Masters_Raghu

Generalized Monte Carlo Tool for Investigating Low-Field
and High Field Properties of Materials Using
Non-parabolic Band Structure Model
by
Raghuraj Hathwar
A Thesis Presented in Partial Fulfillment
of the Requirements for the Degree
Master of Science
Approved June 2011 by the
Graduate Supervisory Committee:
Dragica Vasileska, Chair
Stephen Marshall Goodnick
Marco Saraniti
ARIZONA STATE UNIVERSITY
August 2011

ii
ABSTRACT
In semiconductor physics, many properties or phenomena of materials can be
brought to light through certain changes in the materials. Having a tool to define
new material properties so as to highlight certain phenomena greatly increases the
ability to understand that phenomena. The generalized Monte Carlo tool allows
the user to do that by keeping every parameter used to define a material, within
the non-parabolic band approximation, a variable in the control of the user. A
material is defined by defining its valleys, energies, valley effective masses and
their directions. The types of scattering to be included can also be chosen. The
non-parabolic band structure model is used.
With the deployment of the generalized Monte Carlo tool onto
www.nanoHUB.org the tool will be available to users around the world. This
makes it a very useful educational tool that can be incorporated into curriculums.
The tool is integrated with Rappture, to allow user-friendly access of the tool. The
user can freely define a material in an easy systematic way without having to
worry about the coding involved. The output results are automatically graphed
and since the code incorporates an analytic band structure model, it is relatively
fast.
The versatility of the tool has been investigated and has produced results
closely matching the experimental values for some common materials. The tool
has been uploaded onto www.nanoHUB.org by integrating it with the Rappture

iii
interface. By using Rappture as the user interface, one can easily make changes to
the current parameter sets to obtain even more accurate results.

iv
ACKNOWLEDGMENTS
I am deeply indebted to my guide and mentor, Dr. Dragica Vasileska. Without
her support, advice and optimism this work would not have been possible. Her
expertise in the area helped me to quickly acquire the knowledge I needed to
complete this thesis. I am also grateful for the time and energy she devoted in
discussing my research and editing my thesis.
I am grateful to Dr. Stephen Goodnick and Dr. Marco Saraniti for being a part
of my Graduate Advisory Committe. I would like to extend my appreciation to the
School of Electrical, Computer and Energy Engineering at Arizona State
University for providing me this opportunity to pursue my Master‟s degree. I also
take this opportunity to thank Darleen Mandt and Esther Korner for helping me
with all the official documents.
I would like to thank all my colleagues in the Computational Electronics group
for their help. I am totally indebted to my family for the love and support I
received in my quest for higher education.

v
TABLE OF CONTENTS
Page
CHAPTER 1. INTRODUCTION........................................................................... 1
1.1. INTRODUCTION.............................................................................................. 1
1.2. NEED FOR HIGH FIELD STUDY ...................................................................... 2
1.3. ADVANTAGES OF THE MONTE CARLO METHOD ............................................ 3
CHAPTER 2. THE MONTE CARLO METHOD.................................................. 6
2.1. INTRODUCTION.............................................................................................. 6
2.2. SINGLE PARTICLE MONTE CARLO METHOD................................................ 10
2.3. ENSEMBLE MONTE CARLO METHOD........................................................... 13
2.4. FERMI‟S GOLDEN RULE .............................................................................. 17
2.5. NON-PARABOLIC BANDS ............................................................................. 18
2.6. DEFORMATION POTENTIAL SCATTERING .................................................... 20
2.6.1. Acoustic Phonon Scattering................................................................ 21
2.6.2. Non-Polar Optical Phonon Scattering................................................. 23
2.7. POLAR OPTICAL PHONON SCATTERING....................................................... 26
2.8. PIEZOELECTRIC SCATTERING ...................................................................... 28
2.9. IONIZED IMPURITY SCATTERING ................................................................. 29
2.10. FINAL ANGLE AFTER SCATTERING ............................................................ 30
CHAPTER 3. THE GENERALIZED MONTE CARLO CODE......................... 33
3.1. INTRODUCTION............................................................................................ 33

vi
PAGE
3.2. INPUT PARAMETERS.................................................................................... 33
3.3. CREATING SCATTERING TABLES................................................................. 35
3.4. INITIALIZING ELECTRONS............................................................................ 35
3.5. CARRIER FREE-FLIGHTS.............................................................................. 36
3.6. SCATTERING THE ELECTRON....................................................................... 39
3.7. CALCULATING ENSEMBLE AVERAGES ........................................................ 40
CHAPTER 4. RESULTS...................................................................................... 46
4.1. SILICON....................................................................................................... 46
4.2. GERMANIUM ............................................................................................... 48
4.3. GALLIUM ARSENIDE ................................................................................... 51
CHAPTER 5. THE RAPPTURE INTERFACE................................................... 54
5.1. INTRODUCTION............................................................................................ 54
5.2. INPUT PARAMETER STRUCTURE .................................................................. 54
CHAPTER 6. CONCLUSIONS AND FUTURE WORK.................................... 63
REFERENCES ..................................................................................................... 65

vii
LIST OF FIGURES
Figure Page
1.1 CPU Transistor counts (from www.intel.com)................................................. 2
2.1 Construction of scattering tables (left panel) and scattering tables
renormalization (right panel) ........................................................................ 12
2.2 Block Diagram of the Ensemble Monte Carlo Code ...................................... 14
2.3 Free-Flight Scatter representation of the Monte Carlo Method...................... 16
2.4 The phonon dispersion relation in bulk silicon using the valence force field
model............................................................................................................. 25
3.1 Flow chart for the generalized Monte Carlo code ......................................... 34
4.1 Energy of electrons in eV versus time in seconds (left panel) Drift velocity
the of electrons in m/s versus time in seconds (right panel)......................... 46
4.2 Energy of the electrons versus the applied electric field. Experimental data is
taken from [29]. ............................................................................................ 47
4.3 Drift velocities of the electrons versus the applied electric field. Experimental
data is taken from [29].................................................................................. 48
4.4 Drift velocity versus time for Germanium..................................................... 48
4.5 Population of Valleys versus time at 6kV/cm in Germanium ........................ 49
4.6 Energy versus Applied electric field in Germanium. Experimental data is
taken from [30]. ............................................................................................ 50

viii
Figure Page
4.7 Drift velocity versus applied electric field in Germanium. Experimental is
data taken from [30]...................................................................................... 51
4.8 Energy versus time in Gallium Arsenide........................................................ 52
4.9 Drift velocity of electrons versus applied electric field. Experimental data is
taken from [31]. ............................................................................................ 52
4.10 Energy of electrons in the gamma valley versus applied electric field.
Experimental data is taken from [32]............................................................ 53
4.11 Fraction of electrons in the L valley versus applied electric field.
Experimental data is taken from [32]............................................................ 53
5.1 Material Parameters ........................................................................................ 55
5.2 Simulation Parameters ................................................................................... 56
5.3 Valley Parameters ........................................................................................... 57
5.4 Effective masses and Valley directions ......................................................... 58
5.5 Scattering parameters..................................................................................... 59
5.6 Intervalley phonon parameters........................................................................ 60
5.7 Single simulation output of the tool................................................................ 61
5.8 Multiple simulation outputs from the tool ...................................................... 62

1
CHAPTER 1. INTRODUCTION
1.1. Introduction
Semiconductors have been the focal point of study for electrical transport
from the 20th
century onwards. The main attraction towards these materials was
the ability to change the conductivity of the semiconductor by introducing
dopants and also by applying an electric field. Although the main use of
semiconductor materials in devices started with the invention of the transistor, it
was not the first device to use semiconductors. Metal rectifiers and detectors in
radios called „Cat Whiskers‟, which were primitive forms of modern day Schottky
diodes, were quite common in the beginning of the 20th
century [1]. The
investigation of semiconductor materials could be said to have started with
Russell Ohl of Bell Laboratories when he tried to grow pure crystals of these
semiconductors and analyze their properties. These tests led to the realization of a
diode structure to alter the electrical properties of a material. Building on the
knowledge of how those diodes work, William Schockley, John Bardeen and
Walter Brattain sandwiched two diodes together to create the first transistor in
1947 at Bell Labs. Since then the number of different semiconductor materials has
grown immensely to produce a variety of devices exploiting the individual,
unique advantages of these materials. The experimental success of the
semiconductor industry would not have been successful without the
corresponding success in the understanding of the physical properties such as the
electrical and optical properties of these devices. These properties were

2
investigated in detail in the 1950‟s itself [2], leading to the understanding of the
energy band structures. From a greater understanding of band structures and other
transport properties of these materials, the electrical properties of the devices
created closely followed the theories proposed at that time.
1.2. Need for High Field Study
The study of charge transport in semiconductors is of fundamental
importance both from the point of view of the basic physics and for its
applications to electrical devices [3]. As the need for electrical appliances grew
the need for smaller and faster devices grew as well. As can be seen in Figure 1.1
the number of transistors in a device grew at an amazing rate closely following
Moore‟s law [4].
Figure 1.1 CPU Transistor counts (from www.intel.com)

3
This was made possible by the reduction in the size of the devices. Since the
voltage applied across the devices did not reduce at the same rate, the field
applied across the device increased. Soon after the invention of the transistor in
1947 it was recognized that electric field strengths, so high so as to cause the
devices to no longer obey Ohm‟s Law, were encountered in semiconductor
samples [5-6]. As the requirement of having such high electric fields in
commercial transistors became a possibility the need for new physics to tackle the
working of these devices arose. The field of nonlinear transport which had been
initiated by Landau and Kompanejez [7] entered a period of rapid development
soon after the invention of the transistor and a number of researchers devoted
their efforts to improving the scientific knowledge of this subject. The surge in
research in this field gave way to the realization of new phenomena like the Gunn
Effect [8] which then led to the invention of new devices like the transit-time
device.
1.3. Advantages of the Monte Carlo method
Analyzing charge transport at high electric fields in devices operated in
the on-state is a difficult problem both from the mathematical and physical point
of view. The Boltzmann equation which defines the transport phenomena for
semi-classical cases is a complicated integro-differential equation. Analytic
solutions of the Boltzmann equation can only be obtained for very few cases and
are usually not applicable to real systems. In order to get an analytic result it is

4
necessary to use such drastic approximations that it can no longer be considered
appropriate to describe real device operation. In 1966 Kurosawa proposed the
Monte Carlo technique [9] and Budd proposed the iterative technique [10]. With
these techniques it became clear that, with the use of modern computers, it would
be possible to exactly solve the Boltzmann equation numerically for physical
models of considerable complexity. These two techniques were then developed
further by Price [11], Rees [12] and Fawcett [13]. The Monte Carlo method
became the more popular technique because it is easier to use and has more
physically interpretable results.
Low field properties of the semiconductor can be investigated using the
relaxation time approximation (RTA) for the case when the relevant scattering
processes are either elastic or isotropic. If that is not the case, then Rode‟s
iterative procedure has to be used [14]. The Monte Carlo method which can be
used for calculation of low-field and high-field properties of a semiconductor uses
a different methodology. In fact, in the long-time limit the Monte Carlo method
gives the solution of the Boltzmann Transport Equation (BTE). In the short-time
limit, the Monte Carlo method gives the solution of the Prigogine equation. Monte
Carlo techniques are statistical numerical methods, which are applied to the
simulation of random processes. In fact the Monte Carlo method as a statistical
numerical method was born well before its application to transport problems [15]
and has been applied to a number of scientific fields [16-17]. In case of the charge
transport however the solution of the Boltzmann transport equation is a direct

5
simulation of the dynamics of the carriers in the material. This means that while
the simulation is being run, while the solution is being built up, any physical
information can be easily extracted. Therefore, even though the result of the
Monte Carlo simulation requires a correct physical interpretation, the method is a
very useful tool to achieve real solutions. It permits the simulation of particular
physical situations unattainable in experiments, or even investigation of
nonexistent materials in order to emphasize special features of the phenomenon
under study. This use of the Monte Carlo technique makes it similar to an
experimental technique and can be compared with analytically formulated theory.
A brief overview of the different types of Monte Carlo methods used in
device simulation and their implementation is discussed in Chapter 2. Also in
Chapter 2, the common types of scattering processes and their corresponding
scattering rates derived using the non-parabolic band structure are discussed. In
Chapter 3, the generalized Monte Carlo method for any material and its
implementation is discussed. Simulation results for some common materials like
Silicon, Germanium and Gallium Arsenide obtained using the generalized Monte
Carlo code and are compared with experimental data in Chapter 4. In Chapter 5,
the rapture interface which enables the implementation of the code onto
www.nanoHUB.org is discussed along with some details of its user interface.
Conclusions derived from this research and future directions of research are given
in Chapter 6.

6
CHAPTER 2. THE MONTE CARLO METHOD
2.1. Introduction
The purpose of device modeling is to be able to predict the electrical properties
of materials and devices. This would then permit changing certain parameters to
improve performance. To obtain these electrical properties, one needs to know the
behavior of the particles in the devices, or more specifically in the materials used
in those devices. The Boltzmann transport equation [18-19] ,
( )
(2.1)
is used to obtain this behavior. It governs the carrier transport in materials under
the semi-classical approximation. This equation is essentially a conservation of
volumes in phase space. The left hand side of equation (2.1) consists of three
terms, the first term describes the temporal variation of the distribution function,
the second term describes the spatial variation of the distribution function which
may arise due to temperature or concentration gradients, and finally the third term
describes the effect on the distribution function due to applied fields (electric or
magnetic). On the right hand side we have two terms, the first term describes the
recombination and generation processes and the second term is the collision
integral which describes the scattering processes. As can be seen the Boltzmann
transport equation is a complicated integro-differential equation which if needs to
be solved analytically, it requires many simplifying assumptions which may not
hold in real devices as was mentioned earlier.

7
The Monte Carlo method is a stochastic method used to solve the
Boltzmann transport equation. In order to develop this approach we first write the
Boltzmann equation as in [18],
( ) ( )
( ) ∫ ( ) ( )
(2.2)
where
∫ ( )
(2.3)
is the total scattering rate out of state p for all scattering processes. This motion of
the distribution function is described in six plus one-dimensional phase space,
three in momentum, three in real space and one in time. It is therefore convenient
to describe the motion of the distribution function along a trajectory in phase
space. The variable along this trajectory is taken to be s and each coordinate can
be parameterized as a function of this variable as
( ) ( ) (2.4)
and the partial derivatives are constrained by the relationships
(2.5)
Applying these changes to equation (2.2) we get
∫ ( ) ( )
(2.6)

8
Equation (2.6) is a standard differential equation which can be solved using an
integrating factor which gives
( ) ( ) (∫ ))
∫ ∫ ( ) ( ) ( ∫ )
(2.7)
By a change of variables from ( ) ( ) , the above equation becomes
( ) ( ) (∫ ))
∫ ∫ ( ) ( ) ( ∫ )
(2.8)
The above equation is the Chamber-Rees path integral [20] and is the form of the
Botlzmann equation which can be iteratively solved. In order to make the above
equation solvable a useful mathematical trick introduced by Rees [21] is used in
which we make the complicated energy dependent function into an energy
independent term, thereby making the term inside the integral in equation (1.7)
trivially solvable. This is done by introducing a scattering term called self-
scattering ( ). Self-scattering does not change the momentum or the energy of
the particle and therefore does not change the physics of the particle. What this
term does however is to convert the energy dependent function into an energy
independent term by defining
   ss T O   p p (2.9)
Therefore equation (2.8) becomes

9
*
0
( , ) ( ,0) ' ( , ' ) ( ' )T T
t
t s
f t f e ds d W e s f e s e 
    p p p p p E p E
(2.10)
where
*
( , ') ( , ') ( ') ( ')ssW W   p p p p p p p (2.11)
The first term of equation (2.10) is a transient term while the second term is the
term which can be iteratively solved. If we look at the second term closely the
first integral over represents the scattering of the distribution function out
of state to state ( ). The second integral represents the integration
along the trajectory s and the exponential is just the probability that no scattering
takes place during the time it moves a distance s. Thus if we look at how the
electrons move physically it consists of a scattering event determined by the first
integral and then there is a free-flight motion (no scattering) for a time interval ts.
Rees showed that the time steps ts correlate to 1/ΓT. This free-flight scatter
sequence is the basis of every Monte-Carlo method used in device simulations.
The Monte Carlo method is mainly used in three different styles, the one-
particle Monte Carlo, the ensemble Monte Carlo and the self-consistent ensemble
Monte Carlo. In the one-particle Monte Carlo method a single carrier‟s motion is
tracked for a certain period of time till the particle has reached steady state. This
method is mostly useful to study bulk properties, like the steady state drift
velocity as a function of field.
In the ensemble Monte Carlo method a large ensemble of carriers are
simulated at the same time. This method can be sped up using parallelization and

10
is useful for super-computation. This method is mostly useful for transient
analysis as ensemble averages can be taken at certain time steps during the
simulation.
In the self-consistent ensemble Monte Carlo method, the ensemble Monte
Carlo method is coupled with a Poisson solver or also a Schrödinger solver and is
the most suitable method for device simulations.
2.2. Single Particle Monte Carlo Method
As was mentioned earlier there are free flight times (drift times) and then
scatter sequences in the Monte Carlo method. If [ ( )] is the probability that
an electron in state k suffers a collision during the time interval dt then the
probability that an electron which has had a collision at time t=0 has not yet
undergone another collision after time t is [22],
∫ [ ( )] (2.12)
Therefore, the probability ( ) that the electron will suffer a collision during
around is,
( ) [ ( )] ∫ [ ( )] (2.13)
If is the maximum of [( ( )] in the region of k-space then,
( ) (2.14)
Using a random variable transformation and integrating equation (2.14) on both
sides we obtain,

11
( )
(2.15)
where r is a random number between 0 and 1.
As can be seen the value of ti will have a higher probability of being a value
around 1/ΓT and if we take a large number of particles the average ti will be
around 1/ΓT,
∫ ( )
(2.16)
The total scattering rate is calculated by adding up all the scattering rates of
individual types of scattering as well as the self-scattering rate which are all
energy dependent.
∑ ( ) ( )
(2.17)
where n is the total number of scattering types considered for a particular material
(e.g. acoustic phonon scattering, optical phonon scattering etc..) As you can see
the value of has no upper limit, only a lower limit. It obviously cannot be lower
than ∑ ( ). But as ∑ ( ) is energy dependent it is important that be
greater than ∑ ( ) where is the energy of the scattering table at which the
cumulative scattering rate is the maximum.
Once the particle has drifted it is time to scatter it. The type of scattering
to be used is chosen from the scattering tables. The usual method is to store the

12
scattering values for each type as a fraction of the total scattering rate . The
figure below better explains this method.
Figure 2.1 Construction of scattering tables (left panel) and scattering tables
renormalization (right panel)
A random number (R) is chosen between 0 and 1 and if
∑ ∑
(2.18)
Then scattering type j+1 is chosen. Here of course . If self scattering is
chosen we do nothing and move on. Once enough time has elapsed, the average
carrier velocity is calculated using
∑ ∫ ( )
(2.19)
This average is only valid as long as the distribution function is in steady state.
This prevents the analysis of transient behavior like velocity overshoot effects, or

13
any other non-ergodic process. But using the single particle Monte Carlo method
the steady state velocity, energy and other parameters can be calculated.
2.3. Ensemble Monte Carlo Method
A different approach than what is described in the previous section is
commonly used by most, and that is the ensemble Monte Carlo Method. Instead
of following a single particle for hundreds of thousands of iterations, thousands of
particles can be followed for a much lesser number of iterations. In this thesis the
ensemble Monte Carlo method is adopted. The time coordinate of each electron
must be maintained during the simulation.
The physical quantities such as velocity and energy are averaged over the
whole ensemble at frequent time intervals so as to obtain the time evolution of
these quantities. For example,
( ) ∑ ( ) ( ) ∑ ( )
(2.20)
where N is the number of particles in the ensemble and t is one of the time
intervals at which the ensemble averages are taken. The general block diagram of
an ensemble Monte Carlo code is shown in figure 2.2. In figure 2.2 the initial
distribution of carriers is a Maxwellian distribution at the given temperature.

14
Figure 2.2 Block Diagram of the Ensemble Monte Carlo Code
The ensemble Monte Carlo method does not require steady state
conditions to calculate the ensemble averages and therefore can be used to
investigate transients in devices. The equation (2.21) and equation (2.22)

15
represents an estimate of the true velocity and energy which has a standard error
of
√
where is the variance that is estimated from [23],
{ ∑( ( )) ( ) }
(2.21)
and
{ ∑( ( )) ( ) }
(2.22)
for the velocity and energy calculations respectively. Typically the value of N is
of the order of 104
or 105
. To obtain the time evolution of certain physical
quantities the need to „freeze‟ the simulation comes up. The time steps ∆t at
which the simulation is paused and the ensemble averages taken should not be
much larger than the maximum frequency of scattering. If it is it will cause a
coarsening of the time evolution and a loss of information. If the time step is too
small then it will create noise in the output. Therefore a balance is needed for the
time steps. One usually keeps it at a few femtoseconds. Therefore there are two
different time scales used in this method, the free-flight duration time and the
sampling time. The different time coordinates are shown in the figure below.

16
Figure 2.3 Free-Flight Scatter representation of the Monte Carlo Method
In the above figure N is the total number of particles in the simulation while ts is
the total simulation time and is the free-flight duration time for the ith
particle.
The simulation is paused and the ensemble averages are taken at every as
shown in figure 2.3.

17
2.4. Fermi’s Golden Rule
The scattering processes which interrupt the carrier free-flights are
calculated quantum mechanically. The scattering event is treated by defining a
scattering potential, which is calculated for each type of scattering process. Each
of the different processes, or interactions leads to a different “matrix element”
form in terms of its dependence on the initial wave vector, the final wave vector
and their corresponding energies. The matrix element is given by,
( ) (2.23)
For three dimensional cases the matrix element usually contains the momentum
conservation condition which comes about due to the overlap of the normal Bloch
functions of the electrons.
Solving the time-dependent Schrödinger equation using first-order
perturbation theory leads to the equation for the scattering rate from a state to k
as,
( ) ( ) ( )
(2.24)
Equation (2.24) is called Fermi‟s Golden Rule, where and are the initial and
final states of the carrier, and are the corresponding kinetic energies and
is the phonon energy and ( ) describes the conservation of
energy during the scattering process. The conservation of energy is only valid in
the long-time limit, i.e. when the scattering events are infrequent. The top sign is
for absorption and the bottom sign is for the phonon emission process.

18
The total scattering rate out of a state defined by wave vector and the
energy is obtained by summing over all states in equation (2.24).
( ) ∑ ( ) ( )
(2.25)
In equation (2.25) the sum over all states can be converted to an integral over
giving,
( )
( )
∫ ∫ ∫ ( )
(2.26)
where is the total volume of the crystal and ( ) is given by equation
(2.24). Equation (2.26) is used to calculate scattering rates as a function of energy.
2.5. Non-Parabolic bands
The equation mapping the energy of an electron above the valley minima to its
wave vector k using the parabolic band approximation is,
(2.27)
This approximation is only valid for energies slightly greater than the energy of
the valley minima. For Monte Carlo simulations in which high field transport is
essential this approximation is not accurate enough. To improve accuracy we need
a function that better maps the energy of an electron to its wave vector. A full
band calculation will give us a more accurate mapping of energy and momentum
of an electron but as it is not an analytic function it becomes hard to switch freely
between the energy and momentum which is essential to do in a Monte-Carlo

19
simulation. Therefore, the full band simulations are very computer intensive. In
order to use an analytic approach and still improve upon the accuracy the k.p
method is used to obtain the non-parabolic equation [24],
( )
(2.28)
Here α is a term coming from the k.p method which depends on the material
as,
( )
(2.29)
where is the energy difference between the conduction band and the valence
band at the point, is the electron rest mass and is the conductivity mass.
The above equation is also an approximation valid as long as,
(2.30)
For electron energies in which ~ the above equation fails and a
full band calculation is required to more accurately simulate transport in the
material. Assuming that the above assumptions are valid it is important to note
the changes that the non-parabolic band approximation introduces. The density of
states and the conductivity effective masses are given by,
( ) ( )
(2.31)
For parabolic bands both the masses turn out to be equal to m, but when non-
parabolic bands are used we get,

20
( ) ( ) (2.32)
It is interesting to note that in this method as the electron gains more and more
energy its effective mass increases. This means that the electron becomes
„heavier‟ or reacts slower to the electric field as its energy increases. The change
in density of states effective mass causes the scattering rates to get modified
slightly as compared to the parabolic bands case. The rest of the chapter deals
with the scattering rates commonly used and the formula used to calculate those
rates assuming non-parabolic bands.
2.6. Deformation Potential Scattering
Electrons interacting with the vibrations in the crystal lattice give rise to
deformation potential scattering. These vibrations stress the lattice producing an
elastic strain. When neighboring atoms in a lattice oscillate in the same direction,
they give rise to the acoustic branch in the phonon spectra, giving rise to acoustic
phonon scattering.
When neighboring atoms oscillate in the opposite direction, they give rise
to the optical branch of the phonon spectra, giving rise to non-polar optical
phonon scattering.
The essential concept due to Bardeen and Shockley [25] in calculating the
deformation potentials is that, if the solid is subject to a strain that is a slowly
varying function of position, there will be a change in the energy of the electronic

21
state that is proportional to the strain. Therefore the Hamiltonian due to the
deformation potential electron-phonon interaction is,
(2.33)
where is the strain caused due to the lattice vibrations and is a deformation
potential constant.
2.6.1. Acoustic Phonon Scattering
In order to calculate the scattering rate due to acoustic phonons the „matrix
element‟ given by equation (2.23) needs to be calculated, which means the
Hamiltonian for the electron phonon interaction needs to be calculated. This is
given by equation (2.33). In case of acoustic phonons the volume dilation ( ) is
given by,
( )
(2.34)
where the operator for the lattice displacement vector ( ), that appears in
equation (2.34), is a function of time t and position r, and is given by,
( ) ∑ (√ ) [ ]
(2.35)
where is the phonon wave vector, is the density, is the phonon
energy, is the crystal volume, is the unit polarization vector, and are
the annihilation and creation of phonon operators.

22
This gives the Hamiltonian for the interaction to be,
( ) (2.36)
where is the acoustic deformation potential constant which is an
experimentally determined parameter within certain limits. The final „matrix
element‟ squared given by equation (2.23) which will be used in equation (2.26)
to calculate the scattering rate is,
( ) ( ) ( )
(2.37)
where is the equilibrium number of phonon in a state given by,
(2.38)
Equation (2.37) is further approximated to simplify the calculation. The first
approximation assumes that the acoustic phonon energy is much lesser than the
average energy of an electron at lattice temperature . This means,
(2.39)
The elastic approximation also leads to equation (2.38) becoming
(2.40)
This approximation is called the equipartition approximation and is obviously not
valid at low temperatures. Applying this to equation (2.37) we get,
( ) ( ) ( )
(2.41)

23
The second approximation is to assume that the dispersion relation of
acoustic phonons is linear (Debye limit). This means that,
(2.42)
where is the velocity of sound in the crystal.
This as can be seen in figure 2.9 is a reasonable approximation for low energies
near the gamma point. This approximation further simplifies equation (2.41) to,
( ) ( )
(2.43)
This expression is then substituted into equation (2.24) and the scattering rate is
calculated as a function of energy using the method given in section 2.4. The final
expression of acoustic phonon scattering rate obtained using a non-parabolic band
structure is,
( ) ( ) (
( ) √ ( )
) ( )
(2.44)
2.6.2. Non-Polar Optical Phonon Scattering
To calculate the scattering rate due to non-polar optical phonons, a similar method
to that used previously has to be employed. This means the Hamiltonian for the
electron phonon interaction needs to be calculated. In the case of optical phonons
the neighboring atoms oscillate in the opposite direction, thus directly affecting
the size of the unit cell. Because of this, the volume dilation ( ) is given by,

24
( )
(2.45)
This gives the Hamiltonian for the interaction to be,
( ) (2.46)
where is the optical deformation potential constant which is an experimentally
determined parameter within certain limits. The operator for the lattice
displacement vector ( ) is given by equation (2.35). The final „matrix
element‟ squared given by equation (2.23) which will be used in equation (2.24)
to calculate the scattering rate then becomes,
( ) ( ) ( )
(2.47)
where is the number of phonon in a state given by equation (2.38).
An approximation used here to simply calculations is that for optical
phonons (Einstein model). This basically means that the optical phonons are
dispersionless, this is a reasonable approximation when you look at figure 2.9.
This approximation reduces equation (2.47) to,
( ) ( ) ( )
(2.48)
This expression is then substituted into equation (2.24) and the scattering rate is
calculated as a function of energy using the method given in section 2.4. The final
expression of optical phonon scattering rate (intervalley scattering) from valley
to valley obtained using a non-parabolic band structure is,

25
   
3/22
2 3
(2 ) (1 )1 1
* * *(1 2 )
2 2 4
d f fij j
ij f
ij
m E ED Z
W E n E


 

                  
(2.49)
( )
1
1
ij
B LK T
ijn
e

 

(2.50)
f ij ijE E E   (2.51)
is the deformation potential for transition from valley i to valley j, is the
density of the material and is the number of final valleys to scatter into.
Figure 2.4 The phonon dispersion relation in bulk silicon using the valence force
field model

26
For intervalley processes the interaction described by equation (2.46) is
the zeroth order interaction. Ferry [26] considered the first order interaction
process, for which the matrix element is,
( ) ( ) ( )
(2.52)
which gives the total scattering rate out of state k as,
( )
√
( ) ( )
(2.53)
where
( ) √ ( )( )[ ( ) ( )]
(2.54)
2.7. Polar Optical Phonon Scattering
In polar materials there is an atom with charge greater than four and an atom with
charge less than four. This imbalance causes a net electronic charge transfer from
the atom with greater charge to the atom with lesser charge. For covalent bonds in
which the electrons are shared between different bonding orbitals there is only a
fractional charge transfer given by the „effective charge‟, . This small charge
transfer leads to an effective dipole, which leads to a finite ionic contribution to
the dielectric function.
The lattice vibrations of the crystal cause this dipole to oscillate creating a
scattering potential. The electron-phonon interaction is described by the Fröhlich
Hamiltonian which, including screening of electrons is given by [27],

27
( ) ∑ ( ) √ [ ]
(2.55)
where is the number of atoms in a unit cell. The effective charge is given
by,
(
( )
)
(2.56)
the reduced mass is given by,
(2.57)
and,
√
( )
(2.58)
The final „matrix element‟ squared given by equation (2.23) which will be
used in equation (2.24) to calculate the scattering rate then becomes,
( ) ( ) (
( )
) ( ) ( )
(2.59)
Equation (2.55) was simplified by assuming that in equation (2.52) for
all optical phonons. This expression is then substituted into equation (2.24) and
the scattering rate is calculated as a function of energy using the method given in
section 2.4. The final expression of polar optical phonon scattering is,
( )
√
√
(
( )
) ( ) (
√
) ( )
(2.60)
where

28
( )
(
√
√
)
( )
√
( )
(2.61)
2.8. Piezoelectric Scattering
In polar materials the charge transfer between two atoms creates long-range
macroscopic electric fields which interact with electrons to create a scattering
potential. Due to the acoustic modes of phonons, the strain in the lattice changes
the electric field and creates a new form of scattering called piezoelectric
scattering. The Hamiltonian for this interaction is given by,
( ) ∑ √ [ ]
(2.62)
This leads to the following expression for the matrix element squared for this
scattering mechanism,

29
( ) ( ) ( ) ( )
(2.63)
Assuming the same approximations as those used in acoustic phonon scattering,
namely equipartition approximation and linear dispersion of acoustic phonons we
obtain the following relation for the matrix element squared,
( ) ( ) ( ) ( )
(2.64)
This expression is then substituted into equation (2.24) and the scattering rate
is calculated as a function of energy using the method given in section 2.4. The
final expression of piezoelectric scattering is,
   
23/2
B L
2 4
2 2s
2
2 K T 1
W E * * (1 ) *(1 2 ) *
8 (1 )πρv
pzd
d
D D
eem
E E E
m E E
q q
 

 
    
               
  
(2.65)
where is the velocity of sound in the material, is the piezoelectric coupling
constant and is the high frequency dielectric constant. is given by equation
(2.58).
2.9. Ionized Impurity Scattering
Ionized impurity scattering occurs as the name suggests due to the deflection of
an electron by a Coulomb potential due to ionized impurities. This is an elastic
scattering process and is non-isotropic. The scattering rate due to ionized
impurities obtained using the Brooks-Herring formula for non-parabolic bands is,

30
   
4 3/2
2 4
2 2
2
2 1
* (1 ) *(1 2 ) *
8 (1 )
I d
ds
D D
e N m
W E E E E
m E E
q q
 

 
  
            
  
(2.66)
is the ionized impurity concentration.
The above scattering rates have all been calculated from their respective matrix
elements and the Fermi golden rule under the non-parabolic bands approximation.
2.10. Final Angle after Scattering
An important calculation required is to calculate the final angle after a
scattering takes place. A scattering process is anything that changes the
momentum of the electron. Certain processes are anisotropic by which we mean
they preferably choose certain angles depending on the energy of the electron
while others are isotropic which means they have equal probability to scatter the
electron in any angle. A useful formula to calculate the final angles after
scattering for a certain type of scattering process is [28]
 
 
'
2
0
'2 '
0 0
' '2 '
0 0
0 0
,
,
r d S sin d p dp
dr
d S sin d p dp



  
  



 
 
  
p p
p p
(2.67)
and
 
 
' '2 '
0 0
2
' '2 '
0 0
0 00
,
,
r sin d S d p dp
dr
S sin d d p dp


 
  
  



 
 
  
p p
p p
(2.68)

31
In spherical coordinates all that is required to completely define the coordinates of
the final momentum state is , θ and ϕ. is obtained based on the final energy
after scattering by the formula,
√
( )
(2.69)
while θ is determined by equation (2.68) and ϕ is determined by equation (2.67).
Here ( ) is the matrix element squared for each scattering type. This
is in the same form used in the Fermi‟s golden rule. As all scattering processes
have matrix elements independent of ϕ equation (2.67) can be reduced to,
0 0
2
2
r
d
dr
r



 

 
 
(2.70)
Final Angles for Isotropic Scattering Processes
For isotropic scattering processes like acoustic phonon scattering, non-polar
optical phonon scattering the matrix element is independent of θ. Therefore
equation (2.68) can be reduced to,
0 0
2
1 2
r
sin d
dr
cos r

 


  
 
(2.71)

32
Final Angles for Anisotropic Scattering Processes
The non-isotropic scattering processes are ionized impurity scattering, polar
optical scattering and piezoelectric scattering. Ionized impurity and piezoelectric
scattering have the same relationship with θ so for both processes we can
calculate the final angle using,
22
2
1
1 (1 )
8 (1 )
D
d
r
cos
z r
m E E
z
q


 
 


(2.72)
For polar optical scattering the final angle is calculated using,
( )
( ) ( )
√ ( )( ( ) )
√ ( ) √ ( )( ( ) )
(2.73)

33
CHAPTER 3. THE GENERALIZED MONTE CARLO CODE
3.1. Introduction
The purpose of the generalized Monte Carlo code is to give users the
option of defining their own material or modifying the definition of an existing
material. Technically any number of different materials can be simulated within
the non-parabolic band approximation. This makes the code very versatile and
necessitates a very general method of implementing the code which results in a
large set of input parameters which in turn increases the complexity of the code.
3.2. Input Parameters
In the generalized Monte Carlo code there is a wide range of input
parameters. The input parameters are either loaded from a file or from the
Rappture interface discussed in Chapter 5. In order to be able to define a material
the user can input the number of valleys to be used in the simulation, the number
of sub-valleys (equivalent valleys) within each valley, the direction of those sub-
valleys, the effective masses of electrons in those sub-valley directions and the
energy difference between the bottom of the valleys. In addition to choosing
deformation potential scattering (acoustic and optical), ionized impurity
scattering, polar-optical phonon scattering and piezoelectric scattering the user
can also specify whether the optical phonon scattering within a valley are

34
umklapp processes like it is needed in Silicon for the case of g-phonon intervalley
scattering.
Figure 3.1 Flow chart for the generalized Monte Carlo code

35
3.3. Creating Scattering Tables
The scattering table has to be created and stored separately for each valley
as the scattering parameters depend on the effective mass of the electron which
differs from valley to valley. Also some scattering processes might exist in one
valley but not in the other for e.g. in Germanium f and g type scattering is
considered in the X valley but not in the L valley. Therefore the number of
scattering processes can also vary from one valley to another. The scattering
tables are normalized to the maximum value of the total scattering rate within the
energy range specified by the user which will be different for different valleys.
3.4. Initializing Electrons
The electrons are initialized to a Maxwell distribution at the temperature
T. The formulae used to initialize the electron‟s energy and momentum are,
( )
√
( )
√ ( ) ( ) (3.1)
√ ( ) ( )
( )
The electrons, whose number is defined by the user, are initially placed in the
lowest energy valley which is also defined by the user and is equally distributed to

36
all the sub-valleys present in that valley. Here m is the drift (conductivity)
effective mass used to make the bands isotropic.
3.5. Carrier Free-Flights
In between scattering events the electron is drifted under the applied
electric field. The equation describing the dispersion relation of the electron for a
general sub-valley for non-parabolic bands is,
( )
(3.2)
where are the wave-vectors along the three mutually perpendicular
directions that define the sub-valley and are the effective masses of
the electrons along those directions.
Equation (3.2) represents the dispersion relation for an anisotropic band, which is
the most general case. To calculate the drift velocity equations, equation (3.2)
must first be converted to an isotropic dispersion relation by changing the wave-
vectors to where,
√
(3.3)
here is the conductivity effective mass used for all . Substituting
equation (3.3) into equation (3.2) we get,
( )
( ) (3.4)

37
The above equation now represents the dispersion relation of an electron for a
spherical band with effective mass in all directions. According to Newton‟s
second law of motion - the rate of change of momentum is equal to the force
applied to the electron giving,
(3.5)
where are the electric field magnitudes along the mutually
perpendicular directions that define the sub-valley. is the drift time selected by
equation (2.15). Substituting equation (3.5) into (3.6) we get,
√
√
√
(3.6)
The electric field applied to the device is defined by the user before the simulation
starts along the (x,y,z) coordinate system. In order to drift the electron according
to equation (3.6) we need the electric field magnitudes along the three mutually
perpendicular directions (1,2,3) which define the sub-valley in k-space. In general
the coordinate system (1,2,3) is completely different from the (x,y,z) coordinate
system. Therefore before every electron is drifted, the sub-valley in which the

38
electron currently exists is identified and the coordinate system is transformed
from the (x,y,z) system to the (1,2,3) system which defines that sub-valley in k-
space. The electric field acting on the electron is then given by,
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
(3.7)
where the three mutually perpendicular directions that describe the sub-valley are
[a1,b1,c1], [a2,b2,c2] and [a3,b3,c3]. According to equation (3.6) we also need to
have the wave vectors along the directions [a1,b1,c1], [a2,b2,c2] and [a3,b3,c3]. This
is obtained by doing another transformation which gives,
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
(3.8)
The electron is then drifted according to equation (3.6) and the coordinates system
is transformed back to the (x,y,z) coordinate system using,

39
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
(3.9)
The energy of the electron is then calculated by using,
√
(3.10)
where
( ) (3.11)
where is the mass used in equation (3.3) to make the sub-valley spherically
symmetric. Therefore whenever a change from energy to momentum or vice-
versa is required the mass for that particular sub-valley must be used. In this
code the mass is always the „drift (conductivity) mass‟ of the sub-valley or,
(3.12)
where are the effective masses of the electron along the three
mutually perpendicular directions that define the sub-valley.
3.6. Scattering the Electron
The scattering type is chosen by the method mentioned in the previous
chapter. It would depend on the valley the electron is in at the time the scattering

40
takes place as the scattering tables are different for different valleys. Therefore it
is necessary to keep track of which valley and which sub-valley the electron is in
at all times. If the scattering type chosen is non-polar optical phonon scattering
from valley 1 to valley 2 then the final sub-valley of the electron is randomly
chosen from all the sub-valleys present in valley 2 as they are all at the same
energy level and should therefore have equal probability of being scattered into. If
there is non-polar optical phonon scattering within a valley then the final sub-
valley of the electron will depend on whether f and g type scattering occurs or not.
If there isn‟t any f and g type scattering then the final sub-valley is randomly
chosen from the remaining sub-valleys in that valley. If f and g type scattering is
present and if the f-type scattering process is chosen then the final sub-valley is
randomly chosen from all remaining sub-valleys in the valley which are not in the
same axis as the present sub-valley. In the case of g-type scattering the final sub-
valley is the other sub-valley which lies on the same axis as the present sub-
valley.
3.7. Calculating Ensemble Averages
At the end of every time step the ensemble averages are calculated. This
involves calculating the average drift velocity of the electrons, the average energy
of the electrons and the number of electrons present in each valley and sub-valley.
The average energy is calculated using,

41
〈 〉 ∑
(3.13)
where is the energy of the ith
electron. The average energies within a sub-valley
„j‟ in valley „i‟ is,
〈 〉 ∑
(3.14)
where is the energy of the kth
electron in the jth
sub-valley of the ith
valley
and is the number of electrons in the jth
sub-valley of the ith
valley.
Calculation of drift velocity in many valley semiconductors
In most semiconductors in order to properly simulate high field transport it
is necessary to consider more than 1 conduction band valley. To calculate the drift
velocity along any direction the effective mass along that particular direction is
required. This makes it a little complicated to calculate the average drift velocity
as different valleys are orientated differently in k-space and we only have the
effective mass values along specific directions within each sub-valley. For
example, in GaAs a sub-valley of the L valley lies along the [111] direction. We
know the effective masses along the transverse and longitudinal directions of the
sub-valley but we do not know the effective mass along the [100] direction of that
sub-valley. In order to calculate the drift velocity along the [100] direction we
transform the coordinate system to a system along which we know the effective
masses and then switch back to the original coordinate system.

42
Assume the Monte-Carlo simulation is run on the x,y and z coordinate
system where the x-direction is [100] , y-direction is [010] and z-direction is
[001]. Each sub-valley can be completely described by three mutually
perpendicular axes. Let the three mutually perpendicular directions that describe
the sub-valley be [a1,b1,c1], [a2,b2,c2] and [a3,b3,c3]. The electrons in the Monte-
Carlo simulation will be drifted according to the x,y and z coordinate system so
there will be kx , the component of the wave vector along [100], ky , the
component of the wave vector along [010] and kz , the component of the wave
vector along [001]. Therefore the total wave vector of the electron can be written
as,
̂ ̂ ̂ (3.15)
The drift velocity along [a1,b1,c1] under the non-parabolic band approximation is
calculated using,
( ) (3.16)
where ( ) is the electron dispersion relation for the electrons in the valley.
Equation (3.16) is valid only for spherical valleys, therefore we have to convert
the anisotropic valley to an isotropic valley by using the method described earlier.
After making the valley isotropic the drift velocity is,
[ ]
[ ]
√ [ ]( )
(3.17)

43
where [ ] is the component of the wave vector along [a1,b1,c1] and [ ]
is the effective mass of the electron along [a1,b1,c1], is the conductivity
effective mass used to make the valley isotropic in equation (3.3) and is the
energy of the electron in that valley. Similarly the drift velocity along [a2,b2,c2]
and [a3,b3,c3] is given by,
[ ]
[ ]
√ [ ]( )
[ ]
[ ]
√ [ ]( )
(3.18)
Using a simple transformation of coordinates from x,y and z coordinate system to
the [a1,b1,c1], [a2,b2,c2] and [a3,b3,c3] coordinate system we get,
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
[ ]
√( ) √( ) √( )
(3.19)
The coordinates system is then transformed once again back to the x,y and z
coordinate system to get the drift velocities along the x,y and z directions.

44
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
[ ]
√( )
(3.20)
As [a1,b1,c1], [a2,b2,c2] and [a3,b3,c3] are mutually perpendicular direction we
have,
(3.21)
For N electrons in the simulation the average drift velocity is then calculated as
〈 〉 ∑
〈 〉 ∑
〈 〉 ∑
(3.22)
where , and are the drift velocities of the ith
electron in the x,y and z
directions and will depend on the sub-valley the electron is in at the time of the
calculation.

45
Since it is a Monte Carlo simulation, there is always a certain amount of
error in the final velocities even when steady state is reached. Therefore simply
taking the last value of the steady state velocity is inaccurate. Similar conclusion
holds for the average energies as well. Therefore an average is taken in the last „t‟
seconds of the simulation over all the quantities and it is this average over time „t‟
that is used to plot the velocity versus electric field plots to extract the mobility or
the energy versus electric field plots. The amount of time „t‟ used to take the
average is a user defined value. It is usually one or two picoseconds after steady
state is reached.

46
CHAPTER 4. RESULTS
The generalized Monte Carlo code was used to reproduce the characteristic results
of certain materials to test its capability. There are a set of parameters for each
material which are fitting parameters used to best fit the simulated data to the
experimental data.
4.1. Silicon
Figure 4.1 Energy of electrons in eV versus time in seconds (left panel) Drift
velocity the of electrons in m/s versus time in seconds (right panel)
In figure 4.1 the usual plots of energy versus time and velocity versus time are
shown. As can be seen steady state is achieved quite fast at around 2ps. In figure
4.1 the saturation of the velocity with the increase in electric field can be seen
with the last two values of electric field plotted.

47
Figure 4.2 Energy of the electrons versus the applied electric field. Experimental
data is taken from [29].
In figure 4.2 the velocity for different electric fields is plotted and
compared with experimental data. As can be seen there is very good agreement
between the experimental values and the values obtained from the simulation.
In figure 4.3 a similar plot is plotted between energy and electric field. In
all these plots the electric field is applied in the [111] direction. As can be seen
again there is a very good agreement between the experimental values and those
obtained from the simulation.

48
Figure 4.3 Drift velocities of the electrons versus the applied electric field.
Experimental data is taken from [29].
4.2. Germanium
Figure 4.4 Drift velocity versus time for Germanium

49
The velocity of the electrons and the energy of the electrons are plotted against
time in figure 4.4. In germanium the number of valleys chosen is 3, the L valley,
Gamma valley and the X valley. Due to this the simulation takes longer to reach
steady state as there will be transfer of electrons between the valleys and a
transfer of energy between the valleys. This can be seen when these curves are
compared with those obtained with silicon which uses just 1 valley (the X valley)
in the previous section.
Figure 4.5 Population of Valleys versus time at 6kV/cm in Germanium
In figure 4.5 the number of electrons in each valley is plotted versus time. As can
be seen there are almost no electrons in the gamma valley as its effective mass is
really small. This causes a low density of states causing a low probability of
scattering into that valley.

50
Figure 4.6 Energy versus Applied electric field in Germanium. Experimental
In figure 4.6 and figure 4.7 the energy of the electrons and the steady state drift
velocity of the electrons are plotted against the applied electric field and
compared with experimental values. As can be seen there is a good agreement
between the experimental data and the simulated data. It is also possible to obtain
even more accurate results by tweaking the fitting parameters further more.

51
Figure 4.7 Drift velocity versus applied electric field in Germanium.
Experimental is data taken from [30].
4.3. Gallium Arsenide
In figure 4.8 the drift velocity and the energy of the electrons are plotted against
time. Just as is the case in germanium, in gallium arsenide there are 3 valleys.
Therefore the simulation takes a longer time to reach steady state as can be seen
in these two figures. In figure 4.9 and figure 4.10 the energy of the electrons and
the steady state drift velocity of the electrons are plotted against the applied
electric field and compared with experimental values. In figure 4.11 the fraction
of electrons in the L valley is plotted against the applied electric field.

52
Figure 4.8 Energy versus time in Gallium Arsenide
Figure 4.9 Drift velocity of electrons versus applied electric field. Experimental

53
Figure 4.10 Energy of electrons in the gamma valley versus applied electric field.
Figure 4.11 Fraction of electrons in the L valley versus applied electric field.

54
CHAPTER 5. THE RAPPTURE INTERFACE
5.1. Introduction
Rappture is a toolkit supporting rapid application infrastructure, making it
quick and easy to develop powerful scientific applications. The Rappture toolkit
provides the basic infrastructure for a large class of scientific applications, letting
scientists focus on their core algorithm when developing new simulators [33]. The
tools on www.nanoHUB.org were created using Rappture as the user interface
making the code easily useable and accessible to most people. The same was done
for the generalized bulk Monte Carlo tool.
5.2. Input Parameter Structure
The material can have up to 4 different valleys with each having up to 12
different sub-valleys (equivalent valleys). The Rappture interface is designed to
facilitate the process of entering such a large number of inputs. Some parameters
are not required for certain materials while others are. The tool also has the option
of pre-loading a material‟s input parameters so that the user can run the
simulation of a known material and examine the output. As can be seen in Figure
(5.1), the user can choose the material from the drop down menu and the values
will automatically load themselves. Additional materials can be easily added to
the tool after the fitting parameters have been carefully selected to match
experimental data.

55
Figure 5.1 Material Parameters
The tool guides the user across the various parameters required before starting the
actual simulation. Any electric field direction can be chosen. The other
parameters that can be chosen are also listed. One can also speed up the
simulation by reducing the maximum energy of the scattering table if a low field
simulation is running, or by reducing the number of electrons simulated. As can
be seen in Figure (5.2) there is also the option to choose multiple electric fields

56
each with its own simulation time to output velocity versus field/ energy versus
field/ population versus field curves. The tool uses the previous electric field‟s
simulation final carrier distribution as the starting distribution of the new electric
field simulation. This means that the new simulation need not be run for a long
time as it will reach steady state faster. This speeds up the total time of the
simulation which is critical in Monte Carlo simulations.
Figure 5.2 Simulation Parameters

57
Figure 5.3 Valley Parameters
The valley parameters tab is shown in figure 5.3. The Valley Parameters tab
allows the user to choose the number of valleys and the properties of those
valleys. The number of valley tabs shown depends on the number of valleys
chosen so as to not clutter the page with too much unnecessary information. The
same is done with the number of sub-valley tabs below. There is also the option to
choose whether the valley has f and g type scattering as in Silicon or not as in

58
GaAs. The lowest valley is given a minimum energy and the electrons are initially
all placed in the lowest energy valley.
Figure 5.4 Effective masses and Valley directions
To further simplify the input process there are options to let the simulator
calculate the transverse directions of the sub-valley so that users are not trying to
calculate 3 mutually perpendicular directions. This is shown in figure 5.4. There

59
is also an option to apply the same mass pattern of the 1st
sub-valley to the other
sub-valleys if they are all equivalent. Once these options are ticked the
information boxes not required are automatically shaded out so as to not confuse
the user. So only the parameters required are left open. The above snapshot shows
the L valley description of GaAs.
Figure 5.5 Scattering parameters

60
In this tab (shown in figure 5.5) the user can choose the scattering types, and the
relevant parameters required are automatically shown. As in the previous tab the f
and g type scattering was not included in Valley 1, all the boxes requiring that
information in this valley is shaded out. In this way the user does not get
overwhelmed by unnecessary information.
Figure 5.6 Intervalley phonon parameters
Shown in figure 5.6 is the case where only Zero-Order Intervalley Scattering was
included and only the information that is required is shown. There is also an

61
option to allow all valley transitions to have the same phonon energy/deformation
potential if needed. There are further adjustments made to the input deck e.g. if f
and g type scattering were included in valley 1 there is no need to specify a
phonon energy (the independent f and g type phonon energies would have been
asked in the previous tab) within the valley , so that option would have been
shaded out and so on.
Figure 5.7 Single simulation output of the tool

62
Figure 5.8 Multiple simulation outputs from the tool
Figures 5.7 and 5.8 show the option to run multiple electric fields and observe the
output of each electric field to see whether steady state has been achieved. They
show the ease in which the user can see whether steady state has been reached or
not. The error can also be judged from the outputs which would indicate that the
number of electrons used in the simulation may be lacking.

63
CHAPTER 6. CONCLUSIONS AND FUTURE WORK
This thesis is mainly directed towards creating a research tool and an
educational tool. The tool gives the user the option and freedom to vary
parameters within limits determined by experimental values of the measured
coupling constants and effective masses. Also as this tool will be deployed on
www.nanoHUB.org the reach of this tool will be extensive.
The tool uses a non-parabolic band approximation and incorporates most
types of scattering rates. As of now it has deformation potential scattering
(acoustic and optical), polar optical phonon scattering, piezoelectric scattering and
ionized impurity scattering. Using non-parabolic bands makes the simulation as
accurate as possible without considering a full band relation. Since it uses an
analytical relation between energy and momentum of the electron the simulation
time is low, making it more user friendly as an online tool.
The user interface created using the rapture libraries and an xml file can be
easily modified to add changes to the tool. This makes updating the tool, which
will undoubtedly be needed later, an easy task. Adding new materials to the tool
with pre-defined values is also a simple task. The xml file is automatically created
and saved every time the user runs the tool meaning that one just has to save the
file which has the correct parameters as a new material. This file can then be
loaded any time from the drop down menu to simulate that material.

64
There are some improvements that can be added to this tool. As was
mentioned earlier the tool can be easily updated with regard to adding new
materials or changing the input interface. There is also a plan to add the scattering
rates plots to the output so that the user can see which scattering types dominate
and which do not. A possible but not necessary extension of the tool is
incorporating a full band simulation instead of the non-parabolic band
approximation that was used here. The two have advantages and disadvantages.
The non-parabolic band model is definitely not accurate for very high applied
fields. But the full band calculation, being an equilibrium calculation is also
inaccurate in representing conduction bands, in particular those that lie high in
energy. This limitation of the full band models is not always clearly stated in the
literature and amongst the scientific community. Modeling of holes with a full-
band calculation is a must as holes are accurately represented with a full band
structure. Thus adding hole transport with a full-band model is a possible
extension of the tool.

65
REFERENCES
[1] Braun, Ernest; Stuart MacDonald (1982). “Revolution in Miniature: The
history and impact of semiconductor electronics, 2nd
Ed.” UK Cambridge
Univ. Press 11-12.
[2] Schockley, W. 1951, Bell Syst. Tech. J. 30, 990.
[3] C. Jacoboni and L. Reggiani “The Monte Carlo method for the solution of
charge transport in semiconductors with applications to covalent materials”
646-703.
[4] Keyes, Robert W. (September 2006). “The Impact of Moore’s Law”. Solid
State Circuits Newsletter. Retrieved 28 November 2008.
[5] Davydov, B., 1936, Phys. Z. Sowjetunion 9, 443.
[6] Davydov, B., 1937, Phys. Z. Sowjetunion 12, 269..
[7] Landau, L., and A. Kompanejez, 1934, Phys. Z. Sowjetunion 6, 163.
[8] Gunn, J. B., 1963, Solid State Commun. 1, 88.
[9] Kurosawa, T., 1966, in Proceedings of the International Conference on the
Physics of Semiconductors, Kyoto, J. Phys. Soc. Jpn. Suppl. 21, 424.
[10]Budd H., 1966, in Proceedings of the International Conference on the Physics
of Semiconductors, Kyoto, J. Phys. Soc. Jpn. Suppl. 21, 420.
[11]Price, P. J., 1968, in Proceedings of the 9th
International Conference on the
Physics of Semiconductors, edited by S. M. Ryvkin, p. 173.
[12]Rees, H. D., 1969, J. Phys. Chem. Solids 30, 643.
[13]Fawcett, W., D. A. Boardman, and S. Swain, 1970, J. Phys. Chem. Solids 31,
1963.
[14]A.K. Saxena, M. Mudares, Journal of Applied Physics, Vol.58 Issue 7.
[15]Buslenko, N. P., D. I. Golenko, Yu. A. Schrieder, I. M. Sobol, and V. G.
Sragovich, 1966, “The Monte Carlo method: The Method of Statistical Trials”,
Vol. 87.
[16]Meyer, H. A., 1956, Ed. Symposium on Monte Carlo methods (Wiley, NY).

66
[17]Marchuk, G. I., G. A. Mikhailov, M. A. Nazaraliev, R. A. Darbinjan, B. A.
Kargin, and B. S. Elpov, 1980, “The Monte Carlo Methods in Atmospheric
Optics,” Vol. 12 of Springer Series in Optical Sciences (Springer, Berlin).
[18]D. K. Ferry, “Semiconductors” (Macmillan, New York, 1991).
[19] K. Tomizawa, “Numerical Simulation of Submicron Semiconductor Devices”,
(Artech House, Boston, 1993).
[20] H. D. Rees, J. Phys. C5, 64 (1972).
[21] H. D. Rees, Solid State Commun. A26, 416 (1968).
[22] D. Vasileska, S. M. Goodnick, “Computational Electronics” , Morgan and
Claypool.
[23] “Semiconductor Transport”, D. Vasileska,
http://vasileska.faculty.asu.edu/EEE534.html.
[24] M. Costato, J. Phys. C: Solid State Phys. 5 159 (1972).
[25] J. Bardeen and W. Schockley, Phys. Rev. B, 80, 72-80 (1950).
[26] D. K. Ferry, Phys. Rev. B, Vol. 14, 1605 (1976).
[27] C. Trallero Giner and F. Comas, Phys. Rev. B, 37, 4583-4588 (1988).
[28] Karl Hess, “Monte Carlo device simulation: full band and beyond”, Kluwer
Academic Publishers.
[29] Jacoboni, C., C. Canali, G. Ottaviani, and A. A. Quaranta, Solid State
Electron. 20, 2(1977) 77-89.
[30] Jacoboni C., F. Nava, C. Canali and G. Ottaviani, Phys. Rev. B24, 2 (1981)
1014-1026.
[31] Blakemore, J. S., J. Appl. Phys. 53, 10 (1982) R123-R181.
[32] Pozhela, J. and A. Reklaitis, Solid State Electron. 23, 9 (1980) 927-933.
[33] Michael McLennan (2005), "Add Rappture to Your Software Development,"
http://nanohub.org/resources/240.

Masters_Raghu

More Related Content

Similar to Masters_Raghu (20)

Masters_Raghu