Simulation of Single-Electron

Diss. ETHNo.
13526
Simulation of
single-electron
devices
A dissertation submitted to the
SWISS FEDERAL INSTITUTE OF TECHNOLOGY

ZURICH
for the
degree
ol
Ph. D.
presented by
ANDREAS SCHOLZE
Dipl. Phys., University

citizen of
of Jena
born June 14, 1969
Germany
accepted on the recommendation of Prof. Wolfgang Fichtner, examiner

Prof. Klaus Ensslin, co-examiner
2000
c
i.*
1
*u
U
\
\
11e
\
\
i
'v
Acknowledgements
T wish to thank Prof.
Wolfgang
Fichtner for his
guidance
and
and assistance
during
my years at TIS, and Prof. Klaus Ensslin for
reading
co-examining
the
this thesis.
Special
thanks goes to PD Andreas Schenk for
supervising
scientific aspects of this work.

Thanks to all members of the simulation group: Andreas Wettstein, Bern-
hard Schmithiisen. Bernd
Witzigman.
Andreas
Witzig,
Thomas Krner, Gtz

to
Leonhard, Christian Schuster, Fabian Butler and Frank Geelhaar. I wish

express my with
gratitude for the benefits that I have gained from conversations Michael Stopa (TU Munich), Massimo Macucci (University of Pisa) and
(WIAS Berlin). Furthermore,
for his
I wish to thank Thomas Heinzcl data and Diederick Fokkema
Klaus Grtner
(ETH Zrich) (1SE Inc.) for
help
with the
experimental
his support with numerical software.

to thank the technical staff of
Also, 1 would like
TIS for
providing excellent
working
to
conditions and all other
the excellent
atmosphere
that
people at the I enjoyed very

to
institute for their contribution
much.
I would like to express my

the
appreciation
at the
Prof. Karl Hess who gave
me
opportunity
to
spend
total of five months at the Beckman Institute of
Advanced Science and
Technology
University
of Illinois in Urbana-
Champaign.
This work has been Nano
supported by the Swiss Priority Program in System Technology (MINAST project 7.05 STMNAD).
Micro and
is
11:
in,:
IK
it a i
Contents
Preface
Preface
(German)
Introduction
1.1
Evolution of
single-electron
devices
1.2
Modern semiconductor device simulation
1.3
1.4 1.5
Single-electron Objectives
Eayout
device simulation
of this work and
general
remarks
Introduction to
2.1 2.2
single-electron tunneling
and Coulomb blockade
Introduction Orthodox
2.2.1
2.2.2
theory
of Coulomb blockade
The rules of the game
Single-electron box
2.2.3
2.3 2.4
Single-electron
transistor
Constant interaction model
Example
structures;
vertical quantum dots
vi
Contents
Theory
3.1
3.2
3.3
of
in semiconductors
29
Introduction
Model definition
Master
29
30
32
equation approach
Master
3.3.1
equation
distributions and transition rates
32 33
35
3.3.2
3.3.3
Probability
Detailed-balance
equations
3.3.4
3.4
Stationary
current
36
37
linear-response regime
3.4.1 3.4.2 3.4.3
Linear-response
Limiting
cases
conductance
37 42
Amplitude
and
lineshape
44
Transport
4.1
coefficients
47
Introduction
47
4.2
Equilibrium
4.2.1 4.2.2 4.2.3
4.2.4
carrier densities bandstructure and
48
Energy
applied forces
48
51 54 55
Quasi-equilibrium approximation
Boundary
conditions and Fermi level
Low dimensional electron gases
4.3
Total free 4.3.1

4.3.2
energies
60
60
61
Equilibrium thermodynamics
Flelmholtz free energy and external forces
4.3.3
Helmholtz free energy and
charge quantization
....
63
Contents
vu
4.3.4 4.4
Free energy minimization

rates
65
Tunneling
4.4.1
66 66 69
Bardeen's transfer Haniiltonian method
4.4.2
Adiabatic
theory
for quantum dot systems
.......
Numerics 5.1
5.2 Introduction
73
73
74 74
Space
5.2.1
discretization
Finite differences
5.2.2 5.3
Linear 5.3.1
Discretization of
V-(-7 V)
eigenvalue problems
76
77
77
equation solving
Linear Poisson S chrdinger
and
equation
5.3.2 5.4
equation
79
80
Nonlinear 5.4.1
5.4.2 5.4.3
equation solving
Introduction
80
method
Newton-Raphson BroydeiTs
81 82
method
5.4.4
Globally converging
algorithms
Newton schemes
85
5.5
Self-consistent
5.5.1
87
Predictor-corrector
approach
SIMNAD
87
5.5.2
5.5.3
Algorithm implementation in
Examples
89
92
Simulation
examples
95
6.1
GaAs/AlGaAs heterostructure
95
Vlll
Contents
6.2
Device simulation 6.2.1 6.2.2 6.2.3
98
and
Geometry modeling
Surface and donors Simulation results
grid
98 100 102
115
6.3
Comparison
with measurements
Achievements, failures and prospects
119
Thermodynamic properties
of localized systems
123
Transition rates
B.l
129
129
132
Time-dependent perturbation theory

Effective transition rates
B.2
Density-functional theory
C.l
C.2
135
ilohenberg-Kohn
Kohn-Sham
theorem
136
137
equation
C.3
Local-density approximation
139
Bibliography
141
Curriculum Vitae
149
Preface
repulsion of additional electrons from tiny metallic semiconducting islands (Coulomb blockade) in conjunction with correlated
of
Devices
utilizing
the
or
single electrons from externally controlled reservoirs are called single-electron devices. In contrast to conventional field-effect transistors where switching is accomplished by electrostatically opening a channel and allowing for a macroscopic current to flow, the switching in single-electron transistors (SETs) is based on adding or removing a single electron from a
tunneling
quantum dot, which is the analogue
devices represent the ultimate
to the channel.
As
such, single-electron
physical
limit of miniaturization in electronics
which is
thought
to
be reached within the next decades.
This work is concerned with the numerical devices. The main focus is
on
modeling
of
single-electron
such
as
the simulation of the local
quantities
the
carrier densities and the electrostatic of
potential
and the terminal characteristics
single-electron
After
transistors. introduction into the
giving
of
a
an
by
means
of the so-called orthodox
physics of single-electron devices theory, the physical models used for the
are
description
SET driven in linear response

a
derived.
The distribution
functions and the transition rates for

drain and the
ter
source
quantum dot in equilibrium with the
reservoir and the linearization of the
corresponding
mas
equation
is discussed in detail. The evaluation of the
resulting equation
for
the
linear-response conductance requires the calculation of the self-consistent many-body groundstate of a system of quantized electrons within the frame
work of
density-functional theory
evaluated
on
as
well
as
the calculation of the
tunneling
of
a
rates which are
the basis of Bardeen's transfer-Hamiltonian for
malism.
Finally,
are
the numerical methods used for the
implementation
simulation software,
especially
devised for the simulation of

are
single-electron
for the
transistors,
discussed.
Nonlinear solvers
of
key importance
proposed simulation approach, and therefore, much space is devoted to the quasi-Newton methods employed to solve the nonlinear Schrdinger/Poisson equation
which results from carrier confinement in semiconductor quantum
IX
Preface
dots, -wires, and -wells.
An introduction into methods for

at
solving
and
eigenvalue problems
finite-difference
grids
is
linear-equation given. singleus
Besides the focus of this work
on
basic models and methods for

was
electron device simulation, the simulator
employed
in
comprehensive
modeled
study
of
GaAs/AlGaAs heterostructure SET. The device
was
ing
the realistic
layer
structure and
the exact lateral gate geometry
assuming
which
are
various levels of
tum dot.
quantization
some
in the leads, the
tunneling
barriers and the quan
Despite
assumed to be due to
an
discrepancies with the experimental data oversimplified model for narrow split-gate
structures,
the main features in the Coulomb-blockade characteristics could be

on
explained
the basis of the self-consistent calculations. In addition to the conductance

a
results,
discussion of the
capacitances
of
preprocessed data from the simulation meaningful definition of a quantum dot self-capacitance
given on the basis output. Especially, the problem of a

was
of the system is
studied in
more
detail.
provides a physically sound and numerically stable framework for further investigations of single-electron transistors. The software implementation, for which the preliminary name SIMNAD (Simulation and Modeling of Nanodevices) is used, allows for the simulation of a variety of structures including GaAs/AlGaAs heterostructures and silicon SETs. However, for increased predictive capabilities more inves tigations into various problems including the treatment of narrow split-gate proposed
structures on
This work shows that the
method
active surfaces
or
the role of discrete
doping centers,
for instance,
is
certainly mandatory.
Einleitung
Bauelemente, die auf der Abstossung
lischen oder halbierenden Inseln
reliertem Tunneln einzelner Elektronen
winzigen metal (Coulomb-Blockade) in Verbindung mit kor

von
Elektronen
von
von extern
kontrollierten Reservoiren
basieren, werden Fînzelelektronen-Bauelemente genannt. In herkmmlichen

Feldeffekt-Transistorcn fliesst ein
makroskopischer
von
nur
Strom durch das
ffnen
eines Kanals. Im Unterschied dazu wird in Einzelelektronen-Bauelementen
ein Strom durch die
Bewegung
einem Elektron zwischen einem
Quanten-Punkt,
in der nach wie
dem
Analogon
rasant
zum
Kanal, und der
Umgebung erzeugt.
Daher
stellen Einzelelektronen-Bauelemente einen

vor
physikalisch bedingten Endpunkt
fortschreitenden
Miniaturisierung
der Elektronik
er
dar. Dieser Punkt wird wahrscheinlich innerhalb der nchsten Jahrzehnte

reicht werden.
Diese Arbeit
beschftigt
sich mit der numerischen
Modellierung
von
Einzelelektronen-Bauelementen. Dabei
liegt der Schwerpunkt auf der Simula tion sowohl lokaler Grssen wie der Ladungsdichten und des elektrostatischen Potentials, als auch der Berechnung der Kennlinien von EinzelelektronenTransistoren.
Nach
einer kurzen auf der
Einfhrung
Basis der Modelle
in
die
Physik
die
von
Einzelelektronenwer
Bauelementen
den
sogenannten orthodoxen Theorie
Besehreibung von Einzelelektronen-Transistoren verwendet werden. Die Verteilungsfunktionen und die bergangsraten fr einen Quanten-Punkt im Gleichgewicht mit dem Drain- und dem Source-Reservoir und die Linearisierung der entspre
zur
die
physikalischen
abgeleitet,
chenden Mastcr-Gleichime werden ausfhrlich diskutiert. Die
Linearisierung
der
Master-Gleichung
fhrt auf die
Leitfhigkeit
der
im Rahmen der Linearen-
Antwort-Theorie. Fr die
der selbstkonsistente
ter
entsprechenden Gleichung werden Mehrteilchen-Grundzustand fr ein System quantisierder Transfer-Hamiltonian-Methode nach Bardeen be
Auswertung
Elektronen auf der Basis der Dichtefunktional-Theorie und die Tunnelra
ten unter
Verwendung
rechnet. Schliesslich werden die numerischen Methoden diskutiert, die bei
xi
Xll
Einleitung
der
Implementierung
eines
Programms
zur
Simulation
von
Einzelelektronenvon
Transistoren verwendet wurden. Da die nichtlinearen
Gleichungslser
herausragender Bedeutung fr den gewhlten Simulationsansatz sind, wer den Quasi-Newton-Methoden sehr ausfhrlich behandelt, die bei der Lsung
der nichtlinearen
Schrdinger/Poisson-Gleichung
fr
Ladungstrger
in ein
zur
nu
sperrenden Potentialen verwendet werden. Eine Einfhrung in Methoden Diskretisierung von linearen Operatoren mit Finiten-Differenzen und der
merischen
Behandlung linearer Gleichungssysteme beziehungsweise genwertproblemen ist gegeben.

Der und
von
Ei
Schwerpunkt
fr
dieser Arbeit
die
liegt
auf der
Entwicklung
auf
einen
von
Modellen
Methoden
daraus
Simulation
von
Einzelelektronen-Bauelementen. GaAs/AlGaAs wird auf der
Der
resultierende
Simulator
wird Das
Eiuzelelektronen-Transistor
angewendet.
von
Bauelement
Basis
der realistischen Schichtstruktur und der
exakten
Ga/<?-Geometrie
modelliert, wobei verschiedene Arten
einsperrenden Potentialen sowohl fr Quanten-Punkte, -Drhte und -Trge bercksichtigt werden. Trotz einiger Abweichungen zu den experimentellen Daten, die vermutlich mit den stark vereinfachten Annahmen zusammenhngen, die bei der Behandlung der eng beieinander liegenden split-gates gemacht wurden, knnen die wichtigsten Besonderheiten in der Coulomb-Blockade-Kennlinie auf der
Basis der selbstkonsistenten
Rechnungen
erklrt werden. Zustzlich
zu
den
Leitfhigkeits-Kennlinien werden
eine
die verschiedenen
Kapazitten
an
Hand der
simulierten Daten diskutiert. Insbesondere wird das Problem untersucht, wie
Quanten-Punkt-Selbstkapazitt sinnvoll
zeigt,
dass die
definiert werden kann.

Methode einen
Diese Arbeit
vorgeschlagene
physikalisch
zuknftige Untersuchungen an Einzelelektronen-Transistoren bietet. Die Implementierung, fr die der vorlufige Name SIMNAD (Simulation and Modeling of Nanodevices) ver
wendet
korrekten und numerisch stabilen Rahmen fr
wird, erlaubt die Simulation einer Vielzahl

von
von
Strukturen, einsch
Daten
liesslich
GaAs/AlGaAs Heterostrukturen und Silizium Einzelelektronen-
Transistoren. Um eine bessere

zu
berein Stimmung
nt
experimentellen
zu
erreichen, mssen allerdings weitere Untersuchungen

zum
verschiedenen
Problemen, wie
Beispiel
der
Behandlung
eng beieinander
liegender
Kon
takte auf aktiven Oberflchen oder der Rolle diskreter Dotierzentren, vorge
nommen
werden.
Chapter
Estupor
1
cose non conte.
in'ercm le
was
My
mind
stunned
by
what it did
not
know.
Dantia.
^Purgatorio,
XVvx
Introduction
1.1 Evolution of
single-electron
by
devices
Since the invention of the transistor
John Bardeen and William

to
Shockley
in
1948 constant miniaturization has lead

the LOO
tint
devices with dimensions well below
limit.
soon
Originally designed
to
simply
more
emulate the
vacuum
tube,
the transistor
potential for speed, power consumption and reliability. In its most widespread configuration as metaloxide-semiconductor field-effect transistor (MOSFET) these potentials were proved
to
provide
lot
fully exploited
and initialized what
we now
call the
information something
age.
The relentless quest for device miniaturization is with semiconductor

This search is
that has been
technology since its early evolution starting in the 1950s. driven by a few facts which became obvious at the latest when
the first steps towards
integrated
a
circuits where made in the late 1950s.
First,
the reduced volume of
miniaturized transistor allows for lower power dis lower
sipation. Second,
nects
the
consequently
capacitances
and shorter intercon
lead to faster systems.
ever-increasing chip sizes. taining 100 million transistors

three years.
Finally, relative costs drop with more devices on According to the latest "Toad-map" [1 ], chips con
will start to reach the markets within the next
The end of the millennium
sees
the silicon-based semiconductor
industry
Chapter
1.
Introduction
well at the
height
of its evolution with
even
further transistor miniaturization
ahead. But nevertheless, the

are
scaling
to be
limits for the classical MOSFET structure
already beginning
to
show. Transistors with channel
lengths
of
few tens
car
of
nanometers are
predicted
governed by
the
quantized nature
new
of the
riers. Besides the efforts to
overcome
the limitations intrinsic to the classical
MOSFET structure, there is

not
an
intense search for
device
principles
that
only
allow for the
scaling
limits to be
pushed
even
further but make
use
of
the
quantized
nature of the
charge
carriers evident in small structures.
Since the 1980s,

ory have lead to
a
developments
new
in both semiconductor
field of research
technology
and the
completely
operation
is based
potential
of the
barriers.
focusing on devices whose on the discrete nature of electrons tunneling through thin These devices which exhibit charging effects including
are
Coulomb blockade
referred to
as
single-electron
breathtaking.
devices.
The
potential
single-electron concept
to
is rather
one
In
principle,
it should be
possible
operate in the limit of

Semiconductor
transferred carrier per bit, the ultimate

enormous
limit of semiconductor based

overcome.
logic. However,
for
obstacles have to be
technology
of very
repioducible manufacturing
rate
single-electron devices requires the small conducting islands and their accu
positioning with respect to external electrodes. Tt was not before the mid 1980s that especially lithography technology reached the deep sub-micron range making structuring in the 100 nm range possible. In the mean time,
single-electronics
tures is
has matured and
variety
of concepts and device architec
available. A
typical
device structure is shown in

on
Fig.
1.1. As
are
most
structures in the
field, this device is based
the
TII/V-compomid
material
GaAs
AlGaAs
1 m
Figure
1.1: GaAs/AlGaAs heterostructure quantum dot. The metal gates
are
structured
using electron beam lithography (Stanford University).
1.1. Evolution of
single-electron devices
system GaAs/AlGaAs which has the advantage of providing

a
very clean in
terface, therefore being especially suitable for heterostructure devices. For
long
time, semiconductor heterostructures have been used to achieve two-
dimensional electron gases
(2DEGs)
in HEMTs
(high-electron-mobility
are
tran
sistor). Heterostructure single-electron

ever,
transistors
related to HEMTs, how
with the difference that the 2DEG is

at
patterned by
electrostatic contacts
a
(Schottky contacts)
island, the
regions
so
the top of the device in order to create
small
charged
called quantum dot which is
separated from the
source
and drain
by
tunnel barriers. The electron gas in the quantum dot is
essentially
zero-dimensional
the term
(ODEG) exhibiting properties

atoms
similar to atoms. This led to
artificial
for this kind of structure

were
[2]. GaAs/AlGaAs single-
electron transistor
(SET) prototypes
Therefore,
build
as
early
across
as
1987.
These de
vices had islands several hundreds of nanometers
because of techno
logical
limitations.
they
can
had to be cooled to temperatures below
I K to make observation of
single-electron charging only

be
effects
at
possible.
Devices,
however, whose functionality

atures are somewhat
guaianteed
cryostatic temper
impractical for real-world applications. Consequently, one of the major driving forces behind the development of single-electron de vices was the quest for room temperature operation which requires the island
sizes to be in the sub-100
nm
range.
transistors were realized in silicon.
Just
recently,
the first
single-electron
Silicon is the basic material of modern semiconductor
technology,
20
and there-
nm-/ A*-
Figure
1.2:
Single-electron Right:
transistor m
the device
structure.
SEM
technology. Left: Schematic of micro-graph of the channel after oxidation

SOI
(Princeton Universit\).
Chapter
1.
Introduction
application of the single-electron concept for memory- and switching devices. Figure 1.2 shows one of the most prominent structures processed in silicon-on-insulator
fore, provides
one
of the most
promising paths
towards industrial
(SOI) technology.
is
The central island is formed
by
10
nm
silicon
grain
which
reduced to its final size
by
oxidation. The
size
of the island allows the ob
servation
of the Coulomb blockade of conductance at in

as
general higher
tem
peratures
in GaAs/AlGaAs structures
L3).
a
single-electron transistor memory device f4J with a self-aligned quantum-dot floating gate (Fig. 1.3 left). Due to the self-alignment, the floating gate is accurately posi tioned above the channel wire. If additional charges tunnel from the channel
A similar
technology
was
used for the fabrication of
Si
onto
the floating sate, the threshold voltase shifts and
memory effect simiwas
lar to classical
even
floating gate
room
transistors
is
observed. This device concept

a
able to exhibit
temperature operation which is
prerequisite
for
application [4],[5]. Figure 1.3 (right) shows a side view of a similar structure [6]. Other concepts pursued comprise multiple nano-crystals coupled to the channel of a conventional MOSFET [7] which is especially interesting, since it could minimize offset charge effects.
further industrial
single-electron concept is by no means clear. Several problems are not solved yet. The major flaws of single-electron technology are
The future of the
300
nm
Oxide
Figure 1.3: SEM micro-graphs of quantum dot flash memory devices. Left: Self-aligned floating gate on a narrow channel FET (Fujitsu). Right: View of a quantum dot flash memory device along the channel (IBM).
1.2.
Modem semiconductor device simulation
the
sensitivity
to
random
capacitance
parameters.
How
to raise
background charges, the necessity for lower junction operation temperatures, and the sensitivity to fabrication
exactly
SET devices
can
be used
as
building
blocks for further

several
logic
circuits is yet another unsolved

have been
problem.
Even
though
logic
schemes
proposed, none is presently threatening to oust the semiconduc tor standard CMOS (complementary metal-oxide-semiconductor) technology. Some of the schemes, however, are similar to CMOS technology where bits are represented by voltage levels. Nevertheless, it remains unclear whether these devices have significant gain. Other concepts are based on a bit rep resentation by charges. An intriguing, but as-yet-unrealized proposal in this direction is the quantum-cellular automata (QCA) concept [8]. Mesoscopic cells of quantum dots can be polarized, and since the polarization of a cell can effect that of a nearby cell, a polarization signal can be propagated along a chain of cells. Together with principle problems regarding the read-in or read-out of polarization states and the stability of the states it is not clear that there will be any speed or density advantage and low temperatures seem to be mandatory.
It is clear that silicon MOSFETs have their limits. When these limits will
be reached is not
so
obvious.
The demise of silicon
technology
has been
predicted (and postponed)

semiconductor
many times in the last decades. At the
same
time,
companies have grown increasingly reluctant to work on new technologies whose potential might be exploitable in a far future. In any case, miniaturization will remain an important aspect of VLSI circuits and singleelectron related effects will almost certainly plague, and perhaps, eventually
revolutionize the semiconductor
industry
in the next decades.
1.2
Modern semiconductor device simulation
From the
very
JO
besinnins of semiconductor technolosy it O

OJ
was
thought that
O
im-
merical,
physics-based analysis of devices could help a great deal in their un derstanding. Nowadays, simulation and modeling of semiconductor devices
one
both at the process and the de\ice level has become
important development methodologies industry eventually has developed into an ever-growing industry itself. Complete TCAD (tcchnology-computer-aideddesign) suites, that include process-simulation- and device-simulation tools
in and
of the most
connected to advanced
meshing-
and
optimization
software
are
readily
avail-
Chapter
1.
Introduction
able from many different vendors.
today are essentially partialdifferential equation (PDE) solvers, that are specialized to solve a particular system of PDEs relevant to semiconductor modeling. These so-called semi conductor (device-) equations describe the static and dynamic behavior of car
All of the device simulation tools available
riers in semiconductors under the influence of external fields that

ations from the thermal
cause
devi
as
equilibrium [91.
The system of
equations
known
drift-diffusion equations
consist of the Poisson
equation
+
V-(eV</>)
the
-</Q>-/i
and holes
C),
(1.1)
continuity equations for electrons

_,
on
-qRn
=
V7T
VJP
i +
9P
(Ifff
qBp
and the
current-density equations for

J
=
electrons and holes
qp,nnV
ql)nVn
Jp
where
qfippV
qDpVp.
<f is the electrostatic potential, n and p are the electron- and hole den sities, respectively, G is the total density of ionized donors, J and Jp are the current densities and Rn and Rp can be understood as functions describing generation and recombination of carriers. The material dependent parameters are the permitivity (. the mobilities // and up, and the cam er diffusion con stants Dn and Dp. Numerical solutions of this system of equations, maybe extended by the energy balance equations with the electron and hole temper ature as variables (hydrodynamic model), is the main goal of classical device
simulation.
The drift-diffusion
the flow of
equations are macroscopic equations which describe charged particles using appropriate models for their macroscopic important physical
are
physical
behavior. The most
models
on a
are
for the densities
of states of electrons and holes which
based
bandstructurc model,
rates
the carrier mobilities // and //p, and the and
generation/recombination
are
7?n
Rp.
The semi-classical models for the densities of states of electrons and

based
on
holes
are
the
assumption
an
that the carriers

mass
freely moving particles

Fermi-Dirac statistics.
in
parabolic
bands with
effective
//?*
obeying
1.2. Modem semiconductor device simulation
This
simple assumption, however,
is not sufficient for carriers in

are
narrow
MOS
lower
FET channels for instance. The local carrier densities than the
significantly
riers, i.
classically predicted ones. Needed is a quantum treatment of the car e. the solution of a (usually one-dimensional) Schrdinger equation
quantum effects is still
a
conceptually and numerically dif ficult problem due to the nonlocal relationship between the potential and the quantum mechanically calculated charge density [llj.
[10].
Inclusion of
generation/recombination rates Rn and Rp should be calculated on the basis of physical models. In most cases this is a difficult issue and the agreement of the microscopic model with the experimental data is sometimes rather poor. Common is the application of simple analytical functions containing various adjustable parameters which guarantee the required fit and cause no numerical problems. However, the drawback is obvious. Since a multitude of parameters with no apparent physi cal meaning is available for tuning the functions, almost everything can be cal culated and almost no certainty about the validity of the results can be given. Therefore, the development of advanced physical models for the parameters in the device equations is a mayor quest for nowadays device simulation [12].
A last
The carrier mobilities // and pp and the
point
should be made.
Much of the
success
of device simulation
is due to the
underlying
process simulation.
Process simulation is the first

a
step in modern device modeling. The particular steps in the fabrication of

semiconductor device
deposition, etch ing and implantation of dopands are simulated. The result is a representation of the geometry and a doping (-density) profile which is subsequently used to automatically generate a simulation mesh by means of a mesh-generation software. Process simulation is in particularly demanding since problems as
as
(the semiconductor process) such
moving
boundaries
are
difficult to handle, and instabilities of the simulation

are
due to the mesh
quality
still
rather
common a
mesh has to be
generated manually using

is that
a
problem. Alternatively, the mesh-editor. The disadvantage of

are
this
procedure by
complicated
over
three-dimensional meshes
hard to gen
erate
hand and
control
the mesh
quality
is difficult to achieve.
Also,
manual mesh
generation
makes
within the scope of TCAD
complex optimization geometries experiments (DoE, design-of-experiments) almost
of device
impossible.
Chapter
1.
Introduction
1.3
Single-electron device
equations
simulation
The drift-diffusion
described in the
propriate
mathematical model for
previous section provide no ap single-electron devices. Flere, the transport

tunneling
electrons. If the trans
is dominated
by
the
quantized
an
nature of the
port description for

mechanical
electronic device
requires
the inclusion of quantum-
properties of the carriers, the choice is between two models (Fig. 1.4). Dynamic models on one hand are based on equations of motion for Green's functions or for quantum-phase-space distributions such as Wigner functions. Kinetic models on the other hand use time-independent representa tions of the quantum-mechanical states solving kinetic- or rate-equations.
Transport
stochastic in
arbitrary single-electron
sources
circuits
consisting by
a
of tunnel
junc
tions, capacitors, and voltage
is described
rate
equation for the
probability
distribution
P(AI)
-
^P(K)
at
V [P(M)W(M. K) /-/
My-K
P(K)W(K, M)],
(1.2)
where
YIk P{K)
=:
1 md
W(AL K)
is the rate for the transition from state

a
M to state K
[20]. The
state
variable is the electron number. For

to
given
to
state
of the system the transition rates have
be calculated solution of
according
Eq. (1.2)
system
parameters.
An iterative method based

a
on
for circuits
containing
cially
up to
few tens of islands is based

on a
was
introduced
by
Fonseca et. al.
[14].
An alternative
approach
Monte-Carlo method
[13]. A
commer
available program that combines the master
equation approach
[15].
with the
Monte-Carlo
approach
is SIMON from TU Vienna
All these methods

which treats islands
use
the results of the
phenomenological orthodox theory purely capacitively coupled

given
in
single-electron
or
transitions between
(metallic grains)
(an
its extension towards confined levels, the constant-
interaction model
account
of these models is
Chapter 2).
The
possibility to evaluate complex circuits con taining a multitude of islands and tunneling junctions. Even cotunneling can be included in the calculations. The method is only slowed by the effort that is invested in calculating the transition rates and the probability distributions. Most of the time, analytical capacitance results are used which retain the sim plicity needed for the numerical evaluation. The accuracy strongly depends on the physical models which are used to calculate the capacitances and the tunneling rates.
advantage
of these methods is the
1.3.
Single-electron
device simulation
Device
Modelling
Quantum
Transport
Formalism
?
?
Dynamic
Models
'"1
Kinetic Models
(Coherent Models) Density Matrix Wigner Function

Green's Function
E nvelope Function
Geometry
Structural Parameters "\
\
-*.
/
^
,.
Device
Operating
Numerical
Device
Conditions
A
Material Parameters
,-
jit
Models
Band Structure
Auxiliary Equations
T
Device Terminal
Properties
Figure
1.4:
Classification of methods for
numerical simulations
of quantum
transport in electronic devices.
For semiconductor structures the above mentioned methods

curate
are
not
a
ac
enough, especially
if
discrete energy spectrum is present at
small
island
(then often referred

are
to as
quantum dot).
A self-consistent
on
tion of the confinement
potential
is needed and methods based
descrip densityby
functional theory (DFT)
used. The
screening
of the carriers is treated
solving
an
nonlinear
(Schrdinger-)
a
Poisson
try. This method allows for

the calculation of the exact
detailed
equation for a analysis of single-electron charging

peak heights
realistic geome and
shapes
and
of the Coulomb-blockade
peaks [41].
A drawback of the self-consistent models is the numerical effort.
Such
calculations
As such
extremely costly, especially if employed for they arc mostly limited to a single solid state device,
are
an
3D i.
geometries.
e. to a
single-
analytical solution of the transport equation, Eq. (1.2), in the linear-response regime with respect to the junction voltage exists [26]. However, the hope is that device simulations can help to extract parameters such as the capacitances for the analysis of more complex circuits using
stochastic methods. This is the classical
electron transistor for which
path
in modern semiconductor
10
Chapter
1.
Introduction
device simulation. The first step is the actual device simulation
using
self-
consistent solid-state simulator. Then, in
second step,
a
parameterized model
a
(SPICE-model
more
for instance) is extracted and used in
circuit simulation for
complex
circuit environment.
1.4
Objectives
outlay
and the numerical
This work is concerned with the theoretical
imple mentation of a simulation software, designed to simulate single-electron tran sistors. The developed physical models and numerical algorithms, however, are in principle applicable in the much wider context of nanodevice modeling
and simulation. Two main
objectives
had to be reached.
First, concepts from

that describe the
a
conven
so
tional TCAD had to be advanced towards the nano-scale range,

lutions for the full quantum mechanical of electron transport in
including
equations
of electrons in quantum dots and quantum wires. Second,

nano
sound
properties description particles
structures,
including
the
tunneling
of
through arbitrarily shaped potential barriers, separating the macroscopic regions of the device had to be developed.
The
the quantum dot from
approach
chosen is based
on a
self-consistent
densities and the
potential landscape
conjunction
in
analysis of the charge (almost) arbitrarily shaped semicon

one-
ductor structures. Zero-dimensional confinement in semiconductor quantum

dots
can
be simulated in
with
and two-dimensional confine
ment in
quantum wells and quantum wires.

transistors
This framework is
applied
to
single-electron
els for the
by simulating
the quantum dot
using
various mod
charge density
of zero-dimensional electron gases. The

are
charge
the
den
sity
models for quantum wires- and quantum wells
applied
to
source
Using the results from the self-consistent structure cal culation, the tunneling rates through electrostatically formed barriers are com puted. With the Helmholtz free energy which is calculated from the potential and the charge distribution in the device, the linear-response conductance is computed.
All these concepts had to be
and the drain reservoir.
implemented
even
in
both advanced and
nu
merically
chanical
stable simulation software. State-of-the-art

had
to
algorithms
for nonlinear
me
equation solving
be
pushed
further to include the quantum

a
equations
necessary to describe
single-electron
device. The valid-
1.5.
Layout
of this work and
general
remarks
11
ity
of the simulation
approach
is
proved by simulating example performed.
devices.
GaAs/AlGaAs SET has been modeled and first calculations of the electronic
structure
are
and the conductance characteristic have been

with
The results
compared
Finally,
experimental findings.
was
the software
use
embedded in the TCAD suite of ISE AG Zrich.
This enclosed the
of the ISE data structures in order to utilize the
meshing
and visualization tools part of the ISE TCAD suite.
1.5
Layout
of this work and
general
remarks
This work has been duction to
organized in the following manner: In Chapter 2 an intro single-electron tunneling based on the orthodox theory for metallic
given.
on
grains
with extensions towards finite temperatures and confined states in semi
conductor structures is
single-electron tunneling in semi conductors. Based on a linearized master equation approach, a formula for the linear-response conductance of single-electron transistors is derived and limiting cases are discussed. Chapter
3 concentrates
the
theory
of
Chapter
4 is devoted to the discussion of the two main transport coeffi
cients that enter the transport
equation
derived in the antecedent
chapter.
The
first coefficient is the Helmholtz free energy which is used to evaluate the Gibbs distribution for the quantum dot
occupations.
The second is the tun
neling
rate
for
single-barrier tunneling,
formalism.
account
which is calculated
using Bardeen's
trans fer-Hamiltonian
Chapter
gives
an
of the numerical methods used. Both lineara
and nonlinear solvers
are
needed for the solution of
nonlinear
Schrdinger-
Poisson-type equation in a discretized form. The main focus lies on globally converging Ncwton-Raphson methods that are used to solve the nonlinear sys
tems
of
equations
involved.
Chapter 6 discusses the functioning of the self-consistent simulation proach at a particular example, a GaAs/AlGaAs heterostructure SET.
In
ap
Appendix A,
an
the Gibbs distribution for localized systems under the in

The
fluence of
rates in
equations for the transition single-electron tunneling employing time-dependent perturbation theexternal reservoir is derived.
12
Chapter
1.
Introduction
ory
are
derived in
dations of
Appendix B. Appendix C gives an overview over the foun density-functional theory and the local-density approximation.
some
Finally,
is used
remarks have to be made: The SI

the work. However,
one
(MKSA) system
of units
are
throughout
cm-3
nm
exception
is made. Densities
given
in
in order not to deviate too much from well established
common
usage and
is the distance unit. For all constants such

or
as
the Boltzmann
con
stant, Planck constant
any other constant all units
are
converted to MKS A
units, i.
e.
cm,
nm > m.
Chapter
Qui
2
ogni sospetlo;
morta.
si convien lasciure
ogni villa convien che qui si a

let all your cowardice die
Now here you must leave all distrust behind

on
this spot.
Dante,
> Inferno.
111^.
Introduction to
single-electron tunneling and

C oulomb blockade
2.1 Introduction
The evolution of
to the
consistent
theory for sinsle-electron tunneling
goes
back
early 1950s. An unusual dependence of the resistance of thin metal films was explained by Goiter as due to the charging of small metallic grains [16]. Transferring electrons through the barriers between the grains would
result in
charging
of the
the electrostatic energy
neighboring grains, by
and therefore, in
an
increase of
i?c-^where C is the effective
(2.1) EG
became later known
capacitance
of the island.
as
the Coulomb
charging
energy. This energy suppresses electron transfer unless
13
14
Chapter
2. Introduction to
...
it is
overcome
by
either thermal excitations at
temperature T, i.
e.
T0
-r^-
Be
(2.2)
or
by
an
externally applied voltage
V. i.
e.
Er
suppression of electron transport a new term was coined: Coulomb blockade (CB) of tunneling. Independently from Goiter, this effect was ob Yet another effect, served and explained by Neugebauer and Webb [17]. Coulomb exclusion became known in the late 1960s through the work of GiIf the voltage V is less aever and Zeller [18] and Lambe and Jaklevic [19]. then the threshold voltage Vt. the system is in the Coulomb blockade state. With the voltage V exceeding V\, Eq. (2.3), an electron tunnels through the barrier into the grain. However, because of the Coulomb repulsion it can not be followed by another electron unless the externally applied voltage V is in creased even further by AC e/'C. This can readily be seen by looking at the electrostatic energy of such a grain
For this
with the discrete

T <
charge Q
Ne and the
externally applied voltage V.

known
as
At
and
To
is
the electron number is

a
given
at the minimum electrostatic energy
N(V)
staircase-like function
commonly
the Coulomb staircase.
At temperatures of order of
corners
Tq
the staircase which otherwise exhibits

above
sharp
is washed out,
totally vanishing
Tq.
was
The first and
quantitative theory Shekhter in 1975 [20]. They

as a
for two barrier systems

used
an
equation
similar to
given by Kulik Eq. (2.4), treating
equilibrium Haniiltonian. A master equation for the time evolution of probabilities P(X) which contains all relevant infor mation of the system behavior was derived. This approach is the basis of the so called orthodox theory of single-electron tunneling. The theory of correlated tunneling as is used now was eventually developed by Averin and Likharev [21] in the mid-1980s and almost immediately supported by experiments on single-electron transistors [22]. In the early 1990s the orthodox theory was generalized towards the inclusion of effects of both charge and energy conser vation in semiconductor structures. Based on the pioneering papers of Meir et al. [23], Averin et al. [24] and Beenakker [26] a consistent theory that ex plains the principle features in the characteristics of single-electron tunneling
tunneling perturbation
to the
2.2.
Orthodox
theory of Coulomb
blockade
15
amplitude and lineshape is now well established. Throughout the 1990s this theory was used in many papers on Coulomb blockade devices (see for instance [27]). It proved to be extremely successful in combination with a self-consistent evaluation of the single-particle levels and the total ground state energy of the electrons based on density-functional theory (DPT) (see for instance [45]). Experimen tal findings related to the single-particle level structure such as shell filling effects have been evaluated and at least qualitatively been explained on a selfas
devices such
the width of the conductance
peaks,
their
consistent basis
[28].
2.2
Orthodox
theory of Coulomb
blockade
2.2.1
The rules of the game
theory
island
single-electron tunneling has as its subject the charging of a small (grain, quantum dot etc.). The orthodox theory focuses on the charging
of
of metallic islands with diameters d > 100

island is
given by
the
equation
limit
for
capacitance of 2ixeQtd conducting sphere C

nm.
The
such and
an
can
be estimated to be > 5 aF. The resistance of the
tunneling
barrier
Rt has
to be
above the
quasi-classical
Rf
>
i?Q
/,7r/2f2
6.4 kO,
(2.5)
where
the
i?q
is the resistance quantum.

rate V
=
This bias
is
can
be understood
considering
time The term
not the
tunneling
rfcR\
at some
voltage
r
=
V.
l
--
The
tunneling
between two
subsequent tunneling tunneling
events
eRt/V.
rt
tunneling
10"15
s.
time refers to the characteristic time of

event
junction relaxation,
=
duration of the actual

The
to
itself, which is of order of
F"1
according single
an
uncertainty principle states the duration of the tunneling event rêV > h. Hence. Eq. (2.5) means that rt < r, e. i. tunneling
events are well
separated
in time and
one
speaks
of
sequential tunneling
of
electrons.
Essentially,
this condition
ensures
that the wavefunction of
island electron is well localized there.
resistance
Rt is small compared
occurs
to
single-electron charging
and
tunneling the scale provided by Rq suppression of delocalized states are available for charge
In systems where the
transport1.
'Eq. (2.5)
parency
ensures
localization of electrons
at
the
effectively
suppress the localization and make
conducting island. Barriers with high trans single-electron manipulation possible.
16
Chapter
-15
2.
Introduction to
0.08
SET
operation
6.4
(R>
k)
0)
jT
<0
o
10
16
i\
i
101
0.80
D
\
10
17
1 TH/
cd
\10THz
\100I ll/ \
10
18
Room temperature (300 K)
102
8.00
S-H
\
L0C
101
Resistance
\
HY
LO3
80.0
102
(kO)
Figure
of the parameters for single-electron tunneling parameter r_1 ized by the tunneling frequencies f (i?t-C)~L. Room tempera ture operation is only possible in the lower right-hand side field between the
2.1: Plot
dashed lines.
As
already stated,
be
seen
the
temperature
must
be small
energy exceeds the energy of thermal excitations. it

can
enough that the charging From Eq. (2.1) and Eq. (2.2)
is present for
readily
that the Coulomb blockade
regime
kuT <C
2(7
the temperature
or
(2.6)
litis condition
can
be fulfilled either
by lowering
a
by
de
ex
creasing
the
capacitance
which
at
means
to reduce the island size.
Usually,
periments
are
performed
a
temperatures of
few mK and for structures with
island sizes of
Figure 2.1 shows a plot of the criti cal parameters for single-electron tunneling. For room temperature operation capacitances in the aF-range and island sizes in the sub 100 nm range (sil
few hundred nanometers.
icon:
d < 5 nm)
are
required.
These limitations hindered
experiments
at
and
detailed evaluation of the orthodox

a
theory especially
in semiconductors for
long
time.
Nowadays,
to structure
with advanced nanofabiication
techniques
are
hand
it is
This
possible
is m
islands of known geometry which
separated by
"quantum-electionic devices" which opeiate above the limit given by the tesistance quantum Ot com se, single-election tiansistois use quantum piopeities ot mattet, but so do conventional bipolai oi held-cftcct tiansistois (Likhaiev L25J)
contiast to the held ot
2.2.
Orthodox
theory of Coulomb blockade
17
well defined barriers and
can
be acted upon
by
leads
connecting
the island to
voltage
controlled reservoirs.
single-electron transistors make them candidatedevices for further ultra-dense logic. However, care has to be taken with re spect to power consumption/dissipation. The power p of switching in singleelectron devices is given by the charging energy, Eq. (2.1), divided by the
The small diameters of
switching
time
R{C
IP
ryP
2 C
"=-r=W*Assuming
that silicon-based
(2'7)
technology will still be the technology of future electronic devices, one can take the "road-map" 11] estimate for the year 2012 of 10l2 transistors/chip realized with the same number of single-electron tran sistors. According to Eq. (2.7) a room temperature operating SET would have
a
power
consumption
mean a
of p
10
~8
(I?t
25i?Q
160 kO and (7
aF).
This would
are
total of
104 W/chip
has to be
dissipated
if all transistors
switched at the
same
time. With lower tunnel resistances
R\,
and therefore
higher operational frequencies (Fig. 2.1) the power consumption is increasing linearly. A decrease of the capacitance even leads to a quadratic increase of the power consumption as can be seen from Eq. (2.7).
A last very
important point
at the island
is to be made. At temperatures close to
zero
integer multiple of e. In such systems, tunneling events which occur via two tunneling junctions are not independent the island anymore. The tunneling events are correlated by the need to keep
Kelvin the
charge
is
an
charge
an
integer multiple
of
e.
However, each of the charges
at
the two
junctions can be external gating.
fractional which makes it
possible
to tune
the
junction by
2.2.2
Single-electron
and
box
Charging
energy of
de-charging
of
single
island is related to
changes
an
in the
Hclmholtz free energy F of the system. Within the orthodox model the free
a
single-electron
in terms of
box
(Fig. 2.2)
under the influence of
external
gate is written
macroscopic charges
F(X)-=
dQ(Q),
(2.8)
Chapter
2.
Introduction to
where
potential at the single-electron node and Q the discrete charge. A system of capacitors is described by a capacitance matrix, which relates potentials and charges at the circuit. For the single-electron box as in
4>{Q)
is the
Fi". 2.2 this is
width
Cv
I
\
-C
(Jy
(Jo
Rewriting
of
Eq. (2.9)
in terms of the unknown
quantities (Qg,
Cx-CCoC^1
-CCr
C
CC
which determines the

itance is
sum
potential
sate
=
Co-, and Va is the
single-electron node </>. The gate capac voltase. The total capacitance CV] which is the
at the
of all
capacitances C-%
C +
(7g
is assumed to be
independent
of N.
With
Eq. (2.8)
the free energy takes the form
F(X)
The first term
on
Acr
=
-Nc-
C
CJ]
-V
2(
(2.11)
or
v;
the
right-hand
side represents the electrostatic
charg
ing
energy
U(N)
of A^ electrons
corresponding
to
Eq. (2.1).
The second
term
gives
the work
(potential energy)
voltage source. It is as sumed that only one gate has significant capacitive coupling to the island. An external charge QcyLi C&Vg is intro
done
by
the
duced which
can
be understood
as
an
externally
be varied
induced
polarization charge
can means
which is bound to the gate field and
continuously by
contrast to
of
gate voltage in
restricted to
terms
which is
In
be
of
as
integer multiples of <?. (7ext the free energy can

Xe
=
Figure 2.2: Equivalence a single-electron box.

<-/exext
circuit for
written
F(X)
The last term in
ext, Qcy
2C V!
2C\
A
.
(2.12)
Eq. (2.12)
is
free energy variations with

N electrons at the island is
again independent of changing gate voltage.

Z"-] exp(-J[F(AT)
Figure
The
displays the probability of finding
2.3
given by
--
X)
EFX\).
(2.13)
2.2.
Orthodox
theory of Coulomb blockade
19
where Z is the the reservoir

a net
AMndependent partition function, Ep is the Fermi energy of which can be related to the ground contact of the circuit. Having
allowed numbers of electrons
and N
means
at
of
flowing through the device, the the island must be fluctuating between Ar electrons is then AT 1/2 (Fig. 2.3). This
current
-
1.
The net number

are non-zero
that there
1). 1\X probabilities for two different numbers of electrons, i. e. P(N) an electron Midway between these degeneracy points adding (or removing) This is equivalent to the moves the free energy away from the minimum.
=
condition
F(N)
F(N
1)
p{X)
FJj? (energy-conservation law) for

into this condition
the entire system. Substitution of
Eq. (2.12)
gives
(2.14)
If
eQ ext
Cv
where
EF
is defined
as
the renormalized Fermi energy of the island. energy, the constant

=
one
neglects changes in the Fermi

increments
Fiee
spacing
of the island
charge
by
one
is
simply AVg
e/CR.
Ew
-
en eis y
F
Q
=
-(N
l)t
Q
'=
-
_(\
)e
Q
=
~(N
-JVe
})e
(TV
l)e
Ae
-(TV
l)e
\c
-(
De
Chai ge
Q
Nt
(N
\)e
(N)
Ep
--
Gate
olta.se F
Figure 2.3: Free energies (top): as the gate voltage is increased the charge Q l)e to (N 1/ l)e. for which the energy is minimized changes from (X' the gap in the Fenni Fermi energies (bottom): when Q l/2)e (X

energies vanishes and
current can
flow (after [2f),
20
Chapter
2.
Introduction to
...
2.2.3
Single-electron transistor simple

tunnel
Now consider
junction
The island is located between two It

can
(Fig. 2.4). tunnel junctions with capacities C\ and C2.

one
circuit
containing
island
electrostatically
be controlled
by
the gate
The total
is called
but with
capacitance of the central island is a single-electron transistor. The device

a
capacitance (7g (Fig. 2.4(b)). (A C\ + Co X Cg. Such a setup

=
is reminiscent of
two
MOSFET,
small
conducting
island embedded between
tunnel
junctions,
instead of the usual inversion channel.
First,
circuit
reduced circuit that contains
only
the basic parts of the SET, the
island and the two tunnel junctions in series
analysis
reveals that the
(Fig. 2.4(a)) is considered. Simple voltage drops across the junctions are
,
Vi
C2V
+ Ne
Q(
C\
and
T~
I
CxV-Nc
^
C\]
Qo
(2.15)
Ne-\-Qo. Q2 -Q1 + Q0 the two capacitors Qi and Q2 are given as Q\ The charges at C\Vi and C2V2, and Qo is some background charge which is due to polarization Q2 of the island related to charged defects in the vicinity (not contained in the circuit Fig. 2.4). The electrostatic energy is given as
where V
F1
V2 and the island charge is Q
U(N)
-2 G Co I
=
A>
-Q(
2C\
is
(2.16)
and the work done
by
the
voltage
source
(a)
Ci./
C2.R2
(b)
Ct,Z?i
Ch.Ri
cg
gy
Figure
2.4:
Equivalence
tunnel
circuit
for
SEP structure:
(a) basic
circuit consist
ing of two
junctions, (b) capacitively coupled
SEP. The dashed lines
mark the island.
2.2.
Orthodox
theory
of Coulomb blockade
21
W{NuNz
where
(N1C2XN2C1)eV
Cy.
=
(2.17)
Xi
N2.
ca
N{ and X2
are
the electron numbers at the junctions with N
Additional control of the island
charge
-->
is obtained
by introducing
using
1
on
-^
gate
pacitance Cg (Fig. 2.4(b)).

tions with the substitutions
The free energy is calculated
above equa
C\
=
Ci
a a
oCg,
factor
1.
Co
C2
(1
^)Cg
an(f
Qo
the
~~~>
Qo
+ (
g(^g
f/)
Qey.\
is
=
depending
the circuit. For
the circuit in
question
the choice is
Consequently,
is
the free energy of
capacitively gated single-electron
transistor
FIN1,N2)
(Ci
=
C\)C2V2
(Xc-Qcxi):
[XiC2
+
2CV
N2{Ci
ex
Cg)]cV
(2.18)
with
The condition for the transition from
state with
higher
energy to
state
lower energy
can
be evaluated.
The
change
in free energy for
an
electron
tunneling through
either junction is
e
=
AF(Ni
1, /V2)
-{Ne-Cf,Vg~QQ
C2V)
+
AF(N,, Aro 1)
e
=
ex L2
(~^
C&V&
^0
(C\
Cg)V)
At
zero
temperature only transitions with AF

V
Slope
=
are
allowed. Evaluation of
Cs,
es \
Slope
Ci J-C-.
-Ic
-V
C\
Figure
tion is
Stability diagram for the single-electron transistor. Stable opera only possible in the shaded regions. The system is assumed to be free
i.
e.
2.5:
of background charges,
Qo
0.
22
Chapter
2.
Introduction to
...
this condition leads to the
stability diagram given
in
Fig.
2.5. Stable
operation
is
only possible
for values of V and
Vg
within the shaded
regions.
cir
Figure 2.6(a) displays

cuit
the iV-characteristies for
symmetric junction
R2.
For
(Fig. 2.4(a)),
i.
e.
with
C2
and
h\
\V\
<
e/Cz
as
suppression of tunneling Coulomb blockade. Due to the charging energy, available chemical potentials. Only if the junction voltage V is above the threshold given by the charging
the current is
zero.
The
for low biases known

a
gap has
opened
in the
(a'
energy, the blockade

come
can
be
over
and
current
flows. Then the

an
junction
behaves like resistor. If
electron enters via
junction
one an
other leaves via
is known
as
junction (spatially-)
two.
This
correlated
(b)
tunneling
of electrons.
Figure 2.6(b)
characteristics for
shows
a
the
IVwith
circuit
highly asymmetric junction resis tances, for example R\ <C R2- Car
riers that enter
are
through
one
junction
____
3e
(A
c Zj
kept from escaping though the second junction by the high resis tance. Eventually, the electron will
leave
ever,
Figure 2.6: IV-characteristics of single-electron transistor.

charged
with carrier.
the
through junction
for
most
two,
how
an excess
of the time the island is

more
With
increasing
populate the island. The IV-curve exhibits a staircase-like characteristics, commonly referred to as Coulomb staircase. If the as>mmetry is reversed, i. e. Ri > R2, the island is
bias electrons have the chance to
more
and
more
depleted
and
descending
instead of
an
ascending
Coulomb
staircase appears.
Figure 2.7 (Fig. 2.4(b)) at

the r-axis
shows the
a
1Vg-characteristics
of I
'
bias
voltage
0. This
capacitively coupled SET plot is equivalent to a cut along

a
of
through
the
period c/Cg. In the current peaks to the value Vy2(i?i + Ro). 0 (and T peaks are sharp peaks at V
stability plot (Fig. 2.5). The current oscillates with instable regions, where correlated tunneling occurs,
These
=
the
the
so
called Coulomb blockade
0 K of
as
course), however, with

At finite tempera-
increasing
bias
voltage
the line width increases
well.
2.3.
23
2(lh+B2)
14
3e
Cg
Figure
2.7: J V -characteristics
->
of a single-electron
transistor for
vanishing
junction bias, V
0, and finite temperature.
tures some
electrons
can overcome
the Coulomb
a
charging
energy for values
of
Vg
between the
peaks.
The result is
non-vanishing
conductance in the
blockade
regions. Then, the conductance peaks aquire a finite width. In fact, Fig. 2.7 already shows the characteristics of an SET at finite temperature, since some level broadening is visible. The Coulomb blockade is almost completely
washed out at
sufficiently high temperatures, knT
e2/(7
2.3
The orthodox electron
theory
of Coulomb blockade
was
developed
to
explain single-
charging phenomena observed in small metallic grains. In these grains the level discreteness gains importance only for structures with diam eters below 1 nm. Above this limit, the level separation is in general much smaller than the Coulomb charging energy e2/C's and the free energy is equal
to
a
capacitive
energy. These basic
for the free energy used in the

In
a
assumptions are previous section.
made in the
simple picture
semiconductor structure the island is called quantum dot. It exhibits
discrete level spectrum for much

and the level
larger
island sizes then in metallic systems,
separation
increases
a
dot diameter d.
Assuming by
indirectly proportional to the square of the symmetric square box of a given semiconductor
mass
material characterized
an
effective
nC with
potential
walls of infinite
24
Chapter
2.
Introduction to
..
height,
the level spectrum is calculated
according
to
op
(2.19)
where the lower effective
mass
in semiconductors
(GaAs:
in*
=
0.067) in
1) is the reason comparison to the mass of free electrons used in metals (m* for the significance of the level separation. If the level spacing Ae p\~i sp is of the order of the Coulomb charging energy. At" the Coulomb e2/(7s, blockade is affected by it. A quick estimate can be obtained for a GaAs (tr
13) quantum
dot with
2DEG of d
400
nm.
The
a
charging
is Ac
=
energy
can
be
=
estimated to be 1 meV
(using
the
equation
for
metallic
sphere
=
e2/Cs
e2/27ToTd,). According to Eq.

a
2.19 the level
spacing
0.03 meV. For

2 meV and
diameter of d
50
nm
the level
spacing
would be around Ae
the
charging
energy is estimated to be 8 meV. Island sizes of 50
nm are
well
within the reach of modern
lithography
theory
a
tools.
Even
though
the orthodox
can
still
serve
as
useful model for
complete picture including the interplay of the discrete level spectrum and the capacitive charging, especially of small semiconductor quantum dots is only possible if the electron gas is treated quantum-mechanically. The electrons in a QD comprise a many-body sys tem of strongly interacting Fermions in a complex potential. The quantum
mechanical states of such
a
single-electron charging effects,
system
are
the solutions of
complex system
of
equations taking
all
possible
interactions of the system and of the system with

The issue of
its environment into account.
finding
a
for the quantum states is omitted here and
discrete
meaningful expression set of eigenlevels with

an
energies ep (energies
manner.
relative
to
the Fermi
level) is introduced in
the
sum
"ad-hoc"
energy
The free energy F of the
QD is
of the of all
capacitive
according
particle
be
to the orthodox model and of the

one assumes
energies
occupied singlelevel, which

can
levels. If
that
one
electron
occupies
one
spin-degenerate
of
course,
the
following
o
form of the free energy is obtained
F(N)
i^t AepM; sp.

-
(2.20)
The actual values of ep need to be calculated numerically. Within the ortho dox theory the spacing of conductance peaks is calculated using Eq. 2.20 and
F{N
L)
F(N)
E-y\ yielding the following relation for the spacing of
2.4.
Example
structures: vertical
quantum dots
25
the conductance
peaks
'
e2
CS
:,+i-s,)
O'even)
;2.2i)
C e-yÂV:
Cv
e
Cs
where Al/'
(?' odd),
Vff
C? and V?
marks the center
position
of the ?'th
peak.
The
dependence spacing on whether i is even or odd reflects the spin-degenerated nature of the levels. The separation of the change of the free energies of a QD into a capacitive part which is constant and a fluctuating part related to the discrete spacing of the single-particle levels according to Eq. (2.20) is known as constant interaction (Ci) model.
of the level
2.4
Example
an
quantum dots
In this section
illustration of the main effects related to
neling
in semiconductor nanostructures is
structures in
a
single-electron tun given by shortly discussing vertical

a
quantum dot
which allow
GaAs.
Vertical quantum dots have thin barriers
fore, these
precise control structures are especially

very
of
just
few electrons
(Ar
<
20). There
suitable for
studying
effects related to the
discrete energy spectrum of quantum dots.
Figure 2.8(a)
ture
shows
schematic
diagram
of
vertical two terminal struc

two AlGaAs barriers.
[32j. The InGaAs quantum dot is located between
The two barriers have different widths. The inclusion of Indium reduces the
(b)
UGa V
fnGj V-,
\lGaA>
10
n-GaA>
J.
-1 0
UliUJl
(V)
GATE VOLTAGE
Figure
2.8:
(a):
Schematic
diagram of
for
a
vertical quantum dot.
(b):
The
Coulomb blockade oscillations
500
nm
dot
[32 f.
26
Chapter
2.
Introduction to
bandgap
electrons
so
that the conduction band is below the Fermi level. This accumulated
even
means
that
are
is surrounded
to
by
circular
junction voltage V is zero. The dot Schottky gate placed at the side of the mesa close
gate voltage
when the
the quantum dot. With
Vg applied,
the
quantum dot
can
be
squeezed electrostatically.
Figure
Coulomb measured
the
at
2.8(b)
blockade small
shows
the
1 t)
1?
^"*
oscillations V.
of
Due
to
s
small
=
number
electrons
b
(N
is
...
20) the peak spectrum

influenced
significantly
by
the and
strong
Coulomb
interaction
quantum-mechanical
few-electron
effects in the
Since
the
regime.
electron gas in the dot has the
shape
of
circular disc and the

is harmonic,
confining
\'-0
potential
exhibits
a
the system of symmetry

: i
,
high degree
with
states.
degenerate
single-particle
Somee-Diain
Voltage (mV)
the quam
With the arguments of the

stant
con-
Figure
turn
2.9:
Stability plot for

in
interaction
model,
different
dot
structure
Fig. 2.8 [32],
spacings should lead to dif ferent spacings in the conductance peaks (Eq. (2.21)). Moreover, the
level
shell
structure
of the levels is also

<
c
reflected in the
height of the indi vidual peaks. Strong modulations of both the peak height and the spacing of the conductance peaks, especially for the electron num
bers N
=
-50
...
12
can
be
seen
in the
V(mV)
Fig. 2.8(b). Figure stability plot, i. e.

conductance
2.9
displays
the differential
Figure
2.10:
Coulomb staircase
for
vertical dot
OP/dV
of the structure
[33].
in
Fig.
2.8.
The electron number A' is fixed in each of the white diamondwhere Coulomb blockade is present
areas
shaped
ular
areas,
(Ol/OV
0). The irreg

and the
shape
of these
is also
an
effect related to the level
spacing
2.4.
Example
quantum dots
27
shell-filling. pronounced.
One
can
clearly
see
that the numbers 2, 6 and 12 the
are
especially
an
These numbers
are
magic numbers of
the two-dimensional
harmonic oscillator
potential,
i.
e.
those electron numbers A^ at which
ad
ditional shell in the level spectrum is
occupied.
a
A measurement of the Coulomb staircase for shown in
similar structure
are
[331 is
Fig.
a
2.10.
At
negative voltages C
whenever
an
electrons
accumulated in
the dot and
step
occurs
electron is
trapped.
The few-electron
regime
is
again reflected by
the low number of electrons which leads to modu
lations in the width of the
plateaus
of the staircase due to the strong Coulomb
interaction. The width of the staircase steps is which hints

more on a
shift in the total
larger for low electron numbers capacitance Cyj towards higher values for
electrons.
Is*
iL
"
H
>
|
w
,
(,
' ^
Chapter
And
now
3
note.
Ora incomincian le dolenti
the notes of
anguish
start to
play.
Dante. >lnierno, V<
Theory of single-electron
tunneling
3.1 Introduction
in semiconductors
In the last
chapter
an
introduction to
single
electron
tunneling
was
basis for the theoretical
predictions
was
the so-called orthodox
given. The theory and the

are
constant interaction model. The
shortcomings
of the orthodox
theory
evi
dent.
Equation (2.21)
is sufficient to describe the
spacing
of the conductance
peaks. However, no information about their amplitude and lineshape can be obtained. This requires the solution of a kinetic equation. Furthermore, the assumption that the capacitance of the island is constant and especially that C>j is independent of the electron number is not valid in small semiconductor
quantum dots.
In this
chapter
linear response
theory is discussed, which leads
to an
equation for the conductance of a SET in the limit of vanishingly small sourcedrain voltages. The effects of finite temperatures and discrete level spacings This are taken into account as well as the shape of the tunneling barriers. analysis was first performed by Kulik and Shekhter 120] for metallic islands within the orthodox theory and extended by Averin et al. [24] and Beenakker
[26]
to
semiconductor quantum dots.
29
30
Chapter
3.
Theory
of
3.2
Model definition
The model for
single-electron
a
transistor consists of
confinement
region
Each
(island, quantum dot) with

level carries
one or zero
an
discrete level spectrum eh
{i
0.1,...).
electrons. The
n,
=
occupation of the
?th level is therefore
described
tem, i.
by
integer number,
0,1, and the
occupation of the
is
level sys
e. a
specific occupation
to two
are
number
configuration
and X
=
given by
barriers
the set
{n7}
where the number of electrons is

is
integer
source
]C?
n,. The confined
weakly coupled
electron reservoirs via

named
tunneling
are
region (junctions).
The reservoirs which
and drain
taken to be in thermal
(a)
Gate
capacitor
Tunnel
junctions
(b)
A
^

t>(X)
F<f
m
i
rjeV
co
eV
Figure
3.1:
(a) Schematic drawing of

on
SET.
(b) Potential
cut
through
the
(long dashed lines). The chemical potentials p(N) according to discrete numbers of electrons are indicated by straight lines. Electron s in the reservoirs continuously occupv all energies up to the Fermi levels.
structure.
Electrons
the island occupy discrete levels
3.2. Model definition
31
equilibrium at temperature T and Fermi energy Ep, is applied between the reservoirs is small.
The device is considered to be driven in
tance
i.
e.
the
voltage
V which
linear-response,
)
i.
e.
the conduc
G is defined
as
I/V
in the limit V
0. In the absence of
charging
Ep
in
a more
effects
a net current
flows
with
through
the quantum dot if the Fermi energy
the leads is
aligned
one
one
of the discrete
single-particle
levels. In
general
aligns
set
case, at T
0 K current flows if the Fermi energy in the reservoirs
with
of chemical
potentials ft(N) of the quantum dot. The potentials {//(.A)} is given as the change in the free energies
of the chemical
of the quantum dot when
adding
or
removing
an
electron, and therefore, the
condition for the lift of the Coulomb blockade is
EF^f,{N)
The
F(N)-F{N-i).
by
sets
(3.1)
tunneling through
the two barriers is described

to
for all states k in the quantum dot
the left
(drain-) and
tunneling rates the right (source-)
of
reservoir,
by
the
{F^}. The characteristic energies of the system are given charging energy e2/Cv. the single-particle level separation Ae, the
and
{F^1}
thermal energy
k^T,
X>
and
nFt.
==
The last term denotes the intrinsic width of the
transmission
in the dot
resonance
/zFt
can
^(Fs
4-
Td).
It is assumed that for all levels

resonance
kuT
an
M~\, i.
e.
the finite width of the transmission
through the quantum dot

dot state
be disregarded. This allows to characterize each
by
<C
occupation
are
=
number. Furthermore, it is assumed that the
tion resistances
i.
e.
7-?q
smaller than
quasi-classical (Eq. (2.5) Rt RiR2/(Ri + R2) and the tunneling time rt the junction relaxation time r~ 1. This ensures that the
limit
=
above the
in Section
junc 2.2), F71 is
tunnel
ing
events are well
separated
in time and the current flow does not disturb the

to use a master
equilibrium.
It is therefore
possible
equation
to
describe the
following section it is assumed that inelastic scattering takes place exclusively in the reservoirs and not in the dot. The effect of inelasic
transport. In the
scattering
in the quantum dot is discussed in

an
|26j.
a
The transition of
in either the drain
The initial
or
electron from the kth state in the dot to

source
final state
the
reservoir, cf \
requires
energy conservation.
configuration
of electrons in the quantum dot is charcterized
by
the
Helmholtz free energy
F({ih}.nk
and the final
UX)
F{ni
'î.l.^tr- :N),
configuration by
m
=
F({h,},
0, N
i)
F(m,...
x.
o. m
*1-...
AT
l).
32
Chapter
3.
Theory of single-electron tunneling
...
For
the
given drain-source voltage V

banier is
n
=
the fraction rj of the
voltage that drops
over
source
dot chemical
Rd/(Rd X Rs). According to Eq. (3.1) potential fi(N) is given as

=
the quantum
/x(Ar)
F({n,}.nk
hX)- F{{n,},nk
a
0, N
1).
(3.2)
The energy conservation condition for
transition into
final energy state in
the
source
reservoir is
tl(X)
Similarly,
n(X)
neV.
a
(3.3)
transition from
an
the energy conservation condition for

source,
initial
reservoir state in the
ef (N),
=
to
the Ath dot state reads

+
S1(N)
N is the number of electrons
ii(N
l)
qeV.
event takes
(3.4)
before
the
tunneling
transitions between the dot and the drain reservoir the
place. For the following condition
applies
for the final state
|(N)
(3.5)
ef(N) =ij(X)- {l-ri)cV

and the initial state
e^(Ar)
ef(N)
The parameter //
p(N
i)
(1
ii)eV.
(3.6)
drops
out
of the final
ductance. The derivation of the
equation for the linear-response con conductance expression follows the linc-of-
arguments given by Beenakker [26].
3.3
Master
equation approach
3.3.1
Master
equation
probability
distribution
In
general,
the
P(A7 t)
is
time-dependent property
j\f of the system
in the
to
that describes the
probability
described
for the stochastic variable

a
have the value Ar. Transitions from able Ar to

a
state of the
system described by the vari
state
by
and
M will
produce changes
corresponding
occupation
factors
P(N. t)
P(M. f),
dP(NJ) =-dP{Mj) =P{Xj)W{X>M)dt.
3.3. Master
equation approach
33
For discrete stochastic variables the summation

processes results in the
over
all
possible
transition
following
master
equation
ot
P(N,t)
V [P(XPt)W(M.N) ^^
Myx
P(N,t)W{N,M)\,
(3.7)
probability distribution or transition rate (usually assumed to be time independent) for the transition of the system from state A' to state AT. The probability distribution is normalized according to
where
W(N,M)
is the transition
^P(AA)
x
1.
In
quantum dot the stochastic variable A'' is the electron number. Singleoccur
electron transitions between quantum dot states

trons in the
if the number of elec
dot is increased
can
or
decreased
by
as
one.
the master
equation
therefore be
given
the
stationary solution to solution to the following rate

The
equation
0=
Y,
A'={iU!}
[P{M)W(M.N)- P(N)W(N,M)],
(3.8)
where
W(N,M)
is the total transition
rate
from the iV-electron to the Mdistributions
electron system. The set of discrete
{P(N)}
is the solution of the
non-equilibrium probability master equation.
3.3.2
Probability
distributions and transition rates
stationary equilibrium probability-distribution function Peq({n7}) for particular occupation configurations {/?;} is given by the Gibbs distribution
The
of the
grand
canonical ensemble
theory
Peq({'h})
where
Z-1
exp
| -^-L[* v{h,},7V)
EFN}\
(3.9)
is the free
{n,}
is
energy of this
specific set of occupation numbers, F({nq},N) configuration and Z is the grand partition function
a
z=Y,
X<}
cxp
{-^r^{"'}'x) EfN^}
~
(3-10)
34
Chapter
3.
...
The
equilibrium probability
Peq(N)
=
for the dot
to
contain N electrons is
PeqW
The
Y, Peq({^})VEnt.
(3.11)
single-particle occupation factor for the dot levels is given by the conditional-probability distribution function g{ek\N), i. e. the probability that the th single-particle level in the quantum dot is occupied if the dot contains
N electrons
9{sk\N)
(from
now on
-j^r E ''.({ JW.L A,

reqCv
r X:}
-,
(3.12)
the
symbol gx(?k)
is used
instead). The
sum over
the
occu
pation configurations {r/,} on the right hand side of Fiq. (3.12) is the jointprobability distribution to have AT electrons in the dot and the AAh singleparticle level occupied. The Gibbs distribution for localized systems under
the influence of external reservoirs is discussed in
Appendix
A.
many-particle ground states of the iso lated quantum dot can be derived by treating the tunneling Hamiltonian as a perturbation to the time-dependent Schrdinger equation. An account of this approach is given in Appendix B, The total transition rate W is the sum over the partial transition rates of the reservoirs, i. e. the summation is over the
The transition rates between the
reservoirs
r
{s.r}
W{N. M)
=
]>^ JT71(AT,il/),
rates to or
(3.13)
from
which
are
given
as
sums
over
all
partial tunneling
are
specific
energy levels ek. Transitions to dot states
characterized
by
W1
(AT,
N + 1)
Y rA fttW
* k
fM1
1)
-
9N(k)]
(3.14)
-gN
W>(N-1,N)
This leads to
an
=-
r)/(^Y-
)[!
^(ek)}.
equations decreasing
increase of the electron number in the dot. The next
refer to transitions
from
filled dot states to empty reservoir states
the electron number in the clot
Wl(N,N
1)
rHl
Y V [Jk
-
f(f\(X)
/W(*V
E{;)}gx(sk)
1)
-
W*(N
F 1.
A)
^F)]gN+l(sk),
3.3. Master
equation approach
35
where
c^{
are
the initial and final states in the reservoirs,
is the Fermi-Dirac
distribution for continuous states
f(e
and g is the
tum
EF)
c
=
i + exp
E-p
-t
(3.16)
occupation probability
to
for the discrete energy levels in the quan
dot which is taken
be the Gibbs distribution for discrete
single-particle
[31].
levels
(Eq. (3.12))
in the most
general
case.
For
can
large
electron numbers the
Fermi-Dirac distribution for continuous states
be used instead
3.3.3
Detailed-balance
equations
for
With
the
Eq. (3.14) and Eq. (3.15) the stationary master equation, Eq. (3.8), stationary non-equilibrium distribution function P(N) is
o
=p(n
-1)[i-9y i(ek)]J2nf(t(x
-1;
-
r-ir
-P(N+
l)]T\gx+iXk Y1!^
-
f(A(NX 1)
-
EF]
(3.17)
-P(N)
-P(N)
This
X>
'
">
gN(ek)]Y,Tim(N)
*"\ Mil
EF)
EF)
>(*)] E1"-!1
be recast
/WW
equation can Eq. (3.12) which
using
a
the
leads
to
set
of
conditional-probability distribution equations for the non-equilibrium
probability-distribution
0
=
functions
/?({??,})
.)SnkA
Y P{ni,
k
..
A-1.
0. nk+u
X)rfr/(e(Ar
v
-
1)
Y P(/li
k
>n
J-
L rn^-
'
)(^.o/LrlP
i
^
f(^(N
1)
--
EF
Yp({^})snk^Yrif^x)
k
i
ef)
(3.18)
36
Chapter
3.
Theory
of
...
The first two terms refer to the system

tion
evolving
into the state with the occupa

or
{nt}
either
by adding
out
one
electron to the dot
by removing
one
electron
from the dot to the drain
or source
reservoir. The last two terms refer to the
system evolving
of the state with the
filling an unoccupied level tion (3.18) is equivalent to

dot level k
in the dot
a
occupation {n?} due to and an electron leaving the

equations,
an
electron
dot.
Equa
set
of detailed-balance
one
for each
p(m,...,iik. i,l,nkfl....)
=
Yv\[i
t..
KA(X
i)
i%)\
-
P{nu
<nk
,.
0, nH
..)
VrA/(eI(iV)
1
EF),
(3.19)
using
the notation N
Y2,^k
=
n>-^-
c-
X XI if the Mi state is
is
definitely
occupied, Skti,
and N
N if the A'th
state
unoccupied, Snk)o-
3.3.4
Stationary
to
current
According
rier
Kichhoff 's law the
steady-state
=
current
through
It is
the
source
bar
equals
I
that
through
the drain banier. I
Id.
given by
(3.20)
^]Tp(A)[IFs(Ar. A +
/v
l)
WS{N,N- I)].
equation
Applying
the results of the
previous
section this
can
be written
as
/=Ê/KiM-[i-*)]
X k
-
VI [1
f(cl(N)
E$)]gN(ek)}.
(3.21)
Again, the first term in the sum refers to an electron moving into the previ ously empty state k in the quantum dot. The second term refers to an electron
moving
from
state
k in the dot
to
the final
state
e^
in the
source
reservoir.
3.4.
Linear-response regime
37
3.4
Linear-response
conductance
3.4.1
equations in a general form, a restriction to the case of small source-drain voltages V is made. The conductance in the linear-response regime can be given analytically. As mentioned before, the linear-response conductance is defined as
G
=
Since it is
impossible
to solve
above
lim
v -m
I \
source
(3.22)
and the drain
Equilibrium
i.
e.
conditions
=
are
assumed for the
reservoir,
Pp. In this limit the non-equilibrium probability distri EF EF bution P({n7}) differs very little from the Gibbs distribution at equilibrium Peq({n?}). Therefore, the stationary non-equilibrium probability distribution
is
linearly expanded
in terms of V
p({nj)
Substitution of this
peq(W)(i ^^(K})).
+
(3.23)
expression
1,
A+l,
X
into the balance
equations, Eq. (3.19), yields

}nk
Peq(/*1.
x
1,
1+
cV
A'bT
-
T'(//L)...
-
!,l,/A-rl- )
{ri[l
f(c
qeV)}
r?[l
f(t
e
(I
q)eV)]}
..
Peq(ni,..
.nk.
i.O,
A-ti.
^1
l + -j-ty{7iu.
kB I
,nk_i,0,nk+1,. ..)
x{T*kf(t-\ ,,er)
+ r?/(e-
(X^q)eV)}.
(3.24)
witlK
F{{ih}ynk
l.N X
1)
F({/*,}./7A
0,
X)
PF.
Since
the balance
equations are linearized with respect to the junction voltage V, a Taylor expansion of the Fermi-Dirac factor including only the first-order term can be used in Eq. (3.24), i. e.
/(e
reVWlO-Mer/'tO
=f(t)
1
-
/(<
I X cxp
(3.25)
kXT
XT
38
Chapter
3.
Theory
of
...
and
/(-(I- n)eV)=f(e)-(l-V)eVf(() :/(<

1 +
(L--,)
1
'
CXP
y-^
l
kBT
;
eV
(3.26)
\
1
kBT
The following identities
are
substituted
l-/(e)
^r/'(e
and
/(t)exp( yy )
=
'
(3.27)
1 + exp
kBT
-/(*),
(3.28)
Pcq(ni,.
AU-1- I- A-
)
/U-1.0.
A
,
Pcq('M
!..
.)exp
/CrT
means
(3.29)
that if
Latter describes the so-called detailed-balance symmetry which

the level system is in
occupation
distribution
{/?,},
the
probability
zero
for the kth
level to be
occupied compared
are
to the
probability
that the level is empty differs
by
Boltzmann factor.
Another consequence is that at which would lead to the
temperature
are
no
transitions
possible
previously empty,
whereas at
occupation of the high temperatures both probabilities

leads to
a new
kth level if
equal.
Application
tions,
now
of these
equations
form of the balance equa
for the first-oder
non-equilibrium
correction
^({r^})
to the
equi
librium
probability
...
distribution
Peq({?i,})
l
#(t?i,
cnk
i.l.
n k
x{(Tl+Tt)-r
+
-[r|/,-lt(l-//)]
7?A rl>'
exp exp
kjiT
A-rT
[r^-r?(i-/7)i
>U-
-=^(1
X
ps
-1-0.
)
1 I + oxp
"pd\ i 1 j j T
knP
[Vln-Tdk(l -/;)]
ArT
3.4.
39
All factors
Peq
and
cancel.
Further, all
are
terms
all second order contributions
ignored
and
proportional to eV/k^T, i. a final equation for \I/({r^})
e.
is
obtained
^("i
?H
lX.n k
\ t'
"nd
1
*(m
A-i.o.ui-..)
A,
-
ArrVd
n+q
'?>
(3-3)
for which the solution is
*({.})
cou.t
J2 A
?
\1,
VA -I)t1,
'
pd
(331)
The first-order
77
=
non-equilibrium
correction
T'dn,})
if
/?
=
to
Peq({n^})
and F*
=~
is
zero
if
Tf/(r,
Tf)
for all
are
?.
This is the
case
1/2
Ff,
i.
e.
if
the
tunneling
barriers
identical.
linear-response current through the quantum dot in first order in V can be calculated. Linearization of the current equation, Eq. (3.21), after substitution of Eq. (3.23) and the subsequent use of the laylor expansion
With this result the
of the Fermi-Dirac factor,
Eq. (3.25), yields
~CH E rl^,({,})
I
1111 llll
1 +
\
II
'
lV?*(KK
I
t
{>%,}
x
{Sk o/U
rieV)
Snk^
/(<
>1<>V)]}
and
expanding
the Fermi-Dirac factor
(~=^YYrlP^{{n'])
k
{,}
{<5A
X
o/(0 -6nk i[l
f(c))Xdnk on<.Vf(c)
-
nktlrieVf'(e)
^
^n{>h})nk
*bJ-
o/(0
^({x})S>n
k!
X
i[l
/(e)]
Û({n,}Ku
k'B
orteVffe)
~*({>})îneVf'{t)).
A'BZ
40
Chapter
3.
Theory
of
...
Neglecting
terms
of the order
V2 (the
last two
terms)
one
arrives at
J=-
eV
kn
{J
r ht
x
cy-6>n
+
o/(e)
Ff^6>n if1
+
kXF
,,
"
.f(e)1
ûo^BT//(t)
-+
(5?u.1//A-BT,/,/(f)
-
*(K})<Wo/W
*({"J)<WU
/(*)]}.
bal into
The first two terms cancel because of

ance
symmetry). If
the
one
Eq. (3.27) and Eq. (3.29) (detailed identities Eqs. (3.27)-(3.29) are now substituted
obtains
the
current
equation
e~
Y E rA^q({^}){(û
A
6nk ,U>/"
/(<
1 + exp
{,}
( j^
(3.32)
^({" JM/u of (e)
^(KIKu ,i,/'(()
exp
k~
and
finally
( v
^ "
/vBi
;TrE
A
Erôq({J)^0/(e)
^ *
{n,}

[;/
^(l,
"A-l,
ClUî... .)
*(/?i
/iA-i.0,r7A4i.. )).
(3.33)
where the first two terms in
Eq. (3.32)
combine to
give
Pcq({J)((5A+Au l)/(0
1
-+-
Pcq({?j)û,0//.f(0:
exp
kB l
and the last two terms
give
Peq({n,}W{J)/(0 6nk
X
Lexp
A'B
hit
/'A
Peq(K})ô.f(c
0. /?A
L,
[^(1,..-
.A
i.L/?A-t--.-) -"^(/?]
i.
-*)]
3.4.
41
It is
now
nkto
the
c
=
possible to write replacement N
the current
N
can
equation again always be made, i.

-
in terms of N since for

e.
F({m,},
rtk
1. A-f
1)
F{{nt},nk
V leads
to
0, N)
PF.
(3.34)
for the
Substituting Eq. (3.30) and dividing by linear-response conductance
the final
expression
=ÊE r7CAP-({"'})'"<"/(t)The
(3-35)
particular product
between the Gibbs distribution for the quantum dot
P({n,})
fact that
and the Fermi-Dirac distribution
f(t)
for the reservoirs reflects the

an
tunneling requires
[26J.
an
an
occupied
initial state in the reservoir and
empty
state in the
quantum dot. Eq. (3.35) is the central result of the paper

The
same
of Beenakker
equation
has also been obtained
by
Meir et. al.
[231 by solving
Anderson model in the limit
knT
>>
hTt.
Within the constant-interaction model for semiconductor quantum dots al
ready
introduced in the is
previous chapter,
the difference in the free
energies
using Eq. (2.20)
F(nk
where
tem
LN+l)-F{ntl =0,iV)=fA +r(/y+l)- U{N),

denotes the
(3.36)
U(N)
with N electrons.
capacitive contributions to the free energy of a sys Inserting the Gibbs distribution P(N), Eq. (3.11), and
the
the
the
single-particle occupation numbers gy{sk), Eq. (3.12), linear-response conductance, Eq. (3.35), reduces to
2
oc
equation
for
pS ^
T^Y,i:rx^ûm^gx(x)]
K'B
L
x=o
X^X
f(ek
U(N
1)
U(N)
EF)
(3.37)
or
the
equivalent equation
O
OL
-fiq
Tirl
vKf X
A
iV1
E XXyfUX)gxle'']
k K
[1
f(sk XV(X)
applicability
U(N
I)
EF)].
has
(3.38) already
A first
assumption
that limits the
of these
equations
been made in the model definition:
one assumes
that the quantum mechan
broadening of the quantum dot levels is smaller than the thermal en ergy, hVt <C kifP. In the following section limiting cases for Eq. (3.37) and Eq. (3.38) related to the other significant energy scales are discussed.
ical
42
Chapter
3.
Theory
of
...
3.4.2
Limiting
fcBP
cases
In the limit
X>
Ae, the discrete energy spectrum may be treated

The Gibbs distribution is
as
continuum of states.
Dirac distribution, gjy
(ek)
f{ek
approximated by the Fermip(N)). with a given chemical potential
p>(N)
determined
by
The Gibbs distribution
Peq(A~)
is
taking
its classical form
PclAsdX)
cxp{-[i7(Ar)-f A>- EF)]/kBT}

-
Y exp {-[U(X) X N(fi

x
EF)]/kBT}
equilibrium. The summation over A' in Eq. (3.38) is replaced by an integration over e, multiplied by the density of states p(e). If A-BT <. //, Pp, the density of states and the tunneling rates are evaluated at using that ft(N) const for all N
potential
where //, is the chemical
of the quantum dot in
which contribute to the classical Gibbs distribution. The conductance becomes
Jf^jNXEY E îass(N)g(r(N)
if
U(N
1)
-f M
PF).
(3.39)
A^ft, PF,
with
dyf(y)[L-f{y
If furthermore
x)]
l'^["
charging
energy
can
9X)be
A'BP
^>
r2/CA,
e.
the
ignored,
and
Eq. (3.39)
two
with g
k*T
states that the conductance is
simply
that of the
tunnel barriers in series, i.
7^-^.
if
e2/rS.
As < kBT
<<y
, Pp.
are
(3.40)
The conductances of the
source
and drain barrier, (P and
Gd,
given by
the
thermally averaged
Landauer formula
i'
r
X j
_
'^s
GbU-
/
~T~
h
rr^ d/
.'
.
d-rcl(f)Tde
(3.41)
J0
3.4.
43
where the transmission
T(e)
divided
by
the
probability of a barrier T(s) equals \/hp{e), attempt frequency v{e)

=
the
tunneling
rate
rs-d()
If the
/?rs'd(c-)/)(f).
are
(3.42)
tunneling
rates at
high
barriers
assumed to be
ergy the conductance of the barriers for each side
independent of en becomes Gs,d e2Fs,dp,

=
evaluated at
Pp this leads
to the conductance of the
quantum dot
(j-_e
_
FTd
c2
_ -
PqPd
=
/>rs
+ pd
y f^sTjXfX
if
Clx-
eV^VAe<^BT<p,
Pf-
(3.43)
The conductance
Goo depends solely on the banier height and width which determine the transmission probabilities rPs,d, however, it is independent of the size of the dot. The assumption that the quantum mechanical broadening of the levels should be small, hTi <C kBF. i. e. the levels should be sharply defined together with p if As in Eq. (3.42) requires that 7ls,d <C 1. Hence,
~
the
general
criterion for the
tunneling
rates
of the barriers
implies
that the
conductance of the quantum dot is G <C
e2/h.
In with A"
Coulomb blockade oscillations become visible at low temperatures.

the
regime
where
kBT
to
<C
At. the
term
JVmm in Eq. (3.37)

are
gives
the dominant contribution to the conductance. All
Peq(N)
e.
negligible
small for Ar not
equal
-
Amm
-
or
A"mm
1,
=
so
that
Ninm
-
minimizes the value
A(iV)
0(A)
Q(N
1)
with
0(A)
FIX)
-
NEF. i.
-
A(iA)
and the
eN X
U(N)
U(N
1)
PF,
probability Peq(Ar)
("mm)
'
is
-*
eqcqv"mm'
exp[^0(Amm)/A-B7'] oxp[-n(A*mm)/A-n7,j + exPhO(Aw
l)/kBT]
J V^nnil/
Using
the
equivalence
/(.r)[l-/(.r)]
-A-B77'W.
given
as
(3-44)
the conductance in the low temperature limit is therefore
G^-e2l^-j^f{&mm),
if
A-BPAc.
(3.45)
44
Chapter
3.
...
This formula is
onant
as
equivalent
to
the
one
that describes
includes the
tunneling. However, Eq. (3.45)
thermally broadened res effects of charging as well
the discrete level spectrum.
3.4.3
Amplitude
and
lineshape
With the results from the

imations for the
previous section it is possible to give simple approx amplitude and the lineshape of the conductance oscillations.
regime. As
X
The classical
kBT,
was
first studied
by
Kulik and Shekhter
[20] within
the orthodox
theory.
to
-
A continuum of states is assumed to par
ticipate
in the conductance.
If As <C
the
sum
kBT
<C
e2/(7s, only
one
electron
number, Armm, contributes
in
A(AP)
U(N)
U(N
1)
+ p
Eq. (3.39) minimizing PF. If defining AmiI
the difference
=
A(iVmin),
Eq. (3.39)
reduces to
e2p
xp
FTd
nspd
rsrrt
Amu
(3.46)
*
kBP Vs
X Fd
exp(Amin/A-Br)
and the
exp(-Amin/A;BP)
peak
The final result for the
shape
height
of the
is therefore
Amin/A'B/
Amin
Gmax
sinli(Amm/A'Br)
cosh
\2XkBT J
(3.47)
In the classical
and half the
regime
the maximum conductance is temperature

value of
independent
Eq. (3.43) with negligible charging energy. The reason is the correlation of subsequent tunneling events by the charging energy. One electron can only tunnel onto the quantum dot if another one is tunneling off. Therefore, the probability for one tunnel event decreases to one half. The width of the peaks is linear in temperature.
The situation is
high-temperature
distinctly
<
different in the quantum-
regime. kBT
beled
-C
As
e2/Cv>
The
by Nmm
in
Eq. (3.45).
resonant-tunneling Tunneling occurs through a single level la conductance oscillations and the peak height
or
3.4.
45
G,
are
given by
G
=
G max
-4A-Br/,(Amin)
1
cosli
-2
pd
A,
(3.48)
if
A
<7,
4kBT F
-.
F
V n
The
lineshape in the quantum regime is almost the same as in sical regime, except for the different effective temperatures T*,
2.5 P in the classical and 2 71 in the quantum
the clas which is

maximum
regime.
in
The
peak
Gmax
same
G00(Af/4A,BP)
regime
while
decreases
linearly
with
the quantum
remaining
constant
increasing temperature in the classical regime. At the peaks
time, the peaks broaden and the total
area
under the conductance
remains constant.
It should be noted that above the natural linewidth
theory
is not suitable for the
case,
where
As
a
/if t exceeds
the thermal energy,

can
A?Ft
>
A;BP.
compromise
the
Breit-Wigner
G
formula FTC
be used
9l h
F^ + Fd
(A/2)2+ O72;
energetic
(3.49)
where g is the
degeneracy of
the resonant level and A is the
separa
tion of the resonant level from the Fermi level
[34],
Neglecting
the energy
the the
or
the influence of
on
the temperature
rates and
tunneling
same
assuming
energy
con
C5
rates
as
Ty,
for all levels
as
well
equidistant
const,
levels Ae
ductance
the
can
be evaluated
using
Flo:10J
Eq. (3.37)
strate the
tire
in order to demon
distinct regimes.
3.2 shows the temperature

the conductance Also shown
are
dependence of peak maxima.

the maxima
as
the minima which merge with
Figure
3.2:
Pemperature dependence of of the^

0.01
kBP approaches
In
the maxima and minima blockade
Coulomb
and
e2/CX
(classical regime).
peaks for As
e2/C\2
the resonant tunneling
regime
l\
YL\. (From
Beenakker
f26f.)
46
Chapter
3.
Theory
of
...
peak height increases with reduced temperatures vanishing thermal broadening of the resonance.
kBT
<C Ae the
clue to the
Figure 3.3(a) displays the temperature dependence of the peaks in the clas sical regime. The width increases with P. For low temperatures the peak heights are independent of temperature. The crossover the resonant-tunneling regime is shown in Fig. 3.3(b).
1
,
from the classical to
0 75
AEF/(X/CX
Figure
3.3:
AEF/(X/tX)
Coulomb blockade
Pemperature dependence of the

=
peaks for pa
=
rameters
As
In
O.Olf/Cs
(b)
/// the classical
regime (b).
[a]
to
(a) the temperature

In
increases
regime (a) and in the quantum 0.075 in from kBT / {e2 f Gjf) from kBT/As
=-
2 in
[ff.
the temperature increases
0.5 in
faf
to
15 in
[dj. (From Beenakker [26f.)
Chapter
Transport coefficients
4.1
Introduction
previous chapter a formalism for the calculation of the linear-response transport through a single-electron transistor was derived. The main results are
condensed in
a
In the
equation for the linear-response conductance, Eq. (3.35), for system composed of a quantum dot weakly connected to a drain and a source
an
reservoir via
tunneling
barriers. The
function of the Helmholtz free
linear-response conductance is given as a energies of the system, P({/??}, N), and sets of
tunneling rates tunneling rates

the the
for the two barriers,

are
{Tsk}
and
{Fd}.
The free
energies
and the
termed transport
relevant information about
coefficients since they contain all physical the many-particle ground state of the system and
which
are
tunneling barriers transport equation (Eq. (3.35)).

In
shape
of the
necessary for the evaluation of
Chapter
2 the free energy
was
obtained within the orthodox
theory
and
the constant interaction model at (
zero
temperature, i. C
e.
Ve)2
Â.
A-=l
However, while this simplified model
was
sufficient
to
provide
basic under
standing
tor
of the
key effects
associated with Coulomb blockade in semiconduc
quantum
structures, it will fail to
give precise
results for small structures
47
48
Chapter
4.
strongly influenced by the screened electrostatic potential. Especially, the parameterization of the free energy in a circuit-like approach using capacitances and voltage sources is not a sufficiently accurate model for structures with small quantum dots. More appropriate is the calcu lation of the free energies using a solid state model for the earner densities and the electrostatic potential. Furthermore, the resulting self-consistent potential can also be used to calculate the tunneling rates for realistic electrostatic bar
where the level spectrum is rier
shapes.
In the
the free
following chapter the self-consistent theory for the calculation of energies on one side and the transfer Hamiltonian formalism for the
on
calculation of the tunneling rates
the other side is discussed.
4.2
Equilibrium
carrier densities
4.2.1
Energy bandstructure
and
applied forces
by its bandcrystal potential in
of the crys bands in the
bands which
The electronic structure of

structure.
bulk semiconductor is best described
The bandstructure reflects the

are more or
properties
move.
of the
which the electrons

tal
less free to
The
periodicity
potential in real space allows the calculation of the energy reciprocal space of wavevectors (k-space). This assumes energy are spatially invariant.
An
externally applied force field modifies the bandstructure and leads to spatially varying energy bands. Consider an undisturbed bulk bandstructure P,?o(k), associated with a particular semiconducting material. Force fields ap plied to the bandstructure can act in two ways. Either, they can affect the crys tal potential of the unit cell, and therefore change the bandstructure Pno(k)
such that
E (k. r)
where
EnQ(k)
6En (r).
index,
or
(4.1)
they
can
6En (r) is the perturbation and n the band potential energy T^r) which is superimposed on
produce
the bandstructure such that
,(k.r)
En (k, r)
,2(k.r)+*(r).
an
(4.2)
external
is the total energy of the nth band under the influence of
4.2.
Equilibrium
carrier densities
49
potential.
The exact
Schrdinger equation
for the
crystal
in
force field is
ft4>(k)=$(k),
with
a
(4.3)
Hamilton
operator of the form

r?
U
=
-
2???
V2 +
U(r)
#(r),
(4.4)
where U is the
crystal potential operator. If the externally applied forces are slowly varying over the dimension of the crystal unit cell, Eq. (4.3) transforms equivalent Schrdinger equation
into the
Ui,A(r)=nA(r),
where
(4.5)
a
A(r)
are
Wannier functions, and

e.
PLn is the effective Haniiltonian for
band with the index n, i.
#n=
n2
2
_^V
\m*x{v)
+*(r).
X
(4.6)
of the
Equation (4.6)
has the form
Hn
correspondence principle (-7V
>
En( /V.r) k) yields

+
no
^(r). Application
*Hn(k,r)
The Haniiltonian in
P(k.r)
contains
/(r)
Sn(k,r).
(4.7)
explicit potential term for the peri odic crystal background potential. Electrons (or holes) are described as par ticles with a spatially varying effective mass ???* which is different from the mass m of the free electron. The effective mass depends on the material and Eq. (4.7)
the
of
particular band En atk ko=
which is assumed to be
parabolic
around the minimum
Modifications of the bandstructure due to nonuniform lattice
according
to
SEn(r)
band
with the
locally varying AEn(r). Usually,

The
materials
is
described
the band offset is
by the given for the

can
spacing band offset

conduction
(conduction-band offset).
The valence-band offset energy of the
be determined forces in
e.
bandgap g.
potential
externally applied
semiconductor devices is the
potential
energy of the electrostatic
field, i.
\Ef(r)
where
nian
-qo{r).
potential. Now,
(4.8)
the effective Haniilto
f> is the electrostaticin Eq. (4.7) is written

n (k. r)
or
Hartree
P0(k)
AP (r)
<#(r).
(4.9)
50
Chapter
4.
Transport
coefficients
For
some as
fixed
point in k-space, usually
minimum
(valley), Eq. (4.9)
can
be
written
(r)
where and
v
P (k, r)
to
qf(r).
c.
v;
//=!,...,
the
(4.10)
refers
the lowest conduction band, index.
to
highest
valence band
is the
valley
Figure free
it is
4.1 shows the energy band level
diagram for two diagram

an
materials with different is the
bandstructure.
vacuum
The reference for the band
infinitethe
or
forceexter
Eq.
It is defined
as
the energy
is defined
an
electron would have if

or
completely
free from the influence of the material

vacuum
applied
nal forces.
The local
level
P\
as
the
vacuum
level in the
presence of forces, i.
e.
in the presence of
electrostatic field
ci(r)
Eq -qo{r).
=
(4.11) E\. The

energy needed to
Therefore, in the absence of forces Eq

excite
an
E\
electron from the conduction band to the local
vacuum
level is the
affinity \. It accounts for the work against image forces and sur face dipoles. The work function <f> of the semiconductor is defined as the change in free energy of the material if an electron is added or taken away
electron from the system. The work function
more
changes
with the
doping.
Therefore, it is
practicable
to
characterize the material
using
the electron
affinity.
The
conduction-band offset
APC
used in the band energy,
Eq. (4.9),
is calculated
F(x)
S(x)
Figure 4.1: Energy band diagram for nonuniform material before (left) and after (right) contact. Phe different chemical potentials pL and p2 align to
a
constant Fermi energy
potential if no
equilibrium. external potentials are applied.

m
Ep
Phe
potential cp
is the built-in
4.2.
Equilibrium
carrier densities
51
as
Ac(r)
xi(r)-X2(r),
yo
are
(4.12)
ma
(electron-affinity rule) where \ [ and terials building a heterojunction at r.
the electron affinities of two
Before contact, each of the materials is characterized

tential ft.
to
by
chemical po
The effect of the electrostatic
potential
on
the energy levels due
q<f(r) is described by a locally varying chemical potential, electrochemical potential ( which is defined as follows
=
T^r)
the
(r)=fi(r)-qo(r).
Before contact, the materials
are
(4.13)
by different chemical poten tials, /i, electron affinities, \, and energy gaps g 8Y (Fig. 4.1). After 8C contact, the difference in the chemical potentials leads to electron transfer from right-hand- to the left-hand side minimizing the free energy and estab lishing thermodynamic equilibrium characterized by an uniform electrochem ical potential at a Fermi energy Pp C const. In /i-doped materials the
characterized
=
electron transfer ionizes donors

trons
on
on
the
right-hand
side and creates
excess
elec
the left-hand side. An internal electrostatic
junction which is flat in regions far from the trality holds. The potential difference between the two materials can be under stood as a potential drop (voltage) which is built-in into the junction because of its internal composition (material composition, doping). It is especially not an externally applied voltage.
the
potential is induced across junction where charge neu
4.2.2
Quasi-equilibrium approximation
potential o(r) in a semiconductor structure depends on the charge p(r). In general, p also depends on the current flow. The self-consistent evaluation of potential, carrier densities and current densities is the fundamen tal problem of device theory. However, in many cases an adequate estimate
for the densities and the
This is called
can
The electrostatic
potential
can
be calculated
by neglecting
the current.
be
a
with
quasi-equilibrium approximation. It is assumed that the device subdivided into regions, each of which is locally in thermal equilibrium Fermi level set by the voltage of a contact to which the region is con
assumption
holds
nected. This
exactly
if
no
external
voltages
are
applied
to
the contacts and the structure is
naturally
in thermal
equilibrium.
If different
52
Chapter
4.
Transport
coefficients
applied to the contacts, the regions with different Fermi levels have to be separated by insulating regions where the carrier densities are zero and no current flow is possible.
are
voltages
In the
quasi-equilibrium approximation
one
the system of
equations
for the
potential,
reduces to
the carrier densities, and the current densities
given
in Section 1.2
equation
for the electrostatic
potential and
the carrier
densities,
the non-linear Poisson
equation
-X>.(tV<f>)
which is
a
p[],
(4.14)
subject
to
global equation, i. e. appropriate boundary
which is solved for the whole device domain conditions. The total
charge density
is
/#]
where p is the
q(p[d]
n[0]
X+[<f>]
A, M),
(4.15)
density of holes, ?? the density of the electrons, and AdH, Nff are the densities of the ionized impurities. The equilibrium carrier densities for electrons and holes in bulk regions are given by the following state equations for carriers in parabolic bands
n(r)
p(r)
where
Aî(,?c(r)),
NvTi (;,v(r)).
,,c(r)
ih(r)
(J^h^j (^Z^^)
previously
-
(4.16a) (4-16b)
>
Ec
is the conduction band
edge,
+
which
was
defined
as
c(v)
A
Ec0(K)
AEc(t)
#(r).
(4.17)
by the spa tially invariant term FJcq (k ). The val ence-band edge is given by Ev EgEc and Fi ri is a statistical distribution function which in this particular case is the generic Fermi-Dirac integral of order 1/2. The general form of the FermiDirac integral of order o is
v
=
particular
minimum
in the conduction band is characterized
r^
J
xdt
a
^
1 X
exp( /;)
rt
-l
(4.18)
4.2.
Equilibrium
carrier densities
53
The factor
AP is the effective conduction-band density of stales
Nr
Xirm:tkBry MC7
(4.19)
with
Mc the number of energetically degenerate conduction band valleys and AP is the effective valence-band density of states
'27071* A*BT
/Vc-2
k?
Mv,
(4.20)
It
where
Mv is the number of energetically degenerate valence band valleys.

an
is not useful to introduce
intrinsic
density
a
/^
a
or an
intrinsic Fermi level
Pi
since these the

mass
quantities
are
not constant
through
heterostructure. Furthermore, semiconductor.
action law pn
n2
is not valid in
degenerate
The total ionized
charge density, Eq. (4.15), also contains contributions from the impurity densities of donors Arcj~ and of acceptors Nr. The state
for these
equations
quantities
are
given
as
Nd
(4.21a)
AP
where g a and ga
are
(4.21b)
the
energies
of the
as
energy scale
degeneracy factors for donors and acceptors. The donor levels Ec\ and Ea are defined with respect to the same the conduction band edge Ec and the valence band edge Ev.
for the structure is obtained by
The
screening equation
combining
all equa
which is
as
tions in this section into the non-linear Poisson solved for the electrostatic
are
equation, Eq. (4.14),
potential f>(r).
Material parameters such
APC
determined
by the design of the
structure and the Fermi energy is fixed
by the external circuitry. The solution d(r) has local screening properties: increasing the potential decreases the charge density and vice versa.
54
Chapter
4.
4.2.3
Boundary
conditions and Fermi level
equation is solved within a bounded domain subject to the ap propriate boundary conditions. In principle, two different kinds of boundaries are distinguished. First, real physical boundaries to contacts and interfaces are represented by Diiichlet boundary conditions. Another boundary type is due
to the limited size of the simulation domain. This second
The Poisson
category of artificial
sense
boundaries
(<90a) represents
zero
e.
no
boundary
in the
physical
and is
only
introduced to enable the simulation. i.

e.
Usually,
Neumann
boundary conditions,
normal to the
the condition of
are
gradient
of the electrostatic
potential
boundary
used, i.
nVo(r)
where
n
0,
re
<9Qd,
it is
(4.22) guaranteed
is the unit normal vector of the
boundary. Therefore,
that
no currents are
flowing
in and out the device except contact currents.
physical boundary conditions are either for Of unie contacts (<90o) or for Schottky contacts (ds). For the ith voltage controlled Ohmic contact the boundary condition is
The
<f>^ (r)
where
^ (r)
(j?
0.
re
d0.
(4.23)
(/vjy
is the
externally applied bias,

a
"
and
oj^
is the built-in potential'which
is defined through
local
charge neutrality
condition
using Eq. (4.16)
and
Eq. (4.21).
i.e. 0
p{r)
n(r)
-t
AP/(r)
Xr(T),
act as
c900.
(4.24)
The
terminals, together with the circuit nodes

Fermi levels which
are
are
given
which
in
determined
by
the
charge reservoirs with applied voltages. Terminals

are
connected to the substrate via Ohmic contacts
assumed to be
equilibrium
with the substrate, i.e. the Fermi level in the substrate
region
such
connected to the Ohmic contact is
pinned through
the
applied voltage
that
EF
In
-qog.
can
(4.25)
have differ has been
Other
general,
different
regions
connected to different terminals

a
ent
Fermi levels. However, since
quasi-equilibrium approximation quasi-equilibrium assumption
made, these regions then ha\e wise,

a current
to be
separated by insulating regions.
would flow and the
would be
violated.
4.2.
Equilibrium
carrier densities
55
For
Schottky
barriers the
is used
following
Dirichlet
boundary
condition for the
electrostatic
potential
<!>il\r)
where
^>s
<^(r) + c/4
0$
=0.
re
dQs,
as
(4.26)
the difference of the material
X)
:
is
the
Schottky
barrier of the Mi contact
given
between the work function of the metal and the electron
affinity
the contact is connected to
(Schottky model),
o
(o
G
apo
-*" 1 1
(4.27)
The
height
of the
Schottky
barrier is
characteristic
quantity for
the semicon in
ductor/metal
contact.
Biased
Schottky
contacts are not
generally
rium with the substrate. The Fermi level in the terminal
node, which
equilib is given
applied voltage d)^ is in general different from the substrate Fermi level, which is given by some ground electrode (Ohmic contact) far away from the Schottky contact. Additionally to the Poisson equation, the current equa tions have to be solved to get the bending of the quasi-Fermi potential at the Schottky contact and the contact current right.
by
the
,
4.2.4
Low dimensional electron gases
Kohn-Sham system
The electron
charge
is
in bulk
regions
mensional semi-classical effective
given by Eq. (4.16a) with the three di conduction band density of states Nc. This
parabolic
energy band extrema,
is
approximation
i.
e.
only
a
valid for electrons in
in bands with
localized levels
dimensions has
quadratic dispersion relationship. The charge density for in the bandgap due to carrier confinement in one or more space to be calculated using a different representation of the density
are
of states. Electrons in systems with confinement
described
by
an
equation
of the form
lr
hrv-
\
J
\m*
Veff(r)
c
-
v,(r)
0.
in
O,
(4.28)
where the effective
potential Veff (r) for
electrons is
given
as
Vei(r)-APc(r)^PPH rxc(r).
(4.29)
56
Chapter
4.
Figure
4.2
schematically
shows
possible
subdivision of the simulation do
region and the confine ment region O proper boundary conditions for Eq. (4.28), which is only solved in region Q, have to be supplied. Homogeneous Diiichlet boundary conditions for the eigenfunctions </, are
main. For the interface between the semi-classical
suitable
for
confined the
at
carriers.
Consequently,
sities
of 0
carrier den
vanish
and
a
boundary depletion zone is

confined
the
re
the
Device
Semiclassical carrier densities
surrounding the gion. Naturally,
eigenfunc
QD-ODEG
tions should not feel the bound

ary conditions too much.
Envelope
functions
wave
Although Eq. (4.28) looks like a Schrdinger equation, it is what is called a single-band effective-mass equation (BenDaniel-Duke form
Figure 4.2: Domain partition for the Schrdinger equation. Phe depletion zone around the confined region (ODEG) is indi
cated bv
a
135]).
The
das/ied line.
not
crystal potential is periodic on the atomic scale and does itly. It is incorporated through the position dependent
the conduction band m* and the
set
appear
explic
for
effective
mass
position dependent
an
=
conduction band off
APC. Equation (4.28) is
an
effective
single-particle level characterized by responding envelope wavefuetion (/',

The electron
single-particle equation for the ?'th E7 (V0a) and the cor eigenvalue E,
=
'((Veil).
a
mean-field approximation, i.e. each electron moves in the effective potential produced by all other electrons. The form of the effective potential in Eq. (4.29), however, goes beyond the popular Hartree approximation by incorporating not only the classical interac tion due to the electrostatic potential energy, qcf but an additional term Vxc
which accounts for the quantum mechanical interactions due to
screening is treated within
exchange-
and
Schrdinger type Eq. (4.28) with an appropriate formulation for the exchange- and correlation potential is known within the framework of density-functional theory (DFT) as the Kohn-Sham equation [51]. A short account of DFT is given in Appendix C.
Density-functional theory states and Yxc[n](r) are unique functionals
that the two
correlation. The
potential
terms
qf[n](r)
The
of the electron
charge density (49).

wavefunctions
charge density
is
compiled from
the
single-particle
according
4.2.
Equilibrium
carrier densities
57
to
the quantum mechanical
expression
oo
?7(Veff)(r)
The numbers
Y ^(>eff)k/',(Vcft)(r)|2,
=0
G 0.
(4.30)
N,
are
the
occupation
factors
according
to
iV!(Veff)
where p, denotes the chemical
ferent forms the
/(fJ(Veff)-//),
and
(4.31)
potential,
is the
thermodynamic equilib
d,
rium distribution function for electrons.
The distribution function takes dif
depending
on
the reduced dimension of the electron gas. With
dimensionality
of the electron gas. this is

1
if
if
d d
3
2
/(s)
rjT0
(_-JL_),
kBT
tv=<J-
0
if
d= 1
where
s
is
some
as
positive constant,
T the temperature of the electron gas and
eigenvalue E7 corresponding to the wavefunction ijy used in Eq. (4.30) and the chemical potential, i. e. s Et (VCfr)A*The chemical potential for a given electron number, p(N), is defined by the
is defined
the difference of the
charge conservation law, i.
e.
dr
n(Veti)(r)
V/(,(Veff)
7=0
W).
(4.32)
Electron densities
Confinement is
in two
possible in one space dimension (quantum wells: 2DEG), dimensions (quantum wires: 1DEG) or in all three space dimensions
(quantum dots: ODEG). For zero-dimensional electron gases in quantum dots,

i.e. O C
jR3,
the distribution function is
given by
the Fermi-Dirac distribution
f(s)
2t,F
(pX)
-,
(4.33)
with
E,
s,,
the
single-particle
energy of the iih quantum dot level. For

e.
one-
dimensional electron gases in quantum wires, i.
O C
]i2,
the distribution
58
Chapter
4.
Transport
coefficients
function is
J\
2tt1P
J
a
~T
"3
\kBPj
i
'
(2ml A'bT)
/'^
Klf
-,
'/
Y,
(4.34)
H-expU+^rj
of the
/th
one-
with
E,
6;,
the
dimensional subband.
position dependent subband energy The Schrdinger equation. Eq. (4.28),

=
is solved in slices
alongside
mass
the ID-channel and //C
m\
is the conduction band effective
component
perpendicular
e.
to
. For
two-dimensional electron gas in
quantum wells, i.
O C
A1.
the distribution function is
given
as
/W
2^!ltr^ (^) ^^ln (1

^
+ CXP
())
is solved
'
(4.35)
with
E,
j,
the
position dependent
The
=
subband energy of the
?th twoone-
dimensional subband.
Schrdinger equation, Eq. (4.28),

n
is the component of fhe conduction band rn\ effective mass perpendicular to 0. The factor 2 arises from the spin degen eracy of the eigenstates. g is the valley degeneracy of the particular material.
dimensionally
and ?7?*
Two-dimensional systems with one-dimensional confinement

for instance
are
studied for 2DEGs in semiconductor inversion and accumulation
intensively layers (see
L30J).
dots
are
Quantum
In
some
a
regions
with confinement in all three space dimensions.

are
structures, the electrons

few hundred nanometers
confined in
tiny
areas
of
are
typical
~
sizes
below
(typical
for GaAs systems
100 nm).
be
The energy level spectrum is discrete and the

than there thermal
spacing
of the levels
can
larger
used
broadening.
as
The Fermi-Dirac distribution, which
was
in the
previous
not a
section
distribution function for electrons amongst the

any
more.
levels, is
dot
are
good approximation
The electrons in the quantum

a
isolated from the resenoirs, and therefore, the definition of
Fermi
energy for the dot electrons is somewhat
meaningless. The electron gas in a quantum dot is described by canonic ensemble theory. The electron number is fixed at an integer value N and the occupation factors N, for the electrons in an effective potential VAr are determined by the Gibbs distribution for localized
4.2.
Equilibrium
carrier densities
59
states
g(s, \N)
A,
as
already
introduced in
Chapter
(Eq. (3.12)),
i.
e.
X)
-
Z(X)
exp
XX
kBT
"/A
n>AN
v in
(4.36)
where
Z(N)
is the
A"-dependent partition
function in the canonic ensemble
theory
Z(N)
y
XX
exp
kBl
AT,
t?a
(4.37)
and
X(N)
kBTln{Z(N))
is defined
as
the free energy of the internal
degrees of freedom. Again, the factor 2 arises from the spin degenerate nature of the eigenstates e,, g is the valley degeneracy of the particular material and T the temperature of the electron gas. In practical calculations, the FermiDirac distribution, Eq. (4.33), can be used instead since for larger electron
numbers in
highly populated quantum dots the deviations of the Fermi-Dirac distribution from the more general Gibbs distribution is small [31], However, the difference can be significant for small numbers of X (see Appendix A).
In
one some
cases, the electron gas in
quantum dot is strongly confined in

nature
dimension, and therefore, its quasi-2D
allows
shortcut to the very
expensive solution of the 3D-Schrdinger equation. The wavefunction ip(r) is separable and an one-dimensional equation is solved in the direction of the strong confinement (which is taken to be the ^-direction for instance) at every point
in the
.ry-plane,
1
i.
e.
h2 d
2
0
4
Oz \ mt dz
Vet(x.y,z) C'y(^)^ A^y)i;fy(x
(4.38)
Assuming only a single subband en(x, y), the index n can be dropped and the charge distribution in the .n/-plane is calculated with the effective potential
c(x.y)
eo(.r,(/),
i.e.
r.y)
(pm
(x, y)
=-
sm 0m
{x, y).
(4.39)
The mth
single-particle wavefunction within the single-subband approxima tion is 0m (r) > 0mo(.r. y, z) om (x. y)Co'V {z) an(i m is the energy of the corresponding single-particle level.
=
60
Chapter
4.
Transport
coefficients
4.3
Total free
energies
4.3.1
Equilibrium thermodynamics
The first and second law of
thermodynamics
a
are
condensed in
an
equation
for
the total internal energy E of B
system (Eider's equation) such that
-i>V +
5>AV
on
(4.40)
where the intensive variables (not
depending
the
particle concentration)
temperature T, pressure P,
tensive variables
and chemical
on
(depending
the
potential p, are related to the ex particle concentration) entropy S, volume

p} and of
V, and particle number X,. The index of the chemical potential

the
particle number Xt refers to a particular species of particles With Eq. (4.40) the Helmholtz free energy can be defined as
F= E-TS.
and the Gibbs free energy
in the system.
(4.41)
as
G
both of which have minimal
constant.
E- TS X PC,
at thermal
(4.42)
properties
equilibrium if T
and
A)
are
P is at minimum for constant volume and 67 for constant pressure.
change of the Helmholtz free energy is completely transformed into work. Using Eq. (4.40) a system consisting of a single species of particles with particle number N and the total charge Q qX has the free energy
an
In
isothermal transformation the
F
The Fermi energy when
an
-?7 +
the
(/i-(/o)A
change
in free energy of the
(4.43)
EF is defined
or
as
crystal
electron is added
taken away
qo.
(4.44)
where 0 is the internal electrostatic
potential
in
energy
(built-in potential) of the

no excess
system. The built-in potential is
zero
equilibrium with
carriers
present. Only in this
case
the Fermi energy
equals
the chemical
potential.
The
4.3.
Total free
energies
61
equilibrium. In a non-equilibrium situ ation (externally applied forces) the (non-equilibrium) Fermi energy is often referred to as quasi-Fermi potential C, qf>, where qo is the sum of the p
=
Fermi energy is constant in thermal
internal- and the external
potential. The property Q
EFN is the grand
canonical
potential.
4.3.2
Helmholtz free energy and external forces
To calculate the total
(Helmholtz)
free energy of
of i
i... XI distinct elements with the total
charged system consisting equilibrium charges Q, and the

a
voltages V,
the
following
semi-classical
expression
is used
145]
F=YlQ>Vi-y, / affinait/),
Vi
,--~iJo
(4.45)
where the
provided by the external circuitry (voltage sources). The first term in Eq. (4.45) is the capacitive (potential-) energy U stored in the device. Voltages and charges are related by the elements of the symmetric capacitance matrix C7;, i. e. the charge at the ith clement is related
are
the currents
to the
voltages V3
at all
elements such that

M
The
diagonal
elements The
of the
capacitance
matrix
are
the
capacitances.
or
negative off-diagonal
elements
arc
mutual
positive selfcapacitances
coefficients of ith charge due to

constant.
induction which account for the amount of

the
change
in the
jth voltage
matrix with
if all other
change in the voltages are kept
{Gn}
is the
capacitance
M(M
l)/2 independent
elements.
The device is
separated
into elements which contain
continuous space
charge p (dot, leads, donor layer) and metal plates which are equipotential regions (gates, contacts). The electrostatic energy U is written as
U
-
\ j dvp(r)<f(r) Xlr}Y
V
Q>V>-
{4A1)
il
tutes
The
potential c/>
is the electrostatic
potential,
i.
e.
the solution of the Poisson
equation with
the
charge density
p.
subject to
Diiichlet
boundary
conditions at
62
Chapter
4.
Transport
coefficients
gates and
i.
e.
contacts. The electrostatic
of the Greens function of the Poisson
potential can also be expressed in equation, (7(r, r') l/(c(r') |r

=
terms
-
r'|),
0(r)
The
f dr'G(r.rf)p{r').
(4.48)
integral over the space charges replaces the summation over the M charged elements in Eq. (4.45). The energy contribution from the gates which are kept at constant voltages \ ] is calculated with the surface charges Q, which
are
obtained
using
Gauss law. i.
e.
Ql=
dfo-{r)
with
rr(r)
-<(r)Êl,
OYlf
(4.49)
d,
where
<r(r)
is the surface
charge density
and rif is the unit surface normal
vector of the ?'th
gate
at
<9Q;. Eq. (4.45)

is the work IF which is needed to
The second term in
charge
all
elements which
source
are
connected to external
circuitry (such
are
as
the gates and the
and drain
region)
the
when Ar electrons
transferred to the quantum dot
region.
It is assumed that the

e.
ation times, i.
ideal with
charging is fast compared with the internal relax tunneling time rL. Voltage sources are considered to be
(low-impedance environment). Therefore, it is reasonable to assume that the voltages are kept at constant level before and after the charging Vj(t) \]{0). Using I dQ/dt the current integral in Eq. (4.45), i. e. the work W, for the charging of the ith element is written
zero
internal resistances
Wt
The
Vt
I
7o
df
I,(t')
VfCKt)
be
Q,(0)]
V,QL.
(4.50)
charge
as
at
the ith gate at /
can
arbitrarily
set to
Q7(0)
0. Com
bining
is
these results, the final form of the semi-classical Helmholtz free energy
given
-=
Y
f k HIS
/
O,
dr/)(r)<;)(r'
MUS
;L
'This again
time
is not
Y
.ifnilii.
..Blrf
/drp(r)o(r)/
l- V QX.
/
f,
(4.51)
|,||f
cads
; t
gates
between the
tunneling dilution rt, tunneling e\ents and rt C r
the
moteo\ei it is
the system lelaxation time,
the
(see
Chaptei 2)
4.3. Total free
energies
63
The work needed to
recharge
the ith element
during
single tunneling
event
can
in transition of the dot from the be calculated in

a
AAparticle
e.
to the
(iV + 1)-particle
system
similar manner, i.
AW,=
where
I
7o
dt'I,(t')V,{t')
Q,(N).
The
AQ,Vn
are
(4.52)
surface
AQ7
Q-,(N
-(-
1)
charges Cf,
a
charges
calculated
using Eq. (4.49)
if the ?th element is
metal gate.
4.3.3
Helmholtz free energy and
charge quantization
In the
case
of bound states in
zero-dimensional electron gas the free energy

on
has to be modified
including
the influence of the confined states
the to
tal energy of the quantum dot
[42], [43], [45].

e.
The Helmholtz free energy.
Eq. (4.45), is extended by
two terms, i.
F({r,k}iN)
Y^4
'
'IT
Exc+~Y/Q>V>-Yl / <lf'/,(f')V;(0/
,i(M...ii.i..Llrf
.,**m**KHl
(4.53)
The first
sum over
the bare
(non-interacting)
dot levels
with
particular oc
cupation configuration {/?/,} is the interaction-free kinetic energy of the quan tum dot Ts. The exchange-correlation energy F\c of the electrons in the localdensity approximation (EDA) corresponds to a (local) exchange-correlation
potential ipc according

EDA). The last
free energy is
to
two terms
Eq. (C.ll) (see Appendix C for a discussion of the are the same as in Eq. (4.45) in the previous section.
The total energy contribution of the quantum dot electrons to the Helmholtz
Etat {{'U}-Q dot X^lot)

From the Kohn-Sham
=YUk^
k
"h
o(AdotFdot
Pxo-
(4.54)
equation, Eq. (4.28),
with the Kohn-Sham effective po

are
tential,
Eq. (4.29),
the self-consistent energies
written
as
<V'A.|
^-V
=
Vj
-
-f
AP(r)
qo(r)
ipc(r)|^)-
(4.55)
With the definition of

X
sf
this results in
(!x(iy\
q<f>(r)\vk)
{Vk\Wc{r)\ik}.
(4.56)
64
Chapter
4.
Subsequent
summation
over
k after
weighting
with the
occupation
numbers
ni leads to the
following expression
for the interaction-free kinetic energy
I
xx
k
xY >u
k
/ dr|^|2(/Kr)^ /
h
a
dr
|(/^|2I/xc(r)
occupation configuration dependence of the last two terms is ignored and the discrete occupation numbers /?/, are replaced by the non-integer occupation numbers according to the Gibbs- or Fermi-Dirac distribution, one obtains
If the
Ts(W})
It is not
Ytnk
k
+7
/dr(r)u(r)
9.
a
[ dr n(r)VKC(r).
(4.57)
approximation. However, small thermal variations in the level occupancies have a negligible effect on the self-consistent results (t^, o, p, and Qt) which are implicit functions of the
electron number and the
straightforward to
find
vindication for this
applied voltages [41], [43]. This is especially true for low temperatures, where the level spacing is hardly affected by the tempera ture. The above result is now inserted into Eq. (4.54) where the electrostatic replaced according
to
energy of the quantum dot is
drn(r)c/>(r).
n
(4.58)
The final
expression
for
Ptot
becomes
Ptot(i>u}AP A)
YnXi
-f
zQ
I drn(r)(r)
7^c-
drn(r)rxc(r).
(4.59)
The is
integration
is
over
the quantum dot
area
positive.
The second term in
Eq. (4.59)
is
only and the electron density n the negative electrostatic energy

a mean
of the electrons. This is because the electrostatic energy is counted twice if

the
sum over
the self-consistent
particle
levels from
field
theory
as
the
Kohn-Sham
theory
is included in the total energy calculation.
The last
sum
provides
similar correction for the
exchange-
and correlation energy.

the
The total energy of the quantum dot,

terms in the Helmholtz free energy
EUA. replaces
corresponding
given by Eq. (4.53).
litis leads to the
4.3.
Total free
energies
65
final form of the Helmholtz free energy,

two-terminal
F({nk}, N),
which is used to
com
pute the Gibbs distribution, Eq. (3.9), for the quantum dot occupation and the
linear-response conductance, Eq. (3.37).
In the low temperature

e.
limit,
/?/,.
L for all k < X and nk
0 otherwise, i.
the electrons occupy be denoted
only
the lowest states in the dot and the free energy
can
Fq(X)
with AT
being
an
integer
number and
]P/,
'A
^V-
4.3.4
Free energy minimization
In the
following,
the method used to minimize the free energy to determine the

// of the quantum dot is
equilibrium
ground
state
number of electrons in the dot and the conductance is discussed.
The chemical
potential
and the free energy is
ramped and the self-consistent calculated using the equations from the pre
vious section for
maybe five
values of p above the reservoir Fermi level. The
number of electrons in the quantum dot is
peratures)
and the free energy values
generally non-integer (at finite tem give a parabola if plotted as a function

numbers of electrons,
of ft. Now, the free

determined
energies
at
integer
P({'A },X),
are
Gibbs
in the
by spline interpolation. These free energy values are used in the distribution Eq. (3.11) to calculate the equilibrium number of electrons dot and the conductance according to Eq. (3.35) (note that the Fermi
energy used in these
equations
1
is the reservoir Fermi
energy).
CO
\
\ \
63 0
CD
\
e
(a)
(b)
40
c
Z3
1
1
\
X2
3<D
c
!
E
Z5 C
\
.
30
^
p
CD C CD
f :\
!
CO
G
3
20
1
\
\
_CD
a
c
1
\
LU
L*
10
CD CD
LL
62 0
\
1
II
.1
-10
-0 78
-0 76
60
62
64
60
62
64
Gate
voltage (V)
Electre n tlumber
Figure
4.3:
Left: Coulomb blockade peak and discrete charging (dashed line)
of a GaAs quantum dot at 1 K. Right: free energy curves at the conductance peak (a) and between two peaks (b). Phe zero of the free energy is set arbi
trarily.
66
Chapter
4.
Transport
coefficients
Figure 4.3 shows
an
example of a conductance peak and the voltage.

Also
discrete
charg
ing
of
GaAs quantum dot. The electron number in the dot is increased from
62 to 63
by ramping
the gate
a point half-way to the parabola for the top of the conductance next peak (b). It can be seen that the free energies for electron number 62 and 63 are degenerate at the top of the peak and consequently the equilibrium number of electrons is 62.5. Between the peaks, at point (b), the electron number is an integer value and the minimum of the parabola is to be found at
displayed peak (a) and
are
the two free energy
63.
4.4
Tunneling
rates
4.4.1
Bardeen's transfer Haniiltonian method
The
transfer Haniiltonian including tunneling

able to
formalism
was
introduced
by
of
Bardeen
a
[36]
in 1961.
It has proven to be very successful in the

ena
description
between normal metals and
phenom superconductors and was
number of
Josephson effect [381. The transfer Hamiltonian for malism is based on the separation of the system into subsystems. Consider a two reservoir/barrier system described by the following model Haniiltonian
even
predict
the
={PA iXy
(
/in
ii
it "R-
(4.60)
Each
'Hl and 'Hn

tains
a
are
the Hamiltonians for the

to
an
subsystems.
a
subsystem
con
reservoir connected
electrode and
semi-infinite barrier
Fig.
(4.4). Ol and Ol denote the half-spaces for which the Hamiltonians 'Hi and
fi\\
adequate descriptions. The eigenstates and eigen functions of subsystems are given as the solutions to two Schrdinger equations
are
the two
PPa''o =Eovq,
PlAA =P/P
A
(4.61)
In other words, it is assumed that the electron tunnels from the left hand side to
set of
a
given
state
a
Eq
on
state
F\.
on
the
right
hand side which is part of
eigenstates
of
'HB.
Now, the solution of the
given time-dependent Schrdinger
equation
with the model Hamiltonian
/AyP(P)
-Hv{t)
(4.62)
4.4.
Tunneling
rates
67
(a)
Pn
/~-\
(b)
X^
nh
\J
Pfc 77r
(c)
^\
vy
aR
0,
of the tunnel junction. Phe electron tun nels from a given reservoir state Eq to a discrete quantum dot state P\ (b), (c): Left and right subsystem characterized by the Hamiltonians Hj, and'Hji, respectively. Figure
4.4:
(a): Schematic
view
is written
as a
linear combination of
iJjq
and the various i/'f, which
are
indepen
dent of time
iy(t)â(t)r0c
At t
-=
fXt
+Y^bXt)iye
iEk1
(4.63)
expansion coefficients have 0. Furthermore, the expansion is restricted to first order to vanish, 6(0) in a which leads to a 1 and 0. Inserting the eigenfunction ip(t) into the time-dependent Schrdinger equation yields the following equation for the expansion coefficient
0
one assumes r
=
vq and therefore, the
ink =c'^{EÊo)tMk,
(4.64)
68
Chapter
4.
Transport
coefficients
with the matrix element for the transition
Mo-*-
Mk (4.65)
Mk
fdriliCHh
Po)Po.
The transition rate
IV-u-
^k follows from
i.
e.
F,
The
lim
f^>^?
\bk(t)\2t
-K
F,
'?I\Mk\26(Ek
n
E0).
(4.66)
remaining task is to evaluate the matrix element, Eq. (4.65), which can be done if P0 and the fk are known. Bardeen suggested a more useful form of the matrix element which is derived now. Using the model Hamiltonian, Eq.
(4.60),
in
Eq. (4.65)
Mk*
j dri!>lCHK-Eo)i!>o,
the
(4.67)
where the
integration
is
over
vanishes
on
the left-hand side.
right half-space only since {n\ \% Since (tA \'H Ek [fo) vanishes on
-
Po|0o)
right-
the
hand side, the matrix element
Mk
can
be
approximated by
-
A4
dr
[Xk('H
E0)wo
)(P
Ek)i,Xk)
/ drfc/^Pvo
iXHif
clEoipo
ifoFkff}
All transitions
are
considered to be elastic
tunneling
a
events, i.
e.
if)oFJkifl
i)fPJoiX
and energy is
always
n=r-
conserved. If
model Hamiltonian of the form
^V2
=
Veff(r)
(4.68)
(i.
a
e.
Kohn-Sham
r
Hamiltonian2) PoPef?Xk
is assumed with the effective

i
potential Veff
as
function of
then
fêii )
dr
and the matrix element becomes
Mk
--^r /
2m
[v*kV2vo
ioV2].
that the bandstructure
(4-69)
2Note
constant,
that the form of the Hamiltonian

i.
Eq. (4 68) implies
is
e.
the effective
mass
m"
is
not
spatially dependent.
is
Foi
an extension
ot the
spatially tlieory
towards tunnel junctions with wining bandstructure the eadei
referred to
[371.
4.4.
Tunneling
rates
69
Using
Green's theorem the
integral Eq. (4.69) following
can
be evaluated, and the final
form of the matrix element is the
surface
integral
Mk
kv
2m*
(1/
*"k
On
ipo
df;
~z
(4.70)
On
over
the surface of the
right-hand
side
half-space Or
X
with the vector
being
the unit normal vector of the
surface,
dQB.
4.4.2
Adiahatic
theory for quantum dot systems

[36] is applied in
straightforward man ner to a reservoir/quantum dot system. The quantum dot is separated from the reservoir by a narrow quasi-ID channel. By applying a voltage to an elec trostatic gate above the channel a narrow constriction, Fig. (4.5), is created. Since the constriction is formed electrostatically, its boundaries are smooth and electron scattering can be neglected. An ideal quantum point contact (QPC) between the reservoir region and the quantum dot is formed.
The transfer Hamiltonian formalism
a
The wavefunction inside the constriction is one-dimensional width of the constriction

can
140].
The
not be
considered constant, however. The sim the adiabatic
plest theory
sumes
for this
cross
case
is based
on
approximation
which
as
that the
section of the channel
ligible scattering
value of
x
between the subbands
changes so slowly that only neg occurs. Transport is possible only

occurs
in the .r-direction and confinement of carriers
along
the channel
two-dimensional
?/z-plane. Schrdinger equation

</,
in the
At each
with the
self-consistent effective Kohn-Sham
potential Vcs(x1 energies
z)
for the transverse

is
eigenfunctions f>n (,r;

'
y>
z)
and the subband
(x)
solved, i.
e.
Ir
d2
2/?p
where m*
dy2
ox
Vcij(x. y.z)
en(x) on{.v:y,z)-0,
(4.71)
is the electron effective

mass
mass
tion. The effective i.

e.
is assumed to
perpendicular to the transport direc be constant throughout the /p-plane,

wave
the channel cross-section. The full
function
as
f> satisfying the three-
dimensional
Schrdinger equation
factorizes
if
where the
>>
yk
(.r,
y.
z)
t>
(x:
ty,
z)$h (x),
i.
(4.72)
e.
a-(-*')
are
the solutions of the
coupled-mode equation,
the
70
Chapter
4.
Quantum dot
,r
Classical
E
i
turning point
,T
XL
Xb
Figure
the
4.5:
(a) Schematic
view
of the constriction, (b) Phe barrier potential,

a
energies of
the lowest subband and the quantum dot spectrum, in the reservoir.
is the
classical
turning point
one-dimensional
effective
Schrdinger equation
with the nth subband energy <r
(x)
as
potential
h2
l.ni 2m*
02
4(iv Ox
tn(x)
&(<")
-=-
Evkikix)-
(4.73)
The total energy is
Enk
tn(x)
from
a
h2k2
X
-
(4.74)
Enk
Zllh
Now, the
case
of
an
electron
mo\ing
reservoir state labeled
to
OD-quantum dot process requires

stant
state
sp is considered.
Energ>
conservation in the
to assume
that
Enk
slt. It is convenient
that
,r
tunneling V0f is con

< .rx,
outside
larger
range to the left hand side of the barrier
i.
e.
4.4.
Tunneling
rates
71
const for x < xj, < a, Fig. (4.5). Since the constriction is V0rf,L formed electrostatically, its boundaries are smooth and scattered components (cxp(ikx)) of the wavefunction can be neglected. The wave function com
(x)
ponents in the .f-direction in the
outer
regions
is
are
taken to be
root
plane
waves
k (x)
exp(ikx)
where k
/,'l and k(x)
the
positive
~k2{x)-Ênh
2nP
(.).
particular point
in
(4.75)
The classical
turning point a
e-
in the reservoir is that
where
Hnk
cp> i-
where the
energies
of the electron before and after the tunnel
classically forbidden region starts. Classi cally, the wave is reflected at this point. Quantum mechanically, it can tunnel through the barrier. The WKB approximation is used for the x-component of the wavefunction in xF < x < a, Fig. (4.5), i. e.
process
are
ing
matched and the
k(x)
as
exp[u;(.r)]
with
w(x)
dxf
k(x')
<^
a.
(4.76)
long as k(x) varies slowly enough (in particular, if \dk /dx | WKB approximation is also used in the forbidden region x >
function component in
x
'2).
The
The
wave
is
h(.r)êxp[u>(.r)]
and
with
w{x)
Ja
dx'
k(x')
(4.77)
k(x)
is the
positive
root
^yK2(x)ên(x)
2m*
~Enk
(4.78)
again
the
the
assumption of |dK /dx | X k2. The WKB wave functions in different regions would have to be joined correctly as x increases through classical turning point a.
with the
the matrix element for the transition from the nth trans
to
Using Eq. (4.70),

verse
state in
the constriction
the
pth localized quantum

narrow
dot state that
can
be
a
approximated.
The constriction is assumed to be
enough
only
single
states
transverse state is below the
Fermi level. As the electron

\\
moves
away
from the constriction, the channel becomes

the matrix element is calculated
ider and the number of transverse

a
grows. However, within the constriction
single
mode is present and
only
for the transition from the
ground
state
72
Chapter
4.
(n
at
0).
Both the channel
region
taken
and the quantum dot

over a
region overlap and
the
matrix element is calculated

some
by integrating
as
surface 0 in the
of the
point
x\>
(usually
the
mid-point
yz-pUme barrier, Fig. (4.5)).
This leads to
Mp
Tr
2m"
<4'rb,
ao((;,=N
dydzQ{)(x\yy.
0
-
K{xh)r<p(xh;y, z)
where
y^t/pUpp
cp.
^ ==
-)
(4.79)
K"(,rb)
is defined
by Eq. (4.78)
with
Eqk
element in
Eq. (4.66)
the
tunneling
rate
Yp
is obtained.
Inserting the matrix Equation (4.79) is the
central result of this section.
Chapter
Provando
e
riprovanclo.
By proof
and refutation.
DANTb, >Paradiso, IA
Numerics
5.1
Introduction
Subject of this chapter

tions for the transistors.
are
the numerical methods
employed
to solve the equa
linear-response model which is used to simulate single-electron The basic equation is the nonlinear Poisson equation
V-(tVrf
where the
p
=
+
as
/o[c)]
the
0,
of all
(5.1)
charge density
p is
given
sum
charges
in the device
n X q(p X~). A considerable portion of the charge is cal Xrff culated by solving Schrclinger-type equations for confined electrons whose interaction is not sufficiently described by the classical Coulomb repulsion alone but requires the inclusion of complex quantum-mechanical many-body interactions. In general, it is not possible to lind analytical solutions for this equations. Therefore, the development of efficient numerical methods is of paramount importance for single-electron device simulation.
The numerical solution of
Eq. (4.14) proceeds
in three steps: first the de
vice is subdivided into small elements with constant electrostatic
potential
and
carrier densities. Then the
equations
in these domains
are
approximated by
al
gebraic equations using a finite-difference discretization scheme. This leads to a fairly large system of nonlinear equations for the electrostatic potential f and the charge density p. In a third step this system of equations is solved.
73
74
Chapter
5.
Numerics
5.2
Space
discretization
5.2.1
Finite differences
A discretization
approach in which the differential operators are approximated by differences is known as the method of finite differences (FD). The device domain is partitioned in subclomains separated by grid lines which run paral
lel to the
sented
(Cartesian) coordinate
axes.
Such (/-dimensional
by products
of d one-dimensional
grids can be grids. Consequently, they are
repre
called
tensor-product grids.
The nomenclature for
a
two-dimensional nonuniform
=
grid grid
is shown in
is numbered
Fig. 5.1. In three space dimensions (d by a discrete triple p (?', j, k). The
3)
point
on
the
coordinates of this
rp
(xz,yvzk)
where the elements of this

to
triple
are
point are given by the positions of the
ith, jth and A*th grid line parallel
the
.r-,
y- and -axis,
separation
of the ?'th from the
(i
l)th grid
7
line in the
respectively. The .r-direction is given as

(5.2)
a
IPj
where nT is the
total of
n
j
.iv i
r?,
0,...
na,
highest grid
a
line number in the direction of the r-axis and
--(-1 lines is present in this direction. A scalar
the
grid
is
represented by
ui,).k
quantity ?/(,r, yy z) finite number of sampling points ul))tk with

?-/>/)
at
Hpap-a)
x,
+.r,-ri
zlJjXk
0,.
..
nT
.r,_
+ .r,
U7-
l2.).k
nr,
u
sVvzk
/=-0.... ,nr
--
(5.3)
and
0,
ny and k
ru
if not stated otherwise.
Assuming
can
x
that
is three times
an
be used to find
continuously differentiable, a Taylor expansion approximation for the first order derivative X (x) for
of
(r,_|
x,).
The
Taylor expansion
ii(x/}
is
given
as
4P)
"(*")
(p
A.r)
dx
^XfX^p> (.r) + c7((r?

2 dx-
xf).
(5.4)
Subtracting
an
equivalent equation
for
u(xî)
from
j
definition of
(/
in the middle between .r, and xt
Eq. (5.4) and using the (Eq. (5.3)) the derivative at
5.2.
Space
discretization
75
y,
~i
":
P
1
wF j
"',-1
/7-1
'j-i V 7
X-i
1
*4
t
i;
Figure
node
5.1: Elements and vertices in
finite difference discretization.

elements
The
circles mark the nodes and the dashed
area
is the box volume associated with
i,j.
Phe shaded
areas are
the
different
surrounding
the node
.r,
is given as
Ox"
This leads to
an
'< +
-1.7.A
.
A,;
AM
0((h
x\lh
(5.5)
approximation for
e.
the first order derivative which has second

error
order accuracy, i.
the local truncation
O is of third order. The above

as
definition of the first order derivative is known

mation. It should be mentioned that without the tion the local truncation
error
central-difference approxi central difference approxima
would be of second order.
C7((/?7C)A.
Tensor-product grids have two major drawbacks. Hie first is the poor con trol one has over the point density of the grid. Refining the grid in regions where the gradients of the sampled properties are high always creates an abun dance of unneeded points in other regions. The number of grid points deter
mines both memory and CPU time
requirements
hich
the
can
seriously
limit the
utility
of finite differences. The second is related
to
rectangular properties
of the grid which makes it difficult to model non-rectangular device features.
76
Chapter
5.
Numerics
regularity of the sparse linear systems emerging from the discretization. The simple banded structure of the matrices allows for the use of very simple data structures and algorithms. Finite differ ences are therefore very appealing if a discretization scheme is needed which
can
The main
advantage, however,
is the
be handled with little effort.
5.2.2
Discretization of V
(y V)
such the non-linear Poisson
The
partial differential equations,

and the
are
as
equation,
Eq. (4.14),
eigenvalue problem
characterized
associated with the Kohn-Sham equa

the
same
=
tion, Eq. (4.28),
by
form of the linear operator
VfyV).
of the both
The material
tensor in the Poisson
dependent tensor 7 (")1/fJ)1,^,e{x,y,z} is the dielectric equation with the components yv^ eUfl and the tensor
=
reciprocal effective
the tensor is
a
mass
in the Kohn-Sham i.
e.
cases
symmetric,
y*"
equation, yvp l/m* In 7^. For symmetric tensors
there
always exists
coordinate system in which the tensor is
diagonal
J^(jlJSp,lX,,e{x,yX=
Furthermore, in isotropic media, yz
the
case
fy
0
0
0 \
0
.
7y
(5.6)
Of/
7^
7". This will be assumed to be
in the next steps.
approximation to the differential operator V (7V) can be derived using the box-integration method. An algebraic equation for an operator at a grid point (i. j, k) is derived by integration over the box p
The finite difference
volume
m7>
{.a
^ 'Ai
v>y3-\ ^
y a
yJ+xzk-x ^z ^
zk+\}"
(5.7)
Application
of Gauss law
yields
f
X
,k
drV-(y(r)Xi,(r))=
I
do/
,
d/7(r)Vw(r),
k
(5.8)
which leads to
surface the line
integral
over
the surface of the box defined
planes bisecting
connecting
a vertex to
its next
by the six neighbor (for vertices
5.3.
Linear
equation solving
and
eigenvalue problems
77
on
boundary, only the volume inside is included). symmetric matrix A which operates on the vector u as
the
The result is
real,
:u),
v=.{i y.s}
hf(h/
hX^X2.3,kvi,xk
where
(Au),^^
for
denotes the
row
(column) of the symmetric matrix associated

of the
=
with the box
equation
(?,j, k) multiplied by u, i.e. the left-hand side u in the box (?, j.k). The order of the matrix isn
a
algebraic
for for
nxxny xnz.
Because of the number of
^?,;,A 0M;a
R1, 1R3,
off-diagonal entries the matrix is tridiagonal five-point matrix for il, j k ES2, and a seven-point matrix
e.
i.
the summation in
Eq. (5.2.2)
is restricted for
problems
with
dimensions d Neumann
tions
or
2.
boundary conditions, especially homogeneous boundary equation

for the electrostatic
condi
(used
in the Poisson
are
potential
as
at
artificial
boundaries)
given
for vertices in the surface
region
nVw(r)
They
reduce to u,y],k-\
i
0.
0 B-
(5.9)
in the ^-direction for in
u,
; k
(n
(0, 0,1))
Implementation of Diiichlet boundary conditions in the surface region D is trivially performed by setting iy uq at (07, y, ,zk) G D where uq is , k a physically meaningful value for either Ohmic- or Schottky contacts.
stance.
=
5.3
Linear
equation solving
equation
and
eigenvalue problems
5.3.1
Linear Poisson
The operator of the linear Poisson
equation
is the
.
Laplace operator C
V
t
(\7)
which acts
on
the electrostatic Potential
The dielectric tensor
is
taken for
inhomogeneous
but
isotropic materials,
which makes all entries of
78
Chapter
5.
Numerics
the matrix
equal
to
e.
hand side of the linear the matrix L
Application of Eq. (5.2.2) and integration of the right equation results in a real symmetric matrix problem for
t(f>=. -ftp.
where <p and p
the total linear system in
are
(5.10)
vectors of the vertex
charge density,
and ft is the is
representations of the potential and diagonal matrix of the box volumes. The
to
Eq. (5.10)
equilibrated according
=
(D-l/2LD_1/2)(D1'20)
in order to restrict the
therefore
-D-1'2fipJ
with
diag(L).
0 <
(5.11)
2. and
improve
the
eigenvalue spectrum of the operator conditioning of the matrix [60].
to
A7
<
The solution of the linear Poisson
equilibrated
cretization
matrix L
TL
2LD
equation requires the inversion of the L 2. Due to the fact that the box dis
couples only between different grid points which are neighbors separated by the box faces, the linear system Eq. (5.10) is sparse, i. e. only a small portion of the matrix entries is nonzero. Furthermore, since the FD
method is used, the
nonzero
entries
trix. To limit time and memory

ces
always fill the same positions in the ma requirements for the inversion of linear matri
it is
mandatory
of the
sparsity
YSMP
l
employ data structures and algorithms that make use of problem. A common data structure for sparse matrices is the
to
nonzero
format: all
of the
entries of the matrix
are
stored in
dense vector.
are
In
vector
a
same
length
to
the
corresponding
n
column indices
listed,
are
and in
third vector
row start
pointers
as
the start of each
row
in the other vectors
stored. The
vector has
length
X L, the last entry

row
being
a,
sentinel.
This format is also known
condensed
format
or
( ia, j
a) -format
because of
an
often used
naming
of the three vectors.
The linear system
can
be solved
by
sparse direct methods, i.
e.
Gaus
sian elimination,
or
by
sparse iterative
methods, usually derivatives of the

The matrices
,
conjugate gradient
dimensional
method
(CG) 154].
resulting
from three-
grids
are
often rather
as
large
applied.
large
CG based methods
BiCG
only iterative methods can 1551. CGS [56] and CGSTAB [57],
are
and
be all
combined with ILU
preconditioning [58)
nowadays
methods of choice in
scale device simulations.
Aale Sparse Matrix Package
5.3.
Linear
equation solving
and
eigenvalue problems
79
5.3.2
Schrdinger equation
is called Kohn-Sham
The
Schrdinger equation (which

a
equation
a
in the
case
of the presence of
Kohn-Sham
potential)
constitutes
matrix
eigenvalue
problem for
the Hamilton operator matrix H
H'0
where the operator 'K
En ft'if)
n.
(5.12)
-t
V, the FTamilton operator with the kinetic part
1/2
(l/r/W),
effective
is discretized
using Eq. (5.2.2).
The
inhomogeThe
neous
reciprocal
mass tensor
1/nV
is assumed to be
diagonal.
potential operator V represents the effective (Kohn-Sham-) potential and is given as the product of the unity matrix times the vector of the vertex poten tial V(r) - (IV)l)k, therefore being a contribution to the diagonal of H. En and tpn are the eigenvectors and eigenvalues of the operator.
In order to transform this lem both sides arrives at
are
equation premultiplied by ft
into
x
standard matrix
=
2. Substituting
eigenvalue prob O1'2 one ft
(fr^Hfr12)(o17Vn)
or
En(n1/2if)n)
=
(5.13)
and
Hi/>7)
Enifn
with the still Hermitian matrix H

=
ft"1'"H. ft"1'"
ft1,/2ipn [60]. The matrix is real and symmetric with eigenvectors tfn out an applied magnetic field, which results in real eigenvalues and eigenvec tors. Boundary conditions are such that the eigenfunctions ipn vanish at the boundary of the Schrdinger grid (homogeneous Diiichlet boundary condi tion) which comprises only a small portion of the total grid.
the
problem as with linear equations arises with eigenvalue prob lems. The matrix representations of the operators are too large for direct meth ods. Alternatively, one uses iterative methods which calculate the eigenvalues
The
same
and vectors at the lower bound of the spectrum up to

is
a
desired number. This
reasonable
therefore
approach, since only the lowest eigenvalues are occupied, physically meaningful within a Kohn-Sham framework.
and
Especially, subspace (Krylow subspace)

be
iteration methods have proven to
extremely powerful for large eigenvalue problems arising not only from space discretizations of the Schrdinger equation. A rather new method in
style QR
near a
this context is the Jacobian-Davidson

is able to obtain
a
method
(JDQR) [59]
which
use
group of
eigenvalues
target At without the
of
80
Chapter
5.
Numerics
shift
and invert strategy.
JDQR is
A search
trying
to
find the smallest

is
eigenvalues
which the
of the matrix
(A
AtI).
is
subspace
generated
onto
projected. The much smaller projected eigenproblem is solved and this leads to approximations for the wanted eigenvectors and eigenvalues of the larger problem (Davidson part). Then, a correction equation for a selected eigenpair is considered. The solution of the correction equation defines an orthogonal correction for the current eigenvector approximation (Jacobi part). The correction is used in the expansion of the search subspace and the process is repeated again. JDQR aims to find &max Schur pairs (q, A)
near
given eigenproblem
At.
It consists of and
an
an outer
iteration, in which
a
an
approximate
an
Schur
pair
is
computed
matrix
inner iteration in which

a
linear system is solved. The linear
system is preconditioned by
matrix K which is
(A
AtI).
Deflation is
i.e. the linear system is
applied w henever a projected onto the subspace orthogonal
approximation of the Schur vector has converged,

to the
found
Schur vectors.
5.4
Nonlinear
equation solving
5.4.1
Introduction
Solving
can
nonlinear be
equations
like the nonlinear Poisson
equation
are
is
task that
only
performed numerically. Traditionally,

solves for
all terms
moved to the
left and
one
F(x)
and
R*v
->
RN,
F)
(5.14)
are
where F is the AT-dimensional vector function whose components
the
individual
equations
to
be satisfied
simultaneously.
The functions
17,
are
func
tions of real variables. Furthermore, it is assumed that
they
are
continuous and
diffrend able with respect to these variables. In many tions

are
cases
the vector func
extremely complicated and hence laborious to compute. This has the immediate consequence that during any iterative solution process the bulk of the computing time will be spent in evaluating the vector function. Thus, the most efficient method will be that which requires the smallest number of
function evaluations.
Other than for linear
equations,
root
finding
an
for nonlinear
equations
in
variably proceeds by
iteration.
Starting
from
initial guess vector for x, the
5.4. Nonlinear
equation solving
81
improved until some predetermined convergence criterion is satis fied. Even for smoothly varying functions, most of the convergence behavior of an algorithm depends on the quality of the initial guess. Providing good initial guesses requires at least some insight into the non-linear problem, i. e.
solution is
some
analysis concerning
the
properties
of the vector function F.

semiconductor device simulations method. However, in most
cases
Most of the methods

are
based
on
the
applied today in familiar Newton-Raphson

can
of
actual interest, Newton methods

are
not be
applied
in
pure way since
they
or
not in
general globally converging.

ensure are
The combination with

a
damping-
line-search schemes to
global
bad
convergence is
very
common
practice.
performers essentially in two cases. First, if the initial guess is only a poor approximation to the solution, and second, if the Jacobian matrix can not be given explicitly and a poor approximation to the algebraic expression for the derivatives or numerically calculated derivatives
Newton-Raphson
methods
have to be used.
single-electron and other quantum-median ics-based simulations, the problems arise mainly with the quantized charge densities which can not be given as functions locally dependent on the electrostatic potential. Therefore,
In
the Jacobian matrix has to be
approximated in some way. In most cases the semi-classical expressions for the charge densities are used to approximate the Jacobian entries which are due to quantized charges. However, quadratic con vergence can not be guaranteed with these approximations, and in many cases
the Newton iterations will fail to converge at all.
'fhe
following
schemes
section is
as
mainly dealing
with inexact and
damped Newton-
Raphson
and
arise from
low-temperature
calculations
are
complete
problems
confinement of carriers. The methods
including partial especially designed to
reduce the number of function evaluations the

due to the inexact Jacobians.
required,
and to work around the
5.4.2
Newton-Raphson
method
The most
equation solving in one dimen sion are bisection and bracketing. The only elementary method that works in multidimensional problems is the Newton-Raphson method. The nonlinear equation is linearized in the vicinity of the solution using a truncated laylor important
methods for nonlinear
82
Chapter
5.
Numerics
expansion
F(x
Sx)
F(x)
+ J Sx X
(Sx2),
are
(5.15)
where J is the Jacobian matrix whose elements
the first order derivatives
of the elements of the
vector
function
OF
J^-zr-1By neglecting
set
(5-16)
OXj
+
terms
of order
<5x2 and setting F(x
Sx)
0,
one
obtains
of linear
equations
for the corrections <)x.
-J"xF(x)
that
move
Sx<
(5.17)
to
the vector function closer to
zero
according
Sx.
xnow
x0id 4
(5.18)
In this sense, Newton's method

a
replaces
system of nonlinear equations with
equations, but since the solutions of the two systems are not identical in general, the process must be repeated until a certain accuracy is achieved. The Newton method converges quadratically if applied close to the
set of linear
solution.
5.4.3
Broyden's
updating
method
Secant
methods
The
high
cost per
to
iteration of the Newton method and/or the difficulties for

calculate the Jacobian matrix has lead to alternative meth
secant method. These methods
on
some
problems
an
ods, analogous the one-dimensional

build up
gradually
approximation
to
the Jacobian based
successive evaluations of
the function vector without effort is saved
explicitly calculating
a
derivatives.
Computational
Jacobian
ma
by updating
factorization of the
approximate
trix at each iteration
formula). Because
(employing techniques similar to the Sherman-Morrison of these two features they are often referred to as secant
these methods gen
updating methods. However, the price to be payed is that erally have super-linear and not quadratic convergence.
5.4. Nonlinear
equation solving
83
Broyden's original
method
One of the most effective schemes for secant
updating
is
Broyden
\s method
approximate Jacobian matrix which is updated at each iteration. If the negative inverse of the approximate Jacobian is denoted &-m^ (J^^pmox me /?7tn quasi-Newton step <5x(m) is the solution of
[61], which begins with
=
an
GX)FX) ^Sx{m) ^0,

where and
(5.19)
(he*)
xl'7H
can
l)
x^m) (cf. Eq. (5.18)).

a
In
general, F(x
Sx)
A 0
Eq. (5.15)
be recast into satisfies
quasi-Newton
secant condition which is
such that
F(x
Sx)
Sx(m) XG{m)SF{m)
where
0,
(5.20)
of the secant
SF^
F^r"4L)
F('").
This is the
one
generalization
approximation to the derivative F' in and numerically advantageous to scale
dimension.
It is both convenient
as
the vectors and redefine Sx and SF
S^X)
^^
__
_
,,p(TM+i)
F("7)|lo
and
SF{m)
"
||F0 + l)
F(n0jjo"
(5.21)
Note, that
now
SF(m)
SFX)
l.
Since
G^ generally
fails to
satisfy
is
Eq. (5.20)
used
an
updating
formula for
G^r") minimizing
HG^"1"1)
G("^||2
Q(m + l)
where
Q(m)
(fa(m)
q(w )SF(X)SFXX/
(5-22)
Eq. (5.20). Equation (5.22) is the Broyden updating formula which is often applied using the Sherman-Morrison formula for an analytical inversion of the approximate (quasi-) Jacobian.
is consistent with
G^m+1^
Modifications of
Broyden's
method
Vanderbilt and Louie

method in which
163] introduced
modified version of the
original
previous iteration steps during the updating procedure. Even though the overall performance of the method is improved over the original method, the storing and multiplications
they
retain the informations from all
84
Chapter
5.
Numerics
shortcoming especially in large-scale applica tions. A new updating formula was derived using least-squares minimization of an error function, which includes all previous vectors Sx and SF weighted
of Ar
x
Ar matrices remained
with
factor l for the nth out of??? steps and

error
factor cjq
assigned to
the
error
in the inverse Jacobian. With the
function defined
as
cjqIIG^1)
to
a
G(n||2
Y^n
n1
I S*{n)
G(m+1W' |2,
(5.23)
subject
variational ansatz
according
=
to
DE/dG
x
one
obtains
G^HJ)
with the
(7HL)(B(77"n)
according
and
to
(5.24)
auxiliary
A)
=
matrices A and B
("<
QO")
x^FXXr
B(m+i)
TA"^1
|
=
FX)fXX^
and
where, the
following
notation for the vectors is introduced
(ujn/ujo)F^^
Sx^71)
--
(uj1i/ujo)Sx(7^
x(n) p()
Still,
a
(Sx^l\
AAA)<7rA(,Vxn)
,F(n))
G
(F(1V
X{NX71)
performed to get (B^nH l^) 'is nec in terms of S~F in essary. Johnson [64] suggested an expansion to infinite order order to avoid explicit matrix inversions. By applying the Sherman-MorrisonWoodbury formula2, the following form of (EF'"^L))~~L was derived
matrix inversion which has
to
be
(B(m ID)
I
=
I +
f(m)(m)f(m)T
l
wjth
8(m)
where
-(AF^?,)F(m))a
dM/mxm\
Using
the result in
is
a m
matrix and not
large
A" matrix.
e.
Eq. (5.24).
a new
updating
_
formula is obtained, i.
q(>+A
2Shei
suppose
g<7)
(Qt')]Am)
toi muta e
4.
-yrX))/f>XFXXi\
and both
(5.26)
man-Mon ison-Woodbm
PVxA\
:-Yxl.V
1
^Vx
and
VTA
1TJ)
are
nonsingular,
than
VTr
A"
-"
L(
VT
1) 1VT.
(5.25)
5.4.
Nonlinear
equation solving
85
This method
can
be
improved
even
further if
G^m^
^ is written in terms of
G^L) [64],
i.e.
qO+i)
with the matrices
__
q(i)
2;("^F^n^T
as
(5.27)
\j(X
and
Z^
defined
i
follows
-ry(m)
A<(t)A(n))
VAA
"
2(m)
The
tion.
^fjOA^ j2(m
formula
can
| 0])/J(m).
be
proof of
The
this
new
updating
as
easily
new
method
as
requires
the storage of the and
accomplished by induc current matrix , and the previous

iterations
vectors x
and F
well
the vectors Sx and SF from the

a m x
m
(storage of
The
2(???
1) A-veetors
matrix).
formula
can
following approximation
Q(m + 1)
for the
updating
be made if
Uq <C Ld7)
=
q(D
fiirn)(m)f(m)T^
form of
(52g)
It
only
remains to find
as a
an
appropriate
the
G^.
-
It is
straightforward
to choose it
diagonal
constant matrix
G^
ai and
a
(0
<
cy
<
1).
According (5F(2) ()F(2)
to Powell
<
F(1)
[651, appropriate ()F(1). More elaborate
choice of
should be such that
choices for
G(1)
um.
be
envisaged.
Another
question
is the choice of the

"l
weights
certainly Johnson [64]

can norm
suggested
to use ujm
(dF^ <5F^7^)
2,
i.
e.
the inverse of the
of
the vector function.
5.4.4
Globally converging
Newton schemes
Newton's method may fail to converge when started far from the solution. Un
fortunately,
a
in AAdimensions there is
no
analogy
to bisection that
can
provide
fail-save
hybrid
method when combined with Newton iterations. Neverthe

to
safeguard the Newton's method and open it to a much broader range of applications. Two methods will be discussed her, one is the damped Newton method especially in the form of the Bank-Rose damped
less, it is
possible
Newton method. The other

on
on
is
an
Broyden's
In the
method
as
discussed
extrapolation method which in the previous section.
can
be based
damped
Newton method, the Newton step Sx is calculated

new
as
usual
at everv
calculation, but then the
iterate is taken to be
x0n+i)=x0)
nim)-x(n
with
o<o(m)<l.
(5.29)
86
Chapter
5.
Numerics
where
(m) is
scalar parameter. The motivation is that far from the solution
the Newton step is

tor
oOA
can
be
likely to adjusted in
be unreliable, in most of the such

a
cases
to
big.
The fac
way that
x^777+1)
is
better
to the
solution if the step is
damped. Sufficiently
close to
approximation the solution c\X)
should
approach unity
are
and the convergence properties of the classical Newton
method
retained.
One method to estimate the

Rose
damping
factor
was
introduced
by
Bank and
[66]. They
showed that under certain conditions

one uses
global
and
quadratic
convergence
is achieved if
with Kk such that
a^
satisfies
<
sufficient
decrease condition
cy(,n)S
where S
>
(1
HF^-^lla/'lFÛa)
can
(5.31)
as
0 is
fixed, small number which

noted that
be taken
the machine fast

as
epsilon. iipAOlL
It is
_^
readily
a^ approaches unity sufficiently
||F("7+i)||<>.
a
The method after Bank and Rose is

conductor device
modeling. Other than

not
popular method in the field of semi the extrapolation methods discussed

e.
in the next section, it does
alter the direction of the Newton descent, i.
only
[66].
been
the
length
of the method,
Practical
adapted. Convergence properties therefore, remain predictable and are investigated thoroughly experience with this method which shows its capabilities has
of the correction vector is
reported by
Fichtner and Rose
[67] for instance.
An alternative method to the Bank-Rose
damping
method is based
on an
e.
extrapolation
scheme used to determine the

__
optimal
correction vector, i.
x(m+l)
with
n
x(?7) j-l)/x(m) x(,)
x(")
n)
^x(rnn)\
(5 32)
extrapolation depth. Laux [60] used the Anderson extrapo lation method [68J to extrapolate the correction vector in an Newton-Raphson iteration. This combination proved to be superior in convergence if applied
being
the
to
the nonlinear Poisson

an
equation describing
the
charge density of confined
electrons in
two dimensional channel.
Care should be taken in
damping
methods not
using extrapolation methods since in only the length of the correction vector but general
poorer convergence
contrast to
also its di
com
rection is altered. This leads to in
properties
pared
to
damped
Newton methods.
Especially, quadratic
convergence is not
5.5.
Self-consistent
algorithms
87
achieved in most of the
cases.
However, extrapolation methods
can
be advan
tageous if the
method
vector function is very
expensive
to
evaluate. The Bank-Rose
find
tion
a
requires at least two function evaluations, in many cases even more to an appropriate choice for a777. If, for instance, the evaluation of the func vector includes the solution of Schrdinger equations, this can consume
considerable amount of time.
5.5
Self-consistent
algorithms
equation solvers for the solution of the nonlinear Pois son equation are implemented in the SIMNAD software. Both algorithms are based on Newton-Raphson iterations and a predictor-corrector type approach for the evaluation of the vector function F. They are mainly distinguished by
Tow different nonlinear
the methods used to force and stabilize the convergence. The first
uses
algorithm
the Bank-Rose
damping
method and the second
Broyden extrapolation
steps.
5.5.1
Predictor-corrector
approach
Before the methods and there
implementations are explained in more detail, the main idea of the predictor-corrector method in the solution of the nonlinear Poisson equation involving the Schrdinger equation for confined states is
explained.
As
already
outlined in Section 4.2.4, the electrons in bound states of partial confinement

are
or
sub-
bands in the
by the Kohn-Sham equa tion, Eq. (4.28), for the envelope wavefunctions. The charge densities in re gions with zero-, one- or two-dimensional confinement are given by Eq. (4.30) with the (Fermi-Dirac-) distribution functions Eqs. (4.33), (4.34) and (4.35). Apparently, the exact dependence of the electron densities n on the electro static potential is not known. This results in an incorrect contribution to the Jacobian if a Newton-Raphson method is employed. Newton-Raphson meth ods can only be used without the Schrdinger equation coupled via the charge
case
described
densities, i.
e.
with semi-classical
charge
densities
as
in
Eq. (4.16).
to
In many calculate
approaches potential
is
the
dependence of the electron charge density on the approximated assuming a semi-classical form for n
for instance
electrostatic
On/0f> (see
[60]).
88
Chapter
5.
Numerics
approximate expression (predictor equation) for the quantum electron density n[] in O for which an explicit de pendence on the electrostatic potential can be given. Consequently, the Pois son equation
A
more
advanced solution is to
use an
V-(cV0)
can
/)[?7[o].oAÔ
(5.33)
Newton-Raphson approach (for all charges other then the quantized charges, the semi-classical expressions are used). The predicted result for n is then corrected in an outer iteration, i. e. the Schrdinger equation
be linearized
using
expression for the charge density is evaluated. An expression for a predictor density which depends explicitly on </> was derived by Irellakis et al. [69]. They applied first order perturbation theory to the
is solved and the correct
Schrdinger equation in a quantum wire. The one-dimensional electron charge density is given according to Eq. (4.30) and Eq. (4.34), i. e.
kBT
with the
dimension-dependent factor
'???* knlX
^f*|=- I
2
AYd
2<? I
2?rr?2
(5.35)
A similar
expression (predictor)
is derived
XX^Xt
/A^+^-^x)
in which the subband energy

mented
t,/
from the last outer step (corrector) is aug
by
the
change
in the electrostatic
potential,
i.
e.
fn[o]-^}- q(6-{k)).
This form of the
(5.37)
density
is in
particular
suitable for
since
an
expression
for the contribution to the
Newton-Raphson methods Jacobian can be given

+
AM
0(0
^D,,nW)^
k'li
I
z'
2
|V-A y--^)
V
A'b
(53g)
5.5.
Self-consistent
algorithms
89
In
perturbation ansatz can be extended towards the electron densities in quantum wells (2DEG) and quantum dots (ODEG) re sulting in similar expressions. If the electrostatic potential is sufficiently con verged, it is used as <pA+i) m tjie Schrdinger equation and ??^,+i) is used to
principle,
this first order
calculate The
son
a new
exchange-correlation potential
V^l/h^^^}.
predictor-corrector
can
method for the solution of the nonlinear Pois
equation
=
be summarized
Fpicd
0 is solved
using
the
predictor type equation approximate density h[<p), Eq. (5.36), i. e.

as
a
follows. First,
Pprcd
Then
a
(fV^l+1))+pWl ui):0(k\Hk\t^}l<f>(k+1)].
(5.39)
corrector
step is performed which includes the solution of the
Schrdinger equation with the electrostatic potential solution of the predictor equation (/>^'+i) and the exchange-correlation potential calculated with n/k'^ A
Fr_
-^V
/ V
-VV
m
A?c
#(A'41}
V4c[^+1)]
(Ml)
L(11k+l)ij^+l).
With the exact
(5.40)
the correct
non
(corrector-)
electron
density n[<p\, Eq. (5.34),
linear function vector
Fcorr is evaluated
+
FcolI
(cV<fi{k+1))
are
/>[[^+1);{V>ifr fl),4frtl)}]^(A * l)]^|Fcou||2

<
e.
(5-41)
The iterations
repeated
until
5.5.2
Algorithm implementation
in siMNAD
Figure 5.2 shows a flow chart of a Newton-Raphson method with Bank-Rose damping for the above described nonlinear Poisson equation. The vector func tion F is composed of the \alues of the residual of the Poisson equation at
discrete
grid points so that the order of the nonlinear problem is given by the size of the grid, i. e. by the number of nodes. The argument vector x is the scalar electrostatic potential o at discrete grid points. For the predictor form
of the vector function
Fpiea
the electron
Eq. (5.36) or the plicit dependence on the electrostatic potential according

to
density ??[</>] for a 1DEG is corresponding expressions containing

d?
chosen
the
ex
for 2DEG and ODEG. For
the corrector form of the vector function the exact

to
expression n[</>] according

re-
Eq. (5.34)
is used. All other densities
(electrons outside the confinement
90
Chapter
5.
Numerics
Setup
XA
xmit
Concctot step
Yes
FÂ
=Fcou(x(A;
tf(rn)
p(m
I t)
__
New
ton-Raphson step
=
(A(7
(_J i)(m)F(0
Bank-Rose step
x(m 1-1)
__
x(m)
aA'Ax(m)
Piedictoi step
F(m-+1)
(m)
Am)
Fpred(x("'fl)
Aa,iifA)a:
Figure 5.2: /-'low chart of a Bank-Rose damped Newton-Raphson algorithm for a predictor-corrector t\pe nonlinear problem. Phe i onvergence criterion for the corrector equation Fcoll is S{ and for the predictor equation Fpiej
the convergence criterion
is
s2
to a
Usually,
one
chooses t~2
=-
lei,
i-
<?
the
predictor equation
is
solved
higher accwac\
then the corrector equation.
5.5. Self-consistent
algorithms
91
gion, given
holes and ionized

in Section 4.2.2.
donors)
are
treated within the semi-classical
equations
The second
algorithm
is based
on a
Broyden extrapolation scheme. The new secant approach (BroydeiTs method) described in Section 5.4.3. update vector is extrapolated from the previous steps. This method has the advantage over the Bank-Rose method that the vector function has to be com puted only once during one Newton step, whereas the Bank-Rose method re determine the correct damping quires multiple, but at least two evaluations to factor c\(nt\ This can be extremely costly if the evaluations have to be per 100 000). On the other side 50 000 formed for vectors of high order (N
with
a
~
...
Newton-Raphson method combined Fig. 5.2. Using the multidimensional
Setup
x(0)
=-
xmit
Coiiectoi step
FO")
=Fcoll(xOA)
'
No
I
I
Piedtctoi step
F<'^Fpied(x(A)
'A"
^
^"""\
NO
V.
Newton-Raphson step
dX,n)
(-J
A(m)FA)
Bioxden
extiapolation step
.
X('U)
x(>-
t)
(Sx{ni\ 5F("'\
SXm "'0, $~pX
"))
Figure 5.3: Flow chart of a Newton-Raphson algorithm combined with a Broyden extrapolation scheme for a predictor-corrector type nonlinear prob
lem.
92
Chapter
5.
Numerics
the
use
of the
Broyden extrapolation
n
method
requires storage
of
at
least 2n
vectors
of order N. The number
is the number of
previous
iteration steps
used for the

The
extrapolation.
and the
success
speed
e.
of convergence
heavily depend
on
the choice
of the initial vector ximt, i.
the choice of the initial electrostatic
potential.
Usually, this is not a big issue in quasistationary calculations where a gate voltage is ramped with considerable smoothness and so the boundary condi tions of the Poisson equation change not very much from one run of the solver to the next. In this case the potential from the previous iteration is always a
good estimate. However, at the

7
very
J
beeinniim of the

calculation, when
"
no
in-
previous runs is available, the Eaplace equation, subject to the appropriate boundary conditions is solved and the (homogeneous) potential is
formation from
used
as
the initial choice.
5.5.3
Examples
In order to demonstrate the convergence behavior of the non-linear methods
presented,
performed. A GaAs/AlGaAs structure similar to the one discussed in the next chapter is used. The 2DEG in the GaAs/AlGaAs heterojunction is modeled assuming three independent
a
calculations for
test structure are
1er2 1CA4
=
=
10~6
108 KT10
10
12
_j_
500
1000
1500
2000
CPU time
(s)
Figure
imation
Rose
5.4:
Convergence of the
guess
a.
Bank-Rose
algorithm (o)
curve
with
poor approx
of the initial damping factor
potential.
and the
Phe other
(A) shows the Bank-
Phe dotted lines mark the convergence threshold for
the corrector
equation (c\)
predictor equation (S2), respectively.
5.5.
Self-consistent
algorithms
93
regions
with one-dimensional confinement in the ^-direction. Two

source
related to the drain and
region, respectively,
and the third
regions region is
are
the
quantum dot region.
Figure
son
5.4 shows the convergence for the residual of the non-linear Pois
damping factor a for the Bank-Rose algorithm. The solution of the homogeneous Poisson equation was used as initial estimate for the potential which is a rather poor approximation resulting in a slow con vergence during the first steps. The residual exhibits the typical seesaw-like behavior of predictor-corrector equations since the residual always increases if the corrector equation is solved. A good performance of the Newton-Raphson method is achieved for the predictor equation because of the exact Jacobian involved. The calculation was performed at a temperature of 4.2 K. equation
and the
The result of this calculation is used where the 2DEG model ODEG model
as
||F^'7^ ||2
initial guess for the next

00
run a
previously
used for the quantum dot is 50

...
replaced by
eigenvalues in the quantum dot. The performance of the Bank-Rose algorithm was compared to that of the Broyden-extrapolation algorithm in which the following parameters were 4 and GA1^ 0.5 I. The calculations were now performed 10" used: uq
solving
for
approximately
at
1 K. The
a
where
Broyden-extrapolation algorithm is usually faster in calculations good initial guess potential can be provided, i. e. it is mostly used in
calculations where
a
quasistationary
gate voltage is ramped in small steps and

can
the result from the
previous
calculation
be used
as
initial estimate.
The
10
10
10~8
10
10
AAA
\ /
<
1A
-12
500
1000
1500
2000
2500
CPU time
(s)
Figure 5.5: Comparison of extrapolation algorithm (o).
the Bank-Rose
algorithm (U)
with the
Broyden-
Phe dotted lines mark the convergence threshold and the
for
the corrector
equation (si)
predictor equation (S'i), respectively.
94
Chapter
5.
Numerics
Bank-Rose
algorithm generally proved
to be
more
stable and converges
even
a a
if the initial estimate is rather poor. It is therefore the method of choice if good guess for the potential is needed or no previously calculated results for
structure are available.
Chapter
And here Alice saying to
6
to
begun
in
a
get rallier sleepy, and

sort
went on
cats cat
herself,
cats eat
dreamy
of way,
either
'Do
bats? Do
bats?' and sometimes 'Do bats

answer
eat
cats'?'
for,
vou see, as
she couldn 't

wax
question,
it clidn 't
much
matter
which
she put it.
Liwis Carroll, Alice's Adventures in Wonderlands
Simulation
6.1
examples
GaAs/AlGaAs heterostructure
The most
commonly
used material system for quantum dot structures is
gal
lium arsenide,
GaAs, which combined with the ternary solution aluminum-
gallium arsenide, AlxGai_.,r As. serving as barrier material, allows for the cre ation of very thin epitaxial layers as a result of the almost equal lattice con
stants. The
strong electrostatic confinement
at
the interface between GaAs and

move
AlGaAs
can
be used to constrain electrons to
in
plane,
thus
forming
two-dimensional electron gas The 2DEG

The
can
(2DEG).
any desired
be
given
shape of the gas is defined either which electrostatically deplete the 2DEG underneath. The latter method the advantage that the depletion is not permanent and can be switched
Focal
shape using lithographic methods. by etching a pattern or by metallic gates

has
off.
depletion of the electron gas is associated with a local increase of the electrostatic potential relative to the undepleted region. The resulting potential step in the 2DEG is smooth because of the large depletion length of 100 nm. This is the basis for the so called split-gate technique used to define small
channels
or
dots with variable size and smooth boundaries
by
means
of several
95
96
Chapter
6.
Simulation
examples
r.-T
-,-'
^;A"A
4nm
Alo 32Ga0 esAs

=
^\
\
\
{Nd
16
g x
10i2
>Gao
cm
2)
nm
Alo
'-i
68
As
GaAs substi
ate
Figure
6.1:
Schematic
of the horizontal
structure
of
the GaAs/AlGaAs het
erostructure.
Phe dashed line marks the 2DFG.
gates deposited
on
the top semiconductor
isolated from each other because of the

the metal/semiconductor contact if
layer. Schottky
in
The gates
are
barriers which
automatically are formed by

e.
operated
reverse
bias, i.
by applying
negative voltages
with respect to the semiconductor
ground.
At low temperatures, the distances
conducting electrons are free to move over large (several /./.in) without being scattered and therefore maintain phase
For systems with size
quantization which is connected with a change in the density of states, the Fermi wavelength Ap is important. It de fines the length scale on which the wave nature of a particle becomes obvious. The electron motion is confined and only standing wave solutions, i. e. bound The quantum mechanical wavestates with discrete energies are allowed. function of a particle with the Fermi energy Ap has the typical wave length
Ap
=
coherence.
2tAf
10
and
kp
(2mFX/h2)1^2.
Particles at the Fermi
edge
are es
sential for
Ep
charge transport. In meV and Ap 50 nm.

6.1 shows
an
GaAs heterostructures the
typical
value for
Figure
with
a
example
for
GaAs/AlGaAs heterostructure 170]

nm
patterned top gate.
The 2DEG is situated around 45

a
below the
gate
at the
interface between the GaAs substrate and
AIq 32Gao egAs. The conduction band edge and the

calculated at 1 K
at
are
layer composed of electron charge density

of the conduction band
level
shown in
=
Fig.
the
exposed
zero.
surface (z
0) is
position determined by Fermi
6.2. The
pinning.
The
con
ductance band
to
edge
is
pinned
600 meV above the Fermi level which is taken atomic-force
be
Figure
6.3 shows
an
split-gate structure [71]. The gates consist (Ti) and gold (Au). A 5 nm layer of Ti is first evaporated directly at the GaAs cap-layer. Then a 50-150 nm thick layer of gold is deposited. The Schottky
microscopy (AFM) picture of a of a layered structure of titanium
6.1. GaAs/AlGaAs heterostructure
97
barrier is determined The gates

as
mainly by
the Ti/GaAs contact
(f>s
800
meV).
are
asterisked
the fin
to
06
>
17
~
follows:
is
1 5x10
ger
gate which forms

with
two
E
o
gether
and Q1
the
gates
Q1
quantum point
>,04
E?
1 0x10
w c Q>
contacts
(QPC) by deplet
<5
TJ
iS
02
C
-
ing
electron
charges
in the
00
-
-I
5 0x10
16
O
o
2DEG underneath the gate electrodes.
LU
Usually, negative
are
20
40
z
60
80
100
voltages
to
between 200 meV
(nm)
and 300 meV
sufficient
fully deplete the charges immediately underneath the

gates. A further increase of
the negative voltages leads to
Calculated conduction-band Figure 6.2: edge (solid line) and electron-charge density
(dashed line) of the GaAs/AlGaAs heterostruc

ture
in
Fig. 6.1
in
growth
direction.
or
additional lateral
depletion.
The gate asterisked C is the control-
plunger
a
gate which is used

which is
to vary the
potential
in the quantum dot

an
by applying
tunnel
can
control-gate voltage VA
The gate geometry defines

source
island in the 2DEG
separated
from the drain- and
reservoir
by
two
junc
con-
tions, i.
e. two
quantum point
contacts. The
shape
of the barriers
be
Figure
6.3: AFM
erostructure
picture of the split-gate structure of the GaAs/AlGaAs het SEP in Fig. 6.1. Phe framed region marks the simulated domain. [72f.)
(Courtesy
P. Heinzel
98
Chapter
6.
Simulation
examples
trolled with the gates

controlled
F, Q1
and
F, Q2, respectively.
The island
potential
is
by
the control gate C.

are
Measurements
performed
=
in
dilution
refrigerator
was
temperatures below T
a
1 K. The electron transport
liquid helium at studied by applying

in
small bias
source
voltage (4.3 //V AC)

to
between the drain reservoir to the left and
the
reservoir
the
right.
The current
was
measured
using
standard
lock-in
techniques ([71]
and references
therein).
6.2
Device simulation
6.2.1
Geometry modeling
and
grid
mapped onto a three dimensional non-uniform tensorproduct grid with 69 x 36 x 70 grid points along x x y x z, comprising a total of 173 880 vertex nodes. The total area simulated is a brick-like region of 1000 nmxOOO nmx290 nm. The substrate, which is extending 2.8pm in the
The device in
Fig.
6.1 is
negative 2-direction, boundary conditions

active device
is
effectively
cut-off at 290
as
nm
imposing
von
Neumann
at this surface
well
as
at the surfaces in the lateral .7;in
direct ions where the electric field vanishes at distances far
enough
from the
an
region.
The Fermi level in the substrate is set

e. a
assuming
elec
or a
trode
(ohmic contact) outside the simulated region, i.

connected to
an
metal
backgate
highly doped layer

at
ohmic contact. The
boundary
conditions
Schottky gates (Fig. 6.4(a)) are imposed by use conditions at the relevant nodes (Fig. 6.4(b)) according
the
of Diiichlet
to
boundary
Eq. (4.26).
decomposition of the 2DEG into regions where the cal culation of the electron charge is performed using different models for the electron confinement. Figure 6.4(c) shows the domain decomposition in the 2DEG layer. The quantum dot region (ODEG) is situated in the central region
Critical is the
comprising a total of 12 336 nodes. The wavefunctions and the electron density \anish at the boundaries of this domain. The negative voltages at the split-gates ensure that the area surrounding the ODEG region in the .rty-pane is depleted so that these boundary conditions are
of 585 nmx450 nmx39
nm
adequate. The two quantum wire regions (IDEGs) which are used to selfconsistently model the potential and the 2D-wavefunctions at the quantumpoint contacts (QPCs) are connected to the two quantum well regions (2DEGs)
which
give
the
charge
densities in the drain and
source
regions (leads).
In the
6.2.
Device simulation
99
500
nm
(a)
**i >^*$$$f
^#*^*#*?^4r***^**^**^^f>*^*^*%^^H^
-L
-
El
j^.i^HS.^*$^ | -î*4i-*4^
|
|
-j
X
!-* ,
.
**$***
|...
-i*^^.^.
(b)
-AH'
1
' DEGs
T
Ar
17
,
îiiAriA
TT
a'.X...__
-
XT
+.x:..-.it4l-t4,J
j
i
:;.
.-.+
~~-
ZZX'Â"' :::--Xi:-::^^TT;::: J^--: 4X1. A t T ^Xv-~i AtV A+*- \IL-.T pffif h"
"':
r
.
^T
A.XjXiiiL
l._. i A
lÎTLt
--
ti::pT:
L
.-j|i_r
_
fE:IÊ|
^ :.;-;:
ODEG
X:z:
A
~
4-~i
X::|
^_
1=1
1^
'
'
Al^^-t^
ff|||4fA^1iA Ai|H]t4!ATlHl .:ir".'X:;:itf tt+t-FP":3^4-iic^

j.-..::.
i
(c)
7h
EGs
AÂA
t
tAAx i i
-*-
~"
Xtt "'Alt.
1
"t.i!+i;:XJr
C.U
-
|i
|l
Figure 6.4: (a)- Scheme of pwdiut gnd and Diiichlet
the
split-gate
su m tute
at
(b) Non-uniform
i>ate
tensor
boundai\ nodes
the
layer
(c)
Domain
decomposition of the 2DEG laser
100
Chapter
6. Simulation
examples
yz-plane and the electron charge is calcu lated according to Eq. (4.35). Usually these regions are completely depleted in order to guarantee a quantum-point contact with a sufficiently high barrier.
However, the calculation of the 2D-wavefunctions and of the lD-subband
en
IDEGs the confinement is in the
quantities are needed in the evaluation of the tunneling rates (see Section 4.4, especially Eq. (4.79)). In general, the differ ent regions can overlap. This is mandatory, especially between the ODEG re gion (quantum dot) and the 1DEG regions (lead channels, QPCs), since both
ergies
is necessary since these
Schrdinger equation are needed to evaluate the tunneling rates, i. e. to calculate the overlap integral in the matrix element Eq. (4.79) at some point in the barrier. However, no special boundary conditions for the charge densities are imposed since the QPC is always depleted at least in its central region. Alternatively, open-boundary conditions for the Schrdinger equation in the quantum dot would have to be formulated to account for the propagating components of the otherwise localized states.
solutions for the
6.2.2
Surface and donors
Part of the surface of the device is covered
by
are
metal gates for which Dirichlet
boundary
conditions for
Schottky
surface
contacts
covers a
employed.
The rest is
exposed
GaAs. Since the

the choice of the
exposed
substantial
portion
of the surface,
appropriate boundary potential

are
condition is crucial for the determi
nation of the electrostatic
in the whole device.
The
properties
near
of
the be
exposed
GaAs surfaces
dominated
by
band of surface states
middle of the assumed to

are
bandgap. be pinned to
While the states underneath the metal gate

the metal Fermi level, those at the
can
exposed
surface is
at
assumed to be
modeled
the
pinned at the substrate Fermi level. Fermi-level pinning using Diiichlet boundary conditions for the electrostatic potential
e.
surface, i.
(r)
The conduction band is set to
v>pm-
dftex.
(6.1)
characteristic value above the Fermi level

...
which is assumed at
of
an
Ec
Ep
0.6
0.8 eA for GaAs surfaces. The choice

is
a
appropriate
model for
exposed surfaces
somewhat controversial is
pinning model leads to unrealistic discontinuous steps of the potential at the edges of the Schottky gates. Also, at low temperatures it is difficult to believe that charges from the 2DEG can move to the surface states in response to shifts in the gate voltages in order to maintain the exposed surface at equipotential and in equilibrium with the substrate. An alternative
sue.
The Fermi level
6.2. Device simulation
101
frozen surface model. The surface is treated as a simple dielec tric boundary with a fixed (frozen) surface charge density Qs which does not move in response to the gate voltage. The corresponding Neumann boundary
choice is the
condition reads
On
where
n
is the unit normal vector of the
boundary.
The
the dielectric constant of the semiconductor

tion
can
means
relatively high value of that this boundary condi
0. The frozen surface model, however, approximated by do/On is only valid if the exposed surface is large and smooth [73], [74k If the sepa ration of the gates is small as in split-gate structures, the above approximations
be
are not
valid and the dielectric has to be modeled
as
well for
a more
realistic
description
of the surface
was
[75].
(i-doped at the GaAs/AlGaAs interface by evaporating silicon during the growth process and at the same time stopping the GaAs deposition. The dopands are assumed to be positioned in an atomic mono layer. However, the Coulomb repulsion between the dopands and the high growth temperatures of 600 K lead to a spreading of the layer. The doping3 nm depending on sheet acquires a finite width which is assumed to be 1 the doping density (see [76] and references therein). A typical value for the 2 two-dimensional charge density in the doping layer is 5 x 1012 cm [72].
The structure
...
If
doping
sheet with
constant
assumed, the three-dimensional
density and charge density
an
effective width of 2
x
nm
is
in the sheet is 3
and the
10l9 cm"3.
The discrete nature of the donors is

as
neglected
doping
sheet is treated
jellium.
The
doping density has to be reduced from that grown in the structure since deep levels are neglected. This is reasonable, since half of the electrons from the doping occupy deep levels. These levels are usually found well below the
Fermi level and the associated
charge
can
be assumed to be frozen out. This

..
results in
an
effective
is used to
doping density
of 1.
1.5
1019
cm
3.
The
remaining
adjust the number of electrons in the quantum dot to a particular value. Comparing the charge densities in the doping sheet and in the 2DEG (Fig. 6.2) one notices that the density of the ionized donors is much higher by two orders of magnitudes. Most of the electrons from the donors are obviously absorbed by the surface states.
uncertainty
102
Chapter
6.
Simulation
examples
6.2.3
Simulation results
The conductance characteristics of the heterostructure

a
was
calculated
small
e.
subspace
of the available parameter space. The control
only for gate voltage,
voltage Vg at gate C was ramped from a negative value of 800 mV to the point at which the depletion of the 2DEG underneath the gate is breaking down. Usually, this happens at around 200 mV. All other gates were kept
i. the
fixed at used to
dot.
value of 700 mV. The the
doping density
e.
in the
(Adoping
sheet
was
adjust
charge density,
i.
the number of electrons in the quantum
Figure 6.5 shows along the ^-direction.

Fermi energy is set to
the conduction-band
edge
and the electron densities
The conduction-band offset between GaAs and Al

no
GaAs is 360 meV. There is

zero.
substrate bias in this
example
and the substrate
Displayed are the densities for a parabolic-band model (Bulk), a two-dimensional electron-gas model (2DEG), and a zerodimensional electron-gas model (ODEG). Only the lowest subband is occupied
u.o
!
I
|
|
\
\
\
0 10
0 05
L
^
_
4.0x10
0.6
\
\
0 00
Bulk
-0 05
:
30
%
\
,
:
i
\
>
-0 10
3.0x1017 'E
u
50
70
S 0.4
E?
CD C 1
\
X
-
2DEG
-
ODEG
2.0x10
17
<U
-o c o o
LU
0.2
A*
\
\
,,
LU
0.0
"~~\\T
1
1
1 1
\\
I
1.0x10
--
..~~
III
t,
20
40
z
60
80
(nm)
Figure
6.5:
Conduction-band
edge
and electron
densities
curve
along
the
z-
direction at 1 K. Phe circles at the conduction-band
of the grid planes
in
first
two
subbands
xy-direction. Phe inset displays the (q, c>X corresponding to the 2DEG model.
position wavefunction s of the
show the
6.2.
Device simulation
103
in the 2DEG model. The subband above the conduction-band
energies <fi
at the
and
<f>2
are
31 meV and 55 meV
edge
GaAs/AlGaAs
heterojunction.
by
This
corresponds
Sarma
to
the values of 24 meV and 46 meV calculated

a
Stern and Das
[77] for
similar structure. The difference of the Fermi energy to the
lowest subband is 6 meV.

in the lateral
The conduction-band
v
edge and
the
electron-charge density
n-planc
is shown in
Fig.
a
6.6. The electron
was
calculated
using
ODEG model
density in the center (quantum dot) sohing a 3D-Schrodinger equation. The

was
number of electrons in the quantum dot

gas shows
a
found to be A"
50. The electron
structure which is related to the
shape of
the 3D-wavefunctions
>
Q1
Split-gate structure (top), conduction-band edge (bottom) in a lateral Figure

edge
6.6:
cuts
are
electron
x
density (middle)
and
y-plane.
Phe density and band-
taken 8
nm
below the GaAsXYlGaAs
heterojunction
in the
0DFG/2DEG.
104
Chapter
6.
Simulation
examples
of the electrons which
are
used to calculate the quantum dot
charge density.
is far from be
The form of the conduction band in the quantum dot
region
ing parabolic and the dot electron charge is distorted towards a non-circular shape. The shape of the confinement is determined by the gate geometry in the first place. This finding is in contrast to the calculations of Kumar et. al. [78] which showed a nearly parabolic shape of the confinement poten tial for a similar layer structure. However, they calculated the self-consistent grounclstate for only around 12 gate symmetric in the xj/-plane.
ing the density Fig.
eter
electrons in the quantum dot and for In order to compare with their
was
top
results, the
electron number in the quantum dot

donor
density
in the
adjusted to around 10 electrons us doping sheet as a tuning parameter. The charge

a
in this
case was
found to exhibit
similar
shape
to the one
shown in
6.6
(middle).
The diam
how
of the electron gas,

was
ever,
to
significantly
~
reduced
in the
0 001
-
around
LOO
nm
^-direction.
Measurements in would be
this
low-filling regime
geometry)
difficult (at least with this de

vice because of the increased barrier thickness. The
electron
charge density
more
is
con
>
centrated
in
the
center
of
the
quantum dot and the
3*
>, O)
>~
-0 001
coupling of the single-particle eigenstates to the leads is very

weak.
CD
C
LU
-0 002
single-particle level spectrum and a cut through wavefunctions the along the ,r-direction through the quantum-point contacts (QPC)
The
are
-0 003
shown in
Fig.
not
6.7. The level

-0 004
spectrum
oscillator with
The
a
is
harmonic
associated
x
500
1000
spectrum
two
(nm)
parabolic
confinement.
are
near
first
eigenstates
and
a
Figure
6.7: Cut
through
the
single-particle
the
non-degenerate
level spectrum
and the conduction-band
degeneracy
occurs
only
with the
edge
in the x-direction
along
QPCs.
105
Figure
0 .34; spin-degenerate). The Single-particle levels (\f'j\2, i cut has been taken in the xy-plane 8 nm below the GaAs/AlGaAs interface in the ODEG. The lead channels are situated to the left and the right of each
6.8:
..
individual plot.
plots of the individual wavefunctions labeled 2 and 3 in the xy-plane (Fig. 6.8). The distorted shape of the quantum dot potential is also reflected in the shape of the individual wavefunctions shown in Fig. 6.8. One observes a tendency to
third and the fourth level which
can
also been
seen
in the 2D
quasi-one-dimensional \va\efunction-.stY7r,v. These scars are related to the classical trajectories of particles entering the dot, then bouncing within, and finally exiting. Two pronounced families of scars are visible. The first (SI) is related to the wavefunctions labeled 4, 6, 9. 12, 15, 19, 23, 28, 33 and the second (S2) to 5, 8, 11, 16, 20. 26, 31. The sears of one family have
form
always
the
same
shape
and
position
in the quantum dot, however, the number
of nodes of the wavefunctions increases with
trajectories
of
scars
for S1 and S2
an
are
is
indication that
increasing energy. The classical schematically shown in Fig. 6.9. The occurrence the quantum dot is already acquiring properties
106
Chapter
6.
Simulation
examples
of
disordered
as
ballistic1
structure in which conductance fluctuations
can
be
understood
a
interference of
number of distinct
phase-coherent electrons traversing the dot via classical paths. This is the dominant regime in large,
200) quantum dots.
highly populated (N
>
Figure 6.10 shows the tun neling rates calculated using the
wavefunctions in
sus
the
Fig. 6.8, eigenenergies at a
ver
par
ticular value of the gate
voltage
(V"g
ing
-400 meV). The tunnel
rates
ticular
belonging to one par scar family are linked by

can
dotted lines. It
be
seen
that
especially
which
can
the
rates
related to
same
Figure
the two
6.9:
SI have almost the

be
value
the
Approximate classical paths of wavefunctions scars SI and S2.
explained by
shape
of the wavefunctions which is similar,
leading
to similar values for the
overlap integral
between the lead and the
10
i<r
u
CD
C
05
10u
=3
10
-0,002
0.000
0.002
Energy (eV)
Figure 6.10: Funneling

Fig. 6.8
to
rates, i.
e.
elastic
couplings of the wavefunctions
in
single-particle eigenenergies. Some of the Phe dotted states with the strongest coupling are labeled for comparison. lines link the rates which belong to one of the scar families (SI and S2).
versus
die leads
the
On ballistic
structures
is
the elastic
case m
mean
fiee
path
/ of the
particles
exceeds the structure size.

mean
This, however,
part
can
al\\a\s the
microns.
smalt quantum dots at low temperatures where the
Iree
be
few
107
Q)
-Q
LU
-0 2 Gate
-0
voltage
(V)
Gate
voltage
(V)
Figure
two
6.11:
Electron
charging
(dotted line) and conductance G (solid
line) in units of the conductance quantum
e2/h
versus
the gate
voltage Vg
at
different temperatures.
quantum dot wavefunction. The highest values, however,

wavefunctions which
area,
are
are
related to the
more
uniformly
distributed
over
the quantum dot

states
showing
some
alignment along
the .r-axis (see for
example
18, 24,
29 and 35).
charging of the quantum dot and the con ductance versus the gate voltage. The envelope of the conductance peaks ex hibits some modulation which is more pronounced at the lower temperature 0.7 K). Over the given range the quantum dot is charged, successively (71 increasing the number of electrons from 49 to 66. One of the main reasons for envelope modulation is that at finite temperature more channels are ther 0 K, i. e. when A7 is in the middle of a shell, mally accessible then at T more channels are available for charge transport than if N is just entering or leaving the shell. In Fig. 6.11 a pronounced step in the peak height occurs at Vg 100 mV. On the other side, the conductance peaks show an over all tendency to linearly increase their peak height with an increasing number of electrons in the dot. This can be explained by the stronger coupling of the
Figure 6.11
shows the electron
quantum dot
states to the leads because of the extended size
of the ODEG. Also
visible is the thermal
dependence
of the conductance minimum between the
peaks.
In agreement with the
conductance
At this
at
predictions of Section 3.4.2 the minimum of the peaks approaches the maximum if the temperature is increased.
one notes
point,
that the values of the conductance is very low
even
the conductance
peaks
which have
heights
of
few 10
~5
c2/h.
Indeed, the
108
Chapter
6.
Simulation
examples
-0 6
-0 4
-0 2
Gate
voltage
(V)
Gate
voltage VQ (V)
Figure line) in
two
6.12:
units
Electron
charging
N (dotted line) and conductance G
(solid
at
of the
conductance quantum
c~/h
versus
the gate
voltage Vg
different temperatures.
noise level in conductance measurements

is to be found
at
~
experiments
of this structure |72]

are
10"
e2/h
which
means
that above results
below the
range which
can
be accessed in standard measurements.
Therefore, the doping

a
density
in the
quantum dot
doping sheet was changed and was adjusted to values around A'
the number of electrons in the

=
85 at
negative gate
reason
bias of
Tg
-800 mV. Now, the conductance
peaks
=
exhibit values of the order of
the conductance quantum which is
Gq
A2//?.
The
for this is the with
already
more
mentioned lateral extension of the ODEG if the dot is

The electron-wavcfunction
getting filled
coupling to the leads becomes stronger resulting in higher tunneling rates. Fig. 6.13. The overall values of the tunneling rates are some four magnitudes larger then in the previous case. This corresponds to the experimental observations [72] which show the onset
electrons.
of the oscillations at 1A
=
1000 mV with around N

starts to
at
80 electrons in the
at
quantum dot. The 2DEG below the gates
get depleted
negative gate
-
voltages
at
one
of around 200 mV. Above this value
around -200
...
100 mV
the control gate, the dot starts to extend below the gate.
Experimentally,
significant change in the spacing of the Coulomb-blockade peaks. The spacing becomes smaller because of the much larger capacitance of the extended quantum dot. In the simulations, the region where the electron
observes
a
gas is modeled
as
ODEG has to be extended below the control gate if this
happens.
The calculations here
presented explicitly
exclude this
case
and the
are
conductance calculation
only includes the
range where the gate
regions
fully depleted.
of
a
However, the
beginning
breakdown is
always
detected because
sudden change in the conductance characteristics.
6.2.
Device simulation
109
10
10
10
S1
24
I9 %A
34
40 A
45 52 47 4
59
^
.
Ai*,
A
2
CD
C
106
I
=3
104
t-
10
-0.002
0.000
0.002
0.004
Energy (eV)
Figure
to
rates
6.13:
the leads
which
Punneling rates, i. e. elastic couplings of the dot wavefunctions versus the single-particle eigenenergies. Phe dotted lines link the belong to one of the scar families (SI and S2).
The modulations visible in both

resonant
a
curves
in
Fig.
The
6.12
are
due to coherentas
transport through the quantum dot.

the electrons
are
quantum dot is acting
resonator and
a
reflected
by
the quantum
point
contacts
similar to
Fabry-Perot
interferometer.
The
period
A
of the oscillations is
roughly
can
300 mV. The oscillations attributed to shell fill

reasons.
not be
ing
at
effects for two

is
First,
N
N + 1
shell-filling
mainly
observable
dom
Ar1
low temperatures,
usually
of mK.
inating
the conductance character

a
istics at
few
tens
Sec
Vg
ond, both the upper and the lower
envelope
of the
a
peaks
are
modu
feature
lated which is
significant
also present in the measurements
172] and clearly identified

to
as
due A
A
coherent-resonant transport.
quantum-mechanical interpretation can be given considering the tunnel

ing
rates
for different wavefunction It has
Figure 6.14: Schematic of the chemical potential as function of'Vg. Phe Fermi
level in the leads,
symmetries.
already
been
ar-
E\?,
is indicated.
110
Chapter
6.
Simulation
examples
gued
that those states which

some
are more
situated in the center of the quantum
dot, showing
observed for
alignment
with the .^-direction contribute the most to the

rates
transport by having the highest tunneling
Vp. Strong coupling
can
be
(Fig. 6.13). If one takes the two-fold spin-degeneracy of the states into account, a period for the envelope modulation of around 10 peaks should be visible in Fig. 6.12. In both curves, which are taken at different temperatures, modulations of roughly this period can be observed.
Figure
and the
example
for the states labeled 24, 29, and 34
6.14 shows
schematic of the
quantum-dot
as
=
electron number N
quantum-dot
to the
chemical
orthodox
potential
theory
at
function of the gate

0 K. The chemical
Vg according
approaches
N
voltage potential
as a
the Fermi level, The
Ap, in the leads
at the
step from electron number

can
=
1 to N.
measure
for the
occurring step in the chemical potential self-capacitance of the quantum dot, i. e. Ap,
serve
e2/6\;.
Figure 6.15 shows the gate-voltage dependence of the single-particle eigenvalue spectrum close to the quantum-dot chemical potential //,. The sawtooth-like structure reflects the changes in the electrostatic and the chem ical potential at the electron charging steps. The tilling of a level pushes the energies of the already filled levels energetically lower. This requires addi2.2
56
> x
o) CD c
z
*
A'
/ \
CD
E
C
2.0
z-ry\iXXfX:AX\Xv
J
\'TT
.;
52
o
o
ULI
v/AAAAW
XA
:'\A
AA'T A
_CD
1.8
-0.8
-0.7
-0.6
48 -0.5
Gate
voltage
(V)
Figure
like
6.15:
dot chemical
curve
=
Single-particle eigenvalue spectrum (dotted lines) and quantum 0.8 K. Phe step potential (solid line) versus gate voltage at T
is the electron number in the dot. Phe Fermi energy in the leads is
occurs
FJ-p
0 eV, Phe
crosses
occupation of a previously empty level the chemical potential.
when the level
111
6.2. Device
simulation
effect is related increased. Another at some of the levels are degenerate 6.16). increasing gate voltage (Fig.
at some value
the charging tional energy and of the levels. Some

to
energy is
the
degeneracies
Vg
are
values of
but the
Other levels
This non-degenerate again. filling of states eventually becoming potential due to subsequent the quantum-dot in in the in the shape of to oscillations 6.8). This leads (see Fig. confinement, i. e. with different symmetry
degenerate non-degenerate, becoming
degeneracy
lifts with
of VA and
effect is related
to
changes
proportions
ible
are
of the
spatial dimensions
are
space-symmetries of the potential also change with the gate voltage. which
sizes shells of various
established and broken again.
of the
quantum-dot
Also vis
Figure 6.16:
age
over
at T the whole voltage range
eigenvalue spectrum Complete single-particle

=
versus
gate volt
0.8 K,
112
Chapter
6.
Simulation
examples
Gate
voltage
Vg (V)
Figure
6.17:
Gate
capacitance (A from numerically differentiating the

at
Coulomb staircase
N{VXj
T1
0.8 K in
Fig, 6.12.
The gate
to the
capacitance (A,
i.
e.
the
as
capacitive coupling
of the control gate
quantum dot may be defined

ri
Q
_
di;
where
(6.3)
Q is the quantum dot charge. Therefore, the gate capacitance is calcu lated by numerically differentiating the Coulomb-staircase. Fig. 6.17. Another method to evaluate the gate capacitance is via the spacing of the Coulombblockade peaks. The difference AA between the ATh and the (A7 A l)th peak is used and the gate capacitance is calculated as
in
XV
VJN + V.
UN)
give
(6.4)
similar results.
over
The
capacitance
is
now a
function of A7.
Both methods
to a
The gate
capacitance
increases from 5 aF
value of around 7 aF
are
the
range of the conductance calculations, lliese values

with the measured value which is
=
compared 32 aF [72]. Consequently, the spacing Cg of the conductance peaks is to wide compared \\ ith the measured results. A discussion of possible reasons for this discrepancy is given in the next section. If one now considers a factor 32 aF/5 aF 6 by which the calculated spectra
to
~
small if
are
stretched in
comparison
to
the measured ones, these results
correspond
with each other.
capacitances is only meaningful for conductors, whose electrostatic potential can be specified by one number, the voltage. The con cept of self-capacitances for a semiconductor quantum dot whose dimensions
The concept of
113
LL CO
cd
c
03
-*
'o
CVS co
o
<D
-I.
CD
87
93
96
99
102
Electron number N
Figure
the dot
6.18: Gate
at
capacitance CA
as a
function of the
electron
charge N
at
T- 0.8 K.
comparable with the screening length is therefore a somewhat delicate issue. One possible and still meaningful definition is based on differential ca pacitances 179], i. e. a self-capacitance Cdd is defined as the amount of work, AV, to bring fixed amount of charge, AQ, to the quantum dot. For AV one
are
uses
cAV
-=-
p{X
AN)
of the
p(N
For
(6.5)
where
p(N)
is the chemical
potential
AAparticle system.
is AN
charging the obvious choice for /Y Therefore, the capacitance can be written as
electron
1 and
singlee. AQ
--=
Gdd(X
The quantum dot
//(-V+l)- pfN),
calculated
(6.6)
self-capacitance C^N) (Fig. 6.19)
was
using
potential was extracted from the free energy minimization. The zig-zag shape of the ca pacitance curve is due to even-odd level tilling of the spin-degenerate singleparticle levels in the quantum dot (see also Eq. 2.21 in Chapter 2). This effect has previously been reported in self-consistent structure calculations of semi
conductor quantum dots
the self-consistent structure results. The quantum dot chemical
128).
a
Another method to obtain tioned here. A
quantum dot self-capacitance should be

can
men
self-capacitance C'dd
be calculated
by numerically
differ-
114
Chapter
6.
Simulation
examples
CD
S
Q
CO
CL
CO
75
en
87
90
93
96
99
102
105
108
Electron number N
Figure 6.19: Quantum

number N at the dot
at
dot
self-capacitance Caa
0.8 K.
as a
function of the
electron
entiating
the Coulomb-staircase with respect to the chemical
potential,
i.
e.
QciOg)
The
du
dp
(6.7)
capacitance is now a function of the gate voltage Vr (Fig. 6.20). Striking features in the capacitance curve are the two singularities which result from the two extrema in the quantum-dot chemical potential left and right from the conductance peak. Also, the self-capacitance is negative between the two singularities in the range of the gate voltages where Coulomb blockade is lifted and the quantum dot charge is shifted from N to N -f 1. The negative capacitances reflect the highly unstable regime during the charging of the dot. The dot is free to oscillate between two groundstate energies which correspond to the two electron numbers. Its most stable configuration is acquired between the blockade peaks, where the electron number is integer at low temperatures. In this regime the chemical potential is almost linear and consequently the self-capacitance is almost constant. There, the value of the self-capacitance is around tOO 150 aF which corresponds to the classical electrostatic limit
. ..
(for
the GaAs quantum dot In the classical limit for
13, d
200
nm.
==
27rtQtd
150 aF).
lic
properties
the
large, highly populated quantum self-capacitance is much higher then the
dots with metal

values observed
here. The consequence is that the steps in the chemical
they are proportional to C^1 and the chemical Consequently, the capacitance in the metallic regime
the steps in the
potential vanish since potential becomes a constant.

is constant
as
well since
charging
characteristics of the quantum dot vanish and the in-
6.3.
Comparison
with measurements
115
89.5
z
'
-"<(
""'
'
'
>
-Q
89.0 88.5
-
2.1
CD
E
ZJ c c
cc/
~"
2.0
o Q.
o
o
88.0
87.5
E
CD
1
CD
LU
, i
1.9
' r
i
-
T
0
o 0
CT
O
CD
O c
0.5
A
:
o
A
AXStXl>
A
s?
-x
xx V.
0.0
"
B
5
CO
Q.
-0.5
CO
^
j
-1.0
p
-o
-
O 0 0
^
c
-
-,
a
1
,->
-0 80
-0.78
0.76
-0 74
Gate volta geva
(V)
Figure 6.20: Quantum dot self-capacitance G'dd 0.8 K. voltage Vg at T

=
as
function of the gate
problems of the negative capacitances and makes them a unique feature of single-electron charging and Coulomb blockade invisible in the classical (metallic) regime.
crease
of the island
charge
is linear. This resolves the
6.3
Comparison
with measurements
The measured
linear-response
at constant
conductance
as
function of the
voltage Vc for
gates
were
the simulated structure is shown in
Fig. 6.21
control-gate [72]. All other
kept
voltages defining
The conductance shows dramatic
operating point of the device. oscillations of two orders of magnitude. The

the
peak height is of the order of the conductance quantum (c2/h). This measure ment was performed at a very low temperature of 25 mK. The number of
electrons in the quantum dot
ulated results shown in
was
estimated to be N
in the
75
...
250. The sim

a
Fig. 6.12
previous
section
...
are
obtained with
as
quantum dot with
an
electron number of N
87
110 and
such the sit-
116
Chapter
6.
Simulation
examples
-650
-680
-710
-740
-770
Vc (mV)
Figure 6.21: Coulomb-blockade oscillations measured as a function of the control-gate voltage \f. All other voltages are kept constant. (Paken from
f72]. Courtesy
P
Heinzel.)
uation should
closely
to
resemble the measured
one.
iTie main difference to the
experimental
This
results is the much
larger period of the
conductance oscillations.
corresponds
the much lower simulated gate
capacitance
at the
of 5
...
7 aF
compared
this
to the measured one which is
around 32 aF. A
possible
reason
for
exposed parts a large generally speaking, exposed influence on the general electrostatics of the device, also influencing the cou of all other gates amongst each other and the pling coupling of the gates to the substrate. This is especially evident for narrow split-gate structures as the one
of the GaAs surface. More
the surface has used here
discrepancy
is the
assumption
of Fermi-level
pinning
[75].
Figure 6.22 shows the conductance oscillations over a much wider range of gate voltages and for several temperatures. One observes pronounced enve modulations of the conductance peaks. The period of these modulations lope
is around 50
..
.80 mV. The simulated data
(Fig. 6.12)
is
show
similar mod
as
ulation of the
peak height, however,
the
period
again larger
e.
with the
measured data. Both the measured and the simulated modulations
are
a
due to
coherent-resonant transport between the two QPCs, i. Perot like behavior of the quantum dot, acting as the simulations could not be
because of
Even
cavity.
Fabrythough the
performed
for temperatures lower then 800 mK. the
6.3.
Comparison
with measurements
117
600 mK
3
JZ
350 mK
3
n
I
220 mK
100 mK
25 mK
iini.uiuih
.^
-200
-40O
-600 V.
-800
A 000
(mV)
Figure 6.22: Coulomb-blockade oscillations measured as a function of the control-gate voltage V at various temperatures. Phe conductance curves exhibit strong modulations due to coherent-resonant transport. (Paken from
[72 f. Courtesy
P. Heinzel.)
peaks for different temperatures was observed. In agreement with the theoretical predictions the measurements show a smaller amplitude for the conductance oscillations and a general trend to higher conductance values between the peaks. This trend is confirmed by
same
trend in the
height
of the conductance
the simulations
surements, the
(Fig. 6.11 and Fig. 6.12). Also in agreement envelope modulations become less pronounced.
with the
mea
"~"Tbfw,M
A-
>
H
c
I-v1
s
f
'n\ As A>
l'c
ix"A
'
A "X
t l
X "K
Chapter
Z^
come
7
eon
quel
ehe
lena
affamuita
rix a
itsato
si
fuor dal
pelago
alia
volge all'aeqnaperigliosa
as a
guaia
...
Just
now
swimmer, still with
safe upon the

turn
might
for
one
panting breath, shore, out of the deep last look at the dangerous waters,...
Dan if, Hnferno, 1
Achievements, failures and
prospects
This work describes the theoretical foundations and the numerical
used in
tor
a a
algorithms
simulation software for
has been
implemented
and its
single-electron capabilities are tested at
devices. A prototype simula

a
GaAs/AlGaAs heterostructure SET.
very limited because of its restriction
particular example, To some extend, the approach chosen is to the quasi-equilibrium. Nevertheless,
on
in the context of device simulation which relies
the drift-diffusion equa
tions, inappropriate
to
simulate
single-electron tunneling,
this work
means
considerable progress.
especially emphasized. First, the inclusion of size quantization and many-body effects in the solution of a non-linear Pois son equation and the implementation of stable numerical algorithms has been achieved. This allows for a self-consistent evaluation of single-electron tun neling. Second, a sound method for the calculation of the tunneling rates for
Some aspects should be
soft electrostatic barriers which is based
on
the transfer-Flamiltonian method is

a
included.
Finally,
these results
are
used in
free-energy
minimization proce-
119
120
Chapter
7.
Achievements,
failures and
prospects
dure which allows for the
computation of the discrete charging of the quantum dot and the linear-response characteristics of a SET at finite temperatures and for realistic geometries.
The
software, for which the preliminary

of
name
SIMNAD
(Simulation
and
Modeling
Nanodevices) is used, is embedded in the TCAD suite of TSE AG
(Zrich). This allows for the utilization of the TSE grid editor TED ise which is especially suitable for tensor-product grids. Furthermore, the visualization
of results
can
be done
output files
are
by graphical software such as INSPECT compatible with the ISE datastructures.

of the SIMNAD simulation software
some even
ise
since all
Since the in
development
was
initialized
1997,
at
lot of structures,
with chances to find device

more seems
appli
cations, have been
presented.
It
more
and
to be
clear that any
thing
on
that could be of
practical
interest for
silicon. Simulation of silicon
logical circuits should be based single-electron transistors is well within the quantized
levels for
a
scope of the SIMNAD software. The calculation of the
six-valley
entation.
rather
bandstructure is
1DEG, and
possible in all three confinement models (ODEG, 2DEG) implemented in SIMNAD and for a (100)-substrate ori
most
However, silicon SETs,

cause severe
of them in
that
complicated
three-dimensional numerical
geometries
the
a
SOI-technology, have require huge grids and
sometimes
problems. Also,
the simulation of silicon
quantum dots is often
not
possible using
that allows for
single-subband approximation speed-up

of simula
(Eq. (4.38)
and
Eq. (4.39))
considerable
tions of GaAs/AlGaAs heterostructure SETs.
Apart
gases
are
from silicon, GaAs/AlGaAs based heterostructures
are
still first
choice when
properties
of confined electrons in zero-dimensional electron
investigated. The simulation of these structures can be carried out in a numerically stable and reasonably fast manner. However, there is still need for further investigation of the role of the exposed surface and the dop ing sheet including the doping levels in order to become really predictable. It
should also be mentioned that the numerical
to
algorithms implemented
a
still fail
converge
at
very low temperatures below
few hundred mK. This
regime,
however, is experimentally of particular interest. At higher temperatures, ef

fects related to the
single-particle spectrum
are
often
disguised by
the thermal
broadening
of the levels.
Further work should start
Certainly,
ware
the direction of
by dealing with the shortcomings of this work. further work depends on what the simulation soft
will be used for. If a TCAD-oriented tool is
planned,
some
work has to be
121
devoted to goes
a more
advanced discretization scheme at unstructured
grids
which
simple tensor-product/finite-difference approach used in the present software. The use of automatic meshing, however, requires the formu lation of adequate meshing criteria, especially suitable for the Schrdingerbeyond
the Poisson system. At this consistent scheme for
box discretization for
point the author has no knowledge that a stable, selfthe Schrdinger-Poisson equation system employing a instance on a general, unstructured grid has ever been
implemented.
On the since the
ture
physics side, the inclusion of magnetic application of magnetic fields reveals a lot
magnetic
fields would be desirable

about the electronic struc
of the electronic gases in quantum dots. A whole literature of the discretization of the vector
experimen
tal data related to the effects of
fields in quantum dots is available.
Some work
on
potential
course
at
tensor-product grids
using finite differences has been clone in the test implementation using the object-oriented
is
of this
project [80].
program structure of SIMNAD
already
available.
The SIMNAD code has the tion

one
providing the advanced applica oriented research in academia as well as in industry with more insight into of the most promising branches of modern device physics. The singlecapability
for
electron concept and
single-electron devices
are
certainly beyond the they

may
ultimate
scaling
timate
limits of the conventional CMOS, however,
replacement. Computer
not
simulations and
proof to be its ul complex TCAD experiments

Goal of this work
was cer
will be of
tainly
key importance with this technology. to clarify the question whether or not
the
to
single-electron concept
expect that the basic
re or
will have any
applications.
today
will
It is
search undertaken
only eventually
reasonable
find
use
in further microelectronic-
nanoelectronic devices. Some of the effects discussed in this work may

a more
play
Oth
significant
only
to be
role for conventional
devices, intentionally
or
not.
ers
may
have the
potential
have
for niche
an
electron devices will

to
never
applications. However, that singleimpact on the industry is certainly a verdict

seems
early
proclaimed.
be
The current status of nanodevices

was no reason
somewhat
vacuum
like the situation in the 1940s when there
to
believe that
tubes could
ever
displaced.
A/,i
,v,j
a-
Appendix
of localized systems
Grand canonical systems at

energy
temperature T
a
arc
characterized
a
by
Fermi
E-p,
particle strictly
e.
number A\
volume
A,
and
total energy E. In
equi
is
librium the volume V
ly. the particle

constant as
number N
Ar0, and the
energy
Eq
are
constant.
The entropy of the system

well.
A0
="
^B^'htô
maximal, therefore being
Yq
is the dimension of the classical
phase
space, i.
gives
the number of accessible states of the system. is subdivided into

a
The system
(Fig. A.I)
reservoir R with
set of thermo
dynamic by A", V,
variables N'.
Vf E', Vf S' and
small
subsystem
A characterized
subsystem .4 is embedded in the much larger reservoir R and To, Aq, and PJq are properties of the system as a whole (A A i?).The subsystem exchanges energy and particles with the reservoir. The particle number in the reservoir changes according to the law of equilibrium changes
E, F, S. The
E
'
=
PdS
'
-
PdV
'
EFdX '.
is
so
(A. 1)
small that T
fluctuates
The
and
important assumption
Ep of
is made that the

even
subsystem A
though
the
the reservoir remain fixed

are
subsystem
among its that
possible states which En <A Eq and Ara <<T A't).
labeled
by
a.
It is furthermore assumed
states
can
Even if
some
of the
the above condition, it is
only
small number which
|a) may violate safely be ignored.
124
Appendix
A.
...
Using Eq. (A.I),

spect
to
one
obtains for
re
the constant derivatives with
entropy S' and the par ticle number N'
OE'
OS'
=T
OF/'
ON'
Within
this model the
(A.2)
Gibbs
NA\
distribution for the
occupation numbers of the small subsys tem in particle exchange with

the reservoir
can
Figure A.1:
Reservoir-
(R) subsystem (A)
be derived. The
states in
model and thermodynamic variables.
number of
possible
=
A is
denoted F and that in R is denoted Y'. Therefore, the total number of states
in A A R is IT'
Ft.
=
In
equilibrium
the entropy is maximal,

as a
Ft
Fo and
in
a
St
S X S'
Sq.
The
probability
one
for the system
whole of
being
particular
state
is
Peq
1/Fq.
in 1
The number of choices for the reservoir states
given the subsystem is r< ]\T'a and Ya probability of this state
particular
as
state
|cv)
(since the system is in

be written
state
\Na) En) is Y'a. With ja) by assumption), the
can
ZJCL
to
i.
e.
(A3)
out
of
F0 possible
states
of the total system, there
are
Y'a
states that
find
Ain state
\a). S'a
=
The reservoir entropy
kj^PliiY/ depends
on
the energy and the
particle
number in the
subsystem
Sa
and with
(E0
Ea, A'o
Art),
(A.4)
Sq
k-T hi Yq
it follows that
p/
Sq-S/
It has to be
-,
-kB\n-
t o
-kBlnPa.
(A-5)
emphasized
that
S{)
Sa
is not the entropy of the
subsystem
A.
The entropy of the
subsystem
is
zero,
since A is in
given
state
la).
The
125
equilibrium entropy
of A A R is the average of
a
S'a
over
all
possible
states.
The thermodynamic average of
quantity f
with
is
^^
^^
=
Y,rf<
a
=
Ep
a
1<
(A.6)
which for the entropy S
So
S'a
leads to
=
-A'b
5Z P In Pa,
a
PQ
exp
\-kB (SQ
one
ni
-
'
a.
(A.7)
With the law of

ryl
equilibrium changes, Eq. (A.I),

Ol /E1 o
TP
writes
AT
\T\
ba
(iTo
-/ifj. i\0
AcY)
(Pq^ Aq)
E.
a
-~
\<J
,
y jy/
()Ax
Nn
VAE'
const
piya
N,
(A. 8)
the Gibbs distribution
p
_.
which substituted into
Eq (A.7) gi\es
aj
Pa
pyA
exp
A'bP
kT
(A.9)
using
of the The
the normalization condition for the
Pa and Z is the partition function
grand
canonical ensemble,
i. e.
the
grand
canonical
partition function.
thermodynamic averages N N of the subsystem A Eq. (A.6) giving
of the energy E
are
E and the
particle
number
calculated
using
the Gibbs distribution in
J2EP
aild
-V"-X>aiV
(A. 10)
Inserting the Gibbs distribution A, Eq. (A.7), one obtains

k Phi
and
the
into the equation for the entropy of the system
TS" -A AV
form of the
--
N Ep
--
0,
(ATI)
0 for
consequently to an explicit subsystem A, i. e.

O
=
-
grand
P,
canonical
potential
AAP
--
A'b F In N
or
rjp
exp
kXF
(A.12)
-*wTlnZ.
(A. 13)
126
Appendix
A.
...
thermodynamic variables at fixed V, T, Ep of the system can now be calcu lated by thermodynamic averaging of P, 5. N if the energies and the electron numbers {Ea. Na} for all states |a) of the subsystem A are known.
All
If the
subsystem
a
contains
fixed number of electrons and
no
particle
ex
change
treated
between the reservoir and the

as
subsystem
occurs, the
system has
to be
canonical ensemble.
fixed,
namic
constant
electron number
keeping the chemical potential N for all jo) is assumed according to Nn

~
Instead of
and the
thermodynamic variables of the system are determined by thermody averaging at a particular N. P. V, The thermodynamic potential is the
Helmholtz
free
energy
F(N)
=
F{N)
-kBTTnVexp
/-V
(X
Fl
kT J
(A. 14) (A. 15)
or
F(N)
kBT hi Z{N).
with the canonical
partition function Z(N) and the Gibbs distribution

Pa
Z
x
(AI
Er
exp
kXE
E
^(AT)
5^ exp
kT
(A. 16)
To illustrate the
principle
difference between the Gibbs distribution and the
Fermi-Dirac distribution for

used. Consider
to
a
single-particle
levels, the
with X
two
level system
{Ei,E2}
following example is 1 particles. According

is
Eq. (A. 16)
the
occupation probability
for the level

Eq
given
as
exp
Pa
P{En\N=\)^~
exp
krtT
exp
(A. 17)
If cast into
slightly
different form, this results in
P
1
Fj
=
1 A exp
f\
Fj~^f
kBT*
w;rith
Ep
a
-(Pi
P2)
and
T*
~T,
a
(A. 18)
fictitious tem
which
corresponds
to
Fermi-Dirac distribution, however, at

true
perature PA which is half the
temperature of the electronic system. The
127
Fermi energy
Ep
can
be defined
as
the average of the two
energies P\
and
P2.
For
higher
electron numbers and the number of levels
approaching
infin
ity,
the Gibbs distribution becomes
equivalent
to
the Fermi-Dirac distribution.

can
Numerical
be found in
analysis confirming
[31].
this fact for semiconductor quantum dots
'
\
>
I
-
V>A-
^
v.^
>
^
v #x,
-j
Appendix B
JL JL
Transition rates
Subject
of this
appendix
are
the transition rates between the

on
many-particle
groundstates in the quantum dot. The calculation is based perturbation theory in lowest order in the tunneling terms.
time-dependent
B.l

by
the
The evolution of the Hamiltonian operator H with time is described
time-dependent Schrdinger equation
A?.î4/(f))
Of
P|iF(f)),
of the system.
(B.l)
The time-
where
[^(r))
is the
time-dependent
an
state vector
evolution of the state vector from

can
initial state | ty
(/, ))
to
final state
\fY(tf))
be written
as
\^(tf))û(if.t/)fy(t/)).
with
a
(B.2)
time propagator U of the form

rtf
U(ttJ,)=exv
Jt,
dtH(f
(B.3)
129
130
Appendix
B. Transition rates
which reduces to
U(tfX,)
if the Hamiltonian operator is
exp
--n{H
tt
(B.4)
explicitly time-dependent. The propagator is usually evaluated via expansion of the exponential function in powers of the argument. The direct evaluation of the time propagator is in general a very difficult, if not impossible task. A very common procedure is to split the Hamiltonian into an time-independent part %q for which the stationary
not
Schrdinger equation
,(A
%oW\
E]
AuAO
\yy
can
be solved
exactly
and
an
time-dependent part 'Hpif)
H=--Ho-{-'Hp(t).
The
(B.5)
time-dependent part is treated as weak perturbation to the system. The time propagator is expanded in terms of 'Hp in the basis of eigenstates of PLq. The total propagator is given as a series
'
U(ifztl)=U0(tf.t1)xJ2(
where
_L
iXAtf-fc),
(B.6)
Uq is the propagator of the unperturbed system
U0(tf.tj)
exp
-jUo{if
-i,
(B.l)
andZP the /?th order correction due

Xf
An
to
the presence of the

Xj,
perturbation
U [tf,f
dtn
dtu
dUUo{tf,tn)Hp(tn)
\Uo(taJa-L)'Hp{tn-i)...Hp(tl)Uo(tl,t,).
A small
(B.8)
perturbation
of the system due to

are
Hp
can
be assumed if the matrix
elements of the operator

to evolve in
small. If, furthermore, the system is considered
'short-time'
periods
,^l
then
(0)
(v/;,\nP\v[//,)(tfî1)/hxY
(B.9)
the lowest terms in
n
Consequently,
the series converges \ery fast and
only
a
need to be considered. Since
tunneling
is
always
isolated quantum dot, i.
e.
the
tunneling
terms are
perturbation to the always small compared to
small
B.l.
131
the other energy scales in the system,
only the
first order term in the
expansion
ofU in
terms
does not
t
tf
perturbation PLP needs to be considered. The perturbation explicitly depend on time, it rather depends on the time difference t7. The probability amplitude. A(l), for one tunneling event is
of the
defined
as
,(0)>
-^> Itf40)>
f
AfMîvfMt)^).
(B.10)
the trivial
The
unperturbed operator (Oth order perturbation) generates only time dependence but does not allow for transitions
ef Tf))
However,
one
exp
-X''
()
Alu'(0)).
(B.il)
is interested in transitions between different

?
many-particle
the
eigenstates
where
/ /.
In this case, the lowest
perturbation gives
proba
bility amplitude
Tk .(0)
1
III
AY{t)
(ilf'\^
Jq
di1U0(t-ti)HplMGM[r),
,SX
(B.12)
which becomes after
integrating
l
over
.(i),
Ay/(t)^Mf,t
E(0)
.(0)
exp
l-Eyt)
exp (
-E;'t
(B.l 3)
with
where
Mft1
^(^f\flp\i/0)).
as
Mf)t
is the
perturbation
calculated
the matrix element of the
eigen-
functions of the this
unperturbed
Hamiltonian
(0, Hq. The probability W\ /(f) that
perturbation has caused the transition is \A\ (t)\2, the absolute square of the probability amplitude. This leads to the final form of the transition probability
IF
At)
A
it)
=
\M A
(lA2)sin2[(P|0)^PW)4
l
(B.14)
XX
(E,
-E{
XX\
)m
Assuming 'long enough"

to the transition rate y
times reflected in the
uncertainty
the
relation between
limes leads
time and the energy difference /? AT
(Pf7
as
E) )f,
following
which is defined
It
<
lim
t -+~o
W(f^(t)F
(B.15)
132
Appendix
B.
Transition rates
and therefore
ta
2tt,,. ,.,,(0) ' v /,t| zrliJAl"^W; > h
,,(())
-
#D-
(B-16)
B.2
The
the
expression
a
derived above is
now
used to calculate the transition rates for
many-particle
dot states. The Hamiltonian of the
unperturbed system
one
con
sists of
contribution due to the quantum dot and
contribution for the
source-side reservoir and the drain-side reservoir
respectively
(B.17)
for
Po
The
PdotAP,4 Pa.
eigenvalues
are
Efot,EJ. Ef
and the
each operator the
solutions to the
eigenfunctions j</'fot), [0f), \iftf) time-independent Schrdinger equation
for
and
particular subsystems.
The total energy is

as
P)
Pfot
Pf
The
Pf
the total wavefunction lac tori zes
j*,") )
Ivf01)!^))^1)to the
tunneling
Hamiltonian which acts
as a
small
perturbation
system is
+
np
where
one term
n\
ntd
the
with
nt/d
Hl'-d
H\>/%
(b.is)
(-) describes
tunneling
out
of the dot
Ad=-EAAk>x>
k
.m
cB.i9i
and the term
(+) describes
the
tunneling
into the dot
summing
over
all dot states and lead states. Therefore, it is assumed that tran
occur
sitions between the dot states
solely
are
if
an
electron is added
or
removed
from the dot. Transitions which
keep
the number of electrons fixed and could
be due to
electron-phonon
interactions
neglected
here.
Transitions which reduce the number of electrons in the quantum dot are transitions that move one electron from the dot to the source or drain reser voir.
Inserting
the
tunneling
Hamiltonian from
Eqs. (B.20)
and
Eq. (B.19)
B.2.
133
into
Eq. (B. 13)
leads to
"
<,"
E A'A"fi<A<AA/d,(.c A'XwAx}
k,m
with the matrix elements

D
JJf,i.m
.s/d
-(rdX[\r
"~AC f
s s
xnlX
ddii
+
f
A'doA /
|
,
Ad
^Wl/,X
,d's,
_,,
vd\
=^7"A'/
/
lA/d,AlVTK/d^lV'/
,d
'a
The
or on
=^7
a
expression
for
Mb//
'
contains
product
d k
of
an
operator acting
on
the left
right reservoir
Hubert space
only (A
) and
an
operator acting exclusively

are
the dot Hilbert space

rate.
for the transition

over
all
possible
final
Eq. (B.l 6) The effective transition rate is calculated by summing states in the reservoirs which can be filled leaving an
(cm ). The matrix elements
inserted into
initial dot state empty

s,d
The
following
relations
are
used
(i)Xc \i'X ) =5,

(W \X/dk'cl/dk\vf
/,/,dott
A
i
(completeness
of
\ifû ))
)=[f-,/s/d(PA
(
-EF' )\dkV
,,
li',dot\
(j?dot\c
|CmKm|l f
) Jd\JXm
)0m m',
where
/dot
is the
occupation
distribution function for the quantum dot and

and the
/s/d
is the Fermi-Dirac distribution function for the reservoirs which contains
the reservoir chemical
potential
temperature. Since the initial and the

the electron number AA the
final state in the dot

notation is
are
characterized
by
following
-
adopted
for the transition rate IF,
TF?'~
Ws/rf(A/", iV
1).
Finally,
one
arrives at the
expression
-
Wa'd{N,N
with
1)
XXd^l/d)El- .fs/d(P;/d
m
-
<d)]/do<(0
4
17?;d(PAq d)
~\Jl
\2{IX}ot
Pfot
Ebk/d)
(B.23)
134
Appendix
B.
Transition rates
being
state
the
tunneling
rate for the transition from the rnth dot state to the
~~~
kth
in
in the reservoir. The delta function selects the wavevector k
k(E)
the reservoir and the energy of the

The
crease
same
tunneling
similar
electron is
Pdo1
Efoi.
transitions
procedure
leads to
equation
for transitions which in

are
the number of electrons in the quantum dot, i.e. which electrons from the
source or
that
move
drain reservoir into the dot
W8'd{N, N
1)
X>-dÂ d).fsXi(iA/d
???
"
#f
d)H
Aiot(Pdot)J
(B.24)
The results of Eqs.
(B.23, B.24)
are
known in the literature
as
Fermi's Golden
rule.
Appendix
many-particle problem of interacting electrons in an external potential has been a key problem since the introduction of quantum mechanics in the late 1920s. A first solution in the form of a mean-field theory was provided by the Hartree theory in which the electrons are described by an effective single-particle Schrdinger equation and the effect of the other electrons, the screening, is included via an effective potential, the Hartree potential. While the Hartree theory still failed to include quantum mechanical interactions, the later introduced Hartree-Fock theory expanded the concept and included quan tum mechanical exchange.
The
theory proved to be useful for systems with small num bers of particles. However, the expansion of the many particle wavefunction in Slater-determinants, which even for small systems leads to huge numeri
The Hartree-Fock
problems, made the application to solid-state systems prohibitive [46k An early alternative to Hartree-Fock theory was the Thomas-Fermi theory [47,481 which gave, on a heuristic basis, a description of the physical properties of an
cal electron gas
an
by
use
of
density-functional.
However, it took until 1964 that
formally exact theory of the many-particle problem of a system of inter acting electrons was formulated. In their pioneering paper Flohenberg and
Kohn1 [49J proved
that the is
interacting particles
1998.
ground state properties of an arbitrary system uniquely represented by its charge density.
m
of
1W. Kohn icceivcd the Nobel pn/e
Chemistry
for his woik
on
density-functional theoiy
135
136
Appendix
C.
theory developed by Hohenberg, Kohn and Sham, the density-functional theory (DFT) is given. A more detailed presentation of density-functional concepts can be found in [50k
In the
following
section
short account of the
C.l
Hohenberg-Kohn
theorem
Consider
system of interacting particles described by the many-particle

with the Hamiltonian operator
Schrdinger equation
H
r + VAW
and
P|^)
P|4/)
(C. 1)
where P is the kinetic operator, V the operator of the external
potential, W
the
operator of the electron-electron interaction which,

systems,
can
at least
for semiconductor
assumed to be Coulombic, T' is the many
particle wave-function,
theorem states in is
and P the total energy of the system. The its first part that the external functional of the true
=
Hohenberg-Kohn
=
(T^VlT') potential V (V) ground state charge density2 n

V
=
always
unique
V[n]
X const.
(C.2)
In is
a
a
(total) energy functional PVN is postulated which unique functional of the charge density as well and which, assuming
second statement
a
charge conservation, can be minimized giving the total energy of the system, Ey Ev[n] and Pyjn] < Ey[n') if n' is not the ground state density.
Therefore, the ground
potential and the many-particle Hamilton operator and consequently all ground state properties of the electron gas in a unique way. The restriction to the ground state prop
state
density
determines the external
erties, however, is
The
one
of the mayor weaknesses of

are
properties
of excited states
beyond
the
density-functional theory. scope of the basic theory.
The second part of the
calculate the total energy
Hohenberg-Kohn theorem enables one to explicitly of the ground state. Introducing the following form
of the total energy functional
Ev [n]
fdrn{r)Y(r)
was
E[n]
(C.3)
-"Ongmally density-functional theory
only
introduced for electron gases, however, it is
applicable
to
hole gases
as
well.
C.2.
Kohn-Sham
equation
137
where
F[n)
is
universal functional
to
independent
of the external
potential.
It
contains the kinetic contribution
the total energy and the electron-electron
interaction, and therefore is the
same
for all kinds of electronic systems

n(r)'ii(r' J AÂPxc,
cirdP
, i
vxc)
(C.4)
rx
where
Ps is the functional of the
kinetic energy for

term
with the
charge density
so
n,
the second
is the
non-interacting electrons Hartree energy Ph, and the

as
last term is the
called
exchange=
and correlation energy which is defined

+
tixc
The
(T)-Tshi]
{W)-EH.
(C5)
ex
exchange-
and correlation energy contains the difference between the
pectation
value of the kinetic operator in the
many-particle Hamiltonian, (P),
and the energy
Pa of the non-interacting system and the difference between the
expectation
sition of
value of the operator of the electron-electron interaction
(W)
and
the classical Hartree energy
Ph. This definition of Pxc
ensures
the
Eq. (C.4). The two functionals Ts and known from the beginning. Pxr is given within appropriate approximations the local-density approximation (EDA).
to
F[n] according
decompo Pxc are not

as
C.2
Kohn-Sham
equation
the calculation of the
to
The total
charge density. an appropriate representation One now charge density by assuming a system of non-interacting particles with the density nH. According to the Hohenberg-Kohn theorem the total energy functional is given as
the
tries to find
Hohenberg-Kohn theorem provides the basis for energy using a variational principle with respect
of this
Evjju]
where ns
tions
can
=-
Ps[ns]
dr
iu(t)V5(t)
(C.6)
wavefunc
represented by a sum over squared single-particle fy(r) weighted by an occupation factor A",
be
E^.
It
can
(C.7)
be shown
(Kohn-Sham theorem) that for

a
every system of
so
interacting
state
particles
there exists
single particle potential VB
that the
ground
138
Appendix
C.
charge density
system
ns.
/?
of the
For the
interacting system equals that of the non-interacting effective single-particle potential Fy an effective singlethe Kohn-Sham
particle Schrdinger equation,

h2
2m
equation3
=
can
be formulated
\72
_|_
yjr)
4>,(T)
tA(r).
(C8)
interacting particles in an external potential V is now projected onto a system of non-interacting particles in an effective single-particle poten tial Vg. Equation (C.8) can be solved exactly if the effective potential is known. The kinetic energy functional of non-interacting particles can be written as
The system of
Ts
^ V /dr0;(r)V2(r).
zni
l^
(C.9)
Variation of the functional of the total energy
Eq. (C.3)
with respect to the leads to

an ex
charge density
and consideration of the Kohn-Sham
equation
plicit
formulation for the effective
potential
K(r)
V(r)
dr'
r^7T
jr
Vxc(r).
(CIO)
The second term is the Hartree

trostatic
potential formally given by
as
Vu
potential Fr, which is related to the elec q. The exchange-correlation potential is
rxc(iW^#-
(cid
Equations (C.7). (C.8) and (CIO) describe in an unique manner the system of non-interacting particles w hich is related to the system of interacting particles via the identical groundstate charge density. These equations have to be solved self-consistently. Usually, one begins with an assumed n(r), constructs Vs(r) froniEq. (CIO) and Eq. (C.l 1), and finds a new n(r). The Kohn-Sham total energy (ground-state energy) is given by
#M
Hie
X>T'e<
"
.7
/ drn(r)VH(r)
are
xc-
f dr n(r)V^c(r).
e% and the
(C.12)
interpretation
the
of the
single-particle eigenvalues
the solutions of the
with
m
wavefunctions A which
single-particle Kohn-Sham equation remains

of the free election
AVe
restrict to
\acuum
problem
being
the
mass
C.3.
139
problem. They describe a system of non-interacting (quasi-) particles, but no direct physical meaning for the system of interacting particles can be attached to them4. Nevertheless, they are interpreted as single-particle (orbital-) ener gies which sufficiently accurate describe the bandstructure in the crystalline
a
solid-state.
C.3
relatively straightforward
only possible
difficult to
are
The
formulation of the Kohn-Sham formalism of
DFT is
which
many-particle properties of the electron gas describe, especially the exchange- and correlation inter
because all
action between identical functional is not
particles
are
relocated in the functional

as
PXC. This
explicitly
known and,
already mentioned,
one
depends
on
appropri ate approximation s.

The most
widely
used
approximation
is the
local-density approximation
(EDA)
PL1)A=
where 6XC is the
drexc(n)
n
n(r),
(r)
(C.13)
function of the
density (not
functional) which is defined

a
as
exchangeIt
and correlation energy per electron for
homogeneous
elec
tron gas.
locally approximates the density of the in homogeneous electron gas. This approximation is certainly good in systems where the charge den sity is spatially nearly constant. However, LDA proved to be an astonishingly good approximation
even
in systems where the
mogeneous such
as
molecules
(and of
course
charge density is strongly inhoquantum dots). The exehangeis
and correlation contribution
to
the effective
potential
given according
to
Eq. (Cll)
Ac()
n
(C14)
(r)
The function exc has been determined for

gas
by Ceperly and Alder [521 using a work a parameterized form for the three-dimensional problem was given by Perdew and Zunger [53].
An fact they
Sham effective
are
interacting homogeneous electron Quantum-Monte-Carlo method. In this

an
is
used, which
the
Lagrange-parameters of
the "variational scheme used to derive the Kohn-
potential Eq. (C. 10),
Bibliography
[1] Semiconductor Industry Association (SIA), "The National Technology
Roadmap for
Semiconductors A 1997.
[21 M. A. Kstner. "Artificial atoms". Physics Today, pp. 24-31, January

1993.
[3J
Guo, and S. Y. Chou, "Silicon single-electron quantum-dot transistor switch operating at room temperature", Appl. Phys. Lett., vol. 72,
E.
Zhuang,
L.
pp. 1205-1207,1997.
14] A. Nakajima, T. Futatsugi, K. Kosemura. T. Eukano, and N. Yokoyama,

"Room temperature
aligned floating
15]
E. Guo, E.
dot
operation of Si single-electron memory with selfgate", Appl Phys. Lett., vol. 31, pp. 1742-1744, 1997.
Leobandung. and S. Y. Chou, "A Silicon Single-Electron Tran

at Room
Memory Operating 649-651,1997.

sistor
Temperature", Science,
vol.
275, pp.
[61
J. J.
ature
Weiser, S. Tiwari, S. Rishton. K. Y Fee, and Y Fee, "Room Temper
Operation
of
Quantum-Dot
Flash
Memory",
IEEE Electron Device
Lett., vol. 18, pp. 278-280, 1997.
[7] S. Tiwari, E. Rana, K. Chan. E. Shi. and

pp. 1232-1234, 1996.
Hanafi, "Single charge and confinement effects in nano-crystal memories", Appl. Phys. Lett., vol. 69,
H.
18] P. D. Tougaw, C S. Fent, and W. Porod, "Bistable saturation in coupled quantum-dot cells",,/. Appl. Phys.. vol. 74, pp. 3558-3566, 1993. [9]
W.
Shockley,
Electrons and Holes in Semiconductors, D. Van Nostrand,
Princeton, N.J., 1950.
flO]
F. Stern, "Self-Consistent Results for
/i-Type
Si Inversion
Eayers", Phys.
Rev. B, vol. 5, pp. 4891-4899, 1972.
141
142
Bibliography
fil] A. Wettstein, A. Schenk, A. Scholze, G. Garret on, and W. Fichtner,
"Charge Carrier Quantization Effects in Double-Gated SOT MOSFETs", Meeting Abstracts of Electrochemical Society, vol. 97, pp. 613-621, 1997.
[12] A. Schenk, Advanced Physical Models for Silicon
Device
Simulation,
Springer, Wien,
New
York, 1998.
[13] N. S. Bakhvalov, G. S. Kazacha, K. K. Eikharev, and S. I. Scrduykova,
"Single-electron
-
solitons in one-dimensional tunnel structures", Sov.
Phys.
JFPP, vol. 68. pp. 581-587. 1989.

F. R. C Fonseca, A. N. Korotkov. K. K. Likharev, and A. A. Oclintsov.
[14]
"A numerical
tems", J.
study of the dynamics and statistics of single Appl. Phys., vol. 78, pp. 3238-3251, 1995.
H. Kosina. and S.
electron sys
[15] C Wasshuber,
Selberherr, "SIMON
A Simulator for
on
Single-Electron
Tunnel Devices and Circuits", IEEE Frans,
PCc\D, vol.
16, pp. 937-944, 1997.
[16] C. J. Gorter, "A possible explanation of the
increase of the electric
re
sistance of thin metal films at low temperatures and small field
strengths",
Physica,
vol. 17, pp.
777-780, 1951.
[17] CA. Neugebauer and M. B. Webb, "Electrical conduction mechanism

in ultrathin,
evaporated
metal
films", J. Appl. Phys., vol. 33, pp. 74-82,
1962.
LI8] 1. G. Giaever and

small
H. R. Zeller,
"Tunneling,
vol. 181, pp.
zero-bias anomalies, and
superconductors", Phys. Rev.,
789-799, 1969.
J19] J. Lambe and R.C. Jaklevic, "Charge-quantization studies using tunnel
capacitor", Phys.
[20]
Rev. Lett., vol. 22, pp.
1371-1375, 1969.
I. O. Kulik and R. I. Shekhter. "Kinetic effects in
ness
granular media",
Sov.
phenomena and charge discrete Phys. JETP, vol. 41, pp. 308-316,
-
1975.
[21] D.
V. Averin and K. K. Likharev. "Coulomb blockade of
coherent oscillations in small tunnel
junctions",
./. Low
tunneling, Pemp.Phys.,
and
vol.
62, pp. 345-372, 1986.
[22] T. A. Fulton and G. C Dolan, "Observation of single-electron charging effects in small tunnel junctions", Phys. Rev. Lett., vol. 59, pp. 109-112,
1987.
Bibliography
143
[23] Y Meir, N. S. Wingreen, and P. A. Lee, "Transport through a strongly interacting electron system: Theory of periodic conductance oscillations",
Phys.
[24]
Rev Lett. vol.
66, pp. 3048-3051, 1991. Likharev, "Theory of single-
D. V. Averin, A. N. Korotkov, and K. K.
electron
charging
of quantum wells and dots",
Phys.
Rev. B, vol. 44, pp.
6199-6211, 1991.
[25 ] K. K. Likharev, "Single-Electron Devices and Their Applications", Proc.
of the IEEE, vol. 87.
pp. 606-632. 1999.
[26] C W. J. Beenakker, "Theory of Coulomb-blockade oscillations in the conductance of a quantum dot", Phys. Rev. B, vol. 44, pp. 1646-1656, 1991.
[27] G. Klimeck, G. Chen, and S. Datta, "Conductance spectroscopy in pled quantum dots", Phvs. Rev. B. vol. 50, pp. 2316-2324, 1994.
cou
[28]
M. Macucci, K. Hess, and G. J. lafrate, "Electronic energy spectrum and
the concept of
capacitance
in quantum dots",
Phys.
Rev.
B, vol. 48, pp.
17354-17363,1993.
[29] S.
mer.
I.
Polak, C Den Heijer. W. H. A. Schilders, and P. Markovich, "Semi
conductor device
Methods
modeling from the numerical point Eng., vol 24. pp. 763-838, 1987.
B.
of view", Int. J. Nu-
[30] T. Ando, A.
dimensional
Fowler,
F.
Stern,
"Electronic
vol. 54, pp.
properties
of
two-
systems",
Rev. Mod.
Phys.,
437-672, 1982.
[31] D. Jovanovic and J. P. Leburton, "Self-consistent analysis of singleelectron
charging
effects in
quantum-dot nanostructures". Phvs.
Rev. B, vol.
49, pp. 7474A483, 1994.
[32] S. Tarueha, D. G. Austing.
T. Honda, R. .1.
van
a
der
Hage,
and L. P.
Kouwenhoven, "Shell Filling and
Spin
Effects in
Few Electron
Quantum
Dot", Phys. Rev. Lett., vol. 77, pp. 3613-3616. 1996.
[33] T. Schmidt, M. Tewordt.
R. H. Blick, R. J.
Haug,
D. Pfannkuche. K.
states in a
on
von
Klitzing,
A. Frster, and H. Luth,
"Quantum-dot ground
mag
netic field studied
by single-electron tunneling spectroscopy
double-
barrier heterostructures", Phvs. Rev. B. vol. 51, pp. 5570-5573, 1995.
[34] M. Biittiker, "Role of quantum coherence in series resistors", Phys. Rev.

B, vol. 33, pp. 3020-3026, 1986.
144
Bibliography
[35]
D. J. Ben Daniel and C B.
Duke, "Space-Charge FTfects
on
Electron
Tunneling", Phys. Rev.,

L36]
J.
vol. 152, pp.

a
683-692, 1966.
of view",
Bardeen, "Tunneling from
many-particle point
Phys.
Rev.
Lett., vol. 6. pp. 57-59. 1961.
[37] w. A. Harrison, "Tunneling from
an
Independent-Particle
Point of
View", Phys. Rev., vol 123. pp. 85-89, 1961. [38] B. D. Josephson, "The discovery of tunneling supercurrents",
Rev. Mod.
Phys.,
vol.
46, pp. 251-254. 1974.

En-
[391 K. Hess, Advanced theory of semiconductor devices. Prentice Hall,
glewood Cliffs, N.J.,1988.

[40]
K. A.
Matveev, "Coulomb blockade
at
almost
perfect transmission",
Phys.
[41]
M.
Rev. B, vol.
51, pp. 1743-1751. 1995.
Stopa,
"Coulomb blockade
dot self-consistent level 1993.
amplitudes and semiconductor quantumstructure", Phys. Rev. B, vol. 48, pp. 18340-18343,
[42]
Stopa, R. P. Taylor, Y Aoyagi, and T. Sugano, "Magneto-Coulomb oscillations", Phys. Rev. B. vol. 49, pp. 11488-11491,
J. P. Bird. K. Ishibashi. M.
1994.
[43 ] M. Stopa,
Aoyagi,
and T
Sugano,
"Coulomb oscillation
a
amplitude
enhancement and
envelope
modulation in
magnetic
field and activated
tunneling", Surf. Sei., [44]

M.
vol. 305, pp. 571-575, 1994.
Stopa, Y Aoyagi. and T. Sugano, "Multiple-capacitance magneticfield-dependent Coulomb blockade energetics". Phys. Rev. B. vol. 51, pp.
5494-5497, 1995.
[45]
Stopa, "Quantum dot self-consistent electronic structure and Coulomb blockade", Phys. Rev. B, vol. 54, pp. 13767-13783. 1996.
M. J. C Slater, "A
the
[46]
Simplification
of the Hartree-Fock Method",
Phys.
Rev..
vol.
81, pp. 385-390, 1951. Thomas, "The calculation of atomic fields", Proc. Cambridge Phil.
[47]
L. H.
Soc, vol. 23, pp. 542-547, 1927.
148] E. Fermi, "Eine statistische Methode

schaften des Atoms und ihre
zur
Anwendung
auf die Theorie des
Bestimmung einiger Eigen periodischen
Systems
der Elemente", Z. Physik, \ok 48, pp. 73-79, 1928.
Bibliography
145
[49]
P.
Hohenberg
and W. Kohn.
"Inhomogeneous Electron Gas", Phys. Rev.,
vol.
136, pp. B864-B871, 1964.
[50] S. Lundqvist and N. H. March, Theory of the Inhomogeneous Electron Gas, Plenum, New York, 1983.
[51] W. Kohn and L. J. Sham, "Self-Consistent Equations Including Ex change and Correlation Effects", Phys. Rev., vol. 140, pp. A1133-A1138,
1965T
[52] D. M. Ceperley and B. J. Alder, "Ground State of the Electron Gas by Stochastic Method", Phys. Rev. Lett., vol. 45, pp. 566-569, 1980.
[53] J.
P. Perdew and A.
a
Zunger,
for
"Self-interaction correction to
density-
functional
approximations
many-electron systems", Phys.
Rev. B, vol.
23, pp. 5048-5079, 1981.
154]
M. R. Hestenes and E. Stiefel, "Methods of
conjugate gradients for
solv
ing
[55]
linear
systems",
J. Res. Nat. Bur: Stand., vol.
49, pp. 409-436, 1952.
R. Fletcher,
Proc.
of
the
"Conjugate gradient methods Dundee Biennal Conference on
for indefinite Numerical
systems", Analysis, G.
hi
A.
Watson ed., pp. 409-436. 1975.
[56] P. Sonneveld, "CGS.
fast
systems",
[57] H. A.
ant
SIAM J. Sei. Stat.
Lanczos-type solver for nonsymmetric Comput, vol. 10, pp. 36-52, 1989.
linear
van
der Vorst, "Bi-CGSTAB: A fast and
smoothly converging systems",
vari
of CG-S for the solution of
nonsymmetric
linear
SIAM J. Sei.
Stat.
Comput.,
vol. 13, pp. 631-644, 1992.

van
[58 ] .I.A. Meijerink and H. A.

Math,
der Vorst, "An iterative solution method for

a
linear systems of which the coefficient matrix is
symmetric M-matrix",
of Comp.,
vol. 137, pp. 148-162. 1977.

A.
van
[59] G. L. G. Sleijpen and H.

tion method for linear
vol.
der Vorst, "A Jacobian-Davidson itera SIAM J. Matrix Anal.
eigenvalue problems",
Appl.,
17, pp. 401-425, 1996.
L60] S. E. Laux, "Numerical methods for calculating self-consistent solutions

of electron states in
narrow
channels", in Proc. of NASECODE V, J. J. H.
Miller ed., pp. 270-275, 1987.
L61] C. G. Broyden, "A Class of Methods for Solving Nonlinear Simultane

ous
Equations",
Math.
Comp.,
\ol.
19, pp. 577-593, 1965
146
Bibliography
[62]
P.
Bendt and A.
Zunger,
"New
approach
for
solving
the
density-
functional self-consistent-field
problem", Phys.
Rev. B, vol.
26, pp. 3114-
3137,1982.
[63]
D. Vanderbilt and S. G. Louie, "Total
energies
of diamond
(111) surface
reconstructions
Rev. B. vol.
by
linear combination of atomic orbitals
method", Phys.
30, pp. 6118-6130. 1984.
164]
D. D. Johnson. "Modified
BroydeiFs
method for
accelerating
conver
gence in self-consistent calculations".
Phys.
Rev. B, vol. 38. pp. 12807-
12813,1988.
L65] M. J. D. Powell, "An efficient method for finding the minimum of

function of several variables without
vol.
calculating derivatives", Comput. J.,
7, pp. 155-162,1964.
[66]
R. E. Bank and D. J. Rose, "Global
Approximate
Newton Methods".
Numer Math., vol. 37, pp. 279-295. 1981.
[67] W. Fichtner and D. J. Rose. "On the Numerical Solution of Nonlinear
Elliptic
PDEs
Arising
from Semiconductor Device
Modeling",
in
Elliptic
Problem Solvers, pp. 277-284, Academic Press, 1981.
[68]
D. G. Anderson, "Iterative Procedures for Nonlinear
Integral Expiations",
J. Assoc.
Comp. Mach.,
T.
vol. 12, pp. 547-560. 1965.
[69] A. Trellakis, A.
Gallick, A. Pacelli, and U. Ravaioli,
"Iteration
scheme for the solution of two-dimensional in quantum structures" J.
Schrdfnger-Poisson equations
Appl. Phys.,
vol.
81, pp. 7880-7884, 1997.

at the
[70] This heterostructure has been fabricated by W. Wegscheider
Walter
Schottky
Institut of the Technische Universitt Mnchen.

D. A. Wharam. "Statistics of conductance
[71] F. Simniek T, Heinzel, and

oscillations of
a
quantum dot in the Coulomb-blockade regime", Europhys.
Lett., vol. 38. pp. 123-128, 1997.
[72] T Heinzel, Ein -Elektron -Ladeeffekte
in GaAs-AlGaAs-Heterostrukturen
definierten
Quantenpunkten,
Doctoral
thesis,
Ludwig-Maximilians-
Universitt Mnchen, Mnchen, 1994.
173] J.
H. Davis and I. A.
Larkin, "Modeling the patterned two-dimensional
electron gas: Electrostatics",,/.
Appl. Phys.,
vol. 77, pp. 4504-4512, 1995.
Bibliography
147
[74] I.A. Larkiii and J. H. Davis, "Edge of the two-dimensional electron

in
a
gas
gated heterostructure", Phys.
Rev. B, vol. 52, pp. R5535-R5538, 1995.
[75] M. Chen, W. Porod. and D. J. Kirkner, "Coupled finite element/boundary

element method for semiconductor quantum devices with
exposed
sur
faces", ./. Appl. Phys., vol. 75, pp. 2545-2554, 1994.
[76]
V. Y
Thean, S. Nagaraja, and J. P. Leburton, "Three-dimensional self-
consistent simulation of interface and
dopand disorders
in
delta-doped grid-
gate quantum-dot devices", Phys. Rev. B. vol. 82, pp. 1678-1686, 1997.
[77] F.
Stern
and
S.
Das
Sarma,
"Electron
energy
levels
in
GaAs-
Gai_TAlTAs heterojunctions", Phys.
Rev. B, vol.
30, pp. 840-848, 1984.

a
[78] A. Kumar, S.
dot in
a
E. Laux, and E. Stern, "Electron states in

Rev. B, vol.
GaAs quantum
magnetic field", Phys.

K.
42, pp. 5 166-5175, 1990.

nature
[79] G. J. lafrate,
Hess, J. B. Krieger, and M. Macucci, "Capacitive
of atomic-sized structures",
Phys.
Rev. B. vol. 52, pp.
10737-10739, 1995.
on
[80]
F. M. Buchmann,
Effects of External Magnetic Fields
Quantum Dots,
Semester thesis, Swiss Federal Institute of
Technology (ETFI). Zrich,
1999.
'v%\
J
y>&
*
*,
V,
Jf
':
.
it
Curriculum Vitae
Andreas Scholze
was
born in Arnstadt
(Germany),
on
June 14, 1969. After
finishing secondary school in 1989, he studied Physics at the Universities of Jena (Germany) and Edinburgh (Scotland). He specialized in solid-state the ory submitting a thesis on density-functional based analysis of diamond sur face reconstructions. After graduating from the University of Jena, he joined the Integrated Systems Laboratory (IIS) of the Swiss Federal Institute of Tech nology in Zrich (ETHZ) working towards the Doctoral degree in Electrical Engineering. During the summer of 1997 he stayed as a visiting research as sociate at the Beckman Institute for Advanced Science and Technology at the
University
(M1NAST)
is
on
of Illinois in
K. Hess. Fie is
Urbana-Champaign (UIUC) with the group of Prof. currently working within the framework of a national project
algorithms
for the simula
in the field of nanodevice simulation. The main focus of this work
the establishment of numerical methods and
tion of single-electron devices at the solid-state level.
149

Simulation of Single-Electron

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Simulation of Single-Electron

Uploaded by

Copyright:

Available Formats

Diss. ETHNo.

A dissertation submitted to the

SWISS FEDERAL INSTITUTE OF TECHNOLOGY

Dipl. Phys., University

born June 14, 1969

accepted on the recommendation of Prof. Wolfgang Fichtner, examiner

Fichtner for his

my years at TIS, and Prof. Klaus Ensslin for

thanks goes to PD Andreas Schenk for

scientific aspects of this work.

hard Schmithiisen. Bernd

Thomas Krner, Gtz

Leonhard, Christian Schuster, Fabian Butler and Frank Geelhaar. I wish

(ETH Zrich) (1SE Inc.) for

his support with numerical software.

Also, 1 would like

conditions and all other

people at the I enjoyed very

institute for their contribution

I would like to express my

Prof. Karl Hess who gave

total of five months at the Beckman Institute of

Advanced Science and

Modern semiconductor device simulation

of this work and

and Coulomb blockade

The rules of the game

Constant interaction model

vertical quantum dots

carrier densities bandstructure and

Low dimensional electron gases

Total free 4.3.1

Helmholtz free energy and

Free energy minimization

Bardeen's transfer Haniiltonian method

for quantum dot systems

Device simulation 6.2.1 6.2.2 6.2.3

Achievements, failures and prospects

Time-dependent perturbation theory

limit of miniaturization in electronics

be reached within the next decades.

the simulation of the local

carrier densities and the electrostatic of

and the terminal characteristics

transistors. introduction into the

of the so-called orthodox

SET driven in linear response

functions and the transition rates for

quantum dot in equilibrium with the

reservoir and the linearization of the

is discussed in detail. The evaluation of the

the calculation of the

rates which are

the basis of Bardeen's transfer-Hamiltonian for

the numerical methods used for the

devised for the simulation of

dots, -wires, and -wells.

An introduction into methods for

linear-equation given. singleus

Besides the focus of this work

basic models and methods for

electron device simulation, the simulator