Unterschrift des Betreuers

DIPLOMARBEIT

Numerical Solution of the

Boltzmann Transport Equation
using Spherical Harmonics Expansions

Ausgeführt am

Institut für Analysis und Scientific Computing

der Technischen Universität Wien

unter der Anleitung von

Univ.Prof. Dr.rer.nat. Ansgar Jüngel

durch

Karl Rupp
Eslarngasse 27/19
1030 Wien

Matrikelnummer 0325941
Studienkennzahl 864

Datum Unterschrift
 Abstract

This thesis deals with the numerical solution of the Boltzmann transport equation using an
expansion of the distribution function in spherical harmonics for the purpose of electronic
device simulation. Both the mathematical and physical backgrounds are discussed, then
the Boltzmann transport equation is projected onto spherical harmonics without posing
unnecessary restrictions on the energy band structure. From entropy principles a stabilisa-
tion is found which serves as a Scharfetter-Gummel-like stabilisation for the discretisation.
The finite volume method using a full Galerkin scheme is proposed for the discretisation of
the projected equations, which has the advantage of ensuring current continuity by virtue
of construction. To reduce computational costs and speed up the assembly of the system
matrix, analytical formulae for the integral terms in the discretised equations are derived.
Complexity analysis shows that higher order spherical harmonics expansions suffer from
huge memory requirements, especially for two and three dimensional devices. A compressed
matrix storage scheme is therefore introduced, which reduces the memory requirements for
the storage of the system matrix especially for higher order spherical expansions by up to
several orders of magnitude.
Finally, simulation results for a n+ nn+ -diode prove the applicability of the full Galerkin
method. Self-consistent solutions are obtained by coupling the system of projected equa-
tions with the Poisson equation. The resulting systems of linear equations turn out to be
poorly conditioned, thus preconditioners are proposed and compared.
Contents

Introduction 1

1 Spherical Harmonics 3
1.1 The Laplace Equation in Polar Coordinates . . . . . . . . . . . . . . . . . . 3
1.2 Legendre Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Associated Legendre Functions . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Orthogonal Functions on the Sphere . . . . . . . . . . . . . . . . . . . . . . 12
1.5 The Spherical Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . 17

2 The Boltzmann Transport Equation 22

2.1 Derivation of the Transport Equation . . . . . . . . . . . . . . . . . . . . . 22
2.2 Collision Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Multi-Band Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Average Quantities and Moments . . . . . . . . . . . . . . . . . . . . . . . . 32

3 Spherical Projection of the Boltzmann Transport Equation 36

3.1 The Spherical Harmonics Expansion Equations . . . . . . . . . . . . . . . . 36
3.2 Parity Splitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Entropy Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4 Discretisation 49
4.1 The Finite Volume Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 The Scharfetter-Gummel Discretisation . . . . . . . . . . . . . . . . . . . . 52
4.3 Discretisation of the SHE of the BTE . . . . . . . . . . . . . . . . . . . . . 54
4.4 Rectangular Grids in Two Dimensions . . . . . . . . . . . . . . . . . . . . . 58
4.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5 Reduction of Memory Requirements 66

5.1 The Naive Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.2 Sparsity of Coupling Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Matrix Factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6 Implementation and Results 88

6.1 Problem Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.2 Pitfalls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.3 Results for a n+ nn+ Diode . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

i
CONTENTS ii

7 Outlook and Conclusion 102

7.1 Unstructured and Adaptive Grids . . . . . . . . . . . . . . . . . . . . . . . . 102
7.2 Adaptive Control of the Expansion Order . . . . . . . . . . . . . . . . . . . 103
7.3 Higher Spatial Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.4 Mathematical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.5 Appropriate Linear Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.6 Finite Element Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Bibliography 109
List of Figures

1.1 Selected spherical harmonics. . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.1 Illustration of trajectories in position-momentum space . . . . . . . . . . . . 27

2.2 Multi-band structure of silicon. . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.1 Two methods for the construction of a dual box tessellation from a triangu-
lation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Labels for the Finite Volume Method . . . . . . . . . . . . . . . . . . . . . . 53
4.3 The Bernoulli function Bern(x). . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Labels associated with a rectangular box Bi,j centered at (xi , εj ). . . . . . . 59

6.1 Connection of an n+ nn+ -diode with a MOSFET in inversion . . . . . . . . 89

6.2 Eigenvalues of the system matrix with different preconditioners. . . . . . . . 95
6.3 Comparison of eigenvalues of the system matrix without and with elimination
of odd order unknowns. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4 Self-consistent solution of SHE equations at an applied voltage of 0.3 V. . . 97
6.5 Self-consistent solution of SHE equations at an applied voltage of 1.0 V. . . 98
6.6 Norm of the potential update vector k∆ψs k after each Gummel iteration step
for different externally applied voltages. . . . . . . . . . . . . . . . . . . . . 99
6.7 The potential and its correction at different Gummel iteration steps. . . . . 100
6.8 Coarse grid solutions for a n+ nn+ diode at an applied voltage of 1.0 V. . . 100
6.9 Comparison between electron concentrations obtained from self consistent,
non self consistent and drift-diffusion simulations. . . . . . . . . . . . . . . . 101

7.1 Construction of a prismatic mesh from a lower-dimensional one. . . . . . . . 104

iii
List of Tables

2.1 Parameters for the analytical band structure. . . . . . . . . . . . . . . . . . 32

6.1 Input parameters for the simulation of an n+ nn+ diode. . . . . . . . . . . . 90

6.2 Comparison of condition numbers with different preconditioners. . . . . . . 94

iv
Acknowledgement

Many thanks go to Prof. Ansgar Jüngel for his cooperativeness with the interdisciplinary
topic of this thesis. He has always been eager to hear about the latest progress of my work
and provided me with the highly appreciated freedom that allowed me – driven by my own
curiosity – to find the analytical and numerical results.
I am also indebted to Prof. Klaus-Tibor Grasser from the Institute for Microelectronics.
I remember when he suggested my investigation of the spherical harmonics expansion within
this thesis, then raised his eyebrows and jokingly said that a scientific hybrid like me would
definitely enjoy this topic. Yes, he was right.
A hearty thanks to Philipp Dörsek, who provided the cover page of this thesis, and
Zlatan Stanojevic, who helped me with the interpretation of the simulation results from
the physics point of view.
During my study I have learnt to know many course mates that are now good friends.
Many thanks for making my studies even more pleasurable and for spending a lot of spare-
time together.
I also wish to thank my parents, grandparents and sisters for the support during my
study, even though I have hardly found the time to travel back home to Southern Styria
and stay there for longer than a few days.
Last but not least I thank my girlfriend Doris for all the understanding every time I
went to university earlier than planned just to come back later than expected.

v
Notation

Different notations in the mathematics and the physics community often leads to headache
if a publication from the ’other’ community is read. The notation used in this thesis is a
compromise and tries to fit into both worlds.

Symbol Meaning
R set of real numbers
k·k norm of the argument
|·| modulus of the argument
O(·) Landau symbol
Ω unit sphere
θ colatitude (polar angle)
ϕ longitude (azimuth)
dΩ unit sphere surface element, sin θdθdϕ
eθ , eϕ unit vectors w.r.t. θ and ϕ
∇ Nabla operator
∇ · u = div(u) divergence of u
∆ Laplace operator
∆∗ Beltrami operator
A, B matrices (captial letters, bold)
v, w vectors (lower case)
Pl Legendre polynomial of degree l
Plm associated Legendre function of major index l and minor
index m
Yl,m real valued spherical harmonic of major index l and minor
index m
Xl,m spherical harmonics expansion coefficient to Yl,m of the
quantity X
FΩ {f } spherical Fourier transform
δi,j Kronecker delta
δ(·) delta distribution
T triangulation of the simulation domain
B dual box grid of T
B a box from the dual box grid B
Ps space of (piecewise) polynomials of degree s
Bern x Bernoulli function

N number of degrees of freedom in (x, ε)-space

L spherical harmonics expansion order

vi
NOTATION vii

Symbol Meaning
~ Planck constant divided by 2π
 permittivity (not to be confused with energy ε)
ρ space charge
|q| elementary charge
m∗ effective mass
ε energy (not to be confused with permittivity )
x spatial coordinate within the device
k wave vector
p momentum vector
v(k) (group) velocity
F force
E electric field
Q{·} scattering operator
Qin {·} in-scattering operator
Qout {·} out-scattering operator
QVR {·} velocity-randomising scattering operator
η index of the scattering process
s(x, k, k0 ) scattering rate
n(x, t) particle density
g(ε, θ, ϕ) generalised density of states
j(x, t) current density
h·i moment with respect to the argument
f (x, k, t) distribution function
f˜(x, k, t) generalised energy distribution function
H(f, x, k) entropy
Hl (f, x, k) entropy factor
v l,m (ε) projection of the (group) velocity
Γl,m (ε) spherical harmonics coupling term
VT thermal voltage
Introduction

Semiconductor devices have experienced a tremendous progress in the past decades, follow-
ing Moore’s Law which states that the number of components on a chip doubles approxi-
mately every two years. Since the chip sizes are held more or less constant, Moore’s Law
also states that feature sizes shrink exponentially. At present, MOSFETs with a channel
length of only 20nm are in production.
Due to the high level of miniaturisation, computer aided design of semiconductor devices
is widely used to optimise device layouts. While in the early years of the semiconductor
industry macroscopic models such as the drift diffusion model or the hydrodynamic model
have been sufficient for device simulation, this is not the case anymore for the small feature
sizes used today. As long as quantum mechanical effects can be neglected, the microscopic
behaviour of electrons is in accordance to the Boltzmann Transport Equation (BTE), which
is considered to be the best classical description of electrons.
A direct solution of the BTE has been pursued for several decades and many ingenious
techniques have been developed for this purpose. The most commonly used is the Monte
Carlo method, primarily because the method is very flexible and allows details such as
complicated band structures to be easily incorporated. The main disadvantage of the
Monte Carlo method is its computational expense, especially when attempting to reduce
the statistical noise in the low density tails of the distribution function.
As an alternative to the Monte Carlo method, the spherical harmonics expansion (SHE)
method is considered in this thesis. In contrast to the former, the latter yields determin-
istic solutions of the BTE and is computationally less expensive. The spherical harmonics
expansion method for the simulation of one-dimensional semiconductors was pioneered in
the early 1990’s [15, 17] and since then several scientists around the world have improved
the method in several ways. We mention extensions to higher expansion orders [20, 21],
higher spatial domains [16, 49] and more recently a stabilisation scheme [24]. The spheri-
cal harmonics expansion has also been encountered by mathematicians who have applied
several limits to the BTE and the spherical harmonics expansion in order to derive simpler
models [1, 4, 10, 12].
This thesis is organised as follows: In the first chapter the mathematical background of
spherical harmonics is presented. Most results are known for more than a century, except
for the modern approach of the spherical Fourier transform given in Section 1.5. The proof
of Thm. 13 is contributed by the author of this thesis, mainly because it was not given in
the referenced text-book.
In Chapter 2 the BTE is derived and the individual terms such as the scattering operator
introduced. In contrast to the first chapter, where the emphasis is the introduction of a
solid mathematical foundation, the second chapter aims at the introduction of the necessary
topics from physics. There are no additional contributions by the author in this chapter.

1
INTRODUCTION 2

The BTE is projected onto spherical harmonics in Chapter 3. Additionally, several

intrinsic properties of the BTE are shown to be conserved after the projection onto spherical
harmonics. Finally, the concept of entropies for the continuous SHE equations is introduced,
which later serves as a stabilisation scheme for the discretised equations. The projection
onto spherical harmonics follows the calculations from Jungemann et. al. [24], where several
details of the calculation are filled in. Section 3.3 is compiled from several publications by
Ringhofer [34–37].
Chapter 4 deals with the discretisation of the spherical harmonics expansion equations.
At first a short introduction to the finite volume method is given, then the Scharfetter-
Gummel scheme is briefly discussed to show the necessity of stabilisation schemes for semi-
conductor device modelling. The full Galerkin scheme for the discretisation of the spherical
harmonics expansion equation in the simulation domain as well as the analytical formulae
for simulations in two dimensions are contributed by the author of this thesis.
Based on the Galerkin scheme introduced in Chapter 4, complexity analysis is given
in Chapter 5, which is the author’s own work. On the one hand the coupling terms for
coefficients of spherical harmonics of different order are shown to be sparse for spherical
energy bands, while on the other hand it is shown that the memory requirements can be
reduced considerably if the resulting system matrix is stored in a factored form. Matrix-
vector multiplication is still possible with similar computational costs, which enables the
use of iterative solvers for the solution of the system of linear equations.
In Chapter 6 some numerical results are given. It turned out that the SHE equations
are much harder to implement than for example the drift-diffusion model, from which the
initial guess for the potential is obtained. Moreover, preconditioners are observed to be
necessary to keep the condition number of the resulting system sufficiently low.
Chapter 7 gives an outlook on further ideas for future work on this topic. This thesis
then closes with a conclusion.
Chapter 1

Spherical Harmonics

Many physical quantities such as the potential induced by a charged particle show a radial
symmetry, so that it is often convenient to use a spherical coordinate system. Thus, the
Laplace equation in spherical coordinates was investigated in the 18th century, which has
lead to so-called spherical harmonics. Conversely, spherical harmonics can be used for
the approximation of functions on the sphere, thus they can be seen as three-dimensional
extensions of sines and cosines used in Fourier analysis.
This chapter is devoted to the mathematical background of spherical harmonics and
collects all important mathematical results for later use on the numerical solution of Boltz-
mann’s transport equation using an expansion method. Most of the derivations and results
have been known for more than a century, except for the last section, where newer results
are collected. The reader is referred several times throughout this chapter to the literature
for some of the more technical or lengthy proofs.

1.1 The Laplace Equation in Polar Coordinates

Gauss’ Law states that the electric flux density D is related to the space charge density ρ
by

∇·D =ρ (1.1)

Using the material equation D = E, where  is the permittivity and E is the electric field,
one finds with E = −∇ψ the governing equation for the electric potential ϕ:

−∇ · (∇ψ) = ρ , (1.2)

which is commonly termed Poisson equation. If the permittivity  is a scalar, the homoge-
neous form is known as Laplace’s equation

∆ψ = 0 . (1.3)

In the following we write u instead of ψ, following mathematical conventions. The explicit

form of the Laplace operator ∆ := ∇ · ∇ depends on the underlying coordinate system, in
Cartesian coordinates (x, y, z) the operator is given as
∂2 ∂2 ∂2
∆= 2
+ 2+ 2 . (1.4)
∂x ∂y ∂z