You are on page 1of 223

Electrical transport in Si:P and Ge:P δ -doped

systems

A Thesis
Submitted For the Degree of
Doctor of Philosophy
in the
Faculty of Science

by

Saquib Shamim

Department of Physics
Indian Institute of Science
Bangalore - 560 012
India

June 2015
Declaration

The work reported in this thesis is original and was carried out by me during my tenure
as a PhD student at the Department of Physics, Indian Institute of Science, Bangalore
under the supervision of Dr. Arindam Ghosh.
This thesis has not formed the basis for award of any degree, diploma, associateship,
membership or similar title of any university or institution.

(Saquib Shamim)
June, 2015

Department of Physics,
Indian Institute of Science,
Bangalore - 560012,
India
To

My Parents, brothers and sisters


Acknowledgements

The entire PhD journey would not have been possible without the constant help and
support of many people. Here I take this opportunity to acknowledge them.
The first person that deserves to be mentioned is my supervisor Prof. Arindam Ghosh,
without whom this PhD journey would not have even started. I owe my sincerest gratitude
to him for giving me the opportunity to work under his guidance. His dedication and
enthusiasm towards research has been inspiring me for the last few years. He has always
been ready to discuss results or any new ideas. I am thankful to him for all the fruitful
discussions we had, for actively participating in data analysis, helping me clear my doubts
however trivial it might have been and teaching me things as basic as soldering a wire.
A could not have wished to find a more friendly guide than Arindam da. Apart from
the scientific interactions, he has always participated with us in the table tennis matches,
cricket, football, badminton and even arm wrestling. We never lost an opportunity to
take treats from him for any of his achievements, scientific awards or good publication.
It was because of him that I was able to feel at home in the lab. Once again I thank him
for all his support.
I would like to thank our collaborator Prof. Michelle Simmons from the Centre of
Excellence for Quantum Computation and Communication Technology, University of New
South Wales, and her group for providing all the devices that have been measured in
this work. We had many scientific discussions which were extremely beneficial. I thank
Arindam and Michelle for arranging my visit to UNSW during my PhD tenure. Visiting
her group, seeing how the devices were fabricated and interactions with the researchers
there helped me understand the system much better. Several members of her group needs
to be mentioned here. Dr. Suddhasatta Mahapatra provided all 2D Si:P devices and
had always been ready for discussions. The questions he asked and his suggestions have
always been helpful. Dr. Giordano Scappucci provided the Ge:P devices exactly at the
right time when we struggling to interpret the Si data. Dr. Bent Weber who provided the
1D wires has been a very good friend. I have enjoyed our interactions at UNSW and IISc.
We started the measurements on the wires together and his input has been extremely
helpful. I also thank Dr. Daniel Thompson for the quantum dot devices. Apart from the
scientific collaboration, Bent and Dan have been an excellent companion for lunch and
dinner due to their liking for Indian food.

I also thank Prof. Vijay Shenoy, Prof. Subroto Mukerjee, Prof. Manish Jain, Prof.
Sanjoy Sarker, Prof. J. K. Jain, Prof. Subir Sachdev and Prof. Anindya Das for all
helpful discussions we had. While wiring the dilution refrigerator, the suggestions of Dr.
Johannes Pollanen and Prof. Aveek Bid were extremely beneficial. I had the fortune of
being taught by some of the best teachers during my coursework like Prof. Diptitman
Sen, Prof. Chandan Das Gupta, Prof. K. Rajan, Prof. Subroto Mukerjee, Prof. Vijay
Shenoy and Prof. Venkataraman.

Life in the lab have always been exciting due to the amazing labmates that I was
fortunate to have. When I joined the lab as a summer project student, the seniors in
the lab had always been supportive. Koushik, one of the first persons I interacted with,
guided me during my initial days of the summer project and taught me the basics of low
temperature systems. I thank Koushik and Mohan for helping with the labview codes
and Mohan for reminding us of our duties and responsibilities towards the environment.
Amrita didi and Chandni di gave me initial tips on how to do noise measurements. Even
after going for postdoc, Chandni has been very helpful in proving information about
various scholarships, research positions and conferences. Atin da and Dada taught the
basics of device fabrication like wafer cleaning, graphene exfoliation, e-beam lithography
and thermal deposition. I owe my thanks to Srijit da as I learnt how to do low temperature
transport experiments from him while working on the antidots. His understanding of the
subject and precise way of doing things is something which I look forward to.

Vidya, a very dedicated researcher who wants to everything perfectly has been an
extremely good friend. She is a wonderful person and being of the same batch, we did
many things together like mounting samples in dilution, the electrical wiring, etc. Working
on similar physics, we had innumerable scientific discussions and the questions which she
asked have helped me improve my own understanding of the subject. I interacted with
Dada, another perfectionist, a lot during my research and he is a person whom I can
always ask for help. Being nocturnal researchers helped us to get along very well. Kallol,
my friend for more than a decade, is a brilliant experimentalist and has the capability to
think outside the box. He is an asset to any research group and his ideas have significantly
helped all the lab members. Mitali di’s company has always been enjoyable and she is
a person one can always talk to. Medini di has been a source of motivation for many of
us. Either her own experiments or organization of events, she used to plan every minute
detail in advance. Aamir, the child prodigy, is always energetic to do experiments as well
as organize lab trips. Being in the mess committee, he introduced lot of good items in the
menu. Kimberly, a very good friend, has been my lunch and dinner companion whenever
I am signed out of mess. Both of us are easily bored of mess food and always ready to eat
outside. She is a person who is always ready to help others. Sai, to whom all of us run
to for any problem, has been responsible for managing the SEM and organizing the lab
shifting, designing and getting the infrastructure ready. Paritosh, the network manager
and Saurav, the treat manager have always entertained the lab with their pj competition.
Tanweer, the birthday manager, does the hard work of remembering everyone’s birthday
and organizing birthday cake and treat. Phani, the sports manager, organizes the yearly
lab sports day, which use to leave half of the lab injured. Ranjit or Ranj as I call him,
has been a tough competitor in the badminton matches. I also acknowledge Semonti,
Avradip, Anindita and Tatha for all their help. Ziggy has been a good friend and always
provides useful tips. I would also like to thank all other postdocs; Soumik da (also known
as PD), Jayanta, Partho and project assistants; Komal, Sneha, Anupam, Uma, Priya,
Reema, Bhargava and Ashwani. The young and energetic undergrads of the lab Amogh,
Pranav, Irfan also needs to mentioned. Thanks to Amogh for drawing some schematics
in blender. Our group administrator Swathi needs special mention for processing all the
order for equipments and reimbursement of our bills. I would like thank Saurav, Kimberly,
Aamir, Paritosh and Tatha for proof reading this thesis.

I had a good time with all my IPhD batchmates Nirmalendu, Amit, Ananyo, RPM,
Indrani, Dipankar, Somyadip, Debtosh, Zuala, Pranab, Vineeth and Arun. Debarghya
is one of the first person I became friends with since college. I thank him and RPM for
always being there. I thank all the seniors and friends at IISc, Manas da, whose advice
I always consider very seriously, Debarshini di, who is a very good friend as well, SSR,
Ranjan da, Ketan, Abhiram, Bidya da, Subhro da, Sayantan da, Sayak da, Sumilan da,
Saroj da, Gaurav, Sudeesh, Adhip and others. I must mention my friends Kabir, Niyaz,
Yunus, Nasir, Ata, Umair, Sajid, Jithin, Razzaq, Zaki, Hassan, Shahid, Rustam, Imran,
Altaf, Zahid, Hussain, Khwaja, Abdullah, Younis, Aslam, Farhan for all their help and
support.
I thank the Physics Department office staff for handling all the administrative work
and ensuring that we get our scholarships. I have always bugged Meena madam, Srivatsa
Sir and Bhargavi madam with queries related to scholarships, conference reimbursements
and orders. Mr. Shariff from the Physics workshop has been of tremendous help in
designing and building any infrastructure for lab equipments or custom made setups. I
would like to acknowledge the CCT department for supplying helium without which the
experiments would not gave been possible. I thank Prof. Kasturirangan for his help.
Prof. Venkataraman, the CCT chairman has always been supportive and has helped us
with helium even in crunch situations. I thank Mr. Muniraj for all his efforts to ensure a
continuous supply of liquid helium.
I must also mention my friends and teachers from Presidency College, which was my
second home. I acknowledge the teachings of DRC, DS, AN, PKM, BRC, MA, TSN and
others. Semanti, Srirupa, Saibal, Soham, Nirmalendu have been very good friends and I
remember our college days, badminton and tt matches, tank and quadrangle adda. I also
thank Mayukh, Arka, Joydeep, Saunak and Shakya. My school teachers Mr. Francis,
Mr. T. K. Shah, Mr. Sanyal, Mr D’Costa, Mr. Foran and Mr. Donaghue also deserve
mention.
I have to thank my family without whose support this journey would not have been
possible at all. They have always motivated and inspired me to pursue my PhD and
have kept me away from all distractions. Their patience has been commendable and its
not possible to describe their contribution in words. I thank my mother, father, brother
Aaquib, sisters Zeba, Neda and Nehan for all their support. I also thank Sarfraz, my
brother-in-law and Zehab, my nephew whose actions fill me with joy.
Above all I thank the Almighty for guiding me.
Publications
The work for this thesis has resulted in the following publications:

1. “Thermoelectric properties of electrostatically tunable antidot lattices”, Srijit Goswami,


Christoph Siegert, Saquib Shamim, Michael Pepper, Ian Farrer, David A. Ritchie
and Arindam Ghosh, Appl. Phys. Lett., 97, 132104 (2010).

2. “Suppression of low-frequency noise in two-dimensional electron gas at degenerately


doped Si:P δ layers”, Saquib Shamim, Suddhasatta Mahapatra, Craig Polley,
Michelle Y. Simmons and Arindam Ghosh, Phys. Rev. B, 83, 233304 (2011)

3. “Spontaneous breaking of time reversal symmetry in strongly interacting two dimen-


sional electron layers in silicon and germanium,” Saquib Shamim, Suddhasatta
Mahapatra, Giordano Scappucci, W. M. Klesse, Michelle Y. Simmons and Arindam
Ghosh, Phys. Rev. Lett. 112, 236602 (2014)

4. “Ultra low flicker noise in atomically patterned embedded nanostructures in silicon”,


Saquib Shamim, Bent Weber, Daniel W. Thompson, Michelle Y. Simmons and
Arindam Ghosh (In preparation)

5. “Effect of superconducting contacts on transport in low density δ-doped Si,” Saquib


Shamim, Suddhasatta Mahapatra, Michelle Y. Simmons and Arindam Ghosh (In
preparation)

Conference Papers:

6. “Origin of noise in two dimensionally doped Silicon and Germanium.”, Saquib


Shamim, Suddhasatta Mahapatra, Giordano Scappucci, Craig Polley, Michelle Y.
Simmons and Arindam Ghosh, AIP Conference Proceedings 1566, 413 (2013).

7. “Electrostatic modulation of periodic potentials in a two-dimensional electron gas:


from antidot lattice to quantum dot lattice”, Srijit Goswami, M. A. Aamir, Saquib
Shamim, Christoph Siegert, Michael Pepper, Ian Farrer, David A. Ritchie and
Arindam Ghosh, AIP Conference Proceedings1566, 257(2013)

ix
Abstract

Doped semiconductor systems have for decades provided an excellent platform to


study novel concepts in solid state physics such as quantum hall effect, metal-to-insulator
transition (MIT), weak localization and many body interaction effects. Doped Si, in
particular and doped Ge has been studied extensively to study MIT as a function of
dopant concentration or uniaxial stress. Spin transport phenomena have also been probed
in bulk doped Si. All the previous studies involved bulk doped semiconductors where the
dopants are spread through the bulk of the material. However spatial confinement of
dopants in one or more dimensions may lead to a range of exotic quantum phenomena
such as an absence of Anderson localization in one and two dimensions, hole-mediated
(Nagaoka) ferromagnetism and new modes of quantum transport, when the Fermi energy
lies at or close to centre of the band. Since many of these phenomena are inherent to
lower dimensions, it has been hard to observe these experimentally in bulk doped crystals
of Si and Ge. Recent advances in the monolayer doping techniques with atoms that
closely pack on a surface, has made it possible to design a new class of 2D electron
systems (2DES) in elemental semiconductors, such as Si and Ge, where the dopant (P)
atoms are confined within a few atomic planes. The uniqueness of these systems lies not
merely in the planar doping profile in bulk semiconductors that allow versatile designs
of nanodevices, such as 1D wires, tunnel gaps and quantum dots, but also that it is now
possible to study the interplay of wavefunction overlap and commensurability effects in
2D with unprecedented control. From an application perspective as well these systems
are technologically important as they are aimed at being the building blocks of a solid
state quantum computer. This thesis deals with investigating the electrical transport
properties, both average (resistance) and dynamic (noise) of doped semiconductor systems
in 2D delta layers, 1D wires and 0D quantum dots.
We find that the 2D δ-layers shows suppressed low frequency noise and the Hooge
parameter of delta doped Si is about five to six orders of magnitude lower when compared
to bulk doped Si in metallic regime. At low temperatures, the noise arises in these
systems due to universal conductance fluctuations. For 1D wires as well we find that
the Hooge parameter is one of the lowest among various 1D systems including carbon
nanotubes. We identify that charge traps in the Si/SiO2 are responsible for causing noise
in δ-doped systems. Then we study the noise and transport in 2D delta layers as a
function of doping density (and hence carrier density and interaction). Weak localization
corrections to the conductivity and the universal conductance fluctuations were both found
to decrease rapidly with decreasing doping in the Si:P and Ge:P delta layers, suggesting a
spontaneous breaking of time reversal symmetry driven by strong Coulomb interactions.
At low doping density we observe metal-like dependence of resistance on temperature at
low temperatures, raising the possibility of a metallic ground state in 2D at 0 K in doped
semiconductors. Finally we probe the low density devices (with broken time reversal
symmetry) using superconducting Al as ohmic contacts. Anomalous increase in resistance
below the superconducting transition temperature of Al and magnetoresistance with a
sharp peak at 0 T is observed. Additionally we find that when the Al is superconducting,
there exists a non-local resistance in low doped devices.
Contents

1 Introduction 1
1.1 Strong correlations: Failure of band theory . . . . . . . . . . . . . . . . . . 2
1.1.1 Mott Insulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Anderson Localization . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.3 Metal-Insulator transition in two dimensions . . . . . . . . . . . . . 7
1.2 Low dimensional systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.2 Drude model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.3 Resistivity and conductivity tensor . . . . . . . . . . . . . . . . . . 15
1.2.4 Quantum transport in mesoscopic systems . . . . . . . . . . . . . . 17
1.2.4.1 Effective mass . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.4.2 Characteristic length scales . . . . . . . . . . . . . . . . . 17
1.2.4.3 Weak localization . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.4.3.1 Effect of a perpendicular magnetic field . . 22
1.2.4.4 Universal conduction fluctuations . . . . . . . . . . . . . . 23
1.2.4.4.1 Reduction of UCF in a perpendicular mag-
netic field . . . . . . . . . . . . . . . . . . . . . 23
1.2.4.4.2 Parallel magnetic field . . . . . . . . . . . . . 26
1.2.4.5 Tranport in confined nanostructures . . . . . . . . . . . . 26
1.3 Resistance/Conductance noise . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.3.1 Types of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

xiii
xiv CONTENTS

1.3.1.1 Johnson or Nyquist Noise . . . . . . . . . . . . . . . . . . 28


1.3.1.2 Low frequency or 1/f noise . . . . . . . . . . . . . . . . . 29
1.3.2 Theory of 1/f noise . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4 Motivation of the present work . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5 Systems studied in this thesis . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.5.1 Two dimensional Si:P δ-layers . . . . . . . . . . . . . . . . . . . . . 36
1.5.1.1 Bandstructure of δ-doped Si . . . . . . . . . . . . . . . . . 37
1.5.1.2 Transport characteristics . . . . . . . . . . . . . . . . . . . 38
1.5.2 2D δ-doped layers in Ge . . . . . . . . . . . . . . . . . . . . . . . . 40
1.5.3 One-dimensional wires in Si:P . . . . . . . . . . . . . . . . . . . . . 41
1.5.3.1 Transport properties . . . . . . . . . . . . . . . . . . . . . 42
1.6 Thesis Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2 Experimental techniques 45
2.1 Ultra-low temperature setup: Installation of dilution refrigerator . . . . . . 45
2.1.1 Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.1.2 Wiring for electrical measurement . . . . . . . . . . . . . . . . . . . 47
2.2 Electrical transport measurement techniques . . . . . . . . . . . . . . . . . 50
2.2.1 Resistance measurement . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.2 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.2.1 Noise: Mathematical formulation . . . . . . . . . . . . . . 52
2.2.2.2 Noise: Measurement Scheme . . . . . . . . . . . . . . . . . 54
2.2.2.3 Noise: Commonly used circuits . . . . . . . . . . . . . . . 55
2.2.2.4 Noise figure of the amplifier . . . . . . . . . . . . . . . . . 58
2.2.2.5 Noise: Data acquisition and digital processing . . . . . . . 59
2.2.2.6 Aliasing effect and its removal: Decimation . . . . . . . . 59
2.2.2.7 Noise: Estimation of power spectral density . . . . . . . . 63
2.2.2.8 Time domain analysis: Weiner filtering . . . . . . . . . . . 66
2.3 Device fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
CONTENTS xv

2.3.1 Two dimensional δ-doped layers of phosphorous . . . . . . . . . . . 68


2.3.1.1 Si:P δ-layer . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.3.1.2 Ge:P δ-layer . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3.2 One dimensional Si:P wires . . . . . . . . . . . . . . . . . . . . . . 72

3 Two dimensional P δ-layers in Si and Ge 75


3.1 Overview and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.2 Conductivity: Dependence on carrier density . . . . . . . . . . . . . . . . . 76
3.2.1 Theoretical estimation of momentum relaxation time, τ0 . . . . . . 78
3.3 Conductivity: Dependence on temperature . . . . . . . . . . . . . . . . . . 82
3.4 Conductivity fluctuations: Low Frequency Noise . . . . . . . . . . . . . . . 83
3.4.1 Low frequency noise: 1/f spectrum . . . . . . . . . . . . . . . . . . 84
3.4.2 Low frequency noise : The Hooge relationship . . . . . . . . . . . . 84
3.4.3 Comparison of noise magnitude with bulk doped semiconductors . . 86
3.4.4 Disorder which leads to low frequency noise . . . . . . . . . . . . . 88
3.4.5 Hooge parameter for different densities . . . . . . . . . . . . . . . . 89
3.4.6 Dependence of noise magnitude on temperature . . . . . . . . . . . 91
3.5 Summary of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4 Conductance noise in atomically patterned nanostructures in Si 95


4.1 Overview and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2 One-dimensional wires of P in Si . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.1 Gate voltage characteristics of the 1D wires . . . . . . . . . . . . . 96
4.2.2 Conductance fluctuations: Low frequency noise measurements . . . 98
4.2.2.1 Time series and power spectral density . . . . . . . . . . . 98
4.2.2.2 Comparison of noise magnitude with other 1D systems . . 99
4.2.2.3 Gate Voltage dependence: Noise due to charge traps . . . 100
4.2.2.3.1 Random telegraphic signals . . . . . . . . . . . . 100
4.3 Quantum Dots in Si:P . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.3.1 Gate voltage characteristics of the quantum dots . . . . . . . . . . . 108
xvi CONTENTS

4.3.2 Conductance fluctuations in quantum dots . . . . . . . . . . . . . . 108


4.3.2.1 Time series and power spectral density . . . . . . . . . . . 108
4.3.2.2 Stability in terms of effect of charge traps . . . . . . . . . 111
4.4 Summary of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

5 Transport and Noise in strongly interacting 2D δ-layers 113


5.1 Overview and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2 Magnetoconductivity in transverse magnetic field . . . . . . . . . . . . . . 115
5.2.1 Low perpendicular magnetic field . . . . . . . . . . . . . . . . . . . 118
5.2.2 Reduced quantum correction to conductivity . . . . . . . . . . . . . 119
5.3 Universal Conductance Fluctuations . . . . . . . . . . . . . . . . . . . . . 124
5.3.1 Fluctuation in conductance due to the movement of a single impu-
rity, δG21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.3.2 Magnetic field dependence of UCF . . . . . . . . . . . . . . . . . . 126
5.3.2.1 Reduction of UCF in parallel magnetic field . . . . . . . . 127
5.3.2.2 Low perpendicular magnetic field . . . . . . . . . . . . . . 128
5.3.2.3 High perpendicular magnetic field . . . . . . . . . . . . . . 130
5.4 Role of spins in breaking time reversal symmetry . . . . . . . . . . . . . . 132
5.4.1 Conductivity in parallel magnetic field . . . . . . . . . . . . . . . . 132
5.4.2 Full range fit and extracting the spin scattering time . . . . . . . . 134
5.5 Low density limit and metallic behavior . . . . . . . . . . . . . . . . . . . . 136
5.6 Summary of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

6 Effect of superconducting contacts on transport in δ-layers 139


6.1 DC Current Voltage characteristics . . . . . . . . . . . . . . . . . . . . . . 140
6.2 Resistivity vs temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.3 Magnetoresistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.3.1 Parallel magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.3.2 Perpendicular magnetic field . . . . . . . . . . . . . . . . . . . . . . 144
6.4 Resistance fluctuations or low frequency noise . . . . . . . . . . . . . . . . 147
CONTENTS xvii

6.4.1 Parallel magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . 147


6.4.1.1 Time series of resistance fluctuations . . . . . . . . . . . . 147
6.4.1.2 Integrated variance . . . . . . . . . . . . . . . . . . . . . . 149
6.4.2 Perpendicular magnetic field . . . . . . . . . . . . . . . . . . . . . . 151
6.5 Non-local measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.6 Magnetoresistance and nonlocal measurements on Ge:P δ-layers with su-
perconducting contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.7 Highly doped devices with superconducting contacts . . . . . . . . . . . . . 157
6.8 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.9 Summary of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

7 Conlusions and Outlook 161


7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Scope of future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.2.1 Measurements on gated hall bars of Si:P δ-layers . . . . . . . . . . . 164
7.2.2 Thermal transport in δ-layers . . . . . . . . . . . . . . . . . . . . . 165
7.2.2.1 Thermopower . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.2.2.2 Thermal conductivity . . . . . . . . . . . . . . . . . . . . 165
7.2.3 Dopants arranged in a periodic lattice . . . . . . . . . . . . . . . . 166

A Annealed Silver 167


A.1 Weidemann-Franz law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
A.2 Annealing procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

B Preliminary Thermopower measurements 171


B.1 Introduction to thermopower . . . . . . . . . . . . . . . . . . . . . . . . . . 171
B.2 Measurement of Thermovoltage . . . . . . . . . . . . . . . . . . . . . . . . 173
B.3 Estimation of thermopower using Mott’s relation . . . . . . . . . . . . . . 175
B.4 Thermopower for low density Si:P δ-layer . . . . . . . . . . . . . . . . . . . 177
xviii CONTENTS
Chapter 1

Introduction

One of the most successful theory to describe condensed matter solids is the band theory.
Even in its simplistic form it has been able to explain the properties of many systems but
it has also failed particularly in case of increased interactions which can strongly alter the
states of a system. Another unavoidable ingredient of any real system is disorder which
can fundamentally modify the ground state of a system. The interplay of disorder and
interaction has been a long standing problem in condensed matter physics with several
theoretical attempts in trying to understand the competition between. The phenomena of
Anderson localization which leads to an insulating state in presence of strong disorder, the
Mott insulating regime due to strong electronic correlations and the accompanying metal-
insulator transition have been researched theoretically and experimentally in great detail.
Though many of the predictions of the theory were experimentally verified, there were
observations which suggests the need to go beyond the existing theory of localization.
This gains significance due to technological advancements in realizing low dimensional
systems with varying degrees of disorder and interaction.

In this thesis we will be dealing with semiconductors (Si and Ge) δ-doped with P
atoms in two, one and zero dimensions. We will reporting observations at low doping of
P atoms where the interaction effects are enhanced. Hence in the next few sections we
will review some theoretical and experimental work related to disorder, interactions and
metal-insulator transition.

1
2 Chapter 1. Introduction

1.1 Strong correlations: Failure of band theory

The band theory of solid considers the formation of quasi-continuous energy bands in-
stead of individual electronic states. One of the triumphs of band theory is that it has
successfully classified many solids into metals and insulators by filling up the band with
required number of electrons. Ironically, it is also one of the failures of band theory that
it predicts many solids to be metals while they are actually insulators [1]. A famous
example is of the transition metal oxides like MnO, FeO, CoO and CuO, which should
be metals according to band theory but they are infact insulators. This anomaly arises
due to assumptions of the band theory where is does not take electronic correlations into
account.

Mott-Hubbard
Insulator
Interactions

Anderson
Disorder Insulator

Figure 1.1: A representation of the three parameters namely temperature, disorder and
interactions; an interplay of which determines the state of a material. The origin is state at
zero temperature, non-interacting and free from any disorder state. Actual experimental
systems fall somewhere in the three-dimensional space, away from the origin. Redrawn
from Ref. [2].
1.1. Strong correlations: Failure of band theory 3

In order to understand the electronic states of any material system, it is important


to consider three parameters; namely temperature, disorder and interactions. Figure 1.1
shows the three dimensional space of these parameters. The origin represents the zero
temperature, free from any disorder and non-interacting situation to which most of the
conventional band theory applies. However any actual experimental system can fall any-
where in the three-dimensional space and the nature of electronic states, the mechanism of
localization, etc. depend on an interplay and combination of all these factors. Increasing
the disorder (along the x-axis) also leads to an insulating state called the Anderson insu-
lator, which will be dealt with in section 1.1.2. Increasing electron-electron interactions
(along the y-axis) leads to the formation of a Mott-Hubbard insulator which we discuss
in the next section.

1.1.1 Mott Insulator

In the non-interacting picture, the filling of bands determine whether a system is a metal
or an insulator at 0 K. If the highest filled band is partially filled then its a metal, while for
a completely filled highest occupied band, its an insulator. In other words, we can say that
for an insulator the Fermi lies within the band gap whereas for a metal its inside the band.
This picture though successful for many systems, fails for transition-metal oxides which
have partially filled d-electron bands, yet they are insulators [3, 4]. The importance of
strong electron-electron correlations was then understood and it was realized that strong
Coulomb interactions can give rise to an insulting phase in these systems [5–9]. In order
to understand this, Mott considered a model with a single electronic orbital on each site
which leads to the formation of a single band without taking into account any electron-
electronic interactions. A completely filled band means that each site is occupied by two
electrons with opposite spin. However the two electrons at the same site would feel strong
Coulomb repulsion U between them which can be introduced in the Hubbard hamiltonian
as [10–13]

∑ † ∑ ∑
H = −t (ciσ cjσ + H.c) + µ niσ + U ni↑ ni↓ (1.1)
⟨i,j⟩ iσ i
4 Chapter 1. Introduction

E0+U

E0

࢚Ȁࢁ
Figure 1.2: Schematic of the lower and upper Hubbard band formed due to strong
Coulomb repulsion between two electrons at the same site. The gap is a function of
the ratio t/U where t and U are the overlap integral and on-site Coulomb repulsion re-
spectively. At sufficiently large t/U the two bands merge and the system undergoes a
transition from insulator to metal.

where c†iσ (ciσ ) is the creation (annihilation) operator for at site i, niσ = c†iσ ciσ is the
number operator and t is the overlap integral. This implies that the total energy will
be higher if two electrons are at the same site. The strong Coulomb repulsion results in
splitting of the band into two: a lower band where the electrons occupy an empty site
and an upper band (of higher energy) where the electrons have to occupy a site where an
electron is already present (Fig. 1.2). The gap between the two bands is because of the on-
site Coulomb energy U which is required if an electron has to occupy a site which already
has an electron. For the case of one electron per site (half-filling), the lower band is filled
completely and hence the system is an insulator. Note that the splitting of the band is
determined by the ratio t/U (alternatively the competition between the potential energy
and kinetic energy). When the ratio t/U is sufficiently large the two Hubbard bands
merge into one and the material changes from insulator to metal as shown schematically
in Fig. 1.2
1.1. Strong correlations: Failure of band theory 5

U/t

Mott-Insulator
Metal Metal

Filling Control - MIT

Bandwidth Control- MIT


X<1 X>1
Filling(x)

Figure 1.3: Metal-insulator phase diagram based on the Hubbard model in the plane of
U/t and filling. Two possible routes for the metal-insulator transition (MIT) are shown:
the bandwidth-control MIT and the filling-control MIT. The shaded area is in principle
metallic but under the strong influence of the metal insulator transition, in which carriers
are easily localized by extrinsic forces such as randomness and electron-lattice coupling.
Reproduced from Ref. [14]

In addition to the ratio U/t another important parameter in the Hubbard model is
the band filling x which if tuned suitably can also fuel a Mott metal-insulator transition
(MIT) [14]. A phase diagram of the Mott MIT as a function of U/t and x is shown in
Fig. 1.3. The x = 0 and x = 2 fillings correspond to the band insulator while for the
half-filled case (x = 1), tuning U/t drives the insulator-to-metal transition at a critical
value of U/t (except in the case of perfect nesting, where the critical value U is zero).
This transition at a finite critical U is called a bandwidth control MIT [6, 8]. Another
possible route to the Mott MIT is via the band filling x and is called as the filling control
MIT which has gained prominence in the 3d-electron systems ever since the discovery of
high-temperature superconductivity.
6 Chapter 1. Introduction

The metallic state near the Mott insulating phase exhibits fluctuations and orderings
in spin, charge and orbital degrees of freedom and has properties different from ordinary
metals. The metallic phase near the Mott insulator have been experimentally investigated
in d-electron systems. Some examples include Ti, V, Cr, Mn, Fe, Co, Ni, Cu and Ru
compounds. The rich and exotic phenomena which they exhibit are strong spin and
orbital fluctuations [15, 16], mass renormalization effects [17–19], incoherence of charge
dynamics [20] and phase transitions under control of key parameters such as band filling,
bandwidth and dimensionality [14]. Experimentally these parameters can be varied by
doping, pressure, chemical composition and magnetic field.
The Mott MIT reveals that electron-electron interactions cannot be ignored in sys-
tems where the ratio U/t is large. Infact strong electronic correlations are important
factors to be considered in many aspects other than Mott metal-insulator transition like
magnetism, high-temperature superconductors, Kondo effect, charge ordering, ultra-cold
gases in optical lattices and many others.

1.1.2 Anderson Localization

In 1958, Anderson laid down the foundation for concept of localization due to disorder in
his seminal work on the criteria for diffusion of a particle on a random lattice [21]. He
considered the motion of a quantum particle on a lattice with energy varying randomly
from site to site . The magnitude of the fluctuations in energy for different sites correspond
to disorder. He showed that below a critical disorder strength, the particle was able to
diffuse in the lattice whereas for disorder large than the critical value no diffusion was
allowed. In the former case the state of the particle corresponds to an extended state
(metal) where as the latter situation is described by a localized state (insulator). This
transition from metallic to insulating behavior as a function of disorder is called the
Anderson transition and the insulating phase is known as the Anderson insulator.
It was known that in one-dimensions, all the electronic eigenstates are localized in the
presence of arbitrarily weak disorder. However the scaling theory of localization developed
by Abrahams, Anderson, Licciardello and Ramakrishnan in 1979 surprisingly predicted
that even in two dimensions (2D) any infinitesimal disorder will localize all the electronic
states [22, 23]. The experiments performed at very low temperatures in that decade seem
1.1. Strong correlations: Failure of band theory 7

to confirm the predictions of the scaling theory. A review on the theory and experiments
on localization can be found in Ref [24]. However number of recent experiments have
reported the presence of metallic state in 2D systems at low temperatures which we will
discuss in Section 1.1.3.
The localization effects known for weakly disordered electronic systems is called as
weak localization in which the conductivity of a system decreases logarithmically at low
temperatures due to interference of time reversed trajectories. This effect will be discussed
in Section 1.2.4.3.

1.1.3 Metal-Insulator transition in two dimensions

As discussed in the previous section, scaling theory of localization predicts that there
cannot be a metallic state in two dimensions in zero magnetic field [22]. As the tem-
perature decreases the resistance is expected to increase exponentially (known as strong
localization) or logarithmically (known as weak localization, discussed in Section 1.2.4.3)
finally becoming infinite in the limit T → 0. The theory did not take into account the
interactions between particles. However it was shown by Althsuler, Aronov and Lee [25]
that weak electron-electron interactions would further increase the localization. When the
electron-electron interactions are very large, then even in the presence of a small amount
of disorder the system is expected to be a pinned Wigner crystal which cannot show any
conduction at zero temperature [26]. Hence under all circumstances (weak or strong in-
teractions) a 2D system is expected to be an insulator. This particular scenario was well
supported by experiments for about two decades. The expected logarithmic increase of
resistivity was observed thin metal films [27, 28] and Si MOSFETs [29, 30] while the ex-
ponential increase in resistivity with decreasing temperature was observed for low density
MOSFETs [30].
Though there were some reports by Finkelstein [31] and Castellini [32] on the existence
of metallic state characterized by finite conductivity in 2D at zero temperature due to an
interplay of disorder and interactions, it was not in serious contention as the metallic state
was outside the range of validity of the perturbative theory. However when there were
several experimental reports on the observation of metallic state in 2D, it was realized
that the understanding of the MIT in 2D systems is far from complete [33–43].
8 Chapter 1. Introduction

Kravchenko et al. studied a high mobility two dimensional electron system in low
disordered Si MOSFETs and showed that at low electron density (but higher than the
critical density below which the system becomes an insulator) the resistivity drops by an
order of magnitude as the temperature is decreased below ∼ 1 − 2 K [33]. No evidence
of localization was found till 20 mK. The resistivity as a function of temperature of a
low-disordered Si MOSFET for electron densities ranging from 7.12 − 13.7 × 1010 cm−2
(corresponding to an interaction parameter rs ∼ 15 − 20, where rs is the ratio of electron-
electron interaction energy and Fermi energy) is shown in Fig. 1.4a. For low electron
densities, the usual monotonic increase of resistivity with temperature was observed while
for higher densities, there was a non-monotonicity. The resistivity initially increased
slowly with decreasing temperature till T ∼ 2 K, below which it dropped sharply. At
even higher densities, the resistivity was initially constant and then dropped sharply with
decreasing T exhibiting strong metallic behavior. They interpreted their results as a
possible metal-insulator transition in 2D.

The metallic behavior at low temperatures have been observed in other 2D systems like
a hole gas in GaAs/AlGaAs heterostructures [40]. Hanein et al. reported measurement
on a hole gas with densities in the range 8.9 × 109 cm−2 to 6.4 × 1010 cm−2 corresponding
to rs ∼ 9 − 24. The resistivity as a function of temperature is shown in Fig. 1.4b. At
low hole densities, insulating behavior with resistivity increasing with decreasing T is
seen, whereas for hole densities higher than the critical density, the insulating behavior
is observed at at higher T , but below a few hundred mK, the resistivity drops with
decreasing T . At even higher densities metallic behavior is observed in the entire range of
temperature investigated. Note that a common feature in both the systems (Si MOSFETs
and hole gas in GaAs/AlGaAs heterostructures) is that metallic behavior was observed
in the strongly interacting regime. Additionally, at the density above and below which
metallic and insulating behavior was observed respectively, the resistivity is of the order
of h/e2 .

The low temperature drop in resistivity have been observed in other dilute electron
and hole systems like p-SiGe [44,45], p-GaAs/AlGaAs [46,47] and n-AlAs [48]. There have
been other reports that suggested the drop in resistivity to be a finite temperature effect
and predicted that at even lower temperatures an insulating behavior will be observed [49].
1.1. Strong correlations: Failure of band theory 9

(a) (b)

Figure 1.4: (a) Temperature dependence of resistivity in a dilute low disordered


Si MOSFETs for electron densities ranging from 7.2 − 13.7 × 1010 cm−2 at
zero magnetic field. Reproduced from Ref. [34] (b) Resistivity of a hole gas in
GaAs/AlGaAs, as a function of temperature at various fixed hole densities, p =
0.089, 0.094, 0.099, 0.109, 0.119, 0.125, 0.13, 0.15, 0.17, 0.19, 0.25, 0.32, 0.38, 0.45, 0.51, 0.57,
and 0.64 × 1011 cm−2 at zero magnetic field. There are three distinct regimes: insulating
regime at low densities, a mixed regime at intermediate densities indicated by dashed
lines, and a metallic like regime at high densities. The inset shows schematic of a
p-type inverted semiconductor-insulator semiconductor structure that was used in the
experiments. Reproduced from Ref. [40].

It was found that the metallic behavior observed in the 2D systems at low temperatures
was suppressed when an external magnetic field was applied at arbitrary angle with re-
spected to the plane of the 2D system [50–52]. It was suggested that the suppression of
metallic behavior was correlated with the degree of spin polarization [51, 52].
Recently there has been report on disorder-driven Mott insulator to metal transition
in a single crystal of layered dichalcogenide 1T-TaS2 [53]. As the disorder was increased
by Cu intercalation (which does not change the carrier concentration), the states within
the Mott gap were delocalized leading to a finite dc conductivity. All these observations
have opened up the debate of MIT in 2D. The question as to whether there is true
10 Chapter 1. Introduction

Figure 1.5: Two different scenarios for Mott metal-insulator transition (MIT) in doped
semiconductors. (a) The usual electron transition to a Anderson-Mott insulator which
ultimately goes to a pure Mott insulator. (b) Alternative scenario in which the electrical
MIT occurs at density nc1 resulting in a transition to a quantum spin liquid with gapless
spinon excitations. Another transition occurs at density nc2 to Anderson-Mott insulator.
Reproduced from Ref. [55].

metal-insulator transition at the critical density and if there is a metallic phase at higher
densities remains unanswered experimentally. A detailed review of metallicity in 2D can
be found in Ref. [54].
An alternative route to MIT has recently been predicted in doped semiconductors
such as Si:P and Si:B via an intermediate quantum spin liquid phase between the metal
and Mott-Anderson insulator [55]. A phase diagram depicting two possible routes to MIT
is shown in Fig. 1.5. The conventional picture of electron localization transition to a
Anderson-Mott insulator which finally crosses to a pure Mott insulator as the density is
reduced is shown in Fig. 1.5a. In the alternative scenario, it is proposed that the electrical
MIT occurs at a density nc1 and the system becomes a charge insulator. However there
exists gapless spinon excitations in a phase of quantum spin liquid (Fig. 1.5b). There
have been experimental evidence of such a phase in clean Mott insulator systems with
strong charge fluctuations which destroy magnetic ordering. In a naturally half-filled
doped semiconductor like Si:P, where, near the MIT there are strong charge fluctuations
and the random lattice structure prohibits any magnetic ordering, an exotic quantum spin
liquid phase is likely to exist.
In the last few sections we tried to give an overview of the existing theoretical and
experimental investigations related to disorder, interaction and metal-insulator transition
with particular emphasis on 2D systems. The formation of Anderson insulting ground
1.2. Low dimensional systems 11

state due to disorder and the Mott insulating phase due to strong Coulomb interactions
were discussed. However, the transition from the Anderson localized state to the Mott
insulating phase by tuning the interaction is still unresolved. The existence of a true
metallic ground state in 2D at zero temperature is debatable and the role of the Coulomb
interactions in stabilizing a 2D metal is unclear. A complete microscopic theory to ex-
plain the experimental observations of 2D MIT is lacking. The metallic behavior occurs
at low densities where the interactions are strong, and at resistivity ∼ h/e2 suggesting
that disorder also plays a role. This reiterates the importance of understanding the inter-
play of disorder and interaction in determining the ground state of a system. There exist
conflicting opinions as to whether the observation of MIT in 2D is indeed a signature of
a true zero-temperature quantum phase transition. Possible explanations include a novel
ground state at 0 K of perfect metal, a spin liquid or a Wigner glass, etc [56]. Alterna-
tive explanations suggest that localization will persist at 0 K and the observed metallic
behavior is due to temperature dependent screening, interband scattering, etc [56]. In
other words, there is no consensus on the existence of a quantum critical point in 2D.
Some theories have suggested the existence of a metallic phase which coexists with long
range anti-ferromagnetic order at intermediate disorder between the Anderson and Mott
insulator [57]. Several other concepts such as spin susceptibility, mass enhancement and
occurrence of magnetism needs to be looked into. It is an experimental challenge to inves-
tigate the phenomenology of 2D MIT by fabricating 2D solids in which the doping level
and interacting magnitude can be tuned independently.
The experimental investigation of the interplay of disorder and interaction has pro-
gressed due to the ability to realize low dimensional systems. Below we discuss about
low dimensional quantum systems and key concepts that are essential to understand the
transport phenomena in them.

1.2 Low dimensional systems

A low dimensional system is one in which the motion of electrons (or any other microscopic
degree of freedom like phonons or photons) is restricted along one or more dimensions.
In other words, electrons are not allowed to move in the three dimensional space. When
12 Chapter 1. Introduction

the confinement is along one, two and three dimensions we get a two dimensional (2D),
one dimensional (1D) and a zero dimensional (0D) system respectively. The interest
in low dimensional systems has grown enormously due to new discoveries and exotic
physics which they offer along with the platform for the realization of new state-of-the-art
electronic device architectures. Understanding the effect of disorder and electron-electron
interaction has gained more insight through the investigation of low dimensional quantum
systems. The study of low dimensional quantum phenomena has to led to new fields of
research like the physics of mesoscopic systems. We discuss some important concepts that
are crucial to understanding this thesis below.

1.2.1 Density of states

For a system of free (non-interacting) electrons, the ground state and allowed energy
values can be determined by solving the Schrodinger’s equation. Plane wave solutions
with periodic (or the Bon-von Karman) boundary conditions leads to only certain allowed
values of electron wave vector, k, given by kx = 2πnx /Lx , ky = 2πny /Ly and kz =
2πnz /Lz . The dispersion relation is

~2 k 2
E(k) = , (1.2)
2m

where m is the electron mass, and Lx , Ly and Lz are the dimensions of the system. The
properties of system and its response to various external perturbations depends crucially
on number of states per unit energy at each energy (called the density of states, DOS)
available for the electron. The behavior of the DOS with energy depends on the number
of dimensions in which the electrons are allowed to move. For an n-dimensional wave
wave vector space, the density of states is (2π)−n as can be seen readily from the allowed
values of k.

For the specific case of a two dimensional electron system, where the electrons are
allowed to move freely along two dimensions, the electron waves are freely propagating
1.2. Low dimensional systems 13

(a) 3D DOS
(b) 1D DOS (c) 0D DOS
DOS (Arb units)

E3

E2
2D DOS

E1

Energy, E (Arb units) Energy, E, (Arb Units) Energy, E, (Arb Units)

Figure 1.6: The density of states D(E) for (a) 3D and 2D (b) 1D and (c) 0D.

plane waves in y − z plane (say) and the energy levels are given by,

~2 ky2 ~2 kz2 n2x π 2 ~2 ~2 ky2 ~2 kz2


E(k) = Enx + + = + + , (1.3)
2m 2m 2mL2x 2m 2m

Thus we get a series of subbands, indexed by nx each with a dispersion relation. The
2D DOS, D(E), per unit area and energy, E, is

1 dk 1 dk
D(E) = gs gv 2
2πk = 2gv 2
2πk , (1.4)
(2π) dE (2π) dE

where gs = 2 and gv are the spin and valley degeneracy respectively and k is the wave
vector. For systems with parabolic band structure the electronic spectrum is given by
Eq. 1.2, the DOS is

gv m
D(E) = (1.5)
π~2

The DOS for the 2D case is shown in Fig 1.6a. The DOS increases abruptly at the band
edges and is constant within a subband. The 3D DOS, given by D(E) ∝ E 1/2 is also
shown for comparison.

For a 1D system, the electron waves are freely propagating plane waves only along one
spatial dimension and are confined along other two. The energy levels are given by
14 Chapter 1. Introduction

~2 kz2 n2 π 2 ~2 n2y π 2 ~2 ~2 kz2


E(k) = Enx + Eny + = x 2 + + , (1.6)
2m 2mLx 2mL2y 2m

In this case, there are subbands, indexed by nx and ny , with 1D dispersion. The DOS for
a 1D system is given by

( )1/2
1 dk 1 2m
D(E) = gs gv 2 = gv , (1.7)
(2π) dE π~ E

Figure 1.6b shows the DOS for a 1D system. Note that the DOS diverges at the sub-
band minima. These divergences in the DOS are called van-Hove singularities and are
responsible for many unique properties of 1D systems. For a zero dimensional object, the
electron is confined along all spatial dimensions and no free motion is allowed. For such
a system the states are available only at discrete energies and the DOS is described with
a delta function as shown in Fig. 1.6c.

1.2.2 Drude model

The Drude model was proposed in 1900 by Paul Drude to explain the transport properties
of electrons in materials [58,59]. Considering the metal as a gas of electrons and applying
the kinetic theory of gases, the Drude conductivity was found to be,

ne2 τ0
σD = , (1.8)
m

σD , τ0 and m are the Drude conductivity, momentum (or elastic) relaxation time and
electronic mass respectively and n is the carrier density in the material. The mobility
of the charge carriers is µ = eτ /m. The Drude model neglects electron-electron inter-
actions (independent electron approximation) and electron-ion interaction (free electron
approximation) between collisions.
1.2. Low dimensional systems 15

1.2.3 Resistivity and conductivity tensor

When magnetic field is applied perpendicular to a current carrying 2D (say) conductor an


electric field develops perpendicular to the current. In such a scenario when the electric
field and current are not in the same direction the resistivity ρ and conductivity σ are not
treated as scalars but rather tensors, ρ and σ and can be written as

( )
ρxx ρxy
ρ= (1.9)
ρyx ρyy

( )
σxx σxy
σ= (1.10)
σyx σyy

with the relation

ρ = σ −1 (1.11)

In steady state the rate at which the electrons receive momentum from the external
field is equal to the rate at which they lose momentum due to scattering, that is:

mvd
= e[E + vd × B] (1.12)
τ0

where E and B are the applied electric and magnetic field respectively and vd is the drift
velocity of electrons. Rewriting Eq. 1.12 as,

( )( ) ( )
m/eτ0 −B vx Ex
= (1.13)
B m/eτ0 vy Ey

where the subscript x and y denote the x- and y- components of the respective quantities.
Using J = nevd , where J is the current density, we can write
16 Chapter 1. Introduction

( )( ) ( )
m/eτ0 −B Jx /ne Ex
= (1.14)
B m/eτ0 Jy /ne Ey

Rearranging the above equation, we get

( ) ( )( )
Ex 1 −µB Jx
= σ −1 (1.15)
Ey µB 1 Jy

where σ = neµ. Since we know that

( ) ( )( )
Ex ρxx ρxy Jx
= (1.16)
Ey ρyx ρyy Jy

we get,

ρxx = σ −1 (1.17)

ρyx = −ρxy = (µB)/σ = B/en (1.18)

Thus the Drude model predicts that as a function of perpendicular magnetic field the lon-
gitudinal resistance remains constant while the Hall resistance increases linearly. However
the it has been observed that, in high mobility two dimensional electron systems at low
temperatures and high perpendicular magnetic field, ρxx shows oscillatory behavior and
ρxy shows plateaus at the minima of ρxx which was not explained by the Drude model.
This phenomena known as Quantum Hall effect [60] is beyond the scope of this thesis and
will not be discussed here. Additionally the Drude model does not account for the quan-
tum corrections to conductivity which is seen at low temperatures for diffusive systems,
which we discuss in the next section.
1.2. Low dimensional systems 17

1.2.4 Quantum transport in mesoscopic systems

In this section we will briefly describe concepts that are essential to understand the trans-
port in mesoscopic systems. The concept of effective mass, relevant length scales and the
phenomena of weak localization & universal conductance fluctuations, which are impor-
tant with respect to the our work, are discussed. The phenomena of Coulomb blockade
oscillations in quantum dots or wires will also be introduced.

1.2.4.1 Effective mass

As an electron moves in a system, three fields are acting on it: (i)nearly periodic potential
arising due to the atoms in the crystal (ii) Electron electron interaction potential and (iii)
externally applied electric field. Among these the atomic potential varies more rapidly
than the others. The problem of an electron moving in the presence of atomic potentials
can be approximated by the simpler problem of an electron moving in slow varying fields
only but with a different mass called the effective mass. The dynamics of the conduction
band electron can be described by the equation:

[ ]
(i~∆ + eA)2
Ec + + U (r) Ψ(r) = EΨ(r) (1.19)
2m∗

The above Eq. 1.19, though looks like the Schrödinger equation, is called a single-band
effective mass equation. The lattice potential does not appear explicitly in Eq. 1.19, rather
its effect is incorporated through the effective mass m∗ . For example, for an electron in
GaAs conduction band, the effective mass m∗ ≈ 0.067m where m is the electronic mass
while for electrons in Si, m∗ ≈ 0.26m.

1.2.4.2 Characteristic length scales

For a conductor to show ohmic behavior, its dimensions should be larger than the three
characteristic length scales which we discuss below.

Wavelength: For a 2D system, we know that the Fermi wavevector kF = 2πn.

The corresponding wavelength can be written as λF = 2π/kF = 2π/n. Just to have
an idea of numbers, for an electron density of about 1014 cm−2 , λF is ∼ 2.5 nm. At low
18 Chapter 1. Introduction

temperatures the current is carried mainly by electrons with energy close to the Fermi
energy so that the λF is the relevant length. Electrons with less kinetic energy have longer
wavelengths but they do not contribute to the conductance.
Mean free path: The mean free path is the distance travelled by an electron before
its initial momentum is destroyed, i.e.

l = vF τ0 , (1.20)

where vF = ~kF /m is the Fermi velocity. Typically for n ∼ 1014 cm−2 , vF ∼ 3×107 cm s−1 .
For τ0 ∼ 100 fs, we get l ∼ 30 nm.
Phase breaking length: As the name implies, the phase breaking length lϕ is the
distance travelled by en electron before its phase is randomized. It is related to the phase
breaking time τϕ as

lϕ2 = Dτϕ , (1.21)

where D = vF2 τ0 /2 is the electron diffusivity. The importance of τϕ will be seen in section
on weak localization.

1.2.4.3 Weak localization

Weak localization is a phenomena by which the conductivity of a system starts to decrease


at lower temperatures. It is seen in many diffusive systems at low temperatures and
is explained as due to constructive interference between phase coherent trajectories of
electron waves. In order to understand a basic mechanism of weak localization, consider
the diffusive motion of an electron from point X to Y. The electron can take many different
paths from X to Y (Fig. 1.7a) and the classical probability of an electron going from point
X to point Y is the sum of all possible paths. This can be written as

Cl
PA→B = Σi Pi , (1.22)
1.2. Low dimensional systems 19

where Pi ’s are the probability of the individual paths. The quantum mechanical proba-
bility is however obtained by squaring the sum of amplitudes and is given by

Q
PX→Y = (Σi Ai , )2 (1.23)

For the above case the phase of the waves arriving at Y along different paths (hav-
ing different lengths) are random and hence cancel each other. Thus the cross product
interference terms do not contribute. This however is not true for a path having a self
intersection like the one shown in Fig. 1.7b. The propagator for the two different types
of path (or diffusion modes) shown in Fig. 1.7a and b, considered above are known as
Diffuson (direct paths) and Cooperon (self intersecting path). For simplicity in Fig. 1.7b,
we consider the case where the starting and end points are same, i.e. X = Y and calculate
the probability of an electron to return to origin. A coherent phase relation now exists be-
tween two partial waves propagating clockwise and anticlockwise around the same closed
path having equal amplitude at the origin by the virtue of time reversal symmetry. Hence
the quantum probability to return at the origin is,

PReturn = (A + A)2 = 4A2 (1.24)

(a) (b)

X =Y

Figure 1.7: (a) Multiple paths that an electron can take to go from point X to Y. (b) A
self intersection path which contributes to weak localization.
20 Chapter 1. Introduction

Figure 1.8: The probability distribution of a diffusing electron which starts at r = 0 at


the time t = 0. In case of quantum diffusion (dashed peak) the probability to return to
the origin is twice of that of classical diffusion (full curve). If there is large spin-orbit
scattering, the probability reduces by a factor of two (dotted peak) and yields a weak
antilocalization. Reproduced from Ref. [61]
.

Thus the probability for an electron to return to its starting point is twice of that of
the classical probability (=2A2 ). This implies that the diffusion of an electron from one
point to another is reduced implying a reduction in electrical conductivity or an increase
in electrical resistivity. In Fig. 1.8 the classical and the quantum diffusion probabilities
are plotted qualitatively [61]. For quantum diffusion, the dashed peak at the origin de-
scribes a tendency to remain at or return to the origin leading to increase in resistivity.
However in the presence of large spin-orbit scattering the quantum diffusion gives a re-
duced probability (dotted peak) to return to the origin leading to what is known as weak
antilocalization. Weak localization is observed in systems for which lphi > l because the
electron needs to remember its phase even after multiple scattering.

The free electron weak localization correction ∆σW L for a 2D disordered conductor
has a universal magnitude given by [62–65],
1.2. Low dimensional systems 21

( ) ( )
αe2 τϕ
∆σW L ≈ ln (1.25)
πh τ0

where α is a phenomenological prefactor close to unity and τ0 and τϕ are the momentum
relaxation time and phase breaking time respectively. The τϕ is a temperature dependent
quantity having an inverse power law relation with temperature. Specifically for a 2D
system we have [66],

τϕ ∝ T −1 (1.26)

for dephasing due to electron-electron scattering which is the dominant dephasing mech-
anism at low temperatures. An electron feels the fluctuating electric field produced by all
the other electrons that destroys it phase after a time τϕ . Due to inverse power law de-
pendence of τϕ , the weak localization correction varies logarithmically with temperature.
After the theory of weak localization was developed, there were reports on the logarithmic

(b)

(a)

Figure 1.9: The resistance of thin disordered films as function of the logarithm of the
temperature. (a) a AuPd-film by Dolan et al. [27] and (b) Cu-film by Van den dries et
al. [28]
22 Chapter 1. Introduction

increase of resistance of thin disordered films with temperature as shown in Fig. 1.9a for
AuPd-film [27] and Fig. 1.9b for thin Cu-films [28]. These experimental results appeared
to be an experimental proof of the theory of weak localization. However Altshuler et
al. [25] showed that in disordered electron systems the Coulomb interaction is modified
and it is long ranged which has impact on the DOS as well as the resistance of disordered
2D systems. This leads to the same functional dependence of resistance on temperature
as in case of weak localization. Hence a logarithmic temperature dependence of resistance
is not a conclusive proof of weak localization. The magnetic field dependence of resistance
however has features characteristic of weak localization as discussed below. A detailed
theoretical treatment of the weak localization can be found in Ref [61].

1.2.4.3.1 Effect of a perpendicular magnetic field When a magnetic field is ap-


plied perpendicular to the closed loops, the relative phase of the two waves change by

(a)

(b)

Figure 1.10: (a) The magnetoresistance of a thin Mg-film at different temperatures as a


function of the perpendicular magnetic field. The units of the field are given on the right
of the curves. The points represent the experimental results and the line are theoretical
fits taking into account small spin-orbit scattering [67]. (b) The magnetoresistance of
a 2D electron system in GaAs/AlGaAs heterostructures at different temperatures. The
black solid lines are theoretical fits [68].
1.2. Low dimensional systems 23

2eϕ/~, where ϕ is the magnetic flux [62]. Hence increasing the magnetic field dephases
the waves causing the electrical conductivity to increase. Hence observation of a positive
magnetoconductivity is a characteristic feature of weak localization. The magnetic field
dependence of the weak localization correction is given by the Hikami formula [69] which
for a disordered 2D system is

[ ( ) ( )]
αe2 B⊥ B⊥
∆σW L (B⊥ ) = F −F , (1.27)
πh Bϕ B0

where F (x) = ln(x)+ψ(0.5+1/x), and ψ(x) is the digamma function, B0 is the momentum
relaxation field and Bϕ = ~/4eDτϕ is the phase breaking field with D being the electron
diffusivity.
Figure 1.10a and 1.10b shows the magnetoresistance of thin Mg-films and 2D electron
system in GaAs/AlGaAs heterostructures respectively, at different temperatures in a per-
pendicular magnetic field. The negative magnetoresistance is due to the weak localization
effect. The experimental points are fitted with theory to extract the phase coherent time.

1.2.4.4 Universal conduction fluctuations

Universal conductance fluctuations (UCF) is an essential ingredient of quantum transport


in disordered systems at low temperatures and is analogous to weak localization (WL),
since both UCF and WL are manifestation of the same quantum interference phenomena.
Due to multiple backscattering events, electrical conductance G of a system fluctuates
with an universal magnitude of ∼ e2 /h as a function of magnetic field, Fermi energy or
disorder configuration [70–72]. The UCF provide information on the scattering processes
that is complementary to the time-averaged transport. In order to probe the diffusion
modes (Cooperons and diffusons) we need the UCF magnitude as a function of magnetic
field over a wide range (over the scale of Bϕ and BZ ). Measuring UCF magnitude as a
function of magnetic field is a direct probe to time reversal symmetry.

1.2.4.4.1 Reduction of UCF in a perpendicular magnetic field For a standard


free electron metal the signature of UCF noise is reflected in the noise measurements as
a factor of two reduction in noise magnitude at two characteristic field scales, (i) the
24 Chapter 1. Introduction

Table 1.1: Values of the symmetry parameters k, s & β with and without magnetic field.
k is the no. of independent Eigen values, s is the level degeneracy and β is the Wigner-
Dyson parameter, Bϕ is the phase breaking field and BZ ∼ kB T /gµB B is the Zeeman
field.

k s β ks2 /β Symmetry class


B=0 1 2 1 4 Orthogonal
B ∼ Bϕ 1 2 2 2 Unitary
B ∼ Bc2 2 1 2 1 Unitary

phase breaking field Bϕ , due to the removal of time reversal symmetry [73–77] and (ii)
the Zeeman field scale BZ , where the spin degeneracy is lifted [78].
The reduction of UCF noise in presence of a magnetic field can be understood in terms
of the contribution of paths of an electron moving from an initial position to some final
position [77]. It is well known that there are two kinds of paths in quantum transport
that contribute to the probability of an electron moving from one point to another. One
is the classical diffuson path while another is the self intersecting path called cooperon.
The two paths are shown schematically in Fig. 1.11. Note that for the diffuson, the
original and time reversed path are traversed in the same direction in the interior of
the loop, whereas for the cooperon, the original and the time reversed path are covered
in opposite directions. Application of perpendicular magnetic field removes the time

Diffuson Cooperon

Figure 1.11: Two kinds of paths in quantum transport, diffusons (left) and cooperons
(right). The dark colored lines are paths from one position to another, while the light
colored lines are the time reversed path.
1.2. Low dimensional systems 25

reversal symmetry. At a critical field, Bc1 of the order of Bϕ (which is the field required to
introduce one flux quantum within a phase coherent box) the cooperon contribution gets
∫−
→− →
eliminated. This is due to the additional phase (∝ A . dl ) acquired in a perpendicular
magnetic field. For a diffuson, this phase cancels for the original and the time reversed
path, whereas for a cooperon the phase is nonzero. Thus due to incoherent superposition
between different loops, the cooperon contribution gets cancelled. The contribution from
cooperon and diffuson being equal, the noise reduces exactly by a factor of two when the
cooperon contribution is removed on application of magnetic field.
The reduction in noise in presence of a perpendicular magnetic field can also be un-
( 2 )2 2
derstood from random matrix theory. The conductance fluctuations is ⟨δG2 ⟩ ≈ eh ksβ ,
where k is the number of independent Eigen value sequences, s is the level degeneracy and
β is the Wigner-Dyson parameter which depends on the symmetry class to which the sys-
tem belongs. In presence of time reversal symmetry and weak spin-orbit interaction the
system belongs to orthogonal ensemble, which is characterized by β = 1, whereas when
the time reversal symmetry is broken (due to B ∼ Bϕ ) the system makes a transition to
unitary ensemble for which β = 2. The change of β from 1 to 2 causes the conductance
fluctuations to decrease exactly by a factor of two as shown in the schematic of Fig. 1.12a.

(a) (b)

Figure 1.12: (a) Schematic of reduction of UCF magnitude in a perpendicular magnetic


field at two field scales, the phase breaking field Bϕ and Zeeman field BZ (b) Experimental
data for reduction in UCF as a function of perpendicular magnetic field in mesoscopic
Lithium wires at 4.2 K (filled squares) and 1.6 K (filled circles). The solid line is theoretical
fit. The graph is adapted from Ref. [78]
26 Chapter 1. Introduction

Further increase of B removes the spin degeneracy due to Zeeman splittig (EZ = gµB B),
which changes s from 2 to 1 and k from 1 to 2. This causes the conductance fluctua-
tion to further decrease by a factor of two (Schematic in Fig. 1.12b). It should be noted
that the critical field at which the second factor of two reduction occurs, is determined
by the condition EZ > max{ED , kB T }, where ED = ~D/L2ϕ . Table 1.1 shows how the
various parameters change in presence of a magnetic field which reduces the conductance
fluctuations.
Figure 1.12b shows the measurements of UCF as a function of B⊥ for mesoscopic
lithium wires by Birge, et al. [78]. The UCF reduces by factor of two at two field scales,
Bϕ and BZ . At lower temperatures, since both Bϕ and BZ reduces, the reductions happen
at a lower field scale. The magnetic field dependence of resistance fluctuations has also
been observed in GaAs/AlGaAs heterostructures [79, 80].

1.2.4.4.2 Parallel magnetic field When the magnetic field is applied parallel to the
plane (in the same plane as current) then we observed only a factor of two reduction at
B∥ ∼ BZ due to the removal of the spin degeneracy.

1.2.4.5 Tranport in confined nanostructures

In this section we will briefly discuss about Coulomb blockade oscillations which is ob-
served in quantum dots or wires. A quantum dot is basically an island of electrons with
dimensions such that the energy required to add an extra electron to the quantum dot is
greater than the thermal energy. The Coulomb blockade oscillation have been observed
in the Si:P 1D wires and 0D quantum dots which has been measured in this thesis.

Coulomb blockade oscillations In order to understand the phenomenon of Coulomb


blockade oscillations we a consider a quantum dot, a schematic of which is shown in
Fig. 1.13a and a scanning electron microscopy image in Fig. 1.13b. A quantum dot
is a puddle or an island of electrons which is separated from the source and drain via
tunnel barriers. A capacitively coupled gate tunes the energy levels of the quantum
dot. The occupied states of the quantum dot is separated from the available states by a
1.2. Low dimensional systems 27

(a) (b)

(c) (i) (ii) (d)

Figure 1.13: (a) Schematic of a quantum dot (blue circle) connected to source and drain
via tunnel barriers and capacitively coupled to a gate electrode (b) Scanning tunneling
microscopy image of a quantum dot (taken from Ref. [81]) (c) (i) Energy levels of a
quantum dot showing the charging energy e2 /C and (ii) The gate voltage shifts the energy
levels of the quantum allowing conduction to happen (e) Coulomb blockade oscillations
in a quantum dot as a function of gate voltage.

charging energy e2 /C, where C is the capacitance of the system, as shown schematically in
Fig. 1.13c(i). A small source drain bias leads to a slight difference between the chemical
potential of the source (µS ) and drain (µD ) leads. When there are no available states
between µS and µD , due to the energy gap, then electron transport cannot occur through
the quantum dot and the dot is said to be in Coulomb blockade regime with zero current
(point A in Fig. 1.13d). Application of gate voltage shifts the energy levels of the dot
such that at some point a filled or an empty state can come in between the source and
drain leads such that an electron can tunnel into and out of the dot. Conduction is not
allowed and current can now flow through the dot (point B in Fig. 1.13d). Thus a gate
voltage can be used to tune the number of electrons in the quantum dot.
28 Chapter 1. Introduction

1.3 Resistance/Conductance noise


Resistance noise or fluctuations in electrical resistance are present in almost every system
and is an important probe for the underlying dynamics of charge transport in the system.

1.3.1 Types of noise


In general there are three types of noise that are present in any condensed matter system:

1. Johnson (or Nyquist) noise, which is due to random motion of charge carriers at a
finite temperature and is independent of any voltage across the device. For a device
of resistance R at a temperature T , the Johnson’s noise has a value 4kB T R. Note
that it is independent of frequency and is also called as ’White noise’.

2. Shot noise, which occurs due to discrete nature of charge carriers. The spectral
dependence is given by SI (f ) = 2qI, where q and I are the charge of carrier and
current flowing through the sample respectively

3. Low-frequency (or 1/f) noise, the power spectral density of which has a form SV ∝
1/f α , where α the frequency exponent is in the range 0.8 − 1.2, hence the name
1/f noise. This noise exists in systems in addition to the thermal noise and shot
noise and can be measured as voltage fluctuations when a constant current flows
through the device. Alternatively, a constant bias can be applied to the device and
fluctuations in current can be recorded.

1.3.1.1 Johnson or Nyquist Noise

The Johnson (or Nyquist) noise also called as the thermal noise has a value of 4kB T R was
observed by Johnson in 1928 [82] and theoretically derived by Nyquist [83] in the same
year. Following the derivation in Ref [83], consider a circuit of two identical conductors
each of resistance R connected as shown in Fig. 1.14a. Thermal agitation of carriers in
Conductor I generates an electromotive force V causing a current I = V /2R in the circuit.
This results in power, P = I 2 R being transferred from conductor I to II. Similarly power
is transferred from conductor II to conductor I. Since at equilibrium, both the conductors
are at the same temperature T , the power from conductor I to conductor II is same
1.3. Resistance/Conductance noise 29

as that from conductor II to conductor I and this equality holds at all frequencies [83].
Now consider the circuit shown in Fig. 1.14b, where the conductors are joined by a long
nondissipative line of length l. At equilibrium there are waves traveling from conductor
I to conductor II and vice-versa. Instead of waves traveling in opposite directions, we
consider the vibrations of the line at its natural frequencies. For a frequency range, df ,
the number of nodes are 2ldf /v, where v is the speed of propagation. The total energy of
vibrations is 2lkB T df /v (since energy per degree of freedom is kB T ). Hence the average
power transmitted from one conductor to other within frequency range df is kB T df .
Equating it to I 2 R = V 2 df /4R, we get,

⟨V 2 ⟩ = 4kB T R, (1.28)

(a) (b)

R R R R

Figure 1.14: (a) Two identical conductors of resistance R (b) Two indentical conductors
of resistance R each connected by a non-dissipative line of length l.

1.3.1.2 Low frequency or 1/f noise

This type of noise manifests itself in a wide variety of systems ranging from bulk 3D
semiconductors to thin metallic films to magnetic flux noise in superconductors [84] to
heartbeats [85]. The power spectrum of 1/f type of noise depends quadratically on the
applied voltage across the device and hence can be increased hugely above the thermal
noise and measured. In most solid state systems, 1/f noise arises due to relaxation of
defects with a finite relaxation time.
30 Chapter 1. Introduction

1.3.2 Theory of 1/f noise


This thesis focuses on the study of 1/f noise in Si:P and Ge:P systems and hence we
discuss the various existing theories of 1/f noise. It should however be kept in mind
that none of theories alone can describe the origin of noise in various physical systems.
The mathematical formulation and the techniques used to obtain the noise data from the
sample is discussed in detail in Chapter 3.

1. Hooge model: This empirical model was put forth by Hooge in 1976 to explain
the behavior of noise in homogenous metal and semiconductor systems [86]. Based
on the assumption that noise originates from the bulk, a phenomenological equation
called Hooge relation represents the power spectral density (PSD) as

V2
SV (f ) = γH , (1.29)
Nfα

where V, N and f are the voltage bias across the sample, total number of carriers
and frequency respectively, α is the frequency exponent and γH is called as the
Hooge parameter which quantifies the noise magnitude of a system. The Hooge
model is based on the mobility fluctuations which arise due to scattering of carriers
by the lattice phonon modes, with varying phonon occupation number in different
modes [87–89]. Since the mobility fluctuations due to scattering from impurities
has limited impact in this model, it is mostly applicable to clean and homogeneous
semiconductors.

Though the Hooge model was originally proposed as a mobility fluctuation model
and was thought to be applicable to clean and homogeneous semiconductors, it is
now widely used for a variety of systems like thin metal films [75], graphene [90],
carbon nanotubes [91], nanowires [92, 93], transition metal dichalcogenides [94, 95],
oxides [96], shape memory alloys like NiTi [97], etc. The relations SV ∝ V 2 and
SV ∝ 1/N hold true for many systems and is often used as a starting point for many
noise measurements. The phenomenological parameter γH has been calculated for
many systems and used as a standard to compare the noise magnitude of various
1.3. Resistance/Conductance noise 31

systems.

2. Mc Whorter model: This model also known as the number fluctuation model
was put forth by WcWhorter to explain the 1/f noise behaviour in MOSFETs [98].
It considers the 1/f noise arising due to the slow exchange of carriers between the
conducting channel and oxide traps located at the oxide-semiconductor interface.
Such trapping and detrapping leads to fluctuations in the number of carriers which is
manifested as 1/f noise in drain-source current (IDS ). The PSD of noise calculated
using the McWhorter model is [99],

SI (f ) kB T q 2 Nt (EF )
2
= , (1.30)
IDS 8W Ln2 e2 f

where Nt (EF ) is the oxide trap density near EF , L,W are the length and width
of the channel respectively and n is the carrier number density. As seen from the
2
Eq. 4.2, SI /IDS ∝ 1/n2 is a distinctive feature of this model.

3. Dutta-Horn model: This is a phenomenological model which was originally pro-


posed to explain 1/f noise behavior in thin metallic films [100]. According to this
model 1/f noise arises due to slow relaxation of defects or disorder occuring within
the experimental time scale. For a single defect relaxation time τ with an exponen-
tial autocorrelation function ϕ ∼ e−t/τ , the PSD is a Lorentzian given by,


S(f ) ∝ , (1.31)
1 + (2πf τ )2

For a distribution of relaxation times F (τ ), the PSD becomes

∫ ∞

S(f ) ∝ dτ F (τ ) , (1.32)
0 1 + (2πf τ )2

For F (τ ∼ 1/τ ) within the range τmin ≪ τ ≪ τmax , S(f ) ∼ 1/f in the range
32 Chapter 1. Introduction

−1 −1
τmax ≪ f ≪ τmin . The physical basis for choosing F (τ ) ∼ 1/τ is by considering the
−1
relaxation time to be a product of the microscopic attempt time (τ0 ∼ ωD , where
ωD is the Debye frequency) and the exponential of the activation energy E for defect
migration in the scale of kB T , i.e. τ = τ0 eE/kB T . Thus we get,

∫ ∞
2τ0 eE/kB T
S(f, T ) ∝ D(E)dE, (1.33)
0 1 + (2πf τ0 eE/kB T )2

For a slow varying D(E) at the scale kB T we get S(f ) ∼ 1/f α .

4. Local interference model: This model considers the resistivity fluctuations aris-
ing from the fluctuating interference of electrons in the local environment of a mov-
ing defect. Kogan and Nagaev considered rotations of point defects with symmetry
lower than the crystal [101]. Using dimensional arguments, they calculated the
change in the scattering cross section of a defect when it rotates. Pelz and Clarke
however calculate the change in scattering cross section and predict the magnitude
of resistivity fluctuations as [102],

[ ]
⟨(δρ)2 ⟩ nm
N = (nlβC Λc )2 , (1.34)
⟨R⟩ 2 n

N = n/V is the number of atoms in the sample of volume V , βC and Λc are the
anisotropy parameter and scattering cross section respectively, l is the elastic mean
free path and nm is the concentration of mobile defects which contribute to the local
interference. It is worth mentioning that the Kogan-Nagaev and Pelz-Clarke model
differ only by the anisotropy parameter.

5. Univeral conductance fluctuations: As discussed earlier universal conductance


fluctuations arises due to quantum mechanical interference of backscattered elec-
tronic wavefunctions. This interference is extremely sensitive to the configuration
of disorder/impurities, a change in which causes the conductance to fluctuate [103].
The evidence for 1/f noise due to UCF mechanism have been observed in thin
1.3. Resistance/Conductance noise 33

metal films at low temperatures [75, 104, 105], C-Cu composites [106], amorphous
Ni-Zr [107], bulk-doped Si [76]. The mechanism of conductance fluctuations, its
quantitative estimation and relation to 1/f noise will be discussed in Chapter 3
along with experimental measurements of UCF noise.
34 Chapter 1. Introduction

1.4 Motivation of the present work


The present thesis work has been motivated by two aspects, namely

• Fundamental physics of low dimensional doped semiconductors

• Application of δ-doped devices in quantum circuits

Doped semiconductor systems have for decades provided an excellent platform to


study novel concepts in solid state physics such as quantum hall effect (QHE), metal-
to-insulator transition, weak localization and many body interaction effects. Doped Si,
in particular and doped Ge has been studied extensively to study MIT as a function of
dopant concentration or uniaxial stress. Though there have been reports of metallic state
in two dimensions due to an interplay of disorder and interaction, it is still debatable as to
whether a metallic ground state can be stabilized in a disordered 2D system by Coulomb
interaction. It is known that at large effective on-site Coulomb interaction in the absence
of disorder, the ground state is a Mott insulator. In the presence of disorder, coherent
backscattering of electrons by impurities leads to an insulating state in 2D, known as
the Anderson insulator even without any interactions. The transition from Anderson
localized state to the Mott insulator by varying the interaction is still an open problem
in condensed matter physics [14, 32, 54, 108–112]. All the previous experimental studies
of Si semiconducting systems have involved bulk doped Si where the dopant atoms are
spread throughout the bulk of the material. The 2D systems that were used to study
quantum hall effect and MIT were based on high mobility 2D electron systems in Si
MOSFETs [33, 54], Si/SiGe [113, 114] systems and III-V heterostructures [115–117]. A
gate electrode was used to tune the carrier density and increase the effective Coulomb
interaction. The fundamental limitation of these systems is that they do not allow doping
level and interaction magnitude to be tuned independently.
Progress in monolayer doping techniques have now led to the formation of a new class
of two dimensional electron systems in semiconductors such as Si and Ge where the dopant
phosphorous (P) atoms are confined to a single or few atomic planes. Since each P atom
contributes one electron, the δ-doped Si:P and Ge:P systems are naturally half-filled.
The key advantage of such a system is that the effective on-site Coulomb interaction U/t,
where U is the on-site Coulomb interaction and γ is the overlap integral, can be varied
1.4. Motivation of the present work 35

by changing the doping density of P atoms, while maintaining the condition of half-filling
throughout. At half-filling the interaction effects are enhanced as evidenced in short-range
spin correlations, local moments and Mott-Hubbard charge gap formation [14, 109, 110,
118–120]. This tunability makes δ-doped semiconductor systems an ideal candidate to
explore interaction driven metallicity in 2D. In addition to this, other exotic phenomena
like hole-mediated (Nagaoka) [121, 122] ferromagnetism and a quantum spin liquid phase
as an intermediate phase between the metal and Mott-Anderson insulator have been
predicted to exist in doped semiconductor systems like Si:P [55].

In addition to the fundamental physics aspect detailed above, δ-doped devices offer a
formidable platform in terms of application towards quantum circuits. This is connected
to a proposal by Kane of having a solid state quantum computer in Si by using the nuclear
spin of P atoms as qubits [123]. However this required positioning the dopant atoms with
atomic precision such that they are aligned to nanometer sized gates. The monolayer
doping technique combined with the scanning tunneling microscopy lithography have
overcome this challenge and it has been possible to place P dopant atoms in Si with an
accuracy of 1 nm. This has allowed versatile design of nanodevices such as 1D wires [124,
125], 0D quantum dots [126–128] and tunnel gaps [129]. The fabrication of single or few
donor devices takes us forward to address the solid state quantum bits and perform the
required quantum operation. The advancement in gaining control of the bits at the atomic
level and it tunability however comes at a cost. The devices are now extremely sensitive
to fluctuations (of any parameters). This includes the local electromagnetic environment
of dopants whose influence should be possible to reduce via appropriate design of the
devices. It has been observed that many quantum computing devices exhibit a 1/f
spectra of different variable arising from different physical processes. The low frequency
fluctuations are one of the sources of decoherence in the quantum computing nanodevices.
In order to understand the influence of low frequency fluctuations on quantum bits, we
consider a two-level system with Hamiltonian.

~
H= (εσzP + ΥσxP ) (1.35)
2
36 Chapter 1. Introduction

where ~ε and ~Υ are the diagonal splitting of the levels and their tunnel coupling, and σz,x
are the Pauli matrices. The above Hamiltonian corresponds a pseudospin in a magnetic
field B which can be time dependent. The environment in which the bit is and with which
it can interact creates a stochastic component in B which means stochastic components
in ε(t) and Υ(t) thus destroying the coherent evolution of the qubit [84].
Hence, in view of the significance of these devices towards quantum computation,
an immediate requirement is to know their stability in terms of low frequency noise. It
would be extremely desirable to investigate the conductance fluctuations in the δ-doped
2D layers in Si:P and Ge:P devices and compare them with bulk doped semiconductors
and other 2D systems as well. Additionally as the active device area reduces in size down
to the atomic scale, they become extremely sensitive to the presence of disorder (or charge
traps) which is unavoidable in any real system. In such a scenario, a detailed investigation
of the conductance fluctuations of all the devices ranging from 2D δ-layers, 1D wires and
0D quantum dots is an immediate requirement.

1.5 Systems studied in this thesis


In this section we discuss about the different systems investigated in this thesis. We
briefly describe their bandstructure and important theoretical and experimental works
done previously. Their fabrication process will be discussed in the Chapter 2 which deals
with experimental techniques.

1.5.1 Two dimensional Si:P δ-layers

When silicon is doped with a donor atom like phosphorous, it acquires an excess of free
electrons due to ionization of P atoms. The P atoms can be distributed in the bulk
(bulk doping) or it can be confined to a one atomic plane (δ-doping). In the latter case
we get a 2D electron gas in which the electrons are confined to the Coulomb potential
of the positively charged P ions. Such δ-doped systems have extremely high carrier
densities (∼ 1014 cm−2 ) as compared to other 2D systems like graphene [130], topological
insulators [131] or GaAS/AlGaAs heterostructures [132]. The Si:P δ-doped systems have
been studied theoretically [133–137] and experimentally [138–143] over the past decade.
1.5. Systems studied in this thesis 37

Here we review some of the work that helps in understanding the transport in these
δ-doped systems.

1.5.1.1 Bandstructure of δ-doped Si

It is well known that bulk Si due to its crystal symmetry has six fold valley degeneracy
which can be lifted either by strain or quantum confinement. For a δ-doped Si system
formed by strong confinement in the vertical direction, the 2D bandstructure is a projec-
tion of the bulk bandstructure onto a plane perpendicular to the confinement direction.
The Si:P δ-layers studied here are confined along the [001], it leads two Γ-pockets at k = 0
and four ∆-pockets at lower energy. Qian et al. have used density functional theory to
calculate the bandstructure of the 2D Si:P δ-layers which is shown in Fig. 1.15 [134]. The
zero energy is placed at the conduction band minimum of bulk Si. Due to extremely high
carrier density all the six valleys of Si conduction band are occupied (the lowest subband)
resulting in the Fermi energy of the carriers in the δ-layer to be ∼ 99 meV below the
conduction band minimum of bulk Si as shown in Fig. 1.15. Strong quantum confinement
in the vertical z-direction breaks the six fold valley degeneracy into a set of two Γ-bands
and four fold degenerate ∆-band. As can be seen from Fig. 1.15, the lowest Γ-band has a

Figure 1.15: Band structure of a 2D Si:P δ-layer calculated using density functional theory
by Qian et al.. Reproduced from Ref [134].
38 Chapter 1. Introduction

minimum ∼ 400 meV below the conduction band minimum of bulk Si. Hence the Fermi
energy now lies ∼ 300 meV above this band minimum leading to metallic character of
these δ-doped systems.

1.5.1.2 Transport characteristics

The transport characteristics of the Si:P δ-layers have been studied by conductivity mea-
surements as a function of temperature and magnetic field at the University of New South
Wales over the past decade. The highest density corresponding to a saturation doping
is ∼ 2 × 1014 cm−2 as has been confirmed by Hall measurements. This corresponds to a
conductivity σ > 100 e2 /h. From the Drude model the mobility µ for these δ-layers is
∼ 50 − 100 cm2 V−1 s−1 . A complete list of transport parameters for all the devices studied
in this thesis can be found in Chapter 3.

Figure 1.16: (a) Sheet resistance ρxx of Si:P δ-layers (blue triangles), (b) Mobility µ and
(c) Electronic mean free path l as function of carrier density nHall . Dashed lines are guides
to the eye. Figure reproduced from Ref. [139]
1.5. Systems studied in this thesis 39

Figure 1.17: (a) Magnetoconductivities ∆σ for Si:P δ-layers, (b) Transport (τ ) and (c)
Phase relaxation times (τϕ ) as a function of carrier density nHall , and (c) phase-coherent
length (lϕ ) as a function of nHall . Dashed lines are guides to the eye. Figure reproduced
from Ref. [139]

The resistivity in Si:P δ-layers with different carrier density have been studied by
Goh et al. [139]. Figure 1.16a shows that the resistivity of the δ-layer decreases as the
carrier density increases since now more number of electrons are available to carry current.
The mobility of the system µ = (ρxx nHall e)−1 decreases as nHall increases as shown in
Fig. 1.16b. This can be explained by considering scattering from charged dopants. As
nHall increases, the number of charged dopants increases causing an increase in number
of scattering centers leading to a decrease in µ. The mean free path l on the other hand
increases as nHall increases (Fig. 1.16c), although one would expect it to decrease as the

number of scattering centers increases. However l ∝ µ nHall means that decrease in µ is
more than compensated by an increase in nHall .
40 Chapter 1. Introduction

We now turn to magnetotransport characteristics of Si:P δ-layers [139]. The mag-


netoconductivity for different nHall is shown in Fig. 1.17a. All devices show positive
magnetoconductivity which is characteristic of weak localization. The momentum relax-
ation time τ0 decreases as nHall increases (filled triangles in Fig. 1.17b) as expected since
µ also decreases. The phase breaking time τϕ extracted by fitting the data with Eq. 1.27
increases as nHall increases (filled squares in Fig. 1.17b). This is the case for Nyquist
dephasing due to electron-electron interactions [63]. Note that τϕ is of the order of pi-
coseconds and τ0 few tens of femtoseconds. The phase breaking length lϕ increases on
increasing nHall as can be seen from Eq. 1.21.

1.5.2 2D δ-doped layers in Ge

Another 2D δ doped system can be realized by just replacing silicon with germanium [145–
148]. The fabrication process will be discussed in Chapter 2 on experimental techniques.
Here we review some results pertaining to transport characteristics of this system. The
magnetoconductivity at different temperatures from Ref [144] is shown in Fig. 1.18a. The

(a) (b)

Figure 1.18: (a) Magnetoconductivities ∆σ for Ge:P δ-layer with carrier density n ∼
4.6 × 1013 cm−2 at temperatures ranging from 0.3 to 6 K. The coloured lines are fits to
Eq. 1.27. (b) Phase relaxation time (τϕ ) as a function of temperature T showing that
τϕ ∝ T −1 . Figure reproduced from Ref [144].
1.5. Systems studied in this thesis 41

devices show weak localization behavior with the conductivity correction being higher at
lower T as expected since at phase coherence is more dominant at lower temperatures.
The phase breaking time τϕ extracted by fitting the magnetoconductivity to Eq. 1.27 is
plotted as a function of temperature in Fig. 1.18b. As T decreases, τϕ increases following
the dependence τϕ ∝ T −1 as expected for Nyquist dephasing due to electron-electron
interactions.

1.5.3 One-dimensional wires in Si:P

Apart from the 2D Si:P and Ge:P δ-layers, 1D wires in δ-doped Si has been studied
in this thesis. For 1D wires, apart from the confinement in the vertical direction there
is lateral confinement created by doping in a very thin region ∼ 2 − 5 nm using STM
lithography [124, 125]. In this section we review the theoretical and experimental work
that has already been done for the 1D wires. The bandstructure of these wires calculated
using a self consistent Schrödinger-Poisson solver is shown in Fig. 1.19 [149].
In Fig. 1.19 the evolution of bandstructure is studied as the diameter of the wire

Figure 1.19: Comparison of the band structures of a (a) 1.5 nm (2DR), a (b) 2.3 nm
(3DR) and (c) 4.6 nm (6DR) wide wire. The zero energy is placed at the conduction
band minimum of bulk Si. The figure has been reproduced from Ref [149].
42 Chapter 1. Introduction

increases. For 1.5 nm (Fig. 1.19a), strong band bending happens due to high donor
density and the minimum of the lowest occupied band Γ1 is about 243 meV below the
conduction band minimum of bulk Si. The Fermi energy is ∼ 135 meV above Γ1 band
minimum. As can be seen, there are a total of 6 effective modes, intersecting EF . As the
diameter increases to 2.3 nm (Fig. 1.19b) the number of modes increase 7 and for 4.6 nm
there are 13 modes. For 4.6 nm, the minimum of the lowest band is ∼ 330 meV below
the conduction band minimum of bulk Si. Note that as the diameter of the wire increases
the band structure converges more and more towards the 2D bandstructure.

1.5.3.1 Transport properties

The transport properties of these wires have been studied by Weber et al. [125, 150]. It
was found that even the thinnest wire of diameter 1.5 nm which consists of only two
dimer rows, obeyed Ohm’s law and the resistivity similar to bulk Si of equivalent doping

Figure 1.20: (a) Differential conductance at millikelvin temperatures, comparing the


4.6 nm (W1) and the 1.5 nm (W2) wide wires. (b) and (c) Conductance histograms,
P (Gw ), of both wires, indicating metallic and insulating behavior, respectively. The solid
blue lines are fits to a normal and a log-normal distribution, expected in the metallic and
insulating regimes, respectively. Reproduced from Ref [150]
1.6. Thesis Layout 43

density [125].
Transport measurements showed that the wires were metallic down to millikelvin tem-
peratures. Metallic conduction is a consequence of high carrier densities leading to oc-
cupation of all six valleys. However the 1.5 nm wire shows a breakdown of metallic
conduction at G ∼ < e2 /h due to localization of carriers. Fig. 1.20a shows the conductance
as a function of gate voltage for wires 4.6 nm (W1 with G ≫ e2 /h) and 1.5 nm (W2 with
G∼ < e2 /h) at millikelvin temperatures. Wire W1 shows UCF as confirmed by the normal
probability distribution in Fig. 1.20b. However wire W2 shows shows Coulomb blockade
oscillations as a function of gate voltage and the insulating behavior is confirmed by the
log-normal probability distribution [150].

1.6 Thesis Layout

This thesis comprises the study of δ-doped semiconductors from an application as well
as fundamental physics perspective. We focus on the electronic transport in δ-doped
Si and Ge in two, one and zero dimensions using conventional time-averaged resistance
measurements as well as time-dynamic noise (resistance or conductance fluctuations mea-
surements). Chapter 2 describes the setup, installation and electrical wiring of an ultra
low temperature dilution refrigerator (base temperature ∼ 10 mK) which is essential
for quantum transport measurements of low-dimensional semiconductors. Then it dis-
cusses the measurement techniques, namely resistance and noise measurements, used for
the devices investigated here. Finally, the fabrication procedure of the devices is briefly
described.
In Chapter 3 we study the basic characterization of the 2D Si:P and Ge:P δ-layers in
terms of conductivity and conductivity fluctuations measurement. From low frequency
noise measurements, the noise magnitude is estimated and compared with other doped
semiconductors. The temperature dependence of noise is studied and a possible source of
disorder is discussed.
In Chapter 4 we measure the conductance fluctuation of nanostructures in Si patterned
at the atomic scale. 1D wires as narrow as two dimer rows and 0D quantum dots are
discussed in terms of low frequency noise. We then describe the role of charge traps and
44 Chapter 1. Introduction

the mechanism by which they cause conductance fluctuations.


Chapter 5 deals with quantum measurements of 2D Si:P and Ge:P δ-layers of varying
density (from high doping to low doping) at temperatures down to millikelvin. Magneto-
conductance and universal conductance fluctuations measurement show that low doped
δ-layers behave differently from conventional disordered 2D materials. The interference
effects are suppressed and there is a spontaneous breaking of time reversal symmetry at
low doping when the interactions are enhanced. We discuss the role of local moments
driven spin fluctuations in the breakdown of time reversal symmetry.
In Chapter 6, the effect of coupling of superconducting contacts to the 2D Si:P and
Ge:P δ-layers is discussed. We show that the superconductivity in the contacts affects
the transport only in low doped devices. Magnetoresistance and low frequency noise
are studied across the superconducting transition as a function of magnetic field and
temperature. An alternate probe employing the nonlocal measurement of resistance is
used to understand the nature of transport in this regime.
Chapter 7 concludes this thesis with a summary of the main results and scope for
future directions of research that can be pursued.
In Appendix A, the details of the annealing of silver cold finger used in electrical
wiring of dilution refrigerator is discussed. Appendix B gives the preliminary results of
thermopower measurements in Si:P δ-layers.
Chapter 2

Experimental techniques

This chapter describes the methods and techniques that have been employed to carry out
the experimental work of the thesis. These include the process involved in the installation
of the experimental setup as well as the electrical measurement techniques. Many impor-
tant quantum phenomenon like weak localization, universal conductance fluctuations are
exhibited at low temperatures. The present chapter contains details of the installation
processes that were carried out to have a working ultra-low temperature setup and this
was an integral part of initial PhD work. Apart from these, this chapter also describes
the electrical measurement techniques used to study the transport properties of Si:P and
Ge:P δ-layers. Two kinds of transport techniques are used, namely, resistance measure-
ment (which is time averaged measurement) and low frequency resistance noise (which is
dynamic transport measurement). The basic formulation of noise along with the digital
signal processing techniques are discussed. Finally, we briefly describe the fabrication
procedure for the two-dimensional δ-layers and one dimensional wires.

2.1 Ultra-low temperature setup: Installation of di-


lution refrigerator
To achieve temperatures of the order of few mK we use a dilution refrigerator supplied
by Leiden Cryogenics B.V. The working of a dilution refrigerator relies on the peculiar
properties exhibited by the He3 -He4 mixture at low temperature (∼ 6.4% He3 remains
dissolved in He4 even at absolute zero) [151]. Below 0.8 K, a liquid mixture of the

45
46 Chapter 2. Experimental techniques

Dilution Refrigeration;
Phase separation
1K pot
ot

>90% 3He vapor


still
<1% 3He

heat
exchangers
rs

100% 3He
mixing
chamberr 6% 3He

Figure 2.1: Working principle of a He3 -He4 dilution refrigerator. Pic Courtesy: Arlette
de Waard, Leiden Cryogenics.

two isotopes phase separates into a He3 rich phase and a He4 rich phase. Lowering the
temperature further, results in dilute phase becoming more dilute and concentrated phase
becoming more concentrated. The working is shown schematically in Fig. 2.1.
The setup of the dilution refrigerator can be divided into two parts: (1) Infrastructure,
which involves support framework for maneuvering the system, and (2) Electrical mea-
surement which involves accurate measurement of temperature and wiring for electrical
contacts from the sample to measuring instruments.

2.1.1 Infrastructure

To operate a dilution fridge, the initial step is to cool the fridge to 4.2 K using liquid
helium. For this purpose the fridge has to be lowered into a liquid helium reservoir,
which also contains a 14 T superconducting magnet (from American Magnetics Inc.) for
doing experiments in high magnetic fields. The dewar is lowered into a pit six feet deep
into the ground. To avoid the effects of external vibration, a slab of granite with four
commercial vibration isolators at the corners of the slab was used. Another granite slab
2.1. Ultra-low temperature setup: Installation of dilution refrigerator 47

Gas Handling Rack for


System Pulley Instruments

Insert

Dewar

Figure 2.2: Schematic of the infrastructure designed and built for dilution refrigerator.

was put on the vibration isolators on top of which were two layers of parallel rectangular
strips of rubber in crossed geometry. Finally at the top was a granite slab on which the
dewar rests. An appropriate framework consisting of a pulley and a supporting angle was
designed and machined in the workshop to lower and raise the fridge into and out of the
main dewar. A turbo pump circulates the mixture and in the process gets heated up to
high temperatures. A chiller/pump continuously circulates water around it to prevent it
from getting heated up. A schematic of the supporting framework is shown in Fig. 2.2.

2.1.2 Wiring for electrical measurement

The very first thing that one needs to do is to be able to measure temperatures of the
order of few mK accurately. The thermometer used for this purpose is carbon speer
resistor whose resistance is calibrated for temperature. The resistance is measured using
an AC resistance bridge (Lakeshore 371 AC resistance bridge). One needs to be careful of
grounding issues while measuring such low temperatures and after optimizing the setup
the lowest temperature (T ) achieved was ∼ 10mK (Fig. 2.3).
The coldest part of the fridge is the mixing chamber (T ∼ 10mK), upto which all the
wires are terminated (Fig. 2.4a). But the magnet center is about 40cm below the mixing
48 Chapter 2. Experimental techniques

0.1

TMC (K)

0.01
0 30 60
Time (mins)

Figure 2.3: Temperature vs time plot for a test run of the dilution refrigerator.

chamber. A cold finger was designed for doing experiments in a magnetic field (Fig. 2.4b).
To minimize the heat load and ensure proper thermal anchoring the cold finger was made
of high purity silver, which was annealed to enhance the thermal conductivity. The details
of the annealing procedure have been outlined in Appendix A. The lower part of the cold
finger can hold multiple spring loaded sample holders (charntek). Thus multiple samples
can be mounted simultaneously, in parallel field as well as in perpendicular field. The
wiring from the mixing chamber to the sample holder was done using high conductivity
Copper wires which are thermally anchored to the cold finger using GE varnish (shown in
Fig. 2.4c). Proper thermal anchoring of the wire is very critical since at low temperatures
cooling of the sample occurs only through the wires. Thus the wires should be at the
temperature of the mixing chamber. A picture of cold finger along with the sample
holder after complete wiring is shown in Fig. 2.4d. The electrical wiring of the cold finger
was done in collaboration with Ms. Vidya Kochat and Mr. M. A. Aamir from Department
of Physics, IISc, Bangalore.
At low temperatures (∼ few mK), the electron temperature can be significantly higher
than the temperature indicated by the sensor, which is the lattice temperature. Since the
phonons are frozen at such low temperatures, they cannot cool the electrons. There is no
direct thermal link between the sample and the mixing chamber apart from the Cu wires
that are used for electrical measurements. The heat from the sample is dissipated only
2.1. Ultra-low temperature setup: Installation of dilution refrigerator 49

(a) (b) (c)

Mixing
Mixing chamber
Chamber
Central hole
Central hole for attaching
forthe coldattaching
finger

the cold finger

(d)

Figure 2.4: (a) A photograph of the lower portion of the insert showing the mixing
chamber plate with a central slot for mounting the cold finger. (b) Design of the cold
finger. (c) Photograph showing the Ag cold finger with Cu wires anchored to it. (d) A
picture of the cold finger along with the sample holder after the wiring.

through these wires. Additionally, the electrons can be heated up due to high frequency
electromagnetic waves. All these factors can add up to increase the electron temperature
beyond the lattice temperature. Hence after the complete wiring, it is important to know
the electron temperature that can be achieved. We estimate the electron temperature
of our system by measuring the resistance of a low density Si:P δ-layer as a function of
temperature as shown in Fig. 2.5. The resistance decreases continuously with temperature
down to 60 mK after there is a change in slope (the rate at which the resistance was
decreasing with decreasing temperature has reduced) and the resistance tends to saturate.
The saturation can be due to electron heating effects or due to some sample dependent
property. The exact cause is not known, but this gives us an upper bound of the electron
50 Chapter 2. Experimental techniques

50

R (kW)
45

40 60mK

20 100 500
T (mK)

Figure 2.5: Resistance R of a low density Si:P δ-layer as a function of temperature. The
resistance tends to saturate below 60 mK which we consider to be the upper bound for
electron temperature.

temperature. Hence we believe that the electron temperature of the system is ∼ 60 mK at


maximum although it can be lower than that. We understand that this is a crude method
to determine the electron temperature. The more conventional method of determining
the electron temperature using the Coulomb blockade oscillations of a quantum dot or
using shot noise thermometry has not been done for our system.

2.2 Electrical transport measurement techniques


2.2.1 Resistance measurement
An AC four probe technique using a lockin amplifier (LIA) (SR830) is used to measure the
resistance of the device. A lock-in amplifier uses the phase-sensitive detection technique
to measure signal at a particular frequency and phase. Since it rejects signals at other
frequencies and phase, it is extremely useful in measuring low-noise signals. A schematic
of the measurement circuit is shown in Fig. 2.6. An input voltage Vac is applied across
the sample and a current limiting resistor Rc , using the internal oscillator of the lock-
in amplifier. The current limiting resistor should be much larger than the resistance of
the sample so that a constant current flows through the sample. For our experiments
2.2. Electrical transport measurement techniques 51

Lock-in Output Vac Device VXY

Current limiting
resistor, RC Ground
VXX

Figure 2.6: Circuit diagram for four-probe resistance measurements.

Rc was varied from 1 − 100 MΩ depending on the sample resistance. Typically we kept
Rc > 1000R; where R is the sample resistance. This ensured that a constant current flows
through the sample even when the resistance of the sample changes (say as a function of
magnetic field or temperature). The frequency of the applied voltage was generally kept
in the range 11 − 17 Hz (unless otherwise mentioned). The voltage, V across the sample
is measured using the LIA and resistance is determined as R = V /I. Unless otherwise
mentioned, all the measurements reported in thesis have been carried using the four probe
geometry.
For magnetotransport measurements, the resistance of the device (in hall bar geom-
etry) is measured as a function of perpendicular magnetic field (B⊥ ). The longitudinal
component, Vxx (across the longitudinal probes) and transverse component or Hall volt-
age (across the Hall probes), Vxy = VH is measured. From the longitudinal resistance
Rxx = Vxx /I, we calculate the longitudinal resistivity as ρxx = Rxx W/L, (where L and W
are the length and width of the sample respectively) and Hall resistance as ρxx = VH /I.
The longitudinal conductivity σxx was calculated from the relation σxx = ρxx /(ρ2xx + ρ2xy ).

2.2.2 Noise

Noise or fluctuations in electrical resistance are present in almost every system and can
be an important probe for the underlying dynamics of charge transport in the system.
For a constant current through the device, the resistance fluctuations are manifested as
voltage fluctuations across the device. In general there are three types of noise that are
present in any condensed matter system:
52 Chapter 2. Experimental techniques

1. Johnson (or Nyquist) noise, which is due to random motion of charge carriers at a
finite temperature and is independent of any voltage across the device. For a device
of resistance R at a temperature T , the Johnson’s noise has a value 4kB T R.

2. Shot noise, which occurs due to discrete nature of charge carriers.

3. Low-frequency (or 1/f) noise, the power spectral density of which has a form SV ∝
1/f α , where α, the frequency exponent, is in the range 0.8 − 1.2.

This thesis focuses on the study of 1/f noise in Si:P and Ge:P systems and hence we
describe in detail the mathematical formulation and the techniques used to obtain the
noise data from the sample.

2.2.2.1 Noise: Mathematical formulation

A fluctuating quantity, say voltage V , is recorded in the time domain, whose power
spectrum SV (f ) quantifies the noise in the frequency domain. Given that δV (t) is the
fluctuation in voltage, the power spectral density SV (f ) can be represented as the square
of the amplitude of the Fourier transform of the signal,

( ) ∫ T 2
1 −i2πf t
SV (f ) = lim dt δV (t)e , (2.1)
T →∞ 2T −T

where T is the time window over which the fluctuations are recorded. The power spectrum
can also be represented as the fourier transform of the autocorrelation function (Weiner-
Khintchine theorem) [152], C(τ ) which is given by,
( )∫ T
1
C(τ ) = lim dt δV (t + τ )δV (t). (2.2)
T →∞ 2T −T

The spectral shape depends on the time scales associated with the fluctuating variable,
which is determined by the time scales associated with the physical mechanism that causes
the resistance to fluctuate (defect motion, traps, etc.). For example if certain processes
are occurring at a time scale τ , it determines the spectrum around frequency ∼ 1/(2πτ ).
The finite relaxation time associated with the voltage fluctuation causes power spectrum
to be dependent on frequency. If φ(t) is the relaxation function, the power spectrum can
be written as:
2.2. Electrical transport measurement techniques 53

(a) (b)
10 0.1 Hz
0.2 Hz

1/f
1 1 Hz
1

S(f)
3 Hz
S(f)

-1
t = 1 Hz
0.1 10 Hz

0.1 0.01
0.01 0.1 1 10 0.01 0.1 1 10
f (Hz) f (Hz)

Figure 2.7: (a) Lorentzian power spectrum of a single fluctuator with τ −1 = 1 Hz. (b)
A 1/f noise power spectrum obtained from a superposition of Lorentzians. Here five
Lorentzians with corner frequencies in the range 0.1 ≤ τ −1 ≤ 10 Hz have been added up.

∫ ∞
S(f, T ) = C dte−i2πf t φ(t) (2.3)
−∞

For resistance fluctuations caused due to a single characteristic relaxation time scale τ ,
the relaxation function is the Debye relaxation function, φ(t) ∝ e−t/τ for which the power
spectrum is a Lorentzian given by


S(f ) ∝ (2.4)
1 + (2πf τ )2
Figure 2.7a shows a sketch of S(f ) as a function of f for a single fluctuator with a
relaxation time τ −1 = 1 Hz. At the low frequency end (f ≪ τ −1 ), S(f ) is constant
whereas at the high frequency end (f ≫ τ −1 ), S(f ) ∝ 1/f 2 .
Instead of a single relaxation time, when there is a distribution of relaxation times
described by F (τ ), the power spectrum becomes:

∫ ∞

S(f ) ∝ dτ F (τ ) (2.5)
0 1 + (2πf τ )2
In a general description, a distribution of the type F (τ ) ∝ τ −β will give rise to power
spectrum S(f ) ∝ f −2+β [85]. Thus, a 1/f -type S(f ) can be obtained by adding up few
54 Chapter 2. Experimental techniques

well-spaced Lorentzians. Figure 2.7b shows how a 1/f spectrum can be obtained by
adding up five Lorentzians with 0.1 ≤ τ −1 ≤ 10 Hz.
The spectral shape of SV (f ) gives an idea of the time scales associated with the defect
motion, whereas the magnitude of the power spectral density is governed by the coupling
of electronic parameters (say Fermi energy, etc.) to the defect dynamics. Furthermore, the
dependence of noise with other external fields, like temperature, magnetic field (parallel
and perpendicular), current, etc. can give vital information about the origin of noise,
phase transitions, etc.

2.2.2.2 Noise: Measurement Scheme

Unambiguous, accurate and reproducible measurement of low frequency noise can be quite
tricky and challenging as there are many extraneous sources of noise like bad electrical
contacts (of the sample), electronic circuitry involved in measuring the signal, measuring
instruments and amplifiers, electromagnetic radiation, ground loops, temperature insta-
bility, etc. This acquires more significance when one is dealing with samples which have
extremely low noise levels as it can be easily masked by other external sources of noise.
Measuring instruments and amplifiers can also give rise to spectrum having a 1/f char-
acter. Hence it is of utmost importance that one eliminates all the external sources of
noise in order to obtain clean spectrum arising from the sample itself. In this section we
describe in detail the various steps involved in measuring the noise from the sample. The
process can be divided into two parts: (1) Acquiring the conductance fluctuations from
the sample over a period of time. This is related to the hardware, and requires removal of
background noise coming from amplifiers, electromagnetic interference, temperature in-
stability, grounding issues, etc. and (2) The second part is to obtain the power spectrum

SV(f)
Power spectral
Digital Signal density
Device Processing
∆V (t) f

Figure 2.8: Flowchart of the basic technique used to obtain the power spectrum for any
device.
2.2. Electrical transport measurement techniques 55

-20
10
-18
10 -21
10

SV (V /Hz)
-22
10

2
0.1 1
-19
10

-20
10
0.1 1 10
Frequency (Hz)

Figure 2.9: A typical 1/f spectrum (filled black circles) with a flat background (filled red
squares). The inset depicts the sensitivity of the system to measure 1/f noise much below
the background level. Reproduced from Ref. [153].

from the time series. This is the software aspect of the problem and involves processes
like digitization, digital filtering, fourier transforms, etc.
Fig.2.8 depicts a flowchart of the basic process used to obtain the noise from the
sample. A constant current is passed through the sample using a low noise current
source and the fluctuations in voltage ∆V (t) = V0 (t) − ⟨V ⟩, (where ⟨V ⟩ is the average
voltage) is measured. Instead of this, a constant bias can be applied across the sample and
fluctuations in current can be recorded. For low noise samples, capturing the resistance
fluctuations involves detection of voltages of the order of 0.5 − 5 nV. After the time series
data is acquired, digital signal processing (DSP) techniques (described in this chapter)
are employed to get the power spectrum. The DSP techniques that we use can measure
spectral noise levels lower than SV (f ) ≤ 10−20 V2 Hz−1 till frequencies down to 1 mHz as
shown in Fig. 2.9.

2.2.2.3 Noise: Commonly used circuits

The circuit used for noise measurement depends on the type of device being measured, its
resistance, the noise magnitude, etc. An ac technique put forward by Scofield is commonly
employed for noise measurements [154]. For noise measurements in Si:P and Ge:P δ-layers
56 Chapter 2. Experimental techniques

we use an ac four probe balanced bridge circuit which gives the most optimal data. Below
we discuss the two other common circuits and the reasons why they are not suitable for
us. Then we discuss the circuit used for noise measurements in this work.
The circuit shown in Fig. 2.6 is one of the simple circuits which can be used for noise
measurements. This is just a four probe resistance measurement using lock-in technique.
The sample is biased by a constant current using the sine out of a lock-in amplifier which
is also used to measure the in-phase and out of phase component of resistance. The
resistance can be recorded as a function of time which can be analyzed to get the power
spectrum. However this is not the suitable circuit for low noise samples. For low noise
samples, to have measurable noise we need to have large bias across the sample which
means reducing the sensitivity of the lock-in amplifier. Hence this circuit was not used in
the present case.
The five probe balanced bridge circuit in Fig. 2.10 eliminates the problem of degrading
sensitivity of the lock-in amplifier at high bias since in this case one measures directly
the fluctuation in resistance and hence one can use lock-in amplifier at high sensitivity.
There are five contacts in the sample. The sample forms two arms of a Wheatstone bridge

Lock-in Output Vac

R1 R2

Device

Ground

Ultra-low noise
transformer
preamplifier

To LIA

Figure 2.10: AC five probe balanced bridge circuit for noise measurement.
2.2. Electrical transport measurement techniques 57

with the center contact being grounded. Thus each part of the sample on either side of
the ground contact forms an arm of the bridge. One has to ensure that the resistance of
both the parts of the sample is roughly the same. However this is not a pure four probe
measurement and the contact resistance is not eliminated. Hence this circuit is suitable
for samples which have very low contact resistance. The samples studied in this work
have high contact resistance (2 − 5 kΩ) and hence doing a four probe measurement is
essential.
The two problems discussed previously are eliminated in four probe balanced bridge
circuit (Fig. 2.11). In this circuit the sample and a current limiting resistor (R1 in
Fig. 2.11) form the two arms of the bridge while a decade box (R3) and a wire wound
resistor (R2) make up the other two arms. R1 and R2 were chosen to be wire wound
resistors which have extremely low current noise. However they do have the background
Johnson’s noise. The voltage drop across the sample is amplified by a low noise voltage
preamplifier (SR 560) and the output of the amplifier is balanced against the voltage across
the standard wire wound resistor R2. The voltage across R2 was changed by varying the

Double walled
Faraday cage

R1
C Sample R3
R2

Low noise
preamplifier
SR 560

LIA A LIA B

Figure 2.11: AC four probe balanced bridge circuit for noise measurement.
58 Chapter 2. Experimental techniques

resistance R3 of the decade box. A capacitor C, can be put in parallel to the current
limiting resistor to balance the off phase component. This being a four probe measurement
eliminates the contact resistance and the balanced condition enables the use of lock-in
amplifier at high sensitivity and one can directly record the resistance fluctuations from
the sample over a period of time. The entire circuit including the voltage amplifier is kept
inside a double walled Faraday cage to avoid disturbances due to external electromagnetic
radiation. Unless otherwise mentioned, this circuit has been used to measure noise for all
the devices discussed in this thesis.

2.2.2.4 Noise figure of the amplifier

The measurement of resistance fluctuations or noise often involves detecting voltage fluc-
tuations of the order of few nV or even less. In such cases one often uses amplifiers to
enhance the signal as well as the fluctuations. We have seen in the previous section that
an amplifier was used in the five probe balanced bridge technique as well as the four
probe balanced bridge technique. The choice of amplifier depends on input impedance of
the device that is being measured. Using an appropriate amplifier is important to obtain
clean noise signals from the sample.
Amplifiers themselves have their own noise which is reflected in the measurements.
The output noise of an amplifier depends on the measurement frequency and impedance
connected at its input. A measure of the noise added by the amplifier over and above the
sample’s Johnson noise (given by 4kB T R) as a function of frequency and input impedence
is represented by the Noise Figure (NF) of the amplifier and is given by

N F (f, R)(dB) = 10 log1 0Sv0 (f )/4kB T R (2.6)

The noise figure of transformer preamplifier SR554 and voltage amplifier SR560 is
shown in Fig. 2.12a and 2.12b respectively. The eye of the amplifier is the central region
in the NF where the noise added by the amplifier is minimum. The SR554 is best suited
for samples with low resistances of the order of few ohms to few tens of ohms. The SR560
amplifier on the other hand is best suited for samples having high resistances from few
tens of kΩ to M Ω. Since the resistance of the 2D Si:P and Ge:P δ-layers range from few
kΩ to few tens of kΩ, we have used the SR560 amplifier for noise measurements. The
2.2. Electrical transport measurement techniques 59

(a) (b)

Figure 2.12: Noise figure of (a) Transformer preamplifier SR554 (b) Voltage amplifier
SR560.

measurement frequency used was in the range 200 − 300 Hz which lies in the eye of the
amplifier’s NF.

2.2.2.5 Noise: Data acquisition and digital processing

After obtaining the voltage fluctuations using one of the above circuits, the essential
processes are digitization, decimation and finally fast fourier transform to calculate the
power spectral density. Data is acquired from the LIA using a 16-bit digitizer (Gage
Compuscope 1600) which stores the data in its own memory and then transfers it to the
hard disk of the computer once the acquisition is complete.

2.2.2.6 Aliasing effect and its removal: Decimation

Whenever a continuous signal is digitized at some sampling rate, a common problem is


aliasing which occurs if the sampling is not proper [155, 156]. Figure 2.13 shows some
examples of proper and improper sampling. An analog signal is said to be sampled prop-
erly if one can reconstruct the entire analog signal from the samples. In other words, the
sampled data set contains all the information about the analog signal. In Fig. 2.13a we
have a DC signal (cosine wave of zero frequency) which had been sampled (red dots) at
380 samples/sec. The data is said to be sampled properly because the analog signal can
be reconstructed by joining the sampled points with straight lines. In Fig. 2.13b we have
60 Chapter 2. Experimental techniques

(a) DC signal - Analog frequency = zero (b) Analog frequency = 0.038 of sampling rate

1 1

0 0

-1 -1
Amplitude

0.00 0.01 0.02 0.03 0.00 0.01 0.02 0.03

(c) Analog frequency = 0.19 of sampling rate (d) Analog frequency = 0.95 of sampling rate

1 1

0 0

-1 -1

0.00 0.01 0.02 0.03 0.00 0.05 0.10

Time (s)

Figure 2.13: (a) A DC signal (black line) which is sampled (red points) at 380 samples/sec.
(b) A sine wave of frequency 190 Hz sampled at 5000 samples/sec. (c) A sine wave of
frequency 190 Hz sampled at 1000 samples/sec. (d) A sine wave of frequency 190 Hz
sampled at 200 samples/sec. In this case, sampling is not proper as the sampled points
seem to represent a wave of frequency 10 Hz. Redrawn from Ref [155].

a sampled an analog signal, a sine wave whose frequency is 0.038 of the sampling rate. In
this case though its not as trivial as joining the sampled points with straight lines, the
sampling is said to be proper because no other sine wave or combination of sine waves will
produce the given pattern of samples (within some reasonable constraints). In Fig. 2.13c,
the frequency is 0.19 of the sampling rate and it doesn’t appear that the sampled data
follows the original analog signal. Still the sampling is proper as the information required
to construct the analog signal is present in the sampled data. However, when the analog
frequency is 0.95 of the sampling rate ( 2.13d), then we see that the sampled data rep-
resents a sine wave of frequency different from that of the original analog signal. This
phenomenon in which a sampled data represents a wave of different frequency (from that
of original signal) is called as aliasing. Such a sampling is said to be improper. We now
define the sampling theorem (or Shanon sampling theorem or Nyquist sampling theorem)
which says that a continuous data is properly sampled if the sampling rate is atleast twice
2.2. Electrical transport measurement techniques 61

Spectral band of interest fs = 6Hz

f
-3 -2 0 1 3 4 6 7
-fs/2 fs/2 fs

Figure 2.14: Schematic of spectral relationship showing aliasing. For a sampling rate of
6 Hz, the out of band frequencies 4 Hz and 7 Hz get aliased as −2 Hz and 1 Hz. Redrawn
from Ref [157].

the original frequency.


If a sinewave of frequency f0 is sampled at rate fs , then the sampled values of sinewave
of frequency f0 and a sinewave of frequency f0 ± kfs are indistinguishable [157]. To
illustrate this lets consider a sampling rate of fs = 6 Hz and limit the spectral band
of interest to ±fs /2 = 3 Hz. From Fig. 2.14, we can see that within ±fs /2 there is
energy at −2 Hz and 1 Hz which is aliased from 4 Hz and 7 Hz respectively. It should
be mentioned that the vertical position of the dots (which indicates the amplitude) does
not have any significance, rather their horizontal position indicate which frequencies are
coupled through aliasing. This problem is solved by using low-pass anti-aliasing filters
which attenuate unwanted signals beyond the spectral band of interest.
Another major problem arises due to interference from the power line (for which
f = 50Hz). For a typical bandwidth of (say) 10 Hz, the effective sampling rate according
to sampling theorem should be ≥ 20 Hz. Since this is close to the line frequency of 50Hz,
there can be a large contribution at this frequency which gets aliased in the bandwidth of
measurement. The maximum possible roll-off at the low pass filter output of SR830 lock-
in amplifier is 24 dB/octave which is not sufficient to significantly remove the spurious
contribution due to aliasing. Hence we use digital anti-aliasing filters having much higher
roll-offs (∼ 100 dB/octave). The process is called decimation in the time domain.
62 Chapter 2. Experimental techniques

Figure 2.15: A typical response function of the three stage digital filters that were used.
The dashed line shows the amplitude response function of the low pass filter at the output
of lock-in-amplifier with a roll-off of 24 dB/octave. The solid lines represent the response
functions of the intermediate digital anti-aliasing filters. All the digital filters have better
stop band attenuation than that attained by the lock-in-amplifier filter [158].

The data sampling rate depends on the experimental bandwidth which is determined
by the time constant of the LIA. For a time constant τLIA , the bandwidth, w = 1/2πτLIA .
Initially the data is sampled from the LIA at a rate, fs′ which is many times larger (say
64) than fs (fs ≥ 2w for proper sampling). The roll-off is kept at the maximum value of
24 dB/octave. This removes any component beyond fs /2, with an attenuation of 105 at
fs′ /2, which is the effective bandwidth before decimation. This is then passed through a
digital anti-aliasing filter with a cut-off frequency fs /2 and a very high roll-off. Digital
filtering is beneficial because the maximum roll of LIA is 24 dB/octave whereas the roll-off
for the digital filters which we used were as high as 80 − 100 dB/octave. The process
effectively removes all stray signals beyond fs′ /2. The digital filters used were of the FIR
(Finite Impulse Response) type with a Kaiser window design. The next step is to down-
sample the data by storing say every 64th point to bring down the sample rate to fs ,
which should be atleast 2w to avoid aliasing. This process is called “decimation” and it
2.2. Electrical transport measurement techniques 63

is performed in three stages to optimize for memory usage and speed. For a decimation
factor of 64, we do it in stages of 8, 4 and 2 consecutively. The roll-off the digital filters
were maximum at the third stage (Fig. 2.15). In this way we have a signal upto frequency
fs /2 without any effects of aliasing.

2.2.2.7 Noise: Estimation of power spectral density

The digitized and decimated data was processed to obtain the power spectral density by
a discrete Fourier transform technique known as the “method of averaged periodogram”
developed by Peter D. Welch [159]. In this method the number of computation steps and
computer memory usage are reduced, thus requiring a shorter execution time. The essen-
tial steps of the Welch method are listed below and is shown schematically in Fig. 2.16.
Let, x(j); j = 0, 1, ..., N − 1 be a sample from a stationary, second-order stochastic
sequence. Let SV (f ), |f | ≤ fmax , be the power spectral density of x(j). The data is
divided in segments of length L and the starting point of these segments are D units
apart. Then,

x1 (j) = x(j) j = 0, 1, ........, L − 1 (2.7)


x2 (j) = x(j + D) j = 0, 1, ........, L − 1 (2.8)
xK (j) = x(j + (K − 1)D) j = 0, 1, ........, L − 1 (2.9)

where K is the total number of segments. Thus we get, N = L+(K −1)D. The schematic
of overlapping data segmentation is shown in Fig. 2.16. A modified periodogram for each
segment xK (j) is calculated by the method described below. xK (j) is multiplied with the
Hanning window function w(j) to form the sequences x1 (j)w(j), x2 (j)w(j), ..., xK (j)w(j).
The Hanning function is defined as [160]

1 cos 2πj
w(j) = [1 − ] for j = 0, 1.....L − 1 (2.10)
2 N
Discrete Fourier transforms A1 (n), A2 (n), ..., AK (n) of the K sequences x1 , x2 , ..., xK and
64 Chapter 2. Experimental techniques

Figure 2.16: Schematic of Welch periodogram method used to estimate the power spectral
density of voltage fluctuations from the decimated time series.

for n = 0, 1, 2, ..., L/2 is obtained using the expression:

1∑
L−1
Ak (n) = xk (j)w(j)e−2kijn/L (2.11)
L j=0


where i = −1 and w(j) is the Hanning window function described above. The modified
periodograms are obtained as,

L
Pk (fn ) = |Ak (n)|2 for k = 1, 2......K (2.12)
U
∑L−1
where fn = n(2fmax /L), for n = 0, 1, ..., L/2 and U = (1/L) j=0 w2 (j) is the normal-
ization factor. The discrete frequency steps in the power spectrum are in multiples of
2fmax /L, where fmax = fs /2 (fs is the final effective sampling rate after decimation) is
the maximum frequency till which the spectrum is obtained. After the data acquisition,
while doing the digital signal processing, the parameters L and D are chosen optimally
to get clean data within the range and in minimum computation time. For example, if
L is chosen very large then the resolution of the data will be high, but that reduces the
number of segments K, since N is fixed, and averaging will be poor. Hence one must then
take data for a longer time to get high resolution within the bandwidth. The average of
2.2. Electrical transport measurement techniques 65

these periodograms give the spectral estimate of the decimated time series, i.e.,


K
S(fn ) = Pk (fn ) forf ≤ fmax (2.13)
k=1

In this way we obtain the PSD of the in-phase and quadrature component of δv. δv
contains within it the information about the conductance fluctuations from the sample as
well as the background fluctuations. The PSD of δv is given by

SV (f, δ) ≃ G20 [SV0 (f0 − f ) + (Irms


2
)SR (f ) cos2 δ] (2.14)

where G0 is the product of the gain of the preamplifier and the LIA, SV0 (f ) is the PSD
due to background fluctuations and SR (f ) is the PSD of the resistance fluctuations of the

Lock in Amplifier

X Y

Data acquisition – 16bit Digitizer

DSP and Decimation

Power Spectrum - Welch periodogram

Xpower Ypower

Power Spectral Density = Xpower-Ypower

Figure 2.17: Flowchart showing the process flow during the estimation of power spectral
density from the acquired data.
66 Chapter 2. Experimental techniques

sample, f0 is the excitation frequency and f is the measurement frequency, Irms is the
rms value of the current flowing through the sample and δ is the phase angle of detection
of δv with respect to biasing current. When δ = 0, we have SV (f, 0) = G20 (SV0 + SV (f )),
i.e the in-phase component has contributions from background fluctuations as well as
sample noise. However when δ = π/2, SV (f, π/2) = G20 SV0 i.e the quadrature component
gives only the background noise. On subtracting the PSD of quadrature component from
that of the in-phase component we obtain the noise from the sample. In this way we
get noise from the sample as well as the background fluctuations and this is the greatest
advantage of the ac technique of measuring noise [158, 160]. Fig. 2.17 shows a simple
flowchart depicting the process flow in estimating the power spectral density from the
acquired data. The details can be found in Ref [160].

2.2.2.8 Time domain analysis: Weiner filtering

The previous section focussed on the analysis in the frequency domain to obtain the spec-
tral density of resistance fluctuations. However in many instances it is required to extract
information from the time series data. The time series data was taken for the in-phase
as well as the quadrature component of the voltage fluctuations using a lock-in amplifier.
For situations where the sample noise is much larger than the background noise, only the
in-phase time series data suffices as sample noise. However for low noise samples when
the fluctuations from the sample are comparable to background fluctuations (within the
experimental bandwidth), the in-phase time series data gets corrupted. The in-phase data
contains fluctuations from sample as well as background, while the quadrature component
contains only the background fluctuations. To carry out any time-domain analysis it is
essential to remove the spurious background fluctuations from the in-phase component.
To achieve this we use a Weiner filter, which converts the data into frequency domain
and retains only the fraction of the power spectrum which corresponds to excess noise
coming from the sample. The clean and uncorrupted spectrum is then retransformed
into the time domain signal. Let c(t), u(t) and b(t) be the corrupted, uncorrupted and
background signal in the time domain respectively, with C(f ), U (f ) and B(f ) being their
respective fourier transforms. The filter transfer function in the frequency domain is
defined as
2.2. Electrical transport measurement techniques 67

5
Original data (Sample noise+background)

dV (10-8 V)
-5
Weiner filtered

-10

0 10 20
Time (s)

Figure 2.18: An example of the time series data before and after Weiner filtering. The
traces have been shifted for visual clarity.

|U (f )|2
Φ(f ) = (2.15)
|U (f )|2 + |B(f )|2
This is the fraction of power of the uncorrupted signal in the total signal at each frequency.
Using |C(f )|2 = |U (f )|2 + |B(f )|2 , we get

|C(f )|2 − |B(f )|2


Φ(f ) = (2.16)
|C(f )|2
The above filter transfer function is multiplied by the Fourier transform of the cor-
rupted signal and the resulting signal is inverse Fourier transformed to get the uncorrupted
signal. Here, it is assumed that the uncorrupted and the background signal are uncorre-
lated.
Fig. 2.18 shows an example of a time series data before (black trace) and after Weiner
filtering (red trace) for a low noise sample for which the excess noise from the sample
is comparable to the background fluctuations. Note that the low frequency components,
for which the spectral power of excess noise dominates, are almost unaltered. For the
high frequency components, where the excess noise is suppressed and background noise
dominates, the signal is sharply attenuated.
68 Chapter 2. Experimental techniques

2.3 Device fabrication


This section describes the fabrication of different types of devices investigated in this
thesis. Three systems have been studied: (i) Two dimensional Si:P and Ge:P δ-layers (ii)
One dimensional wires of P in Si and (iii) Zero dimensional Si:P quantum dots. All the
devices that have been investigated in this work have been fabricated by Prof. Michelle
Y. Simmons and her group at the Centre for Quantum Computation and Communication
Technology, University of New South Wales, Sydney, Australia. All the scanning tunneling
microscopy images, that are presented in this thesis, have been taken by Prof. Michelle
Y. Simmons’ group. The fabrication process for the different systems is briefly outlined
below.

2.3.1 Two dimensional δ-doped layers of phosphorous


2.3.1.1 Si:P δ-layer

All the δ-doped Si:P layers studied in this work were fabricated in an ultra-high vacuum
(UHV) variable-temperature scanning tunneling microscope (STM) system with a base
pressure of 5 × 10−11 mbar. The STM-chamber is equipped with a PH3 dosing system
for P doping and a Si sublimation source for homoepitaxial growth. The Samples were
cleaved from a Si(001) wafer with a resistivity of 10 Ωcm. The wafers are cleaned in
standard sulphuric peroxide and RCA-II solvents and loaded in the UHV system. In
UHV, the samples are out-gassed by radiative heating at 350◦ C for 12 hours and by
direct-current heating at ∼ 470◦ C for another 2 hours. A clean 2 × 1-reconstructed
Si(001) surface (Schematic and STM image in Fig. 2.19a) was then prepared by flash-
annealing the samples at ∼ 1150◦ C and cooling the samples slowly (3◦ Cs−1 ) from 900◦ C
to room temperature. The 2 × 1 reconstructed Si(001) surface was then exposed to a
saturation dose of the precursor PH3 gas at room temperature (Fig. 2.19b), resulting in
a surface coverage of 0.37 monolayer (ML) to form the Si:P δ-doped layer. In order to
obtain δ-layers with different doping density the dose time was varied. The P dopants
were incorporated in the surface layer of Si by a thermal anneal treatment at 350◦ C 1 min
(Fig. 2.19c). It has been seen in previous studies that a saturation PH3 dosing leads to
a P dopant density of 0.25 ML after incorporations [161]. The Si:P δ-layer thus formed
2.3. Device fabrication 69

Figure 2.19: Schematic of the fabrication process of two dimensional δ-layers of P in


Si, along with the scanning tunneling microscopy image at various stages. The different
stages of growth shown are (a) A clean Si surface (b) Phosphine dosing (c) Incorporation
of P atoms inside Si matrix and (d) Encapsulation by growing Si on the top.

were encapsulated by growing ∼ 25 nm of Si at 250◦ C (Fig. 2.19d) by molecular beam


epitaxy (MBE). The low temperature of incorporation annealing and subsequent epitaxial
Si growth restricts the dopant segregation to < 2 nm in the growth direction.

For electrical measurements, trench isolated hall bars were formed by electron-beam
lithography (to pattern the hall bar) and reactive ion etching. Ohmic contacts were then
made to the delta layers by thermal evaporation. Three different contact materials have
been used for Si:P δ-layers studied in this thesis: (i) Nickel Silicide (NiSi) (ii) Gold (Au)
and (iii) Aluminium (Al). NiSi and Au contacted δ-layers are identical with respect to
the investigations carried out here. Al, on the other hand becomes superconducting at
T < 1 K and its effect on the transport properties are discussed in Chapter 6. The full
70 Chapter 2. Experimental techniques

details of the fabrication procedure can be found in other references [138, 161, 162].

2.3.1.2 Ge:P δ-layer

The δ-doped Ge:P layers were fabricated in a UHV variable temperature STM-MBE
system with a base pressure < 5 × 10−11 mbar equipped with a Ge evaporator and a PH3
dosing system. The Samples were cleaved from a Sb doped Ge(001) wafer with resistivity
of 1 − 10 Ωcm. To obtain a clean atomically flat Ge(001) surface, a two-step method
of ex-situ chemical passivation and in situ heating procedure was employed. The GeO2
passivation layer was chemically grown via a wet treatment in a HCl:H2 O 36 : 100 bath
followed by H2 O2 :H2 O 7 : 100 bath. The samples were then outgassed at 230◦ C for ∼ 1 hr.
The GeO2 was removed by flash annealing at 760◦ C for 60 secs and then the sample
was slowly cooled at ∼ 2◦ Cs−1 from 600◦ C to room temperature to obtain an ordered
reconstructed surface. The sample was heated to 100◦ C which enhances the reactivity of
the surface and dosed with PH3 at a pressure of 5x10−9 mbar for 10 min. The sample
was heated from 100 to 350◦ C in 5 min and cooled rapidly down to room temperature
to incorporate the P atoms. The Ge:P δ-layer formed was then encapsulated with ∼ 25
nm of intrinsic Ge deposited at 210◦ C a rate of 0.13 Ås−1 by MBE. Figure. 2.20 shows
the STM images at different stages of fabrication of Ge:P δ-doped layers. For electrical
measurements, trench isolated Hall bars were defined by electron-beam lithography and
a CHF3 /CF4 based dry etching. Ohmic contacts to the δ-layer was formed by thermally
evaporating aluminium. The full details of the fabrication procedure can be found in

(a) (b) (c)

Figure 2.20: Scanning tunneling microscopy image at different stages of growth. (a) A
clean atomically flat Ge(001) surface (b) Incorporation of P atoms (c) Incorporation of P
atoms inside Si matrix and (d) After Ge encapsulation.
2.3. Device fabrication 71

(a) (b)

(c) (d)

Figure 2.21: (a) Schematic of the device structure showing the δ-layer(red line) of P atoms
inside Si/Ge. (b) Optical image (false color) of saturation doped Si:P δ-layer. The scale
bar is 100 µm. (c) Optical image (false color) of Si:P δ-layer after the bonding of the
contact pads. The scale bar is 200 µm. Scanning electron micrograph (false color) of a
multi-probe hall bar of a Ge:P δ-layer. The scale bar is 200 µm.

other references [145, 146].


A schematic of a hall bar of 2D Si:P or Ge:P δ-layer is shown in Fig. 2.21a wherein
the δ-layer of P atoms is indicated by a red line. An optical image (false color) of the a
multi-probe saturation doped Si:P δ-layer in hall bar geometry, recorded after deposition
of the ohmic contacts, is shown in Fig. 2.21b. An optical image of a bonded Si:P δ-layer is
shown in Fig. 2.21c. Figure 2.21d shows an scanning electron micrograph of a multi-probe
hall bar of a Ge:P δ-layer.
72 Chapter 2. Experimental techniques

2.3.2 One dimensional Si:P wires

(a) H:Si surface


(b)

STM
(c) Lithography (d)

PH3 dosing and


(e) incorporation
(f)

Figure 2.22: Schematic of the fabrication process of 1D wires along with the corresponding
scanning tunneling microscopy (STM) images. (a) Schematic of H:Si Surface (b) STM
image of H:Si-2×1 surface (c) Schematic of STM lithography to pattern the 1D wire
(d) STM image after hydrogen lithography (e) Schematic of 1D wire formed after P
incorporation (f) STM image after P incorporation. The STM images have been taken
from Ref. [149].

The 1D wires of P in Si have been fabricated using STM lithography. A schematic of


2.3. Device fabrication 73

the various stages of the fabrication process along with the corresponding STM image is
shown in Fig. 2.22. A 2 × 1 reconstructed Si(001) surface at 350◦ C is exposed to atomic
hydrogen for 6 min at pressure ∼ 5 × 10−7 mbar to create a hydrogen passivated H:Si
surface (Schematic and STM image in Fig. 2.22a and b respectively). A STM tip is the
used to pattern the required structure on the hydrogen coated Si surface (Schematic and
STM image in Fig. 2.22c and d respectively). The patterned substrate is then exposed
to a saturation dose of phosphine gas followed by incorporation of phosphorous atom
(Schematic and STM image in Fig. 2.22e and f respectively) using the process described
for the fabrication of δ-layers in Section 2.3.1.1. The detailed fabrication procedure can be
found in Ref [124,125,149]. Note that the same process of STM lithography and δ doping
is used to fabricate other nanostructures like quantum dots of P atoms in Si [126,127,163].
74 Chapter 2. Experimental techniques
Chapter 3

Two dimensional P δ-layers in Si and


Ge: Suppression of low frequency
noise

In this chapter we present the basic characterization of the Si:P and Ge:P δ-layers in
terms of standard time-averaged transport (conductivity) and dynamic transport (noise).
We study the transport properties of Si:P and Ge:P δ-layers using conductance mea-
surements as a function of temperature for devices with varying doping density. The
conductivity decreases logarithmically with decreasing temperature at low temperatures
which indicates weak localization and electron electron interaction. We also study the
low frequency noise in these systems. We find that the devices obey the phenomenolog-
ical Hooge relationship, with a power spectrum that is proportional to 1/f α , where the
frequency exponent α ∼ 1 − 1.2 within the experimental bandwidth over the entire range
of temperature and magnetic field studied here. The noise magnitude of the Si:P δ-layers
was found to be more than six orders of magnitude lower than that of bulk Si:P systems
in the metallic regime and is one of the lowest values reported for doped semiconductors.
The temperature dependence of noise indicates that at low temperatures, the noise in
these devices arise due to mesoscopic fluctuations.

75
76 Chapter 3. Two dimensional P δ-layers in Si and Ge

3.1 Overview and motivation


The proposal by Kane [123] of the possibility of having a solid state quantum computer, in
which the nuclear spin of P atoms in isotopically pure Si can be used as qubits, has led to
extensive research in P-doped Si devices. The major challenge of placing the dopant atoms
precisely in Si has been overcome with the technological advancement of using scanning
tunneling microscopy as a lithographic tool by which single P atoms has been positioned in
Si with an accuracy of less than 1 nm [164]. Combined with molecular beam epitaxy, this
technology has been employed to realize heavily δ-doped planar nanostructures, such as
tunnel gaps [129], nanowires [124, 125] and quantum dots [126–128]. Given the potential
of P-doped devices to shape the future technology, it is important to understand the
transport properties of these systems. The time averaged transport properties of Si:P and
Ge:P δ-doped layers have been studied previously [124, 126, 127, 139, 144], but very little
is known about its long term charge stability, which reflects in low frequency flicker noise
in the electrical transport. Noise in an important tool to probe the local environment of
the dopants in these devices. Though noise has been studied in bulk doped devices, which
show universal conductance fluctuations in the metallic regime [76] and glassy behavior
near the metal-insulator transition [165], noise in δ-layers and nano patterned devices in
Si has not yet been explored. This is paramount to the overall development of devices
with controlled dopant positioning at the nano-scale and in particular, for single dopant
spin based qubits [166].

3.2 Conductivity: Dependence on carrier density


The electrical transport measurements have been carried out using an ac four probe
method described in Chapter 2 section 2.2.1. For resistance measurements, the current was
chosen such that the voltage bias across the sample is less than kB T /e at all temperatures,
to minimize the heating of electrons.
Figure 3.1a shows a schematic of the 2D device structure of a layer of P atoms (red
spheres) sandwiched between Si(or Ge). We have investigated devices grown with different
doping densities of P atoms which leads to different carrier density, n. To determine the
carrier density we perform Hall measurements for all the devices as shown in Fig. 3.2a.
3.2. Conductivity: Dependence on carrier density 77

(a) (b)

Figure 3.1: (a) Schematic showing the 2D device architecture of a layer of P atoms
(red spheres) sandwiched between Si/Ge matrix. (b) Incorporation of P atoms in Si/Ge
tetrahedra. Here rP is the average distance between any two neighboring P atoms, a∗B
is the effective Bohr radius of P atoms in the host lattice, and the ratio rP /a∗B is the
effective dopant separation which determines the extent of wavefunction overlap between
neighboring P atoms.

The maximum density achieved was for saturation doped Si:P devices having n ≈ 2.5 ×
1014 cm−2 while the minimum density measured was n ≈ 3 × 10−13 cm−2 . This range of

density corresponds to rP /a∗B ∼ 0.6 − 2.7 for Si:P and Ge:P devices. Here rP ≈ 2/ πn is
the average distance between neighboring P dopants and a∗B is the effective Bohr radius of
the P dopants in host semiconductor (Fig. 3.1b). The ratio rP /a∗B determines the overlap
between wavefunctions of neighboring P atoms and its effect on transport in the δ-layers
will be discussed in the next chapter.

The Drude conductivity σD , of the δ-layers decreases with decreasing electron density
n, as σD ∝ n3/2 (Fig. 3.2b), indicating the predominance of charged impurity scatter-
ing [167]. Even for the lowest doping, σD was found to be ≫ e2 /h which ensures that
all devices are nominally in the “weakly localized” regime. From the known σD , the mo-
mentum relaxation time τ0 can be calculated using the Drude formula, σD = ne2 τ0 /m∗ ,
78 Chapter 3. Two dimensional P δ-layers in Si and Ge

(a) 60 (b)
T = 4.2K
100
30

sD (e /h)
RXY (W)

2
0

-30
10

-60
-3.0 -1.5 0.0 1.5 3.0 0.5 1 2.5
B^(T) 14 -2
n (10 cm )

Figure 3.2: (a) Hall resistance, RXY , as a function of perpendicular magnetic field. The
solid black lines are linear fits to the data from which the carrier density, n has been
extracted (b) The Drude conductivity, σD , as a function of n. The solid black line shows
that σD ∝ n3/2 . The panel on the right shows the range of rP /aB ∗ and hence the extent
of wavefunction overlap covered in the experiment.

where m∗ is the effective mass of the charge carriers. A short momentum relaxation time
τ0 ∼ 10 − 100 fs indicates strictly diffusive transport in all devices with kB T τ0 /~ ≪ 10−2
over the experimental range of temperature (T ∼ 0.1−4 K). The mean free path, l = vF τ0 ,

where vF = ~kF /m∗ is the Fermi velocity (kF = 2πn is the Fermi wave vector), lies
in the range 5 − 30 nm for these δ-layers. The device nomenclature and the relevant
parameters for the samples that have been investigated in this work can be found in
Table 3.1.

3.2.1 Theoretical estimation of momentum relaxation time, τ0

We have estimated the value of τ0 theoretically assuming that the major contribution
to scattering is due to charged P dopants [167]. We consider the general expression for
scattering times using Born approximation and Fermi’s golden rule as [168],


1 2πnimp d2 k⃗′
= fi (θ)|V (⃗k − k⃗′ )|2 δ(εk − εk′ ), (3.1)
τ0 ~ (2π)2
3.2. Conductivity: Dependence on carrier density 79

Table 3.1: Summary of device properties, where n is carrier density estimated from Hall
measurements at T = 4.2 K, rP is the average distance between any two neighboring P
atoms, rP /a∗B is the effective dopant separation with a∗B being the effective Bohr radius.
a∗B , calculated using the density of states effective mass, was ∼ 0.59 nm for Si and ∼
1.52 nm for Ge. σD is the Drude conductivity, τ0 is the momentum relaxation time
obtained from Drude formula, D is the electron diffusivity, B0 is the field scale associated
with momentum relaxation, l is the mean free path, and W & L are the sample width
and length respectively.

Si Ge
Sample Si HD Si MD Si LD Ge HD Ge MD Ge LD
n(1014 cm−2 ) 2.5 1.1 0.5 1.35 0.46 0.32
rP (nm) 0.74 1.03 1.6 0.97 1.66 2
rP /a∗B 1.2 1.8 2.7 0.6 1.1 1.3
σD (e2 /h) 115 66.7 11.6 77.3 14 9.5
τ0 (f s) 20 24 10 17 9 9
D (ms−1 ) 0.021 0.012 0.002 0.020 0.0037 0.0025
B0 (T ) 0.4 0.56 7.6 0.47 4.8 7.3
W (µm) 2 20 20 20 20 10
L (µm) 1-10 5-100 100 100 100 40
l (nm) 29 24 7 26 8 7

where ⃗k and k⃗′ represent the 2D wavevector of the electron, before and after scattering
respectively. nimp is the impurity density which in the present case is equal to nP , the 2D
dopant density since we have assumed that the charged P dopants are the major source
of scattering. εk = ~2 k 2 /2m is the single particle energy and V (⃗k − k⃗′ ) is the scattering
potential. The angular weighting factor fi (θ), where θ is the scattering angle, incorporates
the effectivity of scattering in the direction of current flow. For transport scattering time,
f0 (θ) = 1 − cos θ, which removes the contribution of small angle scattering. Simplifying
the integration to k and θ, we get
80 Chapter 3. Two dimensional P δ-layers in Si and Ge

∫ ∫ ∞ ( )
1 nP π
′ ′ ~2 k 2 ~2 k ′2
= dθ k dk |V (⃗k − k⃗′ )|2 (1 − cos θ)δ − . (3.2)
τ0 2π~ −π 0 2m 2m

Taking the change in wave vector due to scattering ⃗k − k⃗′ = ⃗q, we get q = 2k sin(θ/2) for
k = k ′ which will be the case due to delta function integration, we get

∫ π
1 mnP
= |V (q)|2 (1 − cos θ)dθ. (3.3)
τ0 π~3 0

Using q = 2k sin(θ/2) the above integration can be written in terms of q as

∫ 2k
1 mnP q 2 dq
= |V (q)|2 √ . (3.4)
τ0 π~3 0 2k 3 1 − (q/2k)2

Since the conduction takes place near the Fermi surface, we can put k = kF , where kF is
the Fermi wave vector. Taking q/2kF = x, we obtain the following expression for τ0 ,


1 4m∗ nP 1
x2
= √ |V (2kF x)|2 dx. (3.5)
τ0 π~3 0 1−x 2

Following Ref. [167] we use

e2 Fi (2kF x)
V (2kF x) = e−2kF x|d| , (3.6)
2ϵr ϵo 2kF x + qT F F (2kF x)

as the screened Coulomb potential for the system where the dopants are at an effective
distance d from the δ-layer. Here ϵr and ϵ are the dielectric constant of the host crystal
and permittivity of free space respectively, m∗ is the effective mass of the charge carriers,
4 2π 2 (1−e−xa/2 )+(xa)2 3(xa)+8π 2 /(xa) 32π 4 (1−e−xa )
Fi (x) = xa 4π 2 +(xa)2
and F (x) = (xa)2 +4π 2
− (xa)2 [(xa)2 +4π 2 ]2 are the form factors
associated with the electron-charged impurity interaction and electron-electron interaction
respectively [167] for the quantum well width a and qT F = g/a∗B is the Thomas-Fermi
wavevector of the system with degeneracy g. Substituting for V (2kF x) in Eq. 3.5, and
3.2. Conductivity: Dependence on carrier density 81

(a) 100 (b) 0.1


SiP - a=1 nm d=0.25 nm Si:P - a=1 nm d=0.25 nm

m (m V s )
-1 -1
t0 (10-14 s)

10

2
0.5
0.01
~n 0.5
n

1
0.01 0.1 1 10 0.2 0.4 0.8 1 2 4
18 -2 18 -2
n (10 m ) n (10 m )

Figure 3.3: (a) Momentum relaxation time, τ0 , as a function of carrier density, n calculated
using Eq. 3.7 for Si:P δ-layer for quantum well width, a = 1 nm, and effective dopant
distance from the centre of the quantum well, d = 0.25 nm. The dotted line shows that
τ0 ∝ n0.5 . (b) The mobility µ as a function of n. The solid olive line is the mobility
estimated using the theoretically calculated τ0 in (a). The filled circles represent the
mobility calculated from the experimentally measured Drude conductivity. The dotted
line shows that µ ∝ n0.5 .

using nP ≈ n (for full dopant activation [140]) we get

∫ 1 [ ]2
1 2~ x2 Fi (2kF x)
= ∗ ∗2 √ dx (3.7)
τ0 m aB 0 1 − x2 x + q0 F (2kF x)

where q0 = qT F /2kF . Substituting for a and d, the above equation was integrated nu-
merically to obtain the absolute magnitude of τ0 . Figure 3.3a shows τ0 as a function of
density n for SiP δ-layer for a = 1 nm and d = 0.25 nm, which are realistic values for
quantum well defined by P dopants in Si or Ge matrix. The calculated n dependence of τ0 ,
within the experimental range goes as, τ0 ∝ n0.5 . From the Drude model σD = neτ0 /m∗,
we expect σD ∝ n3/2 which agrees with the experimental observation (Fig. 3.2b in the
main text). From the calculated τ0 we can obtain the charge impurity scattering limited
mobility using the relation µ = eτ0 /m. Figure 3.3b compares the mobility calculated the-
oretically with the one obtained from experimentally measured σD . Though the function
82 Chapter 3. Two dimensional P δ-layers in Si and Ge

dependence µ ∝ n0.5 holds true for both, the theoretical calculation overestimates the
magnitude of τ0 and hence µ, because other scattering mechanisms which can contribute
to resistance have been ignored.
From Fig. 3.3b it can be seen that the typical mobilities in the Si:P and Ge:P δ-doped
systems fall in the range 50 − 120 cm2 V−1 s−1 , which is much smaller than the mobilities
that have been achieved in Si inversion layers (103 − 104 cm2 V−1 s−1 ) [169,170] or Si/SiGe
modulation doped heterostructures (103 − 105 cm2 V−1 s−1 ) [171, 172]. The reduction in
mobility of Si:P and Ge:P δ-doped layers is due to scattering from the dopants which are
in the plane of the 2D electron layer. For the Si inversion layers and Si/SiGe systems
however, the dopant layer lies away from the 2D electron layer leading to higher mobilities.

3.3 Conductivity: Dependence on temperature


We have carried conductivity measurements as a function of temperature within the range
0.2 − 60 K. The electrical measurements above 4.2 K were carried out in a 4 K dipper
whereas below that were carried out in a 3 He-4 He dilution refrigerator. As a function of
temperature, the conductivity initially increases with decreasing T , confirming metal-like

(a) (b)
Si_HD
112
Ge_HD
0.8
0.96
s (e /h)

s/sD
2

110
0.7

0.92
0.6 Si_LD
108 Ge_LD

0 20 40 60 0.1 1 10
T (K) T (K)

Figure 3.4: (a) Conductivity vs. temperature, T for a highly doped Si:P δ-layer.(b) The
temperature dependence of conductivity, σ (scaled by the conductivity at temperature,
T = 4.2 K) for heavily and lightly doped δ-layers in Si and Ge.
3.4. Conductivity fluctuations: Low Frequency Noise 83

< 12 K (Fig. 3.4a) is associated with


behavior (Fig. 3.4a). The downturn in σ for T ∼
quantum coherence effects. In the quantum coherent regime, electron-electron scattering
is the dominant dephasing mechanism [66,144], and results in a logarithmic T -dependence
of the conductivity as shown in Fig. 3.4b for high and low doped Si:P and Ge:P devices.

3.4 Conductivity fluctuations: Low Frequency Noise

The low frequency noise measurements have been carried out for the first time in these δ-
doped devices. The carrier density n for these devices is ∼ 1013 −1014 cm−2 which is much
larger compared to other two dimensional systems like graphene (n ∼ 1011 − 1012 cm−2 )
or GaAs/AlGaAs based heterostructures (n ∼ 1011 cm−2 ). The dimensions of the devices
that we have measured are in the range 1 − 100 µm (see Table 3.1). This makes the
total number of carriers N = nA, where A is the area between the voltage probes, to
be very large in the δ-layers particularly for the devices with large dimensions. Since
the expected noise power spectrum is inversely proportional to N (Eq. 1.29) [86], it is
expected to be very small for the highly doped δ-layers, making the noise measurements
more challenging. For noise measurements the voltage bias V across the sample was kept
such that V ≪ (kB T /e)L/lϕ , where L and lϕ are the length of the device and the phase
coherence length respectively (lϕ was determined from magnetotransport measurements
as described in the next chapter). This criteria is less stringent than V < kB T /e which
was used for standard transport measurements. Since the power spectrum in these devices
were extremely low we had to use the less stringent criteria (V ≪ (kB T /e)L/lϕ ) for devices
with large dimensions. Since lϕ in these devices were small (∼ 100 − 500 nm), we were
able to perform noise measurements without any appreciable heating of the sample.
We measure the low frequency noise using the technique described in Chapter 2.
Figure 2.11 shows the circuit that has been employed for noise measurements for the 2D
Si:P and Ge:P devices investigated here. The other commonly used circuits namely the
four probe resistance measurement (Fig. 2.6) or the five probe balanced bridge technique
(Fig. 2.10) were not optimal for capturing the conductance fluctuations of the δ-layers.
In the four probe resistance measurement circuit the sensitivity of the lock-in amplifier is
limited by the bias across the sample and low sensitivity does not capture the conductance
84 Chapter 3. Two dimensional P δ-layers in Si and Ge

fluctuations from the sample. Hence this method was not used for the devices measured
here. The five probe balanced bridge does not eliminate the contact resistance of the
device as discussed in Chapter 2 section 2.2.2.3. The contact resistance, resistance of
the voltage arms and the resistance of the area voltage between the voltage probes, all
contribute to the measured voltage in the five probe circuit thus leading to incorrect
estimation of the sample bias. Hence we have used a balanced four probe wheatstone
bridge circuit as detailed in Fig. 2.11 in Chapter 2. This being a four probe measurement
eliminates the contact resistance and the balanced condition enables the use of lock-in
amplifier at high sensitivity. Unless mentioned otherwise, the four probe balanced circuit
of Fig. 2.11 has been used for the noise measurements of the devices in this thesis.

3.4.1 Low frequency noise: 1/f spectrum


Employing the circuit shown in Fig. 2.11, the voltage fluctuations were recorded as a
function of time. A representative decimated time series data obtained by the process
outlined in section 2.2.2 is shown in Fig. 3.5a for Si:P and Ge:P δ-layers at temperature
T = 4.2 K. The power spectral density (PSD), SV , was calculated from the decimated
time series data using digital techniques outlined in Chapter 2. It can be seen that
SV ∝ 1/f α for both Si:P and Ge:P device (Fig. 3.5b), with α in the range 1 − 1.2 over
the full experimental bandwidth. For all the devices, we found SV ∝ V 2 (Fig. 3.6a shows

the bias dependence of ⟨δV 2 ⟩ = SV (f )df for one Si:P and Ge:P device) which ensures
ohmic regime and establishes that measured voltage fluctuations arise due to fluctuations
in the conductivity of the δ layer, i.e. SV /V 2 = Sσ /σ 2 .

3.4.2 Low frequency noise : The Hooge relationship


The frequency and the bias dependence of noise can be combined to normalize the noise
magnitude in terms of the phenomenological Hooge relation, Sσ (f )/σ 2 = γH /nAf α , where
γH , n, and A are the phenomenological Hooge parameter, the areal density of electrons,
and the area of the Hall bar between the voltage probes, respectively. The magnitude of
γH was deduced to be around 3 × 10−5 for the Ge:P δ-layer at T=4.2K. For Si:P δ-layer
γH ∼ 10−6 , which is six orders of magnitude lower than that of bulk doped Si:P systems
degenerately doped to the metallic regime (γH = 0.1 − 2) [173, 174]. Given such a low
3.4. Conductivity fluctuations: Low Frequency Noise 85

a) 4.2K
b)
4.2K
4 -15
10
Ge_HD
10 dV (V)

SV (V Hz )
-16

-1
10
2
1/f

2
-17
-8

10 1.1
1/f
Si_MD
-18
0 10 Si_MD
Ge_HD
-19
10
0 100 200 0.01 0.1 1 10
Time (s) Frequency (Hz)

Figure 3.5: (a) Decimated time series data showing the voltage fluctuations δV as a
function of time at temperature T = 4.2 K and magnetic field B = 0 T. The plots have
been offset vertically for clarity. (b) The power spectral density SV as a function of
frequency for the time series shown in (a). The spectrum have been offset vertically for
visual clarity. The dashed lines show that SV ∝ 1/f .

3
Si_MD 4.2K
Ge_HD
V)
2

2
-17
ádV ñ (10
2

0
0.0 0.2 0.4
-3 2
Bias squared (10 V )

Figure 3.6: The variance ⟨δV 2 ⟩ as a function of bias squared, V 2 for Si MD and Ge HD
at 4.2 K and zero magnetic field. The dashed lines shows linear fits to the data.

value of γH , it is important to establish that we indeed are measuring the noise from the
δ-layer and not from other extraneous sources like amplifier, etc. To achieve this, noise
measurements were performed for different distances between the voltage probes for the
86 Chapter 3. Two dimensional P δ-layers in Si and Ge

a) b)
Si_HD Si_MD
4.2K 10 4.2K
1

2
ds2/s2

ds /s
2
-1
A -1
A

-14
-10

10
10

1
0.1

2 3 4
1 10 10 10 10
2 2
Area (mm ) Area (mm )

Figure 3.7: The normalized variance ⟨δσ 2 ⟩/σ 2 as a function of 1/A, where A is the area
between the voltage probes, for (a) Si HD and (b) Si MD at T = 4.2 K and B = 0 T.
The dashed lines shows that ⟨δσ 2 ⟩/σ 2 ∝ 1/A.

Si:P δ-layer. The result is shown in Fig. 3.7 where ⟨δσ 2 ⟩/σ 2 is plotted against the area (A)

of the δ-layer between the voltage probes for Si HD and Si MD (here, δσ 2 = Sσ (f )df ).
As expected from the Hooge relation ⟨δσ 2 ⟩/σ 2 shows a 1/A dependence, confirming that
the measured resistance fluctuations come from the Si:P δ layer, where different fluctuators
contribute independently to the observed noise magnitude.

3.4.3 Comparison of noise magnitude with bulk doped semicon-


ductors

In Fig. 3.8 we compare the Hooge parameter of our system with the values reported pre-
viously for doped Si. Ghosh et. al. [173] and Kar et. al. [174] have previously measured
highly doped bulk Si:P systems and found fairly large values of γH (0.1 to 2). In compar-
ison to these bulk doped Si:P systems we find that noise in Si:P δ-layers studied in this
work are suppressed by 5 to 6 orders of magnitude. Other references included in Fig. 4 are
for doped thermistors [175], Si MOSFETs [176–179] and piezoresistive cantilevers [180].
The remarkably low value of γH measured here favors the use of Si:P δ-doped devices as
versatile nanoelectronic elements.
3.4. Conductivity fluctuations: Low Frequency Noise 87

Table 3.2: A compilation of Hooge parameter of various 2D systems. FET stands for field
effect transistors, HEMT for high electron mobility transistor and BJT stands for bipolar
junction transistor.

Sample Hooge parameter Remarks Reference


γH
Thin Metal films 10−5 − 10−3 Depends on [181]
temperature
GaAs/AlGaAs 10−5 − 10−3 Depends on [182, 183]
heterostructures temperature, density
GaAs FET 10−8 Short channel FET and [184]
BJT
GaN/AlGaN 10−4 − 10−3 Depends on doping, [185]
heterojunction source-drain bias, gate
FET voltage
AlSb/InAs 5 × 10−4 − ×10−2 Depends on [186–188]
HEMT temperature, mobility,
gate voltage
Single layer 10−4 − 10−2 Depends on [90, 189–191]
graphene temperature
(4.2 − 300 K) and gate
voltage
Multi-layer 10−6 − 10−4 Depends on [90, 190]
graphene temperature
(4.2 − 300 K) and gate
voltage
Suspended 3 × 10−6 − 10−3 Depends on mobility, [90, 192]
graphene temperature
WSe2 0.12 Room Temperature [94]
MoS2 0.005 - 2 Depends on mobility, [95, 193–195]
annealing
Bi2 Se3 0.07 - 2 Depends on mobility [196]
−6 −4
Si:P and Ge:P 10 − 10 Depends on density [197, 198], This work
δ-layer
88 Chapter 3. Two dimensional P δ-layers in Si and Ge

1
10

Piezeresistive cantilevers (T = 300K)


Si MOSFETs (T = 300K)

Si MOSFETs (T = 300K)
Hooge Parameter, gH

Si MOSFETs (T = 297K)

Ge:P d-layer (T = 4.2K)


-1
10

Bulk doped Si:P (T = 4.2K)

Bulk doped Si:P (T = 150K)

Si:P d-layer (T = 4.2K)


Doped Thermistor (T < 0.5K)
-3
10

Si MOSFETs (T = 4.2K)
-5
10

-7
10
1

This Work

This Work
Figure 3.8: Comparison of various reported values of Hooge parameter, γH for doped
silicon. Since the noise magnitude can strongly depend on temperature (T), the values of
T for each reference have been explicitly stated. The x-axis numbers 1 − 8 correspond to
Ref. [173–180] respectively.

Apart from comparing the noise magnitude of δ-doped Si:P and Ge:P layers with bulk
doped semiconductors, it is also important to equate it with other 2D systems and see
where it stands in terms of noisiness. Table 3.2 contains a compilation of Hooge pa-
rameter of various 2D systems which include high mobility systems like GaAs/AlGaAs
and GaN/AlGaN heterostructures, graphene and low mobility materials like transition
metal dichalcogenides (MoS2 , WSe2 ) and topological insulator Bi2 Se3 . Though the mo-
bility of Si:P and Ge:P δ-layers is much lower compared to systems like graphene and
GaAs/AlGaAs heterostructures, they appear to be amongst the quietest in terms of low
frequency flicker noise.

3.4.4 Disorder which leads to low frequency noise

As has been stated in the section 3.2, the charged P dopants scatter the conduction
electrons and limit the mobility of these systems. However, these charged P dopants are
unlikely to be the source of low frequency noise because the activation energy of these
3.4. Conductivity fluctuations: Low Frequency Noise 89

dopants are greater than 600 K and all noise measurements in these devices are carried
out in the temperature range of 20 mK to 4.2 K (apart from the temperature dependence
of noise in Si HD which has been carrier out till 25 K, which is still ≪ 600 K). Hence at
such low temperatures we rule out the charged P dopants as being the source of noise. The
dangling bonds on the surface of the epitaxial Si are known to degrade the performance
of many Si devices. They are unlikely to be a source of noise because as soon as the
devices are taken out of the STM-MBE chamber a native oxide is formed which reduces
the number of dangling bonds drastically. However the oxide formed on the top surface
leads to charge traps at the Si/SiO2 interface. The charged traps in the SiO2 have been
shown to be responsible for noise in mesoscopic devices like graphene [90]. We believe
that charge traps due to formation of native SiO2 are the major source of noise in these
δ-layers. The importance of charge traps and the way in which it can cause noise will be
dealt with in the next chapter where we carry out noise measurements in one-dimensional
Si:P wires and zero dimensional quantum dots.

3.4.5 Hooge parameter for different densities

We have carried out noise measurements of Si:P and Ge:P δ-layers with different doping
densities and estimated the corresponding Hooge parameter. Figure 3.9a shows that as the
density increases the Hooge parameter decreases obeying the function form γH ∝ 1/n1.5 .
It is evident that all the points do not fall on the 1/n1.5 line and it appears that they are
scattered. A closer inspection however reveals that γH is dependent strongly on mobility
rather than density. As the mobility increases, γH goes down obeying the power law
relation γH ∝ 1/µ3 as shown in Fig. 3.9b. It can also be seen from Fig. 3.9b that the Si:P
and Ge:P δ-layers do not fall on the same 1/µ3 curve, rather the points lie on parallel
curves separated vertically.
The particular dependence γH ∝ 1/µ3 has been reported previously in graphene on
Si-SiO2 substrate devices [192]. It is known that noise noise in substrated graphene arises
due to oxide charge traps via the mobility fluctuation mechanism [90]. In this mechanism
a change in configuration of charged traps or their occupation number leads to a change in
scattering crosssection of charge carriers and a fluctuation in its mobility which directly
affects the resistivity or conductivity [90]. As we have discussed, the charge traps are
90 Chapter 3. Two dimensional P δ-layers in Si and Ge

(a) (b)
Hooge Parameter (gH)

4.2 K

-4
10
-4
10
-1.5
n m-3
-5
10
-5
Si:P 10
GeP Si:P
GeP

0.25 0.5 0.75 1 2.5 50 100 150


18 -2 2 -1 -1
n (10 m ) Mobility (cm V s )

Hooge Parameter ( )
Figure 3.9: The Hooge parameter γH as a function of carrier density n for Si:P and Ge:P
δ-layers. The dotted line shows that γH ∝ 1/n1.5 . (b) γH as a function of mobility µ for
Si:P and Ge:P δ-layers. The dotted line shows that γH ∝ 1/µ3 .

the likely source of noise in Si:P and Ge:P δ-layers. The fact that γH ∝ 1/µ3 for the
δ-layers, suggests that similar mechanism of mobility fluctuations might give rise the low
frequency noise in our devices. Another well know mechanism by which the charge traps
cause noise is the number fluctuation or the McWhorter’s model in which there is direct
exchange of charge carriers among the traps and the conduction channel. A distinctive
feature of such a model is that γH ∝ 1/n2 (Eq. 1.30) where as we observe γH ∝ 1/n1.5 .
Hence we rule out the contribution of number fluctuations to noise magnitude. Moreover,
as can be seen from Eq. 1.30, the McWhorter’s model predicts that noise will go down
on decreasing temperature (since the number of traps states in the energy range kB T
decreases). For the Si:P δ-layers however, the noise increases with decreasing temperature,
as will be elaborated in the next section, which is opposite to that of McWhorter’s model.
The model of noise due to mobility fluctuations is consistent this observation as at low
temperatures quantum interference is dominant and the coupling of any defect motion to
noise is much more, i.e. a small change in defect configuration leads to a large fluctuation
in conductance.
It is surprising that these charge traps give rise large noise magnitude in single layer
3.4. Conductivity fluctuations: Low Frequency Noise 91

graphene (γH ∼ 10−3 as seen in Table 3.2) whereas in case of Si:P and Ge:P δ-layers, γH
is in the range 10−6 − 10−4 . One marked difference between the two systems is the fact
that graphene flakes sit directly on SiO2 substrate and hence is physically closer to the
oxide traps. Si:P and Ge:P δ-layers on the other hand are buried 25 nm deep in the Si
substrate, and traps are a result nascent SiO2 which forms as the sample is taken out of
the STM-MBE chamber. The fact that the charge carriers are much further away from
the traps in case of δ-layers, might weaken the impact of the traps leading to extremely
low noise magnitude.

3.4.6 Dependence of noise magnitude on temperature

Noise measurements were performed as a function of temperature from T = 25 K down to


0.3 K for device Si HD. The PSDs at all temperatures show a 1/f dependence (Fig. 3.10a).
By integrating the PSD over the experimental bandwidth we obtain the normalized rela-

tive variance, NG = ⟨δσ 2 ⟩/⟨σ⟩2 = SG /G2 df and plot it as a function of temperature in
Fig. 3.10b. The dependence of noise on temperature is non-monotonous and can be sepa-
rated into two distinct regimes, (1) above 15 K, where the noise increases with increasing
temperature - the noise in this region is due to thermally activated mobile defects and
(2) below 15 K, the noise increases with decreasing temperature and is associated with
quantum interference effects analogous to the upturn in resistance at low temperatures
(Fig. 3.4a). The noise in this region is due to universal conductance fluctuations (UCF)
which increases at lower temperatures.
We concentrate on the temperature dependence of noise in the low T regime where
phase coherent transport occurs and derive the observed dependence theoretically. The
power spectral density of conductance fluctuations (which we measure) is given by

( )∫ T
1
SG (f ) = lim dt⟨δG(t)δG(0)⟩e−i2πf t (3.8)
T →∞ 2T −T

Assuming that each scatterer has a relaxation time τ with a Debye-Lorentzian spectrum
we obtain for SG as [100, 160]

∫ ∞
2 2τ
SG (f ) = (δG) dτ F (τ ) (3.9)
0 1 + (2πf τ )2
92 Chapter 3. Two dimensional P δ-layers in Si and Ge

a) 10 -7
Si_HD 0.3K b) Si_HD
0.75K
2.2K
4.2K

2
7K

ds /s
SG/ G (Hz )

10.7K
-1

-9

2
10 13K 10 1.1
17K T

-10
2

21K

10
24K
-11
10

1
-13
10
0.1 1 10 1 10
Frequency (Hz) T (K)

Figure 3.10: (Color online) (a) The power spectral density (PSD) for Si HD at different
temperatures T ranging from 0.3 K to 24 K. (b) The normalized variance ⟨δσ 2 ⟩/⟨σ⟩2 as
a fucntion of T . The dotted line shows that noise ∼ 1/T in the low T regime.

where F (τ ) is the distribution of relaxation time. Integrating SG (f ) within the exper-


imental bandwidth gives the variance of conductance fluctuations which is plotted in
Fig. 3.10b. In order to obtain the temperature dependence of (δG)2 in the regime of
phase coherent transport, we use the Feng-Lee-Stone (FLS) theory [103] of UCF noise.
According to the FLS theory, the change in conductance caused due to the motion of
a single scatterer by a distance δr within a phase coherent box is given by,
( )2 ( ) ( )2−d
e2 V L
(δG1 ) ≈
2
α(kF δr) (3.10)
h Ni ld l
where kF , V , Ni and l are the Fermi wavevector, system volume, the total number of
sin2 x/2
impurities and the mean free path respectively. The function α(x) = 1 − x2 /4
is the
phase change of the electron due to scattering off the moving impurity at a distance δr.
The factor (L/l)2−d (where L is the size of the system along one of the dimensions and d
is the dimensionality of the system) accounts for the many visits of a Feynman path to a
given site. For a 2D system (as is the case for Si:P and Ge:P δ doped layer) in the x − y
plane (say) the above expression can be simplified as,
( )2
e2 1
(δG1 ) ≈2
α(kF δr) (3.11)
h kF l
where we have used, l = (ζNi /V )−1 , with ζ being the scattering cross-section. In case
3.4. Conductivity fluctuations: Low Frequency Noise 93

of many scatterers moving within the phase coherent box, the conductance fluctuation
due to each scatterer simply add to give the total conductance fluctuations, (δGbox )2 =
(δG1 )2 ns (T )Ωbox , where Ωbox = lϕ2 is the volume of the single phase coherent box and
ns (T ) is the density of the active scatterers. Both lϕ and ns (T ) are temperature dependent
quantities. Substituting for (δG1 )2 using Eq. 3.11 we get,

( )2
e2 1
(δGbox ) ≈
2
α(kF δr) ns (T )lϕ2 (3.12)
h kF l

It is important to mention that for our devices kB T < ~/τϕ for the entire range of
measurement and hence Eq. 3.12 is applicable for the Si:P and Ge:P devices at low
temperatures (for kB T > ~/τϕ , the fluctuations in conductance reduce due to energy
averaging and a multiplicative factor of ~τ −1 /kB T has to be included in Eq. 3.12). As
(δGbox )2 from different boxes are independent of each other, the conductance fluctuations
of the entire sample, (δG)2 is related to (δGbox )2 as,

(δG)2 Ωbox (δGbox )2


= (3.13)
G2 V G2box

where Gbox is the conductance of a single phase coherent box. Substituting for (δGbox )2
using Eq. 3.12, we get

( )2
e2 1 Ly
(δG) ≈
2
α(kF δr) ns (T )lϕ4 (3.14)
h kF l L3x

In the above equation the temperature dependent quantities are ns (T ) and lϕ . As will
be shown in the next chapter, lϕ ∝ T −0.5 . Considering the scatterers to be two-level
systems [100, 199] with a tunneling model of low energy excitations [74], we have ns ∝ T .
Combining the temperature dependence of ns and lϕ we get (δG)2 ∝ T −1 as indeed
observed experimentally (Fig. 3.10b). This agreement indicates that in Si:P and Ge:P
δ-layers, the conductivity fluctuations arise from the FLS mechanism [103].
94 Chapter 3. Two dimensional P δ-layers in Si and Ge

3.5 Summary of the results


1. We have carried out first low frequency noise measurements of Si:P and Ge:P δ-
layers.

2. The devices obey the phenomenological Hooge relationship, with a power spectrum
that goes as 1/f α , where the frequency exponent α ∼ 1−1.2 within the experimental
bandwidth over the entire range of temperature and magnetic field studied here.

3. The noise magnitude of degenerately doped Si:P δ-layers quantified by the Hooge pa-
rameter, γH is ∼ 10−6 which is about five to six orders of magnitude lower than that
of bulk Si:P systems in the metallic regime and is one of the lowest values reported
for doped semiconductors. The noise magnitude depends strongly on mobility and
is higher for devices with lower mobility.

4. The temperature dependence of noise indicates that at low temperatures the noise
in these devices arise due to mesoscopic fluctuations.

5. Charge traps due to native SiO2 formation give rise to low frequency noise via the
mobility fluctuation mechanism similar to graphene on Si/SiO2 substrate.
Chapter 4

Conductance noise in atomically


patterned nanostructures in Si

In this chapter we present the measurements of average conductance and fluctuations


in conductance on nanostructures patterned in Si at the atomic scale. Low frequency
noise measurements on one-dimensional (1D) wires and zero-dimensional (0D) quantum
dots of P atoms in Si indicate that population and depopulation of charge traps causes
electric field fluctuations at the devices leading to fluctuations in conductance. The Hooge
parameter of the 1D wires is one of the lowest among the various 1D systems. We also
study the stability of these atomic scale devices as the surface conditions are altered.

4.1 Overview and motivation

Motivated by the requirement of low resistivity interconnects at the atomic scale, ultra-
thin wires consisting of only two dimer rows of P atoms in Si, have been fabricated using
scanning tunneling microscopy (STM) lithography by Weber, et al. [125]. The scaling
theory of localization says that for 1D systems, even with infinitesimally small amount
of disorder, all state will be localized making the system an insulator. However, the 1D
Si:P wires in Ref [125] show metallic conduction with resistivity values close to that of
bulk Si of equivalent doping density as shown by the purple symbols in Fig. 4.1. This has
been explained by extremely high doping density of P atoms leading to all six valleys of
Si being occupied [150]. Also the wires were made such that the length was smaller than

95
96 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

Figure 4.1: Resistivity of the STM patterned wires (blue and purple diamonds), showing
that it remains constant down to atomic limit with the average value near that for the
silicon bulk resistivity at equivalent doping density (blue line). The inset shows the
resistivity as a function of wire diameter in log-log scale. Reproduced from Ref. [125].

the localization length. Apart from 1D wires, 0D quantum dots have also been realized
through STM lihtography. These devices acquire significance in view of the monolithic
fabrication procedure without involving any electrostatic gating to form the active device.
Given the extreme sensitivity of these atomic scale devices to any disorder or charge traps,
it is of paramount importance to explore the conductance fluctuations in these devices.
Since these devices have immense potential for a quantum computation architecture, it is
crucial to explore the noise magnitude in these nanostructures and compare it with other
systems.

4.2 One-dimensional wires of P in Si

4.2.1 Gate voltage characteristics of the 1D wires

Figure 4.2a and 4.2b show the scanning tunneling microscopy images of the two wires
W1 (diameter = 4.5 nm) and W2 (diameter = 1.5 nm) respectively which have been
used for low frequency noise measurements. The length of wires W1 and W2 are 46 nm
and 49 nm respectively, making the wires almost identical apart from their width. To
4.2. One-dimensional wires of P in Si 97

(a) (b)

Figure 4.2: Scanning tunneling microscope image of (a) Wire W1 of diameter 4.5 nm (b)
Wire W2 of diameter 1.5 nm. Reproduced from Ref. [149].

enable four probe measurements, both the wires have four contacts (source, drain and two
voltage probes). Two side gates coupled capacitively to the wires tune the conductance
of the wires. The conductance as a function of gate voltage at temperatures 4.2 K and
0.1 K is shown in Fig. 4.3a and 4.3b for wire W1 and W2 respectively. Wire W1 with
G > e2 /h is metallic and exhibits universal conductance fluctuations. Wire W2 undergoes
a transition to an insulating state with G ∼ < e2 /h due to localization of charge carriers
and shows Coulomb blockade oscillations as a function of gate voltage as has been shown
earlier [150].

(a) (b)
6 W1
4.2K 1.2 W2 4.2K
0.1K
0.1K

0.8
G (e /h)

G (e /h)

4
2

0.4

2
0.0

-1 0 1 -0.6 0.0 0.6


Vg(V) Vg(V)

Figure 4.3: The conductance, G, as a function of gate voltage, Vg at temperatures 4.2 K


and 0.1 K for (a) Wire W1 and (b) Wire W2.
98 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

4.2.2 Conductance fluctuations: Low frequency noise measure-


ments

4.2.2.1 Time series and power spectral density

The low frequency noise measurements have been carried out in both the wires using the
method described in Chapter 2 Section 2.2.2. The time series of conductance fluctuations
of wires W1 and W2 at different gate voltages are shown in Fig. 4.4a and 4.4b respec-
tively and corresponding power spectral density (PSD) is shown in Fig. 4.4c and 4.4d
respectively. For W1 (diameter 4.5 nm) the conductance fluctuations are recorded at gate
voltages −0.7, −0.1 and 0.8 V. Towards negative gate voltages the time series shows jumps

(a) (c)
-0.7V -0.7V
-3 -6 -0.1V
1.0x10 10
0.8V
SG/G (Hz )
-1

2
dG/G

1/f
-0.1V
2

0.0 10
-8

-3 -10 1/f
-1.0x10 10
0.8V
400 800 0.01 0.1 1 10
(b) (d) Frequency (Hz)
0.02
-0.43V
2 0V
10
-5 1/f 0.41V
SG/G (Hz )
dG/G

-1

0.01 10
-7
2

-9
10
1/f
0.00
-11
10
400 600 800 0.01 0.1 1 10
Time (s) Frequency (Hz)

Figure 4.4: The time-series of conductance fluctuations at different gate voltages for (a)
Wire W1 and (b) Wire W2 at temperature 4.2 K. The power spectral density, SG /G2
obtained from the time series for (c) Wire W1 and (d) Wire W2.
4.2. One-dimensional wires of P in Si 99

in conductance which leads to the deviation of spectrum from 1/f (dark green squares
in Fig. 4.4c). For Vg = 0.8 V, the time series appears to be random without jumps or
switches and the corresponding spectrum is 1/f in nature as shown by orange triangles
in Fig. 4.4c. We observe a similar scenario for the thinner wire W2 where the the time
series shows random telegraphic signals (RTS) for extreme negative gate voltages (blue
lines at Vg = −0.43 V in Fig. 4.4b) where the Coulomb blockade is most prominent. The
corresponding power spectrum is a Lorentzian having a 1/f 2 component at f > 1 Hz and
being nearly frequency independent at lower frequencies (blue squares at Vg = −0.43 V
in Fig. 4.4d) implying the existence of one or few active traps. Note that the presence
of large 2D regions at both the ends of the wires (Fig. 4.2) smears out the telegraphic
signals as these regions have 1/f noise arising out of many traps. At extreme positive
gate voltage (say Vg = 0.41 V) the time series is random and has a 1/f spectrum. The
magnitude of conductance fluctuations is larger at negative gate voltages as the conduc-
tance is much lower there and any small disturbance of disorder leads to a large change
in conductance.

4.2.2.2 Comparison of noise magnitude with other 1D systems

From the power spectral density, the noise magnitude can be calculated in terms of the di-
mensionless phenomenological parameter called the Hooge parameter γH using the Hooge
relation, γH = SG N f α /G2 (Eq. 1.29). Since the PSD of the wires and the dots are not 1/f
in nature the Hooge parameter is frequency dependant. We obtain a γH ∼ 10−5 − 10−6
(at f ≈ 1 Hz) for W1 and γH ∼ 10−3 − 10−6 for W2 for different gate voltages at 4.2 K. A
Hooge parameter of ∼ 10−6 is one of the lowest values reported for any one-dimensional
system including carbon nanotubes [91, 200–205] as shown in the bar graph in Fig. 4.5.
Some of the other systems included in Fig. 4.5 are Si nanowire networks [92], InAs [206],
ZnO [207], Ag [93] and Ni [208] nanowires.
It is to be noted that γH is more strongly dependent on gate voltage for W2. For
the metallic wire W1 for which G > e2 /h the variation in γH is about one order of
magnitude whereas for W2, γH varies by about three orders of magnitude. The weak
dependence of noise magnitude on gate voltage for W1 is due to stronger screening since
W1 is metallic with G > e2 /h. For both the wires the lowest value of γH occurs towards
100 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

3
10
Ni

1
10
Hooge Parameter (gH) CNTs

-1
Ag
10

Si nanowire networks
InAs ZnO
-3 SiP-W2
10

-5 SiP-W1
10

-7
10

Figure 4.5: Comparison of noise magnitude in terms of Hooge parameter γH for vari-
ous one-dimensional (1D) systems. The Si:P 1D wires have the lowest noise among 1D
systems.

the positive gate voltage. In Chapter 3, we reported extremely low noise in degenerately
doped two dimensional δ-layers of Si:P. On reducing the dimensionality from two to one
it is expected that the devices would be more susceptible to disorder and changes in
disorder configuration, particularly when the devices are of atomic scale like the thinner
wire which is just two dimer rows wide. It is therefore surprising that the Hooge parameter
of the wires are of the same order of magnitude as the degenerately doped δ-layers. This
might imply that a common mechanism governs the conductivity fluctuations in embedded
structures of phosphorous in Si.

4.2.2.3 Gate Voltage dependence: Noise due to charge traps

4.2.2.3.1 Random telegraphic signals In this section we explore the random tele-
graphic signals observed in the time series of conductance fluctuations for wire W2 at
4.2 K towards negative gate voltages. We consider a Coulomb blockade peak (first peak
in the Coulomb blockade trace in Fig. 4.6e) and plot the time series of conductance at
three gate voltages: (i) Vg = −0.55 V which is to the left of a Coulomb blockade peak (ii)
Vg = −0.49 V which is at the Coulomb blockade peak and (iii) Vg = −0.47 V which is to
4.2. One-dimensional wires of P in Si 101

(a) Vg=-0.55V
(d)
0.1326 W2
2
G (e /h) 4.2K (i)

0.1323

G
0.1320 3
240 260 280
2
Time (s)
1 4
(b) W2 Vg=-0.49V
0.4368 4.2K
∆Vg ∆Vg
G (e /h)
2

0.4364
Vg(V)
0.4360
Charge Trap (filled)
380 390 400 Charge Trap (empty)
(ii) SiO2
Time (s)
(c) W2
0.369 4.2K
Vg=-0.47V
-
G (e /h)
2

0.368

0.367

240 260 280 300 320 Si substrate 1D Wire


Time (s)

(e) (f)
0.006 3.0x10
-7
RTS_Amp (e /h)

W2 Vg W2
0.6 -0.55 V 4.2K
4.2K
2

f SV / V (Hz )
-1

-0.5
G (e /h)

-7
2.0x10 -0.49
0.003
2

0.4 -0.47
2

-0.45
-7 -0.41
1.0x10
0.2
0.000
0.0
-0.55 -0.50 -0.45 -0.40 -0.35 -0.30 0.01 0.1 1 10
f (Hz)
Vg(V)

Figure 4.6: The time series of conductance fluctuation at 4.2 K for wire W2 at gate
voltages (a) −0.55 V (b) −0.49 V and (c) −0.47 V (d) Schematic of noise due to population
and depopulation of charge traps which gives rise to a fluctuating potential on the wires
(e) The amplitude of random telegraphic signals as a function of gate voltage for wire W2
at 4.2 K. For comparison, the conductance vs gate voltage is plotted on the right axis.
(f) f SV /V 2 as a function of frequency at 4.2 K for wire W2 at different gate voltages.
102 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

the right of a Coulomb blockade peak, in Fig. 4.6a, b and c respectively. The time series
at Vg = −0.55 V shows two or three step processes with the low conductance state being
the preferred one (Fig. 4.6a) while at Vg = −0.47 V the high conductance state is the
preferred one (Fig. 4.6c). At Vg = −0.49 V, where there is a Coulomb blockade peak, the
time series is random without any two or three step processes (Fig. 4.6b). An analysis of
the RTS amplitude as a function of gate voltage shows that the RTS amplitude oscillates
as a function of gate voltage with a period which is half of the period of Coulomb block-
ade oscillations. The data appears a bit noisy but the oscillations and the period will be
clear in a later figure where we plot the integrated variance as a function of gate voltage
(Fig. 4.7).

We explain the above observation of RTS and its characteristics by considering that
these conductance fluctuations arise due to population and depopulation of charge traps.
A schematic of the process is shown in Fig. 4.6d. Consider any Coulomb blockade peak
(shown by the red trace Fig. 4.6d(i)) and the time series that is acquired when the device
is at point 1. A charge trap in the vicinity of the wire gets populated (filled charge
trap in Fig. 4.6d(ii)) and it imparts a potential to the wire such that the effective gate
voltage on the wire changes to point 2 which is a higher conductance state. Population
and depopulation of traps with a particular relaxation time leads to RTS signals with
the lower lower conductance state being the more stable one. When the time series is
acquired at point 3 population of the trap shifts the device to point 4 which is a lower
conductance state. In this case, the higher conductance state is the preferred one. This
mechanism explains the observed RTS and the preferred conductance state as well as the
period of oscillations.

We now obtain the PSD of the time series using the method outlined in Chapter 2.
Figure 4.6f shows f SV /V 2 as a function of frequency at various gate voltages for wire
W2 at 4.2 K. For a Lorentzian power spectrum (where the time series shows RTS) we
expect a peak in f SV /V 2 as seen in Fig. 4.6f for Vg = −0.55, −0.5, −0.47& − 0.45 V. For
Vg = −0.49 V the time was random without any RTS (Fig. 4.6b) and hence f SV /V 2 is flat.
The PSD can be integrated within the experimental bandwidth to obtain the normalized
variance ⟨δG2 ⟩/⟨G2 ⟩ which is plotted as a function of gate voltage in Fig. 4.7. It is evident
that ⟨δG2 ⟩/⟨G2 ⟩ ∝ (dG/dVg )2 for both the wires at all temperatures investigated here
4.2. One-dimensional wires of P in Si 103

-6
(a) (b) 10
W1 30 W1 150
4.2K 0.1K

(dG/dVg) (a.u)
(dG/dVg) (a.u)
-8

ádG2ñ/G2
10
ádG2ñ/G2
20 -7 100

2
10

2
10 50
-8
0 10
-9
10 0
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
(c) (d) 800 10-4 500
W2 W2 0.1K W2 0.1K
80

(dG/dVg) (a.u)
-4
4.2K

(dG/dVg) (a.u)
-6 10
10 -5

ádG2ñ/G2
10
ádG2ñ/G2

2
2
10
-5 400 250
40
-8
10 -6 10
-6
10
0
-7
0 0
10
-10 10 -7
10
-0.6 -0.3 0.0 0.3 0.6 -0.5 -0.4 -0.3 0.2 0.3 0.4 0.5
Vg(V) Vg(V)


Figure 4.7: The normalized variance ⟨δG2 ⟩/⟨G2 ⟩ = SG /G2 df as a function of gate
voltage Vg for (a) Wire W1 at 4.2 K and (b) Wire W1 at 0.1 K (c) Wire W2 at 4.2 K and
(d) Wire W1 at 0.1 K.

(Fig. 4.7a and 4.7b for W1 at 4.2 K and 0.1 K respectively and Fig. 4.7c and 4.7d for W2
at 4.2 K and 0.1 K respectively). Alternatively it can be said that the period of oscillations
of conductance fluctuations is half of the period of Coulomb blockade oscillations as seen
in Fig. 4.6e as well and such periodic dependence of conductance fluctuations with gate
voltage arises due to charge traps [209]. As discussed in the previous chapter, the charge
traps at Si/SiO2 interface are active in causing conductance fluctuations in Si:P devices.
The role of Si/SiO2 traps in the context of low frequency noise has been discussed for
other semiconductor systems like MOSFETs [99, 210] and graphene [90].
From the measured PSD, the equivalent gate voltage noise power spectral density SV G
can be calculated as [99, 211]

( )2
SG I
SV G = 2 (4.1)
G gm

where gm = dI/dVg is the transconductance of the device. The SV G for wire W1 and W2
is shown in Fig. 4.8a and 4.8b respectively, at 4.2 K and 0.1 K. It can be seen that SV G
104 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

is not strongly dependent on gate voltage for the two wires at both temperatures. There
are a few jumps or peaks in SV G at some gate voltages which can arise from features in
dI/dVg , but otherwise its nearly flat as a function of Vg .
A well known mechanism by which the charge traps can cause noise is the number
fluctuation model by McWhorter (Chapter 1 and Ref. [98]). He proposed that electrons
tunnel between the oxide traps and conduction channel giving rise to a fluctuation in
the number of carriers in the channel. The expected gate voltage power spectrum for
McWhorter’s model SVMGW is [99],

kB T e2 Nt (EF )
SVMGW = (4.2)
8W LC 2 αt
where Nt (EF ) is the density of trap states per unit volume per unit energy, C is the
capacitance per unit area of the region between the traps and conduction channel. The
value of αt is typically ∼ 1010 m−1 [99]. At 4.2 K the SVMGW estimated from McWhorter’s
model using Eq. 4.2 exceeds the experimental data by over an order of magnitude as shown
by the dashed black lines in Fig. 4.8a and 4.8b for W1 and W2 respectively. Also it is
evident from Eq. 4.2 that SVMGW is directly proportional to temperature as it depends on
the number of trap states within the energy window kB T . Hence SVMGW is reduced at 0.1 K
as shown by the dashed blue line in Fig. 4.8a and 4.8b. However, the SV G calculated from
noise measurements is greater at 0.1 K than 4.2 K which is opposite to the dependence
expected from Eq. 4.2. Hence we conclude that number fluctuations are not responsible
for noise in these devices.
It has been suggested that dominant contribution to noise in graphene comes from
traps imparting a time varying potential to the charge carriers due to changes in their
configuration or occupation number [90]. In order to verify if trapping-detrapping of
charge carriers among one or few traps give rise to noise we estimate the gate voltage
fluctuations ∆VG2 due to one unit of charge fluctuation in a trap. The charge fluctuation
per unit area, ∆n is 1/A2ef f where Aef f is the effective area within which if a trap is
populated or depopulated, it imparts a voltage to the wire. For the wires at distance d
below the surface, any two traps on the surface, with distance between them less than 2d,
will affect the conductivity of the wire and hence the noise by exchanging a charge among
them. Hence Aef f is the area of the circle with radius d vertically above the wire. This
4.2. One-dimensional wires of P in Si 105

4.2 K; 0 T Mc Whorter model - 4.2 K W1


4.2 K; 12 T Mc Whorter model - 0.1 K
-4 0.1 K; 0.1 T Fluctuating potential due to traps
10
4.2 K; Thermal Cycle
SVG (V )
2

-6
10

-8
10

-0.8 -0.4 0.0 0.4 0.8


Vg(V)

4.2 K Mc Whorter model - 4.2 K W2


0.1 K Mc Whorter model - 0.1 K
-5 Fluctuating potential due to traps
10
SVG (V )
2

-7
10

-9
10

-0.6 -0.3 0.0 0.3 0.6


Vg(V)

Figure 4.8: The power spectral density of gate voltage fluctuations, SV G for (a) Wire
W1 at 4.2 K and 0.1 K and (b) W2 at 4.2 K and 0.1 K calculated using the method
described in text. The dashed black and dashed blue line represent the expected SVMGW
from McWhorter’s model (Eq. 4.2) at 4.2 K and 0.1 K respectively. The dashed olive line
is the expected SV G estimated from Eq. 4.4.

essentially means that any population/depopulation of traps within that region affects
the conductance fluctuations of the wires. This causes an electric field fluctuation,

e 1
∆E = ∆n (4.3)
ϵ 0 ϵ r qT F d
where ϵ0 and ϵr are the vacuum permittivity and dielectric constant of Si respectively.
The factor qT F d arises due to screening, where qT F is the Thomas-Fermi wavevector given
106 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

by qT F = g/a∗B with g being the degeneracy and a∗B , the effective Bohr radius. Using
g = 2 × 6 = 12 where the factor 6 is due to valley multiplicity [150] and substituting the
values of a∗B , ϵ0 , ϵr , Aef f and d = 25 nm as the thickness of the top epitaxial Si, we get
∆E ∼ 5 × 103 Vm−1 . The equivalent gate voltage fluctuations is

∆VG = ∆Ed (4.4)

Using d = 25 nm, we get ⟨∆VG2 ⟩ ∼ 10−8 V2 . This is shown by the dashed olive line in
Fig. 4.8a and 4.8 for W1 and W2 respectively. At 4.2 K the SV G estimated from Eq. 4.4
is of the same order of magnitude as has been obtained experimentally. Note that SV G
calculated by assuming one unit charge fluctuation among the traps is independent of
temperature. The SV G obtained experimentally however is larger at lower temperatures
particularly for W2 (Fig. 4.8b). At lower temperatures, the UCF in wire W1 and Coulomb
blockade in W2 are enhanced and the sensitivity of the wires to changes in configuration
of charge traps is much more than at higher temperatures. Hence the noise magnitude
and corresponding gate voltage fluctuations are larger at lower temperature.
In order to further explore the feasibility of the noise due to population and depop-
ulation of charge traps, we consider the quantum tunneling of charges in and out of the
trap given by,

τT = τ0,T exp(2αT DT ) (4.5)

τT is the characteristic time associated with charge trapping event, DT is the tunneling

distance and αT = (2m∗ ϕB /~2 ), with ϕB being the barrier height. Taking ϕB ∼ 0.1 eV
(conduction band height of Si since the traps are close to the Si/SiO2 as the native
oxide would be few tens of angstrom in thickness), τT = 1 s (from peak in f SV /V 2
in Fig. 4.6f) and τ0,T ∼ 10−12 s (typical phonon frequency in Si is ∼ 1012 Hz), we get
DT ∼ 15 nm. From DT , we calculate the approximate trap density in Si/SiO2 as ntr ∼
1/πDT 2 ∼ 1011 cm−2 which agrees within an order of magnitude with the known values
in literature [212–214]. We have excluded the possibility of thermally activated RTS as at
−1 −1
4.2 K, the activation energy EA calculated using the relation τth = τ0,T exp(−EA /kB T )
is ∼ 0.01 eV which is much less than that expected for the traps in Si/SiO2 .
4.3. Quantum Dots in Si:P 107

4.3 Quantum Dots in Si:P


In order to further study the effect of charge traps on the stability of Si nanostructures,
we study conductance fluctuations in Si:P quantum dots. Figure 4.9 shows the schematic
of the two types of quantum dot that have been measured here: (i) Dot D1 which has a
source, drain, and a side gate (Fig. 4.9a) and (ii) Dot D2 which has, apart from source,
drain and side gate, a SiO2 dielectric on the top of epitaxial Si with a top gate to addi-
tionally tune the energy levels of the dot (Fig. 4.9b). An STM image of dot D1 is shown
Fig. 4.10.

(a) (b)
Source
Dot
Drain

Plunger Gate SiO2

Figure 4.9: (a) Schematic of quantum dot device D1 showing the source, drain and plunger
gate (b) Schematic of quantum dot device D2 which has a SiO2 dielectric and an additional
top gate.

Figure 4.10: Scanning tunneling microscopy image of a quantum dot device D1. Image
reproduced from Ref. [163].
108 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

(a) (b)
D1 0.0004 D1
0.0010 4.2 K 0.2 K
G (e /h)

0.0002
2

0.0005

0.0000 0.0000
-0.6 -0.3 0.0 0.3 0.6 -0.5 0.0 0.5
VPG (V) VPG (V)
(c) (d) 0.06
D2
D2
0.09 4.2K
0.2K
G (e /h)

0.03
2

0.06

0.03
0.00
-4 -3 -2 -1 0 -4 -3 -2 -1 0
VTG (V) VTG(V)

Figure 4.11: Gate voltage characteristics of quantum dot D1 at (a) 4.2 K and (b) 0.2 K.
Top gate characteristics of quantum dot D2 at (c) 4.2 K and (d) 0.2 K.

4.3.1 Gate voltage characteristics of the quantum dots

The gate voltage characteristics of the quantum dots are shown in Fig. 4.11. Device D1
shows Coulomb blockade peaks at 4.2 K and 0.2 K (Fig. 4.11a and 4.11b respectively)
with the conductance going to zero in the blockaded regime. For device D2 we show
the conductance as a function of top gate voltage at 4.2 K and 0.2 K in Fig. 4.11c
and 4.11d respectively. At 4.2 K the conductance of the dot does not go to zero between
the Coulomb blockade peaks, however at lower temperature the conductance goes to zero
in the Coulomb blockade regime at negative gate voltages. A complete analysis of the
Coulomb blockade in these devices can be found in Ref. [163].

4.3.2 Conductance fluctuations in quantum dots


4.3.2.1 Time series and power spectral density

Figure 4.12a and 4.12e shows the time traces of conductance at different gate voltages
(side gate voltage for D1 and top gate voltage for D2) for D1 and D2 respectively. While
4.3. Quantum Dots in Si:P 109

D1 shows one or two random switches, D2 shows well defined RTSs. As discussed earlier
device D2 has a layer of SiO2 on the top Si surface and the coupling of the top gate to

(a) (e) Vg=-3.555V Vg=-3.9745V


Vg=-0.3365V 0.09 Vg=-3.7515V Vg=-4.083V
0.0008 Vg=-0.3065V Vg=-4.171V
Vg=0.30765V

G (e /h)
G (e /h)

Vg=0.40335V
0.06

2
2

0.0004

0.03
0.0000
0 1 2 3 0 3 6 9
Time (hrs) Time (hrs)
(b) (f)
Vg=-0.3365V Vg=-3.555V
0.1 0.01
SG/G (Hz )
-1

SG/G (Hz )
2

-1
2
1/f 1E-3 1/f
2

0.01 1/f 2 1E-4 1/f


1E-5
1E-3 0.01 0.1 1E-4 1E-3 0.01 0.1
(c) 10
Vg=0.30765V
(g)
1 Vg=-3.9745V
SG/G (Hz )
-1

SG/G (Hz )

1
-1

2
2 1/f
1/f 0.01
2

0.1
2

0.01 1E-4 1/f


1/f
1E-3
1E-3 0.01 0.1 1E-4 1E-3 0.01 0.1
(d) (h) 100
Vg=-4.171V
Vg=0.40335V
0.1
SG/G (Hz )

SG/G (Hz )
-1

-1

2 1 2
1/f 1/f
2

0.01
0.01
1/f 1/f

1E-3 1E-4
1E-3 0.01 0.1 1E-4 1E-3 0.01 0.1
Frequency (Hz) Frequency (Hz)

Figure 4.12: (a) Time series of conductance fluctuations at 4.2 K for different gate voltages
for Si:P quantum dot D1. The power spectral density SG /G2 for device D1 at gate
voltages (b) −0.3365 V (c) 0.30765 V and (d) 0.40335 V. (e) Time series of conductance
fluctuations at 4.2 K for different top gate voltages for Si:P quantum dot D2. The power
spectral density SG /G2 for device D2 at top gate voltages (f)−3.555 V (g) −3.9745 V (h)
−4.171 V.
110 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

(a) 2
G (e /h)
-3
0.0 -3
1.0x10

2
G (e /h)
(b) 0.06 0.03

Figure 4.13: Stability of the Coulomb blockade oscillations for (a) Si:P quantum dot D1
(a single gate voltage scan is shown on the left) and (b) Si:P quantum dot D2 (a single
gate voltage scan is shown on the left) at 4.2 K.

oxide traps would also affect the conductance fluctuations. The power spectral density
calculated from the time series (of Fig. 4.12a and 4.12e) are shown in Fig. 4.12b, 4.12c
and 4.12d for gate voltages Vg = −0.3365, 0.30765 and 0.40335 V respectively for device D1
and in Fig. 4.12f, 4.12g and 4.12h for gate voltages Vg = −3.555, −3.9745 and −4.171 V
respectively for device D2. Unlike the case for 2D δ-layers, we do not see a pure 1/f
spectrum for the dots, but rather we see Lorentzian spectrum having a 1/f 2 nature for
cases where the time series shows clear RTSs (Fig. 4.12h for device D2). Since a 1/f
spectrum arises from a wide distribution of relaxation times, it is understood that for
4.4. Summary of the results 111

the nanostructures like 1D wires and 0D quantum dots, only one or few traps are active
(with a particular relaxation time). We thus note that as the dimensionality of the active
device reduces, the number of active traps go down from many to a few.

4.3.2.2 Stability in terms of effect of charge traps

In order to compare the stability of the two quantum dots D1 and D2 we measure suc-
cessive Coulomb blockade oscillations as a function of side gate for D1 and top gate for
D2. The gate voltages are swept continuously from negative side towards the positive
side. A 2D map of the conductance as a function of gate voltage and sweep number is
shown Fig. 4.13a and 4.13b for D1 and D2 respectively. A single gate voltage sweep is
shown for each device on the left of respective 2D plots. It is evident that the device D2 is
more unstable showing more number of switches than the device D1. This can arise from
the traps in the oxide layer at the top Si surface and its coupling to the top gate. This
suggests that the transport characteristics of nano and atomic scale devices is sensitive
to the surface trap states.

4.4 Summary of the results


1. Low frequency noise measurements have been performed on low resistivity 1D wires
and 0D quantum dots of P atoms in Si.

2. Random telegraphic noise (two step processes) have been observed in both the wires
and dots which gives a Lorentzian power spectrum ∝ 1/f α where α ≈ 2 at higher
frequencies in contrast to the 2D δ-layers which has a 1/f spectrum. We find that
as the dimensionality of the device decreases from 2 to 1 and 0, the number of active
traps reduces to one or few.

3. Charge traps in the Si/SiO2 interface are responsible for causing noise in these
devices. Population and depopulation of traps give rise to a fluctuating potential
on the wires causing the conductance to fluctuate.

4. The Hooge parameter of the wires is ∼ 10−6 which is lowest when compared to
other 1D systems. Such low noise among these devices are technologically desirable
112 Chapter 4. Conductance noise in atomically patterned nanostructures in Si

as these are important ingredients of a solid state quantum computer in Si.


Chapter 5

Transport and Noise in strongly


interacting 2D δ-layers in Si and Ge

In this chapter we have carried out magnetoconductivity and noise measurements in


Si:P and Ge:P δ-layers of varying doping density ranging from saturation doping (n ∼
2.5 × 1014 cm−2 ) to low density (n ∼ 2 × 1013 cm−2 ). Magnetoconductivity measurements
in a perpendicular magnetic field has been employed to explore weak localization in these
systems as the interaction increases at lower densities. We use the low frequency noise
as a probe to investigate universal conductance fluctuations. The quantum transport
and noise experiments indicate a strong suppression of quantum interference effects at
low doping densities where the on-site effective Coulomb interaction is large. We could
attribute this to a spontaneous breaking of time reversal symmetry which manifests in an
unambiguous suppression of universal conductance fluctuations (UCF) at zero magnetic
field. Magnetoconductivity in an in-plane magnetic field suggests the role of spin fluctu-
ations in breaking time reversal symmetry. The lowest density device shows metallic like
behavior at low temperatures.

5.1 Overview and motivation

Time reversal symmetry is among the most fundamental and robust symmetries of non-
magnetic quantum systems. Its violation often leads to new and exotic phenomena, par-
ticularly in two dimensions (2D), such as the quantized Hall conductance in semiconductor

113
114 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

heterostructures [60], the quantum anomalous Hall effect in topological insulators [215]
or the predicted chiral superconductivity in graphene [216, 217]. The breaking of time
reversal invariance is experimentally achieved either by an external magnetic field or in-
tentional magnetic doping. In 2D, strong Coulomb interactions can lead to delocalization
of electrons such as, during the transition from the non-interacting (Anderson) localized
phase to the Mott insulating regime [14, 109–112, 118, 218, 219]. However, the nature of
such a transition, particularly in the context of spontaneous symmetry breaking remains
unclear.

Bulk phosphorus-doped silicon and germanium have been extensively researched for
metal to insulator transition in three dimensions in the presence of both disorder and
Coulomb interaction [220–222]. However confining the dopants to one or few atomic
planes (δ−layers) of the host semiconductor has recently led to a new class of 2D electron
system [125,128,139], where electron transport occurs within a 2D impurity band. The im-
purity band is ‘half filled’ because each dopant phosphorus atom contributes one electron
to form the band. The Coulomb interaction is estimated as U/γ where U is the Coulomb
energy required to add an additional electron to a dopant site and γ the hopping integral
between adjacent dopants. This in contrast to dilute 2D electron systems in semiconduc-
tor heterostructures where Coulomb interactions are long range in nature [116, 219]. The
tunability of interactions by varying the P doping density makes δ−doped Si and Ge sys-
tems ideal candidates to explore the rich phenomenology of the half-filled Mott-Hubbard
model in 2D. An in-built electron-hole symmetry coupled with Coulomb interactions can
give rise to several phenomena ranging from Mott metal-insulator transition to spin exci-
tations and magnetic ordering [14, 109, 110, 118, 122].

The key advantage of using both Si and Ge as host semiconductors is the factor
of three difference in the Bohr radius, a∗B , which allows us to achieve a wide range of
average effective dopant separation (rP /a∗B ) within the similar range of doping density

(rP ≈ 2/ πn). As shown in the scale bar of Fig. 3.2b (in Chapter 3), rP /a∗B has an overall
range from ≈ 0.6 to 3. This corresponds to a range of γ ∼ 10−20 meV and ∼ 20−50 meV
for the Ge:P and Si:P devices respectively, assuming hydrogenic orbitals [223]. Since
U ∼ 200 meV and ∼ 50 meV for single P donor in Si and Ge, respectively, the effective
on-cite Coulomb interaction U/γ can be ≫ 1, particularly in lightly doped Si devices.
5.2. Magnetoconductivity in transverse magnetic field 115

5.2 Magnetoconductivity in transverse magnetic field


We consider the transverse magnetic field (B⊥ ) dependence of the quantum correction to
conductivity at temperature T as,

∆σ(B) = σ(B⊥ ) − σ(0) = σQI (B⊥ ) + σcl + σee (5.1)

where σQI (B⊥ ) represents the contribution primarily from quantum interference effects,
σcl is the classical correction to the Drude conductivity σD and σee is the correction due
to electron-electron interaction effect. We consider the different correction terms below.

Classical contribution σcl : This is the classical Lorentz contribution to the conduc-
tivity in presence of a perpendicular magnetic field. The classical contribution can be
estimated by consider the conductivity as (ignoring any other corrections for the time
being) [224],

σD
σ= (5.2)
1 + (ωc τ0 )2

where ωc is the cyclotron frequency and τ0 is the momentum relaxation time. For (ωc τ )2 ≪
1, we get

σcl = σ − σD = −σD ωc2 τ02 (5.3)

Using ωc = eB⊥ /m and τ0 = mσD /ne2 with n being the carrier density, from the Drude
model (Chapter 1) we get

−σD3
2
σcl = 2 2 B⊥ (5.4)
ne

It can be seen from Eq. 5.4 the classical contribution is significant only of highly doped
devices at large B⊥ .
116 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

Electron-electron interaction correction σee : It was shown by Altshuler and


Aronov [25] that Coulomb interactions gives rise to a quantum correction to the con-
ductivity in diffusive regime, which in 2D is given by

σee ≃ (e2 /2π 2 ) ln(kB T τ0 /~) (5.5)

Equation 5.5 predicts a logarithmic temperature dependence of conductivity due to electron-


electron interactions as has been observed in many systems including the Si:P and Ge:P
δ-layers [144]. However since we are doing magnetoconductivity measurements at different
temperatures, we need to know the contribution to the magnetoconductivity tensor due
to electron-electron interaction. Houghton et al. [225] considered the electron-electron
interactions at the lowest order and the impurity scattering was treated in a conventional
diagrammatic technique in the limit kF l ≫ 1 where kF and l are the Fermi wave vector
and mean free path respectively. This condition holds true for all devices measured in
this chapter. For the device with lowest density (∼ 2 × 1013 cm−2 ), kF l ∼ 10. In the
diagrammatic calculation, it turns out that in a strong magnetic field the current vertices
have a additional factor 1/[1 + (ωc τ )2 ], where ωc = eB/m is the cyclotron frequency, as
compared to the weak field case. The contribution of these vertices to the transverse
conductivity is zero while for the longitudinal conductivity the correction becomes

∫ 1 ∫ ∞
ie2 (2EF τ03 )2 [2πN (EF )]2 u2 1 dω 1
σee = 2 2 2 3
(5.6)
2π [1 + (ωc τ0 ) ] 2N1 DH τ Ω 2π 0 [−i(ω + Ω) + q 2 ][−iω + q 2 ]

where DH = (EF τ /m)/[1 + (ωc τ0 )2 ] and 1/τ0 = 2πN1 u2 and it turns out that σee acquires
the same form as the zero field field case in Eq. 5.5. Hence when magnetoconductivity
measurement is carried out at a particular temperature and we consider the difference
σ(B⊥ ) − σ(0), the correction due to electron-electron interaction gets cancelled.
It must be mentioned here that even though the longitudinal and transverse component
of σee does not depend explicitly on magnetic field, the quantities like ρee and the Hall
coefficient RH depend on magnetic field via the zeroth order quantities σxx and σxy through
2 2 2 2
the relations ρxx = σxx /(σxx + σxy ) and ρxy = σxy /(σxx + σxy ). In fact the interaction
5.2. Magnetoconductivity in transverse magnetic field 117

induced magnetoresistance is parabolic in nature and is given by [225, 226]

( ) ( )
ρee (B⊥ ) 1 − (ωc τ0 )2 kB T τ 0
≃− ln (5.7)
ρD πkF l ~

Quantum interference correction σQI (B⊥ ) : This term represents the contribu-
tion to the conductivity primarily from effect of quantum interference between the time-
reversed self-intersecting diffusion trajectories, called the ‘Cooperons’, which are the fun-
damental ingredients of electron localization in disordered media.
In order to get the magnetoconductivity we measure the longitudinal resistivity ρxx and
the Hall resistance ρxy of the device using the ac four probe resistance measurement circuit
(refer Chapter 2 Section 2.2.1 and Fig. 2.6). For all the magnetotransport measurements
reported here, the voltage bias across the sample is less than kB T /e at all temperatures,
to minimize the heating of electrons. Also, the measurements have not been performed
while the magnetic field is being ramped. After reaching the desired magnetic field, the
control switch heater was turned off and then the resistance was recorded after a wait
period of ∼ 60 secs. These precautions were taken to ensure that there is absolutely no
effects of heating or any other anomaly due to a sweeping magnetic field, particularly at
millikelvin temperatures . From the measured Rxx and Rxy , we calculate the longitudinal
and transverse resistivity (ρxx and ρxy respectively). The conductivity σxx can then be
obtained using the relation σxx = ρxx /(ρ2xx + ρ2xy ). Using this procedure the conductivity
was obtained for each magnetic field. Since we will be dealing only with σxx for the rest
of the chapter, the subscript xx will be dropped and we shall refer to the longitudinal
conductivity in a perpendicular magnetic field as σ(B⊥ ).
Figure 5.1a shows conductivity as a function of magnetic field (squares) while the
estimated classical contribution is shown on the right axis (solid line calculated using
Eq. 5.4). After eliminating σcl , we obtain σQI as a function of B⊥ as shown in Fig. 5.1b.
Since a perpendicular magnetic field destroys phase coherence by introducing random
phase difference between the time reversed paths, the net correction to the conductivity
from quantum interference can be obtained as σQI (T ) = σ(B⊥ ≫ Bϕ , T ), where Bϕ =
~/4eDτϕ is the phase breaking field scale, and τϕ and D are the dephasing time and
electron diffusivity, respectively.
118 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

(a) (b)
204 0 sQI = s (B) - s (0) - scl
scl
4

sQI (e /ph)
s (B)

scl (e /ph)
200
s (e /ph)

2
-4

2
2

2
196
-8
0
192
0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10
B^(T) B^(T)

Figure 5.1: (a) The conductivity, σ, as a function of perpendicular magnetic field, B⊥ for
Si MD at 0.28 K along with the classical contribution σcl (solid green line) on the right
axis. (b) The quantum correction to the conductivity σQI calculated using Eq. 5.1 for
Si MD at 0.28 K.

5.2.1 Low perpendicular magnetic field

We have carried out magnetoconductivity measurements of Si:P and Ge:P devices with
different doping densities for T ranging from 0.2 K to 4.2 K. All devices show a dip in
magnetoconductivity at B⊥ = 0 T, which is a hallmark of weak localization behavior. In
Fig. 5.2, ∆σ = σ(B) − σ(0) as a function of B⊥ , is shown for one GeP and two SiP devices
(For the range of B⊥ shown in the figure σcl is negligible). ∆σ increases quadratically at
B < Bϕ and logarithmically for B > Bϕ , which is the conventional WL behavior.
The magnetoconductivity data, in the low field range (B ≪ B0 ) was fitted with the
free electron Hikami formula (Eq. 1.27) to extract Bϕ . In the equation, α is a phenomeno-
logical prefactor which is close to 1, while B0 is calculated using the Drude model. Hence
Bϕ is the only free parameter while fitting. Extracting Bϕ from the low-B⊥ fits, the phase
breaking time can be calculated as Bϕ = ~/4eDτϕ . We have checked the consistency of
the τϕ obtained from Hikami fits with other existing models [227]. We find the values
of τϕ to agree within 10% for Hikami and Kawabata model at all densities [161]. Us-
ing the value of τϕ obtained from the fits, the universal weak localization correction to
conductivity for a diffusive 2D conductor with free electrons, σW L can be calculated as
σW L = (e2 /πh) ln (τϕ /τ0 ).
We find that the phase breaking rate, τϕ−1 ∝ T for all devices [shown in Fig. 5.2b for
5.2. Magnetoconductivity in transverse magnetic field 119

(a) 2.1 0.9 1.2


Si_MD 0.28K Si_LD 0.28K Ge_LD 0.28K
1.1 K 1K 1.0 K
Ds (e /ph) 4.2 K 4.2K 4.2 K
1.4 0.6 0.8
2

0.7 0.3 0.4

0.0 0.0 0.0


-0.1 0.0 0.1 -0.1 0.0 0.1 -0.1 0.0 0.1
B^(T)
(b)
Si_MD Si_LD Ge_LD

10 10
tf (10 s )
11 -1

10
-1

1 1
0.1 1 10 0.1 1 10 0.1 1 10
T (K)

Figure 5.2: (a) The low field magnetoconductivity for devices Si MD (left), Si LD (center)
and Ge LD (right) at three temperatures, T. The solid grey lines are Hikami fits to the
data (Eq. 1.27). (b) The phase breaking rate, τϕ−1 , as a function of T for the three devices.
The solid black lines show that τϕ−1 ∝ T .

Si MD (left), Si LD (middle) and Ge LD (right)] consistent with previous studies [144],


implying that dephasing in Si:P and Ge:P δ-layers arises due to electron-electron scatter-
ing.

5.2.2 Reduced quantum correction to conductivity

The magnetoconductivity, ∆σ for the entire range of B⊥ (till 13 T) is shown Fig. 5.3 for
the Si:P and Ge:P devices. The downturn in magnetoconductivity at large B⊥ for the
heavily doped devices is the classical contribution to the Drude conductivity, σcl , discussed
previously.
In Fig 5.4, σQI (after eliminating the classical contribution, σcl ) for three 2D Si:P
δ-layers at 0.28 K is shown. For comparison σW L (where σW L was estimated using the
120 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

(a) (b)
Si_LD 0.28K Ge_LD 0.28K
7 Si_MD Ge_MD
Si_HD 4 Ge_HD
Ds (e /ph)
2

0
2

-7 0
0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10
B^(T)
B^(T)

Figure 5.3: (a) The conductivity correction ∆σ as a function of perpendicular magnetic


field (B⊥ ) for devices Si HD, Si MD and Si LD at 0.28 K. The plots have been offset
vertically for clarity. (b) ∆σ for Ge:P devices with different doping densities (Ge HD,
Ge MD and Ge LD) at 0.28 K.

(a) (b) (c)


8 8
Si_HD sWL sWL 9 Si_LD sWL
Si_MD
DsAQI
DsAQI
sQI (e /ph)

6 DsAQI
2

4 4
sQI
3

0 0 0
0.01 1 0.01 1 0.1 10
B^(T) B^(T) B^(T)

Figure 5.4: The conductivity correction due to quantum interference, σQI (after eliminat-
ing the classical contribution) as a function of perpendicular magnetic field, B⊥ , at 0.28 K
for the Si:P δ-layers. The expected free electron weak localization correction, ∆σW L is
indicated by the dotted line.
5.2. Magnetoconductivity in transverse magnetic field 121

(a) 1.0
(b) 1.0
Si_HD Si_MD
sQI/sWL

0.5 0.28K 0.5


0.75 0.28K
0.66
Fit
Fit

0.0 0.0
0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10

(c) 1.0 (d) 1.0


Si_LD 0.28K Ge_LD
0.5 0.28K
0.72 0.5
1 1
4.2 4.2
sQI/sWL

Fit Fits
0.5 0.5

0.0 0.0
0.01 0.1 1 10 0.01 0.1 1 10
B^(T) B^(T)

Figure 5.5: The quantum correction to conductivity. σQI (scaled by the free electron
weak localization correction, σW L ) as a function of perpendicular magnetic field, B⊥ , at
different T for (a) Si HD, (b) Si MD, (c) Si LD and (d) Ge LD. The solid lines are fits
using Eq. 5.11.

low-B⊥ magnetoconductivity as explained in the previous section) is shown by dotted


lines for all the three devices. Since the magnitude of σQI at B⊥ ≫ Bϕ represents the net
correction to conductivity due to quantum interference, it is evident from Fig. 5.4 that the
contribution of weak localization on transport decreases with decreasing doping density.
For the device Si HD which has the highest doping density (∼ 2 × 1014 cm−2 ), σQI is seen
to approach σW L whereas the low doped Si LD (∼ 5×1013 cm−2 ), the measured correction
is only ∼ 40% of the expected weak localization correction. It is important to note that
122 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

0
0.28K
SiP
GeP

DsAQI (e /ph)
Old Ref
2

-3

-6

0.5 1 2.5 5
14 -2
n (x 10 cm )

Figure 5.6: (a) The anomalous quantum correction, ∆σAQI , as a function of carrier density,
n, for all devices at 0.28 K. The filled square (green) symbols represent data analysed from
Ref [161]. The solid line is a guide to the eye.

a major shift in the dominant dephasing mechanism in the lightly doped samples is ruled
out because we find τϕ to be similar in magnitude in all three devices, and ∝ T down to
T = 0.2 K (Fig. 5.2b). This confirms the predominance of the electron-electron scattering
mediated dephasing which has been reported earlier in such δ-layers [144]. The measured
suppression of quantum interference correction is present at all temperatures as shown in
Fig. 5.5 for four devices (three Si:P device and one Ge:P device).

The deviation of the quantum interference correction from expected free electron
weak localization correction or the extent of “delocalization”, measured by the differ-
ence ∆σAQI = σQI − σW L , increases monotonically with decreasing doping density for all
devices measured in this work and compiled from earlier study [161] as shown in Fig. 5.6
for T = 0.28 K. It is interesting that the measured difference cluster along a common
trace as a function of n irrespective of carrier mobility, disorder or host material (Si or
Ge).

Figure 5.7 shows net the correction to conductivity from quantum interference effect,
σQI (T ) (scaled by σW L ) as a function of Drude conductivity for all the Si:P and Ge:P
devices investigated here and analyzed from previous work. The filled green squares
represent data analyzed from Ref. [161]. It is evident that as the Drude conductivity
5.2. Magnetoconductivity in transverse magnetic field 123

(a) 1.0 (b) 1.0


T = 0.28K T = 0.28K
SiP SiP
GeP GeP
sQI(B>B0)/sWL

sQI(B>B0)/sWL
Old Ref Old Ref

0.5 0.5

0.0 0.0
1 10 100 1 10 100
2 2
sD (e /h) sD (e /h)

Figure 5.7: (a) The anomalous quantum correction, ∆σAQI , as a function of carrier density,
n, for all devices at 0.28 K. The filled square (green) symbols represent data analysed from
Ref [161]. The solid line is a guide to the eye.

decreases, σQI decreases as well. However the precise functional form in which σQI dies
down as a function of σD is ambiguous since the lowest density sample which we could
measure has σD ∼ 5e2 /h. The different scenarios are illustrated in Fig. 5.7a and 5.7b.
σQI drops linearly with σD as shown by solid red line in Fig. 5.7a which on interpolation
suggests that the quantum correction will vanish for σD ∼ 2e2 /h (dotted red line in
Fig. 5.7a). Another possibility is that σQI decreases gradually (in a non-linear fashion)
as shown by the solid red line in Fig. 5.7b which can be interpolated to indicate that
σQI vanishes for σD ∼ 4e2 /h (dashed red line in Fig. 5.7b). Such a behavior might imply
a quantum phase transition occuring in these systems at lower densities. One might
also be inclined to think about the existence of a minimum metallic conductivity. Note
that these are speculations since we have not been able to measure devices with lower
Drude conductivity. It is also likely that none of the above possibilities hold and σQI
drops down to zero as indicated by the solid gray line in Fig. 5.7b when σD reduces to
< e2 /h. However, within the density range (and hence σD range) we have investigated,
there is unambiguous evidence that the quantum correction to conductivity is less than
the expected free electron weak localization correction even when σD ≫ e2 /h.

The reduced quantum correction cannot be due to finite experimental range (≈ 0 −


124 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

14 T) of B⊥ , which exceeds both Bϕ and B0 (= ~/4eDτ0 , the upper cutoff field due to
momentum relaxation) by factors of 1000 and 2 respectively even for the least doped de-
vices at 0.28 K (Table I in Chapter 3). The spin-orbit interaction is known to be small for
P-doped (bulk) Si and Ge [76, 228], and independent of the density of the dopants. Any
long range magnetic order is also unlikely because the Hall resistance was found to vary
linearly with B⊥ at all T (see Fig. 3.2a in Chapter 3) in all our devices [122]. Alterna-
tively, the weak localization correction can be reduced due to scattering of electrons from
local magnetic moments. These moments serve to remove the time reversal symmetry,
suppressing the coherent back-scattering of electrons. The local magnetic moments are
known to occur in three dimensional P-doped Si in the presence of strong Coulomb inter-
actions close to the metal-insulator transition (MIT) [229–231]. In 2D, the possibility of
localized spin excitations at the Mott transition has been suggested theoretically [14,118],
but without any experimental evidence so far. A suppression of quantum correction to
conductivity has been reported in low density electron gases in Si MOSFETs near the ap-
parent MIT [232], but it remains unclear whether it arises due to temperature dependant
screening of disorder or interaction driven spin fluctuations.

5.3 Universal Conductance Fluctuations

Universal conductance fluctuations are aperiodic fluctuations in conductance ∼ e2 /h as


a function of Fermi energy, magnetic field and disorder ensemble. We cannot do UCF
measurements by varying the Fermi energy as it is extremely difficult to gate these devices
due to their high density. We can measure fluctuations in conductance as a function of
magnetic field (called magneto-fingerprinting) and as function of disorder ensemble by
measuring the fluctuations over a period of time.
Fig. 5.8 shows the magneto-conductivity of Si HD at T = 0.2 K for three distances
between the voltage probes, 1, 5.5 and 30 µm. The width of the hall bar is ≈ 2µm. For
smallest dimension of the device, which is of the order of few phase coherence lengths
(phase coherence length, lϕ is ∼ 450 nm for this device at T = 0.2K as determined from
magnetoconductivity measurements), the conductance fluctuations is ∼ e2 /h. As the
sample dimensions increase, the fluctuations decrease due to ensemble averaging and for
5.3. Universal Conductance Fluctuations 125

Si_HD
1.5
T = 0.2K

1.0
Ds (e /h)
2

0.5 1 mm
5.5 mm
30 mm

0.0
-0.4 -0.2 0.0 0.2 0.4
B^(T)

Figure 5.8: Conductivity correction ∆σ for Si HD in perpendicular magnetic field for


different length (between the voltage probes) of the sample Si HD at 0.2 K

the largest dimension, 30µm which is much greater than lϕ , the conductance fluctuations
are ≪ e2 /h
To probe whether the observed suppression of localization correction indeed manifests
a breaking of the time reversal symmetry, we have measured the UCF as a function of
B∥ and B⊥ from slow time-dependent fluctuations in the conductance (G) of the δ-layers
which represents the ensemble fluctuations via the ergodic hypothesis [74–77, 103, 165].
The time dependant conductance fluctuations are analyzed to obtain the power spectral
density, SG (as has been explained in Chapter 2 and Chapter 3), which on integration

over the experimental bandwidth gives the normalized variance, NG = SG /G2 df =
⟨δG2 ⟩/⟨G⟩2 .

5.3.1 Fluctuation in conductance due to the movement of a sin-


gle impurity, δG21
From the normalized variance, ⟨δG2 ⟩/⟨G⟩2 , we can obtain (δG2ϕ ), using Eq. 3.13. Since
⟨δG2 ⟩/⟨G⟩2 is calculated from 1/f noise measurements it represents fluctuations only
126 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

Table 5.1: Conductance fluctuations due to the movement of a single scatterer. n and
ns are the carrier density and the density of active scatterers respectively. ⟨δGϕ 2 ⟩ is
the conductance fluctuations within a phase coherent box and ⟨δG1 2 ⟩ is the conductance
fluctuations due to the movement of a single scatterer.

Si Ge

Sample Si HD Si MD Si LD Ge HD Ge MD Ge LD

n(1014 cm−2 ) 2.5 1.1 0.5 1.35 0.46 0.32

lϕ at 4.2 K (nm) 102 70 36 150 60 43

(⟨δGϕ 2 ⟩)0.5 (e2 /h) 0.017 0.003 0.002 0.006 0.009 0.01

ns (107 cm−2 ) 1.2 0.04 0.78 0.6 32 47

(⟨δG1 2 ⟩)0.5 (e2 /h) 0.48 0.72 0.2 0.17 0.08 0.1

within the experimental bandwidth. The fluctuations in conductance due to the movement
of a single scatterer, ⟨δG1 2 ⟩ can then be obtained as ⟨δGϕ 2 ⟩ = ns ⟨δG1 2 ⟩lϕ2 , where ns is the
density of active scatterers. An estimate of ns within the experimental bandwidth can
be obtained using the Kogan and Nagaev model [101] according to which ⟨δG2 ⟩/⟨G⟩2 =
ln(f2 /f1 )ns (lδc )2 /LW , where f1 and f2 are the lower and upper limits of the experimental
bandwidth respectively within which the noise measurements were performed (≈ 0.015 Hz
and ≈ 7 Hz respectively), l is the mean free path and δc is the average defect cross section

respectively. Using the experimentally measured ⟨δG2 ⟩/⟨G⟩2 and assuming δc ∼ 1/ n we
can get an estimate of ns and hence ⟨δG1 2 ⟩. The results are given in Table 5.3.1. We see
that for the devices (⟨δG1 2 ⟩)0.5 ∼ e2 /h establishing that the observed noise arises from
mesoscopic fluctuations.

5.3.2 Magnetic field dependence of UCF


As a function of B⊥ , the magnitude of UCF is expected to decrease by an exact factor of
two at two field scales, first at B⊥ ∼ Bϕ when the time reversal symmetry, and hence the
5.3. Universal Conductance Fluctuations 127

Cooperon (self-intersecting diffusion trajectories) contribution, is removed [62, 71, 77] and
second at B⊥ ∼ BZ = kB T /gµB due to removal of spin degeneracy [62, 77, 78], where g
and µB are the g-factor and Bohr magneton respectively (see Chapter 1 for details). As
a function of parallel magnetic field B∥ , the UCF magnitude is expected to reduce only
by a factor of two at B∥ ∼ BZ = kB T /gµB due to removal of spin degeneracy [62, 77, 78].
Below we present a detailed magnetic field dependence of UCF noise and show that it
gives unambiguous evidence of time-reversal symmetry breaking.

5.3.2.1 Reduction of UCF in parallel magnetic field

As a function of parallel magnetic field, B∥ , we observe a factor of two reduction in NG at


B∥ ∼ kB T /gµB which happens solely due to the removal of spin degeneracy. The factor
of two reduction at B∥ ∼ BZ (shown by vertical arrows in Fig. 5.9) for T = 0.5 K and
4.2 K along with the T dependence of NG (shown in Chapter 3 Fig. 3.10) establishes that
the 1/f noise in our devices indeed arises from the UCF mechanism. Also the absolute
magnitude of NG in all devices correspond to the change in conductance by ∼ O[e2 /h]
due to a single fluctuator within a phase coherent box (Section 5.3.1) establishing the
observed noise to be from UCF.

Si-MD 4.2K
1.00
NG(B)/NG(B=0)

0.5K

0.75

0.50

0.01 0.1 1 10
B||(T)

Figure 5.9: NG (B)/NG (B = 0) for Si MD as a function of parallel magnetic field, B∥ , at


4.2 K and 0.5 K. The vertical arrows denote BZ .
128 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

5.3.2.2 Low perpendicular magnetic field

The normalized variance, NG as a function of B⊥ at T = 4.2K is shown in Fig. 5.10a-f for


low magnetic fields (Bϕ < B⊥ < BZ , where BZ is the Zeeman field) for all devices. We
see that NG reduces by a factor of 2 at the scale of Bϕ for Ge HD (Fig. 5.10a). However
for Ge MD and Ge LD we find the NG reduces only by a factor of ∼ 1.5 at the scale
of Bϕ . We observe a similar behavior in Si:P devices. For the highest density Si HD
and Si MD, the reduction factor in NG is ∼ 1.3 and 1.1 respectively (Fig. 5.10d and e
respectively) while for Si LD NG is nearly independent of B⊥ . A closer inspection reveals

(a) 10 (b) 14 (c) 12


Ge_HD Ge_MD Ge_LD

12
NG

8 10
NG

NG
-13
-14

-12
10

10
10

10
6 8

8
4 6
-0.2 0.0 0.2 -1 0 1 -1 0 1
B^(T) B^(T) B^(T)
(d) 5.0 (e) 7 (f) 3.0
Si_HD Si_MD Si_LD

4.5 2.5
NG

NG
NG

-14

6
-14
-10

10

10
10

4.0 2.0

3.5 5 1.5
-0.1 0.0 0.1 -0.5 0.0 0.5 -2 0 2
B^(T) B^(T) B^(T)

Figure 5.10: NG = ⟨δG2 ⟩/⟨G2 ⟩, as a function of perpendicular magnetic field, B⊥ , at 4.2 K


in the low field region (Bϕ < B⊥ < BZ , where BZ is the Zeeman field) for (a) Ge HD (b)
Ge MD (c) Ge LD (d) Si HD (e) Si MD (f) Si LD.
5.3. Universal Conductance Fluctuations 129

NG(B^=0) / NGf
2.0
2.0
Si_LD Ge_LD
1.5
Si_MD Ge_MD
NG(B^)/NGf Si_HD Ge_HD
1.0
1.5
1 2 *
3
rP/aB

1.0

-30 -20 -10 0 10 20 30


B^/Bf

Figure 5.11: NG (B⊥ )/NGϕ as a function of B⊥ (scaled by the phase breaking field, Bϕ ) for
all devices at 4.2 K, where NGϕ = NG (B⊥ ∼ 20Bϕ ). The inset shows NG (B⊥ = 0)/NGϕ
as a function of rP /a∗B .

that the reduction in NG is closely related to the effective dopant separation rP /a∗B . The
reduction factor decreases gradually as the effective dopant separation, rP /a∗B , increases
and for Si LD, NG becomes essentially flat.
To elaborate, we have compiled the B⊥ -dependence of NG normalized by NGϕ , where
NGϕ is the value of NG at B⊥ ≫ Bϕ but < BZ , for all devices in Fig. 5.11. NGϕ was
chosen at B⊥ ∼ 20Bϕ which was < BZ for all the devices at all temperatures. The peak
in NG around B⊥ = 0 is progressively suppressed with decreasing doping density, and
> 1.5, the Cooperon contribution to UCF noise at low B⊥ becomes
eventually for rP /a∗B ∼
immeasurably small, implying a spontaneous breaking of time reversal symmetry even
at B⊥ = 0. The inset of Fig. 5.11 shows the gradual decrease of the reduction factor,
NG (B⊥ = 0)/NGϕ , as a function of rP /a∗B .
130 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

5.3.2.3 High perpendicular magnetic field

a) b) 25
2.0 Ge_HD Si_HD
0.7K 0.28K

NG
20
NG

1.5

-10
-13

15

10
10

1.0
10
0.5
1.0
15 0.75K
4.2K

NG
NG

-10
-13

10

10
10

0.5

5
5
1E-4 1E-3 0.01 0.1 1 10 4.2K
B^(T)
NG 4
-10

c)
10
3
Si-MD
20 0.75K 2
NG
-14

15 1E-3 0.01 0.1 1 10


10

B^(T)
10 d)
12 Ge_LD
4.2K 4.2K
6
NG

9
NG

5
-12
-14

10
10

4 6

3
3
1E-4 1E-3 0.01 0.1 1 10 1E-3 0.01 0.1 1 10
B^(T) B^(T)

Figure 5.12: NG = ⟨δG2 ⟩/⟨G2 ⟩, as a function of perpendicular magnetic field, B⊥ , till


high magnetic fields (> BZ ) for (a) Ge HD at 0.7 and 4.2 K (b) Si HD at 0.28, 0.75 and
4.2 K (c) Si MD at 0.75 and 4.2 K (d) Ge LD at 4.2 K.

Fig. 5.12 shows the B⊥ dependence of NG till high magnetic fields (> BZ ) for various
devices at different temperatures. The lowest temperature at which the noise measure-
ments were performed was limited by the minimum bias at which a measureable SV was
obtained without causing heating or nonlinear effects. Fig. 5.13a-d shows NG (B⊥ )/NGϕ
for the Si:P and Ge:P devices as a function of B⊥ . It is evident that while Ge HD shows
a factor of two reduction at each of the two field scales, Bϕ and BZ , Si HD and Ge LD
shows a reduction in NG by a factor of ≈ 1.5 at the scale of Bϕ and a factor of ≈ 2 at the
5.3. Universal Conductance Fluctuations 131

a) b)
2.0 2.0
Si_HD
Ge_HD
NG(B^)/NGf
0.28K
1.5 4.2K 1.5 0.75K
0.7K 4.2K

1.0 1.0

0.5 0.5

1E-4 0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10

c) d)
2.0 2.0
Ge_LD Si_MD
NG(B^)/NGf

4.2K 0.75K
1.5 1.5 4.2K

1.0 1.0

0.5 0.5

0.001 0.01 0.1 1 10 1E-4 0.001 0.01 0.1 1 10


B^(T) B^(T)

Figure 5.13: NG (B⊥ )/NGϕ as a function of perpendicular magnetic field, B⊥ , till high
magnetic fields (> BZ ) for (a) Ge HD (b) Si HD (c) Ge LD and (d) Si MD at different
temperatures. NG = ⟨δG2 ⟩/⟨G2 ⟩, and NGϕ = NG (B⊥ ∼ 20Bϕ ).

scale of BZ whereas Si MD shows a factor of ≈ 2 reduction only at the scale of BZ . To


compare different devices at same temperature, we compile NG (B⊥ )/NGϕ for the devices
at 4.2 K and 0.75 K in Fig. 5.14a and 5.14b respectively. It is clearly evident that the

a) b)
2.0 2.0
4.2K Ge_HD, T=0.7K
NG(B^)/NGf

Ge_HD Si_HD, T=0.75K


Si_HD Si_MD, T=0.75K
1.5 Si_MD
1.5
Ge_LD

1.0 1.0

0.5 0.5

1E-4 1E-3 0.01 0.1 1 10 1E-4 1E-3 0.01 0.1 1 10


B^(T) B^(T)

Figure 5.14: NG (B⊥ )/NGϕ as a function of perpendicular magnetic field, B⊥ , till high
magnetic fields (> BZ ) for (a) Ge HD, Si HD, Ge LD and Si MD at 4.2 K and (b) Ge HD
at 0.7 K, Si HD and Si MD at 0.75 K. NG = ⟨δG2 ⟩/⟨G⟩2 , and NGϕ = NG (B⊥ ∼ 20Bϕ ).
132 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

breaking of time reversal symmetry on decreasing the dopant density (or increasing the
effective dopant separation) is present at all temperatures investigated here.

5.4 Role of spins in breaking time reversal symmetry

5.4.1 Conductivity in parallel magnetic field

To explore the origin of lifting of the time reversal symmetry in the δ-layers, we subjected
the devices to in-plane magnetic field, B∥ , that resulted in a nonmonotonic magneto-
conductivity in the lightly doped δ-layers. The in-plane magnetoconductivity ∆σ(B∥ ) of
lightly doped δ-layer Si LD at low temperatures is non-monotonic, where an initial nega-
tive magnetoconductivity is followed by a logarithmic increase at larger B∥ (Fig. 5.15a).
The initial negative ∆σ(B∥ ) can be understood from the scattering of electrons by local
spin fluctuations, where with increasing B∥ , a reduction in the spin-flip scattering rate
results in longer dephasing times and stronger localization of the electrons. The positive
magnetoconductivity at larger B∥ represents suppression of weak localization due to the
finite width of the δ-layers [233].
We now try to do a quantitative estimate of the in-plane magnetoresistance observed
in the low doped devices following Ref [233]. Application of an in-plane magnetic field
polarizes the magnetic moments and opens a gap given by Ezimp = g imp µB B∥ , for the
electron spin-flip scattering process. Here g imp is the g-factor of the magnetic impurity.
When the gap Ezimp exceeds kB T then the spin scattering is suppressed. Thus when
B∥ exceeds the characteristic magnetic field Bz = kB T /g imp µB , the magnetic moments
act as non-magnetic scatters having different potential profile for the up and down spin
electrons. The suppression of spin-flip scattering processes restores the weak localization
correction and accounts for the decrease in conductivity (when B∥ > Bz ), which in a 2D
system is given by

( )
e2 τϕ
∆σ(B∥ ≫ Bz ) = σ(B∥ ≫ Bz ) − σ(B∥ = 0) ∼ − ln (5.8)
πh τs

where τs−1 is the spin-flip scattering rate. For τϕ ≫ τs (which holds true for our case as
5.4. Role of spins in breaking time reversal symmetry 133

(a)
Si_LD 0.4
Si_HD 0.2K 0.2K
0.2

Ds (e /h)
2
0.2

Ds (e /h)
0.0 B||(T) 0.75K
-0.5 0.0 0.5
2

0.0
4.2K

-0.2
-10 0 10
B||(T)

(b) (c) 1
0.2 K
4.2K
0.1 0.75K

0.01
-Ds (e /h)

-Ds (e /h)
2

0.01

1E-4
0.001

1E-6
0.01 0.1 1 10 0.1 1 10
B||(T) B||(T)

Figure 5.15: (a) The magnetoconductivity, ∆σ(B∥ ) = σ(B∥ ) − σ(B∥ = 0) in presence of


magnetic field, B∥ , applied parallel to the plane of the δ-layer for Si LD at 0.2 K, 0.75 K
and 4.2 K. The inset shows ∆σ in B∥ for Si HD at 0.2 K. (b) ∆σ as a function of B∥ for
Si LD at 0.2 K and 0.75 K in log-log scale. The solid lines show that ∆σ ∝ B∥ in the region
of negative magnetoconductivity. (c) ∆σ as a function of B∥ for Si LD at 4.2 K in log-log
scale. The solid lines show that ∆σ ∝ B∥2 in the region of negative magnetoconductivity.

can be seen in the Section 5.4.2) we get [233]


134 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

 ( )2
e 2  1 B∥ B∥ < Bz
∆σ(B∥ ) = σ(B∥ ) − σ(B∥ = 0) ≈ − × 6 Bz
(5.9)
h  B∥ B∥ > Bz
Bz

Thus we see that for localized magnetic moments, the magnetoconductivity at low T and
B∥ > Bz is expected to vary linearly with B∥ as

e2 gimp µB B∥
∆σ(B∥ ) ≈ (5.10)
hkB T

where we have used Bz = kB T /g imp µB . At high T , the magnetoconductivity becomes


quadratic in B∥ (Eq. 5.9) because of large characteristic Zeeman field scale.
While we find the negative magnetoconductivity to be indeed proportional to B∥ /T
at 0.2 K and 0.75 K (Fig. 5.15b), and proportional to B∥2 at 4.2K (Fig. 5.15c), the
experimental magnitude of the coefficient e2 gimp µB /hkB was a factor of ∼ 50 smaller
than that expected theoretically (assuming gimp = 2), possibly indicating a many-body
nature of the spin fluctuations. Noticeably, the negative in-plane magnetoconductivity at
low B∥ was completely absent in the heavily doped devices (inset of Fig 5.15a for Si HD),
thus establishing that the spin fluctuations in the low density δ-layers are due to strong
Coulomb interactions and hence observable only in the lightly doped devices.

5.4.2 Full range fit and extracting the spin scattering time

The compelling analogy with the bulk P-doped Si close to MIT provides a “two-fluid”
framework to address transport in our δ-layers. This consists of itinerant electrons in dis-
ordered Hamiltonian and local magnetic moments [229–231]. The interaction between the
local moments and itinerant electrons suppresses localization, although the spin-scattering
process is quasi-elastic (energy exchange ≪ kB T ), causing only minor modification to the
dephasing mechanism (as confirmed by the linear T dependence of τϕ−1 in Fig. 5.2 and
small ∆σ(B∥ )). In addition, the two-fluid model allows a phenomenological generalized
Hikami-Larkin-Nagaoka expression for the total quantum interference correction that in-
cludes the quasi-elastic spin scattering rate (τs−1 ) as,
5.4. Role of spins in breaking time reversal symmetry 135

[ ( ) ( )] ( )
B⊥ B⊥ B⊥
∆σ(B⊥ , T ) = α F −F − βF (5.11)
Bϕ B0 Bs

where α and β are positive constants close to unity, and F (x) = ln(x) + ψ(0.5 + 1/x), with
ψ(x) being the digamma function. As shown by the solid lines in Fig. 5.5, Eq. 5.11 de-
scribes the magnetoconductivity very well over the entire range of B⊥ . The fit parameter
Bs = ~/4eDτs , provides an estimate of the spin scattering time τs . We note the following:
(i) As evident in Fig. 5.16a, τs−1 is more than ten times larger than experimentally mea-
sured τϕ−1 (see Fig. 5.2b for values of τϕ−1 ), confirming that the spin-scattering is mostly
elastic. (ii) Second, τs−1 varies nonmonotonically with n. The filled squares represent
τs−1 analyzed from data of Ref [161]. At low n, τs−1 ∼ n0.5 irrespective of host material,
disorder or carrier mobility, indicating that the number of local spins are only related with
the number of P dopant sites. However τs−1 drops abruptly around n ∼ 1.5 × 1014 cm−2 ,
suggesting a quenching of the spins and commencement of free-electron weakly localized
quantum transport. The T-dependence of τs−1 (Fig. 5.16b), in accordance with the two-
fluid model, shows a power law variation as τs−1 ∝ T p , with p ≈ 0.7. From Ref. [230] we
get τs−1 ∝ T χL (T ), where χL is the local moment contribution to the spin susceptibility.

(a) (b)
10
SiP Si_LD
GeP 10
)

Ge_MD
-1

s (10 s )
s

Old Ref
12 -1
12
ts (x10
-1

t-1

1 1

0.5 1 2.5 5 0.1 1 10


14 -2
n (x10 cm ) T (K)

Figure 5.16: (a) The spin scattering rate, τs−1 , as a function of carrier density, n, for all
devices at 0.28 K. The solid line shows that τs−1 ∝ n0.5 . The filled olive squares mentioned
as Old Ref represent data taken from Ref. [161](b) τs−1 as a function of T for Si LD and
Ge MD. The solid lines show that τs−1 ∝ T 0.7 for both the devices.
136 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

Using the temperature dependence of χL (T ), we get τs−1 ∝ T 1−α , where α > 0 character-
izes the divergence of spin susceptibility [234]. Thus the exponent for susceptibility and
specific heat divergence in the δ-layers is found to be ≈ 0.3, which is about half of that
observed in the bulk Si:P close to MIT [229, 235].
Finally, to estimate the fraction of P-dopants that host a local moment, we compare
the estimated τs−1 in lightly doped Si LD (n = 5 × 1013 cm−2 ) with (1) the total momen-
tum relaxation rate τ0−1 ≈ 1014 s−1 from the experimental Drude conductivity, although
this involves scattering from neutral defects as well, and (2) calculated momentum relax-
ation rate (≈ 2 × 1013 s−1 ) expected purely from the P-dopants (charged impurities) (see
calculation details in Ref [167] and Chapter 3 Section 3.2.1). This gives a bound between
2% − 10% of the P-dopants to host local moments which is consistent with the fraction
expected for half-filled impurity bands in bulk Si:P [231]. Importantly, while the weak
localization correction is reduced only partially ( 40% in Si LD), the UCF noise due to
the Cooperons is completely suppressed for the weakly doped devices. It is possible that
because the UCF noise involves interference between two Feynman propagators, it is more
likely to be affected by the localized spins than the WL correction which is determined by a
single self intersecting propagator. Note that we have not discussed spatial inhomogeneity
or clustering in the distribution of dopants which can lead to coexistence of localized and
delocalized phases [109], impact of multiple valleys [112, 115], or the inter-site Coulomb
interaction [43, 111, 219] which are unlikely to affect the time reversal symmetry.

5.5 Low density limit and metallic behavior

In this section we discuss the results of Si:P δ-layer with lowest doping density of P atoms
(corresponding to n ≈ 2×1013 cm−2 ) yet maintaining a finite electrical conductivity. Below
this density, none of the ohmic contacts were working and no electrical conductance was
observed. It is possible that below this density, the system undergoes a metal-to-insulator
transition or it is also likely that the system is very inhomogeneous and it is not possible
to make ohmic contacts to them. In order to get more insight into the nature of this
low density device we revisit the σD vs n curve by including the lowest density hall bar
(Si LDL). This hall bar has multiple probes with distance between them ranging from
5.5. Low density limit and metallic behavior 137

(a)
100

sD (e /h)
3/2
n

2
10

0.3 1 5
14 -2
n (10 cm )

(b) (c)
L = 10mm
10.8 9
W=2mm
r (kW  -1)

8
9.0
L = 4mm
W=2mm 7
7.2

10 100 1000 10 100 1000


T (mK) T (mK)

Figure 5.17: (a) The Drude conductivity σD as a function of carrier density n including
the lowest density device Si LDL with different distances between the voltage probes,
L = 10, 4 (filled olive circles), 2 (Filled triangles) and 1 µm (Filled squares). The resistivity
ρ as a function of temperature T for a low density hall bar of dimensions (b) 4 × 2 µm
and (c) 10 × 2 µm. The dashed lines show that the resistivity increase as ρ ∝ ln(T ).

L ∼ 1 − 10 µm. The σD vs n (Fig. 5.17a) shows that for large L (4 and 10 µm), this
device follows σD ∝ n3/2 (filled olive circles in Fig. 5.17a). However as the L reduces to 2
and 1 µm (filled triangles and squares respectively in Fig. 5.17a) the conductivity deviates
from the n3/2 curve, having a value lower than expected. For all other devices that have
been investigated in this thesis, the conductivity was independent of device dimensions.

We now concentrate on the section of the hall bar Si LDL with L = 4 and 10 µm
where the σD is nearly independent of dimensions and falls on the n3/2 curve as the rest
138 Chapter 5. Transport and Noise in strongly interacting 2D δ-layers

of the device. The resistivity increases logarithmically with temperature as expected in


the quantum coherent transport regime. However this logarithmic increase continues up
to ∼ 200 mK and ∼ 500 mK for L = 4 and 10 µm respectively, below which there is
metallic like behavior with resistivity decreasing with decreasing temperature (Fig. 5.17b
and c respectively). According to the scaling theory of localization, in the absence of
interactions and spin scattering, there cannot be a metallic ground state in two dimensions
at T = 0 K [22, 23]. Other theoretical studies which take into account the effect of
interactions have predicted the existence of metallic state at T = 0 K [31, 32] and strong
metallic behavior have also been reported in experimental studies of 2D systems with low
disorder [33–43]. Whether low density Si:P δ-layers can have a metallic ground state at
0 K or not, requires further theoretical and experimental studies.

5.6 Summary of the results


1. We have carried out detailed transport studies of Si:P and Ge:P δ-layers with dif-
ferent doping densities using conductivity and universal conductance fluctuations
measurement.

2. The quantum interference correction to conductivity is suppressed at low doping


density (in the strongly interacting regime).

3. The suppression of universal conductance fluctuations for low density devices pro-
vides unambiguous evidence of spontaneous time reversal symmetry breaking.

4. Magnetoconductivity in an in-plane magnetic field indicates local moments at the


P dopant sites break the time reversal symmetry

5. The lowest density device measured (n ∼ 2×1013 cm−2 ) shows metallic like behavior
at low temperatures (< 500 mK) where the resistivity decreases with decreasing
temperature.
Chapter 6

Effect of superconducting contacts


on transport in δ-layers

In this chapter we study the transport properties of low density Si:P and Ge:P δ-layers
with metallic contacts made of a superconducting material. The hall bar dimensions
(length L = 78 nm and width W = 20 nm) are much larger than the superconducting
coherence length which is of the order of a micron. Thus we would expect that any
proximity induced superconductivity will limited to a micron of the δ-layer from the
contacts.

The transport measurements carried out in this chapter involves magnetoresistance


measurements ( in parallel and perpendicular field) and noise as a function of temperature
and magnetic field (parallel and perpendicular). An alternate probe which we employ in
this chapter is nonlocal resistance measurements where the voltage is measured across a
region of the sample far away from the path of the applied current.

Superconducting contacts are made on Si:P and Ge:P δ-layer by using Aluminium
(Al) as material for ohmic contacts. Al is a type I superconductor which undergoes a
transition from normal state to a superconducting state at temperature T < 1 K. The su-
perconductivity can be destroyed by applying a magnetic field (parallel or perpendicular)
above a critical value Bc which is ∼ 10 mT for Al. The energy gap in Al is ∼ 3 × 10−4 eV
and the coherence length is ∼ 1.6 µm.

139
140 Chapter 6. Effect of superconducting contacts on transport in δ-layers

6.1 DC Current Voltage characteristics


The DC current-voltage (I-V) characteristics of the δ-layer were measured in both four
probe and two probe configuration at temperature T = 20 mK using Keithley SMU 2400
which acts as a constant current source while simultaneously measuring the voltage bias
across the sample. The four probe I-V is linear (Fig. 6.1a) and the slope gives a resistance
of ∼ 12.8 kΩ. The two probe I-V is nonlinear near the zero bias region (Fig. 6.1b)
due to the superconductivity in Al contacts. Figure 6.1c shows the two probe I-V in a
perpendicular magnetic field B⊥ = −40 mT which is linear as the superconductivity in
Al contacts is destroyed by the magnetic field.

(a) 0.3 (b) 0.8 (c) 0.7


Four Probe IV Two Probe IV Two Probe IV
20mK 20mK 20mK
B=0T B=0T B^ = -40mT
V (mV)

V (mV)
V (mV)

0.0 0.0 0.0

-0.3 -0.8 -0.7


-20 0 20 -10 0 10 -10 0 10
I (nA) I (nA) I (nA)

Figure 6.1: (a) Current voltage (I-V) characteristic of the device at 20 mK (a) Four probe
configuration at zero magnetic field and (b) Two probe configuration at zero magnetic
field (c) Two probe configuration at perpendicular magnetic field B⊥ = −40 mT.

6.2 Resistivity vs temperature


The resistivity ρ of a low density Si:P and a Ge:P δ-layer (measured using a four probe
configuration with excitation frequency ∼ 220 Hz) as a function of temperature is shown
in Fig. 6.2a and 6.2b respectively. The carriers densities are n = 5 × 1017 m−2 and 3.2 ×
1017 m−2 for the Si:P and Ge:P δ-layer respectively. There are two distinct regimes in the
dependence of ρ on temperature. As T decreases below 4.2 K, ρ increases logarithmically
as expected in quantum coherent transport regime and observed in Chapter 3 and Section
3.3. As T decreases below 0.3 K for the Si:P δ-layer (Fig. 6.2a) and below 0.7 K for the
Ge:P δ-layer (Fig. 6.2b), the resistivity increases sharply. Note that the Tc for Al is ∼ 1 K
6.2. Resistivity vs temperature 141

(a) 3.4 (b) 5.0

r (kW) / box

3.2
4.5
ln (T)
Ge:P
3.0 Si:P -2
17
n = 5 x10 m
-2 3.2 x17 m
4.0
0.01 0.1 1 0.01 0.1 1
T (K) T (K)
(a) (b) 0T
0T
13.0 1mT 12.8 0.5mT
5mT 1mT
7mT 1.6mT
10mT 2mT
12.5 3mT
12mT 12.4
R (kW)

15mT 4mT
20mT 5mT
50mT 10mT
70mT 30mT
12.0 12.0

11.5 B|| 11.6 B^

0.01 0.1 1 0.01 0.1 1


T (K) T (K)
R (k )

Figure 6.2: (a) Resistivity ρ vs temperature T for a Si:P δ-layer having a carrier density
n = 5 × 1017 m−2 . (b) ρ as a function of T for a Ge:P δ-layer having n = 3.2 × 1017 m−2 .
The dotted line shows that ρ ∝ ln(T ). The resistance as a function of temperature for the
Si:P δ-layer at different magnetic fields applied (c) parallel to the plane (d) perpendicular
to the plane.

but the influence of superconductivity in Al contacts on the temperature dependence of


resistivity is visible for lower temperatures. Another surprising aspect is that we see an
increase in resistivity at the onset of superconducting transition whereas naively one would
expect the resistivity to decrease. The sharp increase in ρ on decreasing T continues till
∼ 0.1 K for the Si:P δ-layer after which it saturates (Fig. 6.2a). For the Ge:P δ-layer
we observe a broad peak at ∼ 0.15 K, below which ρ starts decreasing till the lowest
142 Chapter 6. Effect of superconducting contacts on transport in δ-layers

temperature measured (Fig. 6.2b). A small magnetic field gets rid of the sharp increase in
the resistance as well as the resistance peak as shown in Fig. 6.2c and 6.2d for magnetic
field applied parallel and perpendicular to the the plane of the Si:P δ-layer respectively.
The magnetic field required to suppress the effect of superconductivity is > 2 mT in case
of perpendicular field and is > 10 mT in case of parallel field.
A resistance peak near the superconducting transition temperature have been observed
previously in CuZr alloys with dilute magnetic impurities, and the authors highlight the
importance of interaction between superconducting fluctuations and spins [236]. How-
ever the Si:P and Ge:P δ layers, fabricated in ultra-high vacuum environment are not
expected to have any magnetic impurities, though there are evidence of local moment
formation in such low doped devices as explained in Chapter 5. Similar observation of
resistance peak near the superconducting transition has been observed in mesoscopic Al
structures [237–240], in disordered superconducting films of Nb [241] and granular su-
perconductors [242]. A possible explanation was offered for the mesoscopic Al structures
in terms normal-superconducting interfaces and charge imbalance induced by different
critical temperatures for different parts of the structures [237, 239, 240, 243]. An inho-
mogeneous heat transfer or local dissipation in the sample was thought to cause the
anomaly in disordered superconductors [241]. More recently ferromagnetic Co wires with
superconducting tungsten electrodes have shown a resistance peak near the transition
temperature [244] and though the exact cause was unknown, the authors suggest that of
charge/spin imbalance was not the cause of the anomaly.
Since we are doing a four probe measurement of the device in hall bar geometry, we
are not probing the Al contacts but rather only the δ-layer. Hence we believe that our
observations are more likely due an interaction of superconducting fluctuations and local
moments, though it needs to be explored further, theoretically and experimentally.

6.3 Magnetoresistance

6.3.1 Parallel magnetic field

The magnetoresistance in an in-plane magnetic field B∥ with the field direction applied
parallel to the direction is current (schematic in the inset of Fig. 6.3) is shown in Fig. 6.3.
6.3. Magnetoresistance 143

20mK
13.0

RXX (kW)
12.5

12.0

-100 -50 0 50 100


B||(mT)

Figure 6.3: Magnetoresistance in parallel magnetic field at 20 mK. The inset shows a
schematic of magnetic field applied parallel to the plane of the δ-layer with the field
direction being parallel to the direction of current flow.

The following features are noted: (i) There is an anomalous peak in the resistance at
B∥ = 0 where the Al contacts are in superconducting state. (ii) Sharp dips in resistance

(a) (b) 20mK 116mK


-ve to +ve_1 20mK
+ve to -ve_1 186mK 270mK
290mK 320mK
13.0 -ve to +ve_2
420mK 585mK
+ve to -ve_2
R(B) / R(0.1T)

-ve to +ve_3 1.04 650mK 775mK


RXX (kW)

930mK 1.25K
+ve to -ve_3
1.52K

12.5
1.00

12.0
0.96
-100 -50 0 50 100 -100 -50 0 50 100
B||(mT) B||(mT)

Figure 6.4: (a) Hysteresis in magnetoresistance in a parallel magnetic field, B∥ , at 20 mK.


Multiple sweeps in the same field direction fall on top of each other. (b) The scaled
resistance R(B)/R(−0.1 T), where R(−0.1 T) is the resistance at B∥ = −0.1 T, as a
function of B∥ at different temperatures ranging from 20 mK to 1.5 K.
144 Chapter 6. Effect of superconducting contacts on transport in δ-layers

at ∼ ±10 mT which is the critical field at which the Al contacts undergo a transition
from superconducting to normal state. (iii) The cusp beyond ±10 mT where the resistance
increases slowly till B∥ ∼ 50 mT. (iv) The resistance remains constant till B∥ ∼ 100 mT.
In the region where the contacts are superconducting, there is a hysteresis in magne-
toresistance as shown in Fig. 6.4a. As the temperature is increased all the above mentioned
features become weak and for T > 1 K the magnetoresistance is essentially flat having no
dependence on B∥ (Fig. 6.4b).

6.3.2 Perpendicular magnetic field

When the magnetic field is applied perpendicular to the plane of the δ-layer (schematic in
the right inset of Fig. 6.5), a peak in resistance is observed near the B⊥ = 0 region followed
by minimum at B⊥ ∼ ±2−3 mT (Fig. 6.5). Note that this peak in different from the usual
weak localization peak discussed in Chapter 5. The peak observed here is much sharper
and associated with the superconductivity in Al contacts. Similar to the parallel field
magnetoresistance there is a hysteresis in the case of perpendicular magnetic field when
the contacts are superconducting (left inset of Fig. 6.5). The temperature dependence
of magnetoresistance is shown in Fig. 6.6. In order to compare the features at different
temperatures we scale the resistance at different B⊥ by the value at B⊥ = −56 mT
(which is ≫ Bc for Al). As the temperature increases, the peak height decreases and at
T ∼ 480 mK the peak disappears and instead we see a dip in magnetoresistance at B⊥ = 0.
Further increasing the temperature decreases the depth of the resistance minimum and
at T ∼ 1.77 K we observe the usual weak localization behavior in magnetoresistance
expected for a disordered two-dimensional electron gas at low temperatures. The inset
in Fig. 6.6 shows the individual magnetoresistance traces at four temperatures clearly
depicting the two transitions.
The generic features of the magnetoresistance in parallel and perpendicular fields are
the peak at B = 0 T and hysteresis observed in opposite sweep directions. However the
field scales in the parallel and perpendicular orientation are different. It is known that for
thin films the critical magnetic field required to destroy the superconductivity is different
for different field orientations and is smaller when applied magnetic field is perpendicular
to the plane. This is evident in our data as the resistance peak at B = 0 drops at field scale
6.3. Magnetoresistance 145

13.2
13.2 20mK

12.9
RXX (kW)

12.6
12.8 -10 -5 0 5 10

12.4
-50 0 50
B^(mT)

Figure 6.5: Magnetoresistance in a perpendicular magnetic field at 20 mK for up and


down sweep directions of magnetic field. The left inset shows hysteresis near the zero-field
region. The right inset shows the schematic of the magnetic field applied perpendicular
to the plane of the δ-layer.

of ±10 mT and ±2 mT in parallel and perpendicular magnetic field respectively (Fig. 6.3
and 6.5). Additionally we notice that even after the superconductivity in Al contacts is
expected to be destroyed (at ±10 mT and ±2 mT in B∥ and B⊥ respectively), its effect
on the transport in δ-layers is evident till ∼ ±50 mT (cusp like feature) and ∼ ±10 mT
(the small peaks on either side of central maximum) in B∥ and B⊥ respectively. This
is surprising as Al is a type I superconductor with a single critical field beyond which it
should behave as a normal metal. Also, any effect of proximity induced superconductivity
is expected to be limited to about a micron from the Al contacts into the δ-layer. Thus
we do not expect it to manifest itself in a four probe configuration which measures the
resistance of the rectangular mesa between the voltage probes only.
The peculiar magnetoresistance observed is connected to the sharp increase in resis-
tance as seen in Fig. 6.2a and 6.2b, and has been previously observed in superconductor-
semiconductor heterojunction formed by growing single crystal epitaxial Pb on Si sub-
strate [245]. Though a complete satisfactory model was not given, two possible mecha-
146 Chapter 6. Effect of superconducting contacts on transport in δ-layers

25mK 103mK 12.6 25mK


12.3
185mK 212mK
12.20 480mK
1.04 284mK 355mK

R (kW)
12.15
R (B)/R(-56mT)

12.10
410mK 480mK
11.610 1550mK
575mK 700mK
11.592
800mK 900mK 11.54
11.53 1770mK
1000mK 1100mK 11.52
11.51
1210mK 1425mK -40 -20 0 20 40
1.02 1550mK 1770mK B (mT)

1.00

-40 0 40
B (mT)

Figure 6.6: The scaled resistance R(B)/R(−56 mT), where R(−56 mT) is the resistance at
B⊥ = −56 mT, as a function of perpendicular magnetic field B⊥ at different temperatures
ranging from 25 mK to 1.77 K. The inset shows the magnetoresistance at temperatures
25 mK, 480 mK, 1.55 K and 1.77 K.

nisms included increased electron tunneling due to increasing magnetic field and increased
Andreev reflection due to reduction in energy gap with the application of magnetic field.
Another strikingly similar magnetoresistance has been observed in topological insulator
thin films of Bi2 Se3 contacted by superconducting electrode made from Al, In and W [246].
The peak in resistance at B = 0, accompanied by sharp drops at ±5 mT and ±10 mT and
the hysteresis for opposite sweep directions were observed. The results were explained
in terms of interplay between the spin polarized current in topological insulators and
Cooper pairs of the superconducting electrodes. The spin-flip processes that occurs when
the Cooper pairs leak from contacts and spin polarized current flow into the contacts
leads to increase in resistance in the superconducting regime [246]. The exact cause of
the peculiar magnetoresistance in the δ-layers is unknown, though we will discuss few
possibilities at the end of this chapter.
6.4. Resistance fluctuations or low frequency noise 147

6.4 Resistance fluctuations or low frequency noise


To get better insight about the effect of superconductivity on the transport in δ-layers
we have carried out resistance fluctuation measurements as a function of parallel and
perpendicular magnetic field at different temperatures. Any kind of superconducting
fluctuations will have an impact on the fluctuations in resistance which we hope to capture
through the time series of resistance and the power spectral density.

6.4.1 Parallel magnetic field


6.4.1.1 Time series of resistance fluctuations

The time series (after decimation and Weiner filtering as described in Chapter 2) of re-
sistance fluctuations obtained by measuring the four probe resistance of the δ-layer at
T = 20 mK for B∥ = 2, 10, 20, 50 and 100 mT is shown in Fig. 6.7a-e respectively, thus
going across the superconducting transition. At B∥ = 2 mT, where the Al contacts are in
superconducting state, we observe periodic modulations in the time series of resistance.
The power spectral density (PSD) of the resistance fluctuations reveals the periodic fluc-
tuations as peaks at different frequencies. Such periodic modulations in the time series of
resistance have been previously observed in the vortex state of type II superconductors of
2H-NbSe2 [247] and Bi2 Sr2 CaCu2 Oy which is a high temperature superconductor [248],
and is termed as washboard noise. The washboard noise is a manifestation of reordering
of driven vortices on a macroscopic scale. The periodic modulation in the resistance noise
indicates a periodic modulation in the translational velocity of vortex lattice. The peri-
odic modulations in the resistance noise and peaks in power spectral density persist in the
superconducting state and even beyond the critical field of aluminium which is ∼ 10 mT.
This is clearly visible in the time series and corresponding PSD in Fig. 6.7c and 6.7h
respectively. The periodic fluctuations disappear at B∥ = 50 mT and higher, where the
time series is random (Fig. 6.7d and 6.7e) and the PSD is devoid of any peaks as can be
seen in Fig. 6.7i and 6.7j. It is also clear from the time series of resistance fluctuations and
the PSDs that the magnitude of fluctuations are largest in the superconducting regime
and goes down as the magnetic field destroys the superconductivity. This point will be
elaborated in the next section.
148 Chapter 6. Effect of superconducting contacts on transport in δ-layers

At higher temperature ∼ 900 mK which is close to the superconducting transition


temperature of Al (∼ 1K), the time series of the resistance is completely random and
bears no signature of any periodicity at any magnetic field as shown in Fig. 6.8a-d, thus
establishing that the superconductivity in Al contacts greatly influences the states in Si:P
δ-layer.

(a) 96.0 20mK (f)


2mT 10

95.2
5

94.4
0
2340 2360 2380
(b) 95.5 15
10mT (g)
95.0 10

94.5
5

94.0
0
2130 2140 2150 2160

(c)
V Hz )

20mT 3 (h)
-1

83.6
V (mV)

2
83.2
-14
SV (x 10

1
82.8
0
1020 1050 1080 1110
(d) 50mT 0.009
(i)
57.21
0.006

57.20 0.003

0.000
1465 1470 1475
(e) 67.79
(j)
100mT
0.010

67.78
0.005

67.77
0.000
2040 2043 2046 0 2 4 6 8
Time (s) Frequency (Hz)

Figure 6.7: The time series of resistance fluctuations (after decimation and wiener filter-
ing) at 20 mK in a parallel magnetic field, B∥ , of (a) 2 mT (b) 10 mT (c) 20 mT (d)
50 mT and (e) 100 mT and the corresponding power spectral density in (f)-(j)
6.4. Resistance fluctuations or low frequency noise 149

T = 900mK
(a) (b)
0T 5mT
419.33
401.76
V (mV)

419.32
401.75

419.31
401.74

874 876 878 880 506 508 510 512 514


Time (s) Time (s)
(c) (d) 100mT
10mT
421.18 429.24
V (mV)

429.23
421.17

429.22
421.16
756 758 760 762 922 924 926 928 930
Time (s) Time (s)

Figure 6.8: The time series of resistance fluctuations (after decimation and wiener fil-
tering) at 20 mK in a parallel magnetic field, B∥ , of (a) 0 T (b) 5 mT (c) 10 mT (d)
100 mT.

6.4.1.2 Integrated variance

The measured power spectral density was integrated within the experimental bandwidth
to obtain the normalized variance ⟨δR2 ⟩/R2 at each value of B∥ . The result is plotted
for 20 mK and 500 mK in Fig. 6.9a and 6.9b respectively. For comparison the Rxx
vs B∥ is plotted on the left axis for both temperatures. We note the following general
features: (i) The noise variance is enhanced at B∥ = 0 T and remains fairly constant
till B∥ = ±10 mT within which the Al contacts are superconducting and there is a peak
in magnetoresistance. (ii) Beyond ±10 mT the noise decreases with increasing B∥ till
±50 mT. This is the region where there is a cusp-like feature in magnetoresistance where
resistance increase with increasing B∥ . Beyond ±50 mT the noise remain fairly constant.
Note that the resistance also does not change in this region. Thus we note that there is
150 Chapter 6. Effect of superconducting contacts on transport in δ-layers

(a) MR and Noise - 20mK -5 (b) 12.2 MR and Noise - 500mK -8


10 10
13.0

-7
10 12.1
RXX (kW)

RXX(kW)
ádR ñ / R
-9
10
12.5

2
ádR ñ/ R
2

2
-9
10 12.0

12.0 3 3 10
-10

3 2 1 2 3 -11
11.9
2 1 2
10
-100 -50 0 50 100 -100 -50 0 50 100
B||(mT) B||(mT)

(c) 10
-5

20mK
115mK
300mK
-7 500mK
10 900mK
2
ádR ñ / R
2

-9
10

-11
10
-100 -50 0 50 100
B||(mT)

Figure 6.9: (a) The normalized variance, ⟨δR2 ⟩/R2 , as a function of parallel magnetic field,
B∥ , at (a) 20 mK and (b) 500 mK (right axis). For comparison with magnetoresistance in
B∥ at 20 mK is plotted on the left axis. (b) ⟨δR2 ⟩/R2 as a function of B∥ at temperatures
ranging from 20 mK to 900 mK.

a direct correspondence between the noise and resistance as a function of B∥ .


The noise variance as a function of B∥ at different temperatures is shown in Fig. 6.9c.
We observe that the peak in noise at B∥ = 0 T goes down as the temperature increases and
at ∼ 900 mK where the aluminium contacts undergo a transition from superconducting to
6.4. Resistance fluctuations or low frequency noise 151

the normal phase the noise remains constant with B∥ for the entire range of magnetic filed
investigated here. We conclude here that the noise is enhanced by orders of magnitude
in the superconducting regime.

6.4.2 Perpendicular magnetic field

Figure 6.10a shows SV as a function of frequency for B⊥ = 0, 0.5 and 10 mT. At B⊥ = 0


and 0.5 mT when the Al contacts are superconducting, SV shows a peak at low frequencies
(similar to that observed in B∥ in Fig. 6.7) and SV ∝ 1/f α at higher frequencies with
the frequency exponent α ∼ 3. At higher B⊥ (= 10 mT), the superconductivity in Al
contact is destroyed, the magnitude of SV decreases and we obatin SV ∝ 1/f . The noise
variance as a function of B⊥ is shown in Fig. 6.10b with the magnetoresistance plotted
on the right axis (red curve). As observed in the parallel field measurement, there is a
peak in the noise at B⊥ = 0 T. The noise starts decreasing rapidly as B⊥ increases till
∼ ±5 mT after which it remains fairly constant. Once again note the correspondence
with magnetoresistance as the beyond ∼ 5± mT the conventional conventional weak
localization behavior is observed. As the temperature increases the peak in the noise
magnitude at B⊥ = 0 T goes down at ∼ 900 mK the noise remains constant with B⊥
since the superconductivity in Al is destroyed (Fig. 6.10d).
Low frequency noise has previously been measured for tin and lead films near the
superconducting transition [249,250]. It was suggested that noise arises due to fluctuations
in temperature and since near the transition temperature the change in resistance is huge,
a small fluctuation in temperature leads to a large noise. It was also observed that this
model leads to flattening of SV at lower frequencies. The temperature fluctuation model
is unlikely to be the cause of noise in case of Si:P δ-layers. Firstly, we do not observe a
saturation in SV at low frequencies, rather we observe a peak in SV as discussed above.
Secondly, we can see from the time series of resistance fluctuations (Fig. 6.7), the average
change in resistance is > 1% and from the temperature dependence of resistance we
estimate that to a temperature of ∼ 20 mK is required to produce a 1% change in
resistance at ∼ 150 − 200 mK. We have checked that the temperature in our setup does
not fluctuate by such large amount. Hence we can rule out the temperature fluctuations
as being the cause of noise.
152 Chapter 6. Effect of superconducting contacts on transport in δ-layers

Another possibility is a percolative transport in which there is a network of small


superconducting islands and metallic regions [251]. Noise can arise due to metallic re-
gions (classical percolation) or due superconducting regions which switches to normal
state randomly (novel percolation). In both scenarios the noise is expected to vary as
⟨δR2 ⟩/R2 ∝ Rlr,s , where the percolation exponent lr,s ∼ −0.86 and ∼ 1.54 for classical
and novel percolation respectively. For the Si:P δ-layers, the noise increases exponentially
with resistance in the superconducting regime as can be seen in Fig. 6.10c. Hence, per-
colation does not seem to be the cause of noise for the δ-layers in the superconducting

(a) (b) Magneto-resistance


0T Noise
0.5 mT -5
-5 10 20mK
10 10 mT
13.0
SV / V (Hz )

2
-1

ádR ñ / R

RXX (kW)
3
1/f ~ 9-10mT
-7 -7
10 10
2

12.8
2

-9 1/f
10 -9
10 12.6
-11
10
0.01 0.1 1 10 -30 -20 -10 0 10 20 30
Frequency (Hz) B^(mT)
(c) (d)
-5
2.0x10 -5 20mK
10 280mK
500mK
2
ádR2ñ / R2

900mK
ádR ñ / R

-7
-5 10
1.0x10
2

-9
10

0.0
-11
10
11.6 12.0 12.4 12.8 -15 -10 -5 0 5 10 15
R (kW) B^(mT)

Figure 6.10: (a) The power spectral density SV as a function of frequency for different
perpendicular magnetic fields at 20 mK. The olive dotted line shows that SV ∝ 1/f 3 ,
while the blue dotted line shows that SV ∝ 1/f (b) The normalized variance, ⟨δR2 ⟩/R2 ,
as a function of perpendicular magnetic field, B⊥ , at 20 mK (left axis). For comparison
with magnetoresistance in B⊥ at 20 mK is plotted on the right axis. (c) ⟨δR2 ⟩/R2 as a
function of resistance R of the δ-layer at 20 mK. The solid line shows that noise increases
exponentially with resistance. The different values of resistance correspond to different
magnetic fields. (d) ⟨δR2 ⟩/R2 as a function of B⊥ at temperatures ranging from 20 mK
to 900 mK.
6.5. Non-local measurements 153

regime.
At present we do not have a complete picture about the noise mechanism in δ-layers
with superconducting contacts. The vortex noise seem to be a possibility as seen from
washboard effect in Fig. 6.7. Howver, the formation of vortex is a characteristic of type
II superconductors, while Al is a type I superconductor. It is possible that the proximity
induced superconductivity in the δ-layer is of type II nature and the noise is picked up
from there. However a complete picture is lacking.

6.5 Non-local measurements

In this section we describe an alternate probe, called nonlocal resistance measurements


to study the transport behavior in these devices. Nonlocal resistance means existence of
voltage in regions which are outside the path through which the current is flowing. It is
thus in clear contradiction with the very basic Ohm’s law.
At first we study the effect of interchanging the I+ and I− probes. Since we are
doing a four probe ac measurement, it is expected that exchanging the two current probes
among themselves would not have any effect on the measured resistance. However this
does not hold true for low density Si:P δ-layers contacted by Al electrodes. Figure 6.11a
shows the effect of interchanging the current probes on the magnetoresistance in B⊥ . The
top panel shows it in a particular configuration where the peak in resistance is observed
at B⊥ = 0. On interchanging the I+ and I− probes among themselves, the peak at
B⊥ = 0 changes into a dip at B⊥ = 0 (bottom panel). Similar behavior has been observed
for parallel magnetic field as shown in Fig. 6.11 where the peak at B∥ = 0 changes
to a dip (first two panels in Fig. 6.11b). Additionally we observe that for a particular
configuration of I+ and I− contacts, the behavior at B∥ = 0 can be tuned by choosing
from the alternate longitudinal probes of the hall bar. The inset of Fig. 6.11a shows a
scanning electron microscopy image of the device along with the contact nomenclature.
The various contact configuration used for magnetoresistance measurements are written
in the individual graph panels.
Figure 6.12a shows a schematic of nonlocal resistance measurement circuit which was
used for this device. Contacts E1-H1 are used as the current probes. The voltage probes
154 Chapter 6. Effect of superconducting contacts on transport in δ-layers

(a) (b) -0.08 0.00 0.08


-0.08 0.00 0.08
12.6
20mK
13.8
RXX (kW)

11.2 I-K2H1 V-H2K1


I-H1K2
13.2 I-H1K2 V-H2K1
V-H2K1 13.6

RXX(kW)
12.8
12.6

12.9
12 20mK
RXX(kW)

12.6 I-K2H1 V-A1E1


I-H1K2 V-A1E1
11
I-K2H1 12.9
V-H2K1
10 12.6
-0.08 0.00 0.08 -0.08 0.00 0.08
B^(T) B||(T)

Figure 6.11: The magnetoresistance at 20 mK for different contact configurations (in-


terchanging the I+ and I− probes or using the alternate voltage probes for longitudinal
resistance measurement) in (a) Perpendicular magnetic field, B⊥ . The inset shows a scan-
ning electron microscopy image of the device along with the contact contact nomenclature.
(b) Parallel magnetic field, B∥ . Two contact configurations were tried for B⊥ and four
configurations were measured for B∥ .

(A1-H2) are outside the region where charge current flows. An existence of finite voltage
across these probe gives a finite nonlocal resistance in the device. We record the nonlocal
resistance as a function of B∥ at different temperatures ranging from 20 mK to 4.2 K
in Fig. 6.12b. We note that the nonlocal resistance exists only in the region in which
the contacts are superconducting, i.e. within B∥ = ±10 mT and at T < 1 K. The
classical nonlocal resistance expected due to spreading of current paths near the voltage
L = ρe −
cl ( π L
probe region is RN W
), where ρ, L and W are the resistivity, sample length
L to be ∼ mΩ which is much less than the
cl
and width respectively [252] We estimate RN
measured RN L measured in the superconducting regime. Hence we rule out the classical
contribution as a cause of nonlocal resistance.

The fact that the nonlocal resistance exists only in the superconducting regime is seen
6.5. Non-local measurements 155

(a) (b) 600


20mK I-E1H1 V-A1H2
105mK
150mK

RNL--X (ohm)
Length = 78 μm 200mK
Width = 20 μm 400 283mK
436mK
700mK
980mK_1
200 1700mK
4.2K

0
-0.1 0.0 0.1
B||(T)
(c) (d)
600
B = 0T B|| = 100mT
70

400
RNL (W)
RNL (W)

60

200

50

0
0.01 0.1 1 0.01 0.1 1
T (K) T (K)

Figure 6.12: (a) Schematic showing the circuit diagram used for nonlocal resistance mea-
surements. (b) The in-phase component of the nonlocal resistance RN L − X as a function
of parallel magnetic field B∥ at temperatures ranging from 20 mK to 4.2 K. The nonlocal
resistance RN L as a function of temperature for (c) B∥ = 0 T and (d) B∥ = 100 mT.

more clearly in Fig. 6.12c and 6.12d where we plot RN L as a function of temperature
for B∥ = 0 and 100 mT respectively. For B∥ = 0 T the nonlocal resistance increases
by an order of magnitude as T is reduced from 4.2 K to 20 mK. However for B∥ =
100 mT, the nonlocal resistance increases only by about 50 %. Note that even outside the
superconducting regime (B > Bc and T > Tc ), there exists a finite RN L ≫ RN
cl
L , which

we believe, arises due to capacitive coupling. However in the superconducting regime,


the RN L increases drastically, by almost an order of magnitude while the local resistance
increases only by about 10% (Fig. 6.2), and is unlikely to be due to capacitive coupling.
156 Chapter 6. Effect of superconducting contacts on transport in δ-layers

6.6 Magnetoresistance and nonlocal measurements


on Ge:P δ-layers with superconducting contacts
We also study the magnetoresistance and nonlocal transport in low density Ge:P δ-layers
with superconducting Al contacts to see if the effect is limited only to Si:P devices.
We find that superconductivity in the contacts affects the magneto-transport, resistance
fluctuations and nonlocal resistance of Ge:P δ-layers as well. A schematic of a Ge:P
δ-layer with multiple voltage probes is shown in Fig. 6.13a. The carrier density n is
∼ 3×1013 cm−2 . The magnetoresistance in a parallel magnetic field is shown in Fig. 6.13b
and it displays large hysteresis within ±35 mT. At B∥ ∼ ±35 mT the resistance drops in a

(a)

(b) (c) 0
17.5 I L2J2 VF1J1
4.2K
20mK
-300
RXX(kW)

RNL (W)

1.2K
17.0 280mK

-600 20mK

16.5
-900
-0.1 0.0 0.1 -0.1 0.0 0.1
B||(T) B^(T)

Figure 6.13: (a) Schematic of a multiprobe hall bar of a Ge:P δ-layer. The contacts
marked NB were not bonded and hence not available for measurements. (b) The mag-
netoresistance in a parallel magnetic field B∥ for a low density Ge:P δ-layer at 20 mK.
The hysteresis is evident near the zero field region. The red and black arrows indicate
magnetic field sweep directions. (c) The nonlocal resistance RN L as a function of perpen-
dicular magnetic field at temperatures ranging from 20 mK to 4.2 K.
6.7. Highly doped devices with superconducting contacts 157

cusp like fashion similar to Si:P δ-layers. There are additional sharp dips in resistance as
the field reduces to zero. Note that in case of Si:P δ-layer we observed a peak in resistance
at 0 T, whereas in case of Ge:P δ-layers we have observed a minimum in resistance at
0 T. The device also shows a nonlocal resistance when the contacts are superconducting
as shown in Fig. 6.13c.

6.7 Highly doped devices with superconducting con-


tacts
The effect of superconducting contacts is seen only for low doped devices. For highly
doped devices with superconducting contacts we see the usual weak localization behavior

(a) (b) 1416


Ge:P 283mK Si:P 30mK
1905 -2 -2
1.35 e 14cm 2.2e14 cm

1410
RXX (W)

RXX (W)

1890

1404

1875
1398
-0.5 0.0 0.5 -0.5 0.0 0.5
B^(T) B^(T)

(c) 80 Si:P
(d) (e) -80
20mK Ge:P 20mK Ge:P 20mK
-2 -800 -2 -2
n = 5e13 cm n=3.2e13 cm n=4.6e13 cm
RNL (W)

40 -40
-400

0 0
0
-0.05 0.00 0.05 -0.1 0.0 0.1 -0.5 0.0 0.5
B^(T) B^(T) B^(T)

Figure 6.14: Magnetoresistance in a perpendicular magnetic field B⊥ for (a) Ge:P δ-layer
with carrier density n = 1.35 × 1014 cm−2 at 283 mK (b) Si:P δ-layer with carrier density
n = 2.2×1014 cm−2 at 30 mK. The nonlocal resistance RN L as a function of perpendicular
magnetic field at 20 mK for (c) Si:P δ-layer with density n = 5×1013 cm−2 (d) Ge:P δ-layer
with n = 3.2 × 1013 cm−2 and (e) Ge:P δ-layer with density n = 4.6 × 1013 cm−2 .
158 Chapter 6. Effect of superconducting contacts on transport in δ-layers

in perpendicular magnetic field as shown in Fig. 6.14a and b respectively for a highly doped
Si:P and Ge:P device respectively. The nonlocal resistance as a function of magnetic
field for three devices are shown in Fig. 6.14c (low density Si:P δ-layer with n ∼ 5 ×
1013 cm−2 ), 6.14d (low density Ge:P δ-layer with n ∼ 3 × 1013 cm−2 )) and 6.14e (medium
density Ge:P δ-layer with n ∼ 5 × 1013 cm−2 ). It is evident that for low doped Si:P and
Ge:P δ-layers the nonlocal resistance is much larger as compared to the medium density
Ge:P δ-layer. These observations lead us to conclude that the features observed due
to superconducting contacts is intimately connected to the enhanced interaction effects
at lower density which was also responsible for a spontaneous breaking of time reversal
symmetry in low doped devices.

6.8 Discussions

The onset of superconductivity in the Al contacts leads a number of peculiar observations


like the resistance peak at lower temperatures, the unusual magnetoresistance accom-
panied by hysteresis, enhanced noise magnitude in the superconducting regime and the
existence of a finite nonlocal resistance. All these features are present only for low doped
devices in the regime where there is a breakdown of time reversal symmetry due to strong
on-site Coulomb interaction. Highly doped devices do not show any of the above men-
tioned features irrespective of whether the contacts are superconducting or not. In Chap-
ter 5 we have shown that local moments are formed for devices with low doping density.
This leads us to conclude that the unusual features features described in this chapter is
intimately connected to the interplay of superconductivity and the local moments. Infact
the similarity of the temperature dependence of resistance and magnetoresistance with
that observed in ferromagnetic wires and topological insulators with superconducting con-
tacts point towards the possibility of spin-polarized transport in the Si:P δ-layer. It has
been shown that the quasiparticle spin lifetime in superconducting Al is about a million
times longer than in the normal state [253]. Such slow kinetics in the superconducting
state can explain the hysteresis that has been observed in magnetoresistance as well as
enhanced low frequency noise in the superconducting regime. Additionally, it has been
reported that superconducting Al films can have strong spin-orbit coupling [254]. Hence
6.9. Summary of the results 159

the possibility of an effective spin-orbit coupling in the δ-layers leading to the observed
nonlocality should be explored further. It has been predicted that there is a spontaneous
parity breaking in spin-orbit coupled superconductor-ferromagnet heterostructure [255].
The parity violation occurs due to a combination of the in-plane magnetization (exchange
field) and spin-orbit field. Such parity violation can also lead to observed switching of
the resistance peak at B = 0 on interchanging the contact configuration and nonlocal
transport in the δ-layers. However further theoretical and experimental work is required
to arrive at a coherent picture that can explain all the observations. If indeed there is spin
transport or edge state transport that gives rise to nonlocal behavior then a measurement
of thermopower will be crucial as it can probe the spin entropy. A systematic dependence
of non-local resistance for different sizes and different distances from the current path
would also be helpful in understanding the nature of transport. Further, to gauge the
influence of Al contacts, contacts made of other superconducting materials like Pb or
tungsten could be employed to study the effect of superconductivity in the δ-layers.

6.9 Summary of the results


1. We have investigated the effect of superconducting contacts on the transport in Si:P
and Ge:P δ-layers.

2. As the temperature is reduced below the transition temperature of Al, a sharp


increase and an anomalous peak in resistance is observed.

3. Magnetoresistance measurements reveal a departure from conventional weak local-


ization behavior when the contacts become superconducting. Also hysteresis was
observed in the magnetoresistance in both perpendicular and parallel field in the
region where the contacts were superconducting.

4. At the lowest temperature ∼ 20 mK, the noise magnitude was enhanced by five to
six orders of magnitude at B = 0 T as compared to high B > Bc

5. Finite non-local resistance was measurement in both Si:P and Ge:P δ-layers at
B < Bc and T < Tc .
160 Chapter 6. Effect of superconducting contacts on transport in δ-layers

6. All the above mentioned effect of superconducting contacts were observed only in
the low doped devices. Highly doped devices show the conventional weak localiza-
tion magnetoresistance, low noise magnitude and negligible nonlocal resistance even
when the contacts were superconducting.
Chapter 7

Conclusions and Outlook

7.1 Conclusions

In this thesis we have studied the electrical transport in δ-doped semiconductors (Si and
Ge) in two, one and zero dimensions. Apart from conductivity measurements we have
employed low frequency noise measurements to study the conductivity fluctuations and
gain more insight into the nature of transport in these devices. The thesis addresses these
devices from an application perspective as well as fundamental and exiting physics. On
technological grounds, we have looked into the noise magnitude and origin of noise in
these δ-doped devices and compared it with other semiconductors. Our investigation of
noise gains significance form the fact that P-doped Si systems are one of the standard ma-
terials for the semiconductor industry and δ-doped Si are aimed to be active components
in integrated circuits for quantum computation. Apart from application perspective, our
work has also looked into the quantum transport in these devices as the interaction effects
are enhanced at low doping density. We have investigated the fundamental time reversal
symmetry of the hamiltonian, existence of local moments and coupling with the super-
conductivity for device where the effective on-site Coulomb interaction (U/γ, where U is
the on-site Coulomb interaction and γ is the overlap integral) is large. Below we list the
main results and conclusions of individual chapters of this thesis.
In Chapter 3 we have studied the conductivity and low frequency noise in two dimen-
sional P δ-layers in Si and Ge. The conductivity decreases logarithmically with decreasing
temperature at low temperatures which indicates the presence of weak localization and

161
162 Chapter 7. Conlusions and Outlook

electron-electron interaction effects. We have measured the low frequency noise in these
systems and found that the devices obey the phenomenological Hooge relationship, with
a power spectrum that goes as 1/f α , where the frequency exponent α ∼ 1 − 1.2 within the
experimental bandwidth over the entire range of temperature and magnetic field studied
here. The noise magnitude quantified by the Hooge parameter was found to be ∼ 10−6
for the degenerately doped Si:P δ-layers which is about five to six orders of magnitude
lower than that of bulk Si:P systems in the metallic regime and is one of the lowest values
reported for doped semiconductors. The noise magnitude is strongly dependent on mobil-
ity and is larger in devices with lower mobility. It follows a power law relation (∝ 1/µ3 )
which indicates noise due to charge traps as has been observed for graphene on Si/SiO2
substrate. Hence, we believe that charge traps in the Si/SiO2 are the likely sources of
disorder which cause low frequency noise. At low temperatures, the noise was found to
increase as the temperature decreased, the functional dependence being 1/T indicating
that the noise in these devices arise due to universal conductance fluctuations.

Chapter 4 extends the noise measurements to δ-doped devices in one and zero dimen-
sions. Random telegraphic noise (two step process) was observed in the 1D wires and
0D quantum dots which led to a Lorentzian power spectrum proportional to 1/f α where
α ≈ 2 at higher frequencies in contrast to the 2D δ-layers which showed a 1/f spectrum.
Thus we conclude that as the dimensionality of the device decreases from 2D to 1D or 0D,
the number of active traps reduces from many to one or few. The variation of noise with
gate voltage suggests that conductivity fluctuations in such devices arise due to popula-
tion and depopulation of charge traps which causes a fluctuating potential at the devices.
Similar to the 2D δ-layers, we attribute the noise in these devices to the Si/SiO2 charge
traps. We have calculated the voltage fluctuation on the wire when a charge trap gets pop-
ulated or depopulated and compared it to the value estimated from noise measurements.
The agreement between the two verifies the model of noise presented. Measurements on
a quantum dot device which had a SiO2 dielectric deposited on the top Si along with a
metallic top gate, indicates that these devices are extremely sensitive to the surface states.
As a function of top gate, the devices we found to more unstable than an in-plane gated
quantum dot without a top gate. We estimate the Hooge parameter of the 1D wires to
be < 10−6 at positive gate voltages at 4.2 K which is smaller when compared to other 1D
7.1. Conclusions 163

systems such as metal nanowires or carbon nanotubes. A low value of Hooge parameter
for the 2D δ-layers and 1D wires is technologically desirable.

In Chapter 5 we have shown evidence of spontaneous breaking of time reversal sym-


metry in 2D Si:P and Ge:P δ-layers. From the magnetoconductivity measurements in a
perpendicular magnetic field, quantum correction to the conductivity was estimated for
devices with varying doping density and compared to the weak localization correction
expected for a conventional 2D disordered metal. It was observed that for low doped
devices, where the interaction effects are enhanced, the conductivity correction was less
than expected and the weak localization effects are suppressed. Additionally we have seen
that the universal conductance fluctuations were also suppressed as the doping density
was reduced (or alternatively effective dopant separation was increased) which provided
unambiguous evidence of spontaneous breaking of time reversal symmetry in the strongly
interacting regime. In an in-plane magnetic field, a negative magnetoconductivity was
observed near the zero field for low doped devices which suggests that local moments
driven spin fluctuations break time reversal symmetry. We have also seen metallic like
behavior at low temperatures (< 500 mK) where the resistivity decreases with decreasing
temperature for a very low density device (n ∼ 2 × 1013 cm−2 ).

In Chapter 6 we have investigated the effect of superconducting contacts (made by


Aluminium) on transport in two-dimensionally doped Si:P and Ge:P δ-layers. For low
doped devices, at low temperatures, even in a four terminal measurement of resistance
(where the contact resistance is eliminated) and with device dimensions orders of magni-
tude larger than the coherence length, the signature of superconductivity was observed.
As the temperature is reduced below the critical temperature of Al, the four probe resis-
tance of the δ-layer increased sharply and an anomalous peak in resistance is observed.
Magnetoresistance measurements reveal a departure from conventional weak localization
behavior when the contacts become superconducting. Also hysteresis was observed in the
magnetoresistance in both perpendicular and parallel field in the region where the con-
tacts were superconducting. The observation of two critical fields in magnetoresistance
and two critical temperatures in the temperature dependent measurements suggests an
unconventional superconductivity in the δ-layers in the proximity to the contacts. The
noise magnitude was enhanced by five to six orders of magnitude in the superconducting
164 Chapter 7. Conlusions and Outlook

regime. The observation of nonlocal resistance in the devices with superconducting con-
tacts indicates the possibility of edge state transport or spin transport due to an effective
spin-orbit interaction. The fact that all these effect of superconducting contacts are ob-
served only in low doped devices suggests that it is intimately connected to interaction
effects which is also responsible for breakdown of time reversal symmetry.

7.2 Scope of future work

7.2.1 Measurements on gated hall bars of Si:P δ-layers

We have seen in this thesis that for two dimensional δ-layers, as the doping density reduces,
interaction effects lead to breakdown of the fundamental time reversal symmetry. The
metallic like behavior in the temperature dependence of resistance was also observed for
the lowest density (∼ 2×1013 cm−2 ) device that we measured. Additionally, the unusually
high noise, anomalous magnetoresistance and non-local transport due to superconducting
contacts were observed only for low doped devices. It is thus clear that the devices with
low doping have a lot to offer in terms of exotic physics. For the devices measured in this
thesis, different carrier density was achieved by changing the doping density of P atoms.
This is a time consuming process and the density range and values that achieved are rather
limited. For example, the lowest density device which showed electrical conductivity was
∼ 2 × 1013 cm−2 , below which the contacts were simply not working. Also it was not
possible to do measurements (say for example the quantum correction to conductivity)
as a continuous function of density.
The difficulty in gating these δ-layers is the high carrier density (∼ 1013 − 1014 cm−2 )
compared to other 2D systems such as graphene (∼ 1011 − 1012 cm−2 ) or GaAs/AlGaAs
heterostructures. However a possible direction in efforts to gate these 2D δ-layers would
be to put a high-k dielectric such as titanium oxide or Strontium titanate which has
dielectric constants ∼ 100 and 1000 respectively. Since it will still be very difficult to gate
saturation doped δ-layers (n ∼ 2 × 1014 cm− 2), it is wise to start with a low density hall
bar (say n ∼ 4 − 5 × 1013 cm− 2) and then use the gate with a high-k dielectric to fine
tune the density to investigate effects such as quantum correction to conductivity in a
controlled manner. Also such tunability of density would allow us to study the metal-to-
7.2. Scope of future work 165

insulator transition in two dimensionally doped semiconductors. The interplay of disorder


and interaction along with the possibility of a metallic state at low temperatures can be
looked into with much more control.

7.2.2 Thermal transport in δ-layers


7.2.2.1 Thermopower

The thermopower or Seebeck coefficient is defined as

VT h
S = lim (7.1)
∆T →0 ∆T

where VT h is the thermal voltage generated due to a temperature difference ∆T between


two ends of a system. The thermovoltage is generated solely due to the temperature
difference and not due to any charge current. Since it involves flow carriers from hot end
to cold end there is a heat energy transfer and hence an entropy transfer. Infact it can be
shown that S = S/ne where S is the entropy [256]. Hence thermopower is a measure of
entropy per carrier. This particular observation has been employed to study spin entropy
transfer by measuring thermopower [257, 258]. Low density Si:P δ-layers form a strongly
correlated system in 2D where we have observed spontaneous breakdown of time reversal
symmetry driven by spin fluctuations. If at lower densities or in presence of superconduct-
ing contacts we have spin currents in the system, then thermopower measurement can be
used to probe the spin entropy transfer. The thermopower as a function of magnetic field
should go down with increasing magnetic field as the spins will be aligned in the direction
of field.

7.2.2.2 Thermal conductivity

For P-doped semiconductors the impurity band is half-filled and at low densities there
is large on-site Coulomb interaction leading to formation of local moments. Any kind
of magnetic ordering is prevented by random lattice structure. These points give us a
strong reason to look for quantum spin liquids in δ-doped Si [55]. In such a scenario, heat
transport can reveal more insights about the nature of states in the low density Si:P and
166 Chapter 7. Conlusions and Outlook

Ge:P δ-layers. For a quantum spin liquid the electrical and thermal metal-to-insulator
transition (MIT) occur separately. Hence when the electrical MIT occurs making the
system an insulator, the electrical conductivity vanishes but the thermal conductivity
remains metallic. Hence thermal conductivity measurements can probe such a state. In
fact Senthil et al. says “We suggest that the clearest test for spin liquid behavior in doped
semiconductors would come from a careful study of thermal transport near the MIT. A
spinon Fermi liquid would lead to thermal conductivity proportional to T for low T , which,
if observed, would strongly indicate the presence of gapless fermionic excitations” [55]

7.2.3 Dopants arranged in a periodic lattice


In all the two dimensionally doped δ-layers that have been investigated in this thesis, the
phosphorus dopants atoms were randomly distributed. There is no well defined order or
periodicity in the arrangement of dopant atoms. It will be very interesting to investigate
devices where the dopants are arranged in a periodic manner, for example in a square
lattice. In the tight binding approximation, the Fermi surface in a 2D square lattice
displays global nesting and an in-built particle-hole symmetry. The Fermi energy now
lies in the center of the band and that can fundamentally alter or suppress the electron
localization [259]. Coexistence of weak localization and weak anti-localization has been
predicted for a half-filled 2D crystal lattice [260–262]. Along with the normal diffusion
modes (that causes weak localization), a non-trivial diffusive mode also appears via an
Umpklapp scattering at impurities which are protected against backscattering. The fact
that Si has been substituted with the adjacent group phosphorous atoms, leads to purely
off-diagonal disorder with nearest neighbour hopping [261, 263]. Such a system falls into
a symmetry class which is different from the usual Wigner-Dyson classification. It corre-
sponds to one of the chiral symmetry classes [264], as a consequence of which the system
can have very unusual properties like non-ohmic behavior and nonlocal transport.
Appendix A

Annealed Silver as Cold finger

The cold finger of a dilution refrigerator must be made of a material with very high
thermal conductivity. Both copper and silver have very high thermal conductivity. But
copper has a large nuclear magnetic moment and it can get heated up as the magnetic
field is ramped up due to transfer of entropy from the spins. Thus it is not an ideal choice
for making cold finger if magnetic field sweeps are performed. Silver on the other hand,
does not have a large nuclear magnetic moment and hence is a better choice for cold
finger material than copper for magnetic field experiments. The only drawback is that
its much more expensive than copper. The thermal conductivity of as purchased silver
can be increased by about two to three orders of magnitude by annealing it at a high
temperature. Annealing leads to an increase in grain size with the impurities migrating
to the grain boundaries when silver is close to its melting temperature. Hence it is very
important to anneal the silver before it can be used as cold finger.

A.1 Weidemann-Franz law


The thermal conductivity of any material is related to its electrical conductivity by the
Weidemann-Franz Law which is

( )
κ 3 kB
= T (A.1)
σ 2 e

where κ and σ are the thermal and electrical conductivity respectively. Thus the thermal
conductivity of a material can be calculated by measuring the electrical conductivity. A

167
168 Appendix A. Annealed Silver

way by which the thermal conductivity can be estimated is by measuring the residual
resistance ratio (RRR) defined as the ratio R300K /R4K . From Ref. [151] we get that the
thermal conductivity can be written as

RRR
κ= T W/Km (A.2)
0.55

A.2 Annealing procedure

(a)

(b)

Un-annealed

Annealed 5 mm

Figure A.1: (a) The setup showing the annealing furnace and the quartz tube for annealing
the silver rod. (b) Comparison of the grain size for as-purchased and annealed silver. It
is clear that for the annealed rod the grain sizes are visible and are ∼ 1 − 2 mm.

In this section we describe the procedure that was followed to anneal the silver to
increase its thermal conductivity. The silver rod was cleaned with Isopropyl alcohol and
A.2. Annealing procedure 169

and placed in a long quartz tube (closed at one end), which is then placed inside an electric
furnace. The tube is evacuated with a turbo molecular pump. The temperature is then
ramped till 900◦ C at the rate of 10◦ C min−1 . This temperature is maintained for about
36 hours, after which it is ramped down to room temperature at the rate of 20◦ C min−1 .
During the entire annealing process, the pressure was ∼ 5 × 10−7 mbar. The annealing
setup is shown in Fig. A.1a and the comparison of as-purchased and annealed silver is
shown in Fig. A.1b. It is clear that the annealed rod have grains with size ∼ 1 − 2 mm,
which is much larger than the unannealed one.
An estimate of the RRR value after annealing was found by measuring the temperature
dependence of a test piece of silver which was annealed with the same parameters and in
similar conditions as the silver rod. While the RRR for the as-purchased silver was ∼ 5,
after annealing it increased to ∼ 500 indicating that the thermal quality of the sample
has increased massively.
An important point worth mentioning here is that after annealing since the grain sizes
increase, the mechanical strength of silver goes down. It becomes fragile and can be bent
easily and hence it has to be handled very carefully. All machining and threading were
done prior to annealing. After annealing, the silver rod was fixed to the mixing chamber
and wiring was completed as described in Chapter 2.
170 Appendix A. Annealed Silver
Appendix B

Preliminary Thermopower
measurements

In this Appendix we will discuss the results of preliminary thermopower measurements.


These devices were not designed for any thermal measurements and hence the geometry
is not ideal for any quantitative estimation of thermopower. However the qualitative
features can be obtained with present device geometry which can shed some light on
thermal transport in these systems. The measurements are by no means complete but
the main motivation of doing this is to get some basic understanding of heat transport in
these systems and then design devices specifically for thermal measurements.

B.1 Introduction to thermopower


Consider a system (say a Si:P δ-layer) whose two ends are maintained at two different
temperatures T and T + ∆T (by say heating one end with electric current or any other
method) as shown in the schematic in Fig. B.1. Since two ends are at different temper-
atures, they have different Fermi-Dirac distributions as a result of which carriers diffuse
from hot end to the cold end, giving rise to a voltage difference between the two ends.
This voltage generated as a result of temperature difference between two ends is called
thermovoltage VT h . The thermopower or the Seebeck Coefficient can be defined as

VT h
S = lim (B.1)
∆T →0 ∆T

The thermopower is related to the logarithmic derivative of conductivity by the Mott

171
172 Appendix B. Preliminary Thermopower measurements

Cold Hot

T T+∆T
Vth

Figure B.1: Schematic showing a system whose two ends are maintained are two differ-
ent temperatures T and T + ∆T . A thermovoltage is developed by the virtue of the
temperature difference.

relation [265, 266] which is

π 2 kB
2
T d ln σ(E)
S=− (B.2)
3|e| dE E=EF

where the derivative is evaluated at E = EF , with EF being the Fermi energy. Being
proportional to the derivative of conductivity, thermopower is a more sensitive probe to
the transport properties of system than the conductivity itself. From the above equation
we can estimate the thermopower to be [267]

π 2 kB
2
T (p + 1)
S=− (B.3)
3|e|EF

where p is of the order of unity. From Eq. B.3 we can see that the thermopower is
inversely proportional to the EF and hence density. Since the Si:P δ-layers have very high
densities (∼ 1013 − 1014 cm−2 ) as compared to other 2D systems such as graphene and
GaAs/AlGaAs heterostructures, its is expected that the thermopower in these systems
will be very small and challenging to measure.
B.2. Measurement of Thermovoltage 173

B.2 Measurement of Thermovoltage

In this section we discuss thermovoltage measurements on a saturation doped Si:P δ-


layer Si HD having a carrier density n ∼ 2.5 × 1014 cm−2 . The circuit employed for
measuring the thermovoltage in the Si:P δ-layer is shown in Fig. B.2a. An ac current
I = I0 sin(2πf t) at a frequency f is passed through the hall probes H2K1 which sets up a

(a)

(b) (c)

4.2 K 60 Base T
100
40 2
2
I
-Vth (nV)

I
I
20
10

1 2 3 4 5 0.3 0.6 0.9 1.2


I (mA) I (mA)

Figure B.2: (a) The circuit used for measuring the thermovoltage VT h in a 2D Si:P δ-
layer. An ac current at frequency f is passed through any of the hall probes (H2K1 in this
case) and VT h is measured across two longitudinal probes away from the current probes
at frequency 2f . VT h as a function of heating current I at (b) 4 K and (c) Base T. The
dashed red line shows a dependence on I 2 and the dashed blue line shows a dependence
on I.
174 Appendix B. Preliminary Thermopower measurements

temperature gradient along the length of the hall bar. The thermovoltage was measured
across J2K2. The thermovoltage is proportional to the temperature difference which in
turn is proportional to the heat generated. The heat is proportional to I 2 = I02 sin2 (2πf t).
Hence VT h is measured at frequency 2f . The proportionality to I 2 is seen clearly at
temperature T = 4.2 K as shown by the dashed red line in Fig. B.2b. At base temperature
(100 mK) however we see that VT h ∝ I for the entire range shown shown in Fig. B.2c.
The proportionality to I 2 exists if at all towards very low currents only in a very limited
range (first two to three points). The reason for such departure can be due to ∆T ∼ T .
One of the assumptions in Eq. B.1 is that ∆T ≪ T which was true at 4.2 K as seen by
the I 2 dependence. However as the temperature is reduced, VT h also reduces as evident
from Eq B.2, making the measurement of VT h very difficult. To get measurable VT h , we
had to pass large heating current leading to an increase in ∆T such that the assumption
∆T ≪ T no longer holds true. Hence we observe the departure from the expected I 2
dependence in Fig. B.2c.
We have measured VT h as a function of perpendicular magnetic field B⊥ at three
different temperatures, 1.55, 2.1 and 4.2 K. The result is plotted in Fig. B.3a. Since
the resistance as a function of B⊥ shows weak localization behavior (Chapter 5), the
thermovoltage shows signatures of weak localization as a dip in VT h at B⊥ = 0 T with the
dip becoming stronger at lower temperatures. Note that we could not do thermopower
measurements at very low temperatures as the thermovoltage becomes very small. Since
from Eq. B.2 we know that the thermopower is proportional to the logarithmic derivative
of conductivity, we compare the measured B⊥ -dependence of thermovoltage with the

(a) (b)
-60 4.2K -70
50
d lnR/dB^
-VTh (nV)
-VTh (nV)

100 -140
-90

150
1.55 K -210
2.15 K -120
4K
200
-2 0 2 -2 0 2
B^(T) B^(T)

Figure B.3: (a) The thermovoltage VT h as a function of perpendicular magnetic field B⊥


at different temperatures. (b) Comparison of VT h as function of B⊥ with d ln(R)/dB⊥
B.3. Estimation of thermopower using Mott’s relation 175

logarithmic derivative of resistance R in Fig. B.3b. It can be seen that the two curves
(black and red) agree with each other qualitatively.

B.3 Estimation of thermopower using Mott’s rela-


tion
The quantitative determination of thermopower requires that the temperature difference
∆T be known (either measured experimentally or determined theoretically). Conven-
tional measurements of thermopower in systems such as graphene and GaAs/AlGaAs
heterostructures involves putting extra temperature sensors at different regions of the
2D system to determine the temperature. Here we try to measure thermopower without
using any extra temperature sensors. Since the resistance of the Si:P δ-layers varies log-
arithmically with temperature due to weak localization and electron-electron interaction
effects, they can also act as temperature sensors once they have been calibrated. We use
this concept to determine the temperature variation along the length of the hall bar when
once end of it is heated by an ac current.
As discussed earlier the average temperature of a region is estimated by measuring the
resistance of that region and comparing it with the calibration curve recorded earlier. It
is essential that the resistance is measured at the same time when the heating current is
flowing. Figure B.4a shows the measurement circuit which has been used for determination
of temperature of various region of the δ-layer. An ac current heats one end of the δ-
layer (H2K1). Since the resistance needs to measured simultaneously, we pass a small
ac current Is (≪ I, the heating current) at frequency fs which is different from f , the
frequency of the heating current, using contacts C1 and K1 as the I+ and I− probes
respectively. In this circuit, the system has only one ground. Using this configuration
the resistance of different regions is recorded initially at different temperatures for I = 0
so that there is no heating. Thus the calibration curve is obtained. Then the heating
current is passed through the hall probes and the following quantities are measured: (i)
The thermovoltage VT h at 2f , (ii) The sample bias Vs at frequency fs across K2J2 to
determine temperature T1 (Fig. B.4b) of the region where the VT h is measured, and (iii)
The sample bias Vs at frequency fs across L2K2 to determine temperature T2 (Fig. B.4b)
of the cold end. Since this device was not specifically made for thermal measurements,
all the contacts were not bonded. Hence we were restricted by the number of contacts
to determine the temperature gradient along the length of the hall bar. For, example we
176 Appendix B. Preliminary Thermopower measurements

(a)

(b) (c)
241.5
K2J2
R (W)

241.2

240.9

0 1 2
T (K)

Figure B.4: (a) The circuit used for measuring the temperature of different regions of the
Si:P δ-layer. (b) Average temperature at different regions of the δ-layer (c) Resistance of
the region K2J2 as a function of temperature T . This acts as the calibration curve which
will be used for temperature estimation.

could not determine the magnitude of T0 (Fig. B.4b) for the present case. An example of
a calibration curve of the region K2J2 is shown in Fig. B.4c.
Since we were unable to determine the absolute magnitude of ∆T, we adopt a different
method to analyze the thermopower data. We calculate the thermopower S from Eq. B.3
assuming the Mott relation to hold true for these systems. Then knowing S from Mott’s
relation and the experimentally measured VT h , we calculate ∆T from Eq. B.1 and hence
calculate the temperature of the hot end as T0 = T2 + ∆T and see if the values are
reasonable.
B.4. Thermopower for low density Si:P δ-layer 177

Table B.1: Table showing the thermopower SM ott calculated using Mott’s relation, the
sample temperature T1 and temperature of cold end T2 estimated from their respective
calibration curves. The temperature difference ∆T and the temperature of the hot end T2
is calculated from the measured VT h and theoretically calculated S from Mott’s relation.

T1 (K) T2 (K) −VT h (nV K−1 ) −SM ott ∆T (K) T0 (K)


1.03 0.66 21.2 27.54 0.77 1.8
1.5 0.81 46.1 40.11 1.15 2.65
1.72 0.98 76.3 46 1.66 3.38
1.84 1.08 124.8 49.2 2.54 4.38

The results are presented in Table B.1. As can be seen the values of ∆T and T0
are reasonable, although not ideal for Mott’s relation to be valid. As said earlier here
we haven’t performed an ideal thermopower measurement. We have performed some
preliminary measurements and presented a method by which the thermopower can be
measured without involving any extra sensors.

B.4 Thermopower for low density Si:P δ-layer


We have measured the thermopower of a low density Si:P δ-layer Si LDL having a carrier
density n ∼ 2 × 1013 cm−2 . This is the device discussed in Chapter 5 Section 5.5, whose
resistance decreased on decreasing temperature below ∼ 200 − 300 mK. This device struc-
ture is exactly like the one shown in Fig. B.4a, having multiple probes which enables us
to determine ∆T and hence S experimentally.
The dependence of thermopower on temperature is shown in Fig. B.5. We can observe
the following:

1. At ∼ 4.6 K, S changes sign becoming positive which is surprising as the sign of


S is determined by the sign of charge carrier which in case of Si:P δ-layers are
electrons. Hence we expect S to be negative. However it is possible that certain
energy dependent scattering processes are taking place, for example, low energy
carriers (which are moving toward the hot reservoir) encounter less scattering than
the high energy carriers (which are moving away from the hot reservoir) thus leading
to opposite sign of S. Since these are extremely low doped devices (in the regime
178 Appendix B. Preliminary Thermopower measurements

60
S_Measured
S_Mott

-S (mVK )
40
-1

20

4.5 5.0 5.5


T (K)

Figure B.5: The thermopower or the Seebeck coefficient S as a function of temperature


for low density Si:P δ-layer Si LDL having a carrier density n ∼ 2 × 1013 cm−2 . The filled
black squares are experimental values while filled red circles estimated using Eq. B.3.

where time reversal symmetry is broken), we expect local moments to formed at the
dopant sites (due to strong effective on-site Coulomb interaction) which can cause
spin-flip scattering. A non-monotonous temperature dependence of S accompanied
by a sign change, has previously been reported due to spin-flip scattering in Kondo
lattice compounds, [268, 269]. Whether a similar mechanism is responsible for sign-
reversal in S for low-doped Si:P δ-layers requires further investigation.

2. As the temperature increases above 5 K, S becomes negative again and keeps in-
creasing with increasing temperature finally saturating at ∼ 5.5 K. It is surprising
that S saturates at ∼ 60 µWK−1 which is the quantum of thermopower [(kB /e) ln 2]
defined as the entropy per unit charge [256].

3. At the highest temperature (5.5 K), the measured S (filled black squares) is an order
of magnitude large than the value expected from Mott’s relation (filled red circles)

It is worth mentioning here that the temperature of the δ-layer was increased by in-
creasing the heating current. The first thermopower measurements shown some interesting
features as shown above. The high density device Si HD appears to follow Mott’s relation
whereas the low density device Si LDL shows a deviation. As we have already seen that
increased Coulomb interaction in low density devices has consequences on the nature of
B.4. Thermopower for low density Si:P δ-layer 179

transport and fundamental symmetry of Hamiltonian such as the time reversal symmetry,
it is not surprising that thermal measurements show deviation from the Mott’s relation
for diffusive thermopower. However this just the first measurement and any concrete
conclusion will require further detailed theoretical and experimental investigations.
180 Appendix B. Preliminary Thermopower measurements
Bibliography

[1] Mott, N. F. The metallic and nonmetallic states of matter (Ed. P. P. Edwards and
C. N. R. Rao). Taylor & Francis, London (1985).

[2] Logan, D. E., Szezech, Y. H. & Tusch, M. A. Metal-insulator transitions revisited


(ed. P. P. Edwards and C. N. R. Rao),. Taylor & Francis, London (1985).

[3] de Boer, J. H. & Verwey, E. J. W. Semi-conductors with partially and with com-
pletely filled 3d-lattice bands. Proceedings of the Physical Society 49, 59 (1937).

[4] Mott, N. F. & Peierls, R. Discussion of the paper by de boer and verwey. Proceedings
of the Physical Society 49, 72 (1937).

[5] Mott, N. F. The basis of the electron theory of metals, with special reference to the
transition metals. Proceedings of the Physical Society. Section A 62, 416.

[6] Mott, N. F. On the transition to metallic conduction in semiconductors. Canadian


Journal of Physics 34, 1299–1314.

[7] Mott, N. F. The transition to the metallic state. Philosophical Magazine 6, 287.

[8] Mott, N. F. Metal-Insulator Transitions (1990).

[9] Mott, N. F. Metal-insulator transition. Rev. Mod. Phys. 40, 677–683 (1968).

[10] Anderson, P. W. New approach to the theory of superexchange interactions. Phys.


Rev. 115, 2–13 (1959).

[11] Hubbard, J. Electron correlations in narrow energy bands. II. The degenerate band
case. Proceedings of the Royal Society A: Mathematical, Physical & Engineering
Sciences 277, 237.

181
182 BIBLIOGRAPHY

[12] Hubbard, J. Electron correlations in narrow energy bands. III. An improved so-
lution. Proceedings of the Royal Society A: Mathematical, Physical & Engineering
Sciences 281, 401.

[13] Fazekas, P. Lecture Notes on Electron Correlation and Magnetism (World Scientific,
1999).

[14] Imada, M., Fujimori, A. & Tokura, Y. Metal-insulator transitions. Rev. Mod. Phys.
70, 1039–1263 (1998).

[15] Motome, Y. & Imada, M. Superfluid-insulator transition of interacting multi-


component bosons gutzwiller variational and quantum monte carlo study . Journal
of the Physical Society of Japan 65, 2135–2145 (1996).

[16] Motome, Y. & Imada, M. A quantum monte carlo method and its applications
to multi-orbital Hubbard models. Journal of the Physical Society of Japan 66,
1872–1875 (1997).

[17] Fujimori, A. et al. Evolution of the spectral function in Mott-Hubbard systems with
d 1 configuration. Phys. Rev. Lett. 69, 1796–1799 (1992).

[18] Inoue, I. H. et al. Systematic development of the spectral function in the 3d 1 Mott-
Hubbard system Ca1−x Srx Vo3 . Phys. Rev. Lett. 74, 2539–2542 (1995).

[19] Furukawa, N. & Imada, M. Two-dimensional hubbard model metal insulator tran-
sition studied by monte carlo calculation. Journal of the Physical Society of Japan
61, 3331–3354 (1992).

[20] Imada, M. Metal-insulator transition of correlated systems and origin of unusual


metal. Journal of the Physical Society of Japan 64, 2954–2969 (1995).

[21] Anderson, P. W. Absence of diffusion in certain random lattices. Phys. Rev. 109,
1492–1505 (1958).

[22] Abrahams, E., Anderson, P. W., Licciardello, D. C. & Ramakrishnan, T. V. Scaling


theory of localization: Absence of quantum diffusion in two dimensions. Phys. Rev.
Lett. 42, 673–676 (1979).

[23] Lee, P. A. & Ramakrishnan, T. V. Disordered electronic systems. Rev. Mod. Phys.
57, 287–337 (1985).
BIBLIOGRAPHY 183

[24] Kramer, B. & MacKinnon, A. Localization: theory and experiment. Reports on


Progress in Physics 56, 1469 (1993).

[25] Altshuler, B. L., Aronov, A. G. & Lee, P. A. Interaction effects in disordered Fermi
systems in two dimensions. Phys. Rev. Lett. 44, 1288–1291 (1980).

[26] Tanatar, B. & Ceperley, D. M. Ground state of the two-dimensional electron gas.
Phys. Rev. B 39, 5005–5016 (1989).

[27] Dolan, G. J. & Osheroff, D. D. Nonmetallic conduction in thin metal films at low
temperatures. Phys. Rev. Lett. 43, 721–724 (1979).

[28] Van den dries, L., Van Haesendonck, C., Bruynseraede, Y. & Deutscher, G. Two-
dimensional localization in thin copper films. Phys. Rev. Lett. 46, 565–568 (1981).

[29] Bishop, D. J., Tsui, D. C. & Dynes, R. C. Nonmetallic conduction in electron


inversion layers at low temperatures. Phys. Rev. Lett. 44, 1153–1156 (1980).

[30] Uren, M. J., Davies, R. A. & Pepper, M. The observation of interaction and lo-
calisation effects in a two-dimensional electron gas at low temperatures. Journal of
Physics C: Solid State Physics 13, L985 (1980).

[31] Finkel’stein, A. Weak localization and coulomb interaction in disordered systems.


Zeitschrift fr Physik B Condensed Matter 56 (1984).

[32] Castellani, C., Di Castro, C., Lee, P. A. & Ma, M. Interaction-driven metal-insulator
transitions in disordered fermion systems. Phys. Rev. B 30, 527–543 (1984).

[33] Kravchenko, S. V., Kravchenko, G. V., Furneaux, J. E., Pudalov, V. M. & D’Iorio,
M. Possible metal-insulator transition at B =0 in two dimensions. Phys. Rev. B
50, 8039–8042 (1994).

[34] Kravchenko, S. V. et al. Scaling of an anomalous metal-insulator transition in a


two-dimensional system in silicon at B =0. Phys. Rev. B 51, 7038–7045 (1995).

[35] Kravchenko, S. V., Simonian, D., Sarachik, M. P., Mason, W. & Furneaux, J. E.
Electric field scaling at a B = 0 metal-insulator transition in two dimensions. Phys.
Rev. Lett. 77, 4938–4941 (1996).
184 BIBLIOGRAPHY

[36] Brunthaler, G., Prinz, A., Bauer, G. & Pudalov, V. M. Exclusion of quantum
coherence as the origin of the 2D metallic state in high-mobility silicon inversion
layers. Phys. Rev. Lett. 87, 096802 (2001).

[37] Simmons, M. Y. et al. Metal-insulator transition at B = 0 in a dilute two dimen-


sional GaAs-AlGaAs hole gas. Phys. Rev. Lett. 80, 1292–1295 (1998).

[38] Popović, D., Fowler, A. B. & Washburn, S. Metal-insulator transition in two di-
mensions: Effects of disorder and magnetic field. Phys. Rev. Lett. 79, 1543–1546
(1997).

[39] Hanein, Y. et al. Observation of the metal-insulator transition in two-dimensional


n-type GaAs. Phys. Rev. B 58, R13338–R13340 (1998).

[40] Hanein, Y. et al. The metalliclike conductivity of a two-dimensional hole system.


Phys. Rev. Lett. 80, 1288–1291 (1998).

[41] Vitkalov, S. A., Zheng, H., Mertes, K. M., Sarachik, M. P. & Klapwijk, T. M.
Scaling of the magnetoconductivity of silicon MOSFETs: Evidence for a quantum
phase transition in two dimensions. Phys. Rev. Lett. 87, 086401 (2001).

[42] Butko, V. Y. & Adams, P. W. Quantum metallicity in a two-dimensional insulator.


Nature 409, 161–164 (2001).

[43] Clarke, W. R. et al. Impact of long- and short-range disorder on the metallic
behaviour of two-dimensional systems. Nat. Phys. 4, 55–59 (2008).

[44] Coleridge, P. T., Williams, R. L., Feng, Y. & Zawadzki, P. Metal-insulator transition
at B = 0 in p-type SiGe. Phys. Rev. B 56, R12764–R12767 (1997).

[45] Lam, J., D’Iorio, M., Brown, D. & Lafontaine, H. Scaling and the metal-insulator
transition in Si/SiGe quantum wells. Phys. Rev. B 56, R12741–R12743 (1997).

[46] Yoon, J., Li, C. C., Shahar, D., Tsui, D. C. & Shayegan, M. Wigner crystallization
and metal-insulator transition of two-dimensional holes in GaAs at B = 0. Phys.
Rev. Lett. 82, 1744–1747 (1999).

[47] Mills, A. P., Ramirez, A. P., Pfeiffer, L. N. & West, K. W. Nonmonotonic


temperature-dependent resistance in low density 2D hole gases. Phys. Rev. Lett.
83, 2805–2808 (1999).
BIBLIOGRAPHY 185

[48] Papadakis, S. J. & Shayegan, M. Apparent metallic behavior at B = 0 of a two-


dimensional electron system in AlAs. Phys. Rev. B 57, R15068–R15071 (1998).

[49] Simmons, M. Y. et al. Weak localization, hole-hole interactions, and the “metal”-
insulator transition in two dimensions. Phys. Rev. Lett. 84, 2489–2492 (2000).

[50] Dolgopolov V. T., S. A. A. K. S. V., Kravchenko G. V. Properties of electron


insulating phase in Si inversion layers at low temperatures. JETP Lett. 55, 733
(1992).

[51] Simonian, D., Kravchenko, S. V., Sarachik, M. P. & Pudalov, V. M. Magnetic


field suppression of the conducting phase in two dimensions. Phys. Rev. Lett. 79,
2304–2307 (1997).

[52] Pudalov, V., Brunthaler, G., Prinz, A. & Bauer, G. Instability of the two-
dimensional metallic phase to a parallel magnetic field. Journal of Experimental
and Theoretical Physics Letters 65, 932–937 (1997).

[53] Lahoud, E., Meetei, O. N., Chaska, K. B., Kanigel, A. & Trivedi, N. Emergence
of a novel pseudogap metallic state in a disordered 2D Mott insulator. Phys. Rev.
Lett. 112, 206402 (2014).

[54] Abrahams, E., Kravchenko, S. V. & Sarachik, M. P. Metallic behavior and related
phenomena in two dimensions. Rev. Mod. Phys. 73, 251–266 (2001).

[55] Potter, A. C., Barkeshli, M., McGreevy, J. & Senthil, T. Quantum spin liquids
and the metal-insulator transition in doped semiconductors. Phys. Rev. Lett. 109,
077205 (2012).

[56] Ong, N. P. & Bhatt, R. N. More is Different: Fifty Years of Condensed Matter
Physics (Princeton University Press, 2001).

[57] Trivedi, N., Denteneer, P. J. H., Heidarian, D. & Scalettar, R. T. Effect of in-
teractions, disorder and magnetic field in the hubbard model in two dimensions.
Pramana 64, 1051–1061 (2005).

[58] Drude, P. Zur Elektronentheorie der Metalle; II. Teil. Galvanomagnetische und
thermomagnetische effecte. Annalen der Physik 308, 369–402 (1900).
186 BIBLIOGRAPHY

[59] Drude, P. Zur elektronentheorie der metalle. Annalen der Physik 306, 566–613
(1900).

[60] Klitzing, K. v., Dorda, G. & Pepper, M. New method for high-accuracy determi-
nation of the fine-structure constant based on quantized hall resistance. Phys. Rev.
Lett. 45, 494 (1980).

[61] Bergmann, G. Weak localization in thin films: a time-of-flight experiment with


conduction electrons. Physics Reports 107, 1 – 58 (1984).

[62] Beenakker, C. & van Houten, H. Quantum transport in semiconductor nanostruc-


tures. In Semiconductor Heterostructures and Nanostructures, vol. 44 of Solid State
Physics, 1 – 228 (Academic Press, 1991).

[63] Altshuler, B. L. & Aronov, A. G. Electron-Electron Interactions in Disordered


Systems, edited by Efros, A. L. and Pollak, M. (Elsevier, Amsterdam, 1985).

[64] Khmel’nitskii, D. Localization and coherent scattering of electrons. Physica B+C


126, 235 – 241 (1984).

[65] Vollhardt, D. Localization effects in disordered systems 27, 63–84 (1987).

[66] Altshuler, B. L., Aronov, A. G. & Khmelnitsky, D. E. Effects of electron-electron


collisions with small energy transfers on quantum localisation. Journal of Physics
C: Solid State Physics 15, 7367 (1982).

[67] Bergmann, G. Quantitative analysis of weak localization in thin Mg films by mag-


netoresistance measurements. Phys. Rev. B 25, 2937–2939 (1982).

[68] Eshkol, M., Eisenberg, E., Karpovski, M. & Palevski, A. Dephasing time in a
two-dimensional electron Fermi liquid. Phys. Rev. B 73, 115318 (2006).

[69] Hikami, A. I., S. Larkin & Nagaoka, Y. Spin-orbit interaction and magnetoresistance
in the two dimensional random system. Prog. Theor. Phys. 63, 707–710 (1980).

[70] Altshuler, B. L. Fluctuations in the extrinsic conductivity of disordered conductors.


Pis’ma Zh. Eksp. Teor. Fiz. 41, 530–533 (1985).

[71] Altshuler, B. L. & Spivak, B. Z. Variation of random potential and the conductivity
of samples of small dimensions. Pis’ma Zh. Eksp. Teor. Fiz. 42, 363–365 (1985).
BIBLIOGRAPHY 187

[72] Lee, P. A. & Stone, A. D. Universal conductance fluctuations in metals. Phys. Rev.
Lett. 55, 1622–1625 (1985).

[73] Birge, N. O., Golding, B. & Haemmerle, W. H. Electron quantum interference and
1
f
noise in bismuth. Phys. Rev. Lett. 62, 195–198 (1989).

[74] Birge, N. O., Golding, B. & Haemmerle, W. H. Conductance fluctuations and 1/ f


noise in Bi. Phys. Rev. B 42, 2735–2743 (1990).

[75] McConville, P. & Birge, N. O. Weak localization, universal conductance fluctua-


tions, and 1/f noise in Ag. Phys. Rev. B 47, 16667–16670 (1993).

[76] Ghosh, A. & Raychaudhuri, A. K. Universal conductance fluctuations in three


dimensional metallic single crystals of Si. Phys. Rev. Lett. 84, 4681–4684 (2000).

[77] Stone, A. D. Reduction of low-frequency noise in metals by a magnetic field: Ob-


servability of the transition between random-matrix ensembles. Phys. Rev. B 39,
10736–10743 (1989).

[78] Moon, J. S., Birge, N. O. & Golding, B. Observation of Zeeman splitting in universal
conductance fluctuations. Phys. Rev. B 53, R4193–R4196 (1996).

[79] Debray, P., Pichard, J.-L., Vicente, J. & Tung, P. N. Reduction of mesoscopic
conductance fluctuations due to zeeman splitting in a disordered conductor without
spin-orbit scattering. Phys. Rev. Lett. 63, 2264–2267 (1989).

[80] Mailly, D., Sanquer, M., Pichard, J.-L. & Pari, P. Reduction of quantum noise in a
GaAlAs/GaAs heterojunction by a magnetic field: an orthogonal-to-unitary wigner
statistics transition. EPL (Europhysics Letters) 8, 471 (1989).

[81] Goldhaber-Gordon, D. et al. Kondo effect in a single-electron transistor. Nature


391, 156–159 (1998).

[82] Johnson, J. B. Thermal agitation of electricity in conductors. Phys. Rev. 32, 97–109
(1928).

[83] Nyquist, H. Thermal agitation of electric charge in conductors. Phys. Rev. 32,
110–113 (1928).

[84] Paladino, E., Galperin, Y. M., Falci, G. & Altshuler, B. L. 1/f noise: Implications
for solid-state quantum information. Rev. Mod. Phys. 86, 361–418 (2014).
188 BIBLIOGRAPHY

[85] Raychaudhuri, A. K. Measurement of 1/f noise and its application in materials


science. Current Opinion in Solid State and Mat. Sci. 6, 67 (2002).

[86] Hooge, F. 1/f noise. Physica B+C 83, 14 – 23 (1976).

[87] Hooge, F. N. 1/f noise sources. Electron Devices, IEEE Transactions on 41, 1926
(1994).

[88] Jindal, R. P. & van der Ziel, A. Model for mobility fluctuation 1/f noise. Applied
Physics Letters 38, 290–291 (1981).

[89] Hooge, F. N., Kleinpenning, T. G. M. & Vandamme, L. K. J. Experimental studies


on 1/f noise. Reports on Progress in Physics 44, 479 (1981).

[90] Pal, A. N. et al. Microscopic mechanism of 1/f noise in graphene: Role of energy
band dispersion. ACS Nano 5, 2075–2081 (2011).

[91] Lin, Y.-M., Appenzeller, J., Knoch, J., Chen, Z. & Avouris, P. Low-frequency cur-
rent fluctuations in individual semiconducting single-wall carbon nanotubes. Nano
Letters 6, 930–936 (2006).

[92] Heo, K. et al. Large-scale assembly of silicon nanowire network-based devices using
conventional microfabrication facilities. Nano Letters 8, 4523–4527 (2008).

[93] Singh, A., Sai, T. P. & Ghosh, A. Electrochemical fabrication of ultralow noise
metallic nanowires with hcp crystalline lattice. Applied Physics Letters 93, 102107
(2008).

[94] Cho, I.-T. et al. Low frequency noise characteristics in multilayer WSe2 field effect
transistor. Applied Physics Letters 106, 023504 (2015).

[95] Sangwan, V. K. et al. Low-frequency electronic noise in single-layer MoS2 transis-


tors. Nano Letters 13, 4351–4355 (2013). PMID: 23944940.

[96] Sahoo, A., Ha, S. D., Ramanathan, S. & Ghosh, A. Conductivity noise study of
the insulator-metal transition and phase coexistence in epitaxial samarium nickelate
thin films. Phys. Rev. B 90, 085116 (2014).

[97] U, C., Ghosh, A., Vijaya, H. S. & Mohan, S. Signature of martensite transformation
on conductivity noise in thin films of NiTi shape memory alloys. Applied Physics
Letters 92, 112110 (2008).
BIBLIOGRAPHY 189

[98] McWhorter, A. L. Semiconductor surface physics (Philadelphia: University of Penn-


sylvania Press, 1957).

[99] Jayaraman, R. & Sodini, C. A 1/f noise technique to extract the oxide trap density
near the conduction band edge of silicon. Electron Devices, IEEE Transactions on
36, 1773–1782 (1989).
1
[100] Dutta, P. & Horn, P. M. Low-frequency fluctuations in solids: f
noise. Rev. Mod.
Phys. 53, 497–516 (1981).

[101] Kogan, S. & Nagaev, K. On the low-frequency current noise in metals. Solid State
Communications .
1
[102] Pelz, J. & Clarke, J. Quantitative ”local-interference” model for f
noise in metal
films. Phys. Rev. B 36, 4479–4482 (1987).

[103] Feng, S., Lee, P. A. & Stone, A. D. Sensitivity of the conductance of a disordered
metal to the motion of a single atom: Implications for f1 noise. Phys. Rev. Lett. 56,
1960–1963 (1986).

[104] Beutler, D. E., Meisenheimer, T. L. & Giordano, N. Resistance fluctuations in thin


Bi wires and films. Phys. Rev. Lett. 58, 1240–1243 (1987).

[105] Meisenheimer, T. L. & Giordano, N. Conductance fluctuations in thin silver films.


Phys. Rev. B 39, 9929–9936 (1989).

[106] Garfunkel, G. A., Alers, G. B., Weissman, M. B., Mochel, J. M. & VanHarlingen,
1
D. J. Universal-conductance-fluctuation f
noise in a metal-insulator composite.
Phys. Rev. Lett. 60, 2773–2776 (1988).

[107] Alers, G. B., Weissman, M. B., Averback, R. S. & Shyu, H. Resistance noise in
amorphous Ni-Zr: Hydrogen diffusion and universal conductance fluctuations. Phys.
Rev. B 40, 900–906 (1989).

[108] Georges, A., Kotliar, G., Krauth, W. & Rozenberg, M. J. Dynamical mean-field
theory of strongly correlated fermion systems and the limit of infinite dimensions.
Rev. Mod. Phys. 68, 13–125 (1996).

[109] Byczuk, K., Hofstetter, W. & Vollhardt, D. Mott-Hubbard transition versus An-
derson localization in correlated electron systems with disorder. Phys. Rev. Lett.
94, 056404 (2005).
190 BIBLIOGRAPHY

[110] Denteneer, P. J. H., Scalettar, R. T. & Trivedi, N. Conducting phase in the two-
dimensional disordered Hubbard model. Phys. Rev. Lett. 83, 4610–4613 (1999).

[111] Shepelyansky, D. L. Coherent propagation of two interacting particles in a random


potential. Phys. Rev. Lett. 73, 2607–2610 (1994).

[112] Punnoose, A. & Finkel’stein, A. M. Metal-insulator transition in disordered two-


dimensional electron systems. Science 310, 289–291 (2005).

[113] Lai, K., Pan, W., Tsui, D. & Xie, Y.-H. Observation of the apparent metal-insulator
transition of high-mobility two-dimensional electron system in a Si/Si1−x Gex het-
erostructure. Applied Physics Letters 84, 302–304 (2004).

[114] Okamoto, T., Ooya, M., Hosoya, K. & Kawaji, S. Spin polarization and metallic
behavior in a silicon two-dimensional electron system. Phys. Rev. B 69, 041202
(2004).

[115] Gunawan, O. et al. Spin-valley phase diagram of the two-dimensional metal-


insulator transition. Nat. Phys. 3, 388–391 (2006).

[116] Hamilton, A. R., Simmons, M. Y., Pepper, M., Linfield, E. H. & Ritchie, D. A.
Metallic behavior in dilute two-dimensional hole systems. Phys. Rev. Lett. 87,
126802 (2001).

[117] Baenninger, M. et al. Low-temperature collapse of electron localization in two


dimensions. Phys. Rev. Lett. 100, 016805 (2008).

[118] Kohno, M. Mott transition in the two-dimensional Hubbard model. Phys. Rev.
Lett. 108, 076401 (2012).

[119] Macridin, A., Jarrell, M., Maier, T. & Scalapino, D. J. High-energy kink in the
single-particle spectra of the two-dimensional Hubbard model. Phys. Rev. Lett. 99,
237001 (2007).

[120] Preuss, R., Hanke, W., Gröber, C. & Evertz, H. G. Pseudogaps and their interplay
with magnetic excitations in the doped 2D Hubbard model. Phys. Rev. Lett. 79,
1122–1125 (1997).

[121] Bhatt, R. N. & Nielsen, E. Ferromagtism in doped semiconductors without magnetic


ions. International Journal of Modern Physics B 22, 4595 (2008).
BIBLIOGRAPHY 191

[122] Nielsen, E. & Bhatt, R. N. Nanoscale ferromagnetism in nonmagnetic doped semi-


conductors. Phys. Rev. B 76, 161202 (2007).

[123] Kane, B. E. A silicon-based nuclear spin quantum computer. Nature 393, 133–138
(1998).

[124] Rueß, F. J. et al. Narrow, highly P-doped, planar wires in silicon created by scanning
probe microscopy. Nanotechnology 18, 044023 (2007).

[125] Weber, B. et al. Ohms law survives to the atomic scale. Science 335, 64–67 (2012).

[126] Fuhrer, A., Füchsle, M., Reusch, T. C. G., Weber, B. & Simmons, M. Y. Atomic-
scale, all epitaxial in-plane gated donor quantum dot in silicon. Nano Letters 9,
707–710 (2009).

[127] Füechsle, M. et al. Spectroscopy of few-electron single-crystal silicon quantum dots.


Nature Nanotechnology 5, 502–505 (2010).

[128] Fuechsle, M. et al. A single-atom transistor. Nature Nanotechnology 7, 242–246


(2012).

[129] Rueß, F. J., Pok, W., Goh, K. E. J., Hamilton, A. R. & Simmons, M. Y. Electronic
properties of atomically abrupt tunnel junctions in silicon. Phys. Rev. B 75, 121303
(2007).

[130] Novoselov, K. S. et al. Two-dimensional gas of massless dirac fermions in graphene.


Nature 438, 197–200 (2005).

[131] Segawa, K. et al. Ambipolar transport in bulk crystals of a topological insulator by


gating with ionic liquid. Phys. Rev. B 86, 075306 (2012).

[132] Koushik, R. et al. Evidence of gate-tunable topological excitations in two-


dimensional electron systems. Phys. Rev. B 83, 085302 (2011).

[133] Cartoixà, X. & Chang, Y. C. Fermi-level oscillation in n-type δ-doped Si: A self-
consistent tight-binding approach. Phys. Rev. B 72, 125330 (2005).

[134] Qian, G., Chang, Y.-C. & Tucker, J. R. Theoretical study of phosphorous δ-doped
silicon for quantum computing. Phys. Rev. B 71, 045309 (2005).
192 BIBLIOGRAPHY

[135] Carter, D. J., Warschkow, O., Marks, N. A. & McKenzie, D. R. Electronic structure
models of phosphorus δ-doped silicon. Phys. Rev. B 79, 033204 (2009).

[136] Lee, S. et al. Electronic structure of realistically extended atomistically resolved


disordered Si:P δ-doped layers. Phys. Rev. B 84, 205309 (2011).

[137] Drumm, D. W., Hollenberg, L. C. L., Simmons, M. Y. & Friesen, M. Effective mass
theory of monolayer δ doping in the high-density limit. Phys. Rev. B 85, 155419
(2012).

[138] Goh, K. E. J., Oberbeck, L., Simmons, M. Y., Hamilton, A. R. & Clark, R. G.
Effect of encapsulation temperature on Si:P δ-doped layers. Applied Physics Letters
85, 4953–4955 (2004).

[139] Goh, K. E. J., Oberbeck, L., Simmons, M. Y., Hamilton, A. R. & Butcher, M. J.
Influence of doping density on electronic transport in degenerate Si:P δ-doped layers.
Phys. Rev. B 73, 035401 (2006).

[140] Goh, K. E. J., Simmons, M. Y. & Hamilton, A. R. Use of low-temperature Hall


effect to measure dopant activation: Role of electron-electron interactions. Phys.
Rev. B 76, 193305 (2007).

[141] Goh, K. E. J., Simmons, M. Y. & Hamilton, A. R. Electron-electron interactions


in highly disordered two-dimensional systems. Phys. Rev. B 77, 235410 (2008).

[142] Goh, K. E. J., Augarten, Y., Oberbeck, L. & Simmons, M. Y. Enhancing electron
transport in Si:P δ-doped devices by rapid thermal anneal. Applied Physics Letters
93, 142105 (2008).

[143] McKibbin, S. R., Clarke, W. R., Fuhrer, A., Reusch, T. C. G. & Simmons, M. Y.
Investigating the regrowth surface of Si:P δ-layers toward vertically stacked three
dimensional devices. Applied Physics Letters 95, 233111 (2009).

[144] Scappucci, G. et al. Stacking of 2D electron gases in Ge probed at the atomic


level and its correlation to low-temperature magnetotransport. Nano Letters 12,
4953–4959 (2012).

[145] Scappucci, G., Capellini, G., Lee, W. C. T. & Simmons, M. Y. Ultra-dense phos-
phorus in germanium δ-doped layers. Applied Physics Letters 94, 162106 (2009).
BIBLIOGRAPHY 193

[146] Scappucci, G., Capellini, G. & Simmons, M. Y. Influence of encapsulation temper-


ature on Ge:P δ-doped layers. Phys. Rev. B 80, 233202 (2009).

[147] Scappucci, G. et al. A complete fabrication route for atomic-scale, donor-based


devices in single-crystal germanium. Nano Letters 11, 2272–2279 (2011).

[148] Scappucci, G. et al. n-type doping of germanium from phosphine: Early stages
resolved at the atomic level. Phys. Rev. Lett. 109, 076101 (2012).

[149] Weber, B. Towards Scalable Planar Donor-Based Silicon Quantum Computing Ar-
chitectures. Ph.D. thesis, University of New South Wales (2013).

[150] Weber, B., Ryu, H., Tan, Y.-H. M., Klimeck, G. & Simmons, M. Y. Limits to
metallic conduction in atomic-scale quasi-one-dimensional silicon wires. Phys. Rev.
Lett. 113, 246802 (2014).

[151] Pobell, F. The 3He-4He Dilution Refrigerator (Springer Berlin Heidelberg, 1996).

[152] Reif, F. Fundamentals of statistical and thermal physics (McGraw-Hill, New York,
1965).

[153] Chandni, U. One dimensional transport and prospects of structural transitions in


ultrathin metallic wires. Ph.D. thesis, Indian Institute of Science, India (2011).

[154] Scofield, J. H. AC method for measuring low-frequency resistance fluctuation spec-


tra. Rev. Sci. Instrum. 58, 985–993 (1987).

[155] Smith, S. W. The Scientist and Engineer’s Guide to Digital Signal Processing
(California Technical Publishing San Diego, California, 1999).

[156] Defatta, D. J., Lucas, J. G. & Hodgkiss, W. S. Digital Signal Processing (John
Wiley and Sons, New York, 1988).

[157] Lyons, R. Understanding Digital Signal Processing (Prentice Hall, NJ, 2001).

[158] Ghosh, A. Low Frequency Conductance Fluctuations near metal-insulator transition.


Ph.D. thesis, Indian Institute of Science, India (1999).

[159] Welch, P. D. Modern Spectral Analysis (IEEE Press, John Wiley and Sons, New
York, 1978).
194 BIBLIOGRAPHY

[160] Ghosh, A., Kar, S., Bid, A. & Raychaudhuri, A. K. A set-up for measurement of low
frequency conductance fluctuation (noise) using digital signal processing techniques.
ArXiv Condensed Matter e-prints (2004). arXiv:cond-mat/0402130.

[161] Goh, K. E. J. Encapsulation of Si:P devices fabricated by scanning tunnelling mi-


croscopy. Ph.D. thesis, University of New South Wales (2006).

[162] Goh, K. E. J., Oberbeck, L. & Simmons, M. Y. Relevance of phosphorus incor-


poration and hydrogen removal for Si:P δ-doped layers fabricated using phosphine.
physica status solidi (a) 202, 1002–1005 (2005).

[163] Thompson, D. W. Fabricating Atomically-Abrupt, Surface-Gated Devices in Silicon.


Ph.D. thesis, University of New South Wales (2011).

[164] Schofield, S. R. et al. Atomically precise placement of single dopants in Si. Phys.
Rev. Lett. 91, 136104 (2003).

[165] Kar, S., Raychaudhuri, A. K., Ghosh, A., Löhneysen, H. v. & Weiss, G. Observation
of non-gaussian conductance fluctuations at low temperatures in Si:P(B) at the
metal-insulator transition. Phys. Rev. Lett. 91, 216603 (2003).

[166] Hollenberg, L. C. L., Wellard, C. J., Pakes, C. I. & Fowler, A. G. Single-spin readout
for buried dopant semiconductor qubits. Phys. Rev. B 69, 233301 (2004).

[167] Hwang, E. H. & Das Sarma, S. Electronic transport in two-dimensional Si:P δ-doped
layers. Phys. Rev. B 87, 125411 (2013).

[168] Davies, J. H. The Physics of Low-dimensional Semiconductors: An Introduction


(Cambridge University Press, 1998).

[169] Wheeler, R. G. Magnetoconductance and weak localization in silicon inversion


layers. Phys. Rev. B 24, 4645–4651 (1981).

[170] Masaki, K., Taniguchi, K. & Hamaguchi, C. Electron mobility in Si inversion layers.
Semiconductor Science and Technology 7, B573 (1992).

[171] Churchill, A. C. et al. High-mobility two-dimensional electron gases in Si/SiGe


heterostructures on relaxed SiGe layers grown at high temperature. Semiconductor
Science and Technology 12, 943 (1997).
BIBLIOGRAPHY 195

[172] Goswami, S. et al. Controllable valley splitting in silicon quantum devices. Nat
Phys 3, 41 (2007).

[173] Ghosh, A. & Raychaudhuri, A. K. Conductance fluctuations near the Anderson


transition. Journal of Physics: Condensed Matter 11, L457 (1999).

[174] Kar, S. & Raychaudhuri, A. K. Temperature and frequency dependence of flicker


noise in degenerately doped Si single crystals. Journal of Physics D: Applied Physics
34, 3197 (2001).

[175] McCammon, D. et al. 1/f noise in doped semiconductor thermistors. AIP Conference
Proceedings 605, 91–94 (2002).

[176] Adkins, C. J. & Koch, R. H. Noise in inversion layers near the metal-insulator
transition. Journal of Physics C: Solid State Physics 15, 1829 (1982).

[177] Gaubert, P., Teramoto, A., Cheng, W., Hamada, T. & Ohmi, T. Different mecha-
nism to explain the 1/f noise in n- and p-SOI-MOS transistors fabricated on (110)
and (100) silicon-oriented wafers. Journal of Vacuum Science and Technology B 27,
394–401 (2009).

[178] von Haartman, M. et al. Impact of strain and channel orientation on the low-
frequency noise performance of Si n- and p MOSFETs. Solid-State Electronics 51,
771 – 777 (2007).

[179] Marin, M., Allogo, Y. A., de Murcia, M., Llinares, P. & Vildeuil, J. C. Low frequency
noise characterization in 0.13 µm p-MOSFETs. impact of scaled-down 0.25, 0.18 and
0.13 µm technologies on 1/f noise. Microelectronics and Reliability 44, 1077 – 1085
(2004).

[180] Yu, X., Thaysen, J., Hansen, O. & Boisen, A. Optimization of sensitivity and noise
in piezoresistive cantilevers. Journal of Applied Physics 92, 6296–6301 (2002).

[181] Zhigal’skii, G. P. & Jones, B. K. The physical properties of thin metal films. CRC
Press (2003).

[182] Mouetsi, S., Hdiy, A. E. & Bouchemat, M. The 1/f noise in a two-dimensional
electron gas: Temperature and electric field considerations. ACS Nano 92, 3 (2009).
196 BIBLIOGRAPHY

[183] Delseny, C. et al. Excess noise in AlGaAs/GaAs heterojunction bipolar transistors


and associated TLM test structures. Electron Devices, IEEE Transactions on 41,
2000–2005 (1994).

[184] Balandin, A. A. Noise and fluctuations control in electronic devices. American


Scientific Publishers: Valencia, CA (2002).

[185] Balandin, A. et al. Effect of channel doping on the low-frequency noise in


GaN/AlgaN heterostructure field-effect transistors. Applied Physics Letters 75,
2064–2066 (1999).

[186] Kruppa, W. et al. Low-frequency noise in AlSb/InAs and related HEMTs. Electron
Devices, IEEE Transactions on 54, 1193–1202 (2007).

[187] Kruppa, W., Yang, M. J., Bennett, B. R. & Boos, J. B. Low-frequency noise in
AlSb/InAs high-electron-mobility transistor structure as a function of temperature
and illumination. Applied Physics Letters 85, 774–776 (2004).

[188] Kruppa, W., Boos, J., Bennett, B. & Yang, M. Low-frequency noise in alsb/inas
hemts. Electronics Letters 36, 1888–1889 (2000).

[189] Lin, Y.-M. & Avouris, P. Strong suppression of electrical noise in bilayer graphene
nanodevices. Nano Letters 8, 2119–2125 (2008).

[190] Pal, A. N. & Ghosh, A. Ultralow noise field-effect transistor from multilayer
graphene. Applied Physics Letters 95, 082105 (2009).

[191] Liu, G. et al. Low-frequency electronic noise in the double-gate single-layer graphene
transistors. Applied Physics Letters 95, 033103 (2009).

[192] Zhang, Y., Mendez, E. E. & Du, X. Mobility-dependent low-frequency noise in


graphene field-effect transistors. ACS Nano 5, 8124–8130 (2011).

[193] Renteria, J. et al. Low-frequency 1/f noise in MoS2 transistors: Relative contribu-
tions of the channel and contacts. Applied Physics Letters 104, 153104 (2014).

[194] Xie, X. et al. Low-frequency noise in bilayer MoS2 transistor. ACS Nano 8, 5633–
5640 (2014).

[195] Ghatak, S., Mukherjee, S., Jain, M., Sarma, D. D. & Ghosh, A. Microscopic origin
of low frequency noise in MoS2 field-effect transistors. APL Materials 2, – (2014).
BIBLIOGRAPHY 197

[196] Hossain, M. Z. et al. Low-frequency current fluctuations in graphene-like exfoli-


ated thin-films of bismuth selenide topological insulators. ACS Nano 5, 2657–2663
(2011).

[197] Shamim, S., Mahapatra, S., Polley, C., Simmons, M. Y. & Ghosh, A. Suppression
of low-frequency noise in two-dimensional electron gas at degenerately doped Si:P
δ layers. Phys. Rev. B 83, 233304 (2011).

[198] Shamim, S. et al. Origin of noise in two dimensionally doped silicon and germanium.
AIP Conference Proceedings 1566 (2013).

1
[199] Weissman, M. B. f
noise and other slow, nonexponential kinetics in condensed
matter. Rev. Mod. Phys. 60, 537–571 (1988).

[200] Collins, P. G., Fuhrer, M. S. & Zettl, A. 1/f noise in carbon nanotubes. Applied
Physics Letters 76 (2000).

[201] Ishigami, M. et al. Hooge’s constant for carbon nanotube field effect transistors.
Applied Physics Letters 88, – (2006).

[202] Vijayaraghavan, A. et al. Effect of ambient pressure on resistance and resistance


fluctuations in single-wall carbon nanotube devices. Journal of Applied Physics
100, – (2006).

[203] Lin, Y.-M., Tsang, J. C., Freitag, M. & Avouris, P. Impact of oxide substrate on
electrical and optical properties of carbon nanotube devices. Nanotechnology 18,
295202 (2007).

[204] Back, J. H., Kim, S., Mohammadi, S. & Shim, M. Low-frequency noise in ambipolar
carbon nanotube transistors. Nano Letters 8, 1090–1094 (2008). PMID: 18351749.

[205] Tobias, D. et al. Origins of 1/f noise in individual semiconducting carbon nanotube
field-effect transistors. Phys. Rev. B 77, 033407 (2008).

[206] Persson, K.-M. et al. Low-frequency noise in vertical InAs nanowire FETs. Electron
Device Letters, IEEE 31, 428–430 (2010).

[207] Xiong, H. D. et al. Random telegraph signals in n-type ZnO nanowire field effect
transistors at low temperature. Applied Physics Letters 91, 053107 (2007).
198 BIBLIOGRAPHY

[208] Singh, A., Mukhopadhyay, S. & Ghosh, A. Tracking random walk of individual
domain walls in cylindrical nanomagnets with resistance noise. Phys. Rev. Lett.
105, 067206 (2010).

[209] Peters, M. G., Dijkhuis, J. I. & Molenkamp, L. W. Random telegraph signals and
1/f noise in a silicon quantum dot. Journal of Applied Physics 86, 1523–1526 (1999).
ν
[210] Celik-Butler, Z. & Hsiang, T. Y. Spectral dependence of 1/f noise on gate bias
in n-MOSFETS. Solid State Electronics 30, 419–423 (1987).

[211] Sharma, D. et al. Electrical transport and low-frequency noise in chemical vapor
deposited single-layer mos 2 devices. Nanotechnology 25, 155702 (2014).

[212] Gray, P. V. & Brown, D. M. Density of SiO2 Si interface states. Applied Physics
Letters 8 (1966).

[213] Poindexter, E. H. MOS interface states: overview and physicochemical perspective.


Semiconductor Science and Technology 4, 961 (1989).

[214] Yamashita, Y., Asano, A., Nishioka, Y. & Kobayashi, H. Dependence of interface
states in the Si band gap on oxide atomic density and interfacial roughness. Phys.
Rev. B 59, 15872–15881 (1999).

[215] Chang, C.-Z. et al. Experimental observation of the quantum anomalous hall effect
in a magnetic topological insulator. Science 340, 167–170 (2013).

[216] Pathak, S., Shenoy, V. B. & Baskaran, G. Possible high-temperature superconduct-


ing state with a d + id pairing symmetry in doped graphene. Phys. Rev. B 81,
085431 (2010).

[217] Nandkishore, R., Levitov, L. S. & Chubukov, A. V. Chiral superconductivity from


repulsive interactions in doped graphene. Nat. Phys. 8, 158–163 (2012).

[218] Castellani, C., Di Castro, C., Lee, P. A. & Ma, M. Interaction-driven metal-insulator
transitions in disordered fermion systems. Phys. Rev. B 30, 527–543 (1984).

[219] Abrahams, E., Kravchenko, S. V. & Sarachik, M. P. Metallic behavior and related
phenomena in two dimensions. Rev. Mod. Phys. 73, 251–266 (2001).

[220] Rosenbaum, T. F. et al. Metal-insulator transition in a doped semiconductor. Phys.


Rev. B 27, 7509–7523 (1983).
BIBLIOGRAPHY 199

[221] Dai, P., Zhang, Y. & Sarachik, M. P. Electrical conductivity of metallic Si:B near
the metal-insulator transition. Phys. Rev. B 45, 3984–3994 (1992).

[222] Lakner, M. & Löhneysen, H. v. Localized versus itinerant electrons at the metal-
insulator transition in Si:P. Phys. Rev. Lett. 63, 648–651 (1989).

[223] Shklovskii, B. I. & Efros, A. L. Electronic properties of doped semiconductors


(Springer-Verlag, Berlin Heidelberg New York Tokyo, 1984).

[224] Ashcroft, N. W. & Mermin, N. D. Solid State Physics (Thomson Brooks Cole,
2006).

[225] Houghton, A., Senna, J. R. & Ying, S. C. Magnetoresistance and hall effect of a
disordered interacting two-dimensional electron gas. Phys. Rev. B 25, 2196–2210
(1982).

[226] Gornyi, I. V. & Mirlin, A. D. Interaction-induced magnetoresistance: From the


diffusive to the ballistic regime. Phys. Rev. Lett. 90, 076801 (2003).

[227] McPhail, S. et al. Weak localization in high-quality two-dimensional systems. Phys.


Rev. B 70, 245311 (2004).

[228] Dai, P., Zhang, Y., Bogdanovich, S. & Sarachik, M. P. Critical conductivity expo-
nent of Si:P in a magnetic field. Phys. Rev. B 48, 4941–4943 (1993).

[229] Paalanen, M. A., Graebner, J. E., Bhatt, R. N. & Sachdev, S. Thermodynamic


behavior near a metal-insulator transition. Phys. Rev. Lett. 61, 597–600 (1988).

[230] Sachdev, S. Local moments near the metal-insulator transition. Phys. Rev. B 39,
5297–5310 (1989).

[231] Milovanović, M., Sachdev, S. & Bhatt, R. N. Effective-field theory of local-moment


formation in disordered metals. Phys. Rev. Lett. 63, 82–85 (1989).

[232] Rahimi, M., Anissimova, S., Sakr, M. R., Kravchenko, S. V. & Klapwijk, T. M.
Coherent backscattering near the two-dimensional metal-insulator transition. Phys.
Rev. Lett. 91, 116402 (2003).

[233] Meyer, J. S., Fal’ko, V. I. & Altshuler, B. L. Quantum in-plane magne-


toresistance in 2D electron systems. eprint arXiv:cond-mat/0206024 (2002).
arXiv:cond-mat/0206024.
200 BIBLIOGRAPHY

[234] Bhatt, R. N. & Lee, P. A. Scaling studies of highly disordered spin- 21 antiferromag-
netic systems. Phys. Rev. Lett. 48, 344–347 (1982).

[235] Lakner, M. & Löhneysen, H. v. Localized versus itinerant electrons at the metal-
insulator transition in Si:P. Phys. Rev. Lett. 63, 648–651 (1989).

[236] Lindqvist, P., Nordström, A. & Rapp, O. New resistance anomaly in the supercon-
ducting fluctuation region of disordered Cu-Zr alloys with dilute magnetic impuri-
ties. Phys. Rev. Lett. 64, 2941–2944 (1990).

[237] Santhanam, P., Chi, C. C., Wind, S. J., Brady, M. J. & Bucchignano, J. J. Resis-
tance anomaly near the superconducting transition temperature in short aluminum
wires. Phys. Rev. Lett. 66, 2254–2257 (1991).

[238] Vloeberghs, H., Moshchalkov, V. V., Van Haesendonck, C., Jonckheere, R. &
Bruynseraede, Y. Anomalous little-parks oscillations in mesoscopic loops. Phys.
Rev. Lett. 69, 1268–1271 (1992).

[239] Strunk, C. et al. Resistance anomalies in superconducting mesoscopic Al structures.


Phys. Rev. B 57, 10854–10866 (1998).

[240] Arutyunov, K. Y., Presnov, D. A., Lotkhov, S. V., Pavolotski, A. B. & Rinderer,
L. Resistive-state anomaly in superconducting nanostructures. Phys. Rev. B 59,
6487–6498 (1999).

[241] Hua, J. et al. Resistance anomaly in disordered superconducting films. Applied


Physics Letters 90, 072507 (2007).

[242] Rubin, S., Schimpfke, T., Weitzel, B., Voloh, C. & Micklitz, H. Observation of sharp
resistance anomalies near Tc in granular superconductors. Annalen der Physik 504,
492–499 (1992).

[243] Blonder, G. E., Tinkham, M. & Klapwijk, T. M. Transition from metallic to tun-
neling regimes in superconducting microconstrictions: Excess current, charge im-
balance, and supercurrent conversion. Phys. Rev. B 25, 4515–4532 (1982).

[244] Wang, J. et al. Interplay between superconductivity and ferromagnetism in crys-


talline nanowires. Nature Physics 6, 389–394 (2010).
BIBLIOGRAPHY 201

[245] Wang, J. et al. An unusual magnetoresistance effect in the heterojunction struc-


ture of an ultrathin single-crystal Pb film on silicon substrate. Nanotechnology 19,
475708 (2008).

[246] Wang, J. et al. Interplay between topological insulators and superconductors. Phys.
Rev. B 85, 045415 (2012).

[247] Mohan, S. et al. Large low-frequency fluctuations in the velocity of a driven vortex
lattice in a single crystal of 2H-NbSe2 superconductor. Phys. Rev. Lett. 103, 167001
(2009).

[248] Togawa, Y., Abiru, R., Iwaya, K., Kitano, H. & Maeda, A. Direct observation of the
washboard noise of a driven vortex lattice in a high-temperature superconductor,
Bi2 Sr2 CaCu2 Oy . Phys. Rev. Lett. 85, 3716–3719 (2000).

[249] Clarke, J. & Hsiang, T. Y. Low-frequency noise in tin and lead films at the super-
conducting transition. Phys. Rev. B 13, 4790–4800 (1976).

[250] Ketchen, M. B. & Clarke, J. Temperature fluctuations in freely suspended tin films
at the superconducting transition. Phys. Rev. B 17, 114–121 (1978).

[251] Kiss, L. & Svedlindh, P. Noise in high Tc superconductors. Electron Devices, IEEE
Transactions on 41, 2112–2122 (1994).

[252] Mihajlović, G., Pearson, J. E., Garcia, M. A., Bader, S. D. & Hoffmann, A. Negative
nonlocal resistance in mesoscopic gold hall bars: Absence of the giant spin hall effect.
Phys. Rev. Lett. 103, 166601 (2009).

[253] Yang, H., Yang, S.-H., Takahashi, S., Maekawa, S. & Parkin, S. S. P. Extremely
long quasiparticle spin lifetimes in superconducting aluminium using MgO tunnel
spin injectors. Nature Materials 9, 586–593 (2010).

[254] Lo, S.-T., Lin, S.-W., Wang, Y.-T., Lin, S.-D. & Liang, C.-T. Spin-orbit-coupled
superconductivity. Scientific Reports 4, 5438 (2014).

[255] Svetlichnyy, P. M., Jiang, Z. & de Melo, C. A. R. S. Parity violation in ferromagnet-


superconductor heterostructures with strong spin-orbit coupling. ArXiv e-prints
(2014). 1403.6858.
202 BIBLIOGRAPHY

[256] Chaikin, P. M. & Beni, G. Thermopower in the correlated hopping regime. Phys.
Rev. B 13, 647–651 (1976).

[257] Wang, Y., Rogado, N. S., Cava, R. J. & Ong, N. P. Spin entropy as the likely source
of enhanced thermopower in Nax Co2 O4 . Nature 423, 425 (2003).

[258] Scheibner, R., Buhmann, H., Reuter, D., Kiselev, M. N. & Molenkamp, L. W.
Thermopower of a kondo spin-correlated quantum dot. Phys. Rev. Lett. 95, 176602
(2005).

[259] Soukoulis, C. M., Webman, I., Grest, G. S. & Economou, E. N. Study of electronic
states with off-diagonal disorder in two dimensions. Phys. Rev. B 26, 1838–1841
(1982).

[260] Yang, Y. H., Wang, Y. G., Liu, M. & Xing, D. Y. Unitary limit and quantum inter-
ference effect in disordered two-dimensional crystals with nearly half-filled bands.
Phys. Rev. B 68, 045113 (2003).

[261] Nakhmedov, E. P., Feldmann, H., Oppermann, R. & Kumru, M. Effect of sub-
stitutional impurities on the electronic states and conductivity of crystals with a
half-filled band. Phys. Rev. B 62, 13490–13500 (2000).

[262] Yang, Y. H., Xing, D. Y., Liu, M. & Wang, Y. G. Existence of extended states
in disordered two-dimensional crystals with a half-filled band. Phys. Rev. B 66,
205310 (2002).

[263] Hu, W. M., Dow, J. D. & Myles, C. W. Effects of diagonal and off-diagonal disorder
on the Anderson-model densities of states in two and three dimensions. Phys. Rev.
B 30, 1720–1723 (1984).

[264] Evangelou, S. N. & Katsanos, D. E. Spectral statistics in chiral-orthogonal disor-


dered systems. Journal of Physics A: Mathematical and General 36, 3237 (2003).

[265] Cutler, M. & Mott, N. F. Observation of Anderson localization in an electron gas.


Phys. Rev. 181, 1336–1340 (1969).

[266] Mott, N. F. & Jones, M. The Theory and the Properties of Metals and Alloys
(Courier Dover, 1958).
BIBLIOGRAPHY 203

[267] Fletcher, R., Maan, J. C., Ploog, K. & Weimann, G. Thermoelectric properties
of GaAs-Ga1−x Alx As heterojunctions at high magnetic fields. Phys. Rev. B 33,
7122–7133 (1986).

[268] Garde, C. S., Ray, J. & Chandra, G. Resistivity and thermopower studies in the
kondo-lattice Ce3 Sn1−x Inx system. Phys. Rev. B 40, 10116–10118 (1989).

[269] Kowalczyk, A., Tran, V., Tolinski, T. & Miiller, W. Electrical resistivity and ther-
moelectric power of the kondo lattice CeNiAl4 . Solid State Communications 144,
185 – 188 (2007).

You might also like