k
0
, k) ¼
e
2
«
1
s
1
q
2
þb
2
DH
, (4:26)
ß 2008 by Taylor & Francis Group, LLC.
with q ¼jkÀk
0
j, for an electron transition froman initial state at wave vector k to a final wave
vector k
0
. For a dispersion appropriate to the case of Si, described by ellipsoidal (with masses
m
t
and m
l
along the principal axes of the ellipsoids) and nonparabolic (with nonparabolicity
parameter a) equienergy surfaces, the scattering rate for an electron of wave vector k is
1
t
(BH)
(k)
¼
N
imp
g(k)e
4
2
3=2

12
s
m
1=2
d
E
2
b
1 À2aE
0
(k)
1 þ4g(k)=E
b
, (4:27)
where g(k) ¼ (h
2
=2) k
2
t
=m
t
þk
2
l
=m
l

is the parabolic dispersion, k
t
and k
l
being the
components of the k-vector along the transverse and longitudinal axes of the ellipsoid,
respectively, m
d
¼ m
l
m
2
t

1=3
is the DOS effective mass, E
b
¼h
2
b
2
s
=(2m
d
) and N
imp
is the
concentration of ionized impurities per unit volume. The BH approach, therefore, con-
siders the ideal situation of an isolated impurity, but in the presence of a sea of free carriers
able to screen its long-range Coulomb potential.
The opposite approach—historically predating the BH formulation—was followed by
Conwell and Weisskopf. They considered a medium without free carriers, but with many
impurities, so that an electron injected in the medium would feel the bare (i.e., unscreened)
potential of each impurity, but with range truncated at a distance equal to the mean
distance between the impurities, r
0
¼[3=(4pN
imp
)]
1=3
. This is equivalent to an infrared
cutoff employed to regularize the diverging integral resulting from Equation 4.24. The
resulting matrix element takes the form:
M
(CW)
(k
0
, k) ¼
e
2
«
1
s
1 Àcos (r
0
q)
q
2
, (4:28)
which, for ellipsoidal, nonparabolic bands, results in the scattering rate:
1
t
(CW)
(k)
¼
3m
d
e
4
(2p«
1
s
)
2
h
3
I(2kr
0
), (4:29)
having recalled that N
imp
r
3
0
¼ 3=(4p). The function I(x) is defined by
I(x) ¼
1
x

x
xy
dt
(1 Àcos t)
3
t
3
, (4:30)
where in the lower integration limit y ¼(E=E
r
)=[1 þ(E=E
r
)
2
]
1=2
with E
r
¼ e
2
= 8p«
1
s
r
0

.
This results from truncating the impact parameter to a maximum value r
0
. A much
smaller momentum relaxation time is expected from the BH model. The CW approach
appears to cut off artificially small-angle scattering, and, obviously, breaks down when
free carriers are present, a common, important, and interesting situation. For this reason,
the CW model has largely fallen out of favor and most calculations have relied, instead,
on the BH model. But nothing is perfect. The BH formulation applies only to the case of
dilute impurities (range of applicability of the Born approximation as well), thus over-
estimating the mobility at large doping and density concentrations, regime in which the
screening length b
À1
DH
becomes smaller than—or comparable to—the de Broglie wave-
length of thermal carriers. In addition, another problem seems to affect the BH approach
as well. Mahan [21] has noted that employing the BH model to calculate the electron
mobility in heavily doped, charge-neutral samples, for example, the free carriers are all
assumed to contribute to charge transport (thus determining the current flux and so the
mobility) and, at the same time, they are all assumed to fully screen the potential of all
ß 2008 by Taylor & Francis Group, LLC.
impurities, as in Equation 4.25. This double-counting prompted Mahan to perform more
accurate quantum-mechanical calculations. Using a variational approach, he found that the
wave functions of the free carriers do indeed pile up around the impurity potential at low
N
imp
($10
17
cm
À3
in Si at 300 K), but appear to be relatively insensitive to the presence of the
impurities in the opposite limit of large N
imp
. Thus, it appears that the BH model should be
employed at low impurity concentrations, the CW model in the opposite limit of large N
imp
.
Ridley [22], with his statistical screening model, has attempted to merge the two approaches
by considering only the potential due not to the impurity exhibiting the smallest impact
parameter (as inthe CWmodel), but considering the Yukawa-like screenedBHpotential due
to the closest impurity. This leads to a rescaling of the BH scattering rate:
1
t
(Ridley)
(k, r, n)
¼
y
k
d
1Àexp
d
t
(BH)
(k, r, n)y
k
¸
, (4:31)
where
y
k
is the group velocity
d ¼(2pN
imp
)
À1=3
is the average interimpurity distance.
4.3.2 Dielectric Screening
We have made use before of some elementary concepts related to dielectric screening
by free carriers in semiconductors. Since these concepts will be required below, both
in treating screening in low-dimensionality situations as well as to deal with surface
roughness (SR), we review here briefly the simple theory of linear screening.
Let us consider a free semiconductor, i.e., a semiconductor described by the free
Schrödinger equation in a crystal H
0
c
k,m
¼E
m
(k)c
k,m
, where H
0
is the free Hamiltonian
and c
k,m
(r) ¼hrjk, mi the (Bloch) plane waves which are the eigenfunctions of the lattice
Hamiltonian H
0
with eigenvalues E
m
(k) for each band m. If we apply an external
perturbation described by the potential w
(ex)
(r, t), the charges in the semiconductor will
rearrange themselves, causing an additional polarization (or screening) potential
w
(pol)
(r, t) which modifies the original external potential. Thus, the actual potential in
the material will be the sum of the external perturbation and of the polarization potential,
w(r, t) ¼ w
(ex)
(r, t) þw
(pol)
(r, t): (4:32)
We are interested in describing the total potential w, since this is the effective perturbation
acting on the material. We may think of w
(ex)
as the bare potential of a positively charged
donor impurity in n-type Si, described by the 1=r Coulomb potential, while w will be
the screened potential felt by the conduction electrons, resulting from single positive
charge of the impurity ion and from the negative cloud of free conduction electrons
attracted by the impurity. Two approximations are very commonly made to reduce the
complexity of the problem: first, the Fourier components of the various potentials
are considered independently. Cross-terms mixing different wavelengths are ignored,
on the grounds that their phases will vary wildly, resulting in their cancellation. This is
the so-called random phase approximation (RPA). An equivalent way of stating this
approximation is to say that each electron will respond to the average potential caused
by the other electrons. We thus transform this many-body problem to a single-particle
problem, thanks to this mean-field approximation. Second, we assume that both the external
andthe polarization potentials are weak perturbations of the free Hamiltonian H
0
. This is the
so-called linear screening approximation which allows us to express the polarization
ß 2008 by Taylor & Francis Group, LLC.
charge in simple first-order perturbation theory. Thanks to the RPA, we can state the
problem more simply. Let us expand the external perturbation in plane waves:
w
(ex)
(r,t) ¼
¸
q
w
(ex)
q
e
ÀiqÁrÀivt
, (4:33)
having assumed a simple harmonic time dependence. Then, we are interested in finding
the dielectric function «
(r)
(q, v) ¼«(q, v)=«
0
where «
0
is the vacuum permittivity, such that
w
q
¼ w
(ex)
q
þw
(pol)
q
¼
w
(ex)
q
«
(r)
(q,v)
: (4:34)
The usefulness of this expression lies in the fact that, once we know the dielectric function
«(q, v), the knowledge of the simple external potential is sufficient to determine the net
response of the system, without having to reevaluate the self-consistent redistribution of
the charges internally to the system. We proceed as follows: first, we employ the approxi-
mation of linear response to determine the new wave functions of the system under the
perturbation of the net potential w, assuming that it is known. From the perturbed wave
functions, we derive the polarization charge,
r
(pol)
(r,t) ¼
¸
q
dr
q
e
ÀiqÁrÀivt
, (4:35)
and from this the polarization potential. Summing the external and polarization potentials,
we solve for the net potential self-consistently.
Let us now express the wave functions, f
k,m
(r, t), of the perturbed system (lattice
Hamiltonian plus perturbing potential) in terms of the unperturbed wave functions
c
k,m
(r, t) ¼hr, tjk, mi, using first-order perturbation theory:
f
k,m
(r,t) ¼ c
k,m
(r,t) þdc
k,m
(r,t) ¼ c
k,m
(r,t) þ
¸
k
0
m
0
ehk
0
,m
0
jwjk,mi
E
m
(k) ÀE
m
0 (k
0
) þhv þihs
c
k
0
,m
0 (r,t), (4:36)
and
f
k,m
* (r,t) ¼ c
k,m
* (r,t) þdc
k,m
* (r,t) ¼ c
k,m
* (r,t) þ
¸
k
0
,m
0
ehk,mjwjk
0
,m
0
i
E
m
(k) ÀE
m
0 (k
0
) Àhv Àihs
c
k
0
,m
0 * (r,t): (4:37)
From Equation 4.33 and some manipulations with Bloch waves, we have:
hk
0
,m
0
jwjk,mi ¼ w
kÀk
0
¸
G
k
0
þG,m
0
e
i(kÀk
0
)Ár

k,m

: (4:38)
The new charge density is obtained by summing the individual contributions ejf
k,m
j
2
over
all occupied states f
k,m
. Denoting by p(k, m) the occupation of each state—not necessarily
the equilibrium occupation number—we have:
r(r,t) ¼ e
¸
k,m
p(k,m) f
k,m
(r,t)

2
¼ e
¸
k,m
p(k,m) c
k,m
(r,t) þdc
k,m
(r,t)

2
’ r
0
(r,t) þe
2
¸
k,m
p(k,m)c
k,m
* (r,t)
¸
G,k
0
,m
0
w
kÀk
0 k
0
þG,m
0
e
i(kÀk
0
)Ár

k,m

E
m
(k) ÀE
m
0 (k
0
) þhv þihs
c
k
0
,m
0 (r,t) þcc,
ß 2008 by Taylor & Francis Group, LLC.
where cc means complex conjugate. Now notice that under the sums we may swap the
dummy variables k with k
0
and m with m
0
in the cc term. Finally, expressing the product
c
*
k,m
(r,t)c
k
0
,m
0 (r,t) in terms of its Fourier transform,
c
k,m
* (r,t)c
k
0
,m
0 (r,t) ¼
¸
q
k,m e
iqÁr

k
0
,m
0

e
ÀiqÁr
¼
¸
G
0
k,m e
i(k
0
Àk)Ár

k
0
þG
0
,m
0

e
Ài(k
0
Àk)Ár
, (4:39)
and setting q¼k
0
Àk, we get:
dr(r,t) ¼ e
2
¸
k,m
¸
G,q,m
0
[p(k,m) Àp(k þq,m
0
)] k þq þG,m
0
h je
iqÁr
k,m j i

2
E
m
(k) ÀE
m
0 (k þq) þhv þihs
w
q
e
ÀiqÁr
: (4:40)
Comparing Equation 4.35 with Equation 4.39, we have:
dr
q
¼ e
2
¸
G,k,m,m
0
[p(k,m) Àp(k þq,m
0
)] k þq þG,m
0
je
iqÁr
jk,m

2
E
m
(k) ÀE
m
0 (k þq) þhv þihs
w
q
: (4:41)
By Poisson equation, r
2
Á w
(pol)
¼Àdr=«
0
, we can easily express the Fourier components of
the polarization potential in terms of the Fourier components of the charge variation dr
q
:
w
(pol)
q
¼
dr
q
q
2
«
0
: (4:42)
Combining Equations 4.34, 4.41, and4.42, we can finally identify the dielectric function with:
«
(r)
(q,v) ¼ 1 À
e
2
q
2
«
0
¸
G,k,m,m
0
[p(k,m) Àp(k þq,m
0
)] k þq þG,m
0
je
iqÁr
jk,m

2
E
m
(k) ÀE
m
0 (k þq) þhv þihs
: (4:43)
In order to consider the response only of the free carriers, we restrict ourselves to carriers
in a single band, since the corresponding denominator is often (but not always!) the
smallest.
We also consider here only perturbations such that q (G, so that the matrix element
of e
iqÁr
between states in the same band becomes unity, that is,
¸
G
k þq þG,mje
iqÁr
jk,m

2

¸
G
k þq þG,mjk,mj h i j j
2
’ 1: (4:44)
A notable case for which this approximation fails is the case of zone-edge phonons, whose
screening must be treated retaining the matrix element above. Then, dropping for now the
band indices, we can now rewrite Equation 4.43 as follows:
«(q,v)
«
s
¼ 1 À
e
2
q
2
«
s
¸
k
f (k) Àf (k þq)
E(k) ÀE(k þq) þhv þihs
: (4:45)
ß 2008 by Taylor & Francis Group, LLC.
Except for the static dielectric constant «
s
in place of the vacuum value, «
0
, this is exactly the
so-called Lindhard’s expression valid for a free electron gas. As a matter of convenience,
we introduce a screening wave vector b(q, v) and recast Equation 4.45 in the form:
«(q, v) ¼ «
s
1 þ
b
2
(q, v)
q
2
¸
: (4:46)
Thus, the final answer to our initial problem (What is the net potential w(r, t), once the
external perturbation w
(ex)
(r, t) is known?) can be expressed as
w(r, t) ¼
¸
q
w
(ex)
q
«
(r)
(q, v)
e
iq:r
e
ivt
: (4:47)
The evaluation of the electronic component of the dielectric function requires the know-
ledge of the electron distribution function p(k). In transport problems this may actually be
the main unknown, so that self-consistent schemes are required. But even when p is known
(as in thermodynamic equilibrium), the evaluation of the RPA expression Equation 4.43
or even Equation 4.45 is, in general, quite complicated. Therefore, we now consider various
cases in which this expression, ultimately needed to screen the scattering potentials, can
be evaluated in closed form. This can be done by employing a simple effective mass,
parabolic band approximations for the conduction band and assuming that the distribution
function is well approximated by its thermal equilibrium value. Indeed, screening is
particularly important in high-density regions of the devices. In these regions the large
density results in small electric fields and, consequently, in moderate carrier heating. Thus,
we set f(k) ’ f
0
(k).
For the static perturbations of interest here (e.g., impurity or SR scattering), we set hv¼0
and employ the following approximations valid at long wavelengths:
E(k þq) ÀE(k) ’ q Á r
k
E(k), (4:48)
f
0
(k þq) Àf
0
(k) ’ q Á
@f
0
@E
r
k
E(k), (4:49)
so that
«(q,0) ’ «
s
À
e
2
q
2

dED(E) À
@f
0
@E

¼ «
s
þ
e
2
q
2

dED(E) À
@f
0
@E
F

, (4:50)
where D(E) is the DOS at energy E. Thus, we obtain the static screening parameter at long
wavelengths:
b
2
s
¼
e
2
«
s

dED(E)
@f
0
@E
F

¼
e
2
«
s
@n
@E
F

: (4:51)
In the nondegenerate case (i.e., at very high temperatures), @n=@E
F
’ n=(k
B
T) and we have
the Debye–Hückel expression:
b
2
DH

e
2
n
«
s
k
B
T
: (4:52)
ß 2008 by Taylor & Francis Group, LLC.
In the opposite case of a degenerate gas (low temperature), @f=@E
F
’ d(EÀE
F
) and we have
the Thomas–Fermi expression:
b
2
TF

e
2
«
s
D(E
F
) ¼
e
2
«
s
3n
2E
F
: (4:53)
In the more general case of nondegenerate screening at high temperatures (i.e., f
0
(k) ’
exp[ÀE(k)=(k
B
T)]), we can do better, by finding an expression which describes a general
dynamical situation for arbitrary wavelengths, and which reduces to the Debye–Hückel
expression in the static and long-wavelengths limits. Following Fetter and Walecka [23],
we can convert the sum in Equation 4.45 into an integral over k. We obtain:
Re[b
2
(q, v)] ’ b
2
DH
mk
B
T
2h
2
q
2

1=2
 F
m
2k
B
T

1=2
v
q
þ
hq
2m

¸ ¸
ÀF
m
2k
B
T

1=2
v
q
À
hq
2m

¸ ¸ ¸
,
(4:54)
Im[b
2
(q, v)] ’ b
2
DH
pmv
2
2k
B
Tq
2

1=2
exp À
mv
2
2k
B
Tq
2
À
h
2
q
2
8mk
B
T

sinh [hv=(2k
B
T)]
hv=(2k
B
T)
, (4:55)
where
F(x) ¼
1
p
1=2
P

þ1
À1
dy
e
Ày
2
x Ày
(4:56)
is the real part of the plasma dispersion function [24] and m is replaced by the effective
mass of the lowest-lying valley or conduction band (or valence, for holes). Useful asymp-
totic and series expansions for the function F are
F(x) ’
1
x
1 þ
1
2
x
À2
þÁ Á Á

, x )1
2x 1 À
2
3
x
2
þÁ Á Á

, x (1

,
so that
g
1
(x) ’
8p
x
2
1 þ
8p
x
2
þÁ Á Á

, x )1
1 Àx
2p
24
þÁ Á Á , x (1

:
In the static limit, v ! 0, we have:
b(q, v !0)
2
!b
2
DH
g
1
(q‘), (4:57)
where
‘ ¼
2ph
2
mk
B
T

1=2
ß 2008 by Taylor & Francis Group, LLC.
is the thermal wavelength and
g
1
(x) ¼
2p
1=2
x
F
x
4p
1=2

: (4:58)
At long wavelengths, we recover the static Debye–Hückel expression, b(q !0, 0) !b
DH
,
while at short distances screening effects disappear as 1=q
2
. The imaginary part of
the dielectric function is related to the lifetime of the electron response: As a function
of v it peaks at the plasma frequency, while the full-width at half-maximum corresponds to
the lifetime plasmons have before decaying into single-particle excitations (electrons surf-
ing the waves), process called Landau damping. Figure 4.1 shows the dependence of the
static and dynamic screening parameter b(q, v) normalized to the Debye–Hückel value in
the q ! 0 limit . Figu res 4.2 and 4.3 show some of these featur es as well as the validit y of
the high-T, nondegenerate approximation compared to an exact numerical integration
of the RPA expression (Equation 4.43). We see below (Equation 4.148) an alternative
expression for the screening parameter at finite temperatures which is valid also in the
degenerate case. This is required in low-dimensionality situations in which degenerate
statistics plays a larger role.
4.3.3 Full-Band Expression for the Screened Impurity Scattering Rate
In order to calculate the carrier=ionized-impurity scattering rate in the context of our full-
band Monte Carlo program DAMOCLES, we must go back to the original Golden rule
expression:
1
t
imp
(k,r,n)
¼
2p
h
N
imp
¸
G,n
0
h(y Æ, s)

dk
0
(2p)
3
jM(kn,k
0
n
0
,G)j
2
d[E
n
(k) ÀE
n
0 (k
0
)][1 Àf (k
0
,n
0
,r)]
,
(4:59)
where G is a wave vector of the reciprocal lattice, M(kn,k
0
n
0
,G) is the Coulomb matrix
element:
M(kn,k
0
n
0
,G) ¼
Ze
2
«
s
I(kn,k
0
n
0
)
jk Àk
0
þGj
2
þb
2
(q,0)
, (4:60)
10
5
10
6
10
7
10
8
0.0
0.5
1.0
1.5
Si 300 K 10
18
cm
–3
Dynamic
(hw =E
q
)
Static
(hw =0)
Relative momentum q (cm
–1
)
b
2
(
q
,
w
)
/
b
s 2
FIGURE 4.1
Ratio of the static=dynamic (dashed=solid line) screening par-
ameter computed using the RPAin the high-T, nondegenerate
approximation to the Debye–Hückel expression at long wave-
length. The dynamic screening wavevector has been evalu-
ated at the plasma frequency for the indicated carrier
concentration. Note how the Debye–Hückel approximation
overestimates screening at long wavelength. (Reprinted from
Fischetti, M.V., J. Appl. Phys., 85, 7984, 1999. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
where
I(kn, k
0
n
0
) is the overlap integral between the Bloch states
b(q, 0) is the wave vector-dependent static screening parameter evaluated at the
crystal momentum transfer q ¼jkÀk
0
þGj
The function h represents the phase-shift corrections to the first Born approximation, as
tabulated by Mayer and Bartoli [25], at the values y
Æ
¼Æ(1=2)kr
B
for the dimensionless
wave vector, and s ¼2=(r
B
b
s
) for the screening parameter rescaled so to satisfy Friedel’s
sum rule [26]. Here r
B
is the effective Bohr radius, k is the magnitude of the wave vector,
b
DH
is the Debye–Hückel (or, where appropriate, Thomas–Fermi) screening parameter.
The ‘‘þ’’ sign is taken in the case of an attractive interaction (such as for electron-donor or
0 1 2 3 4 5
–10
0
10
20
Si 300 K
n
g
= 10
18
cm
–3
High T limit
Numerical RPA
Re(e)
Im(e)
b
s,3D
q (10
8

m
–1
)
e
g
(
q
,
w

=

w
P
)

(
e
0
)
e
g
(
q

=

b
s
,
3
D
,
w
)

(
e
0
)
0.0 0.5 1.0 1.5
0
5
10
15
20
25
Re(e)
Im(e)
w
P
(a) (b) w (10
14

s
–1
)
FIGURE 4.2
Real and imaginary part of the dielectric constant in Si evaluated as a function of q at v¼0 (top frame) and as a
function of v at q ¼0 (bottom frame). The solid line has been obtained from the full RPA, the dashed line from
the high-T, nondegenerate approximation. The labels b
s,3D
and v P indicate the Thomas–Fermi=Debye–Hückel
wavevector and the plasma frequency, respectively. (Reprinted from Fischetti, M.V., J. Appl. Phys., 85, 7984, 1999.
With permission.)
0 1 2 3 4 5
–10
0
10
20
Si 300 K n
g
=10
20

cm
–3
High–T limit
Numerical RPA
Re(e)
Im(e)
b
s,3D
q (10
9

m
–1
)
e
g
(
q
,
w

=

w
P
)

(
e
0
)
(a) (b)
0.0 0.5 1.0 1.5
0
5
10
15
20
Re(e)
Im(e)
w
P
w (10
15
s
–1
)
e
g
(
q

=

b
s
,
3
D
,
w
)

(
e
0
)
FIGURE 4.3
As in Figure 4.2, but at an electron density of 10
20
cm
À3
. Note that only at densities larger than $10
19
cm
À3
the
nondegenerate limit begins to differ significantly from the exact RPA result, as it is clearly evident here. Yet, even
in this case, the nondegenerate approximation may be considered acceptable. On the other hand, in low-
dimensionality situations the disagreement is more evident already at small sheet densities.
ß 2008 by Taylor & Francis Group, LLC.
hole-acceptor scattering), the ‘‘À’’ sign for a repulsive interaction (such as for electron-
acceptor and hole-donor scattering). The effect of degeneracy, represented by the factor
(1 Àf ) in Equation 4.59 is included by replacing the distribution functions f with their
equilibrium values,
f (k, m, r, t) ’ f
0
[k, m,T
p
(r, t),E
F
(r, t)],
where
f
0
(k, m, T, E
F
) ¼{1 þexp[(E
m
(k) ÀE
F
)=(k
B
T)]}
À1
is the Fermi–Dirac function at tempera-
ture T for band m and Fermi energy E
F
T
p
(r, t) is the equivalent particle temperature yielding the average local particle
density and energy estimated locally during the Monte Carlo simulation
The overlap integral I(kn, k
0
n
0
) is computed from kÁ p perturbation theory [27], while
screening is treated in the high-temperature, nondegenerate limit valid for a single
parabolic and spherical valley, and in the long-wavelength limit (because of the dominance
of small k-vectors in Equation 4.60, so that b(q, v¼0) ’ b
s
, where b
s
is the Debye–Hückel
screening parameter given by Equation 4.51 using the local particle density and tempera-
ture. A sum over the separate electron and hole contributions to our current use of
Equation 4.51 must be understood.
In the actual calculation, a self-scattering algorithm is chosen. First, an appropriate
multiple of Equation 4.63 is employed as a self-scattering rate. For particles undergoing
this self-scattering event, a second self-scattering rate is computed by ignoring overlap
factors and degeneracy:
1
t
imp,self
(k, r, n)
¼
2pZ
2
e
3
N
imp
ha
3
«
2
s
h(y Æ,S)
¸
G
¸
energy conserving cubes j,n
0
S
GR
j,n
0
(w
j
)
jkÀk
j
þGj
2
þb
2
s

2
, (4:61)
where S
GR
j,n
0
(w
j
) is the DOS in cube j
0
and band n
0
resulting from having discritized the entire
B2 into cubes of side Dk¼0.05 (2p=a), where a is the lattice constant at the dimensionless
final energy w
j
, according to the Gilat–Raubenheimer scheme [28], when energies are
measured in eV and crystal momenta in (2p=a)-units. The sum over G vectors is performed
over the first 80–100 vectors, up to a maximum magnitude of about
ffiffiffiffiffi
11
p
(2p=a). The final
state after collision is determined by ignoring phase-shift corrections (and, possibly, cor-
rections due to implementation of the Ridley model), by selecting a random k
j
with
probability density given by the matrix element in Equation 4.61, and by performing a
final rejection=acceptance step on degeneracy and overlap factors.
Whenever transport is simulated using the approximated analytical representation of the
band structure, a self-scattering rate is obtained analytically by ignoring once more degen-
eracy and overlap-factor corrections, considering only intravalley processes, and including
only the G-vector G
u
required to bring kÀk
0
þG
u
inside the first BZ:
1
t
imp,self
(k, r, n)
¼
e
4
Z
2
N
imp
2
9=2
pm
1=2
d,n
«
2
s
h(y Æ, s)g(E)
À3=2
(1 À2aE)

þ1
À1
dm (1 Àm) þ
E
b
2g(E)
¸
À2
(4:62)
so that
1
t
imp, self
(k, r, n)
¼
e
4
Z
2
N
imp
g(E)
1=2
2
3=2
pm
1=2
d,n
«
2
s
E
2
b
h(y Æ, s)
1 þ2aE
1 þ4g(E)=E
b
, (4:63)
ß 2008 by Taylor & Francis Group, LLC.
where E¼E
n
(k) ÀE
bot
is the kinetic energy of the particle in a valley whose energy
minimum is E
bot
, g(E) ¼E(1 ÀaE), E
b
¼h
2
b
2
s
2m
d,v
ð Þ = , and m
d,n
is the DOS effective mass
in band n and in the valley to which k belongs.
The final state is finally obtained by selecting a random azimuthal angle f with uniform
distribution in (0, 2p), and a polar angle u whose cosine, m, is selected with probability
distribution given by the integrand of Equation 4.62.
4.3.4 Additional Corrections to the Impurity Potential
Another issue we may raise is based on the fact that, so far, no distinction has been made
among different impurities. That is, in Si, the repulsive (for electrons) potential due to P is
undistinguishable from the potential due to, say, As. While there is certainly merit to this
observation, the effect (often called central cell correction) is usually small and, in practice
(compared with all other uncertainties affecting the theoretical model) negligible. To see why
this is so, let us consider the case of n-type Si and see how we may introduce the microscopic
nature of the impurity, following the approach of Ref. [29], but differing significantly from
them in some crucial detail. We follow here our own Ref. [30]. First, consider the bare core
potential, V(Z
I
, n; r), of a substitutional donor n-fold ionized, now taken isotropic since only
core electrons are considered. Nara and Morita [31], express this as
V(Z
I
, n; r) ¼ À
e
2
4p«
0
r
[n þ(Z
I
Àn)e
Às
I
r
], (4:64)
where s
I
is now obtained from a fit to the Hartree–Fock potential. Significantly, the
values of the ionic radii 1=s
I
are of the order of 0.01 nm for the species considered in
Ref. [31] (1=s
I
¼0.0124 nm for Si

, 0.0111 nm for P

, 0.0148 nm for As

, and 0.0156 nm
for Sb

). These values are all consistent with the use of the vacuum permittivity «
0
in
Equation 4.64. These very small values of 1=s
I
mean that differences among the various
ion cores are indeed confined to a narrow volume around the nucleus, and so largely
inaccessible to the low-energy free electrons which matter in determining the low-field
mobility.
Having settled on a bare potential, valence electrons are now brought in first as frozen in
their anisotropic, perfect-lattice distribution. Thus, the net potential of the substitutional
impurity, y
I
(r), is obtained by replacing the Si

ion with the n-fold ionized donor, that is,
Equation 4.4 of Nara and Morita [31]:
y
I
(r) ¼ V(Z
I
, n; r) ÀV(Z
Si
, 4; r): (4:65)
Finally, in the linear response approximation, the valence electrons are allowed to readjust
and screen y
I
(r) via the static dielectric function of Si, «
y
(q). Several models have been
proposed for a convenient analytic form for the wavelength-dependent dielectric function
of the valence electrons. Using Nara’s angle-averaged isotropic approximation for his
anisotropic result [32],
«
0
«
y
(q)
¼
Aq
2
q
2
þa
2
þ
Bq
2
q
2
þb
2
þ
Cc
2
q
2
þc
2
, (4:66)
(where A¼1.175, B¼À0.175, C¼«
0

sc
%0.08547, a ¼0.7572, b ¼0.3123, and c ¼2.044, a, b,
and c being in atomic units), the Fourier components of the screened isotropic impurity
potential, y
I
(q), become (accounting also for the wave vector-dependent screening of the
ß 2008 by Taylor & Francis Group, LLC.
free carrier s as give n by Resta [33], but see also Refs. [34,35 and refer ences the rein], for
othe r qualit atively ide ntical expre ssions):
y
I
( q ) ¼
e
2
«
y
( q)
H
I
(q )
q
2
þ «
sc
b
2
(q ) =«
y
( q)
, (4: 67)
where the fre e-carrier screeni ng wave vector is, as usu al, b
2
( q ) ¼ [( e
2
n) =( «
sc
k
B
T )]
[ F
À 1 =2
( h) =F
1 =2
( h)] G( j ,h ), (in the notatio n of Ref. [29] and with k
B
Boltzman n ’s
cons tant) and:
H
I
( q ) ¼ 1 þ ( Z
I
À 5)
q
2
q
2
þ s
2
1
À ( Z
si
À 4)
q
2
q
2
þ s
2
si
: (4 : 68)
In Ref. [29] the differe nt ro le play ed by diffe rent donor ions was lumped into a square
effect ive charge . In our ca se this charge becomes :
Q
I
( q )
2
¼
1 þ q
2
=b
2
( q)
1 þ «
y
( q ) q
2
= «
sc
b
2
( q )

¸
2
H
I
( q)
2
: (4 : 69)
As show n in Figu re 4.4, the angu lar dep endence of the charge Q
(KKS)
1
( q)
2
from Equati on
4.69 using the parame ters s
I
given by Nara and Morita [31], the dependence on the
scattering angle and on the dopant species is very small. For example, in Figure 4.4
the difference of Q
I
between P and As for backward scattering is an insignificant 6%.
Moreove r, as seen in Figure 4 .5 — illustrat ing the static valence-e lectron die lectric functi on
from Nara [32] as a function of the momentum transfer q—the maximum deviation from
0.95
1.00
1.05
1.10
1.15
1.20
1.25
1.30
1.35
Sb
As
P
0 p
From Ref. [30], Equation 8
E = 7k
B
T
E = 7k
B
T
N
D
=10
19
cm
–3
Scattering angle Scattering angle
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
P
As
Sb
0
(a) (b)
p/4 p/4 p/2 3p/4 p/2 3p/4 p
From Ref. [29], Equations 2.8, 2.11
Q
I (
K
K
S
)
(
q
)
2
Q
I

(
q
)
2
FIGURE 4.4
Squared effective charge computed with the model of Ref. [29] (a) and the modified model considered here
(b) as a function of scattering angle for an electron energy of 7k
B
T and for a doping and free-electron concentration
of 10
19
cm
À3
. Note the different ordinate scales. Nonparabolic corrections have been ignored for simplicity.
(Reprinted from Fischetti, M.V. and Laux, S.E., J. Appl. Phys., 89, 1205, 2001. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
isotropy is occurs only about 10%. Thus, the overall correction to the BH model due to the
anisotropy of the dielectric response of the valence electron is comparable to the corrections
due to the differences in doping element, and so negligible in practice.
Additional corrections to the BH model abound in the literature. As we noted above, the
BH model fails at large N
imp
not only for the reasons considered by Mahan [21], but also
because of the failure of the Born approximations. This is seen in an overestimation of the
mobility. For example, in Si and at room temperature, this happens for N
imp
% n>10
17
cm
À3
.
More important, the Born approximation fails to distinguish between scattering with
attractive (situation common in charge neutral samples) or repulsive (in the electron
transport in the p-type base of bipolar transistors) scattering centers. Scattering of an
electron from a positively charged center is enhanced with respect to the result of the
Born approximation, because the electron wave function is more localized near the attract-
ive potential. Vice versa, a negative center (such as an acceptor in p-type regions) pushes
away the electron wave function and the scattering cross section is depressed. These
corrections are sufficiently important in electronic devices to warrant some form of exten-
sion beyond the Born approximation (such as models based on partial-wave analysis
[25,36]) in our transport models. Finally, a major assumption embraced almost universally
(with notable exceptions) is that different scattering potentials due to different ions are
added incoherently. This assumption is made in the spirit of the approximations on which
the BTE is based. It is also justifiable whenever phase-destroying inelastic collisions, such
as scattering with phonons, occur very frequently, so that the probability that a particle
could remember its phase between successive impurity collisions is very small. Yet, it is
not hard to envision situations where the coherent effect of multiple scattering events
may yield noticeable effects in depressing the conductivity via destructive interference.
Claims have been made that even in Si at 300 K, at large impurity concentrations multiple
scattering should be accounted for in order to explain the observed ohmic mobility. The
review paper by Chattopadhyay and Queisser [10], and the many publications cited there
on heavy doping effects which have appeared in the literature should be consulted for
details about these effects.
4.3.5 2D Model
As an example of impurity scattering in low-dimensional situations caused by quantum
confinement, we consider the usual case of a Si channel at the Si–SiO
2
interface and
consider scattering with ionized impurities located in the space charge region, or with
charged defects located in the oxide, or at the interface, corresponding to ionized dopants,
FIGURE 4.5
Isotropic static dielectric function, «
y
, for Si from Ref. [32] plotted
as a function of the magnitude of the scattering wavevector q.
Maximum anisotropy occurs at wavelengths of the order of the
bond length, while an isotropic response is seen around the core
(q ! 1, where « ! «
0
) and at large distances away from the
impurity (q ! 0, where « ! «
sc
). (Reprinted from Fischetti, M.V.
and Laux, S.E., J. Appl. Phys., 89, 1205, 2001. With permission.)
0.0 0.2 0.4 0.6 0.8 1.0
0
3
6
9
12
15
Maximum anisotropy (>10%)
q (2p/a)
e
v
(
q
)
/
e
0
ß 2008 by Taylor & Francis Group, LLC.
fixed oxide charges, and interface traps, respectively. We sketch here only the final
equations, their derivation in terms of the Poisson Green’s function of the system [37]
having been given extensively in Refs. [38,39]. Using again the first Born approximation,
considering again only first-order nonparabolic corrections, and ignoring phase-shift
corrections, the Coulomb scattering rate for an electron of crystal momentum K in the
(x, y)-plane of the interface located at z ¼0 and in subband n can be written as [40]:
1
t
(r)
imp
(K,n)
¼
e
4
N
r
8ph
3
«
2
r
¸
m
m
d,m
k
m
[DE
nm
(K)]juj[DE
nm
(K)]

2p
0
db
0
H
(r)
nm
[Q(b
0
)]

screened
G
m
(b
0
)Q(b
0
)
2
, (4:70)
where the index r runs over impurity, oxide, and interface charges, and where
G
m
(b
0
) ¼
m
d,m
m
1,m
cos
2
b
0
þ
m
d,m
m
2,m
sin
2
b
0
,
Q(b
0
)
2
¼ K
2
À2KK
0
(b
0
) cos (b Àb
0
) þK
0
(b
0
)
2
,
K
0
(b
0
)
2
¼
2m
d,m
DE
nm
(K)[1 ÀaDE
nm
(K) À2aB
m
]
h
2
G
m
(b
0
)
,
DE
nm
(K) ¼ E
n
(K) ÀE
m
,
b ¼ arccos
K
1
K

,
k
m
(E) ¼ 1 À2aE À2aB
m
is a nonparabolic correction in terms of the parabolic subband bottom E
(0)
m
and of the
expectation value hVi
m
of the confining potential energy in the subband m:
B
m
¼ E
(0)
m
ÀV

m
¼ E
(0)
m
À

þ1
À1
dzz
Ã
m
(z)V(z)z
m
(z),
which represents the average effective kinetic energy of an electron confined to the bottom
of the subband n. Finally, m
d,m
, m
1,m
, and m
2,m
are the DOS effective mass in subband m and
the masses along the 1- and 2-axis, respectively.
The unscreened form factor H
(r)
mn
is defined by
N
r
e
2
H
(r)
nm
(Q)
«
2
r
Q
2
¼

þ1
À1
dz
e
N
r
(z
e
) w
nm
(Q; z
e
)

2
, (4:71)
where the matrix elements of the Fourier–Bessel components of the potential due to
charges at z ¼z
e
are defined below by Equations 4.84 through 4.87, so that, assuming a
uniform distribution of charges,
H
(r)
nm
(Q) ¼

1
0
dz

1
0
dz
0
z
m
(z)z
m
(z
0
)I
(r)
Q
(z,z
0
)z
n
(z)z
n
(z
0
): (4:72)
ß 2008 by Taylor & Francis Group, LLC.
The index r runs over impurity, oxide, and interface charges. For scattering with ionized
donors and acceptors, N
D
and N
A
, assumed to be uniformly distributed over the space
charge layer of thickness d
sc
, and using the expression for the potential due to single
charges reported in Ref. [6], one obtains:
«
imp
¼ «
s
, (4:73)
N
imp
¼ (N
A
þN
D
)d
sc
, (4:74)
I
(imp)
Q
(z,z
0
) ¼
jz Àz
0
j þQ
À1

e
ÀQjzÀz
0
j
þ1= 2Q ð Þðe
ÀQ(zþz
0
Þ
þe
Q zþz
0
À2d
sc
ð Þ
Þ
þ ~ «

z þz
0
þQ
À1

e
ÀQ zþz
0
ð Þ
À1=(2Q) e
ÀQ zÀz
0
þ2d
sc
ð Þ
þe
ÀQ z
0
Àzþ2d
sc
ð Þ

þ ~ «
2
e
ÀQ zþz
0
Þ Þ
1 Àe
À2d
sc

= 2Q ð Þ

d
sc
,
(4:75)
where ~ « ¼ «
s
À«
(ox)
s

= «
s
þ«
(ox)
s

, where «
(ox)
s
is the static dielectric constant of SiO
2
. For
scattering with charges uniformly distributed in (Àd
2
, Àd
1
) in the oxide, after integration
over the distribution of the charges:
«
ox
¼ «
s
þ«
(ox)
s

2, (4:76)
N
ox
¼ N
(3D)
ox
(d
2
Àd
1
), (4:77)
I
(ox)
Q
(z,z
0
) ¼ e
ÀQ(zþz
0
)
e
À2Qd
1
Àe
À2Qd
2

= 2Q(d
2
Àd
1
) ½ Š, (4:78)
where N
(3D)
ox
is the volume density of charges in the oxide. Finally, for scattering with
interface charges we have:
«
it
¼ «
s
þ«
(ox)
0

2, (4:79)
N
it
¼ areal density of interface charges, (4:80)
I
(it)
Q
(z,z
0
) ¼ e
ÀQ(zþz
0
)
: (4:81)
Intervalley processes can be considered by accounting for the degeneracy of the final
valleys and modifying the wavevector Q to account for the valley separation in the
Brillouin zone. In general, the larger wavevector in the denominator of Equation 4.70
will reduce the strength of these collisions.
4.3.6 Multi-Subband Linear Screening Approximations
The major computational difficulty is the evaluation of the screened form factor
hH
nm
(Q)i
screened
in Equation 4.70. Assuming static screening, and employing the high-
temperature, nondegenerate limit of static screening when many subbands are signifi-
cantly populated, the squared matrix elements of the screened potential can be written as
N
r
e
2
H
(r)
nm
(Q)

screened
«
2
r
Q
2
¼

þ1
À1
dz
e
N
r
(z
e
) w
(s)
nm
(Q; z
e
)

2
, (4:82)
ß 2008 by Taylor & Francis Group, LLC.
where
w
(s)
nm
(Q; z
e
) ¼

1
0
dzz
n
(z)w
(s)
(Q,z; z
e
)z
m
(z): (4:83)
The potential w
(s)
(Q,z; z
e
) here is the screened Fourier–Bessel component of the unscreened
Coulomb potential for charges located at z ¼z
e
[6]:
w(Q,z; z
e
) ¼
e
«
(ox)
s
þ«
s
e
ÀQjzj
(interface charges, z
e
¼ 0), (4:84)
w(Q,z; z
e
) ¼
e

s
Q
e
ÀQjzÀz
e
j
þ ~ «e
ÀQjzþz
e
j

(ionized impurity at z
e
), (4:85)
w(Q,z; z
e
) ¼
e
«
(ox)
s
þ«
s

Q
e
ÀQjzÀz
e
j
(oxide charge at z
e
): (4:86)
The intrasubband (diagonal) screened matrix elements are obtained from the unscreened
matrix elements
w
mn
(Q; z
e
) ¼

1
0
dzz
n
(z)w(Q,z; z
e
)z
m
(z), (4:87)
by solving the linear problem [6,13,14,38,40,41]:
w
nn
(Q; z
e
) À
¸
l
b
s,l
(Q)
Q
G
nn,ll
(Q)w
(s)
ll
(Q; z
e
), (4:88)
while the intersubband (off-diagonal) matrix elements can be obtained fromthe intrasubband
elements from:
w
(s)
nm
(Q; z
e
) ¼ w
nm
(Q; z
e
) À
¸
l
b
s,l
(Q)
Q
G
nm,ll
(Q)w
(s)
ll
(Q; z
e
), (4:89)
where
G
mn,ll
(Q) ¼

1
0
dz

1
0
dz
0
z
n
(z)z
n
(z)
~
G
Q
(z,z
0
)z
l
(z
0
), (4:90)
is the form factor,
~
G
Q
(z,z
0
) ¼ e
ÀQjzÀz
0
j
þ ~ «e
ÀQ(zþz
0
)
, (4:91)
is the Poisson Green’s function, here given for the case of a single semiconductor–insulator
interface, see Ref. [37] and below (see Equation 4.138) for more general geometries,
b
s,n
(Q) ¼
e
2
n
n

s
k
B
T
g
1
(qL
n
) b
(DH)
s,n
g
1
(qL
n
), (4:92)
ß 2008 by Taylor & Francis Group, LLC.
is the static screeni ng par ameter we have derive d abov e, while the fun ction g
1
( x) is given
by Equati on 4.58, in terms of the plas ma dis persion fun ction F (y), and
b
(DH )
s , n
¼
e
2
2 «
s
@ n
s
@ E
F
(4: 93)
is the De bye –Hückel parame ter for subb and n, a simp le extension of Equation 4.53 in ter ms
of the thermal pop ulatio n in the nth subband , n
n
, and of the thermal wave length of
elect rons in subb and n, L
n
¼ [2 ph
2
=( m
c, n
k
B
T)]
1=2
, where m
c, n
is the conduct ivity ma ss in
the nth subband, and n
s
is the elect ron sheet density.
4.3.7 Ballisti c Limit and the DOS Bottlenec k: The Role of Impurity Scatte ring
A reductio n of the low- field m obility is the fi rst obvi ous and well-stu died effect of carrier-
impu rity scatteri ng. We do not wi sh to discuss her e the releva nce of the low- field mobil ity
to the perform ance of devices at the 10 nm scale. Some of these iss ues can be foun d
in Refs. [42,43] for opp osite views of the probl em. It will suffi ce to say that the
on-p erformanc e of shor t-channe l device is largel y inse nsitive to Coulomb sca ttering on
any type, becaus e of the weaken ing of the scatteri ng rate as some inverse pow er of the
kineti c energy of the carriers ( $ E
À 1=2
in the par abolic -band appro ximation shown in
Equati on 4.2 7 or 4.62). He re, we wou ld lik e to men tion the recent controve rsy surroundi ng
the possibl e role of sca ttering betw een electro ns in the chann el of a thin-insu lator MOSFE T
and ioni zed impurit ies in the deplete d region of the poly crystalline Si (poly -Si) gate [44].
The controve rsy is ba sed on our fund amental ina bility to treat corre ctly dielectri c scr eening
in the highly inho mog eneous electron ga s presen t in the depleted poly -Si gate [45]. The
state of the art bei ng represen ted by the approximat e staircas e (or piecew ise uniform )
appro ach rec ently fo llowed by Ishi hara [46].
A mo re impo rtant aspect of impurit y scatte ring, thi s tim e an unexpe cted and some what
surpri sing bene ficial effect is relate d to the abilit y of mo mentu m rando mizing collision s—
such as elect ron –impu rity coll isions — in the sou rce of a device ope rating clos e to the
ballisti c-trans port regime to sup ply carri ers which can be injected in the channel. The
discuss ion of this iss ue brings us back to the problem known as DOS bot tleneck.
The search for faster devi ces a ble to sustai n Mo ore ’s law is clearly focuse d on semi con-
ductors with a sm all effect ive mass, since this translate s both into a large r ballistic veloc ity
y for a given kinetic energy E, y $ (2 E=m *)
1=2
, and into a reduce d scatte ring rate. Indeed
assum ing at zeroth order that the scatteri ng rate scales with the DOS [47], we would
have 1=t / m
*3 =2
fo r par aboli c bands . Therefor e, III –V c ompound semico nductor s with
effect ive masses muc h smaller than that of Si have been always looked at with gr eat
inter est. The probl em wi th this picture is that wha t matters in determini ng the speed of a
devi ce in a circ uit is the m aximum current the devi ce can sustai n. In the ballisti c li mit, the
cur rent densi ty, given by the produ ct of the carrier densi ty n (scaling as the DOS, / m
*3 =2
)
and their velo city y, scales as ny / m *. In words, in the ballistic limit a larger eff ective
mass is desirable [48]. The problem actually dates back to 1991 [49]. In investigating
the transconductance of nominally identical devices (MOSFETs), but built on different
substrates (n-Si, p-Si, Ge, GaAs, InP, In
0.53
Ga
0.47
As) we found that—with the exception of
In-containing alloys we discuss momentarily—as the channel length approached 50 nm the
performance seemed unrelated to the low-field mobility (and so m*) of the semiconductor
employed. The explanation was two-fold. First, the DOS of face-centered cubic (fcc)
semiconductors at an energy of roughly 0.5 eV above the band extremum (bottom of the
conduction band or top of the valence band) is very similar, as the many satellite valleys
contri bute in rough ly the same way (see Figu re 4.6). Thus, GaA s does no t appe ar to beha ve
ß 2008 by Taylor & Francis Group, LLC.
different ly from Si as el ectron leave the low- mass botto m of the G valle y and transf er to the
heavier-m ass satellite L (or even X) satellite va lleys. Only semico nductors exhibiti ng high-
energy satel lite valle ys (the In-c ontai ning mate rials in Figu re 4.6) ret ain a notic eable
advantage when compar ed to Si. Seco nd, the low DOS of small-mas s materi als implies
that a large swing of the Fermi level is required to change substantially the carrier density in
the channel. This means that a large swing of gate voltage is required to change substantially
the drain current of such a device and the resulting transconductance is depressed.
What was not conside red at the time (wh en 50 nm-l ong devices were on the thre shold of
being cons idered scie nce- fiction al devi ces) was the behavior which devi ces wou ld exh ibit
at even shor ter dime nsions approac hing the ballistic limit . In this case, a second flavo r of
the DOS bottleneck is uncov ered (as shown in more rec ent simulat ions pe rformed all way
to the 15 nm chann el length [42]) . As transpo rt approach es the ba llistic regime , calcul ation s
of the ga te capa cita nce assu ming an equil ibrium distribut ion of carriers even close to the
source =channel junct ion wi ll nec essari ly result in optimi stic overes timati ons. While st ream-
ing ballistical ly, carrier s pop ulate mainly state s of waveve ctor k (or velo city) aligned along
the transpo rt directi on (we call these states longi tudinal k states) . Mo mentu m rando mizing
collisi ons are requi red to repl enish transverse k states. In absence of collisi ons, the densi ty
drops signi fi cantly whe n compar ed to the equilibr ium value . Fa st, light-m ass mate rials
approach the ballisti c regim e at leng ths longe r than heavier-m ass semico nductor s. This
result s in the disappoi nting behavior also of In-based chann els, as see n in Figure 4.7 fo r a
variet y of structur es (bulk, silico n-on-insu lator [SOI], a nd double gate [DG] FETs). This is
indeed con sistent wi th the ide a that a large mass is actually bene ficial as we approach the
ballistic regime .
Comi ng final ly to the role played by impu rities , a third flavo r of the DOS bottlene ck,
which may be labele d as the source st arvation eff ect, emerge s when we conside r in more
detail III –V-ba sed devi ces at the 22 nm no de [50]. A protot ypical device is sch ematica lly
shown in Figu re 4.8. The channel is a 2.5 nm-thic k In
0.53
Ga
0.47
As quantum well under a
thin HfO
2
gate insulator layer and over a 2.5 nm-thick InP barrier layer on an Al
0.48
In
0.52
As
0.00 0.05 0.10 0.15 0.20 0.25 0.30
0
500
1000
1500
n-Si
p-Si
n-GaAs
n-InP
n-InGaAs
n-Ge
300 K
Channel length (μm ) Channel length (μm )
T
r
a
n
s
c
o
n
d
u
c
t
a
n
c
e

(
μ
S
/
μ
m

)

0.00 0.05 0.10 0.15 0.20 0.25 0.30
0
500
1000
1500
77 K
n-Si
p-Si
n-GaAs
n-InP
n-InGaAs
n-Ge
FIGURE 4.6
Large-signal transconductance g
m
as a function of (metallurgical) channel length calculated in 1991 using the
Monte Carlo program DAMOCLES for similarly designed bulk MOSFETs built on different semiconductors. Both
at 300 K (left) and 77 K (right) only In-based devices exhibit improved performance when compared to the Si
benchmark. (Reprinted from Fischetti, M.V. and Laux, S.E., IEEE Trans. Electron. Dev., 38, 650, 1991. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
insula tor. Fina lly, another p
þ
-d oped (5 Â 10
19
cm
À 3
) In
0.53
Ga
0.47
As gr ound-pl ane control s
shor t-channe l effect s. On e sh ould no tice that in this original device design the source and
drain (S=D) con tacts we re n-typ e doped to the density of (2 Â 10
19
impu rities =cm
3
. As
shown in Figure 4.9, the maximum on-current of the device was found to be very disap-
pointing. The DOS bottleneck is clearly in effect here. Yet, several modifications of the
original device design were attempted, in an effort to minimize ballistic transport. Surpris-
ingly, three crucial modifications helped quite dramatically: (1) increasing the doping in the
S=D regions to (5 Â 10
19
dopants=cm
3
(device type 5 in Figure 4.10); (2) employing raised
S=D contacts, exhibiting a larger area (device type 6 in Figure 4.10), and (3) employing a
thicker (5 nm instead of the original 2.5 nm) channel (device type 7 in Figure 4.10).
The explanat ion is on ce more based on the fallacy of assumi ng an equ ilibrium carri er
distri bution, but now not only in the chann el, but also a few diffusion -lengt hs insid e the
hea vily doped sou rce regio n. At or nea r the ballisti c regime , carriers are carried away so
ef ficiently and fast along the chann el that the sou rce region beco mes starved of carrier s
0 20 40 60 80 100
0
5
10
15
InGaAs
Ge
Si
Double gate FET
Gate lentgh (nm)
T
r
a
n
s
c
o
n
d
u
c
t
a
n
c
e
,

g
m

(
m
S
/
μ
m
)

0 20 40 60 80
0
2
4
6
8
10
InGaAs
Ge
InP
Si
Bulk FET
T
r
a
n
s
c
o
n
d
u
c
t
a
n
c
e
,

g
m

(
m
S
/
μ
m
)

T
r
a
n
s
c
o
n
d
u
c
t
a
n
c
e
,

g
m

(
m
S
/
μ
m
)

0 20 40 60 80
0
2
4
6
8
10
Ge
Si
FD–SOI FET
Metallurgical channel length (nm)
FIGURE 4.7
As in Figure 4.6, but now for different device structures (DG at left, bulk, fully depleted SOI at right) and
semiconductors. The trend is similar to what shown in Figure 4.6, but even In-based devices are affected by the
DOS bottleneck (as they operate even closer to the ballistic limit) and show disappointing performance. The
seemingly erratic trend at the shortest dimension is due to different device parameters (S=D doping, channel
thickness) causing different amount of source starvation. (Reprinted from Fischetti, M.V. et al., IEEE Trans.
Electron. Dev., 54, 2116, 2007. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
populating longitudinal k states. Unless the momentum relaxation time in the source
becomes sufficiently short (thanks to more frequent collisions), there will not be carriers
which can be injected into the channel and contribute to the drain current. The source
becomes starved of electrons in longitudinal k-states. Ahigher doping in the source supplies
more carriers which can be scattered and redirected into longitudinal kstates. It also reduces
the momentum relaxation time, as there are more impurities to which electrons can transfer
momentum. Raised S=Dincrease the acceptance cone, effectively enlarging the volume over
which scattering can redirect carriers into the longitudinal k states. Finally, a thicker channel
also helps in widening the acceptance cone.
Al
.48
In
.52
As
undoped
p 5 ϫ10
19
In
.53
Ga
.47
As
InP
HfO
2
n 2 ϫ10
19
n 2 ϫ10
19

3.7 nm
10 nm
3.5 nm
2.5 nm
2.5 nm
Source
Gate
Drain
Substrate
In
.53
Ga
.47
As
FIGURE 4.8
Schematic diagram of the InGaAs-channel device
simulated (along with additional design variations
described in the text) in order to quantify the role
of the DOS bottleneck and source starvation effects.
(Reprinted from Fischetti, M.V. et al., IEDM Tech.
Dig., 109, 2007. With permission.)
0.00 0.25 0.50 0.75 1.00 1.25
0
1
2
3
4
5
Thick (5 nm) channel
Raised S/D
Higher S/D doping
Si (2.5 nm channel)
p
+
under gate
Source extension
Original
InP launcher
L = 20nm V
DS
= 1.0 V
I
D
S

(
m
A
/
μ
m
)
L = 20nm
V
DS
= 1.0 V
0.00 0.25 0.50 0.75 1.00 1.25
10
4
10
2
10
−2
10
0
60 mV/dec
V
GS
–V
T
(V)
I
D
S

(
A
/
m
)
Thick (5 nm) channel
Raised S/D
Higher S/D doping
Si (2.5 nm channel)
p
+
under gate
Source extension
Original
InP launcher
FIGURE 4.9
Calculated turn-on characteristics for the InGaAs device illustrated in Figure 4.8, along with several design
variations. Note how higher S=D doping, thicker channel, and raised S=D regions improve the performance
quite dramatically. Also shown is the behavior of a similar device but with a 2.5 nm-thick Si channel and HfO
2
front (gate insulator) and back barriers. All curves have been shifted at the same threshold for comparison.
(Reprinted from Fischetti, M.V. et al., IEDM Tech. Dig., 109, 2007. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
In conclusion, impurities seem to be highly desirable as we approach the ballistic regime.
Models based on the assumption that carriers are injected from the source in equilibrium
conditions [43] fail to capture this behavior.
4.4 Interface Roughness
Scattering with the roughness (at the atomic scale) of semiconductor–insulator is being
blamed for the disappointing performance of devices at the 10 nm scale [51]. As devices
shrink, so does the thickness of the gate insulator, t
ox
. Since historically t
ox
has been scaled
at a rate much faster than other dimensions and supply bias (since it is the easiest knob
to turn in order to increase the device performance), carriers are confined ever so harder
against the semiconductor–insulator interface and scattering with roughness at this inter-
face becomes a dominant effect.
A full atomistic (or ab initio) treatment of the atomic roughness requires complex
calculations. Recently progress has been made in this area [52]. Evans and coworkers
have employed density functional theory (DFT) at various levels of sophistication to
investigate the structure of specific microscopic defects (suboxide bonds and oxygen
protrusions)—which may be identified as roughness steps at the Si–SiO
2
interface—and
Green’s functions techniques to quantify their effect on the electron mobility, obtaining
results in flattering agreement with experimental observations not only for the relaxed
interface, but also for the strained-Si–SiO
2
interface which had emerged as a puzzle for the
theory [53]. Here, wishing to avoid these complex techniques, we follow, instead and
historically, a phenomenological model proposed originally by Ando [6] which—perhaps
thank to the presence of adjustable parameters—has proven successful in explaining many
features of this scattering process. Mainly, its dependence on the confining field (the
scattering rate increasing roughly with the square of the surface field F
s
). In its original
form this model considered two major broad classes of effects. The first is related to
the modification of the wave function zn(z) of carriers caused directly by the shift of the
location of the interface at a position r on the plane of the interface. We refer to this shift as
to a step at the interface. Microscopically it can be viewed as the presence of an atomic step
between adjacent terraces on an imperfect interface. The matrix element for the scattering
FIGURE 4.10
Calculated maximum large-signal transconductance
extracted from the data shown in Figure 4.9.
(Reprinted from Fischetti, M.V. et al., IEDM Tech.
Dig., 109, 2007. With permission.)
0 1 2 3 4 5 6 7 8
0
2
4
6
8
10
O
r
i
g
i
n
a
l

d
e
v
i
c
e
I
n
P

l
a
u
n
c
h
e
r
S
o
u
r
c
e

e
x
t
e
n
s
i
o
n
P
a
r
t
i
a
l

g
r
o
u
n
d

p
l
a
n
e

d
o
p
i
n
g
R
a
i
s
e
d

S
/
D
T
h
i
c
k

c
h
a
n
n
e
l
H
i
g
h

S
/
D
d
o
p
i
n
g
20 nm InGaAs–HfO
2
MOSFET
Si
40 nm
Device type
T
r
a
n
s
c
o
n
d
u
c
t
a
n
c
e

(
m
S
/
μ
m
)
ß 2008 by Taylor & Francis Group, LLC.
process is thus simply the energy-shift caused by a weaker or stronger confinement when
the interface moves away or closer to the semiconductor. This effect is, sadly, the only effect
usually considered in the literature, with minor exceptions. The second class of effects—
which we show can play a role of the same magnitude—is related to the perturbation
potential due to Coulomb effects. A shift of the charge density of the 2D electron gas which
modifies the potential in the inversion layer at the step; the emergence of dipoles caused by
both the shift of the polarization (or image) charge at the interface between the two
dielectric with different dielectric constants and by the polarization of the vertical wall of
the step induced by the inversion charges and their images. Clearly, the last effects can be
safely ignored in the case of rough interfaces at heterojunctions between III and V com-
pound semiconductors with very similar dielectric constants. This might explain the
reasons why they have been ignored. Yet, in the interesting cases of the Si–SiO
2
(or even
Si=high-k-dielectrics) the potential due to these potentials can be very important.
In its original formulation, Ando’s model was applicable only to a single interface, case
appropriate to bulk MOS devices of the past. Scaled devices of the present technology fight
short-channel effects mainly using thin semiconductor layers (as in SOI or DG devices). The
presence of the bottominterface presents significant challenges froma theoretical perspective.
Obviously, roughness at this interface yields an additional scattering mechanisms. But, even
assuming a perfect bottom interface, the polarization (or image) charges at this interface
modify the magnitude of the Coulomb terms proposed by Ando. In addition, the boundary
conditions at the bottom interface affect also the magnitude of the modifications to the wave
functioncausedby the step(the first class of processes inAndo’s formulation). Thus, revisiting
the problemand extending the model to the case of thin-body devices is of great interest. This
is the major goal of this section which is based heavily on our recent work [20].
4.4.1 Effective Potential in Thin-Body Si
Our prototypical structure is an SOI MOSFET with two Si–SiO
2
interfaces at z ¼0 and t
s
.
We assume that the confinement direction is along the z-axis and a separable dispersion, so
that for ideally smooth interfaces the wave function zn(z) and associated eigenenergy E
ˆ
of
the ˆth subband are obtained from the Schrödinger equation:
H
z
§
n
(z) ¼ E
n
§
n
(z), H
z
¼
h
2
2
@
@z
1
m
z
@
@z
þV(z) þF(z), (4:94)
where the mass along the quantization direction, m
z
, is given by the Si effective mass m
si
z
for
0 <z <t
s
, and by the SiO
2
mass m
ox
z
otherwise. The presence of the interfacial barriers is
described by the barrier potential F(z) ¼V
ox
[Q(Àz) þQ(z Àt
s
)], where V
ox
is the barrier
height, and Q(z) is the Heaviside step function. The effective potential energy V(z) accounts
for three terms: the Hartree potential, the image-potential (which requires some discussion
in the case of thin layers as it differs from the usual case), and an exchange-correlation term:
V(z) ¼ Àef(z) þV
im
(z) þV
xc
(z): (4:95)
The Hartree term Àef is obtained self-consistently from the Poisson equation. The image-
potential arises from the discontinuity of the dielectric constants, «
si
and «
ox
in the Si layer
and in the SiO
2
barriers, respectively, and it is given by [54]:
V
im
(z) ¼
e
2
16p
¸
1
n¼0
~ «
2n
«
si
~ «
z þnt
s
þ
~ «
t
s
Àz þnt
s
þ
2~ «
2
nt
s
þt
s
¸
, (4:96)
ß 2008 by Taylor & Francis Group, LLC.
where, in the usual notation, ~ « ¼(«
si
À«
ox
)=(«
si
þ«
ox
). Finally, for the exchange-correlation
term, V
xc
, here we employ the local-density approximation within the density functional
formalism [55,56] in the form given by [57–60]:
V
xc
(z) ¼ À
e
2
8p«
si
a
2
par
s
1 þ0:7734x ln 1 þ
1
x
¸
, (4:97)
where
a¼(4=9 p)
1=3
r
s
¼a
À1
[4pn(z)=3]
À1=3
x ¼r
s
=21
n(z) is the electron volume density
a ¼4p«
si
h
2
=(m*e
2
) is the effective Bohr radius
The anisotropy of the Si band structure close to the conduction band minima creates some
concerns about the proper choice for the effective mass entering the expression for the
effective Bohr radius. Lacking better justifications, here we choose the (average) conduct-
ivity effective mass m
*
¼ 3=ð2m
À1
t
þm
À1
l
Þ, as suggested in Ref. [61].
These components of the effective potential energy and the resulting total potential are
shown in Figure 4.11. Note that exchange-correlation effects lower the potential energy
near the center of the well, while image-potential effects raise it close to the interfaces. The
net result is a stronger confinement when image and exchange-correlation effects are
accounted for compared to results obtained ignoring these terms.
4.4.2 Roughness as a Perturbation
Let us now move away from the ideal case of smooth interfaces and consider atomic
roughness at the top interface. We describe a single atomic step at position r on the (x, y)-
plane of the interface as a motion of the semiconductor–insulator boundary moves from
z ¼0 to z ¼D(r). Assuming that roughness is determined by random processes, it is safe to
assume that the Fourier components D
q
of D
r
exhibit a power spectrum with zero mean
characterized by the Gaussian model [6]:
D
q

2
¼ pD
2
L
2
e
Àq
2
L
2
=4
, (4:98)
FIGURE 4.11
Effective potential energy V
0
inside the
silicon body and its three components
(Àef, V
im
, and V
xc
) when t
s
¼ 5 nm,
N
s
¼3 Â10
12
cm
À2
, and T¼300 K. The
device is operated in the SG mode. The
reference of the electrostatic energy is
the Fermi energy (E
F
¼ 0 eV). (Rep-
rinted from Jin, S., Fischetti, M.V., and
Tang, T.-W., IEEE Trans. Electron Dev.,
54, 2191, 2007. With permission.)
80
60
40
20
−20
−40
−60
−80
0 1 2 3
Position (nm)
SG mode, t
s
= 5 nm
N
s
= 3ϫ10
12
cm
−2
, T = 300 K
V
im
V
xc
−ef + V
im
+ V
xc
−ef
E
n
e
r
g
y

(
m
e
V
)
4 5
0
ß 2008 by Taylor & Francis Group, LLC.
or by an exponential model [62]:
D
q

2
¼ pD
2
L
2
(1 þq
2
L
2
=2)
À3=2
, (4:99)
depending on its autocorrelation function. The latter seems to better agree with the
experimentally observed power spectrum [62] and will be employed here.
This model of the roughness changes directly the barrier potential to:
F
r
(z) ¼ V
ox
[Q(D
r
Àz) þQ(z Àt
s
)]: (4:100)
This modification corresponds to the first class of processes discussed above. The second
class is related to first-order modifications of the effective potential energy due to Coulomb
effects associated with the distortion of the interface. In order to extract the leading-order
modifications to the wave function (first class) and the leading-order Coulomb terms
(second class), we consider the perturbed Schrödinger equation along the z direction:
H
zr
z
nr
(z) ¼ E
nr
z
nr
(z): (4:101)
Note that now the Hamiltonian H
zr
¼ À
h
2
2
@
@z
1
m
zr
@
@z
þV
r
(z) þF
r
(z), the wave function z
nr
(z),
and the energy level E
nr
become r-dependent. Here m
zr
¼ m
si
z
if D
r
<z <t
s
, and¼m
ox
z
other-
wise, expressing the fluctuation of the effective mass. Retaining only terms of order O(D
q
), we
can express the r-dependent quantities V
r
(z), zn
r
(z), and E
nr
in terms of their Fourier compon-
ents:
V
r
(z) ¼ V(z) þ
¸
q
V
1
q
(z)D
q
e
iq:r
, (4:102)
z
nr
(z) ¼ z
n
(z) þ
¸
q
z
1
nq
(z)D
q
e
iq:r
, (4:103)
E
nr
¼ E
n
þ
¸
q
E
1
nq
D
q
e
iq:r
: (4:104)
The major task consists now in determining the unknown coefficients V
1
q
(z), z
1
nq
(z), and
E
1
nq
. Indeed the matrix element of the scattering potential is obtained from the difference
V
SR
nkn
0
k
0 ¼ H
nkn
0
k
0 ÀH
0
nkn
0
k
0 , (4:105)
where
H
0
nkn
0
k
0 ¼

dr

dz
e
ÀikÁr
ffiffiffiffi
A
p z
n
(z)H
0
R
z
n
0 (z)
e
ik
0
Ár
ffiffiffiffi
A
p , (4:106)
with
H
0
R
¼ À
h
2m
x
@
2
@x
2
À
h
2
2m
y
@
2
@y
2
þH
z
, (4:107)
is the unperturbed Hamiltonian, while
H
nkn
0
k
0 ¼

dr

dz
e
ÀikÁr
ffiffiffiffi
A
p z
nr
(z)H
R
z
n
0
r
(z)
e
ik
0
Ár
ffiffiffiffi
A
p , (4:108)
ß 2008 by Taylor & Francis Group, LLC.
with
H
R
¼ À
h
2
2m
x
@
2
@x
2
À
h
2
2m
y
@
2
@y
2
þH
zr
(4:109)
is the first-order Hamiltonian in the presence of roughness. The matrix elements of the
unperturbed Hamiltonian are easily determined from Equations 4.94 and 4.106:
H
0
nkn
0
k
0 ¼ [E
k
þE
n
]d
nkn
0
k
0 , (4:110)
where E
k
¼

h
2
k
2
x

2m
x

þ

h
2
k
2
y

2m
y

is the kinetic energy of wave vector k and d
nkn
0
k
0 is
the Kronecker delta. The matrix elements of the scattering potential can be obtained to first
order in D
q
from Equations 4.101 through 4.105:
V
SR
nkn
0
k
0 ¼ E
k
a
nn
0
q
þE
k
0 a
n
0
n
q
þE
1
nq
d
nn
0

D
q
, (4:111)
where q¼kÀk
0
, and the coefficient a
nn
0
q
is defined as
a
nn
0
q
¼

dzz
n
(z)z
1
n
0
q
(z): (4:112)
Our next step now is to express the as-yet unknown coefficients E
1
nq
and z
1
nq
(z) in Equation
4.111 in terms of unperturbed quantities. Since first-order corrections to the energy
eigenvalue E
1
nq
can be determined by the zeroth-order wave function, any set of wave
functions correct to zeroth order can be employed. Here, we use the following form [63,64]:
z
0
nr
(z) ¼
t
s
t
s
ÀD
r
¸
1=2
z
n
(~z), (4:113)
~z ¼
t
s
(z ÀD
r
)
t
s
ÀD
r
: (4:114)
This form is quite convenient since it satisfies the following relations:

dzz
0
nr
(z)F
r
(z)z
0
n
0
r
(z) ¼

dzz
n
(z)F(z)z
n
0 (z), (4:115)

dzz
0
nr
(z)
@
@z
1
m
zr
@
@z
z
0
n
0
r
(z) ¼
t
s
t
s
ÀD
r

2

dzz
n
(z)
@
@z
1
m
z
@
@z
z
n
0 (z): (4:116)
The point to note is that we can now obtain the first-order corrections to the eigenvalues E
1
nq
without making any direct reference to the barrier height V
ox
, while accounting automatic-
ally for the discontinuity of the effective mass. Using Equation 4.113, we can express E
nr
as
E
nr
¼

dzz
0
nr
(z)H
zr
z
0
nr
(z) ¼ E
n
þ
¸
q
G
0
nn
þG
q
nn

D
q
e
iqÁr
: (4:117)
Note once more the separation between first class and second class effects. G
0
nn
is related
to the modification of the wave function due to the shift of the boundaries and it is given by
G
0
nn
0 ¼

dzz
n
(z) À
h
2
t
s
@
@z
1
m
z
@
@z
þ
@V(z)
@z
1 À
z
t
s
¸
z
n
0 (z): (4:118)
ß 2008 by Taylor & Francis Group, LLC.
On the other hand, the termG
q
nn
originates from the fluctuations of the potential energy due
to Coulomb effects (second class). It is given by
G
q
nn
0
¼

dzz
n
(z)V
1
q
(z)z
n
0 (z): (4:119)
Thus, we can identify the coefficients E
1
nq
as
E
1
nq
¼ G
0
nn
þG
q
nn
0
: (4:120)
Fromfirst-order perturbation theory, using the above expression, we can find the following
first-order expression for z
nr
(z):
z
n
0
r
(z) ¼ z
0
n
0
r
(z) þ
¸
q
D
q
e
iqÁr
¸
n6¼n
0
G
0
nn
0 þG
q
nn
0
E
n
0 ÀE
n
z
n
(z): (4:121)
Substituting z
0
n
0
r
(z) in this equation as
z
0
n
0
r
(z) % z
n
0 (z) þ
z Àt
s
t
s
@z
n
0 (z)
z
þ
z
n
0 (z)
2t
s
¸
D
r
, (4:122)
we get for the coefficients z
1
n
0
q
(z):
z
1
n
0
q
(z) ¼
z Àt
s
t
s
@z
n
0 (z)
@z
þ
z
n
0 (z)
2t
s
þ
¸
n6¼n
0
G
0
nn
0 þG
q
nn
0
E
n
0 ÀE
n
z
n
(z): (4:123)
Using this expression we can write the coefficients a
nn
0
q
for n 6¼ n as
a
nn
0
q
¼ À

dzz
n
(z) 1 À
z
t
s

@z
n
0 (z)
@z
þ
G
0
nn
0 þG
q
nn
0
E
n
0 ÀE
n
, (4:124)
while a
nn
q
¼ 0 and it can be easily seen that a
nn
0
q
¼ Àa
n
0
n
q
. Finally, inserting Equations 4.120
and 4.124 into Equation 4.111, we obtain the matrix element for SR scattering as
V
SR
nkn
0
k
0 ¼ G
GPN
nn
0 þG
q
nn
0

D
q
, (4:125)
where
G
GPN
nn
0 ¼ (E
n
ÀE
n
0 )

dzz
n
(z) 1 À
z
t
s

@z
n
0 (z)
@z
þG
0
nn
0 : (4:126)
Note that in order to obtain the last expression we have made use of energy conservation
between initial and final states, E
k
þE
n
¼E
k
0 þE
n
0 .
4.4.3 Direct Scattering at Atomic Steps
Let us consider here first the term G
GPN
nn
0 associated with the change of the wave function
due to the shift of the barrier (first class, direct scattering at the step). This is the only term
often considered and it is interesting to see how it is related to the well-known
ß 2008 by Taylor & Francis Group, LLC.
expression originally derived by Prange and Nee. In the limit of an infinitely high-potential
barrier (V
ox
! 1), one can show that (we omit the proof here, see one of the Appendices
of Ref. [20]):
lim
V
ox
!1
G
GPN
nn
0 ¼ G
PN
nn
0 , (4:127)
where G
PN
nn
0 is the well-known Prange–Nee term [3]:
G
PN
nn
0 ¼
h
2
2m
z
@z
n
(0)
@z
@z
n
0 (0)
@z
: (4:128)
We see that the integral expression given by Equation 4.126 reduces to the well-known
Prange–Nee expression, Equation 4.128, in the limit V
ox
! 1. But in the general case of
finite barrier heights, the two expressions G
GPN
nn
0 and G
PN
nn
0 differ and only the former
provides the correct result. This difference is highlighted in Figure 4.12. Here, we show
the diagonal matrix element G
GPN
00
and G
PN
00
in the case of finite barrier heights. Note that in
this case the correct expression should follow dE
0
=dt
s
[65]. This is indeed the case for the
new general expression G
GPN
00
, but the conventional Prange–Nee term G
PN
00
deviates from
the correct behavior as t
s
! 0. This is due to the fact that the original Prange–Nee
expression assumes that the energy of the ground state subband can be neglected com-
pared to the barrier height, so that one can write the following approximation for the wave
function inside the barrier [6,66]:
z
n
(z) % z
n
(0) exp(z=L
n
), (4:129)
where L
n
¼h(2m
z
V
ox
)
À1=2
. This eventually leads to the usual expression
V
ox
z
n
(0)z
n
0 (0) %
h
2
2m
z
@z
n
(0)
@z
@z
n
0 (0)
@z
: (4:130)
FIGURE 4.12
Comparison of the diagonal matrix
element G
GPN
00
and G
PN
00
for the ground
state of the finite square well potential.
The exact solution dE
0
=dt
s
and the
matrix element of the infinite square
well potential are also shown.
The potential barrier height is 3.15 eV
and the effective mass is 0.916m
0
.
(Reprinted from Jin, S., Fischetti, M.V.,
and Tang, T.-W., IEEE Trans.
Electron Dev., 54, 2191, 2007. With
permission.)
10
1
10
0
10
−1
10
−2
10
−3
0.1 1
Well width (nm)
Infinite square well
~t
s
−3
Finite square well potential
V
ox
= 3.15 eV
m
z
= 0.916m
0
dE
0
/dt
s
Γ
GPN
M
a
t
r
i
x

e
l
e
m
e
n
t

(
e
V
/
n
m
)
10
00
Γ
PN
00
ß 2008 by Taylor & Francis Group, LLC.
More correctly, however, one should assume for L
n
:
L
n
¼h[2m
z
(V
ox
ÀE
n
)]
À1=2
: (4:131)
In so doing we see that the factor V
ox
L
n
L
n
0 becomes larger than h
2
=(2m
z
), explaining why
the usual Prange–Nee term gives a smaller matrix element.
Another point of view usually embraced to understand SR scattering in a quantum well
is to consider the fluctuations of the subband energy as the body thickness fluctuates and
look at these fluctuations as scattering agents. This point of view is already included in the
generalized expression for G
GPN
00
. Indeed the matrix element usually associated with thick-
ness fluctuations is generally expressed as [5,67,68]:
V
FL
nkn
0
k
0 % d
nn
0
@E
n
@t
s
D
q
: (4:132)
In the limit of a square well with infinitely high-potential barriers G
GPN
nn
0 becomes:
G
GPN
nn
0 ¼
h
2
p
2
m
z
nn
0
t
3
s
¼
@E
n
@t
s

1=2
@E
n
0
@t
s

1=2
, (4:133)
also containing off-diagonal elements, which yields the well-known t
6
s
dependence of the
mobility [5,68,69]. Our expression for G
GPN
00
contain additional corrections to this term,
corrections arising from the fact that the potential energy V(z) is not constant (flat), but
contains a hump especially noticeable at large carrier densities.
4.4.4 Scattering via Roughness-Induced Coulomb Terms
Let us now consider the second class of processes, those arising from the changes of the
electrostatic potential (in turn due to modifications of the charge) due to the presence of an
atomic step. These are accounted for by the term G
q
nn
0
in Equation 4.125. This scattering
potential V
1
q
(z)D
q
results from three different terms: a term, V
n
q
(z), due to the shift of the
electron (free carrier) sheet density along the z direction as the interface goes through a
step; a term, V
s
q
(z), due to the shift of the polarization charge; and a term, V
im
q
(z), due to a
change of the image-potential. Thus:
V
1
q
(z)D
q
¼ V
n
q
(z) þV
s
q
(z) þV
im
q
(z)

D
q
: (4:134)
The first term is easily evaluated in terms of the Poisson Green’s function appropriate to
the geometry at hand. When the interface shifts by an amount D
q
the change of the sheet
charge, dn
r
(z) ¼n
r
(z) Àn(z), can be approximated to first order as
dn
r
(z) %
t
s
t
s
ÀD
r
n(~z) Àn(z) %
z
t
s
À1

@n(z)
@z
þ
n(z)
t
s
¸
D
r
: (4:135)
For the time being we assume that the number of electrons per unit area remains fixed. This
effect will be considered later when discussing screening effects. Then, the potential V
n
q
due
to the shift dn
r
(z) is immediately given by
V
n
q
(z) ¼ À
e
2
«
si

t
s
0
dz
0
G
q
(z,z
0
)
z
0
t
s
À1

@n(z
0
)
@z
0
þ
n(z
0
)
t
s
¸
, (4:136)
ß 2008 by Taylor & Francis Group, LLC.
where G
q
(z, z
0
) is the Poisson Green’s function which satisfies the equation:
@
2
@z
2
Àq
2

G
q
(z, z
0
) ¼ d(z Àz
0
) (4:137)
for the geometry of interest. In our cases this amounts to satisfying the appropriate boundary
conditions: continuity of the electrostatic potential and the continuity of the normal com-
ponent of the dielectric at each interface. When 0 <z, z
0
<t
s
, G
q
can be written as [16]:
G
q
(z,z
0
) ¼ À
1
2q
1
1 À ~ «
2
e
À2qt
s
e
ÀqjzÀz
0
j
þ ~ «e
Àqjzþz
0
j
þ ~ «e
À2qt
s
e
qjzþz
0
j
þ ~ «e
qjzÀz
0
j

: (4:138)
The second term can be handled by recalling that at the SiÀSiO
2
interface an interface
polarization charge is formed due to the mismatch of the dielectric constants of the two
materials. Its charge density per unit area is given by [70]:
s ¼ À«
0
«
si
À«
ox
«
ox
F
s
, (4:139)
where «
0
is the permittivity of free space and F
s
¼ À@f(0
þ
)=@z the surface electric field. At
a step on the deformed SiÀSiO
2
interface D
r
this interface charge moves from z ¼0 to D
r
.
This causes an effective dipole in the silicon side of the interface located at the SiÀSiO
2
interface given by
p
z
(r) ¼
«
si
«
0
sD
r
¼
«
si
À«
ox
2
ox
«
si
F
s
D
r
: (4:140)
Once more, the electrostatic potential, V
s
q
, due to this dipole can be obtained from the
Poisson Green’s function:
V
s
q
(z) ¼ À
e
«
si
«
si
À«
ox
«
ox
«
si
F
s
@G
q
(z,z
0
)
@z
0

z
0
¼0
¼ e~ «F
s
e
Àqz
þ ~ «e
qzÀ2qt
s
1 À ~ «
2
e
À2qt
s
: (4:141)
Finally, the last term V
im
q
in Equation 4.134 represents the polarization caused, on the
vertical wall of the step, by the image charges. This requires that we calculate to the first
order in D
q
the electrostatic potential caused by a unit charge located in the silicon side of
the interface in the presence of the step. Although straightforward, this derivation involves
quite laborious algebraic work, as it can be seen in one of the Appendices of Ref. [20]. Here,
we simply state the final result which can be cast in the following form:
V
im
q
(z) ¼
e
2
~ «
8p
2

dk
«
si
þ«
ox
«
ox
«
si
þ
k Á k
0
kk
0
¸
e
Àkz
þ ~ «e
kzÀ2kt
s
1 À ~ «
2
e
À2kt
s
¸
e
Àk
0
z
þ ~ «e
k
0
zÀ2k
0
t
s
1 À ~ «
2
e
À2k
0
t
s
¸
, (4:142)
where k
0
¼

k
x
Àq ð Þ
2
þk
2
y

1=2
and kÁ k
0
¼(k
2
þk
02
Àq
2
)=2. Unfortunately, we have not been
able to find any explicit analytic expression exists for the integration above. Thus, we
calculate it numerically and tabulate it as a function of t
s
, z, and q.
As done for the other scattering term, it is instructive to see under which conditions
the Coulomb-related scattering terms reduce to well-known expressions. Indeed, in the
limit t
s
! 1, the matrix element of the SR scattering becomes:
lim
t
s
!1
V
SR
nkn
0
k
0 ¼ lim
t
s
!1
G
GPN
nn
0 þ lim
t
s
!1
G
q
nn
0
¸
D
q
, (4:143)
ß 2008 by Taylor & Francis Group, LLC.
lim
t
s
!1
G
GPN
nn
0 ¼ (E
n
ÀE
n
0 )

dzz
n
(z)
@z
n
0 (z)
@z
þ

dzz
n
(z)
@V(z)
@z
z
n
0 (z), (4:144)
lim
t
s
!1
G
q
nn
0
¼

dzz
n
(z) lim
t
s
!1
V
1
q
(z)z
n
0 (z), (4:145)
lim
t
s
!1
V
1
q
(z) ¼
e
2
«
si
À
1
2q

1
0
dz
0
@n(z
0
)
@z
0
e
ÀqjzÀz
0
j
þ ~ «e
Àqjzþz
0
j

þ ~ «(N
s
þN
d
)e
Àqz
þ
~ «
16p
q
2
Â
K
1
(qz)
qz
À
~ «
2
K
0
(qz)
¸
¸
, (4:146)
where N
s
and N
d
are the number of electrons and depletion charges per unit area,
respectively, and K
0
and K
1
are the modified Bessel functions of the second kind. These
equations are identical to Ando’s equations for the case of inversion layers of bulk
MOSFETs [6]. Thus, the model proposed here can be viewed as an extension of Ando’s
model for finite silicon body thickness, reducing to Ando’s original model in the limit of
infinite body thickness and (for the Prange–Nee term) infinitely high barrier potentials.
4.4.5 Screening
Up to this point we have treated the potential fluctuations caused by interface roughness as
unscreened by the free electrons in the inversion layers. Using the random phase approxi-
mation we can obtained the screened matrix elements
~
V
q;ll
0 from the unscreened ones V
q;ll
0
as [13,14,38]:
~
V
q;ll
0 ¼ V
q;ll
0 þ
¸
m
G
q;ll
0
;mm
b
T
q;mm
~
V
q;mm
, (4:147)
(identical to Equations 4.88 and 4.89) where the form factor G
q,ll
0
;mm
0 is given by Equation
4.90. Rather than using the high-T, degenerate expression for the screening parameter, here
we must use Maldague’s expression valid at finite T also for a degenerate electron gas at
finite-temperature. Indeed, while for bulk MOSFETs at room temperature, one can obtain
b
T
q;mm
in terms of the plasma dispersion function in the nondegenerate limit, as done in
Equations 4.54 and 4.55, in all low-dimensionality situations, and so in the case at hand of
ultrathin-body SOI (UTBSOI) MOSFETs, the DOS is usually smaller than in bulk MOSFETs
because of the large level-splitting between the ground state and the excited states induced
by the quantum confinement. Thus, electrons exhibit a higher degree of degeneracy for a
given density and accounting for degenerate statistics becomes imperative. Therefore, we
compute b
T
q;mm
from the Maldague’s expression as [71]:
b
T
q;mm
(
~
E
F
) ¼

1
0
dE
0
b
0
q;mm
(E
0
)
4k
B
T cosh
2 E
0
À
~
E
F
2k
B
T
, (4:148)
where
~
E
F
¼E
F
ÀE
m
. In the equation above, the screening parameter at zero temperature in
the static limit b
0
q;mm
(
~
E
F
) is given by [72]:
b
0
q;mm
(
~
E
F
) ¼
e
2
r
m
«
si
1 ÀC
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 À
2k
F
q

2

¸
¸
, (4:149)
ß 2008 by Taylor & Francis Group, LLC.
where
r
m
¼g
m
m m=(ph
2
) is the DOS of mth subband
g
m
is the valley degeneracy factor
m
m
¼
ffiffiffiffiffiffiffiffiffiffiffiffi
m
x
m
y
p
is the DOs effective mass
k
F
is the Fermi wavevector
C¼1 if q >2k
F
and C¼0 otherwise
Inserting Equation 4.149 into Equation 4.148, we obtain [73]:
b
T
q;mm
¼
e
2
r
m

si
¸
1 þtanh (y) À

z
0
dx
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Àx=z

cosh
2
(x Ày)

, (4:150)
where
y ¼
~
E
F
=(2k
B
T)
z ¼h
2
q
2
=(16m
op
k
B
T)
m
op
¼2m
x
m
y
=(m
x
þm
y
) is the conductivity effective mass
The integration in this equation is performed numerically. Figure 4.13 shows b
300
0;00
,
the screening parameter of the ground state subband at 300 K in the long wavelength
limit (q !0), as a function of the silicon body thickness for several electron sheet-densities,
assuming that the device is operated in the single gate (SG) mode. We can see that
the nondegenerate statistics significantly overestimates the screening parameter as
the silicon thickness decreases even in the case of relatively moderate electron density
(N
s
¼3 Â10
12
cm
À2
).
Degenerate statistics
Non-degenerate statistics
N
s
= 10
13
cm
−2
10
12
3ϫ10
12
SG-mode, T= 300 K
Ground state, long wavelength
5 10
10
10
10
9
10
8
Silicon body thickness (nm)
S
c
r
e
e
n
i
n
g

p
a
r
a
m
e
t
e
r

(
m

1
)
15
FIGURE 4.13
Screening parameter of the ground state in the long wavelength limit (q ! 0) as a function of the silicon body
thickness for three different electron density (N
s
¼10
12
, 3 Â10
12
, 10
13
cm
À2
). The device is assumed to be operated
in the SG mode at 300 K. (Reprinted from Jin, S., Fischetti, M.V., and Tang, T.-W., IEEE Trans. Electron. Dev., 54,
2191, 2007. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
4.4.6 Mobility Calculation
The low-field mobility for a 2D electron gas can be obtained from the linearization of the
Boltzmann transport equation, as we have outlined above for the 3D case. The general
expression for the xx component of the mobility tensor for nth subband can be derived from
Equation 4.9 and has a form similar to the Kubo–Greenwood expression, Equation 4.12
[13,14,16]:
m
xx
n
¼
eg
n
2h
2
p
2
k
B
TN
n

2p
0
df

1
E
n
dEk
j@E=@k
k
j
@E
@k
x

2
k
t
x
nk
f
E
(1 Àf
E
), (4:151)
where k(E, f) is the wave vector, k ¼jkj, N
n
is the number of electrons in nth subband
including the valley and spin degeneracy, t
x
nk
is the anisotropic momentum relaxation time
for the x component of the momentum in subband n, and f
E
¼{exp[(EÀE
F
)=(k
B
T) þ1]}
À1
is
the Fermi–Dirac distribution function. In the parabolic, ellipsoidal band structure, this
equation becomes:
m
xx
n
¼
er
n
m
x
k
B
TN
n

1
E
n
dE(E ÀE
n
) f
E
(1 Àf
E
)

2p
0
df
(m
f
cos f)
2
pm
x
m
n
t
x
nk
, (4:152)
where m
f
¼(cos
2
f=m
x
þsin
2
f=m
y
)
À1
is the conductivity effective mass along the
fdirection. The momentumrelaxation time t
x
nk
due to SRscattering can be written as [15,16]:
1
t
x
nk
¼
1
2ph
3
¸
n
0
Q(E ÀE
n
0 )

2p
0
df
0
m
f
0
~
V
SR
n
0
k
0
nk

2

1 À
~ t
x
n
0
k
0 y
x
n
0
k
0
~ t
x
nk
y
x
nk
¸
, (4:153)
where
Q(x) is, as before, the Heaviside step function
y
x
nk
¼hk
n
(E,f) cos f=m
x
is the velocity along the x-direction
k
n
(E,f) ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2m
f
(E ÀE
n
)

=h is the magnitude of the wave vector
~ t
x
nk
is the momentum relaxation time due to all scattering processes
The dependence of ~ t
x
nk
inside the square bracket of Equation 4.153 complicates the calcu-
lation, and this dependence has usually been neglected so far. Although SR scattering is
anisotropic, it is elastic and we simplify ~ t
x
n
0
k
0 y
x
n
0
k
0 =~ t
x
nk
y
x
nk
as % y
x
n
0
k
0 =y
x
nk
inside the square
bracket. Finally, the total effective electron mobility is obtained from the weighted average
of the electron mobility in each:
m
xx
eff
¼
1
N
s
¸
n
m
xx
n
N
n
, (4:154)
where N
n
is the electron density of the nth subband.
4.4.7 Simulation Results and Discussion
We now present results regarding the SR-limited electron mobility of UTBSOI MOSFETs
at room temperature (T¼300 K). We have assumed a substrate doping density of
2 Â10
16
cm
À3
and SR parameters L¼1.3 nm and D¼0.47 nm in Equation 4.99 for both
of the top and bottom interfaces. In most cases, we have considered SG SOI MOSFETs with
a silicon dioxide insulator («
ox
¼3.9«
0
), and a bottom oxide thick enough to render the
ß 2008 by Taylor & Francis Group, LLC.
elect ric field wi thin it neglig ible. He re, we also assume an in finite potential barrier height
( V
ox
! 1). Althou gh this assump tion wi ll underes timate the mo bility whe n t
s
! 0, the
result s still illu strate the qualitati vely correct beha vior of the SR-l imited mo bility.
We pres ent first results rega rding the in fluence of the silicon body thi ckness on the
SR-l imited mobilit y. Figu re 4.14 shows the calculate d SR-l imited mob ility as a function of
silicon body thi ckness for three different value s of the el ectron densi ty (N
s
¼ 10
12
, 3 Â 10
12
,
and 10
13
cm
À 2
). At all elect ron dens ities the SR-l imited mo bility exhibi ts a stro ng thi ckness
dep endence as t
s
! 0, but this depen dence weakens , conv erging to the bul k mo bility, when
t
s
!1. In the regi on betw een the se two limits the mo bility exh ibits a sm all maximu m. No te
also that in the first limit of t
s
! 0, the SR-l imited electron mo bility shows the expecte d t
6
s
-
like depen dence due to thick ness fl uctuat ions, more evid ent at the small electro n densi ty of
N
s
¼ 10
12
cm
À 2
. For inc reasing electro n densi ties the thickne ss depen dence on the SR-limite d
mo bility weaken s, as the matrix elem ent is now dominat ed by the fl uctuat ions of the
potenti al energy . The sm all peak of the mobilit y arises from two competi ng effect s:
the already menti oned mo bility degradat ion due to the thickne ss fluct uation as t
s
! 0,
and the mo bility degradat ion due to the increasing occupatio n of the higher -mobili ty
gro und state unprimed valley at the expens e of the low-mo bility prime d valle ys. In Figu re
4.15, we also plo t the indi vidual contri bution of the top and bottom interface s on the SR-
limit ed mo bility for the ca se of N
s
¼ 3 Â 10
12
cm
À 2
. We have st udied the cas es of corre lated,
antico rrelat ed, and uncor related rough ness at both interface s. This is achieved by con sider-
ing the expres sions for G
(GPN)
nn
0
(given by Equatio n 4. 136), a nd the matrix elem ents corre-
spond ing to the pote ntials V
n
q
, V
s
q
, and V
im
q
, given by Equations 4.136, 4.141, and 4.142,
respectively, for scattering at the bottom interface. These will be given by expressions above
quite similar to those given above, but replacing the factors (1 Àz=t
s
) with z=t
s
and account-
ing for a few sign-changes.
To be explicit, the terms reflecting scattering with roughness at the bottom interface will
be given by
G
GPN
nn
0 ¼
h
2
t
s

dz
@
@z
1
m
2
@z
n
0 (z)
@z
¸
þ

dzz
n
(z)
@V(z)
@z
z
t
s
z
n
0 (z)
þ(E
n
ÀE
n
0 )

dzz
n
(z)
z
t
s
@z
n
0 (z)
@z
, (4:155)
FIGURE 4.14
Calculated SR scattering-limited
mobility as a function of silicon body
thickness for three different electron
densities (N
s
¼10
12
, 3 Â10
12
, and 10
13
cm
À2
). The values of the effective field
corresponding to these three electron
densities are approximately 0.077–
0.079, 0.23–0.24, and 0.77 MV=cm,
respectively. (Reprinted from Jin, S.,
Fischetti, M.V., and Tang, T.-W., IEEE
Trans. Electron Dev., 54, 2191, 2007.
With permission.)
10
4
10
3
2 3
t
6
s
4 5
Silicon body thickness (nm)
SG mode, T = 300 K
10
12
3ϫ10
12
S
R
-
l
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2
V

1
s

1
)
6
N
s
=10
13
cm
−2
7 8 9 10
ß 2008 by Taylor & Francis Group, LLC.
V
n
q
( z) ¼
e
2
«
si

dz
0
G
q
( z,z
0
)
z
0
t
s
@ n ( z
0
)
@ z
0
þ
n (z
0
)
t
s
¸
, (4: 156)
V
s
q
¼ Àe ~ «
@ f t
À
s

@ z
e
qz À qt
s
þ ~ «e
À qzÀ qt
s
À~ «
2
e
À 2 qt
s
, (4: 157)
where À @ f( t
À
s
) =@ z is the elect ric field at the botto m int erface , and , fi nally:
V
im
q
(z) ¼
e
2
~ «
8p
3

d k
«
si
þ «
ox
«
ox
«
si
þ
k Á k
kk
0
¸
e
kz À kt
s
þ ~ «e
À kz À kt
s
1 À ~ «
2
e
À 2 kt
s
¸
e
k
0
z À k
0
t
s
þ ~ «e
À k
0
z À k
0
t
s
1 À ~ «
2
e
À 2k
0
t
s
¸
: (4 : 158)
Denoti ng by V
(top)
q
and V
(bottom)
q
the Fourier component s of the scatteri ng potenti al for the
top and botto m inter faces, respec tively , in the evaluati on of the momen tum relaxation rate,
we then conside r either j V
(top)
q
þ V
(bot tom)
q
j
2
fo r cor related rough ness, jV
(top)
q
þ V
(bottom)
q
j
2
for
antico rrelated ro ughness, or, fi nally, j V
(top)
q
þ V
(bottom )
q
j
2
fo r uncorre lated ro ughness. As
expecte d, the c ontribut ion of the bottom interface on the SR-l imited mo bility is appreci able
only when t
s
is sma ller than 6 nm and corre lated ro ughness result s in a min imal mobilit y
degradat ion since thickne ss fluct uations are ab sent in this cas e.
In Figure 4.16, we compar e the SR-limite d elect ron mo bility for SG and DG SOI MOS FETs.
In the t
s
! 0 limit both devices exh ibit ver y similar mo bilities, whil e the SR-limite d mobilit y
in DG SOI MOSFE Ts app roaches the mobilit y of SG SOI MOSFE Ts at one- half the elect ron
densi ty in the t
s
!1limit . Th is is expe cted from the conside ration that in DG SOI MOS FETs
charge transp ort occ urs in two indep endent chann els if t
s
is suf ficiently large. Note also that,
compar ed to the cas e of SG SOI MOSF ETs, the mobilit y enhance ment of DG SOI MOS FETs
in the intermediate range of t
s
is quite noticeable, thanks to the volume inversion, charac-
teristic which has been observed experimentally [74].
Both interfaces
Top interfaces
Bottom interface
SG mode, N
s
= 3 ϫ 10
12
cm
−2
, T = 300 K
t
s
6
1 2
10
4
10
3
10
2
10
1
3
Silicon body thickness (nm)
S
R
-
L
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2

V

1
S

1
)
4 5 6 7 8 9 10
FIGURE 4.15
Calculated SR-limited mobility due to top, bottom, and both interfaces as a function of silicon body thickness for
N
s
¼3 Â10
12
cm
À2
. The contribution of the bottom interface on the SR-limited mobility is appreciable only for t
s
smaller than 6 nm. We also showthe mobility whenthe roughness at the topandbottominterfaces are fully correlated
(the positions of the two interfaces fluctuate in the same direction, crosses), or anticorrelated (the positions fluctuate in
opposite directions, triangles. Note that in the latter case thickness fluctuations vanish. (Reprinted from Jin, S.,
Fischetti, M.V., and Tang, T.-W., IEEE Trans. Electron Dev., 54, 2191, 2007. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
We also show results regarding the influence of the effective field (electron sheet-density)
on the SR-limited mobility, which shows well-known universal characteristics in bulk
MOSFETs [75]. Here, we have defined the effective field as
E
eff
¼
e
N
s

t
s
0
dzn(z)
@f(z)
@z

: (4:159)
Figure 4.17 shows the SR-limited mobility as a function of E
eff
for different silicon
body thickness. As expected, the geometric confinement weakens the dependence of the
SR-limited mobility on E
eff
and the effective field ceases to be a good metric for
FIGURE 4.16
Comparison of the SR scattering-
limited mobility of DG SOI MOSFETs
and SG SOI MOSFETs as a function of
silicon body thickness. (Reprinted
from Jin, S., Fischetti, M.V., and
Tang, T.-W., IEEE Trans. Electron
Dev., 54, 2191, 2007. With permission.)
SG mode, N
s
= 1.5 ϫ 10
12
cm
Ϫ2
T = 300 K
Silicon body thickness (nm)
1 10
SG mode, N
s
= 3 ϫ 10
12
cm
Ϫ2
DG mode N
s
= 3 ϫ 10
12
cm
Ϫ2
S
R
-
L
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2
V
Ϫ
1
s
Ϫ
1
)
10
4
10
3
10
2
10
1
t
6
s
FIGURE 4.17
Calculated SR scattering-limited
mobility as a function of effective
field for four different silicon body
thickness (t
s
¼2, 2.5, 3, and 4 nm).
The slight increase in the mobility
in the low effective field region is due
to the screening effect. (Reprinted from
Jin, S., Fischetti, M.V., and Tang, T.-W.,
IEEE Trans. Electron Dev., 54, 2191,
2007. With permission.)
4
10
4
10
3
10
2
3
2.5
0.1 1
SG mode, T = 300 K
Effective field (MV/cm)
S
R
-
L
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2
V

1
S

1
)
t
s
= 2 nm
ß 2008 by Taylor & Francis Group, LLC.
the charac terization of the mob ility in UTBSOI M OSFETs [66]. We also find a sligh t
increa se of the mo bility in the low E
eff
region for extre mely thi n devices. In Figu re 4.18,
we plo t separatel y the SR -limited mobil ity due to the (gen eralized) Pra nge –Ne e term G
GPN
nn
0 ,
to the potenti al (Coulomb) fl uctuat ion term G
q
nn
0
, and to both term s as a fun ction of effect ive
field for t
s
¼ 2.5 nm. The in fluence of the pote ntial- fl uctuat ion term on the mobilit y is
smaller than the effect of the Prange –Nee term, but the potential (Coulomb) fluct uation
term is nev er negli gible, since it may reduce the mob ility by about 50% with respec t to the
mob ility limit ed by the Prange –Nee alone. Also shown is the unscr eened SR-l imited
mob ility, which, as a con sequenc e of die lectric screeni ng, decreas es mo notoni cally with
increa sing eff ective field .
Fina lly, int eresting results have been obtained by con sidering the infl uence of the
dielectri c cons tant of the insula ting mate rial on the SR-l imited mob ility. Th is issue has
not been addr essed before, perha ps a s a result of havi ng neglecte d Coulo mb terms (thee
image-p otential and the potential - fluct uation ter ms). Intuiti vely, howeve r, we expe ct that
the SR-limited elect ron mob ility will be affec ted by the dielectri c cons tant of the insulatin g
materi al since the difference betwe en the semi conduct or and insula tor die lectric con stants
appe ar in all Coulomb terms. In Equati ons 4.96, 4.141, and 4.142, we see that the image-
potentia l V
im
and the scatteri ng pote ntials V
s
q
and V
im
q
dep end on ~ « ¼ ( «
si
À «
ox
) =( «
si
þ «
ox
).
This par ameter can span the range 1 to À 1 as «
ox
is varied from 0 to 1 , with an obvi ous
sign-c hange occ urring when «
ox
equals «
si
. As «
ox
is inc reased from 0 to «
si
, the repu lsive
image potenti al decreas es and fi nally disappe ar when «
ox
¼ «
si
. When «
ox
> «
si
, on the
contrar y, the image potenti al becomes attra ctive, and the magnitu de inc reases with
increa sing «
ox
. Simila rly, the scatte ring potential V
s
q
and V
im
q
decrease with «
ox
increa sing
from 0 to «
si
, and it vani shes when «
ox
¼ «
si
. For even higher values of «
ox
, the sign of this
term changes , thu s cance ling (i.e., die lectric scr eening due to pol arizati on charges) all ot her
scatte ring pote ntials terms . As a result of the se conside rations we expect that the SR-
limited mobil ity will inc rease with inc reasing «
ox
. This expectatio n is con firme d by the
result s sh own in Figu re 4.1 9. We see that inde ed the SR-limite d mobilit y inc reases with
increa sing dielectri c cons tant «
ox
. In Figu re 4.20, we show the thickne ss depen dence of the
SR-limited mobility for two different dielectric materials (SiO
2
and HfO
2
). It should be
S
R
-
l
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2

V

1

S

1
)
Γ
nnЈ

term only
Γ
nnЈ

term only
Both terms
Both terms, unscreened
0.1
Effective field (MV/cm)
10
4
10
3
10
2
1
SG mode, t
s
= 2.5 nm, T = 300 K
GPN
q
FIGURE 4.18
Calculated SR scattering-limited
mobility due to the potential fluctu-
ation term G
q
nn
0 , generalized Prange–
Nee term G
GPN
nn
0 , and both terms
as a function of effective field for
t
s
¼ 2.5 nm. Also shown is the SR
scattering limited mobility when the
screening is not taken into account.
(Reprinted from Jin, S., Fischetti,
M.V., and Tang, T.-W., IEEE Trans.
Electron Dev., 54, 2191, 2007. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
noted in this discussion that this dependence of the mobility on the dielectric constant of
the insulator should be expected only when the SR parameters remain fixed. Thus, if we
can maintain the same quality of the interface, high-k materials may exhibit higher SR-
limited mobility then SiO
2
. Experimental verification of this trend is not easily obtained,
since the presence of additional scattering mechanisms present in high-k-based MOSFETs
[15] and the difficulty of obtaining smooth interfaces may mask the effect.
S
R
-
L
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2
V

1
s

1
)
0 5 10 15 20
10
12
cm
−2

(0.15 MV/cm)
3ϫ10
12
cm
−2

(0.30 MV/cm)
N
s
= 10
13
cm
−2

(E
eff
~0.84 MV/cm)
Dielectric constant (e
0
)
25 30
Bulk MOSFET, T = 300 K
35 40
10
4
10
3
FIGURE 4.19
Calculated SR scattering-limited mobility of bulk MOSFETs as a function of the dielectric constant of the
insulator («
ox
) for three different electron densities (N
s
¼10
12
, 3 Â10
12
, and 10
13
cm
À2
). (Reprinted from Jin, S.,
Fischetti, M.V., and Tang, T.-W., IEEE Trans. Electron Dev., 54, 2191, 2007. With permission.)
FIGURE 4.20
Calculated SR scattering-limited
mobility as a function of silicon body
thickness for different dielectric
materials (SiO
2
and HfO
2
), where the
same SR parameters are used.
(Reprinted from Jin, S., Fischetti,
M.V., and Tang, T.-W., IEEE Trans.
Electron Dev., 54, 2191, 2007. With
permission.)
S
R
-
L
i
m
i
t
e
d

m
o
b
i
l
i
t
y

(
c
m
2

V

1

s

1
)
10
4
10
3
Silicon body thickness (nm)
SiO
2
(e
ox
= 3.9e
0
)
HfO
2
(e
ox
= 22e
0
)
t
s
6
SG mode, N
s
= 3 ϫ10
12

cm
−2
, T = 300 K
2 3 4 5 6 7 8 9 10
ß 2008 by Taylor & Francis Group, LLC.
4.5 C onclusions
We have conside red two major types of defe cts in semico nductor devi ces (d opants
and int erface rough ness) and have looked at some of the effects they have on electroni c
transpo rt. Regarding the effect of dopant s, bypas sing the obvious effect that
ionize d impu rities have on dep ressing the electron mob ility, we have shown that impurit y
scatte ring in the source of devi ces at the 10 nm scale may be bene fi cial by allowi ng more
electrons to be injected into the quasiballistic channel. This prevents the source from being
starved, that is, from being depleted of carriers occupying those velocity states (longitudinal
k state s) properly oriented to allo w injecti on into the chann el. This is the aspect of the DOS
bottlene ck mo st evid ent in devices ope rating clos e to the ba llistic transp ort regime .
Regardi ng the effect of rough ness at semi cond uctor =insu lator int erface s, we have con sid-
ered the case of thi n Si bodi es (o f inter est in toda y’ s UTBSOI and DG FETs) and exten ded
the well-know n Ando ’s model to this case. In partic ular, we have taken into accoun t all
scatte ring terms, inc luding the of ten ignored Coulom b ter ms due to distorti on of the
inversi on, pol arization and image charge s. Th e new model reduce s to the usual (Ando ’ s)
model in the lim it of infi nite barrier height s (fo r the Pra nge –Ne e term) and body thic kness
(for the Coulo mb ter ms). We have st udied the eff ect of body thickne ss, of scatteri ng at both
inter faces, and of different insu lator dielectric constants. We have see n that the mobilit y
drops signi fi cantly fo r small body thickne ss, that the Coulo mb terms are signi fic ant and
cannot be negle cted, that the degre e of correlat ion of the roughness at both int erfaces
matters signi fi cantly, and fi nally, that insula tors with higher die lectric con stants contri bute
to improve the SR-limited mobilit y (for a fixed rough ness).
Ackn owled gment s
We wo uld like to thank P. Asbec k, Y. Taur, L. Wang, B. Yu, and C. Sachs for dis cussion on
the source st arvation issue outlin ed in Secti on 4.3.7. This wo rk has bee n par tially support ed
by the Semiconductor Research Corporation. An SUR Grant from IBM Corporation is also
acknowledged.
References
1. Conwell, E.M. and Weisskopf, V.F., Theory of impurity scattering in semiconductors, Phys. Rev.,
77, 388, 1950.
2. Brooks, H., Scattering by ionized impurities in semiconductors, Phys. Rev., 83, 879, 1951.
3. Prange, R.E. and Nee, T.W., Quantum spectroscopy of the low-field oscillations in surface
impedance, Phys. Rev., 168, 779, 1968.
4. Saitoh, M., Warm electrons on the liquid
4
He surface, J. Phys. Soc. Japan, 42, 201, 1977.
5. Sakaki, H. et al., Interface roughness scattering in GaAs=AlAs quantum wells, Appl. Phys. Lett., 51,
1934, 1987.
6. Ando, T., Fowler, A.B., and Stern, F., Electronic properties of two-dimensional systems, Rev. Mod.
Phys., 54, 437, 1982.
7. Jacoboni, C. and Reggiani, L., The Monte Carlo method for the solution of charge transport in
semiconductors with application to covalent materials, Rev. Mod. Phys., 55, 345, 1983.
8. Fischetti, M.V. and Laux, S.E., Monte Carlo analysis of electron transport in small semiconductor
devices including band-structure and space-charge effects, Phys. Rev. B, 38, 9721, 1988.
ß 2008 by Taylor & Francis Group, LLC.
9. Kittel, C., Quantum Theory of Solids, Wiley, New York, 1963.
10. Chattopadhyay, D. and Queisser, H.J., Electron scattering by ionized impurities in semi-
conductors, Rev. Mod. Phys., 53, 745, 1981.
11. Kubo, R., Statistical–mechanical theory of irreversible process. I. General theory and simple
applications to magnetic and conduction problems, J. Phys. Soc. Jpn., 12, 570, 1957.
12. Greenwood, D.A., The Boltzmann equation in the theory of electrical conduction in metals, Proc.
Phys. Soc. London, 71, 585, 1958.
13. Fischetti, M.V. and Laux, S.E., Long-range Coulomb interactions in small Si devices. Part I:
Performance and reliability, J. Appl. Phys., 89, 1205, 2001.
14. Fischetti, M.V., Long-range Coulomb interactions in small Si devices. Part II: Effective electron
mobility in thin-oxide structures, J. Appl. Phys., 89, 1232, 2001.
15. Fischetti, M.V, Neumayer, D.A., and Cartier, E.A., Effective electron mobility in Si inversion
layers in metal-oxide-semiconductor systems with a high-k insulator: The role of remote phonon
scattering, J. Appl. Phys., 90, 4587, 2001.
16. Fischetti, M.V. et al., Six-band kÁp calculation of the hole mobility in silicon inversion layers:
Dependence on surface orientation, strain, and silicon thickness, J. Appl. Phys., 94, 1079, 2003.
17. Slater, J.C., Electrons in perturbed periodic lattices, Phys. Rev., 76, 1592, 1949.
18. Luttinger, J.M., The effect of a magnetic field on electrons in a periodic potential, Phys. Rev., 84,
814, 1951.
19. Luttinger, J.M. and Kohn, W., Motion of electrons and holes in perturbed periodic fields, Phys.
Rev., 97, 869, 1955.
20. Jin, S., Fischetti, M.V., and Tang, T.-W., Modeling of surface-roughness scattering in ultrathin-
body SOI MOSFETs, IEEE Trans. Electron Dev., 54, 2191, 2007.
21. Mahan, G.D., Energy gap in Si and Ge: Impurity dependence, J. Appl. Phys., 51, 2634, 1980.
22. Ridley, B.K., Reconciliation of the Conwell-Weisskopf and Brooks-Herring formulae for charged-
impurity scattering in semiconductors: Third-body interference, J. Phys. C: Solid State Phys.,
10, 1589, 1977.
23. Fetter, A.L. andWalecka, J.D., QuantumTheory of Many-Particle Systems, McGraw-Hill, NewYork, 1971.
24. Fried, B.D. and Conte, S.D., The Plasma Dispersion Function, Academic, New York, 1961.
25. Meyer, J.R. and Bartoli, F.J., Phase-shift calculation of ionized impurity scattering in semi-
conductors, Phys. Rev. B, 23, 5413, 1981.
26. Friedel, J., Electronic structure of primary solid solutions in metals, Adv. Phys., 3, 446, 1954.
27. Fischetti, M.V., Laux, S.E., and Crabbé, E., Understanding hot-electron transport in silicon
devices: Is there a short-cut? J. Appl. Phys., 78, 1058, 1995.
28. Gilat, G. and Raubenheimer, L.J., Accurate numerical method for calculating frequency-
distribution functions in solids, Phys. Rev., 144, 390, 1966.
29. Kaiblinger-Grujin, G., Kosina, H., and Selberherr, S., Influence of the doping element on the
electron mobility in n-silicon, J. Appl. Phys., 83, 3096, 1998.
30. Fischetti, M.V., Comment on ‘‘Influence of the doping element on the electron mobility in
n-silicon [ J. Appl. Phys. 83, 3096, (1998)]’’ J. Appl. Phys., 85, 7984, 1999.
31. Nara, H. and Morita, A., Shallow donor potential in silicon, J. Phys. Soc. Jpn., 21, 1852, 1966.
32. Nara, H., Screened impurity potential in silicon, J. Phys. Soc. Jpn., 20, 778, 1965.
33. Resta, R., Dielectric behavior of a doped semiconductor, Phys. Rev. B, 19, 3022, 1979.
34. Resta, R., Thomas-Fermi dielectric screening in semiconductors, Phys. Rev. B, 16, 2717, 1977.
35. Levine, Z.H. and Louie, S.G., New model dielectric function and exchange-correlation potential
for semiconductors and insulators, Phys. Rev. B, 25, 6310, 1982.
36. Bennett, H.S. and Lowney, J.R., Calculated majority- and minority-carrier mobilities in heavily
doped silicon and comparison with experiment, J. Appl. Phys., 71, 2285, 1992.
37. Dahl, D.A. and Sham, L., Electrodynamics of quasi-two-dimensional electrons, Phys. Rev. B, 16,
651, 1977.
38. Fischetti, M.V. and Laux, S.E., Monte Carlo study of electron transport in silicon inversion layers,
Phys. Rev. B, 48, 2244, 1993.
39. Gámiz, F. et al., A comprehensive model for Coulomb scattering in inversion layers, J. Appl. Phys.,
75, 924, 1994.
ß 2008 by Taylor & Francis Group, LLC.
40. Stern, F. and Howard, W.E., Properties of semiconductor surface inversion layers in the electric
quantum limit, Phys. Rev., 163, 816, 1967.
41. Yokoyama, K. and Hess, K., Monte Carlo study of electronic transport in Al
1Àx
Ga
x
As=GaAs
single-well heterostructures, Phys. Rev. B, 33, 5595, 1986.
42. Fischetti, M.V. et al., Theoretical study of some physical aspects of electronic transport in
nMOSFETs at the 10 nm gate length, IEEE Trans. Electron Dev., 54, 2116, 2007.
43. Lundstrom, M. and Ren, Z., Essential physics of carrier transport in nanoscale MOSFETs, IEEE
Trans. Electron Dev., 49, 133, 2002.
44. Krishnan, M. et al., Remote charge scattering in MOSFETs with ultra-thin gate dielectrics, IEDM
Tech. Dig., 571, 1998.
45. Gámiz, F. and Fischetti, M.V., Coulomb scattering in metal-oxide-semiconductor field-effect
transistors: Screening by electrons in the gate, Appl. Phys. Lett., 83, 4848, 2003.
46. Ishihara, T. et al., Model of screening effect on remote Coulomb scattering due to gate impurities
by nonuniform free carrier in poly-Si gate, J. Appl. Phys., 102, 073702, 2007.
47. Shichijo, H. and Hess, K., Band-structure-dependent transport and impact-ionization in GaAs,
Phys. Rev. B, 23, 4197, 1981.
48. Solomon, P.M. and Laux, S.E., The ballistic FET: Design, capacitance and speed limit, IEDM Tech.
Dig., 95, 2001.
49. Fischetti, M.V. and Laux, S.E., Monte Carlo simulation of transport in technologically significant
semiconductors of the diamond and zinc-blende structures. II. Submicron MOSFETs, IEEE Trans.
Electron Dev., 38, 650, 1991.
50. Fischetti, M.V. et al., Simulation of electron transport in high-mobility-MOSFETs: Density of
states bottleneck and source starvation, IEDM Tech. Dig., 109, 2007.
51. Lochtefeld, A. and Antoniadis, D.A., On experimental determination of carrier velocity in deeply
scaled nMOS: how close to the thermal limit? IEEE Electron Dev. Lett., 22, 95, 2001.
52. Evans, M.H. et al., First-principles mobility calculation and atomic-scale interface roughness in
nanoscale structures, Phys. Rev. Lett., 95, 106802, 2005.
53. Fischetti, M.V., Gámiz, F., and Hänsch, W., On the enhanced electron mobility in strained-silicon
inversion layers, J. Appl. Phys., 92, 7320, 2002.
54. Kleefstra, M. and Herman, G.C., Influence of the image force on the band gap in semiconductors
and insulators, J. Appl. Phys., 51, 4923, 1980.
55. Hohenberg, P. and Kohn, W., Inhomogeneous electron gas, Phys. Rev., 136, B864, 1964.
56. Kohn, W. and Sham, L.J., Self-consistent equations including exchange and correlation effects,
Phys. Rev., 140, A1133, 1965.
57. Hedin, L. and Lundqvist, B.I., Explicit local exchange-correlation potentials, J. Phys. C Solid State
Phys., 4, 2064, 1971.
58. Stern, F. and Das Sarma, S., Electron energy levels in GaAs-Ga
1Àx
Al
x
As heterojunctions, Phys.
Rev. B, 30, 840, 1984.
59. Vasileska, D. et al., Calculation of the average interface field in inversion layers using zero-
temperature Green’s function formalism, J. Vac. Sci. Technol. B, 13, 1841, 1995.
60. Gámiz, F. et al., Image and exchange-correlation effects in double gate silicon-on-insulator
transistors, Microelectron. Eng., 72, 374, 2004.
61. Ando, T., Density-functional calculation of sub-band structure in accumulation and inversion
layers, Phys. Rev. B, 13, 3468, 1976.
62. Goodnick, S.M. et al., Surface roughness at the Si(100)–SiO
2
interface, Phys. Rev. B, 32, 8171, 1985.
63. Ishihara, T. et al., Unified roughness scattering model incorporating scattering component
induced by thickness fluctuations in silicon-on-insulator metal-oxide-semiconductor field-effect
transistors, Jpn. J. Appl. Phys., 45, 3125, 2006.
64. Mou, C.-Y. and Hong, T.-M., Transport in quantum wells in the presence of interface roughness,
Phys. Rev. B, 61, 12612, 2000.
65. Penner, U., Rucker, H., and Yassievich, I.N., Theory of interface roughness scattering in quantum
wells, Semicond. Sci. Technol., 13, 709, 1998.
66. Esseni, D., On the modeling of surface roughness limited mobility in SOI MOSFETs and its
correlation to the transistor effective field, IEEE Trans. Electron Dev., 51, 394, 2004.
ß 2008 by Taylor & Francis Group, LLC.
67. Esseni, D. et al., Physically based modeling of low field electron mobility in ultrathin single-and
double-gate SOI n-MOSFETs, IEEE Trans. Electron Dev., 50, 2445, 2003.
68. Uchida, K. and Takagi, S., Carrier scattering induced by thickness fluctuation of silicon-
on-insulator film in ultrathin-body metal-oxide-semiconductor field-effect transistors, Appl.
Phys. Lett., 82, 2916, 2003.
69. Gold, A., Electronic transport properties of a two-dimensional electron gas in silicon quantum-
well structures at low temperature, Phys. Rev. B, 35, 723, 1987.
70. Jackson, J.D., Classical Electrodynamics, Wiley, New York, 1999.
71. Maldague, P.F., Many-body corrections to the polarizability of the two-dimensional electron gas,
Surf. Sci., 73, 296, 1978.
72. Stern, F., Polarizability of a two-dimensional electron gas, Phys. Rev. Lett., 18, 546, 1967.
73. Gold, A. and Dolgopolov, V.T., Temperature dependence of the conductivity for the two-
dimensional electron gas: Analytical results for low temperatures, Phys. Rev. B, 33, 1076, 1986.
74. Prunnila, M. et al., Gate bias symmetry dependence of electron mobility and prospect of
velocity modulation in double-gate silicon-on-insulator transistors, Appl. Phys. Lett., 85,
5442, 2004.
75. Takagi, S. et al., On the universality of inversion layer mobility in Si MOSFETs: Part I-effects of
substrate impurity concentration, IEEE Trans. Electron Dev., 41, 2357, 1994.
ß 2008 by Taylor & Francis Group, LLC.
5
Electrical Characterization of Defects
in Gate Dielectrics
Dieter K. Schroder
CONTENTS
5.1 Introduction........................................................................................................................ 120
5.2 Oxide Defects ..................................................................................................................... 120
5.2.1 Fixed Oxide Charge................................................................................................ 120
5.2.2 Mobile Oxide Charge ............................................................................................. 121
5.2.3 Oxide-Trapped Charge .......................................................................................... 121
5.2.4 E
0
Center ................................................................................................................... 122
5.2.5 Neutral Electron Traps........................................................................................... 122
5.2.6 Interface-Trapped Charge...................................................................................... 124
5.2.7 Border Traps ............................................................................................................ 125
5.2.8 Interface between Two Different Insulators ....................................................... 127
5.3 Measurements .................................................................................................................... 128
5.3.1 Capacitance–Voltage .............................................................................................. 128
5.3.1.1 Theory ......................................................................................................... 128
5.3.1.2 Interface Traps ........................................................................................... 132
5.3.1.3 Border Traps............................................................................................... 134
5.3.2 Conductance ............................................................................................................ 135
5.3.2.1 Interface Traps ........................................................................................... 135
5.3.2.2 Border Traps............................................................................................... 136
5.3.3 Bias-Temperature Stress......................................................................................... 137
5.3.4 Triangular Voltage Sweep..................................................................................... 137
5.3.5 Deep-Level Transient Spectroscopy..................................................................... 138
5.3.5.1 Bulk Traps .................................................................................................. 138
5.3.5.2 Interface Traps ........................................................................................... 139
5.3.5.3 Border Traps............................................................................................... 140
5.3.6 Charge Pumping..................................................................................................... 141
5.3.6.1 Interface Traps ........................................................................................... 141
5.3.6.2 Border Traps............................................................................................... 144
5.3.7 MOSFET Subthreshold Slope................................................................................ 146
5.3.7.1 Interface Traps ........................................................................................... 146
5.3.7.2 Oxide Traps................................................................................................ 147
5.3.7.3 Border Traps............................................................................................... 148
ß 2008 by Taylor & Francis Group, LLC.
5.3.8 DC-IV...................................................................................................................... 148
5.3.9 Stress-Induced Leakage Current......................................................................... 149
5.3.10 Substrate Hot Electron Injection......................................................................... 151
5.3.11 Constant Gate Voltage Oxide Stress .................................................................. 152
5.3.12 1=f Noise................................................................................................................. 152
5.3.13 Electron Spin Resonance...................................................................................... 154
5.4 Summary............................................................................................................................. 157
References.................................................................................................................................... 158
5.1 Introduction
Defects in gate oxides and insulators have been characterized by many techniques, e.g.,
electrical, electron beam, ion beam, x-rays, neutron activation analysis, electron spin
resonance, and others. These methods played a crucial role during the early metal-oxide
semiconductor (MOS) development to determine the origin of oxide and interface charges
that led to unstable MOS devices. Thermally grown oxides on silicon are now quite well
understood and very well controlled during manufacturing. High-K dielectrics, however,
have proven to be less well controlled and more difficult to characterize. This chapter gives
a brief description of the common defects (oxide, border, and interface traps) and then
describes many of the more common electrical characterization techniques used to estimate
their densities and=or energy levels.
5.2 Oxide Defects
This section describes the main defects commonly observed in insulators, some in ther-
mally grown SiO
2
, others in non-SiO
2
insulators, and some generated during irradiation of
MOS devices. Some aspects of oxide breakdown are also discussed as they pertain to oxide
defects.
5.2.1 Fixed Oxide Charge
Fixed oxide charge was identified early during MOS development and was attributed to
excess silicon near the SiO
2
=Si interface in thermally grown oxides [1]. The model was that,
as an oxide grows on a Si wafer, oxygen diffuses through the growing oxide to react at
the interface forming SiO
2
. This leaves some excess Si in the oxide near the interface. As the
growing oxide front moves into the wafer, the excess Si moves with it and defects
associated with this excess Si become positively charged during negative bias stress. The
fixed oxide charge is not in electrical communication with the underlying silicon. These
positively charged defects were given the symbol Q
ss
and later Q
f
. They were originally
referred to as slow states and later Deal called them fixed surface states [2].
Q
f
depends on the final oxidation temperature. The higher the oxidation temperature,
the lower is Q
f
. However, if it is not permissible to oxidize at high temperatures, it is
possible to lower Q
f
by annealing the oxidized wafer in a nitrogen or argon ambient after
oxidation. This has resulted in the well-known Deal triangle, which shows the reversible
relationship between Q
f
and oxidation and annealing [1]. An oxidized sample may be
prepared at any temperature and then subjected to dry oxygen at any other temperature,
with the resulting value of Q
f
being associated with the final temperature.
ß 2008 by Taylor & Francis Group, LLC.
The nature of Q
f
is still not entirelyclear. Inone model, some of the trivalent Si interface traps
are ‘‘tied up’’ in some form of Si–Si bonds that do not interact with the adjacent oxide net-
workbonding [3]. During hydrogen annealing at moderate temperatures (4008C–5008C), Si–Si
bonds break, one terminated with hydrogen and the other free. The free bond interacts
with a neighboring oxygen to form an overcoordinated oxygen with a fixed positive charge.
When the free bond of the Si atom bonds to a neighboring oxygen atom, the electron can
tunnel to the adjacent Si substrate leaving behind a positively charged O

center.
5.2.2 Mobile Oxide Charge
Mobile charge in SiO
2
is due primarily to the ionic impurities Na
þ
, Li
þ
, K
þ
, and
perhaps H
þ
. Sodium is the dominant contaminant. Lithium has been traced to oil in
vacuum pumps and potassium can be introduced during chemical–mechanical polishing.
The practical application of metal-oxide semiconductor field-effect transistors (MOSFETs)
was delayed due to mobile oxide charges in the early 1960s. MOSFETs were very unstable
for positive gate bias but relatively stable for negative gate voltages. Sodium was the first
impurity to be related to this gate bias instability [4]. By intentionally contaminating MOS
capacitors (MOS-Cs) and measuring the gate voltage shift after bias-temperature stress, it
was shown that alkali cations could easily drift through thermal SiO
2
films. Chemical
analysis of etched-back oxides by neutron activation analysis and flame photometry was
used to determine the Na profile [5]. The drift has been measured with the isothermal
transient ionic current, the thermally stimulated ionic current, and the triangular voltage
sweep methods [6].
The mobility of oxide contaminants is given by the expression [7]:
m ¼ m
0
exp(ÀE
A
=kT) (5:1)
where
for Na: m
0
¼3.5 Â10
À4
cm
2
=Vs (within a factor of 10) and E
A
¼0.44 Æ0.09 eV
for Li: m
0
¼4.5 Â10
À4
cm
2
=Vs (within a factor of 10) and E
A
¼0.47 Æ0.08 eV
for K: m
0
¼2.5 Â10
À3
cm
2
=Vs (within a factor of 8) and E
A
¼1.04 Æ0.1 eV
for Cu: m
0
¼4.8 Â10
À7
cm
2
=Vs and E
A
¼0.93 Æ0.2 eV [8]
If the oxide electric field is taken as V
G
=t
ox
, neglecting the small voltage drops across
the semiconductor and gate, the drift velocity of mobile ions through the oxide is
n
d
¼DV
G
=t
ox
and the transit time t
t
is
t
t
¼
t
ox
n
d
¼
t
2
ox
mV
G
¼
t
2
ox
m
0
V
G
exp(E
A
=kT) (5:2)
Equation 5.2 is plotted in Figure 5.1 for the three alkali ions and for Cu. For this plot the
oxide electric field is 10
6
V=cm, a common oxide electric field for such measurements, and
the oxide thickness is 100 nm. For thinner or thicker oxides, the transit times change
according to Equation 5.2. Typical measurement temperatures lie in the 2008C–3008C
range, and only a few milliseconds suffice for Na and Li to transit the oxide. Mobile charge
densities in the 5 Â10
9
–10
10
=cm
2
range are generally acceptable in integrated circuits (ICs).
5.2.3 Oxide-Trapped Charge
Oxide-trapped charge is due to charge trapped in the oxide. This charge is most commonly
electrons and=or holes injected during device operation or during radiation experiments
(see Chapters 6 and 7).
ß 2008 by Taylor & Francis Group, LLC.
5.2.4 E
0
Center
The E
0
center, usually observed in irradiated MOS devices, consists of two Si atoms joined
by a weak, strained Si–Si bond with a missing oxygen atom, sometimes referred to as an
oxide vacancy, shown in Figure 5.2. It is one of the most dominant radiation-induced
defects (Chapter 6). E
0
centers also preexist in oxide films due to the amorphous nature of
SiO
2
and thermodynamic considerations. Each Si atom is back bonded to three oxygen
atoms. It is believed that when a positive charge is captured, the Si–Si bond breaks. Feigl
et al. argued that the lattice relaxation is asymmetrical with the positively charged Si
relaxing into a planar configuration, away from the vacancy and the neutral Si relaxing
toward the vacancy [9]. The annealing characteristics of E
0
centers have been correlated
with positive oxide charge [10].
5.2.5 Neutral Electron Traps
Several models have been proposed to explain oxide breakdown (Chapters 15 through 17).
One of these is the electron trap generation model, based on the principles of perco-
lation theory [11]. This model, originally suggested by Massoud and Deaton [12] and
later verified by other groups [13–15], assumes that neutral electron traps are randomly
generated in the oxide during oxide stressing. It is assumed that traps are continuously
generated during oxide stress until there are sufficient numbers of traps somewhere in
the device that a continuous, conducting path is formed across the oxide and breakdown
occurs. The percolation model can explain the reduced trap density required for breakdown
and the reduced Weibull slope as the oxide becomes thinner. The latter has an important
FIGURE 5.1
Transit times for Na, Li, K, and Cu for an oxide
electric field of 10
6
V=cm. (Reprinted from
Schroder, D.K., Semiconductor Material and
Device Characterization, 3rd edn., Wiley-Inter-
science, New York, 2006. With permission.)
10
Ϫ4
10
Ϫ2
10
0
10
2
10
4
1 2 3 4
T
r
a
n
s
i
t

t
i
m
e

(
s
)
1000/T (K
−1
)
K
Li
Na
t
ox
= 100 nm
V
G
= 10 V
500 300
100
0
Temperature (°C)
Cu
FIGURE 5.2
Model for hole trapping and E
0
center formation in SiO
2
.
(After Caplan, P.J. et al., J. Appl. Phys., 50, 5847, 1979.
With permission.)
h
+
+
Si
Ox
ß 2008 by Taylor & Francis Group, LLC.
influence on the area dependence of breakdown. If the neutral electron traps capture
electrons, they lead to flatband and threshold voltage shifts. That is, in fact, how they are
detected, by filling them with electrons.
Oxide breakdown exhibits a surge in current or a sudden drop voltage during stress
measurements. It has a ‘‘weakest link’’ character or extreme value statistics [16]. The
statistical description is based on the Weibull distribution model in which the cumulative
failure probability F is given by [17]:
F(t
BD
) ¼ 1 Àexp À
t
BD
a

b

(5:3)
where
t
BD
is the time-to-breakdown
a is the t
BD
at the 63rd percentile
b is the Weibull shape factor or the Weibull slope
Equation 5.3 is usually written in the form:
ln(Àln(1 ÀF(t
BD
))) ¼ bln
t
BD
a

(5:4)
with a plot of ln(–ln(1 – F)) versus ln(t
BD
) yielding a straight line. Equations 5.3 and 5.4 also
apply when the time-to-breakdown is replaced by the charge-to-breakdown Q
BD
. The area
dependence is
t
BD
(A
1
) ¼ t
BD
(A
2
)(A
2
=A
1
)
1=b
(5:5)
where A
1
and A
2
correspond to two different areas. Equation 5.5 shows that the area
dependence is not simply linear because it depends on the shape factor b, which in
turn depends on the oxide thickness, i.e., as b decreases, the area dependence becomes
stronger. This makes it very important to specify the area during breakdown measure-
ments. The reduction of b with decreased oxide thickness is attributed to a reduced
number of defects required to trigger a breakdown.
Two techniques have been used to measure the neutral electron trap density. An indirect
measure is stress-induced leakage current (SILC) and a more direct measure is substrate
hot electron injection (SHE) followed by a measurement of the threshold voltage or
flatband voltage shift. Both are described in Section 5.3.
The nature of neutral electron traps is still under debate. A possible defect structure is the
following. It is well established that the E
0
center is formed by breaking the Si–Si bond in an
oxygen vacancy defect, illustrated in Figure 5.3a. The bond breaking is facilitated by
capture of a hole (Figure 5.3b), leaving a positively charged trap and one Si atom with a
dangling orbital containing one unpaired electron. The resonant flipping of the spin of this
unpaired electron gives rise to the E
0
signal in electron spin resonance. Upon electron
capture, the center can return to the E
0
center or the electron from one of the Si atoms
decays to a ground state by joining the unpaired electron of the other Si atom forming a
neutral amphoteric trap (Figure 5.3c) [18]. Capturing a second electron leaves it negatively
charged (Figure 5.3d). It is this electron trapping event that gives rise to the threshold
voltage shifts associated with filled neutral electron traps following electron injection
measurements [19]. Attempts to anneal neutral traps have been only partially successful.
The usual 4008C–4508C=30 min forming gas anneal only anneals a portion of the traps.
ß 2008 by Taylor & Francis Group, LLC.
High-pressure forming gas is more successful in annealing most of the traps. The hydro-
gen-anneal model is illustrated in Figure 5.3e. Once such a trap is annealed by hydrogen
capture, the Si–H bonds may break leading to trap creation, which may be a precursor to
oxide breakdown.
5.2.6 Interface-Trapped Charge
Interface-trapped charges, also known as interface states, interface traps, and fast surface
states (Chapter 7), exist at the SiO
2
=Si interface. They are the result of a structural imper-
fection. Silicon is tetrahedrally bonded with each Si atom bonded to four Si atoms in the
wafer bulk. When the Si is oxidized, the bonding configuration at the surface is as shown in
Figure 5.4a and b with most Si atoms bonded to oxygen at the surface. Some Si atoms bond
to hydrogen, but some remain unbonded. An interface trap is an interface trivalent Si atom
with an unsaturated (unpaired) valence electron usually denoted by Si
3
Si ., where the
‘‘’’ represents three complete bonds to other Si atoms (the Si
3
) and the ‘‘.’’ represents the
fourth, unpaired electron in a dangling orbital (dangling bond). Interface traps, also known
as P
b
centers [20], are designated as D
it
(1=cm
2
eV), Q
it
(C=cm
2
), and N
it
(1=cm
2
). The P
b
ESR spectrum was first observed by Nishi [21] and later identified by Poindexter et al. as a
paramagnetic dangling bond [22,23].
On (111)-oriented wafers, the P
b
center is situated at the Si=SiO
2
interface with its
unbonded central atom orbital perpendicular to the interface and aimed into a vacancy
in the oxide immediately above it, as shown in Figure 5.4a. On (100)-oriented Si, the four
tetrahedral Si–Si directions intersect the interface plane at the same angle. Two defects,
named P
b1
and P
b0
and shown in Figure 5.4b, have been detected by electron spin
resonance. A recent calculation suggests the P
b1
center to be an asymmetrically oxidized
dimer, with no first neighbor oxygen atoms [24]. By 1999, it was unambiguously estab-
lished that both P
b0
and P
b1
are chemically identical to the P
b
center [25]. However, there is
(a)
h
+
e
Ϫ
e
Ϫ
H
H
Si
Ox
− − −
+ +
(b) (c) (d) (e)
FIGURE 5.3
(a) Neutral hole trap (E
0
center, weak Si–Si bond, oxygen vacancy), (b) positive charge (E
g
0
center), (c) neutral
electron–hole trap, (d) negatively charged trap, and (e) hydrogen-annealed trap.
FIGURE 5.4
Structural model of the (a) (111) Si sur-
face and (b) (100) Si surface. (Reprinted
from Schroder, D.K., Semiconductor
Material and Device Characterization,
3rd edn., Wiley-Interscience, New York,
2006. With permission.)
P
b0
Silicon (100)
P
b1
Si O
P
b
Silicon (111)
H
(b) (a)
ß 2008 by Taylor & Francis Group, LLC.
a charge state difference between these two centers indicating P
b0
is electrically active,
while some authors believe the P
b1
to be electrically inactive [26].
Interface traps are electrically active defects with an energy distribution throughout
the Si band gap. They act as generation and recombination centers and can contribute to
leakage current, low-frequency noise, reduced mobility, drain current, and transconduc-
tance. Since electrons or holes occupy interface traps, they become charged and contribute
to threshold voltage shifts. The surface potential dependence of the occupancy of interface
traps is illustrated in Figure 5.5.
Interface traps at the SiO
2
=Si interface are acceptor-like in the upper half and donor-like
in the lower half of the band gap [27]. Hence, as shown in Figure 5.5a, at flatband, with
electrons occupying states below the Fermi energy, the states in the lower half of the
band gap are neutral (occupied donors designated by ‘‘0’’). Those between midgap and
the Fermi energy are negatively charged (occupied acceptors designated by ‘‘À’’), and
those above E
F
are neutral (unoccupied acceptors). For an inverted p-MOSFET, shown in
Figure 5.5b, the fraction of interface traps between midgap and the Fermi level is now
unoccupied donors, leading to positively charged interface traps (designated by ‘‘þ’’).
Hence interface traps in p-channel devices in inversion are positively charged, leading to
negative threshold voltage shifts.
5.2.7 Border Traps
In 1980, a committee headed by Bruce Deal established the nomenclature for charges
associated with the SiO
2
=Si system, i.e., interface trapped, fixed oxide, mobile ionic and
oxide-trapped charge [28]. In 1992, Fleetwood suggested that this list be augmented by
including border traps, which also have been designated as slow states, near-interfacial
oxide traps, E
0
centers, switching oxide traps, and others (Chapter 7) [29,30]. He proposed
border traps to be those near-interfacial oxide traps located within approximately 3 nm
of the oxide=semiconductor interface. There is no distinct depth limit, however, border
traps are considered to be those traps that can communicate with the semiconductor
through capture and emission of electrons and=or holes on the time scale of interest of
measurement.
Oxide, border, and interface traps are schematically illustrated in Figure 5.6a. Defects
at or near the SiO
2
=Si interface are distributed in space and energy and communicate with
the Si over a wide range of time scales. While for interface traps, the communication
of substrate electrons=holes with interface traps is predominantly by capture=emission,
“0”
Acceptors
E
V
E
C
E
i
E
F
D
it
Donors
“0”
“−”
“0”
“+”
“0”
(b) (a)
FIGURE 5.5
Band diagrams of the Si substrate of a p-channel MOS device showing the occupancy of interface traps and the
various charge polarities for a p-substrate with (a) negative interface trap charge at flatband and (b) positive
interface trap charge at inversion. Interface traps are either occupied by electrons (solid circle) or holes (open
circle).
ß 2008 by Taylor & Francis Group, LLC.
for border traps it is mainly by tunneling fromthe semiconductor to the traps andback. Figure
5.6b shows the flatband diagram with interface and border traps occupied by electrons to the
Fermi level E
F
. The border traps are shown over a wide energy range for illustrative purposes
only. The band diagram in Figure 5.6c applies immediately after V
G1
is applied, before
unoccupied border and interface traps have captured electrons. Interface traps now capture
electrons fromthe conduction band, indicated by (ii), and inversion electrons tunnel to border
traps, indicated by (i). Tunnel process (i) is followed by electron capture of lower energy
border traps. In Figure 5.6d, interface and border traps up to E
F
are occupied by electrons
through (ii) electron capture and (iii) tunneling. For ÀV
G2
in Figure 5.6e, electrons tunnel from
border traps to the conduction band (iv), interface traps (v), and the valence band (vi). The
insulator electric field, shown as constant in these figures, will, of course, distort as the border
trap occupancy changes. This change is disregarded here to bring out the main points.
Inversion electron tunneling is a direct tunnel process with time constant [31]
t
t
% t
0
exp(x=l), l ¼
h
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
8m
t
*f
B
(5:6)
where
t
0
is a characteristic time (%10
À10
s)
l is the attenuation length (%10
À8
cm)
m
t
* is the tunneling effective mass
w
B
is the barrier height at the semiconductor=insulator interface
x x x x
Oxide traps
Border traps
Interface
traps
Gate
(a) (b)
Interface
traps
E
C
E
F
E
X
Border
traps
E
V
E
C
(e)
ϪV
G2
(iv)
(v)
E
F
(vi)
(d)
V
G1
(iii)
(i)
(ii)
(c)
V
G1
(i)
(ii)
FIGURE 5.6
(a) Schematic illustration of oxide, border, and interface traps, (b) occupancy at flatband, (c) capture of electrons by
interface traps and tunneling of electrons to border traps from conduction band, (d) border and interface trap
occupied by electrons, and (e) electron tunneling from border traps. The solid circles represent occupied and the
open circles unoccupied traps.
ß 2008 by Taylor & Francis Group, LLC.
The value of t
t
varies from 0.01 to 1 s (100–1 Hz) for x varying from 1.8 to 2.3 nm. Hence,
border traps can be determined to a depth of approximately 2.5 nm from the SiO
2
=Si
interface by measurements for frequencies as low as 1 Hz. Lower frequencies, of course,
allow deeper traps to be characterized, showing that the trap depth that can be character-
ized depends on the measurement frequency. Such measurements include low-frequency
noise, conductance, frequency-dependent charge pumping, and others. The valence band
hole tunnel times are longer than for electrons due to the higher effective mass and barrier
height. Tewksbury and Lee give a more detailed discussion of tunneling [32].
Tunneling from the conduction band into border traps was questioned, as measurements
did not support energy dissipation in the oxide [33]. Tunneling from interface traps is a
two-step process: the electron must be captured from the conduction band before it can
tunnel [34]. The capture time is
t
c
¼
1
s
n
n
th
n
(5:7)
where
s
n
is the capture cross section
n
th
is the thermal velocity
n is the inversion electron density
For strong inversion n¼10
18
–10
19
=cm
3
, and using s
n
¼10
À16
=cm
2
and n
th
¼10
7
cm=s,
t
c
¼10
À9
–10
À10
s. Since the capture and tunnel processes proceed in series, to first order
the time constant is
t
it
¼ t
c
þt
t
(5:8)
and the tunnel time constant dominates for all but the shallowest border traps.
5.2.8 Interface between Two Different Insulators
For a device consisting of two insulators of thicknesses t
1
and t
2
and dielectric constants K
1
and K
2
on a semiconductor, charge accumulates at the interface between the two insulators
as a result of differing conductivities. When a gate voltage is applied to such a two-layer
structure, the two dielectrics will begin to conduct with current densities J
1
and J
2
. Since the
conductivities of the two layers differ from each other, J
1
6¼ J
2
initially, leading to interfacial
charge density Q accumulation at the interface. Eventually, the system reaches steady state
with the same current density flowing through the entire gate stack. This phenomenon is
known as the Maxwell–Wagner instability [35]. With a Maxwell–Wagner instability, it is
not the usual static dielectric constants that determine the electric fields of the two layers in
steady state, but rather the requirement that the same current density flow through each.
Thus, the field in dielectric 1 may differ from VK
2
(t
1
K
2
þt
2
K
1
) by an amount depending on
the magnitude of Q. The interfacial charge density Q, and from it the electric fields, could
be calculated if the current densities J
1
and J
2
were known. Without this knowledge,
VK
2
(t
1
K
2
þt
2
K
1
) is only an approximation. However, when the dielectric constants of
high-K dielectrics are determined experimentally, the interfacial charge is usually ignored,
with the two layers of the gate stack being treated as capacitors in series and the ‘‘effective
dielectric constants’’ include the effects of the interfacial charge.
In this brief defect discussion, radiation-induced defects have been mentioned several
times. Such defects have been fertile ground for studying defects and developing charac-
terization techniques because they can be introduced and annealed at will and their
densities can be very high, facilitating measurements. In contrast, oxide defects in
ß 2008 by Taylor & Francis Group, LLC.
conventional ICs are generally of low density and frequently more difficult to characterize.
Although radiation-induced defects often differ from those induced during normal IC
processing, such processing can introduce its own form of radiation defects. After all,
plasma and reactive ion etching and ion implantation are radiation sources that may
introduce such defects.
5.3 Measurements
5.3.1 Capacitance–Voltage
5.3.1.1 Theory
Capacitance–voltage measurements have played an important role in MOS characteriza-
tion. They can be found in some of the earliest MOS-related papers. Frankl, in 1961,
proposed the use of MOS-C C–V curves to analyze such devices [36]. In 1962, Terman
used C–V measurements to determine surface state densities [37]. One of the first compre-
hensive papers was by the Fairchild group in 1964, the same group that played a large role
in understanding and developing MOS technology [38]. Figure 5.7 shows one of those
early C–V curves. Barrier heights for various gate metals were characterized with C–V
measurements by the same group [39]. A good early overview of the variety of material
and device parameters that can be determined from C–V and C–t measurements is given
by Zaininger and Heiman [40]. And, of course, the entire MOS-C field is very well covered
in the well-known book by Nicollian and Brews [41].
Insulator-related parameters typically determined with C–V measurements are (i) oxide
charge density, r
ox
(x), (ii) interface trap density, D
it
, and (iii) gate–semiconductor work
function difference, w
MS
. The various charges are illustrated in the device cross section in
Figure 5.8a. They are determined from the flatband voltage
V
FB
¼ f
MS
À
1
K
ox
«
0

t
ox
0
xr(x)dx (5:9)
The fixed charge density Q
f
and interface trap charge density Q
it
are assumed to be located
at the SiO
2
=Si interface and the charge density r
ox
(x) in the SiO
2
, leading to the flatband
voltage expression
FIGURE 5.7
Capacitance–voltage characteristics of an
MOS-C: lines, experiment; dots, theory.
N
A
¼1.45 Â10
16
=cm
3
, t
ox
¼200 nm. (After
Grove, A.S. et al., Solid State Electron. 8, 145,
1965. With permission.)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
Ϫ4 2 6 10 14
Low frequency
High frequency
Depletion
V
G
Ϫf
MS
+Q
f
/C
ox
(V)
C
/
C
o
x
ß 2008 by Taylor & Francis Group, LLC.
V
FB
¼ f
MS
À
1
K
ox
«
0

t
ox
0
xQ
f
d(t
ox
)dx À
1
K
ox
«
0

t
ox
0
xr
ox
(x)dx À
1
K
ox
«
0

t
ox
0
xQ
it
(f
s
)d(t
ox
)dx (5:10)
where
t
ox
is the oxide thickness
d is the delta function
Q
it
is a function of surface potential w
s
, as shown in Figure 5.5
V
FB
can be written as
V
FB
¼ f
MS
À
Q
f
K
ox
«
0
t
ox
À
1
K
ox
«
0

t
ox
0
xr
ox
(x)dx À
Q
it
(f
s
)
K
ox
«
0
t
ox
(5:11)
The oxide charge consists of mobile oxide charge (Na, K, etc.), oxide-trapped charge
(electrons and holes), and any other charge that may reside within the oxide.
For uniform oxide charge density, Equation 5.11 becomes
V
FB
¼ f
MS
À
Q
f
þQ
it
(f
s
)
K
ox
«
0
t
ox
À
r
ox
t
2
ox
2K
ox
«
0
)Intercept ¼ f
MS
(5:12)
Equation 5.12 is plotted in Figure 5.9a assuming constant Q
it
. This plot is clearly nonlinear,
making it difficult to extract the various charges. However, w
MS
is given by the V
FB
intercept. Differentiating Equation 5.12 with respect to oxide thickness gives
dV
FB
dt
ox
¼ À
Q
f
þQ
it
K
ox
«
0
À
r
ox
t
ox
K
ox
«
0
)Intercept ¼ À
Q
f
þQ
it
K
ox
«
0
, Slope ¼ À
r
ox
K
ox
«
0
(5:13)
and plotted in Figure 5.9b, yielding Q
f
þQ
it
and r
ox
. The fixed oxide charge and interface
trap densities cannot be determined independently, only their sum. However, Q
it
can be
measured independently by other techniques.
Frequently the oxide charge density in thermally grown SiO
2
in Equation 5.12 is low and
can be neglected. In that case, the V
FB
expression in Equation 5.12 becomes linear with t
ox
.
0 t
ox
x
+
+
+
x
x
x
x
+
+

Q
f
Q
it
(a)
+
+
+
x
x
x
x
+
+

Q
f,HK
Q
it
r
ox r
ox
r
HK
(b)
+
+
+
+
Q
f,ox
t
ox
t
HK
Substrate Substrate

FIGURE 5.8
MOS-C cross section: (a) oxide only and (b) high-K and oxide. The various charges are indicated.
ß 2008 by Taylor & Francis Group, LLC.
An early example is shown in Figure 5.10a, where the thick-oxide V
FB
Àt
ox
plots show the
effect of varying w
MS
and Q
f
þQ
it
[42]. Curve A clearly yields a different intercept than the
other three because it is p-Si, while the other three refer to n-Si. Curves B–D show the effect
of different values of Q
f
. A more recent thin oxide example is shown in Figure 5.10b where
the flatband voltage and gate work functions are extracted from V
FB
–t
ox
plots [43].
For two-layer insulators of Figure 5.8b, consisting of a gate, a high-K insulator, SiO
2
, and
a substrate, the flatband voltage becomes
V
FB
¼ w
MS
À
1
K
ox
«
0

EOT
0
xr(x)dx (5:14)
where
w
MS
is the gate=substrate work function difference
r(x) is the charge density in the two insulators and at their interfaces EOT is the
equivalent oxide thickness
Ϫ1.4
Ϫ1.3
Ϫ1.2
Ϫ1.1
Ϫ1
Ϫ0.9
Ϫ0.8
2ϫ10
Ϫ6
0
(a)
4ϫ10
Ϫ6
6ϫ10
Ϫ6
8ϫ10
Ϫ6
1ϫ10
Ϫ5
F
l
a
t
b
a
n
d

v
o
l
t
a
g
e

(
V
)
Oxide thickness (cm)
f
MS
(b)
2ϫ10
Ϫ6
0 4ϫ10
Ϫ6
6ϫ10
Ϫ6
8ϫ10
Ϫ6
1ϫ10
Ϫ5
Ϫ6ϫ10
4
Ϫ5ϫ10
4
Ϫ4ϫ10
4
Ϫ3ϫ10
4
Ϫ2ϫ10
4
d
V
F
B
/
d
t
o
x

(
V
/
c
m
)
Oxide thickness (cm)
Intercept ~ Q
f
+ Q
it
Slope ~ r
ox

FIGURE 5.9
(a) Flatband voltage and (b) dV
FB
=dt
ox
versus oxide thickness for w
MS
¼À0.85 V, (Q
f
þQ
it
)=q ¼5 Â10
10
=cm
2
and
r
ox
=q ¼10
16
=cm
3
.
Ϫ1.2
Ϫ1
Ϫ0.8
Ϫ0.6
Ϫ0.4
Ϫ0.2
0
0.2
0
(a) (b)
100 200 300
A
B
C
D
F
l
a
t
b
a
n
d

v
o
l
t
a
g
e

(
V
)
Oxide thickness (nm)
Ϫ1
Ϫ0.8
Ϫ0.6
Ϫ0.4
Ϫ0.2
0
0.2
0 5 10 15 20
F
l
a
t
b
a
n
d

v
o
l
t
a
g
e

(
V
)
Oxide thickness (nm)
p
+
Poly-Si
p HfSi
n
+
poly-Si
Undoped HfSi
n HfSi
FIGURE 5.10
Flatband voltage versus oxide thickness for the Al–SiO
2
–Si system. (a) A: p-Si and B, C, D: n-Si. Q
f
þQ
it
vary for
B to D. (From Werner, M.W., Solid State Electron., 17, 769, 1974. With permission.) (b) p-Si with poly-Si and fully
silicided gates. (After Park, C.S. et al., IEEE Electron Dev. Lett., 25, 372, 2004. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
EOT is
EOT ¼ t
ox
þEOT
HK
¼ t
ox
þ
K
ox
K
HK
t
HK
(5:15)
where t
ox
, t
HK
, K
ox
, and K
HK
are the oxide and high-K insulator thicknesses and dielectric
constants. The oxide flatband voltage due to fixed charge density Q
f,ox
at the SiO
2
=Si
interface and charge density r
ox
(x) in the SiO
2
is
V
FB
(SiO
2
) ¼ À
1
K
ox
«
0

EOT
0
xQ
f,ox
d(EOT)dx À
1
K
ox
«
0

EOT
EOT
HK
xr
ox
(x)dx (5:16)
If the bottom insulator is not SiO
2
, then t
ox
is replaced by EOT
ins
¼(K
ox
=K
ins
)t
ins
, where K
ins
and t
ins
are insulator dielectric constant and thickness. Since the charge distribution in
insulators is usually not known, it is frequently assumed to be constant, giving
V
FB
(SiO
2
) ¼ À
Q
f,ox
K
ox
«
0
EOT À
r
ox
2K
ox
«
0
EOT
2
þ
r
ox
2K
ox
«
0
EOT
2
HK
(5:17)
Similarly for the high-K insulator with fixed charge density Q
f,HK
at the HK=SiO
2
interface
and charge density r
HK
(x) in the high-K insulator, V
FB
is
V
FB
(HK) ¼ À
1
K
ox
«
0

EOT
HK
0
xQ
f,HK
d(EOT
HK
)dx À
1
K
ox
«
0

EOT
HK
0
xr
HK
(x)dx (5:18)
giving
V
FB
(HK) ¼ À
Q
f,HK
K
ox
«
0
EOT
HK
À
r
HK
2K
ox
«
0
EOT
2
HK
(5:19)
The total flatband voltage is
V
FB
¼ f
MS
þV
FB
(SiO
2
) þV
FB
(HK) (5:20)
It is obvious from these equations that it is very difficult to determine all four charge
components.
To get an idea of the magnitude of each of the components, let us consider one
particular case: t
ox
¼0.5 nm, t
HK
¼5 nm, K
ox
¼3.9, K
HK
¼25 ) EOT
HK
¼0.78 nm, and
EOT¼1.28 nm. N
f,ox
¼5 Â10
10
=cm
2
and N
f,HK
¼5 Â10
12
=cm
2
give Q
f,ox
¼8 Â10
À9
C=cm
2
,
Q
f,HK
¼8 Â10
À7
C=cm
2
, N
ox
¼10
16
=cm
3
and N
HK
¼5 Â10
19
=cm
3
)r
ox
¼0.0016 C=cm
3
and
r
HK
¼8 C=cm
3
. These charge densities correspond approximately to recent experimental
data [44,45]. Substituting into the flatband equations gives
V
FB
(SiO
2
) ¼À
8Â10
À9
Â1:28Â10
À7
3:45Â10
À13
À
0:0016Â1:64Â10
À14
6:9Â10
À13
þ
0:0016Â6:1Â10
À15
6:9Â10
À13
¼À3:05mV
ß 2008 by Taylor & Francis Group, LLC.
V
FB
(HK) ¼ À
8 Â10
À7
Â0:78 Â10
À7
3:45 Â10
À13
À
8 Â6:1 Â10
À15
6:9 Â10
À13
¼ À251 mV
Clearly the HK charges contribute a much higher flatband voltage shift. Hence, the charges
associated with the thin SiO
2
film are frequently ignored during flatband voltage analyses
of high-K samples.
Equation 5.20 is plotted in Figure 5.11, where in Figure 5.11a t
ox
is held constant and t
HK
is allowed to vary, while in Figure 5.11b t
HK
is constant and t
ox
varies. The points in Figure
5.11b are experimental data with the extracted charges shown in the inset. Note the very
different V
FB
ÀEOT behavior for the two cases. This is mainly due to the different thick-
nesses for these two cases. For EOT¼8 nm, in Figure 5.11a t
ox
¼1 nm and t
HK
¼45 nm,
while in Figure 5.11b t
ox
¼7 nm and t
HK
¼6 nm. The much thicker t
HK
in Figure 5.11a has a
more severe effect on V
FB
than Figure 5.11b.
To measure these devices the oxide is sometimes grown to a certain thickness and
then etched to varying thicknesses across the wafer. In one method, the oxidized wafer is
immersed and slowly withdrawn from a 0.34% HF=H
2
O solution at a constant with-
drawal rate yielding a beveled oxide across the wafer [45]. In another method, the oxide
is step etched [46–48].
5.3.1.2 Interface Traps
MOS capacitance measurements are made at high and low frequencies. Low-frequency
C–V measurements are referred to as quasistatic measurements, first demonstrated in
1968–1970 [49] to measure interface traps. Here, a calculation of the effects of interface
traps on C–V
G
curves is shown to illustrate what one may expect. Figure 5.12a shows
the assumed interface trap density as a function of surface potential for these calculations.
This distribution approximates the interface trap density distribution at the SiO
2
=Si inter-
face with D
it,min
at midgap. Figure 5.12b shows the surface potential versus gate voltage
behavior without and with interface traps. The discontinuity at w
s
% 0.4 V is the result of
the assumption of D
it
being donors in the upper half and acceptors in the lower half of the
band gap. In real devices there is a more gradual transition. The interface trap density
Ϫ2
Ϫ1.5
Ϫ1
Ϫ0.5
0
0.5
0
(a) (b)
2 4 6 8 10
F
l
a
t
b
a
n
d

v
o
l
t
a
g
e

(
V
)
Equivalent oxide thickness (nm)
N
f
= 4ϫ10
11
cm
Ϫ2
N
fHK
= Ϫ8ϫ10
12
cm
Ϫ2
N
ox
=5ϫ10
15
cm
Ϫ3
N
HK
= 4ϫ10
19
cm
Ϫ3
t
ox
=1 nm
Ϫ0.2
Ϫ0.1
0
F
l
a
t
b
a
n
d

v
o
l
t
a
g
e

(
V
)
Equivalent oxide thickness (nm)
N
f
= 4ϫ10
11
cm
Ϫ2
N
fHK
=Ϫ8ϫ10
12
cm
Ϫ2
N
ox
=5ϫ10
15
cm
Ϫ3
N
HK
=4ϫ10
19
cm
Ϫ3
t
HK
=6 nm
0 2 4 6 8 10
FIGURE 5.11
Flatband voltage versus equivalent oxide thickness with (a) t
ox
fixed, vary t
HK
, (b) t
HK
fixed, vary t
ox
for a
TaN=TiN=HfSiO
x
=SiO
2
structure. (Data in the inset after Kaushik, V.S. et al., IEEE Trans. Electron Dev., 53, 2627,
2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
results in a ‘‘stretch-out’’ of the w
s
–V
G
characteristic because charged interface traps lead to
gate voltage shifts.
Figure 5.13 shows the effects of D
it
on (a) low- and (b) high-frequency C–V
G
curves. Here
it is assumed that Q
ox
¼0 in these calculations in order to bring out the effects of interface
traps. Q
ox
would lead to a parallel shift of the curves along the V
G
axis. It is also assumed
0
(a) (b)
5ϫ10
11
1ϫ10
12
1.5ϫ10
12
2ϫ10
12
Ϫ0.2 0 0.2 0.4 0.6 0.8 1
D
i
t

(
1
/
c
m
2

e
V
)
Surface potential (V)
t
ox
= 5 nm
N
A
=10
17
cm
Ϫ3
Donors Acceptors
D
it,min
Ϫ0.3
Ϫ0.1
0.1
0.3
0.5
0.7
0.9
1.1
Ϫ3 Ϫ2 Ϫ1 0 1 2 3
S
u
r
f
a
c
e

p
o
t
e
n
t
i
a
l

(
V
)
Gate voltage (V)
t
ox
=5 nm
N
A
=10
17
cm
Ϫ3
D
it
= 0
D
it
>0
2f
F
f
F
f
s
= 0
FIGURE 5.12
(a) Interface trap distribution and (b) surface potential versus gate voltage. N
A
¼10
17
=cm
3
, t
ox
¼5 nm,
D
it,min
¼2.4 Â10
À11
=cm
2
eV.
0
1ϫ10
Ϫ11
2ϫ10
Ϫ11
3ϫ10
Ϫ11
Ϫ2 Ϫ1 0 1 2 3
C
a
p
a
c
i
t
a
n
c
e

(
F
)
Gate voltage (V)
Before stress
After
stress
A
B
0
Ϫ2 Ϫ1 0 1 2 3
C
a
p
a
c
i
t
a
n
c
e

(
F
/
c
m
2
)
C
a
p
a
c
i
t
a
n
c
e

(
F
/
c
m
2
)
Gate voltage (V) (a) (b)
(c)
t
ox
= 5 nm
N
A
=10
17
cm
Ϫ3
ΔV
f
F
2f
F
D
it
=0
f
s
=0
D
it
>0
0
1ϫ10
Ϫ7
2ϫ10
Ϫ7
3ϫ10
Ϫ7
4ϫ10
Ϫ7
5ϫ10
Ϫ7
6ϫ10
Ϫ7
7ϫ10
Ϫ7
1ϫ10
Ϫ7
2ϫ10
Ϫ7
3ϫ10
Ϫ7
4ϫ10
Ϫ7
5ϫ10
Ϫ7
6ϫ10
Ϫ7
7ϫ10
Ϫ7
Ϫ2 Ϫ1 0 1 2 3
Gate voltage (V)
t
ox
= 5 nm
N
A
=10
17
cm
Ϫ3 f
F
2f
F
D
it
=0
f
s
=0
D
it
>0
FIGURE 5.13
Theoretical (a) low-frequency, (b) high-frequency C–V
G
(D
it,min
¼2.4 Â10
11
=cm
2
eV), and (c) experimental data.
w
s
is the surface potential.
ß 2008 by Taylor & Francis Group, LLC.
that, at high frequencies, the interface traps do not follow the ac gate voltage and do not
contribute a capacitance, so the ‘‘stretch-out’’ is solely due to the effects of D
it
on V
G
. The
acceptor states in the upper half of the band gap lead to a positive gate voltage shift in
inversion, whereas the donor states give the negative V
G
shift in accumulation and
depletion. For the low-frequency curve, on the other hand, there is both ‘‘stretch-out’’
and additional capacitance, as the interface traps are assumed to be able to follow the ac
gate voltage. Figure 5.13c shows experimental data before and after gate oxide stress, with
the stress generating D
it
. Note how the curve broadens as predicted, shifting to the left in
region A and to the right in region B, indicative of amphoteric interface traps. This curve
indicates donor states over slightly more than the lower half of the band gap.
The basic theory, developed by Berglund, compares a low-frequency C–V curve with
one free of interface traps [50]. The latter can be a theoretical curve, but is usually a hf C–V
curve determined at a frequency where interface traps are assumed not to respond. ‘‘Low
frequency’’ means that interface traps and minority carrier inversion charges must be able
to respond to the measurement ac probe frequency. The interface trap density is deter-
mined from the lf capacitance data according to
D
it
¼
1
q
2
C
ox
C
lf
C
ox
ÀC
lf
ÀC
S

(5:21)
where C
it
is related to the interface trap density D
it
by D
it
¼C
it
=q
2
. (We should mention
here that most publications use C
it
¼qD
it
. With D
it
in the usual units of 1=cm
2
eV and
q in C, the units for C
it
in this expression are F=cm
2
C, suggesting that the correct definition
should be C
it
¼q
2
D
it
.) A simplified approach, proposed by Castagné and Vapaille, uses
measured lf and hf C–V curves as [51]:
D
it
¼
C
ox
q
2
C
lf
=C
ox
1 ÀC
lf
=C
ox
À
C
hf
=C
ox
1 ÀC
hf
=C
ox

(5:22)
Equation 5.22 gives D
it
over only a limited range of the band gap, typically from the
onset of inversion, to a surface potential toward the majority carrier band edge where
the ac measurement frequency equals the inverse of the interface trap emission time
constant.
5.3.1.3 Border Traps
C–V
G
curves can also be used to gain information on oxide-trapped charge. For example,
the hump sometimes observed in quasistatic data, illustrated in Figure 5.14, has been
attributed to electron tunneling from the channel into border traps [52]. This device
has an oxide=nitride=oxide gate insulator and the trap density in the nitride or at
the nitride=oxide interface is increased by hot carrier stress. Traps within several kT
about the trap energy participate during the ac measurement, leading to an additional
capacitance. Its magnitude depends on the measurement frequency. Lower frequencies
lead to higher humps as carriers can tunnel deeper into the insulator.
High-frequency C–V
G
measurements also show a capacitance increase for accumu-
lated n-substrate MOS-Cs and inverted n-MOSFETs with Hf-based insulators, attributed
to tunneling from the substrate into border traps [53]. It appears that the charge carriers
tunneling through the thin interfacial oxide must be electrons due to their smaller
effective mass and barrier height compared to holes. The dielectric capacitance
increase at low frequencies, typically 1–100 kHz, has been proposed as a frequency- and
ß 2008 by Taylor & Francis Group, LLC.
voltage-dependent border trap capacitance C
bt
in parallel with the interface trap capaci-
tance C
it
. C
bt
contributes to the measured capacitance only if the border trap char-
ging=discharging can follow the applied ac signal. The frequency=voltage dependence of
C
bt
yields the tunneling distance from the Si substrate and trap energy depth in the HfO
2
.
Related to CÀV
G
measurements are CÀt measurements, commonly used to determine
the generation lifetime [54]. In this technique, the MOS-C is pulsed into deep depletion
and subsequently relaxes to its equilibrium state through electron–hole pair generation.
The generation lifetime so determined is a measure of the substrate purity. This technique
is well understood but irregularities appear when the flatband voltage changes during the
recovery time. This can occur as a result of carrier injection from the semiconductor into
the insulator as, for example, in HfO
2
. An initial capacitance undershoot has been attrib-
uted to electron tunneling into HfO
2
traps and interface trap generation [55].
5.3.2 Conductance
5.3.2.1 Interface Traps
The conductance method proposed by Nicollian and Goetzberger in 1967 yields D
it
in the
depletion and weak inversion portions of the band gap, the capture cross sections for
majority carriers, and information about surface potential fluctuations [56]. The technique
is based on measuring the equivalent parallel conductance of an MOS-C as a function of
bias voltage and frequency. The interface traps are detected through the losses resulting
from changes in their occupancy due to changes in the gate voltage during the ac mea-
surement [57]. A small ac gate voltage leads to an interface trap occupancy change in an
energy interval of a few kT wide centered on the Fermi level by capture or emission
of majority carriers. This capture=emission causes energy loss leading to a conductance
given by
G
P
v
¼
qvt
it
D
it
1 þ(vt
it
)
2
(5:23)
where
v¼2pf
t
it
¼R
it
C
it
is the interface trap time constant ¼[n
th
s
p
N
A
exp(Àqs
s
=kT]
À1
C
it
¼q
2
D
it
1.0
0.8
0.6
0.4
0.2
0
Ϫ4 Ϫ2 0 2 4
Gate voltage (V)
C
/
C
o
x
Stressed
Virgin
Anomalous
“hump”
FIGURE 5.14
MOSFET normalized capacitance before and after
hot carrier injection of 10
16
electrons=cm
2
. (From
Cohen, N.L. et al., IEEE Trans. Electron Dev., 42,
2004, 1995. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
Equation 5.23 is for interface traps with a single energy level in the band gap. Interface
traps at the SiO
2
=Si interface, however, are continuously distributed in energy throughout
the Si band gap leading to a time constant dispersion and giving the normalized conduct-
ance as
G
P
v
¼
qD
it
2vt
it
ln 1 þ(vt
it
)
2

(5:24)
The conductance is measured as a function of frequency and plotted as G
P
=v versus v.
G
P
=v has a maximum at v¼1=t
it
and at that maximum D
it
% 2.9G
P
=qv.
For thin oxides, there may be appreciable oxide leakage current. In addition, the device
has series resistance leading to [58]:
G
P
v
¼
v(G
c
ÀG
t
)C
2
ox
G
2
c
þv
2
(C
ox
ÀC
c
)
2
(5:25)
where
C
c
¼
C
m
(1 Àr
s
G
m
)
2
þ(vr
s
C
m
)
2
, G
c
¼
v
2
r
s
C
m
C
c
ÀG
m
r
s
G
m
À1
(5:26)
where
G
t
is the tunnel conductance
r
s
is the series resistance
C
m
and G
m
are the measured capacitance and conductance
5.3.2.2 Border Traps
Absent tunneling to border traps, the G
p
=v–v peak is symmetrical about the maximum
peak frequency. However, a low-frequency conductance ledge is sometimes observed for
insulators containing oxide traps. This was first reported by Eaton and Sah who observed
a marked asymmetry in their conductance spectrum and attributed this to long time
constants arising from carriers tunneling from interface traps into oxide traps [59].
The conductance spectrum in Figure 5.15 shows such a low-frequency ledge [60]. Such
FIGURE 5.15
G
p
=v versus f for a MOS-C exhibiting a D
it
peak and N
ox
ledge. The lines indicate the
contributions from D
it
¼1.46 Â10
11
=cm
2
eV
and N
ox
¼1.3 Â10
17
=cm
3
eV. (Reused from
Uren, M.J., Collins, S., and Kirton, M.J., Appl.
Phys. Lett., 54, 1448, 1989. With permission.)
10
Ϫ13
10
Ϫ12
10
Ϫ11
10
Ϫ10
10
Ϫ1
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
G
p
/
w

(
S
/
H
z
)
Frequency (Hz)
Due to D
it
Due to N
ot
ß 2008 by Taylor & Francis Group, LLC.
ledges are usually not observed when measurements are confined to frequencies higher
than 0.1–1 kHz.
5.3.3 Bias-Temperature Stress
In the bias-temperature stress (BTS) method the MOS device is heated to 1508C–2508C, and
a gate bias to produce an oxide electric field of around 10
6
V=cm is applied for 5–10 min for
any mobile oxide charge to drift to one oxide interface. The device is then cooled to room
temperature under bias and a C–V
G
curve is measured. The procedure is then repeated
with the opposite bias polarity. The mobile charge is determined from the flatband voltage
shift, according to the equation
Q
m
¼ ÀC
ox
DV
FB
(5:27)
To distinguish between oxide-trapped charge and mobile charge, a BTS test is done with
positive gate voltage. For oxide electric fields around 1 MV=cmmobile charge drifts, but the
electric field is insufficient for appreciate charge injection. If the C–V
G
curve shifts after BTS,
it is due to positive mobile charge. For higher gate voltages, there is a good chance that
electrons and=or holes can be injected into the oxide and mobile charge may also drift,
making that measurement less definitive. For positive gate voltage, mobile charge drift leads
to negative flatband voltage shifts while charge injection leads to positive shifts.
5.3.4 Triangular Voltage Sweep
In the triangular voltage sweep (TVS) method the MOS device is held at an elevated,
constant temperature of 2008C–3008C and the low-frequency C–V
G
curve is obtained by
measuring the current in response to a slowly varying ramp gate voltage [61]. TVS is based
on measuring the charge flow through the oxide at an elevated temperature in response to
an applied time-varying voltage. If the ramp rate is sufficiently low, the measured current
is the sum of the displacement and conduction current due to the mobile charge. If, at
ÀV
G1
, all mobile charges are located at the gate–oxide interface and at þV
G2
all mobile
charges are located at the semiconductor–oxide interface, the mobile charge is determined
by the area under the lf curve according to

V
G2
ÀV
G1
(I=C
lf
Àa)C
ox
dV
G
¼ aQ
m
(5:28)
The hf and lf C–V curves coincide at high temperatures except for the lf hump, due to
mobile charge drifting though the oxide and illustrated in Figure 5.16 [62]. Figure 5.16a
illustrates the effect of different mobile charge densities, and Figure 5.16b the effect of
temperature. Clearly the temperature for this sample must be at least 1208C for all of the
mobile charge to move.
Sometimes two peaks are observed in I–V
G
curves at different gate voltages. These have
been attributed to mobile ions with different mobilities. For an appropriate temperature
and sweep rate, high-mobility ions (e.g., Na
þ
) drift at lower oxide electric fields than
low-mobility ions (e.g., K
þ
) [63]. Hence, the Na peak occurs at lower gate voltages than
the K peak, illustrated in Figure 5.17. Such discrimination between different types of
mobile impurities is not possible with the bias-temperature method. This also explains
why sometimes the total number of impurities determined by the BTS and the TVS
ß 2008 by Taylor & Francis Group, LLC.
methods differ. In the BTS method one usually waits long enough for all the mobile charge
to drift through the oxide. If in the TVS method the temperature is too low or the gate ramp
rate is too high, it is possible that only one type of charge is detected. For example, it is
conceivable that high-mobility Na
þ
drifts but low-mobility K
þ
does not. The TVS method
also lends itself to mobile charge determination in interlevel dielectrics, since a current or
charge is measured instead of a capacitance.
5.3.5 Deep-Level Transient Spectroscopy
5.3.5.1 Bulk Traps
Deep-level transient spectroscopy (DLTS), introduced by Lang in 1972 [64], is commonly
used to determine densities, energies, and capture cross sections of bulk and interface traps
in semiconductors and sometimes for border traps. The device capacitance, current, or
charge is measured as a function of time after pulsing the device between zero=forward
and reverse bias. The traps capture electrons or holes, which they subsequently emit,
leading to the time-dependent capacitance
300
(a) (b)
400
500
600
700
800
900
1000
Ϫ3 Ϫ2 Ϫ1 0 1 2 3
C
a
p
a
c
i
t
a
n
c
e

(
p
F
)
Gate voltage (V)
N
m
=1.3ϫ10
10
cm
Ϫ2
t
ox
=100 nm
4.1ϫ10
9
cm
Ϫ2
hf
lf
0
200
400
600
800
1000
1200
1400
Ϫ4 Ϫ3 Ϫ2 Ϫ1 0 1 2 3 4
C
a
p
a
c
i
t
a
n
c
e

(
p
F
)
Gate voltage (V)
T=120ЊC
100ЊC
80ЊC
60ЊC
10ЊC, 40ЊC
FIGURE 5.16
C
lf
and C
hf
measured at (a) T¼2508C and (b) various temperatures. The mobile charge density is determined
from the area between the two curves. (Reprinted from Schroder, D.K., Semiconductor Material and Device
Characterization, 3rd edn., Wiley-Interscience, New York, 2006. With permission.)
FIGURE 5.17
Ion current normalized by oxide cur-
rent versus gate voltage. The ion
moves to the gate. T¼4238C, a¼0.513
V=s. (Reused from Hillen, M.W.,
Greeuw, G., and Verweij, J.F., J. Appl.
Phys., 50, 4834, 1979. With permission.)
0.8 0.4 0
Na
+
400
300
200
100
0
K
+
Ϫ0.4 Ϫ0.8
Gate voltage (V)
I
o
n

c
u
r
r
e
n
t

(
I



I
o
x
)

(
p
A
)
Ϫ1.2 Ϫ1.6
ß 2008 by Taylor & Francis Group, LLC.
C(t) ¼ C
0
1 À
N
T
2N
D
exp À
t
t
e
¸
(5:29)
with the electron emission time constant t
e
depending on temperature as
t
e
¼
exp((E
c
ÀE
T
)=kT)
g
n
s
n
T
2
(5:30)
where g
n
is a constant for a given semiconductor. Determining the time constant at various
temperatures allows the energy level E
T
, density N
T
, and capture cross section s
n
to be
determined [65].
Example DLTS spectra of iron-contaminated Si are shown in Figure 5.18 [66]. Iron forms
Fe–B pairs in boron-doped p-type Si with a DLTS peak at T % 50 K. When the sample is
heated at 1808C–2008C for a few minutes, the Fe–B pairs dissociate into interstitial iron Fe
i
and substitutional boron, and the DLTS peak for Fe
i
occurs around T % 250 K. After a few
days the interstitial iron again forms Fe–B pairs and the ‘‘Fe–B’’ peak returns while the
‘‘Fe
i
’’ peak shrinks as shown in Figure 5.18.
5.3.5.2 Interface Traps
The instrumentation for interface-trapped charge DLTS is identical to that for bulk deep-
level DLTS, but the data interpretation differs because interface traps are distributed in
energy through the band gap. We illustrate the interface-trapped charge majority carrier
DLTS concept for the MOS-C in Figure 5.19a. For a positive gate voltage most interface
traps are occupied by majority electrons for n-substrates (Figure 5.19b). A negative gate
voltage drives the device into deep depletion, causing electrons to be emitted from
interface traps (Figure 5.19c). Although electrons are emitted over a broad energy spec-
trum, emission from interface traps in the upper half of the band gap dominates. DLTS is
very sensitive, allowing interface trap density determination in the mid 10
9
=cm
2
eV range.
Interface trap characterization by DLTS was first implemented with MOSFETs [67].
Being three-terminal devices, they have an advantage over MOS-Cs. By reverse biasing
the source=drain (S=D) and pulsing the gate, majority electrons are captured and emitted
without interference from minority holes that are collected by the SD. This allows
interface trap majority carrier characterization in the upper half of the band gap.
50 100 150 200 250 300
0
20
40
60
Ϫ
d
C

(
f
F
)
Temperature (K)
As-is
180ЊC/30Њs
As-is; RT/5d
RT/5d
180ЊC/30Њs
Fe-B
Fe
i
FIGURE 5.18
DLTS spectra for iron-contaminated
Si wafer; ‘‘As-is,’’ after 1808C=30 s dissoci-
ation anneal, and room temperature stor-
age for 5 days. (Reprinted after Choi, B.D.,
et al., Jap. J. Appl. Phys., 40, L915, 2001. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
With the SD forward biased, an inversion layer forms, allowing interface traps to be filled
with minority holes. Minority carrier characterization is then possible and the lower half of
the band gap can be explored. This is not possible with MOS-Cs because there is no
minority carrier source other than thermal generation.
MOS-Cs are, nevertheless, used for interface trap characterization [68]. Unlike the
conductance technique, DLTS measurements are independent of surface potential fluctu-
ations. The derivation of the capacitance expression is more complex for MOS-Cs than it is
for diodes. We quote the main results whose derivations can be found in Johnson [69] and
Yamasaki et al. [70] For q
2
D
it
¼C
it
( C
ox
and dC¼C
hf
(t
1
) ÀC
hf
(t
2
) ( C
hf
dC ¼
C
3
hf
K
s
«
0
N
D
C
ox

1
À1
D
it
e
Àt
2
=t
e
Àe
Àt
1
=t
e

dE
it
(5:31)
where
t
e
is the electron emission time
t
1
and t
2
are the sampling times
E
it
is the interface trap energy
If D
it
varies slowly in the energy range of several kT around E
it,max
, it can be considered
reasonably constant and can be taken outside the integral of Equation 5.31 leading to
D
it
¼ À
K
s
«
0
N
D
C
ox
kTC
3
hf
ln(t
2
=t
1
)
dC (5:32)
If the sample contains bulk as well as interface traps, it is possible to differentiate bulk traps
from interface traps by the shape and the peak temperature of the DLTS plot. Interface trap
densities determined by DLTS and quasistatic C–V are shown in Figure 5.20.
5.3.5.3 Border Traps
DLTS has also been used to characterize border traps by using large amplitude filling
pulses, allowing electrons not only to be captured by D
it
but also to tunnel to border traps
[71]. This effect was observed for electron tunneling, but not for holes attributed to the
higher barrier for holes at the SiO
2
=Si interface. The spatial and energetic border trap
distribution is obtained by varying the pulse width and amplitude. In this tunnel DLTS it is
possible to distinguish between interface and border traps.
E
c
E
F
E
V
(a) (b) (c)
ϪV
G2
V
G1
FIGURE 5.19
(a) Flatband, (b) majority carrier capture, and (c) majority carrier emission from interface traps.
ß 2008 by Taylor & Francis Group, LLC.
5.3.6 Charge Pumping
5.3.6.1 Interface Traps
In the charge pumping (CP) method, originally proposed in 1969 [72], the MOSFET source
and drain are tied together and either grounded or slightly reverse biased. The time-
varying gate voltage (square, triangular, trapezoidal, sinusoidal, or trilevel) is of sufficient
amplitude for the surface under the gate to be driven between inversion and accumulation.
The CP current is measured at the substrate, at the S=D tied together, or at the source and
drain separately. For the oxide–semiconductor band diagram in Figure 5.21, the interface
traps are shown by the filled circles representing electron-occupied interface traps and the
open circles representing unoccupied traps. In Figure 5.21a, with the device in inversion,
most of the interface traps are occupied by electrons. The Fermi level is shown as a single
0
1ϫ10
10
2ϫ10
10
3ϫ10
10
4ϫ10
10
5ϫ10
10
6ϫ10
10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
D
i
t

(
1
/
c
m
2


e
V
)
DLTS
E
c
– E (eV)
Quasistatic C–V
FIGURE 5.20
Interface trap density for n-Si measured by the DLTS and quasistatic methods. (Reprinted from Johnson, N.M.,
J. Vac. Sci. Technol., 21, 303, 1982. With permission.)
E
C
(a) (b) (c)
E
F
E
v
To S/D
Channel
Trapped
electrons
Hole
barrier
FIGURE 5.21
Band diagrams for (a) inversion, (b) flatband, and (c) accumulation. The transitions are discussed in the text.
ß 2008 by Taylor & Francis Group, LLC.
level E
F
for simplicity. Of course, the device being in nonequilibrium during the CP
measurement should be represented by two quasi-Fermi levels.
When the gate pulse falls from its high to its low value during its finite transition time,
most channel electrons drift to source and drain and those electrons on interface traps
near the conduction band are thermally emitted into the conduction band (Figure 5.21b)
and also drift to source and drain. Those electrons on interface traps deeper within
the band gap do not have sufficient time to be emitted and remain trapped. Once the
hole barrier is reduced (Figure 5.21c), holes flow to the surface where some are captured
by those interface traps still occupied by electrons and all interface traps are occupied by
holes. As the device reverts back to inversion, some of the holes near the valence band are
emitted to drift to the substrate (not shown), while those remaining on interface traps
are annihilated by electron capture (Figure 5.21a). Hence, only a fraction of interface traps,
those toward the center of the band gap, are not emitted and participate in capture or
recombination processes and lead to CP current. This energy interval depends on the rise
and fall times of the CP waveform. The waveforms can be constant base voltage in
accumulation and pulsing with varying voltage amplitude DV into inversion or varying
the base voltage from inversion to accumulation keeping DV constant. The current satur-
ates for the former, while for the latter it reaches a maximum and then decreases.
From Shockley–Read–Hall recombination statistics the occupancy of interface traps is
determined by carrier capture and emission. Considering capture processes, the variation
of the occupancy during one charge pumping cycle is [73,74]:
DF ¼
[1 Àexp(Àc
n
=2f )][1 Àexp(Àc
p
=2f )]
1 Àexp[À(c
n
=2f ) À(c
p
=2f )]
(5:33)
where
c
n
and c
p
are the capture coefficients for electrons and holes (c
n,p
¼s
n,p
n
th
), s
n,p
are the
capture cross sections and n
th
is the thermal velocity
f is the CP frequency
The charge pumping current then becomes
I
cp
¼ qAf

E
high
E
low
DFD
it
(E)dE (5:34)
where E
high
and E
low
are the Fermi energies for high and low gate bias.
The basic charge pumping technique gives an average value of D
it
over the energy
interval DE. Various refinements have been proposed to obtain energy-dependent interface
trap distributions. For a sawtooth waveform, the recombined charge per cycle, Q
cp
¼I
cp
=f,
is given by [75]:
Q
cp
¼ 2qkTD
it
fA
G
ln n
th
n
i
ffiffiffiffiffiffiffiffiffiffiffi
s
n
s
p
p
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z(1 Àz)

jV
FB
ÀV
T
j
jDV
GS
j f

(5:35)
where
D
it
is the average interface trap density
DV
GS
is the gate pulse peak–peak amplitude
z is the gate pulse duty cycle
ß 2008 by Taylor & Francis Group, LLC.
The slope of a Q
cp
versus log( f ) plot yields D
it
, and the intercept on the log( f ) axis yields
(s
n
s
p
)
1=2
. By varying the gate waveform rise and fall times, one obtains the interface trap
energy distribution. Han et al., using this technique for SiON=HfO
2
insulators, found
higher D
it
in the upper half than in the lower half of the band gap [76]. This also agreed
with V
T
, mobility, and subthreshold slope data, e.g., the V
T
shift of n-MOSFETs was higher
than the flatband voltage shift for these HfO
2
samples.
The interface trap distribution through the band gap and capture cross sections can be
determined with the trilevel waveform with an intermediate voltage level V
step
[77],
illustrated in Figure 5.22. At point (a), the device is in strong inversion with interface
traps filled with electrons. As the waveform changes to (b), electrons are emitted from
interface traps. The gate voltage remains constant to point (c). For t
step
)t
e
, where t
e
is the
emission time constant of interface traps being probed, all traps above E
T
have emitted
their electrons and only those below E
T
are available for recombination when holes come in
to recombine with the electrons at point (d) on the waveform. This gives a charge pumping
current that saturates as t
step
increases. For t
step
<t
e
, fewer electrons have time to be
emitted and more are available for hole recombination giving a correspondingly higher
charge pumping current.
A typical I
cp
versus t
step
plot in Figure 5.23 shows the I
cp
saturation and the t
step
¼t
e
break point. By varying V
step
one can probe interface traps through the band gap. D
it
is
determined from the slope of the I
cp
versus t
step
curve according to the expression [78]:
D
it
¼ À
1
qkTA
G
f
dI
cp
dlnt
step
(5:36)
The charge pumping current is assumed to be due to electron–hole pair recombination at
interface traps. For thin oxides, gate current adds to the charge pumping current. The gate
oxide leakage current can exceed I
cp
. Figure 5.24 shows the effects of gate oxide leakage
current on I
cp
[79]. At sufficiently low frequencies, the gate leakage current dominates and
can be subtracted from the total current.
a
b c
d
V
H
V
T
V
L
t
step
V
step
(a) (b) (c) (d)
E
T
E
c
E
v
FIGURE 5.22
Trilevel charge pumping waveform and corresponding band diagrams. (Reprinted from Schroder, D.K., Semi-
conductor Material and Device Characterization, 3rd edn., Wiley-Interscience, New York, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
5.3.6.2 Border Traps
In the presence of interface and border traps, carriers can tunnel into border traps as the
charge pumping waveform frequency is reduced and Q
cp
¼I
cp
=f, which is constant during
conventional interface trap CP, increases. Paulsen et al. characterized such traps by charge
pumping experiments in which the recombined charge increases at low frequencies [80].
This is attributed to charging and discharging of oxide traps within a tunneling distance of
the SiO
2
=Si interface. In their experiments they determined the trap distances to be 1.2–1.5
nm, based on tunneling time constants of 10
À5
to 10
À3
s.
This effect can be modeled by letting the oxide trap capture cross section be depth
dependent [81]:
s
n,p
(x) ¼ s
n,p
(0) exp(Àx=l) (5:37)
where
s
n,p
(0) is the interface electron=hole capture cross section
l is the attenuation length in Equation 5.6
FIGURE 5.23
I
cp
as a function of t
step
showingt
e
at the point
where I
cp
begins to saturate. (FromSaks, N.S.
and Ancona, M.G., IEEE Electron Dev. Lett.,
11, 339, 1990. With permission.)
0.3
0.4
0.5
0.6
10
Ϫ7
10
Ϫ6
10
Ϫ5
10
Ϫ4
10
Ϫ3
10
Ϫ2
I
c
p

(
p
A
)
t
step
(s)
t
e
FIGURE 5.24
Charge pumping current versus base volt-
age for two voltage pulse heights before
and after gate leakage current correction.
t
ox
¼1.8 nm, f ¼1 kHz. (From Bauza, D.,
IEEE Electron Dev. Lett., 23, 658, 2002. With
permission.)
0
1
2
3
Ϫ1 Ϫ0.8 Ϫ0.6 Ϫ0.4 Ϫ0.2 0
Before
Before
After
After
I
c
p

(
p
A
)
V
base
(V)
ΔV=1 V
0.3 V
ß 2008 by Taylor & Francis Group, LLC.
The variation of the border trap occupancy during one charge pumping cycle, DF
ox
, is that
in Equation 5.33 with a spatially varying s
n,p
. The total charge pumping current then
becomes
I
cp
¼ qAf

E
high
E
low
DFD
it
(E) dE þ

E
high
E
low

t
ox
0
DF
ox
D
bt
(E)dx dE

¸
¸
¸

(5:38)
where D
bt
is the border trap density (1=cm
3
eV). D
bt
is determined from a Q
cp
versus log f
plot according to
D
bt
¼ À
1
q
2
AlDf
s
dQ
cp
dlog f
(5:39)
where Df
s
is the surface potential change during one period and Q
cp
the charge pumped
per cycle. The distance over which oxide traps can be probed for CP where only the
inversion part of the pulse leads to tunneling into traps is [82]:
x
m
% Àlln
s
n
v
th
2f

% 2:2 nm (5:40)
for s
n
¼10
À16
cm
2
and f ¼1 kHz. This shows that traps to distances around 1–2 nm from
the insulator=semiconductor interface can be probed. Example trap distributions for SiO
2
and SiO
2
=Al
2
O
3
insulators are shown in Figure 5.25, illustrating the higher trap density in
Al
2
O
3
compared to SiO
2
. A complication in these measurements is the possible gate
insulator leakage current at low CP frequencies.
In the amplitude sweep charge pumping (ACP) technique, the base voltage is held
constant and the pulse amplitude is gradually increased [83]. The charge-per-cycle
increases with gate voltage amplitude if there is insulator trapping. By using charging
and discharging pulses it is possible to obtain spectroscopic trap information. The trap
energy levels in the insulator are determined by the SiO
2
=Si interface acting as a retarding
potential analyzer and sweeping the discharge voltage to more negative voltages. This
method showed the effective defect energy levels to be close to the Si conduction band.
There is sometimes a geometric component during CP measurements, due to those
channel carriers unable to drift to S=D during the accumulation portion of the pulse.
10
12
10
14
10
16
10
18
10
20
10
22
Ϫ0.6 Ϫ0.5 Ϫ0.4 Ϫ0.3 Ϫ0.2 Ϫ0.1 0
T
r
a
p

d
e
n
s
i
t
y

(
1
/
c
m
3
)
Depth (nm)
SiO
2
Al
2
O
3
Insulator/Si
interface
FIGURE 5.25
Insulator trap density versus insulator
depth from the insulator=Si interface for
Al
2
O
3
and SiO
2
. (From Jakschik, S. et al.,
IEEE Trans. Electron Dev., 51, 2252, 2004.
With permission.)
ß 2008 by Taylor & Francis Group, LLC.
Being unable to return to S=D from where they originated, they are injected into the
substrate to recombine. This component is maximized by using long-channel MOSFETs
and fast rise=fall times [84]. Under these conditions the long channel is rapidly pinched-off,
forcing a large fraction of the inversion charge to recombine in the Si where it is measured
as a substrate current. By grounding both source and drain, the inversion charge density is
N
inv,S=D
; by grounding the source and floating the drain it is N
inv,S
, given by
N
inv,S=D
¼ N
inv
À2N
S=D
, N
inv,S
¼ N
inv
ÀN
S=D
(5:41)
where N
inv
is the true inversion charge density and N
S=D
is the geometric component. From
Equation 5.41
N
inv
¼ 2N
inv,S
ÀN
inv,S=D
(5:42)
yielding the true N
inv
. This technique was used to determine the mobility for SiO
2
=HfO
2
gate insulators, eliminating trapping effects [84].
5.3.7 MOSFET Subthreshold Slope
5.3.7.1 Interface Traps
The effects of D
it
on the MOSFET subthreshold I
D
–V
G
characteristics are illustrated in
Figure 5.26. Similar to the broadening of the C–V
G
curves in Figure 5.13, the subthreshold
characteristic also broadens and only the surface potential region between about midgap
(w
s
¼w
F
) and inversion can be probed. The drain current of a MOSFET in the subthreshold
regime for drain voltages higher than about 4kT=q is [85]:
I
D
¼
Wm
eff
L
kT
q

2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qK
s
«
0
N
A
4f
F

exp
q(V
GS
ÀV
T
)
nkT

(5:43)
where
n¼1 þ(C
b
þC
it
)=C
ox
C
b
is the bulk (substrate) capacitance
The usual subthreshold plot is log(I
D
) versus V
G
with subthreshold swing S being that gate
voltage necessary to change the drain current by one decade, given by
FIGURE 5.26
Theoretical I
D
–V
G
curves for D
it
¼0 and
D
it,min
¼2.7 Â10
10
=cm
2
eV.
10
Ϫ14
10
Ϫ12
10
Ϫ10
10
Ϫ8
10
Ϫ6
10
Ϫ4
0 0.2
f
s
=f
F
0.4 0.6 0.8 1
D
r
a
i
n

c
u
r
r
e
n
t

(
A
)
Gate voltage (V)
f
s
=2f
F
t
ox
= 5 nm
N
A
=10
17
cm
Ϫ3
D
it
=0
D
it
> 0
ß 2008 by Taylor & Francis Group, LLC.
S ¼
ln(10)nkT
q
%
60nT
300
mV=decade (5:44)
with T in Kelvin. The interface trap density is
D
it
¼
C
ox
q
2
qS
ln(10)kT
À 1

À
C
b
q
2
(5:45)
requiring an accurate knowledge of C
ox
and C
b
. The slope also depends on surface
potential fluctuations. This is the reason that this method is usually used as a comparative
technique in which the subthreshold swing is measured, then the device is degraded and
remeasured. The D
it
change is given by
DD
it
¼
C
ox
ln(10)q
2
kT
S
after
ÀS
before
ð Þ (5:46)
The subthreshold MOSFET curves are shown in Figure 5.27 before and after stress, causing
a threshold voltage shift and a slope change.
5.3.7.2 Oxide Traps
Subthreshold measurements are also made to determine oxide charge densities. When the
surface potential coincides with the Fermi level, as shown in Figure 5.28a by w
s
¼w
F
,
interface traps in the upper and lower half in the band gap are neutral, and neither
contributes to a gate voltage shift. The corresponding gate voltage is V
mg
, which is
typically the gate voltage at I
D
% 0.1–1 pA. Increasing the gate voltage from V
mg
to V
T
fills interface traps in the upper half of the band gap with electrons (Figure 5.28b). DE
usually covers the range from midgap to strong inversion. Since at midgap the interface
traps do not contribute any voltage shift, a shift of V
mg
must be due to oxide-trapped
charge, where the threshold voltage due to oxide-trapped charge and the change in oxide
trap charge density as projected to the Si=SiO
2
interface may be estimated according to [86]:
DV
ot
¼ V
mg2
ÀV
mg1
and DN
ot
¼
DV
ot
C
ox
q
(5:47)
10
Ϫ13
10
Ϫ12
10
Ϫ11
10
Ϫ10
10
Ϫ9
10
Ϫ8
10
Ϫ7
10
Ϫ6
0 0.2 0.4 0.6 0.8 1
ID (before)
ID (after)
Shifted
D
r
a
i
n

c
u
r
r
e
n
t

(
A
)
Gate voltage (V)
ΔD
it
=5ϫ10
11
cm
Ϫ2
eV
Ϫ1
Before
stress
ΔV
T
After
stress
FIGURE 5.27
MOSFET subthreshold characteristics
before andafter MOSFETstress. The change
in slope results in a stress-generated
DD
it
¼5 Â10
11
=cm
2
eV. The dashed curve
is the after stress curve shifted to the left to
coincide with the before stress curve at
I
D
¼10
À13
A to bring out the slope change.
(Reprinted from Schroder, D.K., Semicon-
ductor Material and Device Characterization,
3rd edn., Wiley-Interscience, New York,
2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
5.3.7.3 Border Traps
The threshold voltage of high-K dielectric MOSFETs frequently exhibits an instability that
has been explained by charging preexisting defects in the HfO
2
in HfO
2
=SiO
2
dual dielec-
trics with a strong dependence on gate bias and charging (discharging) time [87]. The
measured drive current decay in n-MOSFETs is caused by the continuous increase in V
T
due to a buildup of negative charge, caused by trapping in the HfO
2
. Comparing the
pulsed technique with the dc measurement methods shows the conventional dc techniques
to underestimate the charging effects in SiO
2
=HfO
2
dual layer gate dielectrics. An example
of such dielectric charging is shown in Figure 5.29, clearly illustrating the hysteresis for the
transient measurements due to charge trapping=detrapping. Information about border
traps can be gained from such measurements. The pulsed techniques was recently used
to estimate the capture cross sections in HfO
2
as 10
À16
–10
À14
cm
2
[88].
5.3.8 DC-IV
The DC-IV method is a dc technique to determine D
it
illustrated in Figure 5.30a [89]. With
the source S forward biased, electrons injected into the p-well diffuse to the drain to be
V
mg
V
T
E
c
E
v
E
i
E
F
(a)
f
s
= f
F
f
s
= 2f
F
f
F
ΔE
(b)
FIGURE 5.28
Band diagrams for midgap and threshold voltages. (Reprinted from Schroder, D.K., Semiconductor Material and
Device Characterization, 3rd edn., Wiley-Interscience, New York, 2006. With permission.)
FIGURE 5.29
I
D
–V
G
dc (open circles) and transient (100 ms pulse
width, rise and fall times) characteristics showing
higher transient current and hysteresis due to carrier
trapping=detrapping. (From Kerber, A. et al., IEEE
Intl. Rel. Phys. Symp., 41, 41, 2003. With permission.)
200
150
100
50
0
0 0.5 1.0 1.5
Gate voltage (V)
D
r
a
i
n

c
u
r
r
e
n
t

(
μ
A
)
V
D
=1 V
2.0 2.5
ß 2008 by Taylor & Francis Group, LLC.
collected and measured as drain current I
D
. Some electrons recombine with holes in the
p-well bulk (not shown) and some recombine with holes at the surface below the gate with
only the surface-recombining electrons controlled by the gate voltage. The recombined
holes are replaced by holes from the body contact leading to body current I
B
.
The electron–hole pair surface recombination rate depends on the surface condition.
With the surface in strong inversion or accumulation, the recombination rate is low. The
rate is highest with the surface in depletion [90]. The body current is given by
DI
B
¼ qA
G
n
i
s
r
exp(qV
BS
=2kT) (5:48)
where s
r
is the surface recombination velocity given by
s
r
¼ (p=2)s
0
v
th
DN
it
(5:49)
with s
0
the capture cross section (assuming s
n
% s
p
% s
0
).
When the gate voltage exceeds the flatband voltage, a channel forms between S and D
and the drain current will increase significantly. For V
GB
¼V
T
, the I
D
ÀV
GB
curve saturates.
If charge is injected into the oxide, leading to a V
T
shift, the drain current will also shift. It is
this shift that can be used to determine oxide charge. The interface trap density determined
with the subthreshold slope method samples the band gap between midgap and strong
inversion while the DC-IV body current samples the band gap between subthreshold and
weak accumulation, i.e., surface depletion. By varying the gate voltage, different regions of
the device are depleted (Figure 5.30b) and those regions can be characterized, allowing
spatial D
it
profiling. Experimental DC-IV data are shown in Figure 5.31 for a MOSFET
before and after gate current stress [91]. A clear peak is observed at maximum surface
recombination around V
GB
¼0.
5.3.9 Stress-Induced Leakage Current
An effect frequently observed in thin electric field-stressed oxides is an enhanced gate
oxide current—SILC, defined as the increase of oxide leakage current after high-field stress
(%10–12 MV=cm) compared to before stress and first reported in 1982 [92]. It is typically
B
I
B
p
+
I
D
I
C
x x x
n
+ n
+
n
+
n
+
n
+
I
B
p-well
p-well
p-well
p-well
n-substrate
Sub
(a) (b)
V
GS
=0
V
GS
>0
V
GS
<0
S G D
FIGURE 5.30
(a) MOSFET configuration for DC-IV measurements and (b) cross sections showing the space-charge regions and
the encircled surface generation regions. (Reprinted from Schroder, D.K., Semiconductor Material and Device
Characterization, 3rd edn., Wiley-Interscience, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
observed at low to moderate oxide electric fields (%4–8 MV=cm) and increases markedly
for thinner oxides. However, the SILC decrease for oxides thinner than about 5 nm is
believed to be due to reduced trap generation rates in thin oxides. Different models have
been proposed to explain SILC: interface trap generation, bulk-oxide electron trap gener-
ation, nonuniformities or weak spot formation in the oxide films, trapped holes injected
from the anode. SILC can be best explained by the generation of neutral electron traps in
the oxide, allowing more current to flow through the oxide layer by these traps acting as
‘‘stepping stones’’ by trap-assisted tunneling [93]. The charge-to-breakdown is inversely
related to the defect creation initial rate, which can be measured at low injected charge
where the defect density increases linearly with the injected charge [94]. Hence, SILC can
be used to obtain the oxide defect generation rate. Example SILC curves are shown in
Figure 5.32 [95]. SILC is usually not detected by time-zero or time-dependent oxide
breakdown measurements. It has been proposed as a reliability monitor [96].
FIGURE 5.31
DC-IV measured body currents. Control wafer and stressed
with À12 mA=cm
2
gate current density. V
BS
¼0.3 V,
W=L¼20=0.4 mm, t
ox
¼5 nm. (From Guan, H. et al., IEEE
Electron Dev. Lett., 20, 238, 1999. With permission.)
30
20
10
0
0 1
Gate voltage (V)
B
o
d
y

c
u
r
r
e
n
t

(
p
A
)
No
stress
820 s
stress
Ϫ1
FIGURE 5.32
n-MOSFET normalized SILC variation as a function
of the gate voltage after positive oxide stress.
V
Gstress
, 3.5 V; Q
inj
, 1–1000 C=cm
2
. (After Petit, C.
et al., Microelectron. Reliab., 45, 479, 2005. With per-
mission.)
18
16
14
12
10
8
6
4
2
0
Ϫ2 Ϫ1.5 Ϫ1 Ϫ0.5 0
Gate voltage (V)
Δ
I
G
/
I
G
0
ß 2008 by Taylor & Francis Group, LLC.
SILC has been used to determine the nitride trap density in Si=SiO
2
=Si
3
N
4
=SiO
2
=Si
(SONOS) devices, where the oxide adjacent to the substrate is typically thicker
($3–5 nm) than the oxide ($1 nm) in SiO
2
=high-K dual insulators to improve data
retention in nonvolatile memories. Unlike a MOS-C, the SILC in a SONOS exhibits a
two-stage time dependence, with the first stage exhibiting dc-like SILC characteristic and
the second stage following a 1=t time dependence [97]. The first stage leakage current
is limited by stress-created oxide traps and the second stage by Frenkel–Poole emission
of nitride-trapped electrons. Through appropriate data analysis, the nitride trap density of
7 Â10
12
=cm
2
eV was found to be nearly constant with energy.
5.3.10 Substrate Hot Electron Injection
SHE is the injection of electrons from Si into an oxide at low electric fields. One application
of SHE is to make neutral electron traps visible, by charging them with trapped
electrons. It has been shown that stressing gate oxide at moderate to high oxide electric
fields (5–12 MV=cm) produces traps in the oxide. Since many of these traps are not detected
by conventional measurements, e.g., threshold voltage shifts, it is believed that they are
electron generated, but neutral when unoccupied by electrons. Once occupied by electrons,
they lead to threshold and flatband voltage shifts and can be easily characterized. Electron
injection into SiO
2
was studied initially by Ning because such injected electrons lead to
device degradation and instability, and is a powerful tool for studying electron and hole
traps in thermal SiO
2
films [98].
The concept of the measurement is illustrated in Figure 5.33. With source and drain
grounded, a positive gate voltage normally leads to electron channel formation for
V
G
> 0
V
B
< 0
(a) (b)
V
B
<0
V
n
< 0
V
G
> 0
(c)
V
G
V
B
n n n n r
p p
FIGURE 5.33
(a) Hot electron injection from leakage current, (b) forward-biased diode, and (c) band diagram showing hot
electron injection.
ß 2008 by Taylor & Francis Group, LLC.
grounded substrate. When the substrate is reverse biased with respect to S=D, the channel
disappears. It is possible, however, for thermally generated electrons accelerated toward
the Si=SiO
2
interface, to have sufficient energy to surmount the barrier w
B
at the interface,
as shown in Figure 5.33a and c. These electrons have moderate energy and there is a good
chance that they become trapped in the oxide. A problem with the arrangement in Figure
5.33a is the low density of electrons for injection. The electrons are thermally generated, i.e.,
the leakage current of the gate-induced space-charge region=substrate junction. Ning
proposed to use a MOSFET with the p-substrate replaced by a p-epitaxial layer grown
on an n-substrate. The resulting np junction can then be forward biased to inject electrons
into the p-layer. Such a structure is a natural portion of MOSFETs fabricated in a p-well on
an n-substrate [99]. It is also possible to form an np junction beside the source or drain and
forward bias it, shown in Figure 5.33b [100], or surrounding the MOSFET with an n-type
implant. Either configuration allows independent control of the oxide electric field and the
electron injection current.
5.3.11 Constant Gate Voltage Oxide Stress
Gate oxide integrity (GOI) characterization determines the quality of the gate oxide and
does provide trap densities. It is mentioned briefly here, because it is important to determine
the gate oxide quality. GOI is measured by one of several methods. In the time-zero method,
the gate voltage is swept and the gate current is measured until the oxide breaks down.
In the time-dependent methods, the gate voltage is held constant (constant voltage source,
CVS) and the gate current is measured, or the gate current is held constant (constant current
source, CCS) and the gate voltage is measured, or the gate current is stepped and the gate
voltage is measured. For thin oxides it has been shown that the time-to-breakdown, t
BD
,
increases with decreasing oxide thickness for constant current stress while it decreases for
constant voltage stress [101]. For CCS, the oxide voltage changes during the measurement
and the defect generation rate and critical defect density required for breakdown change.
This makes the CVS technique the more appropriate method for GOI determination.
5.3.12 1=f Noise
Noise measurements can be used to characterize semiconductors. The recent review papers
by Wong [102] and Claeys et al. [103] give a good overview of the present state of noise
theory and measurements. At high frequencies, thermal noise and shot noise dominate.
Both of these noises are fundamental in nature, forming an intrinsic lower noise limit. At
low frequencies, flicker or 1=f noise dominates. Generation–recombination (G–R) noise can
also occur in this frequency range. It is characterized by a Lorentzian spectrum with a
constant plateau at f <f
c
and a 1=f
2
roll-off beyond the characteristic frequency f
c
. In
contrast to the fundamental thermal and shot noise, 1=f and G–R noise depend on material
and semiconductor processing. Only 1=f noise is discussed here since it is sensitive to
interface and border traps.
Low-frequency or flicker noise, first observed in vacuum tubes over 80 years ago [104],
dominates the noise spectrum at low frequencies. It gets its name from the anomalous plate
current ‘‘flicker.’’ Flicker noise is also called 1=f noise, because the noise spectrum varies as
1=f
n
, where the exponent n is very close to unity. Fluctuations with a 1=f power law have
been observed in many electronic materials and devices, including homogenous semicon-
ductors, junction devices, metal films, liquid metals, electrolytic solutions, Josephson
junctions, and even in mechanical, biological, geological, and even musical systems. Two
competing models have been proposed to explain flicker noise: the McWhorter number
fluctuation theory [105] and the Hooge mobility fluctuation theory [106] with experimental
ß 2008 by Taylor & Francis Group, LLC.
evidence to support both models. Christensson et al. were the first to apply the McWhorter
theory to MOSFETs, using the assumption that the necessary time constants are caused by
the tunneling of carriers from the channel into traps located within the oxide (Chapter 7)
[107]. Popcorn noise, sometimes called burst noise or random-telegraph-signal (RTS) noise,
is a discrete modulation of the channel current caused by the capture and emission of as
little as a single channel carrier [108].
RTS presents a serious problem to analog and digital devices as devices are scaled down.
For example, in floating gate nonvolatile memories the capture=emission of even a single
electron affects the threshold voltage and drain current. RTS-induced drain current pro-
duces significant variation in read current [109]. The threshold voltage change
DV
T
¼Àq=AC
ox
increases as the area, A, decreases. The trap properties leading to RTS
can be studied by statistical analysis. One analysis procedure is based on the difference in
the statistical properties of discrete Marcovian telegraph fluctuations and Gaussian back-
ground noise. The average statistical lifetimes and amplitudes of the telegraph signal are
then determined in an iterative way, allowing for analyzing noisy random telegraph
signals with low ratio between the signal amplitude and the intensity of the background
noise that cannot be analyzed by the classical approach. Separation of the time record
enables an in-depth analysis of the spectral properties of the background noise observed
together with the telegraph fluctuations [110].
The MOSFET current is proportional to the product of mobility m
eff
times the charge
carrier density or number N
s
. Low-frequency fluctuations in charge transport are caused
by stochastic changes in either of these parameters, which can be independent (uncorrel-
ated) or dependent (correlated). In most cases, fluctuations in the current, or more specif-
ically in the m
eff
ÂN
s
product are monitored, which does not allow the separation of
mobility from number effects and therefore obscures the identification of the dominant
1=f noise source.
The voltage noise spectrum density is [111]:
S
V
( f ) ¼
q
2
kTl
aWLC
2
ox
f
(1 þsm
eff
N
s
)
2
N
bt
(5:50)
where
l is the tunneling parameter
m
eff
is the effective carrier mobility
a is the Coulombic scattering parameter
N
s
is the density of channel carriers
C
ox
is the gate oxide capacitance=unit area
WL is the gate area
N
bt
is the effective border trap density (1=cm
3
eV) near the interface
In weak inversion, the channel carrier density N
s
is very low (10
7
–10
11
=cm
2
), so the
mobility fluctuation contribution becomes negligible and the second term in the bracket
can be neglected.
It is assumed that the free carriers tunnel to traps in the oxide with a tunneling time
constant, which varies with distance x from the interface. The tunneling parameter and
time constant are given in Equation 5.6. For a trap at 1 nm from the semiconductor
interface, t
T
% 0.5 s or f ¼1=2pt
T
% 0.3 Hz. Hence, a distribution of traps in the oxide
gives rise to a wide range of frequencies and can explain the 1=f dependence. 1=f noise
shows sensitivity to the wafer orientation, which correlates with interface trap density.
Figure 5.34 shows an example of low-frequency noise before and after annealing with the
ß 2008 by Taylor & Francis Group, LLC.
anneal reducing the interface trap density and the l=f noise [112]. The main advantage
of using a noise-based technique is that it can be applied even to very small area devices,
which is not possible with capacitance-based DLTS, and it is well suited for characteri-
zation of electrically active defects in devices with no substrate contact, such as Si-wire
transistors, carbon nanotubes, and others. Low-frequency noise has become a failure
analysis characterization technique, but the noise measurement equipment is quite
complex [113]. A recent low-frequency noise study of the role of fluorine in HfSiON
gate insulators showed that the noise for both n- and p-MOSFETs was reduced by fluorine
due to passivation of traps close to valence and conduction bands [114]. By combining low-
frequency noise with gate leakage currents measurements, information on the trap location
within the insulator was gained.
5.3.13 Electron Spin Resonance
Electron spin resonance (ESR), also known as electron paramagnetic resonance (EPR),
detects species with unpaired electron spins. However, it is almost always possible to
render a defect ESR active by charge injection. Thus, almost all atoms or molecules capable
of charge capture can be observed in ESR after injection of the appropriate charge carriers.
Since ESR is limited to atoms or molecules with unpaired spins, most electronic materials
are invisible to ESR. However, when ESR detects a species it is a powerful tool to
characterize such defects. It proved its worth when it showed the nature of SiO
2
=Si
interface traps to be P
b
centers (Chapter 6). Its sensitivity is generally poorer than electrical
techniques, but it is usually better than Auger electron spectroscopy, secondary-ion mass
spectrometry, x-ray crystallography, and others [20]. Although I–V, C–V, G–V, etc. tech-
niques are generally more sensitive than ESR, they lack specificity in determining the
atomic nature of the point defects. Hence ESR has clear advantages because it provides
atomic identification of point defects and enables distinction between the various types of
positive, negative, or amphoteric charge traps.
The technique depends on the transitions induced between spin states by a magnetic
field perpendicular to electromagnetic energy in the form of microwave signals at frequen-
cies of typically 9–10 GHz. The spin states are induced by the large magnetic field while the
transitions between spin states are induced by the radio frequency electromagnetic energy.
The magnetic field-induced spin state splitting, proportional to the magnetic field, is
illustrated in Figure 5.35. A radio frequency signal incident on the sample containing the
defect under study, leads to a transition from one spin state to another. Quantization of
spin angular momentum leads to the energies of an isolated unpaired electron of
FIGURE 5.34
Low-frequency noise spectra before and after
annealing. W=L ¼10 mm=0.8 mm, t
ox
¼3.3 nm,
V
G
ÀV
T
¼À1.05 V, V
D
¼À0.005 V. (From
Ahsan, A.K.M. and Schroder, D.K., IEEE Elec-
tron Dev. Lett., 25, 211, 2004. With permission.)
10
Ϫ15
10
Ϫ14
10
Ϫ13
10
Ϫ12
10
Ϫ11
10
Ϫ10
10
0
10
1
10
2
10
3
10
4
S
V
G

(
V
2
/
H
z
)
Frequency (Hz)
1000°C
anneal
No anneal
ß 2008 by Taylor & Francis Group, LLC.
E ¼ Æ
1
2
g
e
m
B
B
0
(5:51)
where
g
e
is the free electron g-value (2.00232)
m
B
is the Bohr magneton (9.27 Â10
À24
J=T)
B
0
is the magnetic field
The energy E in Equation 5.51 is the Zeeman energy. The energy level separation DE is
DE ¼ hf ¼ g
e
m
B
B
0
(5:52)
where f is the microwave frequency.
The measured deviation from the free electron g-value is very important in ESR meas-
urements, because it provides information about the paramagnetic center’s electronic
structure. An unpaired electron feels the external magnetic field B
0
and the effects of
local magnetic fields. Hence, the effective magnetic field is
B
eff
¼ B
0
(1 Às) (5:53)
where s, which can be positive or negative, accounts for the local field effects. Hence,
Equation 5.52 becomes
DE ¼ hf ¼ g
e
m
B
B
eff
¼ g
e
m
B
B
0
(1 Às) ¼ gm
B
B
0
(5:54)
g is determined during ESR measurements by knowing B
0
and f at resonance, i.e.,
g ¼
hf
m
B
B
0,res
(5:55)
where B
0,res
is the magnetic field at resonance, as indicated in Figure 5.35.
For a g-factor different from g
e
, the electron magnetic moment=angular momentum ratio,
the electron has gained or lost angular momentum, since the magnetic moment (Bohr
magneton) is constant. The energy gain=loss is due to spin–orbit coupling. With spin–orbit
coupling well understood, the g-factor change provides information about the nature of the
atomic or molecular orbital containing the electron. The measured g-value deviates from g
e
because the local magnetic field differs from the applied magnetic field due to spin–orbit
coupling and=or electron–nuclear hyperfine interactions.
Two measurement approaches are possible: (i) the magnetic field is constant and the
microwave frequency is varied, and (ii) the magnetic field is varied and the frequency is
E
a
= 0.5gm
B
B
0
E
a
= −0.5gm
B
B
0
B
0,res
ΔE = gm
B
B
0
rf-Induced
transition
Magnetic field
E
n
e
r
g
y
FIGURE 5.35
Energy versus magnetic field showing DE at the resonant
magnetic field B
0,res
.
ß 2008 by Taylor & Francis Group, LLC.
constant until resonance is achieved in either case. The latter is preferred. The ESR signal
looks qualitatively like that in Figure 5.36a, showing a peak at a certain magnetic field. To
determine the relevant magnetic field at resonance it is easier to plot the differentiated ESR
signal versus magnetic field, as also illustrated in Figure 5.36a, because the signal is
detected with a lock-in amplifier. Instead of the magnetic field at the peak, one simply
determines the zero-point crossing, indicated by the point.
The ESR spectrometer consists of a frequency-stable radiation source provided by a
Gunn diode and a magnet, which is slowly swept to record a spectrum. By far the most
widely used frequency is X-band, about 9.5 GHz, due to its relatively moderate magnetic
fields and inexpensive Gunn diode microwave sources. Higher frequencies require higher
magnetic fields, leading to higher sensitivity. The most common higher frequency is
35 GHz with a wavelength of 8 mm. Microwave components and samples are thus
correspondingly much smaller than for X-band measurements [115]. The sample is located
in a resonant cavity with dimensions matching the wavelength of the radiation so that a
standing wave pattern is set up in it. Varying the magnetic applied field to bring the
sample to resonance leads to microwave power absorption by the sample. Magnetic field
modulation is essential in ESR to allow for lock-in detection with sensitivities in the mid
10
10
=cm
2
range. The magnetic field is modulated at typically 100 kHz.
ESR has been extensively used to characterize defects in insulators and other materials.
As mentioned earlier, it led to the identification of the nature of interface traps at the
SiO
2
=Si interface. Early irradiation studies used ESR to identify various radiation-induced
defects (Chapter 6) [116]. Defect spectra in SiO
2
are easily separated from those due
to defects in Si since the resonance line shapes of SiO
2
defects are independent of magnetic
FIGURE 5.36
(a) ESR signal versus magnetic field and
derivative of ESR signal versus magnetic
field and (b) ESR signals from oxidized
(111) silicon wafer. (Reused from Caplan,
P.J., Poindexter, E.H., Deal, B.E., and
Razouk, R.R., J. Appl. Phys., 50, 5847,
1979. With permission.)
E
S
R

s
i
g
n
a
l
Magnetic field
E
S
R

s
i
g
n
a
l

d
e
r
i
v
a
t
i
v
e
(a)
(b)
50 G
2.0081
ß 2008 by Taylor & Francis Group, LLC.
field direction, whereas those in Si are not. The magnetic field orientation is due to the
physical defect location at the interface (interface of a crystalline material, hence repeated
oriented lattice structure) or in the oxide (randomly oriented amorphous material). Nega-
tive bias-temperature instability (NBTI) is a significant reliability issue in p-MOSFETs. It
manifests itself as a threshold voltage and drain current degradation and interface trap
generation is one of the culprits. ESR was used early on during NBTI recognition to follow
P
b
center generation by negative gate voltage stress [117].
Corona charging is used routinely in semiconductor characterization [118]. The method
consists of depositing corona charge on the surface of the sample and measuring a variety
of material=device parameters. The corona charge replaces the metal=poly-Si gate. On
oxidized samples, it can generate oxide electric fields as high as 10–15 MV=cm. Since the
deposited ions are thermalized by their passage through a long air path, corona charging
would seem to constitute an almost ideal nondamaging biasing tool. For positive biasing,
the charging process involves hydrogenic species that can interact with the Si=SiO
2
system.
The impact of room temperature corona charging biasing on thermal (100) and (111)
Si=SiO
2
has been probed by ESR [119]. Both structures fully passivated in H
2
as well as
exhaustively H-depleted ones were investigated. At least five types of ESR-active defects
were generated, most likely due to the interaction of atomic hydrogen released during the
corona process with the Si=SiO
2
system. However, for lower corona-induced oxided
electric fields, this biasing technique does not damage the sample [120].
A variation of ESR is spin-dependent recombination (SDR) [121]. In conventional
Shockley–Read–Hall recombination, electrons and holes are captured by deep-level impur-
ities during a recombination event. In SDR, the semiconductor device is placed in a strong
magnetic field that polarizes the spins of the dangling bond electrons as well as the
conduction electrons and the holes. Consider a silicon dangling bond defect with an
unpaired electron when it is electrically neutral. Let the SRH recombination process
begin with electron capture by the defect. If both the dangling bond and electron have
the same spin orientation, the conduction electron cannot be captured, because two
electrons may not occupy the same orbital with the same spin quantum number. Thus,
placing the semiconductor sample in the magnetic field modulates the trap capture cross
section by a combination of magnetic field and microwaves.
5.4 Summary
This chapter has provided an overview of some of the more common defects observed in
MOS devices, divided into oxide, border, and interface traps. While the defects in pure
SiO
2
are quite well understood, that is not the case for high-K=oxide insulators. Not only
are high-K dielectrics more complex, there is an additional interface in such insulators and
the interface and bulk defects are usually higher and less well understood (Chapters 9
through 12). A comprehensive overview has also been provided of a fairly complete set of
electrical characterization methods. In most cases, these techniques, first developed for
SiO
2
, were extended to more complex dielectrics. For some of the newer techniques a brief
theoretical discussion was provided, but for the well-known methods, the reader is direc-
ted to the appropriate references. Wherever possible, applications are provided to interface
oxide and border trap measurements. Other techniques may well be developed in
the future, but for the most part existing techniques, some of which have been around
since the 1960s, have served the IC community well and will continue to do so. Some of the
techniques must be used with caution for advanced ICs because gate oxide leakage current
and other nonidealities make data interpretation more difficult.
ß 2008 by Taylor & Francis Group, LLC.
References
1. Deal, B.E. et al., Characteristics of the surface-state charge (Q
SS
) of thermally oxidized silicon,
J. Electrochem. Soc. 114, 266, 1967.
2. Deal, B.E., Current understanding of charges in thermally oxidized silicon structure,
J. Electrochem. Soc. 121, 198C, 1974.
3. Warren, W.L. et al., A proposed model for positive charge in SiO
2
thin films over-coordinated
oxygen centers, IEEE Trans. Nucl. Sci. 43, 2617, 1996.
4. Snow, E.H. et al., Ion transport phenomena in insulating films, J. Appl. Phys. 36, 1664, 1965.
5. Pliskin, W.A. and Gdula, R.A., Passivation and insulation, in Handbook on Semiconductors, Vol. 3,
S.P. Keller (ed.), North Holland, Amsterdam, 1980 and references therein.
6. Chou, N.J., Application of triangular voltage sweep method to mobile charge studies in MOS
structures, J. Electrochem. Soc. 118, 601, 1971; Derbenwick, G., Mobile ions in SiO
2
: Potassium,
J. Appl. Phys. 48, 1127, 1977; Stagg, J.P., Drift mobilities of Na
þ
and K
þ
ions in SiO
2
films,
Appl. Phys. Lett. 31, 532, 1977; Hillen, M.W., Greeuw, G., and Verwey, J.F., On the mobility of
potassium ions in SiO
2
, J. Appl. Phys. 50, 4834, 1979; Kuhn, M. and Silversmith, D.J., Ionic
contamination and transport of mobile ions in MOS structures, J. Electrochem. Soc. 118, 966, 1971.
7. Greeuw, G. and Verwey, J.F., The mobility of Na
þ
, Li
þ
and K
þ
ions in thermally grown SiO
2
films, J. Appl. Phys. 56, 2218, 1984.
8. Shacham-Diamand, Y. et al., Copper transport in thermal SiO
2
, J. Electrochem. Soc. 140, 2427,
1993.
9. Feigl, F.J., Fowler, W.B., and Yip, K.L., Oxygen vacancy model for the E
1
0
center in SiO
2
, Solid
State Commun. 14, 225, 1974.
10. Lenahan, P.M. andDressendorfer, P.V., Hole traps andtrivalent silicon centers inmetal=oxide=silicon
devices, J. Appl. Phys. 55, 3495, 1984.
11. Shklovskii, B.I. and Efros, A.L., Electronic Properties of Doped Semiconductors, Springer, Berlin,
1984.
12. Massoud, H.Z. and Deaton, R., Percolation model for the extreme-value statistics of dielectric
breakdown in rapid-thermal oxides, Extended Abstracts of the ECS Meeting, San Francisco, 287, 1994.
13. Degraeve, R. et al., A consistent model for the thickness dependence of intrinsic breakdown in
ultra-thin oxides, IEEE IEDM Tech. Dig. 863, 1995.
14. Degraeve, R. et al., New insights in the relation between electron trap generation and the
statistical properties of oxide breakdown, IEEE Trans. Electron Dev. 45, 904, 1998.
15. Stathis, J.H., Quantitative model of the thickness dependence of breakdown in ultrathin oxides,
Microelectron. Eng. 36, 325, 1997.
16. Wolters, D.R. and Van der Schoot, J.J., Dielectric breakdown in MOS devices, Part I: Defect-
related and intrinsic breakdown, Philips Res. Rep. 40, 115, 1985; Wolters, D.R. and Verwey, J.F.,
Breakdown and wear-out phenomena in SiO
2
films, in Instabilities in Silicon Devices, Elsevier,
Amsterdam, 1986, Chapter 6.
17. Wu, E.Y. and Suñé, J., Power-law voltage acceleration: A key element for ultra-thin gate oxide
reliability, Microelectron. Reliab. 45, 1809, 2005.
18. Lelis, A.J. et al., The nature of the trapped hole annealing process, IEEE Trans. Nucl. Sci. 36, 1808,
1989.
19. Walters, M. and Reisman, A., Radiation-induced neutral electron trap generation in
electrically biased insulated gate field effect transistor gate insulators, J. Electrochem. Soc. 138,
2756, 1991.
20. Poindexter, E.H. and Caplan, P.J., Characterization of Si=SiO
2
interface defects by electron spin
resonance, Prog. Surf. Sci. 14, 201, 1983.
21. Nishi, Y., Study of silicon–silicon dioxide structure by electron spin resonance I, Jpn. J. Appl.
Phys. 10, 52, 1971.
22. Caplan, P.J. et al., ESR centers, interface states, and oxide fixed charge in thermally oxidized
silicon wafers, J. Appl. Phys. 50, 5847, 1979.
23. Poindexter, E.H. et al., Interface states and electron spin resonance centers in thermally oxidized
(111) and (100) silicon wafers, J. Appl. Phys. 52, 879, 1981.
ß 2008 by Taylor & Francis Group, LLC.
24. Stirling, A. et al., Dangling bond defects at Si–SiO
2
interfaces: Atomic structure of the P
b1
center,
Phys. Rev. Lett. 85, 2773, 2000.
25. Stesmans, A. and Afanas’ev, V.V., Nature of the P
b1
interface defect in (100) Si=SiO
2
as revealed
by electron spin resonance
29
Si hyperfine structure, Microelectron. Eng. 48, 113, 1999.
26. Stesmans, A. and Afanas’ev, V.V., Electrical activity of interfacial paramagnetic defects in
thermal (100) Si=SiO
2
, Phys. Rev. 57, 10030, 1998.
27. Gray, P.V. and Brown, D.M., Density of SiO
2
–Si interface states, Appl. Phys. Lett. 8, 31, 1966;
Fleetwood, D.M., Long-term annealing study of midgap interface-trap charge neutrality,
Appl. Phys. Lett. 60, 2883, 1992.
28. Deal, B.E., Standardized terminology for oxide charges associated with thermally oxidized
silicon, IEEE Trans. Electron Dev. ED-27, 606, 1980.
29. Fleetwood, D.M., Border traps in MOS devices, IEEE Trans. Nucl. Sci. 39, 269, 1992.
30. Fleetwood, D.M. et al., Effects of oxide traps, interface traps, and border traps on MOS devices,
J. Appl. Phys. 73, 5058, 1993; Fleetwood, D.M., Fast and slow border traps in MOS devices,
IEEE Trans. Nucl. Sci. 43, 779, 1996.
31. Christensson, S., Lundstrom, I., and Svensson, C., Low-frequency noise in MOS transistors:
I-theory, Solid State Electron. 11, 797, 1968.
32. Tewksbury, T.L. and Lee, H.S., Characterization, modeling, and minimization of transient
threshold voltage shifts in MOSFET’s, IEEE J. Solid State Circ. 29, 239, 1994.
33. Burstein, E. and Lundquist, S., Tunneling Phenomena in Solids, Plenum Press, New York, 1969.
34. Fu, H.S. and Sah, C.T., Theory and experiments on surface 1=f noise, IEEE Trans. Electron Dev.
ED-19, 273, 1972.
35. Jameson, J.R. et al., Charge trapping in high-K gate stacks due to the bilayer structure itself,
IEEE Trans. Electron Dev. 53, 1858, 2006.
36. Frankl, D.R., Some effects of material parameters on the design of surface space-charge
varactors, Solid State Electron. 2, 71, 1961.
37. Terman, L.M., An investigation of surface states at a silicon=silicon oxide interface employing
metal-oxide-silicon diodes, Solid State Electron. 5, 285, 1962.
38. Grove, A.S. et al., Investigation of thermally oxidized silicon surfaces using metal-oxide-semi-
conductor structures, Solid State Electron. 8, 145, 1965.
39. Deal, B.E., Snow, E.H., and Mead, C.A., Barrier energies in metal–silicon dioxide–silicon struc-
tures, J. Phys. Chem. Solids 27, 1873, 1966.
40. Zaininger, K.H. and Heiman, F.P., The C–V technique as an analytical tool, Solid State Technol. 13,
25, 1970.
41. Nicollian, E.H. and Brews, J.R., MOS (Metal Oxide Semiconductor) Physics and Technology, Wiley,
New York, 1982.
42. Werner, M.W., The work function difference of the MOS-system with aluminum field plates and
polycrystalline silicon field plates, Solid State Electron. 17, 769, 1974.
43. Park, C.S., Cho, B.J., and Kwok, D.-L., Thermally stable fully silicided Hf-silicide metal-gate
electrode, IEEE Electron Dev. Lett. 25, 372, 2004.
44. Jha, R. et al., A capacitance-based methodology for work function extraction of metals on high-K,
IEEE Electron Dev. Lett. 25, 420, 2004.
45. Kaushik, V.S. et al., Estimation of fixed charge densities in hafnium-silicate gate dielectrics,
IEEE Trans. Electron Dev. 53, 2627, 2006.
46. Wen, H.C. et al., Systematic investigation of amorphous transition-metal-silicon-nitride
electrodes for metal gate CMOS applications, IEEE VLSI Symp. Tech. Dig. 46, 2005.
47. Hou, Y.T. et al., High performance tantalum carbide metal gate stacks for nMOSFET application,
IEEE IEDM Tech. Dig., 31, 2005.
48. Majhi, P. et al. Evaluation and integration of metal gate electrodes for future generation dual
metal CMOS, IEEE Int. Conf. Integr. Circ. Technol. 69, 2005.
49. Castagné, R., Determination of the slow density of an MOS capacitor using a linearly varying
voltage (in French), C.R. Acad. Sci. Paris 267, 866, 1968; Kuhn, M., A quasi-static technique for
MOS C–V and surface state measurements, Solid State Electron. 13, 873, 1970; Kappallo, W.K. and
Walsh, J.P., A current voltage technique for obtaining low-frequency C–V characteristics of MOS
capacitors, Appl. Phys. Lett. 17, 384, 1970.
ß 2008 by Taylor & Francis Group, LLC.
50. Berglund, C.N., Surface states at steam-grown silicon–silicon dioxide interfaces, IEEE Trans.
Electron Dev. ED-13, 701, 1966.
51. Castagné, R. and Vapaille, A., Description of the SiO
2
–Si interface properties by means of very
low frequency MOS capacitance measurements, Surf. Sci. 28, 157, 1971.
52. Cohen, N.L., Paulsen, R.E., and White, M.H., Observation and characterization of near-interface
oxide traps with C–V techniques, IEEE Trans. Electron Dev. 42, 2004, 1995.
53. Wu, W.H. et al., Spatial and energetic distribution of border traps in the dual-layer HfO
2
=SiO
2
high-k gate stack by low-frequency capacitance–voltage measurement, Appl. Phys. Lett. 89,
162911, 2006.
54. Schroder, D.K., Semiconductor Material and Device Characterization, 3rd edn., Wiley-Interscience,
New York, 2006, p. 432.
55. Hall, S., Buiu, O., and Lu, Y., Direct observation of anomalous positive charge and electron-
trapping dynamics in high-k films using pulsed MOS-capacitor measurements, IEEE Trans.
Electron Dev. 54, 272, 2007.
56. Nicollian, E.H. and Goetzberger, A., The Si–SiO
2
interface—electrical properties as determined
by the metal-insulator-silicon conductance technique, Bell Syst. Tech. J. 46, 1055, 1967.
57. Nicollian, E.H. and Brews, J.R., MOS (Metal Oxide Semiconductor) Physics and Technology, Wiley,
New York, 1982, p. 180.
58. Vogel, E.M. et al., Limitations of conductance to the measurement of the interface state density of
MOS capacitors with tunneling gate dielectrics, IEEE Trans. Electron Dev. 47, 601, 2000; Ma, T.P.
and Barker, R.C., Surface-state spectra from thick-oxide MOS tunnel junctions, Solid State
Electron. 17, 913, 1974.
59. Eaton, D.H. and Sah, C.T., Frequency response of Si–SiO
2
interface states on thin oxide MOS
capacitors, Phys. Stat. Sol. 12(a), 95, 1972.
60. Uren, M.J., Collins, S., and Kirton, M.J., Observation of ‘‘slow’’ states in conductance measure-
ments on silicon metal-oxide-semiconductor capacitors, Appl. Phys. Lett. 54, 1448, 1989.
61. Kuhn, M. and Silversmith, D.J., Ionic contamination and transport of mobile ions in MOS
structures, J. Electrochem. Soc. 118, 966, 1971; Hillen, M.W. and Verwey, J.F., Mobile ions in
SiO
2
layers on Si, in Instabilities in Silicon Devices: Silicon Passivation and Related Instabilities,
G. Barbottin and A. Vapaille (eds.), Elsevier, Amsterdam, 1986, p. 403.
62. Stauffer, L. et al., Mobile ion monitoring by triangular voltage sweep, Solid State Technol. 38, S3,
1995.
63. Hillen, M.W., Greeuw, G., and Verweij, J.F., On the mobility of potassium ions in SiO
2
, J. Appl.
Phys. 50, 4834, 1979.
64. Lang, D.V., Deep-level transient spectroscopy: a new method to characterize traps in semi-
conductors, J. Appl. Phys. 45, 3023, 1974; Lang, D.V., Fast capacitance transient apparatus:
Application to ZnO and O centers in GaP p–n junctions, J. Appl. Phys. 45, 3014, 1974.
65. Schroder, D.K., Semiconductor Material and Device Characterization, 3rd edn., Wiley-Interscience,
New York, 2006, p. 270.
66. Choi, B.D. et al., Latent iron in silicon, Jpn. J. Appl. Phys. 40, L915, 2001.
67. Wang, K.L. and Evwaraye, A.O., Determination of interface and bulk-trap states of IGFET’s
using deep-level transient spectroscopy, J. Appl. Phys. 47, 4574, 1976.
68. Schulz, M. and Johnson, N.M., Transient capacitance measurements of hole emission frominterface
states in MOS structures, Appl. Phys. Lett. 31, 622, 1977; Tredwell, T.J. and Viswanathan, C.R.,
Determination of interface-state parameters in a MOS capacitor by DLTS, Solid State Electron. 23,
1171, 1980; Yamasaki, K., Yoshida, M., and Sugano, T., Deep level transient spectroscopy of bulk
traps and interface states in Si MOS diodes, Jpn. J. Appl. Phys. 18, 113, 1979.
69. Johnson, N.M., Measurement of semiconductor–insulator interface states by constant-
capacitance, deep-level transient spectroscopy, J. Vac. Sci. Technol. 21, 303, 1982.
70. Yamasaki, K., Yoshida, M., and Sugano, T., Deep level transient spectroscopy of bulk traps and
interface states in Si MOS diodes, Jpn. J. Appl. Phys. 18, 113, 1979.
71. Lakhdari, H., Vuillaume, D., and Bourgoin, J.C., Spatial and energetic distribution of Si–SiO
2
near-interface states, Phys. Rev. B38, 13124, 1988.
72. Brugler, J.S. and Jespers, P.G.A., Charge pumping in MOS devices, IEEE Trans. Electron Dev.
ED-16, 297, 1969.
ß 2008 by Taylor & Francis Group, LLC.
73. Bauza, D. and Ghibaudo, G., Analytical study of the contribution of fast and slow oxide traps to
the charge pumping current in MOS structures, Solid State Electron. 39, 563, 1996.
74. Weintraub, C.E. et al., Study of low-frequency charge pumping on thin stacked dielectrics,
IEEE Trans. Electron Dev. 48, 2754, 2001.
75. Groeseneken, G. et al., A reliable approach to charge-pumping measurements in MOS tran-
sistors, IEEE Trans. Electron Dev. ED-31, 42, 1984.
76. Han, J.P. et al., Asymmetric energy distribution of interface traps in n- and p-MOSFETs with
HfO
2
gate dielectric on ultrathin SiON buffer layer, IEEE Electron Dev. Lett. 25, 126, 2004.
77. Tseng, W.L., A new charge pumping method of measuring Si–SiO
2
interface states, J. Appl. Phys.
62, 591, 1987; Hofmann, F. and Krautschneider, W.H., A simple technique for determining the
interface-trap distribution of submicron metal-oxide-semiconductor transistors by the charge
pumping method, J. Appl. Phys. 65, 1360, 1989.
78. Saks, N.S. and Ancona, M.G., Determination of interface trap capture cross sections using three-
level charge pumping, IEEE Electron Dev. Lett. 11, 339, 1990; Siergiej, R.R., White, M.H., and
Saks, M.S., Theory and measurement of quantization effects on Si–SiO
2
interface trap modeling,
Solid State Electron. 35, 843, 1992.
79. Bauza, D., Extraction of Si–SiO
2
interface trap densities in MOS structures with ultrathin oxides,
IEEE Electron Dev. Lett. 23, 658, 2002; Bauza, D., Electrical properties of Si–SiO
2
interface traps and
evolution with oxide thickness in MOSFET’s with oxides from 2.3 to 1.2 nm thick, Solid
State Electron. 47, 1677, 2003; Masson, P., Autran, J.-L. and Brini, J., On the tunneling
component of charge pumping current in ultrathin gate oxide MOSFET’s, IEEE Electron Dev.
Lett. 20, 92, 1999.
80. Paulsen, R.E. and White, M.H., Theory and application of charge pumping for the characterization
of Si–SiO
2
interface and near-interface oxide traps, IEEE Trans. Electron Dev. 41, 1213, 1994.
81. Heiman, F.P. and Warfield, G., The effects of oxide traps on the MOS capacitance, IEEE Trans.
Electron Dev. ED-12, 167, 1965.
82. Jakschik, S. et al., Influence of Al
2
O
3
dielectrics on the trap-depth profiles in MOS devices
investigated by the charge-pumping method, IEEE Trans. Electron Dev. 51, 2252, 2004.
83. Cartier, E. et al., Fundamental understanding and optimization of PBTI in nFETs with
SiO
2
=HfO
2
gate stack, IEEE Int. Electron Dev. Meet. 321, 2006.
84. Kerber, A. et al., Direct measurement of the inversion charge in MOSFETs: Application to
mobility extraction in alternative gate dielectrics, IEEE VLSI Symp. Tech. Dig. 159, 2003.
85. Taur, Y. and Ning, T.H., Fundamentals of Modern VLSI Devices, Cambridge University Press,
Cambridge, 1998.
86. McWhorter, P.J. andWinokur, P.S., Simple technique for separating the effects of interface traps and
trapped-oxide charge in metal-oxide-semiconductor transistors, Appl. Phys. Lett. 48, 133, 1986.
87. Kerber, A. et al., Characterization of the V
T
-instability in SiO
2
=HfO
2
gate dielectrics, IEEE Int.
Rel. Phys. Symp. 41, 41, 2003.
88. Zhao, C.Z. et al., Determination of capture cross sections for as-grown electron traps in
HfO
2
=HfSiO stacks, J. Appl. Phys. 100, 093716, 2006.
89. Neugroschel, A. et al., Direct-current measurements of oxide and interface traps on oxidized
silicon, IEEE Trans. Electron Dev. 42, 1657, 1995.
90. Fitzgerald, D.J. and Grove, A.S., Surface recombination in semiconductors, Surf. Sci. 9, 347, 1968.
91. Guan, H. et al., Nondestructive DCIV method to evaluate plasma charging in ultrathin gate
oxides, IEEE Electron Dev. Lett. 20, 238, 1999.
92. Maserijian, J. and Zamani, N., Behavior of the Si=SiO
2
interface observed by Fowler–Nordheim
tunneling, J. Appl. Phys. 53, 559, 1982.
93. DiMaria, D.J. and Cartier, E., Mechanism for stress-induced leakage currents in thin silicon
dioxide films, J. Appl. Phys. 78, 3883, 1995.
94. DiMaria, D.J., Defect generation in field-effect transistors under channel-hot-electron stress,
J. Appl. Phys. 87, 8707, 2000.
95. Petit, C. et al., Low voltage SILC and P- and N-MOSFET gate oxide reliability, Microelectron.
Reliab. 45, 479, 2005.
96. Ghetti, A. et al., Tunneling into interface states as reliability monitor for ultrathin oxides,
IEEE Trans. Electron Dev. 47, 2358, 2000.
ß 2008 by Taylor & Francis Group, LLC.
97. Gu, S.H. et al., Extraction of nitride trap density from stress induced leakage current in silicon-
oxide-nitride-oxide-silicon flash memory, Appl. Phys. Lett. 89, 163514, 2006.
98. Ning, T.H., Hot-electron emission fromsilicon into silicon dioxide, Solid State Electron. 21, 273, 1978.
99. Van den bosch, G., Groeseneken, G., andMaes, H. Direct andpost-injection oxide andinterface trap
generation resulting from low-temperature hot-electron injection, J. Appl. Phys. 74, 5582, 1993.
100. Nissan-Cohen, Y., Shappir, J., and Frohman-Bentchkowsky, D., High field current induced-
positive charge transients in SiO
2
, J. Appl. Phys. 54, 5793, 1983.
101. Wu, E.Y. et al., Challenges for accurate reliability projections in the ultra-thin oxide regime,
IEEE Int. Reliab. Phys. Symp. 37, 57, 1999.
102. Wong, H., Low-frequency noise study in electron devices: review and update, Microelectron.
Reliab. 43, 585, 2003.
103. Claeys, C., Mercha, A., and Simoen, E., Low-frequency noise assessment for deep submicro-
meter CMOS technology nodes, J. Electrochem. Soc. 151, G307, 2004.
104. Johnson, J.B., The Schottky effect in low frequency circuits, Phys. Rev. 26, 71, 1925.
105. McWhorter, A.L., 1=f noise and germanium surface properties, in Semiconductor Surface Physics,
R.H. Kingston (ed.), University of Pennsylvania Press, Philadelphia, 1957, pp. 207–228.
106. Hooge, F.N., 1=f noise is no surface effect, Phys. Lett. 29A, 139–140, 1969; Discussion of recent
experiments on 1=f noise, Physica 60, 130, 1976; 1=f noise, Physica 83B, 14, 1976; Hooge, F.N. and
Vandamme, L.K.J., Lattice scattering causes 1=f noise, Phys. Lett. 66A, 315, 1978; Hooge, F.N., 1=f
noise sources, IEEE Trans. Electron Dev. 41, 1926, 1994.
107. Christensson, S., Lundström, I., and Svensson, C., Low frequency noise in MOS transistors
I. Theory, Solid State Electron. 11, 797, 1968; Christensson, S. and Lundström, I., Low frequency
noise in MOS transistors II. Experiments, Solid State Electron. 11, 813, 1968.
108. Kirton, M.J. and Uren, M.J., Noise in solid-state microstructures: A new perspective on individ-
ual defects, interface states and low-frequency (1=f) noise, Adv. Phys. 38, 367, 1989.
109. Gu, S.H. et al., Read current instability arising from random telegraph noise in localized storage,
multi-level SONOS flash memory, IEEE Int. Electron Dev. Meet. 487, 2006.
110. Yuzhelevski, Y., Yuzhelevski, M., and Junga, G., Random telegraph noise analysis in time
domain, Rev. Sci. Instrum. 71, 1681, 2000.
111. Hung, K.K. et al., A unified model for flicker noise in metal oxide semiconductor field effect
transistors, IEEE Trans. Electron Dev. 37, 654, 1990; A physics based MOSFET noise model for
circuit simulators, IEEE Trans. Electron Dev. 37, 1323, 1990.
112. Ahsan, A.K.M. and Schroder, D.K., Impact of post-oxidation annealing on low-frequency noise,
threshold voltage, and subthreshold swing of p-channel MOSFETs, IEEE Electron Dev. Lett. 25,
211, 2004.
113. Simoen, E., Mercha, A., and Claeys, C., Noise diagnostics of advanced silicon substrates and
deep submicron process modules, in Analytical and Diagnostic Techniques for Semiconductor
Materials, Devices, and Processes, B.O. Kolbesen et al., (eds.), Electrochem. Soc. ECS PV 2003-03,
420, 2003; Härtler, G., Golze, U., and Paschke, K., Extended noise analysis—a novel tool for
reliability screening, Microelectron. Reliab. 38, 1193, 1998.
114. Yasuda, Y. and Hu, C., Effect of fluorine incorporation on 1=f noise of HfSiON FETs for future
mixed-signal CMOS, IEEE Int. Electron Dev. Meet. 06–277, 2006.
115. Atherton, N.M., Principles of Electron Spin Resonance, Ellis Horwood, New York, 1993.
116. Brower, K.L., Lenahan, P.M., and Dressendorfer, P.V., Defects and impurities in thermal oxides
on silicon, Appl. Phys. Lett. 41, 251, 1982.
117. Gerardi, G.J. et al., Generation of P
b
centers by high electric fields: thermochemical effects,
J. Electrochem. Soc. 136, 2609, 1989.
118. Schroder, D.K., Contactless surface charge semiconductor characterization, Mater. Sci. Eng.
B91–92, 196, 2002.
119. Stesmans, A. and Afanas’ev, V.V., Corona charging damage on thermal Si=SiO
2
structures with
nm-thick oxides revealed by electron spin resonance, Microelectron. Eng. 72, 55, 2004.
120. Dautrich, M.S. et al., Noninvasive nature of corona charging on thermal Si=SiO
2
structures,
Appl. Phys. Lett. 85, 1844, 2004.
121. Lepine, D.J., Spin-dependent recombination on silicon surface, Phys. Rev. B6, 436, 1972.
ß 2008 by Taylor & Francis Group, LLC.
6
Dominating Defects in the MOS System:
P
b
and E
0
Centers
Patrick M. Lenahan
CONTENTS
6.1 Introduction and Background.......................................................................................... 164
6.2 ESR....................................................................................................................................... 165
6.2.1 Spin–Orbit Coupling .............................................................................................. 165
6.2.2 Electron–Nuclear Hyperfine Interactions............................................................ 167
6.2.3 Quantitative Analysis (Spin Counting) ............................................................... 171
6.2.4 Spin-Dependent Recombination........................................................................... 172
6.3 Defect Structure.................................................................................................................. 172
6.3.1 P
b
Center Structure................................................................................................. 172
6.3.2 E
0
Center Structure.................................................................................................. 177
6.3.2.1 Structure of Conventional E
0
Centers..................................................... 177
6.3.2.2 Structure of Hydrogen Complexed E
0
Centers ..................................... 180
6.3.2.3 Structure of E
0
d
(or EP) Center ................................................................. 181
6.3.2.4 Structure of Other Less Well-Characterized E
0
Variants Found in Thin Oxide Films...................................................... 182
6.4 Electronic Properties of P
b
and E
0
Centers..................................................................... 182
6.4.1 P
b0
and P
b1
Centers and the Density of States
of Si=SiO
2
Interface Traps...................................................................................... 182
6.4.2 Role of P
b
Centers in Device Performance: The Correspondence
between P
b
Density and Interface Trap Density................................................ 187
6.4.2.1 P
b
Centers in As-Processed Structures................................................... 187
6.4.2.2 P
b
Centers and Radiation Damage ......................................................... 187
6.4.2.3 P
b
Centers Generated by High and Low Oxide Field
Electron Injection....................................................................................... 189
6.4.2.4 P
b
Centers Generated by Hot Carrier Stress
of Short-Channel MOSFETs..................................................................... 189
6.4.2.5 P
b
Centers Generated in the Negative Bias
Temperature Instability ............................................................................ 190
6.4.2.6 Conclusion Regarding P
b
Centers and Si=SiO
2
Instabilities ............... 191
6.4.3 Roles of E
0
Centers in MOS Devices .................................................................... 191
6.4.3.1 E
0
Centers as Oxide Hole Traps .............................................................. 191
6.4.3.2 Neutral Paramagnetic E
0
Centers............................................................ 195
6.4.3.3 Positive but Compensated E
0
Centers .................................................... 196
ß 2008 by Taylor & Francis Group, LLC.
6.4.3.4 Struct ural Changes at E
0
Sites Res ulting
from Repeat ed Char ge Cycl ing ............... ............... ............... ............... ... 197
6.4.3.5 Ne ar Si=Si O
2
Interf ace E
0
Cente rs: Electronic Prop erties ............... ...... 197
6.4.3.6 Hyd rogen Comp lexed E
0
Centers ............... ........... ............... ............... ... 199
6.4.3.7 Ele ctronic Prope rties of E
0
d
(or EP) ............... .......... ............... ............... ... 199
6.4.3.8 Role of E
0
Centers in Ox ide Leakage Currents .... ............... ............... ... 200
6.5 Intrinsic Defe cts and Device Reliability: Physi cally Based
Predictiv e Mo dels and Stat istical Mech anics ............... ............... .. ............... ............... ... 202
6.5.1 Predictin g Oxide Hole Tra pping ............... ............... ........... ............... ............... ... 203
6.5.2 Predictin g Interface Trap For mation ............... ............... ..... ............... ............... ... 205
6.6 Relevan ce of P
b
and E
0
Centers to High Dielectri c
Constan t (Hi gh-K ) MOS FETs ........ ............... ............... ............... ..... ............... ............... ... 209
6.7 Summar y ............... ............... ............ ............... ............... ............... ..... ............... ............... ... 209
Refere nces ............... ............... ............... .... ............... ............... ............... ..... ............... ............... ... 210
6. 1 Intr odu ctio n an d B ack gro un d
Two types of silicon dan gling bond cen ters play domi nating roles in met al=oxide =silicon
(MO S) device limitati ons. Silicon dangling bonds defects at the Si =SiO
2
bound ary,
P
b
centers dominat e Si=Si O
2
interface trapping. Silicon dangling bond defects called
E
0
centers dominate deep levels in the oxide. E
0
centers can also play dominating roles in
the electronic properties of the near Si=SiO
2
interface dielectrics, serving as switching traps
or bor der trap s (Ch apter 7). Unde r some circumst ances, some E
0
centers also act as inter face
traps. Analysis of a combination of electron spin resonance (ESR) and electrical measure-
ments provides a fairly detailed understanding of the relationship between the structure
and electronic properties of the P
b
interface centers and E
0
oxide centers. Straightforward
application of the fundamental principles of statistical mechanics provides a partial explan-
ation of the frequently noted correlation between oxide deep level defects and Si=SiO
2
interface trap defects.
This chapter reviews the spectroscopy, electronic properties, and device relevance of
these two families of defects. Since the understanding of the defects has come from ESR,
the chapter begins with a brief introduction to this technique. It is important to note that
P
b
and E
0
centers are not necessarily the only electrically active point defects in MOS
devices. Although many studies have established the dominating roles of these two
families of silicon dangling bond centers, the limited precision of ESR measurements as
well as that of the relevant electrical measurements precludes concluding with certainty
that either defect is entirely responsible for any device limitation. However, they are clearly
responsible for the majority of the deep levels in most and perhaps all the phenomena
discussed in this chapter. Furthermore, under some circumstances, it is clear that other
defects are playing important roles. For example, the addition of nitrogen introduces
greater complexity in plasma-nitrided gate oxides. Not surprisingly, several very different
defects clearly play important roles in gate stacks based on hafnium oxide=silicon dioxide
systems.
Several decades of ESR [1] studies of the MOS system have established the roles of these
two families of silicon dangling bond defects in MOS device instabilities. At the Si=SiO
2
boundary the P
b
center family dominates interface trapping under many circumstances
[2–32]. Largely due to the simplicity of the ESR spectroscopy, most earlier studies focused
upon the (111) Si=SiO
2
interface, where only one P
b
defect, called simply P
b
, appears [2–15].
ß 2008 by Taylor & Francis Group, LLC.
At the technologically important (100) Si=SiO
2
interface, two P
b
center variants are
important: the P
b0
and the P
b1
centers [5,14,17,19,22–32]. The P
b0
center has been reason-
ably well understood for quite some time because it is essentially identical to the (111)
Si=SiO
2
silicon dangling bond center called P
b
. The relative simplicity of the (111) Si=SiO
2
boundary not only makes sensitive ESR measurement relatively easy; the interface also
makes the analysis particularly easy. This ease of measurement and analysis led to a fairly
complete first-order understanding of the (111) P
b
by the late 1980s. The (111) P
b
and the
(100) P
b0
centers are both silicon dangling bond centers in which the central silicon is back
bonded to three other silicons [3–5,9,17,22,25,27–29]. Both centers’ dangling bond orbitals
are directed along (111) symmetry axes. The centers have quite similar ESR parameters.
The P
b1
center has been less well understood, with gross conflicts appearing in the
literature with regard to its electronic properties [14,28–30,32].
Although several dozen paramagnetic (ESR active) centers have been observed in
MOS oxides, the E
0
center family [33–37] plays a dominating role in a wide variety of
circumstances [10,11,16,18–22,24,38–43]. The centers clearly dominate oxide hole trapping
[10,11,19,21,38,39] and almost certainly play important roles in instabilities resulting from
the presence of very high electric fields such as stress-induced leakage currents [16,42,43].
Both the P
b
and E
0
centers involve unpaired electrons primarily residing in high
p-character wave function orbitals on silicon atoms [9,17,22,25,27–29,34–37,41]. The P
b
centers involve silicons back bonded to three silicons. The E
0
centers involve silicons back
bonded to oxygens.
6.2 ESR
In ESR measurements [1], the sample under study is exposed to a large slowly varying
magnetic field and a microwave frequency magnetic field oriented perpendicularly to the
applied field [1]. Usually the measurements are made at X band: a microwave frequency
n % 9.5 GHz. An unpaired electron has two possible orientations in the large applied field
and thus two possible orientation-dependent energies. (From classical electricity and
magnetism, the energy of a magnetic moment m in a magnetic field H is ÀmÁ H.) Magnetic
resonance occurs when the energy difference between the two electron orientations is equal
to the Planck constant, h, times the microwave frequency. For the very simple case of an
isolated electron, the resonance requirement may be expressed as
hv ¼ g
0
b
e
H (6:1)
where g
0
¼2.002319 and b
e
is the Bohr magneton, eh=4pm
e
, where e is electronic charge
and m
e
is the electron mass. The Bohr magneton is 9.274015 Â10
À28
J=G.
Equation 6.1 describes the resonance condition for an electron that does not otherwise
interact with its surroundings. The structural information provided by ESR is due to
deviations from this simple equation. For the relatively simple trapping centers studied
in MOS systems, these deviations are due to spin–orbit coupling and electron–nuclear
hyperfine interactions.
6.2.1 Spin–Orbit Coupling
The deviations fromEquation 6.1 due to spin–orbit coupling come about because a charged
particle, the electron, traveling in an electric field due to the nuclear charge, experiences a
ß 2008 by Taylor & Francis Group, LLC.
magnetic field B¼EÂv=c
2
, where E is the electric field, v is the velocity, and c is the speed of
light [1]. The spin–orbit interaction may be understood qualitatively (and only qualitatively)
in terms of the Bohr picture: an electron moves about the nucleus in a circular orbit. It would
appear to an observer on the electron that the positively charged nucleus is in a circular orbit
about the electron. (Similarly, it appears to an unsophisticated observer on Earth that the Sun
is in a circular orbit about the Earth.) The nucleus thus generates a local magnetic field that
would scale with the electron’s orbital angular momentum, r Âp, and with the nuclear
charge. One would thus correctly surmise that spin–orbit coupling interactions increase
with increasing atomic number and orbital angular momentumquantumnumber.
In solids, the spin–orbit interaction is quenched, but a second-order effect appears froma
coupling to excited states through spin–orbit coupling generated by the application of a
large magnetic field. This effect scales with the applied magnetic field and depends on the
orientation of the paramagnetic defect in the applied magnetic field. The spin–orbit
coupling may be included in the ESR resonance condition by replacing the constant g
0
of
Equation 6.1 with a g matrix with elements g
ij
. The symmetry of this matrix reflects the
symmetry of the paramagnetic center. Under some circumstances, the symmetry of the
matrix alone allows identification of the defect under study. (The P
b
and P
b0
centers were
fairly convincingly identified on the basis of their g matrix [3–5].)
Perturbation theory allows calculation (with modest accuracy) of the g matrix elements
for both P
b
and E
0
centers [1]. The components of the g matrix are given by
g
ij
¼ g
0
d
ij
À2 l
¸
k
ha L
i
j jkihk L
j

ai
(E
k
ÀE
a
)
(6:2)
where
g
0
is the free electron value 2.0023193
l is the atomic spin–orbit coupling constant
L
i
and L
j
are angular momentum operators appropriate for the x-, y-, or z-directions,
and the summation is over all states k
State jai and energy E
a
correspond to the paramagnetic state for which the g matrix is
calculated. Equation 6.2 provides some very useful (qualitative and semiquantitative)
physical insight in understanding the ESR spectra of both P
b
and E
0
centers. As mentioned
previously, the centers involve electrons residing in high p-character orbitals on silicon
atoms. The angular momentum operator corresponding to the symmetry axis direction of
a p-orbital yields a zero when operating on that orbital [44]. Therefore, the g value corres-
ponding to the magnetic field parallel to the dangling bond orbital, g
k
, should be about
g
0
ffi 2.002. The g value corresponding to the magnetic field perpendicular to this axis, g
?
,
should exhibit the largest deviation fromthis value [44]. Furthermore, since both E
0
and P
b
levels appear within a bandgap, one wouldintuitively expect that the denominator of the far
right termof Equation 6.2 would result in larger deviations of g
?
from g
0
for the wide band
gap (E
g
(SiO
2
) ffi9 eV) E
0
centers than the narrower band gap (E
g
(Si) ffi1.1 eV) P
b
centers. This
is also the case. For an axially symmetric defect, we expect the g corresponding to an angle u
between the applied field and the defect symmetry axis to be:
g ¼

g
2
k
cos
2
u þg
2
?
sin
2
u

1=2
(6:3)
For the most general case of lower symmetry
g ¼

g
2
1
cos
2
a þg
2
2
cos
2
b þg
2
3
cos
2
g

1=2
(6:4)
ß 2008 by Taylor & Francis Group, LLC.
where a, b, and g correspond to the field angle with respect to the principal axis directions
of the g matrix.
6.2.2 Electron–Nuclear Hyperfine Interactions
The other important source of deviation from Equation 6.1 is the hyperfine interaction of
the unpaired electron with nearby nuclei [1,45]. Certain nuclei have magnetic moments; in
metal=silicon dioxide=silicon systems, some examples are
29
Si (spin 1=2),
1
H (spin 1=2),
31
P (spin 1=2), and
14
N (spin 1). A spin 1=2 nucleus has two possible orientations in the
large applied field; a spin 1 nucleus has three possible orientations. Each nuclear moment
orientation corresponds to one local nuclear moment field distribution.
One may envision the nuclear moment interacting with an unpaired electron residing in
a wave function that is a linear combination of atomic orbitals (LCAOs). For all varieties of
P
b
and E
0
centers, a combination of s- and p-type wave functions provide a reasonable
approximation to reality. The LCAO for an unpaired electron can therefore be written as
jai ¼
¸
n
a
n
c
s
si þc
p

pi
¸ ¸
(6:5)
where
jsi and jpi represent the appropriate atomic orbitals corresponding to the nth site
a
n
2
represents the localization on the nth site
c
2
s
and c
2
p
represent, respectively, the amount of s- and p-character of the wave
function on the nth atomic site
With possibly one exception, a
2
1
ffi 1 for all of the P
b
and E
0
centers discussed in this chapter;
that is, the defects’ unpaired electron is reasonably well localized at a single nuclear site
(0:6 < a
1
2
< 1). To first order then, we can nearly always interpret the P
b
and E
0
ESR spectra
in terms of s=p hybridized atomic orbitals localized at a central site. We can understand
the electron–nuclear hyperfine interaction of an electron in an sp orbital by considering the
p and s orbital contributions individually and then add the contributions together. The
electron–nuclear interaction of an electron in a p orbital is anisotropic: it corresponds to an
essentially classical magnetic dipole interaction as schematically illustrated in Figure 6.1.
(a) (b)
FIGURE 6.1
Schematic illustration of an electron in a p orbital interacting with a magnetic nucleus (a) for the nuclear moment
parallel to the symmetry axis and (b) perpendicular to the symmetry axis. (Reprinted from Lenahan, P.M. and
Conley, J.F., J. Vac. Sci. Technol. B, 16, 2134, 1998. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
The interaction is strongest when the field is parallel to the symmetry axis. The sign of
the interaction changes and the magnitude is decreased by one-half when the field is perpen-
dicular to the symmetry axis.
When the electron and nuclear moments are aligned by a strong magnetic field in the
z-direction, a reasonable assumption for all of the ESR results discussed in this chapter,
only the z-component of the dipolar field is important, because the interaction energy
involves a dot product (ÀmÁ H). This z-component, g
n
b
n
(1 À3 cos
2
u)r
À3
, is averaged over
the electronic wave function to produce the dipolar contribution
Dipolar contribution ¼ Àg
n
b
n
1 À3 cos
2
u
r
3

(6:6)
where g
n
is the appropriate nuclear g value.
The electron–nuclear hyperfine interaction of an electron in an s orbital is quite different.
It is larger and it is isotropic. This interaction, often called the Fermi contact interaction, is
illustrated in Figure 6.2. Inspection of the figure as well as the appropriate calculations
indicates that the spherical symmetry of the orbital results in zero interaction with this
field, except for a spherical region about the nucleus with a radius of an imaginary current
loop generating the field of the nuclear moment. Since the s orbital has a nonzero prob-
ability density at the nucleus, a large isotropic interaction results. The s orbital hyperfine
interaction can be computed from an elementary electricity and magnetism calculation
[45]: the magnetic field at the center of a current loop of radius a is given by 2m=a
3
, where m
is the magnetic moment of the current loop. Envision a magnetic moment of the nucleus as
if it were due to a current loop about the circumference of the nucleus. The probability
density of the electron varies little over the volume of the nucleus; take it to be constant,
ja(0)j
2
. Considering then only the fraction of the electron wave function at the nucleus, to
be 4=3pa
3
ja(0)j
2
, the interaction would be
A(isotropic) ¼
4
3
pa
3
a(0) j j
2

2m
a
3

(6:7)
The magnetic moment of the nucleus is the nuclear g factor, g
n
, times b
n
. Thus, the isotropic
or Fermi contact interaction is given by
FIGURE 6.2
Schematic diagramof the isotropic interactionof ans
orbital electron with a magnetic nucleus. (Reprinted
from Lenahan, P.M. and Conley, J.F., J. Vac. Sci.
Technol. B, 16, 2134, 1998. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
A
iso
¼
8p
3
g
n
b
n
a 0 ð Þ j j
2
(6:8)
where ja(0)j
2
represents the unpaired electron probability density at the nucleus.
Both isotropic and anisotropic hyperfine interactions are present for nearly all the
paramagnetic centers studied in oxide thin films on silicon. (The sole exception is atomic
hydrogen.) This is so because the unpaired electron wave functions generally involve both
p- and s-orbital characters. The hyperfine interactions, like the spin–orbit interactions, are
expressed in terms of a 3 Â3 matrix. To a pretty good approximation, E
0
and P
b
centers
have axially symmetric wave functions and thus an axially symmetric hyperfine matrix is
appropriate. (The exception to this rule is the P
b1
center, which has somewhat lower than
axial symmetry in its hyperfine matrix.)
With the magnetic field parallel to the p orbital symmetry axis, the anisotropic coupling of
Equation 6.6 yields (4=5)g
n
b
n
hr
À3
i ¼2A
aniso
. The field perpendicular to the symmetry axis
result s in a n int eraction of hal f the magnitu de and opp osite sign À (2 =5) g
n
b
n
hr
À3
i ¼ÀA
aniso
.
This result is int uitively sat isfying and con sistent with the sketch es of Figure 6.1.
The components of the hyperfine matrix correspond to sums of the isotropic and
anisotropic interactions for the applied field parallel and perpendicular to the unpaired
electron’s orbital symmetry axis:
A
k
¼ A
iso
þ2A
aniso
(6:9)
A
?
¼ A
iso
ÀA
aniso
(6:10)
and thus,
A
iso
¼ (A
k
þ2A
?
)=3 (6:11)
A
aniso
¼ (A
k
ÀA
?
)=3 (6:12)
where
A
aniso
¼
2
5
g
n
b
n
hr
À3
i (6:13)
For a paramagnetic center with an axially symmetric wave function, with a specific
orientation (designated by the angle u between the symmetry axis and the applied field
vector) the resonance condition is
H ¼ H
0
þM
1
A (6:14)
where H
0
¼hn=g
e
b
e
, and M
1
is the nuclear spin quantum number
g ¼

g
2
k
cos
2
u þg
2
?
sin
2
u

1=2
(6:15)
and
A ¼

A
2
k
cos
2
u þA
2
?
sin
2
u

1=2
(6:16)
For the most general low symmetry case,
A ¼

A
2
1
cos
2
a þA
2
2
cos
2
b þA
2
3
cos
2
g

1=2
(6:17)
ß 2008 by Taylor & Francis Group, LLC.
where a, b , and g corre spond to the angl es betwe en the magne tic fi eld and the principa l
axe s of the hyper fine matrix.
Equati ons 6.2, 6.3, and 6.9 throu gh 6 .15 provi de a ver y straigh tforward ba sis for
anal yzing ESR resu lts for axiall y symm etric defects with a spe cifi c orientat ion with
respec t to the appl ied mag netic field. The dangling bon d centers at the Si =SiO
2
boundar y
( P
b
and P
b0
) yield spe ctra rea dily des cribed by Equati ons 6.2, 6.3, and 6.9 throu gh 6.17
since the crys tallini ty of the silicon substrat e provid es a fixed relatio nship betwe en the
appl ied field a nd defect or ientation. (The P
b1
spe ctrum is sligh tly more comple x due to its
lower symmet ry; it has three diffe rent value s for bot h the g and hyper fine matrice s.)
On e impo rtant qualitati ve poi nt to be glea ned from the dis cussion of the hyper fine
inter actions is that, for an elect ron in an sp hyb ridized Si dangling bond orbital, the
spl itting due to the hyper fine inter action will be larges t when the magne tic fi eld poi nts
along the directio n of a dangling bond dire ction and smallest when the fi eld is perp en-
dicu lar to the dangling bon d directi on. Similarl y simple fi rst-orde r logic (d iscussed
previ ously) indi cates that the g value correspond ing to the dangling bon d directi on
wou ld be expe cted to be c loser to the free elect ron value (g
0
ffi 2.002) and the g wo uld
be expe cted to deviate the mo st when the field is perp endicu lar to this dire ction.
The desc ription of ESR spe ctra of defec ts within an amorpho us film is mo re com plex
than is the case for defects withi n a cry stalline environ ment. Within the amorp hous matrix,
all defect orientat ions are equally likely and, due to the lack of long range order, slight
differe nces in local defect geome try may be anticipat ed. The pres ence of defects at all
orientat ions leads to the conti nuous distribut ion of bot h g and A value s fro m g
k
and A
k
to
g
?
and A
?
. The diffe rences in local geome try lead to slight defect to defect variati ons in g
k
,
g
?
, A
k
, and A
?
. (Addi tional c omplexit y, in the form of three g and A terms are requi red for
defects wi th lower than axi al symm etry.)
These com plication s are relative ly easy to deal wi th. The random dis tribution of defect
orientat ion can be dealt with easy in terms of analy tical expre ssions foun d in mo st ESR
textbooks [1]. (For axially symmetric centers, the anticipated ESR patterns are intuitively
obvious. Far fewer centers will have the symmetry axis parallel to the applied field than
perpendicular to it; thus, the ESR spectrum intensity will be far stronger at the A
?
and g
?
values than at A
k
and g
k
.) The slight defect-to-defect variations in g and A values lead to
broadening of the line shapes anticipated for unbroadened tensor components. (The
anal ysis is sch ematica lly illustrat ed in Figu re 6.3.)
The evaluation of ESR hyperfine matrix components usually allows for a reliable and
moderately precise identification of the unpaired electron’s wave function. This is the case
for silicon dangling bond centers like P
b
and E
0
defects. For reasonably clean MOS oxides,
we anticipate significant concentrations of only silicon, oxygen, hydrogen, and under
certain circumstances, nitrogen, phosphorous, and boron. The nuclear moments of these
atoms are all quite different [1,45]. Over 99.9% of oxygen atoms have nuclear spin zero;
over 99.9% of phosphorous, nitrogen, boron, and hydrogen nuclei are magnetic. Silicon is
readily distinguishable; 95% of silicon atoms have nuclear spin zero but 5% of the silicons,
those with
29
Si nuclei, have a nuclear spin of one-half. This 95%spin zero=5%spin one-half
ratio is unique among all the elements in the periodic table. Thus, a three-line pattern with
the two side peaks each corresponding to about 2.5% the integrated intensity of the much
more intense center line, can be convincingly linked to an unpaired electron associated
with a silicon atom.
Having identified the nuclear species involved, a first-order analysis of the unpaired
electron wave function is extremely straightforward in terms of the LCAO picture.
For defects in a crystalline environment, Equation 6.16 can fit to the ESR spectrum for
several values of u. For defects in an amorphous (or polycrystalline) environment, one may
fit the appropriately broadened analytical expressions to the ESR spectra to yield A
?
and
ß 2008 by Taylor & Francis Group, LLC.
A
k
, as illustrated in Figure 6.3. Using Equations 6.9 through 6.13, one then obtains the
isotropic and anisotropic coupling constants A
iso
and A
aniso
. Tabulated values [1] of A
iso
, a
0
,
and A
aniso
, b
0
, calculated for 100% occupation probability can then be utilized to determine
the hybridization and localization of the electronic wave functions. Although the crude
analysis just discussed is not extremely precise it is, to first order, quite reliable for the
relatively simple P
b
and E
0
defects. One should realize that the isolated atomic values
obtained for a
0
and b
0
are themselves only moderately accurate and that placing a silicon
atom in an oxide matrix will inevitably alter the constants somewhat. Nevertheless, a
straightforward analysis of hyperfine parameters provides moderately accurate measure-
ment of hybridization and localization.
6.2.3 Quantitative Analysis (Spin Counting)
In conventional ESR detection, resonance is observed through changes in the quality factor
of a microwave cavity induced by the resonance [1]. Conventional ESR sensitivity is about
10
10
spins. (The precise number depends upon A and g and the coupling of the spins to the
surrounding lattice.) Conventional ESR detection provides a count of paramagnetic defects
g
^
A
^
g
^
(a)
(b)
A
II
g
II
g
II
A
^
A
II
g
II
g
^
FIGURE 6.3
Schematic sketches of ESR absorption (a) and derivative (b) spectra for centers with axial symmetry in an
amorphous matrix. The g
?
, g
k
, A
?
, and A
k
values correspond to the resonance condition when the magnetic
field is parallel (k) or perpendicular (?) to the defect symmetry axes. The sketches in (a) correspond to the
single line anticipated for defects that are not associated with a magnetic nucleus. The sketches in (b) correspond to
the two lines anticipated for defects associated with a spin 1=2 nucleus. (The spin 1=2 nucleus has two possible
orientations in the large applied field thus providing two different hyperfine contributions to the resonance
condition.)
ß 2008 by Taylor & Francis Group, LLC.
within a sample to a precision of about Æ10% in relative number and an absolute precision
of a little better than a factor of 2.
6.2.4 Spin-Dependent Recombination
In semiconductor device studies, an approach very different fromconventional ESR detec-
tion is quite powerful: spin-dependent recombination (SDR) [20,22,23,25,28,29,46,47]. SDR
is an ESR technique in which the spin-dependent nature of recombination events is
exploited so that one may observe ESR through measurement of recombination currents.
A qualitative explanation of SDR, first proposed by Lepine [46], will be sufficient for the
purposes of this chapter. The Lepine model combines the Shockley–Read–Hall (SRH)
[48,49] recombination model and the Pauli exclusion principle.
In the SRH model [48,49], recombination of electrons and holes takes place through
deep level defects in the semiconductor band gap. An electron is trapped at a deep level
defect; then a hole is trapped at the same deep level defect. (The sequence could
obviously be reversed, with a hole being trapped first at the deep level, and then an
electron.) Suppose that the deep level defect is a silicon dangling bond. The dangling
bond has an unpaired electron when it is electrically neutral. In SDR, as in conventional
ESR, the semiconductor device is placed in a strong magnetic field that polarizes the spins
of the dangling bond electrons as well as the conduction electrons and the holes. (That is,
the electron spins tend to orient with the applied magnetic field.) The SRH recombination
process begins with electron capture (or hole capture) at the deep level site. If both the
dangling bond and conduction electrons have the same spin orientation, the conduction
electron cannot be captured at the site because two electrons cannot occupy the same
orbital with the same spin quantum number. Thus, placing the semiconductor sample in
the magnetic field reduces the average trap capture cross section of the traps. In ESR,
electron spins are flipped from one spin orientation to the other when the resonance
condition is satisfied. The spin flipping at the trap site increases probability of oppositely
orientedtraps andconduction electron spins. This process increases the average trap capture
cross section. Thus, if a semiconductor device is placed in a strong slowly varying magnetic
field and is simultaneously subjected to microwave irradiation at frequency n, the recom-
bination current will increase when n and H satisfy the resonance condition. The spin
flipping event induced by ESR increases the possibility of oppositely oriented conduction
electron and deep level defect spins, increasing the likelihood of recombination events. (A
more sophisticatedapproach to SDRhas been developed by Kaplan, Solomon, and Mott [47]
(KSM). The KSM model is not precisely applicable to SDR of either E
0
or P
b
centers.
However, it more nearly approximates reality than the Lepine model. Adetailed discussion
of SDR models and the relevance to MOS defects has been published elsewhere [50].)
The observation of SDR thus, essentially by definition, demonstrates that the defect in
question has levels deep in the band gap. Furthermore, with specific knowledge of the
device in question and the various voltages and currents involved, one may generally,
though only very crudely, identify the physical location and energy levels of the defect
centers observed in SDR.
6.3 Defect Structure
6.3.1 P
b
Center Structure
The chemical and structural nature of P
b
centers has been established by several independent,
consistent, and mutually corroborating studies. Three P
b
variants have been consistently
ß 2008 by Taylor & Francis Group, LLC.
observe d: at (111) Si =SiO
2
inter faces a defe ct called simply P
b
[2 –13,22], and at (100) Si=SiO
2
inter faces, two defects called P
b0
and P
b1
[5,14,19 ,22, 25–31]. The structur e of both the (111)
Si=SiO
2
P
b
and the (100) Si=SiO
2
P
b0
are reasona bly well unde rstood; a more rudime ntary
understan ding of P
b1
exists at this time.
P
b
cen ters we re fi rst observe d by Nishi and cowo rkers [2,3]. Their work focuse d primar -
ily on the (111) Si =SiO
2
system. They showe d that the P
b
centers were at or very nea r to the
Si=SiO
2
inter face and that they possess an axi ally symm etric g matrix with g
k
ffi 2.000 and
g
?
ffi 2. 01 with the symm etry axi s cor respond ing to the (111) directio n. Figu re 6 .4 shows
repres entative P
b
spe ctra taken with the fi eld perpen dicular and parallel to the (111)
surfa ce normal . On the basis of this inf ormatio n, Nishi et al. conclud ed that P
b
centers
are trivale nt silicon center s at or very near the Si =Si O
2
boundar y. The structur e of the (111)
P
b
center is illustrat ed in Figure 6.5.
Later , Cap lan et al. [4,5] obtain ed mo re precis e g matrix parame ters. For the (111) Si =SiO
2
P
b
, they foun d g
k
ffi 2.0014 a nd g
?
ffi 2.0081 with, as Nishi had obse rved previ ously, the
symm etry axi s corre spondin g to the Si (111) dire ction. For the (100) Si =SiO
2
system, the y
foun d two P
b
variants; P
b0
and P
b1
[4,5]. Al though hamper ed by overlap ping spectra,
Capla n et al. were ab le to show that the P
b0
g tensor was quite similar to that of the (111)
Si=SiO
2
P
b
; they foun d g
1
ffi g
k
¼ 2.0015 and g
?
ffi g
2
ffi g
3
, with g
2
¼ 2 .0087 and g
3
¼ 2.0080.
Later , more precis e measure ments by Kim and Lenah an demons trated that the P
b
g matrix,
like that of the (111) P
b
cen ter, is also axi ally symmet ric, with g
k
¼ 2.0015 Æ 0.0002 and
g
?
¼ 2.0080 Æ 0.0002 [19]. A g-map from that work is shown in Figure 6.6. [19] (The
symmetry axes for the P
b0
centers are also [111] directions.) This result was later confirmed
by Stesmans and Afanas’ev who, within experimental error, found the same values:
g
k
¼2.0018 Æ0.0003 and g
?
¼2.0085 Æ0.0003 [26]. These studies strongly indicate that the
(111) P
b
and the (100) P
b0
are essentially identical. (As discussed later in this chapter several
5 G
g =2.0014
g =2.0085
(a)
(b)
FIGURE 6.4
ESR traces of the P
b
center for the
magnetic field (a) parallel and (b) per-
pendicular to the (111) symmetry axis.
(Reprinted from Lenahan, P.M. and
Conley, J.F., J. Vac. Sci. Technol. B, 16,
2134, 1998. With permission.)
O O
Si Si
Si
Si
Si Si
Si
Si
Si Si
FIGURE 6.5
Schematic illustration of a P
b
center at
the (111) Si=SiO
2
interface.
ß 2008 by Taylor & Francis Group, LLC.
other observations also support this conclusion.) The structure of the P
b0
is illustrated in
Figure 6.7.
Poindexter et al. found that the P
b1
center exhibited lower symmetry, with g
1
ffi 2.0076,
g
2
ffi 2.0052, and g
3
ffi 2.0012 [5]. Poindexter et al. reported that the axes corresponding to g
1
are approximately [111] directions, g
2
corresponds to a direction between the [100] and
[111] directions, and g
3
approximately corresponds to [011]. Later, more precise measure-
ments by Stesmans and Afanas’ev refined the original Poindexter et al. results yielding
g
1
¼2.0058, g
2
¼2.0074, and g
3
¼2.0022 [26]. Stesmans et al. found g
1
to correspond to
[0

1

1], g
2
to near [1

1

1], and g
3
to approximately [211].
The close correspondence between the P
b0
and P
b
g matrix components led Poindexter
et al. to propose a near equivalence between the (111) P
b
and the (100) P
b0
, a conclusion that
is clearly correct. The fact that the centers’ matrices have (111) axial symmetry strongly
1.998
2.000
2.002
2.004
2.006
2.008
2.010
0 20 40 60 80 100
g

v
a
l
u
e
f(Њ)
FIGURE 6.6
ESR g value map for P
b0
centers on (100) silicon substrates. The data points correspond to various angles w with
respect to the (100) interface normal. The sample is rotated about the [011] direction. The solid lines are given by the
relation g ¼ g
2
k
cos
2
u þg
2
?
sin
2
u

1=2
with g
k
¼ 2.0015 and g
?
¼ 2.0080, where u is the angle between the magnetic
field and the [111] directions. Note the two different angles involved. The angle u refers to the angle between the
multiple <111>directions inwhichthe P
b0
centers canpoint andthe magnetic field. The angle wis the angle between
the (100) interface normal and the magnetic field. (Reprinted fromKim, Y.Y. and Lenahan, P.M., J. Appl. Phys., 64,
3551, 1988. With permission.)
FIGURE 6.7
Schematic illustration of a P
b0
center at the (100) Si=SiO
2
interface.
O
O
Si
Si
Si
Si
Si
Si
Si
Si
O
O
O
Si
Si
Si
ß 2008 by Taylor & Francis Group, LLC.
supports the idea that they are interfacial silicon dangling bonds. The differences in the
g matrix of P
b1
with respect to P
b0
and P
b
led Poindexter et al. to propose that the defects
have different structures. Although the specific P
b1
model proposed by Poindexter et al.,
a silicon bonded to two silicons and an oxygen [5] is almost certainly incorrect, their
conclusions that the P
b1
is a somewhat different defect is certainly the case.
Any doubt about the basic structure of the (111) P
b
and (100) P
b0
centers has been
resolved by measurements of the hyperfine interactions with
29
Si nuclei. Hyperfine results
have also helped to refine our understanding of P
b1
structure. Representative hyperfine
results for both P
b0
and P
b1
are shown in Figure 6.8. The (111) P
b
hyperfine matrix was first
measured by Brower [9], who found A
k
ffi 152 G and A
?
ffi 89 G. Using hyperfine
parameter calculations tabulated by Weil et al. [1] indicating that ja
0
j ¼1693.3 G and
jb
0
j ¼40.75 G, Equations 6.7 and 6.8 indicate that 12% s-character, 88% p-character, and
58% localization. Brower’s (111) results were confirmed by Jupina and Lenahan [22]. The
(100) P
b0
hyperfine tensor was first measured by Gabrys et al. [25] and more recently by
Stesmans and Afanas’ev [27]. The P
b0
hyperfine parameters reported by Gabrys et al. are
A
?
¼72 Æ3 Gand A
k
¼144 Æ3 G. Stesmans et al. reported A
?
¼75 Æ5 Gand A
k
¼149 Æ4 G.
Utilizing hyperfine parameter calculations tabulated by Weil et al., these parameters
indicate (both studies) 9% s-character, 91% p-character, and (Gabrys) 65% localization or
(Stesmans) 67% localization on the central silicon. So, both studies quite clearly indicate
that the P
b0
, like the (111) P
b
, involves a silicon dangling bond in which the wave function
has very high p-character and is primarily localized on the central silicon. It should be
mentioned that the 5%=95% ratio of side (
29
Si) peaks to center (
28
Si) peaks was reported in
all of these studies. The ffi 5%=95% ratio of hyperfine lines to center line intensity for both
(111) P
b
and (100) P
b0
unequivocally establishes that both centers are silicon dangling
bonds. This is so because, among all possible elements within the system, only silicon
exhibits a 4.7% abundant magnetic nucleus with a nuclear spin of ½. The two possible
orientations of the magnetic silicon nuclei produce two weak side lines with a total
3410
3440
3470
3500
0 30 60 90
Angle q (Њ)
M
a
g
n
e
t
i
c

f
i
e
l
d

(
G
)
FIGURE 6.8
Plot of both P
b0
(open data points) and P
b1
(solid data points) hyperfine lines versus magnetic field orientation.
The sample was rotated in the (110) plane. Angle u is the angle between the [001] direction of the sample and the
applied magnetic field. Note that multiple lines appear because multiple orientations for both P
b0
and P
b1
dangling bonds occur on the (100) interface, the (111) and approximately (211) directions, respectively. (Reprinted
from Mishima, T.D., Lenahan, P.M., and Weber, W., Appl. Phys. Lett., 78, 1453, 2001. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
integr ated intensity corre spondin g to the relative abunda nce of the magne tic nuclei, that is
4.67% , or ab out 5%. The (111) P
b
and (100) P
b0
have very similar hyper fi ne matrice s
that demonstr ate , as had bee n indicated by ear lier g tensor result s, that P
b
and P
b0
are
essen tially identical defects . Of cours e quite subtl e differe nces inevitabl y exist; the (111) P
b
cen ter ’s unpaire d elect ron is in an or bital directed along the (111) surfa ce normal. Th e (100)
P
b0
is also directe d along a (111) dire ction but thi s directi on is obvi ously no t no rmal to
the (100) surfa ce. The ver y high p-charac ter would tend to flatten the bon ding arran gement
with the three ba ck-bonded silico ns. It is easy to unders tand why this sh ould be so. If the
dang ling bon d wave functi on were pure ly p-c haracter , all the s-chara cter wou ld
be ab sorbed by the back bonding, making the back bonding pure planar sp
2
lik e graphit e.
If the dan gling bond wave function were sp
3
, the back bon ding wou ld be essentia lly
tetrahe dral.
The detai led struc ture of the P
b1
center is not so well understoo d; alth ough it to o is
clearl y a silicon dangling bond at the int erface, in which the cen tral silicon ato m is back
bon ded to three othe r silico ns. The zero- order P
b1
bon ding a rrangement was establish ed in
a study by Browe r [17]. Brow er observe d P
b1
(as we ll as P
b0
and P
b
) in oxides grown on
bot h (111) and (100) silicon substrat es in an
17
O enriched atmos phere. Since
17
O pos sesses a
nuc lear mag netic mo ment [1,17], if the P
b
cen ter silicon were bon ded to an oxygen , the
17
O
nuc lear momen t wou ld grea tly broad en the spe ctrum. No ne of the three P
b
variants are
bro adened enough to indicate nearest neighbor oxygen s. Ho wever, all are sligh tly broa-
dened , indicating that the y are all at the interface . Si nce nitrog en and hydrogen a lso
poss ess magne tic momen ts, we kno w that the P
b0
and P
b1
silico ns could not be bon ded
to them either. Since no other atoms are cons istently presen t in num bers suf ficient to
acco unt for the P
b
cen ters we may conclude that all three vari eties of P
b
centers are silicons
back bon ded to silicons at the respectiv e Si =Si O
2
boundar ies.
The P
b1
centers hy per fine inter actions wi th cen tral
29
Si nuclei we re first reporte d by
Stes mans and Afanas ’ev [27] and later by Mishima a nd Lenah an [28] and Mishi ma et al.
[29]. The Stes mans et al. resu lts and Mishi ma et al. result s differ, but only slightl y beyond
the stated error margins of each study. Stesmans et al. repo rted thre e hyper fi ne par ameter s:
A
xx
[1

10], A
yy
[

1

11], and A
zz
[112] to be resp ectively 102, 112, and 167 G. Mishi ma et a l.
reporte d the three hyper fine para meters: A
xx
[1

10], A
yy
[

1

11], and A
zz
[112] to be resp ect-
ively 90, 115, a nd 167 G. In both studi es, the aut hors estima ted that the par ameter s coul d
be in error by up to Æ 3 G, so only the A
xx
parame ters differ by m ore than the state d
expe rimental error, and on ly sligh tly beyo nd those error estimate s. Again, util izing the
hyper fi ne parame ter calcul ations tab ulated by Weil et al. [1] these par ameter s indicate 13%
s-chara cter (Mi shima ) or 14 % s-chara cter (Stes mans) and 87% p-c haract er (Mishima ) or
85% p-charac ter (Stesman s) and 6 0% local ization on the central silico n (Mi shima ) or 57%
local ization on the central silicon (Stes mans). In view of the previo usly discusse d lim its in
the preci sion of this analysis , the conclus ions of the two studi es are essen tially undi stin-
guish able. The P
b1
is thu s also a silicon dangli ng bond interface trap wi th a high
p-c haracter wave function. Unlike the P
b0
, its dangling bond points (approximately)
along a [112] direction, not a [111] direction. The P
b1
center wave function is slightly
more delocalized than that of the P
b0
wave function. It also has somewhat higher
s-character. The paramagnetic (ESR active) status of these defects corresponds to the
electrically neutral case for each center. Since the s-wave function is lower in energy than
the p-wave function, the hyperfine results suggest that the neutral P
b1
level would tend to
be lower in the silicon band gap than the P
b0
levels. Since the P
b1
wave function is more
delocalized than the P
b0
wave function, one might expect the electron correlation energy of
the P
b1
centers to be smaller than that of the P
b
centers. (As discussed later in the chapter,
both of these intuitive arguments turn out to be correct.) The spectroscopic parameters of
the P
b
cen ters are summa rized in Table s 6.1 and 6.2.
ß 2008 by Taylor & Francis Group, LLC.
6.3.2 E
0
Center Struct ure
6.3.2. 1 Structur e of Conven tional E
0
Centers
The most impo rtant oxide trapping cen ters are E
0
cen ters [10,11,1 8–22,38 ,39], whi ch
involve an unpaire d elect ron local ized on a silicon back bonded to three oxygen s [34].
Often , thou gh not a lways, the param agnetic silico n site is coupled to a positive ly charge d
diamagn etic silico n [10,11,1 8–21,35] as shown in Figu re 6.9. A single dang ling bond E
0
is
shown in Figure 6.10. As discusse d later in this chap ter, the cartoon figures of Figure s 6.9
and 6.10 are at least rea sona ble fi rst-orde r approxi matio ns, but m ay no t precis ely des cribe
the mo st importan t E
0
cen ters in MOS oxide s. Th e ear liest st udies of E
0
centers involve large
volume (approx imat ely cubic cen timeter) sample s of both crystal line quar tz and amo rph-
ous SiO
2
[33 –37]. The mo st commo nly repo rted E
0
line shape foun d in the large volume
amorp hous SiO
2
sam ples as well as the mo st commo nly obse rved E
0
line shapes in thin
films closely rese mbles that of a randoml y oriented array of crystalline quartz E
0
1
centers.
This line sh ape is often referred to as E
0
g
. Th e crystal line quar tz E
0
1
center has a g matrix that
is almos t axi ally symmet ric g
1
ffi g
k
¼ 2.00176, g
2
¼ 2.00049, g
3
¼ 2.00029; g
2
ffi g
3
ffi g
?
. The
amorp hous Si O
2
E
0
line sh ape usua lly obse rved to domi nate c orrespond s to a randoml y
TABLE 6.1
29
Si Hyperfi ne Parameters for P
b
and E
0
Centers
(111) P
b
A
?
(G) A
k
(G) % s % p % Loc.
Brower [9] 89 Æ 3 152 Æ 3 12 88 58
(100) P
b0
A
?
(G) A
k
(G) % s % p % Loc.
Gabrys et al. [25] 72 Æ3 144 Æ3 9 91 65
Stesmans and Afanas’ev [27] 75 Æ5 149 Æ4 9 91 67
Mishima and Lenahan [28] and
Mishima et al. [29]
72 Æ3 144 Æ3 9 91 65
(100) P
b1
A
xx
[1

10] A
yy
[

1

11] A
zz
[112] % s % p % Loc.
Stesmans and Afanas’ev [27] 102 Æ3 112 Æ3 167 Æ3 14 86 57
Mishima and Lenahan [28] and Mishima
et al. [29]
90 Æ3 115 Æ3 167 Æ3 13 87 60
E
0
A
?
(G) A
k
(G) % s % p % Loc.
Figure 3 (crudely) 417 483 32 68 80
Griscom et al. [36] (more precisely) 398 464 31 68 79
TABLE 6.2
g Matrix Parameters for P
b
and E
0
Centers
(111) P
b
g
k
(111) g
k
Caplan et al. [4] and Poindexter et al. [5] 2.0014 2.0081
(100) P
b
g
k
(111) g
?
Kim and Lenahan [19] 2.0015 Æ0.0002 2.0080 Æ0.0002
Stesmans and Afanas’ev [26] 2.0018 Æ0.0003 2.0085 Æ0.0003
(100) P
b1
g
xx
[1

10] g
yy
[

1

11] g
zz
[112]
Stesmans and Afanas’ev [26] 2.0058 2.0074 2.0022
E
0
(amorphous SiO
2
) ffi g
k
ffi g
?
Griscom et al. [36] 2.0018 2.0004
ß 2008 by Taylor & Francis Group, LLC.
oriente d array of defe cts with g
k
ffi 2.0018 and g
?
ffi 2.0004. The g
k
ffi 2.0018 is very close to
the free el ectron g
0
¼ 2 .002319. It is worth no ting that this sort of g matrix is con sistent with
the pred ictions of Equ ation 6.2 fo r a dan gling bond defect. An E
0
cen ter line spe ctrum,
measu red on an MOS oxide is illustrat ed in Figure 6.11.
In large volume sam ples, it is quite easy to measu re the hyper fine inter actions of the
unpai red electron wi th the single silicon ato m on which it primar ily resi des [34,36,3 7].
Althou gh the magne tic
29
Si nuclei are only ffi5% abun dant, the number of centers pres ent
in large volume sam ples (typ ically $ 10
16
–10
17
=cm
3
for irr adiated SiO
2
) is more than
suf ficient to gene rate quite strong
29
Si spectra. An ESR spe ctrum taken on such a large
volu me amo rphous sample is sh own in Figure 6.1 2. As dis cussed previousl y one may, by
inspec tion, obt ain a ro ugh estimate of the hybridizat ion and localiza tion of the el ectron
from this spectru m, as schem atically illus trated in Figu re 6.3. Al so, as discusse d previ-
ously, a second integration of the two side peaks (corres ponding to spi n 1 =2 nuc lei) yie lds a
com bined intensity of about 5% of the center line, unamb iguousl y identifyin g the cen ter as
an unpaired spin localized on a silicon atom. An application of the analysis discussed
previously indicates that A
iso
ffi 439 G and A
aniso
ffi 22 G. Recall that the isotropic and
anisotropic coupling constants for an electron 100% localization in a silicon s and p orbital
FIGURE 6.9
Schematic diagram of a positively
charged E
0
center, a hole trapped at an
oxygen vacancy.
Si
Si
O
O
Si
O
Si
Si
O
Si
O
Si
O
Si
+
FIGURE 6.10
Schematic illustration of a neutral E
0
center.
Si
Si
O
O
Si
O
Si
ß 2008 by Taylor & Francis Group, LLC.
are, respec tively , a
0
¼ 16 39.3 G and b
0
¼ 40.75 G. So, if the elect ron were 100% local ized in a
silicon s orbita l one wou ld expe ct an isotrop ic coupling cons tant of a
0
¼ 1639.3 G. Figu re 6.4
indicate s an isotrop ic spl itting of about 439 G; thus, the orbital has ab out 439 =1639 ffi 27% s.
If the electron were 100% localized in a silicon p orbital we would expect A
aniso
¼ b
0
¼ 40.75 G.
Figure 6.4 indic ates ab out 22 G; thus, the orb ital has about 22 =40.75 ffi 54% p. The analy sis
indicate s a localiza tion on the cen ter silicon (54 þ 27) ¼ 81%. Again, this anal ysis is cru de,
but, to first order, qui te reliable. The rea sonablene ss of this estimate is support ed by
compar ison wi th a repre sentativ e calculati on of the wave fun ction anticipat ed for these
E
0
center s that esti mates 32% s-chara cter, 52% p-charac ter, and 84% localiza tion [51].
It has previ ously been men tioned that E
0
cen ters are quite freque ntly posit ively charged
oxyg en vacancie s [10,11,1 8,19,21,3 8,39]. The conne ction betwe en a positivel y charged
oxyg en va cancy and the just compl eted anal ysis is not obvi ous. (In fact, the E
0
spec trum
is not alway s associ ated with the posit ively charge d va cancy [40,43,5 2,53].) Howeve r, a
very plausible explanation for a link between the E
0
spectrum and the positively charged
vacancy was provided before direct experimental evidence linking the center to trapped
holes was available.
1 G
g =2.0005
g ≈2.002
FIGURE 6.11
An 8 G wide ESR scan of a VUV irradiated MOS oxide,
with the spectrometer settings optimized for the E
0
cen-
ter signal. (Reprinted fromLenahan, P.M. and Mele, J.J.,
J. Vac. Sci. Technol. B, 18, 2169, 2000. With permission.)
1000 X 1000 X
60 G
1 X
FIGURE 6.12
ESR trace of the E
0
center in a large volume amorphous SiO
2
sample showing a very narrow center line trace
corresponding to the ffi95% abundant spin zero
28
Si nuclei and two broad lines corresponding to the ffi5%
abundant spin 1=2
29
Si nuclei.
ß 2008 by Taylor & Francis Group, LLC.
In crys talline SiO
2
, Fei gl et al. [35] argu ed that E
0
cen ters are holes trapped in oxyg en
vacanci es. In the Feigl et al. mo del, the unpai red elect ron resides on a neutr al silicon on on e
side of the vacancy. Th e silico n on the other side of the vacancy is pos itively charge d.
Bonde d to jus t three ot her atoms, the positive -side silicon adjusts its pos ition to a flat-
planar arran gement (expected for sp
2
bonding) with its three neighbo ring oxygen s. In
amo rphous SiO
2
, E
0
centers can be positivel y charge d or neutr al when par amagneti c. Th e
Feig l et al. hypot hesi s expl ains how an ESR pattern consiste nt with an el ectron local ized on
a single silicon could cor respond to a posit ively charge d oxyg en vacancy, which wo uld
obvio usly involve two silicons . The elect ron spend s all of its tim e on one of them and is
physi cally separate d from the other due to the relaxation involve d at the posit ively charge d
side of the vacancy. Quite a num ber of studi es indi cate that E
0
cen ters are domi nating
trapped hole centers in high-quality thermally grown oxide films [10,11,18–21,24,38,39].
These E
0
centers are very likely, for the most part, essentially the positively charged Feigl,
Fowle r, and Yip E
0
defects shown in Figure 6.9. It sh ould probabl y be men tioned that
models slightly different from a simple oxygen vacancy have been proposed for the
positively charged E
0
center in amorphous SiO
2
. An example is an E
0
model involving a
neutral silicon dangling bond in which the silicon is back bonded to three oxygens near but
not oriented opposite to a positively charged diamagnetic silicon [54,55].
6.3.2.2 Structure of Hydrogen Complexed E
0
Centers
Two hydrogen complexed E
0
centers have been observed in thin oxide films [18,39,56–58].
They are known as the 74 G doublet and the 10.4 G doublet. Spectra of both hydrogen
complexed E
0
centers are illustrated in Figure 6.13. The term doublet refers to the fact that
both centers have two-line ESR spectra. The two lines arise because the hydrogen nucleus
has a spin of 1=2, and thus can generate two local fields at the defect site [1]. One opposes
FIGURE 6.13
ESR traces of (a) the 74 G doublet and (b) the
10.4 G doublet centers. (Reprinted from
Lenahan, P.M. and Conley, J.F., J. Vac. Sci.
Technol. B, 16, 2134, 1998. With permission.)
10.4 G doublet
74 G doublet
10 G
g =2.0082 (P
b
)
g =2.0005 (EЈ)
g =2.0005 (EЈ)
15 X
5 G
15 X
1X
ß 2008 by Taylor & Francis Group, LLC.
the large applied field; the other adds to it [18,39,56–58]. These hydrogen coupled E
0
centers
have also been observed in large volume ($1 cm
3
) SiO
2
samples.
The 74 G doublet spectrum was conclusively linked to an E
0
=hydrogen complex by
Vitko, who studied large volume ($1 cm
3
) samples [59]. Vitko convincingly established
this link by comparing heavily irradiated SiO
2
samples containing hydrogen and deuter-
ium. The chemistry of ordinary hydrogen and deuterium are virtually identical; however,
the one proton hydrogen nucleus and the proton and neutron deuterium nucleus have
very different magnetic properties. Deuterium has a nuclear spin of 1; hydrogen has a
nuclear spin of 1=2. The deuterium nuclear moment is much smaller than the magnetic
moment of the hydrogen nucleus [1]. Vitko demonstrated that the deuterated SiO
2
samples
exhibited an ESR spectrumprecisely identical to what would be expected with a deuterium
substituting for hydrogen. The reduction in the deuterated sample line spacing, 22.9 versus
74 G, scales perfectly with the anticipated difference due to the different moments of the
two nuclei.
A later study, also involving large volume SiO
2
, by Tsai and Griscom [60] extended and
confirmed the Vitko results. Tsai and Griscom were able to observe the
29
Si hyperfine
interactions of the central silicon of the hydrogen complexed defect. The results of Tsai and
Griscomand Vitko indicate that the 74 G doublet center is a silicon dangling bond defect in
which the central silicon is back bonded to two oxygens and a hydrogen. (A more detailed
description of the defect became available with its observation in Si=SiO
2
systems. This will
be discussed later in the chapter.)
A second hydrogen complexed center, called the 10.4 G doublet, has also been observed
in both large volume samples [60] and in thin films [56,57]. The much smaller splitting
(10.4 vs. 74 G) strongly suggests that the coupled hydrogen is more distant in the 10.4 G
doublet case. Tsai and Griscom [60] have proposed that this center involves a silicon
dangling bond center in which the central paramagnetic silicon is back bonded to three
oxygen; one of the three oxygens is bonded to a hydrogen atom.
6.3.2.3 Structure of E
0
d
(or EP) Center
The E
0
d
(or EP) variant of the E
0
center has been observed in separation by implanted
oxygen (SIMOX) buried oxides, bond and etch back oxides, and thermally grown oxides
[52,61–67]. The spectrum is quite narrow and is centered on g ffi 2.002. In this film, it is
nearly always superimposed on a much stronger conventional E
0
spectrum [61–64]. For
example, Vanheusden and Stesmans [64] estimate that it accounts for about 20% of the E
0
centers in SIMOX buried oxides. It is also much less stable than the conventional E
0
centers;
its intensity drops to a small fraction of its initial amplitude in a time of order 1 day at room
temperature [62,63].
A nearly identical E
0
d
(or EP) spectrum has been observed in large volume SiO
2
samples.
Griscom and Friebele observed an ESR spectrum very similar to the oxide film E
0
d
=EP
pattern in large volume SiO
2
samples [68]. Their initial model was of an unpaired electron
delocalized on a cavity of Cl capped Si atoms [68]. Thin films studies by Vanheusden and
Stesmans [64] as well as Conley et al. [52] effectively rule out the role of chlorine.
Vanheusden and Stesmans [64] as well as Warren et al. [65] proposed that the center
involved a microcluster of four to five silicon atoms but did not offer specific spectroscopic
evidence for this assignment.
On the basis of a close similarity between the conventional E
0
and E
0
d
=EP response to
molecular hydrogen, as well as on the E
0
d
=EP signal response to microwave power, Conley
and Lenahan [66] proposed that the defect must be simpler and more nearly related to the
conventional E
0
centers involving an oxygen-vacancy-like site. ESR measurements by
Zhang and Leisure [69] on large volume samples are sensitive enough to detect the
ß 2008 by Taylor & Francis Group, LLC.
E
0
d
=EP
29
Si hyper fine inter action s. These hyper fi ne measure ments indicate d an interacti on
with two silicon ato ms. Th e obse rvatio ns of Zhang and Leisure [69] as well as Conl ey and
Lenah an [66] are generall y cons istent wi th a defe ct mo del propo sed by Chav ez et al. [70],
which involve s a hole trapped in an unrelaxe d oxygen vacancy . They provide additio nal
sup port for their model by demonstr ating that the model is con sistent with what, at the
tim e, was the best
29
Si hyper fine data on the defe ct. This set of observat ions and c alcula-
tions wo uld see m to have provi ded a reasona bly convin cing identi fi cation of the defect.
Howe ver, recent work by Buscari no et al. [71] on large volu me sample s sugge sts that this
may no t be the case. They find, in contradic tion with the earlier Zhang and Lei sure study
[69], that the E
0
d
unpai red elect ron is in fact sh ared by four electrons . They draw this
conclu sion from a study of E
0
d
29
Si hyper fine inter actions. Th e Buscarino et al. [71] work
sup ports the proposa ls of Vanhe usden and Stesmans [64] and Warre n et al. [65] linki ng the
E
0
d
spe ctrum to a silicon m icroclust er. (T his cluste r is ess entiall y four coup led silicon
dang ling bonds.) Althou gh the prepo ndera nce of evid ence seems to favor the E
0
d
=EP
mo del of a hole trapped in an unr elaxed oxyg en vacancy, the micr ocluster mo dels cannot
be ruled out.
6.3.2. 4 Structur e of Other Less Well -Charac terized E
0
Variant s Found in
Thin Oxide Films
In additio n to the conve ntional E
0
spectru m and the E
0
d
=EP spe ctrum, seve ral othe r less well-
charac terized ESR spe ctra have bee n provisiona lly linke d to the E
0
family of defe cts. (That is,
the spe ctra have bee n provi sionally linked to defe cts involvi ng unpai red elect rons primar ily
occ upying orbita ls of silicon ato ms ba ck bon ded to oxyg ens.) In thin fi lms, these defects
have been provi sionally name d EX, EH, and EP2 [52,72 –74]. Th ey all exhibit g matrice s with
com ponents betwe en 2.000 and 2.005. Griscom and Frie bele [68,75] proposed a nomen clat-
ure for E
0
variants that they had obse rved in large volu me sample s. They called these center s
E
0
a
, E
0
b
, E
0
d
and assigne d the term E
0
g
to the wi dely observe d spe ctrum thi s chap ter ref ers to
as the conve ntional E
0
spe ctrum [74,75] . (This notati on is quite confusing in that the E
0
g
line
shape in thin films can be asso ciated with at le ast two very differe nt struc tures, one
posit ively charged and on e neutr al [10,11,1 9,31,38 ,40,41,43 ]. It is for this rea son that, in
this chapte r, the li ne shape freq uently associ ated wi th E
0
g
and usu ally assume d to corre s-
pond to a hole trapped in an oxyg en vacancy is simply calle d a conve ntion al E
0
c enter.)
The assi gnments of the se less well-ch aracteri zed spectra to E
0
like defects is likel y
rea sonable. Mo st or all of these defects are hole trapping center s. Althou gh it is useful to
be awar e of the existence of these defects , currentl y, little more can be said than that they
are probably oxygen vacancy-related oxide defects.
6.4 Electronic Properties of P
b
and E
0
Centers
6.4.1 P
b0
and P
b1
Centers and the Density of States of Si=SiO
2
Interface Traps
The (100) P
b0
center and the nearly identical (111) P
b
center each have two levels in the
silicon band gap, with an electron correlation energy of about 0.7 eV [7,8,10,11,14,19]. This
was established in a series of experiments in which ESR measurements were made on
Si=SiO
2
structures as a function of gate bias and thus of Si=SiO
2
interface Fermi energy.
From capacitance versus voltage (CV) measurements, the gate bias can be converted to an
interface Fermi energy [7,8,10,11,14,19]. A representative example of such results is shown
in Figu re 6.14.
ß 2008 by Taylor & Francis Group, LLC.
As Figu re 6.14 indicate s, the P
b
(and P
b0
) cen ters are diamag netic and posit ively charge d
when the Ferm i level is nea r the valence band edg e. The defects accept one elect ron,
becomi ng elect rically neutr al and paramagne tic (ESR active) as the Ferm i level mo ves
from about 0.35 eV abov e the v alence band and approach es midgap . The defe cts accept
a second electron when the Ferm i level moves from mid-ga p to ward the cond uction ba nd
edge . The ver y bro ad par amagne tic defe ct populati on amp litude versus bias resp onse
indicate s that the P
b0
defe ct levels a re broad ly distri buted with the þ=0 level ffi 1=3 eV
below midgap and the 0 = – level ffi 1=3 eV abov e midgap .
Some time ago, Ger ardi et al. [14] repo rted a P
b1
par amagne tism v ersus bias resp onse
qualit atively similar to the results illustrat ed in Figure 6.14; ho wever, the y reporte d a
muc h narrower distribut ion with the levels skewed towar d the uppe r part of the silicon
band gap.
A more rec ent study com pared the P
b1
and P
b0
densi ties of st ates usi ng conventi onal ESR
measure ments ver sus applie d voltage on quite thin 3 .3 nm oxide s on silicon [30]. Various
potentia ls were applie d to the bare oxide surface whil e ESR measure ments were made. The
surfa ce pote ntial was appl ied with coro na ions and measure d with a Ke lvin probe elect ro-
static voltm eter. This appro ach had been utilized in ear lier studi es on the (111) P
b
cen ter
and the (100) P
b0
[7,8,10,1 1,14,19] , and the early Ger ardi et al. [14] (100) P
b1
study .
Mea suring the fract ion of P
b0
and P
b1
cen ters with a n unpai red electron as a function of
surfa ce potential provi des a rough compar ison measurem ent of the density of st ates of the
center s as illus trated fo r P
b0
in Figure 6.14. The P
b0
and P
b1
resp onse was monito red on a
Si=SiO
2
structur e in which both centers we re presen t at approxim ately equ al densi ties.
Figure 6.15 shows repres entative ESR spe ctra of P
b0
and P
b1
for four app lied cor ona
biase s. The P
b1
signal is below the detectio n limit at bot h the high est and lowest bias . A plot
of P
b0
and P
b1
ESR amp litude ver sus corona bias is shown in Figure 6.16. The P
b1
signal
P
b
Amplitude versus Fermi energy
0
0.5
1
1.5
2
2.5
3
−0.4 −0.2 0 0.2 0.4
E–E
1
(eV)
P
b

(
1
0
1
2
/
c
m
2
)
FIGURE 6.14
A plot of paramagnetic P
b
concentration plotted versus the Fermi energy of the Si=SiO
2
interface. (Reprinted from
Lenahan, P.M., Microelectron. Eng., 69, 173, 2003. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
appears over a narrow range of bias over which the Si=SiO
2
Fermi level is fairly near the
middle of the silicon band gap. The P
b1
amplitude curve is quite a bit narrower than that of
P
b0
. The P
b1
curve is also shifted negatively with respect to the P
b0
curve. The results of
Figure 6.16 lead to several conclusions regarding the P
b1
density of states. (Although these
conclusions are definitive, they are only semiquantitative.)
3448 3453 3458 3463 3468 3473 3478 3483
Magnetic field (G)
E
S
R

i
n
t
e
n
s
i
t
y

(
a
r
b
.

u
n
i
t
s
)
+1.6 V
+0.8 V
+0.2 V
−0.3 V
P
b0
P
b1
FIGURE 6.15
Four ESR traces taken with different applied corona biases on a 3.3 nm oxide on a (100) Si substrate. The P
b0
and
P
b1
signal amplitudes clearly change with applied corona bias. (Reprinted fromCampbell, J.P. and Lenahan, P.M.,
Appl. Phys. Lett., 80, 1945, 2002. With permission.)
Applied corona bias (V)
−0.5 0.5 1.5
P
b
1

a
n
d

P
b
0

a
m
p
l
i
t
u
d
e

(
a
r
b
.

u
n
i
t
s
)
P
b0
P
b1
FIGURE 6.16
Plot of P
b0
and P
b1
signal amplitudes as a function of applied corona bias on a 3.3 nmoxide on a (100) Si substrate.
The plot indicates that the P
b0
correlation energy is much larger than the P
b1
and that the P
b1
energy distribution is
skewed below midgap. (Reprinted from Campbell, J.P. and Lenahan, P.M., Appl. Phys. Lett., 80, 1945, 2002. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
Earlie r P
b
=coro na bias st udies, such as those illus trated in Figu res 6.15 and 6.1 6,
utilizing much thicker oxides, allowed for a straightforward comparison between CV
determination of the relationship between the Si=SiO
2
interface Fermi energy and the
applied voltage [7,8,10,11,14,19]. Although such a CV=corona ion voltage comparison is,
in principle, possible for thin (3.3 nm) oxides utilized in this case, the precision of both the
measurements and the analysis is lower than had been the case in earlier studies on thicker
oxides. However, reasonably precise knowledge of the P
b0
defect’s response to the Si=SiO
2
interface Fermi energy already exists; this knowledge can be used to crudely calibrate the
P
b1
response.
In examining the results of Figures 6.15 and 6.16, note first that both P
b0
and P
b1
amplitudes grow, and then shrink as the voltage is varied from the most negative to the
most positive polarity; that is, as the Si=SiO
2
Fermi level is advanced fromnear the valence
band edge, through the Si band gap, to near the conduction band edge. (The right and left
sides of the voltage scale correspond to strong inversion and strong accumulation.) This
response indicates that, as has been established earlier for the (111) P
b
and (100) P
b0
, P
b1
defects are amphoteric [7,8,10,11,14,19]. Second, note that the P
b1
curve is much narrower
than that of P
b0
, and that the P
b1
maximum is shifted negatively with respect to the P
b0
curve. The half-maximum width of the P
b0
curve corresponds roughly to the distance
between the peaks of the þ=0 charge and 0=– charge transitions. In band bending (or
relative Fermi energy positions) this corresponds to %0.6–0.7 eV [10,11]. This 0.6–0.7 eV is
approximately the P
b0
electron correlation energy. Since the P
b1
curve has only about
one-half the P
b0
curve width, the P
b1
correlation energy should be roughly half this size
%0.3 eV. This result is qualitatively consistent with the earlier findings of Gerardi and
coworkers [7,8,10,11,14,19]. However, note that the P
b1
curve is clearly shifted negatively
with respect to the P
b0
curve. This indicates that the P
b1
density of states is skewed below
the P
b0
density of states. The P
b0
density of states is approximately centered about midgap.
The results thus indicate rather clearly that the P
b1
density of states is shifted (by several
tenths of an electron volt) below midgap. Although a precise number is beyond the
precision of these data, note that the P
b1
peak is about halfway between the P
b0
peak
and the half amplitude value for the P
b0
curve. This indicates that the P
b1
density of states
shifted about %0.7 eV=4 % 0.2 eV downward relative to P
b0
. This result is not consistent
with the results of Gerardi and coworkers who argued that the P
b1
levels were skewed
toward the upper part of the silicon band gap [7,8,10,11,14,19]. The much narrower P
b1
amplitude versus bias result means that the P
b1
electron correlation energy is much smaller
than the ffi0.7 eV value found for P
b0
. The P
b1
electron correlation energy is thus only a few
tenths of an electron volt. This asymmetric distribution of P
b1
levels results in the occupa-
tion of a significant fraction, perhaps one- to two-thirds of the sites, by net negative charge
with the Si=SiO
2
Fermi level at midgap.
As mentioned previously, the SDR detection technique offers great sensitivity advan-
tages. SDR measurements have also been made comparing the gate bias response of
P
b0
and P
b1
centers. The measurements utilized metal–oxide semiconductor field-effect
transistors (MOSFETs) configured as gate-controlled diodes [28,29,53]. The sensitivity of
the measurements was so high that the
29
Si hyperfine side peaks of P
b0
and P
b1
could be
monitored as a function of gate potential. These side peaks do not overlap, so the separate
responses can be clearly separated [28,29,53]. The transistor gate devices were 10 nmthick.
As Fitzgerald and Grove [76] demonstrated long ago, if a MOS device is configured as a
gate-controlled diode, and the gate and drain to substrate diode is forward biased to a
voltage V
J
, one may probe Si=SiO
2
interface trap recombination by varying the gate
potential. When the diode current is maximized, the electron and hole quasi-Fermi
levels are symmetric about the interface intrinsic level and a range of interface traps
in the middle of the gap about eV
J
wide is surveyed. A shift in gate bias shifts the window
ß 2008 by Taylor & Francis Group, LLC.
of energy (eV
J
) toward the upper or lower part of the gap. SDR measurements of P
b0
and
P
b1
centers have been made as a function of gate bias, with the measurements evaluating
the amplitudes of the P
b0
and P
b1
spectra of
29
Si sites. These signals are widely separated by
the hyperfine interactions.
The results of the measurements are illustrated in Figure 6.17, which shows P
b0
and
P
b1
SDR amplitudes as a function of gate bias. (The source drain to substrate was biased to
0.3 V.) Note that the P
b0
curve is substantially broader than the P
b1
curve. Note also that
the P
b1
curve maximum is shifted negatively with respect to the P
b0
maximum. These
results also indicate that the P
b1
electron correlation energy is smaller than that of P
b0
and
that the levels are more narrowly distributed. Another important point is that strong
signals for both defects appear when the midgap region is explored. Thus, both centers
must have some levels at midgap. (There must clearly be some overlap of the P
b0
levels at
midgap.) The ESR and SDR versus gate bias both clearly indicate that the two interface
defects have a density of states schematically illustrated in Figure 6.18.
FIGURE 6.17
A plot of P
b0
and P
b1
SDR signal hyper-
fine amplitudes as a function of gate
bias for a MOSFET subject to both
gamma and hot carrier damage.
(Reprinted from Mishima, T.D. and
Lenahan, P.M., IEEE Trans. Nucl. Sci.,
47, 2249, 2000. With permission.)
S
D
R

a
m
p
l
i
t
u
d
e

(
a
r
b
.

u
n
i
t
s
)
−0.8 −0.6 −0.4 −0.2 0 0.2
Gate voltage (V)
P
b0
P
b1
FIGURE 6.18
A schematic illustration of the P
b0
and
P
b1
densities of states. (Reprinted from
Campbell, J.P. and Lenahan, P.M.,
Appl. Phys. Lett., 80, 1945, 2002. With
permission.)
D
i
t
D
i
t
P
b0
P
b1
0 0.55 1.1
Energy (eV)
ß 2008 by Taylor & Francis Group, LLC.
6.4.2 Role of P
b
Cent ers in Device Perform ance: The Correspo ndence
betwe en P
b
Densi ty and Interfa ce Trap Density
P
b
center s are importan t because they play dominat ing ro les in Si =SiO
2
interface trap ping
under a wide variet y of circumst ances. They are present at very low densities in
as-pr ocessed devi ces that have been proce ssed or annealed sui tably to passiv ate Si dan-
gling bonds at the inter face, but appea r when devices are subjected to many types of
technol ogical ly rele vant stressi ng. The link betwe en P
b
cen ters and interface traps has been
establish ed throug h many obse rvations of seve ral indep endent gro ups.
6.4.2. 1 P
b
Cent ers in As-Proces sed Structur es
Nish i and coworke rs we re first to no te a corre lation between P
b
densi ties and the dens ity of
Si=SiO
2
inter face trap s, D
it
, in vari ously as-pr ocessed Si =SiO
2
structures of relative ly poor
to ver y poo r interfaci al quality [2,3]. Capla n et al. re fined the pioneeri ng observat ions of
Nish i et al. and demons trated a rough one- to-one corre sponde nce between midgap int er-
face trap densi ty and P
b
dens ity in a variet y of (gen erally high int erface trap densi ty)
Si=SiO
2
st ructures [4]. One of the most signi fi cant observat ions of bot h the Nish i et al. and
Capla n et al. studi es was that exposur e to hydro gen at elevated temperat ure grossly
reduce s the P
b
dens ity. Th is observat ion strong ly implies that, in devi ce quality oxides,
inter face trap densi ties are low because the P
b
cen ters have bee n passiv ated by hydrogen .
6.4.2. 2 P
b
Cent ers and Radi ation Dam age
Sever al early st udies by Lenahan and coworke rs [6 –8,10 ,11] using (111) Si =SiO
2
struc tures,
establish ed that, when MOS devi ces are subject ed to ioni zing rad iation, P
b
centers
are generated in densities that approximately match the average of the interface trap
densities generated in the mid-half band gap. Some representative results are shown in
Figure 6.19. They showed that the annealing characteristics of the P
b
centers and radiation-
induce d int erface trap densi ties are virt ually identical (resul ts shown in Figure 6.20). They
furthermore demonstrated that the densities of radiation-induced P
b
centers could be
strongly influenced by processing variations. Processing yielding low P
b
generation also
0 3ϫ10
5
3ϫ10
6
10
6
10
7
0 3ϫ10
5
3ϫ10
6
10
6
10
7
0
5
10
15
Dose (rad) Dose (rad)
P
b

(
1
0
1
1
/
c
m
2
)
D
i
t

(
1
0
1
1
/
c
m
2
)
(a)
0
5
10
15
(b)
FIGURE 6.19
(a) Average interface trap density in the middle half of the band gap, D
it
, versus radiation dose. (b) Density of
paramagnetic P
b
centers measured with ESR. The absolute value of the concentration may be in error by up to a
factor of 2. The relative concentrations are more accurately determined, ranging from about Æ10% for the higher
concentration to about Æ25% for the smallest concentration. (Reprinted from Lenahan, P.M. et al., IEEE Trans.
Nucl. Sci., 28, 4105, 1981. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
produced low yields of radiation-induced interface trap densities. Processing yielding
higher P
b
densities resulted in proportionately larger interface trap densities. A later
study by Kim and Lenahan [19] extended these results to the more technologically import-
ant (100) Si=SiO
2
system. Some results from these studies are shown in Figures 6.21 and
6.22. As mentioned previously, an interesting aspect of the Kim and Lenahan study was
the observation that the radiation-induced P
b
centers were primarily (although not exclu-
sively) P
b0
defects. (Weaker P
b1
spectra were also generated [19].) The results of these
studies have been confirmed and extended by many other groups.
Miki et al. [21] compared the response of ultradry and steam-grown oxides to ionizing
radiation. They found higher densities of radiation-induced interface traps and higher
densities of radiation-induced P
b
centers in the steam-grown oxides. They also found a
rough numerical correspondence between the ratios of induced P
b
and interface traps as
well as in the absolute numbers of P
b
centers and interface traps. However, since the ESR
measurements were made on soft x-ray irradiated devices and the electrical measurements
were made on devices in which holes were avalanche injected into the oxide (to simulate
the radiation), precise numerical comparisons were not possible.
Awazu et al. [24] have also studied the role of processing parameters on the gener-
ation of P
b
centers by ionizing radiation. Among the oxide processing parameters inves-
tigated were those used in the first study of Lenahan et al. [6]. For these oxides, Awazu
et al. obtained the same P
b
versus dose curve reported in the original 1981 study.
0
10
20
30
40
50
0 50 100 150 200 250
Temperature (ЊC)
0
10
20
30
40
50
0 50 100 150 200 250
Temperature (ЊC)
P
b

(
1
0
1
1
/
c
m
2
)
D
i
t

(
1
0
1
1
/
c
m
2
)
(a) (b)
FIGURE 6.20
(a) Density of paramagnetic radiation-induced P
b
centers as a function of isochronal annealing temperature.
(b) Average radiation-induced interface trap density in the middle half of the band gap D
it
as a function of
isochronal annealing temperature. (Reprinted fromLenahan, P.M. and Dressendorfer, P.V., J. Appl. Phys., 54, 1457,
1983. With permission.)
FIGURE 6.21
Density of radiation-induced P
b0
centers versus mid-half band
gap interface trap density in MOS devices subjected to
60
Co
gamma irradiation. (Reprinted from Kim, Y.Y. and Lenahan,
P.M., J. Appl. Phys., 64, 3551, 1988. With permission.)
0
5
10
0 5 10
D
it
(10
11
/cm
2
)
P
b
0

(
1
0
1
1
/
c
m
2
)
ß 2008 by Taylor & Francis Group, LLC.
Awazu et al. [24] studied oxide s grown on bot h (111) and (100) subs trates. They found
that if an as-proc essed interface had low P
b
density then technol ogically relevant irradi-
ation levels (10 M rad(SiO
2
)) generat ed ver y large ( $ 10
12
=cm
2
) P
b
densi ties. Ho wever, if
the as-proc essed P
b
densities were extre mely high ( $ 2–3 Â 10
12
=cm
2
) the P
b
density was
reduce d. Th ey argu ed that this resu lt shoul d be expecte d from el ementary reaction
theo ry.
Vranch et al. [20] have also inv estigate d the effect s of ionizing radiat ion on the silicon –
silicon dioxi de inter face. They showe d that dose levels of several Mrad (SiO
2
) could gene rate
$ 10
13
P
b
cen ters = cm
2
as measured in a conve ntional ESR measu rement. Their studi es
includ ed conve ntional ES R as well a s SDR. They found (as did several other gro ups) a
strong prep onderan ce of P
b0
centers generated by radiation stressi ng.
6.4.2. 3 P
b
Cent ers Gen erated by High an d Low Oxide Field Electr on Injection
In additi on to ionizing radiat ion, several othe r oxide stressi ng mech anism s have bee n
investi gated by ESR. Mikawa and Lenah an [13,15] showe d that P
b
center s can also be
generat ed by injec ting electro ns into an oxide at low field by inter nal phot oemis sion.
Warren and Lenah an [16] showe d that P
b
cen ters can also be gene rated by high- field
stressi ng oxides. In bot h st udies, a rough (about on e-to-one) correspond ence was observe d
betwe en the densi ties of P
b
c enters gene rated and the densities of interface traps in the
middl e half of the band gap. Results are shown in Figu res 6.23 and 6.24.
6.4.2. 4 P
b
Cent ers Gen erated by Hot Carrie r Stres s of Short- Channe l MOS FETs
Krick et al. [23] and Gabrys et al. [25] have use d SDR to study hot carrier damage centers
create d nea r the dr ain of short-ch annel MOSFE Ts. These studies involved the highest
sensi tivity ES R measure ments ever mad e in con densed matter up to that tim e. Both
Krick et al. and Gabrys et al. generat ed strong P
b
SDR signals by hot hole stressing the
drain regions of the transistors. Krick et al. reported that the P
b0
SDR signal scaled
with increasing interface trap density as measured by charge pumping. The Gabrys
measurements were sufficiently sensitive to allow for the first detection of the
29
Si hyper-
fine side peaks as well as an evaluation of the P
b0
29
Si hyperfine tensor.
0
1
0 50
0.5
100 150 200 250
Temperature (ЊC)
R
e
m
a
i
n
i
n
g

f
r
a
c
t
i
o
n

o
f

P
b
0

a
n
d

D
i
t
P
b0
D
it
FIGURE 6.22
Plot of remaining fractions of P
b0
and D
it
as a function of 1 h isochronal annealing temperature. (Reprinted from
Kim, Y.Y. and Lenahan, P.M., J. Appl. Phys., 64, 3551, 1988. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
6.4.2.5 P
b
Centers Generated in the Negative Bias Temperature Instability
The negative bias temperature instability (NBTI) is perhaps the most important reliability
problem of early twenty-first century complementary MOS (CMOS) integrated circuitry.
NBTI involves a negative threshold voltage shift and a loss in drive current when pMOS-
FETs are subjected to a modest negative bias at moderately elevated ($1208C) temperature
[77–79]. Fujieda et al. [80] utilized conventional ESR measurements on NBTI stressed MOS
capacitors with p-type substrates to show that strong P
b0
and P
b1
spectra are generated
when the structures are subjected to technologically relevant NBTI stressing conditions.
Campbell et al. [81] utilized SDR measurements on NBTI stressed pMOSFETs to show that
strong P
b0
and P
b1
SDR spectra are also generated when the transistors are subjected to
technologically relevant NBTI stressing [81]. Campbell et al. [81,82] further demonstrated a
strong correlation between P
b0
=P
b1
generation and NBTI-induced interface trap density
(per unit energy) D
it
generation.
In a rare exception to the rule Campbell et al. found that in plasma-nitrided
pMOSFETs, P
b
centers do not dominate NBTI-induced D
it
generation [83]. They found
that a different silicon dangling bond center, a silicon back bonded to three nitrogens [83],
a defect called the K center or K
N
center, dominates the NBTI-induced interface trap
generation [83,84].
FIGURE 6.23
Density of trivalent silicon defects at the Si=SiO
2
interface, P
b
, versus interface trap density, D
it
, in
the middle one-half eV of the silicon band
gap. (Reprinted from Mikawa, R.E. and Lena-
han, P.M., Appl. Phys. Lett., 46, 550, 1985. With
permission.)
0
2
4
6
8
10
0 2 4 6 8 10
P
b

(
1
0
1
1
/
c
m
2
)
D
it
(10
11
/cm
2
)
FIGURE 6.24
Density of trivalent silicon defects at the Si=SiO
2
interface versus interface trap density, D
it
, in the
middle half of the silicon band gap. (Reprinted from
Warren, W.L. and Lenahan, P.M., Appl. Phys. Lett., 49,
1296, 1986. With permission.)
0
5
10
0 5 10
D
it
(10
11
/cm
2
)
P
b

(
1
0
1
1
/
c
m
2
)
ß 2008 by Taylor & Francis Group, LLC.
6.4.2. 6 Conclusion Reg arding P
b
Centers and Si =SiO
2
Instabili ties
At lea st six gro ups [6 –8,10,11,1 3,15,18 –25,28] have reported essentia lly incon troverti ble
measure ments on many ox ides demons trating a strong generat ion of P
b
cen ters in device
stressi ng. For radiation- induce d insta bilitie s, all fi ve groups conclude d that P
b
centers
play a dominat ing role. Althou gh these studies have bee n most exten sive fo r ionizin g
radiat ion, one may reasona bly conclud e that P
b
cen ters do indeed play domi nating ro les in
several techno logically importan t instab ilities. The earlier studies of Nish i et al. [2,3] and
Poind exter et al. [5] establ ished that P
b
centers play domi nating ro les in as-proc essed
Si=SiO
2
structur es with relative ly poor inter faces.
6.4.3 Roles of E
0
Cent ers in MOS Devi ces
The roles of E
0
cen ters in high-qu ality thermal ly grown oxide s are fair ly we ll but no t
yet com pletely understo od. The elect ronic beha vior of E
0
centers is summa rized in the
follo wing list:
1. They can act as ho le trap s.
2. They cer tainly play a signi fi cant role in trap -assisted tunneli ng phenomena in
elect rically stresse d oxide s; for example , st ress-indu ced leakage cur rents.
3. It is clear that both pos itively charged and neutr al E
0
cen ters can be generated in
SiO
2
films.
4. Posit ively charge d and com pensated E
0
centers can be generat ed unde r some
circ umstance s.
5. Rep eated electron –ho le fl ooding can, at leas t unde r some circ umstance s, alter E
0
cen ter structure.
6. It is also clear that seve ral E
0
variants can be generat ed in thin films.
7. Sever al vari ants inv olve hydro gen com plexed E
0
center s.
8. Detai led struc ture of seve ral of the variants has yet to be identi fi ed.
6.4.3. 1 E
0
Cent ers as Oxide Hole Traps
E
0
cen ters were probab ly firs t detecte d in thermal ly gro wn oxide s by Marqu ardt and Si gel
[85] who studi ed quite thick (up to 11,000 Å) ox ide films subject ed to qui te high (up to 220
Mrad(S iO
2
)) doses of ionizin g radiat ion. They observe d weak signals in these fi lms which
they attribu ted to E
0
cen ters. Althou gh they did no t report resu lts of elect rical measu re-
ments , the y proposed (corre ctly) that E
0
c enters can be the rmal oxide hole trap s.
The elect ronic prope rties of E
0
centers as hole traps and the ir signi ficance in MOS device
operati on were demonstr ated by Lenah an and Dres sendorfe r [10,11, 38], who made a series
of obse rvatio ns establish ing that E
0
centers are domi nating ho le trap cen ters in a vari ety of
MOS oxide s. The role of E
0
center s in hole trapping was establ ished throu gh several
observat ions, which are as follo ws:
1. Ther e is a ro ugh one-to-o ne cor respond ence betwe en E
0
dens ity and the density of
hole traps in relative ly hard and relative ly soft oxide s gro wn in both steam and dry
oxy gen (see Figure 6.25).
2. Ther e is a ro ugh one-to -one correspo ndence betwe en E
0
density and trapped hole
density in oxides irradiated under positive gate bias over a technologically mean-
ingf ul range of ionizing radiat ion dose (see Figure 6.26).
ß 2008 by Taylor & Francis Group, LLC.
3. MOS ox ide E
0
center s and oxide trapped ho les have the same annealin g respons e
in air (see Figure 6.27).
4. Di stributio n of E
0
centers and trapped holes are virtua lly ide ntical in oxide s
sub jected to ionizin g irradiat ion unde r posit ive gate bias; both are clos e to the
Si =SiO
2
bounda ry (s ee Figu re 6.28).
The results of thos e early studi es have bee n con firmed a nd ext ended in qui te a few later
studi es.
1. Taka hashi and cowo rkers [39] also reporte d an approxi mately on e-to-one cor res-
pond ence betwe en E
0
center s and trapped ho les; the y also repo rted that the distri-
buti on of E
0
cen ters and trapped holes were the same in their irr adiated ox ides.
2. Lip kin et al. [86] also measure d an approxim ately one-to-o ne corre sponde nce
between E
0
density and the density of trapped holes generated in oxides subjected
to 10–20 Mrad(SiO
2
) of gamma radiation.
0
2
4
0 2 4
ΔV
mg
C
ox
/e (10
12
/cm
2
)
E
Ј

(
1
0
1
2
/
c
m
2
)
FIGURE 6.25
Concentration of E
0
centers plotted versus trapped hole density in the oxide (DV
mg
C
ox
=e) for MOS structures
subjected to four sets of processing parameters. All structures were irradiated to 10 Mrad(SiO
2
); thus, differences
in E
0
concentration here are all due to processing variations. (Reprinted from Lenahan, P.M. and Dressendorfer,
P.V., J. Appl. Phys., 55, 3495, 1984. With permission.)
FIGURE 6.26
Distributions of E
0
and DV
mg
C
ox
=e
versus irradiation dose for MOS struc-
tures with oxides grown in dry oxygen
and subjected to a nitrogen anneal.
(Reprinted from Lenahan, P.M. and
Dressendorfer, P.V., J. Appl. Phys., 55,
3495, 1984. With permission.)
0
1
2
3
0 5 10
Dose (Mrad)

E
Ј

a
n
d

Δ
V
m
g

C
o
x
/
e

(
1
0
1
2
/
c
m
2
)
ΔV
mg
C
ox
/e
ß 2008 by Taylor & Francis Group, LLC.
3. Miki et al. [21] compared both E
0
generation and trapped hole generation in
ultradry and steam-grown oxides. They found that their ultradry oxides contained
twice as many E
0
centers as the steam-grown oxides and that the ultradry oxides
also had twice as many trapped holes as the steam-grown oxides. In addition, they
found a rough numerical correspondence between the E
0
densities and trapped
hole densities in the samples investigated. However, since Miki et al. [21] made
electrical measurements on oxides subjected to avalanche injection of holes and
ESR measurements on x-ray irradiated oxides, a precise numerical comparison
between E
0
density and trapped hole density was not possible.
4. Awazu et al. [24] have explored the role of processing parameters on E
0
generation.
They noted that, as Lenahan and Dressendorfer [10,11,38], and Miki et al. [21] had
0
0.5
1
0 100 200 300
Temperature (ЊC)
R
e
m
a
i
n
i
n
g

f
r
a
c
t
i
o
n

o
f

E
Ј

a
n
d

Δ
V
m
g
ΔV
mg

FIGURE 6.27
Plot of remainingfractions of DV
mg
andE
0
as a functionof isochronal annealing time. (ReprintedfromLenahan, P.M.
and Dressendorfer, P.V., J. Appl. Phys., 55, 3495, 1984. With permission.)
0
1.5
3
0 300 600 900 1200
Remaining oxide thickness (Å)
E
Ј

p
r
e
s
e
n
t

i
n

o
x
i
d
e

(
1
0
1
2
/
c
m
2
)
FIGURE 6.28
Distribution of E
0
density in an irradiated oxide. (Reprinted from Lenahan, P.M. and Dressendorfer, P.V., J. Appl.
Phys., 55, 3495, 1984. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
previously reported, the densities of E
0
centers are strongly processing dependent.
Apoint of particular interest in the Awazustudy is their observation that the cooling
rate after high-temperature processing strongly affects E
0
generation. Awazu et al.
concluded that the E
0
centers in their oxides were holes trapped in oxygen vacancies
O
3
Si þÁSi O
3
. Since they observed $1À3 Â10
12
E
0
centers=cm
2
after modest
($1À11 Mrad(SiO
2
)) doses of ionizing radiation, these centers would inevitably be
the dominant hole trap centers in the oxides of their study.
A particularly extensive series of measurements on E
0
centers and oxide trapped holes
was carried out by Kim and Lenahan [19]. They exposed (100) Si=SiO
2
poly-gate MOS
structures to various levels of gamma irradiation while the MOS structures were under
positive gate bias. Their study combined CV and ESR measurements. They found a very
close, near one-to-one, correlation between E
0
density and the density of trapped holes as a
function of total ionizing radiation dose. They subjected some irradiated devices to internal
photoemission of electrons into the oxide. The E
0
centers and holes were annihilated at the
same rate by the internal photoemission, also with an approximately one-to-one corres-
pondence. This result is shown in Figure 6.29. In addition they found that a series of
anneals in air removed both E
0
centers and trapped holes with, within experimental error, a
one-to-one correspondence. This result is illustrated in Figure 6.30. These measurements
clearly demonstrated the primary role of E
0
centers in hole trapping in these oxides.
FIGURE 6.29
Plot of remaining fractions of E
0
and DV
mg
as a
function of UV illumination. (Reprinted from
Kim, Y.Y. and Lenahan, P.M., J. Appl. Phys., 64,
3551, 1988. With permission.)
0.0
0.5
1.0
0 10 20 30 40
UV illumination (h)
R
e
m
a
i
n
i
n
g

f
r
a
c
t
i
o
n

o
f

E
Ј

a
n
d


D
V
m
g
EЈ amplitude
DV
mg
FIGURE 6.30
Plot of remaining fractions of E
0
and DV
mg
as a
function of isochronal annealing (1 h in air) tempera-
ture. (Reprinted from Kim, Y.Y. and Lenahan, P.M.,
J. Appl. Phys., 64, 3551, 1988. With permission.)
0.0
0.5
1.0
0 50 100 150 200 250
Temperature (ЊC)

DV
mg
R
e
m
a
i
n
i
n
g

f
r
a
c
t
i
o
n

o
f

E
Ј

a
n
d


D
V
m
g
ß 2008 by Taylor & Francis Group, LLC.
The result s of multipl e inde pende nt ESR st udies all indicate a dominan t role for E
0
center s in oxide ho le trapping in thermally gro wn ox ide films on silicon. On the basis of
this mutuall y corro borating work one may conclud e that E
0
center s dominat e oxide hole
trapping in a wide variet y of thermally grown ox ide films on silicon.
6.4.3. 2 Neutral Paramagnet ic E
0
Cente rs
Neutr al E
0
c enters can be generat ed in SiO
2
fi lms under a number of circumst ances. Conley
et al. were able to gene rate neutr al E
0
centers in the rmally gr own SiO
2
on silico n by
expos ing the oxide s to extended pe riods of vacuum ultrav iolet (VUV ) lig ht ( hc =l <
10 eV) from a deuteriu m lamp [52]. Exposing the ox ides to the equiva lent of about 500
Mrad(S iO
2
), the y generat ed an ESR line sh ape wi thin experimen tal error ide ntical to that
of the E
0
spec trum associ ated with ho le trapping. Flood ing the oxide with 5 Â 10
13
elec-
trons =cm
2
did little to the ESR amp litude; if the defects were pos itively charged, the
extre mely large Coulomb ic capture cro ss section would have resulted in the annihi lation
of mo st or all of the E
0
amp litude. (Subseque nt hole inje ction signi ficantl y altered the ESR
spectru m gene rating an E
0
variant that will be dis cussed later in this chap ter.)
Warren et al. gene rated neutr al E
0
cen ters by internal ly photoemi tting elect rons into
plasma-e nhan ced che mical vapor deposite d oxide films on silicon [41]. The ultrav iolet light
utilized in the st udy ( hc =l < 5.5 eV) allowed phot oemis sion of elect rons from the silicon
into the oxide but was not suf ficientl y energe tic to generate el ectron –hole pairs in the
(9.0 eV band gap) SiO
2
. CV measu rements (Figure 6.31) showe d that the photoinje cted
electro ns elimina ted trappe d holes initial ly pres ent in the oxide. ESR measu remen ts indi-
cated a strong generat ion of E
0
cen ters as illustrat ed in Figu re 6.32. A plot of E
0
generat ion
versus the flue nce of inje cted elect rons, shown in Figu re 6.33, indicates a very effect ive
mechanism for neutral E
0
center generation in these films.
The generation of neutral E
0
centers in these high hydrogen content oxides would most
plausibly be explained in terms of an E
0
precursor that is not an oxygen vacancy but a
silicon atom back bonded to three oxygens and a hydrogen. (A hydrogen–silicon bond
breaking event would leave behind a single neutral paramagnetic dangling bond.)
0
0.5
1
−40 −30 −20 −10 0 10 20 30
Volts
C
/
C
o
x
(a) (b)
FIGURE 6.31
CV measurements. C–V curve (a) was taken before electron photoinjection. C–V curve (b) was taken after electron
photoinjection. The photoinjected electrons eliminated oxide positive charge. (Reprinted fromWarren, W.L. et al.,
Appl. Phys. Lett., 53, 482, 1988. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
6.4.3.3 Positive but Compensated E
0
Centers
Conley et al. [62,63] reported the generation of positive but compensated E
0
centers in thick
(400 nm) SIMOX buried oxides. These oxides had previously been subjected to an extended
anneal in an inert ambient at a very high temperature (13258C). The oxide exhibited a high
density of E
0
centers when subjected to hole flooding. However, after gamma irradiation
with the oxides unbiased, the oxides exhibited very high E
0
density with essentially no net
positive charge. This result could be interpreted to mean either that the E
0
centers were
neutral or positively charged and compensated. Conley et al. used internal photoemission
to inject electrons into the oxides. The electron injection resulted in the annihilation of a
large fraction of the E
0
centers with a Coulombic capture cross section clearly indicating
that they had been positively charged and compensated. It should be mentioned that
Walters and Reisman [87] have proposed that neutral electron traps may involve electron
trapping at compensated E
0
centers. The Walters–Reisman proposal is generally consistent
with these observations of Conley et al. [62,63].
FIGURE 6.32
ESR measurements. ESR trace (a) was taken
before electron photoinjection. ESR trace (b)
illustrates an E
0
resonance after electron photoin-
jection. Electron injection substantially increased
the E
0
center density. (Reprinted from Warren,
W.L. et al., Appl. Phys. Lett., 53, 482, 1988. With
permission.)
E
S
R

a
m
p
l
i
t
u
d
e
(a)
(b)
2 G
g =2.0005
0
1
2
3
4
5
6
7
0 5 10 15 20 25 30 35 40
Electron density (10
12
/cm
2
)
E
Ј

c
e
n
t
e
r
s

(
1
0
1
2
/
c
m
2
)
Experimental points
Line: Theory
s =1.50ϫ10
−13
cm
2
FIGURE 6.33
Plot of the E
0
density versus injected electron charge density. The circles are the experimental points and the line
was plotted assuming s ¼ 1.5 Â 10
À13
cm
2
. (Reprinted from Warren, W.L. et al., Appl. Phys. Lett., 53, 482, 1988.
With permission.)
ß 2008 by Taylor & Francis Group, LLC.
6.4.3. 4 Structur al Chang es at E
0
Sites Resulti ng from Repe ated Char ge Cycl ing
Conley et al. [62,63] also note d that repe ated cycles of hole a nd electro n floodi ng could
grossl y alter the balan ce betwe en E
0
densi ty and trapped hole densi ty. They no ted an initial
close corre sponde nce betwe en trap ped ho le and E
0
center densi ty but with repe ated
electro n –hole charge injecti on cy cling note d a gradua l increase in E
0
densi ty that was not
accomp anied by posit ive charge . This resp onse is illus trated in Figure 6.34. Su ch a result is
not surprising if many of the E
0
center hole traps are essentially the Feigl–Fowler–Yip like
oxyg en vacancy centers illu strated in Figu re 6.9. These defects wou ld expe rience a large
structural change upon hole capture. With subsequent electron capture, they might not all
be able to return to the initial simple oxygen vacancy configuration.
6.4.3.5 Near Si=SiO
2
Interface E
0
Centers: Electronic Properties
One could reasonably divide electrically active MOS defects into three categories: interface
traps, which can communicate readily with charge carriers in the silicon, oxide traps which
do not communicate with charge in the silicon, and very near interface traps which can, on
fairly long timescales, communicate with charge carriers in the silicon. The near Si=SiO
2
E4 H4 E3 H3 E2 H2 E1 H1 PRE
+2.0
+1.6
+1.2
+0.8
+0.4
0.0
-0.4
-200
-160
-120
-80
-40
0
40
D
V
m
g

(
V
)
T
r
a
p
p
e
d

c
h
a
r
g
e

(
1
0
1
3
/
c
m
2
)
E
Ј

d
e
n
s
i
t
y

(
1
0
1
2
/
c
m
2
)
(a)
E4 H4 E3 H3 E2 H2 E1 H1 PRE
6.0
5.0
4.0
3.0
2.0
1.0
0.0 0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
E
Ј

d
e
n
s
i
t
y

(
1
0
1
7
/
c
m
3
)
(b)
CV data
ESR data
FIGURE 6.34
Effects of charge injection on (a) CV and (b) ESR measurements of unilluminated buried oxides. (Reprinted from
Conley, J.F., Lenahan, P.M., and Roitman, P., IEEE Trans. Nucl. Sci., 39, 2114, 1992. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
inter face traps go by many name s: slo w states, border trap s [88], sw itching traps [89], etc.
(See Chapte r 7.) It is possibl e, eve n likely, that mo re than on e type of nea r Si =SiO
2
inter face
trap can exist in cer tain ox ides under cer tain con ditions . One type of nea r Si =SiO
2
inter face trap has been dire ctly ide nti fied via ESR, E
0
cen ters. Many studies (inv olving
only electri cal measure ments) show that, when some MOS devices are subject ed to ioniz-
ing radiation, CV and current versus voltage characteristics experience a negative voltage
shift, DV, indicating the capture of positive charge in the oxide [90,91]. However, if a
positive gate bias is applied, the magnitude of DV decreases logarithmically in time,
indicating the annihilation or compensation of some of the positive charge via electron
tunneling. If the applied bias is reversed from positive to negative some of the previously
compensated charge returns [88,89,92]. The charge that returns is said to be in switching
(border) traps. An ESR study by Conley et al. [93,94] clearly demonstrates that some E
0
centers can act as switching traps. These centers are presumably very close to the Si=SiO
2
boundary. Results from the Conley et al. [93,94] study are shown in Figure 6.35. The bias
voltages and bias switching times approximately match those of earlier purely electronic
measurements. The first point (Hþ) indicates the E
0
density initially after holes were
injected into the oxide. The second point (01) was taken after 10
5
s with no bias across
the oxide. Point (N1) was taken after a negative gate voltage corresponding to an average
oxide field of 3.5 MV=cm was applied for 24 h. The negative bias increased the number of
paramagnetic E
0
sites. Point (P1) was taken on the same sample after an additional 24 h
under positive gate bias, also corresponding to an average oxide field of 3.5 MV=cm. The
positive bias substantially decreases the density of paramagnetic E
0
centers. Two additional
biasing points indicate the repeatability of this process. Clearly, the spin state and thus
charge state of these E
0
centers can be repeatedly switched with bias; thus, E
0
centers can act
as oxide switching traps. The results of Conley et al. [93,94] are consistent with, and clearly
confirm, the basic premise of the switching trap model proposed earlier by Lelis and
Oldman [89] and Lelis et al. [95]. After hole capture, subsequent electron capture does
not always return the E
0
site involved to its original condition. This irreversibility leads to
the switching behavior. The results of Conley et al. do not, of course, preclude the
possibility that defects other than E
0
centers may act as switching traps. The Conley et al.
[93,94] results extend earlier results generated by Jupina and Lenahan who reported the
0.5
0.6
0.7
0.8
0.9
1
1.1
Bias condition
E
γ Ј


d
e
n
s
i
t
y

(
1
0
1
2
/
c
m
2
)
H+ 01 N1 P1 N2
P2
FIGURE 6.35
E
0
Density versus various 24 h bias sequences. (Reprinted from Campbell, J.P. et al., Appl. Phys. Lett., 90, 123502,
2007. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
SDR detecti on of E
0
center s [22]. Since SDR can only detect defec ts that in some way
commu nicate with Si =SiO
2
inter face charge carri ers, their results stro ngly indicated that
some near Si =Si O
2
interface E
0
center s did indeed beha ve in this way. The Conley et al.
study is also con sistent with an ESR study by Warren et al. [96] which sugge sted, but did
not demonstr ate , that E
0
cen ters may act as sw itching trap s. The expe rimen tally demon-
strated role of E
0
cen ters as swit ching traps or bor der traps also is consiste nt with more
recent pure ly the oretical work [54,55,9 7].
6.4.3. 6 Hydrogen Com plexe d E
0
Cent ers
Two hyd rogen com plexed E
0
vari ants m ay play importan t ro les in device reliabilit y. As
discuss ed earlier in the chap ter, the hyd rogen comple xed cen ters are called the 74 G
doublet and the 10.4 G doublet. ESR traces of bot h defe cts are shown in Figure 6.13 .
Both defe cts were first observe d in cubic centim eter size sampl es [59,60] , and both center s
clearly inv olve an E
0
hyd rogen com plex. As discuss ed previ ously, the 74 G dou blet defect
very lik ely involve s an unpaired electron on a silicon back bon ded to two oxygen s and one
hydro gen. Th e 10.4 G doublet defect ver y likely involve s an unpai red electron on a silicon
back bon ded to three oxygen s with one of the oxyg ens bon ded to a hyd rogen [59,60] . The
74 G doub let was first obse rved in thin oxide films by Takahas hi et al. [39] who generat ed
them a t somew hat elevated tem peratures ( ffi 100 8C) in irrad iated oxide s. Mo re rec ently,
Conley and Lenahan [57,58] observed the ro om tem perature generat ion of both the 74 G
doublet cen ters and 10.4 G doub let cen ters in ox ides subj ected to either VUV ( hc=l 10.2 eV)
or gamma irr adiation. Taka hashi et al. [39] sugge sted that the hydrog en compl ex E
0
defe cts
migh t play a n importan t role in Si =SiO
2
interface trap generat ion. Conley and Lenah an
[57,58] provided strong circ umsta ntial evid ence linki ng E
0
hydro gen coup led centers to
inter face trap generat ion. Se veral (pure ly elect rical m easuremen t) st udies [98,99] had
shown that a molec ular-hyd rogen con taining amb ient lead s to an enha ncement in radi-
ation -induced inter face trap generation. Conley a nd Lenah an [57,58] showe d that exposin g
an oxide previ ously floode d wi th holes (to generat e E
0
cen ters) to an H
2
=N
2
amb ient leads to
a convers ion of conve ntional E
0
centers to 74 G doublet centers and ot her hydrogen c om-
plexed E
0
centers as well as to the generat ion of interface traps. They foun d that the number
of E
0
cen ters con verted to hydrogen com plexed cen ters is approximat ely equal to the number
of inter face trap s generat ed, with the tim e peri od involve d in int erface trap formatio n
approxi mately equal to the time require d to sat urate the E
0
=hyd rogen com plexin g proces s.
6.4.3. 7 Electron ic Prop erties of E
d
0
(or EP)
Conley et al. [52,61 –63,66] investiga ted the elect ronic prope rties of E
0
d
(EP) centers in a
variet y of ox ides. (Conle y et al. chose to call the centers EP since they felt that the
assignme nt as an E
0
vari ant was provisiona l (thu s the P ) at that tim e.) They fl ooded
the oxides with holes and electrons . From plots of ESR inten sity versus injecte d charge
fluence , they were ab le to extract app roximate captur e cro ss secti ons and to dete rmine the
active size s of the captur e cross sectio n and densi ty of precurso rs fo r E
0
d
(or EP) with respec t
to the conve ntion al E
0
cen ter [52]. From etch back measure ments after ho le injecti on, the y
were a lso able to determi ne the spatial density of the centers, findin g them to be highly
concen trated near the Si=die lectric boundar y [52].
Figure 6 .36 illus trates a plo t of conve ntional E
0
(called E
0
g
here) and E
0
d
(refe rred to as EP
here) versus fluence [52]. Th e conventi onal E
0
center has the smaller capt ure cross secti on
for holes , but its precurso r is clearly presen t in high er densi ties in the oxide under st udy.
The E
0
d
(or EP here) amplitude actually drops at higher hole fluence, indicating that it is
muc h less stable than the con vention al E
0
centers. Figure 6.37 illustrat es a plo t of both E
0
ß 2008 by Taylor & Francis Group, LLC.
densities in response to electron injection [52]. The E
0
d
(EP) center has about an order of
magnitude larger capture cross section than the conventional E
0
center (E
0
g
here).
6.4.3.8 Role of E
0
Centers in Oxide Leakage Currents
Oxide leakage currents caused by various types of stressing are a fundamental concern. An
unstressed oxide exhibits current density versus voltage characteristics consistent with the
tunneling current density anticipated from an ideal band diagram. However, if an oxide is
subjected to stressing, for example, by a high electric field across the oxide, damage caused
by the stressing can cause leakage currents to appear at relatively low oxide fields. These
leakage currents have been extensively investigated in many experimental studies of
a purely electronic nature [100–102]. Several investigators have suggested a link between
E
0
centers and these leakage currents [100,101,103]. The leakage currents are likely caused
by inelastic tunneling of conduction band electrons to defect centers in the oxide near the
Si=SiO
2
boundary. Recent work by Takagi et al. [100] suggests that E
0
centers are good
candidates for the centers involved in the inelastic tunneling process.
0
1
2
3
4
5
6
7
8
0 2 4 6 8 10 12 14 16
Injected hole fluence (10
13
/cm
2
)
D
e
f
e
c
t

d
e
n
s
i
t
y

(
1
0
1
1
/
c
m
2
)
EP
E
Ј
g
P
FIGURE 6.36
E
0
gp
and EP defect density versus hole injection fluence (the lines are drawn only as a guide to the eye). (Reprinted
from Conley, J.F. et al., J. Appl. Phys., 76, 2872, 1994. With permission.)
FIGURE 6.37
E
0
gp
and EP defect density versus
injected electron fluence after hole
injection (the lines are drawn only as
a guide to the eye). (Reprinted from
Conley, J.F. et al., J. Appl. Phys., 76,
2872, 1994. With permission.)
0
2
4
6
8
0 2 4 6 8 10 12 14 16
Injected electron fluence (10
13
/cm
2
)
D
e
f
e
c
t

d
e
n
s
i
t
y

(
1
0
1
1
/
c
m
2
)
EP
s = 10
-14

cm
2
s = 10
-13

cm
2
E
Ј
g
P
ß 2008 by Taylor & Francis Group, LLC.
Sever al earlier studies have show n that E
0
cen ters are gene rated in oxide s subjected to
high elect ric field [42,104, 105]. Lenahan and Mele [43] show ed that the generat ion of
electri cally neutr al E
0
center s are acco mpanied by a large (several orders of magni tude)
mono tonic increase in lea kage curren ts and that the dis appearanc e of the se E
0
centers via
low-te mperat ure (200 8 C) annealin g is accomp anied by a monoto nic decre ase in leakage
curren ts. Recent calcu lations by Takag i et al. [100] indi cate that defect densi ties in the
$ 10
18
–10
19
=cm
3
range in the vicini ty of the Si =Si O
2
int erface ( $10
12
–10
13
=cm
2
) would lead
to leakage current densi ties similar to thos e measure d in the study.
In the Lenahan and Mele study, the leaka ge current generat ed with the coro na ions bias
of the ox ide surfa ce was measure d using a Kelv in probe el ectrostati c voltm eter [43]. This
approach , the m easuremen t of oxide currents via coro na ion decay, was pioneered many
year s ago by Weinbe rg et al. [106]. Taking the derivative of potential ver sus tim e, the y
determi ned current ver sus voltage depen dence: C (d V=d t ) ¼ dQ =dt ¼ i . Here C is the oxide
capa citance, «
0
k
SiO
2

t
ox
, whe re e
0
is the permitti vity of free space, k
SiO
2
is the relative
dielectri c constant of SiO
2
, and t
ox
is the oxide thickness; Q is charge; t is time; and i is
curren t [106]. Althou gh thi s m ethod is quite cru de, it provid ed reasona ble semiquan titative
electri cal result s to compar e to the ESR measu rements .
The cen tral resu lts are illustrat ed in Figu res 6.38 and 6.39. Figure 6.38 illustrat es pre-
VUV, post-V UV, and post-VUV -postanne aling ESR traces taken in the vicini ty of the free
electro n g ffi 2.002. Th e m agnetic field is swept over a range that is appro priate for both
P
b
and E
0
c enters. The spe ctromete r settings use d to obt ain the dat a in Figu re 6.38 are less
than optimum for both E
0
and P
b
spectra. The settings represent a compromise, chosen so
that both spectra would be simultaneously visible. Before VUV illumination, no ESR
signals are visible. After illumination, strong E
0
and P
b
signals are present. An anneal of
15 min at 2008C eliminates most of the E
0
centers but does little to P
b
inten sity. Figu re 6.11
illustrates a post-VUV stressing E
0
spectrum taken at a modulation amplitude and
microwave power significantly lower than that of Figure 6.38. These spectrometer settings
are optimized to detect the E
0
center defects.
3440.5 3450.5 3460.5 3470.5 3480.5
Magnetic field (G)
E
S
R

a
m
p
l
i
t
u
d
e

(
a
r
b
.

u
n
i
t
s
)
(a)
(b)
(c)
P
b0
(g =2.0058)
(z.c. g =2.0005) EЈ
FIGURE 6.38
(a) Pre-VUV, (b) post-VUV, and (c) post-VUV postannealing ESR traces (40 Gwide) taken on the 45 nmoxide films
on silicon; the spectrometer settings were chosen so both the Si=SiO
2
interface P
b
centers; and the E
0
centers are
clearly visible. (Reprinted from Lenahan, P.M. and Mele, J.J., J. Vac. Sci. Technol. B, 18, 2169, 2000. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
Figu re 6.39 shows pre-V UV illuminati on, post-VUV illumi natio n, and the pos tillumina-
tion, posts tressing current –voltage cur ves for the 45 nm oxide s. Th e VUV illuminati on
grea tly increa ses leakage cur rents. These curren ts we re greatly redu ced by the subsequ ent
annea l. Pre- and pos t-VUV illumi natio n CV curves indicate d that essen tially no oxide space
charge is create d by the VUV illumi natio n; thus , the ver y large changes in cur rent ver sus
voltage charac teristics illus trated in Figure 6.39 are not due to space charge in the oxides.
Figu re 6.40 dire ctly compar es E
0
densi ties and leakage cur rents in the 45 nm ox ides as a
functi on of VU V illumi natio n. Note the close correspond ence betwe en E
0
center gene ration
and leakage cur rent generat ion. These resu lts clearly demonstr ate a stro ng qualit ative
corre lation betw een neutral E
0
center densi ty and ox ide leak age currents in seve ral sets of
differe nt ox ides. The resu lts provide extrem ely strong, albe it circ umstanti al, evid ence that E
0
defects , as sugge sted by the work of Takag i et al. [100], are largel y resp onsibl e for the leakag e
cur rents due to inelas tic tunneling of silico n cond uction ba nd elect rons through the ir levels.
6.5 Intrinsi c Defects and Devi ce Reliabilit y: Physi cally Based
Predicti ve Models and Statisti cal Mechanics
Althou gh, many ESR studi es of M OS systems are of some gene ral interest as physi cs,
che mistry, and mate rials scie nce, the ir ulti mate sig nificance must relate to their utility:
can these studi es help design better , more reli able, integr ated circuits? Th e answer to thi s
ques tion is almost certainly yes, if the results can be utilized to predic t and m anipula te
defect densities . It is clear that two families of poi nt defects, E
0
centers and P
b
cen ters, play
dominat ing roles in a number of M OS reliabilit y problem s. Materi als scie ntists and engin-
eers have well-developed and widely verified methods of manipulating intrinsic point
defect populations. These methods are based upon the fundamental principles of the
statistical mechanics of solids as well as on basic principles of physical chemistry
0
10
−7
10
−8
10
−9
10
−10
10
−11
10 20 30
Voltage (V)
C
u
r
r
e
n
t

d
e
n
s
i
t
y

(
A
/
c
m
2
)
FIGURE 6.39
Current density versus voltage measurements on the 45 nm oxides before VUV illumination (as-processed), post-
VUV illumination, and post-VUV postanneal at 2008C. Circles, as-processed; squares, post-VUV illumination; and
triangles, post-anneal. (Reprinted from Lenahan, P.M. and Mele, J.J., J. Vac. Sci. Technol. B, 18, 2169, 2000. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
[107,108]. One should be able to ameliorate device reliability problems by applying these
well-established principles to E
0
and P
b
centers.
6.5.1 Predicting Oxide Hole Trapping
Lenahan andConley [109–113] usedthe standardapproach of statistical mechanics [107,108]
to calculate the density of oxygen vacancies in MOSFET oxides, calibrated the parameters of
the expression with ESR measurements, and then tested the validity of calibrated (quanti-
tative) expression on several oxide films. They found a close correspondence between the
calibrated expression and experimental results. A consideration of the basic principles of
statistical thermodynamics tells us that equilibrium occurs when the Gibbs free energy, G,
of a solid is minimized [108]. It can be shown that, for the simplest cases, the minimization of
Gibbs free energy leads to an equilibriumdensity of vacancy sites given by
n ¼ Ne
DS
f
=kÀDH
f
=kT
(6:18)
where
DS
f
represents the nonconfigurational entropy contribution per defect site
DH
f
represents the enthalpy of formation of a defect site
k is the Boltzmann constant
N represents the density of available sites
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0
(a)
(b)
20 40 60 80 100 120 140
VUV illumination time (min)
0 20 40 60 80 100 120 140
VUV illumination time (min)
E
Ј

d
e
n
s
i
t
y

(
1
0
1
2
/
c
m
2
)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
C
u
r
r
e
n
t

d
e
n
s
i
t
y

(
1
0

8

A
/
c
m
2
)
FIGURE 6.40
(a) E
0
density versus VUV illumination time in the 45 nm oxides. (b) Current density plotted at 17 V versus VUV
illumination time for the same oxides. The lines are drawn as a guide for the eye. (Reprinted from Lenahan, P.M.
and Mele, J.J., J. Vac. Sci. Technol. B, 18, 2169, 2000. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
For the pu rposes of thi s disc ussion, the imp ortant points her e are that the noncon fi gura-
tiona l ent ropy contribut ion is large and essen tially temperat ure-inde pende nt, and the D H
f
essen tially represen ts the inc rease in syste m energy cause d by vaca ncy creation of an
unst ressed lattice site minus the strai n energy lost by removal from a com pressed Si O
2
matrix. (This reducti on in D H
f
wou ld be a strain energy $ P d V caused by the eff ective
volu me change resultin g fro m the remov al of the atom from its partic ular locatio n.) As
poi nted out by Oh mame uda et al. [114] thi s strai n energy reductio n will be greatest for
sites near the Si =Si O
2
bounda ry; this energy con tribution shoul d amo unt to several tenths
of an elect ron volt [109 –113]. On e thus expects and finds [10,11,1 9] that the E
0
centers are
primar ily located clos e to the Si =SiO
2
boundar y.
Ant icipatin g the n an oxyg en vacancy =E
0
precu rsor densi ty of the fo rm
n ¼ ae
À b= T
(6: 19)
where the tem perature-i ndepe ndent constant a is given by a ¼ N e
DS
f
=k
and b ¼ D H
f
=k, we
may eval uate the relevant thermod ynamic con stants by makin g measu rements on devices
expos ed to various high -tempera ture annea ls. With a kno wledge of E
0
center hole captur e
cross sectio n [109 –113] and the stand ard analysis of c harge captur e in oxide films, we
wou ld anti cipate that, for a given fluence of ho les through the oxide s
N
th
¼ ae
À b= T
(1 À e
À sh
) (6: 20)
where N
th
is the densi ty of trapped holes and h is the fluence of holes throu gh the ox ide.
Wit h a , b , and s evaluate d from spin resonan ce measu remen ts, the expres sion provides a
nonad justabl e parame ter prediction of ox ide hole trap ping. (Ho wever, due to the m odest
absol ute precis ion of ESR measurem ents, the value of a as dete rmine d st rictly from ESR
coul d be in error by almos t a factor of 2.) The pote ntial valid ity of Equatio n 6.20 was
asses sed [109 –113] throu gh a series of measu remen ts on MOS oxide s subject ed to anneal s
at 875 8 C, 950 8 C, 1025 8C, and 1100 8 C. The oxide s were all grown at 825 8 C and then a
pol ycrystal line silicon gate was deposite d. After gate depositi on the annea ls were carried
out for 30 min in a dry N
2
atmos phere. After the annea ls the capa citors were rap idly pulle d
from the fur nace to que nch in the defe ct densi ties at the anne aling tem perature s. Th e poly-
Si gates were remov ed a nd two sets of measure ments were mad e on the sample s, both after
subj ecting the ox ides to hole floodi ng. To evalu ate the E
0
precu rsor enthal py of creation ,
oxide s of the three higher temperat ure anne aling sample s we re each floode d wi th approxi-
mate ly 2 Â 10
13
holes =cm
2
. The enthalpy was dete rmined from the slo pe of a plo t of the
natural logarithm of E
0
densi ty versus recipro cal tem perature, shown in Figu re 6.41; the
acti vation enthalpy is appro ximately 1.5 Æ 0.1 eV. To test the predic tive capab ility of
Equati on 6.20, holes were inje cted into sam ples subjected to each of the four annealin g
steps ; midgap CV sh ifts, D V
mg
, were plotte d ver sus injec ted hole flue nce. Usin g Equati on
6.20 and taking the trapped holes to be close to the Si =SiO
2
bound ary, the simp le mo del
[109–113] predicts midgap shifts of
DV
mg
¼ qae
Àb=T
C
ox
(1 Àe
sh
) (6:21)
where
q is the electronic charge
C
ox
is the oxide capacitance
all other parameters are as previously defined
Figu re 6.42 compar es the expe rimen tal resu lts and the predic tions of Equatio n 6.16. The
correspondence between prediction and experiment is quite close. It clearly demonstrates
ß 2008 by Taylor & Francis Group, LLC.
that it is possible to predict the response of an oxide from an essentially nonadjustable
parameter fit of a physically based model. The empirical results of Lenahan and Conley
are consistent with more recent theoretical estimates of the enthalpy of E
0
formation. Their
measurement of 1.5 Æ0.1 eV is close to calculated estimates of 1.3 eV by Boureau et al. [115]
and Capron et al. [116]. As demonstrated by Conley et al. [112] the anticipated trap
densities depend on both thermodynamics and kinetics.
6.5.2 Predicting Interface Trap Formation
Straightforward concepts fromthe equilibriumthermodynamics of chemical reactions also
allow one to make some predictions about interface trap formation [110,113,117,118]. As
discussed previously, at least five groups have demonstrated that significant (greater than
25
26
27
28
29
30
0.72 0.76 0.8 0.84 0.88 0.92
1/T (1000/K)
l
n
(
E
Ј
)
FIGURE 6.41
Plot of the natural logarithm of E
0
density versus the reciprocal of annealing temperature. Slope of the line yields
an activation energy of about 1.5 eV. (Reprinted from Lenahan, P.M. and Conley, J.F., J. Vac. Sci. Technol. B, 16,
2134, 1998. With permission.)
0
2
4
6
8
10
12
0 0.5 1 1.5 2 2.5
Hole fluence (10
13
/cm
2
)
875°C
950°C
1025°C
1100°C
D
V
m
g
FIGURE 6.42
Solid lines represent Equation 6.21 evaluated for the various indicated temperatures. Dots represent experimental
results. (Reprinted from Lenahan, P.M. and Conley, J.F., J. Vac. Sci. Technol. B, 16, 2134, 1998. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
or approximately equal to 10
12
=cm
2
) generation of P
b
centers occurs when MOS oxides are
subjected to technologically relevant levels of ionizing radiation. More limited results
indicate P
b
generation resulting from high or low oxide electric field injection of electrons
as well as the injection of hot holes into the oxide from the near drain region of short-
channel MOSFETs. The previously discussed studies of Conley and Lenahan [56–58] show
that E
0
centers rapidly react with molecular hydrogen at room temperature and that this
reaction is accompanied by the simultaneous generation of Si=SiO
2
interface traps (these
are P
b
centers).
It is well established that $5 Â10
12
cm
2
Si=SiO
2
silicon dangling bonds sites are passiv-
ated by hydrogen [2–5]. As pointed out by Conley and Lenahan [110,113,118], in an
unstressed oxide, one would expect to find the circumstances schematically illustrated in
Figure 6.43. A large number of hydrogen dangling bond sites (P
b
H) exist at the Si=SiO
2
boundary while a large number of E
0
precursor sites exist in the oxide. Essentially no
unpassivated silicon dangling bonds exist at the E
0
precursor sites. For purposes of
illustration, the precursor sites are taken to be oxygen vacancy sites. The stressing process
(ionizing radiation, hot carrier injection, high-field stressing, etc.), for whatever reason,
generates significant densities of E
0
center silicon dangling bonds in the oxide. This
situation is schematically illustrated in Figure 6.44. Note the presence of both a large
number of hydrogen passivated silicon dangling bond sites at the Si=SiO
2
boundary and
a large number of unpassivated silicon dangling bond sites in the oxide. Such a combin-
ation of defects centers cannot be in thermodynamic equilibrium. Thermodynamics
requires that a system in equilibrium will reach the lowest Gibbs free energy G [108]:
G ¼ H ÀTS (6:22)
Oxide
Si/SiO
FIGURE 6.43
A schematic illustration of defect–defect precursor structure before stressing. The filled circles indicate silicon
atoms; the small open circles indicate hydrogen atoms. (Reprinted fromLenahan, P.M., Microelectron. Eng., 69, 173,
2003. With permission.)
Oxide
Si/SiO
FIGURE 6.44
A schematic and perhaps somewhat artificial illustration indicating the creation of silicon dangling bonds
(dangling bonds indicated by the gray ovals) in the oxide in the presence of passivated silicon atoms at the
Si=SiO
2
boundary. The filled black circles indicate silicon atoms; the smaller open circles indicate hydrogen atoms.
(Reprinted from Lenahan, P.M., Microelectron. Eng., 69, 173, 2003. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
where
H is the enthalpy
T is the ab solute tem perature
S is the entro py
Enthalp y H is the sum of energy plus press ure tim es volu me
The bond enthal py of a silicon hydro gen bond (P
b
H) at the Si=Si O
2
bound ary must be
very rough ly equ al to the enthalpy of a silico n hyd rogen bond in the ox ide ( E
0
H )
[110,113 ,118]. Th us, the exchange of an inter face dan gling bond hydrog en wi th an oxide
dangling bon d would cost ver y littl e enthal py. The ent ropy of a system is de fi ned as [108]:
S ¼ k ln V (6 : 23)
where
k is the Boltzman n constant
V is the number of micro scopic con fi guration s respons ible for the macros copic system
Consi der a simpli fi ed situati on in whic h all the M Si= SiO
2
int erface dangling bon d site s
are hydrogen passiv ated ( P
b
H ). The con figura tiona l ent ropy of thi s system is simply k ln
1 ¼ 0. Remo ve one hy drogen from any on e of the M site s. The con fi guration al entropy
become s k ln M , a ver y large inc rease. (The rem oval of a second hydrogen would lead to a
con fi guration al entro py of k ln[( M )( M À 1) =2], etc.)
Consi der a simp lified situati on in the oxide in which each of the N E
0
dang ling bond site s
is unpassiv ated. The con fi guration al ent ropy of the se site s is also simply k ln 1 ¼ 0. Add one
hydrogen to any one of the N sites. The configurational entropy contribution increases to
k ln N, again quite a large contribution. (The addition of a second hydrogen would similary
lead to a con figura tion entro py of k ln[( N )(N À 1)=2], etc. ) Th is proce ss is sch ematica lly
illustrat ed in Figure s 6.45 and 6.46.
All EЈ dbs are unpassivated
S (configurational) = k ln(1)
All P
b
dbs are passivated
S (configurational) = k ln(1)
(Total of N)
(Total of M)
¼
¼
FIGURE 6.45
A schematic illustration of an oxide with N E
0
sites all unpassivated and M P
b
sites all of which are passivated. In
both the unpassivated E
0
dangling bond and the passivated P
b
dangling bond cases, the configurational entropy,
S, is given by k ln(1). The filled black circles indicate silicon atoms, the larger open circles indicate oxygen atoms,
the smaller open circles indicate hydrogen atoms, and the gray ovals indicate silicon dangling bonds. (Reprinted
from Lenahan, P.M., Microelectron. Eng., 69, 173, 2003. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
Thus, very basic statistical thermodynamics arguments indicate that the exchange of a
large number of hydrogens from interface dangling bonds centers to oxide dangling bond
centers is thermodynamically favored. This thermodynamic equiliblium can be expressed
in the following equations:
P
b
H þE
0
$P
b
þE
0
H (6:24)
and
P
b
½ Š E
0
H ½ Š
P
b
H ½ Š E
0
½ Š
¼ K (6:25)
In Equation 6.25, the constant K (not the SiO
2
dielectric constant) is defined as fol-
lows [108]:
K ¼ exp
ÀDg
s
kT

(6:26)
where Dg
s
represents a Gibbs free energy per site [108]. As discussed above, the enthalpy
change is near zero.
On the basis of the above discussion K % 1. Thus, if the system can reach some sort of
quasiequilibrium condition, one would, inevitably, observe a fairly close correspondence
between interface trap generation and oxide leakage currents. As noted previously, these E
0
defects are largely responsible for oxide hole trapping under many circumstances and
are likely also responsible for oxide stress-induced leakage currents. As noted previously,
the P
b
centers are quite clearly the dominating source of interface traps under many
Remove one H from the (M) PbH sites;
ΔS (configurational) = k ln(M)
Add one H to the (N) EЈ sites;
ΔS (configurational) = k ln(N)


FIGURE 6.46
A schematic illustration of the effect of the transfer of one hydrogen atom from a passivated interface dangling
bond to an unpassivated oxide dangling bond. Note the large increase in configurational entropy in both
cases. The filled black circles indicate silicon atoms, the larger open circles indicate oxygen atoms, the smaller
open circles indicate hydrogen atoms, and the gray ovals indicate silicon dangling bonds. (Reprinted from
Lenahan, P.M., Microelectron. Eng., 69, 173, 2003. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
circumstances. Thus, the often noted correspondence between oxide defect generation or
population and interface trap generation [119–121] makes physical sense in terms of the
chemistry and energy levels of these defects.
6.6 Relevance of P
b
and E
0
Centers to High Dielectric Constant
(High-K) MOSFETs
As the equivalent oxide thickness of MOS gate stacks approaches fundamental physical
limits, high dielectric constant (high-K) oxide devices are under development [122,123].
The most promising high-K dielectrics are based on HfO
2
and have typically been depos-
ited by the atomic layer deposition (ALD) technique [124]. The P
b
and E
0
families may
remain as important defects in these new device structures because they typically involve
HfO
2
on a very thin ($1 nm) silicon dioxide interfacial layer on the silicon substrate [124].
Not surprisingly, P
b
or P
b
-like centers may be present at the Si=dielectric boundary of these
new devices. E
0
centers may also be present.
Kang et al. [125,126] reportedconventional ESRmeasurements of ALDHfO
2
films on both
(111) and(100) silicon substrates. They detected the presence of P
b
-like interface traps, nearly
like (111) P
b
centers and (100) P
b0
centers, but with small shifts in the g matrix elements
toward higher values (g
k
¼2.0018 and g
?
¼2.0094 in contrast to the Si=SiO
2
values of
g
k
¼2.0013 and g
?
¼2.008). Triplett et al. [127] have also reported observation of P
b
like
centers, also with matrix elements shifted to slightly higher values in HfO
2
-based structures.
Stesmans and Afanas’ev [128] observed both P
b0
and P
b1
centers in HfO
2
-based device
structures. Unlike Kang et al. [125,126] and Triplett et al. [127] they did not observe any
shifts in g matrix elements fromthe values reported for conventional Si=SiO
2
interfaces. Both
Kang et al. [125,126] and Stesmans and Afanas’ev [128] reported that forming gas anneals
(H
2
=N
2
) were effective in reducing the density of the P
b
or P
b
like centers.
E
0
centers may also play a significant role in HfO
2
-based devices. Ryan et al. [129]
recently reported the presence of quite high densities of two types of E
0
centers in the
interfacial SiO
2
layer of HfO
2
-based MOS structures (up to 10
19
=cm
3
). They found high
densities in device structures before any sort of stressing. The E
0
densities depend quite
strongly upon the ALD deposition conditions, and the presence of a metal gate greatly
enhances E
0
generation resulting from postdeposition anneals.
6.7 Summary
Two families of silicon dangling bond defects, P
b
centers and E
0
centers, play extremely
important roles in MOS device limitations. Although a comprehensive understanding of
MOS limitations as well as of these defects has yet to be established, their overall importance
as the dominating interface traps (P
b
) and dominating oxide deep level centers (E
0
) is quite
clear. Under nearly all technologically reasonable stressing conditions, P
b
family defects
dominate interface trap generation. The E
0
family plays a more complex role in device
operation. The defects clearly dominate oxide hole trapping when stressing conditions are
not particularly severe. However, in very heavily stressed oxides or in oxides subjected to
repeated sequences of electron–hole flooding, complex structural changes clearly take place
at the E
0
sites. Since the E
0
defects have levels around the middle of the oxide band gap, they
inevitably play an important role in oxide leakage current phenomena.
ß 2008 by Taylor & Francis Group, LLC.
References
1. Weil, J.A., Bolton, J.R., and Wertz, J.E., Electron Paramagnetic Resonance: Elementary Theory and
Practical Applications, John Wiley & Sons, New York, NY, 1994.
2. Nishi, Y., Study of silicon-silicon dioxide structure by electron spin resonance 1, Jpn. J. Appl.
Phys., 10, 52, 1971.
3. Nishi, Y., Tanaka, K., and Ohwada, A., Study of silicon–silicon dioxide structures by electron
spin resonance 2, Jpn. J. Appl. Phys., 11, 85, 1972.
4. Caplan, P.J. et al., ESR centers, interface states, and oxide fixed charge in thermally oxidized
silicon wafers, J. Appl. Phys., 50, 5847, 1979.
5. Poindexter, E.H. et al., Interface states and electron spin resonance centers in thermally oxidized
(111) and (100) silicon wafers, J. Appl. Phys., 52, 879, 1981.
6. Lenahan, P.M. et al., Radiation-induced trivalent silicon defect buildup at the Si–SiO
2
interface in
MOS structures, IEEE Trans. Nucl. Sci., 28, 4105, 1981.
7. Lenahan, P.M. and Dressendorfer, P.V., Effect of bias on radiation-induced paramagnetic defects
at the silicon–silicon dioxide interface, Appl. Phys. Lett., 41, 542, 1982.
8. Lenahan, P.M. and Dressendorfer, P.V., An electron spin resonance study of radiation-induced
electrically active paramagnetic centers at the Si=SiO
2
interface, J. Appl. Phys., 54, 1457, 1983.
9. Brower, K.L., Si-29 hyperfine-structure of unpaired spins at the Si=SiO
2
interface, Appl. Phys.
Lett., 43, 1111, 1983.
10. Lenahan, P.M. and Dressendorfer, P.V., Paramagnetic trivalent silicon centers in gamma-
irradiated metal-oxide-silicon structures, Appl. Phys. Lett., 44, 96, 1984.
11. Lenahan, P.M. and Dressendorfer, P.V., Hole traps and trivalent silicon centers in
metal=oxide=silicon devices, J. Appl. Phys., 55, 3495, 1984.
12. Poindexter, E.H. et al., Electronic traps and P
b
centers at the Si=SiO
2
interface: band-gap energy
distribution, J. Appl. Phys., 56, 2844, 1984.
13. Mikawa, R.E. and Lenahan, P.M., Structural damage at the Si=SiO
2
interface resulting from
electron injection in metal-oxide-semiconductor devices, Appl. Phys. Lett., 46, 550, 1985.
14. Gerardi, G.J. et al., Interface traps and P
b
centers in oxidized (100) silicon-wafers, Appl. Phys.
Lett., 49, 348, 1986.
15. Mikawa, R.E. and Lenahan, P.M., Electron-spin-resonance study of interface states induced by
electron injection in metal-oxide-semiconductor devices, J. Appl. Phys., 59, 2054, 1986.
16. Warren, W.L. and Lenahan, P.M., Electron-spin-resonance study of high-field stressing in metal-
oxide-silicon device oxides, Appl. Phys. Lett., 49, 1296, 1986.
17. Brower, K.L., Structural features at the Si=SiO
2
interface, Zeitschrift Fur Physikalische Chemie Neue
Folge, Z. Phys. Chem., Neue Folge, 151, 177, 1987.
18. Triplett, B.B., Takahashi, T., and Sugano, T., Electron-spin-resonance observation of defects in
device oxides damaged by soft x-rays, Appl. Phys. Lett., 50, 1663, 1987.
19. Kim, Y.Y. and Lenahan, P.M., Electron-spin-resonance study of radiation-induced paramagnetic
defects in oxides grown on (100) silicon substrates, J. Appl. Phys., 64, 3551, 1988.
20. Vranch, R.L., Henderson, B., and Pepper, M., Spin-dependent recombination in irradiated
Si=SiO
2
device structures, Appl. Phys. Lett., 52, 1161, 1988.
21. Miki, H. et al., Electron and hole traps in SiO
2
-films thermally grown on Si substrates in ultra-dry
oxygen, IEEE Trans. Electron Devices, 35, 2245, 1988.
22. Jupina, M.A. and Lenahan, P.M., Spin dependent recombination—a
29
Si hyperfine study of
radiation-induced P
b
centers at the Si=SiO
2
interface, IEEE Trans. Nucl. Sci., 37, 1650, 1990.
23. Krick, J.T., Lenahan, P.M., and Dunn, G.J., Direct observation of interfacial point-defects gener-
ated by channel hot hole injection in n-channel metal-oxide silicon field-effect transistors, Appl.
Phys. Lett., 59, 3437, 1991.
24. Awazu, K., Watanabe, K., and Kawazoe, H., Generation mechanisms of paramagnetic centers by
gamma-ray irradiation at and near the Si=SiO
2
interface, J. Appl. Phys., 73, 8519, 1993.
25. Gabrys, J.W., Lenahan, P.M., and Weber, W., High-resolution spin-dependent recombination
study of hot-carrier damage in short-channel MOSFETs—Si
29
hyperfine spectra, Microelectron.
Eng., 22, 273, 1993.
ß 2008 by Taylor & Francis Group, LLC.
26. Stesmans, A. and Afanas’ev, V.V., Electron spin resonance features of interface defects in
thermal (100) Si=SiO
2
, J. Appl. Phys., 83, 2449, 1998.
27. Stesmans, A. and Afanas’ev, V.V., Electrical activity of interfacial paramagnetic defects in
thermal (100) Si=SiO
2
, Phys. Rev. B, 57, 10030, 1998.
28. Mishima, T.D. and Lenahan, P.M., A spin-dependent recombination study of radiation-induced
P
b1
centers at the (001) Si=SiO
2
interface, IEEE Trans. Nucl. Sci., 47, 2249, 2000.
29. Mishima, T.D., Lenahan, P.M., and Weber, W., Response to comment on do P
b1
centers have
levels in the Si band gap? Spin-dependent recombination study of the P
b1
hyperfine spectrum,
Appl. Phys. Lett., 78, 1453, 2001.
30. Campbell, J.P. and Lenahan, P.M., Density of states of P
b1
Si=SiO
2
interface trap centers, Appl.
Phys. Lett., 80, 1945, 2002.
31. Lenahan, P.M. and Conley, J.F., What can electron paramagnetic resonance tell us about the
Si=SiO
2
system? J. Vac. Sci. Technol. B, 16, 2134, 1998.
32. Stesmans, A. and Afanas’ev, V.V., Comment on do P
b1
centers have levels in the Si band gap?
Spin-dependent recombination study of the P
b1
hyperfine spectrum, Appl. Phys. Lett., 78, 1451,
2001.
33. Weeks, R.A., Paramagnetic resonance of lattice defects in irradiated quartz, J. Appl. Phys., 27,
1376, 1956.
34. Silsbee, R.H., Electron spin resonance in neutron-irradiated quartz, J. Appl. Phys., 32, 1459, 1961.
35. Feigl, F.J., Fowler, W.B., and Yip, K.L., Oxygen vacancy model for E
0
1
center in SiO
2
, Solid State
Commun., 14, 225, 1974.
36. Griscom, D.L., Friebele, E.J., and Sigel, G.H., Observation and analysis of primary Si-29 hyper-
fine-structure of E
0
center in non-crystalline SiO
2
, Solid State Commun., 15, 479, 1974.
37. Griscom, D.L., E
0
center in glassy SiO
2
—microwave saturation properties and confirmation of
the primary Si-29 hyperfine-structure, Phys. Rev. B, 20, 1823, 1979.
38. Lenahan, P.M. and Dressendorfer, P.V., Microstructural variations in radiation hard and soft
oxides observed through electron-spin resonance, IEEE Trans. Nucl. Sci., 30, 4602, 1983.
39. Takahashi, T. et al., Electron-spin-resonance observation of the creation, annihilation, and charge
state of the 74-gauss doublet in device oxides damaged by soft x-rays, Appl. Phys. Lett., 51, 1334,
1987.
40. Warren, W.L. et al., Neutral E
0
centers in microwave downstream plasma-enhanced chemical-
vapor-deposited silicon dioxide, Appl. Phys. Lett., 53, 482, 1988.
41. Warren, W.L. and Lenahan, P.M., Si-29 hyperfine spectra and structure of E
0
dangling-bond
defects in plasma-enhanced chemical-vapor deposited silicon dioxide films on silicon, J. Appl.
Phys., 66, 5488, 1989.
42. Hazama, H. et al., Proceedings of the Workshop on Ultra Thin Oxides, Jpn. Soc. Appl. Phys., Tokyo,
207, 1998.
43. Lenahan, P.M. and Mele, J.J., E
0
centers and leakage currents in the gate oxides of metal oxide
silicon devices, J. Vac. Sci. Technol. B, 18, 2169, 2000.
44. Atkins, P., Molecular Quantum Mechanics, Chap. 14, 2nd edn., Oxford University Press, Oxford,
1983.
45. Slichter, C.P., Principles of Magnetic Resonance, 2nd edn., Springer-Verlag, Berlin, 1978.
46. Lepine, D.J., Spin-dependent recombination on silicon surface, Phys. Rev. B, 6, 436, 1972.
47. Kaplan, D., Solomon, I., and Mott, N.F., Explanation of large spin-dependent recombination
effect in semiconductors, J. Phys. Lett., 39, L51, 1978.
48. Hall, R.N., Electron-hole recombination in germanium, Phys. Rev., 87, 387, 1952.
49. Shockley, W. and Read, W.T., Statistics of the recombinations of holes and electrons, Phys. Rev.,
87, 835, 1952.
50. Lenahan, P.M. and Jupina, M.A., Spin dependent recombination at the silicon-silicon dioxide
interface, Colloids Surf., 45, 191, 1990.
51. O’Reilly, E.P. and Robertson, J., Theory of defects in vitreous silicon dioxide, Phys. Rev. B, 27,
3780, 1983.
52. Conley, J.F. et al., Observation and electronic characterization of new E
0
center defects in
technologically relevant thermal SiO
2
on Si: an additional complexity in oxide charge trapping,
J. Appl. Phys., 76, 2872, 1994.
ß 2008 by Taylor & Francis Group, LLC.
53. Mishima, T.D., Lenahan, P.M., and Weber, W., Do P
b1
centers have levels in the Si band gap? Spin-
dependent recombination study of the P
b1
hyperfine spectrum, Appl. Phys. Lett., 76, 3771, 2000.
54. Uchino, T. and Yoko, T., Density functional theory of structural transformations of oxygen-
deficient centers in amorphous silica during hole trapping: structure and formation mechanism
of the E
0
(gamma) center, Phys. Rev. B, 74, 2006.
55. Alemany, M.M.G. and Chelikowsky, J.R., Edge-sharing tetrahedra: precursors of the E-gamma
0
defects in amorphous silica, Phys. Rev. B, 68, 2003.
56. Conley, J.F. and Lenahan, P.M., Room-temperature reactions involving silicon dangling bond
centers and molecular-hydrogen in amorphous SiO
2
thin-films on silicon, IEEE Trans. Nucl. Sci.,
39, 2186, 1992.
57. Conley, J.F. and Lenahan, P.M., Molecular-hydrogen, E
0
center hole traps, and radiation-induced
interface traps in MOS devices, IEEE Trans. Nucl. Sci., 40, 1335, 1993.
58. Conley, J.F. and Lenahan, P.M., Radiation-induced interface states and ESR evidence for
room-temperature interactions between molecular-hydrogen and silicon dangling bonds in
amorphous SiO
2
-films on Si, Microelectron. Eng., 22, 215, 1993.
59. Vitko, J., ESR studies of hydrogen hyperfine spectra in irradiated vitreous silica, J. Appl. Phys., 49,
5530, 1978.
60. Tsai, T.E. and Griscom, D.L., On the structures of hydrogen-associated defect centers in irradi-
ated high-purity a-SiO
2
, J. Non-Cryst. Solids, 91, 170, 1987.
61. Conley, J.F., Lenahan, P.M., and Roitman, P., Proceedings of the IEEE SOI Technology Conference,
IEEE, New York, 12, 1991.
62. Conley, J.F., Lenahan, P.M., and Roitman, P., Electron-spin-resonance of separation by
implanted oxygen oxides—evidence for structural-change and a deep electron trap,
Appl. Phys. Lett., 60, 2889, 1992.
63. Conley, J.F., Lenahan, P.M., and Roitman, P., Evidence for a deep electron trap and charge
compensation in separation by implanted oxygen oxides, IEEE Trans. Nucl. Sci., 39, 2114, 1992.
64. Vanheusden, K. and Stesmans, A., Characterization and depth profiling of E
0
-defects in buried
SiO
2
, J. Appl. Phys., 74, 275, 1993.
65. Warren, W.L. et al., Paramagnetic defect centers in BESOI and SIMOX buried oxides, IEEE Trans.
Nucl. Sci., 40, 1755, 1993.
66. Conley, J.F. and Lenahan, P.M., Hydrogen complexed EP (E
0
delta) centers and EP=H
2
inter-
actions—implications for EP structure, Microelectron. Eng., 28, 35, 1995.
67. Conley, J.F. and Lenahan, P.M., Electron spin resonance analysis of EP center interactions with
H
2
: evidence for a localized EP center structure, IEEE Trans. Nucl. Sci., 42, 1740, 1995.
68. Griscom, D.L. and Friebele, E.J., Fundamental radiation-induced defect centers in synthetic
fused silicas—atomic chlorine, delocalized E
0
centers, and a triplet-state, Phys. Rev. B, 34, 7524,
1986.
69. Zhang, L. and Leisure, R.G., The E
0
(delta) and triplet-state centers in x-irradiated high-purity
amorphous SiO
2
, J. Appl. Phys., 80, 3744, 1996.
70. Chavez, J.R., et al., Microscopic structure of the E
0
(delta) center in amorphous SiO
2
: a first
principles quantum mechanical investigation, IEEE Trans. Nucl. Sci., 44, 1799, 1997.
71. Buscarino, G., Agnello, S., and Gelardi, F.M., Investigation on the microscopic structure of
E-delta
0
center in amorphous silicon dioxide by electron paramagnetic resonance spectroscopy,
Mod. Phys. Lett. B, 20, 451, 2006.
72. Stesmans, A., New intrinsic defect in as-grown thermal SiO
2
on (111) Si, Phys. Rev. B, 45, 9501,
1992.
73. Warren, W.L. et al., Hydrogen interactions with delocalized spin centers in buried SiO
2
thin-
films, Appl. Phys. Lett., 62, 1661, 1993.
74. Devine, R.A.B. et al., Point-defect generation during high-temperature annealing of the Si=SiO
2
interface, Appl. Phys. Lett., 63, 2926, 1993.
75. Griscom, D.L., Characterization of 3 E
0
-center variants in x-irradiated and gamma-irradiated
high-purity a-SiO
2
, Nucl. Instrum. Meth. Phys. Res. Sect. B, 229, 481, 1984.
76. Fitzgerald, D.J. and Grove, A.S., Surface recombination in semiconductors, Surf. Sci., 9, 347, 1968.
77. Alam, M.A. and Mahapatra, S., A comprehensive model of pMOS NBTI degradation, Microelec-
tron. Reliab., 45, 71, 2005.
ß 2008 by Taylor & Francis Group, LLC.
78. Chakravarthi, S. et al., A comprehensive framework for predictive modeling of negative bias
temperature instability, in Proceeding of the IEEE International Reliability Physics Symposium, 2004,
p. 273.
79. Huard, V., Denais, M., and Parthasarathy, C., NBTI degradation: from physical mechanisms to
modeling, Microelectron. Reliab., 46, 1, 2006.
80. Fujieda, S. et al., Interface defects responsible for negative-bias temperature instability in
plasma-nitrided SiON=Si(100) systems, Appl. Phys. Lett., 82, 3677, 2003.
81. Campbell, J.P. et al., Direct observation of the structure of defect centers involved in the negative
bias temperature instability, Appl. Phys. Lett., 87, 204106, 2005.
82. Campbell, J.P. et al., Observations of NBTI-induced atomic-scale defects, IEEE Trans. Device
Mater. Reliab., 6, 117, 2006.
83. Campbell, J.P. et al., Location, structure, and density of states of NBTI-induced defects in plasma-
nitrided pMOSFETs, in IEEE International Reliability Physics Symposium, Phoenix, AZ, 2007.
84. Campbell, J.P. et al., Identification of atomic-scale defect structure involved in the negative bias
temperature instability in plasma-nitrided pMOSFETs, Appl. Phys. Lett., 90, 123502, 2007.
85. Marquardt, C.L. and Sigel, G.H., Radiation-induced defect centers in thermally grown oxide
films, IEEE Trans. Nucl. Sci., 22, 2234, 1975.
86. Lipkin, L. et al., Correlation of fixed positive charge and E
0
gamma centers as measured via
electron injection and electron-paramagnetic resonance techniques, J. Electrochem. Soc., 138, 2050,
1991.
87. Walters, M. and Reisman, A., Radiation induced neutral electron trap generation in electrically
biased insulated gate field effect transistor gate insulators, J. Electrochem. Soc., 138, 2756, 1991.
88. Fleetwood, D.M. et al., Border traps—issues for MOS radiation response and long-term relia-
bility, Microelectron. Reliab., 35, 403, 1995.
89. Lelis, A.J. and Oldham, T.R., Time-dependence of switching oxide traps, IEEE Trans. Nucl. Sci.,
41, 1835, 1994.
90. Ma, T.P. and Dressendorfer, P.V., Ionizing Radiation Effects in MOS Devices and Circuits, John
Wiley & Sons, New York, 1989.
91. Oldham, T.R., Ionizing Radiation Effects in MOS Oxides, World Scientific, Singapore, 1999.
92. Fleetwood, D.M. et al., The role of border traps in MOS high-temperature postirradiation
annealing response, IEEE Trans. Nucl. Sci., 40(6), 1323, 1993.
93. Conley, J.F. et al., Electron-spin-resonance evidence for the structure of a switching oxide trap—
long-term structural-change at silicon dangling bond sites in SiO
2
, Appl. Phys. Lett., 67, 2179,
1995.
94. Conley, J.F. et al., Electron spin resonance evidence that E
0
(gamma) centers can behave as
switching oxide traps, IEEE Trans. Nucl. Sci., 42, 1744, 1995.
95. Lelis, A.J. et al., The nature of the trapped hole annealing process, IEEE Trans. Nucl. Sci., 36, 1808,
1989.
96. Warren, W.L. et al., Microscopic nature of border traps in MOS oxides, IEEE Trans. Nucl. Sci., 41,
1817, 1994.
97. Karna, S.P. et al., Electronic structure theory and mechanisms of the oxide trapped hole
annealing process, IEEE Trans. Nucl. Sci., 47, 2316, 2000.
98. Kohler, R.A., Kushner, R.A., and Lee, K.H., Total dose radiation hardness of MOS devices in
hermetic ceramic packages, IEEE Trans. Nucl. Sci., 35, 1492, 1988.
99. Stahlbush, R.E., Mrstik, B.J., and Lawrence, R.K., Postirradiation behavior of the interface state
density and the trapped positive charge, IEEE Trans. Nucl. Sci., 37, 1641, 1990.
100. Takagi, S., Yasuda, N., and Toriumi, A., A new I-V model for stress-induced leakage current
including inelastic tunneling, IEEE Trans. Electron Dev., 46, 348, 1999.
101. McPherson, J.W. and Mogul, H.C., Underlying physics of the thermochemical E model in
describing low-field time-dependent dielectric breakdown in SiO
2
thin films, J. Appl. Phys., 84,
1513, 1998.
102. Rosenbaum, E. and Register, L.F., Mechanism of stress-induced leakage current in MOS capa-
citors, IEEE Trans. Electron Devices, 44, 317, 1997.
103. Dumin, D.J. and Maddux, J.R., Correlation of stress-induced leakage current in thin oxides with
trap generation inside the oxides, IEEE Trans. Electron Devices, 40, 986, 1993.
ß 2008 by Taylor & Francis Group, LLC.
104. Warren, W.L. and Lenahan, P.M., Fundamental differences between thick and thin oxides
subjected to high electric-fields, J. Appl. Phys., 62, 4305, 1987.
105. Warren, W.L. and Lenahan, P.M., A comparison of positive charge generation in high-field
stressing and ionizing-radiation on MOS structures, IEEE Trans. Nucl. Sci., 34, 1355, 1987.
106. Weinberg, Z.A., Johnson, W.C., and Lampert, M.A., Determination of sign of carrier transported
across SiO
2
-films on Si, Appl. Phys. Lett., 25, 42, 1974.
107. Henderson, B., Defects in Crystalline Solids, Crane-Russek, New York, 1972.
108. Chiang, Y.M., Birnie, D., and Kingery, W.D., Physical Ceramics, John Wiley & Sons, New York,
1997.
109. Lenahan, P.M., Conley, J.F., and Wallace, B.D., A model of hole trapping in SiO
2
films on silicon,
J. Appl. Phys., 81, 6822, 1997.
110. Lenahan, P.M. and Conley, J.F., A comprehensive physically based predictive model for
radiation damage in MOS systems, IEEE Trans. Nucl. Sci., 45, 2413, 1998.
111. Conley, J.F. et al., Quantitative model of radiation induced charge trapping in SiO
2
, IEEE Trans.
Nucl. Sci., 44, 1804, 1997.
112. Conley, J.F., Lenahan, P.M., and McArthur, W.F., Preliminary investigation of the kinetics of
postoxidation rapid thermal anneal induced hole-trap-precursor formation in microelectronic
SiO
2
films, Appl. Phys. Lett., 73, 2188, 1998.
113. Lenahan, P.M. et al., Predicting radiation response from process parameters: verification of a
physically based predictive model, IEEE Trans. Nucl. Sci., 46, 1534, 1999.
114. Ohmameuda, T. et al., Thermodynamical calculation and experimental confirmation of the
density of hole traps in SiO
2
-films, Jpn. J. Appl. Phys., Part 2, 30, L1993, 1991.
115. Boureau, G. et al., Thermodynamic analysis of hole trapping in SiO
2
films on silicon, J. Appl.
Phys., 89, 165, 2001.
116. Capron, N. et al., Thermodynamic properties of the Si=SiO
2
system, J. Chem. Phys., 117, 1843,
2002.
117. Lenahan, P.M. and Conley, J.F., A physically based predictive model of Si=SiO
2
interface trap
generation resulting from the presence of holes in the SiO
2
, Appl. Phys. Lett., 71, 3126, 1997.
118. Lenahan, P.M., Atomic scale defects involved in MOS reliability problems, Microelectron. Eng.,
69, 173, 2003.
119. Hu, G.J. and Johnson, W.C., Relationship between x-ray-produced holes and interface states in
metal-oxide-semiconductor capacitors, J. Appl. Phys., 54, 1441, 1983.
120. Dimaria, D.J. and Cartier, E., Mechanism for stress-induced leakage currents in thin silicon
dioxide films, J. Appl. Phys., 78, 3883, 1995.
121. Rofan, R. and Hu, C.M., Stress-induced oxide leakage, IEEE Electron Device Lett., 12, 632, 1991.
122. Wilk, G.D., Wallace, R.M., and Anthony, J.M., High-kappa gate dielectrics: current status and
materials properties considerations, J. Appl. Phys., 89, 5243, 2001.
123. Green, M.L. et al., Ultrathin (<4 nm) SiO
2
and Si–O–N gate dielectric layers for silicon
microelectronics: understanding the processing, structure, and physical and electrical limits,
J. Appl. Phys., 90, 2057, 2001.
124. Kirsch, P.D. et al., Nucleation and growth study of atomic layer deposited HfO
2
gate dielectrics
resulting in improved scaling and electron mobility, J. Appl. Phys., 99, 2006.
125. Kang, A.Y. et al., Electron spin resonance study of interface defects in atomic layer deposited
hafnium oxide on Si, Appl. Phys. Lett., 81, 1128, 2002.
126. Kang, A.Y., Lenahan, P.M., and Conley, J.F., The radiation response of the high dielectric-
constant hafnium oxide=silicon system, IEEE Trans. Nucl. Sci., 49, 2636, 2002.
127. Triplett, B.B. et al., Electron spin resonance study of as-deposited and annealed (HfO
2
)
x
(SiO
2
)
(1–x)
high-kappa dielectrics on Si, J. Appl. Phys., 101, 013703, 2007.
128. Stesmans, A. and Afanas’ev, V.V., Si dangling-bond-type defects at the interface of (100) Si with
ultrathin HfO
2
, Appl. Phys. Lett., 82, 4074, 2003.
129. Ryan, J.T. et al., Electron spin resonance observations of oxygen deficient silicon atoms in the
interfacial layer of hafnium oxide based metal-oxide-silicon structures, Appl. Phys. Lett., 90,
173513, 2007.
ß 2008 by Taylor & Francis Group, LLC.
7
Oxide Traps, Border Traps, and Interface
Traps in SiO
2
Daniel M. Fleetwood, Sokrates T. Pantelides, and Ronald D. Schrimpf
CONTENTS
7.1 Introduction........................................................................................................................ 215
7.2 Measurement Techniques ................................................................................................. 218
7.3 Oxide-Trap Charge Energies............................................................................................ 219
7.3.1 Trapped-Hole Energy Distributions..................................................................... 219
7.3.2 Comparison of Radiation Response and 1=f Noise in SiO
2
.............................. 221
7.3.3 Defect Microstructure............................................................................................. 222
7.3.4 Energy Scale for 1=f Noise..................................................................................... 226
7.4 Dipolar Defects near the Si=SiO
2
Interface .................................................................... 229
7.4.1 Switched-Bias Annealing....................................................................................... 229
7.4.2 Capacitance–Voltage Hysteresis........................................................................... 231
7.4.3 Stable Dipoles in the Near-Interfacial SiO
2
......................................................... 233
7.5 Interface Traps.................................................................................................................... 235
7.5.1 Two-Stage Interface-Trap Buildup Model........................................................... 235
7.5.2 Hydrogen Transport and Reactions..................................................................... 238
7.6 Aging and Scaling Effects................................................................................................. 242
7.6.1 Aging Effects............................................................................................................ 242
7.6.2 Scaling Effects and Emerging Materials .............................................................. 245
7.7 Summary and Conclusions .............................................................................................. 247
Acknowledgments ..................................................................................................................... 248
References.................................................................................................................................... 248
7.1 Introduction
In this chapter, we discuss defects in the critical bulk and near-interfacial SiO
2
regions of
a metal-oxide-semiconductor (MOS) device or integrated circuit (IC). This discussion is
derived mostly from experience in evaluating MOS radiation response, and therefore
applies most directly to the performance and reliability of electronics in radiation environ-
ments. However, the defects that limit the radiation response of a device also can signifi-
cantly affect its reliability outside a radiation environment. Hence, radiation exposure can
be a very effective tool in MOS defect analysis, and the lessons learned from systems that
ß 2008 by Taylor & Francis Group, LLC.
must survi ve harsh radiat ion environ ments often enable insight int o the defects that also
limit MOS perfo rmance and long-ter m reliabilit y.
Whe n a MOS devi ce or IC is expo sed to ionizin g rad iation, elect ron –hole pai rs are
cre ated in the transi stor ga te oxide , and in other (pa rasitic) insula ting laye rs of the devices.
This proce ss is illustrat ed schem atically in Figu re 7.1. Unde r pos itive gate bias at ro om
tem perature, radiation- induce d elect rons rapidly transpo rt to the gate and leave the ox ide,
whil e holes transport slowly to ward the Si . A fract ion of these holes is trappe d near the
Si=Si O
2
inter face, lead ing to a shift in the thresho ld voltage of the transisto r [1]. Du ring
the hole transp ort and trapping proces ses, hyd rogen is rele ased wi thin the oxide, and
under suitable bias cond itions may transp ort to the interface and react wi th Si dan gling
bon ds, formi ng inter face traps [2 –5]. Interface trap s shift transisto r thres hold voltage s and
degrad e chann el carrier mobilit ies. Some positive charge that is trapped nea r the inter face
can induce compe nsating electron traps, which are of ten c alled border traps [6 –9]. Fast er
bor der traps are someti mes mistake n fo r inter face traps in studies of MOS pe rformance ,
reli ability, and radiat ion resp onse [7–9].
In the MOS defect literatu re, there are wide variet ies of nomen clatur es used to charac -
terize defects in mate rials, devices, and ICs. The differe nt terms that are of ten use d to
descri be defects that are simi lar or even identical in micr ostruct ure can vary with the
met hod of charac terizat ion, the effect of the defe ct on the device of interest, the ba ckgroun d
of the inv estigator, the conve ntion of the partic ular technical commu nity, and many
othe r factors. In Figu re 7.2, a repres entative samplin g is provided of many pres ent and
histor ical terms that are used to describ e defects in the Si =Si O
2
system. Th e reader wi ll see a
vari ety of these ter ms use d in this chapte r and book that reflects the divers ity of opinion
and usag e in the modern literature . In Figu re 7.2, the defec ts are group ed sch ematica lly
with their physi cal locati on in the Si bulk , at the Si=Si O
2
inter face, in the nea r-interfac ial
SiO
2
, in the Si O
2
bulk , or (in high ly sca led devi ces), at the ga te =SiO
2
interface . For a mo re
exten sive discussi on of MOS defe ct nomen clatur e, please see Refs. [6 –8, and refer ences
there in].
In this chap ter, we fi rst discuss brie fl y in Se ction 7.2 the m easuremen t techniques
that are used to distinguish the effects of MOS oxide, border, and interface-trap charge.
We focus primarily on thermally stimulated current (TSC) methods to estimate MOS
Gate Oxide Sub
Oxide traps
Border traps
Interface traps
V
+
+
+
+
+
+
+
+
+



+
FIGURE 7.1
Schematic illustration of ionizing radiation effects on a MOS transistor. Irradiation under positive bias leads to the
generation of electron–hole pairs. The electrons transport rapidly to the gate; the holes transport via a random
walk process toward the Si=SiO
2
interface. During hole transport, protons can be liberated which can also
transport to the interface and react with Si–H bonds to form interface traps. A subset of the trapped holes induce
near-interfacial oxide-trap charge (border traps), which are defects that lie within the oxide, but are physically
close enough to the Si that they can exchange charge with the channel on the timescale of the measurements being
performed. (From Fleetwood, D.M., IEEE Trans. Nucl. Sci., 39, 269, 1992. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
oxide -trap charge dens ities and energy dis tributions , the result s of whi ch are then dis-
cussed in detail in Secti on 7.3. We also discuss the micr ostruct ures of defects in the near-
inter facial SiO
2
that contribut e to MOS oxide-tra p charge and low-fre quency no ise. We
discuss the nature of dipoles in the near-i nterfacial SiO
2
in Secti on 7.4; some of these
dipole s are able to exchange charge easily with the unde rlying Si, and others are muc h
more st able ag ainst both bias- and tem perature-d ependen t anne aling. In Se ction 7.5, we
brie fly review the eff ects of hydro gen on the buildu p of MOS inter face-tra p charge . Fina lly,
in Se ction 7.6, we disc uss agin g and devi ce scaling eff ects on the rad iation respo nse and
long-term reliability of MOS devices and ICs. Experimentally observed changes in MOS
radiation response and=or inferred long-termreliability with time are attributed to changes
in distributions of defects and=or impurities (especially hydrogenous species) with time
and=or device thermal history. We conclude that the inferred radiation response or long-
term reliability of devices with mobile impurities (e.g., hydrogen) can change significantly
with time. Hence, efforts to manage MOS performance, reliability, and=or radiation
response via defect engineering should be viewed with caution and checked for long-
term stability. The same kinds of defects are observed in highly scaled devices, e.g., with
ultrathin oxides, as are observed in previous generations of devices that have been studied
extensively. However, their effects on device response can differ—for example, instead
of oxide-trap charge shifting the device threshold voltage in a highly scaled circuit or
device, it instead may increase gate leakage current or device noise. In addition, the
total dose response of many deep submicron ICs is dominated by source-to-drain or
device-to-device leakage currents, rather than charge trapping in ultrathin gate oxides.
In these ICs, charge trapping and defect buildup in field oxides and shallow trench
isolation (STI) regions is important; these isolation oxides frequently exhibit degradation
characteristics that are similar to thick gate oxides from older technologies. Hence, the
Oxide traps
Surface Interface
Gate
SiO
2
Si
traps/states traps/states
Hole traps Bulk traps
Fast Interface
P
b
defects/centers
Electron traps
Mobile ionic charge
neutral electron traps
EЈ defects/centers Neutral centers
Surface
traps/states states dangling Si bonds traps/states
Recombination centers
Bulk Si defects Dopant atoms
Border traps
Fixed oxide charge Switching oxide traps
Anomalous positive charge Slow traps
Slow states Near-interface oxide traps
FIGURE 7.2
Schematic diagram that associates common names applied to defects in MOS devices with SiO
2
gate dielectrics
with their approximate spatial locations. The diagramis not to scale. Although surface traps and states usually are
reserved for cases in which there is truly a surface, in contrast to an interface, these terms are often used
synonymously with interface traps and states. There have been efforts to standardize defect nomenclature with
varying amounts of success. But one still finds all of these terms, and more, in the MOS defect literature.
ß 2008 by Taylor & Francis Group, LLC.
body of kno wledge that has been deve loped over the past 40 or mo re year s on defects in
MOS systems is still quite relevant and applicabl e to modern and future MOS circuits and
devi ces; howeve r, the same defects can lead to differe nt kinds of respo nses, and new
mate rials int roduce new kinds of defects and impurit ies that can also affect M OS perform -
ance , long-ter m reliabilit y, and radiat ion resp onse.
7. 2 Measurement Techniques
Dens ities of net oxide-tra p charge and inter face-tra p charge are most of ten estimate d
via simp le capacita nce –voltage (C –V) and curren t–v oltage ( I–V ) technique s [10 –12]. Th e
mo st popul ar techniq ues rely on the assumpti on that inter face traps are approxi mately
charge neutral at midg ap surfa ce potenti al [11,13,1 4], allowing on e to estimate the net
oxide -trap charge from midgap voltage shift s in C–V or I –V measu remen ts. Becau se the re
can be signi ficant contribut ions to midg ap voltage shifts from nega tive charge in border
traps [6 –9], and becaus e standard C –V and I –V analy sis do not provide informati on
about trapped hole energy distrib utions, other techniques (e.g., TSC [15,16 ], as discusse d
in Secti on 7.3) must be empl oyed to obt ain informati on ab out oxide-tra p charge that is
mo re easi ly interprete d in a strai ghtforward mann er. Char ge-pu mping and c onducta nce
met hods are amo ng othe r tech niques use d to esti mate densities and energy distrib utions
of inter face traps [10 –12,17, 18]. Techn iques that are mo re amenab le to rap id (e.g., less than
$ 1 ms ) measu remen t times are more easily a ble to discri minate the effects of interface trap s
(wh ich typically exc hange charge with the unde rlying Si faste r than bor der traps , which by
de finit ion lie within the oxide , a nd therefore are not in direct elect rical con tact with carrier s
in the Si channel [6 –9]) than a re techniq ues that require longe r measuri ng times. This
differe nce has been exploited to develop met hods to (at least first order) separate the effect s
of inter face and border traps [8,19,20 ], as discuss ed furthe r bel ow.
TSC and related tech niques are often used to study the defec ts and impu rities that limit
the long-t erm reliabilit y and radiat ion toleranc e of electro nic ma terials and devi ces
[15,16,2 1–56]. In MOS capa citor s wi th high-qu ality Si O
2
insulato rs that are thick er than
about 7 nm, it is often possibl e to use TSC to estim ate both the density and energy
distri bution of trap ped posit ive charge fo r devices expos ed to ionizin g radiat ion or high-
fi eld elect rical stress [7,14 –16,24 –27,29 –55]. This is pos sible primar ily becaus e the dominan t
charge traps in the rmal ox ides tend to be at or near the Si =SiO
2
interface [4,5,30] and the
TSC in these mate rials tend s to be almost entir ely due to trappe d hole emission and
transp ort [21,22,2 5–27,30 ,32,33]. For oxide s thi nner than $ 7 nm, dif ficulties in obtaining
high enou gh densi ties of trapped oxide c harge and low enough basel ine cur rents in the
absence of trapped oxide charge, owin g to the natu rally high tun nel currents in ultrat hin
oxide s, make it impo ssible to use this tech nique and less dire ct m ethods to estimate the
oxide -trap charge must be used.
Figu re 7.3 schem atically illustrat es the appl ication of TSC to a MOS capa citor that has
been exposed to ionizin g radiat ion or high -fi eld st ress. Unde r suitable bias cond itions,
radiat ion- or stress-i nduced trapped posit ive charge is trapped in finitesi mally close to the
Si=Si O
2
int erface . Then, ramped tem perature measure ments at a suf ficientl y large negativ e
bias allow the density of trapped posit ive charge to be esti mated by int egrating the current
due to charge transpo rt across the ox ide, and the trap ped-char ge energy distri bution can
be estimate d from the tempe rature dep endence of the current. Becau se interface traps
exchange charge with the Si without displacement current, charge released from interface
traps is not sensed by TSC measurement [27]. Thus, TSC is not useful for characterizing
interface-trap charge. When one exposes MOS capacitors to ionizing radiation or high-field
ß 2008 by Taylor & Francis Group, LLC.
stress, bot h pos itive and nega tive charge s are trap ped in the SiO
2
, although the dominan t
species are almost always positive , and usually dominat ed by trapped holes
[1,4,5,13 ,15,16,29 –34,37, 38,44 –46,49]. Th e currents for typical TSC tend to be quite sm all,
on the order of 0.1 –10 pA for typical devices and irrad iation or stress cond itions [15,16,3 0].
Thus, there are signi fi cant pra ctical dif ficulti es on e must overcom e to be able to accu rately
measure and inter pret TSC in SiO
2
, but in ma ny cas es these diffi culties have bee n
overcom e and accurate esti mates of MOS ox ide-trap charge energy dis tributions and
densi ties have been obtain ed [15,16,3 2].
7.3 O xide-Trap Charge Energies
7.3.1 Trapped -Hole Energy Distrib utions
Oxide traps are distribut ed broad ly in space and energy, as expecte d for an amorp hous
dielectri c layer [1,32]. Figu re 7.4 illus trates trap ped-hol e energ y distrib utions as estimate d
by TSC for several kinds of ox ides that were proces sed in diffe rent ways. The integr ated
TSC provid es a n esti mate of the densi ty of radiat ion-indu ced trapped pos itive charge ,
which is a fun ction of device proce ssing, rad iation dose , bias, dose rate , and several ot her
factors, as discuss ed in detail in Refs. [1,4,5,32 ]. Regardl ess of the trap ped charge densi ties,
the overall shapes of the infer red trapped -charg e energy distrib utions are similar , consist-
ent with the ide a that the dominan t defect types are simi lar in the different ty pes of oxide s
(in most thermal oxides, the dominant hole trap is known to be an O vacancy in SiO
2
[13,40], as discussed further below), at least to first order [13,32]. The energy scale on the
upper x-axis is derived fromthe current–temperature measurements under the assumption
Si
Gate
+ + + + +
− − −
A
Heat
FIGURE 7.3
Schematic illustration of the TSC technique to estimate the density and energy of MOS oxide-trap charge.
Irradiation or high-field stress can lead to hole trapping in the oxide, near the Si=SiO
2
interface, under appropriate
bias conditions (e.g., positive-bias irradiation). When the capacitor is heated under negative bias, holes are emitted
and transport across the oxide. Electrons in border traps do not contribute significantly to the measured TSC,
allowing one to differentiate the density of trapped holes from the net oxide-trap charge. (After Fleetwood, D.M.
et al., Microelectron. Reliab. 39, 1323, 1999. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
that the excess TSC above the background observed in unirradiated devices is caused by
simple hole emission from oxide traps that are distributed in energy and subsequent
hole transport without retrapping across the SiO
2
. With this assumption (which will be
revisited below), a generalized activation model has been developed [32,33] to estimate
effective trapped hole energies for arbitrary heating rates, and including the effects of
electric field-induced barrier lowering. Briefly, the TSC emission process is modeled as
Schottky emission of the trapped hole over a fixed energy barrier, which corresponds to the
effective trap depth in the SiO
2
. The rate at which traps are emptied is described by
@n(w,t)=@t ¼ Àn(w,t) F(t) exp[À(w ÀR(t))=A*(t)] (7:1)
where
n(w,t) is the number of holes that remain trapped per unit area per unit energy
w is the effective trapped hole energy
t is the annealing time
A*(t) ¼kT(t)=q, where k is the Boltzmann constant and T is the absolute temperature
R(t) ¼b(E
ox
)
1=2
, where b¼0.5 (q=p«
ox
)
1=2
for Schottky emission, Àq is the electronic
charge, «
ox
is the SiO
2
dielectric constant, and E
ox
is the oxide electric field applied
during the TSC measurement
F(t) ¼a*T(t)
2
is analogous to the attempt-to-escape parameter n that is typically used to
characterize thermally activated processes that exhibit Arrhenius (e.g., simple
activated emission over a single barrier) response [32,33]
In Equation 7.1, the value of a* is given by [32,33,57]:
a* ¼ 2s
t
g(3k=m
h
)
1=2
(2pm
h
k=h
2
)
3=2
(7:2)
Soft 350 nm, 20 K
Hard 45 nm, 2 M
Wet 23 nm, 2 M
Dry 25 nm, 1 M
AT&T 18 nm, 5 M
1.2 1.4 1.6
Energy (eV)
10
1
0.1
T
S
C

(
p
A
)
0.01
0 50 100 150
Temperature (ЊC)
200 250 300 350
1.8
(I)
(IV)
(II)
(III)
(V)
2.0 2.2
FIGURE 7.4
TSC as a function of oxide processing and radiation dose. Capacitors of different areas from different processes
were irradiated to different doses [rad(SiO
2
)] at electric fields of 1 or 2 MV=cm with 10 keV x-rays to obtain these
data. All TSC spectra were measured at an electric field of À2 MV=cm at a heating rate of $0.11 K=s. (After
Fleetwood, D.M. et al., IEEE Trans. Nucl. Sci., 39, 2192, 1992. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
where
s
t
is the captur e cross section of the dominan t ho le trap in SiO
2
g is the numb er of equiva lent state s into which a hole may be om itted
m
h
is the eff ective mass of a hole in SiO
2
h is the Planck cons tant
To obtain an or der-of-mag nitude esti mate for a*, let s
t
% 10
À13
cm
2
be the hole capt ure
cross sectio n [1,58,59 ], g % 1, and m
h
% m, whe re m is the fre e elect ron mass. Wit h these
simplify ing assump tions, we esti mate a* % 3 Â 10
8
K
À2
s
À 1
. This value of a* cor respond s to
a value of F( t ) of $ 3 Â 10
13
Hz at room temperat ure and $7.5 Â 10
13
Hz a t $ 500 K, which is
the approxima te tem perature at whi ch the TSC curves peak (i.e., the energy at which
the highest trapped ho le emis sion rates are obse rved) in Figu re 7.4. To withi n the unc er-
tainti es in the assump tions underly ing this analysis and the uncerta inties in esti mating F( t )
or n expe rimen tally, thi s va lue is cons istent with experi mental estimate s of n based
on systemat ic variation of the TSC peak posit ion with change s in the heatin g rate
[32,33,5 0], which emphasi zes the self-consi stenc y of the effect ive trapped-h ole energy
scale in Figu re 7.4.
7.3.2 Compa rison of Radiati on Respo nse a nd 1=f Nois e in SiO
2
The fairly straightforw ard mo del presen ted in Secti on 7.3.1 is at best only a fi rst-orde r
descri ption of the defe cts in SiO
2
and their trap ping and emission prope rties, as demon-
strated in rec ent wo rk that com pares the rad iation resp onse, TS C, and low-freq uency noise
of MOS devices with thermal Si O
2
gate oxide s. This com parison provid es signi ficant
insight into oxide-tra p micr ostruct ure a nd the dynami cs of charge trapping and emis sion.
In particul ar, several studies have compar ed the 1=f noise and radiat ion respons e of
MOS devices [7,19,55 ,59 –69]. An empi rically obse rved corre lation betwe en the normaliz ed
1=f noise K of M OS transis tors before irrad iation and thres hold voltage shifts due to net
posit ive radiat ion-indu ced oxide -trap charge DV
ot
is shown in Figu re 7.5 [59,60] . Here
K ¼ S
V
f( V
GS
À V
TH
)
2
=V
DS
2
; f is the frequency, S
V
is the exc ess voltage-n oise pow er spe ctral
densi ty (after correcti on for backgro und leak age), V
TH
is the thres hold voltage, and
V
GS
and V
DS
are the gate and drain v oltages during the noise measu remen ts [61,62] .
Thre shold voltage shift s due to oxide and inter face trap charge were esti mated via the
midgap met hod of Winokur et al. [11].
The correlat ion in Figure 7.5 can be described num erically via the follo wing expression
derive d from a simp le, fi rst-orde r mo del of the no ise that attribu tes the excess low-
frequency noise primarily to fluctuations in the number of channel carriers owing to
trapping and emission events associated with near-interface oxide (border) traps [59–62]:
K ¼
q
2
kTf
ot
t
2
ox
LW s
t
«
2
ox
E
g
ln (t
max
=t
min
)
(7 : 3)
where
f
ot
is the Si O
2
hole-tr apping effi ciency (that is, the probabi lity that a given hole create d
by ionizing radiation exposur e is trapped)
t
ox
is the oxide thickne ss
L is the chann el length
W is the channel width
E
g
is the SiO
2
band gap
t
max
and t
min
are the presumed cutoff times for the noise process [59–62]
ß 2008 by Taylor & Francis Group, LLC.
Equati on 7.3 assu mes (1) defects with similar averag e, eff ective capt ure cross sectio ns s
t
are respons ible for both 1=f noise a nd radiat ion-indu ced-hole trapping, (2) the prei rradia-
tion no ise S
V
is propo rtional to the densi ty of oxide traps , which is in turn prop ortion al
to f
ot
=s
t
, (3) oxide trap s near the Si =SiO
2
interface are distrib uted approximat ely uniform ly
in space and energy, (4) carrier num ber fl uctuations are the dominan t cause of noise ,
and (5) that border trap s have similar capture and emi ssion cross secti ons to the
bulk ox ide traps that are locate d deeper within the SiO
2
laye r [55,59 –62]. A signi ficant
body of work in the literatu re sugge sts strong ly that O vacancie s play a key ro le in thi s
proces s [5,13,40 ,55,63].
7.3.3 Defec t Micr ostruct ure
The crucial signi fic ance of O vacancy- related defects to MOS oxide -trap charge has
been recognize d fo r a long tim e [1,13,40 ,55]. As an example of the signi ficant impact that
O vaca ncies can have on MOS oxide -trap charge, Figure 7.6 com pares electri cal estimate s
of MOS oxide-tra p charge [70] with calcul ations using a simp le Fick ’s law diffusion mo del
of O vacancy formatio n [71], ba sed on the idea that exc ess O vacancy cre ation occurs in
these devi ces primarily as a result of O atoms out-diffu sing fro m the SiO
2
duri ng high-
tem perature process steps after gate oxidati on. Whi le other mechanis ms of O vacancy
formati on in SiO
2
exis t, the cor relation betwe en the experi mental dat a and the calcu lations
in Figure 7.6 provid e st riking con firmatio n of the impo rtance of O vacanci es to MOS
oxide -trap charge, and cons equentl y to MOS 1=f noise [59 –63].
In Figu re 7.7, we show four types of electroni c proces ses of rele vance to the corre lation
betwe en low- frequency noise and MOS radiat ion respons e. Proce ss 1 is an elect ron
neutralizing-trapped positive charge under the proper electric field conditions, which is
known to have a capture cross section that is on the order of $10
À13
cm
2
, consistent with
the above assumptions [1,72,73]. Process 2 in Figure 7.7 depicts hole capture (i.e., valence
10
−8
10
−9
10
−10
K

(
V
2
)
10
−11
0.1 0.3 1.0
−ΔV
ot
(V)
A
B
C
D
E
3.0 10
FIGURE 7.5
Normalized noise magnitude K as a function of threshold-voltage shifts due to radiation-induced oxide-trap
charge DV
ot
for 3 Â 16 mm, nMOS transistors with gate oxides of different thickness (A, D: 32 nm; B, E: 48 nm;
C: 60 nm) and radiation hardness (A–C hard; D, E soft) processed in the same lot. Noise measurements were
performed in the linear region of device operation; values of DV
ot
were obtained from room-temperature
irradiation to 100 krad (SiO
2
) in a Co-60 source at a dose rate of $278 rad (SiO
2
)=s at an oxide electric field of
$3 MV=cm. (From Fleetwood, D.M. and Scofield, J.H., Phys. Rev. Lett., 64, 579, 1990. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
11.0
8.8
6.6
4.4
2.2
0.0
750 800 850
Temperature (ЊC)

Δ
V
o
x

(
V
)

Δ
V
o
t

(
V
)
ΔV
ox
ΔV
ot
900 950 1000
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
FIGURE 7.6
Estimated magnitudes of midgap voltage shifts DV
ot
(squares, right-hand side y-axis) and predicted shifts DV
ox
(circles, left-hand side y-axis), due to Ovacancies in thermal SiO
2
films that were given postgate, 30 min anneals in
N
2
at different temperatures (x-axis). The experimental data are obtained from 1 Mrad (SiO
2
) x-ray irradiations
of capacitors with 45 nm oxides at an electric field of $2 MV=cm, and the model predictions are derived from a
Fick’s value diffusion model. (From Warren, W.L. et al., Appl. Phys. Lett., 64, 3452, 1994. With permission.)
O
Si +
O
O O O O O O O O
O O
O O O O O O
O
O
Si
Si +
Si
Si
Si
Si
O
e−
e−
SiO
2
SiO
2
Si
Si
e−
(2) (1)
(3)
(4)
O
O
O
Si
e−
FIGURE 7.7
Schematic illustrations of (1) an electron from the Si neutralizing an E
0
center (specific type not specified here); (2)
an electron from an E
0
precursor emitted from the O vacancy center, which will leave behind the positively
charged defect shown in (1); (3) noise due to electron exchange between the Si and a positively charged E
0
defect;
and (4) noise due to electron exchange between the Si and a neutral E
0
precursor (O vacancy in SiO
2
). All of these
processes can have similar effective cross sections. (From Fleetwood, D.M. et al., IEEE Trans. Nucl. Sci., 49, 2674,
2002. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
elect ron emission), which has a similar cross secti on [1,58,73 ]. Process es 1 and 2 con stitute a
full captur e and emission cy cle for a defe ct near the Si =SiO
2
interface ; hence , it is qui te
rea sonable that the se kinds of defects, located at suitable positions and energy levels, may
play a key role in MOS 1=f no ise, as long as simi lar kinds of charge captur e and emi ssion
events can origi nate wi thin the Si channel. A comple te noise cycle that starts wi th a
posit ively charge d defect is illus trated by proces s 3 of Figu re 7.7; proces s 4 is a capture
and emis sion seq uence involvi ng a n initial ly neutr al site [55].
At least two pla usible cand idates fo r the O vacancy center s associ ated with proces ses 1–4
in Figure 7.7 have been ide nti fied; E
0
g
and E
0
d
defects are known to be high capture-
cross-section hole traps in SiO
2
[1,40,74]. In several studies, the E
0
d
has been associated
with the dimer O vacancy defect that is illustrated in Figure 7.8a [55,75–80]; however, this
association has been qualified and=or questioned as a result of subsequent embedded
cluster and density functional theory (DFT) calculations [81,82] and recent Si
29
hyperfine
data [83,84]. The E
0
g
defect has been associated with the defects shown in Figure 7.8b and c
[78–82,85–89]. Final and conclusive identification of the E
0
g
and E
0
d
defect microstructures
awaits future work; nevertheless, the dimer and puckered defects depicted in Figure 7.8 are
found via DFT calculations to have several characteristics that emphasize their significance
to comparative studies of MOS radiation response and low-frequency noise.
Recent DFT calculations [78,79] of the energetics of hole and=or electron capture
by neutral and=or positively charged O vacancy centers show that: (1) the vast majority
of O vacancy sites (as many as $90%) in the supercells we have examined are dimers
(Figure 7.8a) after hole capture [79]. This percentage may vary with device processing, but
it suggests that the dimer configuration of the O vacancy center is the more common
vacancy-related defect in SiO
2
. (2) An initially neutral dimer can efficiently capture a hole,
but the energy levels associated with this kind of hole trap typically are relatively shallow
($1.0 eV) compared to those of either of the two puckered configurations (Figure 7.8b
and c, !$3 eV) [79]. An exception is a stretched dimer ($2 eV) with no O available to form
a puckered E
0
g
; such a defect would not be observed in quartz-like SiO
2
, but may well
be present in regions of the oxide of low density [89] and=or near voids. (3) Electron
paramagnetic resonance (EPR) signals fromthe two puckered configurations in Figure 7.8b
O1
O1
O1
Si0
(a) (b) (c)
Si1
Si1
Si0 Si0
O2 O2
O2
O4
O3
O3
O3
Vo
Vo Vo
Si1
FIGURE 7.8
Schematic illustrations of the unpaired electron densities (gray regions) and atomic configurations of (a) a dimer O
vacancy center, (b) a relaxed O vacancy center that has been associated with the E
0
g
defect (the E
0
g4
), and
(c) a second type of O vacancy center that apparently is also associated with the E
0
g
(the E
0
g5
). The differences
between defects (b) and (c) are the coordination of the atom denoted by the arrows and marked ‘‘Si1,’’ with
fourfold coordination in (b) and fivefold coordination in (c). (FromLu, Z.Y. et al., Phys. Rev. Lett., 89, 285505, 2002.
With permission.)
ß 2008 by Taylor & Francis Group, LLC.
and c are likely quite simi lar, since the only difference in defect con figura tion is the
coor dination of the Si ato m faci ng (not ho lding) the unpai red electron that is EPR active.
But the fourf old coor dinat ed, puck ered Si vacancy center (Figu re 7.8b) forms a stable dipole
upon electron capture, while the fivefold coordinated, puckered (Figure 7.8c) does not form
a dipole [79]. Whether an O vacancy center forms a fourfold or fivefold puckered configur-
ation is determined largely by the local atomic spacing and bond angles; these determine
whether O4 has a nearest neighbor Si close enough to bond with the puckered Si1.
DFT calculations also demonstrate that the probability of capturing an electron (e.g.,
from the Si) increases with increasing separation of the Si1–Si0 bond at the center of the
complex shown in Figure 7.8a, as shown schematically in Figure 7.9. At an equilibrium
spacing of $0.25–0.30 nmin the bulk SiO
2
, the electron trapping level for the neutral dimer
in Figure 7.8a is near the SiO
2
conduction band. However, if one significantly stretches
the Si–Si bond to $0.35–0.4 nm, near-midgap states can open up that may at least
metastably capture an electron, with the energetics for capture becoming more favorable
with increasing Si–Si spacing. These kinds of stretched Si–Si bonds are likely exist near the
Si=SiO
2
interface [79,81,82], in which a significant amount of strain must be accommodated
[30,70,89–93]. Moreover, it has been demonstrated that the amount of strain near the
Si=SiO
2
interface can change with radiation exposure [93].
Several kinds of charge exchanges between the Si and O vacancy-related defects such as
those depicted in Figure 7.8 can lead to 1=f noise. (1) In a pMOS transistor, the simple
capture and reemission of a hole from a dimer O vacancy defect near the Si=SiO
2
interface
is an excellent candidate for 1=f noise. This defect has a high effective capture cross section,
a modest barrier for reemission, and thus can contribute to pMOS noise before or after
irradiation [8,55,65]. The charge transfer may occur via simple tunneling, trap-assisted
tunneling, and=or thermal activation. Indeed, this may well be the dominant source of
pMOS noise, although similar transitions with puckered O vacancy defects involving
network relaxation may also play a significant role in pMOS noise. (2) In an irradiated
nMOS transistor, electron exchange with fourfold-coordinated, puckered O vacancy
defects is a possible source of noise. However, dipolar defects in irradiated SiO
2
can be
quite stable, and electrons in some dipolar defects associated with trapped holes are found
in TSC studies not to be reemitted to the Si even when the devices are heated to $1008C
under negative bias [52]. So this is a potential source of noise, but simple capture and
emission of an electron from a fourfold-coordinated, puckered O vacancy defect may not
E
F
Equilibrium dimer
(SiO
2
bulk)
Stretched dimers
(E decreases with
increasing Si–Si
separation)
SiO
2
Si
FIGURE 7.9
Schematic illustration of the energy levels of dimer Ovacancies in bulk SiO
2
with equilibriumSi–Si atomic spacing
of $0.25–0.30 nm and near-interface dimers with stretched Si–Si spacing of $0.35–0.40 nm and a distribution of
energy levels near midgap. (From Fleetwood, D.M. et al., IEEE Trans. Nucl. Sci., 49, 2674, 2002. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
be the only sou rce of noise in the se devi ces. (3) In an nMOS transi stor, one must cons ider
the possible inter conve rsion of fourfo ld and fi vefold-co ordinated, pu ckered O vacancy
defects due to thermal vibration s. Ther mal energy (i.e., phon ons) can cause ato ms to
stret ch and ro tate mo re st rongly with inc reasing temperat ure. If one pictu res the Si1 – O4
atoms bendi ng and stret ching in Figure 7.8b, thi s site may swit ch from a con fi guration
that locally favors fourf old coordinat ion of the Si1 atom into on e that favors fivef old
coor dination, and vice versa. Th is local change in atomi c spacing and bond ang le lead s to
an elect ron trap level in one con figura tion, but not in the othe r. Thus, it wou ld be
energe tically favorabl e for a fourfo ld-coordi nated , puck ered O vacancy defect to releas e
an electron when the O4–Si nea rest neighbors bend to ward and move closer to the Si1
atom in Figure 7.8b to form a fi vefold-co ordinat ed, pu ckered O vacancy defe ct, due to
netwo rk vibra tions and =or vari ations in the local electri c field or strai n. Similarly , it
become s more favo rable to rec apture the electron when the O4 –Si atoms bend and
mo ve away from the Si1. He re, it is the thermally activated ato mic recon fi guration that
trig gers the captur e and releas e of the elect ron. Stretched fivef old-coord inated, pu ckered
O vacancy defects may also exh ibit this kind of behavior nea r a void, or in a low-d ensity
regio n where pucke red defect con figura tions are not geometri cally all owed [79,89] . (4)
Fina lly, it is also poss ible that nMOS transis tor no ise before or a fter irr adiation may
inv olve thermal ly assis ted capt ure and emission of an electro n by a dime r O vacancy.
This again must be mediate d by net work relaxa tion, since there is no el ectron trap st ate
availab le when thi s defect has its equilibr ium Si1–Si0 bon d spacing . Ho wever, esp ecially
near the Si=SiO
2
interface, there likely exist highly strained dimer configurations, shown
via DFT calculati on to have near-mid gap electro n trap levels , as illustrat ed in Figure 7.9.
When an electron is captured, the Si–Si spacing decreases due to the increased electron
density between the two atoms. This leads to a rise in trap energy level and electron
reemission. Although the dimers would have to be configured spatially and energetically
so as to enable these transitions to occur, the timescale for the atomic relaxation process
again is the rate-limiting step in the noise process [55]. In addition to the O vacancy-
related defects considered in Figure 7.8, several hydrogen-related defects may also
contribute to MOS 1=f noise.
7.3.4 Energy Scale for 1=f Noise
After illustrating the link between MOS radiation response and low-frequency noise, we
now consider how the energy scales that are derived for 1=f noise measurements compare
to those expected fromDFT calculations for O vacancies, and also to energy scales derived
from TSC measurements. Ultimately, since the defect energies should not be sensitive to
the precise method of measurement, one expects the results from TSC and noise measure-
ments to be self-consistent. In evaluating the conditions required for the energy scales to be
consistent, additional insight into the underlying physical processes is obtained [55].
The frequency dependence of the measured 1=f noise is defined as
a ¼ À
@ lnS
V
@ lnf
(7:4)
For devices in which the 1=f noise is due to a randomthermally activated process having a
broad distribution of energies relative to kT, Dutta and Horn have shown that the fre-
quency and temperature dependences of the noise are related via [94]:
a(v,T) ¼ 1 À
1
ln(vt
0
)
@ lnS
V
(T)
@ lnT
À1

(7:5)
ß 2008 by Taylor & Francis Group, LLC.
where
v ¼ 2p f
t
0
is the attemp t to escap e frequency fo r the defect
For noise that is demo nstrat ed to satisfy Equati on 7 .5, one can infer the sh ape of the defect
energy distrib ution D ( E
0
) from noise measu remen ts versus temperat ure via:
D ( E
0
) /
v
k
B
T
S
V
( v,T ) (7: 6)
where the defect energy is related to the tem perature and frequency through the simp le
expres sion [94]:
E
0
% ÀkT ln ( vt
0
) (7: 7)
To within a factor of 2p insid e the logarithm (wh ich doe s not sig nificantl y affect the result s
below, and arises only if one cons iders the rad ial frequency v a s opp osed to f as per Dutta
and Horn [94]; alternati ve develop ments om it this factor [95]) , Equation 7.7 essentia lly
resta tes the cond ition that the indivi dual charac teristic times of the noise proces ses t are
related to the acti vation energi es E
0
via the expecte d [55]:
t ¼ t
0
exp ( ÀE
0
=kT ) (7: 8)
Physica lly, if one presumes the noise is the resu lt of thermally acti vated process es, E
0
is the
barrier that the syste m must overcom e for the system to move from one con figura tional
state to the othe r [95]. Hence , Equatio n 7.6 provi des a measu re of the energy requi red not
for the elect ronic transi tions (e.g., the carrier mo tions depict ed in Figure 7.8) that are
associ ated with 1= f noise, but instead provi des insi ght into the energe tics of the the rmally
activate d SiO
2
net work relaxation process es.
Xiong et al. [96,97] have demo nstrated that the 1 =f noise of the transi stors of Figure 7.5
clearly sat isfy the Dutta –Horn frequency –mag nitude corre lation criterion of Equati on 7.5.
This mak es it possible to use Equation 7.7 to ext ract defect energy distrib utions fo r the O
vacancy- related defects causing the noise. Th is energy distrib ution is shown in Figure 7.10
for MOS transistors with 32 nm oxides from the same process lot as those in Figure. 7.5:
(a) before irradiation, (b) after 500 krad (SiO
2
) x-ray irradiation, and (c) after 2008C
postirradiation anneal at 0 V. This energy scale assumes f ffi 1 Hz, and t
0
ffi 1.8 Â10
À15
s
[55]—with the former choice for convenience, and the latter to facilitate comparison to
TSC. The temperature dependence of the attempt-to-escape frequency [32,33] is neglected
in Figure 7.10, as this has only second-order effects on inferred energy scales. Also,
this particular value of t
0
corresponds to the value of n extracted from [50], in contrast
to the slightly lower values inferred in Refs. [32,33], but this difference does not affect
the results.
After annealing, the energy distribution of the defects in Figure 7.10 changes remarkably
from preirradiation values. Evidently the irradiation and anneal have modified the defect
energy distributions, perhaps consistent with work in the literature showing a relaxation of
stress near the Si=SiO
2
interface with irradiation and=or annealing [93,98]. There is a broad
enough range in energy distributions for the dimer and puckered O vacancy-related
defects in DFT calculations (from <0.7 to >3.0 eV [79], with energy spreads for each defect
associated with small variations in configuration) that these O vacancy-related defects may
well be the primary cause for the observed noise [55]. Higher-energy processes tend to
reflect dramatic changes in configuration from one defect type to another (e.g., puckered
ß 2008 by Taylor & Francis Group, LLC.
to unpuck ered), and smaller change s are associ ated with m ore subtle atomi c rea rrange-
men t withi n a singl e defect type [55,79].
Tha t the energy scale fo r the noise measu remen ts in Figu re 7.10 is cons istent with the
energy scale for the TSC measu remen ts in Figure 7.4 is sh own in detai l in Ref. [55]. For
these two energy scales to be con sistent requi res a commonal ity of captur e and emi ssion
proces ses associate d wi th MOS oxide traps and border trap s. Th is impl ies [55] that the
emis sion of trapped holes duri ng TSC is not simple emis sion over a barrier , but is instead
trig gered by the kinds of Si O
2
network relaxa tion that lead to 1=f noise [55,95]. DFT
calcul ations demonstr ate that the energy levels of the dime r and puck ered O vacancy-
related defe cts sh own in Figu re 7 .8 typical ly differ by $1.0 –2.5 eV [79]. This suggest s that
trappe d hole emi ssion may occur at the point duri ng a TSC measure ment when a puckered
O vacancy- related cen ter acquires enou gh thermal energy to con vert to a dime r, after
which the trapped hole is spontan eously emitted. The tri gger for this process appears to
the high vibration al rate inferred for the Si atom puck ered through the plane in the E
0
g
defect [7,50], with hole emission and rebon ding of the Si –Si atom s at a temperat ure that is
high enough to overcom e the barrier to transf orm the pucke red Si in the deeper ho le trap
back int o its or iginal plane in the shallowe r dime r level. De fects will no t alwa ys recon figure
into the ir or iginal (i.e., pre-s tress) cond itions, accountin g fo r the changes in defect energy
distri butions with irrad iation and anneal ing in Figu re 7.10. The O vacancy cen ters in Figu re
7.8a and b there fore form a simple two-l evel system, with both 1 =f noise and TSC
measu remen ts sensi ng changes in atomi c con figura tions that lead to change s in defect
energy levels with time [55].
The above considerations help to resolve the long-standing discrepancies between
thermal (<2 eV) and optical (>3 eV) energies inferred for trapped-hole annealing in the
0.14 0.28 0.7 0.84 0.98 0.42 0.56
50 100
1E−13
1E−14
S
v

(
f
,
T
)

*

f
/
T

(
V
2
/
K
)
1E−15
250
Prerad
Postrad
Postannealing
300 350 150 200
E
0
(eV)
T (K)
FIGURE 7.10
S
V
f=T (proportional to the defect energy distribution) versus T and E
0
for 3 Â 16 mmnMOS transistors with 32 nm
oxides () before irradiation, (.) after 500 krad (SiO
2
) Co-60 irradiation at 6 V bias, and (!) after 24 h of 0 V
annealing at 473 K. The energy scale inferred fromthe Dutta–Horn model is given on the upper x-axis for f ¼ 1 Hz
and t
0
ffi 1.8 Â 10
À15
s. (After Xiong, H.D. et al., IEEE Trans. Nucl. Sci., 49, 2718, 2002. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
literatu re [1,32,33 ]. In opti cal measu rements at ro om temperat ure, there is no t enou gh
thermal energy to enable defect recon figura tion, so on e measu res the dee p ba rrier fo r
trapped -hole emi ssion from a n E
0
g
defect ( > 3 eV, as ver ifi ed by DFT calcul ation [79]) .
At elevated tem perature, the defect rec on figures before thi s emis sion occurs, so it lik ely is
atomi c rec on figura tion that is the rate-limi ting step in emi ssion in this case [55]. This
emphasi zes the impo rtance of match ing the effect ive energy sca le that is most appro priate
to the applicat ion whe n attemp ting to com pare expe rimen tal and theoret ical resu lts.
7.4 D ipolar De fect s ne ar the Si = Si O
2
Interface
7.4.1 Switched -Bias Annea ling
In the dis cussion of defe ct energi es, we dis cussed brie fly two kinds of O vacancy- related
defects that can function a s border traps if located nea r the Si=SiO
2
int erface . We now
discuss other prop erties of border traps and relate d dipolar defe cts in the near-i nterfacial
SiO
2
. Th e presen ce of a signi fi cant densi ty of dip oles in irradiat ed SiO
2
was first demon-
strated by Pepper [99]; ho wever, it was no t studied int ensive ly until later Schwank et al.
[100] sh owed that trapped posit ive charge that was previ ously thought to be rem oved
by posit ive-bias anneal ing could be restor ed during negativ e-bias anneal ing, and
Nissan- Cohen et al. [101] foun d a simi lar rev ersibilit y of net trappe d posit ive charge
after high- fi eld elect rical st ress. Dipoles associ ated with O vacancy- related defects were
studied in detail by Lelis and coworkers [90,102], who proposed that the reversibility of
the positive-charge annealing was caused by an electron tunneling into and out of a Si
atom adjacent to a second, positively charged Si atom, associated with an E
0
g
defect in the
near-interfacial SiO
2
[90], simi lar to the proces ses conside red in Figu re 7.8.
The reversibility in MOS oxide-trap charge when bias is switched after irradiation or
high-field stress can be significant in magnitude. In Figure 7.11, we show the irradiation
and annealing response of devices that were exposed to Co-60 gamma rays under positive
0
−1
−2
−3
−4
10
2
10
3
10
4
10
5
Irradiation and anneal time (s)
45 nm soft oxide
30 k
45 k
90 k
2.75 year
10
6
Δ
V
m
g

(
V
)
10
7
10
8
FIGURE 7.11
Midgap voltage shift as a function of irradiation and annealing time for nMOS transistors with 45 nm oxides that
were intentionally softened with a postoxidation 30 min, 11008C anneal to introduce a large number of O
vacancies. The devices were irradiated at 6 V gate bias with Co-60 gamma rays at 240 rad (SiO
2
)=s; the annealing
temperature was 1008C. (From Fleetwood, D.M., Appl. Phys. Lett., 60, 2883, 1992. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
bias, and then anne aled for 2.75 year s at posit ive bias at 100 8 C [14]. This long, elevated-
tem perature anne al comple tely neutr alized the initi ally high densi ty of positive oxide -trap
charge , but did not reveal any net negativ e charge trap ping even at very long tim es at
nega tive bias at room temperat ure and 100 8 C. These results strong ly reinforce the conclu-
sion that int erface traps are charge neutr al at approximat ely midgap [2,13,14 ], con fi rming
that the midgap voltage sh ift is appro ximate ly equ al to the threshold voltage sh ift due to
oxide -trap charge [11]. The lack of signi fi cant net electron trap ping in the se dielectric layers
is typical of most experi ence with thermal SiO
2
[1,4,32] , showing that the only locations for
elect rons to be trapped in these ox ides with relati vely high captur e cross sectio ns are at
sites that compens ate trapped posit ive charge.
Figu re 7.12 show s that, even after such a long anneal ing period, the pos itive and nega tive
charge s have no t rec ombined , so the defe cts have not truly anneal ed [14,90,1 00 –103].
Inde ed, when the tempe rature is restor ed to 100 8C at nega tive bias in Figu re 7.12, the net
trappe d positive charge in the SiO
2
increa ses dramatical ly. There are two points that are
esp ecially inter esting ab out the reversi bility of the net posit ive oxide -trap charge in Figu re
7.12. First, the amo unt of net trapped positive charge in the two cases in Figu re 7.12 is
nea rly independe nt of dos e, des pite the large difference in initial levels of net pos itive
charge for the devi ces irradiat ed to 30 and 90 krad (SiO
2
) in Figu re 7.11 [34]. Se cond, the net
posit ive oxide -trap charge afte r nega tive-bi as anneal ing fo r the 30 krad (SiO
2
) case in Figu re
7.12 is grea ter than afte r initial irr adiation in Figu re 7.11. Th e latter result is cons istent with
the expe ctation that a signi ficant amoun t of trapped pos itive charge already had been
neutr alized duri ng the irr adiation period [34]. This result is also con sistent with the result s
of TSC experi ments [30,32,4 4,49], as we discuss further below, and is also consiste nt with
the pres ence of dipole s similar to those depict ed in Figu re 7.8.
Add itional sup port for the idea that O vacancy- related defects may lead to the kinds
of oxide-tra p charge rev ersibilit y that is dep icted in Figure s 7.11 a nd 7.12 is illu strated by
the reversib ility in densi ties wi th changing bias observe d in EPR studies by Conl ey et al.
[104]. The key result of this study is sh own in Figure 7.13. The initial E
0
g
densi ty in thi s
study was generat ed via the vacu um ultrav iolet (UV ) injecti on of $ 5 Â 10
13
holes =cm
2
int o
30 K
Δ
V
m
g

(
V
)
25ЊC 100ЊC
90 K
10
3
10
4
10
5
Time (s)
10
6
10
7
0
−0.5
−1
−1.5
−2
FIGURE 7.12
Midgap voltage shift versus negative-bias anneal time and temperature for the devices of Figure 7.9. The 2.75
years, 6 V, 1008C anneal in Figure 7.9 was followed by the room temperature and 1008C anneals at À6 V shown
here. Results for the 45 krad (SiO
2
) irradiations and anneals were similar. (FromFleetwood, D.M. et al., IEEE Trans.
Nucl. Sci., 40(6), 1323, 1993. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
120 nm dry oxide s grown at 1000 8 C. After hole injectio n, bias was applied onto these
oxide s using cor ona ions, establish ing electric fi elds of alternati ng polarity , wi th mag ni-
tudes of $ 3.5 MV=cm. Point N1 in Figu re 7.13 is the first nega tive cor ona ion bias, which
leads to an inc rease in E
0
g
densi ty, follo wed by a large decre ase when the bias is changed
to posit ive (P1 ), wi th this pattern repeati ng for sub sequent nega tive (N2) and posit ive
(P2) cor ona bias. The striking cor relation betwe en the respons e of the E
0
g
density in
Figure 7.13 and the electri cal respons e of MOS devices under simi larly swit ched-b ias
cond itions [5,34,90 ,100 –102 ] is com pelling eviden ce for O vacancy- related dipole s in the
near-i nterfacial Si O
2
.
Reve rsibility in MOS ox ide-trap charge afte r irr adiation or expo sure to high- fiel d stress is
not limit ed to SiO
2
gate oxides. Even larger reversi bility owing to shallo w el ectron traps
has been obse rved in silicon- on-in sulator (SOI) burie d ox ides [105], in silico n nitride [57],
and in some kinds of high- K dielectri cs [106,107 ]. The magnitu de and signi ficance of these
effect s can be minimize d wi th improve d proce ssing tech niques for high - K die lectrics ;
howeve r, the bes t one can appare ntly do is to obt ain respo nse similar to the rmal SiO
2
[108].
7.4.2 Capac itance –Vol tage Hyst eresis
Similar results to the abov e switche d-bias anneal ing respo nse often can be observe d on a
somew hat more rapid tim escale in the hys teresis of MOS capacita nce –voltage ( C–V ) cur ves.
Figure 7.14 illus trates this hysteresi s afte r irr adiation; Figu re 7.15 dis plays simi lar hys teresis
after high field stress, sugge sting similar concentrat ions of border traps are cause d by the
two differe nt stresses [8,44]. Consi stent with the int erpretat ion that the C–V hyster esis
is caused by bor der traps , Figure 7.16 shows that the amoun t of measu red hys teresis
increa ses logarithm ically with inc reasing ramp time during the C–V m easuremen t,
as expe cted for electro ns tunne ling into (at positive bias ) a nd out of (at nega tive bias) border
traps . As was the case for pos tirradia tion swit ched-b ias annea ling, an excel lent corre lation
has been observed between electrical estimates of border-trap densities (proportional to
DV
bt
, which is the difference in midgap voltages for forward and reverse C–V measure-
ments) from C–V hysteresis and measurements of E
0
g
densities via EPR during a room-
temperature annealing sequence for devices with high O vacancy densities [40]. This is
1.1
1.0
0.9
0.8
0.7
0.6
0.5
H+ 01 N1
Bias condition
E
g
Ј

d
e
n
s
i
t
y

(
1
0
1
2

/
c
m
2
)
N2 P1 P2
FIGURE 7.13
E
0
g
defect densities measured via EPR after vacuum UV injection of $5 Â 10
13
holes=cm
2
into 120 nm dry oxides,
followed by two cycles of corona ion biases at alternating biases. (After Conley, J.H. Jr., IEEE Trans. Nucl. Sci., 42,
1744, 1995. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
illus trated in Figure 7.17 for 410 nm thermal oxides grown in dr y O
2
, which were then given
a 1320 8 C anneal to simu late the high-te mperat ure process ing that is used in the manufac-
turin g of SOI buri ed oxide s via the separati on via impl antat ion of ox ygen (SIMO X) pro-
cess [40,109] . The inc rease in both the E
0
g
and bor der-trap densi ties with a nnealing tim e in
Figu re 7.17 is unc ommon, and is attribute d in the se devices to an unus ually high density of
E
d
0
cen ters, due to the extre mely high-tem perature anne al [40]. The pres ence of a large
num ber of E
0
d
defects can grea tly retard ho le transp ort [37,40,5 5,78,79] . Th e correlat ion in
gro wth rates of the border traps and E
0
g
defect densi ties in Figure 7.17 is striking. A simi lar
corre lation betwe en trap ped holes in SiO
2
and dipole formatio n at the Si =SiO
2
interface was
note d by Lai in 1981 in studies of hole trapping and electron injecti on [110]. We note that the
overall density of the E
0
g
defects in these devices is more than an order of magnitude larger
250 Pre-rad/post-TSC
Forward (F)
Reverse (R)
Hysteresis (R − F)
200
150
100
50
0
−3 −2 −1 0
Voltage (V)
C
a
p
a
c
i
t
a
n
c
e

(
p
F
)
1 2
FIGURE 7.14
High-frequency (1 MHz) capacitance–voltage hysteresis curves for n-substrate capacitors with 17 nm thermal
oxides irradiated to 2 Mrad (SiO
2
) at 4 V bias with 10 keV x-rays at a dose rate of 3333 rad (SiO
2
)=s. (From
Fleetwood, D.M. and Saks, N.S., J. Appl. Phys., 79, 1583, 1996. With permission.)
300
Pre-rad/post-TSC
Forward (F)
Reverse (R)
Hysteresis (R − F)
250
150
50
0
−3 −2 −1 0
Voltage (V)
C
a
p
a
c
i
t
a
n
c
e

(
p
F
)
1 2
200
100
FIGURE 7.15
High-frequency (1 MHz) capacitance–voltage hysteresis curves for n-substrate capacitors with 17 nm thermal
oxides subjected to 10 mC=cm
2
Fowler–Nordheimstress. (FromFleetwood, D.M. and Saks, N.S., J. Appl. Phys., 79,
1583, 1996. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
than the density of border traps, consistent with etch back studies that showed a large
amount of bulk positive-charge trapping in these oxides [40]. So the interpretation of this
result is simply that E
0
g
centers that are not close enough to the Si=SiO
2
interface to exchange
charge with the Si do not function as border traps, but act as bulk oxide-trap charge [40,111].
7.4.3 Stable Dipoles in the Near-Interfacial SiO
2
We now present evidence from TSC experiments that strongly suggest there is a second
type of dipole in irradiated or stressed SiO
2
. This defect is apparently distinct from the O
vacancy-related defects discussed above, at least in position, and likely also in microstruc-
0.6
0.5
0.4
0.3
Δ
V
b
t

(
V
)
0.2
0.1
0
0.01 0.10
Ramp rate (V/s)
1.00
FIGURE 7.16
High-frequency C–V hysteresis as a function of
voltage ramp rate for MOS capacitors with 45
nm oxides grown in dry O
2
. These devices
received a 10008C N
2
anneal after oxidation to
increase their O vacancy density. (From Fleet-
wood, D.M. et al., Microelectron. Reliab., 35, 403,
1995. With permission.)
20
18
16
14
10
12
8
6
0 250 500
E
P
R

d
e
n
s
i
t
y

(
1
0
1
1
/
c
m
2
)
E
g
Ј
750
Time (min)
1000 1250
Δ
V
b
t

(
V
)
ΔV
bt (slow)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
FIGURE 7.17
Comparison of E
0
g
defect densities measured via EPR and threshold voltage shifts due to border traps estimated
fromcapacitance–voltage hysteresis for 410 nmoxides annealed at 13208C after gate oxidation and polycrystalline
Si deposition. These devices emulate the thermal history of SIMOX buried oxides. (From Warren, W.L. et al.,
IEEE Trans. Nucl. Sci., 41, 1817, 1994. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
ture. It has been observe d com monly in TSC studies of irr adiated or electric ally stresse d
SiO
2
that the to tal amoun t of trapped pos itive charge estimated by integr ating the TSC
can grea tly exc eed the net posit ive oxide -trap charge esti mated from the midgap voltage in
C –V measure ments, Q
CV
¼ÀC
ox
D V
mg
, where C
ox
is the ox ide capacita nce [24 –27,
29 –32,34,37 ,44 –46,49 –52]. An exa mple is sh own in Figu re 7.18, for radiat ion-hard ened
oxide s with a low O v acancy density [30,31]. Th e dash ed line repre sents Q
CV
( $ 2.3 Â 10
12
cm
À 2
), which is $ 56% of the total pos itive charge one inf ers from TSC
measu remen ts at large negativ e bias ( $4.1 Â 10
12
cm
À2
). The redu ction in charge coll ected
at low elect ric field is a result of space charge eff ects [15,29,3 0].
Tha t there is no signi ficant TSC for posit ive bias in Figu re 7.18 from trapped pos itive
charge is easy to unders tand from Figure 7.3, since unde r positive bias , the trapped
posit ive charge moves on ly an in finit esimal distanc e before reachi ng the Si . It require s
charge movin g over a finite distanc e to produ ce signi ficant TSC [15,24 –27,30]. Howe ver,
the lack of TS C at pos itive bias from the nega tive charge infer red to be in the oxide
provid es sig nificant insight into the nature of the negativ e charge. Inde ed, resu lts simi lar
to tho se in Figu re 7.18 stro ngly suggest that the nega tive charge that must exist in the near-
inter facial Si O
2
after radiation expos ure or high -fi eld stress to accoun t fo r the differe nce
betwe en TSC and C–V estimates of MOS oxide-tra p charge must be unab le to transp ort
across the oxide [29–32,44,46,49,52].
Another interesting property of dipoles in the near-interfacial SiO
2
is illustrated in
Figu re 7.19, whe re we sh ow inferred densi ties of trapped positive and negativ e charge
from TSC and C–V measurements. The trapped negative charge is categorized in the
figure as easily exchanged with the Si (shallow electrons), or difficult to exchange with
the Si (deep electrons). In Figure 7.19, there is very little change in the inferred deep
electron trap density until the postirradiation annealing temperature exceeds 1158C. In
contrast, the inferred density of shallow electrons decreases by more than 30%. Larger
differences in the annealing characteristics of the deep and shallowelectrons are noted after
isothermal annealing at negative bias at $1158C [112]. It is possible that the results in
Figure 7.19 can be explained by an O vacancy-related defect with a single basic micro-
structure (e.g., the E
0
g
) that is distributed in space and energy in such a way as to alter its
8
6
4
2
0
−2
−16 −12 −8 −4 0
TSC bias (V)
45 nm hard oxide
x-ray to 5 Mrad @+5 v
Q (TSC) = 6.5 nC
Q (CV) = 3.6 nC
T
S
C

c
h
a
r
g
e

(
n
C
)
4 8 12
FIGURE 7.18
Integrated TSC, corrected for background leakage, as a function of bias for n-substrate capacitors with 45 nm
oxides grown at 10008C in dry O
2
, after which they received a 9008C forming gas postoxidation anneal (before
polycrystalline Si deposition). The devices were irradiated to 5 Mrad (SiO
2
) with 10 keV x-rays at a dose rate of
5550 rad (SiO
2
)=s at a bias of 5 V. The total positive charge Q
p
estimated fromTSC is compared to the net positive
charge Q
CV
estimated from midgap C–V measurements. (After Fleetwood, D.M., Reber, R.A. Jr., and Winokur,
P.S., IEEE Trans. Nucl. Sci., 38, 1066, 1991. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
annealing properties in the manner required to account for the observed response differ-
ences in Figure 7.19. However, it is also quite possible that hydrogen-related defects
may also be responsible for the stable dipoles in the near-interfacial SiO
2
[110,111]. Hydro-
gen- related defects have also been identified that can contribute to postirradiation or post-
electrical stress reversibility in observed oxide-trap charge [101,113–119], but these
switching defects are obviously not a candidate for the fixed negative charge observed in
the TSC experiments [32,111]. The microstructure of these defects remains an open and
interesting question.
The relative rates of reversibility are compared among oxide, interface, and border traps
in Ref. [20]. Interestingly, in those devices and in recent work on high-K dielectrics [107],
there is a close correspondence between the reversibility in bulk oxide traps and interface
traps, which are dangling Si bonds at the Si=SiO
2
interface. An exchange of protons
between an O vacancy-related defect in the near-interfacial SiO
2
and the interface-trap
site can explain this reversibility in a natural way [107]. Thus, both hydrogenous species
and O vacancy-related defects may play significant roles in the reversibility of MOS oxide-
trap charge in ways that are inextricably linked and affect MOS long-term reliability
and radiation response in significant and fundamental ways that are still under intense
investigation [119–126].
7.5 Interface Traps
7.5.1 Two-Stage Interface-Trap Buildup Model
For the last 30 years, it has been recognized that hydrogenous impurities and Si dangling
bond defects are critical to the formation of interface traps in MOS radiation response
1
0.1
0.01
0 50 100
Anneal temperature (ЊC)
T
r
a
p
p
e
d

c
h
a
r
g
e

d
e
n
s
i
t
y

(
1
0
1
2
/
c
m
2
)
150 200
Positive charge
Deep electrons
Shallow electrons
FIGURE 7.19
Trapped positive charge estimated from the integrated TSC (triangles), estimated density of electrons in border
traps from C–V hysteresis (dots; shallow electrons), and differences between Q
p
estimated from the TSC and Q
CV
(Q
n
¼ Q
p
ÀQ
CV
) as a function of isochronal annealing temperature. To estimate Q
n
(squares, deep electrons), C–V
curves were swept from negative-to-positive bias to exclude contributions of electrons in border traps that
contribute to C–V hysteresis. Capacitors with 45 nm oxides were irradiated to 2 Mrad (SiO
2
) at a dose rate of
1100 rad (SiO
2
)=s at 10 V, after which they were annealed at À10 V for 15 min for the temperatures shown between
irradiation and TSC measurement at À12 V bias. (From Fleetwood, D.M. et al., Appl. Phys. Lett., 74, 2969, 1999.
With permission.)
ß 2008 by Taylor & Francis Group, LLC.
[2,4,5,11 8,127 –13 5]. Figu re 7.20 illustrat es a piv otal expe riment that compares inter face-
trap densities estimate d from high- and low-fre quency C –V method s [10,12] after
Co-60 gamma ( $ 1 MeV) and $ 10 eV UV irrad iation [130]. Energ y from the Co-60
irrad iation is absorbed uniform ly throu ghout the SiO
2
and surrou nding devi ce laye rs.
In contrast, the nonpen etrati ng 10 eV phot ons are ab sorbed on ly at the top surfa ce of the
SiO
2
. Hence , the simi laritie s in esti mated densities and energy dis tributions indicate that
inter face trap s most ly are not create d by the direct interacti on of high ly pene trating
radiat ion with we ak bonds at the Si=SiO
2
interface [130]. Instead, inter face-tra p fo rmation
usu ally occ urs in radiation effects studi es (and many types of high- fi eld stress expe ri-
men ts as well [44]) as a result of a series of proces ses initi ated by the cre ation of
elect ron –hole (e–h) pairs in the SiO
2
, and =or the subseque nt transport of the ho les to
the Si =Si O
2
interface [28,132] . This process is cons istent wi th the schem atic diagram that
was shown in Figure 7.1. Follow -on studi es by Winokur and coworke rs eval uate the
dose , tem perature , proce ssing, radiation bias, and anne aling tim e and bias dep endence s
of radiat ion-indu ced int erface-trap bui ldup in met al gate capa citor s [2,130,1 32–134].
Fro m the se and othe r relate d studies, it was inferred that the acti vation energy for
MOS inter face-tra p buildu p in metal ga te capacitor s is $ 0.8 eV [133]. Moreover, it was
also demons trated that interface -trap formatio n is a two-sta ge proce ss [2,130,1 32 –135].
The first stage of the process is associated with hole transport through the SiO
2
and=or
trapping near the Si=SiO
2
interface, with the accompanying release of a positive ion.
10
14
10
13
10
12
10
11
N
S
S

(
c
m

2

V

1
)
10
10
−1.0 −0.8
E
v
−0.6 −0.4
y
s

(V)
−0.2 0 0.2
E
c
0.4
Post
C
o
60
UV
Pre
FIGURE 7.20
Interface-trap charge densities as a function of surface potential for 83 nmdry oxides with 5000 nmAl gates before
and after $100 krad (SiO
2
) Co-60 irradiation at 10 V, and 90 nm dry oxides with 50 nm Au gates before and after
90 min of $10 eV UV irradiation at 10 V. (FromWinokur, P.S. and Sokoloski, M.M., Appl. Phys. Lett., 28, 627, 1976.
With permission.)
ß 2008 by Taylor & Francis Group, LLC.
The seco nd st age involve s the transport of this ion to the Si =SiO
2
inter face and its
subsequ ent interac tion, leadin g to the formati on of an interface trap. It was argu ed
persua sively in 1980 by McLean that this ion was H
þ
[135]. Th is semiem pirical model
of H
þ
transport and rea ctions captur es many of the essen tial featur es of interface -trap
formatio n, and forms a foun dation for present mo dels of int erface-trap fo rmation in
MOS devices. McL ean also demonstr ate d the consiste ncy of the hydro gen model of
radiat ion-indu ced inter face-tra p formatio n with expe rimen tal results on the temperat ure
depen dence of int erface-trap fo rmation during high - field stress obtain ed by Hu and
Johnso n [135,136 ].
Despi te the succes ses of the two-s tage hyd rogen model , altern ative expl anations for
inter face-tra p buildu p to the hydro gen model of M cLean et al. have been proposed .
Many studi es have obse rved the app arent c onversion of trapped positive charge to int er-
face traps [110,136 –14 4]. Becau se bot h H
þ
ions and trapped ho les are charged pos itively
when transpo rting or trapped in SiO
2
, the interpretat ion of some st udies is ope n to
altern ative expl anation. Mo reover, EPR studies found no link [145] betwe en the reducti on
of holes trappe d in E
0
centers (as no ted in Secti on 7.3, thi s is the mo st com mon type of hole
trap in Si O
2
[1,13,85 –88]) and the buildu p of P
b
inter face defects. Th e P
b
defe ct has been
identi fied as a dan gling Si bon d at the Si O
2
inter face, as is kn own to be the dominan t
radiat ion-indu ced interface trap at the Si=SiO
2
interface [13,87,1 46 –149]. In contrast, st udies
of the time and electri c fi eld depen dences of interface -trap formation in poly crystalline
Si-gate devi ces appeare d at least initial ly to be more consistent with trapped -hole models
than hydrogen model s [150]. Howeve r, a very ca reful set of time, electri c field, bias , and
temperat ure dep endence experi ments by Boe sch [151] and by Saks et al. [152 –159] have
clearly demo nstrat ed that hydrog en releas e in the SiO
2
, transpo rt to the SiO
2
interface , and
subsequ ent reaction at or nea r the interface play a cen tral, enab ling role in rad iation-
induce d inter face-tra p buildu p. We now examin e some of the key results of thi s work.
Figure 7.21 shows the int erface-trap buildu p, m easured via the charge-p umping tech-
nique , for MOS transi stors wi th 42 nm oxides [153]. For on e set of devices, a cons tant 8.4 V
bias was appl ied throu ghout the irrad iation and anneal ing period. The remain der of the
devices we re irrad iated and anneal ed at a bias of À8.4 V, with the appl ied gate bias then
switche d to 8.4 V at vary ing tim es after the radiat ion pulse was delivered . In all switche d-
bias cas es, the bias rem ained nega tive for times muc h longer than that calculate d for hole s
to transp ort out of the oxide and into the gate ( $10 ms) for this oxide thickne ss and elect ric
field [153,160 ]. Clearly, there is signi ficant inter face-tra p formati on in all of the switched
bias cases in Figu re 7.2 1. This indicates that neith er the dire ct inter action of rad iation at the
Si=SiO
2
interface nor the direct c onversion of transpo rting or trap ped holes to interface
traps is the rate-l imiting step in inter face-tra p formatio n in these devi ces. Instead , the
results of these studies are quite consistent with the two-stage buildup of interface traps
in metal gate devices [2,130,132–134], as described by the McLean model [135].
The magnitude and the time dependence of the interface-trap buildup in the devices of
Figure 7.21 have been modeled successfully by Brown and Saks [157] by assuming that the
rate-limiting step for the majority of radiation-induced interface trap buildup is the time it
takes for H
þ
to transport from its point of release in the SiO
2
to its point of reaction at the
Si=SiO
2
interface. Significantly, a similar activation energy is reported for interface-trap
buildup in polycrystalline Si gate devices ($0.8 eV [152,153]) as for metal gate devices
[2,132–134]. An isotope effect in interface-trap formation also has been observed in radi-
ation effects studies, with interface-trap formation occurring on a slower timescale for
deuterium-treated oxides than it does for hydrogen-treated oxides [159]. This is consistent
with the heavier D
þ
taking more time to transport to the interface after release during a
radiation pulse than the lighter H
þ
, once again showing the significance of hydrogenous
species to interface-trap formation.
ß 2008 by Taylor & Francis Group, LLC.
Saks et al. [154] have used isochron al anneal ing measu remen ts after low- temperat ure
irrad iation at pos itive bias to furthe r investig ate the kinetics of rad iation-indu ced inter face-
trap formati on, as shown in Figure 7.22. He re, we observe two dis tinct regi mes of inter face-
trap formatio n. Below 1 50 K, the inter face-tra p formation is rough ly inde pende nt of
annea ling bias (positiv e, nega tive, or zer o), and is attribu ted to the diffus ion of neutra l
H. Abo ve 180 K, the int erface -trap formati on is st rongly affecte d by anneal ing bias
(enhanc ed for posit ive or zero bias, and suppr essed for nega tive bias ), and is attri buted
to proton drift . In the se experi ments, the total fraction of int erface -trap formati on a ssoci-
ate d with the low- temperat ure diffusion mechani sm only correspo nds to about 1%–2% of
the portion of the inter face-tra p densi ty associ ated with proton drift to the int erface. This
af firms that proton drift dominat es over neutr al hy drogen diffusion in ro om-temp erature
inter face-tra p formati on [154].
Sub stantial inter face-tra p buildu p is also reported in oxide s that are fi rst irr adiated or
subj ected to high - field stress, and then are exposed to hydrog en after the radiat ion
expos ure has been st opped [114,116 , 161 –166]. In this case, the expe rimental evid ence
stro ngly sugge sts that neutral hy drogen diffus es into the oxide, cracks at a defect cen ter,
and crea tes atomi c hyd rogen or a proton , which then transports to and reacts at the
inter face. Furthe r, sig nificant interface -trap buildup has been observe d in oxides expo sed
to high fluence s of atomic hydrogen , even in the absence of ionizing radiation expo sure or
high -fi eld stress [167,168 ]. He nce, alth ough alternati ve mo des of inter face-tra p formati on
are poss ible, there is a wealt h of evi dence that sugge sts hydro gen pl ays a key role in
inter face-tra p formati on.
7.5.2 Hydrog en Transpo rt and Rea ctions
Althou gh there remain s little dou bt that hy drogen play s a cen tral role in radiation- induce d
inter face-tra p formati on, the details of the hyd rogen rele ase, transport, and reaction
3.0
PMOS1/2
2.0
1.0
0
10
−1
10
0
10
1
10
2
Time after irradiation (s)
V
gdc
= +8.4 V
Not switched
t
ox
= 42 nm
25 s
t
sw
= 0.1 s
100 s
300 s
Δ
D
i
t

(
ϫ
1
0
1
0
/
c
m
2

e
V
)
/
5
0

k
r
a
d
10
3
10
4
10
5
FIGURE 7.21
Interface-trap charge densities per unit energy as a function of postirradiation annealing time after a 50 krad
(SiO
2
), 1.5 ms radiation pulse froma 40 MeV electron linear accelerator. These devices were polycrystalline Si-gate
MOS transistors with 42 nm oxides, irradiated at either a constant 2 MV=cm oxide electric field, or with a À2
MV=cmelectric field that was switched to 2 MV=cmduring the postirradiation anneal at the times indicated in the
figure. (From Saks, N.S. and Brown, D.B., IEEE Trans. Nucl. Sci., 35, 1168, 1988. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
proces ses rem ain somew hat uncerta in. It is often assu med that hyd rogen releas e occurs
throug hout the bulk of the oxide as a resp onse to hole transp ort [2,132 –135,152 –1 59].
Indeed, such bulk rele ase is con sistent with the depen dence on ox ide thickne ss of the
time requi red for rad iation-indu ced int erface trap bui ldup in the work of Saks et al., as
illustrat ed in Figure 7.23 [153]. The thicker the ox ide, the longer it takes fo r interface trap s
to build up followi ng a rad iation pulse for the devices of Figu re 7.23. This is con sistent with
proton s being rele ased in the bul k of the SiO
2
fo r the se kinds of oxides. Howeve r,
Shaneyf elt et a l. have perform ed a similar study on different oxide s. They foun d min imal
oxide thick ness dep endence, as illustrat ed in Figure 7.24, for dry oxide s [169]. A slightl y
grea ter thickne ss dep endence for the inter face-tra p buildu p rate is obse rved for wet oxide s
by Shaneyf elt and cowo rkers [169], but st ill the dependence of the trap buildup rate
on oxide thick ness is neith er as pronounce d nor a s systemat ic as that obse rved by
Saks et al. [153]. These results, coupled with the strong simi laritie s of the electri c field
depen dences of trapped -positiv e-charge buildup and int erface-trap formatio n in poly crys-
talline Si-gate devi ces [150,170 ], led Shaneyf elt et al. to sugg est that most of the hydrogen
that ul timately lead s to inter face-tra p buildu p in the oxides they have examined is rele ased
when a hole is capt ured at or rele ased from an O vacancy site near an interface (Si or ga te),
as opposed to m ore distribut ed releas e throu ghout the bulk of the Si O
2
[169,170 ]. This
‘‘hole-tr apping =hydrogen transpo rt’’ mo del is qui te similar to the McLean mo del, only
with a different releas e point for the protons. Th e dominan t relea se mechani sm in a
partic ular device depends on the diffe rences in densi ties of hydro gen contai ned wi thin
the bulk SiO
2
network versus the densities that are attached to defect sites (e.g., O
vacancies) near the Si=SiO
2
interface, as well as the dynamics between the transporting
holes and hydrogen species bonded at such different sites.
100
10
0.1
0.01
70 100 200
Anneal temp (K)
t
ox
= 26 nm
0.74 Mrad
LT process
HT
process
PMOS1/6/A7-6
V
g
(ANN)
V
g
(rad) = +5.2 V
Δ
D
i
t

(
ϫ
1
0
1
0
/
c
m
2



e
v
)
300 400
1
0
+10.4
+5.2 V
−5.2
295 K, 60 h
FIGURE 7.22
Interface-trap charge densities per unit energy inferred via the charge pumping technique as a function of
postirradiation annealing temperature and bias for pMOS transistors with 26 nm oxides irradiated to 740 krad
(SiO
2
) at 78 K with Co-60 gamma rays. The annealing time at each temperature was 1200 s, and the devices were
recooled to 78 K for all measurements. (From Saks, N.S., Klein, R.B., and Griscom, D.L., IEEE Trans. Nucl. Sci., 34,
1234, 1987. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
1.0
0.8
0.6
0.4
F
r
a
c
t
i
o
n

o
f

l
a
t
e

D
i
t

g
e
n
e
r
a
t
e
d
0.2
E
ox
= +2 MV/cm
0
10
−1
10
0
10
1
10
2
Time (s)
10
3
10
15
26
42
90
10
4
10
5
t
ox
(nm)
t
½
FIGURE 7.23
Normalized interface-trap charge densities per unit energy as a function of postirradiation annealing time after a
50 krad (SiO
2
) radiation pulse froma 40 MeV electron linear accelerator. A small early component of interface-trap
formation has been subtracted from these results to focus on the two-stage interface-trap formation mechanisms.
These devices were polycrystalline Si-gate MOS transistors with 10–90 nm oxides from a single process lot,
irradiated and annealed at a constant 2 MV=cm oxide electric field. (From Saks, N.S. and Brown, D.B.,
IEEE Trans. Nucl. Sci., 35, 1168, 1988. With permission.)
1.2
1.0
10 MeV LINAC
Constant bias
+1 MV/cm
0.8
0.6
0.4
0.2
0.0
10
0
10
1
10
2
10
3
N
o
r
m
a
l
i
z
e
d

Δ
D
i
t
Time (s)
27.7 nm
35.8 nm
47.6 nm
63.9 nm
104 nm
10
4
10
5
10
6
t
1/2
FIGURE 7.24
Normalized interface-trap charge densities per unit energy as a function of postirradiation annealing time
after radiation pulses from a 10 MeV electron linear accelerator. These devices were polycrystalline Si-gate MOS
transistors with 27.7–104 nmdry oxides froma single process lot, irradiated and annealed at a constant 1 MV=cm
oxide electric field. Doses were 175, 95, 70, 50, and 35 krad (SiO
2
) for the 27.7, 35.8, 47.6, 63.9, and 104 nm oxides,
respectively. (From Shaneyfelt, M.R. et al., IEEE Trans. Nucl. Sci., 39, 2244, 1992. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
In past studies of interface-trap formation mechanisms, it has been often assumed that,
as a transporting proton approached the Si=SiO
2
interface, it is neutralized by an electron
from the Si, diffuses until it is near a Si–H bond, and then reacts via [171–174]:
Si ÀHþH ¼ Si ÀþH
2
(7:9)
While intuitively plausible, this reaction process poses several practical difficulties. First,
atomic hydrogen is highly chemically reactive in SiO
2
and therefore less stable than H
þ
during transport [173]. So there are several competing reactions, including simple
dimerization (HþH¼H
2
), that compete with Equation 7.9 both in the bulk of the
oxide and near the Si=SiO
2
interface. Second, it has also been argued theoretically that
H
þ
is a lower energy state than H in SiO
2
[122,123,175–177]. This suggests that, even if
the atomic H does not react immediately, it would donate an electron and become H
þ
again as soon as kinetically feasible. Consistent with this interpretation, both Afanase’ev
et al. [178] and Vanheusden et al. [179] have obtained evidence that H
þ
has an
approximately 100-times lower cross section for capturing an electron in SiO
2
than
does a trapped hole. Finally, it has always been conceptually difficult to understand
why interface-trap formation would be as efficient as commonly observed under posi-
tive bias during and=or after radiation exposure if the drifting proton were neutralized,
thereby transforming its transport from field-directed drift to a diffusion process [118].
Indeed, evidence of excess hydrogen at or near the Si=SiO
2
interface [180,181] might
even be likely to favor hydrogen motion away from the interface toward areas of lower
concentration in the bulk, in the latter case.
The theoretical and conceptual difficulties associated with Equation 7.9 have been
addressed by Rashkeev and coworkers. DFT calculations [177,182] strongly suggest that
protons interact directly at the Si=SiO
2
interface via the simple reaction:
H
þ
þSi ÀH ¼ Si
þ
þH
2
(7:10)
Theoretical calculations suggest that two electrons leave the Si–H almost simultaneously
when a proton approaches the Si=SiO
2
interface in the vicinity of a passivated dangling Si
bond, forming a H
2
molecule and leaving the resulting dangling bond positively charged,
at least momentarily, as per Equation 7.10. The positive charge on the trivalent Si at the
interface will quickly be transformed to a negative charge following double electron
capture from the inversion charge layer [177,182] when devices are irradiated or subjected
to high-field electrical stress under positive bias. Interface-trap formation under negative
bias requires hydrogenous species to transport from the substrate [183], if positively
charged, or to be neutral [152,154] or negatively charged. The form of the reaction in
Equation 7.10 avoids the above conceptual and theoretical difficulties associated with
electron capture prior to reaction.
Similar kinds of mechanisms to those discussed above have also often been invoked to
explain interface-trap buildup in MOS devices during hot carrier testing and=or high-field
stress [113,174,184–190]. In these experiments a very high electric field is applied, leading
to hot-carrier injection. These hot carriers can deposit enough energy to release hydrogen
from the oxide, or in very thin oxides, perhaps even from the polycrystalline Si gate
region. This release could either be a result of direct interaction, or triggered during
anode hole injection and=or subsequent hole transport, for example [189–192]. After its
release, it is presumed that the subsequent hydrogen transport and reactions at or near
the Si=SiO
2
interface occur very much like that described above for radiation effects
experiments.
ß 2008 by Taylor & Francis Group, LLC.
7.6 Aging and Scaling Effects
7.6.1 Aging Effects
As illus trated in Figure 7.25, MOS ox ide-trap charge is typical ly positive (there by causi ng a
nega tive thre shold voltage shift ) and recovers afte r its creation due to rad iation exposur e or
high -fi eld stress [1,100,1 93 ,194]. Consi stent with the result of Secti ons 7.3 and 7.4, thi s
rec overy is due more to the neutr alizati on of trappe d posit ive charge by compens ating
elect rons than to the true anneal ing of trap ped ho les [31,32,5 2]. Hence , the effects of oxide-
trap charge tend to be mit igated with inc reasing time, in the absence of conti nuing
irrad iation or high - field stress. In contrast, interface trap s ca n conti nue to build up with
inc reasing tim e afte r irradiat ion or elect rical stress, owing to the con tinuing transp ort
and reaction s of hydrog enous species under approp riate bias con ditions , as illu strated in
Figu re 7.26 [2,100,1 93,194]. Borde r traps are known to be le ss signi ficant in these devices
than bulk oxide traps or int erface traps . In nMOS devi ces, inter face traps shift MOS
transi stor thre shold voltages positive ly [2,13]; in pMOS devices, the thres hold voltage
shift s are nega tive, ow ing to the amp hoteric nature of inter face-tra p charge in the Si
band gap [2,13]. Interf ace traps also degrade the transis tor channel mo bility much mo re
stro ngly than oxide-tra p charge , owing to the high er ef ficiency in scatteri ng rates [195,196 ].
These com binations of thresho ld voltage shift s and mo bility degrad ation make devi ce and
IC testing a chal lengi ng and com plex iss ue, altho ugh test met hods have been develop ed to
asses s and assure the suitabili ty of MOS par ts for use in high radiation env ironmen ts
[193,194 ,197].
In Figure s 7.25 and 7.26, the oxide-tra p and inter face-tra p dens ities at long annealin g
tim es afte r pu lsed irrad iation expos ure are similar to those during low-d ose-rat e irrad i-
ation s to the same dose over the same ela psed time. This simi larity sh ows that (1) the re
are no true dose -rate effects in these MOS devices under these irr adiation and annealin g
cond itions [194], and (2) there are no tim e-depend ent agin g effect s that change the
0
−0.4
−0.8
−1.2
−1.6
−2.0
0.1 1.0 10 10
2
10
3
Time (s)
X-ray, 5550 rad (SiO
2
)/s
X-ray, 52 rad (SiO
2
)/s
Cs-137 (0.165 rad/s)
Cs-137 (0.05 rad/s)
LINAC, 2 pulses, 6 ϫ10
9
rad (SiO
2
)/s
Δ
V
o
t

(
V
)
10
4
10
5
10
6
10
7
FIGURE 7.25
Threshold voltage shifts due to oxide-trap charge as a function of irradiation and annealing time for MOS
transistors with 60 nm oxides irradiated to 100 krad (SiO
2
) and annealed at room temperature at an applied
electric field of 1 MV=cm. (From Fleetwood, D.M., Winokur, P.S., and Schwank, J.R., IEEE Trans. Nucl. Sci., 35,
1497, 1988. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
underl ying radiat ion hardne ss of the devi ces during the testing. The latter point is of ten
impl icitly assume d to be true, but is often not the case. Sig nificant changes in MOS ox ide-
trap charge and inter face-tra p charge have been observe d with differe nces in agin g times
and =or the rmal histor ies before irr adiation. For example , Figure 7.27 sh ows a compar ison
of oxide-tra p charge densi ty fo r MOS capac itors with 33.4 nm oxides irrad iated in 1986,
compar ed to devices from the same wafer irradiat ed 15 year s later [198]. In this cas e, less
oxide -trap charge is observe d. Ho wever, a subsequ ent hea t treatment of these capa citors
(proc essed with no passiv ation layers , and given a high-tem perature anne al to inc rease
their O vacancy density [199]) before irr adiation restores the or iginal radiation resp onse. In
this case it is assume d that the absorptio n of mo isture passivate d some hole trapping sites ,
but bak ing the device remov ed the pass ivating spe cies, likely a hydro gen comple x [198].
Aging has also been shown to degrade MOS radiation response. Figu re 7.28 shows
experi ments that were perfo rmed on full y proces sed MOS devices wi th P-glas s (3%
1.2
1.0
0.8
0.6
0.4
0.2
0
0.1 1.0 10 10
2
Time (s)
Δ
V
i
t


(
V
)
10
3
10
4
10
5
10
6
10
7
X-ray, 5550 rad (SiO
2
)/s
X-ray, 52 rad (SiO
2
)/s
Cs-137 (0.165 rad/s)
Cs-137
(0.05 rad/s)
LINAC, 2 pulses, 6 ϫ10
9
rad (SiO
2
)/s
FIGURE 7.26
Threshold voltage shifts due to interface-trap charge as a function of irradiation and annealing time for MOS
transistors with 60 nm oxides irradiated to 100 krad (SiO
2
) and annealed at room temperature at an applied
electric field of 1 MV=cm. (From Fleetwood, D.M., Winokur, P.S., and Schwank, J.R., IEEE Trans. Nucl. Sci., 35,
1497, 1988. With permission.)

1200
0
0.1
0.2
0.3
0.4
0
2001 1986
Dose (krad (SiO
2
))
200 400 600 800 1000
FIGURE 7.27
Oxide trap charge densities for Al-gate MOS capaci-
tors irradiatedwith10 keVx-rays at an appliedelectric
field of $1.5 MV=cm and dose rate of $1 krad
(SiO
2
)=min. (From Karmarkar, A. et al., IEEE Trans.
Nucl. Sci., 48, 2158, 2001. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
1.6
1.4
1.2
High temp no PETS
High temp pets
No PETS
PETS
1988
Δ
V
t
h

(
V
)
Room temp anneal (s) High temp anneal (s)
0.8
0.6
0.4
0.2
0
(a)
10
3
10
4
10
5
10
6
10
7
10
4
10
5
10
6
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
1
High temp no PETS
High temp pets
No PETS
PETS
1988
(b) Room temp anneal (s) High temp anneal (s)
10
3
10
4
10
5
10
6
10
7
10
4
10
5
10
6
2
1.8
1.6
1.4
1.2
1
0.8
0.6
2
1.8
1.6
1.4
1.2
1
0.8
0.6
Δ
V
i
t

(
V
)
High temp no PETS
High temp PETS
No PETS
Pets
1988
(c) Room temp anneal (s) High temp anneal (s)
10
3
10
4
10
5
10
6
10
7
10
4
10
5
10
6
0
−0.1
−0.2
−0.3
−0.4
−0.5
−0.6
−0.7
−0.8
−0.9
−1
0
−0.1
−0.2
−0.3
−0.4
−0.5
−0.6
−0.7
−0.8
−0.9
−1
Δ
V
o
t

(
V
)
FIGURE 7.28
(a) DV
th
, (b) DV
it
, and (c) DV
ot
for 60 nm gate oxide nMOS transistors stored hermetically since 1987 versus
postirradiation anneal time for exposures to 100 krad (SiO
2
). The irradiation and anneal bias was 6 V. (The 1988
data are from Fleetwood, D.M., Winokur, P.S., and Schwank, J.R., IEEE Trans. Nucl. Sci., 35, 1497, 1988.) (From
Rodgers, M.P. et al., IEEE Trans. Nucl. Sci., 52, 2652, 2005. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
P-dop ed Si O
2
) passiv ation laye rs [200]. Devic es teste d for agin g effects in Ref. [200] include
nMOS transis tors proces sed at Sandi a Nati onal Labor atori es in 1984, with poly crystalline
Si ga tes and oxide thick nesses of 3 2 or 60 nm. All irrad iations we re perform ed with a
10 keV x-ray irradiat or at ro om tem perature at 6 V bias. Some devices we re ba ked
before irrad iation. Devic es we re anneal ed at room tem perature and then 1 00 8C at 6 V
bias. Figu re 7.28a compar es thresho ld voltage shifts for nMOS transis tors for 32 nm oxide
parts irr adiated in 1988 [194] and in 2 005 [200]; the latter expo sures were perform ed with
and wi thout expo sure to preirradi ation elevated tem perature stress (PETS ). Th e thres hold
voltage rebou nd fo r par ts irr adiated in 2005 is muc h larger than for parts irrad iated in 1988
(Figure 7.28a ). Figu re 7.28b shows the estimate d threshold voltage sh ifts due to interface -
trap charge DV
it
for the 32 nm parts . By the end of the posti rradiati on room -temp erature
anneal , the value s of D V
it
for par ts not expos ed to PETS are $ 67% greater than the
maximu m DV
it
expe rienced by the se par ts in 1988. Howe ver, when parts are expo sed to
PETS prior to irrad iation, these shifts in magnitu de decre ase substant ially, but are still
grea ter than the 1988 value s [194,200 ,201 ]. Figu re 7.2 8c shows the shifts in thres hold
voltage due to oxide-tra p charge D V
ot
for the 32 nm gate oxide transisto rs [200]. Muc h
less change during irradiat ion and anne aling is obse rved for D V
ot
with agin g or baking
than for DV
it
. Th ese changes in D V
it
have bee n attri buted to mo isture ab sorption and the
subsequ ent transpo rt and reaction s of water in the sensi tive gate oxide region [200,201 ].
7.6.2 Scaling Effects and Emerg ing Mater ials
The total ionizin g dos e response of MOS devices has bee n increa singly less affecte d in
recent year s due to the sca ling dow n of gate oxide thi cknesses [1,4,202 –20 4]. For thicker
oxide s, threshold voltage shifts due to oxide and inter face-tra p charge tend to decrease as
$ 1=t
ox
2
(for othe rwise ide ntical ga te oxide s) [1,4,5]. For thinner oxides (less than $ 6 nm),
the neutralizati on of trapped posit ive oxide -trap charge via tun neling further redu ces
radiat ion-indu ced thresho ld voltage shift s [202,203 ]. Interf ace-trap formatio n follo ws
similar trends with decreas ing oxide thi ckness as oxide -trap charge [2,4,204 ,205]. The
reducti on in int erface trap density also leads to a decrease in mo bility degradat ion
with radiat ion exposur e, and a lesse ning of the requi remen ts to test for int erface -trap
related failures at ver y long times during rad iation hardnes s assuran ce testing that can
be mas ked in sh orter-term tests due to the eff ects of oxide-tra p charge [197,206 ]. Howeve r,
we note that high ly sca led micr oelectro nics devi ces and ICs are increa singly affe cted by
nega tive-bias temperat ure instabilit ies, which may well involve hydro gen releas e wi thin
the Si, follo wed by migrati on to and reaction at the Si =Si O
2
int erface [207 –212]. Th ere is
also eviden ce, at least in altern ative die lectrics to Si O
2
, that this process can be enhanced by
radiation exposure [106,107]. So one should not neglect the effects of interface traps and
trapped positive charge on threshold voltage shifts entirely in ultrathin gate oxides—
alternative mechanisms may become important, especially as operating voltages decrease
and operating speeds increase. These make some devices and ICs susceptible to even very
small threshold voltage shifts.
In addition to threshold voltage shifts, radiation exposure can also lead to enhanced
leakage in ultrat hin MOS gate oxide s. This is shown in Figure 7.29 , for example , for 8 MeV
electron irradiation of 6 nm oxides [213]. For these devices, the primary effect of the
radiation exposure was to increase the gate oxide leakage. This is usually attributed to
trap-assisted tunnelling across the gate oxide, and has also been observed for heavy
ion irradiation [214]. In the case of heavy ion exposure, one can see catastrophic failure for
even light fluences of high-LET ion exposure at high enough electric fields [214,215], but for
lower electric fields, one sees radiation-induced leakage current. More generally, stress-
induced leakage current (for applications outside a radiation environment) [175,216–218],
ß 2008 by Taylor & Francis Group, LLC.
enhanced noise [62,219,220], and oxide breakdown [190,216,221–223] are also extremely
important issues associated with oxide and interface defects in highly scaled MOS devices.
To avoid high leakage currents in highly scaled MOS gate oxides, many electronics
manufacturers are exploring the use of high-K dielectrics as potential alternatives to
thermal SiO
2
, which can show unacceptably high leakage currents for oxide thicknesses
below $1.5–2 nm [224,225]. The use of high-K materials would enable device dielectric
scaling to continue over ranges in which SiO
2
becomes unmanageably thin to manufacture
and=or integrate reliably, by replacing it with a material that is physically thicker, but
equivalent to ultrathin SiO
2
electrically. The radiation responses of high-K dielectrics have
been investigated, and typically are inferior to high-quality thermal SiO
2
, but are improv-
ing with increasing high-K material quality [106–108,226,227], and should be expected to
continue to do so as high-K dielectrics are adopted in commercial manufacturing processes.
For modern commercial microelectronics, the most sensitive regions of the device for
total ionizing dose are the relatively thick isolation oxides. These isolate devices effectively
FIGURE 7.29
Current–voltage measurements for capacitors with
6 nm oxides irradiated from 0 to 50 Mrad (Si) with
8 MeV electrons. (From Ceschia, M. et al., IEEE Trans.
Nucl. Sci., 45, 2375, 1998. With permission.)
10
−4
10
−5
10
−6
10
−7
10
−8
10
−9
10
−10
2

J
g


(
A
/
c
m
2
)
⎢E
ox
⎢(MV/cm)
3 4
4
5 6 7
Dose (Mrad (Si)):
50
Fresh
8
10
−2
10
−3
10
−4
10
−5
10
−6
10
−7
1 M
500 K
200 K
100 K
50 K
30 K
20 K
Pre
10
−8
10
−9
10
−10
10
−11
10
−12
−6 −5 −4 −3 −2 −1 0
V
GS
(V)
I
D
S

(
A
)
1 2 3 4
FIGURE 7.30
Current as a function of voltage for nMOS transistors with unhardened, shallow trench isolation (STI), irradiated
with 10 keV x-rays at 167 rad (SiO
2
)=s. (After Shaneyfelt, M.R. et al., IEEE Trans. Nucl. Sci., 45, 2584, 1998. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
for app lications that do not includ e rad iation expo sure, but of ten trap enough charge to
cause failure s a t relative ly low doses when users attemp t to emp loy these in rad iation
environ ments . An exa mple of this for unhard ened trench isolat ion is see n in Figure 7.30
[228]. These devices wou ld on ly be suitable for very low-d ose radiat ion environ men ts.
Howev er, other com mercial micro electroni cs techno logies with STI have shown more
accepta ble hard ness for space environ ments [229]. Si milar issues are obse rved fo r
the burie d oxide s of com mercial SOI technol ogies [230]. For SOI technol ogies with
harde ned buried oxide s, thou gh, opp ortunities to ope rate the devices in doub le gate
mode, as illu strated in Figu re 7.31 [231], and othe r novel device struc tures make SO I of
conti nuing int erest fo r futur e com mercial and radiat ion-tolerant micro electroni cs.
Ther e are con siderable hardnes s assuran ce challenge s (lot- to-lot variab ility, process
changes made withou t con siderati on for radiation effect s respo nse, etc. ) a ssociate d with
using com mercial micr oelect ronics tech nologi es, as opposed to radiation harde ned or
tolerant devi ces. But the lower the dose of the appl ication and the less critica l the part,
the mo re bene fi ts some com mercial par ts may of fer in ter ms of inc reased functional ity and
reduce d purch ase expe nse. Howeve r, of ten qual ifi cation costs ca n overwhe lm purch ase
price sav ings, so the costs to the full syste m must be conside red as we ll. For thi s reason,
harde ning by design is an increa singly popular approach to deve loping rad iation-
harde ned techno logies [229,232 ] and likely will become an even mo re popular approach
for advance d defense and space systems with high radiat ion requi remen ts.
7.7 Summary and Conclusions
Significant insights have been obtained into the nature of the defects that limit MOS
performance, reliability, and radiation response. Oxide-trap charge in high-quality thermal
V
s
Gate oxide t
ox
t
Si
t
BOX
V
D
Buried oxide
I
D

(
μ
A
)
S D
Si substrate
V
FG
V
BG
7
6
5
4
3
2
1
0
8
7
6
5
4
3
2
0
1
0.0 0.3
Double gate
W/L = 2/10
Single gate
g
m

(
μ
S
)
0.6 0.9
V
FG
(V)
1.2 1.5 1.8 2.1
FIGURE 7.31
(Left) Schematic cross-sectional diagram of a fully depleted SOI MOSFET. (Right) Current–voltage and transcon-
ductance characteristics for fully depleted nMOS transistors operated in double gate mode. Front to back-gate
coupling of the electric fields can lead to improved device mobility and potentially enhanced radiation response in
double gate mode of device operation. (After Jun, B. et al., IEEE Trans. Nucl. Sci., 51, 3767, 2004. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
SiO
2
tends to be dominated by O vacancy-related defects. Interface traps are formed
primarily through the interaction of hydrogenous species, liberated by irradiation or
high-field stress, with Si–H bonds at the Si=SiO
2
interface, leaving a dangling Si bond.
Border traps are near-interfacial oxide traps that can exchange charge with the underlying
Si on the timescale of measurements of interest. These defects can trap charge on timescales
ranging fromnanoseconds to years, the effects of which can often be difficult to describe in
a simple and straightforward fashion. Hence, the most reliable method of ensuring that
devices are not limited by defects is to limit their densities through processing techniques
that limit the number of O vacancies in SiO
2
(especially reducing high-temperature pro-
cessing steps) and the amount of hydrogen in the devices. With these precautions, MOS
devices have been successfully built and integrated into systems that have functioned
successfully for many years in commercial and defense applications. The integration of a
wide variety of new materials into the microelectronics fabrication sequence ensures that
new challenges will continue to arise in future generations of technologies, leading to a
continuing need to develop a deeper understanding of defects in the MOS systemand their
sometimes surprising effects on device and circuit response.
Acknowledgments
Portions of this work have been supported by the Air Force Office of Scientific Research,
the U.S. Navy, the Defense Threat Reduction Agency, and the U.S. Department of Energy.
We wish to thank our many collaborators that have contributed strongly to the efforts
described above. The insights obtained are due to their dedicated efforts.
References
1. McLean, F.B., Boesch, H.E. Jr., and Oldham, T.R., Electron–hole generation, transport,
and trapping in SiO
2
, in Ionizing Radiation Effects in MOS Devices & Circuits, Ma, T.P. and
Dressendorfer, P.V., Eds., Wiley, New York, 1989, Chapter 3.
2. Winokur, P.S., Radiation-induced interface traps, in Ionizing Radiation Effects in MOS Devices &
Circuits, Ma, T.P. and Dressendorfer, P.V., Eds., Wiley, New York, 1989, Chapter 4.
3. Holmes-Seidle, A. and Adams, L., Handbook of Radiation Effects, 2nd edn., Oxford University
Press, Oxford, 2002, Chapter 4.
4. Schwank, J.R., Total dose effects in MOS devices, in 2002 IEEE Nuclear and Space Radiation Effects
Conference Short Course Notebook, Dressendorfer, P.V., Ed., IEEE, Piscataway, New Jersey, avail-
able at nsrec.com, 2002, Chapter 3.
5. Oldham, T.R. and McLean, F.B., Total ionizing dose effects in MOS oxides and devices,
IEEE Trans. Nucl. Sci., 50, 483, 2003.
6. Fleetwood, D.M., Border traps in MOS devices, IEEE Trans. Nucl. Sci., 39, 269, 1992.
7. Fleetwood, D.M. et al., Effects of oxide traps, interface traps, and ‘‘border traps’’ on MOS
Devices, J. Appl. Phys., 73, 5058, 1993.
8. Fleetwood, D.M. et al., Border traps: Issues for MOS radiation response and long-termreliability,
Microelectron. Reliab., 35, 403, 1995.
9. Fleetwood, D.M., Fast and slow border traps in MOS devices, IEEE Trans. Nucl. Sci., 43, 779,
1996.
10. Nicollian, E.H. and Brews, J.R., MOS (Metal Oxide Semiconductor) Physics and Technology, Wiley,
New York, 1982.
ß 2008 by Taylor & Francis Group, LLC.
11. Winokur, P.S. et al., Correlating the radiation response of MOS capacitors and transistors,
IEEE Trans. Nucl. Sci., 31, 1453–1460, 1984.
12. Schroder, D.K., Semiconductor Material and Device Characterization, 3rd edn. Wiley, New York,
2006.
13. Lenahan, P.M. and Dressendorfer, P.V., Hole traps and trivalent Si centers in MOS devices,
J. Appl. Phys., 55, 3495, 1984.
14. Fleetwood, D.M., Long-term annealing study of midgap interface-trap charge neutrality,
Appl. Phys. Lett., 60, 2883, 1992.
15. Reber, R.A. Jr. and Fleetwood, D.M., Thermally-stimulated-current measurements of SiO
2
defect
density and energy in irradiated MOS capacitors, Rev. Sci. Instrum., 63, 5714, 1992.
16. Fleetwood, D.M. et al., Thermally stimulated current in SiO
2
, Microelectron. Reliab., 39,
1323, 1999.
17. Groeseneken, G. et al., A reliable approach to charge-pumping measurements in MOS tran-
sistors, IEEE Trans. Electron Dev., 31, 42, 1984.
18. McWhorter, P.J., Winokur, P.S., and Pastorek, R.A., Donor=acceptor nature of radiation-induced
interface traps, IEEE Trans. Nucl. Sci., 35, 1154, 1988.
19. Fleetwood, D.M., Shaneyfelt, M.R., and Schwank, J.R., Estimating oxide-trap, interface-trap, and
border-trap charge densities in MOS transistors, Appl. Phys. Lett., 64, 1965, 1994.
20. Fleetwood, D.M. et al., Effects of interface traps and border traps on MOS postirradiation
annealing response, IEEE Trans. Nucl. Sci., 42, 1698, 1995.
21. Simmons, J.G. and Taylor, G.W., High-field isothermal currents and thermally stimulated
currents in insulators having discrete trapping levels, Phys. Rev. B, 5, 1619, 1972.
22. Simmons, J.G., Taylor, G.W., and Tam, M.C., Thermally stimulated currents in semiconductors
and insulators having arbitrary trap distributions, Phys. Rev. B, 7, 3714, 1973.
23. Esqueda, P.D. and Das, M.B., Characterization of surface states in HCl grown oxides using MOS
transient currents, Solid State Electron., 23, 365, 1980.
24. Shanfield, Z., Thermally stimulated current measurements on irradiated MOS capacitors,
IEEE Trans. Nucl. Sci., 30, 4064, 1983.
25. Shanfield, Z. and Moriwaki, M.M., Characteristics of hole traps in dry and pyrogenic gate
oxides, IEEE Trans. Nucl. Sci., 31, 1242, 1984.
26. Shanfield, Z. and Moriwaki, M.M., Radiation-induced hole trapping and interface state charac-
teristics of Al-gate and poly-Si gate MOS capacitors, IEEE Trans. Nucl. Sci., 31, 3929, 1985.
27. Shanfield, Z. and Moriwaki, M.M., Critical evaluation of the midgap-voltage-shift method
for determining oxide trapped charge in irradiated MOS devices, IEEE Trans. Nucl. Sci., 34,
1159, 1987.
28. Antonenko, V.I., Zhdan, A.G., and Sul’zhenko, P.S., Identification of spatial localization
of surface states in experiments involving thermally stimulated discharge of a metal-insulator-
semiconductor capacitor, Sov. Phys. Semicond., 22, 473, 1988 [Fiz. Tekh. Poluprovodn., 20, 758,
1988].
29. Shanfield, Z. et al., A new MOS radiation-induced charge: Negative fixed interface charge,
IEEE Trans. Nucl. Sci., 39, 303, 1992 [J. Radiat. Effects Res. Eng., 8, 1, 1990].
30. Fleetwood, D.M., Reber, R.A. Jr., and Winokur, P.S., Effect of bias on thermally stimulated
current (TSC) in irradiated MOS devices, IEEE Trans. Nucl. Sci., 1066, 38, 1991.
31. Fleetwood, D.M., Reber, R.A. Jr., and Winokur, P.S., Trapped-hole annealing and electron
trapping in MOS devices, Appl. Phys. Lett., 60, 2008, 1992.
32. Fleetwood, D.M. et al., New insights into radiation-induced oxide-trap charge through
thermally-stimulated-current measurement and analysis, IEEE Trans. Nucl. Sci., 39, 2192, 1992.
33. Miller, S.L., Fleetwood, D.M., and McWhorter, P.J., Determining the energy distribution of
traps in insulating thin films using the thermally stimulated current technique, Phys. Rev. Lett.,
69, 820, 1992.
34. Fleetwood, D.M. et al., The role of border traps in MOS high-temperature postirradiation
annealing response, IEEE Trans. Nucl. Sci., 40(6), 1323, 1993.
35. Miller, S.L. et al., A general centroid determination methodology, with application to multi-
layer dielectric structures and thermally stimulated current measurements, J. Appl. Phys., 74,
5068, 1993.
ß 2008 by Taylor & Francis Group, LLC.
36. Raynaud, C. et al., Electrical characterization of instabilities in 6H SiC MOS capacitors,
J. Appl. Phys., 76, 993, 1994.
37. Fleetwood, D.M. et al., Physical mechanisms contributing to enhanced bipolar gain degradation
at low dose rates, IEEE Trans. Nucl. Sci., 41, 1871, 1994.
38. Saks, N.S. et al., Radiation effects in oxynitrides grown in N
2
O, IEEE Trans. Nucl. Sci., 41,
1854, 1994.
39. Shaneyfelt, M.R. et al., Effects of burn-in on radiation hardness, IEEE Trans. Nucl. Sci., 41,
2550, 1994.
40. Warren, W.L. et al., Microscopic nature of border traps in MOS oxides, IEEE Trans. Nucl. Sci.,
41, 1817, 1994.
41. Lu, Y. and Sah, C.T., Thermal emission of trapped holes in thin SiO
2
films, J. Appl. Phys., 78,
3156, 1995.
42. Paillet, P. et al., Trapping-detrapping properties of irradiated ultra-thin SIMOX buried oxides,
IEEE Trans. Nucl. Sci., 42, 2108, 1995.
43. Warren, W.L. et al., Electron and hole trapping in doped oxides, IEEE Trans. Nucl. Sci., 42,
1731, 1995.
44. Fleetwood, D.M. and Saks, N.S., Oxide, interface, and border traps in thermal, N
2
O, and N
2
O-
nitrided oxides, J. Appl. Phys., 79, 1583, 1996.
45. Fleetwood, D.M. et al., Radiation effects at low electric fields in thermal, SIMOX, and bipolar-
base oxides, IEEE Trans. Nucl. Sci., 43, 2537, 1996.
46. Fleetwood, D.M., Revised model of thermally stimulated current in MOS capacitors, IEEE Trans.
Nucl. Sci., 44, 1826, 1997.
47. Cai, J. and Sah, C.T., Theory of thermally stimulated charges in MOS gate oxide, J. Appl. Phys.,
83, 851, 1998.
48. Dmitriev, S.G. and Markin, Yu.V., Macroscopic ion traps at the silicon-oxide interface, Semicon-
ductors, 32, 1284, 1998 [Fiz. Tekh. Poluprovodn., 32, 1439, 1998].
49. Fleetwood, D.M. et al., Bulk oxide and border traps in MOS capacitors, J. Appl. Phys., 84,
6141, 1998.
50. Fleetwood, D.M. et al., Effects of isochronal annealing and irradiation temperature on radiation-
induced trapped charge, IEEE Trans. Nucl. Sci., 45, 2366, 1998.
51. Paillet, P. et al., Simulation of multi-level radiation-induced charge trapping and thermally,
activated phenomena in SiO
2
, IEEE Trans. Nucl. Sci., 45, 1379, 1998.
52. Fleetwood, D.M. et al., The role of electron transport and trapping in MOS total-dose modeling,
IEEE Trans. Nucl. Sci., 46, 1519, 1999.
53. Fleetwood, D.M. et al., Dielectric breakdown of thin oxides during ramped current-temperature
stress, IEEE Trans. Nucl. Sci., 47, 2305, 2000.
54. Felix, J.A. et al., Bias and frequency dependence of radiation-induced charge trapping in MOS
devices, IEEE Trans. Nucl. Sci., 48, 2114, 2001.
55. Fleetwood, D.M. et al., Unified model of hole trapping, 1=f noise, and thermally stimulated
current in MOS devices, IEEE Trans. Nucl. Sci., 49, 2674, 2002.
56. Kaschieva, S., Todorova, Z., and Dmitriev, S.N., Radiation defects induced by 20 MeV electrons
in MOS structures, Vacuum, 76, 307, 2004.
57. McWhorter, P.J. et al., Modeling the memory retention characteristics of SNOS nonvolatile
transistors in a varying thermal environment, J. Appl. Phys., 68, 1902, 1990.
58. Tzou, J.J., Sun, J.Y.C., and Sah, C.T., Field dependence of two large hole capture cross sections in
thermal oxide on silicon, Appl. Phys. Lett., 43, 861, 1983.
59. Fleetwood, D.M. and Scofield, J.H., Evidence that similar point defects cause 1=f noise and
radiation-induced-hole trapping in MOS transistors, Phys. Rev. Lett., 64, 579, 1990.
60. Scofield, J.H., Doerr, T.P., and Fleetwood, D.M., Correlation of preirradiation 1=f noise
and postirradiation oxide trapped charge in MOS transistors, IEEE Trans. Nucl. Sci., 36,
1946, 1989.
61. Scofield, J.H. and Fleetwood, D.M., Physical basis for nondestructive tests of MOS radiation
hardness, IEEE Trans. Nucl. Sci., 38, 1567, 1991.
62. Fleetwood, D.M., Meisenheimer, T.L., and Scofield, J.H., 1=f noise and radiation effects in MOS
devices, IEEE Trans. Electron Dev., 41, 1953, 1994.
ß 2008 by Taylor & Francis Group, LLC.
63. Fleetwood, D.M. et al., Enhanced MOS 1=f noise due to near-interfacial oxygen deficiency,
J. Non-Cryst. Solids, 187, 199, 1995.
64. Meisenheimer, T.L. and Fleetwood, D.M., Effect of radiation-induced charge on 1=f noise in MOS
devices, IEEE Trans. Nucl. Sci., 37, 1696, 1990.
65. Meisenheimer, T.L. et al., 1=f noise in n- and p-channel MOS devices through irradiation and
annealing, IEEE Trans. Nucl. Sci., 38, 1297, 1991.
66. Babcock, J.A. et al., Effects of ionizing radiation on the noise properties of DMOS power
transistors, IEEE Trans. Nucl. Sci., 38, 1304, 1991.
67. Simoen, E. et al., DC and low frequency noise characteristics of gamma-irradiated gate-all-around
SOI MOS transistors, Solid State Electron., 38, 1, 1995.
68. Simoen, E. and Claeys, C., The low-frequency noise behavior of silicon-on-insulator technolo-
gies, Solid State Electron., 39, 949, 1996.
69. Simoen, E. et al., Short-channel radiation effect in 60 MeV proton irradiated 0.13 mm CMOS
transistors, IEEE Trans. Nucl. Sci., 50, 2426, 2003.
70. Schwank, J.R. and Fleetwood, D.M., The effect of post-oxidation anneal temperature on
radiation-induced charge trapping in poly-crystalline Si gate MOS devices, Appl. Phys. Lett.,
53, 770, 1988.
71. Warren, W.L. et al., Links between oxide, interface, and border traps in high-temperature
annealed Si=SiO
2
systems, Appl. Phys. Lett., 64, 3452, 1994.
72. Ning, T.H., High-field capture of electrons by Coulomb-attractive centers in SiO
2
, J. Appl. Phys.,
47, 3203, 1976.
73. Fleetwood, D.M., Radiation-induced charge neutralization and interface-trap buildup in MOS
devices, J. Appl. Phys., 67, 580, 1990.
74. Conley, J.F. Jr., et al., Observation and electrical characterization of new E
0
center defects in
technologically relevant thermal SiO
2
on Si: An additional complexity in oxide charge trapping,
J. Appl. Phys., 76, 2872, 1994.
75. Chavez, J.R. et al., Microscopic structure of the E
d
0
center in amorphous SiO
2
: A first principles
quantum mechanical investigation, IEEE Trans. Nucl. Sci., 44, 1799, 1997.
76. Karna, S.P. et al., Electronic structure theory and mechanisms of the oxide trapped hole anneal-
ing process, IEEE Trans. Nucl. Sci., 47, 2316, 2000.
77. Pineda, A.C. and Karna, S.P., Effect of hole trapping on the microscopic structure of oxygen
vacancy sites in a-SiO
2
, J. Phys. Chem. A, 104, 4699, 2000.
78. Lu, Z.Y. et al., Structure, properties, and dynamics of oxygen vacancies in amorphous SiO
2
,
Phys. Rev. Lett., 89, 285505-1, 2002.
79. Nicklaw, C.J. et al., The structure, properties, and dynamics of oxygen vacancies in amorphous
SiO
2
, IEEE Trans. Nucl. Sci., 49, 2667, 2002.
80. Stirling, A. and Pasquarello, A., First-principles modeling of paramagnetic Si dangling-bond
defects in amorphous SiO
2
, Phys. Rev. B, 66, 245201, 2002.
81. Mukhopadhyay, S. et al., Modeling of the structure and properties of oxygen vacancies in
amorphous silica, Phys. Rev. B, 70, 195203, 2004.
82. Uchino, T. and Yoko, T., Density functional theory of structural transformations of oxygen-
deficient centers in amorphous silica during hole trapping: Structure and formation mechanism
of the E
0
g
center, Phys. Rev. B, 74, 125203, 2006.
83. Buscarino, G., Agnello, S., and Gelardi, F.M., Delocalized nature of the E
d
0
center in amorphous
SiO
2
, Phys. Rev. Lett., 94, 125501, 2005.
84. Buscarino, G., Agnello, S., and Gelardi, F.M., Hyperfine structure of the E
d
0
center in amorphous
SiO
2
, J. Phys. Condens. Matter, 18, 5213, 2006.
85. Feigl, F.J., Fowler, W.B., and Yip, K.L., Oxygen vacancy model for E
1
0
center in SiO
2
, Solid State
Commun., 14, 225, 1974.
86. O’Reilly, E.P. and Robertson, J., Theory of defects in vitreous SiO
2
, Phys. Rev. B, 27, 3780,
1983.
87. Warren, W.L. et al., Paramagnetic point defects in amorphous SiO
2
and amorphous Si
3
N
4
thin
films, I. a-SiO
2
, J. Electrochem. Soc., 139, 872, 1992.
88. Rudra, J.K. and Fowler, W. B., Oxygen vacancy and the E
1
0
center in crystalline SiO
2
, Phys. Rev. B,
35, 8223, 1987.
ß 2008 by Taylor & Francis Group, LLC.
89. Mrstik, B.J. et al., A study of the radiation sensitivity of non-crystalline SiO
2
films using
spectroscopic ellipsometry, IEEE Trans. Nucl. Sci., 45, 2450, 1998.
90. Lelis, A.J. et al., The nature of the trapped hole annealing process, IEEE Trans. Nucl. Sci., 36, 1808,
1989.
91. Zekeriya, V. and Ma, T.P., Effect of stress relaxation on the generation of radiation-induced
interface traps in post-metal-annealed Al–SiO
2
–Si devices, Appl. Phys. Lett., 45, 249, 1984.
92. Kasama, K. et al., Mechanical stress dependence of radiation effects in MOS structures,
IEEE Trans. Nucl. Sci., 33, 1210, 1985.
93. Witczak, S.C. et al., Relaxation of the Si–SiO
2
interfacial stress in bipolar screen oxides due to
ionizing radiation, IEEE Trans. Nucl. Sci., 42, 1689, 1996.
94. Dutta, P. and Horn, P.M., Low-frequency fluctuations in solids: 1=f noise, Rev. Mod. Phys., 53,
497, 1981.
95. Weissman, M.B., 1=f noise and other slow, non-exponential kinetics in condensed matter,
Rev. Mod. Phys., 60, 537, 1988.
96. Xiong, H.D. et al., Temperature dependence and irradiation response of 1=f noise in MOSFETs,
IEEE Trans. Nucl. Sci., 49, 2718, 2002.
97. Xiong, H.D., Low-frequency noise and radiation response of buried oxides in SOI nMOS
transistors, IEEE Proc. Circ. Dev. Syst., 151, 118, 2004.
98. Shu, K. et al., Role of stress in irradiation-then-anneal technique for improving radiation
hardness of metal–insulator–semiconductor devices, Appl. Phys. Lett., 61, 675, 1992.
99. Pepper, M., Inversion layer transport and the radiation hardness of the Si–SiO
2
interface,
IEEE Trans. Nucl. Sci., 25, 1283, 1978.
100. Schwank, J.R. et al., Physical mechanisms contributing to device ‘‘rebound,’’ IEEE Trans. Nucl.
Sci., 31, 1434, 1984.
101. Nissan-Cohen, Y., Shappir, J., and Frohman-Bentchkowsky, D., Dynamic model of trapping–
detrapping in SiO
2
, J. Appl. Phys., 57, 2252, 1985.
102. Lelis, A.J. et al., Reversibility of trapped hole annealing, IEEE Trans. Nucl. Sci., 35, 1186, 1988.
103. Roh, Y., Trombetta, L., and DiMaria, D.J., Interface traps induced by hole trapping in MOS
devices, J. Non-Cryst. Solids, 187, 165, 1995.
104. Conley, J.H. Jr. et al., Electron spin resonance evidence that E
g
0
centers can behave as switching
oxide traps, IEEE Trans. Nucl. Sci., 42, 1744, 1995.
105. Schwank, J.R. et al., Generation of metastable electron traps in the near interfacial region of
SOI buried oxides by ion implantation and their effect on device properties, Microelectron. Eng.,
72, 362, 2004.
106. Zhou, X.J. et al., Bias-temperature instabilities and radiation effects in MOS devices, IEEE Trans.
Nucl. Sci., 52, 2231, 2005.
107. Zhou, X.J. et al., Effects of switched-bias annealing on charge trapping in HfO
2
gate dielectrics,
IEEE Trans. Nucl. Sci., 53, 3636, 2006.
108. Lucovsky, G. et al., Differences between charge trapping states in irradiated nano-crystalline
HfO
2
and non-crystalline Hf silicates, IEEE Trans. Nucl. Sci., 53, 3644, 2006.
109. Devine, R.A.B. et al., Point defect generation and oxide degradation during annealing of the
Si=SiO
2
interface, Appl. Phys. Lett., 63, 2926, 1993.
110. Lai, S.K., Two-carrier nature of interface-state generation in hole trapping and radiation damage,
Appl. Phys. Lett., 39, 58, 1981.
111. Fleetwood, D.M. et al., Dipoles in SiO
2
: Border traps or not? in PV 2003–02 Silicon Nitride and
Silicon Dioxide Thin Insulating Films, 7th edn., Sah, R.E. et al. Eds., The Electrochemical Society,
Pennington, NJ, 2003, p. 291.
112. Fleetwood, D.M. et al., Stability of trapped electrons in SiO
2
, Appl. Phys. Lett., 74, 2969, 1999.
113. Trombetta, L.P. et al., An EPR study of electron injected oxides in MOS capacitors, J. Appl. Phys.,
64, 2434, 1988.
114. Stahlbush, R.E., Mrstik, B.J., and Lawrence, R., Post-irradiation behavior of the interface state
density and the trapped positive charge, IEEE Trans. Nucl. Sci., 37, 1641, 1990.
115. Trombetta, L.P., Feigl, F.J., and Zeto, R.J., Positive charge generation in MOS capacitors, J. Appl.
Phys., 69, 2512, 1991.
ß 2008 by Taylor & Francis Group, LLC.
116. Stahlbush, R.E. et al., Post-irradiation cracking of H
2
and formation of interface states in
irradiated MOSFETs, J. Appl. Phys., 73, 658, 1993.
117. Freitag, R.K., Brown, D.B., and Dozier, C.M., Evidence for two types of radiation-induced
trapped positive charge, IEEE Trans. Nucl. Sci., 41, 1828, 1994.
118. Fleetwood, D.M., Effects of hydrogen transport and reactions on microelectronics radiation
response and reliability, Microelectron. Reliab., 42, 523, 2002.
119. Vanheusden, K. et al., Nonvolatile memory device based on mobile protons in SiO
2
thin films,
Nature, 386, 587, 1997.
120. Vanheusden, K. et al., Irradiation response of mobile protons in buried SiO
2
films, IEEE Trans.
Nucl. Sci., 44, 2087, 1997.
121. Fleetwood, D.M. et al., 1=f noise, hydrogen transport, and latent interface-trap buildup in
irradiated MOS devices, IEEE Trans. Nucl. Sci., 44, 1810, 1997.
122. Pantelides, S.T. et al., Reactions of hydrogen with Si–SiO
2
interfaces, IEEE Trans. Nucl. Sci.,
47, 2262, 2000.
123. Bunson, P.E. et al., Hydrogen-related defects in irradiated SiO
2
, IEEE Trans. Nucl. Sci., 47, 2289,
2000.
124. Rashkeev, S.N. et al., Statistical modeling of radiation-induced proton transport in Si: Deactivation
of dopant acceptors in bipolar devices, IEEE Trans. Nucl. Sci., 50, 1896, 2003.
125. Shaneyfelt, M.R. et al., Annealing behavior of linear bipolar devices with enhanced low-dose-
rate sensitivity, IEEE Trans. Nucl. Sci., 51, 3172, 2004.
126. Rashkeev, S.N. et al., Effects of hydrogen motion on interface trap formation and annealing,
IEEE Trans. Nucl. Sci., 51, 3158, 2004.
127. Revesz, A.G., Defect structure and irradiation behavior of noncrystalline SiO
2
, IEEE Trans.
Nucl. Sci., 18, 113, 1971.
128. Revesz, A.G., Chemical and structural aspects of the irradiation behavior of SiO
2
films on silicon,
IEEE Trans. Nucl. Sci., 2102, 24, 1977.
129. Svensson, C.M., The defect structure of the Si–SiO
2
interface, a model based on trivalent silicon
and its hydrogen compounds, in The Physics of SiO
2
and its Interfaces, Pantelides, S.T. Ed.,
Pergamon, New York, 1978, pp. 328–332.
130. Winokur, P.S. and Sokoloski, M.M., Comparison of interface-state buildup in MOS capacitors
subjected to penetrating and nonpenetrating radiation, Appl. Phys. Lett., 28, 627, 1976.
131. Ma, T.P., Scoggan, G., and Leone, R., Comparison of interface-state generation by 25-keV
electron beam irradiation in p-type and n-type MOS capacitors, Appl. Phys. Lett., 27, 61, 1975.
132. Winokur, P.S., McGarrity, J.M., and Boesch, Jr., H.E., Dependence of interface-state buildup on
hole generation and transport in irradiated MOS capacitors, IEEE Trans. Nucl. Sci., 23, 1580, 1976.
133. Winokur, P.S. et al., Field- and time-dependent radiation effects at the SiO
2
=Si interface of
hardened MOS capacitors, IEEE Trans. Nucl. Sci., 24, 2113, 1977.
134. Winokur, P.S. et al., Two-stage process for buildup of radiation-induced interface states, J. Appl.
Phys., 50, 3492, 1979.
135. McLean, F.B., A framework for understanding radiation-induced interface states in SiO
2
MOS
structures, IEEE Trans. Nucl. Sci., 27, 1651, 1980.
136. Hu, G. and Johnson. W.C., Relationship between trapped holes and interface states in MOS
capacitors, Appl. Phys. Lett., 36, 590, 1980.
137. Sah, C.T., Models and experiments on degradation of oxidized silicon, Solid-St. Electron,
33, 147, 1990.
138. Grunthaner, F.J., Grunthaner, P.J., and Maserjian, J., Radiation-induced defects in SiO
2
as
determined with XPS, IEEE Trans. Nucl. Sci., 29, 1462, 1982.
139. Lai, S.K., Interface trap generation in SiO
2
when electrons are captured by trapped holes, J. Appl.
Phys., 54, 2540, 1983.
140. Sabnis, A.G., Characterization of annealing of Co-60 gamma-ray damage at the Si=SiO
2
interface,
IEEE Trans. Nucl. Sci., 30, 4094, 1983.
141. Sabnis, A.G., Process dependent buildup of interface states in irradiated n-channel MOSFETs,
IEEE Trans. Nucl. Sci., 32, 3905, 1985.
142. Grunthaner, F.J. and Gruthaner, P.J., Chemical and electronic structure of the SiO
2
=Si interface,
Mater. Sci. Rep., 1, 65, 1986.
ß 2008 by Taylor & Francis Group, LLC.
143. Kenkare, P.U. and Lyon, S.A., Relationship between trapped holes, positive ions, and interface
states in irradiated Si–SiO
2
structures, Appl. Phys. Lett., 55, 2328, 1989.
144. Pershenkov, V.S. et al., Three-point method of prediction of MOS device response in space
environments, IEEE Trans. Nucl. Sci., 40(6), 1714, 1993.
145. Witham, H.S. and Lenahan, P.M., The nature of the deep hole trap in MOS oxides, IEEE Trans.
Nucl. Sci., 34, 1147, 1987.
146. Nishi, Y., Study of Si–SiO
2
structure by electron spin resonance I, Jpn. J. Appl. Phys., 10, 52, 1971.
147. Poindexter, E.H. et al., Interface states and electron spin resonance centers in thermally oxidized
(111) and (100) Si wafers, J. Appl. Phys., 52, 879, 1981.
148. Brower, K. L., Kinetics of H
2
passivation of P
b
centers at the (111) Si–SiO
2
interface, Phys. Rev. B,
38, 9657, 1988.
149. Kim, Y.Y. and Lenahan, P.M., Electron-spin-resonance study of radiation-induced paramagnetic
defects in oxides grown on (100) Si substrates, J. Appl. Phys., 64, 3551, 1988.
150. Schwank, J.R. et al., Radiation induced interface-state generation in MOS devices, IEEE Trans.
Nucl. Sci., 33, 1178, 1986.
151. Boesch, Jr., H.E., Time-dependent interface trap effects in MOS devices, IEEE Trans. Nucl. Sci.,
35, 1160, 1988.
152. Saks, N.S. and Ancona, M.G., Generation of interface states by ionizing radiation at 80 K
measured by charge pumping and subthreshold slope techniques, IEEE Trans. Nucl. Sci.,
34, 1348, 1987.
153. Saks, N.S., Dozier, C.M., and Brown, D.B., Time dependence of interface trap formation in
MOSFETs following pulsed irradiation, IEEE Trans. Nucl. Sci., 35, 1168, 1988.
154. Saks, N.S., Klein, R.B., and Griscom, D.L., Formation of interface traps in MOSFETs during
annealing following low temperature irradiation, IEEE Trans. Nucl. Sci., 35, 1234, 1988.
155. Saks, N.S. and Brown, D.B., Interface trap formation via the two-stage H
þ
process, IEEE Trans.
Nucl. Sci., 36, 1848, 1989.
156. Saks, N.S. and Brown, D.B., Observation of H
þ
motion during interface trap formation,
IEEE Trans. Nucl. Sci., 37, 1624, 1990.
157. Brown, D.B. and Saks, N.S., Time dependence of radiation-induced interface trap formation in
MOS devices as a function of oxide thickness and applied field, J. Appl. Phys., 70, 3734, 1991.
158. Saks, N.S., Brown, D.B., and Rendell, R.W., Effects of switched gate bias on radiation-induced
interface-trap formation, IEEE Trans. Nucl. Sci., 38, 1130, 1991.
159. Saks, N.S. and Rendell, R.W., The time-dependence of post-irradiation interface trap buildup in
deuterium annealed oxides, IEEE Trans. Nucl. Sci., 39, 2220, 1992.
160. McLean, F.B., Boesch, Jr., H.E., and McGarrity, J.M., Hole transport and recovery characteristics
of SiO
2
gate insulators, IEEE Trans. Nucl. Sci., 23, 1506, 1976.
161. Kohler, R.A., Kushner. R.A., and Lee, K.H., Total dose radiation hardness of MOS devices in
hermetic ceramic packages, IEEE Trans. Nucl. Sci., 35, 1492, 1988.
162. Mrstik, B.J. and Rendell, R.W., Model for Si–SiO
2
interface state formation during irradiation and
during post-irradiation exposure to hydrogen environment, Appl. Phys. Lett., 59, 3012, 1991.
163. Mrstik, B.J. and Rendell, R.W., Si–SiO
2
interface state generation during x-ray irradiation
and during post-irradiation exposure to a hydrogen ambient, IEEE Trans. Nucl. Sci., 38, 1101,
1991.
164. Conley, J.F. and Lenahan, P.M., Roomtemperature reactions involving Si dangling bond centers
and molecular hydrogen in a-SiO
2
thin films on Si, IEEE Trans. Nucl. Sci., 39, 2186, 1992.
165. Conley, J.F. Jr. and Lenahan, PM., Molecular hydrogen, E
0
center hole traps, and radiation
induced interface traps in MOS devices, IEEE Trans. Nucl. Sci., 40, 1335, 1993.
166. Saks, N.S. et al., Effects of post-stress hydrogen annealing on MOS oxides after Co-60 irradiation
or Fowler-Nordheim injection, IEEE Trans. Nucl. Sci., 40(6), 1341, 1993.
167. Cartier, E., Stathis, J.H., and Buchanan, D.A., Passivation and depassivation of Si dangling bonds
at the Si=SiO
2
interface by atomic hydrogen, Appl. Phys. Lett., 63, 1510, 1993.
168. Stahlbush, R.E. and Cartier, E., Interface defect formation in MOSFETs by atomic hydrogen
exposure, IEEE Trans. Nucl. Sci., 41, 1844, 1994.
169. Shaneyfelt, M.R. et al., Interface-trap buildup rates in wet and dry oxides, IEEE Trans. Nucl. Sci.,
39, 2244, 1992.
ß 2008 by Taylor & Francis Group, LLC.
170. Shaneyfelt, M.R. et al., Field dependence of interface-trap buildup in polysilicon and metal gate
MOS devices, IEEE Trans. Nucl. Sci., 37, 1632, 1990.
171. Griscom, D.L., Diffusion of radiolytic molecular hydrogen as a mechanism for the post-
irradiation buildup of interface states in SiO
2
-on-Si structures, J. Appl. Phys., 58, 2524, 1985.
172. Brown, D.B., The time dependence of interface state production, IEEE Trans. Nucl. Sci., 32, 3900,
1985.
173. Griscom, D.L., Brown, D.B., and Saks, N.S., Nature of radiation-induced point defects in amorph-
ous SiO
2
and their role in SiO
2
-on-Si structures, in The Physics and Chemistry of SiO
2
and the Si–SiO
2
Interface, Helms, C.R. and Deal, B.E., Eds., Plenum Press, New York, 1988, pp. 287–297.
174. Zhang, W.D., et al., On the interface states generated under different stress conditions,
Appl. Phys. Lett., 79, 3092, 2001.
175. Blöchl, P.E. and Stathis, J.H., Hydrogen electrochemistry and stress-induced leakage current in
silica, Phys. Rev. Lett., 83, 372, 1999.
176. Bunson, P.E. et al., Ab initio calculations of H
þ
energetics in SiO
2
: Implications for transport,
IEEE Trans. Nucl. Sci., 46, 1568, 1999.
177. Rashkeev, S.N. et al., Defect generation by hydrogen at the Si–SiO
2
interface, Phys. Rev. Lett., 87,
165506, 2001.
178. Afanase’ev, V.V., de Nijs, J.M.M., and Balk, P., SiO
2
hole traps with small cross section,
Appl. Phys. Lett., 66, 1738–1740, 1995.
179. Vanheusden, K. et al., Thermally activated electron capture by mobile protons in SiO
2
thin films,
Appl. Phys. Lett., 72, 28, 1998.
180. Gale, R. et al., Hydrogen migration under avalanche injection of electrons in Si MOS capacitors,
J. Appl. Phys., 54, 6938, 1983.
181. Chen, P.J. and Wallace, R.M., Examination of deuterium transport through device structures,
Appl. Phys. Lett., 73, 3441, 1998.
182. Rashkeev, S.N. et al., Proton-induced defect generation at the Si–SiO
2
interface, IEEE Trans.
Nucl. Sci., 48, 2086, 2001.
183. Tsetseris, L. et al., Common origin for enhanced low-dose-rate sensitivity and bias temperature
instability under negative bias, IEEE Trans. Nucl. Sci., 52, 2265, 2005.
184. Fischetti, M.V., Generation of positive charge in SiO
2
during avalanche and tunnel electron
injection, J. Appl. Phys., 57, 2860, 1985.
185. Do Thanh, L. and Balk, P., Elimination and generation of Si–SiO
2
interface traps by low
temperature hydrogen annealing, J. Electrochem. Soc., 135, 1797, 1988.
186. Stojadinovic, N. and Dimitrijev, S., Instabilities in MOS transistors, Microelectron. Reliab., 29, 371,
1989.
187. Buchanan, D.A. and DiMaria, D.J., Interface and bulk trap generation in MOS capacitors, J. Appl.
Phys., 67, 7439, 1990.
188. DiMaria, D.J., Temperature dependence of trap creation in SiO
2
, J. Appl. Phys., 68, 5234, 1990.
189. Stathis, J.H. and DiMaria, D.J., Identification of an interface defect generated by hot electrons in
SiO
2
, Appl. Phys. Lett., 61, 2887, 1992.
190. DiMaria, D.J., Cartier, E., and Arnold, D., Impact ionization, trap creation, degradation, and
breakdown in SiO
2
films on Si, J. Appl. Phys., 73, 3367, 1993.
191. Al-Kofahi, I.S. et al., Continuing degradation of the SiO
2
=Si interface after hot hole stress, J. Appl.
Phys., 81, 2686, 1997.
192. Degraeve, R., Kaczer, B., and Groeseneken, G., Degradation and breakdown in thin oxide layers:
Mechanisms, models, and reliability prediction, Microelectron. Reliab., 39, 1445, 1999.
193. Winokur, P.S. et al., Total-dose radiation and annealing studies: Implications for hardness
assurance testing, IEEE Trans. Nucl. Sci., 33, 1343, 1986.
194. Fleetwood, D.M., Winokur, P.S., and Schwank, J.R., Using laboratory x-ray and cobalt-60
irradiations to predict CMOS device response in strategic and space environments,
IEEE Trans. Nucl. Sci., 35, 1497, 1988.
195. Sexton, F.W. and Schwank, J.R., Correlation of radiation effects in transistors and integrated
circuits, IEEE Trans. Nucl. Sci., 32, 3975, 1985.
196. McLean, F.B. and Boesch, H.E. Jr., Time-dependent degradation of MOSFET channel mobility
following pulsed irradiation, IEEE Trans. Nucl. Sci., 36, 1772, 1989.
ß 2008 by Taylor & Francis Group, LLC.
197. Fleetwood, D.M. and Eisen, H.A., Total-dose radiation hardness assurance, IEEE Trans.
Nucl. Sci., 50, 552, 2003.
198. Karmarkar, A.P. et al., Aging and baking effects on the radiation hardness of MOS capacitors,
IEEE Trans. Nucl. Sci., 48, 2158, 2001.
199. Fleetwood, D.M. et al., The response of MOS devices to dose-enhanced low-energy MOS
radiation, IEEE Trans. Nucl. Sci., 33, 1245, 1986.
200. Rodgers, M.P. et al., The effects of aging on MOS irradiation and annealing response, IEEE Trans.
Nucl. Sci., 52, 2642, 2005.
201. Batyrev, I.G. et al., Effects of water on the aging and radiation response of MOS devices,
IEEE Trans. Nucl. Sci., 53, 3629, 2006.
202. Saks, N.S., Ancona, M.G., and Modolo, J.A., Radiation effects in MOS capacitors with very thin
oxides at 80 K, IEEE Trans. Nucl. Sci., 31, 1249, 1984.
203. Benedetto, J.M. et al., Hole removal in thin-gate oxide MOSFETs by tunneling, IEEE Trans.
Nucl. Sci., 32, 3916, 1985.
204. Saks, N.S., Ancona, M.G., and Modolo, J.A., Generation of interface states by ionizing radiation
in very thin oxides, IEEE Trans. Nucl. Sci., 33, 1185, 1986.
205. Dressendorfer, P.V., Radiation effects on MOS devices and circuits, in Ionizing Radiation Effects
in MOS Devices & Circuits, Ma, T.P. and Dressendorfer, P.V., Eds., Wiley, New York, 1989,
Chapter 5.
206. Fleetwood, D.M., Winokur, P.S., and Meisenheimer, T.L., Hardness assurance for low-dose
space applications, IEEE Trans. Nucl. Sci., 38, 1552, 1991.
207. Jeppson, K.O. and Svensson, C.M., Negative bias stress of MOS devices at high electric-fields
and degradation of MNOS devices, J. Appl. Phys., 48, 2004, 1977.
208. Blat, C.E., Nicollian, E.H., and Poindexter, E.H., Mechanism of negative-bias-temperature
instability, J. Appl. Phys., 69, 1712, 1991.
209. Ogawa, S. and Shiono, N., Generalized diffusion-reaction model for the low-field charge-
buildup instability at the Si=SiO
2
interface, Phys. Rev. B, 51, 4218, 1995.
210. Schroder, D.K. and Babcock, J.A., Negative bias temperature instability: Road to cross in deep
submicron semiconductor manufacturing, J. Appl. Phys., 94, 1, 2003.
211. Zhou, X.J. et al., Negative bias-temperature instabilities in MOS devices with SiO
2
and
SiO
x
N
y
=HfO
2
gate dielectrics, Appl. Phys. Lett., 84, 4394, 2004.
212. Tsetseris, L. et al., Physical mechanisms of negative-bias temperature instability, Appl. Phys. Lett.,
86, 142103, 2005.
213. Ceschia, M. et al., Radiation induced leakage current and stress induced leakage current in ultra-
thin gate oxides, IEEE Trans. Nucl. Sci., 45, 2375, 1998.
214. Sexton, F.W. et al., Precursor ion damage and angular dependence of single event gate rupture in
thin oxides, IEEE Trans. Nucl. Sci., 45, 2509, 1998.
215. Sexton, F.W. et al., Single event gate rupture in thin gate oxides, IEEE Trans. Nucl. Sci., 44, 2345,
1997.
216. DiMaria, D.J. and Cartier, E., Mechanism for stress-induced leakage currents in thin SiO
2
films,
J. Appl. Phys., 78, 3883, 1995.
217. Depas, M., Nigam, T., and Heyns, M.M., Soft breakdown of ultra-thin gate oxide layers,
IEEE Trans. Electron Dev., 43, 1499, 1996.
218. Ricco, B., Gozzi, G., and Lanzoni, M., Modeling and simulation of stress-induced leakage current
in ultrathin SiO
2
films, IEEE Trans. Nucl. Sci., 45, 1554, 1998.
219. Ralls, K.S. et al., Discrete resistance switching in submicrometer Si inversion layers: Individual
interface traps and low-frequency (1=f?) noise, Phys. Rev. Lett., 52, 228, 1984.
220. Kirton, M.J. and Uren, M.J., Noise in solid-state microstructures—a new perspective on
individual defects, interface states and low-frequency (1=f) noise, Adv. Phys., 38, 367, 1989.
221. Schuegraf, K.F. and Hu, C.M., Hole injection SiO
2
breakdown model for very-low voltage
lifetime extrapolation, IEEE Trans. Electron Dev., 41, 761, 1994.
222. Degraeve, R. et al., New insights in the relation between electron trap generation and the
statistical properties of oxide breakdown, IEEE Trans. Electron Dev., 45, 904, 1998.
223. Stathis, J., Percolation models for gate oxide breakdown, J. Appl. Phys., 86, 5757, 1999.
ß 2008 by Taylor & Francis Group, LLC.
224. Buchanan, D.A., Scaling the gate dielectric: Materials, integration, and reliability, IBM J. Res.
Dev., 43, 245, 1999.
225. Wilk, G.D., Wallace, R.M., and Anthony, J.M., High-K gate dielectrics: Current status and
materials properties considerations, J. Appl. Phys., 89, 5243, 2001.
226. Felix, J.A. et al., Charge trapping and annealing in high-k gate dielectrics, IEEE Trans. Nucl. Sci.,
51, 3143, 2004.
227. Felix, J.A. et al., Effects of radiation and charge trapping on the reliability of high-k gate
dielectrics, Microelectron. Reliab., 44, 563, 2004.
228. Shaneyfelt, M.R. et al., Challenges in hardening technologies using shallow-trench isolation,
IEEE Trans. Nucl. Sci., 45, 2584, 1998.
229. Lacoe, R.C. et al., Application of hardness-by-design methodology to radiation-tolerant ASIC
technologies, IEEE Trans. Nucl. Sci., 47, 2334, 2000.
230. Schwank, J.R. et al., Radiation effects in SOI technologies, IEEE Trans. Nucl. Sci., 50, 522, 2003.
231. Jun, B. et al., Total dose effects on double gate fully depleted SOI MOSFETs, IEEE Trans.
Nucl. Sci., 51, 3767, 2004.
232. Nowlin, N. et al., A total-dose hardening-by-design approach for high-speed mixed-signal
CMOS integrated circuits, in Radiation Effects and Soft Errors in Integrated Circuits and Electronic
Devices, Schrimpf, R.D. and Fleetwood, D.M. Eds., World Scientific, Singapore, 2004, p. 83.
ß 2008 by Taylor & Francis Group, LLC.
ß 2008 by Taylor & Francis Group, LLC.
8
From 3D Imaging of Atoms to Macroscopic
Device Properties
S.J. Pennycook, M.F. Chisholm, K. van Benthem, A.G. Marinopoulos,
and Sokrates T. Pantelides
CONTENTS
8.1 Introduction........................................................................................................................ 259
8.2 Scanning Transmission Electron Microscopy................................................................ 261
8.3 Theoretical Microscope ..................................................................................................... 267
8.4 Structure and Properties of Si=SiO
2
=HfO
2
Gate Stacks
by Aberration-Corrected STEM and Theory ................................................................. 271
8.5 Future Directions ............................................................................................................... 276
Acknowledgments ..................................................................................................................... 277
References.................................................................................................................................... 278
8.1 Introduction
The quest for smaller, faster semiconductor devices is rapidly approaching the so-called
‘‘end of the roadmap’’ for silicon-based devices, in which the thickness of the gate oxide is
becoming too small to maintain the macroscopic properties of that mainstay of the semi-
conductor industry, SiO
2
. As the thickness approaches the size of the intrinsic ring struc-
ture of SiO
2
, the electronic integrity of the gate oxide becomes compromised, and
undesirable characteristics such as leakage and dielectric breakdown cannot be managed
effectively. This situation is the driver for the search for alternative, high dielectric constant
(high-k) gate structures. The introduction of new materials to augment a thin SiO
2
inter-
layer raises the possibility of new kinds of impurities entering either SiO
2
or the Si
substrate, and that dopant atoms may enter new configurations, adversely affecting
macroscopic device properties. It therefore becomes very valuable to be able to characterize
these nanometer scale device structures with probes that are sensitive to individual atoms,
ideally in three dimensions. At the same time, the ability to determine local electronic
structure with similar resolution and sensitivity would reveal any local defect states
introduced into the band gap as a result of these impurity or dopant atoms. In conjunction
with theoretical calculations of electronic structure, it would then become possible to
link the atomic-scale characterization of actual device structures to their macroscopic
ß 2008 by Taylor & Francis Group, LLC.
characteristics, and thus to find the true atomic origins of, for example, high leakage
current or low mobility.
In this chapter, we describe the latest advances in aberration-corrected scanning trans-
mission electron microscopy (STEM), which have in the last few years brought this vision
to a practical possibility. The ability to correct for the primary aberrations of electron lenses
represents a revolution in the field of electron microscopy. The rate of instrumental
advance today is faster than at any time since the invention of the electron microscope in
the 1930s, as shown in Figure 8.1 [1]. Aberrations occur inevitably in round lenses, as a
result of physics, and not only from imperfections, as was recognized by Scherzer over 70
years ago [2]. The spherical aberration of the simple round lens has limited electron
microscope resolution for most of its history, as was appreciated by Feynman in his 1959
lecture, [3] ‘‘There’s Plenty of Room at the Bottom,’’ where he explicitly called for a
100-fold improvement in resolution by overcoming spherical aberration, ‘‘why must the
field be symmetric?’’ Designs for aberration correctors have existed for over 60 years [4],
but they require multipole lenses, which makes such schemes enormously complex.
It was not until the era of the fast computer and the efficient detection of the charge-
coupled device that it has become practically possible to measure aberrations with the
necessary sensitivity, $1 part in 10
7
, and to control all the multipoles to the required
accuracy. Focusing, or tuning, the aberration corrector involves optimization in a space
of 40 or more dimensions, which has proved to be beyond human capability. Therefore, it
is not so surprising that the successful correction of aberrations in electron microscopy has
had to wait for the computer age.
Now, aberration correction has been successfully achieved both in the conventional
transmissionelectronmicroscope (TEM) [5] andinthe STEM[6]. Inthe STEM, direct imaging
of a crys tal lattice has been achieved at sub angstrom resol ution; see Figu re 8.2 [7].
But aberration correction brings more than just a factor of 2 or 3 in resolution. As shown,
in the STEM it also brings a greatly improved signal-to-noise ratio, resulting in the ability
to image individual Hf atoms inside the nanometer-wide gate oxide of an advanced
dielectric device structure [8]. Furthermore, since the lens aperture is much wider after
FIGURE 8.1
Evolution of resolution; squares represent light
microscopy, circles TEM, triangles STEM, solid sym-
bols before aberration correction, open symbols after
correction. (After Rose, H., Ultramicroscopy, 56, 11,
1994. With permission.)
0.1
1
10
100
1000
10
4
10
5
1800 1840 1880 1920 1960 2000 2040
Year
R
e
s
o
l
u
t
i
o
n

(
Å
)
ß 2008 by Taylor & Francis Group, LLC.
aber ration cor rection, the dep th of fi eld is muc h smaller , and it become s pos sible to locate
atoms in three dimension s to a precision bett er than 1 nm in depth and 0.1 nm in lateral
posit ion [9,10]. Three-di mensional mapping of heavy ato ms inside SiO
2
has become pos-
sible, and in princip le, three -dimensi onal spectros copy sh ould be feasible. With thi s level of
sensi tivity, new insights are availabl e into the microscop ic defects in device struc tures and
their infl uence on device charac teristics . Fro m such knowl edge, prop erties can be calcu-
lated from first princip les, ena bling the determinat ion of the atomic-s cale defects resp on-
sible for macros copic device prope rties such as leakage and mobilit y, lea ding to improved
device design .
8.2 S canning T ransmission Electron Microscopy
Figure 8 .3 sh ows a sch ematic diag ram of a STEM. As in any scanning micr oscope, the
image is built up sequent ially, pixe l by pixel, as the probe is sca nned acro ss a spe cimen,
and a variet y of signals can be detecte d to form an image . The two mo st impo rtant imagi ng
signals are the transmitt ed electron signal , which gives a brigh t- field image , and the
scatte red elect ron signal , which give s a dar k- field signal . Th ese two image s are comple -
mentary, giving quite different inf ormatio n about the sample . Th e bright- field image can
be arrang ed to give a high-re solution phase con trast image just as see n in a conve ntional
TEM [11]. A pha se contrast image is a coh erent image fo rmed by int erferenc e between
diffracte d beams, and fringe s in the image can be black or white dep ending on whether the
inter ference is constru ctive or destructi ve, which dep ends sensitiv ely on par ameter s such
as specimen thickness and microscope focus. Consequently there can be significant ambi-
guity on where the atoms actually are, and direct interpretation of atomic structure from
the image is generally not feasible. Instead, simulations of trial structures are usually
performed.
0.78 Å
FIGURE 8.2
Direct imaging of a Si crystal in the <112> projection showing resolution of pairs of Si columns just 0.78 Å apart,
obtained with a VG Microscopes HB603U dedicated STEMequipped with a Nion aberration corrector. (Adapted
from Nellist, P.D. et al., Science, 305, 1741, 2004. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
The annular (or high-angle annular) detector gives an image of quite different character,
which is essentially an atomic-resolution map of the specimen’s scattering power [12,13].
The intensity of electrons scattered by a specimen is approximately proportional to its
mean square atomic number (Z), and also its thickness, at least up to a few tens of
nanometers, by which time most of the beam has been scattered and the image saturates.
The annular dark-field (ADF) image is therefore an incoherent image, which has a directly
interpretable relationship between object and image, just like the image formed with an
optical camera or projector. Because of these characteristics, the image is often referred to
as a Z-contrast image. High-Z atoms can be directly distinguished in the image, and
unexpected configurations can be immediately apparent, properties that have been very
useful in identifying many unexpectedly complex defect arrangements at semiconductor
interfaces [14–20]. Until recently, however, Z-contrast images have been substantially
noisier than the phase contrast images obtained in conventional TEM. This is because
only a small current can be focused into a small spot, which is then scanned sequentially
over the field of view. Conventional TEM images are obtained with a broad, high-current
beam, with all image points being imaged in parallel.
In recent years major improvements in STEMinstrumentation have occurred, including
the successful incorporation of high-resolution STEM capabilities into commercial
GaAs
EELS
spectrometer
Annular
detector
1.4 Å
As Ga
Scanned probe
I ∝ Z
2

Z = 31 Z = 33
Retractable
bright-field
detector
FIGURE 8.3
Schematic of a STEM. A finely focused probe is scanned over a specimen and an image is built up point-by-point.
Simultaneous bright-field and dark-field images are possible or the bright-field detector may be replaced with an
electron spectrometer to provide atomically resolved spectroscopic imaging. (Adapted fromPennycook, S.J. et al.,
Philos. Trans. R. Soc. London, Ser. A, 354, 2619, 1996. With permission).
ß 2008 by Taylor & Francis Group, LLC.
microscope columns with efficient detectors [21]. In particular, the development of aber-
ration correctors has dramatically improved STEM performance [22,23]. For around 60
years, spherical aberration has been the major limitation to microscope resolution. As
shown in Figure 8.4, spherical aberration is the tendency for high-angle rays passing
through the lens to be overfocused, meaning that beyond a certain maximum aperture
they are not usefully contributing to the focused spot but produce an extended tail in the
image. Thus, spherical aberration results in a maximum effective aperture. Microscope
resolution is proportional to l=u, where l is electron wavelength and u is the semiangle of
the objective lens. By correcting for the lowest-order aberrations of the objective lens, the
aperture can be opened up by a significant factor of between 2 and 3, which translates
directly into improved resolution. However, not only has reducing the probe size
enhanced the resolution, but also, the same current can now be focused into a smaller
probe, giving a vastly higher peak intensity in the narrower beam. As a result, the signal-
to-noise ratio in aberration-corrected images is greatly improved compared to uncorrected
images, and the sensitivity to individual high-Z atoms is enormously increased [24–36].
Another key advantage of the STEM is that multiple detectors can be used to obtain
complementary images with pixel-to-pixel correlation. Simultaneously acquired bright-
field and dark-field images can be directly compared, although the optimum focus for
each is slightly different so that a focal series of images may be necessary for quantitative
work [37]. Figure 8.5 shows simultaneously acquired, aberration-corrected phase- and
(a) (b)
FIGURE 8.4
Schematic showing the action of spherical aberra-
tion in overfocusing high-angle rays. (Adapted
from Varela, M. et al., Annu. Rev. Mater. Res., 539,
2005, With permission.)
1 nm 1 nm
(a) (b)
FIGURE 8.5
Simultaneously acquired (a) Z-contrast and (b) phase contrast images of a poly-Si=HfO
2
=SiO
2
=Si gate dielectric
structure obtained with the aberration-corrected HB603U. (Adapted fromRashkeev, S.N. et al., Microelectron. Eng.,
80, 416, 2005; van Benthem, K., Rashkeev, S.N., and Pennycook, S.J., in Characterization and Metrology for ULSI
Technology, American Institute of Physics, Richardson, TX, 2005, 79. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
Z -contrast images of a poly-Si =HfO
2
=SiO
2
=Si gate dielectric structur e in which the different
charac teristi cs a re ver y clear [38,39] . In the pha se con trast image , the SiO
2
shows the
spe ckle pat tern typical of an am orphous materi al, cause d by rando m inter ference effect s
withi n the struc ture, and the crystal line regions sh ow latt ice fri nges. The Z-contr ast image
is dominat ed by the high sca ttering pow er of the HfO
2
( Z ¼ 72 for Hf com pared to 14 for
Si and 8 for O). In the thi n gate oxide laye r there are sm all brigh t spots visi ble that
repres ent single stray Hf atoms [8]. These are no t dete ctable in the phase contrast image.
The exact posit ion of the Si =Si O
2
inter face is diffi cult to dete rmine fro m the pha se
contras t image without extensive simu lations, and it has therefo re bee n propo sed that
micr oscopic measu remen ts of dielectric layer thick ness be dete rmined from intensity trace s
acros s the Z -contr ast images [40]. The incoher ent nature of the Z-co ntrast image makes it
rea sonably immun e to the influence of spe cimen thick ness and inter face rough ness. An
inc oherent image is given by a mathemat ical convo lution of a specimen obje ct functi on
with a resolutio n fun ction of the optical instrum ent. In the cas e of ADF ST EM, the
resol ution function is the int ensity pro fi le of the scanning probe , which is conv oluted
with the spe cimen scatte ring pow er. A use ful con cept is that of an effective probe , which
is the probe pro fi le integr ated through the thick ness of the specimen. In a crystal aligned to
a major zon e axis, the probe can chann el along the colu mns. At the sam e time some
inten sity is sca ttere d to high angles leadin g to probe broad ening. In the amorpho us
oxide there is no chann elin g, but beam broad ening remain s. Becau se of the change in
beam chann eling , the effective probe wi ll change somew hat as it is scanned acros s
the Si =SiO
2
inter face. Ho wever, the se effect s are generall y sm all and normally ignore d.
The metho d proposed for thickne ss determi natio n assu mes the same eff ective probe in the
crys tal and in the oxide. For any rad ially symmet ric probe profi le, a convolu tion with a
step fun ction (for example , the abrupt change in inten sity from Si to Si O
2
) will lead to a
blurring of the image intensity, with the point of inflection locating the interface. Hence,
differentiating the intensity trace will regenerate the formof the effective probe. If now the
interface has structural roughness, this will further broaden the intensity profile. However,
the point of inflection will remain as the mean interface position, which is why the method
is relatively immune to details of the specimen and microscope parameters.
The abov e proce dure is illustrat ed in Figure 8. 6. Differe ntiating the ADF int ensity
profile gives a peak at the interface position. Note, however, that the derivative trace is
somewhat asymmetric. This illustrates the different channeling conditions on either
side of the interface, and is indicative of the potential errors in the method. Also, the
width of the derivative peak, $5 Å, is much broader than the width of the probe, which is
definitely of the order of 1.3 Å, since it is resolving the Si dumbbell separation at 1.26 Å.
This is probably mostly due to interfacial roughness; however, even abrupt interfaces
typically show a transition in intensity that is broader than expected from a simple
convolution model. Such behavior is not predicted even by sophisticated image simula-
tions. The problem is an unexplained background intensity in the images, which is
higher for higher-Z materials, and the background intensity does not change abruptly
across the interface. This issue is often referred to as the Stobbs factor, and is yet to be
resolved [41].
Perhaps the most important simultaneous signals in STEM are obtained by replacing
the bright-field detector with an efficient, parallel detection, electron spectrometer. Then,
it becomes possible to use the ADF image to locate atomic planes or columns and to
perform spectroscopy from specific columns selected from the image. The first atomically
resolved spectroscopic analysis was demonstrated in 1993, in which a characteristic
Co edge was measured plane-by-plane across a CoSi
2
=Si interface [42]. The drop in
measured Co concentration in moving from the last plane of the silicide to the first plane
of the Si well exceeded the 50% points required to demonstrate atomic resolution. Later,
ß 2008 by Taylor & Francis Group, LLC.
column-by -column spe ctrosco py becam e com mon [43– 48], but rec ently, wi th the incorp-
orati on of aberratio n corre ctors int o STE M columns , the impro vement s have become even
more spectacula r. First was the spe ctrosco pic ide nti fi cation of an indivi dual atom insid e a
bulk solid as sh own in Figu re 8.7, [49] and second, the recent achi evement of true,
two-di mensional atomi c-resolu tion maps [50], as opp osed to point spe ctra obtain ed from
columns cho sen from the Z -contrast image , bot h achieveme nts due to the higher signal
levels availab le afte r aber ration corre ction.
Not only does elect ron energy loss spectrosc opy (EELS) provid e com positional inf orma-
tion from the energie s of charac teristi c ab sorption edge s, but also the fi ne structur e on
those edge s gives informati on on local elect ronic struc ture in a similar mann er to x-ray
absorpti on spectros copy (XAS). Figure 8.8 sh ows spectra charac teristic of bulk Si and SiO
2
,
and of the (rough) int erface betwe en the m which show s additi onal peaks due to subo xide
bonding [51]. STEM and EELS have develop ed int o a powerf ul tool for the atomi c and
electro nic charac terizat ion of inter faces in devi ce st ructures [52 –59]. Comp ared with XAS,
EELS has the bene fit of spatial reso lution at the atomi c le vel, and c an probe changes that
occur as the probe is moved acros s an inter face, whe reas XAS provi des an aver age
measure ment. The probl em, as always, is that EELS has poorer count statist ics and so
quan titative analysis is always limit ed by signal-to-n oise.
Where as the mo tivation fo r aber ration corre ction was primar ily the resol ution, implicitl y
the lateral resol ution, a some what unanticipat ed adv antage has been the gain in depth
resolution, which is given by 2l=u
2
. So a factor of 2 or 3 improvement in the transverse
resolution brings also a factor of 4–9 in depth resolution. We have moved from a situation
where the depth of field of the microscope is typically greater than the specimen thickness
(the image is a projection of the specimen) to a situation where the depth of field is usually
Si SiO
2
1 nm
FIGURE 8.6
Z-contrast image of an Si–SiO
2
interface obtained with the uncorrected HB603USTEM. Its probe size of about 0.13
nm is still sufficiently small to resolve the Si dumbbells. A vertically averaged line trace (grey) is fitted with a
smooth curve (dashed line), differentiated (dotted line), and the width can then be compared to the theoretically
calculated probe profile (solid line). Reasons for the increased interface width can be interfacial roughness, and
also probe broadening effects. (Reproduced from Diebold, A.C. et al., Microsc. Microanal., 9, 493, 2003. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
1
2
3
4
5
6
(a)
0.5 nm
(b)
1
2
3
4
5
6
820 850 880
Energy loss (eV)
E
E
L
S

i
n
t
e
n
s
i
t
y
FIGURE 8.7
Spectroscopic identification of a single La atomin a column of Ca in the perovskite CaTiO
3
. (a) Z-contrast image of
CaTiO
3
showing traces of the CaO and TiO
2
{100} planes as solid and dashed lines, respectively. A single La
dopant atomin column 3 causes this column to be slightly brighter than other Ca columns, and EELS shows a clear
La M
4,5
signal (b). Moving the probe to adjacent columns gives reduced or undetectable signals. Results obtained
using a VG Microscopes HB501UX STEM equipped with a Nion aberration corrector. (Adapted from Varela, M.
et al., Phys. Rev. Lett., 92, 095502, 2004. With permission.)
Si
110
1
2
3
4
SiO
2
100
0
I
n
t
e
n
s
i
t
y
Interface
FIGURE 8.8
Si L EELS edges obtained from an Si=SiO
2
interface compared to spectra from Si and SiO
2
far from the interface.
Due to the high band gap in SiO
2
the edge onset is shifted higher in energy. The spectrumfromthe interface shows
the presence of states in the band gap of SiO
2
indicating suboxide bonds. (Reproduced fromPennycook, S.J. et al.,
Encyclopedia of Materials: Science and Technology, Elsevier Science Ltd., Kidlington, Oxford, 2001 p. 1. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
smaller than the specimen thickness, as illustrated in Figure 8.9. This means that we are
able to optically section the specimen, rather as in a confocal optical microscope.
A through-focal series of images becomes a depth sequence of images, which can be
reconstructed to obtain useful information on the three-dimensional structure of the object
[9,10]. The sectioning is opposed by the tendency of a crystal to channel the electrons along
atomic columns [60], and so the method works best in amorphous materials or, if aligned
crystals are used, the columns should be weakly channeling, i.e., low Z.
For more detailed discussions on STEM and the benefits and applications of aberration
correction the reader is referred to several recent reviews [32,34,37,61–63].
8.3 Theoretical Microscope
Density-functional total-energy calculations provide valuable complementary insight
into atomic-scale structure and the only quantitative link to macroscopic properties.
Theory can validate impurity or defect configurations seen in the microscope as stable,
low-energy structures, and provide formation and migration energies, predict local elec-
tronic structure for comparison with EELS data, and, as we show later, even predict
transport properties. This joint approach to understand structure and property relation-
ships has proved fruitful in many semiconductor systems. In some cases theoretical
(a)
1 Å
(b)
0 0
m
n
5
m
n
5
(c)
Ϫ5 Ϫ5
FIGURE 8.9
Probe intensity distributions for three generations of STEM, (a) an uncorrected VG Microscopes HB501UX 100 kV
STEMwith depth of field $45 nm, (b) the aberration-corrected HB603U300 kV STEMwith depth of field $3.7 nm,
and (c) for a next generation fifth-order aberration-corrected 200 kV microscope with predicted depth of field $1
nm. Intensities are shown on a logarithmic scale and curves have been shifted vertically to locate maxima at zero
defocus. (Reproduced fromBorisevich, A.Y., Lupini, A.R., and Pennycook, S.J., Proc. Nat. Acad. Sci. USA, 103, 3044,
2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
predictions stimulate experimental observations, while in other cases experiment comes
first and theory provides the rationalization.
As an example of the additional insights available from complementary theory and
experiment, we present the case of As-doped grain boundaries in Si. This began as a
theoretical study motivated by the observation that experimental segregation energies
for As in Si grain boundaries are around 0.5 eV; whereas, available theory predicted
segregation energies of around 0.1 eV [64]. The reason for this discrepancy was soon
found to be the assumption that impurities would be present in the boundary as isolated
atoms. It was well known that grain boundaries in tetrahedrally coordinated semiconduct-
ors were fully rebonded, i.e., contained no dangling bonds. Priory theory assumed that As
is incorporated in the grain boundary just like in the bulk, i.e., at isolated substitutional
sites. The new hypothesis was that As dopants form pairs, each atom obtaining its
preferred threefold configuration, as shown in Figure 8.10 [65]. The segregation energy
per As atom increased threefold for certain sites. Z-contrast images of As in a Si grain
boundary (see below) confirmed the formation of As pairs in certain sites, as predicted. It
is the environment of the grain boundary that allows the As–As distance to increase,
Si
As
Si
As
(b) (a)
(c)
FIGURE 8.10
(a) Arsenic (As) prefers threefold coordination, so isolated As dopants create Si dangling bonds. (b) As dimers,
however, can lower their energy by relaxing away from each other leaving no dangling bonds. This can only be
accomplished in the grain boundary environment. (c) Plot showing the reduced charge density between the As
dimers compared to between Si atoms. (Adapted from Maiti, A. et al., Phys. Rev. Lett., 77, 1306, 1996. With
permission.)
ß 2008 by Taylor & Francis Group, LLC.
allowi ng the repulsi ve energy of nea r neighbo r As atoms to be avoide d, an exampl e of
coope rative chemica l rebon ding. In the bulk cry stal the re is no such elastic relaxa tion. In
fact, the energy of a dime r is slightl y high er than that of two isolated atoms . Figu re 8.10c
shows the calcul ated charge densi ty betwe en the two atoms of an As dimer, show ing it to
be signi ficantl y redu ced com pared to that in the Si –Si bond.
The theo ry was perform ed on the lowe st peri od grain boundar y to min imize computa-
tiona l cost, a symmet ric S ¼ 5{310} <001 > tilt grain bound ary, which has a miso rientati on
of 368 . Later , expe rimen tal obse rvations were mad e on a lower- angl e grain boundar y [66],
a symm etric 23 8 S¼ 13{510} <001 > tilt grain boundar y. The structur e of thi s bound ary
was foun d to be surprisingl y com plex, a s shown in Figu re 8.11, com prisin g a conti guous
array of six dis location cor es, a perfect edg e dislocat ion (1), and two perfect mix ed type
dislocat ions (1,2) a rranged as a dipole (i.e., the ir dislocat ion con tent cancels and they could
be replace d by perfe ct crys tal withou t changing the miso rientati on of the grain boundar y).
The same sequ ence is then mir rored acros s the boundar y plane , denoted by primed
numb ers. The presen ce of As is reveale d as a higher int ensity in on e of the two m ixed
type dislocat ion cor es, repe ated peri odically along the bou ndary structur e. Clearly, only
two of the six cores forming the bound ary structure contai n As. The extra inten sity in the
image was ab out 18%, indi cating an As concentrat ion of about 5% per colu mn, or about
two As atoms in the spe cimen thi ckness. Th is image was obt ained before ab erration
corre ction, and so single ato m sensi tivity was not pos sible fo r atoms of such low Z
(33 for As compared to 14 for Si).
Theory was then performed on this more complex 238 boundary structure, but as with
the 368 boundary, segregation energies for isolated As atoms never exceeded 0.1 eV.
Although calculations for truly isolated As dimers could not be carried out because the
required supercell would be too large, calculations for As chains (a continuous line of As
3
3
2
2Ј 2Ј
1 1


3Ј 3Ј
0.2 nm
FIGURE 8.11
Z-contrast image of a 238 S¼13{510} <001> tilt grain boundary in Si, with schematic diagram showing the
boundary to comprise three dislocation cores which are then mirrored across the boundary plane. As is seen to
segregate only to one of the three dislocation cores, seen as bright triangles in the image. Results obtained with the
uncorrected HB603U. (Reproduced from Chisholm, M.F. et al., Phys. Rev. Lett., 81, 132, 1998. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
dimers) showed that only in core 2 was there significant segregation energy, in full accord
with experiment. The theory showed further the reason for this. In cores 1 and 3, the
relaxation of a dimer caused rotation toward the tilt axis, pushing the As atoms toward the
neighboring dimer. In core 2, the relaxation rotated the dimer away from the tilt axis,
allowing the As–As separation to increase, thereby lowering the energy. Thus the theory
provides a perfect rationalization of the experimental observations, showing how it is the
fine details of the boundary geometry that controls segregation, and specific sites can be
favored to a significant degree.
Another example where theory rationalized microscopic observation is in the generation
of vacancies by the fast electron beam. The observations were made on a lower-angle 168
S¼25{710} <001> tilt grain boundary where the dislocations formed isolated, but still
complex arrangements, as shown in Figure 8.12 [67]. Again, cores D1 and D2 represent a
dipole, their dislocation content canceling, while cores E1 and E2 are mixed dislocations.
With electron irradiation the intensity of specific columns shaded in the schematic
decreased, indicating atomic displacements, the onset of amorphization. It was clear that
the damage initiated in the vicinity of core E1. To explain these observations vacancy and
interstitial formation energies were computed for various configurations. In the case of
vacancies, all configurations in the boundary had much lower formation energies than in
the bulk, as expected, but in particular, all the low-energy defect structures indeed
occurred in the vicinity of the region seen dark in the micrograph. On the contrary, with
interstitials, no correlation was found to the dark region of the image. It is clear therefore
that the irradiation induces stable vacancy complexes; whereas, the ejected interstitials
diffuse rapidly away.
D1
D2
E1
E2
D1
D2
E1
E2
(a)
0.19 nm
(b)
FIGURE 8.12
Z-contrast image of a 168 S¼25{710} <001> tilt grain boundary in Si and derived structures with increasing
electron irradiation. (a) With little electron irradiation only the three columns shaded show reduced intensity due
to damage. (b) With increasing irradiation the damage spreads. Results obtained with the uncorrected HB603U.
(Adapted from Maiti, A. et al., Appl. Phys. Lett., 75, 2380, 1999. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
8.4 S tructure and Properties of S i= SiO
2
= HfO
2
Gat e Sta cks
by Aberration-Corrected STEM and Theory
One of the mo re promis ing ro utes to overcom e the potenti al ‘‘end of the ro admap ’’ for
shrink ing Si-based devices is to m ove to alternati ve gate dielectric s wi th higher diele ctric
cons tant, the so-called high- k dielectrics , of which HfO
2
is perha ps the most widel y st udied
materi al (Chapt ers 9 throu gh 14). The high- k gives a sm aller effective dielectri c thick ness,
but introd uces new concerns . Key iss ues inc lude wheth er the Si =SiO
2
inter face remains
well-p assivated or wheth er Hf atoms wo uld segre gate to the int erface and degrade its
charac teristics throu gh introduct ion of localize d states, eithe r directl y or throu gh introduc-
tion of dangling bonds. Locali zed state s could act a s sca ttering cen ters for elect rons movin g
in the channel, reduci ng mo bility, or, at a suf fic iently high dens ity, might enable leakage
throug h the thin SiO
2
laye r. To investi gate these eff ects, aberrati on-correcte d STEM com-
bined with the ory is emergi ng as a pow erful combin ation , provid ing the a bility to locate
single Hf ato ms in three dimens ions wi thin the gate ox ide, and throu gh theory, to predic t
macros copic prope rties.
One of the most importan t advantage s of the shor t depth of fo cus in the aberrati on-
corre cted micro scope is with the probe focused insi de the specime n; atoms on the top and
botto m surfa ces are out of focus, and hence invisibl e. Th erefore, a through focal series
ensure s that Hf atoms inside the actu al devi ce struc ture are indeed being exa mined, with
no poss ibility of the atoms being part of any surface layers produ ced duri ng spe cimen
preparat ion. For the pres ent study, HfO
2
fi lms appro ximate ly 3 nm thick were dep osited
on silicon substrat es by atomi c laye r depositi on, and then rap id thermal ly anneal ed at
950 8 C for 30 s in N
2
. The fi lms were then cappe d with undop ed polycrys talline Si .
Sampl es were prep ared purely by mech anical pol ishing to avoid damagin g the thinne st
regions by ion milling. Part of a throug h focal series of image s is shown in Figure 8.13, in
which a bright spot is clearl y seen at a defoc us of zer o and À 5 Å, but fades at lower or
higher defocu s. A line trace across such an image is shown in Figure 8.14, revealin g that
the exc ess intensity of the Hf ato m is comparab le to that of an ent ire column of Si atom s,
which is in ag reement with image simu lations [8,10]. This rules out the possibility that the
spots could be due to a lighter elemen t such as Zr , a known poss ible impu rity in these
systems . It is also appare nt that the inten sity across the oxide is highly nonu niform, which
repres ents the out of focus con tributions of other Hf ato ms a nd par ticularly the HfO
2
at ot her
depths. Becau se of this out of focus backgro und, the atoms are on ly visib le whe n in sharp
focus. Figu re 8.15 shows plots of the excess intensity extracted from such line traces for a few
repres entative ato ms as a functi on of fo cus, or atom depth. Th e plot clearly shows how the
depth of the Hf ato ms can be determi ned to a precis ion of much better than Æ0.5 nm, jus t by
locating the position of the peak inten sity. For atoms that are widely separated , as in the
presen t exam ple, the depth preci sion the refore signi ficantly exc eeds the expecte d depth
resol ution. Th e specime n thick ness is assume d to be app roximate ly equal to the greatest
difference in dep th fo und betw een Hf atoms , in this case 6 nm.
A total of 65 atoms we re analyzed in thi s focal seri es which were randoml y distribut ed in
depth. Intere stingly, the ir lateral position was not found to be rando m. The lateral preci-
sion is an or der of magni tude grea ter than the dep th precision, at a level of 0.1 nm or high er
limited primar ily by specimen drift. Figu re 8.16a shows an image with positions of some
Hf atoms marked, along with a histogram of the lateral positions with respect to the
interface plane. No Hf atoms are present right at the interface, and they prefer to sit at
specific distances from the interface, around 2.5 and 4 Å.
Now it is obvious that a transition from crystalline Si to a truly disordered oxide cannot
occur instantaneously, and some order is therefore expected right near to the last Si plane.
ß 2008 by Taylor & Francis Group, LLC.
Δf = +20 Å Δf = +15 Å Δf = +10 Å
Δf = +5 Å Δf = 0 Å
Δf = −20 Å Δf = −15 Å Δf = −10 Å
Δf = −5 Å
5 Å
FIGURE 8.13
Sequence of frames from a through-focal series of Z-contrast images of an Si=SiO
2
=HfO
2
high-k device structure
showing an individual Hf atom coming in and out of focus (circled). Results obtained with the aberration-
corrected HB603U. (Adapted from van Benthem, K. et al., Appl. Phys. Lett., 87, 034104, 2005. With permission.)
(a) (b)
Si Si Si Si Si
Hf
HfO
2
1 nm
FIGURE 8.14
(a) One frame from the through-focal series showing a single atom with an intensity profile taken from the
rectangular box (b) showing a large nonuniformbackground contribution. (Adapted fromvan Benthem, K. et al.,
Ultramicroscopy, 106, 1062, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
None is a pparent in the Z -contr ast image, ho wever, and so we turn to theo ry. Several
studi es of the Si=SiO
2
interface have been carried out, and here we compar e the expe ri-
mental data to two st udies, the firs t to a study withou t Hf usi ng classical potential s [68]
with a large num ber of atoms , and the seco nd to dens ity functi onal calcul ations inc luding
Hf, using by necess ity a sm aller num ber of atoms [69]. The class ical potenti al calcul ations
give a fair reprod uction of the expe rimen tal Hf histog ram (see Figu re 8.1 6b and c),
there fore sugg esting that the Hf atoms subs titute for Si in the oxide . This was tested
explicitl y by density functi onal theo ry usi ng vari ous crystalline m odels for the interface
studi ed previousl y. In that work it was found energeti cally prefe rable to have abrupt
inter faces, avoidi ng subo xide bon ds [69]. The rati onalizatio n of this result was that Si –Si
bonds are stiff, and the st ructural mismat ch at the int erface can be better accommo dated
by inserti ng an oxyg en atom into any Si –Si bon d that may protr ude into the oxide .
A Si–O –Si bond has a soft angl e at the O atom that allows the bon d angle to vary, as
shown in Figu re 8.17. Two par ticular oxide structur es we re foun d to have the lowe st
energy, tridym ite-like bon ding and quartz-l ike bonding. Quart z-like bon ding results in
tension in the ox ide, in the plane of the inter face, whereas tridym ite-like bonding result s
in compr ession. It was the refore con cluded that the actual int erface struc ture wou ld mo st
likely con sist of an intima te pha se mixture of these two types of bonding to avo id any
long-r ange elastic strains. We therefore look ed at the two models to ext ract the distanc e
from the int erface plane to the first Si ato ms, and fo und 2.3 and 4. 4 Å fo r the quartz-l ike
struc ture, and longe r distanc es for the crys tobalilte- like struc ture, 2.9 and 5.4 Å. Inse rting
these distanc es onto a histog ram, we find that the result is also in rea sonable ag reement
with the data (Figure 8.16d). Furthermore, the theory directly confirmed that substitu-
tional Hf is preferred over interstitial Hf, by at least 4 eV [70]. Interstitial Hf results in
disruption of the oxide lattice, because the Hf atoms steal O atoms from the Si leading to
the creation of dangling bonds. Comparing the energy of substitutional Hf in different
0
200
400
600
800
1000
Ϫ10 0 10 20 30 40 50 60 70
Depth (Å)
E
x
c
e
s
s

i
n
t
e
n
s
i
t
y

(
c
o
u
n
t
s
)
FIGURE 8.15
Plot showing the excess intensity from several individual Hf atoms as a function of depth. (Adapted from van
Benthem, K. et al., Ultramicroscopy, 106, 1062, 2006. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
sites also expl ains why the Hf atoms keep away from the interface . Figure 8.18 shows a
plo t of the energy of fo rmation for substi tutiona l Hf, taking the zer o to be in bul k Si,
which shows how the Hf prefers the mo re ope n oxide struc ture, where it can have O
neighbo rs, and even how it prefe rs to stay away from the inter face plane by a sig nificant
energy, 1.4 eV. Also, since the source of the Hf ato ms is the HfO
2
, we would not
expe ct Hf to migr ate into the Si substrat e. Thus, the the ory comple tely expl ains all the
(b)
Experiment
Quartz
Tridymite
Relative distance (0.1 nm)
#

o
f

S
i

a
t
o
m
s
0
0
2
4
6
8
10
#

o
f

S
i

a
t
o
m
s
0
2
4
6
8
10
#

o
f

H
f

a
t
o
m
s
0
2
4
6
8
10
1 2 3 4 5 6 7 8
Tu and Tersoff (2000)
(c)
(d)
1 nm
(a)
FIGURE 8.16
(a) Image of the device structure with Hf atom positions superimposed. No Hf atoms are found at the Si=SiO
2
interface. (b) Histogram showing the distance of Hf atoms from the Si=SiO
2
interface compared to (c) Si positions
obtained by classical potential calculations (courtesy J. Tersoff) and (d) first-principles calculations of model
abrupt interfaces to crystalline SiO
2
. (Adapted from Marinopoulos, A.G. et al., Phys. Rev. B77, 195317, 2008. With
permission.)
FIGURE 8.17
Schematic diagram showing the additional degree
of freedom obtained with an Si–O–Si bond com-
pared to the stiff Si–Si bond. (Reproduced from
Buczko, R., Pennycook, S.J., and Pantelides, S.T.,
Phys. Rev. Lett., 84, 943, 2000. With permission.)
ß 2008 by Taylor & Francis Group, LLC.
experi mental obse rvations leadin g to a detai led microscop ic explanat ion of the st ructure
and energeti cs of the system.
Theory also gives insi ght into the electroni c prope rties of the mate rial. In Figu re 8.19 , we
plot the densi ties of states around two inter stitial con figura tions of Hf showi ng how the y
overlap the Si ba nd gap and are the refore electri cally active. Also shown is the wave
functi on of the 4d localize d sta te, along with the wave functi on of the bot tom of the Si
cond uction band, showi ng a signi fi cant lateral extent overlap ping the neighbo ring O and
Si atoms. Substituti onal Hf showe d similar beha vior. From the dat a shown in Figu re 8.16,
knowi ng the coordinat es of all the atoms it is simple to deduc e their density, 1.4 nm
À 3
(1.7 Â10
14
cm
À2
), and hence determine their average separation to be about 7 Å. It is clear
Si
Ϫ6
Ϫ4
Ϫ2
0
2
4
Ϫ4 Ϫ2 0
Distance from interface (Å)
Δ
E
s
u
b
s

t

(
e
V
)
2 4 6
SiO
2
FIGURE 8.18
Plot showing the formation energy for substitutional Hf atoms across the interface. Hf atoms at the interface have
1.4 eV higher energy than those in the oxide, explaining why none were found in that position. (Adapted from
Marinopoulos, A.G. et al., Phys. Rev. B77, 195317, 2008. With permission.)
O
O
Hf
Si/SiO
2
interface
(a)
Si band gap
Ϫ2 0
0.0
2.0
4.0
2
Energy (eV)
D
O
S

(
a
r
b
.

u
n
i
t
s
)
4 6
(b)
FIGURE 8.19
(a) Densities of states around two interstitial Hf atoms overlap the Si band gap. (b) Plot of