Lee T Iong Seng Norman

ATOMISTIC CALCULATIONS OF THE MECHANICAL
PROPERTIES OF Cu-Sn INTERMETALLIC COMPOUNDS

LEE TIONG SENG NORMAN
NATIONAL UNIVERSITY OF SINGAPORE
2008
ATOMISTIC CALCULATIONS OF THE MECHANICAL
PROPERTIES OF Cu-Sn INTERMETALLIC COMPOUNDS
LEE TIONG SENG NORMAN
(B.Eng(Hons),NUS)
A THESIS SUBMITTED
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
DEPARTMENT OF MECHANICAL ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE
2008
In my time, I tried to educate our people in an understanding of the
dignity of human life and their right as fellow human beings, and youth
was not only interested but excited about what I consider things that
matter. Things of the spirit, the development of a human being to
his true potential in accordance with his own personal genius in the
context of equal rights of others.
David Saul Marshall (1908-1995)
iv
Acknowledgements
Over the past six years, my supervisors, Dr Lim Kian Meng, Dr Vincent Tan and
Dr Zhang Xiao Wu have provided valuable guidance, advice and support. Very
often, they would ask probing questions that spurred me to think deeper into the
topic at hand or my interpretation of the research results. They have also inspired
me to explore new research areas, often bringing me out of my comfort zone.
My mum has also been a pillar of support. I would come home after taking
the last bus to nd fruits or food on my desk.
I acknowledge the nancial support from the following sources: NUS research
scholarship (2002 - 2004), NUSNNI(2005 - 2006), Institute of Microelectronics
(2002 - 2004). Credit must also be given to Centre for Science and Mathematics,
Republic Polytechnic for employment (2008) and their understanding on the occa-
sions when I was unable to fulll my duties. Financial support came also from the
many opportunities for teaching from the Department of Mechanical Engineering
(especially with Prof CJ Tay and Prof Cheng Li), Professional Activities Centre
and the Bachelor of Technology department.
The facilities provided by NUS have been excellent. The sta at the Science
Library have been most friendly and helpful. I have also made extensive use of the
resources provided by the Supercomputing and Visualization Unit (SVU), and I
thank the sta, Dr Zhang Xinhuai and Mr Yeo Eng Hee for their excellent service.
I would also like to thank the University Health and Wellness centre.
My thanks also goes out to the following members of the scientic community,
v
Dr Alexander Goldberg of Accelrys Inc. and Prof Lee Ming-Hsien of Tamkang
University, Taiwan. Although the need to credit their contributions in the main
text of this thesis did not arise, I appreciate their willingness to respond to my
email queries.
I owe special thanks to my colleagues and lab ocers. Regardless of the time
of day, they would provide useful words of advice and encouragement when the
demands of this research seemed overwhelming. Their knowledge, opinions and
ideas which they shared with me often gave me the needed push to move on. They
also had to put up with my idiosyncracies. In this, I acknowledge Adrian Koh,
Zhang Bao, Alvin Ong, Dr Zhang YingYan, Dr Dai Ling, Dr Deng Mu and Dr
Yew Yong Kin, as well as lab ocers Mr Joe Low, Mr Alvin Goh and Mr Chiam
Tow Jong.
My thanks goes out to my friends for their encouragement and their advice in
the decisions that I have made. Talking to them always helped in seeing things
clearer. I am sure that their prayers helped a lot. Last of all, there were many
occasions when serendipity and decisions that I made in the past (e.g. taking basic
German lessons) played a role in getting the research work done. I recognize the
role of the Creator in all that has happened. Gloria in altissimis Deo.
vi
Contents
List of Tables xv
List of Figures xix
List of Symbols and Acronyms xxiv
1 Introduction 1
1.1 Current Trends in the Electronics Industry . . . . . . . . . . . . . . 1
1.2 Computational Methods . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Elastic Properties of Intermetallic Compounds 8
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Experimental Studies of the Elastic Properties of the Inter-
metallic Compounds . . . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Geometry Optimization . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Making use of the energy . . . . . . . . . . . . . . . . . . . . 15
CONTENTS vii
2.2.4 Elastic Constants . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.5 DFT Calculations of Intermetallic compounds . . . . . . . . 17
2.2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Crystal structure . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Density Functional Theory . . . . . . . . . . . . . . . . . . . 22
2.3.3 Elasticity of Single Crystals . . . . . . . . . . . . . . . . . . 27
2.3.4 Bounds on Polycrystalline Elastic Moduli . . . . . . . . . . . 30
2.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4.1 Crystal structure of the Intermetallic Compounds . . . . . . 33
2.4.2 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.3 Computational Resources . . . . . . . . . . . . . . . . . . . 38
2.4.4 Geometry Optimization . . . . . . . . . . . . . . . . . . . . 38
2.4.5 Accuracy Settings . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.7 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5 Properties of Monoatomic Metals . . . . . . . . . . . . . . . . . . . 44
2.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5.2 Lattice Constants and Elastic Constants of Ag Ni and Cu . 45
2.5.3 Lattice Constants and Elastic Constants of Sn . . . . . . . . 45
2.5.4 Eect of using GGA-generated pseudopotentials . . . . . . . 47
2.5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6 Properties of the Intermetallic Compounds . . . . . . . . . . . . . . 49
2.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6.2 Lattice Constants . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6.3 Internal Crystal Parameters . . . . . . . . . . . . . . . . . . 51
CONTENTS viii
2.6.5 Elastic Anisotropy . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.6 Bounds on Polycrystalline Elastic Moduli . . . . . . . . . . . 57
2.6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.7.1 Limitations of the DFT calculations performed . . . . . . . . 64
2.7.2 Comparison of polycrystalline bounds with nanoindentation
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 Molecular Dynamics Potential For Cu-Sn 70
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.1.2 Molecular Dynamics Method . . . . . . . . . . . . . . . . . . 71
3.1.3 Types of Interatomic Potentials . . . . . . . . . . . . . . . . 73
3.1.4 Challenges with developing an interatomic potential for Cu-Sn 75
3.1.5 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.2 Interatomic Potentials . . . . . . . . . . . . . . . . . . . . . 79
3.2.3 MD simulation of materials with complex structures . . . . . 81
3.2.4 MD simulation of materials with two or more atomic species 83
3.2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.3.1 Notation and Denitions . . . . . . . . . . . . . . . . . . . . 86
3.3.2 Modied Embedded Atom Method . . . . . . . . . . . . . . 89
3.3.3 Optimization Methods . . . . . . . . . . . . . . . . . . . . . 100
3.4 Potential Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.4.1 Predictions of existing parameters . . . . . . . . . . . . . . . 103
CONTENTS ix
3.4.2 Potential Fitting Strategy . . . . . . . . . . . . . . . . . . . 105
3.4.3 Density Functional Theory Calculations . . . . . . . . . . . 108
3.4.4 Fitting database . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4.5 Choice of functions and parameters . . . . . . . . . . . . . . 114
3.4.6 Optimization Methodology . . . . . . . . . . . . . . . . . . . 119
3.5 Results and Validation . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.5.1 Parameters Obtained . . . . . . . . . . . . . . . . . . . . . . 123
3.5.2 Calculation Method . . . . . . . . . . . . . . . . . . . . . . . 124
3.5.3 Minimum Energy Structure . . . . . . . . . . . . . . . . . . 125
3.5.5 Surface Energy . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.5.6 Other Structures . . . . . . . . . . . . . . . . . . . . . . . . 135
3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.6.2 Choice of Potential Parameters . . . . . . . . . . . . . . . . 139
3.6.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4 MD Simulations of Fracture 142
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.2 Experimental studies of the fracture toughness of Cu
6
Sn
5
. 143
4.1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.2.2 Fracture of Metals and Semiconductors . . . . . . . . . . . . 145
4.2.3 Fracture of Intermetallic Compounds . . . . . . . . . . . . . 147
4.2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
CONTENTS x
4.3 Simulation Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3.1 Interatomic Potential . . . . . . . . . . . . . . . . . . . . . . 149
4.3.2 Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3.3 MD Software . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3.4 Unit Cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.4 K
IC
calculation using the tensile loading on a periodic crack . . . . 154
4.4.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.4.3 Calculating the fracture toughness . . . . . . . . . . . . . . 160
4.4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.5 K
IC
calculation using a crack-tip
displacement eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.5.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.5.2 ac plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.5.3 Basal plane . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.6.1 Comparing the two simulation methods . . . . . . . . . . . . 170
4.6.2 Comparison with experimental results . . . . . . . . . . . . 171
4.6.3 Qualitative features of the simulations . . . . . . . . . . . . 172
4.6.4 How realistic is the interatomic potential? . . . . . . . . . . 173
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5 Conclusions 175
5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.2 Recommendations for Future Work . . . . . . . . . . . . . . . . . . 177
Bibliography 180
CONTENTS xi
A Elasticity Formulas 208
A.1 Orientation Dependance Of the Youngs Modulus . . . . . . . . . . 208
A.1.1 Cubic Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . 208
A.1.2 Monoclinic Crystal . . . . . . . . . . . . . . . . . . . . . . . 209
B Internal Relaxations In a Crystal 210
B.1 Honeycomb Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
C MEAM Formulas 212
C.1 MEAM Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
C.2 Cu-Sn Interatomic Distance . . . . . . . . . . . . . . . . . . . . . . 212
C.3 MEAM Lattice Sums for the NiAs Crystal . . . . . . . . . . . . . . 213
C.3.1 Sn atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
C.3.2 Cu atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
C.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
C.4 MEAM Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
C.4.1 Derivative of the Pairwise energy . . . . . . . . . . . . . . . 217
C.4.2 Derivative of the screening function . . . . . . . . . . . . . . 218
C.4.3 Derivative of the Embedding Energy . . . . . . . . . . . . . 218
C.4.4 Derivative of the partial electron densities . . . . . . . . . . 219
C.4.5 Derivative of . . . . . . . . . . . . . . . . . . . . . . . . . 220
D Formation Energy of Cu
6
Sn
5
with DFT 221
E Negative Elastic Constants for a monoclinic crystal 223
E.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
E.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
E.3 Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
E.4 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
E.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
CONTENTS xii
F Review of experimental work 227
F.1 Experimental Methods . . . . . . . . . . . . . . . . . . . . . . . . . 227
F.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
xiii
Summary
Intermetallic compounds such as Cu
6
Sn
5
play a vital role in the reliability of
electronic components. The properties of intermetallic compounds formed at the
solder joint need to be known accurately so that macro-scale computer modelling
can take place to evaluate the reliability of components during the design process.
However, experiments to determine the mechanical properties of Cu
6
Sn
5
, such as
the Youngs Modulus and the critical mode I fracture toughness, have been sparse
and have not produced any denitive value yet. Hence, there is a need to clarify
what the values of these properties are. Therefore, this study investigates the use
of calculations at the atomistic level in order to obtain the mechanical properties
of Cu-Sn intermetallic compounds.
Density Functional Theory (DFT) calculations are performed with the inter-
metallic compounds Cu
3
Sn and Cu
6
Sn
5
to obtain the elastic properties of these
materials. Using a single unit cell, their lattice constants are calculated and shown
to be in agreement with experimental data. The hitherto unknown single-crystal
elastic constants are then calculated. Using these values, the orientation depen-
dence of the single-crystal Youngs Modulus is evaluated. The direction of the
largest value is found to coincide with closely-packed planes. The bounds on the
polycrystalline elastic moduli are also evaluated using the methods of Hill and
Hashin-Shtrikman. These bounds are found to be on the upper range of experi-
mental results that are currently available. This shows that DFT calculations are
a feasible means of predicting the polycrystalline elastic properties of intermetallic
CONTENTS xiv
compounds.
Following which, an interatomic potential for Cu-Sn interactions in the Modi-
ed Embedded Atom Method formalism tailored to the properties of Cu
6
Sn
5
in the
NiAs crystal structure is developed. Using this interatomic potential, Molecular
Dynamics simulations of the fracture of Cu
6
Sn
5
are conducted with thousands of
atoms. Atomic behaviour corresponding to brittle fracture are seen in the simu-
lations. This shows that it is feasible to develop an interatomic potential for the
NiAs crystal structure, and to conduct realistic simulations that can reproduce
qualitatively the properties and behaviour of the brittle intermetallic compound.
xv
List of Tables
1.1 A comparison of the three computational methods . . . . . . . . . 5
2.1 Experimental data on intermetallic compounds investigated in this
Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Constraints on the elastic constants for dierent crystal systems . . 28
2.3 Crystal structure data of the four intermetallic compounds . . . . . 37
2.4 Main computational resources used . . . . . . . . . . . . . . . . . . 38
2.5 Filenames of CASTEP pseudopotentials used . . . . . . . . . . . . 41
2.6 Strain Patterns used in calculating elastic constants . . . . . . . . 42
2.7 Calculated lattice constants and elastic constants (GPa) for Ag, Cu
and Ni using LDA-generated pseudopotentials . . . . . . . . . . . . 46
2.8 Lattice Constants and Elastic Constants (GPa) for -Sn and -Sn
using LDA-generated pseudopotentials (Sn 00.usp). . . . . . . . . . 47
2.9 GGA calculations of the lattice constants and elastic constants (GPa)
of Cu and -Sn with GGA-generated pseudopotentials (Cu 00PBE.usp
and Sn 00PBE.usp). . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.10 Calculated lattice constants . . . . . . . . . . . . . . . . . . . . . . 50
2.11 Computational time in CPU-days for the lattice constants. For
each xc-functional, the values are arranged in order of increasing
accuracy setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.12 Positional Parameters for Cu
6
Sn
5
. . . . . . . . . . . . . . . . . . . 51
LIST OF TABLES xvi
2.13 Positional Parameters for Ni
3
Sn
4
. . . . . . . . . . . . . . . . . . . 52
2.14 Positional parameters for Cu
3
Sn . . . . . . . . . . . . . . . . . . . 52
2.15 Positional parameters for Ag
3
Sn . . . . . . . . . . . . . . . . . . . 52
2.16 Elastic Constants of Cu
3
Sn calculated at dierent accuracy setting.
All values are given in GPa. . . . . . . . . . . . . . . . . . . . . . . 54
2.17 Elastic Constants of Ni
3
Sn
4
calculated at dierent accuracy setting.
All values are given in GPa. . . . . . . . . . . . . . . . . . . . . . . 55
2.18 Elastic Constants calculated for Ag
3
Sn calculated at dierent accu-
racy setting. All values are given in GPa. . . . . . . . . . . . . . . 55
2.19 Computational time in CPU-days for the elastic constants. For
each xc-functional, the values are arranged in order of increasing
accuracy setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.20 Maximum single-crystal Youngs Modulus and their directions . . . 57
2.21 Bounds on the Polycrystalline Moduli (GGA) . . . . . . . . . . . . 62
2.22 Bounds on the Polycrystalline Moduli (LDA) . . . . . . . . . . . . 63
3.1 Physical constants and values used in calculating the Thermal de
Broglie wavelength of Cu . . . . . . . . . . . . . . . . . . . . . . . 72
3.2 The dierent summation symbols in use. . . . . . . . . . . . . . . . 88
3.3 Parameters for Cu-Sn by Aguilar et al.. . . . . . . . . . . . . . . . 103
3.4 Screening parameters used by Aguilar et al. . . . . . . . . . . . . . 103
3.5 Elastic Constants calculated. Calculations are performed with the
GGA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.6 The tting database . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.7 Screening parameters . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.8 Available experimental values of the enthalpy of formation . . . . . 119
3.9 Constraints placed on the tting parameters . . . . . . . . . . . . . 121
3.10 Weights for the items in the tting database . . . . . . . . . . . . . 122
LIST OF TABLES xvii
3.11 Parameters for Cu-Sn obtained from the potential tting process -
Set P1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.12 Parameters for Cu-Sn obtained from the potential tting process -
Set P2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.13 Coordinates of the atoms in Cell A. They are expressed as a fraction
of the cell vectors. The cell lengths are a
3a c. . . . . . . . . 126
3.14 Coordinates of the atoms in Cell B. They are expressed as a fraction
3a c. . . . . . . . . 126
3.15 Lattice constants of the B8 unit cell predicted by the two sets of in-
teratomic potential parameters using the threshold acceptance cal-
culations. For each, three trials are performed. . . . . . . . . . . . 127
3.16 Strain patterns for the various elastic constants . . . . . . . . . . . 128
3.17 Elastic Constants calculated. All values are given in GPa. c
13
is
calculated from the combined elastic constants. . . . . . . . . . . . 130
3.18 Elastic Constants requiring relaxation. All values are given in GPa.
c
11
and c
12
are calculated from the combined elastic constants. . . 131
3.19 Predicted surface energies. All values are in eV/
A
2
. . . . . . . . . 135
3.20 Predicted Energies of other structures . . . . . . . . . . . . . . . . 138
4.1 Experimental data for the fracture toughness K
IC
of Cu
6
Sn
5
. . . . 144
4.2 Atomic-Scale units used in the software . . . . . . . . . . . . . . . 151
4.3 Description of the size of the simulation boxes used in this section.
The pre-crack size and box dimensions are given in terms of unit
cell lengths. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.4 Values of stress and predicted K
IC
for the two cases, assuming that
= c
11
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.5 Dierent box sizes used in the study in this section . . . . . . . . . 164
4.6 Dierent box sizes used in the study in this section . . . . . . . . . 166
LIST OF TABLES xviii
C.1 Position vector components of the atoms in the coordination poly-
hedron of the Sn atom . . . . . . . . . . . . . . . . . . . . . . . . . 214
C.2 Position vector components of the Cu atoms in the coordination
polyhedron of the Cu atom . . . . . . . . . . . . . . . . . . . . . . . 215
C.3 Position vector components of the Sn atoms in the coordination
polyhedron of the Cu atom . . . . . . . . . . . . . . . . . . . . . . . 215
D.1 DFT energies per atom for Cu (FCC) , Sn (DC) and CuSn (B8).
The energy change per atom (E
0
) and cohesive energy (E
coh
) are
then derived from them. . . . . . . . . . . . . . . . . . . . . . . . . 222
F.1 Experimental work on the Youngs Modulus of Cu
6
Sn
5
and Cu
3
Sn.
All values are in GPa. . . . . . . . . . . . . . . . . . . . . . . . . . 231
xix
List of Figures
1.1 Structure of a solder joint . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Illustrating the idea of a lattice with basis. The result is a honey-
comb lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Illustrating the various denitions of the unit cells using the (100)
plane of the FCC crystal. Dierent possibilities for the lattice vec-
tors are also shown. . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Cu
6
Sn
5
in the NiAs-type structure according to Gangulee et al..
The larger spheres represent the Sn atoms and the smaller spheres
represent the Cu atoms. The excess Cu atoms ll the interstitial
sites to make up the 6:5 stoichiometry. . . . . . . . . . . . . . . . . 34
2.4 Unit Cell of Cu
6
Sn
5
according to Larsson et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Cu
atoms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Unit Cell of Cu
3
Sn according to Burkhardt et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Cu
atoms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Unit Cell of Ni
3
Sn
4
according to Jeitschko et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Ni
atoms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
LIST OF FIGURES xx
2.7 Unit Cell of Ag
3
Sn according to Fairhurst et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Ag
atoms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.8 Documentation Provided in LDA-generated pseudopotential les for
Cu (Cu 00.usp) and Sn (Sn 00.usp) . . . . . . . . . . . . . . . . 40
2.9 Flowchart showing the methodology of the calculations in this chap-
ter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.10 Stiness matrix for Cu
6
Sn
5
calculated with the GGA at 320 eV
cuto and 4 4 3 k-point mesh. All values are given in GPa . . 54
2.11 Relation of the coordinate axes to the crystal axes . . . . . . . . . . 56
2.12 Youngs Modulus as a function of crystallographic direction for
Cu
3
Sn. The colours represent the magnitude of the Youngs Mod-
ulus while the x, y, z coordinates represent the crystallographic di-
rections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.13 Youngs Modulus as a function of crystallographic direction for
Cu
6
Sn
5
. The colours represent the magnitude of the Youngs Mod-
ulus while the x, y, z coordinates represent the crystallographic di-
rections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.14 Youngs Modulus as a function of crystallographic direction of vari-
ous planes for Cu
3
Sn. Angles are measured anti-clockwise from the
rst-mentioned axis in each plane. . . . . . . . . . . . . . . . . . . 59
2.15 Youngs Modulus as a function of crystallographic direction of var-
ious planes for Cu
6
Sn
5
. Angles are measured anti-clockwise from
the rst-mentioned axis in each plane. . . . . . . . . . . . . . . . . 59
2.16 The plane corresponding to 55.8
anti-clockwise from the positive

x-axis in Cu
3
Sn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.17 The plane corresponding to 147.6

x-axis in Cu
6
Sn
5
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
LIST OF FIGURES xxi
2.18 A two-dimensional view of the plane marked in Figure 2.17 in Cu
6
Sn
5
.
61
2.19 Cu
6
Sn
5
: Cumulative Distribution Function for the single-crystal
Youngs Modulus of Cu
6
Sn
5
. . . . . . . . . . . . . . . . . . . . . . 68
2.20 Cu
3
Sn: Cumulative Distribution Function for the single-crystal Youngs
Modulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.1 FCC lattice in 2D projection. First neighbour and second neigh-
bour shells of shaded atom are indicated. . . . . . . . . . . . . . . 88
3.2 Illustrating the screening in the MEAM. . . . . . . . . . . . . . . . 92
3.3 Two cases to explain the MEAM Screening Procedure. For Case 1,
C = 3 and for case 2, C =
1
3
. . . . . . . . . . . . . . . . . . . . . . 93
3.4 Relationship between the pairwise, EAM and MEAM potential func-
tionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.5 Evolution of the energy of Cu-Sn B8 unit cell predicted using the
Aguilar MEAM potential . . . . . . . . . . . . . . . . . . . . . . . 104
3.6 Evolution of the lattice constants a and c of Cu-Sn B8 unit cell
using the Aguilar MEAM potential . . . . . . . . . . . . . . . . . . 105
3.7 Energy-Volume calculations for B8 Cu-Sn with DFT. The lines are
a guide to the eye. . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.8 Energy-Volume calculations for B8 Cu-Sn with DFT, enlarged to
show points near equilibrium volume. . . . . . . . . . . . . . . . . 111
3.9 Variation of equilibrium c/a ratio with volume. The lines are a
guide to the eye. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.10 Coordination Polyhedra for Cu and Sn in the B8 unit cell. The basis
atoms are indicated with yellow letters and the rst neighbours are
joined with yellow lines to show the coordination polyhedron. . . . 115
LIST OF FIGURES xxii
3.11 Alternative orthogonal unit cells that have the same symmetry as
the B8 unit cell. The c-direction is perpendicular to the plane of the
diagram. Both Cell A and Cell B have sides of length a
3a c. 125
3.12 Evolution of the lattice constants during the threshold acceptance
process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.13 Strain patterns for the various elastic constants illustrated by show-
ing how the unit cell deforms. The strains for c
11
c
12
and c
44
destroy
the hexagonal symmetry. . . . . . . . . . . . . . . . . . . . . . . . 129
3.14 Unit cell to calculate c
44
. . . . . . . . . . . . . . . . . . . . . . . . 132
3.15 Illustrating the Surface Energy Calculation. Two simulation cells of
dierent boundary conditions are considered, one with fully periodic
boundary conditions, the other with a surface. The surface energy
= (E
1
E
0
)/A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.16 Unit cells used in the DFT calculation of the surface energy (a)
surface normal x (b) surface normal z. The height of the inserted
vacuum is about 8

A. . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.17 Unit cells for dierent structures with 1:1 stoichiometry. For each
structure, the Strukturbericht designation is given followed by the
chemical formula of a typical compound. . . . . . . . . . . . . . . . 137
4.1 Schematic diagram of cracks forming from the corners of the Vickers
indentation. The crack length C is measured from the centre of the
indentation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.2 Flowchart showing the organization of RidgeMD. . . . . . . . . . . 152
4.3 Description of the basal plane and the ac plane. . . . . . . . . . . 153
4.4 Periodic Crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.5 Schematic diagram of the periodic crack simulation box. The side
n of the pre-crack is parallel to the c-axis. . . . . . . . . . . . . . . 155
LIST OF FIGURES xxiii
4.6 Conguration PC1. Snapshot at 20000 time steps with an initial
strain of = 0.20 applied throughout. The sample cleaves between
the immobile layer and the mobile atoms and no crack propagates
from the pre-crack. . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.7 Conguration PC2 with a initial strain of = 0.20 applied through-
out. Snapshots at (a) 6500 time steps (b) 20000 time steps . . . . . 157
4.8 Conguration PC3 with an initial strain of = 0.12 applied through-
out. Snapshots at (a) 1200 time steps (b) 20000 time steps . . . . . 159
4.9 Conguration PC3. Snapshot at 1200 time steps representing the
atoms as circles. All the atoms are superimposed on each other.
The plane at which the crack propagates can be clearly seen. . . . 159
4.10 (a) Array of atoms (b) The simulation box after Sihs equation is
applied to the atoms . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.11 ac plane: Snapshots of atomic conguration for size Small at 20000
time steps at 300 K for (a) K
I
= 0.40 MPa
m(b) K
I
= 0.64 MPa
m166
4.12 ac plane: Snapshots of atomic conguration at 20000 time steps at
300 K for K
I
= 0.80 MPa
m. (a) Size Small (b) Size Large. . . . 167

4.13 ac plane: Size Large, K
I
= 0.80 MPa
m using circles to represent

the atoms. (a) without tails (b) with tails. . . . . . . . . . . . . . 167
4.14 Basal plane: snapshots of atomic conguration for size Small at
20000 time steps at 300 K for (a) K
I
= 0.48 MPa
m(b) K
I
= 0.64
MPa
m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.15 Basal plane: snapshots of atomic conguration for Size Large at
20000 time steps at 300 K for various values of K
I
= 0.64 MPa
m.
(a) 3100 time steps (b) 20000 time steps . . . . . . . . . . . . . . . 169
4.16 Basal plane: Size Large, K
I
= 0.64 MPa
m using circles to repre-

sent the atoms. (a) without tails (b) with tails. . . . . . . . . . . . 170
LIST OF FIGURES xxiv
B.1 Part of the honeycomb lattice. The (x, y) coordinates are given
beside each particle. (a) Before shear, all particles are unit distance
away from each other (b) after shear. The arrows beside particles 3
and 4 shows how they have been displaced . . . . . . . . . . . . . . 211
E.1 (a) (010) plane of Cu
6
Sn
5
(b) Schematic diagram of the (010) plane.
The dashed lines represent the planes of atoms seen in (a) (c) Upon
application of the shear strain
13
, the unit cell deforms according to
the red dotted line. The angle is reduced (d) Bottom left corner
of the unit cell. BC is shortened when is reduced. . . . . . . . . 225
F.1 (a) nanoindentation load-displacement curve (b) Berkovich indenter 228
xxv
List of Symbols and Acronyms
Latin Symbols
A (MEAM) Parameter in Embedding Function
A Surface area
A Amplitude of plane wave
A Indentation Projected Contact Area
a crack half-length
a lattice constant
a
ij
direction cosine between original axis j and rotated axis i
a Primitive lattice vector
B Bulk Modulus
B
0
Hashin-Shktrikman bulk modulus for a hypothetical material
B
R
Reuss average of the Bulk Modulus
B
V
Voights average of the Bulk Modulus
b integer value that denes the points in the reciprocal lattice
b Reciprocal lattice vector

C Vickers Crack length
C (MEAM) Parameter in screening function
C
min
(MEAM) Lower limit on C
C
max
(MEAM) Upper limit on C
C Elastic stiness matrix
LIST OF SYMBOLS AND ACRONYMS xxvi
c
k,G
Coecient for planewaves in the one-electron wavefunction
c lattice constant
c
ij
elastic constant
< c
ij
> average over all crystal orientations of c
ij
D Parameter in Morse potential
d
(h)
j
(MEAM) h-th atomic electron density due to neighbour j
E Strain Energy
E Youngs Modulus
E Energy from Schr odingers Equation
E
coh
Cohesive Energy
E
coh,Cu
Cohesive Energy of Copper
E
d
Total energy calculated by Density Functional Theory
E
i
Youngs Modulus of the indenter
E
r
Indentation Modulus
E
surf
Surface Energy
E
u
Roses Universal Equation of State
E
xc
Exchange-correlation energy functional
F Embedding Function
F Vickers Indentation Load
F
i
Embedding Function of atom i
f Force
f
i
Total force on atom i
f
i
Occupancy of the i-th one-electron wavefunction
f
a,x
Total force on atom a in the x-direction
f
ij
force between atoms i and j
G (MEAM) Function used in calculating electron density
G Shear Modulus
LIST OF SYMBOLS AND ACRONYMS xxvii
G
0
Hashin-Shktrikman shear modulus for a hypothetical material
G
R
Reuss average of the Shear Modulus
G
V
Voights average of the Shear Modulus
G Reciprocal lattice vector

H Hardness
H Hamiltonian operator in Schr odingers Equation

h Planck constant
i
1
i summation index for atoms
j summation index for an atoms neighbours
K
I
Mode I fracture toughness
K
1
Function in Hashin-Shtrikman Bounds
K
2
K
IC
Critical Mode I fracture toughness
k summation index when three-body forces are involved
k Boltzmann constant
k spring constant
k position vector of points in the Brillouin zone
k
w
Wavevector
m mass of atom
N Number of atoms
N (MEAM) Number of interactions for each pair of atomic species
n integer value that denes the points in the conventional lattice
n
ij
Unit vector of the position vector betweeom atoms i and j
P Number of neighbours for each atom pair
p Anisotropic crack-tip displacement eld parameter
q Anisotropic crack-tip displacement eld parameter
LIST OF SYMBOLS AND ACRONYMS xxviii
r Interatomic distance
r Atomic coordinates
r Distance in polar coordinates
r
i
Position of atom i
r
ij
Distance between atoms i and j
r
o
Equilibrium distance between rst neighbours in a crystal
r Position vector
S Elastic compliance matrix
s
ij
Elastic compliance
S
ij
Elastic compliance adjusted for plane strain
S
ij
(MEAM) Combined screening parameter between atoms i and j
s
ijk
(MEAM) Screening parameter between atoms i,j and k
s
1
Anisotropic crack-tip displacement eld parameter
s
1
Complex conjugate of s
1
s Elastic compliance
T Kinetic energy functional
T
s
Kinetic energy operator for a non-interacting system of electrons
t Time
t
(h)
(MEAM) Weighting parameter in function for the h-th partial
electron density
U Electron-electron interactions energy functional
u Displacement due to the anisotropic crack-tip displacement eld
V Volume
V Electron-nucleus interactions energy functional
V
o
Equilibrium Volume per atom
v
s
Eective potential of a system of non-interacting electrons
v speed of stress wave
LIST OF SYMBOLS AND ACRONYMS xxix
W Width of crack sample
X
ik
(MEAM) Parameter required in the calculation of the screening
parameter C
x
a
x-coordinate of atom a
x
ij
x component of the position vector between atom i and atom j
x
Direction cosine in the axis direction

y
a
y-coordinate of atom a
y
ij
y component of the position vector between atom i and atom j
Z (MEAM) Number of rst neighbours
Z Objective function
z
a
z-coordinate of atom a
z
ij
z component of the position vector between atom i and atom j
Greek Symbols
Parameter in Morse potential function
Parameter in Roses Universal Equation of State
lattice constant (angle between a and c)
(h)
(MEAM) Parameter in atomic electron density
(MEAM) Function that combines the partial electron densities
E
0
Energy change due to the formation of a compound
H Enthalpy of formation of a compound
t Time increment
Strain
Angle in polar coordinates
(MEAM) Parameter introduced into Embedding function
LIST OF SYMBOLS AND ACRONYMS xxx
De Broglie wavelength
Poissons ratio
i
Poissons ratio of the indenter
density of the material
Electron density

i
(MEAM) electron density of atom i
(h)
(MEAM) h-th partial electron density
Stress
Pairwise interatomic potential function
Many-body wavefunction
i
one-electron wavefunction of the i-th electron
Acronyms
3D Three dimensions
BCC Body-centred cubic unit cell
DC Diamond cubic unit cell
DFT Density Functional Theory
EAM Embedded Atom Method
FCC Face-centred cubic unit cell
FEM Finite Element Method
f.u. Formula unit
GGA Generalized Gradient Approximation
HK Hohenberg-Kohn
HS Hashin-Shtrikman
KS Kohn-Sham
LDA Local Density Approximation
MD Molecular Dynamics
LIST OF SYMBOLS AND ACRONYMS xxxi
MEAM Modied Embedded Atom Method
pbc Periodic boundary condition
RNG Random Number Generator
SW Stillinger-Weber
tu time unit equivalent to 1.02 10
14
s
UES Universal Equation of State
VI Vickers Indentation
VRH Voight-Reuss-Hill
xc exchange-correlation
1
Chapter 1
Introduction
1.1 Current Trends in the Electronics Industry
Portable electronic products have become ubiquitous in todays society. The con-
sumer demands that such products be small and packed with more features, yet
remain reliable when subject to daily use. Also, as these products continue to be
improved and packed with more features, older products get thrown away. Such
demands of the consumer drives two major trends in the electronics industry that
presents new challenges in the eld of materials science - the use of lead-free solder
and reliability to drop impact.
For many years, Pb-Sn solder alloy was the main solder in use in products.
However, as the amount of electronic waste in landlls increases, concerns grew
that the lead, which is toxic to humans and wildlife, would contaminate the envi-
ronment. As such, legislation has been passed in the European Union to ban lead
from electronics products [1, 2].
Over the years, many lead-free solders have been developed [3, 4], often with
novel compositions involving rare-earth metals [5]. Unlike eutectic PbSn solder
which has been well-studied due to its extensive use in the past, these new sol-
der alloys require extensive testing to understand their mechanical and reliability
1.1. CURRENT TRENDS IN THE ELECTRONICS INDUSTRY 2
properties. In particular, using Pb-free solder poses a dierent set of challenges.
Firstly, previous designs optimized for use with Pb-Sn solder cannot be used for
the new solders [6]. For example, newer Sn-rich solders have higher melting points
than Pb-Sn solders, requiring a higher reow temperature, thus exacerbating the
thermal mismatch stresses [7]. The choice of materials in the rest of the package
must take into account this fact. With Pb-Sn solder, one can draw upon 40 years
of experimental data. With the novel lead-free solders that have been developed,
there is no such benet and much work has to be carried out to understand the
properties of these material systems and their components.
Secondly, with dierent choices of materials, the interfacial reactions are dif-
ferent for each type of solder. A vastly-simplied schematic diagram of a typical
solder joint is shown in Figure 1.1. Along each material interface is a layer of
intermetallic compounds that are formed when the two layers bond together.
Solder
Substrate
Device
Under-Bump
Metallization
ntermetallic
layer
bonding pad ntermetallic
layer
Figure 1.1: Structure of a solder joint
Frear identied the reactions at these interfaces as a key issue in consumer
electronics applications [6], as they have an eect on the reliability and strength
of the solder joint. It was shown by Frear et al. that dierent lead-free solders
1.2. COMPUTATIONAL METHODS 3
produced dierent morphology and microstructure at the solder joint, and resulted
in dierent fatigue behaviour [8]. Ho et al. mentioned that when the solder joint
is subjected to thermal aging, growth of the intermetallic compound Cu
3
Sn was
accompanied by voids at the intermetallic layer [9]. Shear and pull tests have
shown that these voids weaken the solder joint [10].
This weakening of the solder joint is inimical to the reliability of electronic
packages to drop impact, which is an important concern due to the consumer
demand that portable electronic devices remain working when dropped. The in-
termetallic compounds that form at the solder joint play an important role in drop
impact reliability. They are known to be brittle, and studies over the years have
shown that when electronic packages are subjected to drop impact test, cracks are
found at the interface between the intermetallic compounds and the underlying
metallization [11, 12]. It was also demonstrated that fracture occurred through
the voids (caused by growth of Cu
3
Sn) after thermal aging [13].
1.2 Computational Methods
With so many factors that inuence the reliability of package designs, there is a
clear need for tools to enable designs to be evaluated. Experimental testing is
expensive and can only provide a limited amount of data [14]. As such, computa-
tional methods play an integral role in addressing these challenges.
The Finite Element Method (FEM) has established itself to be integral to
the design and development of electronic packaging. Early dynamic FEM studies
simulating the drop-impact behaviour of the solder joint did not include the in-
termetallic layer [15]. However, as the role of the intermetallic layer became clear,
studies taking into account the intermetallic layer have appeared. A FEM model
was developed to study the fracture behaviour of a solder joint and it was shown
that the stress was highest in the intermetallic layer[16]. In the study involving
design optimization, Syed et al. used an FEM model that included an intermetal-
lic layer and showed that designs that reduced the stress in the intermetallic layer
correspondingly proved to be more reliable in drop-testing[17]. Chen et al. stud-
ied the thermal cycling behaviour of a solder joint, showing that the locations of
cracks obtained experimentally were correlated with high-stress locations in the
solder[18].
Essential to successful modelling with FEM is knowledge of accurate material
properties. These studies previously described made use of properties reported by
Fields et al. [19]. Unfortunately, later studies of the Youngs Modulus did not
agree with their values, and the reported values show a wide range. This is also
true for the critical mode I fracture toughness. Details of these studies will be
given in the following chapters.
At the same time, as the length scale in electronic devices continue to be re-
duced, atomic-scale eects will begin to dominate. For example, it was found that
the yield stress of micron-scale Ni wires depended on the sample dimensions [20].
As such, computational tools need to be developed so that predictions of such ef-
fects can be made. Consequently, the challenges that the International Technology
Roadmap for Semiconductors identies are the need for Computational materi-
als science tools to describe material properties, and especially Linkage with
rst principle computation and reduced model (classical MD or thermodynamic
computation) [21].
It is now possible to predict material properties from atomic-level calcula-
tions.One of these tools is calculations based on Density Functional Theory (DFT),
which is essentially a means of solving the Schr odingers Equation [22]. There is
a growing body of evidence that DFT can obtain the elastic constants of existing
materials that correlate very well with those obtained experimentally. Examples
of such works include studies on TiB
2
and MgB
2
[23] and BeO [24]. Some studies
have also made predictions for newly-synthesised materials, like Ti
3
SiC
2
[25], and
Table 1.1: A comparison of the three computational methods
Method Underlying Assumption Size of Problem
Finite Element
Method (FEM)
Continuum approximation to ma-
terials
macroscopic
Molecular Dynamics
(MD)
Classical Mechanics approxima-
tion
thousands of atoms
Density Functional
Theory (DFT)
quantum mechanical approxima-
tions to Schrodingers Equation
tens of atoms
the possible crystal structures of theoretical materials [26].
However, DFT is computationally intensive and calculations typically take
weeks and months on desktop computers to complete, especially for complex struc-
tures with tens of atoms. Most atomic-level processes involve much more atoms
than this, for example, the fracture of a material.
Molecular Dynamics simulation (MD) can bridge the gap between DFT and
FEM. It is essentially a classical mechanics treatment of the forces between atoms
and this reduces the computational eort considerably. A linear spring joining
two particles has the potential energy (x) =
1
2
kx
2
and the force between the two
atoms as F =
d
dx
= kx.
Much of MD simulation work involves nding suitable expressions for (x)
so that certain essential properties of materials are reproduced (eg. elastic con-
stants) and then subsequently conducting simulations with the potential obtained.
The process of nding suitable expressions of (x) involves arbitrary choices as to
the choice of function and which essential properties to reproduce, which would
indicate a reduced predictive capability and accuracy of MD simulations. Never-
theless, for a simulation on the scale of thousands of atoms, there are no better
alternatives, and much simulation work has shown promise.
Putting it all in perspective, the characteristics of the three methods discussed
are summarized in Table 1.1. As these methods are done at dierent length scales,
they complement each other. DFT can calculate the energy of a system accu-
rately but the computational cost restricts the system size to the order of tens of
1.3. OBJECTIVES 6
atoms. MD simulations involve thousands of atoms and can simulate atomic-level
phenomenon, but the process behind obtaining the interatomic potential is often
subject to many arbitrary choices.
1.3 Objectives
The situation regarding the intermetallic compounds in the electronics industry
has been discussed and the computational methods that work at the atomic level
have also been described. The objectives of this thesis are as follows.
1. The elastic properties of intermetallic compounds found in electronic packag-
ing show little agreement, with the values obtained depending on the exper-
imental method used. Methods in computational materials science, such as
DFT, can now calculate the properties of materials. DFT calculations of the
elastic properties of Cu
6
Sn
5
, Cu
3
Sn, Ag
3
Sn and Ni
3
Sn
4
will be performed.
2. The vast majority of MD simulations are concentrated in the cubic metals
and semiconductors, due to the simplicity of the crystal structure and tech-
nological importance. Comparatively less eort has been put in to conduct
simulations of intermetallic compounds. An interatomic potential for Cu
6
Sn
5
in the NiAs structure will be obtained.
3. The interatomic potential obtained will be used to perform MD simulations
of the fracture of Cu
6
Sn
5
.
1.4 Outline of the thesis
Having described the intermetallic compounds and their importance to the perfor-
mance of the solder joint, this thesis will describe calculations performed to obtain
the mechanical properties of Cu
6
Sn
5
and Cu
3
Sn with computational methods.
1.4. OUTLINE OF THE THESIS 7
Calculations to predict the Youngs Modulus of Cu
6
Sn
5
and Cu
3
Sn using Den-
sity Functional Theory are described in Chapter 2. For the single crystal, predic-
tions are made for the structural properties and elastic properties for Cu
6
Sn
5
and
Cu
3
Sn. It will be shown how these results are validated by comparison with ex-
perimental results from polycrystalline specimens. Results for Ag
3
Sn and Ni
3
Sn
4
are also included for comparison.
In order to facilitate MD simulations of Cu
6
Sn
5
, the process of developing an
interatomic potential for Cu
6
Sn
5
in the NiAs-structure is described in Chapter 3.
Parameters in the Modied Embedded Atom Method are obtained by means of
optimizing their values to reproduce a set of energies in a database.
Using the potential developed in Chapter 3, MD simulations of the fracture of
Cu
6
Sn
5
are described in Chapter 4.
These results are summed up in Chapter 5, where possibilities for further re-
search are also laid out.
8
Chapter 2
Elastic Properties of Intermetallic
Compounds
2.1 Introduction
2.1.1 Motivation
The term intermetallic compound refers to compounds that are formed between
two or more metals that have a denite crystal structure at specic concentra-
tions. This chapter describes the calculation of the mechanical properties of four
intermetallic compounds commonly found in electronic packaging, Cu
6
Sn
5
, Cu
3
Sn,
Ag
3
Sn and Ni
3
Sn
4
.
These intermetallic compounds feature prominently in modern life. Apart from
being found in electronic packaging, both Cu
6
Sn
5
and Ag
3
Sn are found in dental
amalgam [27, 28]. Along with Cu
6
Sn
5
[29, 30], Ni
3
Sn
4
[31, 32, 33] is also being
considered as anode materials for lithium-ion batteries. The desire to eliminate
lead because of its toxicity has led to the use of Cu-Sn frangible bullets, within
which Cu
6
Sn
5
and Cu
3
Sn are present[34]. It has also been recently demonstrated
that it is feasible to conduct electroless plating (ie. a method of coating that does
2.1. INTRODUCTION 9
not require passing a current through the set-up) of Sn on Cu substrates, with the
thin Cu-Sn alloy layer formed consisting of Cu
6
Sn
5
and Cu
3
Sn [35, 36, 37].
Within the solder joint, Ni
3
Sn
4
forms at the interface between Sn-based solder
and Ni bonding pads, while Ag
3
Sn forms within the Sn-Ag solder during solder
reow. During the solder reow the solder reacts with a Cu substrate to form Cu-
Sn intermetallic compounds. Studies of the eect of temperature and duration on
the formation of intermetallic compounds have shown that only Cu
6
Sn
5
is formed
initially, while Cu
3
Sn is formed after a longer reaction time [38, 39].
2.1.2 Experimental Studies of the Elastic Properties of the
Intermetallic Compounds
The formation of these intermetallic compounds are necessary for good bonding.
However, being brittle, they compromise the reliability of the joint if present in
excessive amounts. This also becomes more drastic with the miniaturization of
components, as the fraction of the intermetallic layer to the total size of the solder
joint increases.
These material systems are still not well understood with regards to their
properties and their eects on the overall joint reliability. Researchers employing
Finite Element Method studies are beginning to include the intermetallic layer in
the solder joint. Accurate knowledge of their properties is critical if modelling is
to be employed to predict the behaviour of these systems.
There have also been a fair number of studies on the polycrystalline elastic mod-
uli of these intermetallic compounds. Results from these studies are summarized
in Table 2.1. Early studies, such as those by Fields et al. [19] and Subrahmanyan
[40] made use of bulk specimens. More recent studies were conducted on cross-
sectioned diusion couples and the solder joint itself. These studies are discussed
in further detail in Appendix F.
2.1. INTRODUCTION 10
It is clear that these results are plagued with disagreement. Nevertheless, it
is observed that the Youngs Modulus obtained from the nanoindentation method
are generally higher than results from the bulk specimens.
2.1.3 Overview
We have thus seen that there is disagreement regarding the elastic properties of the
Cu-Sn intermetallic compounds. These results are aected by dierent measure-
ment methods and sample preparation methods. In contrast, an innite perfect
crystal is the material in its ideal form, undisturbed by imperfections in sample
preparation methods or by errors introduced in experimental techniques. As such,
its elastic properties can be used to benchmark to evaluate these experimental
properties. At the same time, due to the diculty of growing single-crystals of
these compounds and the complexity of their crystal structures, there are no exper-
imental studies of the single-crystal elastic constants. This chapter thus answers
the question of what the properties of an ideal single-crystal of Cu
6
Sn
5
and Cu
3
Sn
are.
This is done by means of calculating the energy of these systems. Many material
properties can be predicted if the energy of an atomic system can be calculated
accurately. For example, the ground-state crystal structure is the structure with
the lowest free energy, the enthalpy of formation of a compound is obtained from
the dierence in energy between the compound and the individual atomic phases
and the elastic constants are proportional to the second derivative of the energy.
Density Functional Theory is a means of doing so. It enables calculations of
the ground-state energy of an atomic system by means of an approximation to the
Schr odingers Equation. This chapter describes how such calculations are carried
out.
The possibilities of such calculations with Density Functional Theory are
surveyed in section 2.2.
To carry out meaningful calculations, the theoretical foundations need to be
understood, and they are described in section 2.3. Some terms in crystallog-
raphy will be dened, followed by a description of Density Functional Theory
itself and the nature of the approximations made for this work.
The methodology of the calculations employed are described in Section 2.4.
Essential inputs that aect the accuracy of the calculations are explained.
DFT calculations of the lattice constants and elastic constants are presented
in sections 2.5 and 2.6. No direct validation for the elastic constants are
possible due to the lack of published experimental data. Instead, they are
validated by rst calculating bounds on the polycrystalline elastic moduli
and then comparing them with the data in Table 2.1.
T
a
b
l
e
2
.
1
:
E
x
p
e
r
i
m
e
n
t
a
l
d
a
t
a
o
n
i
n
t
e
r
m
e
t
a
l
l
i
c
c
o
m
p
o
u
n
d
s
i
n
v
e
s
t
i
g
a
t
e
d
i
n
t
h
i
s
C
h
a
p
t
e
r
A
g
3
S
n
N
i
3
S
n
4
C
u
6
S
n
5
C
u
3
S
n
Y
o
u
n
g
s
M
o
d
u
l
u
s
(
G
P
a
)
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
7
8
.
9
a
,
8
8
c
,
d
,
9
4
e
1
3
4
i
,
1
3
8
g
,
1
4
2
g
,
1
3
9
c
,
1
4
5
c
,
1
5
8
c
,
1
5
0
e
,
1
5
2
m
8
0
-
1
1
0
q
,
1
1
2
a
,
1
1
9
d
,
1
1
5
g
,
1
1
6
g
,
1
2
5
g
,
1
2
4
m
,
1
6
6
-
1
8
8
p
,
1
1
6
.
9
r
1
2
2
g
,
1
3
4
a
,
1
3
6
g
,
1
4
3
d
,
1
4
3
m
,
1
5
0
e
,
1
1
0
-
1
3
0
q
,
1
3
3
.
4
r
O
t
h
e
r
m
e
t
h
o
d
s
7
0
f
,
7
5
e
,
8
1
.
4
b
9
4
e
,
1
1
8
.
4
b
,
1
3
3
h
8
5
f
,
8
6
h
,
9
7
b
,
1
0
2
.
4
j
,
1
0
2
k
8
0
f
,
1
0
6
.
1
j
,
1
0
8
h
,
1
2
0
e
,
1
2
3
b
,
1
5
3
k
P
o
i
s
s
o
n
s
R
a
t
i
o
0
.
3
4
7
b
,
0
.
3
5
e
0
.
3
1
8
b
,
0
.
2
7
e
,
0
.
3
3
h
0
.
3
0
9
b
,
0
.
3
2
e
,
0
.
3
0
9
h
0
.
3
1
9
b
,
0
.
3
3
e
,
0
.
2
9
9
h
a
R
e
f
[
4
1
]
b
R
e
f
[
4
2
]
c
R
e
f
[
4
3
]
d
R
e
f
[
4
4
]
e
R
e
f
[
4
5
]
f
R
e
f
[
4
6
]
g
R
e
f
[
4
7
]
h
R
e
f
[
1
9
]
i
R
e
f
[
4
8
]
j
R
e
f
[
4
9
]
k
R
e
f
[
5
0
]
m
R
e
f
[
5
1
]
p
R
e
f
[
5
2
]
q
R
e
f
[
5
3
]
r
R
e
f
[
5
4
]
2.2. LITERATURE REVIEW 13
2.2 Literature Review
2.2.1 Introduction
To solve for the energy of an atomic system, one needs only to solve the Schr odingers
Equation, which governs any atomic system. However, it is tractable only for the
hydrogen atom and a number of hypothetical cases. Many approximate methods
were developed to solve it for larger systems, but they all had some major disad-
vantages one way or another. The development of Density Functional Theory by
Hohenberg-Kohn [55] and the elucidation of a possible solution method by Kohn-
Sham [56] were the keys that unlocked the Schr odingers Equation and enabled
accurate calculations of the energy of an atomic system possible.
For the years after Kohn-Shams paper, DFT calculations were performed only
with single-atom systems. Rapid advances in computing technology have made
it possible to perform rst-principles calculations on multi-atom systems such as
intermetallic compounds on computers of modest scale, such as desktop computers
or small clusters.
In general, calculations based on DFT have shown to be able to provide re-
liable predictions of experimental properties, and have seen increased use due to
the decreasing cost vs performance of computing technology. For example, the
lattice constants of many metals have been shown to be accurately predicted. Cal-
culations involving complicated compounds have been shown to be a useful check
against experimental data. It is known that the choice of exchange-correlation
functional can aect the results - generally the Generalized Gradient Approxima-
tion (GGA) is more accurate but the Local Density Approximation (LDA) must
also be considered. The LDA and GGA are described in Section 2.3.2.
This section reviews the current state of research in DFT. The primary output
of any DFT calculation is the energy of the system, thus the various analysis that
can be done as a result of this are rst reviewed.
2.2.2 Geometry Optimization
The geometry optimization procedure is usually the rst step in any DFT calcu-
lation of the energy. In any perfect crystal, the independent degrees of freedom
present are the lattice constants and the internal parameters. The minimum energy
is then obtained by minimizing these degrees of freedom with respect to the DFT
energy. Upon the successful completion of the procedure, the values of the degrees
of freedom obtained constitute the prediction by DFT of the crystal structure.
There is much evidence that the prediction by DFT of the crystal structure
can be accurate. The lattice constants of many elements were calculated [57] to
an accuracy of less than 3%. It was also shown that the GGA typically produces
lattice constants that are larger than the experimental value, and the LDA smaller.
Information on the internal structural parameters can also be obtained. The
investigations by Albe et al. [58] and Weirich [59] into the structural parameters
of various Ti-Se intermetallic compounds showed agreement with experimental
measurements even for complex crystal structures involving up to thirty-one atoms.
It is certainly possible to perform calculations in cases where experimental
information is sparse. Recently, DFT calculations were applied to obtain the
lattice constants, internal parameters and elastic constants of Ti
3
SiC
2
, a newly
synthesized material that shows many desirable properties [60, 25].
The optimization of the geometry need not be conned to perfect crystals.
Most DFT calculations have been with bulk crystals, but there have been some
examples of work within non-bulk conditions. Kang et al. used DFT calcula-
tions of the surface energy of Si and Ge as a yardstick to evaluate the merits of
dierent molecular dynamics potentials of these two atoms [61]. Carling et al.
performed DFT calculations of the vacancy formation energy of Al[62]. Under pe-
riodic boundary conditions, a large supercell of 64 atoms containing one vacancy
was required so that the vacancy would not interact with its copies. They showed
that the DFT prediction of the energy with the GGA was not as accurate as that
obtained with the LDA.
2.2.3 Making use of the energy
For any compound, the ground state is the crystal structure with the lowest en-
ergy. DFT enables the energy of this crystal structure to be calculated to obtain
the enthalpy of formation. Any structure with a higher value of the energy is
eliminated as a candidate for the ground state. Because of the very large number
of possible combinations, predicting the ground-state crystal structure from the
stoichiometry alone is a highly challenging problem.
What is more tractable with existing software and computational resources is to
trawl through several common structures, calculating the energy of each structure
to identify the structure with the lowest energy. The outcome is then compared
with what is observed experimentally. This is the approach taken by Ghosh et al.
in the investigation of the phase stability of Ti-Zn alloys[63]. The ground-state of
four Ti-Zn alloys was correctly predicted. Curtarolo et al. further developed this
idea and reported a set of high-throughput DFT calculations involving 80 alloy
systems[64]. A database of structures corresponding to various stoichimetries was
set up and calculations for each alloy system was done by running through the
database. Curtarolo achieved a successful prediction reliability of more than 90%
(four dierent measures of reliability were reported) with the 14080 calculations
of energies performed in total.
Values of the enthalpy of formation are also used for thermodynamic mod-
elling, in the calculation of phase diagrams (CALPHAD). DFT calculations of the
enthalpy of formation are used to ll in gaps when experimental data are unavail-
able. Turchi et al used DFT to calculate the heat of formation of Ni
2
Cr, Ni
2
Mo
and Ni
2
W alloys for inputs in the calculation of the ternary Ni-Cr-Mo phase dia-
gram [65]. The same was done for Fe
3
AlC to obtain the Fe-Al-C phase diagram
[66].
The idea that the ground state is the crystal structure with the lowest energy
can also be used to predict phase transitions at higher pressures. The crystal struc-
ture with the lowest free energy at any pressure is the one that would be observed.
By calculating the energy of various known Barium Oxide crystal structures as
a function of pressure, Uludogan et al. showed that the pressures at which the
phase transitions are predicted to occur was in good agreement with experimental
observations [67].
If there is some doubt over the crystal structure of a material, the DFT cal-
culations can help identify the possibilities, again using the principle of minimum
free energy. Through the evaluation of several candidate structures, the correct
distribution of B and C atoms in B
4
C was identied [68].
The energies calculated by DFT can also be used to build up a database to be
used in developing interatomic potentials and verifying the potentials for molec-
ular dynamics (MD) simulations. This idea was rst mooted by Robertson et al.
who calculated the energies of Al in various structures with coordination number
ranging from 0 to 12 [69, 70], and showing that it is possible for MD potential
models to reproduce the trends seen. The database of energies was expanded to
include data for all sorts of hypothetical structures [71, 72]. Subsequently, this
idea of including ab-initio energies in a potential tting database was adopted by
several researchers, including Mishin et al. in the development of a potential for
Cu [73] and the Fe-Ni system [74].
2.2.4 Elastic Constants
With DFT calculations, two strategies have been employed to obtain the elastic
constants. The rst strategy is to take the second derivative of the energy with
respect to strain in order to yield the elastic constants. Combinations of strains
are employed to take advantage of symmetry [75]. The second strategy involves
straining the crystal at small strains, and calculating the resulting stresses[76].
In the light of a few conicting experimental reports on the elastic constants of
TiB
2
, Milman et al. performed DFT calculations that made it possible to identify
the set of experimental results that is most likely to be accurate [23]. Similar
calculations for BeO were found to be in good agreement with experimental results
[24].
The elastic constants can also play a role in predicting crystal structures. If
the crystal is stable, then the elastic constants meet some stability criteria that are
described in the theory section. The structure of PtN is subject to some dispute,
hence Patil et al. calculated the elastic constants of a few candidate structures
[77]. Some structures resulted in negative elastic constants, some had all positive
elastic constants but did not meet the stability criteria.
The calculation of the elastic constants also plays a major role in Geophysics,
and has great potential to aid in understanding the dynamics of the materials in
the earths crust [78]. An example of such a calculation is the work by Karki et al.
to calculate the elastic constants, and their dependance on external pressure, of
MgSiO3 which is present in the earths mantle [76]. As the speed of propagation of
elastic waves in a material is related to the elastic constants, the calculations were
used to interpret seismological data on the speed of wave propagation, showing
good agreement within 6-8 % error.
2.2.5 DFT Calculations of Intermetallic compounds
The elastic constants of various ordered intermetallic alloys were rst calculated by
Mehl et al. [79, 80] because no experimental data were available then. Calculations
of the lattice constants, elastic constants and thermal expansion coecient have
been performed for the Laves Phases, intermetallic compounds that crystallize in
the stoichiometry AB
2
[81, 82, 83]. Le page et al. used DFT calculations to pre-
dict the internal coordinate for several intermetallics with CuAl
2
-type structures,
showing good agreement in all cases [84].
DFT calculations on the compounds that are the subject of this thesis are
rare. In a paper describing both computational work and experiments, Ghosh
published the results of calculations of the lattice constants and bulk modulus of
Cu
6
Sn
5
and other Cu-Sn compounds [85]. The lattice constants calculated by DFT
were in good agree with those published experimentally by Larsson [86]. The bulk
modulus obtained from DFT was within the experimental error of the experiments
conducted. The other DFT calculation involving the Larsson structure was by
Sharma et al., whose geometry optimization study of the Larsson cell predicted
only the internal coordinates[30]. This was done by keeping the cell vectors xed.
There have also been calculations on Sn-based oxides. Goniakowski et al. stud-
ied the bulk and surface properties of SnO
2
[87]. They showed that the GGA
functional resulted in surface energies that were 30% lower than that obtained by
the LDA. Meyer et al. calculated the equilibrium structure of SnO and explained
in detail the construction of the Sn pseudopotential in order to obtain results that
correspond to experimental observations [88].
2.2.6 Conclusion
Through this review, it is shown that by just being able to calculate the energy of
a known atomic system accurately, a great deal of information on the properties
and behaviour of materials can be obtained.
In summary, for any crystal of known composition and structure, by calculating
the energy alone, the structural properties (lattice constants, internal parameters)
of any crystal can be predicted. By taking the dierence in energy with the con-
2.3. THEORY 19
stituent atoms, the enthalpy of formation can be predicted. DFT calculations
also make calculations of hypothetical structures possible and these have led to
predictions of new materials and clarications on the structures of existing mate-
rials. Taking the second derivative of the energy with respect to the strain yields
the elastic constants, predictions which can be compared with experimental data.
These capabilities enable three things to happen: new knowledge to be gained from
materials commonly used today, to ll gaps in the knowledge for newly-synthesized
materials, and to identify new materials.
DFT calculations will continue to be an important part of materials research.
It is no wonder that DFT has gone beyond the traditional disciplines of physics and
chemistry to nd applications in geophysics [78], calculation of phase diagrams [65],
development of Lithium-ion batteries [30] and, as this thesis shall demonstrate, in
the eld of electronics packaging.
2.3 Theory
This section explains some technical terms found in the rest of the chapter.
2.3.1 Crystal structure
The vast majority of materials found in nature have their atoms in ordered ar-
rangements, where the patterns are repeated periodically in three dimensions. This
phenomenon is known as a crystal. A crystal is fully specied by two components
1. the basis, which is the repeating unit of atoms, and
2. the underlying Bravais lattice
A bravais lattice is an innite array of points r that satises equation (2.1),
where n
1
, n
2
and n
3
belong to the set of all integers. a
1
, a
2
and a
3
are known as
2.3. THEORY 20
primitive vectors, which are non-unique and can be dened such that they do not
all lie on the same plane.
r = n
1
a
1
+n
2
a
2
+n
3
a
3
(2.1)
As an example, Figure 2.1 shows how a two-atom basis is put on a hexagonal
bravais lattice to form a honeycomb lattice.
Examining the hexagonal bravais lattice in Figure 2.1, one notices that it retains
its symmetry with 60
rotations ie. the crystal overlaps with itself. When the basis
is added to form the honeycomb crystal, the symmetry is obtained only with 120
rotations. As such, dierent crystal structures in nature have dierent symmetry

operations. The complete set of symmetry operations and translations that leave
the crystal invariant is the space group, and the crystal structures of materials
found in nature belong to one of 230 space groups.
A primitive unit cell is the smallest region in the lattice that contains the full
information of the crystal [89]. When it is shifted by the lattice vectors, it will not
overlap itself [90]. A primitive unit cell is not unique - it is possible to dene many
dierent primitive unit cells for a lattice. A denition of a primitive cell that is
unique is the Wigner-Seitz cell, which is the region in space that is closer to that
point than any other point [90].
It is common to use non-primitive unit cells to represent the crystal symmetry.
The face-centered cubic lattice (FCC), the body-centered cubic lattice(BCC) and
the diamond-cubic (DC) lattice are commonly represented by a cubic unit cell,
rather than their respective primitive cells. The sizes of these cells, and of those
with lower symmetry are specied by the lattice constants.
These dierent denitions of the unit cell are illustrated for a two-dimensional
case using the (100) plane of a FCC crystal in Figure 2.2.
2.3. THEORY 21
Figure 2.1: Illustrating the idea of a lattice with basis. The result is a honeycomb
lattice.
Figure 2.2: Illustrating the various denitions of the unit cells using the (100)
plane of the FCC crystal. Dierent possibilities for the lattice vectors are also
shown.
2.3. THEORY 22
Any bravais lattice has a corresponding reciprocal lattice whose primitive vec-
tors are given by
b
1
= 2
a
2
a
3
a
1
. ( a
2
a
3
)
(2.2a)
b
2
= 2
a
3
a
1
a
1
. ( a
2
a
3
)
(2.2b)
b
3
= 2
a
1
a
2
a
1
. ( a
2
a
3
)
(2.2c)
A point

k in the reciprocal lattice can now be written as
k = k
1
b
1
+k
2
b
2
+k
3
b
3
(2.3)
It is observed that

k.r gives integral multiples of 2. This results in another
denition of the reciprocal lattice, where

k is dened to be the set of vectors that
satisfy
exp(i
k.r) = 1 (2.4)
A consequence of this is that each point in the reciprocal lattice corresponds
to a plane in the normal lattice.
The various denitions for the unit cell also apply to the reciprocal lattice.
The Wigner-Seitz cell of the reciprocal lattice is known as the Brillouin zone. This
denition will be applied in the section on Density Functional Theory.
2.3.2 Density Functional Theory
Schr odingers equation
Any atomic system is governed by the Schr odingers equation,
H = E (2.5)
2.3. THEORY 23
where

H is the Hamiltonian operator, is the wavefunction and E is the
energy of the atomic system. The Hamiltonian operator includes operators for
the kinetic energy, the electron-nucleus energy and the electron-electron energy.
All the information that can be gained for any atomic system is contained in the
wavefunction [91]. However, the wavefunction is a function of all the positions
of the nuclei and the electrons, making it intractable except for the simplest cases,
such as the free hydrogen atom, with one electron orbiting a nucleus with a single
positive charge. With most atomic systems, the Born-Oppenheimer approximation
applies. This assumes that the nuclei in an atomic system, being much heavier
than the electrons, move so slowly in comparison to the electrons such that they
are eectively stationary.
Hohenberg-Kohn Functional
In 1964 Hohenberg and Kohn (HK) showed that the energy of a system of electrons
under the Born-Oppenheimer approximation may be expressed as a unique func-
tional of the electron density instead of the wavefunction, and that the electron
density that minimizes the energy is the ground state density [55].
Thus, the electron density , a function of only the spatial coordinates r, re-
places the wavefunction, and the following energy functional
E[] = T[] +U[] +V [] (2.6)
replaces the Schr odingers Equation, where T[(r)] is the kinetic energy func-
tional, U[(r)] is the electron-electron interaction functional and and V [(r)] is
the electron-nucleus interaction functional.
2.3. THEORY 24
Kohn-Sham Method
The HK functional in Equation (2.6) was given only on a conceptual basis, with
nothing said about the exact form of these functionals. In 1965, Kohn and Sham
(KS) proposed that the HK functional be modied as follows,
E[] = T[] +U[] +V [] +E
xc
[] (2.7)
where approximations are made in functionals T[] and U[]. The main idea
provided by KS is that the exchange-correlation functional E
xc
[] will account for
all energies that have been neglected by approximations made in these functionals.
This functional needs to be determined and examples will be given in the later
part of this section.
KS then showed that the electron density can be determined by rst solv-
ing for the set of one-electron wavefunctions
i
using the following form of the
Schr odingers equation for a non-interacting system of electrons
_
T
s
+v
s
(r)
_
i
= E
i
i
(2.8)
where

T
s
is the kinetic energy operator, v
s
(r) is the eective potential that
represents electron-nucleus interactions and the subscript s represents the idea
that these belong to a system of non-interacting electrons. The electron density
can then be obtained by
=
i
f
i
|
i
| (2.9)
where f
i
is the occupancy of each wavefunction. This electron density would
represent the actual electron density of an interacting system, under the condi-
tion that the eective potential v
s
(r) is given by the functional derivatives of the
functionals in Equation (2.7), U[] +V [] +E
xc
[] with respect to .
2.3. THEORY 25
Summary of Approximations
Having made approximations in the functionals, what remains to be done is to
specify appropriate forms of the one-electron wavefunctions
i
, the form of the
exchange-correlation functional E
xc
[] and any other simplifying assumptions.
For the DFT calculations in this chapter, the software CASTEP is used [92, 93].
It makes use of the planewave-pseudopotential method. The term planewave
refers to the choice of planewaves as the basis set for the KS wavefunctions
i
in
Equation (2.8). pseudopotential refers to a simplifying assumption in the func-
tional V [] where only the valence electrons are considered. Finally, two forms of
the exchange-correlation functional are considered, the Local Density Approxima-
tion (LDA) and the Generalized Gradient Approximation (GGA).
Planewaves
Planewaves are waves in three dimensions (3D) and have the typical form
(r) = Aexp(i

k
w
.r) (2.10)
where A is the amplitude, r is the position vector in 3D space and

k
w
is the
wavevector. In one dimension, the crests and troughs of a wave are points, but in
3D, the wavefronts are planes whose normals are the wavevector.
Under periodic boundary conditions, with the choice of planewaves for the KS
wavefunctions, the resulting expression is
i
=
G
c
k,G
exp[i(
k +

G).r] (2.11)
where the wavevectors, (
k +

G), consists of the reciprocal lattice vectors

G and
the position vectors of points in the Brillouin zone

k, known as k-points. c
k,G
are parameters to be determined by solving Equation 2.8.
2.3. THEORY 26
A suciently large set of k-points is able to represent the theoretically innite
number of electrons in a bulk solid. In CASTEP, these points are generated by the
Monkhorst-Pack scheme [94]. The summation in Equation (2.11) is terminated at
a value known as the energy cuto specied by the user. This species the value
of
h
2m
k +

G
2
, and is thus a limit on the wavevectors of the planewaves [95].
Hence, the use of planewaves makes it necessary for the user to specify two
factors that aect the accuracy of the energy obtained from calculations. Hence,
the convergence of the energy of the system must be examined with respect to the
1. size of the k-point mesh, and
2. value of the energy cuto.
Pseudopotential
If all the electrons in the atom were considered, there would be a large number of
planewaves resulting in an impossibly large amount of computational time needed.
As it is known that the bonding in solids depends on the valence electrons most
of the time, the eect of the inner core electrons is replaced by a pseudopotential.
This results in making appropriate modications to the form of V [], and the
electron density is now based just on the valence electrons.
Exchange-correlation functional
Numerous forms have been proposed for the exchange-correlation functional. There
are two main forms,
1. Local Density Approximation (LDA), which is a functional based entirely
upon the electron density
E
xc
[] =
_
f()d
3
r (2.12)
2.3. THEORY 27
The form of the LDA used is that based on the data by Ceperley-Alder for
the homogenous electron gas.
2. Generalized Gradient Approximation (GGA), whose functional depends on
the gradient of the electron density as well
E
xc
[] =
_
d
3
rf(, ) (2.13)
The GGA functional used in this work is by Perdew, Burke and Enzerhof
[96, 97, 98].
2.3.3 Elasticity of Single Crystals
Stiness Matrix
The calculations in DFT produce the ground-state energy of an atomic system,
from which the elastic properties can be obtained by means of a stress-strain linear
t or taking the second derivative of the energy. Before providing details of these
calculations in the next section, the theory of elasticity for crystals is described
here to provide a background.
Using Einsteins indicial notation, the stress-strain relationship is
ij
= c
ijkl
kl
(2.14)
where the indices i, j, k, l indicate summation over 1, 2, 3.
c
ijkl
are the stiness coecients. The stiness matrix formed by c
ijkl
depends
on the symmetry of the crystal. For a crystal of monoclinic symmetry, it is given
as follows
2.3. THEORY 28
Table 2.2: Constraints on the elastic constants for dierent crystal systems
System Constraints on elastic constants
Orthorhombic c
15
, c
25
, c
35
, c
46
= 0
Hexagonal c
15
, c
25
, c
35
, c
46
= 0, c
55
= c
44
,c
11
= c
22
and c
66
=
1
2
(c
11
c
12
)
Cubic c
15
, c
25
, c
35
, c
46
= 0,c
11
= c
22
= c
33
, c
12
= c
13
= c
23
, c
44
= c
55
= c
66
C =
_
_
c
1111
c
1122
c
1133
0 c
1113
0
c
1122
c
2222
c
2233
0 c
2213
0
c
1133
c
2233
c
3333
0 c
3313
0
0 0 0 c
2323
0 c
2312
c
1113
c
2213
c
3313
0 c
1313
0
0 0 0 c
2312
0 c
1212
_
_
(2.15)
Often, a contracted notation is used where only two subscripts are given. The
stiness matrix for the monoclinic crystal is then given as
C =
_
_
c
11
c
12
c
13
0 c
15
0
c
12
c
22
c
23
0 c
25
0
c
13
c
23
c
33
0 c
35
0
0 0 0 c
44
0 c
46
c
15
c
25
c
35
0 c
55
0
0 0 0 c
46
0 c
66
_
_
(2.16)
For crystals of higher symmetry, Table 2.2 lists the various constraints on the
elastic constants.
Compliance Matrix
It is also possible to express strain in terms of stress. The resulting expression is
ij
= s
ijkl
kl
(2.17)
2.3. THEORY 29
where s
ijkl
are the elastic compliances, whose matrix is obtained from inverting
the stiness matrix,
S = C
1
(2.18)
Stability Condition
Having dened the elasticity matrices, it is possible to determine the acceptable
values of the elastic constants. This is necessary as DFT calculations do not always
produce converged values at the rst instance. Using Einsteins indicial notation,
the change in a strain energy of a crystal at the ground state is expressed as
E =
1
2
c
ij
j
(2.19)
Since any deviation from the ground state structures means an increase in the
energy of the crystal, hence E must be zero when the strains are zero and must be
positive when the strains are non-zero. In fullling these conditions, E is said to be
positive denite. For a cubic crystal, Born showed that the following restrictions
on the elastic constants apply
c
11
> 0, c
11
c
12
> 0, c
44
> 0 (2.20)
The physical meaning of these expressions is that a positive strain, normal
or shear, must produce a corresponding positive stress. To satisfy the positive
denite requirement, for crystals of lower symmetry, all the eigenvalues of the
stiness matrix C must be positive.
Orientation Dependence
Once the suitability of the elastic constants obtained has been determined, one
analysis that can be carried out is to examine the orientation dependence of the
2.3. THEORY 30
single-crystal Youngs Modulus. The Youngs Modulus of a single crystal is dened
as
1
s
11
(which is
1
s
1111
in the four-indices notation). This quantity is not invariant
to rotation, and the orientation dependence is given by Nye [99] as
s
1111
= a
1m
a
1n
a
1p
a
1q
s
mnpq
(2.21)
where a
1m
is the direction cosine between the original axis m and the rotated
x
1
axis. This can be expanded to give the expression for a mononclinic crystal,
given in Appendix A.1. From this, the directions where the Youngs Modulus is
highest and lowest can be identied, which suggests the orientations of strongest
and weakest bonding. For example, in the FCC structure, the closest packed plane
is the (111) direction. Many materials, though not all, have the [111] directions as
those with the largest Youngs Modulus [99].
2.3.4 Bounds on Polycrystalline Elastic Moduli
Voight and Reuss Averages
It is possible to relate single crystal properties to that at the macroscopic scale. Let
< c
ij
> denote the average over all orientations of c
ij
. Voight made the assumption
that the strain is constant throughout a polycrystalline body [100]. Using this
assumption, he showed that < c
44
> was equal to that of
1
2
(< c
11
> < c
12
>).
In other words, the averaged elastic constants represent an isotropic material.
Thus, the Voight shear modulus G
V
, dened as < c
44
>, was found to be
G
V
=
1
15
[c
11
+c
22
+c
33
(c
12
+c
13
+c
23
) + 3 (c
44
+c
55
+c
66
)] (2.22)
and the Voight average for the bulk modulus B
V
is
2.3. THEORY 31
B
V
=
1
3
(< c
11
> +2 < c
12
>) (2.23)
=
1
9
[c
11
+c
22
+c
33
+ 2 (c
12
+c
13
+c
23
)] (2.24)
Reuss suggested the alternative assumption of the stress being constant through-
out a polycrystalline body [101]. Similarly considering the orientation averages,
the resulting Reuss shear modulus G
R
was found to be
15
G
R
= 4 (s
11
+s
22
+s
33
) 4 (s
12
+s
13
+s
23
) + 3 (s
44
+s
55
+s
66
) (2.25)
.
and the bulk modulus due to Reuss assumption B
R
is given by
1
B
R
= s
11
+s
22
+s
33
+ 2 (s
12
+s
13
+s
23
) (2.26)
Hill Bounds
Hill showed that the strain energy of a polycrystalline solid under Voights as-
sumption and Reuss assumption would be higher than one without the inuence
of either of these assumptions [102]. Thus, it was established that the Voight aver-
ages and Reuss Averages are upper bounds and lower bounds on the polycrystalline
elastic moduli respectively, i.e.
B
R
B B
V
(2.27a)
G
R
G G
V
(2.27b)
Bounds can also be placed on the Youngs Modulus
2.3. THEORY 32
E
R
E E
V
(2.28)
as a result of the following relationship for the Youngs Modulus E
1
E
=
1
3G
+
1
9B
(2.29)
These bounds are known as the Voight-Reuss-Hill (VRH) bounds.
Hashin-Shtrikman Bounds
In 1962, Hashin and Shtrikman (HS) developed new principles from which bounds
could be derived [103]. HS rst dened a polycrystalline body that undergoes de-
formation as having a stress eld (x) and strain eld (x). They then showed that
if the strain eld (x) were to be applied to a reference body that is homogenous
with shear modulus G
0
and bulk modulus K
0
(that have yet to be determined),
the resulting strain energy functional can be dened in terms of the properties of
this homogenous reference body. Bounds on the the eective elastic moduli of the
polycrystalline body are then obtained by taking the maximum point and mini-
mum point of this functional. This procedure results in the following inequalities
for the eective shear modulus G
and the bulk modulus B
,
B
or K
0
+
3K
1
+ 2K
2
3 +(3K
1
+ 2K
2
)
(2.30a)
G
or G
0
+
K
2
1 + 2K
2
(2.30b)
where
=
3
3B
0
+ 4G
0
(2.31a)
2.4. METHODOLOGY 33
=
3(B
0
+ 2G
0
)
5G
0
(3B
0
+ 4G
0
)
(2.31b)
Whether the inequalities in Equation (2.30) result in upper bounds (i.e. ) or
lower bounds (i.e. ) are determined by appropriate choices of B
0
and G
0
. K
1
and
K
2
have dierent forms for dierent crystal symmetries. Choices for B
0
and G
0
,
and expressions for K
1
, K
2
in the monoclinic system [104], orthorhombic system
[105] and hexagonal system [106] were derived by Watt et al.. These studies,
along with review articles by Watt et al. [107] and Watt [108], also provide data
to support the general observation that bounds produced by the HS method are
always tighter than the VRH bounds.
2.4 Methodology
With the theory explained, this section goes on to describe the tools and procedures
used in obtaining meaningful results with Density Functional Theory.
2.4.1 Crystal structure of the Intermetallic Compounds
Density Functional Theory calculations of any material always begin with the
knowledge of its crystal structure. Fortunately, the crystal structures of the inter-
metallic compounds discussed in this thesis are known. The structural data are
summarized in Table 2.3 and a description of each of them follows.
Cu
6
Sn
5
On the Cu-Sn phase diagram, Cu
6
Sn
5
is denoted as the -phase. It is known that
a phase transformation of the phase occurs at 460K to a low-temperature
-
phase [109]. The crystal structure of Cu
6
Sn
5
has attracted much interest since the
beginning of the 20th century. It was rst thought the structure was a NiAs-type
hexagonal crystal structure, with various studies agreeing on the lattice constants
2.4. METHODOLOGY 34
Figure 2.3: Cu
6
Sn
5
in the NiAs-type structure according to Gangulee et al.. The
larger spheres represent the Sn atoms and the smaller spheres represent the Cu
atoms. The excess Cu atoms ll the interstitial sites to make up the 6:5 stoichiom-
etry.
of a = 4.19

A and c = 5.04

A (Figure 2.3). It was also suggested that the excess
Cu atoms occupy the interstitial sites. The low temperature phase
-Cu
6
Sn
5
was
thought to be a vefold replication of the NiAs cell in both the a direction and c
direction [110, 111, 112, 109, 113].
The structure of
-Cu
6
Sn
5
was rst claried by Larsson et al. [86]. The
structure is monoclinic (Figure 2.4). Chang et al. studied the Cu
6
Sn
5
layers that
were formed at the solder joint and found that the lattice constants were close to
that reported by Larsson et al. [114, 115]. The positions of all the atoms in the
phase are explicitly stated by Larsson et al., hence the calculations in this chapter
will be based on this structure.
Cu
3
Sn
Cu
3
Sn is denoted as the phase, and has a basic Cu
3
Ti-type structure (Figure 2.5)
which crystallizes into a long-period superstructure [116, 117, 110]. Watanabe et
al. claried the nature of this superstructure, showing that the basic unit cell is
orthorhombic involving 80 atoms [116].
2.4. METHODOLOGY 35
Figure 2.4: Unit Cell of Cu
6
Sn
5
according to Larsson et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Cu atoms.
DFT Calculations with 80 atoms will be extremely computationally demanding.
As such, the calculations will be done with the Burkhardt structure.
Ni
3
Sn
4
Interest in the crystal structure of Ni
3
Sn
4
has been limited to studies by Nowotny
[118] and Jeitschko et al [119], both reporting a monoclinic structure of space
group C2/m (Figure 2.6). Furuseth et al. went further and investigated the
eect of temperature on the lattice constants, studying their values from 300K to
1100K[120].
Ag
3
Sn
There is some controversy over the crystal structure of Ag
3
Sn. Fairhurst [121]
reports an orthorhombic structure of space group Pmmn, while Nial et al. [122]
and Ellner et al. [123] suggest an orthorhombic structure of space group Cmcm.
The positional parameters for the latter structure are not available, thus our cal-
culations employ the structure by Fairhurst.
2.4. METHODOLOGY 36
Figure 2.5: Unit Cell of Cu
3
Sn according to Burkhardt et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Cu atoms.
Figure 2.6: Unit Cell of Ni
3
Sn
4
according to Jeitschko et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Ni atoms.
Figure 2.7: Unit Cell of Ag
3
Sn according to Fairhurst et al.. The larger spheres
represent the Sn atoms and the smaller spheres represent the Ag atoms.
2.4. METHODOLOGY 37
T
a
b
l
e
2
.
3
:
C
r
y
s
t
a
l
s
t
r
u
c
t
u
r
e
d
a
t
a
o
f
t
h
e
f
o
u
r
i
n
t
e
r
m
e
t
a
l
l
i
c
c
o
m
p
o
u
n
d
s
C
u
6
S
n
5
(
N
i
A
s
)
C
u
6
S
n
5
(
S
u
p
e
r
c
e
l
l
)
C
u
3
S
n
N
i
3
S
n
4
A
g
3
S
n
C
i
t
a
t
i
o
n
G
a
n
g
u
l
e
e
e
t
a
l
.
[
1
0
9
]
L
a
r
s
s
o
n
e
t
a
l
.
[
8
6
]
B
u
r
k
h
a
r
d
t
e
t
a
l
.
[
1
1
7
]
J
e
i
t
s
c
h
k
o
e
t
a
l
.
[
1
1
9
]
F
a
i
r
h
u
r
s
t
e
t
a
l
.
[
1
2
1
]
S
p
a
c
e
G
r
o
u
p
1
9
4
,
P
6
3
/
m
m
c
1
5
,
C
2
/
c
5
9
,
P
m
m
n
1
2
,
C
2
/
m
5
9
,
P
m
m
n
D
e
s
c
r
i
p
t
i
o
n
H
e
x
a
g
o
n
a
l
M
o
n
o
c
l
i
n
i
c
O
r
t
h
o
r
h
o
m
b
i
c
M
o
n
o
c
l
i
n
i
c
O
r
t
h
o
r
h
o
m
b
i
c
a
(
A
)
4
.
1
9
1
1
.
0
2
2
5
.
5
2
1
2
.
2
1
4
5
.
9
6
8
b
(
A
)
-
7
.
2
8
2
4
.
7
7
4
.
0
6
0
4
.
7
8
0
2
c
(
A
)
5
.
0
4
9
.
8
2
7
4
.
3
5
5
.
2
1
9
5
.
1
8
4
3
)
9
8
.
8
4
-
1
0
5
.
0
-
f
.
u
.
p
e
r
c
e
l
l
2
4
2
2
2
U
n
i
t
c
e
l
l
F
i
g
u
r
e
2
.
3
F
i
g
u
r
e
2
.
4
F
i
g
u
r
e
2
.
5
F
i
g
u
r
e
2
.
6
F
i
g
u
r
e
2
.
7
2.4. METHODOLOGY 38
Table 2.4: Main computational resources used
System Type Manufacturer/Model Central Processing Unit (CPU) Memory
Workstation HP xw6200 2 Intel Xeon 2.4 GHz 2.5 GB
Server HP Integrity rx4640 4 Intel Itanium 2 1.5 GHz 8 GB
2.4.2 Software
The ab-initio calculations are performed with the software CASTEP [92, 93], dis-
tributed with Materials Studio Software suite version 4.0 developed by Accelrys
Inc.
2.4.3 Computational Resources
Table 2.4 lists the main computational resources used for the calculations in this
chapter. Apart from these, a number of Pentium 4 3.06 GHz general-purpose
computers were used as well.
2.4.4 Geometry Optimization
Firstly, the experimentally-determined crystal structure is input into the software.
Using the Broyden-Fletcher-Goldfarb-Smith method (BFGS) [124], the energy of
the system is minimized with respect to the lattice constants and internal atomic
coordinates. This procedure is referred to as Geometry Optimization. The con-
vergence thresholds used were (i) energy change below 0.005 meV/atom, (ii) force
less than 0.01 eV/
A, (iii) stress less than 0.02 GPa and (iv) change in displace-
ment less than 0.0005

A. The result of this procedure are predictions of the lattice
constants and the internal coordinates, both of which may be compared directly
with experimental data.
2.4.5 Accuracy Settings
The accuracy of the geometry optimization calculations depend on four inputs:
2.4. METHODOLOGY 39
1. the exchange-correlation functional,
2. the pseudopotential
3. the k-point setting,
4. the energy cuto.
The exchange-correlation functional accounts for all the energies neglected by
approximations made in the other functionals in Equation (2.7). The choice is
between the GGA and the LDA. Currently, many practitioners of DFT encourage
the comparison of dierent types of exchange-correlation functions so that a sense
of the accuracy of the calculations can be obtained [125]. In this work, the LDA
by Perdew-Zunger [126] and the GGA by Perdew, Burke and Enzerhof [96, 97, 98]
will be used. It is known that the LDA tends to produce larger values of the elastic
constants compared to the GGA.
As mentioned earlier, the pseudopotential replaces the calculation of the core
electrons, leaving the valence electrons. The pseudopotentials used are of the Van-
derbilt ultrasoft-type [127] obtained from the database provided with CASTEP.
Current opinion among DFT practitioners is that with each type of exchange-
correlation functional, a pseudopotential generated with the same functional should
be used [125, 128, 129].
At the same time, it is stated in the CASTEP user manual that among the pseu-
dopotentials in the database, only the LDA-generated ones have been veried with
the GGA. This is seen from the documentation for the LDA-generated pseudopo-
tential of Cu and Sn, reproduced in Figure 2.8. No verication has been performed
for the GGA-generated pseudopotentials [130]. As such, the primary choice in this
work are the LDA-generated pseudopotentials for both GGA and LDA. To verify
the accuracy of these pseudopotentials, calculations for the monoatomic metals
are rst performed to ensure that their experimental properties are reproduced
2.4. METHODOLOGY 40
LDA-generated Sn pseudopotential Sn 00.usp:
Validation test
---------------
#1 SnO2 (SG 136), exp. a=4.7373 and c=3.1864, CASTEP (GGA,
PRECISE) gives 4.709 (-0.6%) and 3.150 (-1.1%)
#2 Sn metal (SG 227), exp. a=6.4912, CASTEP (GGA, PRECISE) gives
6.408 (-1.2%)
LDA-generated Cu pseudopotential Cu 00.usp:
Validation test
---------------
#1 Cu metal, exp.lattice constant a=3.6147, CASTEP with GGA
(FINE cutoff) gives 3.631 (+0.5%)
#2 Cu2O (SG 224), exp.lattice constant a=4.2696, CASTEP with GGA
(PRECISE) gives 4.2533 (-0.4%)
Data taken from Materials Studio Pseudopotential database [130]
Figure 2.8: Documentation Provided in LDA-generated pseudopotential les for
Cu (Cu 00.usp) and Sn (Sn 00.usp)
accurately. The lenames of the LDA-generated pseudopotentials are shown in
Table 2.5.
The k-point mesh and the energy cuto both determine the accuracy of the
one-electron wavefunction given in Equation (2.11). For each exchange-correlation
function, the calculations are repeated at dierent k-point mesh and energy cuto.
This is necessary because it should not be taken for granted that the k-point
setting and energy cuto chosen are adequate[125]. Due to the computational cost
involved, it is not feasible to examine convergence for each input separately, hence
convergence is examined by increasing both the energy cuto and k-point mesh
simultaneously.
2.4. METHODOLOGY 41
Table 2.5: Filenames of CASTEP pseudopotentials used
Element Filename
Ag Ag 00.usp
Cu Cu 00.usp
Ni Ni 00.uspcc
Sn Sn 00.usp
Using the minimized lattice constants, the elastic constants are calculated. This is
done in the same way as Karki et al. [76] using a linear t to the stress vs strain.
Beginning with the structure minimized at each accuracy setting,
1. The unit cell is deformed with various strain modes according to the crystal
structure (listed in Table 2.6). Each strain mode is a combination of strains
applied to the crystal. Four strain modes are necessary to calculate all the
thirteen monoclinic elastic constants, while three modes are needed for the
nine orthorhombic constants.
2. For each strain mode, a set of geometry optimization calculations are per-
formed. The cell vectors are rst strained according to the strain mode.
They are then held xed while the energy of the unit cell is minimized with
respect to the internal atomic positions. This produces the
3. The elastic constants are obtained from a linear t to the stress-strain data.
For each strain mode, four values of strain are chosen arbitrarily at =
0.001, 0.003 by the software CASTEP, resulting in 16 separate geometry op-
timization runs for the calculation of elastic constants of a monoclinic unit cell.
In the energy-minimization runs, the convergence thresholds used were (i) energy
change below 0.001 meV/atom, (ii) force less than 0.002 eV/
A, and (iii) change in

displacement less than 0.0001

A.
2.4. METHODOLOGY 42
Table 2.6: Strain Patterns used in calculating elastic constants
Material Strain Pattern
Monoclinic (1)
11
,
23
, (2)
33
,
12
,(3)
22
, (4)
13
Orthorhombic (1)
11
,
23
, (2)
33
,
12
,(3)
22
,
13
It would be desirable if the elastic constants were also calculated by applying
a strain and taking the second derivative of the energy, so that consistency with
the results presented earlier can be demonstrated. However, this is not done in
this study due to the immense amount of computational time that it involves.
For the orthorhombic cell, a large number of strain values is needed, and the
energies obtained must be converged with respect to the energy cuto and k-
point mesh. At each value of the strain, the internal degrees of freedom must be
relaxed, which is a time-consuming process. Hence, a large number of geometry
optimizations is required, and this is feasible only if larger computing resources
are available.
An example of the scale needed is demonstrated in the study by Beckstein et
al. of the study of orthorhombic PtSi. For the nine elastic constants, they used
a total of fty-nine dierent strain values. A convergence criterion of 1.36 10
5
eV with respect to the k-point mesh was also used[75].
2.4.7 Validation
There are no reported single crystal elastic constants for comparisons to be made; a
similar situation encountered by Holm et al [25]. Nevertheless, indirect validation
can be performed by making use of the elastic constants to perform two sets of
analysis:
1. Elastic Anisotropy, and
2. Analysis of Bounds on Polycrystalline elastic moduli.
2.4. METHODOLOGY 43
The analysis of the elastic anisotropy is performed by using the expression
given in Appendix A.1 to plot the variation of the single-crystal Youngs Modulus
with crystal orientation. Directions of the largest Youngs Modulus then give an
indication of where the bonding is the strongest, and such information can then
be correlated with the packing in the crystal.
At the same time, in actual applications, single crystal properties are normally
not representative of the mechanical properties at larger length scales. Hence,
polycrystalline properties are calculated from the theoretical single crystal prop-
erties. The bounds on the shear modulus (G) and bulk modulus (B) of polycrys-
tals calculated from two approaches are presented - the Voight-Reuss-Hill (VRH)
bounds based on assumptions of uniform stress and strain distribution within the
polycrystal[102] and the Hashin-Shtrikman (HS) bounds derived from variational
principles [103, 104]. For each bound on G and B, the corresponding value of
the Youngs modulus (E) and the Poissons ratio () were obtained. For the HS
method, the formulation by Watt for orthorhombic crystals [105] and monoclinic
crystals [104] are used.
2.4.8 Summary
In this section, the methodology of the calculations employed in this chapter has
been described. The basic type of calculation is the geometry optimization pro-
cedure, of which four inputs determine the accuracy. Applying this procedure in
dierent ways enable predictions of the following
1. lattice constants
2. internal coordinates
3. the elastic constants
2.5. PROPERTIES OF MONOATOMIC METALS 44
Figure 2.9: Flowchart showing the methodology of the calculations in this chapter.
The lattice constants and internal coordinates are compared with experiment,
while further analysis with the polycrystalline bounds and elastic anisotropy are
performed with the elastic constants to validate the calculations. The methodology
is summarized in Figure 2.9
2.5 Properties of Monoatomic Metals
2.5.1 Introduction
In this section, the predictions for the four metals by DFT are examined. There
are two reasons for doing so. In the theory section, it was explained that DFT
calculations still involve making certain approximations and thus, validation is
necessary. Secondly, it is also necessary to have an indication of the typical accu-
racy that such DFT calculations can provide. Working with just the metals can
provide a speedy platform to do so, as DFT calculations involving more than one
or two atoms can be computationally intensive.
2.5.2 Lattice Constants and Elastic Constants of Ag Ni
and Cu
Geometry optimization calculations were performed with both the GGA and LDA
at dierent accuracy settings for Ag, Ni and Cu. With the minimized structures,
the corresponding elastic constants are calculated. These results are shown in
Table 2.7. For each exchange-correlation functional, the calculations are done
with two sets of accuracy settings. One setting is the default ultrane setting
provided by the software, and the other setting is a more stringent setting as shown
in Table 2.7. The LDA-generated pseudopotentials are used with both functionals.
The results show that better agreement with experimental results is obtained
with the GGA. It is known that the LDA tends to have an overbinding eect
[131]. The result of this eect is that the lattice constants are smaller than the
experimental values, and the elastic constants will be higher. This eect is seen in
Table 2.7. In many cases, the LDA elastic constants are at least 50 % higher than
the experimental values, whereas the GGA values are much closer.
2.5.3 Lattice Constants and Elastic Constants of Sn
Having calculated the properties of the transition metals, it is now necessary to
calculate the properties of Sn, which is the common element in the intermetallic
compounds studied in this thesis. Sn exists in two allotropes, -Sn (white tin)
a low-temperature phase with a diamond cubic structure that transforms to -Sn,
tetragonal structure, at 286 K. Calculations are performed for both these forms,
and the results are shown in Table 2.8.
Table 2.7: Calculated lattice constants and elastic constants (GPa) for Ag, Cu and
Ni using LDA-generated pseudopotentials
a (
A) c
11
c
12
c
44
Ag
Expt[132] 4.08 123 92 45.3
GGA 500 eV, 14 14 14 4.107 130.6 85.6 52.5
GGA 330 eV, 11 11 11 4.111 84.1 99.7 28.3
LDA 500 eV, 14 14 14 3.999 190.1 122.3 71.4
LDA 330 eV, 11 11 11 4.002 155.3 134.6 73.4
Cu
Expt[132] 3.62 169 122 75.3
GGA 600 eV, 24 24 24 3.599 171.9 108.2 78.5
GGA 320 eV, 12 12 12 3.609 181.2 114.7 83.4
LDA 600 eV, 24 24 24 3.512 236.6 171.3 86.2
LDA 320 eV, 12 12 12 3.520 233.6 155.0 93.8
Ni
Expt[132] 3.52 247 153 122
GGA 500 eV, 16 16 16 3.534 249.4 171.6 108.0
GGA 330 eV, 12 12 12 3.539 241.8 165.6 107.4
LDA 500 eV, 16 16 16 3.439 333.3 227.8 135.3
LDA 330 eV, 12 12 12 3.442 314.4 227.0 131.2
CASTEP ultrane setting
Table 2.8: Lattice Constants and Elastic Constants (GPa) for -Sn and -Sn using
LDA-generated pseudopotentials (Sn 00.usp).
a (
A) c (
A) c
11
c
33
c
12
c
13
c
44
c
66
-Sn
Expt [133] 6.48 a 69.3 c
11
29.3 c
12
36.2 c
44
GGA S1 6.407 a 69.7 c
11
29.9 c
12
34.5 c
44
GGA S2 6.383 a 76.5 c
11
33.5 c
12
38.7 c
44
LDA S1 6.406 a 71.1 c
11
32.0 c
12
31.1 c
44
LDA S2 6.392 a 75.1 c
11
36.2 c
12
34.2 c
44
-Sn
Expt [132] 5.82 3.175 73.2 90.6 59.8 39.1 21.9 23.8
GGA S3 5.680 3.094 - - - - - -
GGA S4 5.683 3.083 107.5 105.4 47.0 38.6 28.0 32.7
LDA S3 5.706 3.080 - - - - - -
LDA S4 5.694 3.081 105.7 114.1 50.3 37.3 28.7 31.7
S1 - 280eV, 7 7 7 (ultrane)
S2 - 560 eV, 14 14 14
S3 - 280eV, 9 9 6 (ultrane)
S4 - 560 eV, 18 18 12
These sets of calculations performed shows that the GGA and LDA yields
similar values for the lattice constants that are 3% below the experimental values.
Most of the elastic constants calculated for Sn are 25% - 40% larger than their
corresponding experimental value, while c
13
is found to agree very well and c
12
is
17% - 22% lower. For -Sn, the best agreement with experiment in the GGA is
found with c
44
with a dierence of at most 7%, while the other two constants are
systematically higher than their corresponding experimental values by up to 14%.
2.5.4 Eect of using GGA-generated pseudopotentials
In this section, the eect of the untested GGA-generated pseudopotentials on the
lattice constants and elastic constants of Cu and -Sn are investigated. These pseu-
dopotentials are found in the CASTEP les Cu 00PBE.usp and Sn 00PBE.usp.
The results are shown in Table 2.9. The predictions for Cu are very close to exper-
iment, with the lattice constants almost equal and the elastic constants diering
Table 2.9: GGA calculations of the lattice constants and elastic constants
(GPa) of Cu and -Sn with GGA-generated pseudopotentials (Cu 00PBE.usp and
Sn 00PBE.usp).
a (
A) c
11
c
12
c
44
Cu
Expt [132] 3.62 169.0 122.0 75.3
320 eV, 12 12 12 3.636 181.9 119.2 81.9
480 eV, 18 18 18 3.627 189.6 123.4 82.6
-Sn
Expt [133] 6.48 69.3 29.3 36.2
280 eV, 7 7 7 6.674 51.0 24.1 10.2
420 eV, 10 10 10 6.657 55.4 26.3 27.1
by at most 12 %.
The problem lies with the predictions for Sn. The lattice constant predicted
is 2.7% larger, which is within the accuracy expected for GGA calculations [57].
However, the predicted values of c
11
and c
44
are at least 20%-25% lower than the
experimental values. Hence, it is seen here that this performance is worse than
what was obtained previously with the LDA-generated pseudopotential.
2.5.5 Conclusions
In this section, the lattice constants and elastic constants of Ag, Cu, Ni and Sn have
been calculated. These materials are components in the intermetallic compounds
to be investigated in the next section. It has been found that calculations with the
GGA functional yielded properties in good agreement with experimental results.
The choice of the LDA-generated pseudopotential is also validated.
2.6. PROPERTIES OF THE INTERMETALLIC COMPOUNDS 49
2.6 Properties of the Intermetallic Compounds
2.6.1 Introduction
Results using the methodology described previously are now described for the
intermetallic compounds in this section. Apart from Cu
6
Sn
5
and Cu
3
Sn, Ag
3
Sn
and Ni
3
Sn
4
are also included in the investigation. This is done because including
additional compounds into the calculations expands the amount of results available
to evaluate the accuracy of the calculations. Furthermore, they have similar crystal
structures to the Cu-Sn compounds under investigation.
The lattice constants and elastic constants are calculated by the same methods
as in the previous section. From the elastic constants, the single-crystal elastic
anisotropy can be discussed and bounds on the polycrystalline elastic moduli can
be obtained. The calculations that provide the lattice constants also enable the
enthalpy of formation to be calculated.
2.6.2 Lattice Constants
The results for the lattice constants are shown in Table 2.10. Increasing the number
of k-points and cuto energy produced a change of no more than 0.5%, demon-
strating the convergence of the calculations. With the largest energy cut-o and
k-point settings, the values obtained are in excellent agreement with experiment,
being no more than 2% below the experimental values. The GGA functional typ-
ically produces values that are higher than the LDA [128], and the calculated
values show this trend. The computation timings are listed in Table 2.11. The
most computationally intensive calculations were for Cu
6
Sn
5
, which required 14
CPU-days of calculation time and Ni
3
Sn
4
, which required 7.7 CPU-days.
Table 2.10: Calculated lattice constants
Setting a (
A) b (
A) c (
A) (
)
Cu
6
Sn
5
Expt [86] 11.022 7.282 9.827 98.84
GGA (320 eV, 16 k-points) 10.848 7.130 9.643 99.35
GGA (460 eV, 24 k-points) 10.822 7.125 9.630 99.37
LDA (320 eV, 16 k-points) 10.663 7.041 9.536 98.98
Cu
3
Sn
Expt [117] 5.52 4.77 4.35 -
LDA (320 eV, 5 6 5 k-pts) 5.334 4.207 4.592 -
LDA (450 eV, 6 8 7 k-pts) 5.369 4.191 4.565 -
LDA (550 eV, 7 9 9 k-pts) 5.330 4.195 4.593 -
GGA (320 eV, 5 6 5 k-pts) 5.427 4.286 4.659 -
GGA (450 eV, 6 8 7 k-pts) 5.466 4.272 4.626 -
GGA (550 eV, 7 9 9 k-pts) 5.421 4.278 4.656 -
Ni
3
Sn
4
Expt[119] 12.214 4.060 5.219 105.0
LDA (330 eV, 7 7 5 k-pts) 11.850 3.937 5.100 105.29
LDA (450 eV, 9 9 7 k-pts) 11.844 3.945 5.072 105.13
GGA (330 eV, 7 7 5 k-pts) 11.970 3.982 5.156 105.10
GGA (450 eV, 9 9 7 k-pts) 11.952 3.991 5.131 104.84
Ag
3
Sn
Expt[121] 5.968 4.7802 5.1843 -
LDA (330 eV, 4 4 5 k-pts) 5.795 4.648 5.061 -
LDA (600 eV, 7 7 8 k-pts) 5.818 4.651 5.019 -
GGA (330 eV, 4 4 5 k-pts) 5.922 4.733 5.096 -
GGA (600 eV, 7 7 8 k-pts) 5.947 4.737 5.118 -
Table 2.11: Computational time in CPU-days for the lattice constants. For each
xc-functional, the values are arranged in order of increasing accuracy setting
GGA LDA
Cu
3
Sn 0.6, 1.9, 0.5 0.4, 2.0, 2.7
Cu
6
Sn
5
3.9, 14.0 2.9, -
Ni
3
Sn
4
2.8, 7.7 1.1, 2.8
Ag
3
Sn 0.2, 0.9 0.1, 0.6
Table 2.12: Positional Parameters for Cu
6
Sn
5
Site Expt (Ref [86]) GGA 320 GGA 460
Sn1 x 0.39106 0.39096 0.39154
Sn1 y 0.16250 0.16508 0.16290
Sn1 z 0.52867 0.52955 0.52998
Sn2 x 0.28518 0.28724 0.28669
Sn2 y 0.65499 0.65611 0.65726
Sn2 z 0.35792 0.35322 0.35617
Sn3 y 0.79892 0.80190 0.80141
Cu1 x 0.10096 0.10111 0.10151
Cu1 y 0.47297 0.47871 0.47862
Cu1 z 0.20236 0.20193 0.20257
Cu2 x 0.30620 0.30409 0.30459
Cu2 y 0.50404 0.50793 0.50573
Cu2 z 0.60972 0.60523 0.60671
Cu3 0.0 0.0 0.0
CuA y 0.16020 0.16165 0.16160
x = 0.0, z = 0.25
x, y, z = 0.0
2.6.3 Internal Crystal Parameters
As the geometry optimization calculations involve optimizing with respect to any
internal coordinates present, predictions of the internal degrees of freedom can
be obtained at the same time. The calculated positional parameters for Cu
6
Sn
5
(Table 2.12), Cu
3
Sn ( Table 2.14), Ni
3
Sn
4
(Table 2.13) and Ag
3
Sn (Table 2.15) are
compared with experiment. In all cases, the agreement is excellent. Most of the
values agree up to the second decimal place, indicating that the largest dierence
is less than 1 %.
Having obtained the structural properties of the intermetallic compounds, the elas-
tic constants are described next. Starting from the structure optimized previously
obtained at each accuracy setting, the same energy cuto and k-point mesh are
Table 2.13: Positional Parameters for Ni
3
Sn
4
Site Expt (Ref [119]) GGA 330 GGA 450 LDA 330 LDA 450
Ni1 0.0 0.0 0.0 0.0 0.0
Ni2 x 0.2147 0.21525 0.21513 0.21578 0.21552
Ni2 z 0.3369 0.33222 0.33399 0.33433 0.33478
Sn1 x 0.4286 0.42944 0.42922 0.42814 0.42788
Sn1 z 0.6864 0.68583 0.68582 0.68733 0.68751
Sn2 x 0.1718 0.17155 0.17244 0.17136 0.17138
Sn2 z 0.8123 0.80884 0.81210 0.81045 0.81087
All sites have parameter y = 0.0.
x, y, z = 0.0
Table 2.14: Positional parameters for Cu
3
Sn
Site Expt (Ref [117] ) GGA 320 GGA 450 LDA 320 LDA 450
Sn1 z 0.333 0.33038 0.33124 0.33729 0.33489
Cu1 z 0.667 0.68003 0.67924 0.68720 0.68739
Cu2 x 0.250 0.24992 0.25018 0.24962 0.24976
Cu2 z 0.833 0.83732 0.83708 0.84056 0.83908
Sn1: x = 0.0, y = 0.0.Cu1: x = 0.0, y = 0.5.Cu2: y = 0.0.
Burkhardt states that these are approximate values.
Table 2.15: Positional parameters for Ag
3
Sn
Site Expt (Ref [121] ) GGA 330 GGA 600 LDA 330 LDA 600
Sn1 z 0.1687 0.16931 0.17419 0.17123 0.17156
Ag1 x 0.00 -0.00082 -0.00049 -0.00113 -0.00077
Ag1 z -0.3369 -0.33575 -0.33407 -0.33577 -0.33520
Ag2 z -0.1723 -0.16585 -0.16453 -0.16770 -0.16739
Sn1: x = 0.25, y = 0.25.Ag1: y = 0.25.Ag2: x = 0.25, y = 0.75.
used to calculate the elastic constants. For brevity, only the energy cuto is stated
with the exchange-correlation functional in the tabulated results.
Tabulated in this section are the results calculated at various accuracy settings
for Cu
3
Sn (Table 2.16). As Cu
6
Sn
5
required almost two months of CPU time,
only one set of elastic constants were calculated and listed in matrix form in
Figure 2.10. To demonstrate the general applicability of DFT calculations for
elasticity, calculations were also performed for Ni
3
Sn
4
(Table 2.17) and Ag
3
Sn
(Table 2.18). As an indication of the computational eort required, Table 2.19
shows the computer time in CPU-days for the four intermetallics.
From these values, some general observations can be made.
The elastic constants obtained at the lowest energy cuto, which is the ul-
trane setting provided by the software, are unsatisfactory. For Ag
3
Sn, in
the set obtained from the GGA calculations at 320 eV cuto is a negative
elastic constant. In Cu
3
Sn, a large uncertainty is observed for c
33
obtained
with GGA at 320 ev cuto.
The eigenvalues of the stiness matrices formed by these elastic constants
were checked to see if they fullled the stability criterion described in section
2.3.3. No set with negative eigenvalues were found, indicating that these
crystals are predicted to be mechanically stable.
For the set in Ag
3
Sn with the negative elastic constant in Table 2.18 (GGA
330), the eigenvalues are found to be negative. This suggests that the this
particular DFT calculation has not been conducted with sucient accuracy
settings and should be rejected. It can be seen that no negative elastic
constants were found when the accuracy setting was increased (i.e. GGA
500).
The elastic constants obtained with LDA are larger than that of the GGA.
Table 2.16: Elastic Constants of Cu
3
Sn calculated at dierent accuracy setting.
All values are given in GPa.
c
ij
GGA 320 GGA 450 GGA 550 LDA 320 LDA 450 LDA 550
c
11
192.1 197.4 176.1 271.3 233.7 213.1
c
22
222.2 220.6 208.0 269.3 269.2 261.4
c
33
161.5 193.0 191.4 228.1 219.3 229.2
c
44
77.7 58.7 60.7 92.6 68.5 66.5
c
55
43.1 44.1 44.3 58.4 49.7 52.8
c
66
68.1 58.8 69.7 73.6 67.8 83.8
c
12
60.8 87.4 100.6 95.6 110.3 122.7
c
13
105.9 94.8 110.7 102.4 119.1 135.3
c
23
103.0 98.0 95.5 126.5 126.9 114.2
c
ij
=
_
_
156.4 62.2 69.4 0.0 11.3 0.0
165.2 60.6 0.0 16.8 0.0
155.8 0.0 5.6 0.0
42.3 0.0 7.4
51.9 0.0
48.0
_
_
Figure 2.10: Stiness matrix for Cu
6
Sn
5
calculated with the GGA at 320 eV cuto
and 4 4 3 k-point mesh. All values are given in GPa
Again, this reects the over-binding nature of the LDA mentioned earlier.
Some elastic constants for Cu
6
Sn
5
(c
15
) and Ni
3
Sn
4
(c
15
, c
35
) are found to
be negative. They do not compromise the stability of the crystal, and their
implications are discussed in Appendix E.
For validation and analysis in the subsequent sections, for each material, the
set of elastic constants obtained at the largest energy cuto setting will be used.
2.6.5 Elastic Anisotropy
The elastic constants that have been obtained are now used to plot the single-
crystal Youngs Modulus as a function of crystallographic direction for Cu
3
Sn
(Figure 2.12) and Cu
6
Sn
5
(Figure 2.13) with the elastic constants calculated with
the GGA and with the highest accuracy settings. Spherical coordinates are used
Table 2.17: Elastic Constants of Ni
3
Sn
4
calculated at dierent accuracy setting.
All values are given in GPa.
c
ij
GGA 330 GGA 450 LDA 450
c
11
192.8 192.2 209.6
c
22
180.5 176.2 212.3
c
33
209.9 204.6 229.3
c
44
76.2 76.2 80.6
c
55
73.8 77.0 81.1
c
66
69.5 68.3 71.2
c
12
77.7 77.4 92.9
c
13
70.5 82.9 97.9
c
15
-24.6 -25.7 -26.5
c
23
79.6 80.4 102.3
c
25
11.5 12.7 12.4
c
35
-10.6 -10.9 -8.1
c
46
5.1 8.2 15.5
Table 2.18: Elastic Constants calculated for Ag
3
Sn calculated at dierent accuracy
setting. All values are given in GPa.
c
ij
GGA 330 GGA 500 LDA 500
c
11
122.1 125.3 167.0
c
22
125.6 138.5 189.1
c
33
126.6 134.8 171.3
c
44
19.2 31.2 39.1
c
55
-28.3 27.4 31.8
c
66
29.6 32.0 39.1
c
12
61.7 68.4 98.7
c
13
76.3 76.0 101.8
c
23
77.6 76.3 104.0
Table 2.19: Computational time in CPU-days for the elastic constants. For each
xc-functional, the values are arranged in order of increasing accuracy setting
GGA LDA
Cu
3
Sn 2.8,10.7,22.2 3.8,7.7,28.7
Cu
6
Sn
5
58.9 -
Ni
3
Sn
4
2.2,13.2 42.1
Ag
3
Sn 1.3, 3.2 10.2
No LDA calculation was performed for Cu
6
Sn
5
a
c
b
a
c
b
Orthorhombic Monoclinic

Figure 2.11: Relation of the coordinate axes to the crystal axes
with the value of the Youngs Modulus as the radius. is the angle measured
anticlockwise from the positive x-axis and is the angle from the positive z-
axis. In Cu
3
Sn, these coincide with the [100] (a-axis) and [001] (c-axis) directions
respectively, while for monoclinic Cu
6
Sn
5
only the x-axis coincides with the a-axis.
These relationships are illustrated in Figure 2.11. From the plots it is clear that
the intermetallic compounds are highly anisotropic and the Youngs Modulus show
wide variations according to the crystal orientation. From these plots, the value
of the largest Youngs Modulus and their directions are shown in Table 2.20.
For greater clarity, the variation of the Youngs Modulus in various planes are
examined for Cu
3
Sn ( Figure 2.14 ) and Cu
6
Sn
5
(Figure 2.15).
For Cu
3
Sn,it is seen from Figure 2.14 that the direction of the largest Youngs
Modulus is in the x-y plane at = 55.8
, which corresponds closely to the [210]

direction. This direction has an angle of 57.4
with the positive x-axis and corre-

sponds to a closely-packed plane. For Cu
6
Sn
5
, the direction of the global maximum
(is located at 147.6
from the positive x-axis in the xz-plane. This corresponds to

the direction [20
1].
In the analysis of the elastic anisotropy of several intermetallic compounds,
Nakamura observed that generally, there was a correlation between the atomic
packing and the directions of the largest single-crystal Youngs Modulus [134].
This trend is similarly demonstrated for Cu
6
Sn
5
and Cu
3
Sn. In Figure 2.16(a)
Table 2.20: Maximum single-crystal Youngs Modulus and their directions
Value (GPa) Plane Angle
Cu
3
Sn Max 162.09 xy 55.80
Cu
6
Sn
5
Max 152.61 xz 147.60
All angles are measured anticlockwise from the positive x-axis

the unit cell is shown with one edge re-oriented to the [210] direction. Alternating
planes are highlighted in yellow and the plane itself is shown in Figure 2.16(b). It
is thus seen that atoms in this plane are closely-packed with a pseudo-hexagonal
symmetry, with one Sn atom surrounded by six Cu atoms. The stacking sequence
is ABCDABCD, marked out by the three dierent coloured rectangles in (a) and
indicated in (b).
For Cu
6
Sn
5
, planes in the [20
1] direction are highlighted in the unit cell at

Figure 2.17. The packing of these planes is shown in Figure 2.18. Alternate chains
of Cu atoms and Sn atoms in zig-zag arrangement are seen. The coordination of
each atom is rather complex, unlike the straightforward hexagonal arrangement in
Cu
3
Sn.
Hence, the directions of strongest bonding predicted by the anisotropy equa-
tions for the Youngs Modulus is correlated to the nature of the packing in the
crystal structure. Finally, it should be noted that this is not a universal rule and
a few exceptions come to mind. In Molybdenum, which has a BCC structure,
the Youngs modulus is larger in the [100] direction (which is in a plane with
fourfold coordination) than the [110] direction (which is in a plane with sixfold
coordination).
2.6.6 Bounds on Polycrystalline Elastic Moduli
Using another means of indirect validation, the bounds on the polycrystalline
elastic moduli by means of the methods of Hill [102] and Hashin-Shtrikman [103,
Figure 2.12: Youngs Modulus as a function of crystallographic direction for Cu
3
Sn.
The colours represent the magnitude of the Youngs Modulus while the x, y, z
coordinates represent the crystallographic directions.
Figure 2.13: Youngs Modulus as a function of crystallographic direction for
Cu
6
Sn
5
. The colours represent the magnitude of the Youngs Modulus while the
x, y, z coordinates represent the crystallographic directions.
0 50 100 150 200
90
100
110
120
130
140
150
160
170
xy plane
angle
E
0 50 100 150 200
110
115
120
125
130
135
140
145
150
yz plane
angle
E
0 50 100 150 200
95
100
105
110
115
120
125
xz plane
angle
E
Figure 2.14: Youngs Modulus as a function of crystallographic direction of various
planes for Cu
3
Sn. Angles are measured anti-clockwise from the rst-mentioned axis
in each plane.
0 50 100 150 200
110
115
120
125
xy plane
angle
E
0 50 100 150 200
105
110
115
120
125
yz plane
angle
E
0 50 100 150 200
90
100
110
120
130
140
150
160
xz plane
angle
E
Figure 2.15: Youngs Modulus as a function of crystallographic direction of various
planes for Cu
6
Sn
5
. Angles are measured anti-clockwise from the rst-mentioned
axis in each plane.
Figure 2.16: The plane corresponding to 55.8

x-axis in Cu
3
Sn
Figure 2.17: The plane corresponding to 147.6

x-axis in Cu
6
Sn
5
.
Figure 2.18: A two-dimensional view of the plane marked in Figure 2.17 in Cu
6
Sn
5
.
Table 2.21: Bounds on the Polycrystalline Moduli (GGA)
Bound G (GPa) B (GPa) E (GPa)
Cu
6
Sn
5
VRH Upper 47.43 95.75 122.13 0.287
HS Upper 46.42 95.64 119.87 0.291
HS Lower 46.08 95.60 119.11 0.292
VRH Lower 45.03 95.46 116.74 0.296
Cu
3
Sn
VRH upper 52.86 132.10 139.92 0.324
HS upper 51.73 132.03 137.26 0.327
HS lower 51.32 132.00 136.31 0.328
VRH lower 50.02 131.91 133.23 0.332
Ni
3
Sn
4
VRH upper 66.45 117.15 167.66 0.261
HS upper 64.67 116.60 163.75 0.266
HS lower 63.94 116.40 162.15 0.268
VRH lower 62.00 115.94 157.85 0.273
Ag
3
Sn
VRH upper 29.43 94.45 79.99 0.359
HS upper 29.06 94.39 79.07 0.360
HS lower 28.94 94.37 78.78 0.361
VRH lower 28.47 94.30 77.60 0.363
105, 104] are calculated from the single-crystal elastic constants. From these,
comparison can be made with existing experimental data. The bounds are shown
in Table 2.21 and Table 2.22.
For all of these bounds calculated, it is seen that the HS bounds lie within
the VRH bounds, which is an expected outcome as the HS formulation has been
shown to give tighter bounds [108].
As the LDA elastic constants are larger, the bounds correspondingly obtained
are higher (Table 2.22). The VRH bounds on the Youngs Modulus suggest an
average value of 175.1 GPa for Ni
3
Sn
4
, 100.3 GPa for Ag
3
Sn and 161.9 GPa for
Cu
3
Sn. These are respectively 7.6% , 27.3% and 18.3 % higher than the corre-
sponding GGA values, and are beyond the range of experimental results and thus
leads to the conclusion that the LDA calculations over-predict the elastic values.
As such, discussion will proceed with the GGA results only.
Table 2.22: Bounds on the Polycrystalline Moduli (LDA)
Bound G (GPa) B (GPa) E (GPa)
Cu
3
Sn
VRH upper 61.59 159.41 163.68 0.329
HS upper 60.93 158.97 162.09 0.330
HS lower 60.80 158.89 161.76 0.330
VRH lower 60.04 158.44 159.93 0.332
Ni
3
Sn
4
VRH upper 70.45 137.47 180.51 0.281
HS upper 68.42 136.99 175.96 0.286
HS lower 67.69 136.80 174.32 0.288
VRH lower 65.78 136.25 169.76 0.292
Ag
3
Sn
VRH upper 36.86 126.27 100.77 0.367
HS upper 36.70 126.08 100.36 0.367
HS lower 36.68 126.06 100.30 0.367
VRH lower 36.49 125.85 99.82 0.367
For Ag
3
Sn, the VRH bounds for the Youngs Modulus give about 1.5% on
either side of the average value of 78.8 GPa. This average value lies within the
range of experimental results and agreeing most closely with the value of 81.4 GPa
by Ghosh [42]. The bounds also suggest a Poissons Ratio of 0.36 which is in
excellent agreement with the experimental results.
For Cu
3
Sn, the predicted mean Youngs Modulus of 136.8 GPa also lies within
the range of the experimental results obtained by nanoindentation. The HS bounds
suggest a Poissions ratio of 0.33, which is in good agreement with the range of
experimental results (0.299 - 0.33).
The VRH bounds for the Youngs Modulus of Ni
3
Sn
4
are 3% on either side of the
average value of 162.8 GPa. This is closer to the uncertainty usually encountered
in experiments. The average value is higher than the experimental results and lies
closer to the results obtained by nanoindentation. In the case of the Poissons ratio,
there is less agreement. The bounds suggest a value of 0.27. with the experimental
results (Table 2.1) ranging from 0.27 to 0.33.
The bounds for Cu
6
Sn
5
suggest that the Youngs modulus lies at the higher
2.7. DISCUSSION 64
end of the range of experimental values in Table 2.1. The predicted mean Youngs
modulus is 119 GPa, which is in agreement with most experimental values obtained
by nanoindentation. The VRH bounds are about 6 GPa wide, which is of the same
order of magnitude as the experimental error. The Poissons ratio reported by
Fields et al. [19] is 0.309 0.012. The HS bounds suggest a value of 0.29, which
is also in good agreement with the experimental results.
2.6.7 Summary
In this section, the lattice constants and internal crystal parameters have been
calculated with DFT and found to be in excellent agreement with experiment.
With the elastic constants calculated, two sets of analysis were performed. The
elastic anisotropy was analysed and directions of the largest Youngs Modulus are
found to correspond to low-index planes in the crystal. The bounds on the poly-
crystalline elastic moduli were calculated and found to correspond to the results
obtained with nanoindentation.
2.7 Discussion
2.7.1 Limitations of the DFT calculations performed
In this chapter, DFT calculations have been performed on the crystal structure
reported experimentally. The predicted lattice constants are in good agreement
with experimental results and show the expected trend with the LDA and the
GGA functionals. The internal parameters agree to 1%, thus also validating the
accuracy of DFT calculations.
The fact that these crystal structures are observed experimentally shows that
nature assigns them to be the ground state. We did not attempt to show that DFT
predicts these structures to be the ground state. A rigorous search over all possible
2.7. DISCUSSION 65
crystal structures is not feasible as the possibilities are innite. Recognizing this,
researchers such as Ghosh et al. [135] and Curtarolo et al. [64] have chosen to
conduct a targeted search instead, using the structures of other compounds with
similar stoichiometry as a basis for comparison. While such an analysis seems
feasible for Cu
3
Sn as many other intermetallics of 3:1 stoichiometry have high
symmetry and small unit cells (eg. Cu
3
Al and others listed in Ref [136]), this is
not feasible for Cu
6
Sn
5
as a search of the Atlas of Intermetallic Compounds turns
up many structures of low symmetry and extremely large unit cells [118].
The accuracy of the elastic constants can be further assured with a convergence
study. However, this requires larger k-point meshes and energy-cutos over and
above what has already been used. Performing such a study with the intermetallics
themselves will be expensive and will require more resources than what is currently
available.
Nevertheless, eort has been made to obtain elastic constants at accuracy set-
tings that are more stringent than suggested by the software. Hence, it is believed
that these calculated values of the elastic constants are reasonable for two reasons.
In the analysis of the elastic anisotropy, the predicted directions of largest
Youngs Modulus correspond to well-dened low-index planes, a result that
is seen in other intermetallics as well [134].
Using these elastic constants, the polycrystalline bounds actually lie on the
upper range of the experimental data. This point will be discussed further
in the next section.
2.7.2 Comparison of polycrystalline bounds with nanoin-
dentation data
The VRH bounds and HS bounds, as well as the rst-principles calculations, do
not account for the presence of defects in the material. Thus, it is expected that
2.7. DISCUSSION 66
the values of the elastic moduli calculated to lie on the upper end of the range of
experimental values in Table 1, i.e. those from nanoindentation.
On the face of it, such a comparison seems reasonable as nanoindentation tests
involve only a localized region of the material and are hence less suscepible to the
adverse eects of oxidation and porosity than property measurements from bulk
materials testing.
An alternative explanation for the higher values of Youngs Modulus obtained
by nanoindentation was given by Ghosh [42] and Deng et al. [41]. It is believed
that the relatively higher results from nanoindentation could instead reect the
Youngs Modulus of a preferred crystal orientation instead of an average over
many orientations. If this is true, then any comparison between the bounds and
the nanoindentation results will be inappropriate.
However, the experimental evidence available so far only supports this assertion
weakly. This is due to two reasons
1. The experimental conditions in studies of surface texture and nanoindenta-
tion studies dier in the length of the reow time.
2. Studies of nanoindentation do not report the surface texture, nor the grain
size.
In studies of the solder joint with reow times up to 330 s, Prakash and Sritha-
ran reported a surface texture of < 102 > and < 101 > in Cu
6
Sn
5
[38]. Also, Suh
et al. [39] reports of preferred crystallographic orientations between Cu
6
Sn
5
and
Cu in the solder joint in the rst 4 minutes of solder reow that minimizes the
mist between the two crystal structures. However, with reow time extending to
8 mins, Cu
3
Sn is formed between Cu
6
Sn
5
and Cu and the phenomenon of preferred
orientations disappears.
In contrast to these two studies, most studies using nanoindentation cited in
this study report long annealing times in order for the intermetallic layer to grow
2.7. DISCUSSION 67
to a thickness sucient for indentation experiments to be conducted accurately.
Deng et al. [41] and Chromik et al. [44] required two weeks of annealing time
using diusion couples, while Tsai et al., using Sn deposited on Cu lms, required
a total of 140 hours [51]. Only Jang et al. used a reaction time of 2 hours [47].
In these studies, no studies of the crystal orientation were made. There are also
no reports of the mean grain size in these studies, thus nothing is known of the
likelihood of the nanoindentation being done within a single grain (ie. done on a
single-crystal).
Using the data in Figure 2.13, a cumulative distribution function is constructed
for the probability that the Youngs Modulus of any orientation E
will not exceed

a particular value E, ie P(E
E), and shown in Figure 2.19. The Youngs

Modulus obtained with nanoindentation for Cu
6
Sn
5
range from 110 GPa to 125
GPa (Table 2.1). Using this range, it is found that P(110 E 125) = 0.543
ie. 54.3 % of the orientations have a value of the single-crystal Youngs Modulus
between 110 GPa and 125 GPa. This shows that, even if the nanoindentation
results do indicate the Youngs Modulus of any preferred texture, the scatter in
the experimental results point to a large proportion of orientations having the
range of Youngs Modulus.
To summarize, in the study by Suh et al., using reow times of not more
than 4 mins, suggests that any surface texture in Cu
6
Sn
5
disappears with an
increase in reow time, due to the subsequent formation of Cu
3
Sn. Studies using
nanoindentation required a much longer reow time to obtain an intermetallic
layer of sucient thickness. Analysis of the single-crystal Youngs modulus of
the various crystallographic orientations also shows that a large proportion of the
orientations have values that lie within the range of experimental results. As such,
there is no reason so far to believe that a comparison between the polycrystalline
bounds and the nanoindentation data is inappropriate.
2.7. DISCUSSION 68
90 100 110 120 130 140 150 160
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Cumulative Distribution Function for Cu
6
Sn
5
E (GPa)
P
(

E
*

E
)
Figure 2.19: Cu
6
Sn
5
: Cumulative Distribution Function for the single-crystal
Youngs Modulus of Cu
6
Sn
5
90 100 110 120 130 140 150 160 170
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Cumulative Distribution Function for Cu
3
Sn
E (GPa)
P
(

E
*

E
)
Figure 2.20: Cu
3
Sn: Cumulative Distribution Function for the single-crystal
Youngs Modulus
2.8. CONCLUSIONS 69
2.8 Conclusions
Calculations with Density Functional Theory, using the planewave-pseudopotential
method, have been performed to obtain the lattice constants and elastic constants
of four intermetallic compounds, Cu
6
Sn
5
, Cu
3
Sn,Ag
3
Sn and Ni
3
Sn
4
. The bounds
on the polycrystalline elastic moduli obtained using the single-crystal elastic con-
stants calculated lie on the upper end of the range of experimental results, co-
inciding with those obtained with nanoindentation. This indirectly validates the
elastic constants that were calculated for Cu
6
Sn
5
and Cu
3
Sn.
The main contributions of this chapter are as follows.
1. Predictions have been made for the hitherto unknown single-crystal elastic
constants of Cu
6
Sn
5
and Cu
3
Sn.
2. Bounds on the polycrystalline elastic moduli for Cu
6
Sn
5
and Cu
3
Sn have
been obtained using single-crystal elastic moduli calculated with DFT. It
was shown that these bounds coincided with the experimental data obtained
by nanoindentation.
3. The calculated lattice constants were found to be in excellent agreement with
published experimental data, with the GGA functional producing results
that were no more than 2%.
4. From the elastic anisotropy data calculated , it was shown that the directions
of highest single-crystal youngs modulus coincided with planes of closely-
packed atoms.
5. For these intermetallics, the GGA functional produces results closer to ex-
perimental data than the LDA functional.
Parts of this chapter were published in Refs. [137, 138].
70
Chapter 3
Molecular Dynamics Potential
For Cu-Sn
3.1 Introduction
3.1.1 Motivation
Among all the atomistic calculation methods, ab-initio calculations with DFT
stand as one of the most computationally intensive. With several reasonable ap-
proximations to Schr odingers Equation, it can predict the fundamental material
properties of materials quite accurately, as seen from the previous chapter, ob-
taining values of the same order of magnitude as that obtained in experiments.
However, the number of atoms that can be involved in the calculations is very
small, in the order of tens of atoms, and this limits its usefulness signicantly.
Many atomic-scale problems require many more atoms than that currently
achievable by DFT. One example is the fracture of materials. Although it is of-
ten seen to occur on the macroscopic level, it is fundamentally an atomic-level
phenomenon as it involves the breaking of bonds between atoms. Simulating the
propagation of a crack requires a system size of at least thousands of atoms, as
demonstrated in studies like those by Becquart et al. [139]. This would be in-
tractable with ab-initio methods such as Car-Parrinello Molecular Dynamics [140]
unless the most sophisticated computing facilities are employed.
As such, the Molecular Dynamics (MD) method is the method of choice. The
primary input of MD simulations is the interatomic potential, which is the en-
ergy function that describes the behaviour of a system of atoms. The simplest
example is the potential energy function of a linear spring, which has a quadratic
form. Atoms are treated as particles with mass. With this interatomic potential,
Newtons second law is applied to determine how the atoms should move. By
assembling a system of atoms and studying their motion, the system-level phe-
nomena that arises can be studied.
The vast majority of MD studies have been with materials that are techno-
logically important and have crystal structures of high symmetry, such as copper
and silicon. However, intermetallic compounds are slowly gaining technological
importance. The role of Cu
6
Sn
5
in the solder joint has already been described
in Chapter 1, and several other uses of the intermetallic compounds found in the
solder joint have been described in Chapter 2.
The problem of the single-crystal properties of Cu-Sn intermetallic compounds
has been tackled with DFT in the previous chapter. In order to go one step further
and gain an understanding of the atomic-scale phenomena in these intermetallic
compounds, an interatomic potential for the Cu-Sn system that can be used in
MD studies has to be developed. The work in this chapter focuses on the simpler
NiAs structure for Cu
6
Sn
5
reported by Gangulee et al [109].
3.1.2 Molecular Dynamics Method
The use of Newtons Second Law to determine the motion of atoms is an ap-
proximation, given that the interactions between atoms is fully described by the
Table 3.1: Physical constants and values used in calculating the Thermal de Broglie
wavelength of Cu
Constant Symbol Value
Boltzmann Constant k 1.380650410
23
J K
1
Planck Constant h 6.62606896 10
34
J s
1
Atomic mass unit amu 1.660538782 10
27
kg
Mass of a copper atom m 63.546 amu
Schr odingers Equation. The justication for this approximation comes from a
result from the statistical mechanics for ideal gases. The de Broglie hypothesis
states that all matter has an associated wavelength, and for a single particle it is
inversely proportional to the momentum. The analogous concept for an ideal gas
is the thermal de Broglie wavelength which is dened for various temperatures
T as
=
h
2mkT
(3.1)
It has been shown that the quantum nature of the particles becomes unimpor-
tant when is much less than the mean separation between particles, i.e. classical
newtonian mechanics takes over [141]. The value of for a Cu atom can be cal-
culated at 300 K. Using values of the physical constants from Reference [142] and
summarized in Table 3.1, 0.126
A. At 300 K, this comes to approximately 5%

of the nearest-neighbour distance in the FCC Cu crystal of 2.56

A. Therefore, it
is shown that it is valid to use Newtons Second Law as an approximation.
Thus, using Newtons Second Law, the essentials of the Molecular Dynamics
algorithm can be illustrated with a spring-mass system. Consider a linear spring
connecting two particles i and j at a separation of r
ij
. The potential energy and
the force between the particles f is given by
(r
ij
) =
1
2
k(r
ij
r
0
)
2
and f(r
ij
) = k(r
ij
r
0
) (3.2)
where r
0
is the equilibrium distance.
If N particles are connected with springs to each other in three-dimensional
space, then the total potential energy E of the N particles is
E =
j>i
(r
ij
) (3.3)
The force on each pair of interacting particles i, j is given by the derivative of
(r
ij
) with respect to the individual components of the distance vector
f
ij
= (r
ij
) (3.4)
In order to discover the resulting trajectories, Newtons second law must be
applied for the total force on each atom
d
2
r
i
dt
2
=
f
i
m
i
(3.5)
where r
i
is the position of atom i.
Applying the central-dierence approximation to the second derivative results
in the Verlet algorithm [143]
r
i
(t + t) = 2r
i
(t) r
i
(t t) + t
2
f
i
m
i
(3.6)
where r
i
(t) is the position of atom i at time t, and t is a small time increment.
This suggests that, with a suitable choice of the potential energy function (r
ij
),
it is possible to study the trajectories of a collection of atoms once the forces are
known using Newtons Second Law.
3.1.3 Types of Interatomic Potentials
In the previous section, it is seen that the sole input that determines the motions of
the atoms is the potential energy function, which is usually termed the interatomic
potential. Thus, examining the various choices available and understanding their
implications for the resulting behaviour of atoms is important. In this section,
some examples are illustrated.
Pair potentials
The simplest functional form for the energy of a system of atoms is a sum of
pairwise interactions, often termed as a pair potential, which is the expression
in Equation (3.3). The function (r
ij
) can take many forms, such as the Morse
function [144]. Parameters for the Morse function have been proposed to model
FCC metals [145, 146, 147].
Morse function: (r) = D[exp(2(r r
o
)) 2 exp((r r
o
))] (3.7)
However, pair potentials have one fundamental deciency in the case of metals.
It can be shown that the elastic constants with a pair potential is constrained to
c
12
= c
44
. However, most metals do not show this behaviour in nature [132].
Embedded Atom method
The Embedded Atom Method [148] was developed to overcome these disadvantages
by adding a many-body term to the total energy of each atom. The energy of a
system is then expressed as follows
E =
j>i
(r
ij
) +
i
F
i
( ) (3.8)
F
i
( ) is usually termed the embedding function. The argument to the em-
bedding function is dened as follows
=
j
(r
ij
) (3.9)
Initially, is thought of as an electron density, but is later treated by later
authors as another tting function.
The EAM is applicable only to materials with c
12
> c
44
[149]. In order to
model materials with c
12
< c
44
, including Cu
6
Sn
5
, many modications of the
EAM have been proposed. One important modication was proposed by Baskes
in 1992, termed the Modied Embedded Atom Method (MEAM) [150]. It involves
modifying the expression of the electron density in Equation (3.9). The MEAM
will form part of the subject matter of this chapter and details will be given in
later sections.
3.1.4 Challenges with developing an interatomic potential
for Cu-Sn
Having described the types of mathematical models used, the progress in the
scientic community regarding the development of interatomic potentials will be
briey described in order to set the context for the objectives of this study.
Developing an interatomic potential is essentially empirical and a highly arbi-
trary process. A set of mathematical functions with unknown parameters is chosen
within a functional form (e.g. the EAM), and a set of material properties is cho-
sen as the target properties to be reproduced by the potential. An optimization
process is then carried out to nd the parameters that results in the functions
reproducing the target properties.
The vast majority of interatomic potential development work has been for tech-
nologically important materials like silicon and copper, both of which have cubic
crystal structures. Because of the empirical nature of the potential-tting process,
deciencies in existing potentials continued to be reported and new potentials need
to be continually developed. For example, the Stillinger-Weber potential for silicon
[151] was found to result in ductile fracture of silicon [152], and this led to further
attempts to develop potentials that showed the correct brittle behaviour.
A cubic crystal structure has several features that simplify the process of de-
veloping an interatomic potential. Firstly, it has only one lattice constant, and
thus the energy-volume relationship depends only on one degree of freedom. It is
thus much easier to ensure that the potential reproduces the ground-state lattice
constant. Secondly, most of the cubic crystal structures have no internal degrees
of freedom. Hence, zero force on each atom is guaranteed and there is no need
to explicitly ensure that the atoms in the basis are in equilibrium. Even so, this
is no guarantee that the resulting potential predicts the cubic structure to be the
ground state. For example, Heino et al. showed that a potential developed for
FCC copper had a negative stacking fault energy [153]. In other words, the poten-
tial predicts that hexagonal close-packed structure is more stable than the FCC
structure, which is of course erroneous. This is one example which shows that
the process of developing an interatomic potential is never perfect and can lead to
unintended results.
Interatomic potentials have also been developed for many alloy systems, and
one of the systems that has attracted a lot of attention over the past twenty years
is the Ni-Al system. Voter-Chen demonstrated that by taking dierent Embedded
Atom Method (EAM) potentials for Ni and Al and combining them with a rule of
mixtures will not result in a workable potential for Ni-Al alloys [154]. Subsequently,
work has continued on developing new potentials for NiAl, such as Mishins eort
in 2002 [155].
Ideally, an interatomic potential for an A-B alloy system should fulll the fol-
lowing requirements. It should reproduce as many of the equilibrium compounds
that occur between 100% A and 100 % B. For each equilibrium compound, the
crystal structure, enthalpy of formation and elasticity properties should be repro-
duced. Other crystal structures with the same stoichiometry should have a higher
energy.
For the Cu-Sn system, there are many intermetallic compounds, many of
which have complicated crystal structures of low symmetry such as the monoclinic
Cu
5
Sn
4
[156] and the orthorhombic Cu
3
Sn [117, 116]. It would be an immense
task to meet all the ideal requirements stated above. Therefore, in this study, the
potential tting process is focused on one compound, Cu
6
Sn
5
.
The study described in this chapter aims to achieve the following objectives.
The monoclinic structure of Cu
6
Sn
5
will make the tting process extremely compli-
cated as there are 13 elastic constants to reproduce. Therefore, a simpler substitute
will be the hexagonal NiAs structure of Cu
6
Sn
5
reported by Gangulee et al. [109],
which has ve elastic constants. The existing Cu-Sn MEAM potential used in a
study of the deposition of Cu atoms on Sn surfaces [157] cannot reproduce the
NiAs structure accurately, which will be described later in this chapter. Currently,
a search of the scientic literature has not found any MD study that involves com-
pounds with the NiAs structure. Therefore, this study will examine if it is feasible
to develop an interatomic potential that reproduces the experimental lattice con-
stants and the elastic properties of the structure of B8 CuSn. The existing MEAM
potential used by Aguilar et al. [157] will be improved upon. This will be done by
nding new parameters in the MEAM functions that reproduce the properties of
CuSn using an optimization procedure.
Apart from CuSn, there are many intermetallic compounds that have the NiAs
crystal structure as their basis [158, 136]. Many of these intermetallic compounds
show promising uses in technology. One example is MgTe, a semiconductor having
potential applications in optics and high-temperature electronics [159]. The vast
majority of MD simulations have been with technologically important materials
such as copper and silicon. Showing that it is feasible to conduct MD simulations
with the NiAs structure will greatly expand the possibilities of MD simulations,
and broaden our knowledge of these materials.
3.1.5 Overview
The rest of this chapter proceeds as follows. Section 3.2 surveys the existing
scientic literature on the development of interatomic potentials. Methods for
obtaining interatomic potentials for one atomic species are described. Follow-
ing which, complications arising from including more atomic species and complex
crystal structures are reviewed.
In section 3.3, the theoretical aspects are introduced. Basic denitions and
notation are laid out rst. Pertinent aspects of the MEAM are explained and the
MD software written is described. The section concludes with a description of the
various optimization techniques used in this chapter.
Following which, section 3.4 describes the inputs needed to obtain an inter-
atomic potential for Cu-Sn interactions. Results from calculations done to ll in
gaps in experimental data are reported, and choices that need to be made for the
MEAM are described.
The parameters obtained for the interatomic potential, and the validation tests
conducted with it are described in section 3.5. These results are discussed in section
3.6 and summarized in section 3.7.
3.2.1 Introduction
In this literature review, various aspects that concern this work on MD are re-
viewed. The interatomic potential is integral to MD simulations, and an under-
standing of the development of interatomic potentials is important for two reasons.
Firstly, it is to learn what the existing methods are so that a choice could be made.
Secondly, it is necessary to understand the limitations that could occur with the
use of MD potentials, so that a realistic expectation of what could be achieved can
be gained.
Nevertheless, because of its computational eciency relative to ab-intio calcu-
lations, the MD method allows simulation of realistic experimental situations such
as nanowires and nanoindentation and thus facilitates a deeper understanding of
the underlying atomic processes that are taking place.
3.2.2 Interatomic Potentials
The interatomic potential is the sole input that determines the atomic interactions
in the framework of the Molecular Dynamics method. A functional form for the
energy expression is chosen and then the mathematical forms of the individual
component functions are decided upon. Subsequently, to conduct MD simulations,
parameters available in the literature can be chosen or new parameters can be
obtained by tting the functions to a database of properties that the potential is
expected to reproduce.
The rst interatomic potentials for metals appeared in 1959 with parameters
developed for the Morse potential [145]. The pairwise potentials have a funda-
mental deciency in which they constrain the crystal to have elastic constants
c
12
= c
44
. However, an examination of a list of the elastic properties of metals
reveals that such a condition is hardly present [132].
To solve this fundamental aw, the Embedded Atom Method (EAM) by Daw
and Baskes [160, 148] has an additional term to the total energy equation, which
results in c
12
> c
44
[149]. This is satised for many transition metals. For materials
with c
12
< c
44
, many methods have since been proposed to accomodate them, such
as the Modied Embedded Atom Method [150]. Details of these methods will be
discussed in the next section.
An important quality of interatomic potentials is its transferability, which is
the extent of its applicability to diverse situations. Even when existing potentials
for a material exist, a new potential is developed because the potentials show some
deciency in describing some situation that needs to be simulated. For this reason,
technologically important materials such as copper or silicon have many dierent
potentials because of the diverse situations these materials are used in.
In early studies with the EAM, the choices for the mathematical functions were
guided by physical considerations. For example, Voter and Chen chose the Morse
potential for the pairwise function, and an expression for the electron density
that was similar to that found for the orbitals in hydrogen [149]. Most early
development work on interatomic potentials focussed on tting a limited number
of experimental data only with well-dened analytical functions, such as those by
Oh and Johnson [161, 162].
In order to improve the transferability of potentials, improvements continued to
be made to the potential development process, primarily by expanding the number
of material properties to be reproduced. First-principles data would also be used
to ll gaps in the database where no experimental data is available. Consequently,
physical considerations were abandoned and mathematical functions were chosen
based on the exibility that they aorded. One example is the study for Cu
by Mishin et al.. The tting databases was augmented with rst-principles data
and analytical functions with larger sets of parameters were used[73]. These data
consist mainly of energies of various structures and atomic congurations, such as
the surface energy.
Another approach to improve the transferability of interatomic potentials is
known as the force-matching method, developed by Ercolessi and Adams [163,
164, 165]. This approach involves conducting rst-principles molecular dynamics
calculations with various congurations of atoms (e.g. clusters, unit cells). The
data from these simulations are included in the tting database. The tting would
then be done by optimizing the potential functions to reproduce the interatomic
forces in these congurations. Due to the large amount of data involved, cubic
splines with many knots would be used.
A result of all the various methods to develop interatomic potentials in the
quest for better transferability is that for certain technologically-important mate-
rials, such as copper and silicon, a large number of potentials are available in the
literature.
The use of a large database is no guarantee for a wide-ranging transferability.
One example in the literature is the Aluminium potential developed by Ercolessi
and Adams (EA potential) [163] using the force-matching method. Liu et al.
found that the EA potential was decient in predicting the stacking fault energy
[165] and used the force-matching method to adjust the potential. Boyer et al.
compared two potentials for aluminium in predicting the shear stress-displacement
response [166]. The rst potential was developed by Mishin et al [167] using a
database of structural energies, and the other, the EA potential. It was found
that the Mishin potential performed better by giving a prediction of the ideal
shear stress closer to that obtained by Density Functional Theory [166].
At the same time, there is a suggestion to move towards a simpler description
of materials. Baskes proposed a Lennard-Jones EAM model with only two param-
eters [168]. With only two parameters, an understanding of their eect and the
eect of the many-body term on various quantities commonly calculated with MD
can be gained with exceptional clarity [169].
3.2.3 MD simulation of materials with complex structures
The vast majority of MD potentials have been developed for cubic materials due
to the relative simplicity. Recently, there have been some attempts to apply the
MEAM to materials with more complex crystal structures of lower symmetry.
There are several challenges involving this.
Firstly, there is a need to ensure that structures of higher symmetry have
higher energy than the structure found in nature. For example, Si in the diamond
cubic (DC) structure has four rst-neighbours. However, the pairwise potential
cannot reproduce this structure because at the equilibrium distance, the pairwise
potential lowers the energy of the system as more bonds are added. This means
that, under a pairwise potential, the face-centered cubic structure with twelve
rst-neighbours will always have a lower energy than a DC structure. As such,
selecting an appropriate energy functional is crucial.
Secondly, structures of lower symmetry have more properties that need to be
reproduced by the interatomic potential. For example, an orthorhombic crystal
has nine independent elastic constants. There may also be internal degrees of
freedom that must be reproduced. Experimental data may also be sparse and
gaps must be lled by tedious ab-initio calculations. Phase transformations at
various temperatures may exist and ensuring that the potential predicts them can
be tricky.
There have been a few studies in the literature that try to overcome these
challenges. Baskes et al. developed a MEAM potential for Gallium [170], which
has an orthorhombic structure in nature. However, Baskes used a hypothetical
FCC structure for Ga in the tting database, obtaining its properties from ab-
initio calculations. Validation of the resulting potential was then carried out by
predicting the lattice constants and elastic constants of orthorhombic Gallium.
Even though -Sn also has a DC structure like Si, there seems to be far fewer
interatomic potentials developed for Sn. Presumably this is due to the challenge
of reproducing the phase transformation of -Sn to tetragonal -Sn. A potential
with two-body and three-body terms for -Sn was developed by Liu et al. [171].
Ravelo et al. developed a MEAM potential for Sn that predicted with reasonable
accuracy the elastic constants of -Sn (DC) and -Sn (tetragonal, with two lattice
constants) [172]. The phase transition temperature was predicted to be
283 K, in excellent agreement with the experimental value of 286.3K.
3.2.4 MD simulation of materials with two or more atomic
species
Developing a potential for two atomic species has far more challenges than de-
veloping for one. Apart from the same challenges that are found in developing
potentials for low-symmetry structures, there are some additional challenges due
to the existence of compounds formed by both species.
Firstly, a decision has to be made on how the interaction between the two
atomic species is obtained. Very often a larger database has to be involved ie. for
species A and B, and compound A-B. More parameters are needed to accomodate
this larger database. Some authors have also attempted to develop potentials
that are transferable between dierent compounds of AB. This is not a trivial
task, as many intermetallic compounds have compounds with complicated crystal
structures.
The development of the EAM enabled potentials for alloys to be developed.
The Ni-Al system has attracted a great deal of interest from the early days of the
EAM. Foiles et al. showed that it was feasible to use the EAM to describe the
properties of Ni
3
Al [173]. Because of the empirical nature of the tting process,
there is no guarantee that a potential tted to a particular alloy structure will
predict that structure to be the ground state. For example, the energies of defects
could turn out to be negative, which suggests that there is another structure that
is the ground state. This was illustrated by Voter and Chens attempt to develop
an EAM potential for Ni-Al [154]. They found that it was necessary to include
the energies of defects (e.g. anti phase boundary) in the tting database so that
the potential would not predict these energies to be negative.
Potentials have been developed to try to predict the properties of other inter-
metallic compounds and binary alloys. With the MEAM, Baskes et al. developed
a Mo-Si potential and used it to predict the elastic properties of the three stable
intermetallic compounds MoSi
2
, Mo
3
Si and Mo
5
Si
3
[174]. Only the bulk modu-
lus of MoSi
2
and the enthalpy of formation of the three compounds were used in
tting the Mo-Si interactions. Baskes et al. reported that while the agreement
with experiment for the elastic constants of Mo
5
Si
3
was remarkable, predictions
for MoSi
2
included a value of c
44
that was 10% that of the experimental value.
The MEAM study of Ni-Zr alloys by Cherne et al. employed a similar method-
ology in obtaining the Ni-Zr interactions and predicted elastic constants of NiZr
2
that were in good agreement with experiment results[175]. A discrepancy of more
than 90% in the value of c
66
was found, with Cherne et al. suggesting that the
experimental value of 5.62 GPa could possibly be in error. However, it is unclear
as to why no ab-initio calculations were performed to resolve this matter. Excel-
lent predictions for the cohesive energy and enthalpy of formation of various Ni-Zr
phases were obtained, with only minor disagreements for two phases.
MD simulations with the MEAM have also been applied to problems in the
electronics industry. Hai Dong et al. used the Ag-Sn MEAM potential to studying
the interdiusion behaviour of Ag and Sn nanoparticles [176]. Using one sphere
of Ag atoms and one sphere of Sn atoms, they studied how the two nanoparticles
diused into each other at temperatures of 500 K and 800 K. The study con-
cluded that a mixture of Ag and Sn nanoparticles could be feasible in soldering
applications.
Heino was motivated by the possibility of Ta as a diusion barrier in electronics
component to study the diusion properties of the Ta-Si interface with the MEAM
potential [177], showing that the energy of the interface is negatively correlated
to the width of the interface. Motivated by contradictory results in experimen-
tal studies, Kuo et al. studied the nature of the Au-Si interface [178] with MD
simulations and provided a prediction of its nature.
The MEAM has also been used in surface science studies. Aguilar et al studied
the deposition of Sn on Cu surfaces and found that the resulting 2D alloys that
were predicted to form on the surface were consistent with experiments [157]. In
an earlier study by the same authors, the MEAM was found to predict quite well
the diusion activation energies for Cu in Sn and vice versa [179].
There have been even fewer attempts to develop interatomic potentials for
three atomic species. van Beurden et al. developed a MEAM potential for CO-Pt
(carbon monoxide-platinum) interactions by treating the CO molecule as a single
species [180, 181]. The potential was used to study CO-Pt surface reconstruction.
Kuo et al. extended the Au-Si potential developed earlier to a ternary Au-Si-O
system [182]. It involved tting to the structure of quartz which is hexagonal and
has four internal coordinates, a highly challenging process as there is no guarantee
that the forces on each atom will be zero. Consequently, the elastic constants were
not involved in the potential process, presumably to reduce the complexity.
3.2.5 Conclusion
In studies to obtain MD interatomic potentials, advances in both the potential
tting process and the theoretical underpinnings have been seen. The scope of ma-
terial properties to be reproduced have expanded to include interatomic forces cal-
culated from DFT. From the simple pairwise potential for metals in 1959, progress
has seen the development of the MEAM as a functional form to model the atomic
interactions in metals.
Despite such progress, the process to obtain an interatomic potential still in-
volves making many arbitrary choices, both in the mathematical functions and in
the material properties to be reproduced. The empirical nature of this necessarily
leads to a reduced accuracy compared to calculations with the computationally
3.3. THEORY 86
intensive Density Functional Theory. Baskes et al. speaks of a Mortons Fork,
which is a choice between two equally unpleasant alternatives [183]. Compared
to established computational methods at the continuum level such as the Finite
Element Method, the MD method and computational materials science is still in
its infancy. This is despite the scope of MD having grown in tandem with the
advance in computing technology and advances in the theoretical knowledge[184].
This has not deterred researchers from making the best out of the MD method,
as can be seen from the many studies described in this literature review. More
alloy systems are being studied by MD. Kuo et al. pointed out that while their
studies were limited by the relative inaccuracy of the MD method, their results
nevertheless could provide interesting talking points and go some way towards
taking a computational position on the detailed nature of the Au/Si interface
[178]. As the understanding of MD continues to improve, it will slowly improve
on its promise to enable insights to be gained in understanding the behaviour of
materials.
3.3 Theory
3.3.1 Notation and Denitions
Mathematical Notation
The mathematical notations in the MD literature have not been universally ac-
cepted yet. Often, the same quantity is described with various symbols. As such,
it is necessary to set out the denitions of the symbols used in this chapter to
avoid ambiguity.
The position vector between atoms i and j is
3.3. THEORY 87
r
ij
=
_
_
r
1
r
2
r
3
_
_
(3.10)
and the interatomic distance is
r
ij
= | r
ij
| =
_
r
2
1
+r
2
2
+r
2
3
(3.11)
The unit vector is indicated as
x
ij
=
_
_
x
1
x
2
x
3
_
_
=
1
r
ij
_
_
r
1
r
2
r
3
_
_
(3.12)
hence
x
1
=
r
1
r
ij
(3.13)
In this work, the summation notation is somewhat simplied. The limits of
summation are omitted. The summation index i will be used for the atom of
interest, while j will be for its neighbours. It is then understood that the limits
of summation is the total number of atoms and the total number of neighbours
respectively. When , or is used as a summation index, it represents the three
coordinate directions. They are summarized in Table 3.2.
Describing interactions in the crystal
First neighbours or nearest neighbours of an atom refers to the surrounding
atoms that are closest to an atom under consideration. This idea can then be
extended to second neighbours, third neighbours, and so on. This is illus-
trated in Figure 3.1 for the FCC structure, projected in the (100) plane. The rst
3.3. THEORY 88
Table 3.2: The dierent summation symbols in use.
Symbol explanation
i
summation over all the atoms in the system
j
summation over all the neighbours of atom i
j>i
summation over all the neighbours of atom i whose in-
dices are greater than itself
summation over the three coordinate directions.

Figure 3.1: FCC lattice in 2D projection. First neighbour and second neigh-
bour shells of shaded atom are indicated.
3.3. THEORY 89
neighbours and second neighbours are indicated by the circles of increasing radii
respectively.
For the cubic crystal structures introduced in most foundational materials sci-
ence textbooks, their usual descriptions will be used, namely face-centred cubic
(FCC), Body-centred Cubic (BCC) and Diamond-cubic (DC). When there is no
risk of ambiguity, the general description of the seven crystal structure types will
be used, namely tetragonal, trigonal, hexagonal, cubic, triclinic, monoclinic and
orthorhombic. Finally, to identify a specic crystal structure, the Strukturbericht
system of notation will be used.
Denition of physical quantities
One denition of the cohesive energy, E
coh
is given by Marder as the energy
needed in order to rip a sample apart into a gas of widely separated atoms [89].
It can also be thought of as the dierence in energy between a free atom in a
vacuum and an atom in a crystal. The cohesive energy is useful when considered
on a per-atom basis. In MD simulations, the free atom is dened to be at an
energy level of 0 eV. A typical value of the cohesive energy for FCC Cu is -3.62
eV/atom [150]. This means that each atom in the FCC crystal needs 3.62 eV to
become a free atom, and a free atom loses 3.62 eV when it goes into the crystalline
state.
3.3.2 Modied Embedded Atom Method
The EAM has enabled many successful MD simulations of metallic materials to
be carried out. However, materials that have c
12
< c
44
cannot be described by the
EAM. Such materials include the Group IV elements such as Carbon and Silicon.
Earlier potentials for silicon, such as the Stillinger-Weber potential [151], typically
included a three-body term that depended on the angles between atoms so that
3.3. THEORY 90
the c
12
< c
44
can be reproduced. There have been many modications to the EAM
functional form proposed to include angular eects. In 1992, Baskes proposed a
modication which he termed as the Modied Embedded Atom Method (MEAM)
[150].
Screening
The feature that is novel to the MEAM is the screening of atoms. In short, it
involves the calculation of a parameter S
ij
for each interacting pair of atoms, and
this parameter is multiplied to all pairwise functions in the interatomic potential.
The screening procedure begins by considering each of the atoms k that lies
between the atom pair of interest i and j. For each atom k,a parameter C is
calculated. C is derived from the equation of the ellipse i j k whose major
axis has a length of r
ij
, illustrated in Equation (3.14). x and y are the distances
of atom k from the centre of the ellipse.
x
2
+
1
C
y
2
=
_
1
2
r
ij
_
2
(3.14)
When rearranged to involve only the interatomic distances, the expression of
the C parameter is obtained (Equation (3.15a)).
C =
2(X
ik
+X
jk
) (X
ik
X
jk
)
2
1
1 (X
ik
X
jk
)
2
(3.15a)
where X
ik
=
_
r
ik
r
ij
_
2
(3.15b)
and X
jk
=
_
r
jk
r
ij
_
2
(3.15c)
Arbitrary limits are placed on the value of C, where the upper limit is called
C
max
and the lower limit is called C
min
. Atom k is said to fully screen i-j if
C < C
min
, and said to partially screen i-j if C
min
< C < C
max
. There is no
3.3. THEORY 91
screening if C > C
max
. This is illustrated in Figure 3.2.
The eect of the screening i.e. the value of C together with the limits C
min
and C
max
is translated into a factor s
ijk
. Two forms have been proposed and they
are listed in Equations (3.16) and (3.17).
Baskes 1994, Ref. [185] s
ijk
= exp(
_
C
max
C
C C
min
_
2
) (3.16)
Baskes 1997, Ref. [186] s
ijk
= f
_
C C
min
C
max
C
min
_
(3.17a)
where f(x) =
_
_
1 x 1
[1 (1 x)
4
]
2
0 < x < 1
0 x 0
(3.17b)
Finally, the cumulative eect of all the partial screening is represented by a
product of all the s
ijk
into a factor S
ij
. This factor is multiplied to all the functions
in the MEAM that result from the interaction between atoms i and j.
S
ij
=
k
s
ijk
(3.18)
A few observations can be made about the screening procedure. More than one
atom can screen each atom pair. Total screening occurs as long as one screening
atom has C < C
min
. The value of C
min
determines whether an atom pair is
totally screened or not, and hence changing C
min
controls the extent of the nearest-
neighbour interactions. There is no screening when all the C values of the atoms
are larger than C
max
. The extent of partial screening, which is represented by
the value of s
ijk
is controlled by the dierence between C
max
and C
min
. Partial
screening occurs as long as the set of all the screening atoms have C > C
min
. This
discussion is summarized in Equation (3.19). Finally, to turn o screening, one
3.3. THEORY 92
Figure 3.2: Illustrating the screening in the MEAM.
can set C
min
= C
max
= 0.01.
Total Screening: min(C
k
) < C
min
(3.19a)
Partial Screening: min(C
k
) > C
min
(3.19b)
No Screening: min(C
k
) > C
max
(3.19c)
To summarize, the screening procedure is best illustrated by the following
algorithm.
Algorithm 3.1 Screening.
Set S
ik
= 1.
For each atom k that lies between i and j
1. Calculate C
3.3. THEORY 93
i
j
k
r r
r
r
r
3
r
i
j
k
Case 1
Case 2
Figure 3.3: Two cases to explain the MEAM Screening Procedure. For Case 1,
C = 3 and for case 2, C =
1
3
.
2. If C < C
min
, S
ik
= 0 and exit.
3. If C > C
max
go to next atom.
4. If C
min
< C < C
max
, calculate s
ijk
and multiply with S
ij
.
The screening parameters play a role in controlling the extent of the atomic
interactions. Consider the two cases shown in Figure 3.3 . A triangular lattice is
shown, which is the (111) plane in the FCC structure. Two cases of screening are
shown. In each case, the atom marked k screens the pair i and j.
For case 1, the distance i-j corresponds to the rst neighbour distance in an
FCC cell and C = 3. Therefore in order for this interaction not to be screened at
all, C
max
must be set to be less than 3, and in most MEAM studies involving the
FCC structure, it is chosen to be 2.8.
Case 2 shows a third neighbour interaction and C =
1
3
. (The second neighbour
interaction occurs in the (100) plane at a distance of
2r). If the interaction

between atoms is to be allowed up to third neighbours, for partial screening, C
min
must be set below this value of one third.
3.3. THEORY 94
Embedding Function
The next feature of the MEAM that is dierent from the EAM is the design of
the Embedding function. In the MEAM [150], the argument to the embedding
function is expressed as
=
(0)
G() (3.20)
The mathematical expression of , which is the argument of the function G()
is given by
=
t
(1)
(
(1)
)
2
+t
(2)
(
(2)
)
2
+t
(3)
(
(3)
)
2
(
(0)
)
2
(3.21)
In equation (3.21), the parameters t
(1)
, t
(2)
, t
(3)
are weights to be determined in
the tting process. There are four partial electron densities represented by the
terms (
(0)
)
2
, (
(1)
)
2
, (
(2)
)
2
and (
(3)
)
2
. They are dened individually as follows
(0)
=
j
S
ij
d
(0)
j
(3.22a)
(
(1)
)
2
=
j
S
ij
x
d
(1)
j
_
2
(3.22b)
(
(2)
)
2
=
j
S
ij
x
d
(2)
j
_
2
(3.22c)
(
(3)
)
2
=
j
S
ij
x
d
(3)
j
_
2
(3.22d)
Each of the partial electron densities contains a function called the atomic
electron density. These are the functions d
(0)
j
, d
(1)
j
, d
(2)
j
, d
(3)
j
. Each of these func-
tions contains a parameter
(h)
and is dened as follows
3.3. THEORY 95
d
(h)
j
= exp(
(h)
r
) (3.23)
where the index h = 0, 1, 2, 3.
r
is the reduced interatomic distance and is set to be zero at the equilibrium

interatomic distance r
0
. This is obtained as follows,
r
=
r
ij
r
0
1 (3.24)
The partial electron densities also contain the terms x
, x
and x
. These are
the direction cosines for the interatomic distance vector and the subscripts , ,
are summation indices for the three coordinate directions. In other words, the
component of the unit vector in the direction is given by
x
=
r
r
ij
(3.25)
Due to the presence of the direction cosine terms, the partial electron densities
capture the eect of dierent atomic coordination. The same atom in dierent
crystal structures will have dierent atomic coordination. It will be surrounded by
dierent numbers of atoms in dierent ways. For example, an atom in the BCC
structure is surrounded by eight rst neighbours, while it will have twelve rst
neighbours in the FCC structure. The partial electron densities will then be able
to capture the dierence between the two. Thus, it has been commented by van
Beurden et al. that it is more accurate to think of the electron density as an
eective coordination number [187].
Hence, the signicance of the parameter is that it accounts for dierences
in atomic coordination by combining the eect of the four partial electron density
terms. The function G() converts this to a multiplier on
(0)
and two forms have
been proposed:
3.3. THEORY 96
G =
1 + (3.26a)
G =
2
1 + exp()
(3.26b)
To understand the eect of this multiplier, it can be noted that setting G as a
constant value of 1 recovers the EAM. Finally, the form of the Embedding Function
follows Baskes original denition [150],
F( ) = AE
coh

Z
ln(

Z
) (3.27)
where A is a scaling parameter, Z is the number of rst neighbours and E
c
is
the cohesive energy.
Pairwise Function
The pairwise function is typically dened by requiring that the potential reproduce
exactly the Universal Equation of State E
u
(UES) by Rose et al. [188]. The UES
is an energy-volume relationship for the crystal, and it is dened as
E
u
= E
coh
(1 +r
) exp(r
) (3.28)
r
is dened in equation (3.24). Given the Bulk Modulus B, equilibrium volume

per atom V
o
and cohesive energy E
coh
, the parameter in Equation 3.28 combines
these properties as follows
2
=
9BV
o
E
coh
(3.29)
For a monoatomic crystal where each atom has Z rst neighbours at a distance
of r
o
, for each atom, there is a total of Z pairwise energies (r
o
). Since each
pairwise energy is shared between two atoms, the total energy due to the pairwise
3.3. THEORY 97
potential is
1
2
Z(r
o
). Adding the embedding function gives the total energy per
atom. Hence, the following identity exists
E
u
=
1
2
Z(r
o
) +F (3.30)
The pairwise function is then obtained by rearranging the identity above to
obtain
=
2
Z
[E
u
F] (3.31)
Since E
u
is equal to the total energy of the crystal, this procedure ensures that
the cohesive energy of the atom E
coh
is always reproduced exactly. For compounds,
this procedure can be adapted by considering the various types of interactions in
the crystal.
To summarize, the relationship between the three potential models is shown in
Figure 3.4.
Force Expression
The total force on an atom a in a particular direction can be calculated by taking
the derivative of the energy of the entire system with respect to the coordinate of
the same direction. Hence, for the force in the x direction,
f
a,x
=
E
x
a
(3.32)
For a system under the EAM or a pair potential, this can be reduced into the
more ecient manner of calculating the force due to each pair of interacting atoms.
However, this is not possible due to the presence of the screening parameter, S
ij
which is a combination of three-body terms. Therefore, in general, the force on
any atom a is given as
3.3. THEORY 98
E =
j>i
(r
ij
) +
i
F
i
( ) =
(0)
G()
j
(r
ij
) G() =
1 + or
2
1 + exp ()
=
t
(1)
(
(1)
)
2
+t
(2)
(
(2)
)
2
+t
(3)
(
(3)
)
2
(
(0)
)
2
(0)
=
j
S
ij
d
(0)
j
(
(1)
)
2
=
j
S
ij
x
d
(1)
j
_
2
(
(2)
)
2
=
j
S
ij
x
d
(2)
j
_
2
(
(3)
)
2
=
j
S
ij
x
d
(3)
j
_
2
Pairwise EAM MEAM
Figure 3.4: Relationship between the pairwise, EAM and MEAM potential func-
tionals
3.3. THEORY 99
f
a,x
=
j>i
S
ij
x
a
(r
ij
) +
j>i
S
ij
(r
ij
)
x
a
+
i
F
i

i

i
x
a
(3.33)
The derivative of the screening parameter is
S
ij
x
a
=
k
S
ij
s
ijk
s
ijk
x
a
(3.34)
The detailed expressions in these equations are given in Appendix C. These
expressions demonstrate that the force calculation is computationally intensive.
In order to nd the forces on all the atoms, a quadruple for-loop is needed. For
N atoms with P neighbours each, this results in a force calculation that is of the
order O(N
2
P
2
).
Development of the MEAM
Along with the description of the MEAM, the rst parameters for the MEAM
were rst published in 1992 by Baskes[150]. However, some of the
(h)
parameters
were chosen in an arbitrary manner. Hence, many studies that revised the pa-
rameters for specic materials soon resulted. Baskes compared four combinations
of choices of
(1)
,
(3)
, C
min
and G() for Ni [186] in order to determine which
set provided the best predictions of a database of energies. Satou et al. adjusted
some of Baskes original parameters for Vanadium to obtain better predictions of
its elastic properties [189]. Van Beurden et al. found new parameters for Pt using
energies calculated from DFT in the tting database and ignored the screening
procedure, pointing out that it introduced dierent behaviour and physics into
the simulations[187].
There have also been some modications to the functions used in the MEAM.
Baskes initially conceived the MEAM with the atomic interactions limited to rst
3.3. THEORY 100
neighbours [150]. Lee et al. developed MEAM potentials for FCC metals [190]
and BCC metals [191] showing how the functions could be obtained with second
neighbour interactions [192]. Xiaoying et al. modied the Embedding Function
by adding a parameter (Equation (3.35)) in their attempt to develop better
potentials for BCC metals [193].
F( ) = AE
coh

Z
_
ln(

Z
)
_
(3.35)
Heino [177] and Kuo et al. [178, 182] included an additional parameter
0
in
the expression of the electron density (Equation (3.36)).
=

(0)
0
G() (3.36)
Baskes suggested a modication of the form of
_
(3)
_
2
in 1999 as follows [174]
(
(3)
)
2
=
j
x
S
ij
d
(3)
j
_
3
5
j
x
S
ij
d
(3)
j
_
(3.37)
Baskes et al. [170] and Kuo et al. [182] included higher order terms in the
Rose function in developing their potentials for Ga and Si-O respectively. Anglade
et al. made several modications to the functions in the MEAM to allow for
longer-ranged interactions with the screening procedure [194].
Thijsse showed that the Stillinger-Weber (SW) potential is a special case of
the MEAM [195]. This development allows simulations of materials with SW-type
potentials and MEAM potentials to be conducted together.
3.3.3 Optimization Methods
Optimization methods are integral to MD as a problem often posed is to nd the
minimum energy conguration of a number of atoms. A number of optimization
3.3. THEORY 101
methods will be employed in this work.
Nelder-Mead Simplex Method
The Nelder-Mead Simplex method [196] belong to a class of optimization tech-
niques useful in cases where the derivatives are dicult to obtain.
Simulated Annealing
The methods previously described ensure a local minimum, but do not ensure that
a global minimum is obtained. A recent development in optimization techniques
uses random numbers to solve combinatorial optimization problems such as the
travelling salesman problem, which is nding the shortest route through N cities.
Kirkpatrick et al. rst demonstrated the use of the Simulated Annealing tech-
nique to nd the optimal arrangement of chips on a circuit board and a 400-cities
travelling salesman problem [197].
There are two features in the simulated annealing method. The rst feature
is the use of random numbers to perturb the variables, and applying an accep-
tance/rejection criteria to the resulting moves. A key characteristic is that bad
moves, i.e. choices that raise the value of the objective function are allowed when
specic criteria are met. The second feature is a cooling schedule. If the value
of the objective function is thought of as the temperature, a cooling schedule
species the procedure for lowering the temperature of the system, so that the
number of bad moves are progressively reduced.
A simplication of the simulated annealing method was introduced by Dueck,
which was termed Threshold Accepting [198]. It removes the need to arbitrarily
specify a cooling schedule. Instead, it is replaced by a limit called the water
level. Moves that increase the value of the objective function below this limit
are accepted. Every good move that reduces the value of the objective function
results in a small reduction in the value of the water level. The algorithm is listed
3.3. THEORY 102
in Algorithm 3.2. This will be used in searching for the lattice constants that give
the minimum energy for the NiAs crystal structure.
In such techniques, the quality of the random number generator (RNG) is
important. RNGs in computing are generated by mathematical formulae, and
the strings of random numbers generated are repeated periodically. Hence it is
necessary to choose a RNG with a suciently large period, so that within the
requirements of the problem, the numbers generated are close to random. In-built
RNGs provided with the C compiler are insucient, as the period is known to be
2
32
1. The Mersenne twister RNG developed by Matsumoto et al. is used, as
it has a period of 2
19937
1 [199].
Algorithm 3.2 Threshold Accepting
1. Set initial guess x start
2. Set starting values: water level, rain speed
3. Set scale of random numbers: rand scale
4. Set MaximumIterations and Iterations = 0
5. While Iterations < MaximumIterations
(a) Perturb the variables for a new guess. x trial = x start + rand scale
random numbers
(b) Calculate objective function f(x trial).
(c) if (f(x trial) < water level)
i. Reduce the water level. water level = water level - rain speed
ii. Assign x start = x trial
(d) Iterations = Iterations + 1
3.4. POTENTIAL FITTING 103
Table 3.3: Parameters for Cu-Sn by Aguilar et al..
Cu Sn Cu-Sn
E
coh
3.62 3.08 3.5
5.106 6.20 5.38
r
o
2.50 3.44 2.68
A 1.07 1.0 -
Z 12 12 -
(0)
3.62 6.20 -
(1)
2.20 6.00 -
(2)
6.00 6.00 -
(3)
2.20 6.00 -
t
(1)
3.14 6.5 -
t
(2)
2.49 4.5 -
t
(3)
2.95 -0.183 -
Taken from Ref. [157]
Table 3.4: Screening parameters used by Aguilar et al.
C
min
C
max
Cu-Cu-Cu 0.8 2.8
Cu-Cu-Cu 2.0 2.8
Cu-Cu-Sn 2.0 2.8
Cu-Sn-Sn 2.0 2.8
Sn-Cu-Sn 2.0 2.8
Sn-Sn-Sn 2.0 2.8
3.4 Potential Fitting
3.4.1 Predictions of existing parameters
Parameters in the MEAM for Cu-Sn interactions are already available, through
the work of Aguilar et al. in their simulations of the deposition of Cu on Sn [157]
(Tables 3.3 and 3.4).
To evaluate whether these parameters are suitable for simulating Cu
6
Sn
5
, a
geometry optimization calculation using the threshold acceptance method was
carried out using the hexagonal structure reported by Gangulee et al. [109] . The
0 2 4 6 8 10
x 10
4
3
2.5
2
1.5
1
0.5
0
0.5
1
Number of moves
E
n
e
r
g
y

(
e
V
)
Figure 3.5: Evolution of the energy of Cu-Sn B8 unit cell predicted using the
Aguilar MEAM potential
calculation was terminated after 20000 successful moves.
The evolution of energy vs accepted moves is shown in Figure 3.5, and the
evolution of a and c is shown in Figure 3.6. The Gangulee lattice constants are
chosen as the initial conguration, and it can be seen that the energy at this
initial conguration is positive. After about 2000 moves, the energy settles to -2.5
eV/atom, with the lattice constants settling at a = 4.5

A and c = 8.0

A. These
lattice constants are clearly dierent from those reported by Gangulee, which are
a = 4.19

A and c = 5.04

A.
Also, since the energy of the B8 lattice, -2.5 eV/atom, is higher than the
cohesive energy of the individual atomic species, this means the formation energy
is positive. Hence, it is concluded that the MEAM parameters used by Aguilar et
al. cannot be used to simulate Cu
6
Sn
5
. This brings about the necessity to nd
parameters of the MEAM potential that can reproduce the properties of Cu
6
Sn
5
.
0 2 4 6 8 10
x 10
4
3
4
5
6
7
8
9
10
11
12
Number of moves
l
a
t
t
i
c
e

c
o
n
s
t
a
n
t

(
1
0
1

n
m
)
lattice constant a
lattice constant c
Figure 3.6: Evolution of the lattice constants a and c of Cu-Sn B8 unit cell using
the Aguilar MEAM potential
3.4.2 Potential Fitting Strategy
In this section, the potential tting strategy is summarized. Details will be de-
scribed in the following sections.
Firstly, a reference structure is chosen to determine the Cu-Sn interaction
function. Possible choices of the reference structure are compounds which exists
within the alloy system or one that is hypothetical. Either way, the nal inter-
atomic potential obtained will reproduce the properties of the compound. In this
study, since the focus is on the Cu-Sn intermetallic compound, the reference struc-
ture chosen is the NiAs structure reported by Gangulee et al. [109]. This follows
the precedent of Heino, who also chose a hexagonal reference structure to t a
Ta-Si potential [177]. Gangulee et al. reports that the 6:5 stoichiometry is due
to the interstitial sites being partially occupied. To simplify matters, this feature
will be ignored.
The choice of the NiAs structure is also a simplication from the actual struc-
ture reported by Larsson et al.. If the Larsson structure of Cu
6
Sn
5
[86] is chosen
as a reference structure, the resulting minimization will have an immense scope.
Firstly, it will be a fundamental requirement for the potential to reproduce the
eighteen internal coordinates, the four cell parameters, the thirteen elastic con-
stants and the cohesive energy. Secondly, as the monoclinic structure is of low
symmetry, the tting database must include the requirement that the forces on all
44 atoms to be zero. As there are only 23 parameters in the MEAM for two atomic
species, additional parameters to t all these constraints must come from making
suitable modications to the MEAM functions. The optimization algorithm will
take a longer time to sample the parameter space as well.
Hence, from the reference structure chosen, parameters can be obtained for
Roses Universal Equation of State(UES) in Equation 3.28 [188]. In most cases,
experimental data is available which then enables the parameters to be determined
quickly. However, in the case of the NiAs structure, the energy-volume data has
to be obtained from density functional theory calculations. In section 3.3, it was
explained how the UES is used to obtain the pairwise interaction for a single
atomic species. This can also be done for two atomic species to determine the
Cu-Sn interaction function, and details will be elaborated in subsequent sections.
In previous studies for binary alloys, there have been two approaches to obtain
potentials for binary alloys. The rst approach is to obtain potentials separately
for the individual metals, and then obtain the cross potential separately. This
aords convenience as potentials previously developed for the individual metals
can be combined, thus simplifying the potential tting process. This approach
has several precedents in the scientic literature, including the EAM potential for
Ni-Al by Voter et al.[154] and the MEAM study by Kim et al. for Cu-Zr [200]
and the MEAM study by Lee for Fe-H [201]. In the MEAM studies, the original
parameters for the individual atoms were left unchanged, while the C
min
and C
max
parameters were adjusted. Instead of simultaneously adjusting these parameters,
each parameter was adjusted individually to t dierent properties in the tting
database. The second approach is to simultaneously t all the potential functions.
The rationale for this is to allow more exibility in the tting process, especially
if a large database of properties is to be reproduced. This approach has also been
taken in the EAM by Mishin et al. for the Ni-Al alloy [155]. In the MEAM, van
Beurden et al. carried out a simultaneous tting for the parameters in the CO-Pt
potential [180].
Essentially, the rst approach means that the shapes of the functions are al-
ready determined before the tting process. Adjusting the C
min
and C
max
param-
eters merely controls the values of these functions and the interatomic distances at
which they apply. In contrast, adjusting the parameters of the MEAM functions
allows the exibility to change their shape, which then improves the exibility of
the interatomic potential functions to adjust during the tting process.
In this study, the latter approach is chosen because it is felt that the unique
reference structure chosen requires the exibility of the functions to change their
shape. The set of adjustable parameters that is chosen comes from both the
parameters for Cu and Sn as well. They are simultaneously tted such that the
potential reproduces a database of properties. The C
min
and C
max
values remained
xed, similar to the approach taken by Heino [177]. This means that the original
parameters for Cu and Sn will now be changed to t the properties of the Cu-Sn
structure.
In summary, the strategy for potential tting in the MEAM is as follows:
1. The B8 NiAs unit cell is chosen as the reference structure to represent the
Cu-Sn interactions.
2. The Energy-Volume relationship and elastic constants are calculated for the
reference structure (Section 3.4.3).
Table 3.5: Elastic Constants calculated. Calculations are performed with the
GGA.
Energy cuto,
k-point mesh
c
11
(GPa)
c
33
(GPa)
c
44
(GPa)
c
12
(GPa)
c
13
(GPa)
B
(GPa)
320,101007 71.1 118.3 51.0 124.7 77.3 91.0
320,141410 143.3 158.9 60.6 60.1 53.1 86.4
320,171712 148.8 184.4 62.5 63.8 50.0 89.7
500,101007 122.1 191.0 44.7 104.3 55.1 95.8
500,141410 141.3 168.1 58.5 85.4 57.3 90.7
3. With the data obtained, the tting database is assembled (Section 3.4.4).
4. Choices are made regarding the various MEAM functions and parameters to
be tted (Section 3.4.5)
5. Constraints on the parameters are identied and the objective function to
be minimized is constructed (Section 3.4.6).
3.4.3 Density Functional Theory Calculations
With a reference structure chosen, its properties must be obtained such that the
Roses UES (Equation (3.28)) can be included in the interatomic potential. Gan-
gulee et al. reports only the lattice constants, which provides the equilibrium
volume [109]. However, the elastic constants and the energy-volume relationship
are not available. Hence, the calculations that obtain these properties are described
in this section.
Elastic Constants
DFT calculations of the elastic constants are examined at a variety of k-points and
energy cut-o settings, and summarized in Table 3.5. Only the GGA is used, as
it has been shown in Chapter 2 that calculations of the elastic constants with the
GGA produce results that are closer to experiment.
As the elastic constants are obtained from a stress-strain linear curve t, the
energies must be calculated more accurately than the lattice constants. Therefore,
the k-point settings for calculating the lattice constants are insucient for the
elastic constants.The results for a k-point mesh of 10107 dier vastly from the
results with a k-point mesh of 141410 and 171712, which have a similar
order of magnitude.
For the tting database, the results from 500 eV, 141410 are chosen as they
are calculated with the most stringent accuracy settings.
Energy-Volume Relationship
The equation of state can be obtained by performing DFT calculations to obtain
the energy as a function of volume[187]. However, because the B8 cell has two
lattice parameters a and c, for each value of the volume, the values of c and a that
minimize the energy need to be obtained.
The procedure by Uludogan et al. was followed [67]. The energy is calculated at
various values of the external pressure to the unit cell. More points were sampled
at lower pressures to capture the behaviour close to the equilibrium volume. For
each point, the energy was minimized with respect to the lattice constants a and
c. To check the convergence of the calculations, two accuracy settings were used,
the default Ultrane setting provided by CASTEP (320 eV, 7 7 6) and a
stricter setting comprising of energy cuto 600 eV and a k-point mesh of 14 14
10.
The outcome is shown in Figure 3.7. It can be seen that the choice of the
accuracy setting had an eect on the energy. To examine the dierence, the rst
10 points closest to the equilibrium volume are shown in Figure 3.8. It is shown
that the Ultrane setting did not produce satisfactory behaviour.
The variation of the c/a ratio with volume is shown in Figure 3.9. As the
pressure increases, the c/a ratio increases by less than 1 percent. It can thus be
assumed that the c/a ratio remains constant. Employing this assumption, the
54 56 58 60 62 64 66 68 70 72
0
0.5
1
1.5
Volume (A
3
)
E

E
m
i
n

(
e
V
)
UltraFine
600 eV cutoff
Figure 3.7: Energy-Volume calculations for B8 Cu-Sn with DFT. The lines are a
guide to the eye.
curve in Figure 3.7 is tted to the Rose UES (Equation (3.28)), so as to obtain
the parameter for Cu-Sn interactions. By minimizing the least-squares residual
with the Nelder-Mead simplex method, it was found that = 5.388.
3.4.4 Fitting database
With the necessary ab-initio calculations performed, the tting database can now
be constructed. The properties of Cu and Sn themselves will be excluded from
the database because the objective of this work is for the properties of Cu
6
Sn
5
only. For example, including only their elastic constants c
11
, c
12
and c
44
will add
24 energies to the tting database. This will expand the database to a size larger
than the number of adjustable parameters available.
The primary inputs in the tting database are the elastic constants. The elastic
constants are, of course, the second derivative of the total energy of the system
(Equation (3.8)) with respect to the appropriate strains. With the complexity
68.6 68.8 69 69.2 69.4 69.6 69.8 70 70.2 70.4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
x 10
3
Volume (A
3
)

E

E
m
i
n

(
e
V
)
UltraFine
600 eV cutoff
Figure 3.8: Energy-Volume calculations for B8 Cu-Sn with DFT, enlarged to show
points near equilibrium volume.
54 56 58 60 62 64 66 68 70 72
1.23
1.235
1.24
1.245
1.25
1.255
Volume (A
3
)
c
/
a

r
a
t
i
o
UltraFine
600 eV cutoff
Figure 3.9: Variation of equilibrium c/a ratio with volume. The lines are a guide
to the eye.
of the MEAM potential, deriving analytical expressions requires much tedious
algebra.
Instead, a nite dierence approximation is used for each elastic constant.
Firstly, c
11
and c
33
are used directly. c
12
is included in c
aa
, which is dened in
Equation (3.38a). This elastic constant is the result of simultaneously lengthening
the 1-direction and shortening the 2-direction by the same amount, eectively
distorting the basal plane in the B8 unit cell. c
13
is included in c
cc
, which is dened
in Equation (3.38b). This elastic constant is a result of a simultaneous lengthening
of the 1-direction and shortening of the 3-direction by the same amount, eectively
changing the c/a ratio. In the tting database, each of these elastic constants is
represented by two energies at a strain = 0.001.
c
aa
= 2 (c
11
c
12
) (3.38a)
c
cc
= 2 (c
11
+c
33
2c
13
) (3.38b)
It must be mentioned that the strains applied for these elastic constants break
the hexagonal symmetry of the crystal. In actual fact, strains can be applied that
retain the hexagonal symmetry of the crystal [202]. These strains would result in
a dierent combination of elastic constants. The program for the tting database
was designed before this idea was understood. Nevertheless, it will be shown in
later sections that this aw has little eect on the values of the elastic constants
reproduced by the potential.
The shear elastic constant c
44
is excluded because calculating its value accu-
rately requires relaxation of the forces (e.g. with conjugate gradient methods),
which will unduly prolong the optimization calculation. Instead, predicting its
value will be part of the validation process.
Due to the procedure described in section 3.3.2, there is no need to explicitly
include the cohesive energy in the tting database.
Table 3.6: The tting database
Description Energy (eV)
c
11
,
1
= +0.001 -3.429992
c
11
,
1
= 0.001 -3.429992
c
33
,
3
= +0.001 -3.429990
c
33
,
3
= 0.001 -3.429990
c
aa
,
1
=
2
= +0.001 -3.429994
c
aa
,
1
=
2
= 0.001 -3.429994
c
cc
,
1
=
3
= +0.001 -3.429989
c
cc
,
1
=
3
= 0.001 -3.429989
a = 3.05

A, c = 9.25

A -2.85
E
a
= 0.0 at a = 4.19

A NA
E
c
= 0.0 at c = 5.04

A NA
Only the non-zero strains are listed
The equilibrium lattice constants are chosen to be a = 4.19

A and c = 5.04

A,
the experimental lattice constants reported by Gangulee et al. [109]. To enable
these values to be reproduced, the condition is imposed that the rst derivative of
the energy with respect to the lattice constants are zero, i.e.
E
a
= 0 (3.39a)
E
c
= 0 (3.39b)
In an initial optimization run with the four elastic constants and the two
derivative conditions, the potential obtained was found to predict the structure
at a = 3.05

A and c = 9.25

A to be more stable than the structure at a = 4.19
A and c = 5.04

A. The decision was then taken to include this unexpected struc-
ture in the tting database. The energy of this unexpected structure was calculated
by DFT and found to be -2.85 eV. The entire tting database is summarized in
Table 3.6.
3.4.5 Choice of functions and parameters
Various options are available for several of the components in the MEAM for-
malism. Also, a decision must be made regarding the parameters involved in the
tting process. The choices made for them are described in this section.
Choice of C
min
and C
max
Parameters related to the screening procedure are rst described. Due to the
small size of the tting database, the screening parameters C
min
and C
max
were
not included in the set of adjustable parameters. To simplify the tting process, it
is desirable to have a xed form of
CuSn
throughout. This is done by choosing
C
min
and C
max
such that interactions for each atom are limited to rst neighbours.
This was also the approach taken by Heino in tting a potential for the hexagonal
TaSi
2
[177].
It has been shown that the values of C
min
and C
max
decide the extent to which
each atom interacts with its neighbours. Thus, if these parameters were allowed
to vary during the tting, additional terms might appear in
CuSn
due to more
neighbours being taken into account, which would complicate the tting process
as steps must be taken to account for them.
It was found that the values chosen by Ravelo et al. can largely be retained
with one small modication, and are listed in Table 3.7. C
max
for Cu-Cu-Cu has
to be reduced to 2.7 to ensure that the Cu atom could have interactions with its
two nearest Cu neighbours.
These parameters constrain the interactions in the following manner. In the
perfect crystal, the Cu atom in the basis has 2 Cu rst neighbours and six Sn rst
neighbours, while the Sn atom in the basis has six Cu rst neighbours, forming a
wedge-shaped coordination polyhedron with equilateral triangles on the top and
bottom. This is illustrated graphically in Figure 3.10, where the basis atoms are
Figure 3.10: Coordination Polyhedra for Cu and Sn in the B8 unit cell. The basis
atoms are indicated with yellow letters and the rst neighbours are joined with
yellow lines to show the coordination polyhedron.
Table 3.7: Screening parameters
C
min
C
max
Cu-Cu-Cu 2.0 2.7
Cu-Cu-Cu 2.0 2.8
Cu-Cu-Sn 2.0 2.8
Cu-Sn-Sn 2.0 2.8
Sn-Cu-Sn 2.0 2.8
Sn-Sn-Sn 2.0 2.8
indicated with yellow letters and the rst neighbours are joined with yellow lines
to show the coordination polyhedron. Both atoms do not have partially-screened
second neighbours.
Choice of screening function
Having decided the values of C
min
and C
min
, the next choice to be made is for the
function for s
ijk
. The choice by Aguilar et al. (Equation (3.16)) is retained.
Choice of G()
With the screening function, the partial electron densities given in Equation (3.22)
are calculated and combined in the function in Equation (3.21). Following which,
the choices by Aguilar et al. for the function G() for both Cu and Sn are retained
and stated in Equation 3.40.
Cu: G() =
1 + (3.40a)
Sn: G() =
2
1 + exp()
(3.40b)
Form of
CuSn
For the monoatomic metals Cu and Sn, is obtained using Equation (3.31). The
process of obtaining the cross potential
CuSn
, the pairwise function that de-
scribes interactions between Cu and Sn, is now described.
The rst constraint that is imposed is that the Roses UES E
u
is exactly repro-
duced for the reference structure. Hence, the following expression can be written
down
E
u
=
j>i
(r
ij
) +
i
F
i
( ) (3.41)
To obtain the expression, it is noted that Equation (3.41) can be simplied
for a perfect lattice. Let the interactions be limited to those with the nearest
neighbours. The NiAs unit cell has just two atoms in its basis, one Cu and one
Sn [118], as the other atoms shown in Figure 3.10 arise out of the symmetry of
the crystal. Let the number of interactions of each atom be N
CuCu
, N
CuSn
, N
SnSn
.
The total energy of the reference structure is then expressed in Equation (3.42).
E
u
=
1
2
[N
CuCu
CuCu
+N
SnSn
SnSn
+N
CuSn
CuSn
+F
Cu
+F
Sn
] (3.42)
The form of
CuSn
depends on the extent of the nearest-neighbour interactions
allowed by the screening parameters. Also, Heino showed that it requires an
intimate understanding of the type of nearest-neighbour interactions in the choice
of the reference structure [177].
The choice of the screening parameters in Table 3.7 result in the following
unscreened interactions: six Cu-Sn interactions (N
CuSn
= 6), 2 Cu-Cu interactions
(N
CuCu
= 2) and no Sn-Sn interactions (N
SnSn
= 0). Substituting these numbers
into Equation (3.42), the following expression for
CuSn
is obtained,
CuSn
=
1
3
E
u
1
6
F
Cu
1
6
F
Sn
1
3
CuCu
(3.43)
In their commentary on the appropriate choice of a reference structure, Kuo et
al. observed that restricting interactions to rst neighbours works well for struc-
tures that are reasonably tightly packed, but is not sucient for open structures
like silica [182]. They pointed out that increasing the number of neighbours
complicates the expression for the pair function.
Choice of adjustable parameters
The majority of the adjustable parameters are found in the expressions of the
embedding function F and its argument . For the embedding function in Equation
(3.27), A is allowed to vary. Z is kept at 12, which retains the value used originally
in the potentials for the individual metals. This is possible because Z merely acts
as a scaling factor for , and the t
(h)
parameters can play the same role too.
For the partial electron densities in Equation (3.22), it is found that not all
of them have an eect on the value of . If only rst neighbours are considered,
the lattice sums in Equation 3.22 can be worked out. The details are described
in Appendix C.3. It turns out that for the Cu atom, (
(1)
)
2
and (
(3)
)
2
are zero
for all c/a ratios while only (
(1)
)
2
is zero for the Sn atom. Therefore the values
by Aguilar et al. for
(1)
Cu
,
(3)
Cu
,
(1)
Sn
, t
(1)
Cu
, t
(3)
Cu
and t
(1)
Sn
are retained. The remaining
parameters will be adjusted in the tting process.
Parameters in the Rose Universal Equation of State
The pairwise functions require the Rose UES (Equation (3.28)) as an input. For
Cu and Sn, the parameters , r
o
, E
coh
in the Rose UES for the individual metals
as determined by Aguilar et al. are retained.
For the Rose UES for CuSn, = 5.388 having been determined by the DFT
calculations in section 3.4.3.
The cohesive energy E
coh,CuSn
can be calculated by making use of the avail-
able experimental values of the enthalpy of formation H. Since H is usually
measured at room temperature and not at 0 K, the resulting value of the cohesive
energy is an estimate.
In this case, DFT calculations are not employed to obtain the 0 K formation
energy, as the evidence available so far is that DFT calculations of such a quantity
are inconsistent. DFT calculations by Ghosh obtained values of -4.0 kJ/mol and
-3.2 kJ/mol [85], while the calculations performed in this thesis (details are shown
in Appendix D) obtained a value of -9.9 kJ/mol.
Hence, the experimental values of the enthalpy of formation are used. There
are two sets of values and they are summarized in Table 3.8. From the data in the
table, H can be taken to be approximately 7.1 kJ/mol, or 0.074 eV/atom
(since 1 kJ/mol is approximately equal to 0.0104 eV/atom).
There are two atoms per f.u. in the B8 unit cell i.e. one atom of Cu and Sn
each. Therefore, the cohesive energy per f.u. for B8 CuSn is written as 2E
coh,CuSn
.
Table 3.8: Available experimental values of the enthalpy of formation
Author Temperature (K) H (kJ/mol)
Flandorfer et al. (Ref [203]) 298 6.1 1
Gangulee et al. (Ref [109]) 273 7.04 0.06
The enthalpy of formation can then be written as
H =
1
2
[2E
coh,CuSn
E
coh,Cu
E
coh,Sn
] (3.44)
which results in the following expression for the cohesive energy
E
coh,CuSn
=
1
2
[E
coh,Cu
+E
coh,Sn
+ 2H] (3.45)
where the inputs to the energy are the published values of Cohesive Energy of
E
coh,Cu
= 3.62 eV and E
coh,Sn
= 3.08 eV [150]. To two decimal places,
E
coh,CuSn
= 3.43 eV/atom.
The parameter r
o
= 2.70

A, which represents the rst-neighbour distance in the
reference structure. For B8 CuSn, it depends on the value of the lattice constants
a and c and is given by Equation (3.46). Its derivation is found in Appendix C.2.
r
CuSn
=
_
a
2
3
+
c
2
16
(3.46)
3.4.6 Optimization Methodology
Objective function
The goal now is to nd the parameters for the MEAM functions such that they re-
produce as accurately as possible the energies and properties in the tting database
in Table 3.6. This is an optimization problem and the rst step is now to construct
a function to be minimized in order that this goal can be achieved. This function
is known as the objective function.
One common way of constructing the objective function is to consider the size
of the dierences between the target values in the tting database and the calcu-
lated values. There are also constraints on the values of the function parameters
themselves due to the design. With these two considerations, the following objec-
tive function was used to obtain the parameters
Z = Z
p
+Z
penalty
(3.47)
Z
p
is the weighted sum of the squares of the dierences between the targeted
tting properties in the database p
k,database
and the properties calculated by any
arbitrary choice of parameters p
k,calc
. Mathematically, it is expressed as follows
Z
p
=
k
w
k
(p
k,database
p
k,calc
)
2
(3.48)
where w
k
are the individual weights for each dierence. Values of the weights
w
k
are given in Table 3.10.
Z
penalty
is a penalty function which adds a large value to the objective function
when any of the constraints are violated and it is dened as follows.
Z
penalty
=
_
_
0 if no constraints are violated
1 10
10
if any constraint is violated
(3.49)
In summary, the task at hand is now to nd the set of parameters for the
MEAM functions that minimizes the objective function Z. The set of parameters
that minimizes Z will give us an MEAM interatomic potential that reproduces the
values in the tting database for Cu-Sn in Table 3.6.
Constraints on parameters
The constraints placed on the parameters are now described in this section. Initial
optimization runs without constraints resulted in values that were extremely large
Table 3.9: Constraints placed on the tting parameters
Parameter min max
(h)
1.0 9.0
t
(2)
Cu
0.0 -
A 0.5 2.0
or extremely small. Hence, constraints were placed on some of the parameters.
A survey of literature values for
(h)
reveals that they are mostly of the order of
magnitude of 1, with values ranging from 1 to 9. Hence,
(h)
was constrained to
between 1 and 9.
Most of the values of A published for various MEAM potentials were close to
1.0. van Beurden et al. goes further to say that this value of A avoids anomalies
such as negative values for the unrelaxed vacancy formation energy in the EAM
limit [187]. Kuo and Clancy placed the constraint 0.8 < A < 1.2 during their
parameter search for the Si-O potential. In this work, this constraint is widened
to 0.5 < A < 2.0.
Finally, because of the choice of function of G for Cu, t
(2)
Cu
was constrained to
be greater than zero. These constraints are summarized in Table 3.9.
Choice of weights
The magnitude of the weights decide the accuracy to which the energies in the
tting database (Table 3.6) are reproduced. To reproduce the elastic constants
accurately, the energy must be accurate up to 1 10
6
eV. Hence, a relatively
large weight of 5 10
9
was chosen for all the energies that were related to the
elastic constants. In contrast, the requirement for the unexpected structure of
a = 3.05

A, c = 9.25

A is less stringent. It will be acceptable as long as it has a
higher energy than the experimental structure, thus its weight was chosen to be
a low value of 1. A weight of 10 was found to work well for the rst derivative
Table 3.10: Weights for the items in the tting database
Item Weight
Finite Dierence Energies 5 10
9
E
a
,
E
c
= 0 10
Energy of a = 3.05

A, c = 9.25

A structure 1
requirements. The weights used are summarized in Table 3.10.
Choice of Optimization technique
The objective function is minimized by the Nelder-Mead simplex method [196] as
implemented in the MATLAB function fminsearch.
Implementation
A set of matlab codes were written to carry out the optimization. These consisted
of subroutines for the following
Input le. This is where the inputs (eg. the tting database) are specied.
Calculating the Objective Function. The value of the objective function is
calculated according to the following algorthm
Algorithm 3.3 MEAM Objective Function
1. Read the trial parameters
2. Generate the MEAM functions in tabulated form
3. Calculate each item in tting database
4. Calculate Z
Calculating the MEAM Energy and Forces.
The subroutines were designed to be exible in accommodating the size of the
tting database as well as the number of adjustable parameters.
3.5. RESULTS AND VALIDATION 123
Table 3.11: Parameters for Cu-Sn obtained from the potential tting process - Set
P1
Cu Sn Cu-Sn
E
coh
3.62 3.08 3.43
5.106 6.20 5.388
r
o
2.50 3.44 2.70
A 0.5076 2.0000 -
Z 12 12 -
(0)
2.1611 2.3932 -
(1)
2.20 6.00 -
(2)
9.0000 2.0265 -
(3)
2.20 1.0000 -
t
(1)
3.14 6.5 -
t
(2)
11.6444 24.9504 -
t
(3)
2.95 -9.1503 -
3.5 Results and Validation
3.5.1 Parameters Obtained
Typically, more than one set of parameters will minimize the objective function
and satisfy the constraints. Two sets of parameters were obtained, which are
termed Set P1 (Table 3.11) and Set P2 (Table 3.12). For both sets, the C
min
and
C
max
values have previously been xed and are found in Table 3.7.
The similarities in both sets are as follows. The same values of E
coh
, and
r
o
are used as they were determined by DFT calculations. Very similar values
were also obtained for
(2)
Cu
,
(0)
Sn
and
(3)
Sn
. The dierences lie in the values of the
parameters A, t and the rest of the parameters that were adjustable.
The potential tting process is inherently arbitrary, starting from the choice
of the properties in the tting database, right down to the form of some of the
functions to be used in the MEAM potential. Hence, for the rest of this section,
calculations will be performed to understand the features and characteristics of
these parameters, so as to make a choice regarding the better set and to understand
where their limitations lie.
Table 3.12: Parameters for Cu-Sn obtained from the potential tting process - Set
P2
Cu Sn Cu-Sn
E
coh
3.62 3.08 3.43
5.106 6.20 5.388
r
o
2.50 3.44 2.70
A 0.3181 1.6228 -
Z 12 12 -
(0)
2.7993 2.3878 -
(1)
2.20 6.00 -
(2)
9.0000 2.3066 -
(3)
2.20 1.0580 -
t
(1)
3.14 6.5 -
t
(2)
19.0486 2.2802 -
t
(3)
2.95 -4.9919 -
3.5.2 Calculation Method
A full-scale computer program was written to perform MD simulations with the
MEAM. Full details will be described in the next chapter but some features relevant
for the calculations in this section will be described.
Inputs for each calculation, including the coordinates, function parameters and
boundary conditions are specied in an input le that is read by the program. Sub-
routines were written to implement the MEAM energy equations and the MEAM
force equations. Other subroutines that perform the threshold accepting calcula-
tions and the conjugate gradient minimization then call these subroutines.
The computer program was designed only to accommodate an orthogonal sim-
ulation box. In other words, the unit cells used must have mutually perpendicular
cell vectors. This poses problems with the hexagonal structure of the B8 unit cell.
However, using the idea that a unit cell is not unique (previously described in
Chapter 2), an orthogonal unit cell that retains the hexagonal symmetry can be
used.
How this is derived is illustrated in Figure 3.11. In this gure, the crystal is
shown in the basal plane and each white tile represents one B8 unit cell. The
Cell A
Original B8
unit cell
Cell B
Figure 3.11: Alternative orthogonal unit cells that have the same symmetry as the
B8 unit cell. The c-direction is perpendicular to the plane of the diagram. Both
Cell A and Cell B have sides of length a
3a c.
atoms are omitted for clarity. One of these tiles is highlighted in grey, and the
directions of the a vector are shown. The c-axis is perpendicular and goes out of
the page.
Two possible orthogonal unit cells are shown as a rectangular box shaded in
lighter grey. Both have dimensions of a
3a. Cell A has sides parallel to the a-axis

and c-axis and thus will be used as a substitute for the B8 cell for all calculations
except when specied. The internal coordinates of the atoms are shown in Table
3.13. Cell B is 60
clockwise from Cell A. Due to its orientation, it can used for

calculations for the elastic constant c
44
, and its internal coordinates are shown in
Table 3.14.
3.5.3 Minimum Energy Structure
In order to nd the minimum energy structure predicted by the potential above, a
threshold acceptance calculation was carried out. Using a and c as the adjustable
variables, the objective function was the cohesive energy itself. Therefore, the
water level is the threshold for the cohesive energy. An orthogonal simulation
box size of 4a 4
3a 4c was used, consisting of a total of 64 cells of Cell A or

Table 3.13: Coordinates of the atoms in Cell A. They are expressed as a fraction
3a c.
Species x y z
Sn 0 2/3 1/4
Sn 0 1/3 3/4
Sn 1/2 5/6 3/4
Cu 0 0 0
Cu 0 0 1/2
Sn 1/2 1/6 1/4
Cu 1/2 1/2 0
Cu 1/2 1/2 1/2
Table 3.14: Coordinates of the atoms in Cell B. They are expressed as a fraction
3a c.
Species x y z
Sn 0 1/3 1/4
Sn 0 2/3 3/4
Cu 0 0 0
Cu 0 0 1/2
Sn 1/2 1/6 3/4
Sn 1/2 5/6 1/4
Cu 1/2 1/2 0
Cu 1/2 1/2 1/2
Table 3.15: Lattice constants of the B8 unit cell predicted by the two sets of
interatomic potential parameters using the threshold acceptance calculations. For
each, three trials are performed.
Trial a (
A) c (
A) E
coh
(eV/atom)
Set P1
1 4.113271 5.132905 -3.434442
2 4.113377 5.133412 -3.434442
3 4.113349 5.133667 -3.434442
Set P2
1 4.118914 5.105980 -3.432454
2 4.118905 5.106012 -3.432454
3 4.118911 5.105947 -3.432454
128 B8 unit cells.
Because of the use of random numbers in the threshold acceptance algorithm,
three trials of 50000 moves each were conducted. The results of the trials are
summarized in Table 3.15, and the typical evolution of the lattice constants during
each trial is shown in Figure 3.12.
It can be seen that each trial produces similar results and it is also noted that
their values have a dierence of not more than 2% compared to the experimental
lattice constants of a = 4.19 eV c = 5.04 eV [109]. The cohesive energy of both
trials also agree to two decimal places with the value of 3.43 eV/atom that was
used in the tting database.
The elastic constants are calculated by applying a series of appropriate strain
patterns and tting the resulting energy-strain curve to a quadratic polynomial.
The strain patterns are similar to what was carried out in Chapter 2, and are
derived based on the strain energy expression of a single crystal that was described
in section 2.3.3. They are listed in Table 3.16 and illustrated in Figure 3.13.
0 5000 10000 15000
3.8
4.0
4.2
4.4
4.6
4.8
5.0
5.2
5.4
Number of Moves

L
a
t
t
i
c
e

c
o
n
s
t
a
n
t

(

A
n
g
s
t
r
o
m

)
a
c
Figure 3.12: Evolution of the lattice constants during the threshold acceptance
process
Table 3.16: Strain patterns for the various elastic constants
Elastic Constants Strain Pattern
c
11
+c
12

1
= ,
2
=
c
11
+c
12
+ 2c
33
4c
13

1
= ,
2
= ,
3
= 2
c
33

3
=
c
11
c
12

1
= ,
2
=
c
44

4
=
c
33
3
c
11
+c
12
2

2
3
c
11
+c
12
+2c
33
-4c
13
c
11
-c
12
2
Unit Cell
a
a
c
a
a
a
60
c
44
60
2

2
Figure 3.13: Strain patterns for the various elastic constants illustrated by showing
how the unit cell deforms. The strains for c
11
c
12
and c
44
destroy the hexagonal
symmetry.
Table 3.17: Elastic Constants calculated. All values are given in GPa. c
13
is
calculated from the combined elastic constants.
c
ij
P1 P2 DFT
c
11
+c
12
232.9 216.1 226.7
c
11
+c
12
+ 2c
33
4c
13
214.7 211.4 334.0
c
33
111.1 121.4 168.1
c
13
60.1 61.9 57.3
Calculating c
11
+c
12
, c
13
, c
33
The strains used in calculating the rst three elastic constants retain the hexagonal
symmetry of the crystal, and so do not require any energy minimization at each
step, thus simplifying the calculation. The strains for c
11
+c
12
changes the length
of the two sides of the unit cell that has length a while keeping c constant. c
11
+
c
12
+ 2c
33
4c
13
is calculated by lengthening a and simultaneously shortening c
(and vice versa). Finally, c
33
is obtained by varying the value of c. c
13
can then
be easily obtained from these values. The results for both potentials are shown in
Table 3.17.
It can be seen that the calculated values for both potentials for c
11
+ c
12
and
c
13
are quite close to the input values in the tting database. However, the value
for c
33
is 28 % to 33% lower than the input value.
Calculating c
11
c
12
, c
44
There are two more elastic constants to be calculated, c
12
and c
44
. It is not possible
to calculate c
12
directly, and it must be one by obtaining c
11
c
12
rst. Johnson
points out correctly that calculating c
11
c
12
and c
44
accurately requires an energy
minimization procedure [202]. This is because the strains required for these elastic
constants break the hexagonal symmetry of the crystal, thus exposing the atoms
to forces. The possibility of internal relaxation in the crystal due to a reduction
in symmetry is further explained in Appendix B. The calculations are done with
Table 3.18: Elastic Constants requiring relaxation. All values are given in GPa.
c
11
and c
12
are calculated from the combined elastic constants.
c
ij
P1 P2 DFT
c
11
c
12
negative 34.6 55.9
c
44
negative 34.3 58.5
c
11
NA 125.3 141.3
c
12
NA 90.8 85.7
Cell A, applying the same strain patterns too without any loss in the hexagonal
symmetry.
In order to calculate c
44
, a shear strain has to be applied to the unit cell,
including the cell vectors. If Cell A is used, the cell vectors will no longer be
perpendicular to each other. It is possible to handle a simulation box that is
non-orthogonal, but the program written for MD simulations did not include this
capability. This was because the vast majority of MD simulations are done with
an orthogonal simulation box.
It is observed that a shear strain applied to the B8 cell results in a orthorhombic
distortion of Cell B, with the cell vectors remaining perpendicular to each other.
This is illustrated in Figure 3.14. The sheared B8 unit cell is indicated in red lines
in part (d), superimposed over the original B8 lattice for contrast. It is noted that
the distance AB is reduced to AB while the longer diagonal BC is lengthened to
BC. The two sides remain perpendicular to each other.
Therefore, this indicates that the calculation of c
44
can be achieved without
making special provisions for non-orthogonal cell vectors by using Cell B. One
side of Cell B is shortened while the other side is lengthened by the same amount.
Since this destroys the hexagonal symmetry, relaxation of the internal atoms was
performed.
For both c
11
c
12
and c
44
, two sets of calculations are perform using a strain of
= 0.001. The resulting values are shown in Table 3.18. Together with c
11
+c
12
,
individual values of c
11
and c
12
can then be obtained.
(a) Original B8
unit cell
(c) B8 Unit cell under

shear strain
A
B
C
D
A'
B'
C'
D'
(b) Cell B
(d) Cell B under shear strain
Figure 3.14: Unit cell to calculate c
44
For the potential Set P1, the minimized energy for both elastic constants and
both values of strains were found to be lower than the cohesive energy by approx-
imately 0.0001 eV/atom. In other words, the relaxed value of c
11
c
12
and c
44
turns out to be negative.
The results for Set P2 are better, as both the elastic constants were found to
be positive after relaxation. c
11
was found to be 125.3 GPa, which is 12% lesser
than the DFT value, while that for c
12
is 6% higher than the DFT value. Thus,
they are close to the DFT values.
For c
44
, it is predicted to be 42% lower than the DFT value. This large dif-
ference is not surprising as c
44
was not included in the potential tting process.
Nevertheless, it is still predicted to be positive. This is one result in favour of Set
P2.
Energy = E
0
pbc
pbc
pbc
pbc
pbc
pbc
pbc
pbc
Surface.
Area =: A/2
Surface.
Area =: A/2
pbc
pbc
Legend:
A total exposed
surface area
pbc periodic
boundary
Energy = E
1
Figure 3.15: Illustrating the Surface Energy Calculation. Two simulation cells
of dierent boundary conditions are considered, one with fully periodic boundary
conditions, the other with a surface. The surface energy = (E
1
E
0
)/A
3.5.5 Surface Energy
The surface energy can be calculated with the typical three-dimensional box of
atoms. It is calculated by taking the dierence in energy between a system with a
surface E
1
, and a system with periodic boundary conditions in all three directions
E
0
, which is a multiple of the cohesive energy. This dierence is then divided by
the total exposed surface area A to obtain the surface energy. This is illustrated
in Figure 3.15 and expressed mathematically in Equation (3.50).
Exposing a surface means that the surface atoms will have forces on them.
However, for ease of calculation, no relaxation is performed.
E
surf
=
E
1
E
0
A
(3.50)
In order to be able to assess how good the predictions of the MD surface energy
are, the surface energy is also calculated by DFT. Since calculations in CASTEP
are carried out with the unit cell under periodic boundary conditions, a surface is
introduced by putting in a suciently large volume of empty space in the unit cell
between the atoms and the cell boundaries. This is so that the periodic images
c
nserted
vacuum
a
a
c
a
a
nserted
vacuum
(a) (b)
Figure 3.16: Unit cells used in the DFT calculation of the surface energy (a) surface
normal x (b) surface normal z. The height of the inserted vacuum is about 8

A.
of the surfaces do not interact with each other. These unit cells are illustrated in
Figure 3.16.
To obtain energies that are directly comparable with the unrelaxed surface
energies obtained with MD, no geometry optimization is performed with DFT.
Instead, the energy of the unit cell itself is calculated.
The reduction in the surface energy due to relaxation cannot be predicted in
advance. Therefore, to obtain the relaxed surface energy, geometry optimization
is then carried out. The cell vectors are xed, thus adjusting only the positions of
the atoms. For the DFT calculations, only the GGA is used with the Ultrane
setting provided by the CASTEP software.
The results are shown in Table 3.19. The unrelaxed MD surface energies are
positive, thus suggesting that the potential predicts that a system with a surface
is less stable than one with periodic boundaries. The unrelaxed energies are also
of the same order of magnitude compared to the DFT energies.
Table 3.19: Predicted surface energies. All values are in eV/
A
2
Surface Normal Set P1 Set P2 DFT
Unrelaxed
x 0.101 0.108 0.156
z 0.133 0.114 0.153
Relaxed
x 0.054 0.087 0.077
z 0.094 0.113 0.074
It is possible that after relaxation, the surface energy could turn negative. MD
calculations to obtain the relaxed surface energy were then conducted to see if this
possibility can be eliminated. In order to relax the surface, a quenching procedure
was used in the manner suggested by H akkinen et al. [204]. Whenever a force
component is in the opposite direction to the corresponding velocity component,
the velocity component itself is set to zero. This allows a minimum energy cong-
uration to be reached. An initial temperature of 100 K was given to the system
and the atoms were allowed to move according to Newtons Second Law for 1000
time steps before the quenching procedure was applied for a further 49000 time
steps. The minimum energy obtained in this time period is then used to calculate
the relaxed surface energy. The results are also compiled in Table 3.19.
From the results shown, the values of the surface energy remain positive after
relaxation, thus eliminating the possibility mentioned earlier. The relaxed values
are also rather close to the DFT results with P2 for x and P1 for z showing the
closest predictions.
3.5.6 Other Structures
Having compared the predictions of the surface energy, this section answers the
question of whether the MD potential will predict structures other than the Gan-
gulee structure to be of lower energy. This is important for the following reasons.
During the potential tting process, the elastic constants and the derivative
condition help to ensure that there is a local minimum. However, there is no
known way of guaranteeing that an interatomic potential will predict any specic
crystal structure to be the ground state. The only solution employed has been to
include other higher-energy structures in the tting database, such as defects and
other crystal structures. Therefore, checking the predictions of the interatomic
potential for other structures will lead to an understanding of its capabilities and
help in future potential tting eorts.
The cohesive energy of four other structures that exists for compounds of 1:1
stoichiometry are examined using the potential developed. These structures are
shown in Figure 3.17. They are compared with the cohesive energy of the B8
structure. If the potential truly predicts the B8 structure to be the ground state,
all these other structures should have a higher energy. The threshold accepting
method is used to search for the lattice constants that minimize the energy. They
are compared with DFT calculations using the GGA at ultrane setting. In both
cases, once the energy has been found, the dierence in energy per atom E is then
calculated. A positive dierence is a desirable outcome, while a negative dierence
suggests that the particular structure is more stable than the B8 structure of
Gangulee. The results are shown in Table 3.20.
From the results there are several similarities between the predictions from
potential P1 and P2. For the lattice constants, both potential have predictions
that are strikingly close to the DFT results, except P1 in the case of the B3(ZnS)
structure. Both P1 and P2 predict the B1(NaCl) structure to be less stable than
the B8 structure, though the energy dierence was one order of magnitude lesser
than the DFT prediction. The two potentials also predict the L10(CuAu) structure
to be more stable than the B8 structure.
What sets P1 and P2 apart are their predictions for the B3(ZnS) structure.
For P1, the threshold accepting calculation for the B3 structure did not terminate
when a minimum was found. Instead, it terminated when the lattice constants
CuSn_B1
Page 1 of 1
Printed by Materials Studio at 6:47:07 PM on 9/2/2008
(a) B1 (NaCl)
CuSn_B2
Page 1 of 1
(b) B2 (NiAl)
CuSn_B3
Page 1 of 1
(c) B3 (ZnS)
CuSn_L10
Page 1 of 1
(d) L10 (CuAu)
Figure 3.17: Unit cells for dierent structures with 1:1 stoichiometry. For each
structure, the Strukturbericht designation is given followed by the chemical formula
of a typical compound.
3.6. DISCUSSION 138
Table 3.20: Predicted Energies of other structures
Structure Method E (eV/atom) a (
A) c (
A)
B1 (NaCl) DFT +1.050 5.308 -
Set P1 +0.077 5.493 -
Set P2 +0.178 5.462 -
B2 (NiAl) DFT - - -
Set P1 -0.291 3.223 -
Set P2 -0.365 3.199 -
B3 (ZnS) DFT +0.550 5.767 -
Set P1 Very negative very small -
Set P2 +1.121 6.099 -
L10 (CuAu) DFT +1.071 4.218 3.860
Set P1 -0.278 4.261 3.672
Set P2 -0.457 4.310 3.747
The geometry optimization did not converge
No minimum was found.
resulted in the atoms being less than 1

A apart. The energy of the crystal was
very negative. In other words, the potential predicts that there is no metastable
conguration for the B3 structure and the atoms just collapse into each other if
placed in this conguration. In contrast, P2 makes similar predictions to DFT. It
predicts that the B3(ZnS) structure will be less stable than the B8 structure, and
the lattice constant is close to that of the DFT prediction. The predicted energy
dierence is 100% higher than that of the DFT value.
3.6 Discussion
3.6.1 Introduction
This chapter has described the process of obtaining an interatomic potential for
the B8 Cu-Sn intermetallic compound with the MEAM. The process involved
constructing the objective function, whose value measures how good a current set
of parameters is in reproducing a certain set of physical properties. The potential
parameters are then obtained by searching for the set of parameters that minimize
3.6. DISCUSSION 139
the objective function. From the tting process, two sets of potential parameters,
P1 and P2 were generated.
3.6.2 Choice of Potential Parameters
Since two sets of parameters were generated, P1 and P2, a choice must be made. It
is known that the elastic constants of a crystal depend on the interatomic potential
and the crystallographic symmetry. Max Born proved this result for a pair poten-
tial in 1940 [205], while Daw and Baskes derived a similar expression for the EAM
[148]. Therefore, reproducing the elastic constants of a material as accurately as
possible is important for an interatomic potential.
From the elastic constants predicted by P1 and P2, it is clear that P2 is the
better choice. All ve elastic constants, including c
11
c
12
and c
44
can be calculated
with P2 and all of them show positive values, whereas P1 predicted negative values
of these two constants. Using P2, c
12
and c
13
were closest to the values used in the
tting processing, being higher by 6.0 % and 8.0 % respectively. c
11
was found to
be 11.5 % lower while c
33
was lower by 27.8%. This is a reasonable outcome, given
that the elastic constants were not included directly in the objective function.
There was no guarantee during the tting process that c
44
would turn out to
be positive, as c
44
was not included in the objective function and. Hence it is not
surprising that P1 predicts c
44
to be negative. What is then remarkable is that P2
predicts c
44
to be positive.
These successes with the elastic constants for potential P2 suggest that it is
indeed feasible to construct an interatomic potential for the B8 crystal with the
MEAM.
3.6. DISCUSSION 140
3.6.3 Limitations
Since pairwise potentials were developed for metals in the 1960s, much progress
has been made in devising suitable functional forms to describe the various types
of bonding, such as the Stillinger-Weber potential for covalent bonding in silicon,
and the EAM and MEAM for metals. However, what has not changed is the
inherently arbitrary nature of the potential tting process, especially in the choice
of properties to reproduce and the choice of the various mathematical functions in
the functional forms.
The choice of properties included in the tting database is limited and this
necessarily limits the areas in which the potential is applicable. To expand this,
the size of the tting database can be increased by increasing the number of ad-
justable parameters, either by introducing new mathematical functions or using
cubic splines. However, this would increase the complexity of the optimization
process as the algorithm would have to sample a larger parameter space. Since
this is an initial study, the properties are limited to the fundamental properties of
any crystal, which are the elastic constants and the lattice constants.
The initial set of parameters used by Aguilar et al. [157] for the deposition of
Sn on Cu were put together from parameters designed for the metals individually.
In this study, the decision was taken to adjust all the parameters so that they are
optimized only for the properties of the B8 CuSn structure. None of the properties
of Cu and Sn were included in the tting database. The implication of this choice
is that the resulting potential generated cannot be used to simulate systems with
the individual atomic species, examples of which include a CuSn-Cu interface.
Although the objective function during the tting process includes constraints
on the rst derivatives and second derivatives of the energy (i.e. the elastic con-
stants), these merely guarantee a local minimum and have no bearing on whether
the B8 structure is a global minimum.
3.7. SUMMARY 141
Therefore, when other structures with 1:1 stoichiometry are considered (Figure
3.17), there is absolutely no guarantee that any of them would be predicted to be
less stable than the B8 structure. Hence, the results in Table 3.20, which show
the L10 structure and B2 structure as more stable than the B8 structure, are not
surprising. This implies that this MD potential cannot be used to predict the
enthalpy of formation for other Cu-Sn intermetallic compounds.
What is remarkable though is that the lattice constants for three of the four
structures considered were close to the DFT results. Also, the B1 structure and
B3 structure were predicted to be less stable. These positive results suggest the
possibility that in future renements of the tting process, the energies of these
structures can be included in the tting database.
Another remarkable point is that the surface energies calculated for both P1
and P2 were positive, even after relaxation and of similar order of magnitude to
DFT calculations. Since the surface energy is positive, this suggests the possibility
of performing MD simulations involving surfaces, such as fracture.
3.7 Summary
A MEAM interatomic potential for Cu-Sn designed for the B8 crystal has been
obtained. It has been shown that the MEAM potential for Cu-Sn published by
Aguilar et al. [157] cannot reproduce the properties of the B8 CuSn compound
that were calculated by DFT. This chapter describes the process of obtaining a new
interatomic potential to describe the interactions in the B8 intermetallic compound
CuSn, which is a simplied structure for Cu
6
Sn
5
. Two sets of parameters were
obtained, and after validation tests involving the elastic constants, surface energy
and other structures with 1:1 stoichiometry, the parameters in Table 3.12 were
chosen. This shows that it is feasible to use the MEAM to describe the interactions
in a B8 structure, which has hitherto not been demonstrated.
142
Chapter 4
MD Simulations of Fracture
4.1 Introduction
4.1.1 Motivation
The proper use of materials in many everyday situations requires a proper un-
derstanding of their strength properties, such as the yield stress and the fracture
toughness. This is to ensure that loads do not exceed what the material can
withstand, and structures and components do not fail catastrophically.
It has been discussed in Chapter 1 that a layer of intermetallic compound exists
within a solder joint (Figure 1.1), and that failure of the joint often occurs at this
layer. Computer modelling of the solder joint could be more eective and useful
in design if accurate fracture properties of the components in the solder joint are
available, including the intermetallic compound. Unfortunately, for Cu
6
Sn
5
, such
information is sparse and shows little agreement.
It now seems that atomic-level calculations can go some way towards comple-
menting experimental data. Since DFT tends to provide more accurate material
properties, it would be ideal if calculations using this method could be performed
to provide some predictions of the fracture toughness. However, due to its high
computational cost, less accurate methods such as MD have to be employed.
An interatomic potential for Cu-Sn interactions has been developed in Chapter
3. This chapter will describe how this potential is used in MD simulations to
predict the fracture properties of Cu
6
Sn
5
.
4.1.2 Experimental studies of the fracture toughness of
Cu
6
Sn
5
There have been some studies investigating the mode I fracture toughness (here-
after known as fracture toughness) K
IC
of Cu
6
Sn
5
, and their results are sum-
marized in Table 4.1. Three of the studies indicate a K
IC
in the range of 2.3 -
2.8 MPa
m. These ve studies employ two methods, the buckling of thin lms

and Vickers indentation. Given that there are only ve experimental studies, no
denitive conclusion can yet be made regarding the actual value of K
IC
.
In the buckling of thin lm method used by Balakrisnan [206] and Chen et
al. [207], the intemetallic compound Cu
6
Sn
5
is formed by sputtering atomic Cu
and Sn in the ratio of 6:5 on a polymer substrate, which is then annealed to
remove residual stresses and to ensure that the intermetallic compound is formed
completely. The sample is then subjected to a compressive load that is gradually
increased until the sample buckles. The strain at which the buckling occurs is then
used to calculate the critical energy release rate using bimaterial fracture theory,
from which the fracture toughness K
IC
can be obtained.
The Vickers indentation method to obtain the fracture toughness is semi-
empirical. The surface of the sample is rst polished. The surface is then indented
by a pyramidal indenter. During this process, a load is applied to the indenter.
Usually, cracks are seen to form from the corners of the indentation. This is shown
schematically in Figure 4.1. The lengths of these cracks are measured, and then
the the fracture toughness is calculated by one of several empirical calibration
Figure 4.1: Schematic diagram of cracks forming from the corners of the Vickers
indentation. The crack length C is measured from the centre of the indentation.
Table 4.1: Experimental data for the fracture toughness K
IC
of Cu
6
Sn
5
Author Method Results (MPa
m)
Chen et al., Ref [207] buckling of thin lm 2.63 0.15
Ghosh, Ref [42] Vickers indentation several tests, 2.43 - 2.80
Balakrisnan et al., Ref [206] buckling of thin lm 2.36 0.15
Kang et al., Ref [209] Vickers indentation 0.48 0.09
Fields et al., Ref [19] Vickers indentation 1.4 0.3
equations available. In all the Vickers indentation studies described in Table 4.1,
the authors used the following equation developed by Anstis et al. [208]
K
IC
= 0.016
_
E
H
__
F
C
1.5
_
(4.1)
where E is the Youngs Modulus, H is the hardness, F is the indentation load
and C is the crack length measured from the centre of the indentation (Figure 4.1).
Because this equation is empirical, the resulting value of the fracture toughness is
at best considered an estimate.
4.1.3 Overview
Having examined the experimental data currently available for the fracture tough-
ness of Cu
6
Sn
5
, existing work done using MD to calculate the fracture toughness
of various materials will be reviewed in section 4.2. Results using two methods
of calculating the fracture toughness will then be described. Section 4.4 describes
the outcome of using a periodic array of cracks, while section 4.5 shows how the
calculations are done using a crack-tip displacement eld. These results will be
discussed in section 4.6, and the main ndings of this chapter will be summarized
in section 4.7.
4.2.1 Introduction
In this section, literature regarding the molecular dynamics simulation of fracture
will be reviewed. The fracture of materials ultimately involves the breaking of
interatomic bonds to enable the crack to propagate. In addition, processes at the
atomic level are also involved, such as the formation of dislocations and voids.
MD simulation, being a technique at the atomic level, readily lends itself to the
simulation of such phenomena.
A review of the studies on metals and semiconductors will rst be conducted.
Such studies are often instructive in understanding the various types of computa-
tional techniques and the eects of the choice of the interatomic potential. This is
followed by a review of studies with intermetallic compounds. Due to their unique
crystal structures, this presents a dierent set of challenges. Finally, this section
concludes with a summary of the main points.
4.2.2 Fracture of Metals and Semiconductors
The potential for MD simulation to study fracture has been recognized as early
as 1983, when Decelis et al. studied the fracture of -Fe and Cu using pairwise
potentials [210]. The simulations showed a cleavage fracture for -Fe and the
blunting of the crack tip for Cu, thus corroborating what is known about the
typical fracture behaviour of these materials.
At that time, the inadequacy of such potentials in describing metals was al-
ready known, leading to the development of the EAM. Using an EAM potential
for Al, Hoagland et al. built three models with the atoms arranged in dierent
crystallographic orientations [211]. A mode I crack was introduced by means of
the anisotropic crack-tip displacement eld given by Sih et al. [212]. They found
that not all orientations showed crack propagation, as one orientation showed a
blunting of the crack tip instead. A similar study for Ni was conducted by Daw et
al., who investigated the eect of mixed mode loading [213].
In recent years, with the rapid improvements in computing technology, it has
been possible to conduct MD simulations with millions of atoms. One example
is the study by Latapie et al., who conducted simulations of the fracture of -Fe
with up to two million atoms [214]. With such a large number of atoms, it was
possible to introduce grains. The simulations showed that the crack propagated
across grain boundaries, and void coalescence was identied as a mechanism for
this phenomenon.
Perhaps the material that has attracted the most interest in atomistic studies of
fracture is silicon due to its brittle nature and technological importance. A notable
series of studies by Holland and coworkers investigated the dynamic fracture of
silicon by conducting MD simulations [152, 215] and designing an experiment that
was very close to the MD simulation conditions [216]. They showed that the
Stillinger-Weber potential had to be modied in order for the MD simulations to
show brittle behaviour. Even with the modication, their experimental results for
the minimum energy for crack propagation did not agree with the MD simulations.
Later, Swadener et al. discovered that the MEAM potential for silicon gave brittle
behaviour and better quantitative agreement with the experimental results [217].
These results show the importance of understanding the properties of the in-
teratomic potential chosen for any MD simulation. Kang et al investigated the
fracture of silicon nanowires using MD simulations with ve dierent potentials
for silicon [61]. They found that not all potentials gave brittle behaviour, and the
MEAM potential was found to be the most realistic. Calculations of the fracture
toughness K
IC
using a crack-tip displacement eld with three dierent potentials
also showed the same conclusion regarding the MEAM potential [218].
So far, most MD studies calculating the fracture toughness have used the crack
tip displacement eld as a starting condition. One study involving silicon with the
MEAM potential used a rectangular simulation box with a notch [219]. Fujii et
al. calculated the fracture toughness based on the stress at the onset of crack
propagation using the formula
K = F
t (4.2)
where t is the length of the notch and F is a correction factor. This is not a
unique procedure, as several similar formulas for various geometries are found in
the handbook by Tada et al. [220].
4.2.3 Fracture of Intermetallic Compounds
In comparison to that for metals and semiconductors, MD studies on the fracture
of intermetallic compounds have been sparse.
One of the rst MD studies of the fracture of intermetallic compounds was
by Becquart et al. [139]. Using the crack tip displacement eld by Sih et al.,
they investigated the dependence of the K
IC
on temperature and crystallographic
orientation for RuAl and Nb
3
Al. They also observed the possible slip systems and
dislocation emission phenomena.
Also using the Sih displacement eld, Dang et al. studied the crack-tip be-
haviour of Ti-V alloys at various compositions [221]. Apart from slip systems
and dislocation phenomena, they reported phase transformations occurring in the
region behind the crack tip. Using the crack data, they suggested an optimal
composition for the alloy to maximize the toughness.
Hirosaki et al. predicted the mode I fracture toughness of Si
3
N
4
[222], moti-
vated by the diculty of obtaining experimental results for single-crystal Si
3
N
4
.
While other studies used the anisotropic set of equations, in their MD study, they
used the crack tip displacement eld for an isotropic solid to create a crack. This
would represent an approximation to the actual anisotropic displacement eld.
R osch et al. used Lennard-Jones type potentials to study the brittle fracture
of the intermetallic compound Al-Mn-Pd [223], which shows a unique quasicrystal
arrangement. Using a rectangular geometry with a layer of atoms removed, the
MD simulations of the crack propagation showed a brittle behaviour that did not
come with any emission of dislocations. The study also claried the role of internal
clusters of atoms in determining the crack propagation path.
4.2.4 Conclusions
The value of MD simulations to study the mechanical behaviour of materials has
been recognized [224]. The advantage is primarily in viewing on the computer what
is dicult or almost impossible to determine experimentally. This review has given
many examples of this, such as the mechanism of fracture across grains in -Fe,
several studies in calculation the fracture toughness and the phase transformations
that occur in the Ti-V alloy system.
It has also been shown that not all interatomic potentials for the same material
show the right mechanical behaviour. This is certainly due to the empirical manner
in which potentials are developed. Hence, while MD simulations can give valuable
qualitative information, this pitfall must always be borne in mind.
4.3. SIMULATION PROCEDURE 149
4.3 Simulation Procedure
4.3.1 Interatomic Potential
The interatomic potential developed in Chapter 3 that is tted to the hexagonal
NiAs structure reported by Gangulee et al. [109] is used. The interstitial sites are
not lled to obtain the 6:5 stoichiometry, thus the ratio of the Cu atoms to Sn
atoms is 1:1.
4.3.2 Visualization
Atoms will be represented in two ways, spheres and circles. Spheres give a good
view of the three-dimensional arrangement of the atoms, but do not show how the
layers of atoms are stacked. Unlled circles can show such information and show
clearly how the arrangement of atoms has evolved. In this chapter, larger yellow
spheres and green circles will represent Sn atoms, while smaller blue spheres and
blue circles will represent Cu atoms. Information on how the atoms have moved
to their present position will be shown by tails, which are lines beginning at the
atoms initial positions and ending at their current positions.
4.3.3 MD Software
Many research groups have made their own in-house MD software publicly avail-
able, such as XMD [225] and LAMMPS [226]. However, none of these software
supported the MEAM potential at the beginning of this project. The MEAM
functionality was implemented in LAMMPS only in 2007. As such, the decision
was taken to develop an in-house code. No comparison or benchmarking with
LAMMPS was performed.
Developing an in-house code has several advantages. Intimate familiarity with
the MD algorithm and techniques can be gained during the programming process.
The code can easily be modied as the author is the designer of the code itself.
The following paragraphs describe the essential features of the software.
The inputs to the MD simulation has to be organized in an eective and user-
friendly manner. This is done through an input le using descriptive keywords,
similar to that in XMD.
The atoms are generated by the user specifying the basis and the lattice vec-
tors. Currently, the code only supports a cubic basis. A feature that allows atom
to ll non-cubic volumes in space, such as a square pyramid, hemisphere was im-
plemented to enable nanoindentation simulations to be conducted in future.
In simulations like nanoindentation and the tensile test, blocks of atoms need
to be displaced at time intervals. The user rst denes the dierent sets of atoms
in the simulation. Then, for each set, the user can specify all the points in time
and the amount of displacement in each coordinate direction that the set of atoms
is to be moved.
Both the Velocity-Verlet method [227] and the Gear Fourth-Order predictor-
corrector method [228] to numerically integrate Newtons second law were imple-
mented. To enable calculations that involve temperature control, the Nose-Hoover
thermostat method [229] and the velocity-scaling method proposed by Woodcock
[230] were included.
E = (MLT
2
) (L) T =
_
L
2
M
E
(4.3)
Atomic-scale units were used in the coding. The unit of energy was xed as the
electron-Volt (eV), the unit of length was xed as the angstrom (
A) and the unit of

mass was xed as the atomic mass unit (amu). One time unit (tu) can be worked
according to Equation (4.3) and is equivalent to 1.01805 10
14
s. Equivalent
units of force and pressure can also be derived and they are listed in Table 4.2.
The value of the Boltzmann constant, necessary for converting kinetic energy into
Table 4.2: Atomic-Scale units used in the software
Quantity Atomic-Scale unit SI equivalent
Energy (E) eV 1.6021764610
19
J
Mass (M) amu 1.6605388610
27
kg
Length (L)

A 1 10
10
m
Force (F) eV/
A 1.602 10
9
N
Pressure(P) eV/
A
3
1.602 10
11
Pa
Time (T) tu 1.01805 10
14
s
temperature, in the atomic scale units is then 8.617343 10
5
eV/K [142].
The program is written in a serial manner and does not take advantage of
the availability of dual CPUs. When development work began, Intel had not yet
introduced hyperthreading, a feature that enabled a single CPU to process two
sets of instructions simultaneously, into its mass-market Pentium 4 processors.
Finally, the standard techniques to speed up MD calculations were imple-
mented. The function les are stored in a tabulated form with evenly-spaced
nodes. A neighbour-list is maintained for each atom. For the MEAM and EAM,
quantities calculated in the rst for-loop which are repeated in the second for-loop
are stored. The organization of the program is shown in Figure 4.2.
The most critical component of the MD program is the subroutine that calcu-
lates the forces. In order to check that the forces have been programmed correctly,
the forces are estimated by a nite dierence method and compared with the calcu-
lated forces. The nite dierence procedure is as follows. To calculate the force in
the x direction, an atom is displaced a small displacement dx in both the positive
and negative directions and the energy of the entire system for each displacement
E
up
, E
dn
is correspondingly recorded. Then the force on the atom is
f
x
=
E
x
i

E
up
E
dn
2dx
(4.4)
Figure 4.2: Flowchart showing the organization of RidgeMD.
basal plane
Unit Cell
a
a
60
a
a
a
c
ac plane
Figure 4.3: Description of the basal plane and the ac plane.
4.3.4 Unit Cell
It is more convenient to use the rectangular form Cell A of the NiAs unit cell
that is specied in Section 3.5.2. This unit cell is orthogonal with cell lengths
a a
3 c, where a = 4.1189

A and c = 5.1095

A. Hence, all references to unit
cells in this chapter will be to this particular form of the unit cell.
The arrangement of the unit cells to form the sample for MD simulation shall
be known as the simulation box. Hence the size of the simulation box can both
refer to the dimensions of the sample or the number of unit cell lengths on each
side.
For the simulations in sections 4.4 and 4.5, two orientations for the crack will
be used. They are the basal plane and the ac plane. The ac plane is formed
by the a-axis and c-axis. This is illustrated in Figure 4.3.
4.4. K
IC
CALCULATION USING THE TENSILE LOADING ON A
PERIODIC CRACK 154
Figure 4.4: Periodic Crack

4.4 K
IC
calculation using the tensile loading on
a periodic crack
4.4.1 Method
A periodic array of cracks is illustrated in Figure 4.4. For this situation, an ana-
lytical solution for the fracture toughness for a given level of stress is described by
Tada et al. [220]. K
I
is given by
K
I
=
_
W tan
a
W
(4.5)
where a/W is the ratio of the crack half-length to the width of the sample and
is the far-eld normal stress. This expression suggests that for the same a/W
ratio, a wider sample (i.e. larger W) requires a lower stress to fracture.
A periodic crack can be studied using MD simulations as periodic boundary
conditions can be used. A schematic diagram is given in Figure 4.5. Firstly, a
perfect crystal is assembled. A pre-crack is created by removing a n 1 row of
4.4. K
IC
PERIODIC CRACK 155
pbc
pbc
mmobile atoms
(one unit cell thick)
Atoms free to
move
mmobile atoms
(one unit cell thick)
pre-crack is
created by
removing unit
cells across the
thickness
aa
c
Figure 4.5: Schematic diagram of the periodic crack simulation box. The side n
of the pre-crack is parallel to the c-axis.
unit cells through the thickness of the block. The size of n varies according to the
requirements of the simulation, as the dimensions of the block. Comparing this to
Figure 4.4, the half-crack length a in Equation (4.5) is given by n/2.
Periodic boundary conditions (pbc) is applied in the direction that is parallel
to the crack, thus creating a periodic array of cracks. Pbc is also applied to the
direction perpendicular to the plane of the crack. The boundary condition in the
remaining direction remains free. One layer of atoms on either side of the block of
atoms are prescribed to have zero velocity throughout the simulation.
A strain in the direction perpendicular to the crack is applied to the entire
block of atoms. Hence, the coordinate of every atom is multiplied by (1 +). This
step completes the specication of the starting conguration of the atoms.
The atoms are then allowed to move according to Newtons second law. The
temperature of the system is maintained at 300 K by scaling the velocities of
the atoms using the velocity scaling method proposed by Woodcock [230]. This
method directly scales the velocities at each time step to achieve the desired sys-
tem temperature. Time integration was carried out using the Gear Fourth-Order
method with a time increment of t = 0.1 tu, which is approximately 1 fs.
Several simulations are done with dierent starting values of . Typically, at
4.4. K
IC
PERIODIC CRACK 156
Table 4.3: Description of the size of the simulation boxes used in this section. The
pre-crack size and box dimensions are given in terms of unit cell lengths.
Label Pre-crack size Box Dimensions Number of atoms a/W
PC1 1 1 21 2 13 4352 0.039
PC2 2 1 21 2 14 4672 0.071
PC2L 3 1 31 2 21 10368 0.071
PC3 3 1 21 2 13 4300 0.115
PC4 4 1 29 2 18 8288 0.111
low values, the crack remains stable. The strain value at which the crack is seen
to propagate is recorded. At this point, the stress can then be identied and the
K
IC
then identied using Equation (4.5).
4.4.2 Results
For the simulations in the ac plane, the number of unit cells in the c-direction
determines the value of W. Various box sizes are used and they are described in
Table 4.3.
Initially, it was found that creating a pre-crack of only one unit cell size i.e. case
PC1 is insucient. After applying the strain and allowing the atoms to move, no
crack propagated from the initial pre-crack. Instead, the system fractured between
the immobile layers and the atoms that were free to move. This is illustrated in
Figure 4.6. Hence, further simulations with pre-cracks of 2 1 and 3 1 unit
cells i.e. cases PC2 and PC3 were carried out instead.
For case PC2 (2 1 pre-crack), it was found that the crack propagated only
at a strain of = 0.20. The outcome is shown in Figure 4.7. Simulations with a
lower values of strains merely showed the pre-crack relaxing without any signs of
crack propagating.
For the 3 1 pre-crack (case PC3), it was found that the crack propagated
at a strain of = 0.12. This is lower than for the 2 1 case (case PC2), and is
4.4. K
IC
PERIODIC CRACK 157
mmobile atoms
mmobile atoms
1 1 Pre-crack
a
c
Figure 4.6: Conguration PC1. Snapshot at 20000 time steps with an initial strain
of = 0.20 applied throughout. The sample cleaves between the immobile layer
and the mobile atoms and no crack propagates from the pre-crack.

Figure 4.7: Conguration PC2 with a initial strain of = 0.20 applied throughout.
Snapshots at (a) 6500 time steps (b) 20000 time steps
4.4. K
IC
PERIODIC CRACK 158
an expected result as a larger pre-crack means a weaker sample. The outcome is
shown in Figure 4.8. As before, simulations with a lower values of strains merely
showed the atoms on the surfaces created by the pre-crack relaxing without any
signs of crack propagating. For this case, the snapshot prior to fracture is plotted
again by representing atoms with circles and shown in Figure 4.9. The plane at
which the crack occurs can be clearly seen, and the fracture can be seen to be a
clean separation of planes of atoms.
In both cases, the crack propagates in the same direction as the crack and a
relatively at fracture surface is seen. There is little sign of plastic deformation or
phase transformation, as the diagrams do not show any change in the arrangement
of atoms.
Considering the periodic boundary conditions imposed on these simulations, it
is akin to a series of voids that coalesce into a crack upon loading. Hence, it is to
be expected that the crack will travel horizontally to join the next void and this
is exactly what is seen in the simulations.
As mentioned earlier, Tadas expression predicts that, for the same a/W ratio,
a larger W would result in a smaller needed to cause fracture. Using the data
for PC3 and PC4 from Table 4.3, the fracture stress is predicted to be 86.6%
lower. Assuming that a linear relationship between stress and strain is valid at
this point, this suggests that the failure strain for PC4 will be = 0.104. Hence,
three simulations were conducted at strains of = 0.09, 0.10, 0.11 for size PC4.
Fracture was seen for = 0.10, 0.11 but not seen for = 0.09, thus conrming the
prediction by the equation.
The fracture strain for case PC2 is rather large compared to cases PC3 and
PC4. To examine the size eect, simulations for case PC2L were performed. The
pre-crack size is 3 1 i.e. 1.5 times longer than that of PC2. The value of W is
increased by the same proportion, thus keeping the same a/W ratio. The fracture
strain for PC2L was found to be = 0.12, which leads to a smaller fracture stress
4.4. K
IC
PERIODIC CRACK 159
Figure 4.8: Conguration PC3 with an initial strain of = 0.12 applied through-
out. Snapshots at (a) 1200 time steps (b) 20000 time steps
Figure 4.9: Conguration PC3. Snapshot at 1200 time steps representing the
atoms as circles. All the atoms are superimposed on each other. The plane at
which the crack propagates can be clearly seen.
4.4. K
IC
PERIODIC CRACK 160
Table 4.4: Values of stress and predicted K
IC
for the two cases, assuming that
= c
11
Sample a/W (GPa) K

I
(MPa
m)
PC2 0.071 0.20 25 1.01
PC2L 0.071 0.12 15 0.741
PC3 0.115 0.12 15 0.752
PC4 0.111 0.10 12.5 0.722
(15 GPa) compared to PC2 (25 GPa). This suggests that the box size used for
PC2 was insucient to achieve a successful crack simulation with a/W = 0.071.
All these results are summarized in Table 4.4.
4.4.3 Calculating the fracture toughness
Once the fracture strain has been obtained, to calculate the critical fracture tough-
ness, the stress in each of the three cases must be obtained.
A straightforward way to estimate the stress is to take the failure strain and
multiply it by c
11
, observing that the strains in other directions are zero. This
makes the assumption that the far-eld stress required to cause failure stress is
similar to that of a perfect sample without any void. The predicted critical fracture
toughness is summarized in Table 4.4.
It is seen that the results for PC2L, PC3 and PC4 all agree with each other,
while the result for PC2 is higher. This suggests that a pre-crack size of 2 1 is
not sucient to conduct a successful periodic crack simulation.
4.4.4 Summary
Using MD simulations at a temperature of 300 K, the fracture toughness has been
estimated using a periodic array of cracks and Tadas equation given in Equation
(4.5). From these simulations, the failure strain was found and the failure stress
estimated. With the failure stress, using Tadas solution given in Equation (4.5),
4.5. K
IC
CALCULATION USING A CRACK-TIP
DISPLACEMENT FIELD 161
Figure 4.10: (a) Array of atoms (b) The simulation box after Sihs equation is
applied to the atoms
the critical fracture toughness was estimated to lie in the range of 0.72 - 0.75
MPa
m for a crack lying in the ac plane.

4.5 K
IC
calculation using a crack-tip
displacement eld
4.5.1 Method
Having calculated the critical fracture toughness using the periodic crack solution
by Tada [220], another method of calculating it is explored, using the crack-tip
displacement eld solution by Sih et al. [212]. This method of calculating the
fracture toughness was also used by Becquart et al. [139].
In this method, an array of atoms is formed by building a block of unit cells.
One layer of atoms on either side of the array are prescribed to have zero velocity
throughout the simulation. This is illustrated in Figure 4.10(a). Unlike the study
4.5. K
IC
by Becquart, these zero-velocity layers are not displaced during the simulation.
The only function of this layer is to preserve the shape of the sample of atoms,
and they play no further role in the simulation.
A crack is then inserted by displacing all the atoms, including the immobile
atoms, according to the crack-tip displacement eld for an anisotropic material
[212] for a given value of K
I
. The crack-tip displacement eld is then applied
using the following set of equations. For each atom the displacement u
x
and u
y
are given by
u
x
= K
I
(2r)
1
2
Re
_
1
s
1
s
2
[s
1
p
2
(cos +s
2
sin )
1
2
s
2
p
1
(cos +s
1
sin )
1
2
]
_
(4.6a)
u
y
= K
I
(2r)
1
2
Re
_
1
s
1
s
2
[s
1
q
2
(cos +s
2
sin )
1
2
s
2
q
1
(cos +s
1
sin )
1
2
]
_
(4.6b)
where r and are polar coordinates of the atoms.
To obtain the variables s
1
, s
2
in Equation 4.6, the roots of the following char-
acteristic equation are taken, where s is the variable.
S
11
s
4
S
16
s
3
+ (2S
12
+S
66
)s
2
2S
26
s +S
22
= 0 (4.7)
The coecients are the elastic compliances corrected for plane strain, i.e.
S
ij
= s
ij
s
i2
s
i3
s
33
(4.8)
Solving Equation (4.7) results in four complex roots, s
1
, s
1
, s
2
, s
2
. To obtain
the variables p
1
, p
2
, q
1
, q
2
, the following expressions are applied
4.5. K
IC
p
1
= S
11
s
2
1
+S
12
S
16
s
1
(4.9a)
p
2
= S
11
s
2
2
+S
12
S
16
s
2
(4.9b)
q
1
=
S
12
s
2
1
+S
22
S
26
s
1
s
1
(4.9c)
q
2
=
S
12
s
2
2
+S
22
S
26
s
2
s
2
(4.9d)
The resulting conguration is shown in Figure 4.10(b). Once the crack tip dis-
placement eld has been obtained, boundary conditions can be decided. Periodic
boundary conditions are applied only in the direction of the short side of the block,
so as to mimic an innitely thick block of atoms. The boundary conditions in the
other directions remain free. This step completes the specication of the initial
conguration of the atoms.
After all the initial conditions are specied, the atoms are then allowed to
move according to Newtons Second Law for 20000 time steps, during which the
behaviour of the crack is observed. During this period, the layer of zero-velocity
atoms acts as a boundary to prevent the upper layers of mobile atoms from moving
too far away. The temperature of the system is maintained at 300 K by Wood-
cocks velocity-scaling method. The Gear Fourth-Order method was used for time
integration with a time increment of t = 0.1 tu.
Several separate simulations are performed with dierent displacement elds
that correspond to dierent values of K
I
. When the crack is seen to propagate,
this enables the value of K
IC
to be identied. The simulations are performed in
two planes, the basal plane and the ac plane, at a temperature of 300 K xed by
the velocity-scaling method. The unit cell that is specied in Section 3.5.2 is used.
4.5. K
IC
Table 4.5: Dierent box sizes used in the study in this section
Label Unit cells Approximate lengths Number of atoms
Small 17 2 14 70

A 14

A 70

A 3808
Large 24 2 19 100

A 14

A 100

A 7296
4.5.2 ac plane
Two types of box sizes were used in the simulations described in this section. For
ease of description, the box sizes are either called Small or Large and their
dimensions are given in Table 4.5.
Using size Small, simulations were conducted at various values of K
I
and the
crack was found to propagate at K
I
= 0.80 MPa
m. The results for K

I
=
0.40, 0.64, 0.80 MPa
m will be shown in this section. Figure 4.11(a) shows the

outcome when an initial displacement according to K
I
= 0.4 MPa
m is applied.
In this case, the crack did not propagate, but some signs of the crack branching
into two directions can be seen. The atoms behind the crack is seen to rearrange
themselves into a cubic arrangement.
Starting from K
I
= 0.64 MPa
m, the branched crack is seen to propagate but

does not cleave the sample. This is shown in Figure 4.11(b). The crack is seen to
blunt when it stops.
With an initial displacement of K
I
= 0.80 MPa
m, the two branched cracks

are seen and the lower crack propagates across the sample to cleave it into two.
This is shown in Figure 4.12(a). The upper crack does not propagate, thus leaving
the portion of the atoms behind the crack to swing about.
In MD simulations, it is possible that the limited system size can inuence
the phenomena that are displayed. For example, the crack could have propagated
through because of the short distance between the crack tip and the free surface
on the rear.
In order to examine the eect of size on this phenomenon, the simulation for
4.5. K
IC
K
I
= 0.80 MPa
m was repeated using size Large, shown in Figure 4.12(b). The

crack branching is still seen, as well as the phase transformation. In this case, since
there is more material behind the crack, the material behind the crack remains
relatively xed.
To go deeper into the phenomenon shown in this simulation, the atoms in
Figure 4.12(b) are represented using circles instead of spheres. This is shown in
Figure 4.13(a). It can be seen where the atoms retain their ordered arrangement,
and where the atoms lose it. The region where the atoms show a cubic-type
arrangement is also clearly seen.
Figure 4.13(b) shows tails, which are lines indicating how each atom has
travelled from its original position. This gives an indication of how the cracks
form, and whether there are any slip planes.
A slip plane can be identied by two adjacent regions of atoms where the tails
point in dierent directions. From the gure, no slip planes can be seen as all the
tails indicate that the atoms are relaxing inwards towards the immobile atoms.
The tails also indicate that the lower branched crack is a result of two layers of
atoms travelling apart, while the upper crack is formed by just one layer of atoms
moving inwards. This thus explains why the upper crack does not propagate.
The critical fracture toughness predicted from section 4.4 was in the range of
0.72 - 0.75 MPa
m. Using a conguration with starting K

I
= 0.72 MPa
m, the
branched cracks were seen but did not propagate across the sample.
4.5.3 Basal plane
Two types of box sizes were used in the simulations described in this section. For
ease of description, the box sizes are either called Small or Large and their
dimensions are given in Table 4.5.
Similar to the previous section, simulations were run at various values of K
I
4.5. K
IC
(a) (b)
Figure 4.11: ac plane: Snapshots of atomic conguration for size Small at 20000
I
= 0.40 MPa
m(b) K
I
= 0.64 MPa
m
Table 4.6: Dierent box sizes used in the study in this section
Label Unit cells Approximate lengths Number of atoms
Small 17 10 3 70

A 14

A 70

A 4080
Large 24 14 3 100

A 14

A 100

A 8064
4.5. K
IC
(a) (b)
Figure 4.12: ac plane: Snapshots of atomic conguration at 20000 time steps at
300 K for K
I
= 0.80 MPa
m. (a) Size Small (b) Size Large.

(a) (b)
Figure 4.13: ac plane: Size Large, K
I
= 0.80 MPa

the atoms. (a) without tails (b) with tails.
4.5. K
IC
and crack propagation was seen from K
I
= 0.64 MPa
m onwards. Results for

K
I
= 0.48, 0.64 MPa
m will be presented.
The snapshot at 20000 time steps for K
I
= 0.48 MPa
m is shown in Figure
4.14(a). The crack is seen to move for a short distance before blunting. No crack
branching is seen compared to the previous results.
For K
I
= 0.64 MPa
m, the crack propagates but stops short of the free surface,

as illustrated in Figure 4.14(b). When the box of size Large was used at the same
value of K
I
, some interesting phenomena could be seen. A void rst forms behind
the crack tip, before the crack nally propagates through it and cleaves the sample
into two. Regions of plastic deformation can be seen on both sides of the crack
surface. This is illustrated in Figure 4.15.
A representation using circles is shown in Figure 4.16(a). From the arrangement
of the atoms, the ordered arrangement is retained in most places. At the region
where the crack propagates through, the atoms show a disordered arrangement.
The tails are attached to the atoms in Figure 4.16(b). It is no surprise that the
tails are longest at where the crack propagates through. The atoms naturally have
to move a longer distance. They indicate a clean separation of the crack surfaces
without any dislocations or slips occurring.
4.5.4 Summary
MD simulations have been performed to predict the critical fracture toughness
using the crack-tip displacement eld method. The calculations suggest that in
the basal plane, the K
IC
= 0.64 MPa
m, and in the ac plane, the K

IC
= 0.80
MPa
m. Some of the phenomena that was seen include the formation of a void
behind the crack tip, and branched cracks. In all the simulations, no dislocations
were seen.
4.5. K
IC
(a) (b)
Figure 4.14: Basal plane: snapshots of atomic conguration for size Small at 20000
I
= 0.48 MPa
m(b) K
I
= 0.64 MPa
m.

Figure 4.15: Basal plane: snapshots of atomic conguration for Size Large at 20000
time steps at 300 K for various values of K
I
= 0.64 MPa
m. (a) 3100 time steps

(b) 20000 time steps
4.6. DISCUSSION 170
(a) (b)
Figure 4.16: Basal plane: Size Large, K
I
= 0.64 MPa

the atoms. (a) without tails (b) with tails.
4.6 Discussion
4.6.1 Comparing the two simulation methods
Using the potential for Cu-Sn developed in Chapter 3, fracture simulations have
been conducted using two methods. Both a periodic crack conguration and a
crack-tip displacement eld have been used. From these methods, predictions
have been made regarding the critical fracture toughness, and it is seen that both
methods do not produce identical values.
It is an expected result that the two methods do not produce identical values.
Due to the periodic boundaries in the periodic crack simulation, the crack that
exits one side of the box has to enter at the same position on the other side. In
contrast, there are no constraints of this kind in the crack-tip displacement eld
method. That is why branched cracks are seen in the latter method.
For the ac plane, both methods have produced rather similar values for the
4.6. DISCUSSION 171
predicted critical fracture toughness. The value of K
IC
= 0.80 MPa
m obtained
from the crack-tip displacement eld is not too far away from the range of 0.72 -
0.75 MPa
m obtained by the periodic crack method. Combining the results of

the two methods, it can be said that the critical fracture toughness is in the range
of 0.72 - 0.80 MPa
m.
4.6.2 Comparison with experimental results
When comparing this range to the experimental results in Table 4.1, it is on the
lower end of the range of results. The results summarized in this table are too
widely spread and sparse to indicate any denitive experimental value. It is seen
that the buckling of thin lm results tends to produce higher values while the
Vickers indentation (VI) method produces results across the range of values.
There seems to be some consistency among the two results using the buckling
of thin lm method. The two values obtained are 2.63 0.15 MPa
m(Ref [207])
and 2.36 0.15 MPa
m(Ref [206]). In the experimental study by Balakrisnan et

al. [206], the crack was seen to propagate between grain boundaries.
In the VI method, the crack lengths that result from the indentation are mea-
sured and then input into one of several empirical calibration equations available.
All the authors that use the VI method mention that their values are just estimates
[19, 209, 42]. Recently, doubt has been cast on the validity of the VI method in
measuring the critical fracture toughness. Quinn et al. argued that the VI test
does not produce a reliable fracture toughness value[231]. They conducted VI
tests on a standard silicon nitride sample whose K
IC
value had been proved to
be constant. They showed that dierent calibration equations resulted in dierent
predicted values of the K
IC
, all of which did not agree with the standard value.
Quinn et al. also pointed out that the behaviour of cracks from an indentation
typically shows complex behaviour that does not correspond to the behavior seen
4.6. DISCUSSION 172
in other fracture toughness testing methods. This was also observed by Ghosh
in his study of intermetallics. Very often, cracks would not propagate from the
corners of the indentation, or the cracks would branch after leaving the indentation
[42].
All the three VI method studies use the same calibration equation, but they
dier in how the sample was produced. The study by Fields et al. that used
bulk specimens made by powder metallurgy provides no information on whether
the indentation cracks occurred within a single grain [19]. Kang et al. used
a diusion couple to make the intermetallic layer and mentions that the cracks
occurred within a single grain [209]. Ghosh produced homogenous samples by
means of casting and annealing, and mentions that the cracks occurred within a
single grain [42]. Ghoshs results for K
IC
show a range of 2.43 - 2.80 MPa
m,
while Kangs result is the lowest at 0.48 MPa
m.
Therefore, the results obtained in this chapter (0.72 - 0.80 MPa
m) seem to
be closest to Kangs results. Kang also mentions that the crack propagated within
a single grain. This is similar to how the crack propagates within a single-crystal
sample shown in the simulations here. However, what prevents any rm conclusion
from being made is the fact that in light of Quinns study, it is not known how
accurate the calibration equation used is.
4.6.3 Qualitative features of the simulations
Having compared the experimental results with the values obtained from the sim-
ulation results, it is not known how accurate the values are. However, it is also
possible to examine the simulations qualitatively.
Both the methods have their unique features. In the crack-tip displacement
eld sample, the distance between the crack tip and the free surface at the rear
is very short. It was reported by Ghosh that the indentation cracks were arrested
4.6. DISCUSSION 173
by the grain boundary [42]. This phenomenon cannot be reproduced by this sim-
ulation as the sample is a perfect crystal without any grain boundaries. Thus, the
fracture toughness values obtained in the simulations are for single-crystal samples
that are perfect.
Branched cracks were seen in the fracture simulations using the crack-tip dis-
placement eld. Currently, no information from the experimental results are avail-
able to validate this phenomenon, as the thin lm experiments saw intergranular
cracking. It is therefore dicult to make direct comparisons between the numer-
ical results obtained here and the sparse experimental results that are currently
available.
The boundary conditions in the periodic crack simulations suggest the crack
propagating in a network of voids in a thin lm. Thus it may be argued that the
periodic crack simulations have some similarity to the bulk, as far as the presence
of voids is concerned. Typically, bulk intermetallic samples have some percentage
of porosity. For conguration PC3 and PC4, the porosity can calculated by taking
the ratio of the volume of the unit cells removed to the volume of the simulation
box, and the resulting value is about 1.1%. This value is comparable to the porosity
of 4.8% reported for a bulk sample of Ag
3
Sn made by hot-pressing[232].
4.6.4 How realistic is the interatomic potential?
Having considered the qualitative features of the simulation, the input that ul-
timately controls what the simulations display is the interatomic potential. The
interatomic potential was constructed empirically and is a rst attempt at obtain-
ing one that ts the properties of Cu-Sn intermetallic compounds. Using a limited
set of properties, it was tted to the NiAs unit cell reported by Gangulee et al.
[109], while the unit cell in the correct 6:5 stoichiometric ratio is given by Larsson
et al. [86].
4.7. SUMMARY 174
However, the simulation results have produced much evidence that the poten-
tial is realistic in some way. It has been already shown in Chapter 3 that the
surface energy is positive. In both types of simulations, no evidence of slip or dis-
locations which are signs of plastic deformation are seen as the crack propagated.
The fracture in the periodic crack simulations occurs cleanly between planes of
atoms. This ties in with the understanding that there is little plastic deformation
for brittle materials.
In other words, the interatomic potential indeed causes the simulations to show
behaviour typical of a brittle material. All this is encouraging evidence to under-
take further work in improving the interatomic potential.
4.7 Summary
Using the interatomic potential developed in Chapter 3, the critical fracture tough-
ness has been calculated by two methods, periodic cracks and the crack-tip dis-
placement eld. It was found that the value of K
IC
lies in a range of 0.72 - 0.80
MPa
m. This value cannot be validated by the experimental results yet. The

simulations show typical brittle behaviour, which corresponds to what is known
regarding Cu-Sn intermetallic compounds.
175
Chapter 5
Conclusions
5.1 Conclusions
Calculations at the atomic level have been performed in order to clarify conict-
ing experimental data for the mechanical properties of Cu-Sn intermetallic com-
pounds. Firstly, DFT calculations were employed to elucidate the elastic proper-
ties of Cu
6
Sn
5
and Cu
3
Sn. Following which, an MD potential was developed for
Cu
6
Sn
5
. This potential was then used to conduct MD simulations to understand
the fracture properties of Cu
6
Sn
5
.
It was shown that DFT calculations can yield accurate crystal properties for
Cu
6
Sn
5
and Cu
3
Sn. The calculated lattice constants and internal crystal parame-
ters showed good agreement with experimental results. This validates the accuracy
of DFT in predicting crystal properties for these intermetallic compounds.
Following which, DFT calculations were performed to obtain the single crystal
elastic constants for Cu
6
Sn
5
and Cu
3
Sn. These results are the rst set of values that
are available, as hitherto no experimental results exists. Using the values found,
it was found that the directions of the largest single-crystal Youngs Modulus
coincided with planes of closely-packed atoms. Even though the elastic constants
cannot be directly validated, this provides evidence of the consistency of these
5.1. CONCLUSIONS 176
values.
From these single crystal properties, bounds on the polycrystalline elastic mod-
uli were obtained using the method of Hill [102] and Hashin-Shtrikman [103]. It
was found that these bounds lie on the upper range of the experimental results.
These calculations involve perfect crystals and no defects are present. Hence, it can
be concluded that DFT calculations of the elastic properties with the Hill bounds
can provide the theoretical upper limit of the polycrystalline elastic properties of
these materials.
An interatomic potential tailored to the properties of Cu
6
Sn
5
in the NiAs struc-
ture for MD simulations was developed in Chapter 3 as the existing Cu-Sn potential
by Ravelo and Baskes [157] was found unsuitable. Using optimization methods,
parameters in the MEAM formulation were found that enabled the potential to
reproduce a database of material properties. This database was constructed by in-
cluding experimental data of the lattice constants and data of the elastic constants
obtained from DFT calculations. Two sets of parameters were obtained.
Tests were carried out on both sets of parameters to validate their properties
and to make a choice. The set of parameters that was nally chosen reproduced
the lattice constants and elastic constants that were initially included in the tting
database. The predictive power of the potential was then tested by calculating
the surface energy of two surfaces, properties that were not included in the tting
database. The energies were found to be positive and close to the DFT predictions.
This is sucient evidence to come to the conclusion that it is feasible to develop
a potential for the NiAs unit cell using the MEAM formulation.
Since an interatomic potential is available, it is now possible to simulate the
material behaviour of Cu
6
Sn
5
at atomic length scales. As fracture is a atomic-
level phenomenon, simulations of the fracture toughness was carried out using
two methods, using a periodic crack and a crack-tip displacement eld. In both
methods, the crack was seen to propagate by cleavage of atomic planes without
5.2. RECOMMENDATIONS FOR FUTURE WORK 177
any evidence of the emission of dislocations. This is behaviour consistent with
a brittle material. Hence, this shows that it is feasible to simulate the fracture
of intermetallic compounds with MD. This also suggests that the MD potential
developed can also reproduce the qualitative brittle behaviour of the intermetallic
compound Cu
6
Sn
5
.
The two methods were also used to predict the critical fracture toughness K
IC
.
It was found to lie in the range of 0.72 - 0.80 MPa
m, which is on the lower end

of the range of experimental data. No conclusions can be drawn regarding the
accuracy of this prediction, as the experimental data is sparse and do not show
agreement.
Nevertheless, it has been shown that it is feasible to perform DFT calculations
and MD simulations to obtain the mechanical properties of Cu-Sn intermetallic
compounds. With improvements to these computational methods, it is envisaged
that they can be used as tools in the design of electronic components at length
scales where atomic-scale eects will become important.
5.2 Recommendations for Future Work
The DFT calculations in this study have focussed on Cu
6
Sn
5
and Cu
3
Sn, com-
pounds that are found in solder joints that make use of Sn-based solder. Such
calculations can be extended to Au-based and In-based intermetallic compounds
as solders containing these elements are playing a greater role in electronic pack-
aging.
The DFT calculations can be taken beyond merely calculating the elastic prop-
erties. Since the thermal loading in electronic packages is an important factor,
calculations of the thermal expansion coecient can be performed. An investiga-
tion can be performed to extend the method developed by Jin et al., developed to
calculate the thermal expansion coecient of cubic metals, to the lower-symmetry
intermetallic compounds [233].
For the MD potential developed in Chapter 3, a small database consisting of
seven quantities were used. This certainly has scope for expansion, given that
the MEAM formulation has 23 parameters. The choice depends on the type of
simulations that is to be carried out. A potential to simulate the Cu-Cu
6
Sn
5
inter-
face would require the properties of Cu to be explicitly included in the database
too. It is also possible to include the properties of Cu
3
Sn, and also the energies
of other crystal structures with similar stoichiometric ratio, in light of the results
in section 3.5.6. An expanded database beyond 23 quantities can also lead to the
theoretical development of the MEAM, as suitable modications to the MEAM
functions must be investigated to expand the number of adjustable parameters.
With an improved MD potential for Cu-Sn, MD simulations of nanoindentation
can be conducted. The results can be compared with the experimental studies on
Cu
6
Sn
5
making use of nanoindentation. This would enable further validation of
the potential to be carried out, and to examine the atomic-level processes that
occur during the procedure.
In Chapter 4, it has already been seen that the number of atoms does have an
eect on the results. Due to limitations in the computational resources available,
larger simulations were not performed. Future work can involve increasing the size
of these simulations to investigate the size eect.
There are also several possibilities for experimental work. As mentioned in sec-
tion 2.7.2, some researchers have suggested the possibility that the Youngs Modu-
lus obtained from nanoindentation does not reect the average over all orientations.
Experimental work could be designed to test this hypothesis, and nanoindentation
simulations with MD could also be conducted to provide predictions.
More work could be done to provide experimental data on the fracture tough-
ness of Cu
6
Sn
5
and Cu
3
Sn. This is necessary due to the sparsity of the experimen-
tal results. Also, as mentioned in section 4.1.2, out of the ve results available,
three employ the Vickers indentation method, whose accuracy and relevance in
determining the fracture toughness has been questioned.
180
Bibliography
[1] D. R. Frear and S. Thomas. Emerging materials challenges in microelectron-
ics packaging. MRS Bulletin, 28(1):6874, January 2003.
[2] E. P. Wood and K. L. Nimmo. In search of new lead-free electronic solders.
Journal of Electronic Materials, 23(8):709713, 1994.
[3] Sung K. Kang and Amit K. Sarkhel. Lead (Pb)-free solders for electronic
packaging. Journal of Electronic Materials, 23(8):701707, 1994.
[4] Judith Glazer. Microstructure and mechanical properties of Pb-free solder
alloys for low-cost electronic assembly: A review. Journal of Electronic
Materials, 23(8):693700, 1994.
[5] M. A. Dudek, R. S. Sidhu, and N. Chawla. Novel rare-earth-containing
lead-free solders with enhanced ductility. JOM, 58(6):5762, Jun 2006.
[6] D. R. Frear. Issues related to the implementation of Pb-free electronic sol-
ders in consumer electronics. Journal of Materials Science: Materials in
Electronics, 18:319330, 2007.
[7] Laura J. Turbini. Processing and material issues related to lead-free sol-
dering. Journal of Materials Science: Materials in Electronics, 18:147154,
2007.
BIBLIOGRAPHY 181
[8] D. R. Frear, J. W. Jang, J. K. Lin, and C. Zhang. Pb-free solders for ip-chip
interconnects. JOM, 53(6):2838, Jun 2001.
[9] C. E. Ho, S. C. Yang, and C. R. Kao. Interfacial reaction issues for lead-free
electronic solders. Journal of Materials Science: Materials in Electronics,
18:155174, 2007.
[10] Kejun Zeng, Roger Stierman, Tz-Cheng Chiu, and Darvin Edwards. Kirk-
endall void formation in eutectic SnPb solder joints on bare Cu and its eect
on joint reliability. Journal of Applied Physics, 97:024508, 2005.
[11] Kinuko Mishiro, Shigeo Ishikawa, Mitsunori Abe, Toshio Kumai, Yutaka Hi-
gashiguchi, and Ken ichiro Tsubone. Eect of the drop impact on BGA/CSP
package reliability. Microelectronics Reliability, 42:7782, 2002.
[12] Jin-Wook Jang, Ananda P. De Silva, James E. Drye, Steve L. Post, Nor-
man L. Owens, Jong-Kai Lin, and Darrel R. Frear. Failure morphology after
drop impact test of ball grid array (BGA) with lead-free Sn-3.8Ag-0.7Cu
and eutectic SnPb solders. IEEE Transactions on Electronics Packaging
Manufacturing, 30(1):4953, January 2007.
[13] T. T. Mattila and J. K. Kivilahti. Reliability of lead-free interconnections
under consecutive thermal and mechanical loadings. Journal of Electronic
Materials, 35(2):250256, 2006.
[14] Darrel Frear, Harold Morgan, Steven Burchett, and John Lau, editors. The
Mechanics of Solder Alloy Interconnects. Van Nostrand Reinhold, 1994.
[15] T. Sogo and S. Hara. Estimation of fall impact strength for BGA solder
joints. In International Conference on Electronic Packaging, pages 369373,
2001.
BIBLIOGRAPHY 182
[16] Hwa-Teng Lee, Ming-Hung Chen, Huei-Mei Jao, and Tain-Long Liao. In-
uence of interfacial intermetallic compound on fracture behaviour of solder
joints. Materials Science and Engineering A, 358:134141, 2003.
[17] Ahmer Syed, Seung Mo Kim, Wei Lin, Jin Young Kim, Eun Sook Sohn, and
Jae Hyeon Shin. A methodology for drop performance modeling and appli-
cation for design optimization of chip-scale packages. IEEE Transactions on
Electronics Packaging Manufacturing, 30(1):4248, January 2007.
[18] H. T. Chen, C. Q. Wang, and M. Y. Li. Numerical and experimental analysis
of the Sn3.5Ag0.75CU solder joint reliability under thermal cycling. Micro-
electronics Reliability, 46:13481356, 2006.
[19] R.J. Fields and S.R. Low. Physical and mechanical properties of inter-
metallic compounds commonly found in solder joints. World Wide Web,
http://www.metallurgy.nist.gov /mechanical properties/ solder paper.html.
[20] Michael D. Uchic, Dennis M. Dimiduk, Jerey N. Florando, and William D.
Nix. Sample dimensions inuence strength and crystal plasticity. Science,
205:986989, 13 August 2004.
[21] ITRS. International technology roadmap for semiconductors 2006 update:
Modeling and simulation. Technical report, ITRS, 2006.
[22] Andrew R. Leach. Molecular Modelling: principles and applications. Pearson
Education, second edition, 2001.
[23] V. Milman and M.C. Warren. Elastic properties of TiB
2
and MgB
2
. Journal
of Physics: Condensed Matter, 13:55855595, 2001.
[24] V. Milman and M.C. Warren. Elasticity of hexagonal BeO. Journal of
Physics: Condensed Matter, 13:241251, 2001.
BIBLIOGRAPHY 183
[25] Bengt Holm, Rajeev Ahuja, and B orje Johansson. Ab initio calculations of
the mechanical properties of Ti
3
SiC
2
. Applied Physics Letters, 79(10):1450
1542, September 2001.
[26] Jin-Cheng Zheng, Michael C. Payne, Yuan Ping Feng, and Adele Tzu-Lin
Lim. Stability and electronic properties of carbon phosphide compounds
with 1:1 stoichiometry. Physical Review B, 67:153105, 2003.
[27] Tejpal Kaur Hooghan, Russell F. Pinizzotto, John H. Watkins, and Toru
Okabe. A study of a low copper dental amalgam by analytical transmission
electron microscopy. Journal of Materials Research, 11(10):24742485, 1996.
[28] J. V. Wood and S. C. King. Production of dental alloy by the melt technique.
Journal of Materials Science, 13:11191124, 1978.
[29] D. Larcher, L. Y. Beaulieu, D. D. MacNiel, and J. R. Dahn. In situ x-ray
study of the electrochemical reaction of Li with
-Cu
6
Sn
5
. Journal of the
Electrochemical Society, 147(5):1658 1662, 2000.
[30] S. Sharma, L. Fransson, E. Sj ostedt, L. Nordstr om, B. Johansson, and K. Ed-
str om. A theoretical and experimental study of the lithiation of
-Cu
6
Sn
5
in
a lithium-ion battery. Journal of the Electrochemical Society, 150(3):A330
A334, 2003.
[31] I. Amadei, S. Panero, B. Scrosati, G. Cocco, and L. Schini. The Ni
3
Sn
4
intermetallic as a novel electrode in lithium cells. Journal of Power Sources,
143:227230, 2005.
[32] Xin-Qun Cheng and Peng Fei Shi. Electroless cu-plated Ni
3
Sn
4
alloy used as
anode material for lithium ion battery. Journal of Alloys and Compounds,
391:241244, 2005.
BIBLIOGRAPHY 184
[33] Heon-Young Lee, Serk-Won Jang, Sung-Man Lee, Seung-Joo Lee, and Hong-
Koo Baik. Lithium storage properties of nanocrystalline Ni
3
Sn
4
alloys pre-
pared by mechanical alloying. Journal of Power Sources, 112:812, 2002.
[34] S. Banovic. Microstructural characterization and mechanical behaviour of
Cu-Sn frangible bullets. Materials Science and Engineering A, 460-461:428
435, 2007.
[35] Yutaka Fujiwara. Sn deposition onto Cu and alloy layer growth by a contact
immersion process. Thin Solid Films, 425:121126, 2003.
[36] Takuma Katase, Ryochi Kurosaki, Kuniaki Murase, Tetsuji Hirato, and Ya-
suhiro Awakura. Formation of Cu-Sn alloy layer by contact deposition using
quaternary ammonium-imide-type ionic liquid. Electrochemical and Solid
State Letters, 9(4):C69C72, 2006.
[37] Kuniaki Murase, Ryochi Kurosaki, Takuma Katase, Hiroyuki Sugimura, Tet-
suji Hirato, and Yasuhiro Awakura. Electrochemical alloying of copper sub-
strate with tin using ionic liquid as an electrolyte at medium-low tempera-
tures. Journal of the Electrochemical Society, 154(11):D612D616, 2006.
[38] K.H. Prakash and T. Sritharan. Textured growth of cu/sn intermetallic
compounds. Journal of Electronic Materials, 31(11):12501255, 2002.
[39] J. O. Suh, K. N. Tu, and N. Tamura. Preferred orientation relationship
between Cu
6
Sn
5
scallop-type grains and Cu substrate in reactions between
molten Sn-based solders and Cu. Journal of Applied Physics, 102:063511,
2007.
[40] B. Subrahmanyam. Elastic moduli of some eutectic alloy systems. Transac-
tions of the Japan Institute of Metals, 13:8992, 1972.
BIBLIOGRAPHY 185
[41] X. Deng, M. Koopman, N. Chawla, and K. K. Chawla. Youngs modulus of
(Cu,Ag)-Sn intermetallics measured by nanoindentation. Materials Science
and Engineering A, 364:240243, 2004.
[42] G. Ghosh. Elastic properties, hardness and indentation fracture toughness
of intermetallics relevant to electronic packaging. Journal of Materials Re-
search, 19(5):1439 1454, May 2004.
[43] K. Mohankumar and A. A. O. Tay. Nanoindentation study of the Pb-free
solders in ne pitch interconnects. In 2004 Electronics Packaging Technology
Conference, pages 483489. IEEE, 2004.
[44] R. R. Chromik, R. P. Vinci, S. L. Allen, and M. R. Notis. Nanoindentation
measurements on Cu-Sn and Ag-Sn intermetallics formed in Pb-free solder
joints. Journal of Materials Research, 18(9):22512261, 2003.
[45] H. J. Albrecht, A. Juritza, K. M uller, J. Sterthaus, J. Villian, and
A. Vogliano. Interface reactions in microelectronic solder joints and asso-
ciated intermetallic compounds: An investigation of their mechanical prop-
erties using nanoindentation. In 2003 Electronics Packaging Technology Con-
ference, pages 726731. IEEE, 2003.
[46] B. Subrahmanyam. Elastic moduli of some complicated binary alloy systems.
Transactions of the Japan Institute of Metals, 13:9395, 1972.
[47] Guh-Yaw Jang, Jyh-Wei Lee, and Jenq-Gong Duh. The nanoindentation
characteristics of Cu
6
Sn
5
and Cu
3
Sn and Ni
3
Sn
4
intermetallic compounds in
the solder bump. Journal of Electronic Materials, 33(10):11031110, 2004.
[48] Zhong Chen, Min He, Bavani Balakrisnan, and Chan Choy Chum. Elasticity
modulus, hardness and fracture toughness of Ni
3
Sn
4
intermetallic thin lms.
Materials Science and Engineering A, 423:107110, 2006.
BIBLIOGRAPHY 186
[49] R. Cabarat, L. Guillet, and R. Le Roux. The elastic properties of metallic
alloys. Journal of the Institute of Metals, 75:391402, 1975.
[50] L. M. Ostrovskaya, V. N. Rodin, and A. I. Kuznetsov. Elastic properties of
intermetallic compounds produced by vacuum deposition. Soviet Journal of
Non-Ferrous Metals (Tsvetnye Metally), 26:9092, 1985.
[51] Iting Tsai, Enboa Wu, S. F. Yen, and T. H. Chuang. Mechanical properties
of intermetallic compounds on lead-free solder by moire techniques. Journal
of Electronic Materials, 35(5):10591066, 2006.
[52] Dongil Kwon. A Guide to Lead-Free Solders, chapter Intermetallic Formation
and Growth, pages 97126. Springer-Verlag, London, 2007.
[53] H. J. Albrecht, T. Hannach, A. H ase, A. Juritza, K. M uller, and W. H.
M uller. Nanoindentation: a suitable tool to determine local mechanical
properties in microelectronic packages and materials. Archives of Applied
Mechanics, 74:728738, 2005.
[54] Ping-Feng Yang, Yi-Shao Lai, Sheng-Rui Jian, Jiunn Chen, and Rong-
Sheng Chen. Nanoindentation identications of mechanical properties of
Cu
3
Sn,Cu
6
Sn
5
, and Ni
3
Sn
4
intermetalic compounds derived by diusion cou-
ples. Materials Science and Engineering A, 485:305310, 2008.
[55] P. Hohenberg and W. Kohn. Inhomogenous electron gas. Physical Review,
136(3B):B864B871, 1964.
[56] W. Kohn and L.J. Sham. Self-consistent equations including exchange and
correlation eects. Physical Review, 140(4A):A1133A1138, 1965.
[57] V. Milman, B. Winkler, J.A. White, C.J. Pickard, M.C. Payne, E.V.
Akhmatskaya, and R.H.Nobes. Electronic structure, properties and phase
BIBLIOGRAPHY 187
stability of inorganic crystals: A pseudopotential plane-wave study. Inter-
national Journal of Quantum Chemistry, 77:895910, 2000.
[58] Karsten Albe and Thomas E. Weirich. Structure and stability of - and
-Ti
2
Se. electron diraction versus density-functional theory calculations.
Acta Crystallographica Section A, A59:1821, 2003.
[59] Thomas E. Weirich. First-principles calculations as a tool for structure vali-
dation in electron crystallography. Acta Crystallographica Section A, A60:75
81, 2004.
[60] R. Ahuja, O. Eriksson, J.M. Wills, and B. Johansson. Electronic structure
of Ti
3
SiC
2
. Applied Physics Letters, 76(16):22262228, April 2000.
[61] K. Kang and W. Cai. Brittle and ductile fracture of semiconductor nanowires
- molecular dynamics simulations. Philosophical Magazine, 87(14-15):2169
2189, 2007.
[62] Karin Carling, G oran Wahnstr om, Thomas R. Mattsson, Ann E. Matts-
son, Nils Sandberg, and G oran Grimvall. Vacancies in metals: From
rst-principles calculations to experimental data. Physical Review Letters,
85(18):38623865, 2000.
[63] G. Ghosh, S. Delsante, G. Borzone, M. Asta, and R. Ferro. Phase stability
and cohesive properties of ti-zn intermetallics: First-principles calculations
and experimental results. Acta Materialia, 54:49774997, 2006.
[64] Stefano Curtarolo, Dane Morgan, and Gerbrand Ceder. Accuracy of ab
initio methods in predicting the crystal structures of metals: a review of 80
binary alloys. Computer Coupling of Phase Diagrams and Thermochemistry,
29:163211, 2005.
BIBLIOGRAPHY 188
[65] Patrice E. A. Turchi, L. Kaufman, and Zi-Kui Liu. Modeling of Ni-Cr-Mo
based alloys: Part I - phase stability. Computer Coupling of Phase Diagrams
and Thermochemistry, 30:7087, 2006.
[66] P. Maugis, J. Lacaze, R. Besson, and J. Morillo. Ab initio calculations of
phase stabilities in the Fe-Al-C system and CALPHAD-type assessment of
the iron-rich corner. Metallurgical and Materials Transactions A, 37A:3397
3401, 2006.
[67] Mustafa Uludogan, Tahir Cagin, Alejandro Strachan, and William A God-
dard III. Ab-initio studies of pressure induced phase transitions of BaO.
Journal of Computer-Aided Materials Design, 8:193202, 2001.
[68] James E. Saal, Shunli Shang, and Zi-Kui Liu. Structural evolution of boron
carbide via ab initio calculations. Applied Physics Letters, 91:231915, 2007.
[69] I.J. Robertson, M.C. Payne, and V. Heine. Multi-atom bonding in aluminium
over a wide range of coordination number. Europhysics Letters, 15(3):301
306, 1991.
[70] I.J. Robertson, Volker Heine, and M.C. Payne. Cohesion in aluminium sys-
tems: A rst-principles assessment of Glue schemes. Physical Review Let-
ters, 70(13):19441947, 1993.
[71] I.J. Robertson, D.I. Thomson, V. Heine, and M.C. Payne. A database of
structural energies of aluminium from ab initio calculations. Journal of
Physics: Condensed Matter, 6:99639972, 1994.
[72] M.C Payne, I.I. Robertson, D. Thomson, and V. Heine. Ab initio databases
for tting and testing interatomic potentials. Philosophical Magazine B,
73(1):191199, 1996.
BIBLIOGRAPHY 189
[73] Y Mishin, M.J.Mehl, D.A. Papaconstantopoulos, A.F Voter, and J.D. Kress.
Structural stability and lattice defects in copper: Ab initio, tight-binding
and embedded-atom calculations. Physical Review B, 63:224106, 2001.
[74] Y. Mishin, M. J. Mehl, and D. A. Papaconstantopoulos. Phase stability in
the fe-ni system: Investigation by rst-principles calculations and atomistic
simulations. Acta Materialia, 53:40294041, 2005.
[75] O. Beckstein, J.E. Klepeis, G.L.W. Hart, and O. Pankratov. First princi-
ples elastic constants and electronic structure of -Pt
2
Si and PtSi. Physical
Review B, 63:134112, 2001.
[76] B. B. Karki, L. Stixrude, S. J. Clark, M. C. Warren, G. J. Ackland, and
J. Crain. Elastic properties of orthorhombic MgSiO
3
perovskite at lower
mantle pressures. American Mineralogist, 82:635638, 1997.
[77] S. K. R. Patil, S. V. Khare, B. R. Tuttle, J. K. Bording, and S. Kodambaka.
Mechanical stability of possible structures of PtN investigated using rst-
principles calculations. Physical Review B, 73:104118, 2006.
[78] Bijaya B. Karki, Lars Stixrude, and Renata M. Wentzcovitch. High-pressure
elastic properties of major materials of earths mantle from rst principles.
Reviews of Geophysics, 39(4):507534, 2001.
[79] M.J. Mehl, D.J. Singh, and D.A. Papaconstantopoulos. Properties of ordered
intermetallic alloys: rst-principles and approximate methods. Materials
Science And Engineering A, 170:4957, 1993.
[80] Michael J. Mehl, Barry M. Klein, and Dimitris A. Papaconstantopoulous.
First principles calculations of elastic properties. In J.H. Westbrook and
R.L. Fleischer, editors, Intermetallic Compounds. Vol 1: Principles, pages
195210. John Wiley and Sons Ltd, 1994.
BIBLIOGRAPHY 190
[81] H. Anton and P.C. Schmidt. Theoretical investigations of the elastic con-
stants in laves phases. Intermetallics, 5:449465, 1997.
[82] B. Mayer, H. Anton, E. Bott, M. Methfessel, J. Sticht, J. Harris, and P.C.
Schmidt. Ab-initio calculation of the elastic constants and thermal expansion
coecients of laves phases. Intermetallics, 11:2332, 2003.
[83] B. Mayer, M. Methfessel, M. Schott, and P.C. Schmidt. Theoretical study of
the elastic constants of o-stoichiometric alloys. Intermetallics, 12:333339,
2004.
[84] Y. Le Page, John R. Rodgers, and Peter S. White. Ab initio modelling of in-
termetallics with CuAl
2
structure type: computed powder patterns from cell-
and-type CRYSTMET entries. Journal of Applied Crystallography, 35:364
367, 2002.
[85] G. Ghosh and M. Asta. Phase stability, phase transformations, and elas-
tic properties of Cu
6
Sn
5
: Ab initio calculations and experimental results.
Journal of Materials Research, 20(11):31023117, 2005.
[86] Ann-Kristin Larsson, Lars Stenberg, and Sven Lidin. The superstructure of
domain-twinned
-Cu
6
Sn
5
. Acta Crystallographica Section B, 50:636643,
1994.
[87] J. Goniakowski, J. M. Holender, L. N. Kantorovich, M. J. Gillan, and J. A.
White. Inuence of gradient corrections on the bulk and surface properties
of TiO
2
and SnO
2
. Physical Review B, 53(3):957960, 1996.
[88] Madeleine Meyer, Giovanni Onida, Maurizia Palummo, and Lucia Rein-
ing. Ab initio pseudopotential calculation of the equilibrium structure of
tin monoxide. Physical Review B, 64:045119, 2001.
[89] Michael P. Marder. Condensed Matter Physics. John Wiley and Sons, 2000.
BIBLIOGRAPHY 191
[90] Neil W. Ashcroft and N. David Mermin. Solid State Physics. New York,Holt,
Rinehard and Winston, 1976.
[91] Klaus Capelle. A birds-eye view of density-functional theory. arXiv.
[92] M.D. Segall, Philip J.D. Lindan, M.J. Probert, C.J. Pickard, P.J. Hasnip,
S.J. Clark, and M.C. Payne. First-principles simulation: ideas, illustrations
and the CASTEP code. Journal of Physics: Condensed Matter, 14:2717
2744, 2002.
[93] Stewart J. Clark, Matthew D. Segall, Chris J. Pickard, Phil J. Hasnip, Matt
I. J. Probert, Keith Refson, and Mike C. Payne. First principles methods
using CASTEP. Zeitschrift f ur Kristallographie, 220:567570, 2005.
[94] Hendrik J. Monkhorst and James D. Pack. Special points for brillouin-zone
integrations. Physical Review B, 13:51885192, 1976.
[95] M.C Payne, M.P. Teter, D.C. Allan, T.A. Arias, and J.D. Joannopoulos.
Iterative minimization techniques for ab initio total-energy calculations:
molecular dynamics and conjugate gradients. Reviews of Modern Physics,
64(4):10451097, 1992.
[96] John P. Perdew, Kieron Burke, and Matthias Enzerhof. Generalized gradi-
ent approximation made simple. Physical Review Letters, 77(18):38653868,
1996.
[97] John P. Perdew, Kieron Burke, and Matthias Enzerhof. Errata: Generalized
gradient approximation made simple. Physical Review Letters, 78(7):1396,
1997.
[98] John P. Perdew, Kieron Burke, and Matthias Enzerhof. Comment on
generalized gradient approximation made simple. Physical Review Letters,
80(4):890891, 1998.
BIBLIOGRAPHY 192
[99] J. F. Nye. Physical Properties of Crystals. Clarendon Press, Oxford, 1957.
[100] W. Voight. Ueber die beziehung zwischen den beiden elastict atsconstanten
isotroper k orper. Annalen Der Physik, 38:573587, 1889.
[101] A. Reuss. Berechnung der iessgrenze von mischkristallen auf grund der plas-
tizit atsbedingung f ur einkristalle. Zeitschrift f ur Angewandte Mathematik
und Mechanik, 9(1):4958, 1929.
[102] R. Hill. The elastic behaviour of a crystalline aggregate. Proceedings of the
Physical Society of London. Section A, 65(5):349354, 1952.
[103] Z. Hashin and S. Shtrikman. A variational approach to the theory of the
elastic behaviour of polycrystals. Journal of the Mechanics and Physics of
Solids, 10:343352, 1962.
[104] J. Peter Watt. Hashin-shtrikman bounds on the eective elastic moduli
of polycrystals with monoclinic symmetry. Journal of Applied Physics,
51(3):1520, 1980.
[105] J. Peter Watt. Hashin-shtrikman bounds on the eective elastic moduli
of polycrystals with orthorhombic symmetry. Journal of Applied Physics,
50(10):62906295, 1979.
[106] J. Peter Watt and Louis Peselnick. Clarication of the hashin-shtrikman
bounds on the eective elastic moduli of polycrystals with hexagonal, trig-
onal and tetragonal symmetries. Journal of Applied Physics, 51(3):1525,
1980.
[107] J. Peter Watt. The elastic properties of composite materials. Reviews of
Geophysics and Space Physics, 14(4):541563, 1976.
BIBLIOGRAPHY 193
[108] J. Peter Watt. POLYXSTAL: a FORTRAN program to calculate average
elastic properties of minerals from single crystal data. Computers and Geo-
sciences, 13(5):441462, 1987.
[109] A. Gangulee, G.C. Das, and M.B. Bever. An x-ray diraction and calori-
metric investigation of the compound Cu
6
Sn
5
. Metallurgical Transactions,
4:20632066, September 1973.
[110] J.D. Bernal. The complex structure of the copper-tin intermetallic com-
pounds. Nature, 122(3063):54, July 14 1928.
[111] Arne Westgren and G osta Phragmen. Rontgenanalyse der kupfer-
zinnlegierungen. Zeitschrift f ur anorganische und allgemeine chemie,
175(1):8089, 1928.
[112] O. Carlsson and G. H agg. Zur kenntnis des kristallbaues einiger kupfer-
zinnphasen (on the crystal structure of some copper-tin phases). Zeitschrift
f ur Kristallographie, 83:308317, 1932.
[113] B. Peplinski, G. Schulz, D. Schultze, and E. Schierhorn. Improved x-ray
powder diraction data for the disordered -Cu
6
Sn
5
alloy phase. Materials
Science Forum, 228-231:577582, 1996.
[114] Tao-Chih Chang, Moo-Chin Wang, and Min-Hsiung Hon. Eect of Ag-
addition on the structures of intermetallic compounds and the adhesion
strength of the Sn-9Ag-xAg/Cu interface. Journal of Crystal Growth,
252:391400, 2003.
[115] Tao-Chih Chang, Moo-Chin Wang, and Min-Hsiung Hon. Crystal growth of
the intermetallic compounds at the Sn-9Ag-xAg/Cu interface during isother-
mal aging. Journal of Crystal Growth, 252:401412, 2003.
BIBLIOGRAPHY 194
[116] Y. Watanabe, Y. Fujinaga, and H. Iwasaki. Lattice modulation in the long-
period superstructure of Cu
3
Sn. Acta Crystallographica, Section B, 39:306
311, 1983.
[117] Walter Burkhardt and Konrad Schubert.

Uber messingartige phasen mit
A3-verwandter struktur. Zeitschrift f ur Metallkunde, 50:442452, 1959.
[118] J.L.C Daams, P.Villars, and J.H.N. van Vucht. Atlas of Crystal Structure
Types for Intermetallic Phases, volume 2. ASM International, 2nd edition,
1992.
[119] W. Jeitschko and B. Jaberg. Structure renement of Ni
3
Sn
4
. Acta Crystal-
lographica B, 38:598600, 1982.
[120] Sigrid Furuseth and Helmer Fjellvag. Structural properties of Ni
3+x
Sn
4
. Acta
Chemica Scandinavica A, 40:695700, 1986.
[121] C. W. Fairhurst and J.B. Cohen. The crystal structures of two compounds
found in dental amalgam: Ag
2
Hg
3
and Ag
3
Sn. Acta Crystallographica B,
28:371378, 1972.
[122] O. Nial, A. Almin, and A. Westgren. R ontgenanalyse der systeme gold-
antimon und silber-zinn. Zeitschrift f ur Physikalische Chemie, B14:8190,
1931.
[123] Martin Ellner and Eric J. Mittemeijer. In situ and ex situ investigation of
the displacive phase transformations Ag
3
Sn(h) Ag
3
Sn(l) and Ag
3
Sb(h)
Ag
3
Sb(l). Zeitschrift f ur Kristallographie, 218:675682, 2003.
[124] Thomas H. Fischer and Jan Alml of. General methods for geometry and wave
function optimization. Journal of Physical Chemistry, 96:97689774, 1992.
BIBLIOGRAPHY 195
[125] Ann E. Mattsson, Peter A Schutz, Michael P Desjarlais, Thomas R Mattsson,
and Kevin Leung. Designing meaningful density functional theory calcula-
tions in materials science. Modelling and Simulation in Materials Science
and Engineering, 13:R1R31, 2005.
[126] J. P. Perdew and Alex Zunger. Self-interaction correction to density-
functional approximations for many-electron systems. Physical Review B,
23(10):50485079, 1981.
[127] David Vanderbilt. Soft self-consistent pseudopotentials in a generalized
eigenvalue formalism. Physical Review B, 41(11):78927895, 1990.
[128] Fernando Nogucira, Alberto Castro, and Miguel A.L. Marques. A tutorial
on density functional theory. In C. Fiolhais, F. Nogueira, and M. Marques,
editors, Lecture Notes In Physics. A Primer in Density Functional Theory.
Springer, 2003.
[129] M. Fuchs, M. Bockstedte, E. Pehlke, and M. Scheer. Pseudopotential study
of binding properties of solids within generalized gradient approximations:
The role of core-valence exchange correlation. Physical Review B, 57(4):2134
2145, 1998.
[130] Accelrys Inc. Materials studio 4.0. World Wide Web,
http://www.accelrys.com.
[131] B. Winkler. An introduction to computational crystallography. Zeitschrift
f ur Kristallographie, 214:506527, 1999.
[132] E.A. Brandes and G.B. Brook, editors. Smithells Metals Reference Book.
Butterworth-Heinemann Ltd, 7th edition, 1992.
[133] D. L. Price, J. M. Rowe, and R. M. Nicklow. Lattice dynamics of grey tin
and indium antimonide. Physical Review B, 3(4):12681279, 1971.
BIBLIOGRAPHY 196
[134] Morihiko Nakamura. Elastic properties. In J.H. Westbrook and R.L. Fleis-
cher, editors, Basic Mechanical Properties And Lattice Defects of Intermetal-
lic Compounds, pages 124. John Wiley and Sons Ltd, 2000.
[135] G. Ghosh, S. Vaynman, M. Asta, and M. E. Fine. Stability and elastic
properties of L1
2
(Al, Cu)
3
(Ti, Zr) phases: Ab initio calculations and ex-
periments. Intermetallics, 15:4454, 2007.
[136] Gregory S. Rohrer. Structure and Bonding in Crystalline Materials. Cam-
bridge University Press, Cambridge, UK, 2006.
[137] N. T. S. Lee, V. B. C. Tan, and K. M. Lim. First-principles calculations
of structural and mechanical properties of Cu
6
Sn
5
. Applied Physics Letters,
88:031913, 2006.
[138] N. T. S. Lee, V. B. C. Tan, and K. M. Lim. Structural and mechanical
properties of Sn-based intermetallics from ab-initio calculations. Applied
Physics Letters, 89:141908, 2006.
[139] C.S. Becquart, D. Kim, J.A. Rifkin, and P.C. Clapp. Fracture properties of
metals and alloys from molecular dynamics simulations. Materials Science
and Engineering A, 170:8794, 1993.
[140] R. Car and M.Parrinello. Unied approach for molecular dynamics and
density-functional theory. Physical Review Letters, 55(22):24712474, 1985.
[141] Jean Pierre Hansen and Ian R. McDonald. Theory of Simple Liquids. Aca-
demic Press, 1976.
[142] National Institute of Science and Technology. The NIST ref-
erence on constants, units and uncertainty. World Wide Web,
http://physics.nist.gov/cuu/index.html.
BIBLIOGRAPHY 197
[143] Loup Verlet. Computer experiments on classical uids. I. Thermodynami-
cal Properties of Lennard-Jones Molecules. Physical Review, 159(1):98103,
1967.
[144] Philip M. Morse. Diatomic molecules according to the wave mechanics.II.
vibrational levels. Physical Review, 34:5764, 1929.
[145] L.A. Girifalco and V.G Weizer. Application of the morse potential function
to cubic metals. Physical Review, 114(3):687690, 1959.
[146] R.C. Lincoln, K.M. Koliwad, and P.B. Ghate. Morse-potential evaluation
of second and third-order elastic constants of some cubic metals. Physical
Review, 157(3):463466, 1967.
[147] Frederick Milstein. Applicability of exponentially attractive and repulsive
interatomic potential functions in the description of cubic crystals. Journal
of Applied Physics, 44(9):38253832, 1973.
[148] Murray S Daw and M.I. Baskes. Embedded-atom method: Derivation and
application to impurities, surfaces and other defects in metals. Physical
Review B, 29(12):64436453, 1984.
[149] Arthur F. Voter. The embedded-atom method. In J.H. Westbrook and R.L.
Fleischer, editors, Intermetallic Compounds. Vol 1: Principles, pages 7790.
John Wiley and Sons Ltd, 1994.
[150] M.I. Baskes. Modied embedded-atom potentials for cubic materials and
impurities. Physical Review B, 46(5):27272742, 1992.
[151] Frank H. Stillinger and Thomas A. Weber. Computer simulation of local
order in condensed phases of silicon. Physical Review B, 31(8):52625271,
1985.
BIBLIOGRAPHY 198
[152] Dominic Holland and Michael Marder. Ideal brittle fracture of silicon studied
with molecular dynamics. Physical Review Letters, 80(4):746749, 1998.
[153] P. Heino, L. Perondi, K. Kaski, and E. Ristolainen. Stacking fault en-
ergy of copper from molecular dynamics simulation. Physical Review B,
60(21):1462514631, 1999.
[154] Arthur F. Voter and Shao Ping Chen. Accurate interatomic potentials for
Ni, Al and Ni
3
Al. In Materials Research Society Symposium Proceedings,
volume 82, pages 175180, 1986.
[155] Y Mishin, M.J.Mehl, and D.A. Papaconstantopoulos. Embedded-atom po-
tential for B2-NiAl. Physical Review B, 65:224114, 2002.
[156] A.-K. Larsson, L. Stenberg, and S. Lidin. Crystal structure modulations in
-Cu
5
Sn
4
. Zeitschrift f ur Kristallographie, 210:832837, 1995.
[157] Jose F Aguilar, R Ravelo, and M.I. Baskes. Morphology and dynamics of
2D Cu-Sn alloys on (100) and (111) Cu surfaces. Modelling Simul. Mater.
Sci. Eng., 8:335344, 2000.
[158] Sven Lidin and Ann-Kristin Larsson. A survey of superstructures in inter-
metallic nias-Ni
2
In-type phases. Journal of Solid State Chemistry, 118:313
322, 1995.
[159] S. Duman, S. Ba gci, H. M. T ut unc u, and G. P. Srivastava. First-principles
studies of ground-state and dynamical properties of MgS, MgSe, and MgTe
in the rocksalt, zinc blende, wurtzite and nickel arsenide phases. Physical
Review B, 73:205201, 2006.
[160] Murray S Daw and M.I. Baskes. Semiempirical, quantum mechanical cal-
culation of hydrogen embrittlement in metals. Physical Review Letters,
50(17):12851288, 1983.
BIBLIOGRAPHY 199
[161] R.A. Johnson. Analytic nearest-neighbour model for fcc metals. Physical
Review B, 37(8):39243931, 1988.
[162] D.J. Oh and R.A. Johnson. Simple embedded atom method model for fcc
and hcp metals. Journal of Materials Research, 3(3):471478, 1988.
[163] F.Ercolessi and J.B. Adams. Interatomic potentials from rst-principles cal-
culations: the force-matching method. Europhysics Letters, 26(8):583588,
1994.
[164] Xiang-Yang Liu, F.Ercolessi, and J.B. Adams. EAM potential for magnesium
from quantum mechanical forces. Modelling and Simulation in Materials
Science and Engineering, 4:293303, 1996.
[165] Xiang-Yang Liu, F.Ercolessi, and J.B. Adams. Aluminium interatomic po-
tential from density functional theory calculations with improved stacking
fault energy. Modelling and Simulation in Materials Science and Engineer-
ing, 12:665670, 2004.
[166] Robert D. Boyer, Ju Li, Shigenobu Ogata, and Sidney Yip. Analysis of shear
deformations in Al and Cu: empirical potentials versus density functional
theory. Modelling and Simulation in Materials Science and Engineering,
12:10171029, 2004.
[167] Y. Mishin, D. Farkas, M. J. Mehl, and D. A. Papaconstantopoulos. Inter-
atomic potentials for monoatomic metals from experimental data and ab
initio calculations. Physical Review B, 59(5):3393 3407, 1999.
[168] M. I. Baskes. Many-body eects in fcc metals: A Lennard-Jones embedded
atom potential. Physical Review Letters, 83(13):25922596, 1999.
[169] S. G. Srinisavan and M. I. Baskes. On the Lennard-Jones EAM potential.
Proceedings of the Royal Society of London A, 460:16491672, 2004.
BIBLIOGRAPHY 200
[170] M.I. Baskes, S.P. Chen, and F.J. Cherne. Atomistic model of gallium. Phys-
ical Review B, 66:104107, 2002.
[171] X.H. Liu, Z. Zhen, and J.C Zhang. New potential energy functions for
diamond and -tin crystals. Chinese Physics Letters, 15(5):360361, 1998.
[172] R Ravelo and M Baskes. Equilibrium and thermodynamic properties of grey,
white and liquid tin. Physical Review Letters, 79(13):24822485, 1997.
[173] S.M. Foiles and M.S. Daw. Application of the embedded atom method to
Ni
3
Al. Journal of Materials Research, 2(1):515, 1987.
[174] M.I. Baskes. Atomistic potentials for the molybdenum-silicon system. Ma-
terials Science and Engineering A, 261:165168, 1999.
[175] F. J. Cherne, M. I. Baskes, and R. B. Schwarz. Atomistic simulations of the
phase stability and elastic properties of nickel-zirconium alloys. Journal of
Non-Crystalline Solids, 317:4551, 2003.
[176] Hai Dong, Lianhua Fan, Kyoung sik Moon, C P Wong, and M I Baskes.
MEAM molecular dynamics study of lead free solder for electronic packaging
applications. Modelling and Simulation in Materials Science and Engineer-
ing, 13:12791290, 2005.
[177] P. Heino. Atomistic potential for TaSi system: application to low-index
interfaces. Modelling and Simulation in Materials Science and Engineering,
10:403412, 2002.
[178] Chin-Lung Kuo and Paulette Clancy. MEAM molecular dynamics study of
a gold thin lm on a silicon substrate. Surface Science, 551:3958, 2004.
[179] R. Ravelo, J. Aguilar, and M.I. Baskes. Molecular dynamics studies of thin-
lms of Sn on Cu. In Materials Research Society Symposium Proceedings,
volume 492, pages 4348, 1998.
BIBLIOGRAPHY 201
[180] P. van Beurden, H. G. J. Verhoeven, G. J. Kramer, and B. J. Thijsse. Atom-
istic potential for adsorbate/surface systems: CO on Pt. Physical Review B,
66:235409, 2002.
[181] P. van Beurden, B. S. Bunnik, G. J. Kramer, and A. Borg. Mechanism and
dynamics of the CO-induced lifting of the Pt(100) surface reconstruction.
Physical Review Letters, 90(6):066106, 2003.
[182] Chin-Lung Kuo and Paulette Clancy. Development of atomistic MEAM po-
tentials for the silicon-oxygen-gold ternary system. Modelling and Simulation
in Materials Science and Engineering, 13:13091329, 2005.
[183] M. I. Baskes, S. G. Srinivasan, S. M. Valone, and R. G. Hoagland. Multistate
modied embedded atom method. Physical Review B, 75:094113, 2007.
[184] Richard LeSar, Daryl C. Chrzan, and Ernest Orlando. Is computational
materials science overrated? Materials Today, 2(3):21, 1999.
[185] M.I. Baskes, J.E. Angelo, and C.L. Bisson. Atomistic calculations of com-
posite interfaces. Modelling and Simulation in Materials Science and Engi-
neering, 2:505518, 1994.
[186] M.I. Baskes. Determination of modied embedded atom parameters for
nickel. Materials Chemistry and Physics, 50:152158, 1997.
[187] P. van Beurden and G.J. Kramer. Parametrization of modied embedded-
atom-method potentials for Rh, Pd, Ir, and Pt based on density functional
theory calculations, with applications to surface properties. Physical Review
B, 63:165106, 2001.
[188] James H Rose, John R Smith, Francisco Guinea, and John Ferrante. Univer-
sal features of the equation of state of metals. Physical Review B, 29(6):2963
2969, 1984.
BIBLIOGRAPHY 202
[189] Manabu Satou, Sidney Yip, and Katsunori Abe. Molecular dynamics simula-
tion of vanadium using an interatomic potential tted to nite temperature
properties. Journal of Nuclear Materials, 307-311:10071010, 2002.
[190] Byeong-Joo Lee, Jae-Hyeok Shim, and M.I. Baskes. Semiempirical atomic
potentials for the fcc metals Cu, Ag, Au, Ni, Pd, Pt, Al and Pb based on rst
and second nearest-neighbour modied embedded atom method. Physical
Review B, 68:144112, 2003.
[191] Byeong-Joo Lee, M.I. Baskes, Hanchul Kim, and Yang Koo Cho. Second
nearest-neighbour modied embedded atom method potentials for bcc tran-
sition metals. Physical Review B, 64:184102, 2001.
[192] Byeong-Joo Lee and M.I. Baskes. Second nearest-neighbour modied
embedded-atom-method potential. Physical Review B, 62(13):85648567,
2000.
[193] Xiaoying Yuan, Kunio Takahashi, Yingchen Yin, and Tadao Onzawa. De-
velopment of modied embedded atom method for a bcc metal: lithium.
Modelling and Simulation in Materials Science and Engineering, 11:447456,
2003.
[194] Pierre-Matthieu Anglade, Gerald Jomard, Gregory Robert, and Gilles Zerah.
Computation of the peierls stress in tantalum with extended-range modied
embedded atom method potential. Journal of Physics: Condensed Matter,
17:20032018, 2005.
[195] B. J. Thijsse. Relationship between the modied embedded-atom method
and stillinger-weber potentials in calculating the structure of silicon. Physical
Review B, 65:195207, 2002.
BIBLIOGRAPHY 203
[196] J. A. Nelder and R. Mead. A simplex method for function minimization.
Computer Journal, 7:308313, 1965.
[197] S. Kirkpatrick, C.D. Gelatt, and M.P. Vecchi. Optimization by simulated
annealing. Science, 220:671680, 1983.
[198] Gunther Dueck and Tobias Scheuer. Threshold accepting: A general purpose
optimization algorithm appearing superior to simulated annealing. Journal
of Computational Physics, 90:161175, 1990.
[199] Makoto Matsumoto and Takuji Nishimura. Mersenne twister: A 623-
dimensionally equidistributed uniform pseudo random number generator.
ACM Transactions on Modeling and Computer Simulation, 8(1):330, 1998.
[200] Young-Min Kim and Byeong-Joo Lee. A modied embedded-atom method
interatomic potential for the Cu-Zr system. Journal of Materials Research,
23(4):10951104, 2008.
[201] Byeong-Joo Lee. A modied embedded-atom method interatomic potential
for the Fe-C system. Acta Materialia, 54:701711, 2006.
[202] R.A. Johnson. Internal relaxation in the HCP lattice. Modelling and Simu-
lation in Materials Science and Engineering, 1:717722, 1993.
[203] H. Flandorfer, U. Saeed, C. Luef, A. Sabbar, and H. Ipser. Interfaces in
lead-free solder alloys: Enthalpy of formation of binary Ag-Sn,Cu-Sn and
Ni-Sn intermetallic compounds. Thermochimica Acta, 459:3439, 2007.
[204] H. H akkinen and M. Manninen. Computer simulation of disordering and
premelting of low-index faces of copper. Physical Review B, 46(3):17251743,
1992.
[205] Max Born. On the stability of crystal lattices. i. Proceedings of the Cambridge
Philosophical Society, 36:160172, 1940.
BIBLIOGRAPHY 204
[206] B. Balakrisnan, C. C. Chum, M. Li, Z. Chen, and T. Cahyadi. Fracture
toughness of Cu-Sn intermetallic thin lms. Journal of Electronic Materials,
32(3):166171, 2003.
[207] Zhong Chen and Zhenghao Gan. Fracture toughness measurement of thin
lms on compliant substrate using controlled buckling test. Thin Solid Films,
515:33053309, 2007.
[208] G. R. Anstis, P. Chantikul, B. R. Lawn, and D. B. Marshall. A critical eval-
uation of indentation techniques for measuring fracture toughness: I, direct
crack measurements. Journal of the American Ceramic Society, 64(9):533
538, 1981.
[209] J. S. Kang, R. A. Gagliano, G. Ghosh, and M. E. Fine. Isothermal solid-
ication of Cu/Sn diusion couples to form thin-solder joints. Journal of
Electronic Materials, 31(11):1238, 2002.
[210] Benito deCelis, Ali S. Argon, and Sidney Yip. Molecular dynamics simulation
of crack tip processes in alpha-iron and copper. Journal of Applied Physics,
54(9):48644878, 1983.
[211] R.G Hoagland, M.S Daw, S.M. Foiles, and M.I. Baskes. An atomic model
of crack tip deformation using an embedded atom potential. Journal of
Materials Research, 5(2):313324, Feb 1990.
[212] G. C. Sih, P. C. Paris, and G. R. Irwin. On cracks in rectilinearly anisotropic
bodies. International Journal of Fracture Mechanics, 1:189203, 1965.
[213] Murray S. Daw, Stephen M. Foiles, and Michael I. Baskes. The embedded-
atom method: a review of theory and applications. Materials Science Re-
ports, 9:251310, 1993.
BIBLIOGRAPHY 205
[214] A. Latapie and D. Farkas. Molecular dynamics investigation of the fracture
behavior of nanocrystalline -Fe. Physical Review B, 69:134110, 2004.
[215] Dominic Holland and Michael Marder. Erratum: Ideal brittle fracture of sil-
icon studied with molecular dynamics. Physical Review Letters, 81(18):4029,
1998.
[216] Jens A Hauch, Dominic Holland, M.P. Marder, and Harry L. Swinney. Dy-
namic fracture in single crystal silicon. Physical Review Letters, 82(19):3823
3826, 1999.
[217] J.G. Swadener, M.I. Baskes, and M. Nastasi. Molecular dynamics simulation
of brittle fracture in silicon. Physical Review Letters, 89(8):085503, 2002.
[218] Nicholas P. Bailey and James P. Sethna. Macroscopic measure of the cohesive
length scale: Fracture of notched single-crystal silicon. Physical Review B,
68:205204, 2003.
[219] T. Fujii and Y. Akiniwa. Molecular dynamics analysis for fracture behaviour
of single crystal silicon thin lm with micro notch. Modelling and Simulation
in Materials Science and Engineering, 14:S73S83, 2006.
[220] Hiroshi Tada, Paul C. Paris, and George R. Irwin. The stress analysis of
cracks handbook, chapter Part III. Two-Dimensional Stress Solutions for Var-
ious Congurations with Cracks, page 170. ASME Press, New York, 3rd
edition, 2000.
[221] P. Dang and M. Grujicic. The eect of crack-tip material evolution on frac-
ture toughness: An atomistic simulation study of the ti-v alloy system. Acta
Materialia, 45(1):7587, 1997.
BIBLIOGRAPHY 206
[222] Naoto Hirosaki, Shigenobu Ogata, and Hiroshi Kitagawa. Molecular dy-
namics study of the stress distribution near a crack tip in -silicon nitride.
Materials Science Research International, 5(4):253257, 1999.
[223] Frohmut R osch, Christoph Rudhart, Johannes Roth, and Hans-Rainer Tre-
bin. Dynamic fracture of icosahedral model quasicrystals: A molecular dy-
namics study. Physical Review B, 72:014128, 2005.
[224] M.I. Baskes, R.G. Hoagland, and A. Needleman. Summary report: com-
putational issues in the mechanical behaviour of metals and intermetallics.
Materials Science and Engineering A, 159:134, 1992.
[225] J.A. Rifkin. Centre for materials simulation, university of connecticut. World
Wide Web, http://www.ims.uconn.edu/centers/simul/.
[226] Sandia National Laboratories. Lammps. http://lammps.sandia.gov/.
[227] William C. Swope, Hans C. Andersen, Peter H. Berens, and Kent R. Wilson.
Computer simulation method for the calculation of equilibrium constants for
the formation of physical clusters of molecules: application to small water
clusters. Journal of Chemical Physics, 76(1):637649, 1982.
[228] H.J.C. Berendsen and W.F. van Gunsteren. Practical algorithms for dynamic
simulations. In G. Ciccotti and W.G. Hoover, editors, Molecular-Dynamics
Simulation of Statistical-Mechanical Systems, pages 4365, 1986.
[229] Shuichi Nose. Constant temperature molecular dynamics methods. Progress
of Theoretical Physics Supplement, 103:146, 1991.
[230] L.V. Woodcock. Isothermal molecular dynamics calculations for liquid salts.
Chemical Physics Letters, 10(3):257261, 1971.
APPENDIX 207
[231] George D. Quinn and Richard C. Brandt. On the vickers indentation fracture
toughness test. Journal of the American Ceramic Society, 90(3):673680,
2007.
[232] X. Deng, N. Chawla, K. K. Chawla, and M. Koopman. Deformation be-
haviour of (Cu,Ag)-Sn intermetallics by nanoindentation. Acta Materialia,
52:42914303, 2004.
[233] Hong Mei Jin and Ping Wu. First principles calculation of thermal expansion
coecient. part 1. cubic metals. Journal of Alloys and Compounds, 343:71
76, 2002.
[234] D.C. Rapaport. The Art of Molecular Dynamics Simulation. Cambridge
University Press, Cambridge,UK, 1995.
[235] M. F. X. Wagner and W. Windl. Lattice stability, elastic constants and
macroscopic moduli of NiTi martensites from rst principles. Acta Materi-
alia, 56:62326245, 2008.
[236] L. E. McNeil and M. Grimsditch. Elastic constants of As
2
S
3
. Physical Review
B, 44(9):41744177, 1991.
[237] David R. Lide, editor. CRC Handbook of Chemistry and Physics. CRC Press,
84th edition, 2003.
[238] Anthony C. Fischer-Cripps. Nanoindentation. Springer, New York, 2004.
208
Appendix A
Elasticity Formulas
A.1 Orientation Dependance Of the Youngs Mod-
ulus
From Nye [99], the Youngs Modulus of a crystal is given by
1
s
1111
. The orientation
dependence of s
1111
in the form of Einsteins indicial notation is given by
s
1111
= a
1m
a
1n
a
1p
a
1q
s
mnpq
(A.1)
A.1.1 Cubic Crystal
For a cubic crystal, Equation A.1 is expanded to give
s
1111
= s
1111
_
a
4
11
+a
4
12
+a
4
13
_
+ (2s
1122
+ 4s
2323
)
_
a
2
11
a
2
12
+a
2
11
a
2
13
+a
2
12
a
2
23
_
(A.2)
a
11
, a
12
and a
13
are the direction cosines of the vector lying on the new axis x
1
with respect to the existing axes x
1
, x
2
and x
3
. Given that the direction cosines
also form the components of the unit vector, then the following identity can be
APPENDIX A 209
written down
a
2
11
+a
2
12
+a
2
13
= 1 (A.3)
.
Therefore, the rst term in Equation A.2 can be simplied into
a
4
11
+a
4
12
+a
4
13
=
_
a
2
11
+a
2
12
+a
2
13
_
2
2
_
a
2
11
a
2
12
+a
2
11
a
2
13
+a
2
12
a
2
23
_
(A.4)
= 1 2
_
a
2
11
a
2
12
+a
2
11
a
2
13
+a
2
12
a
2
23
_
(A.5)
By convention, reducing the number of indices from four to two means that
s
11
= s
1111
, s
12
= 4s
1122
and s
44
= 4s
2323
. Hence, Equation A.2 is simplied to
s
11
= s
11
2
_
s
11
s
12
+
1
2
s
44
_
_
a
2
11
a
2
12
+a
2
11
a
2
13
+a
2
12
a
2
23
_
(A.6)
A.1.2 Monoclinic Crystal
The work done in this thesis has been with orthorhombic and monoclinic crystal
structures. Equation A.1 can also be expanded for a monoclinic crystal, shown in
Equation A.7. This expression can also be used for an orthorhombic crystal by
setting s
15
, s
25
, s
35
, s
46
= 0.
s
11
= a
4
11
s
11
+ a
4
12
s
22
+ a
4
13
s
33
+
2a
2
11
a
2
12
s
12
+ 2a
2
11
a
2
13
s
13
+ 2a
2
12
a
2
13
s
23
+
a
2
12
a
2
13
s
44
+ a
2
11
a
2
13
s
55
+ a
2
11
a
2
12
s
66
+
2a
3
11
a
13
s
15
+ 2a
11
a
2
12
a
13
s
25
+ 2a
11
a
3
13
s
35
+ 2a
11
a
2
12
a
13
s
46
(A.7)
210
Appendix B
Internal Relaxations In a Crystal
B.1 Honeycomb Lattice
It is mentioned by Johnson that the possibility of internal relaxations within the
unit cell arises when there is more than one atom per primitive unit cell [202].
This Appendix demonstrates this for a honeycomb lattice, which is made up of
two particles per bravais lattice point.
Part of the honeycomb lattice is shown in Figure B.1(a). The coordination of
atom 1 is an equilateral triangle made up of atoms 2,3 and 4. This coordination
is similar to the coordination polyhedron of the Sn atom in Figure 3.10, which
consists of two planes of Cu atoms arranged in an equilateral triangle.
Assume that for this particular honeycomb lattice, the atoms are of unit dis-
tance apart and have only rst neighbour interactions under a pairwise potential.
Hence, the force on atom 1 depends only on atoms 2,3 and 4. Since the interatomic
distances r
12
, r
13
and r
14
are the same, the forces are all the same and balanced,
hence the net force on atom 1 is zero.
When a shear strain is applied, causing the new x-coordinate x
= x + y,
atom 3 shifts to the right and atom 4 shifts to the left. This is shown in Figure
B.1(b). r
13
is lengthened while r
14
is shortened. It should be noted that this
APPENDIX B 211
1 2
3
4
(-1,0)
(0,0)
(0.5,0.866)
(0.5,-0.866)
1 2
3
4
(-1,0)
(0,0)
(0.5+ 0.866,0.866)
(0.5-0.866,-0.866)
1
1
1
1
r
13'
r
14'
(a) (b)
Figure B.1: Part of the honeycomb lattice. The (x, y) coordinates are given beside
each particle. (a) Before shear, all particles are unit distance away from each other
(b) after shear. The arrows beside particles 3 and 4 shows how they have been
displaced
shear deformation destroys the 120
rotational symmetry as the angles within

the honeycomb are no longer 120
. Neglecting higher powers of strain, the new

interatomic distances are
r
13
=
1 + 0.866 (B.1)
r
14
=
1 0.866 (B.2)
Let the pairwise forces be f
12
,f
13
and f
14
, all three of which would have dif-
ferent values depending on the choice of interatomic potential. The net force on
atom 1 is now
f
1
= f
12
_
_
1
0
_
_
+
f
13
13
_
_
0.5 + 0.866
0.866
_
_
+
f
14
14
_
_
0.5 0.866
0.866
_
_
(B.3)
Thus, Equation B.3 shows that the possibility of a net force exists.
212
Appendix C
MEAM Formulas
C.1 MEAM Energy
The energy of the system of atoms under the MEAM is given by
E =
j>i
S
ij
(r
ij
) +
i
F(
i
) (C.1)
where (r
ij
) is the pairwise function, S
ij
is the screening function,
i
is the
electron density and F is the embedding function. Detailed expressions of these
terms are given in section 3.3.2.
C.2 Cu-Sn Interatomic Distance
In the NiAs crystal, there are two atoms in the basis. From the Atlas of Inter-
metallic Compounds [118], the Cu atom is located at (0,0,0) and the Sn atom is
APPENDIX C 213
located at (1/3,2/3,1/4). These coordinates are given in terms of the cell vectors
a
1
= (1/2)a
i + (
3/2)a
j (C.2)
a
2
= (1/2)a
i (
3/2)a
j (C.3)
a
3
= c
k (C.4)
The vector between the Cu and Sn atoms is
r
CuSn
= (1/2)a
3/6a
j + 1/4c
k (C.5)
Hence, this shows that the distance between the Cu atom and Sn atom is
r
CuSn
=
_
(1/4)a
2
+ (1/12)a
2
+ (1/16)c
2
=
_
(1/3)a
2
+ (1/16)c
2
(C.6)
C.3 MEAM Lattice Sums for the NiAs Crystal
The aim of this section is to show that through the analysis of the coordination
polyhedron, it can be shown that some of the partial electron densities of the
MEAM for the NiAs remain zero regardless of the value of the lattice constants a
and c. Since it is known that there is one Cu atom and one Sn atom in the basis,
the coordination polyhedron for each of them can be considered separately.
C.3.1 Sn atom
Using the data found in the Atlas [118], the components of the position vectors
from the Sn atom are given in Table C.1.
All the Cu atoms have the same distance from the Sn atom. Thus the value of
the atomic electron density d
(h)
j
is not relevant in the lattice sums as its value can
be taken outside of the summation. From the data, it can be easily seen that
APPENDIX C 214
Table C.1: Position vector components of the atoms in the coordination polyhedron
of the Sn atom
x y z
Cu 0.0 -0.577350a -0.25c
Cu 0.0 -0.577350a 0.25c
Cu 0.5a 0.288675a -0.25c
Cu -0.5a 0.288675a -0.25c
Cu -0.5a 0.288675a 0.25c
Cu 0.5a 0.288675a 0.25c
x = 0,
y = 0,
z = 0 (C.7)
thus leading to the conclusion that (
(1)
)
2
for Sn is zero for all a and c.
With the data, the following can be shown
x
2
= a
2
,
y
2
= a
2
,
z
2
= 0.375c
2
(C.8)
x
2
y = 0.288a
3
,
y
3
= 0.288a
3
(C.9)
thus (
(2)
)
2
and (
(3)
)
2
for Sn are not zero.
C.3.2 Cu atom
The coordination polyhedron for the Cu basis atom contains both Cu atoms and
Sn atoms. In this case, the sums for the Cu-Cu interactions and Cu-Sn interactions
will have to be examined separately. The position vectors of the Cu atoms in the
coordination polyhedron are given in Table C.2
It can be seen that all the sums for Cu-Cu are zero except
z
2
= 0.5c
2
(C.10)
For the Cu-Sn interactions, all the Sn atoms have the same distance from the
APPENDIX C 215
Table C.2: Position vector components of the Cu atoms in the coordination poly-
hedron of the Cu atom
x y z
Cu 0.0 0.0 -0.5c
Cu 0.0 0.0 0.5c
Table C.3: Position vector components of the Sn atoms in the coordination poly-
hedron of the Cu atom
x y z
Cu 0.0 0.577350a 0.25c
Cu -0.5a 0.288675a -0.25c
Cu 0.5a 0.288675a -0.25c
Cu -0.5a -0.288675a 0.25c
Cu 0.5a -0.288675a 0.25c
Cu 0.0 -0.577350a -0.25c
Cu atom. Similarly, the value of the atomic electron density d
(h)
j
is not relevant in
the lattice sums. Their position vectors are shown in Table C.3.
From this data, it is found that all the lattice sums for (
(1)
)
2
and (
(3)
)
2
are
zero for all a and c. The non-zero sums are
x
2
= a
2
,
y
2
= a
2
,
z
2
= 0.375c
2
(C.11)
C.3.3 Conclusion
It has thus been shown that when the lattice constants a and c are varied, for the
Sn atom, (
(1)
)
2
remains zero, while for the Cu atom, (
(1)
)
2
and (
(3)
)
2
remain
zero.
APPENDIX C 216
C.4 MEAM Forces
This section of the Appendix presents the force equations from the MEAM. These
expressions are needed in order to write the force subroutine needed for MD simu-
lations. Typically, when interatomic potentials are published, the energy equation
is published, from which the force expressions must be derived by the user in order
to implement it in a computer code.
To obtain the force expressions, two methods can be used. The most common
way is to take the derivative of the total energy of the system E with respect to
the interatomic distance x
ij
. Hence, the force on atom i in the direction x due to
atom j is
f
x,ij
=
E
x
ij
(C.12)
For a pairwise potential, this is given by
f
x,ij
=
(r
ij
)
x
ij
(C.13)
while for the EAM, the pairwise force is
f
x,ij
=
_
(r
ij
)
x
ij
+
F
i

i
j
x
ij
+
F
j

j
i
x
ij
_
(C.14)
where the symbols have the same meaning as in section 3.1.
Another way is to take the derivative with respect to the atomic coordinates
themselves. This is done when three-body terms are present in the potential
[234]. Since such terms are found in the MEAM, this method is used. Hence, the
analytical expression of the pairwise force force on atom a in the x direction f
x,a
is obtained by the partial derivative of Equation (C.1).
Let the MEAM energy in Equation (C.1) be expressed as
E = E
pair
+E
embed
(C.15)
APPENDIX C 217
where
E
pair
=
j>i
S
ij
(r
ij
) (C.16)
E
embed
=
i
F
i
(
i
) (C.17)
Then the force on each atom is
f
x,a
=
_
E
pair
x
a
+
E
embed
x
a
_
(C.18)
Hence, the contribution of the pairwise energy and the embedding energy to
the force will be looked at separately.
C.4.1 Derivative of the Pairwise energy
The derivative of the pairwise energy is
E
pair
x
a
=
j>i
S
ij
x
a
(r
ij
) +S
ij
(r
ij
)
r
ij
(C.19)
For the second term, it is non-zero only when i, j = a and S
ij
is treated as a
constant. Hence the forces due to this term can be considered separately by doing
a pairwise calculation instead, i.e.
f
ij,x
= S
ij
(r
ij
)
r
ij
x
ij
r
ij
(C.20)
This means that the force on the atoms due to the pairwise energy can be
calculated in two separate parts. The rst part is to calculate the pairwise forces
using Equation C.4.1. The second part is done by calculating the forces on each
individual atom through the rst term of Equation C.19, which is the force due
APPENDIX C 218
to the derivative of the screening function. Its expression will be given in the
following section.
C.4.2 Derivative of the screening function
The screening function is given by
S
ij
=
k
s
ijk
(C.21)
.
Dierentiating it with respect to the x-coordinate of atom a gives
S
ij
x
a
=
k
s
ijk
x
a
S
ij
s
ijk
(C.22)
.
By observing that this derivative is non-zero when i, j, k = a, only one terms
remains when the atom index a lies within the set of indices k, ie.
S
ij
x
a
=
s
ijk
x
a
S
ij
s
ijk
(C.23)
.
By writing a separate subroutine to calculate s
ijk
,
s
ijk
x
a
can be calculated by
nite dierences.
C.4.3 Derivative of the Embedding Energy
The derivative of the embedding energy is
E
embed
x
a
=
i
F
i

i
_
(0)
x
a
+
(0)
G
x
a
_
(C.24)
Note that the force on atom a has to take into account the embedding function
APPENDIX C 219
for all atoms due to the three-body screening parameter within
(0)
and . While
G
is easy to obtain,

x
a
is complicated due to the design of the partial electron
densities. The derivative of the partial electron density is introduced in section
C.4.4. This is needed for

x
a
which will be given in section C.4.5.
C.4.4 Derivative of the partial electron densities
Overall Derivative
In this section, the derivative for the rst partial electron density will be illustrated,
as the expressions for the other terms are similar. Let a
represent the direction

cosine terms. Then it can be written for the rst partial electron density
_
(1)
_
2
=
j
a
S
ij
d
(1)
j
_
2
(C.25)
and its derivative is
(1)
_
2
x
a
=
2
_
j
a
S
ij
d
(1)
j
__
a
d
(1)
j
S
ij
x
a
+S
ij
d
(1)
j
a
x
a
+a
S
ij
d
(1)
j
x
a
_
(C.26)
Similarly, the calculation of this term can be separated into two parts, one
holding S
ij
constant, and the other considering the derivative of S
ij
.
Direction Cosines
The derivative of the direction cosines can then be obtained. Consider the following
term as an example
a
x
=
x
ij
r
ij
(C.27)
APPENDIX C 220
with the individual terms being dened as
x
ij
= x
i
x
j
(C.28)
r
ij
=
_
x
2
ij
+y
2
ij
+z
2
ij
(C.29)
In these equations, x
i
is the x-coordinate of atom i. Then
a
x
x
i
=
1
r
ij
x
2
ij
r
3
ij
(C.30a)
a
x
y
i
=
y
2
ij
r
3
ij
(C.30b)
a
x
z
i
=
z
2
ij
r
3
ij
(C.30c)
C.4.5 Derivative of
With the derivatives of the partial electron densities, the derivative of ( Equation
(3.21) ) is
i
x
a
=
3
h=1
t
(h)
(
(h)
)
2
x
a
(
(0)
)
2
2
i
(0)
x
a
(0)
(C.31)
This expression concludes this section on the MEAM force expression.
221
Appendix D
Formation Energy of Cu
6
Sn
5
with
DFT
The formation energy at 0 K is calculated as follows. Geometry optimizations are
performed on the ground state unit cells of Cu, Sn (i.e. -Sn) and the B8 structure
of CuSn. Using both the GGA and the LDA, the convergence was examined with
two accuracy settings. The outcome of the Geometry Optimizations are the DFT
energy of the unit cell per formula unit (f.u.) E
d
. The change in energy per atom
E
0
is then given by
E
0
=
1
2
[E
d,CuSn
E
d,Cu
E
d,Sn
] (D.1)
The numerical results from these equations are shown in Table D.1. A range
from -0.103 eV/atom to -0.163 eV/atom is found, which is -9.93 kJ/mol to -15.73
kJ/mol. This is not in agreement with Ghoshs results of -3.2 kJ/mol to -4.0
kJ/mol [85]. One possible reason for this are the dierences in DFT software used
and the choice of the pseudopotential. It must be noted that the inconsistency
is limited to the formation energy. Predictions by Ghosh of the lattice constants
were found to be in good agreement.
APPENDIX D 222
Table D.1: DFT energies per atom for Cu (FCC) , Sn (DC) and CuSn (B8). The
energy change per atom (E
0
) and cohesive energy (E
coh
) are then derived from
them.
xc func-
tional
E
d,Cu
(eV/f.u.)
E
d,Sn
(eV/f.u.)
E
d,CuSn
(eV/f.u.)
E
0
(eV/atom)
GGA-1 -1352.93580 -97.98137 -1451.12352 -0.103
GGA-2 -1352.95687 -98.00744 -1451.17037 -0.103
LDA-1 -1349.66483 -97.47340 -1447.46404 -0.163
LDA-2 -1349.68385 -97.53613 -1447.51244 -0.146
Accuracy settings. 1: 320eV, 7 7 6, 2: 600eV, 14 14 10
f.u. - formula unit
223
Appendix E
Negative Elastic Constants for a
monoclinic crystal
E.1 Introduction
In Tables 2.17 and Figure 2.10, the elastic constant c
15
for Cu
6
Sn
5
and c
15
, c
35
for
Ni
3
Sn
4
were found to be negative. Typically, in crystals of orthorhombic or higher
symmetry, all the entries in the stiness matrix will be positive. This is so that
the crystal will be mechanically stable. Therefore, it seems out of order to nd
constants with negative values in the stiness matrix for these materials. This
appendix will explain why these negative values do not pose a problem. Using the
terminology by Wagner et al., the elastic constants c
15
, c
25
, c
35
and c
46
are termed
as non-orthorhombic elastic constants [235].
E.2 Literature Review
Negative values of the elastic constants from experimental studies have been re-
ported. McNeil et al. reported negative values of c
15
and c
35
for the elastic con-
stants of As
2
S
3
[236]. Furthermore, the CRC Handbook of Chemistry and Physics
APPENDIX E 224
contains data on the elastic constants of monoclinic crystals, some of which have
negative non-orthorhombic elastic constants [237].
Negative values have also been reported in other DFT studies. Wagner et al.
studied various forms of the intermetallic compound NiTi [235]. For one form with
a monoclinic crystal structure, a negative c
35
was obtained, with the rest of the
elastic constants being positive. An analysis of the stability criteria found that
this particular form remained mechanically stable.
In summary, negative values of the non-orthorhombic elastic constants have
been reported in the scientic literature, for both experimental studies and DFT
calculations.
E.3 Eigenvalues
The stability requirement for the elastic constants has described in section 2.3.3.
In short, for deformations of the crystal to result in a positive energy change, the
eigenvalues of the stiness matrix must be positive.
For the constants given in Figure 2.10, the eigenvalues for Cu
6
Sn
5
are found
to be 37.2, 43.9, 53.1, 89.6, 108.36 and 287.4. Thus, it is proven that the elastic
constants suggest a mechanically stable crystal, in spite of the negative value of
c
15
.
E.4 Implications
The negative value of c
15
found for Cu
6
Sn
5
is considered in this section. A negative
c
15
means that a shear strain
13
would result in a negative normal stress
11
. In
Cu
6
Sn
5
, this can be explained by observing the arrangements of the planes of
atoms. The (100) plane in Cu
6
Sn
5
is shown in Figure E.1(a). Clearly dened
planes of atoms can be seen and they are schematically represented by dashed
APPENDIX E 225
Figure E.1: (a) (010) plane of Cu
6
Sn
5
(b) Schematic diagram of the (010) plane.
The dashed lines represent the planes of atoms seen in (a) (c) Upon application
of the shear strain
13
, the unit cell deforms according to the red dotted line. The
angle is reduced (d) Bottom left corner of the unit cell. BC is shortened when
is reduced.
lines in Figure E.1(b). When a shear strain
13
is applied, the unit cell changes
shape according to the red dotted line in Figure E.1(c). This means that the angle
is reduced, thus shortening the dashed lines. Eectively, the planes of atoms
represented by the dashed lines are compressed in the 1-direction. This can be
illustrated further in Figure E.1(d), which shows the bottom left corner of the unit
cell. When decreases, side AC goes nearer to the vertical and hence side BC
reduces in length. Thus, upon application of a shear strain
13
, certain planes of
atoms become compressed, and this explains the negative value of the constant
c
15
.
APPENDIX E 226
E.5 Summary
Negative values of the non-orthorhombic elastic constants for Cu
6
Sn
5
and Ni
3
Sn
4
have been found from DFT calculations in this work. Such negative values have
also been found in the scientic literature. This appendix has shown that a neg-
ative value of c
15
for Cu
6
Sn
5
has a physical cause. Hence, Negative values of the
non-orthorhombic elastic constants do not suggest a mechanically unstable crystal.
227
Appendix F
Review of experimental work
F.1 Experimental Methods
The experimental studies done to date to nd the Youngs Modulus are listed in
Table F.1 for both Cu
6
Sn
5
and Cu
3
Sn. A variety of methods have been used to
make samples of the intermetallics to be tested, such as casting and annealing, dif-
fusion couples, powder metallurgy and thin lms from electroplating and vacuum
deposition.
Casting and annealing involves mixing amounts of Cu and Sn in the right
proportions. The mixture is then heated for a period of time to cause the chemical
reaction to occur so as to obtain the intermetallic compound. Annealing is also
done to ensure that the resulting sample is homogenous.
With casting and annealing, more than one phase could be present even after
annealing. To overcome this, Fields et al. made ne powders of Cu
6
Sn
5
by reacting
small drops of liquid Cu and Sn in an inert atmosphere[19]. A bulk sample was
then made by compressing the powders at high pressure, at temperatures that do
not cause any further chemical reaction.
A bulk sample does not reproduce the conditions at the solder joint.In order
to do so, the diusion couple was used. This involves sandwiching a solder layer
APPENDIX F 228
loading
unloading
ndentation Load
P
ndentation Depth
h
(a) (b)
Figure F.1: (a) nanoindentation load-displacement curve (b) Berkovich indenter
between two layers of Cu at temperatures of about 300

C to obtain a thin layer
of the intermetallic compound. The sample is then cross-sectioned to reveal the
layers, and nanoindentation is used to obtain the properties of the intermetallic
layer.
From a bulk sample, most authors used the resonance method. This involves
propagating a stress wave in the sample, and by measuring the speed of this wave
v, the Youngs Modulus E can be obtained from the expression
E = v
2
(F.1)
where is the density of the material. In the case of Ostrovskaya et al. [50], as a
thin lm sample was used, transverse vibrations of the sample was used instead.
To measure the properties of the intermetallic layer in the diusion couple,
nanoindentation is used [238]. In nanoindentation, a diamond tip is pressed into
the surface of a material with increasing load to a user-specied maximum value,
APPENDIX F 229
before being withdrawn. The shape of the tip used in the nanoindentation studies
in Table F.1 is the Berkovich indenter, which is a three-sided pyramid as shown
in Figure F.1(b). As the load increases, the penetration or displacement into the
sample increases. After the test is completed, a load-displacement curve such
as the one shown in Figure F.1(a) is obtained. The indentation modulus E
r
is
obtained by
E
r
=
1
2
A
dP
dh
(F.2)
where
dP
dh
is the gradient of the unloading curve at maximum load as shown in
Figure F.1(a). A is the contact area of indentation. Together with known values
of the Youngs Modulus E
i
and the Poissons Ratio
i
of the tip, the Youngs
Modulus of the sample E can be obtained as follows
1
E
r
=
1
2
E
+
1
2
i
E
i
(F.3)
This expression also requires the Poissons ratio of the sample, which is ap-
proximated to be 0.33 by Deng et al. [41] and Chromik et al. [44].
F.2 Discussion
From the information in Table F.1, it can be seen that there is little agreement in
the experimental values for both Cu
3
Sn and Cu
6
Sn
5
. The dierence between the
smallest value and largest value is over 50 GPa for both compounds.
However, some trends can be discerned. For both Cu
6
Sn
5
and Cu
3
Sn, the re-
sults from resonance measurement and compression testing on bulk samples pro-
duced lower values compared to those from nanoindentation. This seems intuitive
as bulk samples will usually contain defects such as voids and impurities that will
aect the results, while nanoindentation probes the properties of the materials
APPENDIX F 230
at such small length scales such that these imperfections are not likely to aect
the results. The studies by Chromik et al.[44], Deng et al. [41], Jang et al. [47]
and Yang et al. [54] show good agreement with each other. These three studies
are similar in performing nanoindentation on the intermetallic layer in diusion
couples.
The DFT calculations in Chapter 2 predict that the Youngs Modulus of Cu
3
Sn
> Cu
6
Sn
5
. This is also what all of the nanoindentation studies and some of the
studies on bulk samples in Table F.1 show. On the other hand, the studies by
Cabarat et al. and Subrahmanyan et al. have similar values for Cu
6
Sn
5
and Cu
3
Sn.
Thus, Fields et al. raised concerns that their samples could have contained more
than one phase due to the sample preparation process [19].
APPENDIX F 231
T
a
b
l
e
F
.
1
:
E
x
p
e
r
i
m
e
n
t
a
l
w
o
r
k
o
n
t
h
e
Y
o
u
n
g
s
M
o
d
u
l
u
s
o
f
C
u
6
S
n
5
a
n
d
C
u
3
S
n
.
A
l
l
v
a
l
u
e
s
a
r
e
i
n
G
P
a
.
A
u
t
h
o
r
S
a
m
p
l
e
p
r
e
p
a
r
a
t
i
o
n
m
e
t
h
o
d
M
e
a
s
u
r
e
m
e
n
t
m
e
t
h
o
d
R
e
s
u
l
t
s
,
C
u
6
S
n
5
R
e
s
u
l
t
s
,
C
u
3
S
n
F
i
e
l
d
s
e
t
a
l
.
,
R
e
f
.
[
1
9
]
P
o
w
d
e
r
m
e
t
a
l
l
u
r
g
y
,
B
u
l
k
C
o
m
p
r
e
s
s
i
o
n
t
e
s
t
i
n
g
8
5
.
6
1
0
8
S
u
b
r
a
h
m
a
n
y
a
n
,
R
e
f
.
[
4
6
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
R
e
s
o
n
a
n
c
e
8
5
8
0
C
a
b
a
r
a
t
e
t
a
l
.
,
R
e
f
.
[
4
9
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
R
e
s
o
n
a
n
c
e
1
0
2
.
4
1
0
6
.
1
O
s
t
r
o
v
s
k
a
y
a
e
t
a
l
.
,
R
e
f
.
[
5
0
]
V
a
c
u
u
m
D
e
p
o
s
i
t
i
o
n
o
f
T
h
i
n
F
i
l
m
R
e
s
o
n
a
n
c
e
1
0
2
1
5
3
A
l
b
r
e
c
h
t
e
t
a
l
.
,
R
e
f
.
[
4
5
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
R
e
s
o
n
a
n
c
e
9
0
1
2
0
C
h
r
o
m
i
k
e
t
a
l
.
,
R
e
f
.
[
4
4
]
D
i
u
s
i
o
n
c
o
u
p
l
e
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
1
9
1
4
3
A
l
b
r
e
c
h
t
e
t
a
l
.
,
R
e
f
.
[
4
5
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
3
0
1
5
0
D
e
n
g
e
t
a
l
.
,
R
e
f
.
[
4
1
]
D
i
u
s
i
o
n
c
o
u
p
l
e
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
1
2
.
3
1
3
4
J
a
n
g
e
t
a
l
.
,
R
e
f
.
[
4
7
]
D
i
u
s
i
o
n
c
o
u
p
l
e
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
1
6
.
3
,
1
2
5
1
3
5
.
7
J
a
n
g
e
t
a
l
.
,
R
e
f
.
[
4
7
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
1
4
.
9
1
2
1
.
7
G
h
o
s
h
,
R
e
f
.
[
4
2
]
C
a
s
t
i
n
g
a
n
d
a
n
n
e
a
l
i
n
g
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
9
6
.
9
1
2
3
T
s
a
i
e
t
a
l
.
,
R
e
f
.
[
5
1
]
E
l
e
c
t
r
o
p
l
a
t
e
d
f
o
i
l
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
2
4
1
4
3
A
l
b
r
e
c
h
t
e
t
a
l
.
,
R
e
f
.
[
5
3
]
D
i
u
s
i
o
n
c
o
u
p
l
e
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
8
0
,
9
0
,
1
1
0
1
1
0
,
1
3
0
Y
a
n
g
e
t
a
l
.
,
R
e
f
.
[
5
4
]
D
i
u
s
i
o
n
c
o
u
p
l
e
N
a
n
o
i
n
d
e
n
t
a
t
i
o
n
1
1
6
.
9
1
3
3
.
4

Lee T Iong Seng Norman

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Lee T Iong Seng Norman

Diunggah oleh

Hak Cipta:

Format Tersedia

ATOMISTIC CALCULATIONS OF THE MECHANICAL

PROPERTIES OF Cu-Sn INTERMETALLIC COMPOUNDS

anti-clockwise from the positive

anti-clockwise from the positive

m. (a) Size Small (b) Size Large. . . . 167

m using circles to represent

m using circles to repre-

b Reciprocal lattice vector

G Reciprocal lattice vector

H Hamiltonian operator in Schr odingers Equation

k position vector of points in the Brillouin zone

Direction cosine in the axis direction

rotations. As such, dierent crystal structures in nature have dierent symmetry

and the bulk modulus B

A, and (iii) change in

, which corresponds closely to the [210]

with the positive x-axis and corre-

from the positive x-axis in the xz-plane. This corresponds to

All angles are measured anticlockwise from the positive x-axis

1] direction are highlighted in the unit cell at

anti-clockwise from the positive

anti-clockwise from the positive

will not exceed

E), and shown in Figure 2.19. The Youngs

A. At 300 K, this comes to approximately 5%

summation over the three coordinate directions.

2r). If the interaction

is the reduced interatomic distance and is set to be zero at the equilibrium

is dened in equation (3.24). Given the Bulk Modulus B, equilibrium volume

3a. Cell A has sides parallel to the a-axis

clockwise from Cell A. Due to its orientation, it can used for

3a 4c was used, consisting of a total of 64 cells of Cell A or

(c) B8 Unit cell under

m. These ve studies employ two methods, the buckling of thin lms

A) and the unit of

Figure 4.4: Periodic Crack

Sample a/W (GPa) K

m for a crack lying in the ac plane.

m. The results for K

m will be shown in this section. Figure 4.11(a) shows the

m, the branched crack is seen to propagate but

m, the two branched cracks

m was repeated using size Large, shown in Figure 4.12(b). The

m. Using a conguration with starting K

m. (a) Size Small (b) Size Large.

m using circles to represent

m onwards. Results for

m, the crack propagates but stops short of the free surface,

m, and in the ac plane, the K

m. (a) 3100 time steps

m using circles to represent

m obtained by the periodic crack method. Combining the results of

m(Ref [206]). In the experimental study by Balakrisnan et

m. This value cannot be validated by the experimental results yet. The

m, which is on the lower end

rotational symmetry as the angles within

. Neglecting higher powers of strain, the new

represent the direction

Anda mungkin juga menyukai