Modeling and Optimization of High Aspect Ratio Plasma Etching

by

Florian Krüger

A dissertation submitted in partial fulfillment

of the requirements for the degree of
Doctor of Philosophy
(Electrical and Computer Engineering)
in the University of Michigan
2024

Doctoral Committee:

Professor Mark J. Kushner, Chair

Professor Karthik Duraisamy
Associate Professor Bryan R. Goldsmith
Professor L. Jay Guo
 Florian Krüger

[email protected]

ORCID iD: 0000-0003-2761-905X

© Florian Krüger 2024

 Dedication

To Daria. Obviously.

ii
 Acknowledgements

Firstly, I wish to thank my supervisor Mark Kushner who has been a constant and

inexhaustible source of knowledge and advice. Scientific rigor, integrity and work ethic are best

taught by example, and I can only hope to emulate some of the dedication to those principles I

have witnessed over the last years.

Secondly, I would like to thank my thesis committee members Professor Karthik

Duraisamy, L. Jay Guo and Bryan R. Goldsmith. I would also like to thank my project collaborators,

Hyunjae Lee (Samsung), Sang Ki Nam (Samsung), Mingmei Wang (Lam Research/Tokyo

Electron), Du Zhang (Tokyo Electron), Minjoon Park (Tokyo Electron), Andrew Metz (Tokyo

Electron) and Pingshan Luan (Tokyo Electron) for their support as well as valuable insight and

data.

I am indebted to the entirety of the support staff at the University of Michigan

with special thanks Julia Falkovitch-Khain who continually shielded me from all bureaucratic

trials and tribulations so I could wholly focus on my academic work.

I wish to acknowledge all fellow members of the Computational Plasma Science and

Engineering Group at the University of Michigan, who have shared with me their time, experience

and, most important of all, many (probably too long) lunch breaks, some of which helped sustain

my sanity more than I care to admit. In alphabetical order: Amanda Lietz, Chenhui Qu, Evan Litch,

Eve Lanham, Jordyn Polito, Juliusz Kruszelnicki, Kseniia Konina, Mackenzie Meyer, Sanjana

Kerketta, Scott Doyle, Shuo Huang, Tugba Piskin, Xifeng Wang, Yifan Gui.

iii
 Because my education did not begin in Michigan, I want to acknowledge my previous

advisors Jan Trieschmann, Julian Schulze, Ralf Peter Brinkmann, Sebastian Wilczek and Thomas

Mussenbrock who introduced me to the world of science and quite literally, changed my life’s

trajectory.

Thank you to my parents and family who supported me and made sure that my educational

and personal life stayed on track although, at times, I tried my best to derail it.

Last but not least, I must thank my extraordinary wife, Daria, for her love, patience, support

and trust without which I would have achieved nothing. While this piece of writing may be the

token of one chapter’s end, I can’t wait for countless more.

iv
 Table of Contents

Dedication.................................................................................................................................. ii

Acknowledgements ................................................................................................................... iii

Table of Contents ........................................................................................................................ v

List of Tables .............................................................................................................................. x

List of Figures ............................................................................................................................ xi

List of Appendices .................................................................................................................. xvii

List of Abbreviations..............................................................................................................xviii

List of Symbols ........................................................................................................................ xix

Abstract .................................................................................................................................. xxii

Chapter 1 : Introduction .............................................................................................................. 1

1.1 The Fourth State of Matter ...............................................................................................1

1.2 Basic Properties of Low-Temperature Plasmas ................................................................4

1.3 Boltzmann’s Equation .....................................................................................................4

1.4 Conservation Equations ...................................................................................................6

1.4.1 Particle Conservation ..............................................................................................6

1.4.2 Momentum Conservation ........................................................................................6

1.4.3 Energy Conservation ...............................................................................................7

1.5 Maxwell–Boltzmann Distribution ....................................................................................7

1.6 Thermal Non-Equilibrium ...............................................................................................8

1.7 Electron Collision Cross Sections ....................................................................................9

1.8 Rate Coefficients and Collision Rates ..............................................................................9

v
1.9 Quasineutrality .............................................................................................................. 10

1.10 Ambipolar Diffusion .................................................................................................... 11

1.11 Plasma Sheath ............................................................................................................. 12

1.12 Capacitively Coupled Plasmas ..................................................................................... 14

1.13 Electrical Characteristics ............................................................................................. 15

1.14 Heating Mechanisms ................................................................................................... 18

1.15 Power Matching .......................................................................................................... 19

1.16 Dual and Multi Frequency RF Power ........................................................................... 21

1.17 Plasma Asymmetry and DC self-bias ........................................................................... 22

1.18 Generalized Asymmetry .............................................................................................. 25

1.19 Electrical Asymmetry .................................................................................................. 27

1.20 Voltage Waveform Tailoring ....................................................................................... 28

1.21 Ion Energy Distributions .............................................................................................. 30

1.22 Electronegative Plasmas .............................................................................................. 32

1.23 Plasma in Semiconductor Processing ........................................................................... 34

1.24 Adsorption................................................................................................................... 39

1.24.1 Physisorption ...................................................................................................... 39

1.24.2 Chemisorption .................................................................................................... 40

1.25 Etching and Sputtering................................................................................................. 40

1.25.1 Direct Sputtering ................................................................................................. 41

1.25.2 Direct Chemical Etching ..................................................................................... 42

1.25.3 Ion Enhanced Etching ......................................................................................... 43

1.26 Aspect Ratio Dependent Etch ...................................................................................... 44

1.27 Intra Feature Charging ................................................................................................. 46

1.28 Sidewall Passivation .................................................................................................... 48

vi
 1.29 Computational Modeling ............................................................................................. 49

1.30 Goals and Scope of this Dissertation ............................................................................ 51

1.31 References ................................................................................................................... 53

Chapter 2 : Description of the Models ....................................................................................... 63

2.1 Hybrid Plasma Equipment ............................................................................................. 64

2.1.1 Electron Energy Transport Module (EETM) ......................................................... 65

2.1.2 Fluid Kinetics Poisson Module (FKPM)................................................................ 67

2.1.3 Surface Kinetics Module (SKM) ........................................................................... 72

2.1.4 Plasma Chemistry Monte Carlo Module (PCMCM) .............................................. 73

2.2 Monte Carlo Feature Profile Model (MCFPM) .............................................................. 74

2.2.1 Energetic Particle Surface Reaction ...................................................................... 75

2.2.2 Electrostatic Charging ........................................................................................... 76

2.2.3 Crosslinking Module ............................................................................................. 77

2.3 References ..................................................................................................................... 80

Chapter 3 : Electric Field Reversals Resulting from Voltage Waveform Tailoring in Ar/O2
Capacitively Coupled Plasmas Sustained in Asymmetric Systems ............................................. 81

3.1 Introduction ................................................................................................................... 81

3.2 The Ar Base Case .......................................................................................................... 86

3.3 EFR with Ar/O2 Gas Mixtures ....................................................................................... 92

3.4 Concluding Remarks ................................................................................................... 102

3.5 References ................................................................................................................... 105

Chapter 4 : Voltage Waveform Tailoring for High Aspect Ratio Plasma Etching of SiO2 using
Ar/CF4/O2 Mixtures: Consequences of Ion and Electron Distributions on Etch Profiles .......... 107

4.1 Introduction ................................................................................................................. 107

4.2 Description of the Reactor Scale Model ....................................................................... 108

4.3 Description of the Feature Scale Model ....................................................................... 110

vii
 4.4 Plasma and Etch Properties for 1000 W of VWT Power .............................................. 113

4.5 Plasma and Etch Properties for 2000 W of VWT Power .............................................. 127

4.6 Concluding Remarks ................................................................................................... 135

4.7 References ................................................................................................................... 138

Chapter 5 DC Self-Bias and Ion Energy at Very Low Driving Frequencies. ............................ 139

5.1 Introduction ................................................................................................................. 139

5.2 Simulation Setup and Geometry .................................................................................. 140

5.2.1 Reactor Scale Setup ............................................................................................ 140

5.2.2 Feature Scale Setup ............................................................................................. 142

5.3 Reactor Scale Plasma Properties with Constant Applied Voltage ................................. 144

5.4 Reactor Scale Plasma Properties with Constant Applied Power ................................... 155

5.5 SiO2 Etching at Constant Power................................................................................... 163

5.6 Concluding Remarks ................................................................................................... 166

5.7 References ................................................................................................................... 168

Chapter 6 Autonomous Hybrid Optimization of a SiO2 Plasma Etching Mechanism................ 171

6.1 Introduction ................................................................................................................. 171

6.2 Description of the Optimization Scheme ...................................................................... 175

6.3 Reactor Scale Plasma Simulation ................................................................................. 177

6.4 Surface Simulation Setup............................................................................................. 181

6.5 Target Metrics and Loss Function ................................................................................ 186

6.6 Gradient Descent ......................................................................................................... 189

6.7 Nelder-Mead Optimization .......................................................................................... 195

6.8 Hybrid GD-NM Optimization ...................................................................................... 198

6.9 Transferability of the Converged Mechanism............................................................... 203

6.9.1 Variation of O2 in Flow ....................................................................................... 204

viii
 6.9.2 Variation of Low Frequency Power..................................................................... 207

6.10 Optimization Based on Multiple Features .................................................................. 209

6.11 Concluding Remarks ................................................................................................. 214

6.12 References ................................................................................................................. 216

Chapter 7 Summary and Outlook ............................................................................................ 224

7.1 Summary ..................................................................................................................... 224

7.2 Future Work ................................................................................................................ 226

Appendices ............................................................................................................................. 229

ix
 List of Tables

Table 6.1: Base case fluxes to wafer ........................................................................................ 180

Table 6.2: High-level description of the SiO2 and AC etch mechanism. ................................... 185

Table 6.3: Tuning parameters for gradient descent optimization .............................................. 191

Table 6.4:Target metrics for gradient descent optimization ...................................................... 191

Table 6.5: Final tuning parameters after coupled optimization. ................................................ 200

Table 6.6: Final tuning parameters after coupled optimization of feature grid. ......................... 211

Table B.1: Complete surface reaction mechanism ................................................................... 232

x
 List of Figures

Figure 1.1: States of matter .........................................................................................................2

Figure 1.2. Plasma parameter space [16]. ....................................................................................3

Figure 1.3: Maxwell-Boltzmann distribution corresponding to three different temperatures. ........7

Figure 1.4: Electron impact cross sections for e-Ar collisions. Reproduced from Petrović et al. ...9

Figure 1.5: Schematic of a plasma bounded by two parallel grounded metal plates and the
resulting density b) as well as potential c) distribution throughout the gap. ................................ 12

Figure 1.6: Schematic of a capacitively coupled plasma and the corresponding potential
distribution at 4 different times of the RF period. ...................................................................... 14

Figure 1.7: Circuit diagram of capacitively coupled plasma. ...................................................... 18

Figure 1.8: Circuit diagram of the matching network in conjunction with the plasma. ............... 20

Figure 1.9: Schematic of an asymmetric CCP with two different electrode sizes with A 1 < A2
and the blocking capacitor CB at the powered electrode. ............................................................ 22

Figure 1.10: Voltage waveforms VVWT(t) based on (1.56) for 𝜑 = 0, 45, 90, 135 and 180°. ...... 29

Figure 1.11 a) IEDs at the powered electrode of a rf CF 4 discharge driven at 13.56 MHz from
Kuypers and Hopman [129]. b) Particle in Cell simulation results showing IEDs of helium ions
hitting the target electrode of rf discharges driven at frequencies from 1 MHz to 100 MHz.
Reproduced from Kawamura et al [130]. ................................................................................... 32

Figure 1.12: Schematic representation of a MOSFET ................................................................ 34

Figure 1.13: Manufacturing of strucured integrated circuit devices through deopsition, etch and
pattern transfer: a) metal film deposition; b) photoresist deposition; c) patterned optical
exposure; d) photoresist development; e) film etch; f) photoresist removal. Based on [16]. ....... 35

Figure 1.14: a) Isotropic etch processes result in an uncontrolled and ultimately undesirable
feature. b) Ion bombardment enables anisotropic etching. ......................................................... 37

Figure 1.15: Cross section SEM image of memory channel hole pattern after etching [131]. ..... 38

Figure 1.16: Potential diagram of a generic physisorption and chemisorption process. .............. 39

xi
Figure 1.17: Different types of material removing surface reactions. a) physical sputtering, b)
thermal etching, c) chemical removal of ion activated species and d) physical removal of
chemically modified species. ..................................................................................................... 41

Figure 1.18: Etching yield of polysilicon and silicon dioxide by 100 eV Ar and Cl as a function
of ion incident angles for saturated chlorine fluxes. Reproduced from Chang et al [93]. ............ 42

Figure 1.19: Critical ion angle of incidence for direct hit on the etch front for a) low and b) high
aspect ratio feature. ................................................................................................................... 45

Figure 1.20: Surface charge effects responsible for notching, trenching and twisting in HAR etch
processes. .................................................................................................................................. 46

Figure 1.21: Sidewall passivation during the feature etch process. ............................................. 48

Figure 2.1: Flowchart of coupled modeling suite consisting of HPEM and MCFPM.................. 63

Figure 2.2: Schematic representation of the crosslinking process. .............................................. 78

Figure 3.1: Geometry and circuit used in the model. a) Schematic of the dual frequency
capacitively coupled plasma reactor used in this investigation. b) Equivalent circuit diagram of
the reactor. Source power at 80 MHz is applied to the top electrode. The 1 MHz tailored
waveform is applied to the bottom electrode.............................................................................. 84

Figure 3.2: Plasma properties as a function of phase angle for the pure argon plasma with a
constant high frequency power of 100 W and constant bias power of 1 kW. a) Applied voltage
amplitudes for the high frequency source (Vhf) and amplitude of the low frequency bias (V0).
Top and bottom electrode. b) DC self-bias. ............................................................................... 87

Figure 3.3: Plasma properties for the pure argon plasma averaged across the wafer as a function
of height above the wafer and time during the low frequency,1 MHz cycle. Values are shown
(left-to-right) for phase angles of φ = 0, 45, 90, 135, 180°. a) Axial component of the electric
field, b) Ar+ density, c) electron density and d) applied voltage waveform on the substrate. The
Ar+ and electron densities are plotted on a log-scale spanning 2 decades with the maximum value
noted in each image................................................................................................................... 88

Figure 3.4: Properties of electrons incident onto the wafer for the pure argon plasma. a) Electron
energy and angular distributions for different phase angles plotted on a log-scale spanning 2
decades. b) Mean electron energy and mean electron angle of incidence as a function of phase
angles. ....................................................................................................................................... 91

Figure 3.5: Spatially averaged densities of Ar+, e-, O2+ and O-, and electronegativity of Ar/O2
plasmas as a function of the O2 mole fraction with a phase angle of φ = 0°. ............................... 93

Figure 3.6: Plasma properties as a function of phase shift φ for different Ar/O 2 ratios when
keeping source and bias powers constant. a) DC self-bias produced on the powered substrate. b)
Applied voltage amplitude V0. .................................................................................................. 94

xii
Figure 3.7: Radial average of the vertical component of the electric field as a function of time
during a single 1 MHz cycle and distance from bottom electrode different gas mixtures. Ar/O2 =
a) 100/0, b) 90/10, c) 75/25, d) 50/50. e) Normalized applied voltage V0. Results are shown for
phase shifts of (left-to-right columns) φ = 0, 45, 90, 135 and 180°. The dotted line is for E = 0,
the separation between negative electric fields pointing down and positive pointing up. ............ 95

Figure 3.8: Plasma properties for Ar/O2 = 50/50 and a phase shift of φ = 0°. a) Radial average of
the Ar+ density as a function of time during a single 1 MHz RF cycle plotted on a log-scale over
2-decades. b) O- density (log-scale, 2-decades) and c) electron density (log-scale 3-decades). The
maximum density is noted in each image. d) e) Normalized voltage amplitude V0..................... 98

Figure 3.9: Electron energy and angular distributions incident onto the wafer surface for different
Ar/O2 ratios plotted on a log-scale over 2 decades with phase shifts of a) φ = 0° and b) φ = 180°.
The EAD increases in energy and narrows in angle with increasing O2 fraction, an effect that is
most prominent for a) φ = 0°. .................................................................................................. 100

Figure 3.10: Properties of the electron velocity distribution incident onto the wafer for different
Ar/O2 ratios as a function of phase shifts φ. a) Mean electron energy and b) mean angle of
incidence. Decreasing phase angle increases the incident electron energy and narrows the angular
distribution, with the effect being most prominent for large O2 mole fractions. ........................ 101

Figure 4.1: Schematics of the reactor. a) The two frequency CCP has the VWT power applied to
the lower electrode and a sinusoidal, 80 MHz waveform applied to the top. b) Electrical
schematic of the system. From F. Kruger et al, Plasma Sources Sci. Technol. 30, 085002 (2021).
Reproduced by permission of IOP Publishing Ltd. .................................................................. 109

Figure 4.2: Initial geometry used in for feature profile simulations. A 700 nm mask defines a 100
nm opening to 3000 nm thick SiO2. The simulations are for trench etching, performed in 3-
dimensions having a finite depth. ............................................................................................ 112

Figure 4.3: Voltage amplitudes for the VWT bias (Vbias) and top power (Vtop), and DC self-bias
as a function of phase angle. a) PVWT = 1000 W and b) PVWT = 2000 W. ................................. 114

Figure 4.4: Vertical electric field component as a function of time and distance from wafer for
PVWT = 1000 W for phase angles of a) 0°, b) 45°, c) 90°, d) 135° and e) 180°. The shape of the
voltage waveform applied to the bottom electrode is shown in each image. ............................. 116

Figure 4.5: Properties of ions and electrons incident onto the wafer for PVWT = 1000 W. a) IEAD
and b) EEAD for phase angles of φ = 0°, 45°, 90°, 135° and 180°. The distributions are separately
normalized and plotted on a log-scale over 2 decades. c) Mean ion energy as a function of phase
angle φ .................................................................................................................................... 118

Figure 4.6: Fluxes of reactive species onto the wafer as a function of phase angle φ. a) Neutrals
and b) ions. ............................................................................................................................. 119

Figure 4.7: Feature properties for PVWT = 1000 W following 15 minutes of etching a) Feature
profiles for different phase angles (φ = 0°, 45°, 90°, 135° and 180°). b) Comparison of etch result
and electric potential for (left) self-consistent VWT produced EEAD ...................................... 121

xiii
Figure 4.8: Spatial distribution of average electron energies incident on the surface of the feature.
The feature properties are for PVWT = 1000 W and φ = 0° after 15 minutes of etching. Results are
shown for initial EEADs produced with VWT and for thermal electron having a temperature. of
3 eV. a) When including acceleration in electric fields produced by feature charging. b) Without
acceleration due to feature charging. The view is looking obliquely into the 3-dimensional
feature having reflective boundary conditions front-and-back. ................................................. 126

Figure 4.9: Vertical electric field component as a function of time and distance from wafer for
PVWT = 2000 W for phase angles of a) 0°, b) 45°, c) 90°, d) 135° and e) 180°. The shape of the
voltage waveform applied to the bottom electrode is shown in each image. ............................. 129

Figure 4.10: Electron density for PVWT = 2000 W for phase angle φ of a) 0°, b) 90° and c) 180°).
The densities are plotted on a log-scale over 2-decades. .......................................................... 130

Figure 4.11: Energy and angular distributions for charged particles incident onto the wafer for
PVWT = 2000 W for phase angles of φ = 0°, 45°, 90°, 135° and 180°. a) Ions and b) electrons.
The distributions are separately normalized and plotted on a log-scale over 2 decades. ............ 131

Figure 4.12: Feature properties for PVWT = 2000 W following 15 minutes of etching a) Feature
profiles for different phase angles (φ = 0°, 45°, 90°, 135° and 180°). b) Comparison of profiles
obtained with VWT generated EEADs and synthetic thermal electron distribution for phase
angles φ of 0°, 90° and 180°. ................................................................................................... 134

Figure 5.1: Reactor geometry and circuit diagram. .................................................................. 141

Figure 5.2: Feature scale model initial conditions for simulating plasma etching of SiO2. PR
represents photoresist. ............................................................................................................. 144

Figure 5.3: Plasma properties of the Ar/CF4/O2 CCP base case with VT = 300 V, V0 = 1500 V,
f0 = 1 and φ = 0°. Time averaged a) electron density ne, b) ionization source by bulk electrons
Sb, c) ionization source by secondary electrons Ss, and d) electron temperature, Te. ................. 145

Figure 5.4: Average electron density as a function of phase shift φ for fundamental frequencies
f0 = 1 MHz to 10 MHz at constant applied voltage................................................................... 146

Figure 5.5: Power deposition for a) top electrode and b) bottom electrode as a function of phase
shift φ for fundamental frequencies f0 = 1 MHz to 10 MHz. .................................................... 147

Figure 5.6: Plasma properties as a function of phase shift φ for different fundamental frequencies
f0 = 1 MHz to 10 MHz for constant voltage. a) Vdc. b) Mean ion energy incident onto the wafer.
............................................................................................................................................... 149

Figure 5.7: Sheath properties as a function of fundamental frequency f0 for φ = 0° at constant

voltage. a) Sheath thickness ds and b) sheath ion inertia coefficient S...................................... 152

Figure 5.8: Incident average energy of select ion species as a function of mass. ...................... 153

xiv
Figure 5.9: IEADs as a function of phase shift φ for fundamental frequencies a) f0 = 1 MHz and
b) f0 = 10 MHz at constant applied voltage. Values are plotted on a 2-decade log scale. .......... 154

Figure 5.10: Applied voltage for constant power (2000 W) as function of phase shift φ for
fundamental frequencies f0 = 1 to 10 MHz applied to the bottom electrode. a) Top electrode and
b) bottom electrode. ................................................................................................................ 156

Figure 5.11: Mean electron density as a function of phase shift φ for fundamental frequencies f0
= 1 MHz to 10 MHz at constant applied power........................................................................ 157

Figure 5.12: Sheath properties as a function of fundamental frequency f0 for φ = 0° at constant

power. a) Sheath thickness ds and b) sheath ion inertia coefficient S. ....................................... 158

Figure 5.13: Plasma parameters as a function of phase shift φ for fundamental frequencies f0 = 1
MHz to 10 MHz at constant power. a) DC self-bias and b) Mean ion energy delivered to the
wafer. ...................................................................................................................................... 159

Figure 5.14: IEADs as a function of phase shift φ with constant power for fundamental
frequencies a) f0 = 1 and b) 10 MHz. Values are plotted on a 2-decade log scale. .................... 160

Figure 5.15: Fluxes to the wafer as a function of phase shift φ for fundamental frequencies f0 =
1 to 10 MHz while keeping power constant. a) Total polymerizing flux, b) O-atom flux, c) total
ion flux and d) ratio of polymerizing flux to ion flux. .............................................................. 161

Figure 5.16: Predictions for etch profiles in SiO2 at constant power with fundamental frequency
f0 = 1 MHz and varying phase angles φ. .................................................................................. 163

Figure 5.17: Predictions for etch profiles in SiO2 at constant power with fundamental frequency
f0 = 10 MHz and varying phase angles φ.................................................................................. 165

Figure 6.1: Global schematic of the optimization process. ....................................................... 176

Figure 6.2: CCP reactor geometry and circuit. ......................................................................... 177

Figure 6.3: Plasma properties for the base case: a) Electron density [e], b) electron temperature
Te and c) negative fluorine ion Density [F-], d) ionization rate by bulk electrons, e) ionization
rate by secondary electrons...................................................................................................... 179

Figure 6.4: Ion Energy-Angular Distribution sampled at the wafer surface during the base
case. ........................................................................................................................................ 181

Figure 6.5: Feature scale simulation setup consisting of a SiO2 substrate with an amorphous
carbon mask. ........................................................................................................................... 182

Figure 6.6: Etch features for different O based polymer etch probabilities 0.5% (left) and 2.0%
(right). ..................................................................................................................................... 184

Figure 6.7: Target metrics used to evaluate the experimental matching. ................................... 186

xv
Figure 6.8: SEM image resulting from the base case experiment. ............................................ 188

Figure 6.9: Outline of the gradient descent method for a 1D parameter space. ......................... 189

Figure 6.10: Results from the gradient descent optimization. a) minimal epochal loss. b) values
of select metrics as a function of epochs. c) select tuning parameters as a function of epochs. . 192

Figure 6.11: The finite resolution of the loss function leads to convergence issues. ................. 193

Figure 6.12: Evolution of the total loss for different learning rates η = 1x10-4, 5x10-4, 1x10-3 and
5x10-3...................................................................................................................................... 194

Figure 6.13: Examples of the geometric transformations of the parameter simplex in a 2D

parameter space; reflection, extension; contraction; shrinkage. ................................................ 195

Figure 6.14: Flow chart describing the Nelder-Mead optimization algorithm. .......................... 197

Figure 6.15: Results from the combined gradient descent and Nelder-Mead optimization. a)
minimal epochal loss. b) values of select metrics as a function of epochs. c) select model
parameters as a function of epochs. ......................................................................................... 199

Figure 6.16: Final features after etch completion for different parameter sets as a function of
epochal evolution and the actual experimental target feature. .................................................. 201

Figure 6.17: Fluxes incident on the wafer as a function of the O2/C4F6 feedstock gas ratio. ..... 204

Figure 6.18: a) MCFPM features resulting from the O2/C4F6 variation. b) SEM images of the
features resulting from the corresponding experiment.............................................................. 205

Figure 6.19: IEADs resulting from the Plf variation with Plf = 0, 4, 6 and 8 kW. ...................... 207

Figure 6.20: a) MCFPM features resulting from the Plf variation. b) SEM images of the features
resulting from the corresponding experiment. .......................................................................... 208

Figure 6.21: Feature scale simulation setup consisting of a SiO2 substrate with an amorphous
carbon mask and four separate features in a regular grid. ......................................................... 210

Figure 6.22: Results of the feature array optimization. a) minimal epochal loss. b) values of select
metrics as a function of epochs. c) select model parameters as a function of epochs. ............... 212

Figure 6.23: Final feature grids after etch completion for different parameter sets as a function
of epochal evolution and the actual experimental target feature. .............................................. 213

Figure A.0.1: Statistical run-to-run variation of the final etch profiles for the PVWT = 1000 W and
φ = 0° case. a) 15 min run and b) 30 min to produce over etch into the stop layer. For each series
of profiles, each case had a different seed for the random number generators, in addition to
having the inherent statistical variations in the parallel computing environment. ..................... 231

xvi
 List of Appendices

Appendix A: Statistical Variation in Predicted Feature Profiles ............................................... 229

Appendix B: Surface Reaction Mechanism ............................................................................. 232

xvii
 List of Abbreviations

DC Direct Current
AC Alternating Current
ICP Inductively Coupled Plasma
CCP Capacitively Coupled Plasma
RF Radio Frequency
LF Low Frequency
HF High Frequency
LTP Low Temperature Plasma
IMN Impedance Matching Network
EAE Electrical Asymmetry Effect
VWT Voltage Waveform Tailoring
MOSFET Metal Oxide Field Effect Transistor
PECVD Plasma Enhanced Chemical Vapor Deposition
ALD Atomic Layer Deposition
ALE Atomic Layer Etching
RIE Reactive Ion Etching
PR Photo Resist
AC Amorphous Carbon
HAR High Aspect Ratio
ARDE Aspect Ratio Dependent Etching
HPEM Hybrid Plasma Equipment Model
EETM Electron Energy Transport Module
FKPM Fluid Kinetics Poisson Module
SKM Surface Kinetics Module
SSBM Surface Site Balance Model
PCMCM Plasma Chemistry Monte Carlo Module
MCFPM Monte Carlo Feature Profile Model
HVM High Volume Manufacturing

xviii
 List of Symbols

𝑡 Time
𝑛𝑒 Electron density
𝑛𝑖 Ion density
𝑛𝑔 Neutral gas density
𝑇 Temperature
𝑇𝑒 Electron temperature
𝑇𝑖 Ion temperature
𝑇𝑔 Neutral gas temperature
𝛼 Ionization degree
𝑚 Mass
𝑚𝑒 Electron mass
𝑚𝑖 Ion mass
𝑓 Frequency
𝑓0 Fundamental frequency
𝜔 Angular frequency
𝜔𝑝𝑒 Electron plasma frequency
𝜆𝐷 Debye length
𝑘𝑏 Boltzmann constant
𝑒 Elementary charge
𝜀 Permittivity
𝜀0 Vacuum permittivity
𝜌 Charge density
Φ Electric potential
𝑉 Voltage
𝑉1 Voltage across sheath 1
𝑉2 Voltage across sheath 2
𝑉̅1 Mean voltage across sheath 1
𝑉̅2 Mean voltage across sheath 2
𝑉̂1 Maximum voltage across sheath 1
𝑉̂2 Maximum voltage across sheath 2
𝑉𝑟𝑓 RF voltage at generator
𝑉̃𝑚𝑎𝑥 Maximum of applied voltage at generator
𝑉̃𝑚𝑖𝑛 Minimum of applied voltage at generator
𝑉𝑑𝑐 DC self-bias voltage
𝑉𝑉𝑊𝑇 Tailored voltage waveform
𝐽 Current density

xix
𝐽1 Current density in sheath 1
𝐽2 Current density in sheath 2
𝐼 Electric current
𝐼1 Current in sheath 1
𝐼2 Current in sheath 2
𝑬 Electric field strength
𝑅 Electrical resistance
𝑅𝑃 Plasma bulk resistance
𝐿 Electrical inductance
𝐿𝑃 Plasma bulk inductance
𝐶 Electrical capacitance
𝐶1 Capacitance of sheath 1
𝐶2 Capacitance of sheath 2
𝐶𝐵 Capacitance of blocking capacitor
𝑍 Electrical impedance
𝑍𝐿 Total load impedance
𝑍0 Generator Output impedance
𝑄 Electric charge
𝑄1 Total electric charge in sheath 1
𝑄2 Total electric charge in sheath 2
𝜎 Electrical conductivity
𝜎𝑑𝑐 DC conductivity
Γ𝑅 power reflection coefficient
𝑆𝑜ℎ𝑚̅ Time averaged ohmic heating
𝑆𝑠𝑡𝑜𝑐 ̅ Time averaged stochastic heating
𝛽 Discharge asymmetry parameter
𝑙 Discharge gap
A Area
A1 Area of electrode 1
A2 Area of electrode 2
s1 Width of sheath 1
s2 Width of sheath 2
ŝ1 Maximum width of sheath 1
ŝ2 Maximum width of sheath 2
Γ Surface Fluxes
Γ𝑖 Ion flux to wafer surface
Γ𝑒 Electron flux to wafer surface
𝜙 Fluxes
E𝐴 Etch rate of material “A”
𝜉𝐴,𝐵 Etch selectivity of material “A” vs “B”
𝑘𝑖 Reaction rate coefficient of reaction i
𝜈𝑖 Total rate of collision i
𝜇 Mobility
𝜇𝑒 Electron mobility

xx
𝜇𝑖 Ion mobility
𝐷 Diffusion coefficient
𝐷𝑒 Electron diffusion coefficient
𝐷𝑖 Ion diffusion coefficient
𝐷𝑎 Ambipolar diffusion coefficient

xxi
 Abstract

As critical dimensions of semiconductor devices shrink, feature densities increase and the

geometries become more complex, the manufacturing processes are required to consistently

improve and innovate. Low temperature plasma based processes are required to etch nanometer

scale features with high aspect ratios through multi-material stacks as fast as possible, while

maintaining high uniformity and a high yield over a 300 mm wafer. The quality of the surface etch

is highly dependent on the energy and angular distribution of the charged particles, ions and

electrons originating from the gas phase, incident on the wafer surface.

To control their dynamics, the use of complex tailored voltage waveforms was

investigated. The tailored waveform consisted of a sinusoidal harmonic wave and its higher

harmonics. Coupled reactor and surface scale simulations were performed to investigate the

respective physical regimes. The Hybrid Plasma Equipment Model (HPEM) was utilized to

simulate the gas phase and discharge physics in a capacitively coupled plasma operated in the low-

pressure regime. Investigated feed gas mixtures include Ar/O2, Ar/O2/CF4 as well as Ar/O2/C4F6.

Based on the HPEM results, the Monte Carlo Feature Profile Model (MCFPM) was used to

simulate the feature etch process into SiO2.

It was found that some degree of control of charged particle dynamics is possible by

adjusting the phase of higher harmonics φ through the resulting generation of electrical asymmetry

and electric field reversal. These general trends were present in most considered configurations,

however the nature of the interaction between ions and the generated DC self-bias were found to

xxii
be context dependent with respect to its effects on ion energy. Two distinct regimes were

identified. Average ion energy onto the wafer is strongly correlated to the DC self-bias at high f0,

whereas in the low frequency regime this correlation is weak. Average ion energy onto the wafer

is instead dominated by dynamic transients in the applied voltage waveforms.

In all cases however, the trends produced in the gas phase translated to significant

differences in the feature properties, strongly suggesting that voltage waveform tailoring

constitutes a potent concept for etch process control.

Additionally, as many other simulation concepts, the MCFPM is critically dependent on

the reaction mechanism representing the physical processes occurring between plasma produced

reactant fluxes and the surface represented by the reaction probabilities, yields, rate coefficients,

threshold energies etc. The increasing complexity of the structures being fabricated, new materials

and novel gas compositions for plasma produced radical fluxes to the wafer also increases the

complexity of the reaction mechanism used in feature scale models, and the difficulty in

developing the fundamental data required for the mechanism. This challenge is further exacerbated

by the fact that acquiring these fundamental data through more complex computational models or

experiments is often limited by cost, technical complexity or inadequate models. Methods to

automate the selection of fundamental data in a reduced reaction mechanism for feature scale SiO2

plasma etching using a fluorocarbon gas mixture is discussed. By matching predictions of etch

profiles to experimental data using a gradient descent / Nelder-Mead method hybrid optimization

scheme these methods produce a reaction mechanism that replicate the experimental training data

as well as experimental data using a related but different etch processes.

xxiii
 Chapter 1 : Introduction

Plasma is everywhere. Physically, as the most prevalent state of ordinary matter throughout

the universe in the form of stars, lightning and auroras [1–4]; dramatically, but also quietly in terms

of how its technological utilization has a seldom recognized presence in all our lives. Parts for

furniture, appliances and cars are precisely cut by plasma torches and gas-discharge lasers [5,6].

Machines and tools are hardened by it [7], while it (at least for now) lights streets, offices and train

stations [8,9]. Plasma can make water drinkable [10], provide nourishment for plants [11] and treat

some of our ills [12–14]. Beyond terrestrial uses, plasma-based propulsion is essential in long-

term orbital and interplanetary space fare. Certainly, advances in nuclear fusion research inspire

hope that the understanding of plasma can contribute to the creation of a long-term sustainable

energy source in a (hopefully) not too distant future. However, undeniably one of the most

important contributions of technological plasmas is their established and ever increasingly

important role in the manufacturing of advanced semiconductors. Their processing has motivated

intense interdisciplinary research endeavors of tens of thousands of scientists as well as engineers

(this one included) and directly contributed to the proliferation of ever faster and more efficient

electronics and computing devices.

Without technological plasmas there would be no information age!

1.1 The Fourth State of Matter

Plasma is often referred to as the fourth state of matter. State transitions of matter can be

understood as responses to energy that is added to a system. The most common states of matter

1
and respective state transitions are shown in Figure 1.1. Generally, the lowest energy state a

substance can occur as is a solid. A solid’s structure is defined by the molecules’ outer shell

electrons interacting over small distances and forming strong chemical bonds, resulting in a

materials ability to maintain constant volume, density and a fixed shape. Adding sufficient energy

into the system, for example in the form of heat, causes the breaking of these bonds (melting) and

a phase transition into the fluid state, losing the ability to withstand shear forces and maintain a

fixed shape. At sufficiently high energies molecules are no longer constrained by the cohesive

forces and can move freely in a gaseous state (vaporization). If molecules accrue enough energy it

can lead to ionization where a neutral particle splits into a charged ion(+) electron(-) pair. A

partially or fully ionized gas is called “plasma”. Due to the presence of mobile charge carriers,

plasmas can react to electromagnetic fields on macroscopic scales and carry electric currents.

Figure 1.1: States of matter

Beyond these basic characteristics, the plasma state spans an enormous range of physical

regimes, each exhibiting drastically different properties. [15,16] A basic categorization is often

done based on a plasma’s number density and temperature which can range from in space plasmas

with low densities (1 cm-3) and low temperatures (1 - 100 eV) to fusion plasmas with high densities

2
(1015 cm-3) and high temperatures (> 10 keV) [17]. An overview of different plasma regimes is

depicted in Figure 1.2.

Figure 1.2. Plasma parameter space [16].

The focus of this thesis lies on low temperature plasmas (LTP) with electron temperatures

between 1 and 10 eV operated at low pressures (10s of mTorr). These plasmas are often referred

to as glow discharges due to their propensity to emit visible light and the fact that they are typically

maintained via electric discharge configurations, where the plasma is sustained by coupling energy

by electromagnetic or static means.

3
1.2 Basic Properties of Low-Temperature Plasmas

For plasma to exist, ionization is necessary. The ionization process produces a population

of electrons with the number density 𝑛𝑒 , one of the most fundamental plasma properties often

referred to as simply the plasma density. The degree of ionization 𝛼 is defined as the ratio of the

number densities of ionized to non-ionized neutral gas species, 𝑛𝑖 and 𝑛𝑔 , respectively.

𝑛𝑖
𝛼= (1.1)
𝑛𝑖 + 𝑛𝑔

The types of plasmas investigated in this work are considered weakly ionized plasmas with

𝛼 < 1%. The primary source of ionization in most plasmas is based on electron impacts either

directly

𝑒 + 𝐴 → 𝐴+ + 𝑒 + 𝑒, (1.2)

or as a dissociative ionization

𝑒 + 𝐴𝐵 → 𝐴+ + 𝐵 + 𝑒 + 𝑒. (1.3)

Both reactions are highly dependent on the electron’s energy which, for a single collision

process needs to be greater than the ionization threshold of species A, 𝜀𝑖𝑜𝑛,𝐴 .

1.3 Boltzmann’s Equation

Kinetically, a system of particles is fully described by its distribution function in six-

dimensional phase space

𝑓(𝐱, 𝐯, t) (1.4)

with particle positions x and velocities v. The change to a system’s distribution function can be

described by 3 major components

4
 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓
=( ) +( ) +( ) (1.5)
𝜕𝑡 𝜕𝑡 𝑓𝑜𝑟𝑐𝑒 𝜕𝑡 𝑑𝑖𝑓𝑓 𝜕𝑡 𝑐𝑜𝑙𝑙

where the "force" term corresponds to external forces, for example in the form of external fields

exerting a temporal change on the particles. The "diff" term represents the diffusion of particles,

and "coll" is the collision term – accounting for the forces acting between particles in collisions.

The evolution of this phase space state is described by Boltzmann’s Equation:

𝜕𝑓 𝐹 𝜕𝑓
+ 𝐯∇𝑟 𝑓 + ⋅ ∇𝑣 𝑓 = | (1.6)
𝜕𝑡 𝑚 𝜕𝑡 𝑐𝑜𝑙𝑙

Interparticle collisions can cause rapid, effectively discontinuous, changes in the

𝜕𝑓
distribution and are collectively represented in the collision term | . The details of the
𝜕𝑡 𝑐𝑜𝑙𝑙

collision term depend on the specific application but can contain number losses (species absorbing

collisions), number gains (species generating reactions), and instantaneous changes to velocity

(elastic collisions) or speed (inelastic collisions). In the case of low temperature plasmas, the

external force is dominated by electromagnetic fields leading to the commonly used two term

approximation:

𝜕𝑓 𝑒(𝐄 + 𝐯 × 𝐁) 𝜕𝑓
= −𝐯∇𝑟 𝑓 − ⋅ ∇𝑣 𝑓+ | (1.7)
𝜕𝑡 𝑚 𝜕𝑡 𝑐𝑜𝑙𝑙

Macroscopic quantities can be derived from the distribution function such as the number

density

𝑛(𝐫, 𝑡) = ∫ 𝑓𝑑 3 𝑣, (1.8)

and the particle flux

𝚪(𝐫, 𝑡) = 𝑛𝐮 = ∫ 𝐯𝑓𝑑 3 𝑣. (1.9)

5
1.4 Conservation Equations

Formulations for fundamental conservation requirements can be obtained through the

moments of Boltzmann’s equation.

1.4.1 Particle Conservation

The 0th moment yields the continuity equation which governs the particle conservation and

by extension the total mass balance of the system.

𝜕𝑛
+ ∇ ∙ (𝑛𝐮) = 𝐺 − 𝐿 ( 1.10)
𝜕𝑡

A change in local density is the consequence of a divergence of particle flux, i.e., more particles

flow into a test location than flow out or due to local generation or loss, given by G and L

respectively. G and L originate from the collision operator and are thus linked to the same physical

processes that cause a change in particle number.

1.4.2 Momentum Conservation

The first moment yields the momentum conservation equation.

𝜕𝐮
𝑚𝑛 [ + (𝐮 ∙ ∇)] = 𝑞𝑛(𝐄 + 𝐮 × 𝐁) − ∇ ∙ 𝚷 + 𝑓|𝑐𝑜𝑙𝑙 (1.11)
𝜕𝑡

Temporal or spatial changes to the momentum carrying flux, nu, described on the left-hand side,

are the result the external E and B fields, the pressure gradients, described by the divergence of

the pressure tensor 𝚷 and the momentum changes because of collisions.

6
1.4.3 Energy Conservation

The second moment yields the energy conservation equation.

𝜕 3 3 𝜕 3
( 𝑝) + ∇ ∙ 𝑝𝐮 + p∇ ∙ 𝐮 + ∇ ∙ 𝐪 = ( 𝑝)| (1.12)
𝜕𝑡 2 2 𝜕𝑡 2 𝑐𝑜𝑙𝑙

3 3
Here 2 𝑝 is the thermal energy density and consequently 2 𝑝𝐮 describes the thermal energy

flux. p∇ ∙ 𝐮 is the local heating and cooling due to compression and expansion.

1.5 Maxwell–Boltzmann Distribution

A system that is dominated by elastic momentum transfer collisions will eventually reach

a steady state that is represented by the Maxwell–Boltzmann distribution

3⁄
𝑚 2 𝑚𝑣 2
𝑓𝐵 (𝐯) = ( ) 4𝜋𝑣 2 𝑒𝑥𝑝 (− ). (1.13)
2𝜋𝑘𝐵 𝑇 2𝑘𝐵 𝑇

The specific distribution is dependent on the species temperature. Boltzmann distributions

corresponding to 3 different temperatures T1, T2 and T3 with T1 < T2 < T3 are shown in Figure 1.3.

Figure 1.3: Maxwell-Boltzmann distribution corresponding to three different temperatures.

7
 For higher temperatures, the distribution extends further to high velocities, while still

containing a finite low temperature component. Systems without external power losses and gains

eventually reach this state through mutual interaction of particles, referred to as thermalization i.e.,

systems in thermal equilibrium generally follow the Maxwell–Boltzmann distribution.

1.6 Thermal Non-Equilibrium

Temperature of a system is directly related to the species’ thermal velocities 𝑣𝑡ℎ .

2
3𝑘𝑏 𝑇 𝑣𝑡ℎ 𝑚
𝑣𝑡ℎ = √ ⇒𝑇= (1.14)
𝑚 3𝑘𝑏

When power is coupled into plasma via electromagnetic or electrostatic means, it results

in heating of the charged particles only. Thermalization i.e., the process of physical bodies reaching

thermal equilibrium through mutual interaction, occurs mostly through elastic collisions. Due to

the differences in thermal velocity 𝑣𝑡ℎ and the subsequent difference in collision frequency,

electrons thermalize much more rapidly than the heavier ions. Because of this, in the low-pressure

regime, electrons are mostly considered in thermal equilibrium with themselves, whereas ions are

often not. Energy transfer between electrons and ions through elastic collisions is generally rather

inefficient due to the large difference in masses and thus they are not in thermal equilibrium with

each other as well. Finally, due to the low collisionality between ions and neutrals because of the

low pressure they, too, are generally considered to not be in thermal equilibrium with each other

yielding three distinct temperatures of electrons ions and the neutral gas, 𝑇𝑒 , 𝑇𝑖 , and 𝑇𝑔 ,

respectively, generally with 𝑇𝑒 ≫ 𝑇𝑖 > 𝑇𝑔 .

8
1.7 Electron Collision Cross Sections

The probability of electrons to undergo reactions with collision partners is a function of

energy which is mostly determined by electron energy. This relation is often represented in the

form of energy dependent cross sections 𝜎(𝜀) like the one presented in Figure 1.4 for electron

impact cross sections with Ar. Every type of interaction has its own energy criteria, and thus

possesses its own cross section. The most relevant are usually the momentum transfer 𝜎𝑒 ,

excitation 𝜎𝑒𝑥𝑐 𝑖 , and ionization 𝜎𝑖𝑜𝑛 𝑖 . Notably, momentum transfer can occur at every energy to

some degree while excitation and ionization cannot occur below certain thresholds which stem

from the atomic/molecular energy level structure.

Figure 1.4: Electron impact cross sections for e-Ar collisions. Reproduced from Petrović et al.

1.8 Rate Coefficients and Collision Rates

The effective rate of electron impact reactions is given by its rate coefficient

𝑘𝑖 = ∫ 𝑓 (𝜀 ) 𝑣 𝜎𝑖 (𝜀 ) 𝑑𝜀, (1.15)

9
which is a function of the electron energy distribution 𝑓(𝜀 ) as well as the respective energy

dependent reaction cross section 𝜎𝑖 (𝜀 ). The total rate at which collisions occur is a function of the

densities of the respective reactant particles as it is proportional to the mean particle distance and

mean free path. The total rate of collision ‘i’ occurring between electrons and gas species ‘j’ is

then given by

𝜈𝑗𝑖 = 𝑛𝑒 𝑘𝑗𝑖 𝑛𝑗 (1.16)

1.9 Quasineutrality

Most plasmas are typically quasineutral i.e., the amount of positive charge is equal to the

amount of negative charge on macroscopic scales. In terms of the space charge distribution

quasineutrality represents the lowest energy state and thus deviations from it are restored in a self-

regulating manner. The presence of space charge regions leads to the generation of an electric

potential that attracts species of the opposite charge, thus neutralizing the space charge. However,

quasineutrality can be broken over small time or length scales [15,16,18] when a force is applied,

or an external potential is present. The time scale is defined by the plasma frequency, derived by

determining the frequency of electrostatic oscillation of charged species displaced from neutral. In

LTPs, the plasma frequency is the electron plasma frequency ωpe, given by

𝑒 2 𝑛𝑒
𝜔𝑝𝑒 = √ , (1.17)
𝜀0 𝑚𝑒

where e is the fundamental charge, ε0 is the permittivity of free space, and me is the electron mass.

The length scale over which quasineutrality can be broken is the Debye length λD, an estimate of

the electrostatic shielding distance given by

10
 𝜀0 𝑘𝐵 𝑇𝑒
𝜆𝐷 = √ 2 , (1.18)
𝑒 𝑛𝑒

where kB is Boltzmann’s constant and Te is electron temperature. The convention in plasma physics

is to refer to the electron temperature as kBTe in units of eV.

1.10 Ambipolar Diffusion

Macroscopic fluxes can be caused by external fields and or diffusion based on pressure

gradients

Γ = ±𝜇𝑛𝐄 − 𝐷∇𝑛, (1.19)

where 𝜇 and D are the macroscopic mobility and diffusion constant, which are unique to each

species. The 𝜇𝑛𝐄 term describes particles being accelerated by an electric field under the influence

of collisions and its inertia. The 𝐷∇𝑛 term describes movement down a pressure gradient under

the influence of thermal motion. In most plasmas, the positive and negative species have different

masses, with the electron dominated negative charge carriers being significantly more agile. This

means that the mean velocity of positive charges is different than the mean velocity of negative

charges. Charge carrier separation, caused by different diffusion speeds, leads to the generation of

an electric field that effectively couples positive and negative charge carrier transport, to a degree.

This effect is referred to as ambipolar diffusion and described by

Γ = −𝐷𝑎 ∇𝑛, (1.20)

with the ambipolar diffusion coefficient

𝜇𝑖 𝐷𝑒 + 𝜇𝑒 𝐷𝑖
𝐷𝑎 = . (1.21)
𝜇𝑖 + 𝜇𝑒

11
1.11 Plasma Sheath

Whenever plasma is in contact with a surface a region of space charge called the “plasma

sheath” is generated. This sheath region is of central importance to this work as plasma-surface

interactions are critically dependent on its properties. To illustrate its basic structure, it is

instructive to consider a simplified case of an electropositive plasma bounded by grounded metal

plates depicted in Figure 1.5a.

Figure 1.5: Schematic of a plasma bounded by two parallel grounded metal plates and the resulting density b) as
well as potential c) distribution throughout the gap.

12
 Losses of charged particle species to the walls occur at different rates. Electrons which are

typically much lighter and faster than ions (me/mi = 7.2 × 10-4 in the exemplary case of Argon and

Te/Ti ≥ 10) are lost to the surface at a significantly higher rate. Typical number densities of ions

and electrons as a function of distance are shown in Figure 1.5b. The center region of the plasma,

the “bulk”, is generally considered to be quasineutral (ne≈ni). Because of wall losses, particle

densities decrease in proximity to the boundary surfaces with. Due to the electrons’ higher loss

rate their density ne in the boundary region is lower than the ions’ ni creating an electron-depleted

positive space charge region (shaded area).

The spatial extend of this break in quasineutrality is generally on the order of the Debye

length λD. This space charge ρ = ni - ne causes an electric potential ϕ in accordance with Poisson’s

equation with the electrical permittivity ε [19–21]:

𝜌
∇2 ϕ = − (1.22)
𝜀

ϕ is depicted in Figure 1.5c. Due to the positive space charge the plasma potential 𝑉𝑝 is

positive with respect to the ground reference and the corresponding electric field 𝐸 points from the

plasma bulk towards the surface. The sheath electric field in turn accelerates positive ions towards

the said surface while simultaneously confining electrons to the bulk. This directional ion

acceleration within the plasma sheath is the basis for many technological applications of plasmas,

including the anisotropic etching of materials which is a particular focus of this work and

introduced in Section 0.

13
1.12 Capacitively Coupled Plasmas

Since charge is lost to the walls and to ion-electron recombination, a continuous power

input is required to sustain plasmas. Although not necessarily thermal in nature, this is generally

referred to as plasma heating as it transfers kinetic energy to the charged and neutral species. The

magnitude of plasma heating and the underlying mechanism play an essential role in determining

basic plasma properties such as plasma density, temperature, potential and a host of downstream

properties.

Figure 1.6: Schematic of a capacitively coupled plasma and the corresponding potential distribution at 4 different
times of the RF period.

14
 One type of plasma heating with extraordinary technical relevance is through capacitive

coupling [22–43]. Capacitively coupled plasmas (CCP) are heated by interfacing with large area

electrodes. The simplest setup resembles the planar configuration shown in Figure 1.5a with the

exception that a radio frequency (RF) alternating current (AC) voltage is applied to at least one of

the electrodes. This setup is schematically depicted in Figure 1.6 where an AC voltage source with

𝑉𝑟𝑓 (𝑡) = 𝑉0 sin(𝜔𝑡) is connected to the left electrode. Because the plasma potential is always

positive with respect to the boundaries the RF oscillation at the electrode propagates to the plasma

and sheath potentials leading to oscillations in the sheath potential and thickness. Thus, the plasma

potential with respect to the electrodes oscillates while maintaining a mostly flat structure

throughout the plasma bulk i.e., the bulk electric field remains negligible. To illustrate this

behavior, the plasma potential as a function of gap position z, is shown in Figure 1.6 for two 4
1 3
different points in the RF-period with 𝜔𝑡 = 0, 𝜋, 𝜋 and 2 𝜋. Note that while always positive with
2

respect to both electrodes, sheath voltages at the grounded and driven electrode are not always

identical.

1.13 Electrical Characteristics

In order to examine the behavior of capacitive discharges, a suitable model of its electrical

characteristics is imperative. Due to their differences in charge density and potential configuration

the bulk and sheath regime are considered separately.

i) The bulk region of the plasma is generally well described by assuming a high electron

density, negligible internal electric fields and a conduction current that is much greater

than any present displacement current. Its impedance is given by

𝑋𝑝 = 𝑅𝑝 + 𝑗𝜔𝐿𝑝 (1.23)

15
 which consists of the bulk plasma resistance 𝑅𝑝 in series with an inductance 𝐿𝑝 . 𝑅𝑝 is

defined by the bulk’s dc conductivity 𝜎𝑑𝑐 , its width d and effective electrode facing

area A:

1 𝑑 𝑚𝑒 𝜈𝑚 𝑑
𝑅𝑝 = = 2 (1.24)
𝜎𝑑𝑐 𝐴 𝑒 𝑛𝑒 𝐴

analogous to the Drude theory in metals [44,45], with the momentum transfer collision

frequency 𝜈𝑚 [16,18]. While the physical origin of the bulk plasma inductance 𝐿𝑝 is

not in line with the common provenance of inductances i.e., the magnetic self-induction

in current carrying coils, it nevertheless resembles the systems response to changes in

current based on the electron’s inertia. 𝐿𝑝 is given by

𝑅𝑝 𝑚𝑒 𝑑
𝐿𝑝 = = 2 . (1.25)
𝜈𝑚 𝑒 𝑛𝑒 𝐴

ii) The sheath region of the plasma is mostly defined by displacement current as the low

electron density does not allow for significant amounts of conduction current. Most of

the applied voltage is dropped across the two sheaths. While the entire CCP setup

resembles a plate capacitor, the sheaths themselves effectively do, too; two areas of

high conductivity (electrode and plasma bulk) separated by a low conductivity gap

(sheath) across a relatively large area. The capacitances at both sheaths, 𝐶1 and 𝐶2 ,

respectively, must be considered as non-ideal because the sheath oscillation causes a

non-constant gap width and conductivity. Although the conduction current in each

sheath is small, it is still determined by the balance between ion and electron

contributions. Due to their high mass and inertia, the ion current 𝐼𝑖̅ is generally assumed

to be static with

16
 𝐼𝑖̅ = 𝑒𝑛𝑒 𝑢𝐵 𝐴, (1.26)

where 𝑢𝐵 is the Bohm velocity [46–49]. While this assumption is specifically

challenged as part of this thesis it is still valid in the current context as the ion’s inertia

is still significantly higher than the electrons’ and the ion current is thus much less

modulated.

To balance the continuous ion current, the time averaged electron current must

be identical. Because of their low inertia, electron transport in the sheath is strongly

dependent on the sheath potentials temporal and spatial dynamics which are depicted

in Figure 1.6 for 4 different time points in the RF-cycle, 𝜔𝑡 = 0, 𝜋⁄2 , 𝜋 and 3𝜋⁄2

The electrons are confined by the sheath potential for most of the RF period and thus

their current is zero. However, for brief time periods the applied RF voltage causes the

plasma potential at the respective sheaths to be small enough to allow for electron

current to occur. This moment is referred to as the sheath collapse. In the depicted

example it occurs at 𝜔𝑡 = 0 at the powered electrode (x = 0) and at 𝜔𝑡 = 𝜋 at the

grounded electrode (x = 𝑙). Since the sheath voltage collapses to zero at the time that

the electrons are transferred to the plate, this acts like an ideal diode across the sheath

whose preferred direction of current flow is into the plasma. In total, sheaths can be

modeled as a non-ideal capacitance, an ideal current source (ion current) and an ideal

diode, connected in parallel.

Combining these representations leads to a simplified circuit model shown in Figure 1.7 in

which the two sheaths are in series with the bulk conductance and resistance. The voltage 𝑉𝑟𝑓 is

applied across it.

17
 Figure 1.7: Circuit diagram of capacitively coupled plasma.

1.14 Heating Mechanisms

Understanding the electrical properties of CCPs enables some insight into the plasma

heating required to sustain the discharge. Two different heating mechanisms are responsible for

most of the power coupling in CCPs. While both are ultimately based on the acceleration of

electrons in electric potentials and their subsequent momentum transfer collisions, the details have

relevant consequences in terms of the involved scaling laws and control capabilities.

i) Ohmic heating – the oscillating voltage across the entire discharge causes an equivalent

electron current that, through collisions transfers energy to heavy particles [16]. The

̅
time average power deposition per unit area 𝑆𝑜ℎ𝑚 can be described by

18
 1 2 𝑑 1 𝑚𝑒 𝜈𝑚 𝑑
̅
𝑆𝑜ℎ𝑚 = 𝐽 = 𝐽2 2 (1.27)
2 𝜎𝑑𝑐 2 𝑒 𝑛𝑒

with current density 𝐽. Because 𝜈𝑚 strongly depends on the gas pressure ohmic heating

is not very effective in the low-pressure regime.

ii) Stochastic heating – Due to the changing applied voltage electrons can diffuse into the

active sheath area at the powered electrode when the electron confining electric field is

at its minimum. When the sheath voltage increases again it accelerates electrons back

into the plasma bulk. Because electron heating effectively only occurs during the sheath

expansion this heating mode is also known as sheath expansion heating. Assuming a

Maxwellian electron distribution and sheath densities according to Child’s law the

stochastic heating term can be approximated by

1 𝑚 𝜈̅𝑒 2
̅
𝑆𝑠𝑡𝑜𝑐 = 𝐽 (1.28)
2 𝑒 2 𝑛𝑒

with the mean thermal electron velocity 𝜈̅𝑒 . Applying Child’s law, the current density

can be approximated with 𝐽2 = 1.73𝑒𝜀𝜔2 √𝑇𝑒 𝑉𝑛𝑠 with the applied voltage amplitude

𝑉 and the RF driving frequency 𝜔, yielding

𝑚
̅
𝑆𝑠𝑡𝑜𝑐 ∝ √𝑇𝑒 𝜀𝑉𝜔2 (1.29)
𝑒

as the primary scaling behavior [50–55]. Stochastic heating is expected to be sensitive

to changes in applied voltage as well as driving frequency.

1.15 Power Matching

When transmitting power from an external source to a load with complex impedance (such

as a capacitively coupled plasma), some amount of power can be reflected if the output impedance

19
of the generator is not matched to that of the plasma system. The power reflection coefficient, Γ𝑅 ,

is the ratio of power reflected from the plasma reactor relative and power delivered by the power

supply through a transmission line and given by

Z𝐿 − Z0
Γ𝑅 = (1.30)
Z𝐿 + Z0

where Z𝐿 is the impedance of the load (in this case, the combined impedance of the plasma reactor

and the plasma) and Z0 is the output impedance of the power supply and the transmission line

[19,56,57]. Typically, to minimize Γ𝑅 an impedance matching network (IMN) (often referred to as

match box) between the plasma and the generator is employed. A simple L-type IMN is shown in

Figure 1.8.

Figure 1.8: Circuit diagram of the matching network in conjunction with the plasma.

It consists of a series inductance L𝑚1 as well as a parallel and series capacitance, C𝑚1 and

C𝐵 , respectively. C𝐵 , called the blocking capacitor as it blocks DC current flow, can affect the

plasma dynamics in vital ways by charging up and creating a large negative DC potential at the

powered electrode. This effect is induced by a property called asymmetry.

20
1.16 Dual and Multi Frequency RF Power

The frequency dependence of the capacitive heating modes has inspired the use of dual

frequency CCPs, where the power is coupled via two independent low and high frequency

components [58,59]. The total applied voltage becomes

𝑉𝑟𝑓 (𝑡) = 𝑉𝑙𝑓 cos(2𝜋𝑓𝑙𝑓 𝑡) + 𝑉ℎ𝑓 cos(2𝜋𝑓ℎ𝑓 𝑡), (1.31)

where 𝑉𝑙𝑓 and 𝑉ℎ𝑓 are the voltage amplitudes of the low and high frequency component with

respective frequencies of 𝑓𝑙𝑓 and 𝑓ℎ𝑓 . The basic idea of using two different frequencies

simultaneously was to achieve independent control over both the ion flux and ion bombardment

energy onto the substrate electrode. The ion flux is mostly a function of the bulk plasma density

whereas the ion energy directly mirrors the sheath voltage. It has been shown that higher driving

frequencies generally lead to increased power deposition into the electrons which in turn leads to

increased ionization and higher plasma densities and ion fluxes [59–61]. For an equal power

deposition, reducing the driving frequency requires an increase in the applied voltage. Since this

voltage almost entirely drops across the sheath, an electron starved region, this leads to ion

acceleration inside the sheath and consequently allows for some ion energy control. In general,

0.5–13.56 MHz sources are used as the LF source while 27.12– 81.36 MHz sources are used as

the HF source [62–64]. While some differential control of ion flux and ion energy can be achieved,

in real world application the LF and HF components are not fully independent. They interact

indirectly through the respective changes to plasma properties such as plasma density, sheath

voltage, sheath width, electron temperature etc. and also create novel non-linear effects that are

not present in the single frequency scenario.

21
1.17 Plasma Asymmetry and DC self-bias

Most capacitive discharges are asymmetric in one or many respects. The simplest and most

common one is geometrical asymmetry. Because reactor walls in CCPs are generally grounded,

the effective grounded area is larger than the powered area, even if both electrodes have the same

dimensions. A schematic of an asymmetrical CCP setup is shown in Figure 1.9, where the powered

electrode with area A1 is on the left-hand side of the discharge gap and the grounded surface with

area A2 is on the right-hand side; note that A1 <A2.

Figure 1.9: Schematic of an asymmetric CCP with two different electrode sizes with A 1 < A2 and the blocking
capacitor CB at the powered electrode.

22
 The most relevant voltages in the system are the voltages across the sheaths, 𝑉1 and 𝑉2 ,

respectively, the voltage across the blocking capacitor 𝑉𝐵 as well as the voltage generated by the

source 𝑉𝑟𝑓 = 𝑉0 sin (𝜔𝑡). The time average voltage drop across a sheath is well approximated by

its capacitive behavior with

Q
V∝ , (1.32)
𝐶

where Q is the total charge contained within the sheath space charge region, and

A
C=𝜀 . (1.33)
𝑑

The conservation of total sheath charge is a fairly good assumption as DC current is not

permitted to flow into and out of the system and the bulk region is mostly quasineutral. It thus

follows for the two sheaths with (1.32) and (1.33)

𝑄1 = 𝑄2 ∧ 𝐴1 < 𝐴2 ⇒ 𝑉1 > 𝑉2 . (1.34)

Because no DC voltage is generated by the ideal AC source, this leads to a closed loop over which

according to Kirchhoff’s law [65] the sum of all remaining voltages must equal zero. Accounting

for the opposing polarity of the voltages across the sheath this gives:

𝑉1 − 𝑉2 + 𝑉𝐵 = 0 (1.35)

Considering (1.34), this results in the requirement of a negative voltage across the blocking

capacitor called the DC self-bias, 𝑉𝑑𝑐 .

𝑉𝑑𝑐 = 𝑉𝐵 = −(𝑉1 − 𝑉2 ) < 0 (1.36)

The presence of this DC self-bias is well documented [66–69] and is of high relevance as

effectively increases the total potential drop across the powered sheath, as is depicted in Figure

1.9b. This increased potential directly translates to increased directional acceleration of ions

towards the surface, a highly sought after property in CCPs used for semiconductor manufacturing.

23
This high practical relevance has driven considerable research aimed at understanding and

modeling the system’s response to asymmetry. In practice, the presented derivation contains some

critically weak assumptions such as the assumed constant sheath widths d and consequently the

sheath capacitances. Both are, in fact, critically dependent on the sheath voltage itself leading to

nonlinear capacitances and time-varying density profiles in the sheaths. Early model descriptions

used the Child-Langmuir law [70,71] to describe the sheath behavior [68]. The total rf current in

sheath 1 is given by

I1 = ∫ J1 (𝒙) 𝑑 2 𝑥. (1.37)
𝐴1

Because the rf current density is proportional to the sheath width s1 and the voltage drop

is independent of the position along the surface 𝒙 with

𝑉̅1
J1 (𝒙) ∝ (1.38)
𝑠1 (𝒙)

and by using an ion density relation derived from Child’s law

⁄ 3
𝑉̅1 2
𝑛𝑖1 (𝒙) ∝ 2 (1.39)
𝑠1 (𝒙)

s1 can be expressed in terms of 𝑛𝑖1 . Substituting it in (1.38), (1.37) gives

1⁄ 1⁄
I1 ∝ 𝑉̅1 4
∫ 𝑛𝑖1 2 (𝒙) 𝑑 2 𝑥 (1.40)
𝐴1

and similarly

1⁄ 1⁄
I2 ∝ 𝑉̅2 4
∫ 𝑛𝑖2 2 (𝒙) 𝑑 2 𝑥 (1.41)
𝐴2

for the other sheath. Assuming RF current continuity this gives

24
 1⁄ 4
2 2
̅1
V ∫𝐴 𝑛𝑖2 (𝒙) 𝑑 𝑥
( 2 1 )
̅2 = (1.42)
V ⁄2 2
∫𝐴1 𝑛𝑖1 (𝒙) 𝑑 𝑥

Making the, admittedly weakly motivated, assumption of a uniform ion density with 𝑛𝑖1 (𝑥 ) =

𝑛𝑖2 (𝑥 ) = 𝑛𝑖0 we get

̅1
V A2 4
̅2 = (A1 )
V
(1.43)

This scaling law clearly shows a non-linear relation of the DC self-bias (which is the difference of

̅1 and V
V ̅2) and the electrode area ratio. However, the model details were not in accordance with

many experimental findings [66–69] which typically found

̅1
V A2 𝑞
̅2 ≈ (A1 )
V
(1.44)

with q ≤ 2.5.

1.18 Generalized Asymmetry

The theory of asymmetric discharges was later expanded upon by Heil et al [72] who

considered the total unbalanced charge in a sheath given by

𝑠
𝑄
= 𝑒 𝑛𝑖 (𝑧)𝑑𝑧.
∫ (1.45)
A 0

When sheath 1 is at its maximum extension 𝑠̂1 the opposing sheath is at its minimum width and

can in good approximation be considered fully collapsed i.e., it contains no charge. This in turn

means that all unbalanced charge is contained in sheath and its maximum voltage 𝑉̂1 is given by

𝑒 𝑠̂1
𝑉̂1 = − ∫ 𝑛𝑖 (𝑧)𝑧 𝑑𝑧. (1.46)
𝜀 0

25
This maximum charge 𝑄̂ can alternatively be expressed as a function of the mean charge density

in the sheath

𝑄̂ 1 𝑠̂1
= 𝑒𝑠̂1 𝑛̅1 , with 𝑛̅1 = ∫ 𝑛 (𝑧)𝑧 𝑑𝑧. (1.47)
𝐴 𝑠̂1 0 𝑖

The maximum sheath voltage can now be expressed as

𝑒 2
𝑉̂1 = − 𝑠̂ 𝑛̅ 𝐼 , (1.48)
2𝜀 1 1 𝑠

where the sheath integral 𝐼𝑠 is a dimensionless quantity that depends only on the normalized profile

of the sheath density. Combining (1.47) and (1.48) the two sheath voltages can be expressed as
2
𝑒 𝑄̂ 𝐼𝑠1
𝑉̂1 = − ( ) (1.49)
2𝜀 𝐴1 𝑛̅1

and
2
𝑒 𝑄̂ 𝐼𝑠2
𝑉̂2 = − ( ) . (1.50)
2𝜀 𝐴2 𝑛̅2

The absolute of the sheath voltage ratios yields the asymmetry parameter

𝑉̂2 𝑛̅1 A1 2 𝐼𝑠2

𝛽=| |= ( ) . (1.51)
𝑉̂1 𝑛̅2 A2 𝐼𝑠1

In literature this parameter is often also referred to as 𝜀 [24,73–78]. To maintain

consistency throughout this thesis and in acknowledgement of the fact that the discipline of

electrical engineering already has a sufficient number of physical properties being referred to as

‘e’, ‘E’ or ‘𝜀’, ‘𝛽’ will be used exclusively. Through the asymmetry parameter, experimentally

established scaling laws with respect to the area ratio (with q ≤ 2.5) are much better represented.

Perhaps more notably, this formulation hints at additional factors that can cause asymmetry in a

discharge, in addition to the geometric asymmetry of the reactor. A host of different sources for

discharge asymmetry are subjects of active research. This includes asymmetry induced by certain

26
magnetic field configurations, differing electrode material properties such as secondary electron

emission coefficients, electron beam injection or special applied RF voltages. Some of these

approaches are of particular interest as they are not fixed (in contrast to the reactor geometry) and

possibly allow for active discharge control. Asymmetry based on asymmetric applied voltages will

play a central role in this thesis.

1.19 Electrical Asymmetry

Under certain conditions asymmetry, i.e., the generation of a DC self-bias voltage, can be

induced even in perfectly geometrically symmetric systems by applying asymmetric voltages. Its

occurrence is referred to as the electrical asymmetry effect (EAE). While this effect has, quite

possibly inadvertently, been present in many multi frequency CCPs, its active consideration and

theoretical description goes back to foundational work by Heil, Schulze, Czarnetzki, Brinkmann

and Mussenbrock [72–74].

The maximum absolute sheath voltage is defined by the minimum voltage at the powered

electrode, keeping in mind the polarity of the sheath voltages i.e. the plasma potential is always

positive with respect to the enclosing electrodes, it is a combination of the minimum of the applied

RF voltage 𝑉̃𝑚𝑖𝑛 and the DC self-bias:

𝑉̂1 = 𝑉̃𝑚𝑖𝑛 + 𝑉𝑑𝑐 (1.52)

Similar reasoning for the grounded electrode and the maximum of the applied voltage 𝑉̃𝑚𝑎𝑥

leads to a convenient expression for the DC self-bias

𝑉̃𝑚𝑎𝑥 + 𝛽𝑉̃𝑚𝑖𝑛
𝑉𝑑𝑐 = − (1.53)
1+𝛽

which, in the case of a single frequency sinusoidal applied voltage 𝑉𝑟𝑓 (𝑡) = 𝑉0 sin(𝜔𝑡), reduces

to

27
 1+𝛽
𝑉𝑑𝑐 = − (1.54)
1+𝛽

In this scenario, no electrical asymmetry is generated. More general, no time varying applied

voltage that is harmonically symmetric ( 𝑉𝑠𝑦𝑚 (𝜑 + 𝜋) = −𝑉𝑠𝑦𝑚 (𝜑) ) or anti-symmetric (

𝑉𝑎𝑠𝑦𝑚 (𝜑) = −𝑉𝑎𝑠𝑦𝑚 (−𝜑) ) with respect to phase 𝜑 = 𝜔𝑡 leads to the generation of a DC self-

bias in an otherwise symmetric system. Ergo, in order to evoke electrical asymmetry, a voltage

that is neither symmetric nor anti-symmetric must be applied.

1.20 Voltage Waveform Tailoring

While there is, technically, an infinite number of functions that mathematically meet these

requirements, arbitrary voltage functions can lead to technical difficulties in their generation and

impedance matching, especially at high powers. It is for this reason that for technological

applications the focus has been on harmonically constructed waveforms where the applied voltage

waveform is comprised of a superposition of a fundamental sine wave and its higher harmonics.
𝑁

𝑉𝑉𝑊𝑇 (𝑡) = 𝑉𝑡𝑜𝑡 ∑ 𝑎𝑘 cos (𝑘𝜔0 𝑡 + 𝜑𝑘 ) (1.55)

𝑘=1

Here, 𝜔0 = 2𝜋𝑓0 is the fundamental frequency, 𝑘 denotes the harmonic order (𝑘 = 1 is

considered the fundamental), N is the total number of harmonics used, 𝑎𝑘 is the relative amplitude

of each harmonic, 𝜑𝑘 is the relative phase shift of each harmonic and 𝑉𝑡𝑜𝑡 is the total applied

voltage amplitude. Intentional purpose-oriented tailoring of applied voltages or voltage waveform

tailoring (VWT) has applications beyond the generation of a DC self-bias, however, in the

presented thesis this will be its main purpose, which in turn guides some principles guiding its

design and construction. Because odd harmonics do not contribute to the DC self-bias, they are

avoided. The technical complexity increases with the number of applied harmonics, so their

28
number is limited to 5 in this work. Lastly, for a given 𝑉𝑡𝑜𝑡 it exists a solution that optimizes the

difference between its extrema, given by

𝑁
𝑁−𝑘+1
𝑉𝑉𝑊𝑇 (𝑡) = 𝑉𝑡𝑜𝑡 ∑ cos(𝑘𝜔𝑡 + 𝜋𝑘𝜑). (1.56)
𝑁 (𝑁 + 𝑘 )
𝑘=1

The resulting time dependent voltage waveforms are shown in Figure 1.10 for different

phase shifts 𝜑 = 0, 45, 90, 135 and 180°.

Figure 1.10: Voltage waveforms VVWT(t) based on (1.56) for 𝜑 = 0, 45, 90, 135 and 180°.

29
 Due to their resemblence the waveform resulting from 𝜑 = 0° is often reffered to as “peak”

waveform whereas the 𝜑 = 180° case is known as as “valley” waveform. In a geometrically

symmetric CCP discharge the peak and valley waverfom have been shown to produce maximum

absolute value for 𝑉𝑑𝑐 with identical magnitude and opposing polarities; negative for 𝜑 = 0° and

positve for 𝜑 = 180°, respectively. The 𝜑 = 90° waveform is symmetric and thus does not

produce a DC self-bias. However, the assumption geometric symmetry is not in line with real

reactors and a superposition of geometric and electrical asymmetry is to be expected.

1.21 Ion Energy Distributions

Surface processes, especially ones relevant to the etching of semiconductors, are sensitive

to the energy and angle of incidence of ions hitting the surface. For this reason, the control and

tailoring of ion angular and energy distributions is of high technical relevance. Ions are generally

accelerated to the electrode by the sheath electric field. The maximum energy ions can gain by

traversing the sheath is the sheath voltage. However, due to the temporal dynamics of the sheath

not all ions gain the same amount of energy but rather obtain a characteristic ion energy and

angular distribution IEAD corresponding to the properties of the ion sheath traversal. It is

instructive to consider two specific edge cases.

i) The thick sheath or high frequency limit. Here, the sheath is considered thick enough so

that ions can only cross its distance on time scales much larger than the RF period. Without

considering ion collisions with the background gas, this idealized conception yields

monoenergetic ions corresponding to the mean sheath voltage.

ii) In the thin sheath or low frequency limit, the ion energy distribution at the surface perfectly

matches the temporal behavior of the sheath voltage. Ions are considered to be mobile

30
 enough to traverse the surface on timescales much smaller than the RF period, gaining

energy corresponding to the momentary sheath voltage value.

Realistically however, while slow and inert, compared to electrons, ions may possess some

limited ability to react to the changing sheath potential. This ability is a function of their weight,

the sheath thickness and the driving frequency. For single frequency RF discharges this can result

in characteristic bimodal distributions such as those depicted in Figure 1.11 for different particle

masses a) and varying driving frequencies b).

Given the use of two or more frequencies, as is the case with voltage waveform tailoring,

this simple scaling behavior vanishes, and a more detailed analysis of the sheath dynamics

becomes necessary. A detailed analysis of the ion sheath transit behavior resulting from the use of

tailored voltage waveforms will be provided in this work.

31
Figure 1.11 a) IEDs at the powered electrode of a rf CF4 discharge driven at 13.56 MHz from Kuypers and Hopman
[129]. b) Particle in Cell simulation results showing IEDs of helium ions hitting the target electrode of rf discharges
driven at frequencies from 1 MHz to 100 MHz. Reproduced from Kawamura et al [130].

1.22 Electronegative Plasmas

Some of the theory provided above either directly or indirectly assumes specific relations

between the negative charge carries; electrons (light and high energy) the positive charge carriers,

ions (heavy and medium to high energy) and neutrals (heavy and low energy) which relied on the

presence of only positive ions. Under certain conditions, however, significant amounts of negative

ions may be present, depleting electrons, and shifting the mass and inertia ratios of negative and

positive charge carriers. The extreme case, in which no electrons are present, and the charge

32
balance is provided by negative ions, is called an ion-ion plasma. The intermediate regime is

characterized by the ratio of negative ions and electrons, called plasma electronegativity. Negative

ions are generally produced by direct electron attachment

𝑒 + 𝐴 + 𝑀 → 𝐴− + 𝑀, (1.57)

where a third body M is required to conserve energy and momentum, dissociative attachment

𝑒 + 𝐴𝐵 → 𝐴− + 𝐵, (1.58)

or dissociative ion pair formation

𝑒 + 𝐴𝐵 → 𝐴− + 𝐵 + + 𝑒. (1.59)

In either case, the propensity for electron attachment is species specific. The atomic species with

the highest specific electronegativity according to the Pauling scale, are F and O, both of which

can be expected to be present in relevant quantities due to the usage of oxygen and fluorocarbons

as feedstock gases in this thesis.

33
1.23 Plasma in Semiconductor Processing

Fundamentally, semiconductor devices used as logic (processors), memory, power

applications, light emission (LED) or absorption (photovoltaic) are just materials with different

electrical properties, some of which are semiconductors, arranged in specific geometric

configurations to elicit specific physical behavior. Take for example a (by modern standards)

simple device such as a planar metal oxide field effect transistor (MOSFET), shown in Figure 1.12.

Figure 1.12: Schematic representation of a MOSFET

Doped n-type regions embedded in a p-type substrate are contacted by high conductivity

materials and a thin oxide film insulates the gate from the channel region below. Its electrical

properties are solely determined by the makeup and geometry of the different materials.

Generating these types of structures with suitable precision, repeatability at an appropriate cost

and volume is of central importance to the semiconductor industry and by extension to much of

the modern world.

The scale of modern devices can be on the order of only several nano meters, and they

continue to shrink. Since no mechanical manufacturing can practically be employed on these

scales, semiconductor devices are created by means of chemical, thermal and optical processes.

Plasmas play a role in a host of these processes such ion implantation used for doping [79], film

deposition via sputter processes [80,81], plasma enhanced chemical vapor deposition (PECVD)

[82,83] or atomic layer deposition (ALD) [84,85] as well as material removal by direct or

34
downstream chemical etching, through reactive ion etching (RIE) [86,87] or related processes such

as atomic layer etching (ALE) [88–91]. A typical set of steps to create a patterned film is shown

in Figure 1.13.

Figure 1.13: Manufacturing of strucured integrated circuit devices through deopsition, etch and pattern transfer: a)
metal film deposition; b) photoresist deposition; c) patterned optical exposure; d) photoresist development; e) film
etch; f) photoresist removal. Based on [16].

Starting with a blank substrate, a homogeneous film is deposited in a), followed by the

deposition of a photoresist (PR) film b). Photoresist is a light sensitive material that changes its

chemical structure when it is exposed to certain types of light, making it easier (positive PR) or

harder (negative PR) to etch. In c) parts of the photoresist are exposed to light based on a pattern

that partially blocks it. The exposed PR is subsequently removed through chemical etching, leaving

35
behind a patterned mask d). In e), through plasma based anisotropic etch processes this pattern is

transferred to the film by removing the material that is no longer covered and protected by the

photoresist. Finally, the photoresist is removed, leaving behind only the patterned film in f). While

this is an extremely simplified and idealized depiction of the deposition-litho-etch cycle it

nevertheless highlights some key requirements:

i) Sufficiently high etch and deposition rates as well as the ability to control them

precisely. High rates are highly desirable as this allows for throughput and subsequently

a higher production volume.

ii) High etch-selectivity between the different materials, “A” and “B”, 𝜉𝐴,𝐵 , which is

E
defined as the ratio of their respective etch rates”, 𝜉𝐴,𝐵 = EA . For example, in the model
B

process discussed above, high selectivity is required between the exposed and non-

exposed PR, the PR and the film as well as the film and the substrate.

iii) Controlled anisotropy of the etch process, since a purely isotropic etch process does

not allow for precise transfer of mask patterns and as shown in Figure 1.14. An isotropic

etch, as depicted in sub-figure a), causes undercut (etching underneath the mask), non-

straight sidewalls and overall insufficient control over the shape of the feature.

36
 Figure 1.14: a) Isotropic etch processes result in an uncontrolled and ultimately undesirable feature. b)
Ion bombardment enables anisotropic etching.

Technological plasmas are a well-established solution that meets these criteria due to some

of their fundamental working principles. The plasma’s ability to etch is not necessarily native to

the used neutral, unexcited gases. Often, (relatively) stable feedstock gases are used, and the

reactive components are generated within the etch chamber through dissociation or other

modification as a direct result of the highly energetic particles inside the plasma. Because of this,

the reactivity and selectivity of the etchant gas can be precisely controlled (including time

dependent control) via the electric power deposition into the plasma. Short lived species that

otherwise could not be supplied in a stable form can be utilized as they are actively generated on

site. Some of the removal processes are synergetic processes requiring high energy ions as well as

thermal neutrals. Because ions are non-neutral, their trajectories are sensitive to electric fields.

Imparting directionality onto the ions through acceleration in electric fields is the source for

anisotropic etching in most plasma systems. While this is not a complete list of reasons, it

37
nevertheless highlights the utility and relevance of plasmas in semiconductor etch processes. Due

to the shrinking dimensions of process nodes and trends toward ever increasingly complex 3-

dimensional structures the etching of high aspect ratio (HAR) features has become an area of

substantial technological challenges. HAR features in this context are defined as structures in a

substrate that possess narrow plasma facing openings with large etch depths. An example of HAR

features used for DRAM applications is shown in Figure 1.15. These high aspect ratio features are

more susceptible to geometric defects and reductions in etch rate, leading to increased process

requirements with respect to variability, ion energy distribution, etch rate etc.

Figure 1.15: Cross section SEM image of memory channel hole pattern after etching [131].

38
1.24 Adsorption

In the context of this work, adsorption is the adhesion of atoms, ions or molecules from a

gas to a surface and constitutes the basis for deposition and some surface modification steps. In

general, all adsorption processes can be described as an incident particle’s capture in a potential

well close to the surface. The depth, distance and general shape of that potential are determined by

the physical principle on which the adsorption is based. Potential diagrams corresponding to

generic physisorption and chemisorption processes are shown in Figure 1.16

Figure 1.16: Potential diagram of a generic physisorption and chemisorption process.

1.24.1 Physisorption

Physisorption, short for physical adsorption, is a surface adsorption mechanism based on

the Van der Waals Force. Physisorption possesses very low binding energy ( ∆𝜀𝑝ℎ𝑦𝑠 = 10 −

100 meV ) which means that is generally considered non-permanent as even room temperature

thermal energies can cause desorption. However, since this type of bond is rather universal with

respect to bonding partners and non-local it allows for significant surface mobility and surface

diffusion of physiosorbed species. Through this, physisorption sometimes acts as an intermediate

step where the adsorbed particle finds a stronger surface bond type after diffusing to a more

appropriate site. [92]

39
1.24.2 Chemisorption

Chemisorption is based on the formation of chemical bonds between surface and incident

species. In contrast to physisorption it is selective with respect to bonding partners as the

availability and nature of the bond is determined by the configuration of both partners outer shell

electron. Generally, the bonds are much stronger energy ( ∆𝜀𝑐ℎ𝑒𝑚 ≥ 500 meV ) and depending on

the interacting species, multiple bonds can be formed further increasing the total binding

energy.[92]

1.25 Etching and Sputtering

Different mechanisms for the removal of material can be used to achieve desirable etch

characteristics and control. Semiconductor manufacturing heavily relies on physical sputtering,

chemical etching and a combination of the two in the form of ion enhanced chemical etching.

Physical processes such as sputtering are usually anisotropic but fairly slow with very little control

over selectivity. Chemical etch processes can be tuned to be very selective, provide high etch rates

but usually act almost completely isotopically. The goal is to combine the benefits of multiple

fundamental basic mechanism to achieve a reliable, fast and efficient etch process with high

control over anisotropy and selectivity. Some basic processes relevant to plasma based material

removal processes are depicted in Figure 1.17

40
 Figure 1.17: Different types of material removing surface reactions. a) physical sputtering, b) thermal etching, c)
chemical removal of ion activated species and d) physical removal of chemically modified species.

1.25.1 Direct Sputtering

Sputtering is the ejection of surface species due to the bombardment by ions or fast neutrals.

The incident particle collides with the surface and transfers its energy to it. Above a certain

threshold energy 𝜀𝑡ℎ,𝑠𝑝𝑢𝑡 this leads to the ejection of on or more surface particles:

𝐴(𝑠) + 𝑀 + → 𝐴(𝑔) + 𝑀(𝑔) (1.60)

The ratio of the number of incident and ejected particles is called the sputter yield 𝛾𝑠𝑝𝑢𝑡 , which is

generally a function of energy and angle. While angular dependences vary based on specific details

such as the incident and sputtered species, crystalline structure or molecular composition, the

specific case or Cl sputtering SiO2, depicted in Figure 1.18 can nevertheless provide some insight.

Generally, sputter yield falls off drastically at narrow angles, because very little energy is

transferred in these glancing collisions. The maximum sputter yield is generally not a normal

incidence, because while it is optimal for energy transfer, it is not guaranteed to provide the best

41
result in terms of material ejection as the collisional cascade penetrates very deep and energy is

lost to a point where the energy transfer to the particles located on the surface is reduced. [16,93,94]

Figure 1.18: Etching yield of polysilicon and silicon dioxide by 100 eV Ar and Cl as a function of ion incident
angles for saturated chlorine fluxes. Reproduced from Chang et al [93].

1.25.2 Direct Chemical Etching

The process of a surface species 𝐴(𝑠) reacting with a gas phase species 𝐵(𝑔) creating a

volatile compound 𝐴𝐵(𝑔)

𝐴(𝑠) + 𝐵(𝑔) → 𝐴𝐵(𝑔) (1.61)

is referred to as direct etching or thermal etching. This process is almost invariably isotropic

because the relevant etchant species possess a uniform, thermal angular distribution. Thermal etch

processes are necessarily confined to a singular step but rather tend to occur in multiple subsequent

surface modification steps with the final one only forming a volatile compound:

𝐴(𝑠) + 𝐵(𝑔) → 𝐴𝐵(𝑠) (1.62)

𝐴𝐵(𝑠) + 𝐵(𝑔) → 𝐴𝐵2(𝑠) (1.63)

42
 𝐴𝐵2(𝑠) + 𝐵(𝑔) → 𝐴𝐵3(𝑠) (1.64)

𝐴𝐵3(𝑠) + 𝐵(𝑔) → 𝐴𝐵4(𝑔) (1.65)

For most processes the rates of reactivity of the surface decreases with increasing

attachment, i.e., the rate of reaction of (1.65) is lower than (1.64). Pure, thermal etch processes are

predominantly limited by the availability of the educts, given by the reactant fluxes to the surface

and density of available surface sites.

1.25.3 Ion Enhanced Etching

Ion enhanced etching or reactive ion etching refers to the simultaneous use of ion surface

bombardment and chemical etching with the aim to combine the high selectivity and etch rate of

chemical processes with the directionality of physical sputtering. This is commonly achieved by

using feed gas mixtures that contain or produce chemically active radicals in conjunction with a

CCP or CCP-like etch reactor where radicals and highly energetic ions are simultaneously

impinging on the wafer surface. In most scenarios, the removal process consists of multiple steps.

𝐴(𝑠) + 𝑀+ → 𝐴∗(𝑠) (1.66)

𝐴∗(𝑠) + 𝐵(𝑔) → 𝐴𝐵(𝑔) (1.67)

One of the mechanisms, depicted in Figure 1.17c) relies on ion surface activation. The high

energy ions carry and deposit enough energy to activate the surface. The nature of the surface

activation can differ, depending on the process, but is generally based on the breaking of bonds, to

lower the total surface bonding energy, and increase reactivity through the creation of dangling

bonds. The activated surface sites subsequently react with gas phase radicals similar to the

mechanism discussed in Section 1.25.2.

43
 Alternatively, the role of the ion bombardment can be the removal of more inert species

that were created by prior isotropic processes.

𝐴(𝑠) + 𝐵(𝑔) → 𝐴𝐵(𝑠) (1.68)

𝐴𝐵(𝑠) + 𝑀+ → 𝐴𝐵(𝑔) (1.69)

As discussed in Section 1.25.2 the reactivity of surface compounds can decrease as they form more

saturated species. Often, this process coincides with a reduction of surface bonds, which reduces

the total surface bonding energy, making them easier to remove by physical means such as

sputtering. In many scenarios, the total etch rate of the combined process can exceed the sum of

the individual processes. This synergy is generally highly desirable as it has desirable effects on

process times and etch anisotropy.

1.26 Aspect Ratio Dependent Etch

In the context of the ever-increasing requirements for aspect ratio of the etched features, a

major technical challenge is posed by aspect ratio dependent etching (ARDE) or RIE-lag. In HAR

etch process, as the etch front propagates downwards, a drastic reduction in etch rate is often

observed [95–97]. Generally, this is a consequence of transport limitations of neutrals as well as

ions. Neutral transport to the bottom of the feature is limited by a host of factors. In scenarios

where the mean free path of a neutral thermal particle is on the same scale or larger than geometric

restrictions, as is the case in narrow high aspect ratio features, transport is governed by Knudsen

diffusion which severely limits neutral gas transport deep into the feature. This is coupled with the

loss of radicals during sidewall collisions, the number of which increases with increases etch depth

and aspect ratio [98,99]. It has been shown that to some extent the ARDE can be controlled via the

gas pressure and substrate temperature [98,100,101]. Through an increase in pressure the lack of

44
gas transport to the feature bottom can be compensated by simply supplying more total flux, while

the lower temperature reduces the surface reactivity and reduces the loss due to surface reactions.

The ion transport effect on ARDE is mostly related to ion losses to the wall. The maximum

angle at which an incident ion is able to directly hit the etch front is the critical ion angle Θ𝑐 , shown

in Figure 1.19 for a low a) and high b) aspect ratio feature.

Figure 1.19: Critical ion angle of incidence for direct hit on the etch front for a) low and b) high aspect ratio feature.

As the feature etch progresses, the effective aspect ratio of the feature increases and Θ𝑐

decreases accordingly which in consequence leads to a decrease of total ion flux to the etch front.

This dependence on the ion angular distribution for HAR is one key factor for the requirement of

narrow ion distributions. The reduction of etch rate due to lack of neutral and ion transport to the

bottom of the feature are respectively known as the neutral and ion starved regime.

45
1.27 Intra Feature Charging

The highly anisotropic, near normal, high-energy ions can penetrate deeply into HAR

features whereas the electrons typically have nearly thermal and angularly broad distributions onto

the wafer [102]. When etching dielectric (or low conductivity materials), these conditions result in

the lower echelons of the feature being charged positively and the upper echelons being charged

negatively [103,104]. This differential charging generates electric field components within the

feature [103–105] The intra-feature electric fields can deviate ion trajectories from the vertical

which can lead to undesired consequences including a reduction of the ion flux at the bottom of

the feature [106], and profile distortion such as notching, twisting and bowing [103–105,107–110]

as shown in Figure 1.20.

Figure 1.20: Surface charge effects responsible for notching, trenching and twisting in HAR etch processes.

Several remedies have been investigated to minimize differential charging, including

operation with ever higher ion energies to minimize the effects of ion deflection. However, this

46
approach must overcome several challenges. Increasing the ion energy increases the power density

at the wafer surface, leading to wafer heating which, unchecked, can stress the thermal budget of

the process [22]. In HVM (high volume manufacturing) additional wafer cooling is applied. While

this strategy has been very successful it adds technological complexity to the etch system and is

ultimately subject to diminishing returns due to the limited heat transport of coolants and the wafer

itself.

Another challenge is that the sheath thickness is a function of the sheath voltage and, by

extension, is a function of the CCP power. If the source power and plasma density are relatively

constant, increasing the incident ion energy by increasing the applied bias voltage will ultimately

lead to a thicker and more collisional sheath. The more collisional sheath will produce a lower

energy and broader angular distribution of the incident ion flux. The higher ion energies are also

incident onto the photoresist or other hard masking materials, resulting in higher physical

sputtering rates, which then decrease the selectivity of the dielectric etch with respect to the mask.

Maintaining high selectivity is critical during the typically long etch times required for HAR

features [111].

Another approach to mediating differential charging is to produce enhanced fluxes of

negative charge carriers into the feature. Once such method utilizes power-pulsing to generate an

ion-ion plasma during the inter-pulse period. Acceleration of negative ions into the feature during

the interpulse period helps alleviate positive charging [112]. This technique is challenged by the

difficulty of sustaining a thin sheath during the low-plasma density afterglow, and so the ion

acceleration through the sheath is collisional. Another technique utilizes a negative DC bias on the

top electrode to generate an electron-beam-like flux of electrons into the feature [102].

47
1.28 Sidewall Passivation

While ion assisted etch processes possess some intrinsic anisotropy, some etching of the

sidewall occurs, nevertheless. This is due to non-ideal ion distributions that hit the sidewalls to

some (lesser) degree, a non-zero etch rate even without ions or a combination of the two. Often

this is addressed by sidewall passivation, either as an intrinsic part of the process or by intentional

design. Adsorption of passivating species on the sidewalls, as depicted in Figure 1.21 can produce

a protective layer that slows down or stops the lateral attack of etchant species and reduces

undercutting of the mask and bowing. The chemical makeup of this passivation layer is process

dependent and can be the result of polymer deposition from Fluorocarbon gas mixtures [113] or

the formation of a chemically inert oxide layer [114]. In many scenarios the sidewall passivation

mechanism is not self-limiting and excessive deposition of the passivating species causes a

reduction of the feature diameter known as necking or in extreme cases a complete clog of the

feature. This is especially relevant in HAR processes where the diameter of the opening is very

limited to begin with.

Figure 1.21: Sidewall passivation during the feature etch process.

48
1.29 Computational Modeling

Plasma processes are complex systems, usually operated in very controlled environments

that are sensitive to changes in geometry, materials and operating conditions which makes direct

measurements notoriously difficult. The measurement of some of the most fundamental plasma

properties is often performed using probes such as Langmuir probes [115,116], resonant probes

[117,118] or retarding field analyzers [119,120], which perturb the plasma and generally lack fine

spatial or temporal resolution. While some non-invasive techniques like optical measurements

[121–127], or electrical analysis exist, they can lead to significantly increased technical complexity

as well as cost and may still not capture all relevant quantities. Some of the plasmas fundamental

mechanisms may never be directly measured in a reasonable processing setup which has for a long

time been a major driver behind the development of computational models that aim to accurately

represent the inner workings of plasma processes, allowing for functional and most importantly,

flexible insight into otherwise opaque plasma processes. This culminates in the ultimate vision of

“digital twins” where entire etch processes are perfectly reproduced in simulations, allowing their

development and optimization to be performed on their digital counterpart ahead of or in tandem

with the actual physical entity, saving time, cost and providing absolute insight [128]. While this

vision has not fully materialized, computational modeling has nevertheless become an integral part

of academic and industrial research efforts in the field of low temperature plasmas and

semiconductor manufacturing processes. In reality, both surface and gas phase properties and

behavior are fundamentally defined by atomic or molecular interactions which, due to the sheer

number of particles, cannot be reasonably modeled in a direct approach on macroscopic scales

such as reactor or wafer dimensions, which are on the order of tens of cm. Because of this, the

particle interactions must either be reduced by clustering similar particles into superparticles on

49
static cells (kinetic approach) or reducing their transport behavior to idealized quantities such as

densities, fluxes, temperatures etc. (fluid approach). While the nature of the assumptions made

differ in detail and derivation, they nevertheless require the usage of condensed macroscopic

interaction parameters such as rates and probabilities which are ultimately supplied to the

simulation in the form of external datasets, either based on experimental data or from mor complex

ab initio models. The quality of these datasets can critically affect the validity of simulation results

and thus great care must be taken to ensure that they accurately reflect real world behavior.

50
1.30 Goals and Scope of this Dissertation

This work aims to identify and computationally test techniques to improve upon critical

aspects of HAR feature etch processes into SiO2 using fluorocarbon gas mixtures. This includes

the usage of special tailored voltage waveforms as a source of plasma heating and how they affect

neutral and charged particle dynamics as well as subsequent effects on feature etch performance

and wafer charging. Previous work has shown that the use of voltage waveform tailoring can be

used to control the DC self-bias and ion as well as electron incident energy onto the surface. The

motivation of these investigations was the control of HAR etching and mediation of feature charge

effects, but it has thus far not been applied to 2D systems or relevant gas compositions and neither

has the direct link to etch behavior been causally verified. To that end, this thesis employs well

tested computational models to

i. Verify the occurrence of the electrical asymmetry effect and electric field reversal,

induced by the usage of tailored voltage waveforms, in a different computational

model.

ii. Explore effects of the VWT that require a 2-dimensional setup such as effects on

discharge homogeneity, self-consistent treatment of geometry (especially geometric

asymmetry) and radial structures in plasma properties.

iii. Expand the tested range of gas chemistries to include electronegative gases and

fluorocarbons used in semiconductor fabrication.

iv. Provide a direct causal link between charged particle dynamics and etch behavior by

simulating HAR surface etch processes directly coupled to the gas phase simulations.

v. Test the established and extended theories of the link between ion energy and DC self-

bias in technologically relevant low frequency regimes.

51
 vi. Additionally, to address the stated issue of a model’s dependence on physical

parameters, this thesis provides a framework for the tuning and optimization of model

parameters by developing a semi-autonomous physics informed hybrid optimization

scheme and employing it to optimize a fluorocarbon based SiO2 etch mechanism.

In Chapter 2, the computational models used in this thesis are described. The Hybrid

Plasma Equipment Model (HPEM) was used to model the gas phase plasma phenomena and the

Monte Carlo Feature Scale Model (MCFPM) was used to represent the surface etch processes.

The use of tailored voltage waveforms for the generation of directional electrons and ions

is investigated in Chapter 3 with special focus on the gas composition in terms of electronegativity.

Building on the findings of Chapter 3, the investigation is expanded to appropriate etch gas

mixtures and includes the examination of feature etch processes. The results are presented in

Chapter 4.

The link between the ion energy control through plasma asymmetry and resulting DC-self

bias is investigated for the low frequency regime in Chapter 5.

Chapter 6 introduces and investigates the use of an automatic optimization scheme for

surface processes.

A summary of the studies in this thesis is presented in Chapter 7. Improvements to the

models and potential future studies to expand upon the work in this thesis are also discussed.

52
1.31 References

[1] Michael. Zeilik, E. v. P. (Elske van P. Smith and E. v. P. (Elske van P. Smith, Introductory

astronomy and astrophysics, Saunders College Pub (1987).

[2] P. Schneider, Extragalactic Astronomy and Cosmology Extragalactic Astronomy and

Cosmology, Springer Berlin Heidelberg (2015).

[3] A. R. Choudhuri, Astrophysics for Physicists Astrophysics for physicists, Cambridge

University Press (2010).

[4] B. W. Carroll and D. A. Ostlie, An Introduction to Modern Astrophysics, Second Edition

An Introduction to Modern Astrophysics, Second Edition, Cambridge University Press

(2019).

[5] M. Boulos, P. Fauchais and E. Pfender, Thermal Plasmas: Fundamentals and Applications,

(1994).

[6] V. A. Nemchinsky and W. S. Severance, J Phys D Appl Phys 39, R423 (2006).

[7] E. Menthe, K. T. Rie, J. W. Schultze and S. Simson, Surf Coat Technol 74–75, 412 (1995).

[8] J. F. Waymouth and F. Bitter, J Appl Phys 27, 122 (1956).

[9] A. Reatti, IEEE Trans Power Electron 15, 361 (2000).

[10] A. Kumar, N. Kumar, G. N. Pandey, al -, R. Gao, S. Mosquera-Romero, E. Ntagia, N.

Takeuchi and K. Yasuoka, Jpn J Appl Phys 60, SA0801 (2020).

[11] N. Puač, M. Gherardi and M. Shiratani, Plasma Processes and Polymers 15, 1700174

(2018).

[12] T. Von Woedtke, A. Schmidt, S. Bekeschus, K. Wende and K. D. Weltmann, In Vivo

(Brooklyn) 33, 1011 (2019).

[13] S. Bekeschus, T. von Woedtke, S. Emmert and A. Schmidt, Redox Biol 46, 102116 (2021).

53
[14] Z. Chen, G. Garcia, V. Arumugaswami and R. E. Wirz, Physics of Fluids 32, 111702 (2020).

[15] F. F. Chen, Introduction to Plasma Physics and Controlled Fusion, Springer (2006).

[16] M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges and Materials

Processing: Second Edition Principles of Plasma Discharges and Materials Processing:

Second Edition, John Wiley and Sons (2005).

[17] P. M. Bellan, Fundamentals of Plasma Physics, Cambridge University Press (2006).

[18] P. Chabert, T. V. Tsankov and U. Czarnetzki, Plasma Sources Sci Technol 30, 024001

(2021).

[19] John David Jackson, Classical Electrodynamics, Wiley (1998).

[20] D. J. Griffiths, Introduction to Electrodynamics Introduction to Electrodynamics,

Cambridge University Press (2017).

[21] E. M. Purcell and D. J. Morin, (2013).

[22] B. Wu, A. Kumar and S. Pamarthy, Journal of Applied Physics 108, 51101 (2010).

[23] S. J. Doyle, A. R. Gibson, R. W. Boswell, C. Charles and J. P. Dedrick, Plasma Sources Sci

Technol 29, 124002 (2020).

[24] P. Saikia, H. Bhuyan, M. Escalona, M. Favre, B. Bora, M. Kakati, E. Wyndham, R. S. Rawat

and J. Schulze, J Appl Phys 123, 183303 (2018).

[25] E. Kawamura, V. Vahedi, M. A. Lieberman and C. K. Birdsall, Plasma Sources Sci. Technol

8, (1999).

[26] D. Levko, R. R. Upadhyay, K. Suzuki and L. L. Raja, Journal of Vacuum Science &

Technology B 41, 12205 (2023).

[27] E. Kawamura, V. Vahedi, M. A. Lieberman and C. K. Birdsall, Plasma Sources Sci Technol

8, R45 (1999).

54
[28] S. D. Baalrud, B. Scheiner, B. T. Yee, M. M. Hopkins and E. Barnat, Plasma Sources Sci

Technol 29, 053001 (2020).

[29] M. S. Benilov and N. A. Almeida, Phys Plasmas 26, 123505 (2019).

[30] O. Murillo, A. S. Mustafaev and V. S. Sukhomlinov, Technical Physics 64, 1308 (2019).

[31] I. Adamovich, S. D. Baalrud, A. Bogaerts, P. J. Bruggeman, M. Cappelli, V. Colombo, U.

Czarnetzki, U. Ebert, J. G. Eden, P. Favia, D. B. Graves, S. Hamaguchi, G. Hieftje, M. Hori,

I. D. Kaganovich, U. Kortshagen, M. J. Kushner, N. J. Mason, S. Mazouffre, S. M. Thagard,

H. R. Metelmann, A. Mizuno, E. Moreau, A. B. Murphy, B. A. Niemira, G. S. Oehrlein, Z.

L. Petrovic, L. C. Pitchford, Y. K. Pu, S. Rauf, O. Sakai, S. Samukawa, S. Starikovskaia, J.

Tennyson, K. Terashima, M. M. Turner, M. C. M. Van De Sanden and A. Vardelle, Journal

of Physics D: Applied Physics 50, 323001 (2017).

[32] S.-J. Chung, P. Luan, M. Park, A. Metz and G. S. Oehrlein, Journal of Vacuum Science &

Technology B 41, 062201 (2023).

[33] Z. H. Bi, Y. X. Liu, W. Jiang, X. Xu and Y. N. Wang, Current Applied Physics 11, S2

(2011).

[34] Z. Donkó, A. Derzsi, I. Korolov, P. Hartmann, S. Brandt, J. Schulze, B. Berger, M. Koepke,

B. Bruneau, E. Johnson, T. Lafleur, J. P. Booth, A. R. Gibson, D. O’Connell and T. Gans,

Plasma Phys Control Fusion 60, 014010 (2017).

[35] T. Gans, J. Schulze, D. O’Connell, U. Czarnetzki, R. Faulkner, A. R. Ellingboe and M. M.

Turner, Appl Phys Lett 89, 261502 (2006).

[36] S. Rauf, K. Bera and K. Collins, Plasma Sources Sci Technol 19, 015014 (2009).

[37] W. Jiang, X. Xu, Z. L. Dai and Y. N. Wang, Phys Plasmas 15, 33502 (2008).

[38] T. Kitajima, Y. Takeo, N. Nakano and T. Makabe, J Appl Phys 84, 5928 (1998).

55
[39] J.-K. Liu, E. Kawamura, M. A. Lieberman, al - and S. Rauf, Plasma Sources Sci Technol

29, 095019 (2020).

[40] S. Rauf, K. Bera and K. Collins, Plasma Sources Sci Technol 17, 035003 (2008).

[41] G. Alcouffe, M. Cavarroc, G. Cernogora, F. Ouni, A. Jolly, L. Boufendi and C. Szopa,

Plasma Sources Sci Technol 19, 015008 (2009).

[42] A. Agarwal, S. Rauf and K. Collins, Plasma Sources Sci Technol 21, 055012 (2012).

[43] T. Lafleur, P. Chabert and J. P. Booth, Plasma Sources Sci Technol 23, 035010 (2014).

[44] P. Drude, Ann Phys 308, 369 (1900).

[45] P. Drude, Ann Phys 306, 566 (1900).

[46] R. P. Brinkmann, J Phys D Appl Phys 44, 042002 (2011).

[47] I. Langmuir, Physical Review 33, 954 (1929).

[48] L. S. Hall, Phys Fluids 4, 388 (1961).

[49] K. U. Riemann, J Phys D Appl Phys 24, 493 (1991).

[50] M. A. Lieberman, IEEE Transactions on Plasma Science 26, 955 (1998).

[51] E. Kawamura, M. A. Lieberman and A. J. Lichtenberg, Phys Plasmas 21, 123505 (2014).

[52] Z. Wang, A. J. Lichtenberg and R. H. Cohen, IEEE Transactions on Plasma Science 26, 59

(1998).

[53] M. A. Lieberman, IEEE Transactions on Plasma Science 16, 638 (1988).

[54] J. Schulze, B. G. Heil, D. Luggenhölscher, R. P. Brinkmann and U. Czarnetzki, J Phys D

Appl Phys 41, 195212 (2008).

[55] E. Kawamura, M. A. Lieberman and A. J. Lichtenberg, Phys Plasmas 13, 53506 (2006).

[56] C. Qu, S. J. Lanham, S. C. Shannon, S. K. Nam and M. J. Kushner, J Appl Phys 127, 133302

(2020).

56
[57] K. Kurokawa, IEEE Trans Microw Theory Tech 13, 194 (1965).

[58] Z. H. Bi, Y. X. Liu, W. Jiang, X. Xu and Y. N. Wang, Current Applied Physics 11, S2

(2011).

[59] H. H. Goto, H.-D. Löwe, T. Ohmi and H.-D. Lowe, Journal of Vacuum Science &

Technology A 10, 3048 (1992).

[60] W. Tsai, ; G Mueller, ; R Lindquist, ; B Frazier, ; V Vahedi, ) G Mueller, R. Lindquist, B.

Frazier and V. Vahedi, Journal of Vacuum Science & Technology B: Microelectronics and

Nanometer Structures Processing, Measurement, and Phenomena 14, 3276 (1996).

[61] H. H. Goto, H. D. Lowe and T. Ohmi, IEEE Transactions on Semiconductor Manufacturing

6, 58 (1993).

[62] A. Derzsi, E. Schüngel, Z. Donkó and J. Schulze, Open Chem 13, 346 (2015).

[63] Y. Xu, 徐轶君, X. Wu, 吴雪梅, C. Ye and 叶超, Plasma Science and Technology 15, 1066

(2013).

[64] S. Rauf, P. Tian, J. Kenney and L. Dorf, Journal of Vacuum Science & Technology B 40,

32202 (2022).

[65] S. Kirchhoff, Ann Phys 140, 497 (1845).

[66] R. A. Gottscho, G. R. Scheller, D. Stoneback and T. Intrator, J Appl Phys 66, 492 (1989).

[67] T. Kaneda, T. Kubota, M. Ohuchi and J. S. Chang, J Phys D Appl Phys 23, 1642 (1990).

[68] S. Maniv, J Appl Phys 63, 1022 (1988).

[69] R. A. Gottscho, G. R. Scheller, D. Stoneback and T. Intrator, J Appl Phys 66, 492 (1989).

[70] C. D. Child, Physical Review (Series I) 32, 492 (1911).

[71] I. Langmuir, Physical Review 2, 450 (1913).

57
[72] B. G. Heil, U. Czarnetzki, R. P. Brinkmann and T. Mussenbrock, J Phys D Appl Phys 41,

165202 (2008).

[73] J. Schulze, E. Schüngel, Z. Donkó and U. Czarnetzki, Plasma Sources Sci Technol 20,

15017 (2011).

[74] U. Czarnetzki, J. Schulze, E. Schüngel and Z. Donkó, Plasma Sources Sci Technol 20,

024010 (2011).

[75] J. Schulze, Z. Donkó, B. G. Heil, D. Luggenhölscher, T. Mussenbrock, R. P. Brinkmann

and U. Czarnetzki, J Phys D Appl Phys 41, 105214 (2008).

[76] P. Hartmann, L. Wang, K. Nösges, B. Berger, S. Wilczek, R. P. Brinkmann, T.

Mussenbrock, Z. Juhasz, Z. Donkó, A. Derzsi, E. Lee and J. Schulze, J Phys D Appl Phys

54, 255202 (2021).

[77] J. Schulze, E. Schüngel and U. Czarnetzki, J Phys D Appl Phys 42, 092005 (2009).

[78] T. Lafleur, Plasma Sources Sci Technol 25, 013001 (2015).

[79] S. Qin, Proceedings of the International Conference on Ion Implantation Technology

(2014).

[80] P. J. Kelly and R. D. Arnell, Vacuum 56, 159 (2000).

[81] Y. Yamamura and H. Tawara, At Data Nucl Data Tables 62, 149 (1996).

[82] G. Lucovsky, Adv Funct Mater 6, 55 (1996).

[83] L. Martinu, O. Zabeida and J. E. Klemberg-Sapieha, Handbook of Deposition Technologies

for Films and Coatings: Science, Applications and Technology 392 (2010).

[84] M. Leskelä and M. Ritala, Angewandte Chemie International Edition 42, 5548 (2003).

[85] R. W. Johnson, A. Hultqvist and S. F. Bent, Materials Today 17, 236 (2014).

58
[86] M. Huff, L. Romano and K. Jefimovs, Micromachines 2021, Vol. 12, Page 991 12, 991

(2021).

[87] F. Laermer, S. Franssila, L. Sainiemi and K. Kolari, Handbook of Silicon Based MEMS

Materials and Technologies 417 (2020).

[88] C. M. Huard, Y. Zhang, S. Sriraman, A. Paterson, K. J. Kanarik and M. J. Kushner, Journal

of Vacuum Science & Technology A: Vacuum, Surfaces, and Films 35, 031306

(2017).

[89] G. Antoun, T. Tillocher, P. Lefaucheux, J. Faguet, K. Maekawa and R. Dussart, Scientific

Reports 2021 11:1 11, 1 (2021).

[90] J. K. Kim, S. Il Cho, S. H. Lee, C. K. Kim, K. S. Min and G. Y. Yeom, Journal of Vacuum

Science & Technology A: Vacuum, Surfaces, and Films 31, 061302 (2013).

[91] S. Dallorto, A. Goodyear, M. Cooke, J. E. Szornel, C. Ward, C. Kastl, A. Schwartzberg, I.

W. Rangelow and S. Cabrini, Plasma Processes and Polymers 16, 1900051 (2019).

[92] P. W. Atkins, J. De Paula and J. Keeler, Oxford University Press, Oxford Physical

Chemistry, (2017).

[93] J. P. Chang and H. H. Sawin, Journal of Vacuum Science & Technology B:

Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena 19,

1319 (2001).

[94] G. Packard, A. Rosenfeld, G. S. Oehrlein and S. Hamaguchi, Plasma Sources Sci Technol

27, 023001 (2018).

[95] C. M. Huard, Y. Zhang, S. Sriraman, A. Paterson and M. J. Kushner, Journal of Vacuum

Science & Technology A: Vacuum, Surfaces, and Films 35, 5 (2017).

59
[96] R. L. Bates, M. J. Goeckner and Lawrence. J. Overzet, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 32, 51302 (2014).

[97] L. Meng, J. Li, C. Zhao and J. Yan, ECS Solid State Letters 3, Q25 (2014).

[98] T. Panagopoulos and T. Lill, Journal of Vacuum Science & Technology A 41, 33006 (2023).

[99] S. Gruener and P. Huber, Phys Rev Lett 100, 064502 (2008).

[100] T. Lill, ; M Grimbergen, ; D Mui, ) M Grimbergen and D. Mui, Journal of Vacuum Science

& Technology B: Microelectronics and Nanometer Structures Processing, Measurement,

and Phenomena 19, 2123 (2001).

[101] (2002).

[102] V. M. Donnelly and A. Kornblit, Journal of Vacuum Science & Technology A: Vacuum,

Surfaces, and Films 31, 050825 (2013).

[103] B. Radjenović and M. Radmilović-Radjenović, Engineering 06, 1 (2014).

[104] C. Han, Z. Wu, C. Yang, L. Xie, B. Xu, L. Liu, Z. Yin, L. Jin and Z. Huo, Semicond Sci

Technol 35, 045003 (2020).

[105] B. M. Radjenović, M. D. Radmilović-Radjenović and Z. L. Petrović, IEEE Transactions on

Plasma Science 36, 874 (2008).

[106] H. Jansen, M. De Boer, R. Wiegerink, N. Tas, E. Smulders, C. Neagu and M. Elwenspoek,

Microelectron Eng 35, 45 (1997).

[107] R. J. Belen, S. Gomez, M. Kiehlbauch and E. S. Aydil, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 24, 350 (2006).

[108] N. Negishi, M. Miyake, K. Yokogawa, M. Oyama, T. Kanekiyo and M. Izawa, Journal of

Vacuum Science & Technology B, Nanotechnology and Microelectronics: Materials,

Processing, Measurement, and Phenomena 35, 051205 (2017).

60
[109] G. S. Hwang, Journal of Vacuum Science & Technology B: Microelectronics and

Nanometer Structures 15, 70 (1997).

[110] T. Nozawa, T. Kinoshita, T. Nishizuka, A. Narai, T. Inoue and A. Nakaue, Jpn J Appl Phys

34, 2107 (1995).

[111] D. Zhang, S. Rauf and T. Sparks, IEEE Transactions on Plasma Science 30, 114 (2002).

[112] A. Agarwal, S. Rauf and K. Collins, J Appl Phys 112, 33303 (2012).

[113] G. S. Oehrlein and Y. Kurogi, Materials Science and Engineering: R: Reports 24, 153

(1998).

[114] M. Haverlag, ; G S Oehrlein, ; D Vender, G. S. Oehrlein and D. Vender, Journal of Vacuum

Science & Technology B: Microelectronics and Nanometer Structures Processing,

Measurement, and Phenomena 12, 96 (1994).

[115] J. Hopwood, C. R. Guarnieri, S. J. Whitehair and J. J. Cuomo, Journal of Vacuum Science

& Technology A 11, 152 (1993).

[116] R. L. Merlino, Am J Phys 75, 1078 (2007).

[117] J. Oberrath, M. Friedrichs, J. Gong, M. Oberberg, D. Pohle, C. Schulz, C. Wang, P.

Awakowicz, R. P. Brinkmann, M. Lapke, T. Mussenbrock, T. Musch and I. Rolfes, IEEE

Transactions on Plasma Science 49, 3293 (2021).

[118] N. S. J. Braithwaite and R. N. Franklin, Plasma Sources Sci Technol 18, 014008 (2008).

[119] Y. X. Wei, S. Q. Liu, X. X. Li, H. L. Shen, M. G. Huang and P. K. Liu, Nucl Instrum

Methods Phys Res A 785, 123 (2015).

[120] Y. Yuan, J. Bansky, J. Engemann and A. Brockhaus, Surf Coat Technol 74–75, 534 (1995).

[121] E. Slikboer, K. Acharya, A. Sobota, E. Garcia-Caurel and O. Guaitella, Sci Rep 10, 2712

(2020).

61
[122] B. Huang, C. Zhang, I. Adamovich, Y. Akishev and T. Shao, Plasma Sources Sci Technol

10, 044001 (2020).

[123] T. Darny, G. Bauville, M. Fleury, S. Pasquiers and J. Santos Sousa, Plasma Sources Sci

Technol 30, 105021 (2021).

[124] L. Invernizzi, N. Sadeghi, F. P. Sainct and P. Guillot, Plasma Sources Sci Technol 31,

035002 (2022).

[125] S. Schröter, A. Wijaikhum, A. R. Gibson, A. West, H. L. Davies, N. Minesi, J. Dedrick, E.

Wagenaars, N. De Oliveira, L. Nahon, M. J. Kushner, J. P. Booth, K. Niemi, T. Gans and

D. O’Connell, Physical Chemistry Chemical Physics 20, 24263 (2018).

[126] C. Y. T. Tschang, R. Bergert, S. Mitic and M. Thoma, J Phys D Appl Phys 53, 215202

(2020).

[127] G. Nayak, M. S. Simeni, J. Rosato, N. Sadeghi and P. J. Bruggeman, J Appl Phys 128,

243302 (2020).

[128] D. R. Gunasegaram, A. B. Murphy, A. Barnard, T. DebRoy, M. J. Matthews, L. Ladani and

D. Gu, Addit Manuf 46, 102089 (2021).

[129] A. D. Kuypers and H. J. Hopman, J Appl Phys 67, 1229 (1990).

[130] E. Kawamura, V. Vahedi, M. A. Lieberman and C. K. Birdsall, Plasma Sources Sci Technol

8, R45 (1999).

[131] Y. Kihara, M. Tomura, W. Sakamoto, M. Honda and M. Kojima, 2023 Symposium on VLSI

Technology and Circuits, (2023).

62
 Chapter 2 : Description of the Models

This thesis is based on computational models that capture the gas phase dynamics and

surface evolution of the coupled etch process. Due to the significant difference in relevant time

and length scales, the coupled system consists of two discrete models representing the gas phase

and surface evolution separately. The modeling suite used in this work is schematically shown in

Figure 2.1.

Figure 2.1: Flowchart of coupled modeling suite consisting of HPEM and MCFPM.

63
 The Hybrid Plasma Equipment Model (HPEM) is used to simulate the reactor scale gas

phase and is described in Section 2.1. The surface incident fluxes of electrons, ions and neutrals

as well as their energy and angular distribution are generated by the HPEM and used as input for

the feature scale surface model. The Monte Carlo Feature Scale Model (MCFPM) simulates the

actual etch process in three dimensions and generates, among others, the temporal evolution of the

feature geometry, composition and electric potential within it. It is described in Section 2.2.

2.1 Hybrid Plasma Equipment

The Hybrid Plasma Equipment Model (HPEM) is a 2-dimensional model which resolves

plasma phenomena in a time-slicing approach and incorporates fluid treatment for all species with

the option for limit kinetic representation of charged species. HPEM as a whole is based on a

modular paradigm where each module addresses specific physical regime and is coupled to others

by exchanging physical quantities – electric and magnetic fields, densities, rate coefficients etc.

The major modules used in this work are the Fluid Kinetics Poisson Module (FKPM), the Electron

Energy Transport Module (EETM), and the Plasma Chemistry Monte Carlo Module (PCMCM)

[1].

The coarse program flow is depicted in Figure 2.1. The electron distributions relevant to

their impact reactions are obtained in the EETM, by employing a kinetic description using a Monte

Carlo approach. The electron impact reaction rates k (r ,  ) are used to determine source terms for

electron impact reaction S (r ,  ) . The spatially resolved source terms are used as inputs to the Fluid

Kinetics Poisson Module (FKPM). FKPM calculates the densities N (r ) , velocities v (r ) , and the

temperatures T (r ) of the heavy as well as their reaction rate coefficients and source functions. The

charged particle species densities are used to determine the local space charge density and based

64
on this, and the boundary conditions set by the metals (grounded or driven) Poisson’s equation is

solved to provide the electrostatic electric field ES (r ,  ) and the potential  P (r ,  ) . The spatial

field configuration is passed to the Plasma Chemistry Monte Carlo Module (PCMCM). PCMCM

tracks the trajectories of select species using a kinetic Monte Carlo method and records their energy

and angular distribution with which they hit certain surfaces (In this thesis it is exclusively used

for the bottom mounted wafer).

The specific time step requirements of the respective modules can be different, so the A

time slicing approach is employed in which the submodules communicate on the basis of a global

iteration which is generally defined by the longest required timestep.

2.1.1 Electron Energy Transport Module (EETM)

The EETM calculates the rate coefficient from electron impact reactions k (r ,  ) and

corresponding source function S (r ,  ) by using the electrostatic electric field ES (r ,  ) , as well as

heavy particle densities N (r ) provided by the FKPM. In this thesis The EETM then obtains the

electron properties by using the electron Monte Carlo Simulation (eMCS) to compute electron

energy distribution functions (EEDFs). This includes the treatment of secondary electrons that are

emitted from surfaces subject to ion bombardment.

The eMCS is a fully kinetic approach in which electrons are represented by

pseudoparticles. Pseudoparticle refers to an infinitesimally small cluster of particles that has

physical properties (mass, charge etc.) that correspond to a large number of particles. This allows

for kinetic treatment without the need to simulate every individual electron while still effectively

reproducing macroscopic quantities such as number density charge density etc. The electrons are

initialized according to a Maxwellian distribution. Spatially they are initially distributed according

65
to the background fluid ion density. In the electrostatic scenario, electrons are acted upon by the

electric fields

dve e
= E, (2.15)
dt me

where ve is the electron velocity and E is the total local electric fields (electromagnetic and static).

The electrons are first sectioned by energy into coarse non uniform bins, e.g., 0-5, 5-10, 10-50, 50-

300, and 300-1000 eV. the lower energy ranges (< 10 eV) are typically associated with bulk

electrons while the higher ones correspond to beam or secondary electrons which were accelerated

by sheath electric field. Following the coarse sorting, the collision frequency is calculated on a

finer bin resolution according to

12
 3 
i =  i   ijk Nj , (2.16)
 me  j ,k

where  i is the average energy in bin i,  ijk is the cross-section of the species j of process k at

energy  i , and Nj is the density of the species j. To ensure a constant time step without collision

rate dependence, a null collision approach is applied in the eMCS, which yields a constant timestep

t = − ln(r ) /  mj , where r is a random number distributed within (0, 1) and is the maximum collision

frequency in energy range j. A null collision frequency is defined as the difference between the

maximum collision rate and the real collision frequency. Each timestep, based on a random

number, the occurrence of a collision is checked. In the absence of a collision (a null collision) the

electron velocity and energy remain unchanged. Otherwise, the electron energy is modified according

to the inelastic or elastic nature of the collision and the electron trajectory is scattered [2]. To obtain

the EEDFs, electron energies are recorded and binned by energy (i) and space (l)

66
  1  
Fil =  w j   i   i  −  i   ( rl  r ) − rl  , (2.17)
j  2  

where wj is the relative particle weight, which is determined by

• the weight of the pseudoparticle i.e. the number of actual particles it represents,
• the time step used to advance the particle trajectory and
• spatial weighting.

After each time step the obtained EEDFs f e ( , r ) are normalized according to

 F  =  f ( , r ) 
i
ij i
i
e
12
i  i = 1 . (2.18)

While the kinetic treatment for bulk electrons is optional, it is always performed for beam

and secondary electrons as their distributions are always assumed to be non-Maxwellian and highly

directional. They are tracked until they are removed via gas phase or surface collisions or until

their energy falls below the lowest inelastic collision threshold in which case, they are transformed

into bulk electrons by adding them to the bulk electron source term.

Kinetic treatment allows for highly precise tracking of electron properties that is entirely

self-consistent and does not require any assumptions about their distribution. The quality of the

kinetic approach is strongly dependent on the statistical quality which generally means that a large

number of pseudoparticles is required. This can lead to high computational cost which, fortunately,

is very compatible with parallel treatment which has been implemented into HPEM using the

OpenMP API [3].

2.1.2 Fluid Kinetics Poisson Module (FKPM)

The Fluid Kinetics Poisson Module (FKPM) is responsible for the handling of the heavy

particle densities, electron energy equation and the electrostatic potential. If the kinetic treatment

of electrons is not required, alternatively the FKPM determines electron transport properties and

67
the electron rate coefficients by solving Boltzmann’s equation for a range of values of the reduced

electric field (E/N). The electron distribution function fe ( , r ,  ) is obtained by solving the

Boltzmann equation using the two-term approximation given by equation (1.7). A table is

generated and its interpolated values are referenced during the execution of the code.

Consequently, the electron temperature (Te) is obtained using the electron power balance equation

3
( ne kBTe )
2 = Te +   ( eTe ) = Pe , (2.20)
t

with the Boltzmann constant kB , the thermal conductivity κ the electron flux e and the total power

deposited into electrons

Pe = j e  E = ee  E . (2.21)

The electron continuity equation (see Eq 1.10) is solved in either a drift-diffusion formulation

given by

e = ee ne E − Dene , (2.22)

where µe is electron mobility, and De is the electron diffusion coefficient or in the Scharfetter-

Gummel (S-G) expression. The S-G method is widely used to describe the transport of charged

particles in the fluid model [4–6]. With a S-G expression, the flux between the mesh points i and

i+1 is

 D[ni +1 − ni exp(x)]
i +1 2 = , (2.23)
1 − exp(x)

with

  − i 
 = −e  i +1 , (2.24)
 x 

68
where D is the averaged diffusion coefficient between vertex i and i+1,  and x are the averaged

mobility and the length of this interval, and Φi is the potential on vortex i. To attain the heavy

particle densities, the coupled system consisting of the continuity, momentum, and energy

equations is solved. For species i with a source term Si

N i
= −  i + Si , (2.25)
t

 i ( Ni vi ) 1 q
= = − (kNiTi ) −   ( Ni vi vi ) + i Ni ( Es + vi  B)
t t mi mi

mj
−  i −  Ni N j (vi − v j )vij , (2.26)
j mi + m j

Ni i
= − Ti − pi vi −  ( Ni vi i ) + qii E − (ivi )
t

− kmij Ni N j i +  kmjl N j Nl  mjl , (2.27)

m, j m , j ,l

where  is the flux, N is the density, v is the velocity, m is the mass, T is the temperature,  is

the viscosity, p is the pressure, and ε is the energy. Since the electrostatic potential is solely

determined by the charge density and the boundary conditions, it is solved for using Poisson’s

equation


  (t + t ) = −  (t + t ) = −  ( t ) − t , (2.28)
t t +t

where  is the charge density. The total charge density is comprised of the surface charge on

materials and the sum of all gas phase charge carries,

 ( t ) = m ( t ) +  qi Ni (t ) , (2.29)
i

69
where the first term is the charge density from the solid material, and the second term is the charge

from the plasma. Employing a semi-implicit approach for solving Poisson’s equation, allows the

use of time steps t larger than the dielectric relaxation time, which would otherwise be required


by explicit methods [7]. If Scharfetter-Gummel fluxes are used, the term can be written as
t t +t

 m (t ')   
= − qe, j     e (t ) + e [ s (t + t ) −  s (t )] 
t t +t t   

 t  i (t ) 
− qi     i (t ) + , (2.30)
i  2 t 

where qe, j represents the charge of electrons and ions, respectively,  S is the electrostatic

potential. t’ means that the charge density is evaluated at the current time step t, but the potential

is evaluated at the time step t + t .

e
The Jacobian term from Eq. (2.30) is addressed via a perturbation method, where a


small fractional potential change (typically = 5%) ΔΦ is applied within Δt by solving two first-order

partial derivatives of the electron flux with respect to the potential

i , j i , j (i +1, j + i +1, j ) − i , j (i +1, j )

= , (2.31)
i +1, j i +1, j

where i and j are the coordinates of the mesh in the radial and axial direction. The discretized

equations mentioned above are solved using either Successive-Over-Relaxation or direct sparse

matrix method [8,9].

In addition to the plasma, the potential calculation includes all solid materials. The FKPM

discriminates between two types of materials. Dielectrics are included in the solution or the

potential and are thus defined by their charge density, conductivity and permittivity. The total

70
material charge is defined by the sum of charged particle fluxes impinging on it, the secondary

electrons emitted from the surface, and the conduction current within the material.

Metals are considered to be free of internal electric fields and perfectly equipotential

throughout their entire volume. Metal materials effectively act as Dirichlet boundary conditions

where the magnitude of the potential is externally defined. On grounded metals it is set to 0 while

the potential on powered materials is defined by the applied potential and, if directly connected to

the blocking capacitor, the DC self-bias Vdc.

V ( t + t ) = VRF (t ) + Vdc , (2.32)

where VRF (t ) is the rf potential at time t. As discussed in the Introduction, Vdc is a function of the

systems total asymmetry and is numerically determined by evaluating the total accumulated charge

on the capacitor

1   ( E  nˆ ) 
Vdc =
C   m   (q
i
i
j
j j + q ij )  nˆ +
t
 dt ,

(2.33)

where C is the magnitude of the blocking capacitance, i is the index of the materials, and j is the

index of the particles. The first term on the right-hand side is the conduction current carried by ion

and secondary electron fluxes. γij is the secondary electron emission coefficient of material i when

bombarded with particle j. n̂ is the normal vector of the electrode surface. The second term is the

displacement current on the electrode. The contribution of each metal surface is defined by whether

it is directly connected to the grounded (𝑚𝑖 = −1) or powered side (𝑚𝑖 = 1) of the circuit. The

power on the electrode is

1 
 j (r ,t ) +
(
d  E (r ,t ) ) dtdA ,
Prf =
  V (t )
 dt 
(2.34)
 

71
where V(t) is the time-dependent voltage, j is the conduction current density, τ is the rf integration

time, and A is the surface area of the electrode. The second term in Eq. (2.34) is the displacement

current on the electrode, where ε is the electrode permittivity and E is the time-varying electric

field.

2.1.3 Surface Kinetics Module (SKM)

Surface interactions can significantly affect the discharge composition and by extension its

dynamics as a whole either in the form of surface losses or chemical interaction. This is especially

relevant for reactor setups with high surface to volume ratios as is the case with most CCP setups.

Simple surface reactions can be addressed using the Surface Kinetics Module (SKM). Lasting

surface modification and resulting changes to reactivity can be included via the Surface Site

Balance Model (SSBM) within SKM. The SSBM tracks fractional surface site coverage and

implements it into the reaction rate equations in the form of a variable sticking coefficient.

Modification of the surface, for example by attachment and subsequential site blocking changes

the local site coverage and sticking coefficient. [10]

The flux ( in ) incident to a surface is provided by FKPM. A surface reaction generally has

the form

A(g) + B(s) → C(s) + D(g) + E(g), (2.38)

where g stands for gas phase species, and s denotes the surface sites. The rate of ith reaction between

gas phase species A and surface sites B on material m is:

Rim =  i Am Bm , (2.39)

72
where αi is the reaction rate of ith reaction, ϕAm is the flux of gas phase species A on material m,

θBm is the fractional coverage of surface site B on material m which is updated according to the

sum of all relevant reaction rates.

The surface coverage ratio, of all the surface sites, is attained by integrating the rate

equations of the corresponding sites using a third-order Runge-Kutta technique. The fractional

surface coverage along with the gas phase fluxes from the surface, out , is returned to the FKPM,

through which the surface reactions feedback to the plasma. The reaction probability of gas phase

species on the surface is the summation of the rates of reactions with it as a reactant.

2.1.4 Plasma Chemistry Monte Carlo Module (PCMCM)

The Plasma Chemistry Monte Carlo Module (PCMCM) is used to record the energy

angular distributions of neutrals as well as ions incident on predefined relevant surfaces, most

notably the wafer surface in this thesis. In the present scenario it is executed at the end of the

simulation, after the plasma has reached a quasi-steady state and thus does not couple its

information back into the gas phase simulation, though it can also be executed throughout the

simulation when necessary. The PCMCM takes as input the particle densities and field

configuration. As the name suggests, it is based on a kinetic Monte Carlo approach.

Pseudoparticles are launched spatially distributed according to their density and at velocities based

on their Maxwellian temperature distribution obtained from the FKPM. The particles are tracked

until they hit a surface and are, if they are charged, subject to the forces of the electric field. Similar

to the eMCS, collisions in the PCMCM are also handled using the null-collision technique.

The time step used to integrate the trajectory is based on the time to traverse a grid cell, the

mean time between collisions and the RF period. It is set to be a fraction of the smallest of the

aforementioned time scales; the fraction is usually between 0.2 and 0.5 within the plasma bulk.

73
Because the charged particles are accelerated to high energies as they cross the sheath potential

this technique ensures an accurate representation of particle velocity and energy. Although

generally, neutrals are close to thermally distributed as they make contact with the surface, it is

possible to encounter ‘hot neutrals’ which can result from a highly energetic ion neutralizing in a

charge exchange collision while maintaining most of its kinetic energy. As both ions and hot

neutrals can carry a significant amount of energy to the surface both are recorded as the Ion/Neutral

energy angular distribution (IEAD and NEAD respectively).

2.2 Monte Carlo Feature Profile Model (MCFPM)

The Monte Carlo Feature Profile Model (MCFPM) is a voxel-based feature scale simulator

used to investigate the physics of feature scale etch processes. It has the option to model 2D and

3D processes, while in this thesis, only the 3D geometry is used [11,12]. In the MCFPM, the

surface material is represented by cubic voxels. Each voxel preserves its assigned material

properties. Pseudoparticles representing gas phase species are launched with energies and angles

sampled from the EADs obtained from the HPEM. The trajectories of the incoming particles are

integrated with charged particles subject to acceleration by electric fields, until a collision with a

surface occurs. Based on the species of the colliding pair, the incident energy of the gas phase

species, and the reaction probability, the outcome of the bombardment is determined. The specific

surface reaction is chosen based on a stochastic Monte Carlo approach. Depending on the reaction,

the voxel representing the surface site is removed (physical or chemical sputtering), chemically

modified (passivation), or a voxel is added on top of the site (deposition). Species reflected from

or produced at the surface, such as a sputter or chemical etch product, are returned to the gas phase,

and their trajectories are tracked until they are consumed in a subsequent reaction at a surface or

leave the feature. Although the capability exists in the MCFPM to have electron stimulated surface

74
chemistry or electron implantation, in this thesis, electrons do not modify the surface other than

by depositing their charge.

Apart from the state of material identities, the MCFPM tracks a variety of other quantities

such as the position of the etch front, the spatially resolved magnitude of surface fluxes, the change

to energy and angular particle distribution as a function of feature height, the flux into and out of

the simulation domain, the electric charge and potential ( see 2.2.2 ) as well as the state of surface

cross linking ( see 2.2.3 ). The boundaries in r and y direction (horizontal as well as normal to the

viewing plane, respectively) can be reflective or periodic while the top and bottom act as perfect

sinks with the added caveat that the simulation is stopped once a particle reaches the lower

boundary.

2.2.1 Energetic Particle Surface Reaction

For reactions between thermal neutrals and surface sites the reaction probability is usually

assumed to be independent or the angular dependence and carried kinetic energy because it is

negligibly small compared to relevant energy thresholds. However, for highly energetic particles,

such as ions and hot neutrals, properties such as the rate of reaction, sputter yield and details of the

scattering can strongly depend on the specific incident energy and angle [13–17]. In MCFPM this

dependence is implemented by scaling the respective reaction probabilities with functional

representations of angle and energy

n
 E − Eth 
p( Ei , ) = p0  i  f ( ) , (2.40)
 Er − Eth 

where Ei is the particle energy, θ is the incident angle, p0 is the reference probability, Er is the

reference energy, Eth is the threshold energy, and f(θ) is the angular dependent function. When

75
bombarding on the surface, energetic particles can cause physical sputtering or chemical enhanced

etching, any of which may have different angular dependence.

In high aspect ratio features, energetic particles are likely to undergo collisions with

the sidewall, during which they lose a fraction of their kinetic energy and are scattered. The energy

of a particle after a surface collision Es is

 E − Ec    − c 
Es ( ) = Ei  i    , (2.41)
 Ets − Ec   90 − c 

where Ets is the threshold energy for specular scattering, Ec is the cut off energy of diffusive

scattering, and θc is the lower angular boundary for specular scattering. Particles with Ei > Ets

undergo a purely specular scattering event, i.e., the preserve all the energy and the exit angle equals

the angle of incidence. Particles with Ei < Ec, or θ < θc are treated as partially or fully be diffusively

scattered, that is to say, they lose a fraction of their energy and gain a randomly assigned diffusive

angular component.

2.2.2 Electrostatic Charging

Electrostatic charging of features results from ions or neutrals which neutralize upon

striking surfaces and deposit their charge at the impact site. Charge is then retained on the voxel

upon which it is deposited until neutralized by an opposing charge or transported via conductive

charge transport. In the case of the removal of a charged voxel, the charge is retained in the mesh

by redistributing it to the adjacent mesh cells. At sufficiently spatially disparate charge distribution,

the resulting electric potential and electric fields affect the trajectories of charged particles via the

addition of a Lorentz force term.

Based on the assumption that the time averaged flux of charged species onto the surface is

net neutral, the electron flux is adjusted accordingly to enforce that criterion globally. Local charge

76
transport is not balanced explicitly and is computed self consistently. A detailed description of the

MCFPM charging mechanism is given in Wang et al [18] and briefly summarized below. The time

rate of change of charge density in numerical cell k, k (C/cm3) is

d k qw
=  i i −   k k E , (2.35)
dt i Vk

where E = − is the electric field,  is the electric potential, and the sum is over incident

particles i having weighting wi and carrying charge qi. The numerical cell has volume Vk and the

charge has electrical mobility k in that cell. The electric potential is obtained by implicitly solving

Poisson’s equation, −   =  , using finite volume techniques. Computationally, this is

performed using the Successive-Over-Relaxation algorithm employing a parallel red-black

technique. Each material included in the simulation is assigned a dielectric constant and mobilities

for positive and negative charge transport, corresponding to the hole and electron mobilities.

Compared to the other aspects of the simulation, directly solving Poisson’s equation is extremely

computationally expensive. In order to reduce the computational cost, the evaluation of the charge

and potential is only performed after a large amount of charged particles have deposited their

charge. Reflective boundary conditions for electric potential are used in the lateral (x-y)

dimensions and a zero-gradient boundary condition is used on the top surface of the computational

domain. The bottom of the computational domain is grounded. All positive ions neutralize upon

their first collision with a surface and return as a hot neutral that, apart from no longer being affected

by the electric field usually behave identically to their ion counterpart.

2.2.3 Crosslinking Module

A surface species’ (especially polymer’s) physical and chemical properties can be

dependent on mutual links with neighboring species. This can result in changes to the reactivity

77
due to a lack of available bonds or more resistance to physical processes due to increases in total

surface bond strength. Since these bonds can be broken by exposure to highly energetic particles

or radiation, for example through exposure to plasma, this spatially discriminate activation can

result in anisotropic shapes if deposits in the neck area of the feature. The general workings of the

implanted crosslinking model are shown in Figure 2.2; the crosslink creation during the polymer

deposition is depicted in the left column (a-d) and the removal in the right (e-g).

Figure 2.2: Schematic representation of the crosslinking process.

78
 Crosslinking occurs during the deposition of eligible materials (Figure 2.2a) and b). Each

material has a maximum number of crosslink partners associated with it, which is based on the

number of available bonds (3 in the example depicted in Figure 2.2). During deposition bonds to

random eligible cell neighbors can be formed, increasing the respective crosslink number, which

is tacked for every cell (Figure 2.2c). If crosslinking occurred, the respective cell identities are

changed to represent the physical differences and incur the changes to the physical properties cell

(Figure 2.2d).

Crosslinks can be broken by impinging particles such as ions, hot neutrals and photons

(Figure 2.2e) The bond breaking is represented by incrementing the affects surface cell’s bond

number (Figure 2.2f) and its reversion back to an unlinked state (Figure 2.2g).

79
2.3 References

[1] M. J. Kushner, J Phys D Appl Phys 42, 194013 (2009).

[2] S. H. Song and M. J. Kushner, Plasma Sources Sci Technol 21, 055028 (2012).

[3] L. Dagum and R. Menon, IEEE Computational Science and Engineering 5, 46 (1998).

[4] R. Chandra, Morgan Kaufmann (2001).

[5] C. M. Snowden, Published in 1986 in Singapore by World scientific 222 (1986).

[6] D. L. Scharfetter and H. K. Gummel, IEEE Trans Electron Devices 16, 64 (1969).

[7] P. L. G. Ventzek, T. J. Sommerer, R. J. Hoekstra and M. J. Kushner, Appl Phys Lett 63,

605 (1993).

[9] W. H. Press, S. A. Teukolsky, W. T. Vetterling and B. P. Flannery,

[10] D. Zhang and M. J. Kushner, Journal of Vacuum Science & Technology A 19, 524 (2001).

[11] A. Sankaran and M. J. Kushner, Journal of Vacuum Science & Technology A 22, 1242

(2004).

[12] C. M. Huard, Y. Zhang, S. Sriraman, A. Paterson, K. J. Kanarik and M. J. Kushner, Journal

of Vacuum Science & Technology A: Vacuum, Surfaces, and Films 35, 031306

(2017).

[13] C. F. Abrams and D. B. Graves, J Appl Phys 86, 2263 (1999).

[14] C. F. Abrams and D. B. Graves, J Appl Phys 86, 5938 (1999).

[15] G. S. Oehrlein, P. J. Matsuo, M. F. Doemling, N. R. Rueger, B. E. E. Kastenmeier, M.

Schaepkens, T. Standaert and J. J. Beulens, Plasma Sources Sci Technol 5, 193 (1996).

[16] Y. Yamamura and H. Tawara, At Data Nucl Data Tables 62, 149 (1996).

[17] P. J. Kelly and R. D. Arnell, Vacuum 56, 159 (2000).

[18] M. Wang and M. J. Kushner, J Appl Phys 107, 023309 (2010).

80
 Chapter 3 : Electric Field Reversals Resulting from Voltage Waveform Tailoring in Ar/O2
Capacitively Coupled Plasmas Sustained in Asymmetric Systems

3.1 Introduction

As described in Chapter 1, plasma etching of dielectrics such as SiO2 and Si3N4 is an

integral process in nearly all micro- and nanometer scale electronics fabrication [1] and the etching

of high aspect ratio (HAR) features is becoming an increasingly critical process due to the

development of 3-dimensional structures such as 3D-NAND memory which requires contact holes

through more than 256 alternating layers of SiO2 and Si3N4 with a total aspect ratio of up to

100.[2,3] Typical operating conditions are tens of mTorr gas pressure with hundreds of watts to

many kW power deposition over a 30 cm diameter wafer. [4,5] These conditions produce ion

fluxes to the wafer of 1015-1016 cm-2s-1.[6] To enable these ion fluxes (or hot neutral fluxes after

ion scattering from sidewalls) to reach the bottom of HAR features with sufficient energy and

narrow angular distribution to continue the etch, applied voltages to the substrate are as high as 5

kV while endeavoring to minimize ion collisions inside the sheath.[7]

The electric field in the sheath above the wafer points towards the wafer surface during the

majority of the RF cycle, accelerating positive ions to high energies and narrow angular spreads.

This same electric field confines electrons to the bulk plasma. It is only during a small fraction of

the RF cycle at which time the sheath collapses that electron flux from the bulk plasma reaches

the wafer.

81
 In a CCP of the type used for plasma etching of dielectrics in microelectronics fabrication

(pressures of tens of mTorr, frequencies of 1 – 100 MHz), the net charged flux to dielectric surfaces

in contact with the plasma must sum to zero over the RF cycle in the steady state,
T

 (  (t ) −  (t ) ) dt = 0
0
e i
(1)

where, T is the RF period. Γe(t) and Γi(t) are the time dependent fluxes to the surface of electrons

and positive ions, acknowledging that the flux of negative ions is negligible in the absence of

pulsing. The electric field in the sheath and presheath usually point towards surfaces to accelerate

ions out of the plasma and confine electrons. It is only during a small portion of the anodic part of

the RF cycle that the sheath voltage decreases sufficiently to enable electrons to reach the substrate.

These electrons arrive at the substrate with largely thermal, isotropic velocity distributions.

Voltage waveform tailoring (VWT) is technique that is able to generate directional (anisotropic),

high energy electron fluxes onto the substrate through promoting an electric field reversal (EFR)

in the presheath above the substrate [8,9]. EFR refers to the electric field in the presheath pointing

into the plasma, as opposed to pointing towards the surface. In VWT, a non-sinusoidal voltage is

applied to the substrate, typically using several harmonics of a fundamental frequency.[8] EFR in

the presheath ultimately results from the requirement that the time average of positive and negative

fluxes to the substrate must balance. Only during the sheath collapse during the anodic part of the

cycle are electrons able to diffuse to the surface. Under certain conditions, this diffusive electron

transport may not be sufficient to satisfy the local charge balance. This lack of ability for diffusive

electrons to reach the surface in sufficient numbers may be due to short sheath collapse times,

thick sheaths, or magnetically or collisionally hindered transport. For these conditions, negative

space charge in the presheath produces an EFR, which slows the transport of ions and speeds the

transport of electrons towards the wafer. Electric field reversal has been observed computationally

82
as well as experimentally.[8–11]

With its ability to produce anisotropic energy and angular distributions (EADs) of electrons

onto the substrate which can penetrate deeply into HAR features, EFR produced by voltage

waveform tailoring has been proposed as a (partial) remedy for charging inside HAR features.[8]

In principle, if the flux of positive ions and negative electrons that penetrate deeply into features

can be balanced, the detrimental effects of intra-feature charging can be minimized. Since the

production of EFR requires manipulating the sheath potential during the RF cycle, the ion EADs

are also affected, and so independent control of the EADs of both electrons and ions is a challenge.

The dynamics of VWT, and EFR in particular, are sensitive to the geometry of the plasma

reactor and the electronegativity of the plasma. In this chapter, we discuss results from a

computational investigation of electric field reversal in electronegative CCPs sustained in Ar/O 2

mixtures in a plasma reactor similar to those used in HVM. In these simulations, the power

deposition is held constant as frequency content is varied to acknowledge that in HVM, power

(and not voltage) is usually what is controlled in HAR etching. The consequences of EFR on the

EADs of ions and electrons onto the wafer surface are discussed. We found that the consequences

of VWT are not limited to electron dynamics but also affect the EADs of ions mainly through the

electrical asymmetry effect and the resulting change in the DC self-bias. The observed electric

field reversal is almost entirely due to the inertia limited lack of electron conduction current and

the decrease in sheath collapse time.

The computational platform used in this investigation, the Hybrid Plasma Equipment

Model (HPEM) and has been previously discussed in Chapter 2. The outcome of the investigation

for the Ar-only base case is discussed in Section 3.2 with a focus on the influence of the phase

angle φ as of the harmonic components of the waveform. In Section 3.3 the influence on EFR of

83
added oxygen, an electronegative species, to the gas mixture is discussed. Concluding remarks are

in Section 3.4.

The reaction mechanism used for pure Ar plasmas is the same as that described in Ref.

[12]. The mechanism for Ar/O2 plasmas is a subset of that described in Ref. [13] which involves

only Ar and O containing species. The mechanism for reactions between only O, O2 and its excited

and ion species is described in Ref. [14]. To decrease computational cost, excited states for O and

O2 were lumped into a single state.

A schematic of the cylindrically symmetric CCP reactor used in this investigation is shown

in Figure 3.1a).

Figure 3.1: Geometry and circuit used in the model. a) Schematic of the dual frequency capacitively coupled plasma
reactor used in this investigation. b) Equivalent circuit diagram of the reactor. Source power at 80 MHz is applied to
the top electrode. The 1 MHz tailored waveform is applied to the bottom electrode.

The reactor, modeled after multi-frequency CCPs used in in industrial etch applications,

consists of two parallel plate electrodes with a diameter of 300 mm separated by a 2.8 cm gap. The

84
top electrode also serves as a showerhead gas inlet. A 0.7 mm thick silicon wafer is mounted on

the bottom electrode. A focus ring made of silicon and quartz to improve discharge uniformity

surrounds the wafer. The dielectric constants of these components are ε/ε0 = 4 for the outer quartz

ring and ε/ε0 = 11.8 for the Si ring. While the conductivity of the quartz is negligible, that of the Si

wafer is 0.05/Ω-cm. The feedstock gas is a mixture of Ar and O2 flowing at 500 sccm through the

top electrode, with oxygen fractions of 0 to 50%. The reactor pressure is held constant at 40 mTorr

at the location of the pressure sensor near the pump port. This is accomplished by throttling the

rate of pumping. Secondary electron emission due to ion bombardment of surfaces is included with

a coefficient of 0.15.

The wafer is naturally included in the self-consistent electric potential and current

calculations by specifying its permittivity and conductivity, whose values are adjusted during

execution of the model to accurately represent the actual thickness of the wafer. The conductivity

of the wafer is large enough that the voltage drop across the wafer is only a few volts, and does

not meaningfully affect the DC bias.

A general electric circuit diagram is shown in Figure 3.1b. The VWT power supply is

connected to the bottom electrode through a blocking capacitor CB of 500 nF. In principle, in the

quasi-steady state, the DC bias should be independent of the value of the blocking capacitor

provided that the RC time constant of the plasma-capacitor series impedance is large compared to

transients in current. The capacitance used here is a balance of there being an acceptably short

charging time and long enough RC time constant so that there is little variation in the DC bias

during the RF cycle. It may appear in the reactor schematic that the high frequency electrode is

touching a grounded metal. Computationally, a zero-conductivity dielectric lies between the two

metal materials. As a result, only displacement current flows between the metals that are separated

85
by what appears to be a perfect capacitor. This is analogous to a dark-space shield. Averaged over

a radio frequency cycle, there is no net power transfer nor net current passing between the two

metals.

To maintain a relatively constant plasma density, 100 W of power is coupled into the

discharge through the top electrode using a sinusoidal voltage Vhf with a frequency of 80 MHz. To

control the dynamics of charged particles impinging on the wafer surface, a customized voltage

waveform, V(t), is applied to the bottom electrode. The waveform is described by equation (1.56)

with frequency f0 = 1 MHz and N = 4 consecutive higher harmonics. The power coupled through

the bottom electrode was held constant at 1 kW by adjusting V0.

3.2 The Ar Base Case

Vhf and V0 are shown in Figure 3.2a as a function of the phase angle φ. The voltage

waveforms for these cases are shown in Figure 3.3d for phase shifts of φ = 0°, 45°, 90°, 135° and

180°. The high frequency power is constant at 100 W. Since the phase shift only applies to the

tailored waveform on the bottom electrode and the plasma density remains fairly constant, Vhf also

remains relatively constant for all φ, varying by less than 10%. Even with the low frequency power

being constant at 1 kW, due to the changes to the electron dynamics in the sheath region, there is

some variation in V0, increasing from 1296 V to 1650 V or about 27%.

The DC self-bias as a function of the phase angle φ is shown in Figure 3.2b for the argon

base case conditions. The decrease in the magnitude of the DC self-biases (less negative) with

increasing phase angle φ can be directly attributed to the EAE. At φ = 0° with Vmax  Vmin (see

Figure 3.3d) the electrical asymmetry of the applied voltage is at its maximum, and the contribution

of the EAE to the DC bias is most negative. The asymmetry is zero when φ = 90° with Vmax = Vmin

(Figure 3.3d) where the contribution of the EAE to the dc bias is also zero. The asymmetry is at

86
its minimum at φ = 180° with Vmax  Vmin (see Figure 3.3d), at which point the contribution of the

EAE to the DC bias is most positive.

The axial components of the electric field between the two electrodes, Ez(z, t), radially

averaged across the wafer, are shown in Figure 3.3a for phase shifts of φ = 0°, 45°, 90°, 135° and

180°. The electric fields are plotted as a function of time during a single 1 MHz cycle. The dotted

line in the images is the Ez(z, t) = 0 contour. The respective voltage waveforms are shown in Figure

3.3d.

Figure 3.2: Plasma properties as a function of phase angle for the pure argon plasma with a constant high frequency
power of 100 W and constant bias power of 1 kW. a) Applied voltage amplitudes for the high frequency source (V hf)
and amplitude of the low frequency bias (V0). Top and bottom electrode. b) DC self-bias.

87
Figure 3.3: Plasma properties for the pure argon plasma averaged across the wafer as a function of height above the
wafer and time during the low frequency,1 MHz cycle. Values are shown (left-to-right) for phase angles of φ = 0,
45, 90, 135, 180°. a) Axial component of the electric field, b) Ar+ density, c) electron density and d) applied voltage
waveform on the substrate. The Ar+ and electron densities are plotted on a log-scale spanning 2 decades with the
maximum value noted in each image.

88
 In plasmas which do not have EFR, Ez(z, t) in the sheath at the bottom electrode will point

towards the electrode at all times, which corresponds to a negative value of E z in these images.

However, even with pure argon, there can be an EFR at the peak of the positive excursion of the

applied voltage waveform. With φ = 0° the EFR occurs during the sheath collapse at around 0.5

μs when the applied voltage is at its maximum positive value and a positive electric field

component (pointing in +z direction) is present. This EFR results, in part, from the short duration

of the positive voltage excursion of the waveform. During more than 80% of the RF cycle, the

applied voltage is negative, with there being less than 20% of the RF cycle being amenable to

electron collection. Given that the electrons are mostly thermal, diffusion alone would not provide

sufficient current to the wafer to balance the ion current during this short period. An additional

drift component is required, which for electrons requires a net electric field pointing upwards

(positively), away from the electrode. The reversal in the electric field provides additional electron

acceleration to overcome the confining potential and inertial effects to enable the charge and

current balance conditions. The generation of the EFR is a natural outcome of the calculation due

to solution of Poisson’s equation and the charge transport equations.

The densities of Ar+ and electrons as a function of height averaged across the wafer are

shown in the Figure 3.3b and Figure 3.3c as a function of time and for different values of φ.

Maximum ion densities increase from 2.8 1010 cm-3 for φ = 0° to 6.9 1010 cm-3 for φ = 180°.

Recall that the power deposition at both frequencies is being held constant. With a decrease in the

magnitude of Vdc (less negative) with increasing φ (see Figure 3.2), less fractional power is

dissipated by ion acceleration through the sheath. The response of the system is to increase V0 and

Vhf (the latter being a small increase) to increase plasma density to maintain a constant power. In

spite of its large mass, there is a small response of the Ar+ density profile to the sheath dynamics,

89
with ions being accelerated into the lower sheath during the local maxima of the cathodic portion

of the low frequency RF cycle. Ions are accelerated into the upper sheath during the local maxima

of the anodic cycle. These responses are afforded, in part, by the low fundamental frequency of 1

MHz. Such responses would not be expected, for example, at 13.56 MHz. In contrast, the spatial

distribution of electron density in the sheaths is strongly modulated during both the RF cycle and

with different values of φ. Electrons are excluded from the lower sheath during the local maxima

of the cathodic portion of the low frequency RF cycle and excluded from the upper sheath during

local maxima of the anodic portion of the cycle.

A phase angle of φ = 0° produces the “peak” waveform (Figure 3.3d, left-most sub-image)

which approximates a delta function for higher harmonic orders (N→ꝏ). As such, it creates the

shortest period of sheath collapse having the largest time derivate in electric field (dE z/dt), both of

which favor EFR. With larger phase angles, this abrupt sheath collapse becomes more gradual and

longer, providing electrons with more time to diffuse to the surface, which reduces the need for

EFR to balance current. With φ = 180°, the voltage waveform resembles the “valley” waveform

(Figure 3.3d, right-most sub-image) which results in a plasma sheath that is at its minimum voltage

and width most of the time – the opposite of the φ = 0 waveform. This change in φ from 0 to 180

produces a decrease of the EFR, with its elimination for φ = 180. A portion of this reduction (and

eventual elimination) of the EFR with increasing phase angle is due to the decrease of the

magnitude of the DC self-bias (less negative) which reduces electron repulsion from the sheath.

Although the instantaneous electron diffusion flux from the bulk plasma may appear to be

sufficient to balance current, that flux is actually limited by the ambipolar-like electric fields in the

presheath and by the finite time required for electrons to cross the sheath. The secondary electrons

emitted from the surface generate a net positive current into the wafer when the sheath expands,

90
increasing the requirements to balance current by the bulk electrons during sheath collapse.

The EADs of electrons incident onto the wafer for the pure argon plasma for different phase

angles φ are shown in Figure 3.4a. The mean electron energy and angle of incidence are shown in

Figure 3.4b as a function of phase angle.

Figure 3.4: Properties of electrons incident onto the wafer for the pure argon plasma. a) Electron energy and angular
distributions for different phase angles plotted on a log-scale spanning 2 decades. b) Mean electron energy and
mean electron angle of incidence as a function of phase angles.

91
 The trends in the maximum strength of the EFR translate to the EADs. The electron

temperature in the bulk plasma is 3.1- 3.3 eV, which is approximately the temperature of the

electrons incident onto the wafer for φ = 180° which has the minimum EFR. This similarity,

including the broad angular distribution, indicates that the electrons are simply diffusing into the

sheath with there being little drift component. With decreasing φ, conditions which increase the

EFR, there is an increase in directional electron acceleration onto the wafer. With decreasing phase

angle φ, the EADs increase in energy while becoming more narrow in angle. As the phase angle

decreases from φ = 180° to 0°, the mean energy of electrons incident onto the wafer increases from

7.1 eV to 17.3 eV, while the mean angle of incidence decreases from 26.8° to 15.1°. These results

suggest that high fluxes of directional electrons onto the substrate can be generated by EFR.

3.3 EFR with Ar/O2 Gas Mixtures

The addition of oxygen to the argon gas mixture enables the generation of negative ions

and the transition of the electropositive Ar discharge to an electronegative plasma. The densities

of Ar+, O2+, O- and electrons, as well as the discharge electronegativity are shown in Figure 3.5 as

a function of the O2 mole fraction in the inlet flow for otherwise the base-case conditions with φ =

0°. Electronegativity in this case is the ratio of densities of [O-]/[e].

Through dominantly dissociative electron attachment of O2 to form O-, the electron density

decreases with increasing O2 inflow while that of O- increases. At 50% O2, the total negative charge

in the plasma bulk is nearly equally due to O- and electrons. Total positive ion density decreases

from 2  1010 cm-3, for the pure Ar discharge to 1.4  1010 cm-3 at 50% O2, a consequence of more

power being dissipated in non-ionizing collisions with the increase in O2 fraction. The total

discharge electronegativity increases from to 1.05 at 50% O2.

92
 Due to the change in electron density and the subsequent changes in power deposition, the

DC self-bias is a sensitive function of the Ar/O2 ratio. Vdc as a function of the voltage phase angle

φ for Ar/O2 ratios of 100/0, 90/10, 75/25 and 50/50 is shown in Figure 3.6a. The general trend of

decreasing magnitude of Vdc (less negative) with increasing phase angle φ occurs for all gas

mixtures. This trend is a result of the EAE.

Figure 3.5: Spatially averaged densities of Ar+, e-, O2+ and O-, and electronegativity of Ar/O2 plasmas as a function
of the O2 mole fraction with a phase angle of φ = 0°.

A second major trend is an increase of the magnitude of Vdc (more negative) with increasing

O2 mole fraction. Due to the electronegativity of O2 containing plasmas and dissipation of power

in non-ionizing collisions with O2, increasing the mole fraction of O2 requires higher applied

voltages to maintain a constant power deposition. The total voltage amplitude V0 as a function of

the phase angle 𝜑 for different Ar/O2 ratios is shown in Figure 3.6b. At φ = 0°, V0 increases by

approximately 18% from 1300 V with Ar/O2 = 100/0 to 1545 V at Ar/O2 = 50/50. For a given

electrical or geometrical asymmetry Vdc is proportional to V0, and so on this basis alone, the

magnitude of the Vdc should increase (become more negative) with increasing O2 mole fraction.

The axial component of the electric field between the two electrodes, Ez(z, t), radially

93
averaged across the wafer is shown in Figure 3.7. Each column represents a constant phase angle

with φ = 0°, 45°, 90°, 135° and 180°. Each row in turn represents cases at constant gas ratio and

varying phase angle with for Ar/O2 = 100/0 (Figure 3.7a), 90/10 (Figure 3.7b), 75/25 (Figure 3.7c)

and 50/50 (Figure 3.7d). The voltage waveforms are shown in the bottom row (Figure 3.7e). The

dotted line in the images is the Ez(z, t) = 0 contour.

Figure 3.6: Plasma properties as a function of phase shift φ for different Ar/O2 ratios when keeping source and bias
powers constant. a) DC self-bias produced on the powered substrate. b) Applied voltage amplitude V0.

94
Figure 3.7: Radial average of the vertical component of the electric field as a function of time during a single 1 MHz
cycle and distance from bottom electrode different gas mixtures. Ar/O2 = a) 100/0, b) 90/10, c) 75/25, d) 50/50. e)
Normalized applied voltage V0. Results are shown for phase shifts of (left-to-right columns) φ = 0, 45, 90, 135 and
180°. The dotted line is for E = 0, the separation between negative electric fields pointing down and positive
pointing up.

95
 The magnitude of the EFR increases with increasing oxygen mole fraction for all phase

angles φ. The EFR has a peak magnitude of 47 V/cm (E/N = 3650 Td, 1 Td = 10 -17 V-cm2) in the

absence of oxygen for φ =0, and a peak value of 94 V/cm (7300 Td) with Ar/O2 = 50/50 while the

EFR extends deeper into the plasma. In pure argon, the sheath thickness during the cathodic phase

is 9.5 mm and the extent of the EFR from the substrate during the anodic phase is 6 mm. For Ar/O2

= 50/50, the sheath thickness is 15 mm and the extent of the EFR is 12 mm. Due to the lower

applied sinusoidal voltage no electric field reversal occurs at the top electrode. In general, EFR

occurs when electron transport to the surface has a limited time window during the anodic portion

of the RF cycle or is impeded by other means. Since mobility µi is inversely proportional to mass

and the generation of negative O- ions produces a decrease in the electron density, the average

mobility of negative charge carriers decreases. This decrease in mobility then requires a larger

extracting electric field in the form of the EFR. With increase in O2 fraction from zero to 50%, the

electron density decreases to about 40% of that in pure argon. With the sum of the positive ion

fluxes decreasing to 80% of their values with pure argon, the EFR increases by a factor of 2 (47

V/cm to 94 V/cm) to extract the needed electron current.

As with the pure argon base cases, the magnitude of the EFR decreases with increasing

phase angle for all Ar/O2 gas mixtures. The temporal modulation of the zero electric field contour

in the bulk plasma also varies with phase angle. This modulation indicates that the effects of the

VWT are not limited to the sheath and could be responsible for small modulation that occurs in

electron temperature. To some degree, this modulation is required by current continuity.

Displacement current through the sheath is proportional to dE/dt – current which must be matched

largely by conduction current in the bulk plasma. As dE/dt in the sheath varies with phase angle,

there must be variation in conduction current in the bulk plasma which in turn requires modulation

96
in the bulk electric field. The modulation of the E=0 contour is more pronounced at higher Ar/O 2

ratios, indicating that the changes to the sheath potential structure (leading to more displacement

current) which increase with the O2 content extend beyond the immediate sheath region to

modulate conduction current in the bulk plasma.

The time dependent densities of Ar+, O- and electrons are shown in Figure 3.8 for φ = 0°

and Ar/O2 = 50/50. Due to their low mobility and their being trapped in the positive plasma

potential, the O- ions (Figure 3.8b) do not follow the modulated sheath and are confined to the bulk

plasma at all times. There is some perturbation of the O- density at the peak of the cathodic cycle

at the top electrode, however the negative ions do not significantly contribute to the transport of

negative charge to the surface. In contrast to the stationary O- density profiles, the electron density

profile, shown in Figure 3.8c, responds quickly to the applied potential. The electrons are confined

to the plasma bulk during most of the RF period and only extend to the surface at the time of sheath

collapse at 0.5 µs. As a result of this discrepancy in dynamic behavior between electrons and O-,

increasing the mole fraction of O- in the system effectively reduces the transport of negative

charges to the surface, thereby necessitating a larger EFR during the portion of the RF cycle during

which there is sheath collapse.

With an increase of the magnitude of the DC self-bias Vdc and the applied voltage V0, the

sheath potential and sheath width increase. The maximum width of the sheath increases from

approximately 9 mm for pure argon to 15 mm for Ar/O2 = 50/50. This increase in sheath width

increases collisionality (reducing electron mobility) and increases the distance electrons need to

traverse during the sheath collapse in order to reach the surface. Both effects limit electron

transport and increase the need for and magnitude of the EFR.

97
Figure 3.8: Plasma properties for Ar/O2 = 50/50 and a phase shift of φ = 0°. a) Radial average of the Ar+ density as a
function of time during a single 1 MHz RF cycle plotted on a log-scale over 2-decades. b) O- density (log-scale, 2-
decades) and c) electron density (log-scale 3-decades). The maximum density is noted in each image. d) e)
Normalized voltage amplitude V0

98
 These two mechanisms (the reduction in the density of mobile charge carriers by the

increasing negative ion density and the confinement of electrons by the applied tailored voltage

waveform) at least partially add, thereby increasing the magnitude and extent of spatial penetration

of the EFR. Comparing the electric field configurations over the range of phase angles reveals a

strong dependence of the EFR on gas mixture. With Ar/O2 = 50/50, the magnitude of the EFR

decreases with increasing phase angle – decreasing from 94 V/cm at φ = 0° to 11 V/cm at φ =

180°, a total change of 83 V/cm. For the same phase angle shift for the pure argon plasma, the

change in magnitude of the EFR is only 11 V/cm.

The interaction between the two mechanisms producing EFR also results in enhanced

electron acceleration towards the wafer. The electron EADs incident onto the wafer are shown in

Figure 3.9 for φ = 0° (maximum EFR) and for φ = 180° (minimum EFR) for different Ar/O2

feedstock mixtures. With φ = 180°, the EFR is nominal, resulting in the electron EADs being

largely thermal with average energies < 10 eV for bulk electron temperatures of 3.2 – 3.5 eV.

There is a small increase in incident electron energy and narrowing of angular distribution for

Ar/O2 = 50/50 where the magnitude of the EFR begins to increase. With φ = 180° when most of

the negative charge is carried by heavy O- ions, electron transport is still sufficient to balance

positive ion transport due to the sheath being collapsed during most of the RF cycle. There is little

additional electron acceleration by an EFR necessary to balance the positive ion flux.

With φ = 0°, the EFR is well established with peak magnitudes reaching nearly 100 V/cm

for Ar/O2 = 50/50. The maximum electron energy incident onto the wafer extends from 30 eV for

pure argon to 100 eV for Ar/O2 = 50/50. The angular distribution narrows from ±45 degrees for

pure argon to a ±20 degree spread for Ar/O2 = 50/50. The wide range of electron energy, while the

minimum energy is nearly constant for oxygen containing mixtures, suggests that the electron

99
attracting sheath is the equivalent of the thin-sheath-limit for accelerating ions into the electrode.

That is, the electron transit time across the sheath is short compared to the RF period, and so

electrons are accelerated to the instantaneous sheath potential. The lower energy component of the

EAD is likely sustained by continuing thermalization of electrons through collisions inside the

sheath.

Figure 3.9: Electron energy and angular distributions incident onto the wafer surface for different Ar/O 2 ratios
plotted on a log-scale over 2 decades with phase shifts of a) φ = 0° and b) φ = 180°. The EAD increases in energy
and narrows in angle with increasing O2 fraction, an effect that is most prominent for a) φ = 0°.

100
 Figure 3.10: Properties of the electron velocity distribution incident onto the wafer for different Ar/O2 ratios as a
function of phase shifts φ. a) Mean electron energy and b) mean angle of incidence. Decreasing phase angle
increases the incident electron energy and narrows the angular distribution, with the effect being most prominent for
large O2 mole fractions.

The mean electron energies at the wafer surface as a function of the phase angle,  ( ) , for

different Ar/O2 mixtures are shown in Figure 3.10a. As with the EADs, the trends correlate with

the magnitude and depth into the plasma of the EFR. Mean energies for φ = 180° have a nominal

dependence on gas mixture, extending from 6.2 eV for pure argon to 9.7 eV for Ar/O2 = 50/50.

Although this is a change of a factor of 1.5, the absolute change is only 3.5 eV. Mean energies for

101
φ = 0° extend from 16.7 eV for pure argon to 57.7 eV for Ar/O 2 = 50/50, an increase by a factor

of 3.4 and absolute change of 41 eV. The same mechanisms producing a decrease in mean energy

produces a decrease in the mean angle with which electrons strike the waver, as shown in Figure

3.10b as a function of the phase angle and gas mixture. For a given Ar/O2 mixture, the mean angle

increases with increasing φ. As with the mean energy, these changes are most pronounced at the

highest oxygen content. For Ar/O2 = 50/50 the mean angle increases from 6.4° at φ = 0° to 20.3°

at φ = 180°, a 3-fold increase.

3.4 Concluding Remarks

The use of voltage waveform tailoring (VWT) to generate and control electric field

reversals (EFRs) during the sheath collapse in low pressure capacitively coupled RF plasmas

sustained in Ar/O2 mixtures was computationally investigated. The purpose of this investigation

is to assess the potential of this technique to accelerate electrons into sub-micron HAR etch

features to compensate positive surface charging inside these structures. The effects of the Ar/O2

feed gas ratio as well as voltage waveforms on the electron energy-angular distributions (EAD)

incident onto the wafer surface were investigated while keeping the power deposition constant, as

would be the practice in industry. To maintain the plasma density and reduce its dependence on

the voltage waveform, a high frequency RF voltage was applied to the top electrode. The substrate

bias waveform consisted of a sinusoidal signal and its first 4 harmonics with the even harmonics

having a variable phase shift φ. The strength of the EFR increases with the O2 mole fraction and

with a decrease in φ. The correlation with the O2 mole fraction is a consequence of the

electronegative nature of the oxygen. The increase in O2 mode fraction reduces the electron density

while increasing negative ion density, which results in a decrease in the average negative charge

carrier mobility. To maintain constant power, a larger applied voltage is required.

102
 The consequences of the phase angle on EFR can be attributed to two major effects: i) the

electrical asymmetry effect is maximum at φ = 0° (“peak” waveform), leading to a large DC self-

bias (most negative) and thick sheath and ii) the time that the sheath collapses is minimized at φ =

0° and maximized at φ = 180°. Both of these effects enhance the EFR. Since the ion flux to the

substrate must be balanced by electrons on a time average while electron transport to the substrate

by diffusion is insufficient to do so in the presence of these transport limiting factors.

For a fundamental driving frequency of 1 MHz and a power of 1 kW applied to substrate,

the bias waveform enables significant control of the electron energy and angle distributions onto

the wafer surface. The relative and absolute changes in mean electron energy were strong functions

of the Ar/O2 feed gas ratio and most pronounced in the Ar/O2 =50/50 mixture, while being

significant in all configurations that were investigated. The fact that electronegative gases increase

the magnitude of the EFR, and its ability to accelerate electrons into the wafer, suggests that

process gas compositions used in semiconductor plasma etching processes, which often contain

strongly electronegative species, can potentially profit from this mechanism as well.

The details of plasma etching processes critically depend on the ion energy and angular

distributions onto the wafer. Although VWT is able to control the electron EADs onto the wafer,

this cannot be done independently of also changing the ion EADs. Changes in the VWT

configuration change the DC bias which translates into changing the average ion energy onto the

wafer. The shape of the ion EAD will depend on the number of harmonics and the phase angle.

Nevertheless, some of the scaling of the EAD described in this work could allow for favorable

synergies with current trends in semiconductor manufacturing. Higher applied voltages, as used in

plasma etching of HAR features, combined with a voltage waveform resembling the peak-

waveform, will produce high energy, narrow angle fluxes of electrons into the wafer surfaces.

103
These fluxes have the properties required to address intra-feature charging.

To evaluate the efficacy of VWT in the context of actual plasma etching processes, it is

important to investigate relevant process conditions using halogen containing gas mixtures

combined with feature scale simulations that include the effects of the electron and ion EADs on

charge accumulation, feature deformation and depth limited etch rates. There are still open

questions with respect to the feasibility of plasmas sustained by complex waveforms at very high

powers and the technical difficulties related to power matching, filtering, phase locking and

metrology as well as economic considerations.

Due to its potential versatility, future implementations of VWT may be not limited to

conventional CCPs but could find application in different contexts such as in inductively coupled

plasmas. These applications could be for similar surface charge remediation on biased electrodes

or the prevention of material erosion near the antennas caused by high energy ions in E-mode

operation during pulsed operation. A broader set of waveform configurations should be

investigated as the physical mechanisms discussed here are, in principle, not dependent on any

specific waveform. These investigations may include the use of non-consecutive harmonics,

pulsed higher harmonics, transient phase shifts or entirely non-harmonic waveforms.

104
3.5 References

[1] V. M. Donnelly and A. Kornblit, Journal of Vacuum Science & Technology A: Vacuum,

Surfaces, and Films 31, 050825 (2013).

[2] H. Tanaka, M. Kido, K. Yahashi, M. Oomura, R. Katsumata, M. Kito, Y. Fukuzumi, M.

Sato, Y. Nagata, Y. Matsuoka, Y. Iwata, H. Aochi and A. Nitayama, Digest of Technical

Papers - Symposium on VLSI Technology 14 (2007).

[3] Y. Li, IEEE Solid-State Circuits Magazine 12, 56 (2020).

[4] B. Wu, A. Kumar and S. Pamarthy, Journal of Applied Physics 108, 51101 (2010).

[5] S. Rauf and A. Balakrishna, Journal of Vacuum Science & Technology A: Vacuum,

Surfaces, and Films 35, 021308 (2017).

[6] G. Le Dain, A. Rhallabi, C. Cardinaud, A. Girard, M.-C. Fernandez, M. Boufnichel and F.

Roqueta, Journal of Vacuum Science & Technology A 36, 03E109 (2018).

[7] J. C. Wang, W. Tian, S. Rauf, S. Sadighi, J. Kenney, P. Stout, V. S. Vidyarthi, J. Guo, T.

Zhou, K. Delfin, N. Lundy, S. C. Pandey, S. Guo and G. S. Sandhu, Plasma Sources Sci

Technol 27, 94003 (2018).

[8] P. Hartmann, L. Wang, K. Nösges, B. Berger, S. Wilczek, R. P. Brinkmann, T.

Mussenbrock, Z. Juhasz, Z. Donkó, A. Derzsi, E. Lee and J. Schulze, J Phys D Appl Phys

54, 255202 (2021).

[9] U. Czarnetzki, D. Luggenhölscher and H. F. Döbele, Plasma Sources Sci Technol 8, 230

(1999).

[10] Y. X. Liu, Q. Z. Zhang, J. Liu, Y. H. Song, A. Bogaerts and Y. N. Wang, Appl Phys Lett

101, 114101 (2012).

105
[11] S. Sharma and M. M. Turner, J Phys D Appl Phys 47, 285201 (2014).

[12] P. Tian and M. J. Kushner, Plasma Sources Sci Technol 24, 34017 (2015).

[13] S. Huang, C. Huard, S. Shim, S. K. Nam, I.-C. Song, S. Lu and M. J. Kushner, Journal of

Vacuum Science & Technology A 37, 031304 (2019).

[14] D. S. Stafford and M. J. Kushner, J Appl Phys 96, (2004).

106
 Chapter 4 : Voltage Waveform Tailoring for High Aspect Ratio Plasma Etching of SiO2
using Ar/CF4/O2 Mixtures: Consequences of Ion and Electron Distributions on Etch
Profiles

4.1 Introduction

In this Chapter we discuss results from a computational investigation of a multi-frequency

CCP sustained in an Ar/CF4/O2 mixture using a geometry similar to those found in HVM. High

frequency power applied to the top electrode is used to sustain the plasma while VWT is used to

define the low-frequency bias voltage applied to the bottom electrode. The consequences of the

electrical asymmetry effect and the electric field reversal on the IEADs and EEADs onto the wafer

surface are discussed. The resulting charged and neutral particle fluxes and distributions are then

used in self-consistent simulations of etching a HAR feature into SiO2 to investigate the ability of

VWT to control feature properties.

We found that in spite of the complicating factors of geometry, gas mixture and multiple

frequencies, VWT allowed for effective control of IEADs and EEADs onto the wafer, though this

control could not be produced independently. Optimizing the waveform to produce a desired

EEAD inevitably affects the IEAD. Peak and mean energies of positive ions and of electrons onto

the wafer are strong functions of the shape of the applied voltage waveform. With the narrow angle

and energetic EEADs produced by VWT, our investigation did show a decrease in intra-feature

charging, as measured by a decrease in the positive electrical potential. However, the overall

characteristics of the final etch feature were dominated by the ion dynamics.

Synthetic changes to only the EEADs resulted in minor changes of the final etched feature.

107
For example, there were not significant differences in feature quality between an isotropic, thermal

EEAD onto the wafer corresponding to an electron temperature of a 3-4 eV, and the narrow,

energetic EEAD produced by VWT. We attribute this result to the fact that thermal electrons are

accelerated into the feature by the electric fields produced by the positive, in-feature electric

potential. Ultimately, in a quasi-steady state, the flux of positive and negative charge into the

feature must be equal. This is achieved by producing a positive potential in the feature, which then

accelerates thermal electrons into the feature. The narrow energetic EEADs produced by VWT are

also accelerated into the feature. However, their initially higher wafer directed velocities require a

smaller positive potential in the feature to achieve the needed flux balance. The end result is a

decrease in the intra-feature potential relative to that produced with thermal electrons that is

approximately equal to the average energy of the VWT produced EEAD.

The computational platform used for the reactor scale plasma transport, the Hybrid Plasma

Equipment Model (HPEM) and the feature scale model, the Monte Carlo Feature Profile Model

are described in Chapter 2. The reactor scale setup is described in Section 4.2 and the feature scale

setup is described in Section 4.3. The influence of phase angle φ of the voltage waveform on

charged particle distributions onto the wafer and etch profiles with low bias power is discussed in

Section 4.4 and with high bias power is discussed in Section 4.5. Concluding remarks are in

Section 0.

4.2 Description of the Reactor Scale Model

A schematic of the cylindrically symmetric CCP reactor used in this investigation is shown

in Figure 4.1. The reactor, modeled after multi-frequency CCPs used in industrial HVM

applications, consists of two parallel plate electrodes with a diameter of 300 mm separated by a

2.8 cm gap. A silicon wafer is mounted on the bottom electrode and a showerhead gas inlet is

108
distributed across the top electrode. A focus ring made of silicon and quartz to improve discharge

uniformity surrounds the wafer. The dielectric constants of these components are ε/ε0 = 4 for the

outer quartz ring and ε/ε0 = 11.8 for the Si ring. While the conductivity of the quartz is negligible,

that of the Si wafer is 0.05/Ω-cm. The feedstock gas Ar/CF4/O2 = 75,/15/10 flows at 500 sccm

through the top electrode. The reactor pressure is held constant at 40 mTorr at the location of the

pressure sensor near the pump port by throttling the rate of pumping.

Figure 4.1: Schematics of the reactor. a) The two frequency CCP has the VWT power applied to the lower electrode
and a sinusoidal, 80 MHz waveform applied to the top. b) Electrical schematic of the system. From F. Kruger et al,
Plasma Sources Sci. Technol. 30, 085002 (2021). Reproduced by permission of IOP Publishing Ltd.

The VWT power supply is connected to the bottom electrode through a blocking capacitor

of 500 nF. In principle, in the quasi-steady state, the DC bias should be independent of the value

of the blocking capacitor provided that the RC time constant of the plasma-capacitor series

impedance is large compared to transients in current. The capacitance used here is a balance of

there being an acceptably short charging time and long enough RC time constant so that there is

109
little variation in the DC bias during the RF cycle.

Sheath properties are in part a function of plasma density at the edge of the sheath which

then determines sheath thickness. To maintain a relatively constant plasma density and to minimize

sheath variation based on changes in plasma density, 500 W is coupled into the plasma through

the top electrode using a sinusoidal voltage waveform with a frequency of 80 MHz. This power

was maintained by continually adjusting the amplitude of the voltage at 80 MHz.

To control the dynamics of charged particles impinging on the wafer surface, a customized

voltage waveform, V(t), was applied to the bottom electrode. The waveform is described by

equation (1.56) with frequency f0 = 1 MHz and N = 4 consecutive higher harmonics,

Here, V0 is the total applied voltage amplitude, k is the harmonic order, N is the total number

of frequencies, and the relative phase shift of each harmonic frequency k is φk. In this work the

number of harmonics is N = 5. The consequences of VWT were investigated by varying the phase

shift of the even harmonics φ for k = 2, 4. The power coupled through the bottom electrode, P0,

was initially held constant at 1 kW by adjusting V0.

4.3 Description of the Feature Scale Model

In this study, we investigated the plasma etching of SiO2 in a Ar/O2/CF4 gas mixture using

the MCFPM which is described in Chapter 2. A detailed description of the reaction mechanism is

given by Huang et al. [1]. Generally, all incoming ions or hot neutrals resulting from neutralization

of ions when striking surfaces can physically sputter every material according to the incident

particles mass, angle and energy as well as the surface binding energy of the material. Radical

CxFy species can lead to the deposition of polymer, the main contributor to sidewall passivation,

which in turn is removed by sputtering and etching by oxygen radicals. CxFy radicals can react

with the SiO2 surface to create a layer of SiO2CxFy complexes, significantly lowering the binding

110
energy and making the complex easier to chemically sputter.

The electric potential in the feature was computed by collecting charge on surfaces and

solving Poisson's equation. The relative permittivity for SiO2 used in solution of Poisson's equation

was r = 4.0, and for photoresist, polymer and SiO2CxFy complexes was r = 3.0. The electric

potential then provides electric fields used to advance the trajectories of charged particles. All

positive ions neutralized when striking a surface, and if reflected, continued as a neutral. Electrons

were collected on surfaces with unity probability. For this study, the SiO2 and photoresist had

essentially zero conductivity (10-10/-cm). Polymer (510-6/-cm) and SiO2CxFy complexes

(110-6/-cm) had finite but small conductivity. These conditions generally resulted in charge

being resident only on surfaces or buried under polymer or redeposited etch products if deposition

occurred on top of charged surfaces.

The geometry used for the feature evolution is shown in Figure 4.2. The feature is a trench

consisting of a 3000 nm thick SiO2 layer over a Si stop-layer, covered by 100 nm thick photoresist

with a 100 nm opening. This geometry produces an aspect ratio of about 30 for an etched trench

down to the stop layer. The simulation was performed in 3-dimensions as a finite depth trench.

The mesh consists of 60 cells wide by 750 cells tall by 10 cells deep, producing cubic voxels with

5 nm edge length. Reflective boundary conditions are used along the y-axis normal to the geometry

shown in Figure 4.2. Although we are simulating a nominally 2-dimensional feature (a trench),

performing the simulation in 3-dimensions results in a more realistic representation of sidewall

roughness. When simulating a trench in 2-dimensions, sidewall roughness is effectively uniform

(and infinite) in the unresolved depth. In 3-dimensions, sidewall roughness is finite. The etching

was performed for a fixed process time of 15 minutes.

111
 Figure 4.2: Initial geometry used in for feature profile simulations. A 700 nm mask defines a 100 nm opening to
3000 nm thick SiO2. The simulations are for trench etching, performed in 3-dimensions having a finite depth.

The Monte-Carlo method which is used in the MCFPM inherently produces statistical

variation in the details of the features that are produced. This run-to-tun variation can occur

naturally in parallel computations by differences in the timings of the processors. The statistical

variation features can be demonstrated by changing the seeds for the random number generators

on separate runs. The features shown here are typical, average features in which outliers due to

statistical variation have been excluded. The statistical run-to-run differences in features are

discussed in Appendix A.

112
4.4 Plasma and Etch Properties for 1000 W of VWT Power

In industrial applications of plasma etching, it is most common to control power deposition

in the plasma as opposed to voltage amplitude. To make more direct connection to these

applications, the power applied through the VWT on the bottom electrode was first held constant

at PVWT = 1000 W by adjusting the voltage amplitude V0 while varying the phase φ from 0 – 180o.

Although this power appears large, by industrial processing standards for HAR etching of 30 cm

diameter wafers, 1000 W is a moderately small power [1–3]. The applied voltage amplitudes

necessary to maintain constant power coupling through the top electrode (500 W), bottom

electrode (PVWT = 1000 W) and the resulting DC self-bias on the bottom electrode are shown in

Figure 4.3 as a function of phase angle. At constant power, the amplitude of the high frequency

voltage applied to the top electrode is a weak function of the phase angle and is nearly constant at

Vhf = 120 V. The DC self-bias, however, is strongly coupled to φ with its magnitude decreasing

(becoming more positive) by 88% from Vdc= -2480 at φ=0° to Vdc= -290 at φ=0°.

113
 Figure 4.3: Voltage amplitudes for the VWT bias (Vbias) and top power (Vtop), and DC self-bias as a function of
phase angle. a) PVWT = 1000 W and b) PVWT = 2000 W.

Earlier works have described the trend of how DC bias depends of the shape of the applied

waveform through the generation of the electrical asymmetry effect (EAE) [4,5]. (See Sections

1.17 and 1.18 for details). The formulation for the EAE does not specifically account for a second

powered electrode having a unique frequency, such as used here. However, due to the higher

frequency and lower power of this second RF bias, the voltage amplitude applied to the top

electrode is smaller than that applied to the lower electrode. Sinusoidal voltages alone do not

induce electrical asymmetry effects and so we do not expect large perturbations to the EAE process

by the high frequency source. With the phase shift applying only to the tailored waveform on the

114
bottom electrode, Vhf remains relatively constant for all φ, varying by less than 10%. Due to the

changes to the electron dynamics in the sheath region some variation does occur in V0, decreasing

from 2800 V to 2200 V or about 27% from  = 0 to 180.

The DC self-bias on the lower electrode, Vdc, is an important parameter in determining the

energy of incident positive ions. For high frequency excitation where the energy width of the IEAD

is small compared to the RF voltage amplitude, the centroid in energy of the IEAD typically

increases with the magnitude of Vdc. With low fundamental frequencies (thin sheath limit), ions

transiting the sheath respond to the sheath dynamics during the RF period. (This is nearly always

true for electrons.) In this low frequency limit, the correlation between IEAD and Vdc is not straight

forward. The radially averaged axial electric field as a function of time and distance from the

substrate, Ez(z, t), is shown in Figure 4.4 for phase shifts of φ = 0° to 180°. In CCPs without an

electric field reversal, Ez(z, t) near the bottom electrode sheath will point towards the electrode

(negative z direction) at all times. However, with φ = 0°, an EFR occurs during the sheath collapse

at around 0.5 μs when the applied voltage is at its maximum positive value and there is a positive

electric field component (pointing in +z direction). This EFR results, in part, from the short

duration of the positive voltage excursion of the voltage waveform and the correspondingly short

time during which the sheath collapses. In order to produce the required electron flux to the surface

to maintain a charge-neutral flux to the substrate, an EFR occurs to accelerate electrons towards

the surface. This field reversal is most pronounced in the φ = 0° case (69 V/cm) as this phase has

the shortest sheath collapse and so requires the largest instantaneous electron flux to the surface.

As the duration of the sheath collapse increases with increasing φ and becomes less abrupt, the

magnitude of the instantaneous electron flux that is required to balance charge decreases, and so

the magnitude of the EFR reversal also decreases.

115
Figure 4.4: Vertical electric field component as a function of time and distance from wafer for PVWT = 1000 W for
phase angles of a) 0°, b) 45°, c) 90°, d) 135° and e) 180°. The shape of the voltage waveform applied to the bottom
electrode is shown in each image.

The modulation of the sheath as a function of the phase of the harmonics translates into

significant variation of IEADs incident onto the wafer, as shown in Figure 4.5a. These IEADs are

collected by summing over all ions striking the wafer for φ = 0° to 180°. The IEADs generally

decrease in maximum energy with increasing φ, following the trend in Vdc. The mean energy of

the ions also decreases with increasing φ, as shown in Figure 4.5c. Aside from the general decrease

in energy, the shape of the IEADs reflect the temporal dynamics of the sheath. The IEAD for φ =

116
0° contains a dominant peak at 3000 eV, a result of the sheath maintaining a nearly constant

potential and thickness for the majority of the RF cycle. (See Figure 4.4.) The IEADs for φ = 45°,

90° and 135° contain more complex multi-modal structures with broader spreads in energy. The

structure in the IEADs reflects the length of time the sheath is maintained at a given potential and

thickness. The IEAD for φ = 180 ° has a single dominant peak at approximately 400 eV, as the

sheath has one of two quasi-binary states – fully collapsed or fully expanded.

The EEADs incident onto the wafer surface are shown in Figure 4.5b for φ = 0° to 180°.

The electric field reversal for small values of φ produces a significant acceleration of electrons

normal to the surface, generating an EEAD with a higher energy and narrower angular distribution

that might occur with low frequency, sinusoidal waveforms. (Recall that for a purely thermal

distribution of electrons onto the surface, the mean energy will be about the electron temperature,

3-4 eV, and the angular spread would span -90 to +90.) The EEAD for φ = 0°, where the EFR

has is largest magnitude, extends to 125 eV, considerably higher than the thermal bulk temperature

of about 3.2 eV. (Although the electron energy distribution is non-Maxwellian in the bulk plasma,

for convenience we will refer to the electron flux incident onto the electrode in the absence of

significant EFR as being the thermal electron flux or resulting from thermal electrons.) The

downward trend of the magnitude of the electric field reversal with increasing φ in turn leads to a

decrease in the electron acceleration normal to the surface, and so to an overall reduction in

electron energy incident onto the surface. With φ = 180°, the EEADs incident onto the surface

closely resemble a thermal distribution, though with a somewhat narrower angular distribution.

The high energy, low angle electrons are likely the result of electrons accelerated in the sheath at

the opposing electrode. Due to the two-decade log scale of the image, these structures are visually

overemphasized.

117
Figure 4.5: Properties of ions and electrons incident onto the wafer for PVWT = 1000 W. a) IEAD and b) EEAD for
phase angles of φ = 0°, 45°, 90°, 135° and 180°. The distributions are separately normalized and plotted on a log-
scale over 2 decades. c) Mean ion energy as a function of phase angle φ

118
 In addition to the charged particle EADs onto the surface, the silicon-dioxide etching

mechanism depends on the magnitude of the individual fluxes of charged and neutral species onto

the wafer. In particular, the ratio of polymer depositing and polymer removing species is important

in dielectric etching in fluorocarbon containing plasmas. The time averaged fluxes of neutral and

ion species onto the wafer as a function φ are shown in Figure 4.6.

Figure 4.6: Fluxes of reactive species onto the wafer as a function of phase angle φ. a) Neutrals and b) ions.

With changes in φ there are systematic changes in EADs of charged particles onto the

surface. However, the neutral fluxes are fairly stable with moderate variation with changes in phase

φ for most species. With the high frequency power being held constant, the rates of electron impact

dissociation and ionization are nearly constant, producing nearly constant fluxes. The small

119
variation in neutral fluxes with φ can be attributed to the contribution to dissociation by secondary

electrons emitted from the wafer and accelerated by the sheath into the bulk plasma. With the

electron energy distribution of the secondary electrons being functions of sheath dynamics, rates

of dissociation will also vary.

Similar trends apply for the fluxes of ions to the wafer as a function of φ, which also show

little variation with phase both in terms of overall magnitude as well as the relative ratios of ion

fluxes. The one notable exception is the flux of O+ which increases for intermediate values of φ.

The atomic weight of O+ is the lightest of ions having a significant flux to the surface and so will

respond more quickly to changes in the sheath (and presheath) compared to other ions. There is

also likely a component of ionization of atomic oxygen by high energy secondary electrons. Given

that the fluxes of Ar+ and CF3+ are 2 orders of magnitude higher than that of O+, we do not expect

this sensitivity of the flux of O+ on phase angle to have a large effect on etch properties.

The feature profiles for trenches in SiO2 after 15 min of plasma etching are shown in Figure

4.7a for phase angles of φ = 0° to 180°. Varying φ for constant power produces significant

differences in etch depth, ranging from 1750 nm for φ = 0° to an immediate etch stop due to

excessive polymer buildup at φ = 180°. Due to the similar absolute fluxes of reactants for all phase

angles, this trend in etch depth is a consequence of the differences in charged particle energy

distributions onto the wafer resulting from the change in applied voltage waveforms. The higher

energy of the incident ions at low phase angles produces higher rates of removal of passivating

polymer films and higher rates of sputtering of SiO2, directly or by chemically assisted processes.

The etch stop at φ = 180° is a consequence of the lower ion energies that are unable to balance the

deposition of polymer by removal of polymer by sputtering. A net etch at φ = 180° would be

possible for lower rates of polymer deposition (lower incident fluxes of CF x radicals).

120
Figure 4.7: Feature properties for PVWT = 1000 W following 15 minutes of etching a) Feature profiles for different
phase angles (φ = 0°, 45°, 90°, 135° and 180°). b) Comparison of etch result and electric potential for (left) self-
consistent VWT produced EEAD

121
 The higher incident energies of the positive ions also enable them to partially overcome

positive electrical potential that builds up inside features. This positive potential slows ions,

thereby reducing their sputter yield with possible deflection of their trajectories inside the features,

which leads to twisting. One of the original motivations for VWT was to promote in-feature charge

neutralization on surfaces by narrowing the angular distribution of electrons to better match that

of the ions. These narrow and energetic EEADs presumably would penetrate deeper into the

feature than is possible by the broad, low energy thermal EEADs produced by conventional CCPs.

The desired end result would be a charge neutral feature.

To test this hypothesis, one would need to perform reactor scale simulations where the ion

and neutral fluxes, and IEADs remain the same, while the EEADs varied from low energy, broad

angle to high energy, narrow angle. From our attempts to find such conditions, we conclude that

the likelihood is small of being able to hold the IEAD constant while varying the EEAD. The very

conditions that modify the EEADs also modify the IEADs. A large set of computational

experiments were performed to find combinations of voltage, power and waveforms that would

maintain the same IEAD while toggling between a nearly thermal EEAD, and an EFR enhanced

EEAD. We were not able to find such conditions.

Instead, to test the hypothesis that EEADs produced by electric field reversals can decrease

the magnitude of positive charge in the feature the following process was used. Etch profiles were

compared when using EEADs produced from using VWT and a synthetic thermal EEAD having

a 3 eV electron temperature. The fluxes of radicals and ions, and IEADs, were the same for both

cases. These reference simulations were performed for φ = 0°. For these conditions, the differences

in EEADs between those generated by VWT and synthetic thermal distributions are the greatest.

The VWT case, φ = 0° also resulted in the overall most desirable feature profile.

122
 The etch profiles and corresponding final electric potential are shown in Figure 4.7b for

the (left) self consistently generated VWT produced EEAD and (right) for the synthetic thermal

EEAD. Both cases have both negative and positive charge distributed throughout the feature,

though with net negative charge at the top of the features and net positive charge deeper in the

feature. The end result is positive potential inside the feature. The maximum potential with the

VWT produced EEAD is 227 V and the maximum potential produced with thermal electrons is

346 V. The larger positive potential with thermal electrons reduces the final etch depth by 30% by

slowing the positive ions incident into the feature. This result confirms the initial hypothesis that

directional electrons produced by the EFR can be utilized to at least partially neutralize intra-

feature surface charging and lessen the decrease in etch rate resulting from that charging.

The maximum ion energy incident into the feature exceeds 3000 eV. If there was no

electron neutralization of positive charge in the feature, the positive potential would increase to

nearly this maximum value to prevent positive charge from being continually collected. Even when

using the thermal electron EEAD, the potential inside the feature is significantly smaller than the

maximum ion energy. The thermal electrons initially have isotropic trajectories which dominantly

strike the top of the feature. However, the buildup of positive potential in the feature produces

electric fields directed upwards that then accelerate these thermal electrons into the feature. The

thermal electrons accelerated into the feature then produce significant neutralization of positive

charge on the sidewalls.

In the steady state, the positive potential inside the feature increases to a large enough

magnitude so that the thermal electrons accelerated into the feature are able to neutralize

incremental additions to positive charge. This transient is directly analogous to the charging of a

dielectric surface in contact with a plasma to a negative potential that balances the flux of electrons

123
and ions. The use of the VWT produced EEAD, with its initially higher energy and velocities

towards the wafer, and narrower angular distribution, enables this balance of electron and ion

fluxes to be achieved with a lower positive potential. The difference in electric maximum potential

in the features between the thermal EEAD and the VWT produced EEAD, 120 V, corresponds to

approximately the difference in the maximum energy of the VWT produced and thermal EEADs.

Although this is not a general result, the difference in potential would not be expected to exceed

this value.

When using the VWT generated and thermal EEADs, the location of maximum potential

is not at the bottom of the feature but at depths of 1.2 m for the VWT EEAD and 0.77 m for the

thermal EEAD. The charging of the mask is nominal compared to the energies of ions and EFR

accelerated electrons. The mask charging may be more important for low aspect ratio (AR) features

where the voltages and powers are lower. However, for HAR features, this is typically not an issue.

The maximum of the positive potential is a function of AR. As the feature is etched deeper and the

AR increases, the maximum in positive potential transitions from the bottom of the feature at low

AR to the side of the feature as the AR increases [1]. For small AR, ions even with a finite angular

spread can directly strike the bottom of the feature, and so the maximum positive potential is on

the bottom of the feature. For HAR features, ions having a finite angular spread will strike the

sidewalls, and positive charging will occur dominantly on the sidewalls. The neutralized particles

then proceed as hot neutrals deeper into the feature. The shadowing by the fairly thick photoresist

produces a view angle of 3.8 (VWT) to 4.3 (thermal) which is in line with the ion angular

spreads.

The charging of the feature occurs over many RF cycles. There are no specific in-feature,

intra-cycle dynamics of the electric field as the charging and electric fields are the results of

124
averaging charged particle fluxes over these many RF cycles. For bias frequencies of many MHz

and above, distinct cyclic in-feature electric field dynamics will likely only occur when using

pulsed power.

Electron trajectories inside the feature depend on their initial distribution in energy and

angle as well as the acceleration resulting from electric fields in the feature produced by charge

deposition. In the case with the highest electron energy incident onto the wafer (φ = 0°) the mean

energy of the EEAD is 62 eV. Given the maximum electric potential in the feature is 280 V, the

electron trajectories are still likely dominated by the electric fields produced by charging. To verify

this likelihood, electron energies incident onto the surface inside the feature were spatially

sampled. To enable a side-by-side comparison of these energies, the following process was

followed. The feature shape and potential configuration resulting from the φ = 0° case after 15

minutes of etching was used and kept constant. (That is, no further evolution of the feature or

potential was allowed.) Initial electron trajectories were sampled from the EEAD produced by

VWT or from a thermal distribution having an electron temperature of 3 eV. The electron

trajectories were followed into the feature, and the electron energies striking the surface were

recorded and averaged at each spatial location. Tests were performed for VWT produced EEAD

and thermal distributions when including acceleration from the charge produced electrics fields,

and when excluding that acceleration.

125
 Figure 4.8: Spatial distribution of average electron energies incident on the surface of the feature. The feature
properties are for PVWT = 1000 W and φ = 0° after 15 minutes of etching. Results are shown for initial EEADs
produced with VWT and for thermal electron having a temperature. of 3 eV. a) When including acceleration in
electric fields produced by feature charging. b) Without acceleration due to feature charging. The view is looking
obliquely into the 3-dimensional feature having reflective boundary conditions front-and-back.

The resulting spatially resolved average electron energies incident on the feature surface

are shown in Figure 4.8. The average energies shown in Figure 4.8a are for initial EEADs (VWT

and thermal) when including acceleration in the electric fields produced by charging. The average

energies striking the surface for VWT and thermal distributions are spatially similar and

126
qualitatively resemble the electric potential shown in Figure 4.7b. These distributions largely result

from acceleration of the incident electrons by the in-feature electric fields. The average energy of

the electrons striking the surface for the initially thermal electrons is maximum at about 240 eV,

which corresponds to the maximum positive potential inside the feature. The average energy for

the electrons launched from the VWT produced EEAD is maximum at about 275 eV. This value

is larger than for the thermal electrons by 35-40 eV due to the electrons selected from the VWT

produced EEAD beginning their trajectories with significant initial velocities.

To reinforce this finding, average electron energies striking the surface of the feature were

computed for the VWT produced EEAD and for thermal electrons when excluding acceleration

by electric fields, as shown in Figure 4.8b. The average energies for the VWT produced EEAD

range up to 35 eV and those for the thermal EEAD range up to 6 eV, both values largely reflecting

the initial distributions.

These findings suggest that for reasonably large positive potentials produced inside

features, the high energy VWT produced EEADs will not dominate the resulting feature properties

compared to EEADs produced using conventional waveforms. In both cases, in the steady state

electrons are accelerated into the feature in such a manner as to neutralize the incremental charge

produced by fluxes of positive ions. This is particularly true for positive potentials that are much

larger than the average energy of the incident EEADs.

4.5 Plasma and Etch Properties for 2000 W of VWT Power

While maintaining 500 W of power on the top electrode, the power delivered by the bottom

electrode was increased to PVWT = 2000 W to be more aligned with industrial processes. The

resulting voltage amplitudes for the top and bottom electrodes, and the DC self-bias are shown in

Figure 4.3b as a function of phase angle. The top electrode voltage amplitude remains nearly

127
unchanged compared to the PVWT = 1000 W cases with there being little variation with phase angle

φ. In order to accommodate the increase in power, the bottom electrode voltage amplitude

increased by a factor of 1.5-2 compared to the PVWT = 1000 W cases. The low- and high-power

cases have nearly identical trends with phase angle, which suggests that these trends are the result

of the sheath dynamics rather than being of stochastic origin. Similar to the 1000 W cases, Vdc

decreases with increasing phase angle as a consequence of the EAE. The large magnitude of Vdc

correlates to the larger applied voltage amplitude.

The radially averaged, axial electric fields as a function of time during the RF cycle for

PVWT = 2000 W are shown in Figure 4.9 for different phase angles φ. Overall, the sheath dynamics

and electric field reversals as a function of phase angle are similar to the PVWT = 1000 W cases

(Figure 4.4) with the EFR decreasing with phase. One exception is the φ = 0° case whose EFR is

reduced compared to φ = 45° at same power and also reduced compared to the φ = 0° case at 1000

W. This apparent outlier is the result of the geometrical constraints of the plasma reactor and of

the finite electrode gap (distance between bottom and top electrodes). For example, the spatially

resolved electron densities averaged over the RF period are shown in Figure 4.10 for φ = 0°, 90°

and 180°. While the peak electron densities vary by only about 10% the spatial distributions vary

considerably with change in φ.

128
Figure 4.9: Vertical electric field component as a function of time and distance from wafer for PVWT = 2000 W for
phase angles of a) 0°, b) 45°, c) 90°, d) 135° and e) 180°. The shape of the voltage waveform applied to the bottom
electrode is shown in each image.

With low values of φ, the time averaged sheath thickness is larger, which reflects the larger

(more negative) DC bias and voltage amplitude. The sheath thickness then occupies a significant

fraction of the gap. The compressed bulk plasma is then more sensitive to, for example,

discontinuities between the top electrode and the bounding dielectric insulator where electric field

enhancements occurs. The end result is a less spatially uniform plasma. We acknowledge that, if

129
unaddressed, this expansion of the sheath could potentially affect the spatial uniformity of the

wafer processing, and that this effect may be amplified by this particular reactor configuration.

Several reactor configurations were investigated (e.g., adjusting the shape and diameter of the

focus rings, and top electrode) to confirm that the VWT scaling discussed here, and sheath

expansion and plasma compression are general trends. There are many actions, such as adjusting

the process parameters (gas composition, gas pressure, flow rates, power) and/or reactor geometry

that could be used to recoup the trends with phase angle that were obtained at the lower power.

This optimization is beyond the scope of this work. That said, this demonstration emphasizes the

concept that design of plasma processing reactors must be performed from a systems perspective.

Optimizing one aspect of the design, such as EFR, may ultimately be limited by other aspects of

the design, such as gap dimension.

Figure 4.10: Electron density for PVWT = 2000 W for phase angle φ of a) 0°, b) 90° and c) 180°). The densities are
plotted on a log-scale over 2-decades.

130
 The similarities to the trends of the lower power cases are also reflected in the IEADs to

the wafer as a function of phase angle, are shown in Figure 4.11a.

Figure 4.11: Energy and angular distributions for charged particles incident onto the wafer for P VWT = 2000 W for
phase angles of φ = 0°, 45°, 90°, 135° and 180°. a) Ions and b) electrons. The distributions are separately normalized
and plotted on a log-scale over 2 decades.

131
 While compared to the lower power cases, the ion energies are higher overall due to the

larger voltage amplitudes, the peak energies still decline with increase in the phase angle from a

maximum of about 4500 eV for φ = 0° to 3000 eV at φ =180°. Comparing the cases with highest

and lowest ion energies (φ = 0° and φ = 180°), the peak ion energy was reduced by approximately

85% at 1000 W and reduced by 38% for 2000 W. This degree of ion energy control should enable

some degree of customizing fluxes for energy-sensitive surface processes. In particular, the

distribution of high-energy ions compared to low energy ions within the IEAD is sensitive to phase

angle.

These trends for the IEADs are in contrast to the much lighter and more agile electrons,

whose EEADs to the wafer are shown in Figure 4.11b. The EEADs display the same general trends

as in the low power case. The EEADs have high energies and narrow angular spreads for low

values of φ compared to thermal distributions due to the electric field reversal produced by VWT.

The energies decrease and angular spread increases with increasing phase angle φ. The φ = 0° case

is, again, a notable exception, having a lower peak energy than a larger phase angle. This exception

correlates with the smaller EFR described above. The suppressed electric field reversal due to

geometric confinement of the sheath leads to reduced electron acceleration, resulting is the

comparatively lower electron energy.

The IEADs, EEADs and fluxes for different phase angles for 2000 W bias power were used

as input to otherwise identical MCFPM simulations. The final etch profiles are shown in Figure

4.12a. Significant etch depths were reached for all phases, in deference to the PVWT = 1000 W

cases where an etch stop occurred due to excessive fluxes of polymer forming radicals, or lack of

polymer sputtering ions, at large phase angle. There are, however, significant differences in etch

depth and overall feature quality. The φ = 0° case produces the deepest overall etch while having

132
perhaps a tolerable amount of bowing. The φ = 45° case has less bowing but lower etch rate. These

trends result from φ = 0° case having the higher ion energies (faster overall etch rate) while the φ

= 45° has a narrower ion angular distribution (less bowing), albeit at a lower energy. Some

component of the narrower feature could be a result of the EEAD at φ = 45° being more energetic

which enables some non-negligible decrease in in-feature charging. The shape of the EEAD is

important to moderately increasing etch rate and moderately modifying sidewall slope through

neutralizing charge. However, the final feature quality is dominated by the EAD of ions rather than

the electrons.

Analogous to the low power case, MCFPM simulations using synthetic thermal EEADs

were performed, and the resulting profiles are compared to etch profiles resulting from VWT

produced EEADs in Figure 4.12b for φ = 0°, 90° and 180°. In the φ = 0° and 180° cases the

differences between the profiles obtained with thermal and VWT EEADs are small and below the

statistical threshold discussed in Appendix A. The profile obtained with the VWT EEAD for φ =

90° is deeper with a similar amount of tapering compared to the thermal EEAD. These trends result

from the EEADs onto the wafer for these conditions having the highest energy and lowest angular

spread for intermediate values of φ. This correlation between the improvement of the final etched

feature and the energy and directionality of the electrons reinforces the conclusion that these

directional electrons enhance the natural neutralization of charge that is produced by acceleration

of thermal electrons into the feature. However, the properties of the EEADs do not dominate the

shape of the feature.

133
Figure 4.12: Feature properties for PVWT = 2000 W following 15 minutes of etching a) Feature profiles for different
phase angles (φ = 0°, 45°, 90°, 135° and 180°). b) Comparison of profiles obtained with VWT generated EEADs
and synthetic thermal electron distribution for phase angles φ of 0°, 90° and 180°.

134
4.6 Concluding Remarks

To investigate the consequences of voltage waveform tailoring (VWT) on etching of SiO 2

using a capacitively coupled Ar/CF4/O2 plasma, computational investigations of the reactor scale

and feature scale processes were performed. While applying 500 W at 80 MHz to the top electrode

with the goal of sustaining a constant plasma density, the bottom electrode was driven using VWT

composed of consecutive harmonics with a fundamental frequency of 1 MHz at powers of 1000

W and 2000 W. The phase of the even harmonics φ was varied in order to generate electrical

asymmetry as well as electric field reversal in the bottom sheath. It was found that changing φ

enables control of charged particle dynamics incident onto the wafer surface which in turn

translates to significant changes in etch profiles. At low phase angle the EAE as well as the EFR

are most significant, which leads to increased energy and directionality of ions as well as electrons

onto the wafer.

These favorable distributions in energy and angle increase etch rate by partially reducing

the negative effects of intra-feature charging. The directional electrons are better able to reach the

lower echelons of the feature and partially neutralize positive charges. The higher energy ions have

larger rates of chemical sputtering while also being more able overcome the remaining deflecting

electric fields produced by in-feature charging. Overall, for otherwise equal processing conditions

and time, low φ cases produced higher etch rates and more favorable sidewall slope. For low bias

power, the reduction in ion energy with large values of phase angle resulted in an etch stop due to

the large incident flux of polymerizing radicals. At low bias power, having higher ion energies,

sputtering of the polymer enabled competitive etching even at high phase angle.

A key finding of this investigation is the VWT produced EEADs having large energy and

narrow angle do not produce the expected significant reductions of in-feature charging compared

135
to thermal EEADs. The VWT produced EEADs did produce somewhat higher etch rates and lower

in-feature potentials. However, the buildup of positive charge in the feature accelerates thermal

electrons into the feature to energies commensurate with that of the positive potential. The VWT

produced EEADs are also accelerated into the feature by this positive potential. If the positive

potential is significantly larger than the average energy of the VWT produced EEAD, there is little

difference in feature properties between having thermal or VWT generated EEADs. Both are

accelerated into the feature by the positive charge to sufficient energy to neutralize, in the steady

state, the incremental positive charge brought into the feature by ions.

Another key finding is the need for a systems perspective in process design and in

employing VWT in particular. The results of this study showed that at low power, the VWT

generated EEADs produced a beneficial effect, enabling higher etch rates for a given power

compared to conventional sinusoidal excitation. Conversely, the same etch rate could be sustained

at a lower power. We also found that this benefit was not universal. At higher powers, we found

that the gap dimension (distance between wafer and top electrode) ultimately limited the ability of

VWT to generate electric field reversal and energetic fluxes to the wafer. Low phase angle

produces a larger (more negative) DC bias, which thickens the sheath. For a narrow gap CCP, the

sheath may consume a large fraction of the gap, which then interferes with the formation of EFR.

The fundamental physics of VWT generated EEADs applies to a wide range of powers. However,

as with many excitation schemes in plasma etching, the benefits of VWT are likely process

dependent, being sensitive to geometry, frequency, pressure and gas mixture.

In the ideal, one would like to tailor an IEAD to best produce the desired etch feature, and

then tune the feature profile by tailoring the EEAD. Unfortunately, independently producing

specified IEAD and EEAD with the same waveform is difficult. Changing the waveform to

136
generate, for example, the desired EEAD will also affect the IEAD. There may be opportunity to

separately tailor the IEAD and EEAD by using binary-pulsed system. The system would consist

of two separate waveforms alternately applied, with one waveform optimized to produce the

desired IEAD and the second to produce the desired EEAD.

137
4.7 References

[1] S. Huang, C. Huard, S. Shim, S. K. Nam, I.-C. Song, S. Lu and M. J. Kushner, Journal of

Vacuum Science & Technology A 37, 031304 (2019).

[2] J. K. Kim, S. Il Cho, S. H. Lee, C. K. Kim, K. S. Min and G. Y. Yeom, Journal of Vacuum

Science & Technology A: Vacuum, Surfaces, and Films 31, 061302 (2013).

[3] J. K. Kim, S. H. Lee, S. Il Cho and G. Y. Yeom, Journal of Vacuum Science & Technology

A: Vacuum, Surfaces, and Films 33, 021303 (2014).

[4] U. Czarnetzki, J. Schulze, E. Schüngel and Z. Donkó, Plasma Sources Sci Technol 20, 024010

(2011).

[5] J. Schulze, Z. Donkó, B. G. Heil, D. Luggenhölscher, T. Mussenbrock, R. P. Brinkmann and

U. Czarnetzki, J Phys D Appl Phys 41, 105214 (2008).

138
 Chapter 5 DC Self-Bias and Ion Energy at Very Low Driving Frequencies.

5.1 Introduction

Recent industry trends have favored the use of very low frequency (VLF) biases. VLF,

typically frequencies below 1-2 MHz, correspond to the thin sheath limit in which the ion entering

the sheath will cross the sheath in a small fraction of the RF cycle. In this regime, maximum ion

energies incident onto the wafer extend to Vrf-Vdc, where Vrf is the amplitude of a sinusoidal bias

and Vdc is the negative self-bias on the blocking capacitor. At the same time, the use of voltage

waveform tailoring (VWT), is being investigated to customize those ion energy distributions.

VWT consists of a bias which is the sum of multiple harmonic frequencies on a single electrode.

Due to the added technical complexity and lack of fundamental understanding of its role in

modifying ion and electron dynamics in and near the plasma sheath, VWT is only now being

broadly adopted by industry [1–7].

The direct measurement of ion energies during industrial plasma etching processes is

usually not performed due to the added technical complexity or to avoid disturbing the plasma. As

a result, the DC self-bias, Vdc is frequently used as a proxy for ion energy onto the wafer. This is a

good approximation if the sheath is collisionless and the system is operating in the thick sheath

limit, where the ion requires many RF cycles to cross the sheath. For these conditions, the ion

energy distribution (IED) striking the wafer has a narrow width in energy centered on -Vdc. This is

typically the condition intended for control of the IED through VWT.

In this chapter, we discuss results from a computational investigation into the relationship

139
between Vdc and the IED for capacitively coupled plasmas employing VWT over a large range of

fundamental bias frequencies. The computed reactive fluxes to the substrate and their energy

distributions were used as input to a profile simulator to assess their influences of these reactive

fluxes on HAR features etched into SiO2. We found that at very low frequencies the ions are able

to dynamically react to the applied potential and as a consequence incident ion energy onto the

wafer can decouple from the trends suggested by Vdc.

The reactor and feature scale models (HPEM and MCFPM) are described in Chapter 2

Brief descriptions of reactor and feature scale simulation setup and geometry employed in this

investigation are in Section 5.2. The results of the reactor scale gas phase simulations are presented

and discussed for constant bias voltage amplitude in Section 5.3 and for constant bias power in

Section 5.4. Results are discussed for simulated profiles of HAR trenches etched in SiO2 using the

reactive fluxes produced by these waveforms are discussed in Section 5.5. Concluding remarks are

in Section 5.6.

5.2 Simulation Setup and Geometry

5.2.1 Reactor Scale Setup

The investigation of the reactor scale plasma dynamics was performed using the Hybrid

Plasma Equipment Model (HPEM) [8–10] which is described in Section 2.1. The reaction

mechanism used for Ar/O2/CF4 plasmas is described by Huang et al [11].

A schematic of the cylindrically symmetric CCP reactor used in this investigation is shown

in Figure 5.1. The reactor, modeled after multi-frequency CCPs used in industrial plasma etching

applications, consists of two parallel plate electrodes with a diameter of 30 cm separated by a 2.8

cm gap. A silicon wafer is mounted on the bottom electrode and a showerhead gas inlet is

distributed across the top electrode. A focus ring made of quartz to improve uniformity of fluxes

140
to the wafer surrounds the substrate. The dielectric constant of the quartz ring is ε/ε 0 = 4. While

the conductivity of the quartz ring is negligible, that of the Si wafer is 0.05/Ω-cm. The feedstock

gas is an Ar/CF4/O2 =75/15/10 mixture flowing at 500 sccm through the showerhead. The reactor

pressure is held constant at 40 mTorr at the location of a pressure sensor near the pump port. This

is accomplished by throttling the rate of pumping.

Figure 5.1: Reactor geometry and circuit diagram.

Voltage Waveform Tailoring (VWT) is provided by a power supply connected to the

bottom electrode through a blocking capacitor of 100 nF. As such, we are not addressing distortion

of the waveform originating from the power supply that may occur by the impedance of the

transmission line and substrate. In quasi-steady state operation, the value of Vdc should be

independent of the value of the blocking capacitor provided that the RC time constant of the

plasma-capacitor series impedance is large compared to transients in current. The capacitance used

here is a balance of there being an acceptably short charging time and long enough RC time

constant so that there is little variation in the Vdc during the RF cycle.

141
 The plasma is largely sustained by a sinusoidal voltage with amplitude VT applied to the

top electrode with a fundamental frequency of 80 MHz. To control the dynamics of charged

particles impinging on the wafer surface, a customized voltage waveform, V(t), is applied to the

bottom electrode. The waveform consists of a fundamental sine wave with a frequency f0 = 1 MHz

to 10 MHz and consecutive harmonics according to (1.56):

In this work the number of harmonics is N = 5. The consequences of VWT were

investigated by varying the phase shift of the even harmonics φ for k = 2, 4. The resulting

waveforms are equivalent to those shown in Figure 1.14 for phase shifts of φ = 0°, 45°, 90°, 135°

and 180°. With the shape and frequency of the applied voltage being closely linked to the plasma

properties, two scenarios were investigated:

a) Voltages are held constant at VT = 300 V and V0 = 1500 V, as phase shift φ and fundamental

frequency f0 are varied.

b) Applied voltages are dynamically adjusted to maintain a constant power deposition of PT

= PB = 2000 W, as phase shift φ and fundamental frequency f0 are varied.

5.2.2 Feature Scale Setup

The fluxes and particle energy and angular distributions to the wafer obtained from HPEM

are used as input to the Monte Carlo Feature Profile Model (MCFPM) to simulate the evolution of

etch features in semiconductor device materials [11,12]. The MCFM is described in Chapter 2.

The mechanism for etching SiO2 in an Ar/O2/CF4 gas mixture used in this work is described

in Ref. [11]. All incoming ions or hot neutrals resulting from neutralization of ions striking surfaces

can physically sputter all materials according to the incident particle's mass, angle and energy as

well as the surface binding energy of the material. Formation of a complex between fluorocarbon

radicals and SiO2 lowers the binding surface energy, thereby reducing the chemical sputtering

142
threshold. CxFy radicals deposit a thin polymer layer (at most a few nm) which is the primary

source of sidewall passivation within the feature, whose thickness is controlled by etching with

oxygen radicals.

To demonstrate the consequences of changes ion energy and angular distributions (IEADs)

resulting from varying f0 and phase angle, simulations were performed of etching a trench through

3000 nm of SiO2 covered by a 700 nm thick mask with a 100 nm opening; and terminated by a

100 nm thick Si stop layer. The total aspect ratio is approximately 30. The geometry used for the

feature evolution is shown in Figure 5.2. The simulation was performed in 3-dimensions with a

mesh consisting of 60 cells wide by 750 cells tall by 20 cells deep using periodic boundary

conditions, producing cubic voxels with a 5 nm side length. The etching was performed for a fixed

time of 10 minutes.

The intent of the simulation is a uniform trench perpendicular to the plane shown in Figure

5.2 which could, in principle, be performed in 2-dimensions. However, we have found that more

robust results are produced when performing a finite depth 3-dimensional simulation. In 2-

dimensions, roughness or statistical variation in surface properties are extended infinitely into the

third dimension, which can systematically prejudice the outcome, for example, overemphasizing

twisting. The finite depth 3D simulations, even with periodic boundary conditions, largely removes

these systemic biases.

143
Figure 5.2: Feature scale model initial conditions for simulating plasma etching of SiO 2. PR represents photoresist.

5.3 Reactor Scale Plasma Properties with Constant Applied Voltage

With constant applied voltage amplitudes of VT = 300 V and V0 = 1500 V, the fundamental

frequency f0 was varied from 1 MHz to 10 MHz. For each frequency a sweep across the phase

angle φ was performed with φ = 0°, 45°, 90°, 135° and 180°. The electron density for the base case

with f0 = 1 MHz and φ = 0° is shown in Figure 5.3a. The cycle averaged electron density at mid-

gap is 3.9  1010 cm-3, sustained by an electron temperature of 3.1 eV. Ionization at mid-gap is

approximately 90% by bulk electrons and 10% by secondary, sheath accelerated electrons. The

electronegativity (negative ion density/electron density) at mid-gap is 0.3. Electron density ne and

temperature Te are fairly homogeneous as a function of radius as shown in Figure 5.3a and d. The

ionization is dominated by the bulk electrons. The spatial electron source distribution, generated

144
by bulk electrons Sb. closely resembles those of ne and Te, as shown in Figure 5.3b. The smaller,

but still significant amount of ionization by secondary electrons S s is shown in Figure 5.3c. Since

the bulk ionization source is mostly a consequence of electrons accelerated by sheath motion, its

maximum is located at the sheath edges. Ss results from electrons accelerated by the total sheath

potential and is maximum close to the bottom electrode.

Figure 5.3: Plasma properties of the Ar/CF4/O2 CCP base case with VT = 300 V, V0 = 1500 V, f0 = 1 and φ = 0°.
Time averaged a) electron density ne, b) ionization source by bulk electrons Sb, c) ionization source by secondary
electrons Ss, and d) electron temperature, Te.

The different voltage waveforms that result from changing the phase angle φ can alter the

discharge dynamics through differences in sheath expansion heating, electron confinement and

modifying surface processes such as ion and electron induced secondary electron emission. While

the constant sinusoidal voltage applied to the top electrode is meant to provide a constant

background ionization and electron heating, some variation in electron density ne as a function of

φ nevertheless occurs and is shown in Figure 5.4 for fundamental frequencies f0 = 1 MHz to 10

145
MHz. Generally, ne is a function of the total input power at frequencies that produce electron

heating. With the top electrode powered at 80 MHz, the majority of this power is expended in

electron heating. With varying harmonic content with changing φ on the bottom electrode, power

deposition into electrons also varies, directly translating to changes in electron heating and

ionization.

Figure 5.4: Average electron density as a function of phase shift φ for fundamental frequencies f0 = 1 MHz to 10
MHz at constant applied voltage.

The decrease in electron density with increasing phase angle is at least partially the

consequence of less efficient electron confinement. Electrons are largely confined to the bulk

plasma by electric fields in the sheaths which point towards surfaces during the majority of the RF

cycle. Electrons typically reach the surface (to balance positive ion current) by diffusion across the

sheath during a short period during the anodic portion of the cycle. If transport by diffusion is

insufficient to balance currents, an electric field reversal may occur to accelerate electrons towards

the surface [13]. For φ = 0° (peak waveform), the plasma sheath is nearly at its maximum width

with the most negative sheath potential for the majority of the RF cycle, allowing electron transport

to the surface only during the brief collapse in the sheath at the peak of the anodic portion of the

146
cycle. The opposite is true for φ = 180° where the plasma sheath is collapsed (anode-like) for most

of the RF period, allowing electrons to transport to the bottom surface for a significantly longer

fraction of the cycle.

With constant voltage (VT = 300 V, V0 = 1500 V), power deposition from the top electrode

is a function of the bulk plasma properties and not independent of the bottom electrode power

coupling. Power deposition as a function of phase angle φ and fundamental frequency f0 are shown

in Figure 5.5a for the top electrode (PT) and Figure 5.5b for the bottom electrode (P0).

Figure 5.5: Power deposition for a) top electrode and b) bottom electrode as a function of phase shift φ for
fundamental frequencies f0 = 1 MHz to 10 MHz.

147
 Higher fundamental frequencies f0 on the bottom electrode produce more efficient electron

heating and higher power deposition for a fixed voltage. For our conditions, electron power

deposition is dominated by stochastic heating resulting from sheath expansion (or sheath speed)

which is more rapid and occurs more often at higher frequencies. Increasing the fundamental

frequency f0 of the voltage on the bottom electrode therefore leads to a significant increase in

power deposition, PB. However, this relation is non-linear. A ten-fold increase in frequency f0 (1

MHz to 10 MHz) only results in a 1.5 (φ = 180°) to 2 (φ = 0°) increase in PB. These trends in power

deposition with frequency directly translate to the corresponding trends in ne as shown in Figure

5.4. Smaller phase angles have more rapidly expanding sheaths producing more electron heating,

leading to an increase in PB. With constant top voltage VT, additional electron heating produced by

PB which increases electron density translates to an increase in power deposition by the top

electrode, PT.

As described in Chapter 1, the dependence of Vdc on the shape of the applied waveform

through the generation of the electrical asymmetry effect (EAE) given by Eq. (1.53) [14–17]. This

relation implies that the magnitude of Vdc is largest (most negative) for φ = 0° and decreases with

increasing φ (becoming more positive). The values of Vdc produced by the model, shown in Figure

5.6a as a function of phase shift φ for fundamental frequencies f0 = 1, 2, 5 and 10 MHz, generally

verify this scaling. |Vdc| tends to decrease with increasing φ for all frequencies (the negative Vdc

becoming more positive). A few discrete exceptions occur, for example f0 =5 MHz, φ = 0°; and f0

= 1 MHz, φ = 180°. These exceptions are likely due to effects such as contributions of Vdc to

secondary electron emission produced ionization, relative contributions of conduction and

displacement current in a geometrically complex reactor and electronegativity of the plasma which

are not explicitly included in the theory leading to Eq. (1.53).

148
 The mean ion energies striking the wafer are shown in Figure 5.6b as a function of phase

shift φ for fundamental frequencies f0 = 1, 2, 5 and 10 MHz. The expectation is that mean ion

energy should closely track the magnitude of Vdc with a change phase shift φ. This expectation is

met for the highest frequency 10 MHz. At lower frequencies, beginning with 5 MHz, the

correlation of mean ion energy with magnitude of Vdc as a function of phase shift φ begins to break

down. At 1 MHz, the mean ion energy increases with phase shift while magnitude of Vdc is constant

or is decreasing.

Figure 5.6: Plasma properties as a function of phase shift φ for different fundamental frequencies f0 = 1 MHz to 10
MHz for constant voltage. a) Vdc. b) Mean ion energy incident onto the wafer.

149
 Recall that Vdc results from the requirement that the current flowing to each electrode

(powered and grounded) be equal over the RF cycle. In our model, currents are computed at the

electrode surface. If that surface is in direct contact with the plasma, the current consists of both

conduction current (charged particles flowing to the surface) and displacement current (

( )
jD =   E t ). If the electrode is in contact with a conductive material that is in contact with the

plasma, such as the wafer, the current collected by the electrode consists of conduction current

(through the wafer) and displacement current. If the electrode is buried beneath or within non-

conducting dielectrics, the current the electrode collects is only displacement current.

Contributions of displacement current to the bottom electrode may originate from charging of the

top and sides of the focus ring. With the ratio of conduction to displacement current being functions

of frequency which differs for top and bottom electrodes, there is no expectation that Vdc should

be independent of frequency.

Another factor that may affect Vdc with frequency is electron heating. With electron heating

being dominated by sheath expansion, higher bias frequencies f0 produce more electron heating

adjacent to the sheath at the bottom electrode for any given voltage amplitude. With increasing f0

this increase in local as well as bulk electron density ne leads to a decrease in shielding lengths λD

and a decrease in the sheath width d𝑠 at the bottom electrode. The end result is a change in the

ratio of surface adjacent ion densities (nsp \ nsg). As described by Eq. 1.51 and 1.53 this change

directly translates to a change in the Vdc.

An ion’s ability to react to temporally changing electric fields in the sheath is related to the

time required for the ion to cross the sheath, T, compared to the RF period, 1/RF, described by

the sheath ion inertia coefficient,

150
 1/2
 2d 2 m 
S = T RF =  RF  s i  (5.1)
 qVS 

where 𝑑𝑠 is the average sheath thickness, VS is the average sheath potential, mi is the ion mass and

q is the elementary charge. For S > 1, ions typically do not dynamically respond to changes in the

electric field that occur in the sheath (high frequency, large mass, thick sheath) . Ions transiting

through sheaths whose oscillation frequency produces S >> 1 arrive at the surface with an energy

close to the average sheath potential. Although this scaling assumes a sinusoidal voltage with a

single ion species, it can nevertheless serve as a general guide to understanding the frequency

dependent trends in average ion energy onto the wafer shown in Figure 5.6b.

S depends on the value of the sheath width. Defining a precise criterion for location of the

sheath-bulk plasma transition is an active area of research [18–27] and beyond the scope of this

work. In this work the sheath edge is defined as the location at which the sheath electric field has

decayed to 10-3 of its maximum value. The resulting sheath thickness is shown in Figure 5.7a as a

function of the fundamental frequency f0 with φ = 0°.

Since the sheath thickness is generally inversely proportional to plasma density at the

sheath edge, which increases with f0, the sheath thickness decreases with increasing frequency. S

as a function of f0 for φ = 0° is shown in Figure 5.7b. Consistent with the premise above, non-

dynamic ion transport through the sheath with S > 1 occurs for f0 = 10 MHz. Dynamic transport

behavior where ions react to time variations of electric field in the sheath with S < 1 occurs for f0

= 1 MHz. As such, at the high end of frequencies the ion dynamics are dominated by time averaged

sheath properties, such as the DC self-bias, Vdc. It is for that reason, that mean ion energy as a

function of phase angle for f0 = 10 MHz scales proportionally to the magnitude of Vdc, as shown

in Figure 5.6b.

151
 Figure 5.7: Sheath properties as a function of fundamental frequency f0 for φ = 0° at constant voltage. a) Sheath
thickness ds and b) sheath ion inertia coefficient S.

Moving to lower fundamental frequencies f0 the trend for average ion energy as a function

of phase angle φ reverses. At low f0 the average ion energy increases with increasing φ. Lowering

the fundamental frequency results in the sheath modulation time scales approaching a regime in

which ions can dynamically react to transients in the electric fields. For example, compare the

voltage waveforms shown in Figure 1.14 for φ = 0° (peak) and φ = 180° (valley). Assuming

dynamically reacting (positive) ions (S < 1), the maximum ion flux to the electrode occurs when

the applied voltage is most negative. At φ = 0° the applied potential is negative with respect to the

plasma for most of the RF period. However, the minimum sheath potential, Vmin, has a smaller

magnitude than for φ = 180°. It is this ability of the ions to (at least partially) react to applied

152
voltage transients that results in their incident energy onto the surface to scale with the DC self-

bias at high frequencies and the applied minimum potential at low frequencies.

The incident average energies of select ion species as a function of atomic mass for f0 = 1

MHz and φ = 180° are shown in Figure 5.8.

Figure 5.8: Incident average energy of select ion species as a function of mass.

Due to their lower inertia, the lighter O+ ions can more rapidly react to changes in the

electric field than heavier species such as CF3+. Consequently, the lighter ions can reach an energy

closer to the maximum sheath potential drop during the small fraction of the RF period when the

potential is most negative. While the average ion energies differ by approximately 60 eV which is

only about 5% of the total, these trends illustrate the relation between the average ion energy and

the ion's ability to dynamically react to transient fields. This scaling is also shown by the trends in

the IEADs collected at the wafer surface shown in Figure 5.9 for f0 = 1 MHz and 10 MHz for phase

shifts φ = 0, 45, 90, 135 and 180°. At the lowest frequency the increase average energy with φ is

mirrored in the IEAD where the mean energy is roughly equal to -Vdc - |Vmin|.

153
 As with the respective average energies onto the surface, there are significant differences

in the trends for IEADs as a function of φ between the low and high fundamental frequencies.

Overall, with f0 = 10 MHz both maximum and average energies trend downwards with increasing

φ while maintaining an approximately constant spread in energy.

Figure 5.9: IEADs as a function of phase shift φ for fundamental frequencies a) f0 = 1 MHz and b) f0 = 10 MHz at
constant applied voltage. Values are plotted on a 2-decade log scale.

154
 Two distinct regimes for the behavior of average ion energy and IEAD with respect to the

`phase angle (and Vdc) can be defined: The first is the steady state ion regime having S >1 (large

f0) in which average ion energy onto the surface is well correlated to the magnitude of Vdc, with the

average energy being maximum at φ = 0° and minimum at φ = 180°. The second is the dynamic

ion regime having S < 1 (small f0) where average ion energy and the IEAD are dominated by the

temporal response of the ions to the applied voltage waveforms – with the average ion energy

being minimum at φ = 0° and maximum at φ = 180°.

5.4 Reactor Scale Plasma Properties with Constant Applied Power

To better align with industry practice where power (as opposed to voltage) is the control

variable, simulations were performed in which the total power deposition was held constant for

both electrodes while varying frequency and phase angle of the waveform on the bottom electrode.

This was achieved by adjusting the voltage on the top electrode, VT and bottom electrode, V0, to

deliver power depositions of PT = P0 = 2000 W.

The voltage amplitudes applied to the top electrode VT are shown in Figure 5.10a as a

function of phase shift φ for fundamental frequencies f0 = 1 MHz to 10 MHz. The mean electron

density as a function of φ for f0 = 1 to 10 MHz is shown in Figure 5.11. When delivering constant

power, VT is a weak function of φ for a given f0. However, the VT decreases with increasing f0. For

a given f0 with power delivered though the bottom electrode being held constant, the plasma

density and so total impedance of the system does not significantly vary with φ. Consequently, the

change in VT with φ to deliver constant power is small. However, with increasing f0, the fraction

of power delivered by the bottom electrode devoted to ion acceleration decreases and that delivered

to electron heating increases, which for constant power, produces an increase in electron density

and decrease in impedance. A lower VT is then required to deliver constant power.

155
Figure 5.10: Applied voltage for constant power (2000 W) as function of phase shift φ for fundamental frequencies
f0 = 1 to 10 MHz applied to the bottom electrode. a) Top electrode and b) bottom electrode.

The voltage applied to the bottom electrode, V0, with respect to f0 and φ is shown in Figure

5.10b for constant power deposition. Power deposition for constant voltage and f0 generally

decreases with increasing φ, as shown in Figure 5.5. To recoup this decrease in power to maintain

constant power, V0 increases with increasing φ. The thickness of the sheath adjacent to the bottom

electrode ds and the sheath ion inertia coefficient S as a function of f0 are shown in Figure 5.12 for

φ = 0° at constant voltage. With ds being a function of the electron density, which increases with

156
f0, the sheath thickness decreases. This reduction in ds contributes to a decrease in S. However, this

effect is dominated by the changes in f0 and Vs, ultimately resulting in an increase of S with f0. This

trend indicates that with constant power, a transition also occurs from the steady state ion regime

to the dynamic ion regime occurs as a function of f0.

Figure 5.11: Mean electron density as a function of phase shift φ for fundamental frequencies f0 = 1 MHz to 10 MHz
at constant applied power.

The DC self-bias Vdc as a function of phase angle φ for f0 = 1 MHz to 10 MHz is shown in

Figure 5.13a while maintaining constant power. The trends with φ are similar to those when

keeping voltage constant, shown in Figure 5.6a, where the magnitude of Vdc decreases (becomes

more positive) with φ due to the electrical asymmetry effect. However, when holding power

constant, the amplitude of V0 decreases with increasing f0 due to there being a larger proportion of

power being more efficiently dissipated by electron heating. With a decrease in amplitude of V0,

the magnitude of Vdc decreases, becoming more positive.

157
 Figure 5.12: Sheath properties as a function of fundamental frequency f0 for φ = 0° at constant power. a) Sheath
thickness ds and b) sheath ion inertia coefficient S.

The average ion energies onto the wafer are shown in Figure 5.13b as a function of phase

shift φ for f0 = 1 MHz to 10 MHz. The corresponding IEADs as a function of phase shift φ for f0 =

1 MHz and f0 =10 MHz are shown in Figure 5.14. As when holding V0 constant, for higher

frequencies the average ion energies scale with the magnitude of Vdc due to the inability of the ions

to dynamically react to the transients in applied voltage. For f0 = 10 MHz and f0 = 5 MHz this leads

to a decrease in average energy onto the wafer with increasing phase angle. As with constant

voltage, at low frequencies f0, the average ion energies are poorly correlated (and, in fact, anti-

correlated) with Vdc and instead reflect the maximum sheath potentials.

158
Figure 5.13: Plasma parameters as a function of phase shift φ for fundamental frequencies f0 = 1 MHz to 10 MHz at
constant power. a) DC self-bias and b) Mean ion energy delivered to the wafer.

As a result of the applied voltage no longer being constant, the resulting average ion

energies are strong functions of phase shift φ for fundamental frequencies f0. For example, the

mean energies at 1 MHz are larger than those at 10 MHz by a factor of 2 for φ = 0 ° and 4.5 at φ

= 180 °. These similarities and differences in mean ion energy between the constant voltage and

constant power cases translate to the respective IEADs. At the lowest frequency (f0 = 1 MHz) the

IEADs for φ = 0 and 180° are more monoenergetic, reflecting the applied waveforms. The shapes

of the applied voltage for these phase shifts roughly consist of a plateau and a singular well-defined

short excursion to a maximum at φ = 0° and to a minimum at φ = 180°. The intermediate cases (φ

159
= 45, 90 and 135°) do not have similar monoenergetic structures as the applied waveforms

themselves do not predominantly consist of a stable voltage plateau or singular peak.

Desired anisotropic etch profiles require a balance of polymer deposition and activation

energy delivered by ions and hot neutrals whose rates are largely determined but the magnitude of

the reactive fluxes arriving on the surface.

Figure 5.14: IEADs as a function of phase shift φ with constant power for fundamental frequencies a) f0 = 1 and b)
10 MHz. Values are plotted on a 2-decade log scale.

160
Figure 5.15: Fluxes to the wafer as a function of phase shift φ for fundamental frequencies f0 = 1 to 10 MHz while
keeping power constant. a) Total polymerizing flux, b) O-atom flux, c) total ion flux and d) ratio of polymerizing
flux to ion flux.

161
 The generation of oxide-polymer-complexes which enables selective removal of the SiO2

is correlated to the incident flux of the polymerizing CxFy gas phase species, which are shown as

a function of φ and f0 in Figure 5.15a. The overall trends are that polymerizing fluxes are weak

functions of phase angle φ while generally increasing with increasing f0 on the bottom electrode.

With increasing f0, a larger proportion of bias power is dissipated by electron heating, which then

produces a larger rate of dissociation of the CF4 feedstock gas. With increasing phase angle, the

sheath is collapsed for a greater fraction of the cycle, thereby moving the effective (time average)

sheath edge closer to the wafer. This shift in sheath edge places radical production closer to the

wafer, and so increases fluxes.

The thickness of the polymer layer by deposition is balanced by chemical, isotropic etching

by O radicals and anisotropic sputtering by directional ions. The time and spatial average fluxes

to the wafer of atomic oxygen, O, the most prevalent oxygen radical are shown in Figure 5.15b.

The general trends reflect those of the polymerizing fluxes (Fig 16a), fluxes that increase with φ

and f0.

The magnitude of the ion flux, shown in Figure 5.15c, has a first order effect on etch rates

and feature quality. Applied electric fields directly influence charged particle transport, whereas

electron impact dissociation, excitation and ionization are at least one step removed as these

phenomena occur as a result of electron collisions following their acceleration by electric fields.

The dominant mechanism for ion power deposition is through sheath acceleration which, assuming

a collision-less sheath, is proportional to the product of ion flux and incident ion energy. Assuming

a constant fraction of power deposition by ion acceleration, a decrease in incident ion energy would

necessitate an increase in ion flux to maintain the desired power. This is the trend for f 0 =5 MHz

and 10 MHz in which the trend of ion-flux with phase angle is opposite that of the ion energy. This

162
explanation fails to capture the low frequency behavior in ion flux where an increasing fraction of

the constant power deposition is due to electron heating.

5.5 SiO2 Etching at Constant Power

To characterize the consequences of reactant fluxes and IEADs on SiO2 etch properties, profile

simulations were performed for f0 = 1 MHz and 10 MHz for phase angles φ = 0° to 180° while

maintaining constant power. The resulting features for f0 = 1 MHz are shown in Figure 5.16.

Figure 5.16: Predictions for etch profiles in SiO2 at constant power with fundamental frequency f0 = 1 MHz and
varying phase angles φ.

163
 For constant processing time, etching through the entire 3000 nm thick SiO 2 layer only

occurred for φ = 180° while for φ = 0° the final etch depth is 1600 nm. The total relative etch rates

are 0.53, 0.6, 0.78, 0.93 and 1.0 for φ = 0°, 45°, 90°, 135°, and 180° respectively. This trend is

directly correlated to the trend in incident ion energy. Higher ion energies generally more rapidly

remove SiO2 by direct or chemically enhanced sputtering and retain their ability to do so after

losing energy to grazing sidewall collisions. The profiles produced by IEADs at higher φ have

more desirable characteristics such as straighter sidewalls and less overall bowing. This benefit is

a direct consequence of the narrower angular distribution of the incident ions. The SiO 2 etch

mechanism contains a SiO2-polymer complex which requires fluorocarbon radical fluxes as

reactants. Since these fluxes have a small increase with φ (Figure 5.15a), the increased availability

of reactants could also play a role in the increased etch rate if the etch progression is flux limited

as is often the case in HAR features.

The results of the feature etching for f0 = 10 MHz are shown in Figure 5.17. The overall

etch rates are similar to those at f0 =1 MHz while the incident ion energies are significantly lower

overall. The trend in etch depth with phase angle is the opposite to that for f0 = 1 MHz. With f0 =

10 MHz, etch rates mildly decrease with increasing phase angle φ while increasing for f0 = 1 MHz.

These trends most directly follow from both the average ion energies and maximum ion energies

trending higher with increasing phase angle φ at f0 = 1 MHz while decreasing at f0 = 10 MHz. A

secondary effect is that ion fluxes are nearly constant with increasing phase angle at f0 = 10 MHz

while increasing at f0 = 1 MHz. With chemical and physical sputtering rates depending on ion

energy as 1/2, large increases in ion energy are required for significant increases in etch rates. That

said, the likelihood for specular scattering from side walls increases with increasing ion energy,

and so more energy is retained deeper into the feature upon grazing collisions with sidewalls. On

164
the other hand, for otherwise constant, non-rate limiting conditions, etch rates increase linearly

with increases in ion flux.

Figure 5.17: Predictions for etch profiles in SiO2 at constant power with fundamental frequency f0 = 10 MHz and
varying phase angles φ.

While certainly an important parameter, ion energy is not the sole determining factor for

etch rate. The etch process is based on a sensitive balance of surface passivation, activation and

removal by fluxes of neutrals, ions and hot neutrals (generated by ions neutralizing during surface

collisions). For each fundamental frequency, the ratio of these fluxes as function of φ is relatively

constant, as shown in Figure 5.15d. Although ion energy and fluxes determine overall rate of

165
etching, the shape of the feature (e.g., sidewall slope, bowing) depend on relative rates of

passivation by deposition, etching and sputtering. For example, for f0 = 10 MHz, in spite of the

higher etch rate with increasing phase angle, the features generally have less bowing, an effect that

may be attributable to a larger ratio of polymerizing flux to ion flux. Large fluxes of passivating

species typically produce more tapered features.

5.6 Concluding Remarks

Coupled reactor and feature scale simulations were performed to investigate the

consequences of the fundamental driving frequency f0 on the relation between DC self-bias Vdc,

incident ion energy onto the wafer and reactive fluxes in dielectric etch processes using tailored

voltage waveforms as a power source. For a set of waveforms based on consecutive harmonics for

which the relative phase angle φ was varied from 0 to 180°, gas phase simulations were performed

using fundamental frequencies f0 = 1, 2, 5 and 10 MHz. While plasma conditions were found to

differ when holding either voltage or power constant, the trends in Vdc, were qualitatively similar

for the two scenarios. The magnitude of Vdc decreases (a negative bias becoming more positive)

with increasing φ for all f0 at constant voltage as well as power. Due to ion inertia, in the high

frequency regime ion fluxes to the substrate are dominated by time average quantities such as the

Vdc self. At low f0, the ions are able to react transient characteristics in the sheath such as local

extremes in the sheath potential directly produced by the applied tailored voltage waveforms.

The consequences of inertial effects on IEADs as a function of sinusoidal bias frequency

are well known. The thick sheath regime corresponding to high frequency and large ion mass

produces a single peaked IEAD. The thick sheath regime corresponding to low frequency and

small mass produces a double peaked IEAD. The extension of these dependencies to VWT power

sources is not straight forward due to the intrinsically more complex sheath structure and harmonic

166
content. The results of this study suggest that scaling of Vdc produced by processes akin to electrical

asymmetry, a common goal sought when using voltage waveform tailoring, is particularly

sensitive to the frequency regime of f0. Low values of f0 can result in significantly different trends

of incident ion energy as a function of phase angle compared to high values of f0. Incident ion

energy is closely correlated to the Vdc self-bias at high values of f0, maximum at φ = 0° and

minimum at φ = 180°. For low values of f0, this correlation dissipates, and incident ion energy is

instead dominated by the sheath dynamics of the applied voltage waveforms, maximum at φ =

180° and minimum at φ = 0°.

The trends in ion energy (and IEADs) as a function of f0 and φ directly impact the HAR

etch process. Although the observations and conclusions made in this work are highly dependent

on power deposition, plasma density, chemical composition and geometry, these observations also

open additional avenues for process control. With 3-dimensional structures and atomic layer

resolution already dominating industrial plasma etching processes, additional control strategies are

required to achieve the desired critical dimensions. With frequency agile power supplies becoming

more available, the combination of using VWT while varying fundamental frequency and phase

angle provide new control opportunities.

167
5.7 References

[1] G. A. Skarphedinsson and J. T. Gudmundsson, Plasma Sources Sci Technol 29, 084004

(2020).

[2] S. Brandt, B. Berger, E. Schüngel, I. Korolov, A. Derzsi, B. Bruneau, E. Johnson, T. Lafleur,

D. O’Connell, M. Koepke, T. Gans, J. P. Booth, Z. Donkó and J. Schulze, Plasma Sources

Sci Technol 25, 17 (2016).

[3] P. Hartmann, L. Wang, K. Nösges, B. Berger, S. Wilczek, R. P. Brinkmann, T.

Mussenbrock, Z. Juhasz, Z. Donkó, A. Derzsi, E. Lee and J. Schulze, J Phys D Appl Phys

54, 255202 (2021).

[4] Z. Donkó, A. Derzsi, M. Vass, J. Schulze, E. Schuengel and S. Hamaguchi, Plasma Sources

Sci Technol 27, 104008 (2018).

[5] T. Lafleur, Plasma Sources Sci Technol 25, 013001 (2015).

[6] J. Wang, S. Dine, J.-P. Booth and E. V. Johnson, Journal of Vacuum Science & Technology

A: Vacuum, Surfaces, and Films 37, 021303 (2019).

[7] F. Schmidt, J. Schulze, E. Johnson, J. P. Booth, D. Keil, D. M. French, J. Trieschmann and

T. Mussenbrock, Plasma Sources Sci Technol 27, 095012 (2018).

[8] C. Qu, Y. Sakiyama, P. Agarwal and M. J. Kushner, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 39, 52403 (2021).

[9] S. J. Lanham, J. Polito, Z. Xiong, U. R. Kortshagen and M. J. Kushner, J Appl Phys 132,

73301 (2022).

[10] S. Huang, S. Shim, S. K. Nam and M. J. Kushner, Journal of Vacuum Science & Technology

A: Vacuum, Surfaces, and Films 38, 023001 (2020).

168
[11] S. Huang, C. Huard, S. Shim, S. K. Nam, I.-C. Song, S. Lu and M. J. Kushner, Journal of

Vacuum Science & Technology A 37, 031304 (2019).

[12] C. M. Huard, S. Sriraman, A. Paterson and M. J. Kushner, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 36, 06B101 (2018).

[13] J. Schulze, Z. Donkó, B. G. Heil, D. Luggenhölscher, T. Mussenbrock, R. P. Brinkmann

and U. Czarnetzki, J Phys D Appl Phys 41, 105214 (2008).

[14] B. G. Heil, U. Czarnetzki, R. P. Brinkmann and T. Mussenbrock, J Phys D Appl Phys 41,

165202 (2008).

[15] U. Czarnetzki, J. Schulze, E. Schüngel and Z. Donkó, Plasma Sources Sci Technol 20,

024010 (2011).

[16] J. Schulze, E. Schüngel, Z. Donkó and U. Czarnetzki, Plasma Sources Sci Technol 20,

15017 (2011).

[17] S. J. Doyle, A. R. Gibson, R. W. Boswell, C. Charles and J. P. Dedrick, Plasma Sources Sci

Technol 29, 124002 (2020).

[18] L. P. Beving, M. M. Hopkins and S. D. Baalrud, Plasma Sources Sci Technol 31, 084009

(2022).

[19] B. G. Heil, R. P. Brinkmann and U. Czarnetzki, J Phys D Appl Phys 41, 225208 (2008).

[20] T. Shirafuji and K. Denpoh, Jpn J Appl Phys 57, 06JG02 (2018).

[21] B. Mancinelli, L. Prevosto, J. C. Chamorro, F. O. Minotti and H. Kelly, Plasma Chemistry

and Plasma Processing 38, 147 (2018).

[22] O. Murillo, A. S. Mustafaev and V. S. Sukhomlinov, Technical Physics 64, 1308 (2019).

[23] M. S. Benilov and N. A. Almeida, Phys Plasmas 26, 123505 (2019).

169
[24] S. D. Baalrud, B. Scheiner, B. T. Yee, M. M. Hopkins and E. Barnat, Plasma Sources Sci

Technol 29, 053001 (2020).

[25] R. P. Brinkmann, J Phys D Appl Phys 42, 194009 (2009).

[26] R. P. Brinkmann, J Phys D Appl Phys 44, 042002 (2011).

[27] M. Klich, J. Löwer, S. Wilczek, T. Mussenbrock and R. P. Brinkmann, Plasma Sources Sci

Technol 31, 045014 (2022).

170
 Chapter 6 Autonomous Hybrid Optimization of a SiO2 Plasma Etching Mechanism

6.1 Introduction

The computational modeling of plasma etching processes is receiving new interest at least

in part due to its ability to aid in the design and understanding of semiconductor device

manufacturing and provide the basis for machine learning based optimization of processes [1–5].

In simulation of plasma etching, the framework typically includes a reactor scale model and a

feature scale model [6–8]. The reactor scale model provides reactive fluxes, including their energy

and angular distributions, that are used as input to the feature scale model [9][10]. There are several

classes of feature scale models – using level-set-methods [11–14], statistical voxel approaches

[15–17], and molecular dynamics (MD) [18,19]. Although MD methods are the most fundamental,

they are currently computationally limited in the size of the feature that can be simulated and by

availability of inter-particle potentials for complex chemistries. As a result, most full-feature scale

capable models employ level set or statistical voxel methods with, in most cases, less fundamental

reaction mechanisms.

Due to limits resulting from computational complexity and sometimes incomplete

understanding of the physical process involved in plasma etching, feature profile simulations

typically make use of simplified or reduce reaction mechanisms that use rate coefficients, reaction

probabilities, threshold energies and energy and angular scaling to represent reactions between gas

phase and surface species. The outcome of these reactions are the addition of material (deposition),

removal of material (etching or sputtering) and its modification (passivation, implantation)

[16,20,21]. The reaction mechanism and the coefficients and parameters that quantify the

171
mechanism are ideally based on either experimental observations or more complex computational

models that do not allow for an efficient real-time evaluation.

Rapidly and accurately producing reaction mechanisms for profile simulation for novel

materials, gas mixtures and temperature regimes, as in cryogenic etch processes [22–24], is

challenging. This is particularly the case in data starved regimes, and so there is a need to robustly

and efficiently adjust and improve reaction mechanisms. Classical optimization techniques as well

as machine learning (ML) based approaches is an area of active research [4,25–28].

Applying ML techniques for process design is an expanding area of research where the

increasing number of tunable parameters and complexity as well as highly non-linear behavior has

rendered exhaustive sampling and rudimentary ‘design of experiment’ methods ineffective.

Model-based feedback has proven effective at enabling precise, repeatable and stable process

control [29–31]. Here the same issues with respect to data availability and consistency apply and

the control models, too, are sometimes based on machine learning methods as well. This approach

to virtual metrology has proven fairly effective as the efficient models often allow for rapid real

time evaluation [32–36]. Other applications include the development of surrogate methods where

statistical models are trained on actual experimental data or synthetically computed data to

simulate the process as a whole or act as a sub model in a larger context [37,38], a practice

especially common in large scale simulations of fusion plasmas [39–45]. Ideally, these models are

capable of reproducing reactor and feature scale etch processes either directly as an outcome of

the surrogate model or by interpolating between known results [46,47]. Generally, these types of

approaches have underlying model structures that are valid in their training parameters space,

however the parameters that are the outcome of the underlying model do not provide physical

insight. Many of these data driven ML approaches are themselves fundamentally limited by the

172
availability or accuracy of the data used for training purposes. Producing this data is itself costly

and time consuming, while the resulting trained ML based model may have limited applicability

to process conditions outside the training set.

The challenge of needing large training sets of data has led to the use of physics informed

ML and optimizations techniques where a lack of data is, to some degree, mitigated by the explicit

use of physical formulations and models to reduce the degrees of freedom in the ML-derived

model. Using this approach, a reaction mechanism can be built using known forms of probabilities

of gas-surface interactions which may have poorly known coefficients. Using ML-methods to

derive the physics-based coefficients should, in principle, produce a reaction mechanism that

extends between the conditions of the training set.

These challenges specifically apply to the modeling of high aspect ratio (HAR) plasma

etch processes [48–51]. The aspect ratio (AR) is the ratio of depth and to the width, most often

given by the opening of a mask material. These processes are of relevance to the creation of vias,

contact holes and isolation trenches [50]. Features are now produced for 3D memory devices

having ARs approaching 100. Ideal HAR features have vertical side walls with widths replicating

the mask with a minimum of bowing [52] or tapering [53]. There are many process-dependent

challenges with make these outcomes difficult. For example, polymer deposition is a critical

process in dielectric plasma etching in fluorocarbon gas mixtures [54,55]. A critical process

challenge is the control polymer deposition in the upper regions of the mask during the etch

process, which may result in necking and clogging [46,56,57]. Necking is the narrowing of the

opening in the mask by polymer deposition. Clogging is closing of the mask opening by the

deposition. Both necking and clogging result from the same processes with different degrees of

severity. Clogging will stop etching while necking can limit neutral gas transport to the bottom,

173
trap etched surface species and shadow ions and photons into the feature. Modeling these

processes, critical to feature optimization, is additionally challenged by lack of fundamental data.

Physics informed ML approaches to process development have taken the form of compact

or reduced models. For example, through ML methods, reduced reaction mechanisms can be

generated [58–60]. In a reduced reaction mechanism, species or reactions for which the outcome

is weakly dependent are removed from a comprehensive mechanism, resulting in a model that is

computationally more efficient. Reduced reaction mechanisms, though more rapidly executing,

are generally valid over a limited range of operation, such as power, gas mixture or pressure.

The optimization of the reaction mechanism itself has been investigated previously by

similar means for simulations based on Monte Carlo approaches as well as simulations based on

cellular automatons [61–65]. Optimization schemes such as evolutionary algorithms and particle

swarm methods can require many evaluations of the model. This large number of evaluations can

lead to a large computational expense or needing to reduce the domain over which the optimization

is being performed, either spatially or in terms of interaction complexity (e.g. limiting the number

of species).

In this paper, we discuss the development and application of a semi-autonomous

optimization technique using a coupled gradient descent – Nelder Mead approach to derive

physics-based coefficients for reactions mechanisms used in voxel-based feature simulation of

plasma etching. The demonstration system is SiO2 HAR plasma etching in a fluorocarbon gas

mixture. The reactor scale simulations that provide reactant fluxes were performed with the Hybrid

Plasma Equipment Model (HPEM). The feature scale simulations were performed with the Monte

Carlo Feature Profile Model (MCFPM) described in Chapter 2.

A high-level overview of the optimization scheme is given in Section 6.2. The process to

174
be optimized as well as the simulation tools used for the gas phase and feature scale simulation are

discussed in Section 6.3 and 6.4. Metrics for the optimization process are defined in Section 6.5.

The optimization schemes are described in Sections 6.6 and 6.7 and the hybrid approach is

motivated. Results of the optimization combined optimization are discussed in Section 6.8. To test

the transferability of the derived mechanism, etch processes outside its original training regime

were simulated, with those results being presented in 6.9. The simultaneous optimization of

multiple features is investigated in Section 6.10. Concluding remarks are in Sec. 6.11.

6.2 Description of the Optimization Scheme

The goal of this work is to optimize the reaction mechanism that describes plasma-surface

interactions for SiO2 etching to best replicate experimental data, which is provided in the form of

scanning electron microscopy (SEM) images. The optimization is centered around adjusting

physics parameters, in the reaction mechanism, represented by scalar values pi, each representing

a physical quantity, contained in a parameter set

 p1 
 
p
p= 2 (6.1)
 
 
 pn 

The size of the parameter space scales like O(cn) making exhaustive search algorithms

impractical for even moderately large numbers of parameters. Enforcing constraints and relations

based on prior physical knowledge of the likely range of pi can help to reduce the total parameter

space and enable the use of conventional optimization techniques. This physical knowledge is, in

our case, implemented by using a well-established surface mechanism having poorly known

reaction probabilities, and applying physically reasonable boundaries to those poorly known

coefficients. The optimization scheme is based on a hybrid approach, combining a gradient descent

175
method [27,66,67] with a Nelder-Mead optimization scheme [68–70]. A top-level overview of the

method is shown in Figure 6.1.

Figure 6.1: Global schematic of the optimization process.

The optimization setup requires external input (located on the left-hand side of Figure 6.1)

which in the case are the reactant fluxes of radicals and ions, and their energy and angular

distributions (EADs), that are incident onto the wafer being processed. These fluxes and EADs

were provided by the HPEM using reactor conditions that produced the features that were imaged

by the SEMS. A legitimate issue is the sensitivity of the optimized feature scale reaction

mechanism to the accuracy of fluxes and EADs provided by the HPEM. Although this is an

important discussion, it is beyond the scope of this particular investigation which is focused on

developing the feature scale reaction mechanism. As such, the fluxes and EADs from the HPEM

are treated as the ground truth. Details of the experiment and gas phase simulation are provided in

176
Section 6.3.

The gas phase fluxes and EADS are fed into MCFPM which performs the etch process

simulations outlined in Section 6.4. The result of these simulations is compared to the features in

the SEM images. Based on a set of geometric metrics, discussed in Section 6.5, a model error is

determined. The data produced by the MCFPM and their differences to experimental data

effectively act as the loss function to be minimized. The model error is used as the basis for the

optimizer which iteratively adjusts the model parameters p to minimize the loss. This loop is

repeated until the system converges.

6.3 Reactor Scale Plasma Simulation

The reactor scale gas phase simulations were performed using the Hybrid Plasma

Equipment Model (HPEM), which is described in Section 2.1. The HPEM was used to model a

capacitively coupled plasma sustained in a C4F6/Ar/O2 gas mixture at 10 mTorr. In the base case

the total gas flows were C4F6/Ar/O2 = 140/100/105 sccm. The reactor configuration is shown in

Figure 6.2.

Figure 6.2: CCP reactor geometry and circuit.

177
 Two electrodes with a radius of 15 cm are separated by a 4 cm gap. Dual radio frequency

(RF) power VRF is applied to the bottom electrode. The voltage waveform consists of low and high

frequency components, Vrf(t) = Vlf sin(2πflft) + Vhf sin(2πfhft) with flf = 1 MHz and f hf = 40 MHz.

For the base case low frequency voltage, Vlf and high frequency voltage Vhf, were adjusted to

supply powers of Plf = 8.0 kW and Phf = 2.5 kW. A negative DC bias of -500 V was applied to the

top electrode, to deliver 650 W. The reaction mechanism is based on previous work [16,51].

178
 Figure 6.3: Plasma properties for the base case: a) Electron density [e], b) electron temperature Te and c) negative
fluorine ion Density [F-], d) ionization rate by bulk electrons, e) ionization rate by secondary electrons.

The spatial distributions of the electron density [e], electron temperature, T e, negative

fluorine ion density [F-], ionization by bulk electrons Sb, and ionization by secondary electrons Ss

are shown in Figure 6.3.

Due to the large power input, the electron density is high, with a maximum value of 9.5 x

1010 cm-3. The bulk electron temperature, shown for electron densities > 1.0  109 cm3, is 3.4 to

3.8 eV over the wafer. High energy electrons are initially produced by secondary electron emission

by ion bombardment, followed by acceleration in the sheaths. The electron source generated by

bulk electrons Sb, mirrors that of [e] and Te. Most of the ionization is produced by bulk processes.

Ionization produced by sheath accelerated beam electrons plays a secondary but nevertheless

important role in the overall ionization dynamics. Fluorine and oxygen containing plasmas can

have significant densities of negative ions. The density of F-, shown in Figure 6.3c, has the largest

density of negative ion in this process with a maximum density of 2.2 x 10 10 cm-3. Due to the

plasma having on the average, a positive electric potential, the F- ions are confined to the center of

the plasma. The negative ions have only a small overall effect on plasma transport dynamics to the

wafer due to its negligible density close to the sheath region where its effect on electron heating

dynamic would be the most important.

Plasma-surface interactions are almost entirely described by the fluxes of reactive species

and ions, and their energy and angular distributions onto the surface. The fluxes of the most

significant neutral and ion species incident onto the wafer at a radial position of 7.5 cm are listed

in Table 6.1.

179
 Table 6.1: Base case fluxes to wafer

Species Flux (cm-2s-1)

C3F4 9.5 x 1016
C2F3 6.8 x 1016
CF 4.4 x 1016
CF2 9.4 x 1016
CF3 8.4 x 1015
O 7.7 x 1016

These fluxes include the fluorocarbons radicals most responsible for polymer deposition

as well as atomic oxygen which etches and removes the polymer. The ratio of the flux of ions and

neutrals to the wafer is an important parameter in the overall process dynamics. The anisotropy of

the process is almost solely enabled by the directionality of ions incident onto the surface whereas

the fluxes of neutral fluorocarbon radicals determine passivation. This is especially true for HAR

processes in which the overall performance requires a precise balance of deposition and removal

of polymerizing radical fluxes, both of which are either indirectly or directly dependent on the ion

flux. neutral as well as ion fluxes.

In addition to the magnitude of the ion flux, the influence of energetic positive ion fluxes

incident onto the surface and their transport through HAR features is sensitive to their energy and

angular distribution (IEAD). The combined IEAD of all positive ion species is shown in Figure

6.4. Due to the large applied low frequency power, the sheath potential and commensurate DC

self-bias, positive ions are accelerated vertically into the wafer while traversing the sheath. This

leads to a narrow angular distribution of ions striking the wafer with energies up to 4800 eV, both

properties being desirable in the context of HAR etch processes.

180
 Figure 6.4: Ion Energy-Angular Distribution sampled at the wafer surface during the base case.

6.4 Surface Simulation Setup

The process investigated in this work is a HAR plasma etch into SiO2 using an amorphous

carbon (AC) mask, schematically shown in Figure 6.5. A SiO2 substrate is covered by a 850 nm

thick AC film patterned to contain an ideal, straight walled, opening with an initial width of 90

nm. The etch was performed for 60 seconds.

181
 Figure 6.5: Feature scale simulation setup consisting of a SiO2 substrate with an amorphous carbon mask.

The surface simulations were performed using the Monte Carlo Feature Scale Model

(MCFPM) [16,50,74,75]. MCFPM is described in Section 2.2. The SiO2 reaction mechanism is

based on previous work by Huang et al [51] and is described in detail in [16]. A high-level overview

of the mechanism is listed in Table 6.2. SiO2 can be removed through physical sputtering by

energetic ions and hot neutrals. The sputtered products can be redeposited on other surfaces.

Unsaturated fluorocarbons can chemisorb on the SiO2 to form an oxide -fluorocarbon complex.

This complex is in turn easier to sputter based on a modified threshold (reduced total binding

energy) and an overall higher reaction probability as the site has a lower binding energy. Additional

182
polymer can be deposited on top of the complex as well as other surfaces, providing a physical

barrier to further etching and providing sidewall passivation. The polymer can be removed by

physical sputtering or chemical etching by oxygen radicals. The only O-containing radical having

significant fluxes to the surface is ground state atomic oxygen.

The gas phase reactions with the mask and substrate are ultimately contained in a single

global reaction mechanism where they interact and share certain reactions, such as polymer

deposition. A high-level overview of the AC mask reaction mechanism is in Table 2 and the full

mechanism is described in Appendix B. The mechanism includes polymer deposition by CxFy

precursors on AC and on prior deposited polymer. Similar to the etch mechanism for SiO2, polymer

is removed by oxygen radicals and physical sputtering. The AC can be sputtered by ions and hot

neutrals as well.

The polymer, as deposited, consists of individual CxFy radicals and so is strictly an

assembly of monomers. This radical based film can subsequently crosslink to create an actual

polymeric material that is more resistant to sputtering. Ion bombardment can then break bonds

(chain scission) to produce lower molecular weight polymer.

A polymer’s physical and chemical properties can be dependent on crosslinks with

neighboring species. The manner of linkage determines the reactivity of the polymer due to, for

example, a lack of available radical sites for or more resistance to physical processes due to an

increase in total bond strength. These crosslinks can be broken by exposure to energetic particles

or radiation, for example through exposure to plasma. Since these energetic particles are typically

delivered anisotropically to the surface, this spatially discriminate activation can result in shaping

of the polymer deposition. The developed crosslinking mechanism used in this work is described

in Section 2.2.3.

183
 Since the control of necking and clogging by polymer deposition is of special interest, we

discuss it as an important control parameter. The amount of polymer growth is determined by the

ratio of deposition and removal. A steady state polymer thickness occurs when these contributions

balance. Polymer removal occurs by sputtering and O-radical based etching. In oxygen rich gas

mixtures, polymer removal can be dominated by O-radical based etching, mostly ground state

atomic oxygen.

Figure 6.6: Etch features for different O based polymer etch probabilities 0.5% (left) and 2.0% (right).

184
 In the mechanism discussed here, necking and clogging (the amount of polymer deposition

in the mask region) can ultimately be controlled by the reaction probability of the O based polymer

etch. The final etch profiles for otherwise identical process conditions are shown in Figure 6.6

while varying the probability of polymer etching by O-atoms. An etch probability of 0.5% results

in net polymer growth which ultimately leads to a complete clog at the top of the feature. Higher

removal probabilities, while still producing significant necking, does not fully clog the feature and

allows for continued etching throughout the entire process.

Table 6.2: High-level description of the SiO2 and AC etch mechanism.

Species
M Ion or hot neutral AC Amorphous carbon mask
S Surface Site (g) Gas phase
P Polymer (s) Solid
PC Crosslinked polymer

Reaction Description
SiO2(s) + M(g) → SiO2(g) + M(g) Physical sputtering of SiO2

SiO2(g) + S(s) → SiO2(s) + S(s) SiO2 redeposition

SiO2(s) + CxFy(g) → SiO2CxFy(s) SiO2 -Fluorocarbon complex formation

SiO2CxFy(s) + M(g) → SiO2CxFy(g) + M(g) Complex sputtering

S(s) + CxFy(g) → S(s) + P(s) Polymer deposition on Surfaces

P(s) + CxFy(g) → P(s) + P(s) Polymer deposition on Polymer
P(s) + O(g) → COFy(g) Polymer etch by O
AC(s) + CxFy(g) → AC(s)+ P(s) Polymer deposition on AC
P(s) + M(g) → P(g) + M(g) Polymer sputtering
AC(s) + M(g) → AC(g) + M(g) AC sputtering

P(s) + P(s) → PC(s) + PC(s) Crosslinking

P(s) + P(s) + M(g) → P(s) + P(s) + M(g) Breaking of Crosslinking

185
6.5 Target Metrics and Loss Function

A scalar metric is needed to evaluate the quality of the simulation-experiment matching

when tuning the reaction mechanism. In this work the loss function L(p) of a given parameter set

p is the RMS error between the simulation and experiment of geometric measures, or shape

parameters, of the shape of the etched feature, shown in Figure 6.7.

Figure 6.7: Target metrics used to evaluate the experimental matching.

While these shape parameters are not independent of each other, they nevertheless each

aim to represent a specific characteristic of the feature that is of technical relevance. The width of

the mask opening wm including deposition stands as a measure of the necking and clogging, a

186
process which is dependent on the ration of the fluxes of polymer depositing to removing species.

The narrowed opening can impede neutral gas transport into the feature, trap etch products inside

the feature and shadow the trajectories ions and photons. Although the shape (curvature) of the

polymer deposition is not a shape parameter in this investigation, the curvature of the polymer

affects the angle of the trajectory with which ions (hot neutrals) reflect from its surface deeper into

the surface. An improved approach might include shape (curvature) in the optimization process.

The width at the top of the feature, wt, aims to capture undercutting (etching under the

mask) that mostly occurs with excessive isotropic etching of the substrate or with there being

insufficient passivation of the surface of the substrate compared to that of the mask material. The

maximum width of the feature wf, captures the occurrence of bowing, which can be caused by lack

of passivation or broad ion angular distributions (bowing).

Mask selectivity refers to the relative rate of etching of the mask compared to the substrate.

Masks for HAR etching must have high selectivity (low etch rate) compared to the substrate so

that that the mask has a reasonable small thickness. The height of the mask hm (height is thickness

of the mask plus thickness of the substrate) must be controlled to ensure that the mask has a critical

thickness at the end of the etch. The thickness of the mask determines the degree to which ions

having broad angular spread are shadowed by and reflect off the mask prior to entering the feature.

The depth of the final feature, hf, fundamentally defines the characteristics of HAR features

and for a given process time determines the etch rate. Undesired etch defects such as twisting [76]

can be captured by measuring the asymmetry, ah, across the vertical center axis (dotted vertical

line in Figure 6.7.

The loss function is then

L( p) = awm ( wˆ m − wm )2 + awt ( wˆ t − wt ) 2 + awf ( wˆ f − w f ) 2 + ahm (hˆm − hm ) 2 + ahf (hˆ f − h f )2 + aah (aˆh − ah ) 2 (6.2)

187
where the hat accent (^) denotes the target value of a property. Each singular error component can

be weighted in accordance with its importance or sensitivity by the respective weight factors ai. In

this work all components are equally weighted with the exception of the total asymmetry which

has a higher weighting. The total asymmetry is computed based on the total number of asymmetric

cells which has a substantially different scale.

For any given evaluation of the loss function L( p) , MCFPM has to be executed to simulate

a full etch, which carries with it a computational cost. The method of evaluating and minimizing

L( p) through the optimization algorithm should have the goal of minimizing the number of

evaluations. In this work, the target metrics are based on the scanning electron microscopy (SEM)

image, shown in Figure 6.8. This image was produced after the process described in Section 6.3

was performed in a reactor that on which the simulated geometry was based.

Figure 6.8: SEM image resulting from the base case experiment.

188
6.6 Gradient Descent

Gradient descent (GD) is a method for minimizing a multivariate error function, which is

convenient due to its conceptual simplicity, universality and fast convergence. The GD method

has been applied to a wide variety of physical and engineering problems [66,67,77–81]. In short,

GD is based on following the gradient of a loss function L to its minimum as shown in Figure 6.9.

Figure 6.9: Outline of the gradient descent method for a 1D parameter space.

189
 For a given set of parameters pi the selection of the next set of parameters be expressed as

pi +1 = pi − L(pi ) , (6.3)

where η is the learning rate which controls the effective step size and the rate of descent. The loss

function consists of a set of discrete points that results from successive runs of the MCFPM and

so is not natively differentiable. A finite difference scheme was used to estimate the partial

gradients in every dimension by evaluating L(p) at a test location L(p′)

L(pj ) = L(p + x j pj ) (6.4)

for every dimension of the parameter space n and computing the relative delta ΔL/ Δpj. Effectively,

the discrete gradient descent formulation becomes

n
L n
L(pi ) − L(pi )
pi +1 = pi −   x j = pi −   x j (6.5)
j =1 p j j =1 pij − pji

A useful modification to the gradient descent method is the addition of a momentum term,

that enables the method to overcome local extrema and avoid convergence around a non-global

minimum. Analogous to the momentum in Newtonian physics, momentum in GD adds an inertia

term to the updated parameters. The previous gradient is remembered and the update on the next

iteration is determined by the effective gradient gi which is a linear combination of the actual

gradient and the previous update:

pi +1 = pi − gi (6.6)

with

gi =  gi −1 + L(pi ) (6.7)

where δ is the momentum decay coefficient which determines the overall strength of the

momentum term. Akin to a heavy ball, rolling down a gravitational potential (referred to as heavy

190
ball method) the gradient descent is less affected by small scale perturbation or “roughness” of the

loss function surface.

Based on this methodology, an initial optimization was performed. The tuning parameters as well

as their intervals are listed in Table 6.3. The target metrics obtained from experimental data are

listed in Table 6.4. The optimization was performed for a total of 100 epochs. (An epoch is single

round of evaluating the loss function with a given set of parameters.)

Table 6.3: Tuning parameters for gradient descent optimization

Symbol Description Min. value Max. value

ps,SiO2 Sputter probability of SiO2 0.0 0.3
ps,SiO2CFXY Sputter probability of SiO2-polymer complex 0.1 0.5
pp,SiO2 SiO2-polymer-complex formation probability 0.1 0.5
pe,poly O based polymer etch 0.0 0.5
pd,poly-AC Polymer deposition probability on mask 0.0 0.5

Table 6.4:Target metrics for gradient descent optimization

Symbol Description Target value

wm Width of mask opening 45 nm
wt Width at the top of the feature 90 nm
wf Maximum width of the feature 90 nm
hf Etch depth 825 nm
hm Remaining mask height 850 nm
ah Asymmetry 0
The evolution of the total loss function is shown in Figure 6.10a) as a function of elapsed

epochs. The total loss quickly decreased during the first 20 epochs which indicates a partially

successful optimization. This convergence mirrored by the evolution of the target metrics, shown

in Figure 6.10b). The three metrics, etch depth hf (blue squares), minimum mask opening wm (red

pyramid) and maximum feature width wf (green delta) quickly approach their target values,

represented by dashed horizontal lines. Some of the trends are correlated to the underlying

adjustments to the parameters defining the etching mechanism, a selection of which are shown in

Figure 6.10c).

191
Figure 6.10: Results from the gradient descent optimization. a) minimal epochal loss. b) values of select metrics as a
function of epochs. c) select tuning parameters as a function of epochs.

The increase in final predicted etch depth, for example, can be explained by changes to the

complex formation probability and its sputter probability both of which increase in tandem with

the increase in final etch depth. However, beyond epoch 20, however, the clear trend towards error

192
minimization breaks down and instabilities both in the model error as well as in the metrics emerge.

This breakdown results from the underlying definition of the loss function.

The step size with which the parameter space is traversed is inherently tied to the size of

the gradient (Eq. (6.6)) and should diminish as one approaches the minimum (zero gradient). L(p)

is derived from predicted profiles produced by the MCFPM which has finite grid resolution – the

dimension of the voxel. As the step size diminishes, producing a small change in the physics

coefficients, the physical change in the predicted profile may not exceed the size of the voxel. At

this point, the loss function is no longer differentiable, as shown in Figure 6.11.

Figure 6.11: The finite resolution of the loss function leads to convergence issues.

This condition can lead to an absence of a gradient when the change in the profile stays

within a single voxel, producing a breakdown of the functional relation between p and L(p) = L =

constant. Another outcome is an overestimation of the gradient when a small change p causes the

previous static profile t change to a different voxel.

The GD method provides rapid initial convergence but is potentially erratic when

approaching the optimum. The behavior of the GD and control of the speed of the initial descent

193
speed is a function of the learning rate η. The evolution of L(p) for different learning rates η =

1x10-4, 5x10-4, 1x10-3 and 5x10-3 is shown in Figure 6.12.

Figure 6.12: Evolution of the total loss for different learning rates η = 1x10-4, 5x10-4, 1x10-3 and 5x10-3.

The rate of descent is correlated to η – a smaller learning rate leads to smaller step sizes

used to traverse across the parameter space and, consequently, a slower overall descent. Due to the

limitations of the method that result from the erratic differentiation near convergence (an outcome

of the finite voxel size), all optimizations eventually result in a convergence failure, independent

of the learning rate.

Despite the limitations imposed by the finite difference approach, the GD method has the

ability to overcome local minima (to some degree) while having rapid and controllable initial

descent. This rapid initial convergence is particularly important in minimizing the number of loss

function evaluations and so minimizing computational cost). As a result, one is motivated to retain

the GD method to approach the optimized values, followed by a more robust, though possible more

slowly converging, optimization method. In this work the Nelder-Mead method was selected to

perform this task.

194
6.7 Nelder-Mead Optimization

The Nelder-Mead method is a gradient free optimization algorithm that has been used in a

wide variety of optimization problems [82–88]. Conceptually, the Nelder-Mead (NM) method

represents the tuning parameters as a point in n-dimensional solution space, where n is the number

of tuning parameters to be optimized. A simplex, an object with n+1 vertices, each representing a

different set of parameters, is employed to traverse the solution space through geometric

transformations along its centroid x0.

Figure 6.13: Examples of the geometric transformations of the parameter simplex in a 2D parameter space;
reflection, extension; contraction; shrinkage.

An example scenario is shown in Figure 6.13 with n = 2 – a 2D solution space where the

simplex is a triangle:

195
 • Sorting: The points are sorted according to their loss function from best (lowest loss) to
worst (highest loss): L(u) < L(v) < L(w)
• Reflection: Reflect the worst point, w, across the centroid; r = x0 + α(x0 - w), with α > 0.
• Extension: Extend the reflection beyond point r; e = x0 + γ(r - x0), with γ > 1.
• Contraction: Generate two points, c0 and c1, that represent contractions of r towards w;
c0 = x0 + ρ(r - x0) and c0 = x0 + ρ(w - x0) with 0 < ρ < 0.5
• Shrinkage: Shrink the simplex towards the best point; w’ = u + σ(w - u) and v’ = u + σ(v
- u), with 0 < σ < 1

Using these simple geometric transformations, the NM algorithm consists of a specific set

of decisions, which leads to an iterative loop that requires at least one successive evaluation of the

loss function per step and a maximum of 4 per epoch. A flow chart of the Nelder-Mead algorithm

is shown in Figure 6.14. Every epoch begins by sorting the simplex points according to their loss

function value, with L(u) < L(v) < L(w). Reflection is always the first geometric operation to be

tested. If the reflected point, r, is better than the previous second point but worse than the best (

L(u) < L(r) < L(v) ), the reflected point replaces the previously worst point ( w = r ) and the epoch

ends. If r is better than the previously most preferred point ( L(r) < L(u) ) the extension point e is

tested and added to the simplex if it outperforms r ( L(e) < L(r) ). If the reflected point r is worse

than all other simplex points, the contraction is performed. If neither c0 nor c1 have a lower loss

than r, the simplex is shrunk and the next epoch is initiated. This loop continues until the maximum

number of iterations is reached or the specified convergence has been reached. Since the individual

parameters represent physical quantities, additional constraints can be placed on the values of the

parameters. For example, sputter yields less than zero or reaction probabilities greater than one

can be accounted for by limiting respective transformations to valid ranges.

196
 Figure 6.14: Flow chart describing the Nelder-Mead optimization algorithm.

Since NM is a direct search method (only relies on direct evaluations of the loss function

itself), NM does not suffer from the same issues that arise with GD which relies on discrete

differentiation. NM is regarded as heuristic optimization in this context. Although convergence is

197
technically not formally guaranteed, from a practical perspective NM does nearly always

converge. This convergence results from the finite grid size. The loss function effectively becomes

insensitive to perturbations below a certain threshold. Once the optimization process switches from

GD to NM, no instabilities or non-stationary convergences were encountered.

The use of NM as the only convergence method is somewhat vulnerable to the presence of

local minima. That is, NM may converge around a non-global minimum. This propensity for local

convergence increases for higher order problems as the size of the parameter space increases. This

tendency for local convergence is greatly reduced by initializing the first simplex in the vicinity of

the absolute minimum as provided by the initial GD method. That is, the first simplex is not

initialized at a random location. As a result, the NM portion of the optimization does not need to

sample complete coverage of the parameter space nor conduct an exhaustive search has to be

required. For our conditions, the sequential GD-NM approach minimized the computational load

by reducing the number of loss function evaluations compared to random scatter initialization or

swarm optimization methods.[89]

6.8 Hybrid GD-NM Optimization

Subsequently executing GD and NM optimization algorithms produces a hybrid scheme

whose goal is to address the weakness of each singular approach while capturing their positive

qualities. The convergence and output parameters produced by the hybrid optimization scheme are

shown in Figure 6.15 as a function of epoch for the same conditions as Figure 6.10 a) minimum

epochal error, b) selected metrics and c) evolution of the tuning parameters.

198
Figure 6.15: Results from the combined gradient descent and Nelder-Mead optimization. a) minimal epochal loss. b)
values of select metrics as a function of epochs. c) select model parameters as a function of epochs.

199
 The evolution of the total model error has 3 stages. The first stage, I, consists of the initial

rapid descent and nearly monotonic reduction of the loss function, enabled by the gradient descent

with an appropriate learning rate. Starting at approximately epoch 20, stage II suffers from

convergence instabilities which is reflected in fluctuations in the target metrics (Figure 6.15b) and

model parameters (Figure 6.15c). At epoch 100, stage III, the gradient descent is terminated and

based on its results, the NM algorithm is initiated. The fluctuations are rapidly dampened, and a

stable convergence commences. The error drops to approximately 3x10 -3 which given the grid

resolution is on the order of a single cell or less. This minimization in error is mirrored in the target

metrics (Figure 6.15b) which converge to the desired values. The final parameter set is listed in

Table 6.5.

Table 6.5: Final tuning parameters after coupled optimization.

Tuning Parameter Value

ps,SiO2 0.0852
ps,SiO2CFXY 0.1471
pp,SiO2 0.278
pe,poly 0.0423
pd,poly-AC 0.094

The definition of model physics parameters based on this optimizer algorithm is based on

the reduced scalar representation of the feature. The method will not capture effects that are not

included in the target metrics. For example, select features produced by MCFPM after the full etch

process are shown in Figure 6.16 as a function of epochal progression. (These features are

predictions at the end of the full etch period for a particular set of model parameters corresponding

to an epoch. The sequence of features is not the temporal evolution of the feature itself). The SEM

from which the target metrics were extracted is also shown (grayscale).

200
Figure 6.16: Final features after etch completion for different parameter sets as a function of epochal evolution and
the actual experimental target feature.

Here, too, the optimization stages can be identified. During epochs 0 – 20, belonging to

stage I (initial descent), the monotonic behavior is reflected in the increasing etch depth and

widening of the necking. The increased etch rate directly correlates with increasing oxide-polymer

complex formation, its removal and increases in the SiO2 sputter rate. The decreased polymer

deposition at the top of the mask is a consequence of the increasing polymer etch by oxygen.

Stage II (instabilities) produces erratic behavior in the predicted profiles with an absence

of clear trends. Between epochs 20 and 100 the etch depth and polymer deposition vary seemingly

randomly without clear trends. With stage III and the change to the Nelder-Mead algorithm for

epochs 100-200 mechanism converges, with a lack of non-monotonic evolution of the metrics.

201
Overall, the final simulated etch feature at epoch 200 is in good agreement with the experimental

counterpart in terms of the metrics used in the optimization process. There are, however, some key

differences between the predicted profiles and the experimental SEM. For example, there are

differences in the vertical position of the minimum in necking and the taper of the mask. The

mismatch in these properties is due to the fact that they were not part of the optimization metrics.

Assuming that the optimization process produced good model parameters, these differences

indicate might that a) the mechanism lacks processes that would otherwise determine necking

location or taper, b) the non-optimized physics parameters are not accurate and/or c) the solution

is not unique and a second solution might better capture these phenomena. These results stress the

importance of including metrics that address the most critical properties of the features in the

optimization process, as well as the physics in the reaction mechanism that correlate to those

properties. For example, taper of the feature is known to be sensitive to the chemical sputtering

probability as a function of angle of incidence of energetic particles. This physics parameter was

included in our mechanism but was not part of the optimization process.

In this context, metrics and the optimization processes are as a whole coupled, requiring at

least a partial simultaneous evaluation. That is, independently optimizing each metric will not

produce the desired result. This coupling can be demonstrated by the relationship between necking

and clogging and feature evolution. Necking or, in the extreme, clogging reduces or entirely

impedes the transport of reactive species and ions into the feature, slowing the etch process. In the

event of clogging, the etch process is stopped which prevents evaluation and optimization of the

etch process as a whole. Since in this work metrics are only derived from the final feature, there is

no mechanism to differentiate between two fully clogged features based on the mask metrics alone.

The width of mask opening is wm = 0 in both cases even if the rate of the deposition were different

202
and the clogging occurred at different times during the etch. However, differences in rates of, for

example, deposition can affect the time during which the feature is not clogged, which translates

to differences in total etch depth. Through this mechanism, the etch depth acts as a secondary

metric for the clogging mechanism, without which prior attempts to optimize the isolated mask

mechanism failed.

Overall, the hybrid optimization scheme performed satisfactory in that it produced a

mechanism that reproduces the desired feature with high accuracy. However, aspects of the process

have potential for improvement. The epoch at which switching between GD and NM methods

should be carefully chosen. Epochs spent in Stage II produce little additional information and are

not a good use of computing resources. Such hybrid optimization schemes should include a

mechanism or criterion to identify convergence failure of the GD and automatically switch to the

secondary algorithm. A careful choice of the needed convergence would also minimize the number

of epochs spent in Stage III.

6.9 Transferability of the Converged Mechanism

A test of whether the derived model parameters are physically relevant is to evaluate their

transferability to related processes. Given the small sample set, overfitting or circumstantial

unphysical results are of major concern. The term related processes was used deliberately. The

parameters that have been derived from the optimization process are physics parameters that

should, in principle, apply to a wide range of process conditions. However, there is a practical

limit. For example, introduction of new gases or new method of excitation (inductively coupled vs

capacitively coupled) would likely exceed that limit or require additional parameters in the

optimization process.

203
6.9.1 Variation of O2 in Flow

To test the applicability of the derived model parameters, a process with different input

power (Plf = 6.0 kW and Phf = 2.5 kW) was simulated with differing oxygen to fluorocarbon

feedstock gas ratios (O2/C4F6 = 0.5, 1, 1.5 and 2.5). All other conditions were kept the same as the

base case. The most significant changes compared to the base case were in the particle fluxes

incident on the wafer surface, which are shown as function of the feed gas ratio in Figure 6.17.

Figure 6.17: Fluxes incident on the wafer as a function of the O2/C4F6 feedstock gas ratio.

The variation in the O2/C4F6 ratio was achieved, experimentally and in the simulation, by

adjusting the oxygen inflow only. As a consequence, the fluxes of the major contributing

fluorocarbons remain nearly constant. The O flux increases from 4.1 x 1016 to 1.5 x 1017 cm-2s-1, a

factor of 3.6, with increase of the oxygen inflow by factor of 5. These trends indicate that mole

fractions of feedstock gases do not translate one-to-one to the surface fluxes. The features

generated by MCFPM and their experimental counterparts are shown in Figure 6.18.

204
Figure 6.18: a) MCFPM features resulting from the O2/C4F6 variation. b) SEM images of the features resulting from
the corresponding experiment.

205
 Although the reaction mechanism used in MCFPM was not specifically tuned for these

process conditions, relevant trends were qualitatively reproduced qualitatively. Trends in polymer

deposition on the mask are reproduced. At low O2 fraction and low O-flux, the O based etching of

deposited polymer is low which leads to full clogging of the feature in both the experiment and

simulation. With increasing O2 the rate at which polymer is removed increases and the necking is

reduced, being essentially absent for the O2/C4F6 = 2.5 case.

The lack of polymer film on the mask leads to increased mask erosion with increasing O 2

inflow. The amount of mask erosion in the simulated features is not in quantitative agreement with

the experimental results. This lack of quantitative agreement is likely due to not having the

necessary physical processes in the mechanism despite not being specifically required for the

initial optimization process. For example, direct oxidation of the AC mask by O2 was not included

in the mechanism but may become important at larger O2 flow rates. The etch depth is related to

the mask necking through reduction in particle transport into the feature and through etch stop with

full clogging. The full clogging of the feature for O2/C4F6 = 0.5 is reproduced by the simulation,

however, etch depth for the fully clogged feature is not reproduced as this depth depends on when

the feature was clogged. The experiments show maximum etch depth for O2/C4F6 = 1.5, which is

not the case for the simulation. Overall, however, the predicted trends generally agree with

experiments, which is not necessarily the expectation given the fact that the mechanism was tuned

for only a single different process. The emergence of these untrained trends is a consequence of

using a ‘physics informed’ optimization approach where the derived parameters are physics

parameters and not simply fitting parameters. These trends also rely on the fluxes produced by the

HPEM being truth and accurately representing experimental fluxes, which adds additional

variability to the process.

206
6.9.2 Variation of Low Frequency Power

As second test of the model parameters derived from the optimization process was

prediction of feature profiles while varying the low frequency power, P lf = 0, 4, 6 and 8 kW.

Low frequency power is often used to control the ion energy at the wafer surface. IEADs onto the

wafer are a first-order concern in HAR etching.

Figure 6.19: IEADs resulting from the Plf variation with Plf = 0, 4, 6 and 8 kW.

The IEADs produced by HPEM for this range of Plf are shown in Figure 6.19 with there

being direct correlation between Plf and the mean and maximum energies of ions at the wafer.

While the IEAD for Plf = 0 kW has the lowest energies, with electrons exceeding 500 eV the high

frequency RF power alone nevertheless produces high energy ions. The maximum angular spread

has little variation as a function of Plf, with the exception of Plf = 0 kW which has significant

207
broadening of the angular distribution. Using the derived model parameters, MCFPM simulations

were performed for the different values of Plf = 0. Those results and the experimental SEM

references are shown in Figure 6.20.

Figure 6.20: a) MCFPM features resulting from the Plf variation. b) SEM images of the features resulting from the
corresponding experiment.

208
 The profiles, experiment and simulation, for Plf = 0 kW produce total clogging of the mask

opening, indicating that ion energy plays an important role in removing excess polymer. The

remaining cases (Plf = 4, 6 and 8 kW) have unclogged features and full etching with unexpected

little variation as a function of LF power. In the experimental data a doubling of the P lf (4 kW to

8 kW) produces few differences in the final features. These trends indicate that above a certain

threshold energy the etch progression and the mask removal process are not ion starved, but rather

limited by neutral gas transport. To some degree this trend is reproduced by the simulations where

etch depth does increase with increasing low frequency power however the rate of increase is

substantially sublinear. These outcomes indicate that the effect of ion energy (for example in

sputter yield or related processes) might be overestimated in the mechanism. Again, this assumes

that the results produced by the HPEM are truth.

6.10 Optimization Based on Multiple Features

Because Monte Carlo based simulations are inherently subject to statistical variations, the

fact that the loss function is based on a single feature’s geometry can result in overfitting and

misinterpretation of the target metrics. To partially mitigate and roughly estimate the level of

overfitting, instead of matching to a single features etch profile, the optimization was performed

on a set of four features arranged in a grid as shown in Figure 6.21.

209
Figure 6.21: Feature scale simulation setup consisting of a SiO2 substrate with an amorphous carbon mask and four
separate features in a regular grid.

The individual features are Identical to those used earlier in this work and the distance

between them is 40 nm. Using this geometry, the total loss is set to be the average of the individual

features. Because the particles launched from the top of the simulation domain are launched based

on a shared random number pool, this effectively induces statistical variation between the features.

Apart from the changes in feature geometry, the optimization setup remains unchanged. Analogous

to the results depicted in Section 6.8, the epochal evolution of the loss function is depicted in

Figure 6.22a. As with the single feature optimization, 3 distinct stages in the optimization can be

identified. The initial descent (Stage I) seems to commence more rapidly, despite identical learning

rates. This might be the result of the result of the induced statistical variation that, to a degree,

remedies some of the initial overfitting to statistical noise and enables a smoother gradient descent.

While the convergence failure (Stage II) still occurs, the random oscillation of the loss seems to

210
be significantly dampened compared to Figure 6.15a. This is likely due to the fact that the incorrect

gradient calculations are partially negated, as they are unlikely to occur in all features at the same

time. The final convergence (Stage III), too, is similar to the single feature optimization in that it

rapidly converges to a stable value and although the final loss is slightly larger, it is still on the

order of a single cell. This behavior is confirmed by considering the average of the metrics. Shown

in Figure 6.22b. All considered metrics reach very good final parity with the respective targets

while their initial approach is faster and the intermediate non-convergence oscillations are less

erratic and of smaller magnitude. The same set of tuning parameters is shown in Figure 6.22c. and

while their temporal evolution is very different from that observed in the single feature case the

final values are very close. The final values of the feature grid-based simulation are listed in Table

6.6.

Table 6.6: Final tuning parameters after coupled optimization of feature grid.

Tuning Parameter Value

ps,SiO2 0.0909
ps,SiO2CFXY 0.1384
pp,SiO2 0.2729
pe,poly 0.0628
pd,poly-AC 0.0842

The features produced as a function of epoch are depicted in Figure 6.23. As with the single

feature optimization the initial mechanism produces clogged features, caused by insufficient

polymer etching, which leads to an early etch stop. With progressing epochs (left to right in Figure

6.23), the net polymer deposition is reduced and the etch rate and etch depth increased. At epoch

100, immediately before stage III, the scalar metrics (Figure 6.22b) and 2d feature representation

both indicate a good match in etch depth but excessive feature bowing that is effectively removed

with subsequent ongoing optimization.

211
Figure 6.22: Results of the feature array optimization. a) minimal epochal loss. b) values of select metrics as a
function of epochs. c) select model parameters as a function of epochs.

212
Figure 6.23: Final feature grids after etch completion for different parameter sets as a function of epochal evolution
and the actual experimental target feature.

While these findings indicate that the optimization based on feature grids is generally

preferable to that using only a single feature, these improvements must be weighed against the

increase in computational cost that is incurred by the larger geometry and subsequent increase in

total cells. A detailed investigation of this relation could be of considerable interest but is beyond

the scope of the presented work.

213
6.11 Concluding Remarks

This chapter presented a two-stage hybrid optimization scheme aimed at optimizing

physical parameters in a reaction mechanism that is used in profile simulations of a high aspect

ratio etch into a SiO2 substrate using a C4F6/Ar/O2 gas mixture. The process is performed in a dual

frequency capacitively coupled plasma reactor with independently controlled high frequency and

low frequency powers applied to the bottom electrode. The employed optimization scheme

consisted of consecutive execution of a finite difference gradient descent and Nelder-Mead

algorithm. Doing so allowed taking advantage of some of the respective desirable qualities while

at least partially negating some of the respective drawbacks. The fast, initial descent of the GD

method was maintained while its inevitable instability was dampened by the successive NM

process. The mechanism was matched to best represent geometric properties of the final etched

feature extracted from scanning electron microscopy images. Using appropriate target metrics, the

optimization produced a mechanism that was able to reproduce the experimental reference with

sub 1% error, based on the employed metrics.

Outside the training parameter space using different gas compositions and power, the

mechanism captured relevant trends. In a variation of the O2/C4F6 ratio it reproduced feature

clogging onsets and, to a lesser degree, necking and mask erosion trends. Similarly, a variation of

the low frequency power produced a comparable onset of clogging at low enough powers which

was also captured by the mechanism. Trends in feature etch rate (or lack thereof) are reproduced

to a lesser degree.

Overall, the presented approach shows promising potential in that with a very

limited set of data, just a single SEM image, it was able to reproduce a good match to the training

target and was able to reproduce some relevant trends outside of its immediate training regime.

214
This is thanks to the ‘physics informed’ optimization approach which, in this given scenario,

employed a base mechanism that already contained explicit formulations of the relevant physical

processes and was thus conceptually more universal than the singular case to which it was matched.

Despite the given encouraging results this should only be considered a proof of concept as

many details of the optimization scheme still require improvement. This includes a more effective

criterion for the algorithm switching as well as a robust formulation to detect convergence and

initiate an early stoppage to reduce unnecessary computing overhead. Another potentially

worthwhile improvement would be the expansion of the metrics to include not only a single

process condition but rather match to one or more parametric sweeps. Alternatively, the inclusion

of multiple different etch times per process condition could proof useful to include data about the

temporal feature evolution and, since the data set is effectively increased, partially mitigate

overfitting which is of great concern whenever only a small set of data is used.

215
6.12 References

[1] V. M. Donnelly and A. Kornblit, Journal of Vacuum Science & Technology A: Vacuum,

Surfaces, and Films 31, 050825 (2013).

[2] I. Adamovich, S. D. Baalrud, A. Bogaerts, P. J. Bruggeman, M. Cappelli, V. Colombo, U.

Czarnetzki, U. Ebert, J. G. Eden, P. Favia, D. B. Graves, S. Hamaguchi, G. Hieftje, M. Hori,

I. D. Kaganovich, U. Kortshagen, M. J. Kushner, N. J. Mason, S. Mazouffre, S. M. Thagard,

H. R. Metelmann, A. Mizuno, E. Moreau, A. B. Murphy, B. A. Niemira, G. S. Oehrlein, Z.

L. Petrovic, L. C. Pitchford, Y. K. Pu, S. Rauf, O. Sakai, S. Samukawa, S. Starikovskaia, J.

Tennyson, K. Terashima, M. M. Turner, M. C. M. Van De Sanden and A. Vardelle, Journal

of Physics D: Applied Physics 50, 323001 (2017).

[3] G. Packard, A. Rosenfeld, G. S. Oehrlein and S. Hamaguchi, Plasma Sources Science and

Technology 27, 023001 (2018).

[4] Y. Suzuki, S. Iwashita, T. Sato, H. Yonemichi, H. Moki and T. Moriya, 2018 International

Symposium on Semiconductor Manufacturing (ISSM), 2018-Decem, 1–4 (2018).

[5] R. Anirudh, R. Archibald, M. S. Asif, M. M. Becker, S. Benkadda, P. T. Bremer, R. H. S.

Bude, X. Zhang, et al., IEEE Transactions on Plasma Science 51, 1750 (2023).

[6] A. Sankaran and M. J. Kushner, Journal of Vacuum Science & Technology A 22, 1242

(2004).

[7] R. J. Hoekstra, M. J. Grapperhaus and M. J. Kushner, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 15, 1913 (1998).

[8] C. M. Huard, Y. Zhang, S. Sriraman, A. Paterson and M. J. Kushner, Journal of Vacuum

Science & Technology A: Vacuum, Surfaces, and Films 35, 05C301 (2017).

216
[9] T. Iwase, Y. Kamaji, S. Y. Kang, K. Koga, N. Kuboi, M. Nakamura, N. Negishi, T. Nozaki,

S. Nunomura, D. Ogawa, M. Omura, T. Shimizu, K. Shinoda, Y. Sonoda, H. Suzuki, K.

Takahashi, T. Tsutsumi, K. Yoshikawa, T. Ishijima and K. Ishikawa, Japanese Journal of

Applied Physics 58, SE0802 (2019).

[10] N. Kuboi, https://doi.org/10.1117/1.JMM.22.4.041502 22, 041502 (2023).

[11] J. A. Sethian and D. Adalsteinsson, IEEE Transactions on Semiconductor Manufacturing

10, 167 (1997).

[12] M. Chopra, S. Helpert, R. Verma, Z. Zhang, X. Zhu, al Meghali Chopra and R. Bonnecaze,

https://doi.org/10.1117/12.2297482 10588, 132 (2018).

[13] T. Shimada, T. Yagisawa and T. Makabe, Japanese Journal of Applied Physics, Part 2:

Letters 45, L132 (2006).

[14] Y. G. Yook, H. S. You, J. H. Park, W. S. Chang, D. C. Kwon, J. S. Yoon, K. H. Yoon, S.

S. Shin, D. H. Yu and Y. H. Im, Journal of Physics D: Applied Physics 55, 255202 (2022).

[15] H. Tsuda, Y. Takao, K. Eriguchi and K. Ono, Japanese Journal of Applied Physics 51,

08HC01 (2012).

[16] S. Huang, C. Huard, S. Shim, S. K. Nam, I.-C. Song, S. Lu and M. J. Kushner, Journal of

Vacuum Science & Technology A 37, 031304 (2019).

[17] C. M. Huard, Y. Zhang, S. Sriraman, A. Paterson, K. J. Kanarik and M. J. Kushner, Journal

of Vacuum Science & Technology A: Vacuum, Surfaces, and Films 35, 031306

(2017).

[18] H. Tsuda, K. Eriguchi, K. Ono and H. Ohta, Applied Physics Express 2, 116501 (2009).

[19] N. A. Mauchamp and S. Hamaguchi, Journal of Vacuum Science & Technology A 40,

53004 (2022).

217
[20] C. M. Huard, S. Sriraman, A. Paterson and M. J. Kushner, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 36, 6 (2018).

[21] C. Qu, Y. Sakiyama, P. Agarwal and M. J. Kushner, Journal of Vacuum Science &

Technology A: Vacuum, Surfaces, and Films 39, 52403 (2021).

[22] R. Dussart, R. Ettouri, J. Nos, G. Antoun, T. Tillocher and P. Lefaucheux, Journal of

Applied Physics 133, 113306 (2023).

[23] G. Antoun, T. Tillocher, P. Lefaucheux, J. Faguet, K. Maekawa and R. Dussart, Scientific

Reports 2021 11:1 11, 1 (2021).

[24] R. Dussart, T. Tillocher, P. Lefaucheux and M. Boufnichel, Journal of Physics D: Applied

Physics 47, 123001 (2014).

[25] A. Mesbah and D. B. Graves, Journal of Physics D: Applied Physics 52, 30LT02 (2019).

[26] A. D. Bonzanini, K. Shao, D. B. Graves, S. Hamaguchi and A. Mesbah, Plasma Sources

Science and Technology 32, 024003 (2023).

[27] R. Anirudh, R. Archibald, M. S. Asif, M. M. Becker, S. Benkadda, P. T. Bremer, R. H. S.

Bude, X. Zhang, et al., IEEE Transactions on Plasma Science 51, 1750 (2023).

[28] J. Moyne and J. Iskandar, Processes 2017, Vol. 5, Page 39 5, 39 (2017).

[29] D. Gidon, D. B. Graves and A. Mesbah, Plasma Sources Science and Technology 26,

085005 (2017).

[30] T. F. Edgar, S. W. Butler, W. J. Campbell, C. Pfeiffer, C. Bode, S. B. Hwang, K. S.

Balakrishnan and J. Hahn, Automatica 36, 1567 (2000).

[31] M. Hankinson, T. Vincent, K. B. Irani and P. P. Khargonekar, IEEE Transactions on

Semiconductor Manufacturing 10, 121 (1997).

218
[32] D. Stokes and G. S. May, IEEE Transactions on Semiconductor Manufacturing 13, 469

(2000).

[33] J. H. Xia, Rusli and A. S. Kumta, IEEE Transactions on Plasma Science 38, 142 (2010).

[34] W. Q. Xiong, Y. Qiao, L. P. Bai, M. Ghahramani, N. Q. Wu, P. H. Hsieh and B. Liu, IEEE

Transactions on Semiconductor Manufacturing 34, 207 (2021).

[35] J. H. Xia, Rusli and A. Kumta, IEEE Transactions on Plasma Science 38, 1091 (2010).

[36] C. E. Davis and G. S. May, IEEE Transactions on Electronics Packaging Manufacturing 31,

104 (2008).

[37] P. Seleson, M. Mustafa, D. Curreli, C. D. Hauck, M. Stoyanov and D. E. Bernholdt,

Computer Physics Communications 279, 108436 (2022).

[38] T. Gergs, B. Borislavov and J. Trieschmann, Journal of Vacuum Science & Technology B

40, 12802 (2022).

[39] G. Dong, X. Wei, J. Bao, G. Brochard, Z. Lin and W. Tang, Nuclear Fusion 61, 126061

(2021).

[40] S. Morosohk and E. Schuster, Contributions to Plasma Physics 63, e202200153 (2023).

[41] C. Ma, B. Zhu, X. Q. Xu and W. Wang, Physics of Plasmas 27, 42502 (2020).

[42] A. Merlo, D. Böckenhoff, J. Schilling, U. Höfel, S. Kwak, J. Svensson, A. Pavone, S. A.

Lazerson and T. S. Pedersen, Nuclear Fusion 61, 096039 (2021).

[43] J. C. Martínez-Loyola, A. A. Siller-Ceniceros, M. E. Sánchez-Castro, al -, S. M. Alia, S.

Pylypenko, Y. Liu, C. Akcay, L. L. Lao and X. Sun, Nuclear Fusion 62, 126067 (2022).

[44] P. Rodriguez-Fernandez, A. E. White, A. J. Creely, M. J. Greenwald, N. T. Howard, F.

Sciortino and J. C. Wright, Fusion Science and Technology 74, 65 (2018).

[45] S. Dasbach and S. Wiesen, Nuclear Materials and Energy 34, 101396 (2023).

219
[46] Y. Kim, S. Lee, T. Jung, B. Lee, N. Kwak and al Yongjin Kim,

https://doi.org/10.1117/12.2087765 9428, 9 (2015).

[47] Y. Yang and Y. Xu, Journal of Vacuum Science & Technology B 41, 52602 (2023).

[48] M. Omura, J. Hashimoto, T. Adachi, Y. Kondo, M. Ishikawa, J. Abe, I. Sakai, H. Hayashi,

M. Sekine and M. Hori, Japanese Journal of Applied Physics 58, SEEB02 (2019).

[49] K. J. Owen, B. VanDerElzen, R. L. Peterson and K. Najafi, Proceedings of the IEEE

International Conference on Micro Electro Mechanical Systems (MEMS) 251 (2012).

[50] B. Wu, A. Kumar and S. Pamarthy, Journal of Applied Physics 108, 51101 (2010).

[51] S. Huang, S. Shim, S. K. Nam and M. J. Kushner, Journal of Vacuum Science & Technology

A: Vacuum, Surfaces, and Films 38, 023001 (2020).

[52] Z. Hu, H. Shao, J. Li, P. Lai, W. Wang, C. Li, Q. Yan, X. He, J. Li, T. Yang, R. Chen and

Y. Wei, Journal of Vacuum Science & Technology A 41, 63113 (2023).

[53] J. Bobinac, T. Reiter, J. Piso, X. Klemenschits, O. Baumgartner, Z. Stanojevic, G. Strof, M.

Karner and L. Filipovic, Micromachines 2023, Vol. 14, Page 665 14, 665 (2023).

[54] R. P. Brinkmann, J Phys D Appl Phys 44, 042002 (2011).

[55] W. Park, J. Han, S. Park and S. Y. Moon, Vacuum 216, 112466 (2023).

[56] X. Xiao, X. Ke, B. Su and H.-Y. Zhang, ECS Transactions 104, 201 (2021).

[57] T. Matsushita, T. Matsumoto, H. Mukai, S. Kyoh, K. Hashimoto Takaya Matsushita and K.

Hashimoto, https://doi.org/10.1117/12.2085628 9428, 15 (2015).

[58] A. A. Kaptanoglu, K. D. Morgan, C. J. Hansen and S. L. Brunton, Physical Review E 104,

015206 (2021).

[59] E. P. Alves and F. Fiuza, Physical Review Research 4, (2020).

[60] K. Willcox and J. Peraire, https://doi.org/10.2514/2.1570 40, 2323 (2012).

220
[61] Y. Xing, M. A. Gosálvez, K. Sato, M. Tian and H. Yi, Journal of Micromechanics and

Microengineering 22, 085020 (2012).

[62] M. A. Goslvez, N. Ferrando, Y. Xing, P. Pal, K. Sato, J. Cerdá and R. Gadea, Journal of

Micromechanics and Microengineering 21, 065017 (2011).

[63] Y. Li, Y. Xing, M. A. Gosalvez, P. Pal and Y. Zhou, 2013 Transducers and Eurosensors

XXVII: The 17th International Conference on Solid-State Sensors, Actuators and

Microsystems, TRANSDUCERS and EUROSENSORS 2013 1087 (2013).

[64] C. C. Liu, Y. Li, Y. S. Yang, C. Y. Chen and M. H. Chuang, International Conference on

Simulation of Semiconductor Processes and Devices, SISPAD 2020-September, 355

(2020).

[65] T. Xiao and D. Ni, Processes 2021, Vol. 9, Page 151 9, 151 (2021).

[66] M. S. Daoud, M. Shehab, H. M. Al-Mimi, L. Abualigah, R. A. Zitar and M. K. Y. Shambour,

Archives of Computational Methods in Engineering 2022 30:4 30, 2431 (2022).

[67] M. Jalalitabar, I. G. Tsoulos, Y. Tian, Y. Zhang and H. Zhang, Mathematics 2023, Vol. 11,

Page 682 11, 682 (2023).

[68] L. ; Zhou, X. ; Zhou, C. A. Yi, L. Zhou, X. Zhou and C. Yi, Electronics 2023, Vol. 12, Page

994 12, 994 (2023).

[69] S. Takenaga, Y. Ozaki and M. Onishi, Optimization Letters 17, 283 (2023).

[70] Z. Huang, S. Yan, M. Rosenbusch, S. Yan and Z. Huang, Authorea Preprints (2023).

[71] M. J. Kushner, Journal of Physics D: Applied Physics 42, 194013 (2009).

[72] T. Piskin, Y. Qian, P. Pribyl, W. Gekelman and M. J. Kushner, Journal of Applied Physics

133, 173302 (2023).

221
[73] C. Qu, S. J. Lanham, S. C. Shannon, S. K. Nam and M. J. Kushner, Journal of Applied

Physics 127, 133302 (2020).

[74] D. Zhang, S. Rauf and T. Sparks, IEEE Transactions on Plasma Science 30, 114 (2002).

[75] R. J. Hoekstra, M. J. Kushner, V. Sukharev and P. Schoenborn, Journal of Vacuum Science

& Technology B: Microelectronics and Nanometer Structures Processing, Measurement,

and Phenomena 16, 2102 (1998).

[76] M. Shen, T. Lill, J. Hoang, H. Chi, A. Routzahn, J. Church, P. Subramonium, R.

Puthenkovilakam, S. Reddy, S. Bhadauriya, S. Roberts and G. Kamarthy, Japanese Journal

of Applied Physics 62, SI0801 (2023).

[77] H. J. M. Shi, M. Qiming Xuan, F. Oztoprak and J. Nocedal, Optimization Methods and

Software 38, 289 (2023).

[78] E. J. Paul, M. Landreman and T. Antonsen, Journal of Plasma Physics 87, 905870214

(2021).

[79] D. Shin and S. J. Hong, Journal of Vacuum Science & Technology B 41, 64002 (2023).

[80] J. A. Désidéri, Comptes Rendus Mathematique 350, 313 (2012).

[81] S. Ruder, (2016).

[82] R. R. Barton and J. S. Ivey, https://doi.org/10.1287/mnsc.42.7.954 42, 954 (1996).

[83] K. R. Mahmoud, Electromagnetics 31, 578 (2011).

[84] I. Fajfar, Á. Bűrmen and J. Puhan, Optimization Letters 13, 1011 (2019).

[85] C. Audet and C. Tribes, Computational Optimization and Applications 71, 331 (2018).

[86] M. A. Luersen and R. Le Riche, Computers & Structures 82, 2251 (2004).

[87] Y. Ozaki, M. Yano and M. Onishi, IPSJ Transactions on Computer Vision and Applications

9, 1 (2017).

222
[88] Y. Lee, A. Resiga, S. Yi and C. Wern, Journal of Manufacturing and Materials Processing

2020, Vol. 4, Page 66 4, 66 (2020).

[89] E. Zahara and Y. T. Kao, Expert Systems with Applications 36, 3880 (2009).

[90] J. D. Kress, ; D E Hanson, ; A F Voter, ; C L Liu, X.-Y. Liu, ; D G Coronell, ) D E Hanson,

A. F. Voter, C. L. Liu and D. G. Coronell, Journal of Vacuum Science & Technology A 17,

2819 (1999).

[91] J. P. Chang and H. H. Sawin, Journal of Vacuum Science & Technology A 15, 610 (1997).

223
 Chapter 7 Summary and Outlook

7.1 Summary

In Chapter 1, an overview of physical properties and mechanisms relevant to plasma-based

etching of semiconductors was given. This included basic plasma and surface etch phenomena as

well as specific applications such as CCPs powered by tailored voltage waveforms operated in

Fluorocarbon gas mixtures. A brief literature overview of the field of semiconductor

manufacturing and the narrow subject of this thesis was given. Finally, the role of computational

modeling as well as current challenges with respect to the fundamental data availability was

presented to motivate the content of this thesis.

In Chapter 2, the computational models utilized in this work were introduced. The Hybrid

Plasmas Equipment Model, HPEM, was used to simulate the gas phase as well as discharge

dynamics and collect the magnitude as well as energy and angular distributions of fluxes to the

wafer surface. These fluxes were used as input for the surface simulations using the Monte Carlo

Feature Profile Model, MCFPM, which determines the temporal evolution of the surface during

the etch process based on a predefined reaction mechanism.

In Chapter 3, the use of tailored voltage waveforms in geometrically asymmetric

capacitively coupled plasmas sustained in Ar/O2 at 40 mTorr was computationally investigated

with the goal of shaping the EAD of electrons and ions incident onto the substrate to address

differential charging. The tailored waveform consisted of a sinusoidal wave and its higher

harmonics with a fundamental frequency of 1 MHz. It was found that electric field reversals in the

224
sheath and presheath can occur during the anodic portion of the cycle. The electric field reversal

increases the energy and decreases the angular spread of electrons incident onto the substrate. The

magnitude of the electric field reversal can be controlled by the phase angle of the even harmonics

and the gas composition. Due to its electronegative nature, increasing mole fractions of O2 impedes

electron transport to the surface which further increases the electric field reversal.

In Chapter 4, the previous results were expanded upon by investigating the plasma etching

of HAR features into SiO2 in a similar geometry and power delivery configuration but using a

Ar/CF4/O2 mixture. It was similarly found that some degree of control of the IEADs and EEADs

is possible by adjusting the phase of higher harmonics φ through the resulting generation of

electrical asymmetry and electric field reversal. However, the IEADs and EEADs cannot easily be

separately controlled. The control of IEADs and EEADs are inherently linked. The highest quality

feature was obtained with a phase angle φ = 0° as this value generated the largest (most negative)

DC self-bias and largest electric field reversal for accelerating electrons into the feature. That said,

the consequences of voltage waveform tailoring (VWT) on etched features was dominated by the

change in the IEADs. Although VWT does produce EEADs with higher energy and narrower

angular spread, the effect of these electrons on the feature compared to thermal electrons is not

large. This smaller impact of VWT produced EEADs is attributed to thermal electrons being

accelerated into the feature by electric fields produced by the positive in-feature charging.

In Chapter 5, the relation between ion energy and DC self-bias was investigated in the

context of a varying fundamental frequency f0 for capacitively coupled plasmas sustained in

Ar/CF4/O2 and how those trends translate to a high aspect ratio etching of trenches in SiO2. f0, was

varied from 1 MHz to 10 MHz, and the relative phase from 0 to 180°. Two distinct regimes were

identified. Average ion energy onto the wafer is strongly correlated to the DC self-bias at high f0,

225
with there being a maximum at φ = 0° and minimum at φ = 180°. In the low frequency regime this

correlation is weak. Average ion energy onto the wafer is instead dominated by dynamic transients

in the applied voltage waveforms, with a maximum at φ = 180° and minimum at φ = 0°. The trends

in ion energy translate to etch properties. In both the high and low frequency regimes, higher ion

energies translate to higher etch rates and generally preferable final features, though behaving

differently with phase angle.

In Chapter 6, an autonomous optimization scheme for reaction mechanisms was

introduced to address the issue of data availability and accurate reproduction of physical etch

processes. A method to automate the selection of fundamental data in a reduced reaction

mechanism for feature scale SiO2 plasma etching using a fluorocarbon gas mixture was discussed.

This was done by matching predictions of etch profiles, obtained from MCFPM simulations to

experimental data using a gradient descent / Nelder-Mead method hybrid optimization scheme.

These methods produce a reaction mechanism that replicates the experimental training data as well

as experimental data using a related but different etch processes.

7.2 Future Work

While this thesis has presented promising evidence to highlight the upside potential for

technological applications of voltage waveform tailoring, further research will certainly be

required to solidify these observations. It would be instructive to expand upon this work by

broadening the parameter space in terms of pressure, power, reactor geometry, and gas

composition. The use of VWT was aimed at improving etch dynamics in Fluorocarbon etching of

SiO2, but the gained insights could potentially translate to other related systems in which the

control of electron and ion EADs is of great importance. This may include similar etch processes

of dielectrics or other materials using entirely different gas compositions.

226
 The use of tailored voltage waveforms has been intensely studied using a host of theoretical

and computational approaches. While many of these have shown great promise with respect to the

possible their technological applications, many challenges remain with respect to a widespread use

of this or related techniques. One of these challenges is the adequate matching of these complex

voltages which is only further compounded by recent trends to very high powers that limit the

tolerable relative amount of reflected power.

Conceptually, the use of complex waveforms is not limited to planar CCPs as presented in

this work. Since many other kinds of technological plasmas rely on RF power coupling as a source

of plasma heating it is not unreasonable to expect useful applications of similar techniques in a

host of different plasma sources and operating regimes. In closely related systems, like for example

inductively coupled plasmas with RF biases, similar behavior when non sinusoidal waveforms are

applied, could open up analogues avenues for process control. However, this may not be true for

other types of discharges that operate in entirely different regimes with respect to the density and

temperature of the plasma as well as neutral feedstock gas. So, while it is unlikely that the same

working principles produce identical effects, it is nevertheless possible that voltage waveform

tailoring can produce significant effects in these systems as well that warrant further investigations.

The optimization scheme presented in Chapter 6 should at this stage be considered a proof

of concept, giving many potential areas of improvement.

The target metrics used in this work were able to represent certain aspects of the etch

simulation with high accuracy but were lacking in others. The optimization scheme would greatly

benefit from the inclusion of metrics that better capture the mask necking and the geometry in the

lower echelons of the feature such as micro trenches and flatness of the etch front. The metrics

considered in this work were based on a 2-dimensional representation of a vertical slice through

227
the feature. It might prove useful to implement 3-dimensional target metrics such as symmetry

along the y-axis (normal to the current viewing plane) or changes in feature circularity, where

applicable.

For the optimization itself the testing and direct comparison of other algorithms might

provide additional improvements in terms of accuracy and convergence behavior. Additionally,

the formulation and implementation of a rigorous convergence criterion will likely prove very

effective in reducing the overall computational expense of the optimization process.

Finally, the optimization of the surface reaction mechanism has relied on the assumption

that the HPEM data accurately represents the physical ground truth. To some degree, HPEM is

subject to the same fundamental issues with respect to its parameterized data and imperfect

benchmarking to experiments. Thus, a natural progression of this work could be the extensions of

a similar optimization scheme to include gas phase reaction mechanisms and tune it using HPEM

and appropriate target metrics.

228
 Appendices

Appendix A: Statistical Variation in Predicted Feature Profiles

The Monte-Carlo method employed in the MCFPM produces profiles that are inherently

subject to statistical variations. This is especially the case when charging processes are included

which can act as a feedback loop which amplifies statistical variations. These variations result

from the sequence of random numbers that are used to initially generate particle trajectories

towards the surface. Quantitatively different features are produced by changing the seed that is

used to initialize the random number generators. Even when keeping the same seeds, when

executing the simulation in a parallel computing environment, there are nearly unavoidable

differences in sequencing of the particle trajectories tracked on different processors. Different

random number generators are used for each parallel thread to minimize these sequencing issues,

but they will occur.

To ensure the overall validity of results produced when statistical variations occur, it is

imperative to verify that the random run-to-run variations are of significantly smaller magnitude

than the effects discussed. This includes the overall trends in feature properties as a function of

phase angle φ as well as the isolated effects produced by surface charge neutralization by the

incident electron flux, perhaps most sensitive to these statistical variations.

229
 To assess the statistical variations, a series of identical simulations was performed while

varying the random number seed. These cases also include the inherent statistical variations that

occur in the parallel computing environment. The resulting profiles for the φ = 0°, PVWT = 1000 W

series are shown in Figure A.0.1a for 15 minutes of etching. The maximum difference from the

mean etch depth for this series is less than 5% of the total etch. This variation is less than the

change in etch depth and profile shape produced by the change in IEADs as a function of φ

observed in this work. Similar conclusions extend to variation between the profiles produced with

VWT produced and thermal EEADs.

Statistical variations of this type are not limited to numerical simulations but also occur in

actual etch processes, producing a certain degree of feature-to-feature non-uniformity. This

statistical feature-to-feature variation results from the small size of features, which in turn produces

statistically different fluxes of reactant species into the feature. A common approach to remedy

these unwanted statistical variations in HVM processes is over-etching. In over-etching, a terminal

or etch stop layer is located at the desired final etch depth. For any given process, the etch rate into

this layer is significantly lower than for the overlying substrate. This means that once the stop layer

is reached during the process, vertical etching effectively stops (or is significantly slowed) while

etching of the tapered side walls continues. Extending the etch process beyond this contact time,

over-etching, partially remedies some of the statistical variations in feature profile by straightening

the sidewalls of the feature. A series of profiles are shown in Figure A.0.1b produced with different

random number seeds, analogous to Figure A.0.1a, while etching to the stop layer and over-etching

for a total of 30 minutes. Through this over-etching, the variations in etch depth and profile are

reduced. Where applicable, etch stop layers and over-etching can suppress the statistical run to run

differences in feature properties.

230
Figure A.0.1: Statistical run-to-run variation of the final etch profiles for the PVWT = 1000 W and φ = 0° case. a) 15
min run and b) 30 min to produce over etch into the stop layer. For each series of profiles, each case had a different
seed for the random number generators, in addition to having the inherent statistical variations in the parallel
computing environment.

231
 Appendix B: Surface Reaction Mechanism

Table B.0.1 contains the surface reaction mechanism used in this work after convergence.

The reaction probability p0 is modified according to Eq. (2.40) if angular or energy dependence

of the reaction is present. In that case,  th ,  0 and q define the energy dependence and ∠ defines

the nature of the angular dependence, with ∠=1 corresponding to the results obtained by [1] and

∠=2 corresponding to the results obtained by [2].

If no values for p0 ,  0 ,  th n and ∠ are provided, the reaction has not energy or angular

dependence. The superscripts ‘+’, ‘#’ and ‘*’ identify positive ions, hot neutrals and exited species,

respectively.

The subscript ‘(s)’ refers to a surface species, while ‘(xs)’ is a crosslinked surface species.

Not listed below are surface neutralization reactions that occur for every charge species. EP is a

generic etch product that is assumed to be inert and not tracked further.

Table B.0.1: Complete surface reaction mechanism

Reaction p0  th n 0 ∠
CF(s) + CF → CF(s) + CF(s) 0.1
CF(s) + CF2 → CF(s) + CF2(s) 0.1
CF(s) + CF3 → CF(s) + CF3(s) 0.1
CF(s) + C2F3 → CF(s) + C2F3(s) 0.1
CF2(s) + CF → CF2(s) + CF(s) 0.1
CF2(s) + CF2 → CF2(s) + CF2(s) 0.1
CF2(s) + CF3 → CF2(s) + CF3(s) 0.1
CF2(s) + C2F3 → CF2(s) + C2F3(s) 0.1
CF3(s) + CF → CF3(s) + CF(s) 0.1

232
CF3(s) + CF2 → CF3(s) + CF2(s) 0.1
CF3(s) + CF3 → CF3(s) + CF3(s) 0.1
CF3(s) + C2F3 → CF3(s) + C2F3(s) 0.1
C2F3(s) + CF → C2F3(s) + CF(s) 0.03
C2F3(s) + CF2 → C2F3(s) + CF2(s) 0.03
C2F3(s) + CF3 → C2F3(s) + CF3(s) 0.03
C2F3(s) + C2F3 → C2F3(s) + C2F3(s) 0.03
AC(s) + CF → AC(s) + CF(s) 0.2
AC(s) + CF2 → AC(s) + CF2(s) 0.2
AC(s) + CF3 → AC(s) + CF3(s) 0.2
AC(s) + C2F3 → AC(s) + C2F3(s) 0.2
AC(xs) + CF → AC(xs) + CF(s) 0.2
AC(xs) + CF2 → AC(xs) + CF2(s) 0.2
AC(xs) + CF3 → AC(xs) + CF3(s) 0.2
AC(xs) + C2F3 → AC(xs) + C2F3(s) 0.2
CF(xs) + CF → CF(xs) + CF(s) 0.02
CF(xs) + CF2 → CF(xs) + CF2(s) 0.02
CF(xs) + CF3 → CF(xs) + CF3(s) 0.02
CF(xs) + C2F3 → CF(xs) + C2F3(s) 0.02
CF2(xs) + CF → CF2(xs) + CF(s) 0.02
CF2(xs) + CF2 → CF2(xs) + CF2(s) 0.02
CF2(xs) + CF3 → CF2(xs) + CF3(s) 0.02
CF2(xs) + C2F3 → CF2(xs) + C2F3(s) 0.02
CF3(xs) + CF → CF3(xs) + CF(s) 0.02
CF3(xs) + CF2 → CF3(xs) + CF2(s) 0.02
CF3(xs) + CF3 → CF3(xs) + CF3(s) 0.02
CF3(xs) + C2F3 → CF3(xs) + C2F3(s) 0.02
CF(s) +O → EP 0.0423
CF2(s) +O → EP 0.0423
CF3(s) +O → EP 0.0423
C2F3(s) +O → EP 0.0423
CF(xs) +O → EP 0.0423
CF2(xs) +O → EP 0.0423
CF3(xs) +O → EP 0.0423
AC(s) +O → CO 1.00E-05
AC(xs) +O → CO 1.00E-05
AC(s) + Ar+ → C + Ar# 0.001 200 0.4 250 1
AC(s) + F2+ → C + F2# 0.001 200 0.4 250 1
AC(s) + F+ → C + F# 0.001 200 0.4 250 1
AC(s) + O2+ → C + O2# 0.001 200 0.4 250 1
AC(s) + O+ → C + O# 0.001 200 0.4 250 1

233
AC(s) + CF3+ → C + CF3# 0.001 200 0.4 250 1
AC(s) + CF2+ → C + CF2# 0.001 200 0.4 250 1
AC(s) + CF+ → C + CF# 0.001 200 0.4 250 1
AC(s) + C2F3+ → C + C2F3# 0.001 200 0.4 250 1
AC(s) + C2F4+ → C + C2F4# 0.001 200 0.4 250 1
AC(s) + C2F5+ → C + C2F5# 0.001 200 0.4 250 1
AC(s) + C3F5+ → C + C3F5# 0.001 200 0.4 250 1
AC(s) + C3F6+ → C + C3F6# 0.001 200 0.4 250 1
AC(s) + C3F7+ → C + C3F7# 0.001 200 0.4 250 1
AC(s) + C4F7+ → C + C4F7# 0.001 200 0.4 250 1
AC(s) + C4F8+ → C + C4F8# 0.001 200 0.4 250 1
AC(s) + Ar# → C + Ar# 0.001 200 0.4 250 1
AC(s) + F2# → C + F2# 0.001 200 0.4 250 1
AC(s) + F# → C + F# 0.001 200 0.4 250 1
AC(s) + O2# → C + O2# 0.001 200 0.4 250 1
AC(s) + O# → C + O# 0.001 200 0.4 250 1
AC(s) + CF3# → C + CF3# 0.001 200 0.4 250 1
AC(s) + CF2# → C + CF2# 0.001 200 0.4 250 1
AC(s) + CF# → C + CF# 0.001 200 0.4 250 1
AC(s) + C2F3# → C + C2F3# 0.001 200 0.4 250 1
AC(s) + C2F4# → C + C2F4# 0.001 200 0.4 250 1
AC(s) + C2F5# → C + C2F5# 0.001 200 0.4 250 1
AC(s) + C3F5# → C + C3F5# 0.001 200 0.4 250 1
AC(s) + C3F6# → C + C3F6# 0.001 200 0.4 250 1
AC(s) + C3F7# → C + C3F7# 0.001 200 0.4 250 1
AC(s) + C4F7# → C + C4F7# 0.001 200 0.4 250 1
AC(s) + C4F8# → C + C4F8# 0.001 200 0.4 250 1
AC(xs) + Ar+ → C + Ar# 0.001 200 0.4 250 1
AC(xs) + F2+ → C + F2# 0.001 200 0.4 250 1
AC(xs) + F+ → C + F# 0.001 200 0.4 250 1
AC(xs) + O2+ → C + O2# 0.001 200 0.4 250 1
AC(xs) + O+ → C + O# 0.001 200 0.4 250 1
AC(xs) + CF3+ → C + CF3# 0.001 200 0.4 250 1
AC(xs) + CF2+ → C + CF2# 0.001 200 0.4 250 1
AC(xs) + CF+ → C + CF# 0.001 200 0.4 250 1
AC(xs) + C2F3+ → C + C2F3# 0.001 200 0.4 250 1
AC(xs) + C2F4+ → C + C2F4# 0.001 200 0.4 250 1
AC(xs) + C2F5+ → C + C2F5# 0.001 200 0.4 250 1
AC(xs) + C3F5+ → C + C3F5# 0.001 200 0.4 250 1
AC(xs) + C3F6+ → C + C3F6# 0.001 200 0.4 250 1
AC(xs) + C3F7+ → C + C3F7# 0.001 200 0.4 250 1

234
AC(xs) + C4F7+ → C + C4F7# 0.001 200 0.4 250 1
AC(xs) + C4F8+ → C + C4F8# 0.001 200 0.4 250 1
AC(xs) + Ar# → C + Ar# 0.001 200 0.4 250 1
AC(xs) + F2# → C + F2# 0.001 200 0.4 250 1
AC(xs) + F# → C + F# 0.001 200 0.4 250 1
AC(xs) + O2# → C + O2# 0.001 200 0.4 250 1
AC(xs) + O# → C + O# 0.001 200 0.4 250 1
AC(xs) + CF3# → C + CF3# 0.001 200 0.4 250 1
AC(xs) + CF2# → C + CF2# 0.001 200 0.4 250 1
AC(xs) + CF# → C + CF# 0.001 200 0.4 250 1
AC(xs) + C2F3# → C + C2F3# 0.001 200 0.4 250 1
AC(xs) + C2F4# → C + C2F4# 0.001 200 0.4 250 1
AC(xs) + C2F5# → C + C2F5# 0.001 200 0.4 250 1
AC(xs) + C3F5# → C + C3F5# 0.001 200 0.4 250 1
AC(xs) + C3F6# → C + C3F6# 0.001 200 0.4 250 1
AC(xs) + C3F7# → C + C3F7# 0.001 200 0.4 250 1
AC(xs) + C4F7# → C + C4F7# 0.001 200 0.4 250 1
AC(xs) + C4F8# → C + C4F8# 0.001 200 0.4 250 1
CF(s) + Ar+ → EP + Ar# 0.9 20 0.5 500 1
CF(s) + Ar+ → AC(s) +F + Ar# 0.01 20 0.5 500 1
CF(s) + Ar# → EP + Ar# 0.9 20 0.5 500 1
CF(s) + Ar# → AC(s) +F + Ar# 0.01 20 0.5 500 1
CF(s) + CF3+ → EP + CF3# 0.9 20 0.5 500 1
CF(s) + CF3+ → AC(s) +F + CF3# 0.01 20 0.5 500 1
CF(s) + CF3# → EP + CF3# 0.9 20 0.5 500 1
CF(s) + CF3# → AC(s) +F + CF3# 0.01 20 0.5 500 1
CF(s) + CF2+ → EP + CF2# 0.9 20 0.5 500 1
CF(s) + CF2+ → AC(s) +F + CF2# 0.01 20 0.5 500 1
CF(s) + CF2# → EP + CF2# 0.9 20 0.5 500 1
CF(s) + CF2# → AC(s) +F + CF2# 0.01 20 0.5 500 1
CF(s) + CF+ → EP + CF# 0.9 20 0.5 500 1
CF(s) + CF+ → AC(s) +F + CF# 0.01 20 0.5 500 1
CF(s) + CF# → EP + CF# 0.9 20 0.5 500 1
CF(s) + CF# → AC(s) +F + CF# 0.01 20 0.5 500 1
CF(s) + F+ → EP + F# 0.9 20 0.5 500 1
CF(s) + F+ → AC(s) +F + F# 0.01 20 0.5 500 1
CF(s) + F# → EP + F# 0.9 20 0.5 500 1
CF(s) + F# → AC(s) +F + F# 0.01 20 0.5 500 1
CF(s) + F2+ → EP + F2# 0.9 20 0.5 500 1
CF(s) + F2+ → AC(s) +F + F2# 0.01 20 0.5 500 1
CF(s) + F2# → EP + F2# 0.9 20 0.5 500 1

235
CF(s) + F2# → AC(s) +F + F2# 0.01 20 0.5 500 1
CF(s) + C2F5+ → EP + C2F5# 0.9 20 0.5 500 1
CF(s) + C2F5+ → AC(s) +F + C2F5# 0.01 20 0.5 500 1
CF(s) + C2F5# → EP + C2F5# 0.9 20 0.5 500 1
CF(s) + C2F5# → AC(s) +F + C2F5# 0.01 20 0.5 500 1
CF(s) + C2F4+ → EP + C2F4# 0.9 20 0.5 500 1
CF(s) + C2F4+ → AC(s) +F + C2F4# 0.01 20 0.5 500 1
CF(s) + C2F4# → EP + C2F4# 0.9 20 0.5 500 1
CF(s) + C2F4# → AC(s) +F + C2F4# 0.01 20 0.5 500 1
CF(s) + C2F3+ → EP + C2F3# 0.9 20 0.5 500 1
CF(s) + C2F3+ → AC(s) +F + C2F3# 0.01 20 0.5 500 1
CF(s) + C2F3# → EP + C2F3# 0.9 20 0.5 500 1
CF(s) + C2F3# → AC(s) +F + C2F3# 0.01 20 0.5 500 1
CF(s) + C3F5+ → EP + C3F5# 0.9 20 0.5 500 1
CF(s) + C3F5+ → AC(s) +F + C3F5# 0.01 20 0.5 500 1
CF(s) + C3F5# → EP + C3F5# 0.9 20 0.5 500 1
CF(s) + C3F5# → AC(s) +F + C3F5# 0.01 20 0.5 500 1
CF(s) + C3F6+ → EP + C3F6# 0.9 20 0.5 500 1
CF(s) + C3F6+ → AC(s) +F + C3F6# 0.01 20 0.5 500 1
CF(s) + C3F6# → EP + C3F6# 0.9 20 0.5 500 1
CF(s) + C3F6# → AC(s) +F + C3F6# 0.01 20 0.5 500 1
CF(s) + C3F7+ → EP + C3F7# 0.9 20 0.5 500 1
CF(s) + C3F7+ → AC(s) +F + C3F7# 0.01 20 0.5 500 1
CF(s) + C3F7# → EP + C3F7# 0.9 20 0.5 500 1
CF(s) + C3F7# → AC(s) +F + C3F7# 0.01 20 0.5 500 1
CF(s) + C4F7+ → EP + C4F7# 0.9 20 0.5 500 1
CF(s) + C4F7+ → AC(s) +F + C4F7# 0.01 20 0.5 500 1
CF(s) + C4F7# → EP + C4F7# 0.9 20 0.5 500 1
CF(s) + C4F7# → AC(s) +F + C4F7# 0.01 20 0.5 500 1
CF(s) + C4F8+ → EP + C4F8# 0.9 20 0.5 500 1
CF(s) + C4F8+ → AC(s) +F + C4F8# 0.01 20 0.5 500 1
CF(s) + C4F8# → EP + C4F8# 0.9 20 0.5 500 1
CF(s) + C4F8# → AC(s) +F + C4F8# 0.01 20 0.5 500 1
CF(s) + O+ → EP + O# 0.9 20 0.5 500 1
CF(s) + O+ → AC(s) +F + O# 0.01 20 0.5 500 1
CF(s) + O2+ → EP + O2 # 0.9 20 0.5 500 1
CF(s) + O2+ → AC(s) +F + O2# 0.01 20 0.5 500 1
CF(s) + O# → EP + O# 0.9 20 0.5 500 1
CF(s) + O# → AC(s) +F + O# 0.01 20 0.5 500 1
CF(s) + O# → CF + O# 0.1
CF(s) + O2# → EP + O2 # 0.9 20 0.5 500 1

236
CF(s) + O2# → AC(s) +F + O2# 0.01 20 0.5 500 1
CF2(s) + Ar+ → EP + Ar# 0.9 20 0.5 500 1
CF2(s) + Ar+ → CF(s) +F + Ar# 0.01 20 0.5 500 1
CF2(s) + Ar# → EP + Ar# 0.9 20 0.5 500 1
CF2(s) + Ar# → CF(s) +F + Ar# 0.01 20 0.5 500 1
CF2(s) + CF3+ → EP + CF3# 0.9 20 0.5 500 1
CF2(s) + CF3+ → CF(s) +F + CF3# 0.01 20 0.5 500 1
CF2(s) + CF3# → EP + CF3# 0.9 20 0.5 500 1
CF2(s) + CF3# → CF(s) +F + CF3# 0.01 20 0.5 500 1
CF2(s) + CF2+ → EP + CF2# 0.9 20 0.5 500 1
CF2(s) + CF2+ → CF(s) +F + CF2# 0.01 20 0.5 500 1
CF2(s) + CF2# → EP + CF2# 0.9 20 0.5 500 1
CF2(s) + CF2# → CF(s) +F + CF2# 0.01 20 0.5 500 1
CF2(s) + CF+ → EP + CF# 0.9 20 0.5 500 1
CF2(s) + CF+ → CF(s) +F + CF# 0.01 20 0.5 500 1
CF2(s) + CF# → EP + CF# 0.9 20 0.5 500 1
CF2(s) + CF# → CF(s) +F + CF# 0.01 20 0.5 500 1
CF2(s) + F+ → EP + F# 0.9 20 0.5 500 1
CF2(s) + F+ → CF(s) +F + F# 0.01 20 0.5 500 1
CF2(s) + F# → EP + F# 0.9 20 0.5 500 1
CF2(s) + F# → CF(s) +F + F# 0.01 20 0.5 500 1
CF2(s) + F2+ → EP + F2# 0.9 20 0.5 500 1
CF2(s) + F2+ → CF(s) +F + F2# 0.01 20 0.5 500 1
CF2(s) + F2# → EP + F2# 0.9 20 0.5 500 1
CF2(s) + F2# → CF(s) +F + F2# 0.01 20 0.5 500 1
CF2(s) + C2F5+ → EP + C2F5# 0.9 20 0.5 500 1
CF2(s) + C2F5+ → CF(s) +F + C2F5# 0.01 20 0.5 500 1
CF2(s) + C2F5# → EP + C2F5# 0.9 20 0.5 500 1
CF2(s) + C2F5# → CF(s) +F + C2F5# 0.01 20 0.5 500 1
CF2(s) + C2F4+ → EP + C2F4# 0.9 20 0.5 500 1
CF2(s) + C2F4+ → CF(s) +F + C2F4# 0.01 20 0.5 500 1
CF2(s) + C2F4# → EP + C2F4# 0.9 20 0.5 500 1
CF2(s) + C2F4# → CF(s) +F + C2F4# 0.01 20 0.5 500 1
CF2(s) + C2F3+ → EP + C2F3# 0.9 20 0.5 500 1
CF2(s) + C2F3+ → CF(s) +F + C2F3# 0.01 20 0.5 500 1
CF2(s) + C2F3# → EP + C2F3# 0.9 20 0.5 500 1
CF2(s) + C2F3# → CF(s) +F + C2F3# 0.01 20 0.5 500 1
CF2(s) + C3F5+ → EP + C3F5# 0.9 20 0.5 500 1
CF2(s) + C3F5+ → CF(s) +F + C3F5# 0.01 20 0.5 500 1
CF2(s) + C3F5# → EP + C3F5# 0.9 20 0.5 500 1
CF2(s) + C3F5# → CF(s) +F + C3F5# 0.01 20 0.5 500 1

237
CF2(s) + C3F6+ → EP + C3F6# 0.9 20 0.5 500 1
CF2(s) + C3F5# → CF(s) +F + C3F5# 0.01 20 0.5 500 1
CF2(s) + C3F6# → EP + C3F6# 0.9 20 0.5 500 1
CF2(s) + C3F6# → CF(s) +F + C3F6# 0.01 20 0.5 500 1
CF2(s) + C3F7+ → EP + C3F7# 0.9 20 0.5 500 1
CF2(s) + C3F7+ → CF(s) +F + C3F7# 0.01 20 0.5 500 1
CF2(s) + C3F7# → EP + C3F7# 0.9 20 0.5 500 1
CF2(s) + C3F7# → CF(s) +F + C3F7# 0.01 20 0.5 500 1
CF2(s) + C4F7+ → EP + C4F7# 0.9 20 0.5 500 1
CF2(s) + C4F7+ → CF(s) +F + C4F7# 0.01 20 0.5 500 1
CF2(s) + C4F7# → EP + C4F7# 0.9 20 0.5 500 1
CF2(s) + C4F7# → CF(s) +F + C4F7# 0.01 20 0.5 500 1
CF2(s) + C4F8+ → EP + C4F8# 0.9 20 0.5 500 1
CF2(s) + C4F8+ → CF(s) +F + C4F8# 0.01 20 0.5 500 1
CF2(s) + C4F8# → EP + C4F8# 0.9 20 0.5 500 1
CF2(s) + C4F8# → CF(s) +F + C4F8# 0.01 20 0.5 500 1
CF2(s) + O+ → EP + O# 0.9 20 0.5 500 1
CF2(s) + O+ → CF(s) +F + O# 0.01 20 0.5 500 1
CF2(s) + O2+ → EP + O2 # 0.9 20 0.5 500 1
CF2(s) + O2+ → CF(s) +F + O2# 0.01 20 0.5 500 1
CF2(s) + O# → EP + O# 0.9 20 0.5 500 1
CF2(s) + O# → CF(s) +F + O# 0.01 20 0.5 500 1
CF2(s) + O# → CF + O# 0.1
CF2(s) + O2# → EP + O2 # 0.9 20 0.5 500 1
CF2(s) + O2# → CF(s) +F + O2 #
0.01 20 0.5 500 1
CF3(s) + Ar+ → EP + Ar# 0.9 20 0.5 500 1
CF3(s) + Ar+ → CF2(s) +F + Ar# 0.01 20 0.5 500 1
CF3(s) + Ar# → EP + Ar# 0.9 20 0.5 500 1
CF3(s) + Ar# → CF2(s) +F + Ar# 0.01 20 0.5 500 1
CF3(s) + CF3+ → EP + CF3# 0.9 20 0.5 500 1
CF3(s) + CF3+ → CF2(s) +F + CF3# 0.01 20 0.5 500 1
CF3(s) + CF3# → EP + CF3# 0.9 20 0.5 500 1
CF3(s) + CF3# → CF2(s) +F + CF3# 0.01 20 0.5 500 1
CF3(s) + CF2+ → EP + CF2# 0.9 20 0.5 500 1
CF3(s) + CF2+ → CF2(s) +F + CF2# 0.01 20 0.5 500 1
CF3(s) + CF2# → EP + CF2# 0.9 20 0.5 500 1
CF3(s) + CF2# → CF2(s) +F + CF2# 0.01 20 0.5 500 1
CF3(s) + CF+ → EP + CF# 0.9 20 0.5 500 1
CF3(s) + CF+ → CF2(s) +F + CF# 0.01 20 0.5 500 1
CF3(s) + CF# → EP + CF# 0.9 20 0.5 500 1
CF3(s) + CF# → CF2(s) +F + CF# 0.01 20 0.5 500 1

238
CF3(s) + F+ → EP + F# 0.9 20 0.5 500 1
CF3(s) + F+ → CF2(s) +F +F #
0.01 20 0.5 500 1
CF3(s) + F# → EP + F# 0.9 20 0.5 500 1
CF3(s) + F# → CF2(s) +F + F# 0.01 20 0.5 500 1
CF3(s) + F2+ → EP + F2# 0.9 20 0.5 500 1
CF3(s) + F2+ → CF2(s) +F + F2# 0.01 20 0.5 500 1
CF3(s) + F2# → EP + F2# 0.9 20 0.5 500 1
CF3(s) + F2# → CF2(s) +F + F2# 0.01 20 0.5 500 1
CF3(s) + C2F5+ → EP + C2F5# 0.9 20 0.5 500 1
CF3(s) + C2F5+ → CF2(s) +F + C2F5# 0.01 20 0.5 500 1
CF3(s) + C2F5# → EP + C2F5# 0.9 20 0.5 500 1
CF3(s) + C2F5# → CF2(s) +F + C2F5# 0.01 20 0.5 500 1
CF3(s) + C2F4+ → EP + C2F4# 0.9 20 0.5 500 1
CF3(s) + C2F4+ → CF2(s) +F + C2F4# 0.01 20 0.5 500 1
CF3(s) + C2F4# → EP + C2F4# 0.9 20 0.5 500 1
CF3(s) + C2F4# → CF2(s) +F + C2F4# 0.01 20 0.5 500 1
CF3(s) + C2F3+ → EP + C2F3# 0.9 20 0.5 500 1
CF3(s) + C2F3+ → CF2(s) +F + C2F3# 0.01 20 0.5 500 1
CF3(s) + C2F3# → EP + C2F3# 0.9 20 0.5 500 1
CF3(s) + C2F3# → CF2(s) +F + C2F3# 0.01 20 0.5 500 1
CF3(s) + C3F5+ → EP + C3F5# 0.9 20 0.5 500 1
CF3(s) + C3F5+ → CF2(s) +F + C3F5# 0.01 20 0.5 500 1
CF3(s) + C3F5# → EP + C3F5# 0.9 20 0.5 500 1
CF3(s) + C3F5# → CF2(s) +F + C3F5# 0.01 20 0.5 500 1
CF3(s) + C3F6+ → EP + C3F6# 0.9 20 0.5 500 1
CF3(s) + C3F5# → CF2(s) +F + C3F5# 0.01 20 0.5 500 1
CF3(s) + C3F6# → EP + C3F6# 0.9 20 0.5 500 1
CF3(s) + C3F6# → CF2(s) +F + C3F6# 0.01 20 0.5 500 1
CF3(s) + C3F7+ → EP + C3F7# 0.9 20 0.5 500 1
CF3(s) + C3F7+ → CF2(s) +F + C3F7# 0.01 20 0.5 500 1
CF3(s) + C3F7# → EP + C3F7# 0.9 20 0.5 500 1
CF3(s) + C3F7# → CF2(s) +F + C3F7# 0.01 20 0.5 500 1
CF3(s) + C4F7+ → EP + C4F7# 0.9 20 0.5 500 1
CF3(s) + C4F7+ → CF2(s) +F + C4F7# 0.01 20 0.5 500 1
CF3(s) + C4F7# → EP + C4F7# 0.9 20 0.5 500 1
CF3(s) + C4F7# → CF2(s) +F + C4F7# 0.01 20 0.5 500 1
CF3(s) + C4F8+ → EP + C4F8# 0.9 20 0.5 500 1
CF3(s) + C4F8+ → CF2(s) +F + C4F8# 0.01 20 0.5 500 1
CF3(s) + C4F8# → EP + C4F8# 0.9 20 0.5 500 1
CF3(s) + C4F8# → CF2(s) +F + C4F8# 0.01 20 0.5 500 1
CF3(s) + O+ → EP + O# 0.9 20 0.5 500 1

239
CF3(s) + O+ → CF2(s) +F + O# 0.01 20 0.5 500 1
CF3(s) + O2+ → EP + O2 # 0.9 20 0.5 500 1
CF3(s) + O2+ → CF2(s) +F + O2# 0.01 20 0.5 500 1
CF3(s) + O# → EP + O# 0.9 20 0.5 500 1
CF3(s) + O# → CF2(s) +F + O# 0.01 20 0.5 500 1
CF3(s) + O# → CF + O# 0.1
CF3(s) + O2# → EP + O2 # 0.9 20 0.5 500 1
CF3(s) + O2# → CF2(s) +F + O2 #
0.01 20 0.5 500 1
C2F3(s) + Ar+ → EP + Ar# 0.8 50 0.5 500 1
C2F3(s) + Ar# → EP + Ar# 0.8 50 0.5 500 1
C2F3(s) + CF3+ → EP + CF3# 0.8 50 0.5 500 1
C2F3(s) + CF3# → EP + CF3# 0.8 50 0.5 500 1
C2F3(s) + CF2+ → EP + CF2# 0.8 50 0.5 500 1
C2F3(s) + CF2# → EP + CF2# 0.8 50 0.5 500 1
C2F3(s) + CF+ → EP + CF# 0.8 50 0.5 500 1
C2F3(s) + CF# → EP + CF# 0.8 50 0.5 500 1
C2F3(s) + F+ → EP + F# 0.8 50 0.5 500 1
C2F3(s) + F# → EP + F# 0.8 50 0.5 500 1
C2F3(s) + F2+ → EP + F2# 0.8 50 0.5 500 1
C2F3(s) + F2# → EP + F2# 0.8 50 0.5 500 1
C2F3(s) + C2F5+ → EP + C2F5# 0.8 50 0.5 500 1
C2F3(s) + C2F5# → EP + C2F5# 0.8 50 0.5 500 1
C2F3(s) + C2F4+ → EP + C2F4# 0.8 50 0.5 500 1
C2F3(s) + C2F4# → EP + C2F4# 0.8 50 0.5 500 1
C2F3(s) + C2F3+ → EP + C2F3# 0.8 50 0.5 500 1
C2F3(s) + C2F3# → EP + C2F3# 0.8 50 0.5 500 1
C2F3(s) + C3F5+ → EP + C3F5# 0.8 50 0.5 500 1
C2F3(s) + C3F5# → EP + C3F5# 0.8 50 0.5 500 1
C2F3(s) + C3F6+ → EP + C3F6# 0.8 50 0.5 500 1
C2F3(s) + C3F6# → EP + C3F6# 0.8 50 0.5 500 1
C2F3(s) + C3F7+ → EP + C3F7# 0.8 50 0.5 500 1
C2F3(s) + C3F7# → EP + C3F7# 0.8 50 0.5 500 1
C2F3(s) + C4F7+ → EP + C4F7# 0.8 50 0.5 500 1
C2F3(s) + C4F7# → EP + C4F7# 0.8 50 0.5 500 1
C2F3(s) + C4F8+ → EP + C4F8# 0.8 50 0.5 500 1
C2F3(s) + C4F8# → EP + C4F8# 0.8 50 0.5 500 1
C2F3(s) + O+ → EP + O# 0.8 50 0.5 500 1
C2F3(s) + O2+ → EP + O2 # 0.8 50 0.5 500 1
C2F3(s) + O# → EP + O# 0.8 50 0.5 500 1
C2F3(s) + O2# → EP + O2 # 0.8 50 0.5 500 1
CF(xs) + Ar+ → EP + Ar# 0.6 50 0.5 500 1

240
CF(xs) + Ar+ → CF(s) + Ar# 0.3 8 0.5 500 1
CF(xs) + Ar# → EP + Ar# 0.6 50 0.5 500 1
CF(xs) + Ar# → CF(s) + Ar# 0.3 8 0.5 500 1
CF(xs) + CF3+ → EP + CF3# 0.6 50 0.5 500 1
CF(xs) + CF3+ → CF(s) + CF3# 0.3 8 0.5 500 1
CF(xs) + CF3# → EP + CF3# 0.6 50 0.5 500 1
CF(xs) + CF3# → CF(s) + CF3# 0.3 8 0.5 500 1
CF(xs) + CF2+ → EP + CF2# 0.6 50 0.5 500 1
CF(xs) + CF2+ → CF(s) + CF2# 0.3 8 0.5 500 1
CF(xs) + CF2# → EP + CF2# 0.6 50 0.5 500 1
CF(xs) + CF2# → CF(s) + CF2# 0.3 8 0.5 500 1
CF(xs) + CF+ → EP + CF# 0.6 50 0.5 500 1
CF(xs) + CF+ → CF(s) + CF# 0.3 8 0.5 500 1
CF(xs) + CF# → EP + CF# 0.6 50 0.5 500 1
CF(xs) + CF# → CF(s) + CF# 0.3 8 0.5 500 1
CF(xs) + F+ → EP + F# 0.6 50 0.5 500 1
CF(xs) + F+ → CF(s) + F# 0.3 8 0.5 500 1
CF(xs) + F# → EP + F# 0.6 50 0.5 500 1
CF(xs) + F# → CF(s) + F# 0.3 8 0.5 500 1
CF(xs) + F2+ → EP + F2# 0.6 50 0.5 500 1
CF(xs) + F2+ → CF(s) + F2# 0.3 8 0.5 500 1
CF(xs) + F2# → EP + F2# 0.6 50 0.5 500 1
CF(xs) + F2# → CF(s) + F2# 0.3 8 0.5 500 1
CF(xs) + C2F5+ → EP + C2F5# 0.6 50 0.5 500 1
CF(xs) + C2F5+ → CF(s) + C2F5# 0.3 8 0.5 500 1
CF(xs) + C2F5# → EP + C2F5# 0.6 50 0.5 500 1
CF(xs) + C2F5# → CF(s) + C2F5# 0.3 8 0.5 500 1
CF(xs) + C2F4+ → EP + C2F4# 0.6 50 0.5 500 1
CF(xs) + C2F4+ → CF(s) + C2F4# 0.3 8 0.5 500 1
CF(xs) + C2F4# → EP + C2F4# 0.6 50 0.5 500 1
CF(xs) + C2F4# → CF(s) + C2F4# 0.3 8 0.5 500 1
CF(xs) + C2F3+ → EP + C2F3# 0.6 50 0.5 500 1
CF(xs) + C2F3+ → CF(s) + C2F3# 0.3 8 0.5 500 1
CF(xs) + C2F3# → EP + C2F3# 0.6 50 0.5 500 1
CF(xs) + C2F3# → CF(s) + C2F3# 0.3 8 0.5 500 1
CF(xs) + C3F5+ → EP + C3F5# 0.6 50 0.5 500 1
CF(xs) + C3F5+ → CF(s) + C3F5# 0.3 8 0.5 500 1
CF(xs) + C3F5# → EP + C3F5# 0.6 50 0.5 500 1
CF(xs) + C3F5# → CF(s) + C3F5# 0.3 8 0.5 500 1
CF(xs) + C3F6+ → EP + C3F6# 0.6 50 0.5 500 1
CF(xs) + C3F6+ → CF(s) + C3F6# 0.3 8 0.5 500 1

241
CF(xs) + C3F6# → EP + C3F6# 0.6 50 0.5 500 1
CF(xs) + C3F6# → CF(s) + C3F6# 0.3 8 0.5 500 1
CF(xs) + C3F7+ → EP + C3F7# 0.6 50 0.5 500 1
CF(xs) + C3F7+ → CF(s) + C3F7# 0.3 8 0.5 500 1
CF(xs) + C3F7# → EP + C3F7# 0.6 50 0.5 500 1
CF(xs) + C3F7# → CF(s) + C3F7# 0.3 8 0.5 500 1
CF(xs) + C4F7+ → EP + C4F7# 0.6 50 0.5 500 1
CF(xs) + C4F7+ → CF(s) + C4F7# 0.3 8 0.5 500 1
CF(xs) + C4F7# → EP + C4F7# 0.6 50 0.5 500 1
CF(xs) + C4F7# → CF(s) + C4F7# 0.3 8 0.5 500 1
CF(xs) + C4F8+ → EP + C4F8# 0.6 50 0.5 500 1
CF(xs) + C4F8+ → CF(s) + C4F8# 0.3 8 0.5 500 1
CF(xs) + C4F8# → EP + C4F8# 0.6 50 0.5 500 1
CF(xs) + C4F8# → CF(s) + C4F8# 0.3 8 0.5 500 1
CF(xs) + O+ → EP + O# 0.6 50 0.5 500 1
CF(xs) + O+ → CF(s) + O# 0.3 8 0.5 500 1
CF(xs) + O2+ → EP + O2 # 0.6 50 0.5 500 1
CF(xs) + O2+ → CF(s) + O2# 0.3 8 0.5 500 1
CF(xs) + O# → EP + O# 0.6 50 0.5 500 1
CF(xs) + O# → CF(s) + O# 0.3 8 0.5 500 1
CF(xs) + O# → CF + O# 0.1
CF(xs) + O2# → EP + O2 # 0.6 50 0.5 500 1
CF(xs) + O2# → CF(s) + O2# 0.3 8 0.5 500 1
CF2(xs) + Ar+ → EP + Ar# 0.6 50 0.5 500 1
CF2(xs) + Ar+ → CF2(s) + Ar# 0.3 8 0.5 500 1
CF2(xs) + Ar# → EP + Ar# 0.6 50 0.5 500 1
CF2(xs) + Ar# → CF2(s) + Ar# 0.3 8 0.5 500 1
CF2(xs) + CF3+ → EP + CF3# 0.6 50 0.5 500 1
CF2(xs) + CF3+ → CF2(s) + CF3# 0.3 8 0.5 500 1
CF2(xs) + CF3# → EP + CF3# 0.6 50 0.5 500 1
CF2(xs) + CF3# → CF2(s) + CF3# 0.3 8 0.5 500 1
CF2(xs) + CF2+ → EP + CF2# 0.6 50 0.5 500 1
CF2(xs) + CF2+ → CF2(s) + CF3# 0.3 8 0.5 500 1
CF2(xs) + CF2# → EP + CF2# 0.6 50 0.5 500 1
CF2(xs) + CF2# → CF2(s) + CF2# 0.3 8 0.5 500 1
CF2(xs) + CF+ → EP + CF# 0.6 50 0.5 500 1
CF2(xs) + CF+ → CF2(s) + CF# 0.3 8 0.5 500 1
CF2(xs) + CF# → EP + CF# 0.6 50 0.5 500 1
CF2(xs) + CF# → CF2(s) + CF# 0.3 8 0.5 500 1
CF2(xs) + F+ → EP + F# 0.6 50 0.5 500 1
CF2(xs) + F+ → CF2(s) + F# 0.3 8 0.5 500 1

242
CF2(xs) + F# → EP + F# 0.6 50 0.5 500 1
CF2(xs) + F# → CF2(s) + F# 0.3 8 0.5 500 1
CF2(xs) + F2+ → EP + F2# 0.6 50 0.5 500 1
CF2(xs) + F2+ → CF2(s) + F2# 0.3 8 0.5 500 1
CF2(xs) + F2# → EP + F2# 0.6 50 0.5 500 1
CF2(xs) + F2# → CF2(s) + F2# 0.3 8 0.5 500 1
CF2(xs) + C2F5+ → EP + C2F5# 0.6 50 0.5 500 1
CF2(xs) + C2F5+ → CF2(s) + C2F5# 0.3 8 0.5 500 1
CF2(xs) + C2F5# → EP + C2F5# 0.6 50 0.5 500 1
CF2(xs) + C2F5# → CF2(s) + C2F5# 0.3 8 0.5 500 1
CF2(xs) + C2F4+ → EP + C2F4# 0.6 50 0.5 500 1
CF2(xs) + C2F4+ → CF2(s) + C2F4# 0.3 8 0.5 500 1
CF2(xs) + C2F4# → EP + C2F4# 0.6 50 0.5 500 1
CF2(xs) + C2F4# → CF2(s) + C2F4# 0.3 8 0.5 500 1
CF2(xs) + C2F3+ → EP + C2F3# 0.6 50 0.5 500 1
CF2(xs) + C2F3+ → CF2(s) + C2F3# 0.3 8 0.5 500 1
CF2(xs) + C2F3# → EP + C2F3# 0.6 50 0.5 500 1
CF2(xs) + C2F3# → CF2(s) + C2F3# 0.3 8 0.5 500 1
CF2(xs) + C3F5+ → EP + C3F5# 0.6 50 0.5 500 1
CF2(xs) + C3F5+ → CF2(s) + C3F5# 0.3 8 0.5 500 1
CF2(xs) + C3F5# → EP + C3F5# 0.6 50 0.5 500 1
CF2(xs) + C3F5# → CF2(s) + C3F5# 0.3 8 0.5 500 1
CF2(xs) + C3F6+ → EP + C3F6# 0.6 50 0.5 500 1
CF2(xs) + C3F6+ → CF2(s) + C3F6# 0.3 8 0.5 500 1
CF2(xs) + C3F6# → EP + C3F6# 0.6 50 0.5 500 1
CF2(xs) + C3F6+ → CF2(s) + C3F6# 0.3 8 0.5 500 1
CF2(xs) + C3F7+ → EP + C3F7# 0.6 50 0.5 500 1
CF2(xs) + C3F7+ → CF2(s) + C3F7# 0.3 8 0.5 500 1
CF2(xs) + C3F7# → EP + C3F7# 0.6 50 0.5 500 1
CF2(xs) + C3F7# → CF2(s) + C3F7# 0.3 8 0.5 500 1
CF2(xs) + C4F7+ → EP + C4F7# 0.6 50 0.5 500 1
CF2(xs) + C4F7+ → CF2(s) + C4F7# 0.3 8 0.5 500 1
CF2(xs) + C4F7# → EP + C4F7# 0.6 50 0.5 500 1
CF2(xs) + C4F7# → CF2(s) + C4F7# 0.3 8 0.5 500 1
CF2(xs) + C4F8+ → EP + C4F8# 0.6 50 0.5 500 1
CF2(xs) + C4F8+ → CF2(s) + C4F8# 0.3 8 0.5 500 1
CF2(xs) + C4F8# → EP + C4F8# 0.6 50 0.5 500 1
CF2(xs) + C4F8# → CF2(s) + C4F8# 0.3 8 0.5 500 1
CF2(xs) + O+ → EP + O# 0.6 50 0.5 500 1
CF2(xs) + O+ → CF2(s) + O# 0.3 8 0.5 500 1
CF2(xs) + O2+ → EP + O2 # 0.6 50 0.5 500 1

243
CF2(xs) + O2+ → CF2(s) + O2# 0.3 8 0.5 500 1
CF2(xs) + O# → EP + O# 0.6 50 0.5 500 1
CF2(xs) + O2+ → CF2(s) + O2# 0.3 8 0.5 500 1
CF2(xs) + O# → CF + O# 0.1
CF2(xs) + O2# → EP + O2 # 0.6 50 0.5 500 1
CF2(xs) + O2+ → CF2(s) + O2# 0.3 8 0.5 500 1
CF3(xs) + Ar+ → EP + Ar# 0.6 50 0.5 500 1
CF3(xs) + Ar+ → CF3(s) + Ar# 0.3 8 0.5 500 1
CF3(xs) + Ar# → EP + Ar# 0.6 50 0.5 500 1
CF3(xs) + Ar# → CF3(s) + Ar# 0.3 8 0.5 500 1
CF3(xs) + CF3+ → EP + CF3# 0.6 50 0.5 500 1
CF3(xs) + CF3+ → CF3(s) + CF3# 0.3 8 0.5 500 1
CF3(xs) + CF3# → EP + CF3# 0.6 50 0.5 500 1
CF3(xs) + CF3# → CF3(s) + CF3# 0.3 8 0.5 500 1
CF3(xs) + CF2+ → EP + CF2# 0.6 50 0.5 500 1
CF3(xs) + CF2+ → CF3(s) + CF3# 0.3 8 0.5 500 1
CF3(xs) + CF2# → EP + CF2# 0.6 50 0.5 500 1
CF3(xs) + CF2# → CF3(s) + CF2# 0.3 8 0.5 500 1
CF3(xs) + CF+ → EP + CF# 0.6 50 0.5 500 1
CF3(xs) + CF+ → CF3(s) + CF# 0.3 8 0.5 500 1
CF3(xs) + CF# → EP + CF# 0.6 50 0.5 500 1
CF3(xs) + CF# → CF3(s) + CF# 0.3 8 0.5 500 1
CF3(xs) + F+ → EP + F# 0.6 50 0.5 500 1
CF3(xs) + F+ → CF3(s) + F# 0.3 8 0.5 500 1
CF3(xs) + F# → EP + F# 0.6 50 0.5 500 1
CF3(xs) + F# → CF3(s) + F# 0.3 8 0.5 500 1
CF3(xs) + F2+ → EP + F2# 0.6 50 0.5 500 1
CF3(xs) + F2+ → CF3(s) + F2# 0.3 8 0.5 500 1
CF3(xs) + F2# → EP + F2# 0.6 50 0.5 500 1
CF3(xs) + F2# → CF3(s) + F2# 0.3 8 0.5 500 1
CF3(xs) + C2F5+ → EP + C2F5# 0.6 50 0.5 500 1
CF3(xs) + C2F5+ → CF3(s) + C2F5# 0.3 8 0.5 500 1
CF3(xs) + C2F5# → EP + C2F5# 0.6 50 0.5 500 1
CF3(xs) + C2F5# → CF3(s) + C2F5# 0.3 8 0.5 500 1
CF3(xs) + C2F4+ → EP + C2F4# 0.6 50 0.5 500 1
CF3(xs) + C2F4+ → CF3(s) + C2F4# 0.3 8 0.5 500 1
CF3(xs) + C2F4# → EP + C2F4# 0.6 50 0.5 500 1
CF3(xs) + C2F4# → CF3(s) + C2F4# 0.3 8 0.5 500 1
CF3(xs) + C2F3+ → EP + C2F3# 0.6 50 0.5 500 1
CF3(xs) + C2F3+ → CF3(s) + C2F3# 0.3 8 0.5 500 1
CF3(xs) + C2F3# → EP + C2F3# 0.6 50 0.5 500 1

244
CF3(xs) + C2F3# → CF3(s) + C2F3# 0.3 8 0.5 500 1
CF3(xs) + C3F5+ → EP + C3F5# 0.6 50 0.5 500 1
CF3(xs) + C3F5+ → CF3(s) + C3F5# 0.3 8 0.5 500 1
CF3(xs) + C3F5# → EP + C3F5# 0.6 50 0.5 500 1
CF3(xs) + C3F5# → CF3(s) + C3F5# 0.3 8 0.5 500 1
CF3(xs) + C3F6+ → EP + C3F6# 0.6 50 0.5 500 1
CF3(xs) + C3F6+ → CF3(s) + C3F6# 0.3 8 0.5 500 1
CF3(xs) + C3F6# → EP + C3F6# 0.6 50 0.5 500 1
CF3(xs) + C3F6+ → CF3(s) + C3F6# 0.3 8 0.5 500 1
CF3(xs) + C3F7+ → EP + C3F7# 0.6 50 0.5 500 1
CF3(xs) + C3F7+ → CF3(s) + C3F7# 0.3 8 0.5 500 1
CF3(xs) + C3F7# → EP + C3F7# 0.6 50 0.5 500 1
CF3(xs) + C3F7# → CF3(s) + C3F7# 0.3 8 0.5 500 1
CF3(xs) + C4F7+ → EP + C4F7# 0.6 50 0.5 500 1
CF3(xs) + C4F7+ → CF3(s) + C4F7# 0.3 8 0.5 500 1
CF3(xs) + C4F7# → EP + C4F7# 0.6 50 0.5 500 1
CF3(xs) + C4F7# → CF3(s) + C4F7# 0.3 8 0.5 500 1
CF3(xs) + C4F8+ → EP + C4F8# 0.6 50 0.5 500 1
CF3(xs) + C4F8+ → CF3(s) + C4F8# 0.3 8 0.5 500 1
CF3(xs) + C4F8# → EP + C4F8# 0.6 50 0.5 500 1
CF3(xs) + C4F8# → CF3(s) + C4F8# 0.3 8 0.5 500 1
CF3(xs) + O+ → EP + O# 0.6 50 0.5 500 1
CF3(xs) + O+ → CF3(s) + O# 0.3 8 0.5 500 1
CF3(xs) + O2+ → EP + O2 # 0.6 50 0.5 500 1
CF3(xs) + O2+ → CF3(s) + O2# 0.3 8 0.5 500 1
CF3(xs) + O# → EP + O# 0.6 50 0.5 500 1
CF3(xs) + O2+ → CF3(s) + O2# 0.3 8 0.5 500 1
CF3(xs) + O# → CF + O# 0.1
CF3(xs) + O2# → EP + O2 # 0.6 50 0.5 500 1
CF3(xs) + O2+ → CF3(s) + O2# 0.3 8 0.5 500 1
AC(s) +F → CF(s) 0.01
AC(xs) +F → CF(s) 0.001
CF(s) +F → CF2(s) 0.01
CF2(s) +F → CF3(s) 0.01
CF3(s) +F → CF4 0.01
AC(s) +C → AC(s) + AC(s) 0.01
SiO2(s) +C → SiO2(s) + AC(s) 0.01
CF2(s) +C → CF2(s) + AC(s) 0.01
SiO2(s) + Ar+ → SiO2*(s) + Ar# 0.9
SiO2(s) + Ar+ → Ar# + SiO2 0.0852 70 1 140 1
SiO2(s) + F+ → SiO2*(s) + F# 0.9

245
SiO2(s) + F+ → F# + SiO2 0.0852 70 1 140 1
SiO2(s) + F2+ → SiO2*(s) + F2# 0.9
SiO2(s) + F2+ → F2# SiO2 + F2# 0.0852 70 1 140 1
SiO2(s) + O2+ → SiO2*(s) + O2# 0.9
SiO2(s) + O2+ → O2# + SiO2 0.0852 70 1 140 1
SiO2(s) + O+ → SiO2*(s) + O# 0.9
SiO2(s) + O+ → O# + SiO2 0.0852 70 1 140 1
SiO2(s) + CF3+ → SiO2*(s) + CF3# 0.9
SiO2(s) + CF3+ → CF3# + SiO2 0.0852 70 1 140 1
SiO2(s) + CF2+ → SiO2*(s) + CF2# 0.9
SiO2(s) + CF2+ → CF2# + SiO2 0.0852 70 1 140 1
SiO2(s) + CF+ → SiO2*(s) + CF# 0.9
SiO2(s) + CF+ → CF# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F5+ → SiO2*(s) + C2F5# 0.9
SiO2(s) + C2F5+ → C2F5# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F4+ → SiO2*(s) + C2F4# 0.9
SiO2(s) + C2F4+ → C2F4# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F3+ → SiO2*(s) + C2F3# 0.9
SiO2(s) + C2F3+ → C2F3# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F5+ → SiO2*(s) + C3F5# 0.9
SiO2(s) + C3F5+ → C3F5# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F6+ → SiO2*(s) + C3F6# 0.9
SiO2(s) + C3F6+ → C3F6# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F7+ → SiO2*(s) + C3F7# 0.9
SiO2(s) + C3F7+ → C3F7# + SiO2 0.0852 70 1 140 1
SiO2(s) + C4F7+ → SiO2*(s) + C4F7# 0.9
SiO2(s) + C4F7+ → SiO2 + C4F7# 0.0852 70 1 140 1
SiO2(s) + C4F8+ → SiO2*(s) + C4F8# 0.9
SiO2(s) + C4F8+ → C4F8# + SiO2 0.0852 70 1 140 1
SiO2(s) + Ar# → SiO2*(s) + Ar# 0.9
SiO2(s) + Ar# → Ar# + SiO2 0.0852 70 1 140 1
SiO2(s) + F# → SiO2*(s) + F# 0.9
SiO2(s) + F# → F# + SiO2 0.0852 70 1 140 1
SiO2(s) + F2# → SiO2*(s) + F2# 0.9
SiO2(s) + F2# → F2# SiO2 + F2# 0.0852 70 1 140 1
SiO2(s) + O2# → SiO2*(s) + O2# 0.9
SiO2(s) + O2# → O2# + SiO2 0.0852 70 1 140 1
SiO2(s) + O# → SiO2*(s) + O# 0.9
SiO2(s) + O# → O# + SiO2 0.0852 70 1 140 1
SiO2(s) + CF3# → SiO2*(s) + CF3# 0.9
SiO2(s) + CF3# → CF3# + SiO2 0.0852 70 1 140 1

246
SiO2(s) + CF2# → SiO2*(s) + CF2# 0.9
SiO2(s) + CF2# → CF2# + SiO2 0.0852 70 1 140 1
SiO2(s) + CF# → SiO2*(s) + CF# 0.9
SiO2(s) + CF# → CF# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F5# → SiO2*(s) + C2F5# 0.9
SiO2(s) + C2F5# → C2F5# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F4# → SiO2*(s) + C2F4# 0.9
SiO2(s) + C2F4# → C2F4# + SiO2 0.0852 70 1 140 1
SiO2(s) + C2F3# → SiO2*(s) + C2F3# 0.9
SiO2(s) + C2F3# → C2F3# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F5# → SiO2*(s) + C3F5# 0.9
SiO2(s) + C3F5# → C3F5# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F6# → SiO2*(s) + C3F6# 0.9
SiO2(s) + C3F6# → C3F6# + SiO2 0.0852 70 1 140 1
SiO2(s) + C3F7# → SiO2*(s) + C3F7# 0.9
SiO2(s) + C3F7# → C3F7# + SiO2 0.0852 70 1 140 1
SiO2(s) + C4F7# → SiO2*(s) + C4F7# 0.9
SiO2(s) + C4F7# → SiO2 + C4F7# 0.0852 70 1 140 1
SiO2(s) + C4F8# → SiO2*(s) + C4F8# 0.9
SiO2(s) + C4F8# → C4F8# + SiO2 0.0852 70 1 140 1
SiO2*(s) + Ar+ → Ar #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + CF3+ → CF3 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + CF2+ → CF2 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + CF+ → CF #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + F2+ → F2 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + F+ → F#
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + O2+ → O2 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + O+ → SiO2 + O# 0.0852 70 1 140 1
SiO2*(s) + O+ → O# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F4+ → C2F4 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F3+ → C2F3 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F3+ → SiO2 + C2F3# 0.0852 70 1 140 1
SiO2*(s) + C2F5+ → C2F5# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F5+ → C3F5 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F6+ → C3F6 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F7+ → C3F7# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C4F7+ → C4F7 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C4F8+ → C4F8 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + Ar# → Ar #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + CF3# → CF3 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + CF2# → CF2 #
+ SiO2 0.0852 70 1 140 1

247
SiO2*(s) + CF# → CF# + SiO2 0.0852 70 1 140 1
SiO2*(s) + F2# → F2 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + F# → F#
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + O2# → O2# + SiO2 0.0852 70 1 140 1
SiO2*(s) + O# → SiO2 + O# 0.0852 70 1 140 1
SiO2*(s) + O# → O# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F4# → C2F4 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F3# → C2F3 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C2F3# → SiO2 + C2F3# 0.0852 70 1 140 1
SiO2*(s) + C2F5# → C2F5# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F5# → C3F5 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F6# → C3F6# + SiO2 0.0852 70 1 140 1
SiO2*(s) + C3F7# → C3F7 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C4F7# → C4F7 #
+ SiO2 0.0852 70 1 140 1
SiO2*(s) + C4F8# → C4F8 #
+ SiO2 0.0852 70 1 140 1
SiO2(s) + CF → SiO2CF(s) 0.278
SiO2(s) + CF2 → SiO2CF2(s) 0.278
SiO2(s) + CF3 → SiO2CF3(s) 0.2
SiO2(s) + C2F3 → SiO2C2F3(s) 0.2
SiO2(s) + C2F4 → SiO2C2F4(s) 0.001
SiO2(s) + C3F5 → SiO2C3F5(s) 0.001
SiO2(s) + C3F6 → SiO2C3F6(s) 0.001
SiO2*(s) + CF → SiO2CF(s) 0.8
SiO2*(s) + CF → SiO2(s) + CF 0.2
SiO2*(s) + CF2 → SiO2CF2(s) 0.85
SiO2*(s) + CF2 → SiO2(s) + CF2 0.15
SiO2*(s) + CF3 → SiO2CF3(s) 0.9
SiO2*(s) + CF3 → SiO2(s) + CF3 0.1
SiO2*(s) + C2F3 → SiO2C2F3(s) 0.9
SiO2*(s) + C2F3 → SiO2(s) + C2F3 0.1
SiO2*(s) + F → SiO2(s) +F 1
SiO2*(s) + O → SiO2(s) +O 1
SiO2CF(s) + Ar+ → SiF + CO2 + Ar# 0.1471 35 1 140 2
SiO2CF2(s) + Ar+ → CO2 + Ar# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + Ar+ → CO2 + Ar# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + Ar+ → SiOCF3(s) + CO + Ar# 0.1471 35 1 140 2
SiO2C2F4(s) + Ar+ → SiOCF3(s) + COF + Ar# 0.1471 35 1 140 2
SiO2C3F5(s) + Ar+ → SiO2CF(s) + C2F4 + Ar# 0.1471 35 1 140 2
SiO2C3F6(s) + Ar+ → SiO2CF3(s) + C2F3 + Ar# 0.1471 35 1 140 2
SiO2CF*(s) + Ar+ → CO2 + Ar# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + Ar+ → CO2 + Ar# + SiF2 0.1471 35 1 140 2

248
SiO2CF3*(s) + Ar+ → CO2 + Ar# + SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + Ar+ → SiOCF3*(s) + CO + Ar# 0.1471 35 1 140 2
SiO2C2F4*(s) + Ar+ → SiOCF3*(s) + COF + Ar# 0.1471 35 1 140 2
SiO2C3F5*(s) + Ar+ → SiO2CF*(s) + C2F4 + Ar# 0.1471 35 1 140 2
SiO2C3F6*(s) + Ar+ → SiO2CF3*(s) + C2F3 + Ar# 0.1471 35 1 140 2
SiO2CF(s) + Ar# → SiF + CO2 + Ar# 0.1471 35 1 140 2
SiO2CF2(s) + Ar# → CO2 + Ar# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + Ar# → CO2 + Ar #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + Ar# → SiOCF3(s) + CO + Ar# 0.1471 35 1 140 2
SiO2C2F4(s) + Ar# → SiOCF3(s) + COF + Ar# 0.1471 35 1 140 2
SiO2C3F5(s) + Ar# → SiO2CF(s) + C2F4 + Ar# 0.1471 35 1 140 2
SiO2C3F6(s) + Ar# → SiO2CF3(s) + C2F3 + Ar# 0.1471 35 1 140 2
SiO2CF*(s) + Ar# → CO2 + Ar# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + Ar# → CO2 + Ar #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + Ar# → CO2 + Ar #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + Ar# → SiOCF3*(s) + CO + Ar# 0.1471 35 1 140 2
SiO2C2F4*(s) + Ar# → SiOCF3*(s) + COF + Ar# 0.1471 35 1 140 2
SiO2C3F5*(s) + Ar# → SiO2CF*(s) + C2F4 + Ar# 0.1471 35 1 140 2
SiO2C3F6*(s) + Ar# → SiO2CF3*(s) + C2F3 + Ar# 0.1471 35 1 140 2
SiO2CF(s) + O+ → SiF + CO2 + O# 0.1471 35 1 140 2
SiO2CF2(s) + O+ → CO2 + O# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + O+ → CO2 +O #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + O+ → SiOCF3(s) + CO + O# 0.1471 35 1 140 2
SiO2C2F4(s) + O+ → SiOCF3(s) + COF + O# 0.1471 35 1 140 2
SiO2C3F5(s) + O+ → SiO2CF(s) + C2F4 + O# 0.1471 35 1 140 2
SiO2C3F6(s) + O+ → SiO2CF3(s) + C2F3 + O# 0.1471 35 1 140 2
SiO2CF*(s) + O+ → CO2 +O #
+ SiF 0.1471 35 1 140 2
SiO2CF2*(s) + O+ → CO2 + O# + SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + O+ → CO2 +O #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + O+ → SiOCF3*(s) + CO + O# 0.1471 35 1 140 2
SiO2C2F4*(s) + O+ → SiOCF3*(s) + COF + O# 0.1471 35 1 140 2
SiO2C3F5*(s) + O+ → SiO2CF*(s) + C2F4 + O# 0.1471 35 1 140 2
SiO2C3F6*(s) + O+ → SiO2CF3*(s) + C2F3 + O# 0.1471 35 1 140 2
SiO2CF(s) + O# → SiF + CO2 + O# 0.1471 35 1 140 2
SiO2CF2(s) + O# → CO2 +O #
+ SiF2 0.1471 35 1 140 2
SiO2CF3(s) + O# → CO2 + O# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + O# → SiOCF3(s) + CO + O# 0.1471 35 1 140 2
SiO2C2F4(s) + O# → SiOCF3(s) + COF + O# 0.1471 35 1 140 2
SiO2C3F5(s) + O# → SiO2CF(s) + C2F4 + O# 0.1471 35 1 140 2
SiO2C3F6(s) + O# → SiO2CF3(s) + C2F3 + O# 0.1471 35 1 140 2
SiO2CF*(s) + O# → CO2 + O# + SiF 0.1471 35 1 140 2

249
SiO2CF2*(s) + O# → CO2 + O# + SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + O# → CO2 +O #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + O# → SiOCF3*(s) + CO + O# 0.1471 35 1 140 2
SiO2C2F4*(s) + O# → SiOCF3*(s) + COF + O# 0.1471 35 1 140 2
SiO2C3F5*(s) + O# → SiO2CF*(s) + C2F4 + O# 0.1471 35 1 140 2
SiO2C3F6*(s) + O# → SiO2CF3*(s) + C2F3 + O# 0.1471 35 1 140 2
SiO2CF(s) + O2+ → SiF + CO2 + O2# 0.1471 35 1 140 2
SiO2CF2(s) + O2+ → CO2 + O2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + O2+ → CO2 + O2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + O2+ → SiOCF3(s) + CO + O2 # 0.1471 35 1 140 2
SiO2C2F4(s) + O2+ → SiOCF3(s) + COF + O2 # 0.1471 35 1 140 2
SiO2C3F5(s) + O2+ → SiO2CF(s) + C2F4 + O2# 0.1471 35 1 140 2
SiO2C3F6(s) + O2+ → SiO2CF3(s) + C2F3 + O2# 0.1471 35 1 140 2
SiO2CF*(s) + O2+ → CO2 + O2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + O2+ → CO2 + O2 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + O2+ → CO2 + O2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + O2+ → SiOCF3*(s) + CO + O2# 0.1471 35 1 140 2
SiO2C2F4*(s) + O2+ → SiOCF3*(s) + COF + O2# 0.1471 35 1 140 2
SiO2C3F5*(s) + O2+ → SiO2CF*(s) + C2F4 + O2# 0.1471 35 1 140 2
SiO2C3F6*(s) + O2+ → SiO2CF3*(s) + C2F3 + O2# 0.1471 35 1 140 2
SiO2CF(s) + O2# → SiF + CO2 + O2# 0.1471 35 1 140 2
SiO2CF2(s) + O2# → CO2 + O2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + O2# → CO2 + O2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + O2# → SiOCF3(s) + CO + O2 # 0.1471 35 1 140 2
SiO2C2F4(s) + O2# → SiOCF3(s) + COF + O2 # 0.1471 35 1 140 2
SiO2C3F5(s) + O2# → SiO2CF(s) + C2F4 + O2# 0.1471 35 1 140 2
SiO2C3F6(s) + O2# → SiO2CF3(s) + C2F3 + O2# 0.1471 35 1 140 2
SiO2CF*(s) + O2# → CO2 + O2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + O2# → CO2 + O2 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + O2# → CO2 + O2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + O2# → SiOCF3*(s) + CO + O2# 0.1471 35 1 140 2
SiO2C2F4*(s) + O2# → SiOCF3*(s) + COF + O2# 0.1471 35 1 140 2
SiO2C3F5*(s) + O2# → SiO2CF*(s) + C2F4 + O2# 0.1471 35 1 140 2
SiO2C3F6*(s) + O2# → SiO2CF3*(s) + C2F3 + O2# 0.1471 35 1 140 2
SiO2CF(s) + CF3+ → SiF + CO2 + CF3# 0.1471 35 1 140 2
SiO2CF2(s) + CF3+ → CO2 + CF3# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF3+ → CO2 + CF3 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF3+ → SiOCF3(s) + CO + CF3# 0.1471 35 1 140 2
SiO2C2F4(s) + CF3+ → SiOCF3(s) + COF + CF3# 0.1471 35 1 140 2
SiO2C3F5(s) + CF3+ → SiO2CF(s) + C2F4 + CF3# 0.1471 35 1 140 2
SiO2C3F6(s) + CF3+ → SiO2CF3(s) + C2F3 + CF3# 0.1471 35 1 140 2

250
SiO2CF*(s) + CF3+ → CO2 + CF3# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF3+ → CO2 + CF3 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF3+ → CO2 + CF3 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF3+ → SiOCF3*(s) + CO + CF3# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF3+ → SiOCF3*(s) + COF + CF3# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF3+ → SiO2CF*(s) + C2F4 + CF3# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF3+ → SiO2CF3*(s) + C2F3 + CF3# 0.1471 35 1 140 2
SiO2CF(s) + CF3# → SiF + CO2 + CF3# 0.1471 35 1 140 2
SiO2CF2(s) + CF3# → CO2 + CF3# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF3# → CO2 + CF3 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF3# → SiOCF3(s) + CO + CF3# 0.1471 35 1 140 2
SiO2C2F4(s) + CF3# → SiOCF3(s) + COF + CF3# 0.1471 35 1 140 2
SiO2C3F5(s) + CF3# → SiO2CF(s) + C2F4 + CF3# 0.1471 35 1 140 2
SiO2C3F6(s) + CF3# → SiO2CF3(s) + C2F3 + CF3# 0.1471 35 1 140 2
SiO2CF*(s) + CF3# → CO2 + CF3# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF3# → CO2 + CF3 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF3# → CO2 + CF3 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF3# → SiOCF3*(s) + CO + CF3# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF3# → SiOCF3*(s) + COF + CF3# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF3# → SiO2CF*(s) + C2F4 + CF3# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF3# → SiO2CF3*(s) + C2F3 + CF3# 0.1471 35 1 140 2
SiO2CF(s) + CF2+ → SiF + CO2 + CF2# 0.1471 35 1 140 2
SiO2CF2(s) + CF2+ → CO2 + CF2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF2+ → CO2 + CF2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF2+ → SiOCF3(s) + CO + CF2# 0.1471 35 1 140 2
SiO2C2F4(s) + CF2+ → SiOCF3(s) + COF + CF2# 0.1471 35 1 140 2
SiO2C3F5(s) + CF2+ → SiO2CF(s) + C2F4 + CF2# 0.1471 35 1 140 2
SiO2C3F6(s) + CF2+ → SiO2CF3(s) + C2F3 + CF2# 0.1471 35 1 140 2
SiO2CF*(s) + CF2+ → CO2 + CF2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF2+ → CO2 + CF2 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF2+ → CO2 + CF2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF2+ → SiOCF3*(s) + CO + CF2# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF2+ → SiOCF3*(s) + COF + CF2# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF2+ → SiO2CF*(s) + C2F4 + CF2# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF2+ → SiO2CF3*(s) + C2F3 + CF2# 0.1471 35 1 140 2
SiO2CF(s) + CF2# → SiF + CO2 + CF2# 0.1471 35 1 140 2
SiO2CF2(s) + CF2# → CO2 + CF2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF2# → CO2 + CF2 #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF2# → SiOCF3(s) + CO + CF2# 0.1471 35 1 140 2
SiO2C2F4(s) + CF2# → SiOCF3(s) + COF + CF2# 0.1471 35 1 140 2
SiO2C3F5(s) + CF2# → SiO2CF(s) + C2F4 + CF2# 0.1471 35 1 140 2

251
SiO2C3F6(s) + CF2# → SiO2CF3(s) + C2F3 + CF2# 0.1471 35 1 140 2
SiO2CF*(s) + CF2# → CO2 + CF2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF2# → CO2 + CF2 #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF2# → CO2 + CF2# + SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF2# → SiOCF3*(s) + CO + CF2# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF2# → SiOCF3*(s) + COF + CF2# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF2# → SiO2CF*(s) + C2F4 + CF2# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF2# → SiO2CF3*(s) + C2F3 + CF2# 0.1471 35 1 140 2
SiO2CF(s) + CF+ → SiF + CO2 + CF# 0.1471 35 1 140 2
SiO2CF2(s) + CF+ → CO2 + CF# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF+ → CO2 + CF #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF+ → SiOCF3(s) + CO + CF# 0.1471 35 1 140 2
SiO2C2F4(s) + CF+ → SiOCF3(s) + COF + CF# 0.1471 35 1 140 2
SiO2C3F5(s) + CF+ → SiO2CF(s) + C2F4 + CF# 0.1471 35 1 140 2
SiO2C3F6(s) + CF+ → SiO2CF3(s) + C2F3 + CF# 0.1471 35 1 140 2
SiO2CF*(s) + CF+ → CO2 + CF# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF+ → CO2 + CF #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF+ → CO2 + CF #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF+ → SiOCF3*(s) + CO + CF# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF+ → SiOCF3*(s) + COF + CF# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF+ → SiO2CF*(s) + C2F4 + CF# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF+ → SiO2CF3*(s) + C2F3 + CF# 0.1471 35 1 140 2
SiO2CF(s) + CF# → SiF + CO2 + CF# 0.1471 35 1 140 2
SiO2CF2(s) + CF# → CO2 + CF# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + CF# → CO2 + CF #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + CF# → SiOCF3(s) + CO + CF# 0.1471 35 1 140 2
SiO2C2F4(s) + CF# → SiOCF3(s) + COF + CF# 0.1471 35 1 140 2
SiO2C3F5(s) + CF# → SiO2CF(s) + C2F4 + CF# 0.1471 35 1 140 2
SiO2C3F6(s) + CF# → SiO2CF3(s) + C2F3 + CF# 0.1471 35 1 140 2
SiO2CF*(s) + CF# → CO2 + CF# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + CF# → CO2 + CF #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + CF# → CO2 + CF #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + CF# → SiOCF3*(s) + CO + CF# 0.1471 35 1 140 2
SiO2C2F4*(s) + CF# → SiOCF3*(s) + COF + CF# 0.1471 35 1 140 2
SiO2C3F5*(s) + CF# → SiO2CF*(s) + C2F4 + CF# 0.1471 35 1 140 2
SiO2C3F6*(s) + CF# → SiO2CF3*(s) + C2F3 + CF# 0.1471 35 1 140 2
SiO2CF(s) + F+ → SiF + CO2 + F# 0.1471 35 1 140 2
SiO2CF2(s) + F+ → CO2 + F# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + F+ → CO2 +F #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + F+ → SiOCF3(s) + CO + F# 0.1471 35 1 140 2
SiO2C2F4(s) + F+ → SiOCF3(s) + COF + F# 0.1471 35 1 140 2

252
SiO2C3F5(s) + F+ → SiO2CF(s) + C2F4 + F# 0.1471 35 1 140 2
SiO2C3F6(s) + F+ → SiO2CF3(s) + C2F3 + F# 0.1471 35 1 140 2
SiO2CF*(s) + F+ → CO2 +F #
+ SiF 0.1471 35 1 140 2
SiO2CF2*(s) + F+ → CO2 + F# + SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + F+ → CO2 +F #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + F+ → SiOCF3*(s) + CO + F# 0.1471 35 1 140 2
SiO2C2F4*(s) + F+ → SiOCF3*(s) + COF + F# 0.1471 35 1 140 2
SiO2C3F5*(s) + F+ → SiO2CF*(s) + C2F4 + F# 0.1471 35 1 140 2
SiO2C3F6*(s) + F+ → SiO2CF3*(s) + C2F3 + F# 0.1471 35 1 140 2
SiO2CF(s) + F# → SiF + CO2 + F# 0.1471 35 1 140 2
SiO2CF2(s) + F# → CO2 +F #
+ SiF2 0.1471 35 1 140 2
SiO2CF3(s) + F# → CO2 + F# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + F# → SiOCF3(s) + CO + F# 0.1471 35 1 140 2
SiO2C2F4(s) + F# → SiOCF3(s) + COF + F# 0.1471 35 1 140 2
SiO2C3F5(s) + F# → SiO2CF(s) + C2F4 + F# 0.1471 35 1 140 2
SiO2C3F6(s) + F# → SiO2CF3(s) + C2F3 + F# 0.1471 35 1 140 2
SiO2CF*(s) + F# → CO2 + F# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + F# → CO2 +F #
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + F# → CO2 +F #
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + F# → SiOCF3*(s) + CO + F# 0.1471 35 1 140 2
SiO2C2F4*(s) + F# → SiOCF3*(s) + COF + F# 0.1471 35 1 140 2
SiO2C3F5*(s) + F# → SiO2CF*(s) + C2F4 + F# 0.1471 35 1 140 2
SiO2C3F6*(s) + F# → SiO2CF3*(s) + C2F3 + F# 0.1471 35 1 140 2
SiO2CF(s) + F2+ → SiF + CO2 + F2# 0.1471 35 1 140 2
SiO2CF2(s) + F2+ → CO2 + F2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + F2+ → CO2 + F2#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + F2+ → SiOCF3(s) + CO + F2# 0.1471 35 1 140 2
SiO2C2F4(s) + F2+ → SiOCF3(s) + COF + F2# 0.1471 35 1 140 2
SiO2C3F5(s) + F2+ → SiO2CF(s) + C2F4 + F2# 0.1471 35 1 140 2
SiO2C3F6(s) + F2+ → SiO2CF3(s) + C2F3 + F2# 0.1471 35 1 140 2
SiO2CF*(s) + F2+ → CO2 + F2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + F2+ → CO2 + F2#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + F2+ → CO2 + F2#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + F2+ → SiOCF3*(s) + CO + F2# 0.1471 35 1 140 2
SiO2C2F4*(s) + F2+ → SiOCF3*(s) + COF + F2# 0.1471 35 1 140 2
SiO2C3F5*(s) + F2+ → SiO2CF*(s) + C2F4 + F2# 0.1471 35 1 140 2
SiO2C3F6*(s) + F2+ → SiO2CF3*(s) + C2F3 + F2# 0.1471 35 1 140 2
SiO2CF(s) + F2# → SiF + CO2 + F2# 0.1471 35 1 140 2
SiO2CF2(s) + F2# → CO2 + F2# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + F2# → CO2 + F2#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + F2# → SiOCF3(s) + CO + F2# 0.1471 35 1 140 2

253
SiO2C2F4(s) + F2# → SiOCF3(s) + COF + F2# 0.1471 35 1 140 2
SiO2C3F5(s) + F2# → SiO2CF(s) + C2F4 + F2# 0.1471 35 1 140 2
SiO2C3F6(s) + F2# → SiO2CF3(s) + C2F3 + F2# 0.1471 35 1 140 2
SiO2CF*(s) + F2# → CO2 + F2# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + F2# → CO2 + F2#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + F2# → CO2 + F2#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + F2# → SiOCF3*(s) + CO + F2# 0.1471 35 1 140 2
SiO2C2F4*(s) + F2# → SiOCF3*(s) + COF + F2# 0.1471 35 1 140 2
SiO2C3F5*(s) + F2# → SiO2CF*(s) + C2F4 + F2# 0.1471 35 1 140 2
SiO2C3F6*(s) + F2# → SiO2CF3*(s) + C2F3 + F2# 0.1471 35 1 140 2
SiO2CF(s) + C2F5+ → SiF + CO2 + C2F5# 0.1471 35 1 140 2
SiO2CF2(s) + C2F5+ → CO2 + C2F5# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C2F5+ → CO2 + C2F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C2F5+ → SiOCF3(s) + CO + C2F5# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F5+ → SiOCF3(s) + COF + C2F5# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F5+ → SiO2CF(s) + C2F4 + C2F5# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F5+ → SiO2CF3(s) + C2F3 + C2F5# 0.1471 35 1 140 2
SiO2CF*(s) + C2F5+ → CO2 + C2F5# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F5+ → CO2 + C2F5#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F5+ → CO2 + C2F5# + SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F5+ → SiOCF3*(s) + CO + C2F5# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F5+ → SiOCF3*(s) + COF + C2F5# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F5+ → SiO2CF*(s) + C2F4 + C2F5# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F5+ → SiO2CF3*(s) + C2F3 + C2F5# 0.1471 35 1 140 2
SiO2CF(s) + C2F5# → SiF + CO2 + C2F5# 0.1471 35 1 140 2
SiO2CF2(s) + C2F5# → CO2 + C2F5# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C2F5# → CO2 + C2F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C2F5# → SiOCF3(s) + CO + C2F5# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F5# → SiOCF3(s) + COF + C2F5# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F5# → SiO2CF(s) + C2F4 + C2F5# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F5# → SiO2CF3(s) + C2F3 + C2F5# 0.1471 35 1 140 2
SiO2CF*(s) + C2F5# → CO2 + C2F5# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F5# → CO2 + C2F5#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F5# → CO2 + C2F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F5# → SiOCF3*(s) + CO + C2F5# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F5# → SiOCF3*(s) + COF + C2F5# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F5# → SiO2CF*(s) + C2F4 + C2F5# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F5# → SiO2CF3*(s) + C2F3 + C2F5# 0.1471 35 1 140 2
SiO2CF(s) + C2F4+ → SiF + CO2 + C2F4# 0.1471 35 1 140 2
SiO2CF2(s) + C2F4+ → CO2 + C2F4# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C2F4+ → CO2 + C2F4#
+ SiF3 0.1471 35 1 140 2

254
SiO2C2F3(s) + C2F4+ → SiOCF3(s) + CO + C2F4# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F4+ → SiOCF3(s) + COF + C2F4# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F4+ → SiO2CF(s) + C2F4 + C2F4# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F4+ → SiO2CF3(s) + C2F3 + C2F4# 0.1471 35 1 140 2
SiO2CF*(s) + C2F4+ → CO2 + C2F4# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F4+ → CO2 + C2F4#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F4+ → CO2 + C2F4#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F4+ → SiOCF3*(s) + CO + C2F4# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F4+ → SiOCF3*(s) + COF + C2F4# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F4+ → SiO2CF*(s) + C2F4 + C2F4# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F4+ → SiO2CF3*(s) + C2F3 + C2F4# 0.1471 35 1 140 2
SiO2CF(s) + C2F4# → SiF + CO2 + C2F4# 0.1471 35 1 140 2
SiO2CF2(s) + C2F4# → CO2 + C2F4# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C2F4# → CO2 + C2F4#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C2F4# → SiOCF3(s) + CO + C2F4# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F4# → SiOCF3(s) + COF + C2F4# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F4# → SiO2CF(s) + C2F4 + C2F4# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F4# → SiO2CF3(s) + C2F3 + C2F4# 0.1471 35 1 140 2
SiO2CF*(s) + C2F4# → CO2 + C2F4#
+ SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F4# → CO2 + C2F4# + SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F4# → CO2 + C2F4#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F4# → SiOCF3*(s) + CO + C2F4# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F4# → SiOCF3*(s) + COF + C2F4# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F4# → SiO2CF*(s) + C2F4 + C2F4# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F4# → SiO2CF3*(s) + C2F3 + C2F4# 0.1471 35 1 140 2
SiO2CF(s) + C2F3+ → SiF + CO2 + C2F3# 0.1471 35 1 140 2
SiO2CF2(s) + C2F3+ → CO2 + C2F3#
+ SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C2F3+ → CO2 + C2F3# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C2F3+ → SiOCF3(s) + CO + C2F3# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F3+ → SiOCF3(s) + COF + C2F3# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F3+ → SiO2CF(s) + C2F4 + C2F3# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F3+ → SiO2CF3(s) + C2F3 + C2F3# 0.1471 35 1 140 2
SiO2CF*(s) + C2F3+ → CO2 + C2F3# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F3+ → CO2 + C2F3#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F3+ → CO2 + C2F3#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F3+ → SiOCF3*(s) + CO + C2F3# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F3+ → SiOCF3*(s) + COF + C2F3# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F3+ → SiO2CF*(s) + C2F4 + C2F3# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F3+ → SiO2CF3*(s) + C2F3 + C2F3# 0.1471 35 1 140 2
SiO2CF(s) + C2F3# → SiF + CO2 + C2F3# 0.1471 35 1 140 2
SiO2CF2(s) + C2F3# → CO2 + C2F3# + SiF2 0.1471 35 1 140 2

255
SiO2CF3(s) + C2F3# → CO2 + C2F3# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C2F3# → SiOCF3(s) + CO + C2F3# 0.1471 35 1 140 2
SiO2C2F4(s) + C2F3# → SiOCF3(s) + COF + C2F3# 0.1471 35 1 140 2
SiO2C3F5(s) + C2F3# → SiO2CF(s) + C2F4 + C2F3# 0.1471 35 1 140 2
SiO2C3F6(s) + C2F3# → SiO2CF3(s) + C2F3 + C2F3# 0.1471 35 1 140 2
SiO2CF*(s) + C2F3# → CO2 + C2F3# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C2F3# → CO2 + C2F3#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C2F3# → CO2 + C2F3#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C2F3# → SiOCF3*(s) + CO + C2F3# 0.1471 35 1 140 2
SiO2C2F4*(s) + C2F3# → SiOCF3*(s) + COF + C2F3# 0.1471 35 1 140 2
SiO2C3F5*(s) + C2F3# → SiO2CF*(s) + C2F4 + C2F3# 0.1471 35 1 140 2
SiO2C3F6*(s) + C2F3# → SiO2CF3*(s) + C2F3 + C2F3# 0.1471 35 1 140 2
SiO2CF(s) + C3F5+ → SiF + CO2 + C3F5# 0.1471 35 1 140 2
SiO2CF2(s) + C3F5+ → CO2 + C3F5# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F5+ → CO2 + C3F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F5+ → SiOCF3(s) + CO + C3F5# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F5+ → SiOCF3(s) + COF + C3F5# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F5+ → SiO2CF(s) + C2F4 + C3F5# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F5+ → SiO2CF3(s) + C2F3 + C3F5# 0.1471 35 1 140 2
SiO2CF*(s) + C3F5+ → CO2 + C3F5# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F5+ → CO2 + C3F5#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F5+ → CO2 + C3F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F5+ → SiOCF3*(s) + CO + C3F5# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F5+ → SiOCF3*(s) + COF + C3F5# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F5+ → SiO2CF*(s) + C2F4 + C3F5# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F5+ → SiO2CF3*(s) + C2F3 + C3F5# 0.1471 35 1 140 2
SiO2CF(s) + C3F5# → SiF + CO2 + C3F5# 0.1471 35 1 140 2
SiO2CF2(s) + C3F5# → CO2 + C3F5# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F5# → CO2 + C3F5#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F5# → SiOCF3(s) + CO + C3F5# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F5# → SiOCF3(s) + COF + C3F5# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F5# → SiO2CF(s) + C2F4 + C3F5# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F5# → SiO2CF3(s) + C2F3 + C3F5# 0.1471 35 1 140 2
SiO2CF*(s) + C3F5# → CO2 + C3F5# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F5# → CO2 + C3F5#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F5# → CO2 + C3F5# + SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F5# → SiOCF3*(s) + CO + C3F5# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F5# → SiOCF3*(s) + COF + C3F5# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F5# → SiO2CF*(s) + C2F4 + C3F5# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F5# → SiO2CF3*(s) + C2F3 + C3F5# 0.1471 35 1 140 2
SiO2CF(s) + C3F6+ → SiF + CO2 + C3F6# 0.1471 35 1 140 2

256
SiO2CF2(s) + C3F6+ → CO2 + C3F6# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F6+ → CO2 + C3F6#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F6+ → SiOCF3(s) + CO + C3F6# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F6+ → SiOCF3(s) + COF + C3F6# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F6+ → SiO2CF(s) + C2F4 + C3F6# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F6+ → SiO2CF3(s) + C2F3 + C3F6# 0.1471 35 1 140 2
SiO2CF*(s) + C3F6+ → CO2 + C3F6# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F6+ → CO2 + C3F6#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F6+ → CO2 + C3F6#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F6+ → SiOCF3*(s) + CO + C3F6# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F6+ → SiOCF3*(s) + COF + C3F6# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F6+ → SiO2CF*(s) + C2F4 + C3F6# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F6+ → SiO2CF3*(s) + C2F3 + C3F6# 0.1471 35 1 140 2
SiO2CF(s) + C3F6# → SiF + CO2 + C3F6# 0.1471 35 1 140 2
SiO2CF2(s) + C3F6# → CO2 + C3F6# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F6# → CO2 + C3F6#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F6# → SiOCF3(s) + CO + C3F6# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F6# → SiOCF3(s) + COF + C3F6# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F6# → SiO2CF(s) + C2F4 + C3F6# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F6# → SiO2CF3(s) + C2F3 + C3F6# 0.1471 35 1 140 2
SiO2CF*(s) + C3F6# → CO2 + C3F6# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F6# → CO2 + C3F6#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F6# → CO2 + C3F6#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F6# → SiOCF3*(s) + CO + C3F6# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F6# → SiOCF3*(s) + COF + C3F6# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F6# → SiO2CF*(s) + C2F4 + C3F6# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F6# → SiO2CF3*(s) + C2F3 + C3F6# 0.1471 35 1 140 2
SiO2CF(s) + C3F7+ → SiF + CO2 + C3F7# 0.1471 35 1 140 2
SiO2CF2(s) + C3F7+ → CO2 + C3F7# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F7+ → CO2 + C3F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F7+ → SiOCF3(s) + CO + C3F7# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F7+ → SiOCF3(s) + COF + C3F7# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F7+ → SiO2CF(s) + C2F4 + C3F7# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F7+ → SiO2CF3(s) + C2F3 + C3F7# 0.1471 35 1 140 2
SiO2CF*(s) + C3F7+ → CO2 + C3F7#
+ SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F7+ → CO2 + C3F7# + SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F7+ → CO2 + C3F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F7+ → SiOCF3*(s) + CO + C3F7# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F7+ → SiOCF3*(s) + COF + C3F7# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F7+ → SiO2CF*(s) + C2F4 + C3F7# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F7+ → SiO2CF3*(s) + C2F3 + C3F7# 0.1471 35 1 140 2

257
SiO2CF(s) + C3F7# → SiF + CO2 + C3F7# 0.1471 35 1 140 2
SiO2CF2(s) + C3F7# → CO2 + C3F7# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C3F7# → CO2 + C3F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C3F7# → SiOCF3(s) + CO + C3F7# 0.1471 35 1 140 2
SiO2C2F4(s) + C3F7# → SiOCF3(s) + COF + C3F7# 0.1471 35 1 140 2
SiO2C3F5(s) + C3F7# → SiO2CF(s) + C2F4 + C3F7# 0.1471 35 1 140 2
SiO2C3F6(s) + C3F7# → SiO2CF3(s) + C2F3 + C3F7# 0.1471 35 1 140 2
SiO2CF*(s) + C3F7# → CO2 + C3F7# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C3F7# → CO2 + C3F7#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C3F7# → CO2 + C3F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C3F7# → SiOCF3*(s) + CO + C3F7# 0.1471 35 1 140 2
SiO2C2F4*(s) + C3F7# → SiOCF3*(s) + COF + C3F7# 0.1471 35 1 140 2
SiO2C3F5*(s) + C3F7# → SiO2CF*(s) + C2F4 + C3F7# 0.1471 35 1 140 2
SiO2C3F6*(s) + C3F7# → SiO2CF3*(s) + C2F3 + C3F7# 0.1471 35 1 140 2
SiO2CF(s) + C4F7+ → SiF + CO2 + C4F7# 0.1471 35 1 140 2
SiO2CF2(s) + C4F7+ → CO2 + C4F7# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C4F7+ → CO2 + C4F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C4F7+ → SiOCF3(s) + CO + C4F7# 0.1471 35 1 140 2
SiO2C2F4(s) + C4F7+ → SiOCF3(s) + COF + C4F7# 0.1471 35 1 140 2
SiO2C3F5(s) + C4F7+ → SiO2CF(s) + C2F4 + C4F7# 0.1471 35 1 140 2
SiO2C3F6(s) + C4F7+ → SiO2CF3(s) + C2F3 + C4F7# 0.1471 35 1 140 2
SiO2CF*(s) + C4F7+ → CO2 + C4F7# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C4F7+ → CO2 + C4F7#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C4F7+ → CO2 + C4F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C4F7+ → SiOCF3*(s) + CO + C4F7# 0.1471 35 1 140 2
SiO2C2F4*(s) + C4F7+ → SiOCF3*(s) + COF + C4F7# 0.1471 35 1 140 2
SiO2C3F5*(s) + C4F7+ → SiO2CF*(s) + C2F4 + C4F7# 0.1471 35 1 140 2
SiO2C3F6*(s) + C4F7+ → SiO2CF3*(s) + C2F3 + C4F7# 0.1471 35 1 140 2
SiO2CF(s) + C4F7# → SiF + CO2 + C4F7# 0.1471 35 1 140 2
SiO2CF2(s) + C4F7# → CO2 + C4F7# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C4F7# → CO2 + C4F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C4F7# → SiOCF3(s) + CO + C4F7# 0.1471 35 1 140 2
SiO2C2F4(s) + C4F7# → SiOCF3(s) + COF + C4F7# 0.1471 35 1 140 2
SiO2C3F5(s) + C4F7# → SiO2CF(s) + C2F4 + C4F7# 0.1471 35 1 140 2
SiO2C3F6(s) + C4F7# → SiO2CF3(s) + C2F3 + C4F7# 0.1471 35 1 140 2
SiO2CF*(s) + C4F7# → CO2 + C4F7# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C4F7# → CO2 + C4F7#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C4F7# → CO2 + C4F7#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C4F7# → SiOCF3*(s) + CO + C4F7# 0.1471 35 1 140 2
SiO2C2F4*(s) + C4F7# → SiOCF3*(s) + COF + C4F7# 0.1471 35 1 140 2
SiO2C3F5*(s) + C4F7# → SiO2CF*(s) + C2F4 + C4F7# 0.1471 35 1 140 2

258
SiO2C3F6*(s) + C4F7# → SiO2CF3*(s) + C2F3 + C4F7# 0.1471 35 1 140 2
SiO2CF(s) + C4F8+ → SiF + CO2 + C4F8# 0.1471 35 1 140 2
SiO2CF2(s) + C4F8+ → CO2 + C4F8#
+ SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C4F8+ → CO2 + C4F8# + SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C4F8+ → SiOCF3(s) + CO + C4F8# 0.1471 35 1 140 2
SiO2C2F4(s) + C4F8+ → SiOCF3(s) + COF + C4F8# 0.1471 35 1 140 2
SiO2C3F5(s) + C4F8+ → SiO2CF(s) + C2F4 + C4F8# 0.1471 35 1 140 2
SiO2C3F6(s) + C4F8+ → SiO2CF3(s) + C2F3 + C4F8# 0.1471 35 1 140 2
SiO2CF*(s) + C4F8+ → CO2 + C4F8# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C4F8+ → CO2 + C4F8#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C4F8+ → CO2 + C4F8#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C4F8+ → SiOCF3*(s) + CO + C4F8# 0.1471 35 1 140 2
SiO2C2F4*(s) + C4F8+ → SiOCF3*(s) + COF + C4F8# 0.1471 35 1 140 2
SiO2C3F5*(s) + C4F8+ → SiO2CF*(s) + C2F4 + C4F8# 0.1471 35 1 140 2
SiO2C3F6*(s) + C4F8+ → SiO2CF3*(s) + C2F3 + C4F8# 0.1471 35 1 140 2
SiO2CF(s) + C4F8# → SiF + CO2 + C4F8# 0.1471 35 1 140 2
SiO2CF2(s) + C4F8# → CO2 + C4F8# + SiF2 0.1471 35 1 140 2
SiO2CF3(s) + C4F8# → CO2 + C4F8#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3(s) + C4F8# → SiOCF3(s) + CO + C4F8# 0.1471 35 1 140 2
SiO2C2F4(s) + C4F8# → SiOCF3(s) + COF + C4F8# 0.1471 35 1 140 2
SiO2C3F5(s) + C4F8# → SiO2CF(s) + C2F4 + C4F8# 0.1471 35 1 140 2
SiO2C3F6(s) + C4F8# → SiO2CF3(s) + C2F3 + C4F8# 0.1471 35 1 140 2
SiO2CF*(s) + C4F8# → CO2 + C4F8# + SiF 0.1471 35 1 140 2
SiO2CF2*(s) + C4F8# → CO2 + C4F8#
+ SiF2 0.1471 35 1 140 2
SiO2CF3*(s) + C4F8# → CO2 + C4F8#
+ SiF3 0.1471 35 1 140 2
SiO2C2F3*(s) + C4F8# → SiOCF3*(s) + CO + C4F8# 0.1471 35 1 140 2
SiO2C2F4*(s) + C4F8# → SiOCF3*(s) + COF + C4F8# 0.1471 35 1 140 2
SiO2C3F5*(s) + C4F8# → SiO2CF*(s) + C2F4 + C4F8# 0.1471 35 1 140 2
SiO2C3F6*(s) + C4F8# → SiO2CF3*(s) + C2F3 + C4F8# 0.1471 35 1 140 2
SiOCF3(s) + Ar+ → Ar# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + Ar# → Ar #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF3+ → CF3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF3# → CF3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF2+ → CF2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF2# → CF2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF+ → CF# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + CF# → CF# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + F+ → F#
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + F# → F#
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + F2+ → F2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + F2# → F2 #
+ CO + SiF3 0.1471 35 1 140 2

259
SiOCF3(s) + C2F5+ → C2F5# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C2F5# → C2F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C2F4+ → C2F4 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C2F4# → C2F4# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C2F3+ → C2F3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C2F3# → C2F3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F5+ → C3F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F5# → C3F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F6+ → C3F6 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F6# → C3F6 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F7+ → C3F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C3F7# → C3F7# + CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C4F7+ → C4F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C4F7# → C4F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C4F8+ → C4F8 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + C4F8# → C4F8 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + O+ → O #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + O# → O #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + O2+ → O2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3(s) + O2# → O2# + CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + Ar+ → Ar #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + Ar# → Ar #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF3+ → CF3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF3# → CF3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF2+ → CF2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF2# → CF2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF+ → CF #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + CF# → CF# + CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + F+ → F#
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + F# → F#
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + F2+ → F2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + F2# → F2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F5+ → C2F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F5# → C2F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F4+ → C2F4 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F4# → C2F4# + CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F3+ → C2F3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C2F3# → C2F3 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C3F5+ → C3F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C3F5# → C3F5 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C3F6+ → C3F6 #
+ CO + SiF3 0.1471 35 1 140 2

260
SiOCF3*(s) + C3F6# → C3F6# + CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C3F7+ → C3F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C3F7# → C3F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C4F7+ → C4F7# + CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C4F7# → C4F7 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C4F8+ → C4F8 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + C4F8# → C4F8 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + O+ → O #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + O# → O #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + O2+ → O2 #
+ CO + SiF3 0.1471 35 1 140 2
SiOCF3*(s) + O2# → O2 #
+ CO + SiF3 0.1471 35 1 140 2
SiO2CF(s) + F → SiO2CF2(s) 0.1
SiO2CF*(s) + F → SiO2CF2*(s) 0.1
SiO2CF2(s) + F → SiO2CF3(s) 0.1
SiO2CF2*(s) + F → SiO2CF3*(s) 0.1
SiO2C2F3(s) + F → SiO2C2F4(s) 0.1
SiO2C2F3*(s) + F → SiO2C2F4*(s) 0.1
SiO2C3F5(s) + F → SiO2C3F6(s) 0.1
SiO2C3F5*(s) + F → SiO2C3F6*(s) 0.1
SiO2CF(s) + CF → SiO2C2F3(s) 0.0002
SiO2CF*(s) + CF → SiO2CF*(s) + CF(s) 0.002
SiO2CF2(s) + CF → SiO2C2F3(s) 0.0002
SiO2CF2*(s) + CF → SiO2CF2*(s) + CF(s) 0.002
SiO2CF3(s) + CF → SiO2C2F4(s) 0.0002
SiO2CF3*(s) + CF → SiO2CF3*(s) + CF(s) 0.002
SiO2C2F3(s) + CF → SiO2C3F5(s) 0.002
SiO2C2F3*(s) + CF → SiO2C2F3*(s) + CF(s) 0.002
SiO2C2F4(s) + CF → SiO2C3F5(s) 0.0002
SiO2C2F4*(s) + CF → SiO2C2F4*(s) + CF(s) 0.002
SiO2C3F5*(s) + CF → SiO2C3F5*(s) + CF(s) 0.002
SiO2C3F6*(s) + CF → SiO2C3F6*(s) + CF(s) 0.002
SiOCF3(s) + CF → SiOCF3(s) + CF(s) 0.002
SiOCF3*(s) + CF → SiOCF3*(s) + CF(s) 0.002
SiO2CF(s) + CF2 → SiO2C2F3(s) 0.00015
SiO2CF*(s) + CF2 → SiO2CF*(s) + CF2(s) 0.0015
SiO2CF2(s) + CF2 → SiO2C2F4(s) 0.00015
SiO2CF2*(s) + CF2 → SiO2CF2*(s) + CF2(s) 0.0015
SiO2CF3(s) + CF2 → SiO2C2F4(s) 0.00015
SiO2CF3*(s) + CF2 → SiO2CF3*(s) + CF2(s) 0.0015
SiO2C2F3(s) + CF2 → SiO2C3F5(s) 0.00015
SiO2C2F3*(s) + CF2 → SiO2C2F3*(s) + CF2(s) 0.0015

261
 SiO2C2F4(s) + CF2 → SiO2C3F6(s) 0.00015
SiO2C2F4*(s) + CF2 → SiO2C2F4*(s) + CF2(s) 0.0015
SiO2C3F5*(s) + CF2 → SiO2C3F5*(s) + CF2(s) 0.0015
SiO2C3F6*(s) + CF2 → SiO2C3F6*(s) + CF2(s) 0.0015
SiOCF3(s) + CF2 → SiOCF3(s) + CF2(s) 0.0015
SiOCF3*(s) + CF2 → SiOCF3*(s) + CF2(s) 0.0015
SiO2CF(s) + CF3 → SiO2C2F4(s) 0.0001
SiO2CF*(s) + CF3 → SiO2CF*(s) + CF3(s) 0.001
SiO2CF2(s) + CF3 → SiO2C2F4(s) 0.0001
SiO2CF2*(s) + CF3 → SiO2CF2*(s) + CF3(s) 0.001
SiO2CF3(s) + CF3 → SiO2C2F4(s) 0.0001
SiO2CF3*(s) + CF3 → SiO2CF3*(s) + CF3(s) 0.001
SiO2C2F3(s) + CF3 → SiO2C3F6(s) 0.0001
SiO2C2F3*(s) + CF3 → SiO2C2F3*(s) + CF3(s) 0.001
SiO2C2F4*(s) + CF3 → SiO2C2F4*(s) + CF3(s) 0.001
SiO2C3F5*(s) + CF3 → SiO2C3F5*(s) + CF3(s) 0.001
SiO2C3F6*(s) + CF3 → SiO2C3F6*(s) + CF3(s) 0.001
SiOCF3(s) + CF3 → SiOCF3(s) + CF3(s) 0.001
SiOCF3*(s) + CF3 → SiOCF3*(s) + CF3(s) 0.001
SiO2CF(s) + C2F3 → SiO2C3F5(s) 0.0001
SiO2CF*(s) + C2F3 → SiO2CF*(s) + C2F3(s) 0.001
SiO2CF2(s) + C2F3 → SiO2C3F5(s) 0.0001
SiO2CF2*(s) + C2F3 → SiO2CF2*(s) + C2F3(s) 0.001
SiO2CF3(s) + C2F3 → SiO2C3F6(s) 0.0001
SiO2CF3*(s) + C2F3 → SiO2CF3*(s) + C2F3(s) 0.001
SiO2C2F3*(s) + C2F3 → SiO2C2F3*(s) + C2F3(s) 0.001
SiO2C2F4*(s) + C2F3 → SiO2C2F4*(s) + C2F3(s) 0.001
SiO2C3F5*(s) + C2F3 → SiO2C3F5*(s) + C2F3(s) 0.001
SiO2C3F6*(s) + C2F3 → SiO2C3F6*(s) + C2F3(s) 0.001
SiOCF3(s) + C2F3 → SiOCF3(s) + C2F3(s) 0.001
SiOCF3*(s) + C2F3 → SiOCF3*(s) + C2F3(s) 0.001

B.1 References
[1] J. D. Kress, ; D E Hanson, ; A F Voter, ; C L Liu, X.-Y. Liu, ; D G Coronell, ) D E Hanson,

A. F. Voter, C. L. Liu and D. G. Coronell, Journal of Vacuum Science & Technology A 17,

2819 (1999).

[2] J. P. Chang and H. H. Sawin, Journal of Vacuum Science & Technology A 15, 610 (1997).

262