Understanding and Improving the Environmental Dependent Tribology and Thermal Stability of Hydrogenated Amorphous Carbon by Using Silicon and Oxygen As Dopants

University of Pennsylvania ScholarlyCommons

Publicly Accessible Penn Dissertations

2018

Understanding And Improving The Environmental Dependent Tribology And Thermal Stability Of Hydrogenated Amorphous Carbon By Using Silicon And Oxygen As Dopants

James Hilbert University of Pennsylvania, [email protected]

Follow this and additional works at: https://repository.upenn.edu/edissertations

Part of the Mechanical Engineering Commons

Recommended Citation Hilbert, James, "Understanding And Improving The Environmental Dependent Tribology And Thermal Stability Of Hydrogenated Amorphous Carbon By Using Silicon And Oxygen As Dopants" (2018). Publicly Accessible Penn Dissertations. 3128. https://repository.upenn.edu/edissertations/3128

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/edissertations/3128 For more information, please contact [email protected]. Understanding And Improving The Environmental Dependent Tribology And Thermal Stability Of Hydrogenated Amorphous Carbon By Using Silicon And Oxygen As Dopants

Abstract Hydrogenated amorphous carbon (a-C:H) films are a class of non-crystalline materials composed of hydrogen and carbon bonded in both sp2 and sp3 configurations. These films are notable for their high hardness (10–16 GPa), low roughness, chemical inertness, and good tribological performance (low friction and wear). This combination of favorable properties has led to promising applications in diverse fields, including automotive and aerospace components, biomedical devices, and computer hard drives. However, a-C:H becomes unstable above 150 °C, preventing its use in important technological applications, such as heat assisted magnetic recording (HAMR) disk drives. Also, the low friction and wear is only maintained in dry and vacuum conditions. To understand and address these limitations, the effect of adding silicon and oxygen to a-C:H films is considered, since prior experimental evidence shows that this can significantly increases thermal stability, and help maintain low friction and wear in humid environments. However, the mechanisms and extent of these improvements are unknown. Friction and wear testing were performed on a-C:H doped with Si and O (a-C:H:Si:O) in a range of environments. Friction coefficientsaried v from approximately 0.05 in dry environments (RH < 5%) to 0.15 in humid air, better than prior observations for undoped a-C:H films. The friction and wear behavior is controlled by adhesive interactions leading to the development of transfer films on the steel counterface. Possible mechanisms underlying this behavior are discussed. Annealing experiments showed significant improvements in thermal stability up to 450 °C. In order to understand the atomistic origins of this enhanced thermal stability, reactive molecular dynamics (MD) simulations using the ReaxFF potential were performed. The primary thermal degradation pathway in undoped a-C:H was observed to be the breaking of tensile strained C-C bonds. The presence of Si suppresses this mechanism by decreasing the frequency of highly strained C-C bonds in the unannealed structure. This is due to the longer C-Si equilibrium bond length compared to C-C bonds. The activation energy for rehybridization could be modeled using the same methods as in prior experiments and produced good agreement between the experimental and simulation results.

Degree Type Dissertation

Degree Name Doctor of Philosophy (PhD)

Graduate Group Mechanical Engineering & Applied Mechanics

First Advisor Robert W. Carpick

Keywords Diamond-like carbon, Molecular dynamics, Thermal stability, Tribology

Subject Categories Mechanical Engineering

This dissertation is available at ScholarlyCommons: https://repository.upenn.edu/edissertations/3128 UNDERSTANDING AND IMPROVING THE ENVIRONMENTAL DEPENDENT

TRIBOLOGY AND THERMAL STABILITY OF HYDROGENATED AMORPHOUS

CARBON BY USING SILICON AND OXYGEN AS DOPANTS

James Andrew Hilbert

A DISSERTATION

Mechanical Engineering and Applied Mechanics

Presented to the Faculties of the University of Pennsylvania

Partial Fulfillment of the Requirements for the

Degree of Doctor of Philosophy

2018

Supervisor of Dissertation

Robert W. Carpick, John Henry Towne Professor and Chair, Mechanical Engineering and Applied Mechanics

Graduate Group Chairperson

Kevin Turner, Professor, Mechanical Engineering and Applied Mechanics

Dissertation Committee Jennifer R. Lukes, Professor, Mechanical Engineering and Applied Mechanics David J. Srolovitz, Joseph Bordogna Professor, Materials Science and Engineering Julien Fontaine, CNRS Researcher, Laboratoire de Tribologie et Dynamique des Syst`emes UNDERSTANDING AND IMPROVING THE ENVIRONMENTAL DEPENDENT

TRIBOLOGY AND THERMAL STABILITY OF HYDROGENATED AMORPHOUS

CARBON BY USING SILICON AND OXYGEN AS DOPANTS

2018

James Andrew Hilbert ACKNOWLEDGEMENTS

I would like to thank the many people who contributed to and supported this research.

Most importantly, I thank my advisor, Professor Robert Carpick, for his amazing advice, guidance, and patience in pursuing my Ph.D. His support has been invaluable in this en- deavor.

I am also grateful to Professor Jennifer Lukes for her collaboration on this project and mentoring me in computer simulations and Dr. Julien Fontaine for his advice and hosting me during a wonderful study abroad at Ecole´ Centrale de Lyon. I thank Professor David

Srolovitz as well for his great feedback on this research and serving on my thesis committee.

I would also like to thank all of the current and past members of the Carpick Research

Group, who have helped me in my work and been excellent friends and colleagues. In particular, I thank Professor Filippo Mangolini and Dr. Brandon McClimon, who collab- orated on this project with me, for their excellent work and feedback. Also, Dr. Komlavi

Medard Koshigan and Dr. Nick Garabedian have contributed greatly to this research as well through discussions and experimental assistance.

I also thank Dr. Matthew Brukman for his support in use of instrumentation at the Singh

Center for Nanotechnology, and Dr. Mehdi Zanjani and Paul Barclay for their willingness to help me with use of computer resources for simulations.

The staff of the Mechanical Engineering and Applied Mechanics department, Maryeileen

Griffith, Peter Litt, Sue Waddington Pilder, and Desirae Johnson have also provided great support in logistics for this work.

This work was also supported by funding from the National Science Foundation, Grant

No. DMR-1107642 and Advanced Storage Technology Consortium [ASTC] (Grant No.

2011-012), as well as through use of the Singh Center for Nanotechnology, supported by NSF

Grant ECCS-1542153. Also funding from Coopérationet MobilitéInternationales Rhône-

Alpes (CMIRA) 2015 and Programme Avenir Lyon Saint-Etienne (PALSE) for funding covering travel to Ecole´ Centrale de Lyon is acknowledged.

iii ABSTRACT

UNDERSTANDING AND IMPROVING THE ENVIRONMENTAL DEPENDENT

TRIBOLOGY AND THERMAL STABILITY OF HYDROGENATED AMORPHOUS

CARBON BY USING SILICON AND OXYGEN AS DOPANTS

James A. Hilbert

Robert W. Carpick

Hydrogenated amorphous carbon (a-C:H) films are a class of non-crystalline materials composed of hydrogen and carbon bonded in both sp2 and sp3 configurations. These films are notable for their high hardness (10–16 GPa), low roughness, chemical inertness, and good tribological performance (low friction and wear). This combination of favorable properties has led to promising applications in diverse fields, including automotive and aerospace components, biomedical devices, and computer hard drives. However, a-C:H becomes unstable above 150 ◦C, preventing its use in important technological applications, such as heat assisted magnetic recording (HAMR) disk drives. Also, the low friction and wear is only maintained in dry and vacuum conditions. To understand and address these limitations, the effect of adding silicon and oxygen to a-C:H films is considered, since prior experimental evidence shows that this can significantly increases thermal stability, and help maintain low friction and wear in humid environments. However, the mechanisms and extent of these improvements are unknown. Friction and wear testing were performed on a-C:H doped with Si and O (a-C:H:Si:O) in a range of environments. Friction coefficients varied from approximately 0.05 in dry environments (RH < 5%) to 0.15 in humid air, better than prior observations for undoped a-C:H films. The friction and wear behavior is controlled by adhesive interactions leading to the development of transfer films on the steel counterface. Possible mechanisms underlying this behavior are discussed. Annealing experiments showed significant improvements in thermal stability up to 450 ◦C. In order to understand the atomistic origins of this enhanced thermal stability, reactive molecular dynamics (MD) simulations using the ReaxFF potential were performed. The primary thermal degradation pathway in undoped a-C:H was observed to be the breaking of tensile strained C-C bonds.

iv The presence of Si suppresses this mechanism by decreasing the frequency of highly strained

C-C bonds in the unannealed structure. This is due to the longer C-Si equilibrium bond length compared to C-C bonds. The activation energy for rehybridization could be modeled using the same methods as in prior experiments and produced good agreement between the experimental and simulation results.

v TABLE OF CONTENTS

ACKNOWLEDGEMENTS ...... iii

ABSTRACT ...... iv

TABLE OF CONTENTS...... vi

LIST OF TABLES...... ix

LIST OF ILLUSTRATIONS ...... x

CHAPTER 1: Introduction ...... 1

1.1 Classification of DLCs...... 2

1.2 Historical background and synthesis ...... 5

1.3 Applications for DLC ...... 6

1.4 Overview of the theories for tribological behavior in DLC...... 8

1.5 Thermal stability of DLC ...... 16

1.6 Heat-assisted magnetic recording...... 17

1.7 Improving the tribology and thermal stability of a-C:H through the addition

of Si and O ...... 19

1.8 Investigating the tribological behavior and thermal stability of a-C:H:Si:O . 23

CHAPTER 2: Tribological behavior of a-C:H:Si:O as a function of environment and

load...... 26

2.1 Experimental samples ...... 27

2.2 Description of gas blowing tribometer ...... 28

2.3 Basic principles of atomic force microscopy ...... 29

2.4 Brief description of the Hertz theory of contact mechanics...... 33

2.5 Environmental dependent tribology of a-C:H:Si:O...... 35

vi 2.6 Wear track topography measured with AFM...... 40

2.7 Importance of transfer film in achieving low friction for a-C:H:Si:O . . . . . 44

2.8 Load dependent tribology of a-C:H:Si:O in humid air...... 50

2.9 Investigating changes in film structure through Raman spectroscopy . . . . 55

2.10 Modeling wear of a-C:H:Si:O as a function of the energy dissipated by friction 57

2.11 Conclusions...... 62

CHAPTER 3: Experimental investigation into the thermal stability of a-C:H:Si:O . 64

3.1 Raman spectra of carbon ...... 64

3.2 Raman spectroscopic analysis of annealed a-C:H and a-C:H:Si:O ...... 69

3.3 Review of XPS and NEXAFS experiments...... 73

3.4 Nanoscale heating with thermal AFM probes...... 76

3.5 Conclusions...... 83

CHAPTER 4: Molecular dynamics simulations of a-C:H:Si:O using the ReaxFF po-

tential...... 87

4.1 Description of the ReaxFF potential...... 90

4.2 Formation of a-C:H:Si:O through liquid quenching ...... 92

4.3 Validation of the ReaxFF potential and simulation procedure ...... 95

4.4 Atomic structure of a-C:H:Si:O...... 96

CHAPTER 5: Atomistic origins of enhanced thermal stability from Si and O . . . 100

5.1 Constant temperature annealing of a-C:H and a-C:H:Si:O ...... 100

5.2 Si doping increases thermal stability through a reduction in bond length

disorder...... 103

5.3 Changes in Si bonding at high temperature ...... 105

5.4 Clustering and ordering of sp2 carbon ...... 106

5.5 Effect of varying Si content on the structure and thermal stability . . . . . 107

5.6 Comparison between simulations and experimental results ...... 108

5.7 Modeling the distribution of activation energies from heating ramps . . . . 110

vii 5.8 Effect of stress on activation energy distribution ...... 115

5.9 Conclusions...... 117

CHAPTER 6: Conclusions...... 121

6.1 Main results and implications for the design and use of a-C:H:Si:O coatings 121

6.1.1 Environmental dependence of the friction and wear for a-C:H:Si:O . 121

6.1.2 Dependence of the friction and wear on load ...... 123

6.1.3 a-C:H:Si:O as a thermally stable alternative to a-C:H ...... 125

6.1.4 Atomistic origins of the enhanced thermal stability through the addi-

tion of silicon...... 126

6.2 Future work...... 128

6.2.1 Tribometer testing in well controlled humidity ...... 128

6.2.2 Tribometer testing with different balls ...... 128

6.2.3 Localized mapping of mechanical and structural properties in wear

tracks ...... 129

6.2.4 Experimental investigation of the effect of stress on thermal stability 131

6.2.5 MD simulations of thermal stability with oxygen or water atmosphere 131

6.2.6 Simulating the tribological behavior of a-C:H:Si:O ...... 132

BIBLIOGRAPHY ...... 133

viii LIST OF TABLES

TABLE 1.1: Comparison of properties for single crystal diamond, ta-C, a-C:H [1,

3, 18]...... 4

TABLE 1.2: Typical friction coefficients for CVD diamond, a-C, and a-C:H coat-

ings in different environments [10]...... 11

TABLE 1.3: Comparison of properties for a-C:H:Si:O with a-C:H [3, 88]. The

composition of a-C:H:Si:O can vary greatly depending on deposition

methods. Typically films contain 5–35 at. % Si and 3–30 at.%O. 22

TABLE 2.1: Summary of coefficient of friction values for a-C:H:Si:O in humid air

(30-50% RH) vs. steel from prior work...... 26

TABLE 2.2: Atomic composition of a-C:H:Si:O samples used in this thesis. . . . 27

TABLE 3.1: G peak position and intensity ratios (I(D)/I(G)) for Raman spectra

acquired from annealed a-C:H and a-C:H:Si:O...... 69

TABLE 4.1: Bonds between elements in simulated a-C:H:Si:O before annealing. 97

ix LIST OF ILLUSTRATIONS

FIGURE 1.1: Schematic diagrams showing the sp3 and sp2 hybridization states

for carbon. In the sp3 hybridization state, carbon where it forms

4 σ-bonds in a tetrahedral arrangement. The solid and striped

wedges represent bonds going into and out of the page. In the sp2

hybridization state, carbon forms three in plane σ-bonds and the

fourth valence electron is part of a delocalized π-bond...... 3

FIGURE 1.2: Ternary diagram showing different forms of DLC films based on

relative concentrations of sp2 to sp3-bonded carbon and hydro-

gen [18, 20]. There are three main regions shown. First, along

the leftmost axis is hydrogen free amorphous carbon, ranging from

mostly sp2 to mostly sp3 bonding. Films with low sp3 fraction are

referred to as amorphous carbon (a-C), while films with high3 sp

fraction are referred to as tetrahedral amorphous carbon (ta-C).

Second, at the bottom right of the phase diagram are compounds

with such high hydrogen content that the material cannot form a

connected network, but instead remains as gas molecules. Finally,

in the center is hydrogenated amorphous carbon (a-C:H), which

can range from materials with only 20% H content up to materials

where close to 60% of the atoms are hydrogen...... 4

FIGURE 1.3: sp3 fraction vs. ion energy for ta-C. The data points are experimen-

tal observations, while the line is a theoretical prediction based on

the subplantation process, where an atom penetrates the surface re-

sulting in a local density increase and residual compressive stress.

Local relaxation at the impact site is assumed to take place during

a thermal spike lasting only a few picoseconds [19]...... 6

x FIGURE 1.4: (a) Schematic showing the design of modern hard disks. The mag-

netic spacing is the distance between the magnetic layer and the

read/write elements. This distance includes the air gap, a molec-

ular lubricant layer, and a DLC coating on both the head and

disk. Typically ta-C is used on the head, and a-C:H on the disk.

Maximum achievable storage density is a function of the magnetic

spacing, as shown in (b). This requires reducing the thickness of

the DLC coatings to only a few nm while maintaining uniform cov-

erage. Figures from ref [8] ...... 9

FIGURE 1.5: Schematic of the three general origins of macroscopic friction forces,

abrasion, shearing, and adhesion [38]...... 10

FIGURE 1.6: Schematic diagram showing the general trends in the frictional be-

havior of DLC as a function of humidity [10]...... 11

FIGURE 1.7: Steady state coefficient of friction in vacuum for several different a-

C:H films as a function of the viscoplastic exponent. A correlation

is found to exist between the films viscoplastic character and the

coefficient of friction, and films with mostly plastic behavior cannot

achieve superlow friction [50]...... 14

FIGURE 1.8: Schematic of the HAMR process. Heating the magnetic material

reduces its coercivity, allowing recording. The material then cools

rapidly to the storage temperature, locking in the recorded data [77]. 18

FIGURE 1.9: Schematic of the structure of a-C:H:Si:O. The material is made

of two networks, one of (a-C:H) and the other of (a-Si:O). The

connection between the two networks is proposed to be primarily

Si-C bonds [14, 89, 96]...... 21

xi FIGURE 2.1: (a) Image showing the setup of the gas blowing tribometer used

for measurements of friction and wear in different environments. A

schematic of the setup is shown in (b). Normal force was applied via

weights and an actuator controlled the sample motion. Two nozzles

directed the flow of gas at 5L/min to the contact perpendicular to

the sliding direction to control the environment...... 28

FIGURE 2.2: Optical microscope image showing an example of a multisection

wear track produced in dry N2. The number of sliding cycles for each section is labeled. The total length of the wear track is 3 mm.

The image was produced by stitching together individual high mag-

nification images...... 29

FIGURE 2.3: Top: Diagram of contact mode AFM setup. A cantilever with a

sharp tip is in contact with the sample surface. A laser is focused on

the back of the cantilever and reflects into a photodetector. Normal

forces move the cantilever vertically, while lateral forces twist the

cantilever left and right. The sample is moved horizontally with

a piezoelectric scanner and the tip is rastered over the surface to

produce an image [101] Bottom: Image of the end of a typical AFM

cantilever showing the sharp tip. Tip radii can be < 10 nm. . . . . 30

FIGURE 2.4: Coefficient of friction vs. sliding cycles obtained in three different

environments, laboratory air (40% RH), dry air (< 5% RH), and

dry nitrogen (< 5% RH) for NCD Technologies a-C:H:Si:O. In each

environment the coefficient of friction initially drops during arun-

in period, followed by stabilization. Friction is lowest in dry N2 and highest in humid air. The arrows indicate the approximate location

where the film began to be worn through, based on measurements

of maximum wear depth. As the films begin to wear through the

coefficient of friction becomes unstable and begins to increase. . . 36

xii FIGURE 2.5: Optical microscope images of tribofilms on steel balls formed in

(a) laboratory air (40% RH), (b) dry air (< 5% RH), and (c) dry

nitrogen (< 5% RH). The formation of tribofilms is known to be

important for achieving low friction and wear...... 37

FIGURE 2.6: Wear track dimensions measured with AFM for laboratory air (40%

RH), dry air (< 5% RH), and dry nitrogen (< 5% RH). (a) Max-

imum wear track depth as a function of sliding cycles. The total

a-C:H:Si:O film thickness was 40 nm (indicated by the gray line).

Maximum wear track depth increases with the number of cycles.

(b) Wear track width as a function of sliding cycles. The wear

tracks formed in dry environments have similar width, which is

significantly larger than the wear tracks formed in humid air.The

wear track width is primarily determined at the start of sliding

by contact mechanics, which are effected by the properties of the

tribofilms. In dry environments a wide transfer film formed, ex-

panding the contact diameter. (b) Wear track cross sectional area

as a function of sliding cycles. The overall amount of wear is great-

est in humid air and least in dry N2...... 39 FIGURE 2.7: AFM topography images of wear tracks formed on NCD Technolo-

gies a-C:H:Si:O in dry nitrogen (a) and dry air (b) after 50 sliding

cycles. Wear tracks in both dry environments appeared similar,

but with higher wear in dry air than dry N2. In both cases the maximum wear depth occurs at the side of the contact due to the

formation of narrow grooves along the edge of the wear track, while

the center of the wear track is a flat plateau. This unusual topog-

raphy is highly reproducible for a-C:H:Si:O...... 40

xiii FIGURE 2.8: Line profile from AFM topography measurement of wear track

formed in dry nitrogen after 50 sliding cycles. Large regions of

the wear track are raised compared to the as-received surface, indi-

cating either the formation of a tribofilm on the a-C:H:Si:O surface

or a structural change resulting in near surface expansion. . . . . 41

FIGURE 2.9: Magnified AFM scans of wear tracks produced in dry nitrogen after

50 sliding cycles. (a) Topography and (b) friction for a 5 × 5 µm2 area in the center of the wear track. Material removal occurs

at localized regions through the breaking of adhesive junctions.

There is higher friction inside the worn areas compared to the sur-

rounding surface. (c) Topography and (d) friction for a magni-

fied. 1 5 × 1.5 µm2 area surrounding a single wear site. The surface has significant texturing, with clear friction contrast. Interestingly,

other than the elongation of the wear events parallel to the sliding

direction, the modified wear track surface appears isotropic. There

is no clear indication of sliding direction...... 42

FIGURE 2.10: Magnified AFM scans of wear tracks produced in dry air after50

sliding cycles. (a) Topography and (b) friction for a 5 × 5 µm2 area in the center of the wear track. Magnified wear track topography

is similar to sliding in dry nitrogen, however, the relative ratio

of higher friction adhesive wear events to low friction wear track

surface is larger than in dry nitrogen. This is consistent with the

higher friction and increased wear observed when oxygen is present.

(c) Topography and (d) friction for a magnified. 1 5 × 1.5 µm2 area. 44 FIGURE 2.11: Wear track formed in humid air (40% RH) after 50 sliding cycles.

The wear tracks formed in humid air are significantly different than

what was observed in dry environments. The tracks are both deeper

and narrower, with different topography inside the wear track. . . 45

xiv FIGURE 2.12: Magnified AFM scans of wear tracks produced in humid air af-

ter 50 sliding cycles. (a) Topography scan of 5 × 5 µm2 area. (b)

1.5 × 1.5 µm2 topography scan of the same area. The detailed features on the wear track surface are significantly different in humid

air. The localized wear events and isotropic surface texture are

no longer present, instead the dominant topography is ridges and

grooves running parallel to the sliding direction...... 45

FIGURE 2.13: Coefficient of friction vs. sliding cycles obtained in three different

environments, laboratory air (40% RH), dry air (< 5% RH), and

dry nitrogen (< 5% RH) for DLN 360 a-C:H:Si:O. Humid air be-

havior is similar to previous measurements. However, the friction

in dry air rapidly becomes very high. Post sliding imaging of the

steel balls revealed no transfer film buildup. The lowest coefficient

of friction was measured in dry nitrogen, however, the behavior was

unstable...... 46

FIGURE 2.14: (a) Optical microscope image of tribofilm formed on the steel ball

after sliding in humid air on DLN 360 a-C:H:Si:O. (b) Steel ball

after sliding in dry air. No tribofilm is present, and there appears

to be wear of the steel flattening the end of the ball. There isa

ring of transfered material outside of the contact zone, suggesting

material transfer from the film to the ball did occur, but was not

strongly adherent enough to remain in the contact. This set of

conditions results in high friction and wear...... 47

FIGURE 2.15: Extreme example of the unstable frictional behavior of DLN 360

a-C:H:Si:O in dry nitrogen...... 47

xv FIGURE 2.16: Coefficient of friction vs. sliding cycles for DLN 360 a-C:H:Si:O in

dry nitrogen. The low friction behavior was highly unstable. In

two cases the test was stopped while the friction was high (solid

lines), and in two cases the test was stopped while the friction was

low (dashed lines). Imaging of the balls revealed that transfer films

were present for the test stopped while the friction was low, but

not for the tests stopped while the friction was high...... 48

FIGURE 2.17: (a) Optical microscope image of steel ball after sliding in dry nitro-

gen after the test was stopped when friction was high. Similar to

the case for dry air, no tribofilm is present in the contact. (b) Opti-

cal microscope image of steel ball after sliding in dry nitrogen after

the test was stopped when friction was low. In this case a tribofilm

is found on the ball. The contact area is much smaller and there

is no evidence of wear of the steel. An explanation for the friction

instability observed in dry nitrogen is that the tribofilms are only

weakly adherent to the steel. If a film is present, friction and wear

is low, however, the film can be removed from the ball, causing an

increase in friction. A new film can form however, which will cause

a return to low friction behavior...... 49

FIGURE 2.18: Coefficient of friction for DLN 360 a-C:H:Si:O sliding in humid air

(25–30% RH) as a function of load and cycles. The steady state

friction coefficient is observed to decrease with load. These tests

were also performed as multisection wear tracks; the spikes in the

data (which are most noticeable at low loads) are artifacts due to

the changing track length...... 50

xvi FIGURE 2.19: Steady state coefficient of friction for DLN 360 a-C:H:Si:O sliding

in humid air (25–30% RH) as a function of load. The data is well

fit with a power law. For a single asperity contact, the coefficient

of friction is expected to scale with real area of contact. Based

on Hertz contact mechanics this suggests the coefficient of friction

should scale as load to the −1/3 power. The discrepancy between

the observed behavior and Hertz model predictions is likely due to

underestimation of the real area of contact when using Hertz theory. 51

FIGURE 2.20: (a) AFM topography image of wear track formed with a 1000 mN

load after 100 cycles. (b) Line profile showing the wear track cross

section. Again the maximum wear depth occurs at the edge of the

contact in two narrow grooves, while the central region forms a

level plateau...... 52

FIGURE 2.21: (a) Wear track width as a function of cycles. The average width is

set by the contact mechanics, which depends on the properties of

the transfer film. The transfer film forms within the first 10cycles,

after which the contact diameter only increases slightly with slid-

ing cycles. (b) Wear track width as a function of load. The dashed

line shows the predicted contact diameter based on Hertz contact

mechanics. At low loads the wear track width closely follows the

Hertz model predictions, while at higher loads the model underes-

timates the contact diameter. This is most likely due to the softer

tribofilm being deformed at high loads causing it to spread outand

expand the contact area...... 54

xvii FIGURE 2.22: (a) Wear track median depth as a function of cycles. Due to the

presence of the previously discussed grooves on the sides of the wear

tracks, the median depth provides a more physically meaningful

parameter rather than the mean. As expected the depth scales

linearly with the number of cycles. (b) Wear track depth as a

function of load. The median depth is observed to be independent

of load...... 55

FIGURE 2.23: (a) Raman spectra showing as-received film (DLN 360 a-C:H:Si:O)

compared to wear track surfaces formed under different loads. Spec-

tra are generally similar, with small differences. In order to more

clearly see the changes, difference spectra between the wear track

surfaces and the as-received film are shown in (b). In general, an

increase in peak intensity and a shift to lower wavenumbers of the

peak position indicate an increase in sp2 carbon, while an increase

in the slope of the background indicates an increase in passivation

of dangling bonds or clustering of sp2 carbon...... 56

FIGURE 2.24: Raman spectra acquired from tribofilms formed on steel balls after

sliding DLN 360 a-C:H:Si:O in humid air at different loads com-

pared to the as-received film. There is a massive increase in inten-

sity under the same acquisition conditions, and no evidence of the

peaks associated with DLC. It is not possible to derive bonding

composition from these spectra, but the Raman data clearly in-

dicates that substantial chemical changes occur to the a-C:H:Si:O

that is worn away and incorporated into the tribofilm...... 57

xviii FIGURE 2.25: Wear track cross sectional area vs. load × number of cycles. Ar-

chard’s wear equation predicts that the cross section area should

be a linear function of load × number of cycles, where the slope

is the dimensional wear coefficient. This model predicts the data

should collapse on a single line. However, it is observed instead

that the wear coefficient is inversely proportional to the load. . . . 59

FIGURE 2.26: Wear volume vs. energy dissipated by friction. Modeling wear as

proportional to the energy dissipated by friction reasonably col-

lapses the wear volume data onto a single curve, although some

variation still remains...... 60

FIGURE 2.27: Wear volume vs. energy dissipated by friction for NCD Technologies

a-C:H:Si:O (a) and DLN 360 a-C:H:Si:O (b) in different environments. 61

FIGURE 3.1: Comparison of typical Raman spectra for several forms of car-

bon. The spectra are dominated by the G and D peaks, at about

1550 cm−1 and 1360 cm−1. The G peak represents the stretching

vibration of all sp2-bonds, while the D peak represents the breath-

ing mode of sp2 carbon rings [18, 113] In monocrystalline graphite

only the G peak is present, and the breathing mode vibrations are

not active. In microcrystalline-graphite the D peak breathing mode

vibrations become active due to the presence of grain boundaries.

In amorphous carbon a single peak is seen, which is a convolution

of the G and D peaks. As can be seen from the comparison of

Raman spectra for a-C, a-C:H, and ta-C, the shape of this peak is

dependent on the sp2/sp3 ratio, hydrogen content, and clustering

of the sp2 phase. Figure reproduced from Ref [18]...... 66

xix FIGURE 3.2: Schematic illustrating the change in G peak position and I(D)/I(G)

ratio as a function of ordering and sp3 fraction from Ferrari and

Robertson [113]. For crystalline graphite there is no D peak, so

I(D)/I(G) is 0. For nanocrystalline graphite the presence of disor-

der due to grain boundaries gives rise to the D peak, and the G peak

shifts to higher wavenumbers. As the sp3 fraction increases there

is a decrease in the I(D)/I(G) ratio, while the G peak position ini-

tially declines and then increases. This progression, referred to as

the amorphitization trajectory is true for as-deposited films. When

annealed DLC films will progressively order, and3 thesp fraction

will decrease. This is referred to as the ordering trajectory, and

does not follow the same path. For ordering, both the I(D)/I(G)

ratio and the G peak position increase with the sp2 fraction. Figure

reproduced from Ref [125]...... 67

FIGURE 3.3: Raman spectra for a-C:H (a), and a-C:H:Si:O (b), showing varia-

tion of the D and G peaks with temperature. For a-C:H at 300 ◦C

there is a substantial increase in the intensity of the D peak and

a shift of the G peak to higher wavenumbers. At 350 ◦C there is

complete film removal as indicated by the lack of a characteristic

DLC peak in the Raman spectra. For a-C:H:Si:O in (b) the in-

tensity of the D peak is increasing as well, but shows significantly

less change than for a-C:H. The G peak shifts similarly to a-C:H,

but occurs at higher temperatures. After annealing to 350 ◦C film

volatilization has not occurred for a-C:H:Si:O, in contrast to a-C:H.

Shifts in the G peak position and I(D)/I(G) ratio indicate changes

to the structure of a-C:H/a-C:H:Si:O [18, 113, 125]...... 70

xx FIGURE 3.4: Multiwavelength Raman spectra from annealed DLN 360 a-C:H:Si:O

acquired at four different wavelengths as indicated in the figure.

The position of the G peak was determined by fitting the spectra

with two Gaussian peaks and a linear background. The peak po-

sition was used to compute the sp3 fraction based on Cui et al.

[127]...... 72

FIGURE 3.5: sp3 fraction vs. temperature for annealed DLN 360 a-C:H:Si:O de-

termined from multiwavelength Raman...... 73

FIGURE 3.6: High resolution XPS spectra of a-C:H:Si:O showing the C1s spec-

trum (left) and the Si2p spectrum (right) as a function of annealing

temperature. The peaks assigned to carbon bound to oxygen are

due to the contamination layer, which desorbs after annealing. An

increase in the peaks assigned to π-π∗ shake-up satellites suggests

and ordering and clustering of sp2 carbon. The curve fitting of

the Si2p spectra shows an increase in the fraction of Si in higher

oxidation states. Figure from Ref [121]...... 75

FIGURE 3.7: Fraction of sp2 carbon as a function of annealing temperature cal-

culated from NEXAFS spectra. The data was fit to determine the

distribution of activation energies for the conversion of sp3 to sp2

carbon (see insert) [67, 121] ...... 75

FIGURE 3.8: SEM Image of a typical heated AFM cantilever. Current flows

through the low resistance cantilever legs and the high resistance

tip, resulting in heat generation at the end of the cantilever. The

image on the right shows a zoomed view of the sharp probe tip.

(Images from Anasys)...... 76

xxi FIGURE 3.9: Resistance vs. temperature for a typical heated AFM cantilever.

The polymer calibration curve is determined by measuring the

melting point of three different polymer samples with the heated

probe. Raman calibration data points are based on the shift in the

Stokes peak position with temperature...... 78

FIGURE 3.10: Thermal circuit model used by Nelson and King to calculate the

interface temperature rise resulting from a heated AFM probe [133]. 79

∗ FIGURE 3.11: Dependence of θInterface on stress for a range of tip radii. The stress is computed from Hertz contact mechanics based on the applied

load. Larger tips achieve greater interface temperature rise. . . . . 83

FIGURE 3.12: AFM scans of a-C:H (top) and a-C:H:Si:O (bottom) before and

after scanning with heated cantilevers operating at 450 ◦C (115 ◦C

interface temperature). Film volatilization is observed for a-C:H

resulting in a depression after scanning. No change is observed for

a-C:H:Si:O under the same conditions...... 84

FIGURE 3.13: Depth of film removal as a function of interface temperature for

a-C:H and a-C:H:Si:O after a single scan with a heated AFM probe. 85

xxii FIGURE 4.1: Comparison of sp3 content obtained with different simulation tech-

niques to experimental values for a-C. The color of the symbols in-

dicates the simulation method (as shown in the legend), while the

shape of the symbol represents the reference for the data. DFT sim-

ulations most accurately reproduce the experimental trends. The

Tersoff potential consistently overpredicts the sp3 fraction, while

the REBO potential significantly underpredict the3 sp fraction.

The EDIP potential provides the best match to experimental data,

followed by ReaxFF. A modified version of the REBO potential

with the cutoffs replaced by environmentally dependent screening

functions (REBO+S) performs similarly to ReaxFF. Experimental

data (gray symbols) from Refs ◦ [148], [149], △ [150] and simu-

lations (colored symbols) from Refs ◦ [147], [146], △ [151, 152],

▽ [145], [143]...... 89 FIGURE 4.2: Radial distribution function (RDF) for a-C simulated with different

potentials employing short range cutoffs compared to experimen-

tal neutron diffraction data. The RDF is a histogram representing

the average number of atoms at a given separation distance. The

RDF for crystalline materials shows sharp peaks at the first, sec-

ond, third, and so on nearest neighbor locations. For amorphous

materials these peaks are broadened due to disorder. For DLC

the first neighbor peak is centered near 1.5 A,˚ corresponding to the

length of C-C σ bonds, while the second neighbor peaks is near

2.5 A.˚ Unphysical peaks between the first and second neighbor lo-

cations are present corresponding to the cutoff distance employed

in each method [144]...... 90

xxiii FIGURE 4.3: Comparison between DLC structures produced with the ReaxFF

Si/C/H/O force field and experimental data. ReaxFF slightly un-

derestimates the sp3 fraction at a given density for a-C, but gen-

erally reproduces the experimental trends. The addition of H (30-

40% in experiments vs. 35% in simulations) decreases the density

while increasing the sp3 fraction, and the simulated a-C:H closely

matches the experimental measurements. Two different a-C:H:Si:O

samples are also shown. In (A) the composition is 33.5% H, 6.4%

Si, and 2.5% O, matching the DLN 180 samples, while in (B) the

composition is 47% H, 10% Si, and 15% O, matching the NCD Tech

samples. Experimental data for a-C is from Refs ◦ [148], [149],

and △ [150], and a-C:H data is from Ref [171]...... 96 FIGURE 4.4: Effect of quenching time on resulting a-C:H:Si:O structures. (a) sp3

fraction vs. quenching time for two different initial configurations.

The change in sp3 fraction with quenching time is small com-

pared to the effect of the randomization of the initial configura-

tion. (b) Average potential energy (eV) per atom vs. quenching

time. The change in energy with increased quenching time is minor. 97

xxiv FIGURE 4.5: (a) Simulated structure of a-C:H:Si:O. The liquid quenching proce-

dure results in a fully amorphous simulation cell. The silicon and

oxygen are intermixed in the a-C:H network. (b) Snapshot of a-

C:H:Si:O showing only the carbon and hydrogen bonded to carbon.

The structure is composed of a dense a-C:H network. The majority

of the hydrogen (91.5 ± 1.0%) is bonded to carbon. (c) Snapshot of

a-C:H:Si:O showing the silicon and oxygen. The silicon is dispersed

throughout the a-C:H matrix. Oxygen typically bonds to silicon

(50.4 ± 4.0% of oxygen bonds are to silicon), forming small regions

of amorphous silica. There is a small amount of hydrogen bonded

to silicon, which is also shown in (c)...... 98

FIGURE 4.6: Distribution of bond lengths (a) and angles (b) for sp3-carbon be-

fore annealing. The average bond lengths are Csp3-H: 1.09 A,˚ Csp3-

C: 1.54 A,˚ and Csp3-Si: 1.98 A.˚ The average bond angle is 108.5◦. 99

FIGURE 5.1: Sample of simulated annealing data from MD for a-C:H:Si:O (a)

and a-C:H (b). The temperature was ramped to the value indicated

in the legend over 5 ps. There is an initial drop in sp3-bonded C,

followed by stabilization. The amount of rehybridization increases

with temperature...... 101

FIGURE 5.2: Distribution of sp3-C:C bond lengths at 300 K for a-C:H:Si:O (top)

and a-C:H (bottom). The light colored region shows the bonds

which broke resulting in rehybridization from sp3 to sp2-C when

annealed at 1100 K for 100 ps. Rehybridization was primarily due

to tensile strained C-C bonds...... 103

xxv FIGURE 5.3: (a) Increase in sp2 carbon fraction as a function of annealing tem-

perature for a-C:H and a-C:H:Si:O. Annealing was for 100 ps. The

lines are guides for the eye. There is a greater degree of rehy-

bridization for a-C:H. Error bars show the max and min values

observed over three replications with randomized initial configura-

tions. (b) C-C bond length histograms for a-C:H and a-C:H:Si:O.

The increase in rehybridization is attributed to the greater fraction

of highly strained bonds in a-C:H compared to a-C:H:Si:O as seen

by comparing the histograms. This is due to the presence of longer

C-Si bonds in a-C:H:Si:O...... 104

FIGURE 5.4: Change in Si bonding for a-C:H:Si:O with annealing temperature.

Error bars show the max and min values observed over three repli-

cations with randomized initial configurations...... 106

FIGURE 5.5: Relative frequency of sp2-bonded neighbors to sp2 carbon, for a-

C:H and a-C:H:Si:O. When annealed for 100 ps there is an increase

in the frequency of sp2 carbon neighbors indicating a growth of sp2

clusters. Lines are guides for the eye...... 107

FIGURE 5.6: Increase in the number of six-member sp2 carbon rings with an-

nealing time for a-C:H and a-C:H:Si:O at 825 ◦C. In both cases

there was only one six-member sp2 carbon ring in the unheated

structure. When annealed at 825 ◦C, the number of six-member

sp2 carbon rings increased rapidly for a-C:H, suggesting clustering

and ordering of the sp2 phase. In contrast, only a small increase

was observed for a-C:H:Si:O...... 108

xxvi FIGURE 5.7: (a) sp3-carbon fraction as a function of Si content in simulated

a-C:H:Si:O. At a given density, the sp3 fraction significantly in-

creases with Si content. (b) Percent of sp3 C-C bonds that are

highly strained (bond length greater than 1.6 A)˚ as a function of Si

content. As the Si content increases, the fraction of highly tensile

strained C-C bonds decreases, from 18% at 0% Si to 2% at 25% Si. 109

FIGURE 5.8: (a) Fit of temperature ramps at three different heating rates using

a single set of fitting parameters (Ea = 0.96 ± 0.38 eV). (b) Fit of a single heating ramp (at 8 × 1012 K/s) with a single distribution

fit compared with two distribution fits. The activation energies

computed by each fit are given in the text...... 113

FIGURE 5.9: Single distribution fits for temperature ramps for a-C:H and a-C

at a heating rate of 8 × 1012 K/s. The mean activation energy for

rehybridization is lower for a-C:H than a-C, with both being lower

than a-C:H:Si:O (Ea = 0.96 ± 0.38 eV)...... 114 FIGURE 5.10: Simulation data and single distribution activation energy fits for

a-C:H:Si:O with in-plane compressive stress. The fitted activation

energy (shown in the legend) is strongly dependent on stress. . . . 116

FIGURE 5.11: Activation energy for sp3 to sp2 conversion vs. stress for a-C:H:Si:O.

The error bars represent the fit standard deviation for the activa-

tion energy distribution, not the confidence interval for the mean

activation energy. The large increase in standard deviation at

1.5 GPa is due to the small amount of total rehybridization, which

makes fitting the activation energy distribution difficult...... 118

xxvii CHAPTER 1: Introduction

Diamond-like carbon (DLC) thin films have attracted wide technological interest due to their exceptional properties, including high mechanical stiffness and strength, low coefficient of friction, low roughness, and chemical inertness [1,2]. These coatings have found uses in a wide variety of applications, particularly those demanding outstanding tribological performance, such as coatings for high-performance engine components [3,4], biological implants [4–6], protective coatings for hard disks [7,8], and microelectromechanical systems [9]. However, they suffer from some significant limitations. First, depending ontheir composition, these films can become unstable above 150 ◦C[10], preventing their use in important technological applications, such as heat assisted magnetic recording (HAMR) disk drives [11]. Second, the friction and wear of DLC is strongly environmentally dependent, which prevents their use as protective coatings in applications where the humidity may vary such as use in aerospace components which need to function in and outside the atmosphere [12]. Thus, further research is needed to understand and potentially address these limitations.

One way to overcome the limitations of DLCs is through doping the films with other elements. Numerous dopants are possible, and can lead to modifications of a wide range of properties including hardness, residual stress, electrical conductivity, and tribological behavior [1, 13]. The goal of this thesis is to advance the understanding of modified DLC coatings doped with silicon and oxygen (referred to as a-C:H:Si:O), which have shown considerable promise as more thermally stable alternatives to DLC, including maintaining low friction in both dry and humid environments [14–17]. Despite the prior studies on the properties of a-C:H:Si:O that have shown improvements in tribology and thermal stability, the exact origins, mechanisms, and extent of these improvements are unknown. This thesis will aim to address these questions through a combination of experimental measurements and molecular dynamics simulations.

In order to understand the dominant mechanisms controlling the friction and wear of a-C:H:Si:O in different environments, macroscopic tribometer testing was performed ontwo

1 different a-C:H:Si:O samples in three environments: dry nitrogen, dry air, and humidair.

The scaling of friction and wear with load was also investigated. Characterization of the wear tracks through atomic force microscopy (AFM) was used to quantify wear rates and observe nanoscale surface modification that occurs during sliding. The thermal stability of a-C:H:Si:O was investigated through annealing experiments and compared to undoped a-C:H. Raman spectroscopy was used to characterize the structural changes and thermal degradation pathways in a-C:H:Si:O. Additionally, heated AFM probes, which allow for rapid heating and cooling of a nanoscale contact area, were used to probe the thermal stability at time and length scales relevant to applications such as HARM disk drives. Fi- nally, reactive MD simulations using the ReaxFF potential were performed on a-C:H and a-C:H:Si:O in order to understand the atomistic origins of this enhanced thermal stability.

The effect of silicon content was also investigated. The simulations provided physical insight into the effect of silicon doping on thermal stability. The activation energy for rehybridization was also modeled using the same methods as for the experiments and produced good agreement between the experimental and simulation results.

1.1. Classification of DLCs

Carbon can create a wide variety of structures due to its ability to form several different bonding configurations. The two dominant bonding states of carbon in such materials are the sp3 and sp2 hybridizations (the sp1 state is not commonly found in solid carbon materials except for polymers). Schematics of these configurations are shown in Figure 1.1.

In diamond, carbon exists purely in the sp3 hybridization state, where it forms 4 σ-bonds in a tetrahedral arrangement. These strong C-C σ-bonds result in its exceptional mechanical properties, such as high elastic modulus and hardness. In graphite, carbon exists in the sp2 hybridization state, where it forms three in plane σ-bonds, and the fourth valence electron forms a delocalized π-bond. This creates a layered structure with strong in plane bonds and weak van der Waals interactions between the planes [18].

The name DLC refers to a broad class of carbon-based amorphous materials with a range

2 C C

sp3 sp2 Figure 1.1: Schematic diagrams showing the sp3 and sp2 hybridization states for carbon. In the sp3 hybridization state, carbon where it forms 4 σ-bonds in a tetrahedral arrangement. The solid and striped wedges represent bonds going into and out of the page. In the sp2 hybridization state, carbon forms three in plane σ-bonds and the fourth valence electron is part of a delocalized π-bond.

of different compositions and bonding [8, 19]. These materials possess many of the same

exceptional mechanical properties as diamond (high hardness and elastic modulus) as well

as a low coefficient of friction in many environments, making them attractive materials for

tribological coatings [1]. DLC consisting of only carbon is known as amorphous carbon (a-

C) and contains carbon in a mixture of sp2 and sp3 bonding configurations. Typically, these films with high3 sp fraction (> 70%) are referred to as tetrahedral amorphous carbon (ta-C).

Often DLC films contain substantial fractions of hydrogen as well (20–60 atomic %), inwhich case these materials are called hydrogenated amorphous carbon (a-C:H) or hydrogenated tetrahedral amorphous carbon (ta-C:H) if the sp3 fraction is also large. These various forms

of DLC can be displayed in a ternary phase diagram as shown in Figure 1.2. Table 1.1 gives

typical values for composition, density, Young’s modulus, hardness, friction coefficients,

and degradation temperature for two forms of DLC in comparison to diamond. DLC’s

mechanical, physico-chemical and tribological properties strongly depend on the chemical

composition (namely hydrogen content) and structure of the material (sp2/sp3 content) [1,

19]. The diamond-like properties, such as high Young’s modulus and hardness, arises from

the C-C sp3 bonds and therefore varies with the film’s C-C sp3 fraction. Hydrogen, while

forming C-H bonds that tend to increase the total sp3 fraction, does not contribute to

linkages in the network and thus tends to decrease the mechanical properties, resulting in

softer, lower density films at the highest hydrogen contents [8, 19].

3 sp3

ta-C ta-C:H

HC polymers sputtered a-C(:H) a-C:H no lms glassy carbon graphitic C sp2 H Figure 1.2: Ternary diagram showing different forms of DLC films based on relative concentrations of sp2 to sp3-bonded carbon and hydrogen [18, 20]. There are three main regions shown. First, along the leftmost axis is hydrogen free amorphous carbon, ranging from mostly sp2 to mostly sp3 bonding. Films with low sp3 fraction are referred to as amorphous carbon (a-C), while films with high sp3 fraction are referred to as tetrahedral amorphous carbon (ta-C). Second, at the bottom right of the phase diagram are compounds with such high hydrogen content that the material cannot form a connected network, but instead remains as gas molecules. Finally, in the center is hydrogenated amorphous carbon (a-C:H), which can range from materials with only 20% H content up to materials where close to 60% of the atoms are hydrogen.

Table 1.1: Comparison of properties for single crystal diamond, ta-C, a-C:H [1,3, 18]

Property Diamond ta-C a-C:H Density 3.515 g/cm3 2.3–3.1 g/cm3 1.2–2.2 g/cm3 Young’s Modulus 1100 GPa 650–800 GPa 50–150 GPa Hardness 100 GPa 40–80 GPa 10–20 GPa Degradation 600 ◦C (w/ oxygen) 600 ◦C 150 ◦C Temperature (burning→C-O) (sp3→sp2) (loss of H, sp3→sp2) µ (Humid) 0.01–0.02 0.04–0.1 0.1–0.5 µ ( Dry ) >1 >1 0.001–0.01 C Hybridization 100% sp3 50–80% sp3 30–50% sp3 H content Negligible <5 at.% 30–50 at.%

4 1.2. Historical background and synthesis

The first synthesis of DLC was reported in the 1950s [1,2]. However, little further research took place at that time. The first comprehensive studies of DLC began inthe

1970s [21], with research interest significantly increasing in the 1990s, resulting in a large number of key publications that laid the groundwork for comprehensive studies of the structure and properties of DLC [1,2, 22–26]. Over this time, improvements in thin film deposition systems allowing inexpensive and high quality coatings to be produced led to steadily increasing popularity of DLC films as protective coatings in industry [1].

DLC films can be produced through a wide variety of methods, including both plasmava- por deposition (PVD) and chemical vapor deposition (CVD) processes [3]. These include ion beam, sputtering, filtered cathodic arc (FCVA), pulsed laser deposition, plasma-enhanced chemical vapor deposition (PECVD), and plasma immersion ion implantation and deposition (PIIID). These techniques offer various advantages and disadvantages including quality of the coatings, cost, growth rate, deposition temperature, and line-of-sight vs. non-line- of-sight coating [3, 18, 27]. The sp2/sp3 ratio and hydrogen content is dependent on the precursor gas compositions and the deposition parameters used (pressure, ion impingement energy, and surface power density at the substrate) [18, 27].

The exact atomistic physics by which the structure of DLC arises from the deposition process are not certain. However, some general mechanisms have been proposed [18]. The sp3 fraction is known to depend on the ion energy during deposition. Initially the sp3

fraction increases with energy, reaching a maximum around 100 eV and then decreasing,

as shown in Figure 1.3[18, 28, 29]. It has been shown that the ion energy is high enough

such that atoms are implanted into the surface, rather than depositing on top, in a pro-

cess referred to as subplantation [30]. This subplantation process is considered critical for

achieving the films diamond-like3 sp bonding. It has been suggested that compressive stress

is required to stabilize the sp3 phase, which is created during deposition by the local density

increases caused by carbon atoms implanting into the surface [28, 31, 32]. In addition, the

energy of the impacting atoms creates a localized thermal spike, which allows for relaxation

5 Figure 1.3: sp3 fraction vs. ion energy for ta-C. The data points are experimental observations, while the line is a theoretical prediction based on the subplantation process, where an atom penetrates the surface resulting in a local density increase and residual compressive stress. Local relaxation at the impact site is assumed to take place during a thermal spike lasting only a few picoseconds [19]. of neighboring atoms during a rapid quenching period on the order of picoseconds [28, 31].

On the basis of this description of the deposition process, the dependence of the sp3 fraction on ion energy has been explained as follows [19, 29, 33]: For low ion energies (< 30 eV), the atoms will not have enough energy to penetrate the surface, and instead deposit on top, leading to the formation of highly sp2 bonded films. With increasing ion energy (50–300 eV) subplantation will occur, resulting in a local density increase and quenched in compressive stress, favoring the formation of sp3 carbon. However, if the energy is too high (> 300 eV), the larger thermal spikes produced will allow greater structural relaxation, reducing compressive stress and favoring the formation of sp2 carbon.

1.3. Applications for DLC

The applications for DLC are widespread, including both current industrial uses and developing technologies. In the automotive industry, DLC coatings have been applied to a variety of engine components that operate in the boundary lubrication regime, where the lubricant oil films are not thick enough to separate the surfaces. These include pistons,

6 camshafts, tappets, and gears. These DLC coatings serve to reduce friction losses, giving rise to improvements in fuel economy and reductions in CO2 emissions, in addition to providing increased wear protection [3,4, 34]. DLC coatings are also an integral component in the design of modern high pressure diesel fuel injectors [3,4]. In the textile industry, DLC coatings are applied to components of machines to reduce wear and corrosion from contact with yarn fibers [4] Due to DLC’s intrinsic corrosion and wear resistance, it has been applied as a protective coating on optical materials and electronic devices including laser barcode scanners [4, 35]. DLC is also used as a coating for injection molding equipment, where the low adhesion between DLC and the polymers used improves the release behavior [4]. Several commercial applications exist as well. A variety of different DLC coated razor blades are available, which offer improved performance and lifetime [3,4]. Another application for

DLC is in the coating of plastic bottles used for beverages, particularly carbonated soft drinks and beer. In this case DLC provides a transparent, food safe barrier to diffusion of carbon dioxide and oxygen, which can significantly improve shelf life [8].

A number of studies have investigated the biocompatibility of DLC and found it to be suitable for use in medical implants and devices [4–6]. Current applications include coating of implants which are in contact with blood such as heart valves and stents, where it has been shown to reduce platelet adhesion and prevent diffusion of metal ions into the bloodstream [4,5]. Due to its chemical inertness and wear resistance, it is considered a promising material for coating orthopedic implants, where wear particles produced by sliding and corrosion at contact points within the implant result in adverse health effects and failures of the implants [4,5].

There are other promising developing applications as well. In the design of microelectromechanical systems (MEMS) there is interest in DLC as a material from which to fabricate devices. This is due to the combination of high wear resistance, hydrophobic surfaces, and chemical inertness, and active research is ongoing into the ability to fabricate

MEMS structures from DLC [9]. DLC is also being studied as a possible solid lubricant for use in moving mechanical assemblies in space applications where the use of liquid lubricants

7 is prohibited by the vacuum conditions [12].

One critical application in which DLC films have been widely used is as the protective overcoat on magnetic storage media, such as hard disks [1,4,7,8, 36, 37]. Modern hard disks store data in a Co-based magnetic layer mounted on a substrate and covered with a hydrogenated amorphous carbon (a-C:H) overcoat and 1-2 monolayers of a lubricant film. a-C:H is used since it forms a smooth, chemically inert film which protects the magnetic media against corrosion and mechanical contact with the read-write head [7,8]. A schematic showing the design of the current head/disk interface is shown in Figure 1.4a. The storage density of magnetic disks has been increasing at a rate of 100% per year and storage densities of over 200 Gbits/inch2 have been achieved and demonstrations have shown densities over

400 Gbits/inch2 are possible [8]. To put this in context, today a commercially available hard drive with a storage density of 140 Gbits/inch2 can store the equivalent of 25 CDs worth of data in a single square inch of space. In order to achieve higher storage density the vertical distance between the read head and the magnetic media, called the magnetic spacing, must be reduced. This requires protective a-C:H films that are only a few nanometers thick [8].

Figure 1.4b shows the magnetic spacing required for a given storage density based on current recording methods, as well as the corresponding thickness for the DLC overcoat.

1.4. Overview of the theories for tribological behavior in DLC

In a broad sense, tribological coatings can be divided between soft and hard coatings.

Soft coatings usually provide low friction, but limited protection against wear, while hard coatings provide wear resistance, but have higher coefficients of friction [38]. DLC, however, is known to provide both superlow friction (with coefficients of friction less than 0.01 being recorded in some testing environments) [38, 39] and high wear resistance [10], and this combination of properties has led to its great success in a wide variety of tribological applications. Conventional theories for solid lubrication include plastic shear in soft metals and the shearing of atomic layers in lamellar solids. However, since DLC is amorphous it does not possess the multiple slip systems that are present in soft metals or the ordered

8 (a) (b)

Figure 1.4: (a) Schematic showing the design of modern hard disks. The magnetic spacing is the distance between the magnetic layer and the read/write elements. This distance includes the air gap, a molecular lubricant layer, and a DLC coating on both the head and disk. Typically ta-C is used on the head, and a-C:H on the disk. Maximum achievable storage density is a function of the magnetic spacing, as shown in (b). This requires reducing the thickness of the DLC coatings to only a few nm while maintaining uniform coverage. Figures from ref [8]

lamellar structure that allows for low friction in materials such as graphite or MoS2 [38, 40]. Thus, an important research question is: what is the primary causes of friction and key lubrication mechanism that give rise to superlow friction in DLC?

In general, the friction between two solids in contact arises from interactions (mechanical, physical, or chemical) between the surfaces. These give rise to three phenomena that control the friction: abrasion, shearing, and adhesion, shown schematically in Figure 1.5.

Abrasion refers to the scratching of the surface by sharp asperities on the harder surface or wear debris, shearing refers to the plastic flow at the interface, and adhesion refers to the breaking of adhesive junctions due to bonds between the surfaces [38]. The relative strength of these three phenomena will depend on the sliding interface, and some may dominate, while others may be neglected. For the case of DLC we can consider the likely relative importance of these mechanisms in contributing to the friction and wear. It is generally accepted that the dominant contribution to the fiction and wear in DLC comes from adhesion [1, 38, 40]. Abrasion is argued to be unlikely in the case of DLC, since the surfaces have high hardness and high strain tolerance [38]. Additionally, DLC typically forms very smooth coatings with low RMS roughness [41], leading to minimal abrasion due

9 Figure 1.5: Schematic of the three general origins of macroscopic friction forces, abrasion, shearing, and adhesion [38]. to sharp asperities. Shearing is believed to occur between the DLC surface and a sp2 rich transfer film on the countersurface (to be discussed in detail later), however, this shearing component would only be significant if adhesion was large [38]. This leads to the conclusion that adhesion is then the dominant factor in the friction of DLC. Carbon surfaces can interact with countersurface by a variety of different bonds, ranging from very strong σ-

bonds to very weak van der Waals forces, depending on the nature of the contact [1, 10].

This means there can be substantial variations in adhesion forces, which are the fundamental

contribution to friction.

These adhesive interactions give rise to an extremely complex frictional behavior for DLC

that depends heavily on the nature of the film and the sliding environment. For example,

Figure 1.6 shows the dependence of friction on relative humidity for both a-C and a-C:H. A

brief explanation for this behavior is the following. For the case of a-C in a dry environment,

high friction occurs due to high adhesion resulting from strong σ-bonds formed between

carbon dangling bonds at the interface, where those dangling bonds were formed by the

applied stresses during sliding. If the film contains hydrogen, it can passivate these dangling

bonds, resulting in low adhesion due to weak interactions between hydrogen terminated

surfaces [10]. If water is in the environment, it can form bonds with the unsaturated carbon

atoms, forming, e.g., C=O or C-OH bonds on the surface. The formation of these C=O or

C-OH bonds eliminates the strong σ-bonds, reducing friction for the case of a-C. However,

for the case of a-C:H these C=O or C-OH bonds interact more strongly than the hydrogen

10 Figure 1.6: Schematic diagram showing the general trends in the frictional behavior of DLC as a function of humidity [10]. terminated bonds, resulting in an increase in friction [10]. These phenomena show that interactions at the nano and atomic scale are critical for understanding the macroscale friction of DLC. Table 1.2 gives typical values of the coefficient of friction for diamond, a-C, and a-C:H in various environments. These large changes in tribological behavior with environment can present problems for the use of DLC in several applications. For example, in designing solid lubricants for use in space, components need to be able to function in a wide range of environments. Devices must be tested before launch in ambient air, while operating environments can range from ultrahigh vacuum (UHV) to a highly reactive atomic oxygen atmosphere [12, 42].

The dominant mechanism in controlling the friction of DLC is the passivation of carbon dangling bonds at the sliding interface. High hydrogen content a-C:H (∼ 40% H) will have superlow friction (µ ∼ 0.001–0.02) in vacuum and inert environments [10]. This is

Table 1.2: Typical friction coefficients for CVD diamond, a-C, and a-C:H coatings indif- ferent environments [10].

Property Diamond a-C/ta-C a-C:H/ta-C:H µ ( Vacuum ) 0.02–1 0.3–0.8 0.007–0.05 µ ( Dry N2 ) 0.03 0.6–0.7 0.001–0.15 µ ( Dry air ) 0.08–0.1 0.6 0.025–0.2 µ (Humid air) 0.05–0.15 0.05–0.2 0.02–0.5

11 due to the weak van der Waals interactions between the hydrogen terminated C-H bonds at the interface leading to low adhesive interactions between the sliding surfaces [10, 12,

43]. During sliding, however, surface hydrogen will be released due to stress and frictional heating. For a-C:H films with low hydrogen content, this will result in a loss of superlow friction after some time [44]. In order to maintain superlow friction, there needs to be a high enough concentration of hydrogen in the film to replenish the surface passivation, either through the release of free hydrogen in the film, or through diffusion of hydrogen to the surface [44].

In the presence of humidity or oxygen, however, the friction for a-C:H tends to increase, and can see coefficients of friction as high as 0.5 in humid air[10]. In the case of humid environments, experiments have shown that the presence of water molecules disrupts the

C-H van der Waals interactions that give rise to superlow friction by replacing the -H termination of carbon dangling bonds with -OH termination, leading to stronger hydrogen bonding or dipole interactions at the interface [10, 43, 45]. These results are supported by density functional theory (DFT) simulations of a-C surface passivation in contact with aluminum, which have found and increase in the work of separation from 0.02 J/m2 to

0.2 J/m2 when -H passivation was replaced by -OH passivation [46]. For the case of oxygen environments results are inconsistent, with some experiments finding higher coefficients of friction compared to vacuum [43], while others have seen no effect47 [ ]. A possible explanation for these results could be due to the amount of C-O bonding at the interface. On the basis of DFT simulations looking at the effect of oxygen passivation of a-C surfaces [48], it has been claimed that for a-C:H with high initial C-H bonding oxygen can passivate the small number of dangling bonds produced during sliding. However, if the dominant passivation mechanism becomes C-O bonding, the coefficient of friction will increase, since the larger van der Waal radius of oxygen prevents full passivation, and C-O-C bonds can form across the interface, increasing friction and adhesion [48].

Unlike a-C:H, for a-C in vacuum or inert environments, high friction and wear is observed due to strong covalent bonding caused by unpassivated carbon at the sliding interface [1,

12 10, 40, 45]. In this case DFT simulations have calculated a work of separation of 4.5 J/m2, far higher than the 0.02 J/m2 found for -H passivation [46]. The high friction observed in vacuum is reduced if hydrogen gas is present to provide passivation of dangling bonds produced by sliding. Simulations have shown that dissociation of H2 at the sliding interface is energetically favorable [46], and experiments of a-C sliding in H2 gas have shown low friction and spectroscopic measurements have confirmed hydrogen bonding in the wear track [38]. In ambient environments with the presence of water vapor, the friction of a-C decreases compared to vacuum. This is due to C-O and C-OH passivation which produces behavior in humid air similar to a-C:H [1, 10, 40, 45, 46, 49].

Beyond the effects of adhesive interactions and passivation of interfacial bonds, other important factors controlling the friction and wear of DLC include transfer film formation, mechanical properties of the film and transfer film, and rehybridization of carbon bonding in the wear track. Transfer films, also known as tribofilms, refer to films formed onaslid- ing partner due to material transfer from the other. When sliding against metal surfaces, such tribofilm formation is typical for all DLCs, and generally consists of material transfer from the DLC film to the metal surface [26, 40, 50, 51]. It has been shown in experiments that this tribofilm formation is a dominant process during the run-in period when friction decreases at the start of sliding [12, 26] and is necessary for superlow friction [50]. Once formed, sliding takes place at the DLC/tribofilm interface, and therefore friction is controlled by the interfacial shear strength between the DLC and the tribofilm [40, 51]. It has been hypothesized that the formation and properties of these tribofilms are influenced by environmental species as well, further contributing the the environmental dependent friction observed for DLC. It has been suggested based on experiments sliding a-C:H in humid air that the formation of a tribofilm is inhibited by the presence of water vapor, andthe tribofilms formed in humid environments have higher interfacial shear strength [26, 47], contributing to the large increase in coefficient of friction for a-C:H with humidity.

The mechanical properties of DLC have also been shown to be important for achieving low friction [50, 52]. In particular, strong correlations have been shown to exist between

13 the viscoplastic nature of a-C:H films and the coefficient of friction in vacuum. Viscoplastic materials are ones in which the hardness increases with strain rate. This can be quantified by the equation:

x H = H0 · ε

where H is the hardness, H0 is a constant, ε is the strain rate, and x is a constant known as the viscoplastic exponent. When x = 0, the hardness does not depend on strain rate, so the behavior is purely plastic. Increasing values of x then indicate a greater viscoplastic character [50, 52]. It has been shown that only a-C:H films with sufficiently high vis- coplasticity were able to achieve superlow friction in vacuum [50, 52]. Figure 1.7 shows the correlation between viscoplastic exponent and coefficient of friction for several different a-C:H films [50]. It is hypothesized that since viscoplastic behavior is due to greater ease of network relaxation, the more viscoplastic films are more capable of surface relaxation during sliding allowing the formation of lower friction interfaces [50, 52].

Figure 1.7: Steady state coefficient of friction in vacuum for several different a-C:H filmsas a function of the viscoplastic exponent. A correlation is found to exist between the films viscoplastic character and the coefficient of friction, and films with mostly plastic behavior cannot achieve superlow friction [50].

14 The final mechanism often discussed in the tribology of DLC is carbon rehybridization in the wear track, which is the subject of many, often conflicting, results. Several early studies of DLC observed an increase in sp2 carbon in the wear track and tribofilm based on

Raman spectroscopic measurements. This is due to frictional heating overcoming the energy barriers for sp3 to sp2 conversion, which are reduced by stress at the interface [53, 54]. On this basis it was proposed that the mechanism for low friction in DLC was the formation of crystalline graphite near the film surface, and shear was accommodated by interlayer sliding of these graphite crystals. This mechanism was referred to as ”graphitization” [53, 55–57].

However, molecular dynamics (MD) simulations and better spectroscopic measurements do not support this hypothesis. Near Edge X-Ray Absorption Fine Structure (NEXAFS) spectroscopy, which can distinguish clearly between disordered sp2 carbon and graphite, has

shown no ordered graphite is produced in ta-C wear tracks [49], and transmission electron

microscope (TEM) images have also confirmed a disordered structure [51]. In addition,

multiple MD simulations have observed rehybridization from sp3 to sp2 carbon in simulated

sliding of DLC, but no crystalline structure was formed, the sliding interface remained

amorphous [58–61]. Some studies have also claimed to find an increase in sp3 carbon in

the wear tracks on the basis of Raman spectroscopy [51, 62]. It is proposed this is due to

the bonding of hydrogen to sp2 carbon, converting them to sp3 sites, resulting in a near

surface increase in hardness and density [62]. However, these findings are not certain, since

accurately determining the sp3-C fraction from Raman data is difficult, and it is unclear

why an increase of C-H bonds near the surface would produce the observed increase in

mechanical properties, since such bonds do not contribute to the interconnectivity of the

carbon network.

The impact of surface rehybridization on tribology is also unclear. Many sources claim

that an increase in sp2-C would result in increased adhesion and friction due to “strong” π-π

interactions between sp2-C aligned in parallel [1, 40]. It is suggested that these interactions

could have strength similar to C-C covalent bonds [1]. This contention is primarily based

on observations that graphite has high friction in vacuum, which has been explained by

15 claims that strong π-π interactions exist between graphite layers in vacuum, while in humid air water could enter between the layers and weaken these π-π interactions, resulting in

easy shear [63]. However, this is not well supported. It has been shown by surface X-ray

diffraction measurements that the interlayer spacing in graphite is not changed between

vacuum and ambient environments [64], which refutes the hypothesis that water weakens

the binding force between graphite layers, and lends support to the alternative explanation

that graphite layers have intrinsically weak interactions and the high friction in vacuum is

due to dangling bonds at edge sites [65]. This would also be consistent with other sources

that claim the π-π interactions strength is closer in magnitude to van der Waals forces, and

far weaker than covalent bonds [66]. In addition, contrary to claims that sp2-C aligned in

parallel across the sliding interface increases friction, MD simulations of self-mated a-C:H

sliding have found that during run-in sp2-C near the surface rotates parallel to the sliding

direction, which decreases bonding between the two surfaces [59].

1.5. Thermal stability of DLC

Unfortunately, DLC films are known to have limited thermal stability, starting attem-

peratures as low as 100 ◦C in some cases [10, 18, 67]. The thermal degradation pathways for

DLC have been widely investigated, and are thought to consist of three primary mechanisms:

the direct conversion of sp3 to sp2 carbon, dehydrogenation (for a-C:H), and oxidation (in

aerobic conditions) [10]. The conversion from sp3 to sp2 carbon is frequently referred to

as “graphitization”. However, experimental evidence does not suggest the formation of or-

dered graphite, only a transition to a highly sp2 hybridized amorphous network [68]. The

environmental conditions strongly affect the reaction pathway and its kinetics, for example,

in aerobic conditions the thermo-oxidative degradation of DLC results in the formation

of volatile species (e.g., CO, CO2), leading to rapid film disintegration at higher temperatures [69–71]. The general trend at elevated temperature is decreased mechanical properties

and increased wear, due to the transformation from sp3 to sp2 bonded structure and the

oxidation of the film [10, 69, 72].

16 a-C:H tends to have significantly lower thermal stability compared to ta-C, and studies have shown thermal stability decreases with increasing hydrogen content [73, 74]. How- ever, in these studies a mechanism was not proposed for the physical basis of hydrogens effect on thermal stability. Experiments have seen conversion from3 sp to sp2 carbon for a-C:H starting between 150 and 200 ◦C[69, 73, 75]. MD simulations of a-C:H with differing hydrogen content observed rehybridization starting at 1000 ◦C for films with 0% H and decreasing to 325 ◦C for films with 50% H[74]. Loss of hydrogen is also observed in experiments [69, 73, 75], and this is hypothesized to be due to the breaking of sp3 C-H bonds and the release of trapped free hydrogen in the film [73]. Loss of hydrogen has been observed in

MD simulations as well, however, it does not occur until much higher temperatures (400–

2000 ◦C, depending on H content) than experiments [74]. For ta-C rehybridization typically does not occur until higher temperatures (600 ◦C) [68]. Based on a quantitative determination of the sp3 fraction from NEXAFS measurements [68], the thermally induced conversion from sp3 to sp2 carbon in vacuum has been modeled using reaction rate theory as a first order reaction with a distribution of activation energies [67]. This gave an activation energy of 3.5 ± 0.9 eV [68].

1.6. Heat-assisted magnetic recording

The low thermal stability of DLC, and particularly of a-C:H, limits the use of these films in high temperature applications. One emerging technology where this isofimme- diate concern is the development of high storage density heat-assisted magnetic recording

(HAMR) disk drives [11, 76]. As discussed previously in Section 1.3, DLC overcoats are a critical component in the design of modern disk drives. However, scaling current hard disk technology to higher storage densities is not possible, since the resulting magnetic grains would be too small to hold their magnetization against thermal agitation. A temperature stability factor can be defined as the ratio of magnetic energy to thermal energy:

K V u ≥ 70 kbT

17 where Ku is the materials magnetic anisotropy energy, V is the volume of a grain, and T is the ambient temperature. The value of 70 roughly corresponds to a disk lifetime of 10 years [77]. Therefore, in order to maintain this ratio while reducing grain volume, high anisotropy materials must be used. However, this presents challenges, since such materials cannot have their magnetization set by the available recording head field in the first place.

A promising solution is to use heat-assisted magnetic recording (HAMR). In HAMR, a laser is used to locally heat the magnetic material above the Curie point, reducing the coercivity of a single domain below the available magnetic field, allowing recording on high anisotropy materials [77, 78]. This is shown schematically in Figure 1.8. This approach requires local temperatures over 300 ◦C during the recording process [78]. Thus, the low thermal stability of a-C:H presents a problem for the use of conventional coatings in the design of HAMR disk drives, so these new recording techniques require a new generation of protective films.

Heating Media Storage Temp. Cooling Media Coercivity

Write Available Head Field Temp.

Temperature

Figure 1.8: Schematic of the HAMR process. Heating the magnetic material reduces its coercivity, allowing recording. The material then cools rapidly to the storage temperature, locking in the recorded data [77].

18 1.7. Improving the tribology and thermal stability of a-C:H through the addition of Si and O

One way to overcome the limitations of a-C:H (high friction in humid environments and low thermal stability), is through doping the films with other elements. Numerous dopants are possible and have been investigated in the literature including Ti, B, S, Si, O, Cr, F,

W, and N [1, 13]. These can lead to modifications of a wide range of properties including hardness, residual stress, electrical conductivity, and tribological behavior. For example, doping with metal has been shown to increase electrical conductivity [13, 79], while the incorporation of N can increase thermal stability and reduce residual stress, but decreases hardness [71]. Silicon doping in particular has attracted research interest as a means to reduce residual stress [80–82], increase thermal stability [70, 83], and lower friction in humid environments [81, 84–86], while maintaining high hardness [81]. A variety of experimental techniques (Raman, Fourier-transform infrared (FTIR), and X-ray photoelectron (XPS) spectroscopy) have shown the inclusion of Si (∼10 atomic %) in DLC increases the fraction of sp3 carbon [70, 80, 83, 87]. It has been proposed that this is due to silicon not forming

π bonds, and thus increasing the overall sp3 character of the film [70, 83]. In addition, Si- doping delays the typical rehybridization of sp3 carbon and formation of graphitic clusters observed for a-C:H until higher temperatures [70, 83, 87]. In terms of tribology, the inclusion of Si has been found to decrease the coefficient of friction in humid air compared to a-C:H.

Experiments with a-C:H:Si (10 atomic % Si) showed similar behavior to a-C:H in dry environments, but a lower coefficient of frictionµ ( = 0.01) in humid air [84]. Based on ab initio MD simulations it has been proposed that the inclusion of Si increases hydrophilicity and decreases the energy barrier for water dissociation leading to Si-OH groups at the surface [85, 86]. It has been proposed that these Si-OH groups serve as anchor points for other water molecules through hydrogen bonding, leading to the formation of a water layer that can result in decreased friction [85, 86].

One particularly interesting and important candidate for a next-generation coating is

SiOx-doped a-C:H (a-C:H:Si:O). Such SiOx-doped a-C:H is commonly produced through

19 PECVD using hexamethyldisiloxane (HMDSO, C6H18OSi2) as a precursor [88, 89]. From a production standpoint, this offers several advantages compared to pure silicon doping,

since HMDSO has low toxicity and flammability [88], compared to the silane precursors

used to produce a-C:H:Si, which are highly flammable and an explosion hazard [88, 89].

This material was first developed in the early 1990s and was referred to as diamond-like

nanocomposite (DLN) [90]. It has been claimed that a-C:H:Si:O possess many of the same

favorable properties as Si-DLC, including low residual stress, higher thermal stability, and

low coefficient of friction in humid air[14–16]. In addition a-C:H:Si:O has good adhesion

to a variety of substrates, including metals, semiconductors, ceramics, and plastics [14].

a-C:H:Si:O has been explored for many of the same applications as conventional DLC,

including the automotive and aerospace industries, biomedical applications, and the design

of MEMS [4, 42, 91–93].

A number of studies have investigated the structure of a-C:H:Si:O through Raman,

FTIR, and XPS spectroscopy [89, 94–96]. The films are found to primarily consist ofC-C,

C-H, Si-O, and Si-C bonding [89, 95]. For a series of films deposited with varying Si andO

content (7–22 at. % Si, 16–46 at. % O), it was found that increasing Si favors sp3 carbon

bonding and reduces residual stress, similar to a-C:H:Si [84, 89]. Oxygen preferentially

bonds to Si, replacing Si-H bonds, and at high levels of Si and O leads to the formation

of a Si-O-Si network [89]. Si-C bonds form the connectivity between the a-C:H and Si:O

networks, and there is minimal C-O bonding [89, 96]. Based on Raman spectroscopy of

films with differing Si content, it was proposed that for low Sicontent(< 13 at. %) the

structure consists of a single phase, while for films with high Si content (> 13 at. %) a large

increase in the photoluminescent background is hypothesized to be caused by a segregated

SiOx phase [94]. Based on these measurements, a-C:H:Si:O is proposed to consists of a diamond-like net-

work of amorphous carbon and hydrogen (a-C:H) and a second quartz-like network of silicon

and oxygen (a-Si:O) (see Figure 1.9)[14, 89, 96]. These two networks are intermingled and

result in a purely amorphous structure with interbonding between the two networks due

20 Carbon Silicon Oxygen Hydrogen

a-(C:H0.15)0.7 a-(Si:O0.3)0.3

Figure 1.9: Schematic of the structure of a-C:H:Si:O. The material is made of two networks, one of (a-C:H) and the other of (a-Si:O). The connection between the two networks is proposed to be primarily Si-C bonds [14, 89, 96]. to Si-C bonding [95, 96]. The structure of a-C:H:Si:O shown in Figure 1.9 is proposed to offer several key advantages over a-C:H. First, the intermingling of the a-Si:O network results in more structural degrees of freedom, allowing a-C:H:Si:O to adopt structures with lower residual stress (< 1 GPa) than a-C:H (several GPa) [89]. Additionally, the satura- tion of bonds with O causes a-C:H:Si:O to resist further oxidation. Finally, Si favors sp3 hybridization and bonds to C in the film forcing a higher3 sp -C fraction and preventing carbon rehybridization at high temperatures [14, 15, 89]. Table 1.3 gives typical properties of a-C:H:Si:O compared to a-C:H.

Investigations into the thermal stability of a-C:H:Si:O have suggested it is thermally stable up to 400 ◦C in aerobic and 600 ◦C in anaerobic environments [15, 97]. Raman spectroscopy measurements have suggested changes in film structure beginning around 300 ◦C in air, and 400 ◦C in inert environments, however, the mechanical properties did not decrease until 400–600 ◦C depending on the environment [15, 97]. Also, a-C:H undergoes significant reactions with oxygen when annealed in air starting at 200 ◦C, resulting in film loss due to the formation of CO2. For a-C:H:Si:O annealed in the same conditions minimal film loss was observed up to 300 ◦C[97]. It was proposed that oxygen preferentially reacts with Si rather than C, delaying film volatilization [97].

21 Table 1.3: Comparison of properties for a-C:H:Si:O with a-C:H [3, 88]. The composition of a-C:H:Si:O can vary greatly depending on deposition methods. Typically films contain 5–35 at. % Si and 3–30 at. % O.

Property a-C:H:Si:O a-C:H Young’s Modulus 40–160 GPa 50–150 GPa Hardness 6–30 GPa 10–20 GPa Degradation > 300 ◦C 150 ◦C Temperature (uncertain) (loss of H, sp3→sp2) µ (Humid) 0.04–0.1 0.1–0.5 µ ( Dry ) 0.01–0.03 0.001–0.01 C Hybridization unknown 30–50% sp3 H content 10–40 at.% 30–50 at.%

Similar to silicon doped a-C:H, a-C:H:Si:O films are proposed to have improved tribological behavior in humid environments. A number of studies have measured the coefficient of friction for a-C:H:Si:O in humid air and found values between 0.04 and 0.2 depending on the film composition and sliding conditions [16, 17, 91, 96, 98, 99]. For example, the coefficient of friction for low Si content a-C:H:Si:O sliding against 52100 steel in humidair

(30–50% RH) with a 10 N load was found to be between 0.04 and 0.08 [16, 91], while a coefficient of friction between 0.05 and 0.1 was found when sliding against WCballs(5N load) [96]. There has been some investigation of the effect of Si content on the friction of a-C:H:Si:O, with one study finding a coefficient of friction between 0.05 and 0.15 forfilms with low (< 5%) Si in humid air (50% RH), which decreased to 0.04–0.08 for films with moderate Si content (18%). However, when the Si content was increased further (36%), the coefficient of friction increased to 0.1–0.2 [99].

However, there is still a lack of knowledge regarding the structural composition and the physical basis for the observed effects of Si and O on macroscopic properties. For instance, the ratio of sp2 to sp3-bonding or the thermal degradation pathways are not known. There has been some effort to understand the improved tribological properties in humid environments. Scharf et al. [17] conducted sliding test between a-C:H:Si:O and steel in well controlled environments (dry N2, dry air (< 1% RH), and humid air (50% RH) with loads ranging from 50 to 1000 mN. At a load of 100 mN, the steady state coefficient of

22 friction was 0.02 in dry N2, 0.05 in dry air, and increasing to 0.22 in humid air. It was observed that the friction displayed non-Amontonian behavior (the coefficient of friction decreased with load), and it was proposed that this behavior could account for the higher coefficient of friction measured in humid air (0.22) compared to prior works using ahigher load [16, 17]. It was proposed that the environmental effect on friction was controlled by modifying the chemical composition and resulting interfacial shear strength of the tribofilms formed in different environments. In dry environments, carbon rich tribofilms with large quantities of C-H bonding were present with a low shear strength (9 MPa), while in humid air SiOx rich tribofilms with a much higher shear strength (78 MPa) were observed[17]. Work by Koshigan et al. [100] investigated the tribological behavior of a-C:H:Si:O in vacuum and low partial pressures of hydrogen and oxygen gas. It was observed that in high vacuum or low partial pressures of either gas (< 10 mbar), friction was high (µ > 1).

However, after a critical partial pressure was reached (between 10 and 50 mbar), friction dropped to below 0.1. A spectroscopic investigation of the wear tracks and steel balls revealed that under low gas pressures adhesive junctions formed between the a-C:H:Si:O and the steel, leading to transfer of steel from the ball to the a-C:H:Si:O flat, resulting in high friction. For the case of high gas pressure, initial adhesive junctions still form, however, the material transfer is from the a-C:H:Si:O to the steel, leading to the formation of a carbon rich tribofilm film. The low steady state friction then results from theweak interaction and low shear strength between this tribofilm and the a-C:H:Si:O.

1.8. Investigating the tribological behavior and thermal stability of a-C:H:Si:O

Despite the prior studies on the properties of a-C:H:Si:O that have shown improvements in tribology and thermal stability, the exact origins, mechanisms, and extent of these improvements are unknown. For the purpose of designing improved tribological coatings this presents a problem, since a lack of a physical understanding of the structure-property relationships in a-C:H:Si:O prevents informed coating design. For example, the optimal quantities of Si and O are not known, and are typically a side effect of the precursor

23 molecules used, rather than a design consideration.

As stated above, in this thesis, the tribological behavior and improved thermal stability of a-C:H:Si:O compared to a-C:H will be investigated through a combination of experimental measurements and atomistic simulations. To explore improvements in tribological properties, friction and wear testing was performed on two different a-C:H:Si:O samples with varying atomic composition over a range of environments and loads. Coefficients of friction between approximately 0.03–0.05 in dry environments (RH < 5%) to 0.13 in humid air when sliding against steel were observed. This is better than prior observations for undoped a-C:H films, which typically have friction coefficients close to 0.2 inhumid air (although considerable variation exists) [10]. The scaling of friction with load was also investigated and showed a power law decrease of the friction coefficient vs. load. The wear rate of a-C:H:Si:O, which is important for understanding the durability of these coatings, has generally not been carefully investigated in previous literature. Here, the formation of wear tracks was measured as a function of sliding cycles and the wear rate was shown to be correlated with the energy dissipated by friction. The results demonstrate that friction and wear behavior is controlled by the formation of adhesive junctions between the a-C:H:Si:O films and the steel, and by the development of transfer films on the steel counterface. Pos- sible mechanisms occurring during run-in will be discussed.

To explore the thermal stability, annealing experiments compared undoped a-C:H and a-C:H:Si:O. The a-C:H:Si:O films showed significant improvements in thermal stability at temperatures up to 450 ◦C. Spectroscopic analysis using multiple complementary techniques was used to determine the thermal degradation mechanisms in a-C:H:Si:O in both vacuum and aerobic environments. Thermal decomposition of a-C:H:Si:O in vacuum followed similar pathways to a-C:H (loss of hydrogen and conversion from sp3 to sp2 bonding). However, the

temperatures required for this thermal decomposition to occur were substantially increased.

In aerobic environments, the inclusion of Si and O was shown to significantly reduce film

decomposition through oxidation by the formation of a protective Si:O2 surface layer, which

limits carbon volatilization. Modeling of quantitative measurements of the carbon sp3

24 fraction was used to determine activation energies for the thermally induced sp3-to-sp2

conversion for both a-C:H and a-C:H:Si:O and revealed the addition of Si and O increases

the mean activation energy for rehybridization.

In order to understand the atomistic origins of this enhanced thermal stability, reactive

MD simulations using the ReaxFF potential were performed on a-C:H and a-C:H:Si:O. As in

experiments, the simulated a-C:H:Si:O demonstrated increased thermal stability compared

to conventional a-C:H. The primary thermal degradation pathway in undoped a-C:H was

observed to be the breaking of tensile strained C-C bonds resulting in a transformation of

sp3 to sp2-hybridized carbon. The presence of Si suppresses this mechanism by decreasing

the frequency of highly strained C-C bonds in the unannealed structure. This is due to the

longer C-Si equilibrium bond length compared to C-C bonds, which allows the Si-doped

films to accommodate higher structural disorder. By linearly varying the temperature in

the simulations the activation energy for rehybridization could be modeled using the same

methods as for the experiments and produced good agreement between the experimental

and simulation results.

25 CHAPTER 2: Tribological behavior of a-C:H:Si:O as a function of environment and load

As described in Section 1.4, the tribological behavior of DLC is a complex process dominated by adhesive interactions, which strongly depend on the composition of the film

(e.g., a-C vs. a-C:H) and the environment (e.g., air, vacuum, humidity). Although a-C:H films have found wide success in many applications [3,4,8, 12], the strong environmental dependence of the friction, in particular the large increase in friction and wear with humidity, is a major limitation for the use of a-C:H in important technological applications, such as aerospace components that involve operation in terrestrial (humid) environments [10, 12,

42]. This has motivated ongoing research into improving the tribological behavior of a-C:H, particularly in humid environments. a-C:H:Si:O has attracted considerable interest as an alternative to a-C:H coatings which can be used in many of the same applications while maintaining a reasonably low coefficient of friction even in humid environments [4, 14, 16,

42, 92, 99].

A number of prior studies have investigated the friction of a-C:H:Si:O, primarily in humid air (30-50% RH) when sliding against steel [16, 17, 91, 99]. A summary of some results is given in Table 2.1. Several early studies noted significantly lower coefficients of friction in humid environments for a-C:H:Si:O compared to a-C:H [16, 91, 99]. However, this behavior was not universal, and in some cases moderately higher friction coefficients were observed in humid air [17]. These different results are attributed to the different normal stress at the contact and will be addressed in Section 2.8. Much of this prior work is based a-C:H:Si:O samples with similar composition, and the exact mechanisms controlling the

Table 2.1: Summary of coefficient of friction values for a-C:H:Si:O in humid air (30-50% RH) vs. steel from prior work.

Reference Applied Normal Stress (GPa) COF [16] 0.8 0.05 [91, 99] 0.8 0.04–0.08 [17] 0.3–0.6 0.22–0.1

26 friction and wear are uncertain. Therefore, further research into the tribological behavior of a-C:H:Si:O, including samples with different compositions, is important for understanding of the tribology of a-C:H:Si:O and to see if reported behavior is universal to this class of materials over a range of compositions.

2.1. Experimental samples

For the work in this thesis, several a-C:H:Si:O samples with different compositions were investigated. Two different a-C:H:Si:O samples were purchased from Sulzer Metco (now

Oerlikon Balzers, Pf¨affikon, Switzerland). These samples were2 µm thick films deposited using PECVD from an HMDSO source gas and (at the time of purchase) were called DLN

180 and DLN 360. This product line is now listed as Balinit Dylyn and may have dif-

ferent composition than the samples used in this work. A third a-C:H:Si:O sample was

obtained from NCD Technologies (Madison, WI, USA). This sample was deposited using

PIIID from an HMDSO source gas. This film was a custom order and not part of any

standard product line. The film was 40 nm thick and deposited on a silicon wafer. The

atomic composition of each samples was determined through a combination of X-ray photo-

electron (XPS) spectroscopy, forward recoil elastic scattering (FRES), secondary ion mass

spectrometry (SIMS), Rutherford backscattering spectrometry (RBS), and hydrogen for-

ward scattering (HFS). The XPS and FRES was performed by Dr. Filippo Mangolini and

J. Brandon McClimon using equipment at the University of Pennsylvania, while the SIMS

and RBS/HFS was performed by Evans Analytical Group (Sunnyvale, CA, USA). Table 2.2

shows a summary of the atomic compositions for each sample.

The DLN 180 composition has been the subject of much prior tribology research [14,

Table 2.2: Atomic composition of a-C:H:Si:O samples used in this thesis.

Sample C % H % Si % O % DLN 180 57.0 ± 1.0 34.0 ± 1.0 6.0 ± 1.0 3.0 ± 1.0 DLN 360 44.0 ± 5.0 30.0 ± 5.0 18.0 ± 5.0 7.0 ± 5.0 NCD Tech 25.1 ± 0.1 47.0 ± 2.0 10.3 ± 0.1 14.6 ± 0.1

27 16, 17, 91, 99, 100]. As such, in this work the tribological experiments focused on DLN 360 and NCD Technologies a-C:H:Si:O while being compared to prior literature on DLN 180 a-C:H:Si:O in order to both further the understanding of the tribology of a-C:H:Si:O and to the effect of different film compositions on the tribological behavior. DLN 180samples were investigated later as the experimental reference for the thermal stability work.

2.2. Description of gas blowing tribometer

In order to investigate the tribology of a-C:H:Si:O in a range of environments a custom built reciprocating motion tribometer was used, located at the Laboratoire de Tribologie et

Dynamique des Syst`emesin Lyon, France during a visit from September–November 2015.

The tribometer had the ability to flow gas into the contact area to control the sliding environment between laboratory air, dry air, and dry nitrogen. A photograph of this setup is shown in Figure 2.1. The a-C:H:Si:O samples were slid against 3 mm radius 52100 steel ball bearings under a 0.5 N load. This corresponds to a mean Hertzian normal stress of

0.3 GPa The sliding speed was constant at 3 mm/s. The a-C:H:Si:O and the steel balls were cleaned prior to testing by sonicating in acetone and then ethanol. The relative humidity

(a) (b)

Figure 2.1: (a) Image showing the setup of the gas blowing tribometer used for measurements of friction and wear in different environments. A schematic of the setup is shownin (b). Normal force was applied via weights and an actuator controlled the sample motion. Two nozzles directed the flow of gas at/ 5L min to the contact perpendicular to the sliding direction to control the environment.

28 of the laboratory air varied, but was typically around 40% and always fell between 23–42%.

Environmental conditions were controlled by the flow of gas into the contact/ at5L min.

Nozzles located to both sides of the contact area directed the flow, which was perpendicular

to the sliding direction. Two gases were used, dry nitrogen and dry air. A hygrometer

was used to measure the relative humidity in the contact region while gas was flowing and

confirmed it to be less than 5%. For each set of conditions, the sliding tests were performed

at least 3 times to check for reproducibility. In the analysis that follows only one set of

representative data is shown.

Typically, when performing tribometer experiments, the coefficient of friction is mea-

sured for each sliding cycle, but wear (or other changes to the surface) can only be measured

for the final wear track after all sliding is done. One solution to this is to produce what

is known as a multisection wear track, or a “stripe test”. In this case, the track length

(initially 3 mm) is reduced after a set number of cycles, while the speed is held constant.

This periodic reduction in track length continues throughout the test so that the final wear

track is made up of a number of sections showing the progression of wear. An example

illustrating such a wear track is shown in Figure 2.2.

2.3. Basic principles of atomic force microscopy

One of the best tools for analysis of wear tracks post-sliding is the atomic force micro-

scope (AFM). The basic principle of the AFM is to measure the interaction between a sharp

Figure 2.2: Optical microscope image showing an example of a multisection wear track produced in dry N2. The number of sliding cycles for each section is labeled. The total length of the wear track is 3 mm. The image was produced by stitching together individual high magnification images.

29 probing tip and a sample surface. A schematic of an AFM setup is shown in Figure 2.3. In

AFM, the probing tips can have very low radii of curvature (as small as 10 nm), which allows for potentially sub-nanometer lateral resolution. The tip is attached to a flexible cantilever produced through microfabrication techniques, which is typically thin (0.5–1 µm), with a low spring constant (0.1–50 N/m), and a high resonance frequency (10 kHz – 1 MHz) [102].

In response to the normal and lateral forces between the tip and the sample, the cantilever is deflected. Most modern AFMs use the optical beam deflection technique to detect the cantilever deflection. In this scheme, a laser is focused on the back of the cantilever and reflects into a four quadrant photodetector which tracks the motion of the cantilever. Im- ages are acquired by rastering the tip over the sample and recording the normal and lateral deflection of the cantilever as a function of the position [102].

Typical tip-sample forces are between 10−11 and 10−6 N. In comparison, at separations

Figure 2.3: Top: Diagram of contact mode AFM setup. A cantilever with a sharp tip is in contact with the sample surface. A laser is focused on the back of the cantilever and reflects into a photodetector. Normal forces move the cantilever vertically, while lateral forces twist the cantilever left and right. The sample is moved horizontally with a piezoelectric scanner and the tip is rastered over the surface to produce an image [101] Bottom: Image of the end of a typical AFM cantilever showing the sharp tip. Tip radii can be < 10 nm.

30 of a few Angstroms,˚ the interaction between covalently bonded atoms is of the order 10−9 N.

Thus an AFM can be used for non-destructive measurement [103]. Frictional forces create lateral deflections of the cantilever and can be measured as well. This work was carriedout using a Dimension Icon AFM operating in contact mode (Bruker Corporation, Billerica,

MA, USA).

The raw output of the AFM is a signal in volts, and requires calibration to be converted to real units. Typical values of interest are the topography in nm or µm, the normal and frictional forces in nN, the contact diameter in nm, and the contact stress in GPa.

First, the normal deflection of the cantilever and the topography signal can be converted to units of length by measuring the deflection sensitivity of the cantilever, defined asthe change in photodetector voltage per nm of deflection. In performing this measurement, the amount of cantilever deflection is determined through a prior calibration of the piezoelectric scanner that moves the cantilever or some form of alternative sensor, such as a capacitive sensor [103]. Then if the normal spring constant of the cantilever is known, the normal deflection can be converted to normal force. There are several methods todeter- mine to normal spring constant, including calculations from beam bending formulas based on the cantilever dimensions, deflection of a reference cantilever with known spring constant, Sader’s method [104] (which is based on the hydrodynamic behavior of the cantilever oscillating in air), and the thermal tune method (which relates the spring constant to the mean square displacement of the cantilever oscillations in air) [105]. The thermal tune method is built into the Dimension Icon AFM, and was the primary method used for this work.

In general, the lateral force has two components, the friction force and a geometric component, when scanning on sloped surfaces. Generally, it is only the friction force which is of interest. The geometric component of the lateral force can be eliminated by subtracting the lateral force measurements in two opposite scan directions (referred to as the trace and retrace directions). Since the friction force has opposite signs depending on the sliding direction while the geometric component has the same sign in either direction, this subtrac-

31 tion will eliminate the geometric component while doubling the friction force. Calibrating the lateral deflection is more challenging, since the lateral deflection sensitivity isdifficult to measure and most of the normal force calibrations do not work for the lateral force [101].

One method to overcome these difficulties is to compute a single lateral force sensitivity

(which is a combination of the deflection sensitivity and the lateral spring constant) by scanning a sample with known slopes and taking advantage of the geometric component of the lateral force. This method is known as the wedge method [101].

Finally, the contact diameter and stress can be computed through contact mechanics if the tip radius of curvature is known. This can be measured via transmission electron microscopy (TEM), however, this is very time consuming. Instead the tip radius is often estimated from a procedure known as blind tip reconstruction (BTR) [106]. This procedure involves scanning a rough surface and observing that the resulting image is actually a combination of the tip and sample geometry. Mathematically, the topography image produced is a dilation of the real topography by the tip apex geometry. In the BTR method, an iterative algorithm is used to solve for the tip shape consistent with measured topography image [106].

Other than AFM, a second method to measure sample topography is scanning white- light interferometry. Interferometry provides 3D topography data based on the phase difference between light reflected from the sample surface and light from a reference mirror. This phase difference is used to determine the sample height at each pixel[107]. The advantages of interferometry are excellent vertical resolution (< 0.1 nm) and the ability to rapidly image large (> 100 × 100 µm) areas. However, disadvantages compared to AFM include limited lateral resolution (as an optical technique interferometry is diffraction-limited resulting in a lateral resolution of approximately 600 nm), and inability to measure any surface forces. In addition interferometry can be limited by the optical properties of the sample. This is particularly relevant for the NCD Technologies a-C:H:Si:O. This film was semi-transparent and only 40 nm thick on a polished silicon substrate. Because of this light can be reflected from both the film surface and the silicon substrate, resulting in artifacts in the interferometry

32 data for this sample.

2.4. Brief description of the Hertz theory of contact mechanics

In the study of tribology, theories of contact mechanics, which describes the stress and deformations that arise when two surfaces are brought into contact, are of great importance.

In particular, the problem of a sphere in contact with a flat surface is critical for describing both the tribological contact and the contact between the AFM probe and the surface. The first solution to this problem was by Heinrich Hertz in 1882 who considered the solution to the problem of two contacting spheres [108]. A detailed description of this solution and its derivation is provided by Johnson [109], here a brief description of the important results will be presented.

The Hertz model is derived for two contacting spheres under the assumptions of smooth, macroscopic contact between the two bodies undergoing elastic deformation which are ho- mogeneous, isotropic, and linear elastic. Furthermore, it is assumed there is no adhesion between the two surfaces and the deformations are small compared to the geometry of the contacting objects. Although the solution is for two spheres, this reduces easily to the case of a sphere on flat contact by letting the radius of the second sphere go to infinity. Hertz showed a solution for the normal pressure distribution within the contact that satisfies the displacement boundary conditions is given by [108, 109]:

2 1/2 p = p0{1 − (r/a) }

where p0 is the maximum pressure occurring at the center of the contact, r is the radial distance for the point of contact, and a is the radius of the contact area. From this the contact parameters of interest (contact radius a, indentation depth δ, and mean contact

33 pressure pm) can be derived and are given by:

3RL1/3 a = (2.1a) 4Ec a2 9L2 1/2 δ = = 2 (2.1b) R 16REc L 1 4E 2/3 p = = c L1/3 (2.1c) m πa2 π 3R

where R is the radius of curvature of the sphere, L is the normal load, and Ec is a composite elastic modulus based on the two contacting materials defined as:

1 1 − ν2 1 − ν2 = 1 + 2 Ec E1 E2 where E and ν represent the elastic modulus and Poisson’s ratio of the two materials in contact. The maximum contact pressure, occurring at the center of the contact is equal to

3pm/2. Some authors include a factor of 1/2 in the definition of the composite modulus [110], which results in a factor of two difference in the constants for the above equations.

Although the Hertz model is a reasonable description of contact mechanics, there are a number of limitations. The effect of shear forces and sliding between the two surfaces is not considered; it is assumed there is not slip between the surfaces (i.e., high local

friction); there is no adhesion force; the contact indentation is assumed to be much smaller

than the contact radius, and the contact radius much smaller than the sphere radii; and it

assumes only elastic deformations. All of these are often not true in tribological contacts. In

addition, it is assumed that the two surfaces can be considered as infinite half spaces, which

may not be valid for thin films on substrates. Later work has proposed a number ofmore

complex models that account for some of these limitations by relaxing the assumptions made

in the Hertz solution. For example, the Johnson-Kendall-Roberts (JKR) theory and the

Derjaguin-M¨uller-Toporov (DMT) theory include the effects of adhesion, although they do

so in different ways [111]. In general, the JKR theory works better for soft contacts, while the

DMT theory is preferred for contacts between hard materials with minimal deformation due

34 to adhesion [111]. The effect of frictional forces and sliding contacts has also been considered.

For contact between identical materials the normal pressure and contact area are the same as given by the Hertz model, although other stress components will be affected [109]. For dissimilar materials the deviation from the Hertz solution for normal pressure and contact area depends on the difference between the shear modulus and Poisson’s ratio of the two materials, but is generally small [109].

For this work, the basic Hertz theory is used as an approximation for computing contact area and normal pressure. Even though it is known that many of the assumptions in this model will not be true in a tribological contact, the model can still provide a reasonable approximation of the contact parameters. In addition, during sliding material transfer from the sample surface to the ball is known to produce third body films in the contact, which can have very different and hard to measure mechanical properties. Thus, although more complex contact mechanics models exist, there is limited improvement in predictions that can be achieved since the formation of transfer films and significant plastic deformation make any model only a rough approximation.

2.5. Environmental dependent tribology of a-C:H:Si:O

Samples of NCD Technologies a-C:H:Si:O were run in three different environments; laboratory air (40% RH), dry air (< 5% RH), and dry nitrogen (< 5% RH). A summary of the friction data is shown in Figure 2.4. The contact pressure computed from Hertz contact mechanics was 0.3 GPa. The sliding distance was progressively decreased during the experiments to produce multisection wear tracks. In each case the coefficient of friction initially drops during a period of time called “run-in”, and then stabilizes (typically in this case after 50–100 cycles). The coefficient of friction is, on average, 0.03 in dry nitrogen.

This increases to an average value of 0.05 in dry air once stabilized, and to 0.13 in humid air, although there is substantial variation seen in humid air. The amount of wear is also substantially different between each environment, with higher wear in humid and dry air, and a much lower wear rate in dry nitrogen. Arrows in Figure 2.4 show the approximate

35 0.25 Air Dry Air 0.20 N2

0.15

0.10

0.05 Coefficient of Friction (-)

0.00 0 400 800 1200 Number of Cycles

Figure 2.4: Coefficient of friction vs. sliding cycles obtained in three different environments, laboratory air (40% RH), dry air (< 5% RH), and dry nitrogen (< 5% RH) for NCD Technologies a-C:H:Si:O. In each environment the coefficient of friction initially drops during a run-in period, followed by stabilization. Friction is lowest in dry N2 and highest in humid air. The arrows indicate the approximate location where the film began to be worn through, based on measurements of maximum wear depth. As the films begin to wear through the coefficient of friction becomes unstable and begins to increase. location where the film first began to wear through (the point where the maximum wear depth becomes greater than or equal to the film thickness). As can be seen, this is strongly correlated with a rapid increase in friction.

Generally these results are consistent with prior literature on the friction of a-C:H:Si:O in different environments. Scharf et al. [17] measured the coefficient of friction between DLN

180 a-C:H:Si:O and steel in dry nitrogen, dry air, and humid air (50% RH) with a contact pressure of 0.3 GPa (the same as used here). The coefficients were 0.02 in dry nitrogen,

0.05 in dry air, and 0.22 in humid air. These environmental-dependent differences were attributed to changes in the interfacial shear strength of the tribofilms produced in each environment. In dry environments, the tribofilms were carbon-rich with large quantities of C-H bonding, resulting in a low shear strength (9 MPa), while in humid air SiOx-rich tribofilms formed with a much higher shear strength (78 MPa). The results hereonthe much more silicon and oxygen rich NCD Technologies a-C:H:Si:O is in excellent agreement

36 with these results, indicating that these mechanisms may be universal to a-C:H:Si:O and not limited to the specific atomic compositions studied.

Prior studies have reported much lower coefficients of friction (0.04–0.08) in humid environments [16, 91, 99]. This difference, however, is attributed to the much higher loads used (contact stresses over 0.8 GPa) [17]. This behavior will be discussed later in the context of the load dependent tribology results presented in Section 2.8.

It has previously been shown that for DLCs in general, the formation of a tribofilm on the steel surface is important for achieving low friction [12, 26, 50]. After the run-in period, sliding occurs at the DLC / tribofilm interface, and thus the steady state friction is controlled by the interfacial interactions between the tribofilm and the DLC [40, 51]. In all cases the formation of a tribofilm was observed on the steel ball. Optical microscope images of representative examples are shown in Figure 2.5 for each environment.

Characterization of the properties of these tribofilms was not a major component of the work in this thesis, however, detailed analysis of these films was carried out by McClimon et al. [112] and the key details will be briefly mentioned. The tribofilms formed in all three environments were surprisingly thick, ranging from 0.5–2 µm when measured with AFM (prior reports of DLC tribofilm thickness are in the range of 100 nm or less). Nanoinden-

tation performed on these tribofilms showed they were significantly softerE ( = 1–30 GPa,

H = 0.3–3.5 GPa) compared to the as-received film (E = 120 GPa, H = 11 GPa). There

(a) (b) (c)

Figure 2.5: Optical microscope images of tribofilms on steel balls formed in (a) laboratory air (40% RH), (b) dry air (< 5% RH), and (c) dry nitrogen (< 5% RH). The formation of tribofilms is known to be important for achieving low friction andwear.

37 was a large degree of lateral inhomogeneity in the films mechanical properties. Although there is a large degree of scatter in the data, the tribofilms formed in humid air generally have a higher average modulus and hardness compared to films formed in dry nitrogen and dry air. On the basis of spectroscopic measurements, the tribofilms are proposed to have a partially polymeric structure.

Using AFM, wear track dimensions were measured as a function of environment and sliding cycles. The results are shown in Figure 2.6. Since the NCD Technologies a-C:H:Si:O films were only 40 nm thick, the samples could be worn through relatively rapidly. The point at which the maximum wear depth reached the film thickness was observed to be well correlated with the sudden increase in friction seen in Figure 2.4 at the end of each test. As shown in Figure 2.6a, the maximum wear depth increases approximately linearly with the number of cycles, but at much different rates in each environment. In dry nitrogen, steady state, low friction sliding could be maintained without the film wearing through for over

1000 cycles, compared to less than 200 cycles in humid air.

The wear track width is primarily determined by contact mechanics, which depends on the mechanical properties of the tribofilms. As can be seen in Figure 2.6b, the wear tracks formed in dry environments have similar width, which is significantly larger than the wear tracks formed in humid air. It is proposed that the softer tribofilms formed in dry environments are more easily compressed, expanding the contact area, compared to the harder tribofilms formed in humid air. Although wear track width does increase over time, it is not strongly dependent on the number of cycles, and clear environmental differences are present even after only 10 cycles. This indicates that the tribofilm forms almost immediately after the start of sliding and requires minimal run-in. Finally, the wear track cross sectional area is shown in Figure 2.6c. The cross-sectional area is proportional to the worn volume, which is typically the quantity of interest in characterizing a material’s wear resistance.

The overall rate of wear is much higher in humid air and dry air compared to dry nitrogen.

38 60 80 Air Air Dry Air Dry Air 50 N2 70 N2 40 60 30 50 20

10 40 Wear Track Width (µm) Maximum Wear Depth (nm) 0 30 0 400 800 1200 0 400 800 1200 Number of Cycles Number of Cycles (a) (b)

1.0 )

2 0.8

0.6

0.4

0.2 Air

Wear Track Area (µm Dry Air 0.0 N2

0 400 800 1200 Number of Cycles (c)

Figure 2.6: Wear track dimensions measured with AFM for laboratory air (40% RH), dry air (< 5% RH), and dry nitrogen (< 5% RH). (a) Maximum wear track depth as a function of sliding cycles. The total a-C:H:Si:O film thickness was 40 nm (indicated by the gray line). Maximum wear track depth increases with the number of cycles. (b) Wear track width as a function of sliding cycles. The wear tracks formed in dry environments have similar width, which is significantly larger than the wear tracks formed in humid air. The wear track width is primarily determined at the start of sliding by contact mechanics, which are effected by the properties of the tribofilms. In dry environments a wide transfer film formed, expanding the contact diameter. (b) Wear track cross sectional area as a function of sliding cycles. The overall amount of wear is greatest in humid air and least in dry N2.

39 2.6. Wear track topography measured with AFM

In order to gain understand more of the mechanisms controlling the friction and wear behavior in different environments, AFM topographic and friction images were obtained on the different wear tracks. These revealed interesting nanoscale features that provide insight into the mechanisms occurring during run-in and controlling the friction and wear.

Figure 2.7 shows AFM topography images of wear tracks formed in dry nitrogen and dry air after 50 sliding cycles. Both dry environments are considered together since the nanoscale behavior is similar. For both cases, the maximum wear depth occurs in a very narrow region at the extreme edges of the contact as seen by the formation of a set of narrow grooves, approximately 2–3 nm deep. The center of the wear track is flat, and the wear depth in the center, ∼1 nm, is significantly less than the maximum wear depth in the grooves. A particularly clear example of this topography for a much deeper wear track is shown later in Figure 2.20b. Wear events inside the track appear to occur at localized points where a few nanometers of material is removed.

For the case of dry nitrogen, where the wear volume is smallest, it is interesting to

10.5 nm 17.6 nm

20 µm 0.0 nm 20 µm 0.0 nm (a) (b)

Figure 2.7: AFM topography images of wear tracks formed on NCD Technologies a-C:H:Si:O in dry nitrogen (a) and dry air (b) after 50 sliding cycles. Wear tracks in both dry environments appeared similar, but with higher wear in dry air than dry N2. In both cases the maximum wear depth occurs at the side of the contact due to the formation of narrow grooves along the edge of the wear track, while the center of the wear track is a flat plateau. This unusual topography is highly reproducible for a-C:H:Si:O.

40 note that the wear track surface is elevated compared to the surrounding film. A line profile showing this is in Figure 2.8. This suggests that structural changes are taking place causing expansion in the near surface region. One possibility is rehybridization from sp3 to sp2 carbon, caused by stress and frictional heating, which results in a reduction in density. Many prior experimental studies on the tribology of DLC have shown spectroscopic evidence an increase in sp2 carbon in the wear track [49, 53, 56, 57]. In addition, MD simulations of sliding DLC surfaces have also seen increased sp2 carbon at the interface [58–

61]. Furthermore, simulations by Pastewka et al. [59] have shown that the sp2 carbon rings at the sliding interface are preferentially oriented to be parallel to the surface of the film. It is expected that rehybridization 3from sp to sp2 carbon in such a manner

would result in expansion normal to the wear track surface, because of the larger van

der Waals spacing between planar sp2-bonded regions compared to covalent C-C bond

lengths [67]. Spectroscopic evidence in support of this proposed mechanism will be shown

later in Section 2.9. It is hypothesized that the same processes are occurring in the dry air

wear tracks as well, but because of the higher wear rate an elevated surface is not observed.

0 Depth (nm) -2

-4 0 20 40 60 Position (µm)

Figure 2.8: Line profile from AFM topography measurement of wear track formed indry nitrogen after 50 sliding cycles. Large regions of the wear track are raised compared to the as-received surface, indicating either the formation of a tribofilm on the a-C:H:Si:O surface or a structural change resulting in near surface expansion.

41 Higher resolution magnified images for the dry nitrogen wear tracks are shown in Fig- ure 2.9 and for dry air in Figure 2.10 along with corresponding friction images. The wear track topography shows clear evidence for the dominant role of adhesive interactions in the tribology of a-C:H:Si:O. Material removal is observed to occur at localized regions consistent with the idea of forming and breaking adhesive junctions. There is higher friction inside the worn areas compared to the surrounding surface.

7.0 nm 0.26 V

2 µm 0.0 nm 2 µm 0.12 V (a) (b)

6.3 nm 0.28 V

500 nm 0.0 nm 500 nm 0.10 V (c) (d)

Figure 2.9: Magnified AFM scans of wear tracks produced in dry nitrogen after 50 sliding cycles. (a) Topography and (b) friction for a 5 × 5 µm2 area in the center of the wear track. Material removal occurs at localized regions through the breaking of adhesive junctions. There is higher friction inside the worn areas compared to the surrounding surface. (c) Topography and (d) friction for a magnified 1.5 × 1.5 µm2 area surrounding a single wear site. The surface has significant texturing, with clear friction contrast. Interestingly, other than the elongation of the wear events parallel to the sliding direction, the modified wear track surface appears isotropic. There is no clear indication of sliding direction.

42 The wear tracks formed in dry air display similar topography to the tracks formed in dry nitrogen. However, the area ratio of higher friction regions to the lower friction regions in the wear track is larger than in dry nitrogen. This topography is consistent with the higher friction and increased wear observed when oxygen is present. These changes can be attributed to the increased reactivity of the dry air environment, which includes oxygen.

Spectroscopic analysis of DLN 180 a-C:H:Si:O wear tracks by Koshigan et al. [100] has shown a significant increase in C-O bonding in wear tracks formed while sliding in oxygen gas. The reaction between environmental oxygen and the wear track surface can break C-C bonds leading to increased wear. This would explain the higher ratio of worn surface in for the wear tracks formed in dry air.

Figure 2.11 shows an AFM topography image of a wear track formed in humid air after

50 sliding cycles, and magnified AFM scans are shown in Figure 2.12. There are several differences between the wear tracks formed in dry and humid environments. In humidair, the tracks are deeper and narrower. The localized wear events and surface texture are no longer present, while the dominant topographic features are ridges and grooves running parallel to the sliding direction. There are several possible explanations for the observed differences in humid air. First, as mentioned previously the average mechanical properties

(modulus and hardness) are higher for the tribofilms formed in humid environments [112].

The harder regions in the humid air tribofilms could give rise to a larger abrasive component to the wear and increased difficulty to shear the film at the interface. In addition, the higher modulus is related to a more compact tribofilm which increases the mean contact pressure compared to dry environments where the soft tribofilms significantly expand the contact area, distributing the load. Finally, the increased reactivity of the environment due to the presence of water vapor could increase the tribochemical reactions that facilitate wear, similar to the observations for the effect of oxygen.

43 11.0 nm 0.055 V

2 µm 0.0 nm 2 µm 0.015 V (a) (b)

6.6 nm 0.045 V

500 nm 0.0 nm 500 nm 0.01 V (c) (d)

Figure 2.10: Magnified AFM scans of wear tracks produced in dry air after 50 sliding cycles. (a) Topography and (b) friction for a 5 × 5 µm2 area in the center of the wear track. Magnified wear track topography is similar to sliding in dry nitrogen, however, the relative ratio of higher friction adhesive wear events to low friction wear track surface is larger than in dry nitrogen. This is consistent with the higher friction and increased wear observed when oxygen is present. (c) Topography and (d) friction for a magnified. 1 5 × 1.5 µm2 area.

2.7. Importance of transfer film in achieving low friction for a-C:H:Si:O

Tribometer experiments were also performed on DLN 360 a-C:H:Si:O in each environ-

ment. A summary of the friction results is shown in Figure 2.13. In this case substantially

different behavior was observed compared to the NCD Technologies samples. Friction in

humid air was similar to previously discussed measurements. However, the friction in dry

nitrogen was highly unstable, while the friction in dry air was extremely high.

An understanding of the cause of this behavior can be found by observations of the balls

44 33.2 nm

20 µm 0.0 Figure 2.11: Wear track formed in humid air (40% RH) after 50 sliding cycles. The wear tracks formed in humid air are significantly different than what was observed in dry environments. The tracks are both deeper and narrower, with different topography inside the wear track.

12.0 nm 7.6 nm

2 µm 0.0 nm 500 nm 0.0 nm (a) (b)

Figure 2.12: Magnified AFM scans of wear tracks produced in humid air after 50 sliding cycles. (a) Topography scan of 5 × 5 µm2 area. (b)1 .5 × 1.5 µm2 topography scan of the same area. The detailed features on the wear track surface are significantly different in humid air. The localized wear events and isotropic surface texture are no longer present, instead the dominant topography is ridges and grooves running parallel to the sliding direction. post-sliding. Figure 2.14 shows optical microscope images of the contact area on the steel balls after sliding in humid air and dry air. In the case of humid air, a tribofilm is present, similar to the case for the NCD Technologies samples shown in Figure 2.5. However, in dry air no tribofilm is present, and there appears to be wear of the steel (there aresmall scratches in the steel running parallel to the sliding direction and large contact area is only possible for a flattened sphere). There is a ring of transferred material outside of the contact

45 0.6 Dry Air Nitrogen Lab Air 0.4

0.2 Coefficient of Friction (-)

0.0 0 100 200 300 Number of Cycles

Figure 2.13: Coefficient of friction vs. sliding cycles obtained in three different environments, laboratory air (40% RH), dry air (< 5% RH), and dry nitrogen (< 5% RH) for DLN 360 a- C:H:Si:O. Humid air behavior is similar to previous measurements. However, the friction in dry air rapidly becomes very high. Post sliding imaging of the steel balls revealed no transfer film buildup. The lowest coefficient of friction was measured in dry nitrogen, however,the behavior was unstable.

zone, suggesting some material transfer from the film to the ball did occur, but wasnot

strongly adherent enough to remain in the contact. A similar dependence of the formation

of a tribofilm on the environment was observed by Koshigan etal.[100] on DLN 180 a-

C:H:Si:O. In that study, sliding in high vacuum resulted in coefficients of friction greater

than 1. It was observed that in these conditions, strong adhesive junctions were formed

between the a-C:H:Si:O and the steel which resulted in wear of the steel, rather than the

formation of a tribofilm.

This relationship between the tribofilm and friction can be clearly established by looking

at the data from sliding in dry nitrogen. As mentioned previously, this resulted in the lowest

friction, but the behavior was unstable. An extreme example of this friction instability is

shown in Figure 2.15. The friction initially is stable at a low value (close to 0.05), but then

rapidly increases. After each spike, the friction begins a new run-in, and returns to a low

value.

To investigate the reason for this behavior a set of experiments was performed where the

46 (a) (b)

Figure 2.14: (a) Optical microscope image of tribofilm formed on the steel ball after sliding in humid air on DLN 360 a-C:H:Si:O. (b) Steel ball after sliding in dry air. No tribofilm is present, and there appears to be wear of the steel flattening the end of the ball. There is a ring of transfered material outside of the contact zone, suggesting material transfer from the film to the ball did occur, but was not strongly adherent enough to remain inthe contact. This set of conditions results in high friction and wear.

0.5

0.4

0.3

0.2

0.1 Coefficient of Friction (-)

0.0 0 100 200 300 Cycle Number

Figure 2.15: Extreme example of the unstable frictional behavior of DLN 360 a-C:H:Si:O in dry nitrogen.

47 test was stopped in both the low friction and high friction state, as shown in Figure 2.16.

Optical microscope images of the steel balls for each case are shown in Figure 2.17. When the test was stopped while friction was high, no tribofilm is present in the contact. The contact area appears similar to the case in dry air. However, when the test was stopped while friction was low, a tribofilm is found on the ball. The contact area is much smaller and there is no evidence of wear of the steel. An explanation for the friction instability observed in dry nitrogen is that the tribofilms are only weakly adherent to the steel. Ifa film is present, friction and wear is low, however, the film can be removed from theball, causing an increase in friction. A new film can form however, which will cause a return to low friction behavior.

One question to consider is why the tribofilm formation and adhesion is inhibited in

DLN 360 a-C:H:Si:O compared to the NCD Technologies samples and literature results for

DLN 180 films [17, 100]. The reasons for this behavior are unknown and would require further study. However, it can be reasonably expected that this is related to the atomic

0.6

0.4

0.2 Coefficient of Friction (-)

0.0 0 100 200 300 Number of Cycles

Figure 2.16: Coefficient of friction vs. sliding cycles for DLN 360 a-C:H:Si:O in dry nitrogen. The low friction behavior was highly unstable. In two cases the test was stopped while the friction was high (solid lines), and in two cases the test was stopped while the friction was low (dashed lines). Imaging of the balls revealed that transfer films were present for the test stopped while the friction was low, but not for the tests stopped while the friction was high.

48 (a) (b)

Figure 2.17: (a) Optical microscope image of steel ball after sliding in dry nitrogen after the test was stopped when friction was high. Similar to the case for dry air, no tribofilm is present in the contact. (b) Optical microscope image of steel ball after sliding in dry nitrogen after the test was stopped when friction was low. In this case a tribofilm is found on the ball. The contact area is much smaller and there is no evidence of wear of the steel. An explanation for the friction instability observed in dry nitrogen is that the tribofilms are only weakly adherent to the steel. If a film is present, friction and wear is low, however, the film can be removed from the ball, causing an increase in friction. A new filmcanform however, which will cause a return to low friction behavior. composition of the DLN 360 a-C:H:Si:O, which had the highest silicon content and the lowest ratio of O:Si. For example, Koshigan et al. [100] hypothesizes that for DLN 180 a-C:H:Si:O, tribofilm formation is facilitated by rehybridization 3from sp to sp2 carbon in the wear track due to the effects of stress and frictional heating. 2This sp surface layer is easily etched by environmental species such as oxygen leading to material transfer from the film to the ball. This results in the buildup of2 asoft,sp rich tribofilm that facilitates easy shear. In the case of the DLN 360 samples, the high silicon content and low O:Si ratio suggests the highest fraction of C-Si bonding. As will be shown in Chapter5, C-Si bonding significantly favors sp3 carbon hybridization and enhances thermal stability by increasing the mean activation energy for sp3 to sp2 conversion. Then a possible explanation for the difficulty in tribofilm formation when sliding on DLN 360 a-C:H:Si:O is that theformation of sp2 rich carbon is inhibited by the high fraction of Si-C bonding.

49 2.8. Load dependent tribology of a-C:H:Si:O in humid air

With assistance from Nikolay Garabedian, Ph.D. student at the University of Delaware in the group of Prof. D. L. Burris, a reciprocating ball on flat tribometer was used to investigate the load dependence of DLN 360 a-C:H:Si:O in laboratory air (∼ 30% RH). This tribometer did not have environmental control, but otherwise was similar to the previously described gas blowing tribometer used for the environmental tribology results, including the ability to produce multisection wear tracks. For these tests a-C:H:Si:O was slid against 1/4 inch (6.35 mm) diameter 52100 steel ball bearings at a speed of 3 mm/s, and the load was varied between 84 and 5000 mN. This corresponded to a Hertzian contact pressure between

0.15 and 0.6 GPa.

The coefficient of friction vs. sliding cycles for each load is shown inFigure 2.18. Al- though noisy, the data show a general trend of coefficient of friction decreasing with load.

For these films, since the total film thickness was2 µm, the coatings never wore through, unlike the 40 nm thick NCD Technologies samples shown in Figure 2.4.

0084 mN 0750 mN 0150 mN 1000 mN 0360 mN 2015 mN 0.5 0530 mN 5000 mN

0.4

0.3

0.2

0.1 Coefficient of Friction (-)

0 0 200 400 600 800 1000 Cycles (-) Figure 2.18: Coefficient of friction for DLN 360 a-C:H:Si:O sliding in humid air (25–30% RH) as a function of load and cycles. The steady state friction coefficient is observed to decrease with load. These tests were also performed as multisection wear tracks; the spikes in the data (which are most noticeable at low loads) are artifacts due to the changing track length.

50 From the data in Figure 2.18, the steady state coefficient of friction was defined asthe average coefficient of friction from 200 to 1000 cycles. Figure 2.19 shows a plot of this steady state coefficient of friction vs. load. The friction coefficient decreases from 0.27 at84mNto

0.13 at 5000 mN. The data are very well fit with a power law function: µ ∼ L−0.179. For a single asperity contact, the friction force is expected to scale with real area of contact.

Based on Hertz contact mechanics the contact area is given by Equation 2.1a and scales with load to the 2/3 power. Since the coefficient of friction is defined as the friction force divided by the normal load, this suggests the coefficient of friction should scale as loadto the −1/3 power. The discrepancy between the exponent in the observed behavior and the

Hertz model predictions is likely due to underestimation of the real area of contact when using Hertz theory, which does not account for the presence of the tribofilm.

Similar power law scaling of the friction coefficient with load has been observed for other solid lubricant coatings such as some DLCs and MoS2 [17]. The coefficient of friction in humid air measured here is higher than what is reported in many prior papers on the friction of DLN 180 a-C:H:Si:O in humid air, which report friction coefficients less than

0.3

0.25

0.2

0.15

0.1

Coefficient of Friction (-) Friction of Coefficient 0.05

0 0 1000 2000 3000 4000 5000 Load (mN)

Figure 2.19: Steady state coefficient of friction for DLN 360 a-C:H:Si:O sliding in humid air (25–30% RH) as a function of load. The data is well fit with a power law. For a single asperity contact, the coefficient of friction is expected to scale with real area of contact. Based on Hertz contact mechanics this suggests the coefficient of friction should scale as load to the −1/3 power. The discrepancy between the observed behavior and Hertz model predictions is likely due to underestimation of the real area of contact when using Hertz theory.

51 0.1 [16, 91, 99]. However, in those papers, a higher load range was used, resulting in contact stresses greater than 0.8 GPa. Since the coefficient of friction scales with load, these results can be consistent with prior literature when extrapolated to a higher load range, which gives a coefficient of friction equal to 0.1 at a contact pressure. of0 95 GPa. This is still slightly higher than the previously reported values. However, those measurements were on a different composition a-C:H:Si:O and at different relative humidity. Similar behavior is observed by Scharf et al. [17] on DLN 180 a-C:H:Si:O, where the coefficient of friction showed a power law decrease from 0.22 at a contact pressure of 0.3 GPa to 0.09 at a contact pressure of 0.6 GPa, although they only considered three different loads.

AFM measurements were performed to quantify the amount of wear and specific details of the wear track topography. In general, the wear tracks appear similar to previous experiments in humid air and the qualitative features did not depend on load. A representative example of the wear track topography is shown in Figure 2.20. As before, the maximum wear occurred at the sides of the wear track in two grooves, with a level central plateau.

This topography was generally observed for all loads and cycles, but was most pronounced for higher loads after 50–100 cycles.

163 nm 0

-50

-100 Depth (nm)

-150 0 20 40 60 20 µm 0 nm Position (µm) (a) (b)

Figure 2.20: (a) AFM topography image of wear track formed with a 1000 mN load after 100 cycles. (b) Line profile showing the wear track cross section. Again the maximum wear depth occurs at the edge of the contact in two narrow grooves, while the central region forms a level plateau.

52 The exact cause of these grooves is unknown, however, we can propose two possibilities.

First, this groove formation could be facilitated by enhanced etching of the film near the edge of the contact zone, due to increased exposure to environmental gases while experiencing normal and shear stresses. This would be consistent with the idea that the presence of reactive species, such as oxygen and water vapor, can facilitate the breaking of C-C bonds in a-C:H:Si:O [100]. Second, the grooves could be due to mechanical stress. For a Hertzian contact, the radial stress component becomes tensile outside the contact area, with the maximum tensile stress occurring at the edge of the contact [109]. If the wear of a-C:H:Si:O is greater when stressed in tension, this stress state could lead to higher wear at the edge of the track. Although this tensile region extends around the circumference of the contact, due to the motion of the ball, the material parallel to the edge of the wear track is exposed to a greater sliding duration of tensile stress compared to the center of the contact.

From the AFM wear track measurements, it was possible to quantify how the wear track topography scaled with load and number of cycles. Since the total film thickness was 2 µm interferometer measurements could also be used to determine wear track topography, and agreed with AFM results. This was used for determining wear track width at the higher loads, where the wear tracks did not fit in the maximum AFM scan size (95 × 95 µm2). Figure 2.21 shows the dependence of wear track width on sliding cycles and load. The track width is determined by contact mechanics (which depends on the mechanical properties of the tribofilms). Tribofilm formation occurs within the first 10 cycles, after whichthe contact width shows only a slight increase with sliding cycles. However, the track width shows a strong dependence on the load, as would be expected based on contact mechanics.

The dashed line in Figure 2.21b shows the contact diameter predicted by the Hertz model, given by Equation 2.1a. At low loads the track width agrees well with the Hertz predictions, while at high loads the model significantly underestimates the contact diameter. This is expected to occur due to deformation of the soft tribofilm, which expands the contact area.

The fact that the real contact diameter is larger than the Hertz model predictions at high loads explains the shifted power law scaling of the friction coefficient shown in Figure 2.19.

53 150 150 0084 mN 0150 mN

0360 mN m) m) 0530 mN 100 100 0750 mN 1000 mN 2015 mN 5000 mN 0010 Cycles 50 50

0050 Cycles Average width ( width Average Average width ( width Average 0100 Cycles 0500 Cycles 1000 Cycles 0 0 0 200 400 600 800 1000 0 1000 2000 3000 4000 5000 Cycles (-) Load (mN) (a) (b)

Figure 2.21: (a) Wear track width as a function of cycles. The average width is set by the contact mechanics, which depends on the properties of the transfer film. The transfer film forms within the first 10 cycles, after which the contact diameter only increases slightly with sliding cycles. (b) Wear track width as a function of load. The dashed line shows the predicted contact diameter based on Hertz contact mechanics. At low loads the wear track width closely follows the Hertz model predictions, while at higher loads the model underestimates the contact diameter. This is most likely due to the softer tribofilm being deformed at high loads causing it to spread out and expand the contact area.

An increased real area of contact vs. load will result in a more positive power law exponent, as is observed.

Figure 2.22 shows the wear track depth as a function of sliding cycles and load. Due to the unique wear track topography (see Figure 2.20), the median depth is plotted, since it is more representative of wear track geometry than the mean. As would be expected, the wear depth scales linearly with the number of sliding cycles. However, the median depth is observed to be independent of the load. The general explanation for this behavior, which will be investigated further in Section 2.10, is that with increasing load the wear volume increases, but this is accommodated by the increase in wear track width, resulting in constant wear depth.

54 350 350 0084 mN 0010 Cycles 300 0150 mN 300 0050 Cycles 0360 mN 0100 Cycles 250 0530 mN 250 0500 Cycles 0750 mN 1000 Cycles 200 1000 mN 200 2015 mN 150 5000 mN 150

100 100

Median Depth (nm) Depth Median (nm) Depth Median 50 50

0 0 0 200 400 600 800 1000 0 1000 2000 3000 4000 5000 Cycles (-) Load (mN) (a) (b)

Figure 2.22: (a) Wear track median depth as a function of cycles. Due to the presence of the previously discussed grooves on the sides of the wear tracks, the median depth provides a more physically meaningful parameter rather than the mean. As expected the depth scales linearly with the number of cycles. (b) Wear track depth as a function of load. The median depth is observed to be independent of load.

2.9. Investigating changes in film structure through Raman spectroscopy

One question of interest is whether chemical changes are occurring in both the worn surface and the tribofilms in addition to the mechanical changes observed. For example, frictional heating in combination with stress could result in structural transformations such as the conversion from sp3 to sp2 carbon. Increased sp2 bonding in DLC wear tracks has been proposed as an important part of the run-in process leading to low friction [53, 56, 57].

NEXAFS measurements on worn ta-C [49] and DLN 180 a-C:H:Si:O [100] have previously shown an increase in sp2 carbon near the surface.

Raman spectroscopy is a useful tool commonly used to characterize carbon-based materials. A detailed description of the theory of Raman spectroscopy as it applies to DLCs is given in Chapter3, where it is used to investigate the thermal degradation pathways in a-C:H:Si:O. Briefly, the Raman spectra for DLC based films is characterized byabroad peak between 1300 and 1600 cm−1, which is typically fit with two Gaussian components called the D and G bands. This peak is due to vibration modes of sp2 carbon bonds, but can be related to the sp3 fraction [18, 113]. Increases in the intensity of this peak and

55 shifts to lower wavenumbers can suggest an increase in sp2 carbon, while increases in the photoluminescence (PL) background slope of the spectra indicate a reduction in dangling bonds due to increased passivation or clustering of the sp2 carbon [18, 76, 113].

Figure 2.23 shows Raman spectra acquired from as-received DLN 360 a-C:H:Si:O compared to the worn surface in the center of wear tracks at different loads after 1000 sliding cycles. There are only minor changes observed between the spectra acquired inside and outside of the wear tracks. However, Raman is not a surface-sensitive technique, so structural changes near the surface are only a small part of the total contribution to the spectrum.

These changes are more clearly resolved in the difference spectra shown in Figure 2.23b.

The increase in peak intensity and shift to lower wavenumbers of the peak position indicates an increase in sp2 carbon, while the increase in the slope of the background suggests an increase in passivation of dangling bonds or clustering of sp2 carbon.

Raman measurements were also acquired from the tribofilms produced at different loads.

a.u.)

(

i i 0

te 10 As-received te

In 530 mN In 530 mN 1000 mN 1000 mN -10 2015 mN 2015 mN 5000 mN 5000 mN 0 500 1000 1500 2000 2500 500 1000 1500 2000 2500 -1 -1 Raman Shift (cm ) Raman Shift (cm ) (a) (b)

Figure 2.23: (a) Raman spectra showing as-received film (DLN 360 a-C:H:Si:O) compared to wear track surfaces formed under different loads. Spectra are generally similar, with small differences. In order to more clearly see the changes, difference spectra betweenthe wear track surfaces and the as-received film are shown in (b). In general, an increase in peak intensity and a shift to lower wavenumbers of the peak position indicate an increase in sp2 carbon, while an increase in the slope of the background indicates an increase in passivation of dangling bonds or clustering of sp2 carbon.

56 Example spectra are shown in Figure 2.24. These spectra show significant changes from the as-received film and are no longer characteristic of DLC. It is difficult to characterize the specific structure of the tribofilms from these spectra, but they indicate that tribofilm formation is not simply material from the film transferred to the ball, but also involves significant structural and chemical changes. One possibility is segregation oftheSiOx or an increase in SiOx concentration in the tribofilms. This has been clearly shown by McClimon et al. [112] based on electron energy loss spectroscopy (EELS) measurements of tribofilms formed when sliding against NCD Technologies a-C:H:Si:O. In Raman spectroscopy, SiOx exhibits strong photoluminescence at higher oxygen ratios (x ≥ 1.3) [114], and an order of magnitude increase in the PL background intensity for a-C:H:Si:O films with varying silicon content has been attributed to segregation of the SiOx phase [94].

2.10. Modeling wear of a-C:H:Si:O as a function of the energy dissipated by friction

When discussing wear, a parameter of interest is frequently a materials wear rate (W ), which is defined as the volume of material removed per unit of sliding distance: W = V/s.

Figure 2.24: Raman spectra acquired from tribofilms formed on steel balls after sliding DLN 360 a-C:H:Si:O in humid air at different loads compared to the as-received film. Thereis a massive increase in intensity under the same acquisition conditions, and no evidence of the peaks associated with DLC. It is not possible to derive bonding composition from these spectra, but the Raman data clearly indicates that substantial chemical changes occur to the a-C:H:Si:O that is worn away and incorporated into the tribofilm.

57 From this definition, a simple model of wear was proposed by Archard [115] based on the assumption of hemispherical wear particles with volume, V ∼ 2πa3/3. Where a is the contact radius for a single asperity. From this, after summing over all asperities in contact one obtains: V A W = = κ r s 3 where Ar is the total real area of contact and κ is a proportionality constant describing the probability that a given contact point results in wear. For a material with hardness H undergoing primarily plastic deformation Ar = L/H, which then gives:

L V = Ks H where V is the wear volume (m3), s is the sliding distance (m), L is the load (N), H is

the hardness (Pa), and K is a proportionality constant called the wear coefficient. This

has become known as the Archard wear equation [102, 110]. Frequently however, since the

hardness can be difficult to determine accurately (especially for thin films) a modified wear

equation is used with the form:

V = ksL (2.2) where k is a dimensional wear coefficient [102, 110]. Here k is typically expressed in units of mm3/Nm, and V is in mm3. In the literature this wear coefficient is used to describe and quantify a material’s wear resistance. Generally, a material would be considered to have very good wear resistance if the dimensional wear coefficient is about− 10 7 or lower [10, 110].

For example, review articles frequently describe the wear resistance of DLC with different compositions or in different environments in terms of the dimensional wear coefficient k [10].

It has been noted, however, that Archard’s wear equation is purely empirical and wear coefficients computed from it may not be transferable to different sliding conditions [102].

For example, the wear rate of DLC is known to be correlated with the friction, and thus strongly depends on the environment [10].

From the wear measurements discussed in Section 2.8 the wear coefficient of DLN 360

58 a-C:H:Si:O in humid air was computed. When computing the wear coefficient from Equa- tion 2.2 it is convenient to note that the sliding distance is equal to the track length (ℓ)

times the number of sliding cycles, s = Nℓ, while wear volume is equal to the cross-sectional

area times the number of sliding cycles as well, V = Aℓ. Therefore Equation 2.2 can be rewritten as:

A = kNL which is easier to apply in the case of a multisection wear track. Figure 2.25 shows a plot of wear track cross sectional area vs. load × number of cycles, where the wear coefficient is given by the slope. Archard’s wear model predicts the data should collapse to a single line.

However, it is clearly observed that the wear coefficient decreases with increasing load.

In general this suggests that the wear coefficient is a poor method by which to quantify the wear resistance of a-C:H:Si:O, since the value is only meaningful in the context of a specific set of experimental conditions. For example, in a review of wear performance of

DLC coatings by Ronkainen and Holmberg [10], the reported wear coefficients from different studies are all acquired at different loads, making comparison between the different DLC

k = 4.6 10-5 mm3/Nm 150 30 k = 7.6 10-6 mm3/Nm

) 5000

2 m 0084 mN 20 0150 mN 0360 mN 0530 mN 0750 mN Worn Area ( Area Worn 10 1000 mN 2015 mN 5000 mN 0 0 1 2 3 4 5 Load x Cycles ( 106 mN)

Figure 2.25: Wear track cross sectional area vs. load × number of cycles. Archard’s wear equation predicts that the cross section area should be a linear function of load × number of cycles, where the slope is the dimensional wear coefficient. This model predicts the data should collapse on a single line. However, it is observed instead that the wear coefficient is inversely proportional to the load.

59 compositions or sliding environments questionable.

It would be better to quantify wear in a more physically meaningful manner. One possibility, motivated by the fact that wear in DLC is generally correlated with the coefficient of friction [10], would be to model wear as proportional to the work done by the friction force [116, 117]: Z ℓt V = k F (ℓ) dℓ (2.3) 0 where F (ℓ) is the friction force as a function of sliding length. This approach has the advantage of taking into account the changing friction force that occurs due to run-in and tribofilm formation [116, 117]. Previously this method has been applied to ta-C coatings, where the wear was shown to be the result of tribochemical reactions [118]. From this, the energy wear coefficient was found as a function of temperature and used to computean activation energy for tribochemical wear [118]. Wear volume vs. frictional energy is plotted in Figure 2.26. As can be seen, such an approach reasonably collapses the data onto a single line, although some variation still exists.

Additionally, it is interesting to note that in some cases this approach appears to work for environmentally-dependent friction and wear as well. Figure 2.27a shows wear volume vs. energy dissipated by friction for the NCD Technologies a-C:H:Si:O wear data from Sec-

) 3

m 3

3 10 0084 mN 2 0150 mN 0360 mN 0530 mN 1 0750 mN 1000 mN

Worn Volume ( Volume Worn 2015 mN 5000 mN 0 0 10 20 30 40 50 Dissipated Energy (J)

Figure 2.26: Wear volume vs. energy dissipated by friction. Modeling wear as proportional to the energy dissipated by friction reasonably collapses the wear volume data onto a single curve, although some variation still remains.

60 tion 2.5. As can be seen, although the wear rate in each environment (shown in Figure 2.6c) is different, when plotted as a function of work done by friction, the data collapse toasingle trend. However, for the DLN 360 a-C:H:Si:O samples from Section 2.7 (Figure 2.27b), it is clear that a single trend is not observed. A reasonable explanation for this difference exists. In the case of the NCD Technologies a-C:H:Si:O, in all environments the tribological behavior was controlled by adhesive transfer from the film to the steel, resulting inthe formation of a tribofilm. Steady-state sliding then took place at the a-C:H:Si:O / tribofilm interface. However, for DLN 360 a-C:H:Si:O well adhered tribofilms did not form in dry environments. When a tribofilm did not form, sliding took place at the a-C:H:Si:O /steel interface, resulting in much higher friction and wear, including wear of the steel. This difference in the tribological interface (a-C:H:Si:O / tribofilm vs. a-C:H:Si:O / steel) could substantially change the energetics of the wear process.

500 20 Air Air )

Dry Air 3 Dry Air 400 ) 3 N2 15 N2 µm 3 300 10 200

100 5 Wear Volume (µm Wear Volume (x10 0 0 0 2 4 6 8 10 0 20 40 60 80 100 Dissipated Energy (mJ) Dissipated Energy (mJ) (a) (b)

Figure 2.27: Wear volume vs. energy dissipated by friction for NCD Technologies a-C:H:Si:O (a) and DLN 360 a-C:H:Si:O (b) in different environments.

61 2.11. Conclusions

The tribology of a-C:H:Si:O was investigated through ball on flat reciprocating motion tribometer experiments over a range of loads and environments. The use of multisection wear tracks allowed the progression of wear to be monitored in addition the coefficient of friction. NCD Technologies a-C:H:Si:O displayed similar environmental trends to a-C:H, however, the increase in friction coefficient with humidity (0.03–0.05 in dry environments vs. 0.13 in humid air) was more moderate. This behavior is consistent with prior literature on DLN 180 a-C:H:Si:O which also shows a lower coefficient of friction in dry environments (0.02–0.05) compared to humid air (0.22) [17]. Analysis of wear track topography through AFM clearly shows wear is progressing through dominantly adhesive interactions

(formation and breaking of adhesive junctions in the film). Wear increases in more reactive environments, such as dry air compared to dry nitrogen, consistent with the idea of reactive environmental species etching carbon bonds in the film. In AFM measurements this is evident due to an increase of adhesive wear events present in the topography images.

The run-in process for a-C:H:Si:O consists of surface modification (possibly due to sp3

to sp2 conversion) and tribofilm formation. These run-in processes, which are critical to achieving low steady state friction and wear, depend on film composition and structure.

In the case of the high silicon DLN 360 a-C:H:Si:O stable tribofilm formation was difficult in dry environments, preventing the typical low friction behavior. Further work is needed to determine the precise mechanisms behind the inhibited tribofilm formation for DLN

360 films, however, one possible hypothesis is that the increased Si-C bonding limitedthe formation of sp2 carbon during sliding.

Friction vs. load measurements on DLN 360 a-C:H:Si:O in humid air showed non-

Amontonian behavior (the coefficient of friction had a power law dependence on load).

The friction coefficient decreases from 0.27 at a load of 84 mN to 0.13 at 5000 mN.Thisis expected based on contact mechanics for single asperity contacts, which suggests the coefficient of friction should scale as L−1/3 due to the dependence of real area of contact on load.

In the case of a-C:H:Si:O, the extremely smooth surfaces and soft tribofilms result in such

62 scaling remaining true even for macroscale contacts. However, due to the growth of the contact area caused by tribofilm deformation, the measured scaling exponent is −0.179 instead of −1/3. The power law decrease of friction coefficient with load explains the higher friction coefficients measured here in humid air compared to previous literature, which reported values between 0.04 and 0.08 from experiments performed at higher loads [16, 91, 99].

Wear rates for a-C:H:Si:O were considered in the context of Archard’s wear equation, which predicts that the worn volume should be proportional to the sliding distance times the load. The constant of proportionality is known as the wear coefficient, and is frequently used in the literature to report on and compare materials wear resistance. However, this model does not adequately describe the wear data on a-C:H:Si:O since the wear coefficient measured for DLN 360 films depends strongly on the load. A better approach, motivated by the correlation between friction and wear, is to consider the wear as a function of the work done by the friction force, which gives an approximately linear fit to the data. This same approach works for the NCD Technologies a-C:H:Si:O environmentally dependent friction as well, but not for the DLN 360 samples, likely due to the different energetics of the friction and wear process when sliding on bare steel.

63 CHAPTER 3: Experimental investigation into the thermal stability of a-C:H:Si:O

As discussed in Section 1.5, the low thermal stability of a-C:H is a significant limitation for many technological applications. The thermal degradation is a combination of multiple processes including the direct conversion of sp3 to sp2 carbon, dehydrogenation, and oxidation (in aerobic conditions) [10]. A number of different experimental techniques can be used to probe the bonding and structure in carbon based materials, including Ra- man spectroscopy, X-ray photoelectron spectroscopy (XPS), near edge X-ray absorption fine structure (NEXAFS) spectroscopy, IR spectroscopy, and nuclear magnetic resonance

(NMR) spectroscopy [18, 76, 119, 120]. Different techniques probe different aspects of the structure and bonding; therefore in order to understand the thermal degradation of a-C:H:Si:O in comparison to a-C:H it is necessary to make use of a variety of experimental techniques. For this work Raman spectroscopy, which is a common technique used to probe carbon bonding in DLCs, was used to investigate bulk changes in structure for annealed a-

C:H:Si:O in comparison to a-C:H. Complimentary to this work, Mangolini et al. [119][121] used highly surface sensitive XPS and NEXAFS spectroscopy to perform quantitative measurements of the atomic composition and sp3 fraction. Finally, heated AFM probes, which allow for rapid heating of a nanoscale contact area were used to demonstrate the significant difference in thermal stability between a-C:H and a-C:H:Si:O even for short and localized heating.

3.1. Raman spectra of carbon

Raman scattering, which is the inelastic scattering of photons by fundamental vibrational excitations in molecules and solids, is a standard non-destructive technique for the characterization of carbon-based materials due to its sensitivity to probe sp2 and sp3- bonding of the carbon atoms [18]. In Raman spectroscopy, a laser is focused on the sample surface. The majority of the light is elastically scattered, with no shift in energy. This is called Rayleigh scattering [122]. A small fraction of the light, however, is inelastically scattered with an energy shift related to vibrational excitations in the sample, giving rise

64 to the Raman spectrum. The peaks due to scattered photons which have lost energy are referred to as Stokes peaks, while peaks due to scattered photons which have gained energy are referred to as anti-Stokes peaks [83, 122].

Visible Raman spectra (excitation with visible light) for DLC are dominated by two broad bands called the G-peak and the D-peak appearing at about 1550 cm−1 and 1360 cm−1

respectively [18, 113, 122]. The names D and G originally stood for diamond and graphite,

however, the Raman cross-section of the sp2 phase is much higher (50–250 times) than that

of the sp3 phase and in actuality both peaks are due to sp2 carbon [18, 122]. The G peak

represents in-plane displacements of the C-C sp2-bonds, while the D peak represents the

breathing mode of six-member sp2 carbon rings [18, 113, 122, 123]. In monocrystalline

graphite only the G peak is present, and the vibrations giving rise to the D peak are

forbidden. The D peak vibrations require disorder, such as the presence of grain boundaries

in graphite [113, 122, 124]. From this the name G is often said to stand for “graphite”, and

D for “disorder” [18, 122]. However, it is important to remember that while the D mode

vibrations are only due to sp2 carbon in six-member rings (with signal observed only if there

are defects in the structures), the G mode vibrations arise from any form of sp2 bonding,

not just graphite [18, 113]. Figure 3.1 shows characteristic Raman spectra for several forms

of carbon.

For DLC, the D and G peaks are very broad and produce a single convoluted peak. The

shape of this peak depends on a number of factors including the clustering of the sp2 carbon,

the bond disorder, the presence of sp2 rings or chains, and the sp2/sp3 ratio [113, 124, 125].

In order to interpret the Raman spectra for DLC, the single peak is generally fit with

two components, corresponding to the D and G peaks. Often the spectra is fit with two

Gaussian peaks, however a common alternative is Breit-Wigner-Fano (BWF) function for

the G peak and a Lorentzian for the D peak [113, 124, 125]. Most commonly, the spectra

are interpreted in terms of two parameters derived from these fits: the position of theG

peak maximum, and the ratio of the intensity of the D peak to the G peak (I(D)/I(G)).

For the intensity ratio, some authors use the ratio of peak areas, while others use the ratio

65 Figure 3.1: Comparison of typical Raman spectra for several forms of carbon. The spectra are dominated by the G and D peaks, at about 1550 cm−1 and 1360 cm−1. The G peak represents the stretching vibration of all sp2-bonds, while the D peak represents the breathing mode of sp2 carbon rings [18, 113] In monocrystalline graphite only the G peak is present, and the breathing mode vibrations are not active. In microcrystalline-graphite the D peak breathing mode vibrations become active due to the presence of grain boundaries. In amorphous carbon a single peak is seen, which is a convolution of the G and D peaks. As can be seen from the comparison of Raman spectra for a-C, a-C:H, and ta-C, the shape of this peak is dependent on the sp2/sp3 ratio, hydrogen content, and clustering of the sp2 phase. Figure reproduced from Ref [18]. of peak heights [7].

Based on experimental observations, Ferrari and Robertson [113] proposed a model describing the change in G peak position and I(D)/I(G) ratio as a function of ordering and sp3 fraction. A schematic of this model is shown in Figure 3.2. There are two trajectories shown, referred to as the amorphization and the ordering trajectory. The amorphization trajectory is relevant for as-deposited films where deposition processes such as ion implantation lead to increases in the sp3 fraction, while the ordering trajectory is relevant for changes

in film structure with annealing. For crystalline graphite, there is no D peak, so I(D)/I(G)

is 0. For nanocrystalline graphite, the presence of disorder due to grain boundaries gives

66 Figure 3.2: Schematic illustrating the change in G peak position and I(D)/I(G) ratio as a function of ordering and sp3 fraction from Ferrari and Robertson [113]. For crystalline graphite there is no D peak, so I(D)/I(G) is 0. For nanocrystalline graphite the presence of disorder due to grain boundaries gives rise to the D peak, and the G peak shifts to higher wavenumbers. As the sp3 fraction increases there is a decrease in the I(D)/I(G) ratio, while the G peak position initially declines and then increases. This progression, referred to as the amorphitization trajectory is true for as-deposited films. When annealed DLC films will progressively order, and the sp3 fraction will decrease. This is referred to as the ordering trajectory, and does not follow the same path. For ordering, both the I(D)/I(G) ratio and the G peak position increase with the sp2 fraction. Figure reproduced from Ref [125]. rise to the D peak, and the G peak shifts to higher wavenumbers. For the amorphization trajectory, as the sp3 fraction increases there is a decrease in the I(D)/I(G) ratio, while the

G peak position initially declines and then increases. For ordering, both the I(D)/I(G) ratio and the G peak position increase with the sp2 fraction. The numbers in Figure 3.2 are only meant to illustrate qualitative changes in the Raman spectra. The specific values of the G

67 peak position and the I(D)/I(G) ratio depend on the DLC film being studied and the laser wavelength used in the Raman system.

Often a photoluminescence (PL) background is also observed in the Raman spectra [76,

94]. Frequently, the PL spectrum is quantified in terms of the ratio of the slope ofthePL background under the D and G peaks (S) to the intensity of the G peak (I(G)). For the case of a-C:H, the PL background increases in intensity with increasing passivation of dangling bonds by C-H bonds and clustering of the sp2 carbon [76]. In the case of a-C:H:Si:O, however, there is another possible contribution to the PL background due to SiOx, which exhibits strong photoluminescence at higher oxygen ratios (x ≥ 1.3) [94, 114]. Thus an increase in the PL background for a-C:H:Si:O could also be due to increasing oxidation or segregation of the SiOx phase. The intensity of the Raman spectra should have units of photons s−1 cm−2 [126], but in practice, since the measured intensity is strongly dependent on the spectrometer and the operating conditions it is not comparable between different experiments and is simply reported in arbitrary units. However, for spectra acquired in the same experimental setup relative intensity is meaningful and can convey some useful information. For example, since the Raman spectra are due to vibrations of sp2 carbon, a large increase in peak intensity could indicate a higher fraction of sp2 bonded carbon, provided the same acquisition conditions were used.

In the present work, Raman spectra were acquired using a NTEGRA Spectra confocal

Raman (NT-MDT, Zelenograd, Moscow, Russia) using a 532 nm laser source. The maximum laser power was kept under 300 µW to avoid sample damage, which was confirmed by observing no change in the spectra over time while exposed to the laser. Prior to measurement, the calibration of the Raman system was checked by collecting reference spectra from highly ordered pyrolytic graphite (HOPG) and Si samples. When relevant for comparing intensity the acquisition time, laser power, and focus were kept constant to produce comparable results. Spectra were acquired from a minimum of five points on the sample surface and showed lateral uniformity; representative spectra are shown in the figures.

68 3.2. Raman spectroscopic analysis of annealed a-C:H and a-C:H:Si:O

Samples of HGST a-C:H and DLN 180 a-C:H:Si:O were annealed up to 350 ◦C in aerobic conditions (ambient laboratory air) for 1 hour using a hotplate. Raman spectroscopic measurements were performed on the as received and annealed samples to characterize bulk changes in carbon bonding. Raman spectra acquired from annealed a-C:H and a-

C:H:Si:O samples are shown in Figure 3.3. The spectra were fit using a quadratic baseline and two Gaussian peaks corresponding to the D and G bands. Table 3.1 gives the values of the G peak positions and intensity ratios (I(D)/I(G)) from the fits.

The a-C:H spectra in 3.3a show a shift of the G peak to higher wavenumbers and a significant increase in the I(D)/I(G) ratio occurring when the films were annealed to300 ◦C.

In comparison, for a-C:H:Si:O, the Raman spectrum remains relatively unchanged between the as received sample and the sample annealed at 300 ◦C. There is a slight shift in the position of the G peak and an increase in the I(D)/I(G) ratio. When annealed at 350 ◦C, more noticeable changes begin to appear in the Raman spectrum, with a continuing shift in the G peak and a larger increase in the I(D)/I(G) ratio as seen in Table 3.1, however, still not to the extent as in the case of a-C:H at 300 ◦C.

The changes in the Raman spectra indicate that a-C:H:Si:O is stable between 300 ◦C

and 350 ◦C, and is beginning to show signs of thermal degradation at 350 ◦C, while a-C:H

has already begun to significantly thermally degrade when annealed to300 ◦C, with film

volatilization occurring at 350 ◦C. As discussed in Section 3.1 an increase in the G peak

position and the I(D)/I(G) ratio has been associated with a decrease in sp3 content, however

Table 3.1: G peak position and intensity ratios (I(D)/I(G)) for Raman spectra acquired from annealed a-C:H and a-C:H:Si:O.

Temperature a-C:H a-C:H:Si:O Pos(G) (cm-1) I(D)/I(G) Pos(G) (cm-1) I(D)/I(G) RT 1540 0.92 1524 1.30 300 ◦C 1560 3.09 1536 1.75 350 ◦C - - 1550 2.73

69 2000 5000

350 °C

.) 5000 .) 5000

.u .u

a a

( (

ty ty

i i

s s

n n

te te

In 300 °C In 300 °C

2000 5000 G

RT RT

500 1000 1500 2000 2500 500 1000 1500 2000 2500 -1 -1 Raman Shift (cm ) Raman Shift (cm ) (a) (b)

Figure 3.3: Raman spectra for a-C:H (a), and a-C:H:Si:O (b), showing variation of the D and G peaks with temperature. For a-C:H at 300 ◦C there is a substantial increase in the intensity of the D peak and a shift of the G peak to higher wavenumbers. At 350 ◦C there is complete film removal as indicated by the lack of a characteristic DLC peak in theRaman spectra. For a-C:H:Si:O in (b) the intensity of the D peak is increasing as well, but shows significantly less change than for a-C:H. The G peak shifts similarly to a-C:H, butoccurs at higher temperatures. After annealing to 350 ◦C film volatilization has not occurred for a-C:H:Si:O, in contrast to a-C:H. Shifts in the G peak position and I(D)/I(G) ratio indicate changes to the structure of a-C:H/a-C:H:Si:O [18, 113, 125]. it does not unambiguously define a change in the2 sp /sp3 ratio, since other factors, such as an increase in the clustering of the sp2 phase could also account for such as change [18, 113, 125].

Thus, the Raman spectra provide qualitative evidence for a potential conversion of sp3 to sp2-bonding and clustering of sp2 carbon as mechanisms for the thermal degradation of a-

C:H:Si:O beginning to occur between 300 and 350 ◦C, while significant thermal degradation

70 of a-C:H begins a temperatures below 300 ◦C and complete film volatilization occurs between

300 and 350 ◦C.

It is not possible to accurately quantify the sp3 fraction from Raman spectra at a single wavelength [113, 124]. However, empirical relations between Raman spectra and the sp3 carbon fraction have been obtained based on spectra acquired at multiple wavelengths. For a-C:H with 25–50% hydrogen, Cui et al. [127] reported a linear relationship between the sp3 fraction and the dispersion rate of the G peak (Disp(G)) given by:

sp3 fraction = 2.5 · Disp(G) − 0.07 (3.1) where Disp(G) is defined as:

Pos(G) − Pos(G) -1 Disp(G) = 2 1 [cm / nm] λ2 − λ1

Here Pos(G)1 and Pos(G)2 are the positions of the G peak at wavelengths λ1 and λ2. Although Disp(G) is claimed to be constant, and thus can be computed from any two wavelengths [127], due to experimental noise and errors in fitting the G peak position, it is more accurate to use a range of wavelengths and compute Disp(G) from a linear fit of

Pos(G) vs. λ. Equation 3.1 is purely empirical, but based on experimental measurements it had an accuracy of ±0.06 for a wide range of a-C:H samples [127].

Samples of DLN 360 a-C:H:Si:O were annealed at 250 ◦C and 330 ◦C for 1 hour in laboratory air using hotplates. After annealing Dr. Komlavi Medard Koshigan assisted with access to a multiwavelength Raman spectrometer in the Laboratoire Hubert Curien in

Saint Etienne, France. This system was used to acquire spectra at four different wavelengths,

325 nm, 442 nm, 488 nm, and 633 nm. Representative spectra are shown in Figure 3.4.

To compute the sp3 fraction based on Equation 3.1 the position of the G peak at each wavelength was determined by fitting the spectra with two Gaussian peaks and a linear background. Then the G peak dispersion (Disp(G)) was computed from a linear fit of the

G peak position vs. wavelength.

71 300 300 20°C 325 nm 20°C 442 nm 250°C 250°C 330°C 330°C

.u .u 200 200

(

te te 100 100

0 0 500 1000 1500 2000 500 1000 1500 2000 -1 -1 Raman Shift (cm ) Raman Shift (cm ) 1500 25 20°C 488 nm 20°C 633 nm 250°C 250°C

330°C .) 20 330°C

.u 1000

a 0 15

(

s 10

te 500 n

In 5

0 0 500 1000 1500 2000 500 1000 1500 2000 -1 -1 Raman Shift (cm ) Raman Shift (cm )

Figure 3.4: Multiwavelength Raman spectra from annealed DLN 360 a-C:H:Si:O acquired at four different wavelengths as indicated in the figure. The position of the Gpeakwas determined by fitting the spectra with two Gaussian peaks and a linear background. The peak position was used to compute the sp3 fraction based on Cui et al. [127].

Figure 3.5 shows the computed sp3 fraction vs. temperature determined from multiwavelength Raman. As can be seen, the as-received film is computed to have a 3high sp fraction

(65% sp3 carbon), with minimal change when annealed to 250 ◦C. At 330 ◦C there is a larger decrease in sp3 fraction, however, the film remains primarily3 sp bonded. It is important to remember that there is a good deal of uncertainty in these values. First, Equation 3.1 is an empirical relationship derived for a-C:H with varying hydrogen content. Although the a-C:H:Si:O investigated is primarily a-C:H, it is not clear what effect the inclusion of

72 Si and O has on the sp3 vs. Disp(G) relationship. Second, as mentioned Section 3.1, the

Raman spectra is dominated by contributions from sp2 carbon. For the highly sp3 bonded

DLN 360 a-C:H:Si:O, the overall intensity of the D and G peaks is reduced relative to the background, and the spectra were hard to accurately fit with two peaks.

3.3. Review of XPS and NEXAFS experiments

Due to the multiple contributions to the location and intensity of the D and G peak in the Raman spectra it is difficult to extract quantitative results. In addition, no information can be gained about the silicon and oxygen bonding. To develop a quantitative understanding of how the structure of a-C:H:Si:O changes with annealing, X-ray photoelectron spectroscopy (XPS) and near edge X-ray absorption fine structure (NEXAFS) spectroscopy was performed by Mangolini et al. [119] on HGST a-C:H and DLN 180 a-C:H:Si:O [121].

Although this work is not part of this thesis, it will be discussed in some detail as it provides the basis for comparisons between experimental observations and the MD simulations presented in Chapter5. The XPS spectroscopy was performed using a custom-built XPS system in the Carpick Research Group at the University of Pennsylvania; the details of

65 C 3 60

55 Percent sp 50

45 0 100 200 300 400 Temperature (°C)

Figure 3.5: sp3 fraction vs. temperature for annealed DLN 360 a-C:H:Si:O determined from multiwavelength Raman.

73 this system and operation are provided elsewhere [119, 128]. Heating experiments were performed in the XPS chamber in ultra-high vacuum conditions. The a-C:H:Si:O samples were annealed in steps of 50 ◦C from 150 ◦C to 500 ◦C for 1 hour, after which XPS spectra were acquired. XPS is able to provide a quantitative determination of the C, O, and Si concentration and the bonding configuration of these elements in the near surface region.

The concentration of H cannot be determined by XPS.

Upon annealing at temperatures over 200 ◦C a slight decrease in carbon concentration and an increase in the oxygen concentration was observed, which was attributed to reactions with adsorbed water and residual oxygen in the vacuum chamber. The high resolution spectra of the C1s peak (Figure 3.6) showed a progressive decrease in the peaks assigned to carbon bound to oxygen. It was hypothesized that carbon-oxygen bonds in the unheated sample were due to a contamination layer from air exposure, suggesting a desorption of the contamination layer with annealing. The low percentage of carbon-oxygen bonds in the MD simulations (< 2%) discussed in Section 4.4 support this assumption. At higher temperatures two additional components were observed in the C1s spectra, which were assigned to π-π∗ shake-up satellites for ordered sp2 carbon. This indicates clustering and ordering of the sp2 carbon when a-C:H:Si:O is annealed over 200 ◦C. Figure 3.6 also shows

the fitting of the Si2p spectra with annealing. The fraction of silicon in higher oxidations

states was observed to increase with annealing temperature [121].

One important structural characterization for amorphous carbon films is the fraction

of sp2 and sp3 bonding. This cannot be accurately quantified from the XPS C1s spectra,

however, NEXAFS spectra of the carbon K-edge allow the relative fraction of sp2 carbon

to be calculated from the intensity of the peak due to the C1s→ π∗ transition [121]. The

fraction of sp2 carbon calculated from the NEXAFS spectra is shown in Figure 3.7. The sp2

carbon fraction is observed to increase with annealing. Based on the sp3 fraction computed

from NEXAFS measurements an activation energy for thermally induced rehybridization

was computed based on a model assuming rehybridization from sp3 to sp2 could be treated

as a first order reaction with a Gaussian distribution of energy barriers[67, 119, 121]. The

74 EASseta h aawsftt eemn h itiuino ciaineege for energies of activation distribution the to determine sp of fit conversion was the data The spectra. sp NEXAFS of Fraction 3.7: Figure to to annealing. assigned after assigned desorbs peaks peaks which The layer, the contamination temperature. the in to annealing increase due of and An are (left) function oxygen spectrum to a C1s bound as the carbon showing (right) a-C:H:Si:O spectrum of Si2p spectra XPS the resolution High 3.6: Figure rcino ii ihroiainsae.Fgr rmRf[121]. Ref from Figure states. oxidation higher in Si of fraction sp of clustering

2 Intensity 296 abn h uv itn fteS2 pcr hw nices inthe anincrease shows spectra Si2p the of fitting curve The carbon. 150°C 250°C 350°C 450°C RT 20 Cps 20 Cps 20 Cps 20 Cps 20 Cps 3 Binding Energy(eV) osp to O-C=O 2 292 shake-up π

2 Fraction of sp Carbon C=O - 0.45 0.50 0.55 0.60 0.65 0.70 2 π abna ucino neln eprtr acltdfrom calculated temperature annealing of function a as carbon * C-O 121] [67, insert) (see carbon 288

N(E ) 284

0.0 0.1 0.2 a 100 π 0 - Temperature (°C) 280 π ∗ 2 E 75 hk-pstlie ugssadodrn and ordering and suggests satellites shake-up a 200 (eV) E 4 a

=3.0±1.1 eV Intensity 6 112 150°C 250°C 350°C 450°C RT 10 Cps 10 Cps 10 Cps 10 Cps 10 Cps 300 8 Binding Energy(eV) 108 Si(IV) Si(III) 400 104 100 Si(II) Si(I)+ Si-C 96 details of this model are provided later in Section 5.7. This results in an activation energy of 3.0 ± 1.1 eV for thermally induced rehybridization in a-C:H:Si:O [121]. Following the same methodology, Mangolini et al. [119] computed an activation energy of 2.6 ± 1.2 eV for a-C:H, showing that the addition of Si and O increases the mean activation energy for rehybridization.

3.4. Nanoscale heating with thermal AFM probes

One important difference between the annealing experiments and HAMR applications is that the HAMR heating is over a short timescale and highly localized. One approach that allows an approximation of the HAMR conditions is the use of heated AFM cantilever.

AFM cantilevers can be produced with integrated resistive heaters made from doped silicon, which allows the cantilevers to reach temperatures over 500 ◦C. The temperature of the heater is controlled by varying the applied voltage. An image of a typical heated AFM cantilever is shown in Figure 3.8. These cantilevers can be used to rapidly heat a sample with nanometer lateral resolution. Such cantilevers are produced commercially and were purchased from Anasys Instruments (nano-TA, Anasys Instruments, Santa Barbara, CA,

USA). In a typical AFM set-up, the probe tip velocity would be on the order of 1–100 µm/s,

Figure 3.8: SEM Image of a typical heated AFM cantilever. Current flows through the low resistance cantilever legs and the high resistance tip, resulting in heat generation at the end of the cantilever. The image on the right shows a zoomed view of the sharp probe tip. (Images from Anasys)

76 and the contact area diameter, estimated from contact mechanics models, would be on the order of 10–100 nm. Therefore, the heating time for any individual spot on the sample surface would be on the order of a tenth of a second to milliseconds.

Since the heater temperature cannot be directly measured in the AFM, precise calibration is necessary before use. The resistance of the doped silicon cantilever is known to be a function of temperature [129]. In general, the cantilever resistance will initially increase with temperature due to decreasing carrier mobility caused by increased electron-phonon scattering. This increase in resistance with temperature is well fit by a quadratic. The resistance will the reach a peak, generally between 450 ◦C and 650 ◦C, after which the resistance decreases with increasing temperature. This is a result of the thermal excitation of electrons into the silicon conduction band [129–131]. In a typical AFM set-up, a controller varies the applied heating voltage while measuring the cantilever resistance. If the relationship between resistance and temperature is known, then the temperature can be accurately controlled.

There are several methods reviewed in the literature to determine the cantilevers resistance as a function of temperature [129, 131, 132]. The most accurate calibration technique is using Raman spectroscopy to determine the probe temperature based on shifts in the location and width of the Stokes peak. The peak location as a function of temperature is given by Beechem et al. [131] as:

ω = −A(Tω − T0) + ω0

where ω is the peak position at temperature Tω and ω0 is the peak position at reference

−1 temperature T0. A is a calibration constant given as 0.022 cm /K. Although the peak position is easy to determine, error in this calibration is introduced if the material is stressed, since the peak position is also linearly dependent on stress [131]. The peak width is related to temperature through a quadratic relationship and is independent of stress, provided the stress in the probed volume is uniform [131]. This would make peak width the preferable method for determining temperature, however, in practice accurate fitting of the peak width

77 is difficult and requires long acquisition times. A review of calibration techniques byNelson and King [129] found that for heated AFM cantilevers, the error in determining the peak width was greater than the error introduced by the effect of stress when using the peak location.

In a second method, known as the melting standard calibration, the probe is heated until it is observed to melt a polymer with a known melting temperature. This is repeated for three different polymers and the resulting data is fit with a quadratic. Melting standard calibration is less time consuming, but suffers from inaccuracies due to the reduction in melting temperature resulting from confinement effects [129]. For this work, Raman calibration based on the shift in the position of the Stokes peak was used to check the results of the melting standard calibration. A typical calibration curve for the cantilevers used is shown in Figure 3.9. Since there was no significant difference observed between the calibration methodologies, only the melting standard calibration was used for additional probes due to the decreased time required.

Once the cantilever is calibrated, the temperature of the heater can be controlled. How-

3 . 5 P o l y m e r c a l i b r a t i o n R a m a n p e a k s h i f t

) 3 . 0 Ω k (

2 . 5 e c

n 2 . 0 a t s i 1 . 5 s e

R 1 . 0

0 . 5 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 T e m p e r a t u r e ( o C )

Figure 3.9: Resistance vs. temperature for a typical heated AFM cantilever. The polymer calibration curve is determined by measuring the melting point of three different polymer samples with the heated probe. Raman calibration data points are based on the shift in the Stokes peak position with temperature.

78 ever, it is the temperature at the interface between the tip and the substrate that is of interest, which is generally not the same as the heater temperature. This temperature cannot be calibrated or directly measured. Therefore, heat transfer models must be used to calculate the interface temperature. Heat transfer from the cantilever to the substrate is primarily due to conduction through the air from the cantilever legs, conduction through the solid-solid contact at the probe tip, and conduction through the liquid meniscus that may be present [132] This heat transfer was modeled using a thermal circuit analysis developed by Nelson and King [133] and shown in Figure 3.10. For a typical heated AFM cantilever, the thermal resistance of the air gap is orders of magnitude smaller than the thermal resistance through the tip. Therefore, the majority of the heat is transferred to the substrate through the air. However, due to the nanometer sized contact area, the heat flux at the tip sample interface is large, resulting in the interface temperature rise (θInterface) being much higher than the temperature rise due to air conduction (θSurface). At the interface, the heat transfer is due to conduction through the solid contact and the liquid meniscus. The solid-solid contact generally has higher thermal conductance than the liquid film, resulting in the dominant heat transfer being through the solid [132].

Heat Generation

θHeater

RTip RGap

RContact θSurface θInterface R R Sub,2 Thin Film Sub θ=0 θ=0 Substrate

Figure 3.10: Thermal circuit model used by Nelson and King to calculate the interface temperature rise resulting from a heated AFM probe [133].

79 Based on this model, the interface temperature rise can be calculated if the thermal resistances RTip, RContact, and RSub are known. These can be computed from the thermal conductances of the tip and substrate, the cantilever/tip geometry, and the applied load.

A detailed description of these calculations is provided by Nelson and King [133]. Briefly, the result of this analytic model is that the non-dimensional interface temperature rise

∗ (θInterface), defined as, ∗ θInterface TInterface − T0 θInterface = = θH THeater − T0 is a function of six dimensionless parameters, as shown in equation 3.2,

∗ ksub Rbktip,d kgap D a L θInterface = f , , , , , (3.2) ktip,d d ktip,d d d d

where ksub is the substrate thermal conductivity (W/mK), ktip,d is the thermal conductivity

2 at the end of the conical tip (W/mK), Rb is the thermal contact impedance (m K/W), d is the cross sectional diameter at the end of the cone (m), kgap is the thermal conductivity of the ambient environment (W/mK), D is the tip base diameter (m), a is the contact diameter (m), and L is the tip height (m).

From the variables in equation 3.2, d, D, and L are based on the cantilever geometry. Rb, the contact impedance, is based on literature values. It is difficult to estimate accurately, but tends to fall between 10−7 and 10−9 m2 K/W, and does not vary greatly depending on contacting materials [133]. The contact diameter (a) can be found from contact mechanics.

In this case the Hertz contact mechanics model was used to compute the contact diameter and maximum stress in the contact [102, 109].

In the model developed by Nelson and King [133], the calculation of the contact diameter a assumes the tip geometry is a truncated cone with a hemispherical cap (i.e. it is assumed that the tip diameter of curvature dsphere is always equal to d, the cross sectional diameter at the end of the cone). This is a reasonable approximation for sharp tips, however, gives unrealistic geometry for blunt tips. As an AFM tip wears, the geometry should approach that of a flat punch (i.e. as a tip wears, dsphere → ∞, but the cross sectional diameter of

80 the cone is finite); this is not true for the assumed tip geometry, which makes themodel unsuitable for blunt probes. Here the model was expanded to allow for blunt probes by relaxing these geometric assumptions (dsphere ̸= d). The thermal conductivity of the substrate is of particular importance in determining

the interface temperature. For DLCs the thermal conductivity generally depends on the sp3

fraction and hydrogen content. Polymeric films, with low3 sp fraction and high hydrogen

content have thermal conductivities on the order of 0.2 W/mK [134, 135], while highly sp3

bonded ta-C has thermal conductivities ranging from 2–10 W/mK [18, 125, 134, 136]. a-

C:H falls in the middle of these values, with thermal conductivities typically in the range of

0.7–1.3 W/mK [125, 134, 135]. The thermal conductivity of amorphous SiO2 films has also been measured, and is similar to a-C:H, with values ranging from 0.8–1.4 W/mK [136–138].

The thermal conductivity of a-C:H:Si:O is more uncertain. Only one paper could be

found that measured the thermal conductivity of a-C:H:Si:O, which reported a value of

76 ± 10 W/mK [139]. This value appears far too high, since it is two orders of magnitude

larger than reported values for a-C:H and amorphous SiO2 films. A possible reason for this discrepancy could be incorrect experimental techniques applied in [139]. The high thermal

conductivity of a-C:H:Si:O was measured using scanning thermal microscopy, a technique

that uses the heat flux from a heated AFM tip to compute the thermal conductivity based

on a calibration curve. However, this method has been shown to be unsuitable for samples

with high thermal conductivity (> 10 W/mK), since the heat flux becomes independent of

samples thermal conductivity [140–142]. In the case of the a-C:H:Si:O measurements in

[139], the calibration curve was based on GaAs, Ge, Si and Al; all of which are materials

with thermal conductivity > 10 W/mK. In addition, these materials often have oxide layers with far lower thermal conductivity than the bulk values [142], which could be influencing the calibration. For these reasons, in this work the thermal conductivity of a-C:H:Si:O was assumed to be in the range of 0.7–1.4 W/mK, which is consistent with the thermal

conductivity values for a-C:H and a-SiO2.

The thermal conductivity of the tip (ktip,d) is different from the bulk thermal conduc-

81 tivity of silicon due to restriction in heat flow due to the nano-scale cross sectional area, while the thermal conductivity of the substrate (ksub) must be modified to account for the reduction in substrate thermal resistance for thin films on high thermal conductivity substrates. The equations used for these corrections are described in the paper by Nelson and

King [133].

∗ Once the parameters in equation 3.2 are known, θInterface can be found by solving for the temperature distribution through the conical tip. MATLAB was used to analyze the

∗ ∗ effect of error in the input parameters on θInterface. It was found that the error in θInterface is dominated by error in Rb and ksub. Error in all other input parameters has negligible effect on the accuracy of the model. This analytic model was compared to finite difference simulations of heat transfer through the tip, and shown to agree with the results [133].

∗ For a silicon tip on a-C:H the normalized interface temperature rise (θInterface) is between ∗ 0.1 and 0.4 depending on the tip radius and load. The dependence of θInterface on stress for a range of tip radii was computed using MATLAB and is shown in Figure 3.11. Larger tips and higher stress results in increased heating due to a larger contact area between the tip and the substrate.

Heated probes were used to scan a 5 × 5 µm2 area on both materials. Based on the thermal circuit model and the dimensions of the cantilevers used the normalized interface temperature rise was calculated to be 0.22. The cantilevers used had a maximum control- lable heater temperature of 450 ◦C, resulting in a 115 ◦C interface temperature. Figure 3.12 shows the before and after survey scans (taken at room temperature and low load) for both materials. The scans were performed at 250 nN applied load and a 1 Hz scan speed.

Volatilization of the a-C:H film was observed after scanning. In comparison, the a-C:H:Si:O sample was stable under the same conditions. To confirm the a-C:H film removal was due to thermal degradation and not stress alone, control scans were performed at the same load and no film removal was observed. However, this does not rule out the possibility thatthe amount of film removal was increased due to stress. Although the depth of film volatilization is only 1.5 nm, in applications such as HAMR disk drives the total film thickness would

82 0.4 ) t n i * θ (

m 0.3 e T

e c a f r e t 0.2 n I

e Tip Radius z i l

a 160 nm

m 0.1 r 80 nm o

N 40 nm 20 nm 0.0 0.0 0.5 1.0 1.5 2.0 Stress (GPa)

∗ Figure 3.11: Dependence of θInterface on stress for a range of tip radii. The stress is computed from Hertz contact mechanics based on the applied load. Larger tips achieve greater interface temperature rise. be on the order of 2–4 nm [8], where 1.5 nm of film volatilization would be significant.

The amount of film volatilization was measured as a function of interface temperature.

The results are shown in Figure 3.13. As can be seen, a-C:H volatilization occurs even for the lowest interface temperature increases, compared to no change for a-C:H:Si:O, and the amount of volatilization increases with temperature. Repeated scans over the same region were also considered and showed a progressive increase in film removal for a-C:H, and no measurable change for a-C:H:Si:O.

3.5. Conclusions

A variety of experimental techniques were used to probe the thermal stability of a-

C:H:Si:O in comparison to technologically relevant a-C:H used in the current generation of hard disk drives, and similar to that used in many other applications. Long duration

(1 hour) annealing in laboratory air up to 350 ◦C showed complete film volatilization for a-C:H compared to moderate changes at 350 ◦C for a-C:H:Si:O as measured by Raman spectroscopy. Characterization of the Raman spectra based on fitting the D and G peaks

83 As deposited After heating 3.5 nm (a) (b) Before heated scan After heated scan a-C:H

2.0µm 2.0µm 0 nm

4 nm (c) (d) a-C:H:Si:O

2.0µm 2.0µm 0 nm Figure 3.12: AFM scans of a-C:H (top) and a-C:H:Si:O (bottom) before and after scanning with heated cantilevers operating at 450 ◦C (115 ◦C interface temperature). Film volatilization is observed for a-C:H resulting in a depression after scanning. No change is observed for a-C:H:Si:O under the same conditions. suggests thermal degradation through rehybridization from sp3 to sp2 carbon and clustering of sp2 carbon.

This is in agreement with quantitative XPS and NEXAFS spectroscopy performed by

Mangolini et al. [121]. XPS measurements showed clustering and ordering of the sp2 carbon and an increase in the silicon oxidation state. NEXAFS spectra were used to quantify the sp3 carbon fraction and observed an increase in sp2 carbon with annealing. Modeling of the

84 1.5 a-C:H a-C:H:Si:O

1.0

Depth (nm) 0.5

0.0 20 40 60 80 100 120 Approximate Interface Temperature (°C)

Figure 3.13: Depth of film removal as a function of interface temperature for a-C:H and a-C:H:Si:O after a single scan with a heated AFM probe. activation energy for thermally induced rehybridization based on a Gaussian distribution of energy barriers resulted in an activation energy of 3.0 ± 1.1 eV for a-C:H:Si:O [121] compared to 2.6 ± 1.2 eV for a-C:H [119], showing that the addition of Si and O increases the mean activation energy for rehybridization.

Heated AFM probes can be used to investigate thermal stability at time and length scales relevant to applications such as HARM disk drives. Although the heat transfer between the tip and sample can only result in limited heating, technologically significant amounts of film volatilization were observed for a-C:H, even at low temperatures and increased with heater temperature (although this is also influenced by the applied stress). Under the same conditions no changes were observed for a-C:H:Si:O.

This combination of experimental characterization techniques clearly shows the addition of even small amounts of silicon and oxygen can substantially increase the thermal stability of a-C:H, overcoming a major technological limitation for these films. However, the atomistic mechanisms responsible for this enhanced thermal stability is not clear. In order to understand the mechanisms responsible for the effect of silicon and oxygen on the thermal stability of a-C:H:Si:O molecular dynamics simulations were employed. The simulation

85 techniques are introduced in Chapter4 and the results are presented in Chapter5.

86 CHAPTER 4: Molecular dynamics simulations of a-C:H:Si:O using the ReaxFF potential

Molecular dynamics (MD) simulations are a valuable tool to gain insight into the atomic interactions that influence the structure/property relations in a-C:H:Si:O. By gaining the ability to see every atom, the results of a simulation can help explain and understand experimental results. Additionally, in the context of HAMR disk drives, the temperature pulses would be on the order of picoseconds, a timescale that is experimentally challenging, but accessible in simulations. There are several approaches to modeling DLC including density functional theory (DFT) based methods and empirical potentials. DFT simulations have shown good ability to reproduce experimental results. However, due to computational cost, these methods are limited to only a few hundred atoms and short timescales [143, 144].

This is problematic for the study of amorphous systems, where a large degree of structural variation exists, which may not be accurately captured in smaller simulation domains. In addition, simulations of tribological properties typically require longer timescales, which are inaccessible in DFT methods. Therefore a number of empirical potentials have often been used to model DLC, including Tersoff, REBO, ReaxFF, and EDIP [143–146]. How- ever, some of these potentials have not been parameterized for amorphous carbon, and in particular, few simulations of a-C:H:Si:O using these potentials exist. Thus it is important to investigate their ability to reproduce the structure of DLC. Several papers [143, 145, 146] have studied the ability of these potentials to model a-C, typically be looking at how well the potentials can reproduce the experimentally measured relationship between sp3 fraction and density [18].

A comparison between different potentials for how the3 sp fractions depend on density, and experimental results, are shown in Figure 4.1. In general, wide variation exists in the simulated structures depending on the potential used. DFT-based methods most accurately reproduce experimental trends (sp3 fraction vs. density), although they still slightly underpredict the sp3 fraction [143, 147]. In terms of empirical potentials however, the experimental trends are often not well reproduced. Several reviews of the capability of different

87 potentials to model the structure of a-C can be found in the literature [143, 145, 146], which compared the Tersoff, REBO, ReaxFF, EDIP, LCBOP, and COMB potentials. Overall it was found that the EDIP potential (which was parameterized specifically for describing amorphous carbon) provided the best match to experimental data [143, 145], followed by ReaxFF. The Tersoff potential was only suitable for modeling a-C at high density> (

2.6 g/cm3), due to significant overestimation of the sp3 fraction at lower densities, while the REBO potential was only suitable for low density a-C (< 2.4 g/cm3), due to significant underestimation of the sp3 fraction. EDIP and ReaxFF produce reasonable results over a range of densities, with ReaxFF being better suited to lower densities (< 2.6 g/cm3) due to increasing underestimation of the sp3 fraction with increasing density [146]. The

LCBOP and COMB potentials perform poorly, and do not produce accurate structures at any density [145].

The most common potential used in the literature is REBO, with several studies investigating the tribology and thermal stability of DLC [58, 59, 61, 74, 153–155]. For example,

Pastewka et al. [59] modeled the tribology of self-mated a-C:H at different hydrogen contents and observed the influence of surface passivation and rehybridization3 fromsp to sp2 carbon in reducing friction. Quan et al. [74] modeled the thermal stability of a-C:H as a function of H content and saw a decrease in thermal stability with increasing hydrogenation, in agreement with experiments.

As mentioned previously however, the REBO and Tersoff potentials have been shown to incorrectly predict the structure of a-C. This has been attributed to the use of short range cutoff functions in these potentials that limit the range of interactions between atoms [144, 147]. This results in incorrect descriptions of bond breaking [147] and atoms becoming trapped in unphysical positions between the expected first and second neighbor locations (see Figure 4.2)[144]. Pastewka et al. [147] has shown that by replacing the fixed cutoff distance with an environmentally-dependent screening function (REBO+S), boththe incorrect description of bond breaking and the underestimation of the sp3 fraction in a-C are resolved. However, in order to model a-C:H:Si:O, the potential used must be parametrized

88 100 Experimental REBO REBO+S 80 Tersoff ReaxFF

) EDIP

( 60 DFT

3 40

0 2.0 2.5 3.0 3.5 3 Density (g/cm )

Figure 4.1: Comparison of sp3 content obtained with different simulation techniques to experimental values for a-C. The color of the symbols indicates the simulation method (as shown in the legend), while the shape of the symbol represents the reference for the data. DFT simulations most accurately reproduce the experimental trends. The Tersoff potential consistently overpredicts the sp3 fraction, while the REBO potential significantly underpredict the sp3 fraction. The EDIP potential provides the best match to experimental data, followed by ReaxFF. A modified version of the REBO potential with the cutoffs replaced by environmentally dependent screening functions (REBO+S) performs similarly to ReaxFF. Experimental data (gray symbols) from Refs ◦ [148], [149], △ [150] and simulations (colored symbols) from Refs ◦ [147], [146], △ [151, 152], ▽ [145], [143]. for C, H, Si, and O, which significantly limits the choice of potential. For this work, the

ReaxFF potential was used, with a Si/C/H/O force field developed by Newsome et al.

[156], which is the only potential including all of these elements. The fitting parameters are derived from experimental data and quantum mechanical simulations. The bond-order for- mulation used by ReaxFF to determine interactions based on local environment is similar to the screening functions proposed by Pastewka et al. [147] to improve the REBO potential, and it has demonstrated the ability to model the structure of amorphous carbon [151, 152].

89 Figure 4.2: Radial distribution function (RDF) for a-C simulated with different potentials employing short range cutoffs compared to experimental neutron diffraction data. TheRDF is a histogram representing the average number of atoms at a given separation distance. The RDF for crystalline materials shows sharp peaks at the first, second, third, and so on nearest neighbor locations. For amorphous materials these peaks are broadened due to disorder. For DLC the first neighbor peak is centered near1.5 A,˚ corresponding to the length of C-C σ bonds, while the second neighbor peaks is near 2.5 A.˚ Unphysical peaks between the first and second neighbor locations are present corresponding to thecutoff distance employed in each method [144].

4.1. Description of the ReaxFF potential

ReaxFF makes use of bond order-dependent energy terms to describe reactive processes and allows for bond breaking and formation during simulations [157–159]. The potential was developed initially for hydrocarbons to allow for the study of reactions in systems too large for quantum chemical methods [157]. The framework for the potential is designed to be very general and includes energy terms corresponding to covalent terms, coulomb interactions, and non-bonding forces. A fundamental idea behind the potential is bond order. Instead of using fixed connectivity, the bond order between each pair of atomsis updated continuously. Then, all of the energy terms describing covalent interactions are expressed in terms of bond order, so that these contributions to the total energy can be introduced and removed smoothly with bond formation and dissociation [157, 160]. It is assumed that the bond order between a pair of atoms can be computed based only on the distance between them, as given in equation 4.1, where the terms represent the σ, π, and second π bonds. Here rij is the distance between a pair of atoms, and the pbo and r0 terms

90 are fitting parameters.

pbo,2 pbo,4 pbo,6 ′ rij rij rij BOij = exp pbo,1 · σ + exp pbo,3 · π + exp pbo,5 · ππ (4.1) r0 r0 r0

The potential has separate parameterizations of equation 4.1 for each possible type atomic pair in the system. The parameters are determined by fitting to DFT calculations on a large

′ set of training molecules [157, 160]. To correct for overcoordination BOij is multiplied by a correction factor, the functional form of which is given in the original paper on ReaxFF by van Duin [157].

ReaxFF then expresses the potential energy of the system as a sum, given by,

Esystem(rij, rijk, rijkl, qi,BOij) = Ebond + Eover + Eunder + Elp + Eval

+ Epen + Etor + Econj + EHbond (4.2)

+ EvdWaals + ECoulomb where the terms are a function of two, three, and four body interactions, as well as the charge on an atom, and the bond order. Ebond is the bond energy, Eover and Eunder are energy penalties for over- or under-coordination, Elp is the energy associated with lone pairs,

Eval is valence angles, Epen is a penalty for atoms with two double bonds, Etor is torsion angles, Econj is conjugated bonds, EHbond is the energy in hydrogen bonding, and EvdWaals and ECoulomb are the energy of the non-bonded interactions. The functional forms of these terms are given by van Duin [157]. The potential is very flexible, but the trade off is that the method is approximately 50–200 times slower than traditional MD. This is due to the complex functional form of the potential needed to model bond breaking and formation, as well as the charge equilibration that must take place at every time step. Also, the maximum time-step is between 0.1 and 0.5 fs [161, 162]. Given these considerations ReaxFF could be thought of as falling in between quantum mechanical simulations and conventional MD. It scales better than DFT calculations, but due to the complex potential and small time-step is limited in size compared to conventional MD, and is difficult to run simulations reaching

91 nanoseconds. For example, based on benchmarking simulations, ReaxFF is approximately

170 times slower than the REBO potential, but between 105–106 times faster than DFT calculations, depending on the system size [163]. In addition, the computational cost of

ReaxFF scales approximately linearly with the number of atoms, unlike DFT, which scales

2.5 roughly as Ne , where Ne is the number of electrons [163]. There are few papers that have made use of ReaxFF to model amorphous carbon [144–

146, 151, 152, 164], and in general these have shown good ability to reproduce experimental results. For example, ReaxFF simulations of the tribology of ta-C showed high friction in vacuum due to dangling bonds at the interface. However, when hydrogen or water was introduced between the surfaces it was found to dissociate and passivate these dangling bonds with -H or -OH groups, reproducing the experimentally observed tribology of ta-

C[151, 152]. None of these simulations included elements other than carbon in the film, however, it is suggested by Newsome et al. [156] that the Si/C/H/O force field used here is transferable to systems such as silicon oxides, diamond, and graphite.

4.2. Formation of a-C:H:Si:O through liquid quenching

In general, there are two commonly used methods to generate DLC in MD simulations, liquid quenching and simulated deposition [144]. In the liquid quenching method, an initial crystalline lattice is heated until it melts, typically between 5000–8000 K. The system is equilibrated at this temperature for a few picoseconds to randomize the structure. Finally the liquid is rapidly quenched to the desired temperature. It has been claimed that the precise shape and procedure used for the quenching does not substantially affect the resulting amorphous structure [144]. However, if the quench is too fast, then the resulting structures can contain a very high number of energetically unfavorable bonding configurations. Thus, some consideration should be given to the cooling rate chosen, and what effect it has on the output.

In the literature, a wide range of quenching times have been used, from 0.5 to 100 ps, with starting temperatures ranging from 5000–8000 K. This corresponds to quenching rates

92 between 60 and 10 000 K/ps [143, 145, 146, 154, 165]. For the work in this thesis, quenching rates between 300 and 6000 K/ps were used. This is consistent with multiple simulation- based studies of the structural properties of amorphous carbon. For example, Ref. [145] used a liquid quenching procedure with a quench rate of 6000 K/ps for multiple potentials, including ReaxFF. Ref. [146] used a liquid quenching procedure with a quench rate of 3000 K/ps and concludes that the ReaxFF potential produces realistic structures for densities less than 2.6 g/cm3. Refs. [143] and [165] used a liquid quenching procedure with a quench rate of 10 000 K/ps for both empirical methods and density functional theory simulations and show good agreement between the resulting structures and experimental data. Finally,

Zhang et al. [152] considered the effect of quenching rate on the ta-C structure predicted by ReaxFF and found no difference in3 sp fraction between simulations using a quench rate of 100 K/ps vs. 1400 K/ps. One reason for the apparent lack of effect of quenching time is the variability introduced by randomization of initial positions. It has been shown that a-C structures simulated with identical quenching procedures can have significant variation in the resulting sp3 fraction when the initial configuration is randomized [143, 146]. This variation is larger than the effect of even large changes in quenching rate.

It should be considered whether the success of liquid quenching in producing accurate

DLC structures is due to some meaningful physics or coincidental. As discussed in Sec- tion 1.2, it has been proposed that the impact of energetic ions during film deposition can produce picosecond thermal spikes which are important for determining the film’s structure [28, 31]. According to Marks et al. [166], these thermal spikes can provide a physical basis for liquid quenching and explain why it should be expected to produce realistic DLC samples.

The alternative method used to generate DLC films in simulations is by simulating the deposition procedure itself [144, 167]. This method has the advantage of following the dynamics of the film growth. However, this method is computationally slow compared to liquid quenching, and thus undesirable when the only concern is the resulting structure.

Since the goal of the simulations here is to study the properties of a-C:H:Si:O and not the

93 formation process, it is advantageous to use the liquid quenching procedure if possible, as long as it produces structures representative of the real material, due to its simplicity and shorter computation time than a simulated deposition.

For this work the simulations were carried out using the molecular dynamics program

LAMMPS (Sandia National Laboratories, http://lammps.sandia.gov) [168, 169]. A description of the typical liquid quenching procedure used is as follows: the initial simulation cell consisted of 2744 atoms selected to match the experimental atomic composition of the DLN

180 samples ([C] = 57%; [H] = 34%; [Si] = 6%; [O] = 3%, as given in Table 2.2) at a density of 2.3 g/cm3. Periodic boundary conditions were used in X, Y, and Z directions, and the melting and quenching were carried out in the NVE ensemble. The time step for all simulations was 0.1 fs. The initial temperature was set to 6000 K, which was observed to be sufficient to melt the initial structure, and was maintained for 3–5 ps. The temperature was then reduced to 300 K over 1–20 ps by either velocity rescaling or a Berendsen thermostat, and the resulting structure was equilibrated at 300 K for 5 ps. The simulated a-C:H:Si:O was then further relaxed in an NPT ensemble with a Nose-Hoover thermostat and barostat to 300 K and 0–1.5 GPa external pressure. The damping constant for the thermostat was set to 100 time steps (10 fs) and 1000 time steps (100 fs) for the barostat. This reduced the density to 1.9 g/cm3, which is within 7% of the experimental value. Finally, the periodic boundary conditions in ±Y were removed to form free surfaces, which were relaxed by running in the NPT ensemble for 5 ps. In addition, a control sample of a-C:H was produced following the same procedure to match the composition and density of the a-C:H samples used by HGST. The final density for the a-C:H sample simulated. was2 3 g/cm3. Values extracted from the simulation results are averages over 5 ps, and the simulations were repli- cated three times with randomized initial configurations to confirm reproducibility ofthe results. For analysis and visualization purposes, a pair of atoms was considered bonded if the bond order between them as calculated by the ReaxFF potential was greater than 0.3, which is a standard value used in other ReaxFF simulations with similar parameters [170].

The choice of bond order cutoff was also validated by comparing to a distance based cutoff

94 for bonding and showed good agreement.

4.3. Validation of the ReaxFF potential and simulation procedure

Although literature results have shown a reasonable ability for ReaxFF to model amorphous carbon, the results depend on the specific parameterization used, and no papers looked at DLC containing Si, or O. Therefore it is important to validate that both the potential and liquid quenching procedure can accurately model a-C:H, and a-C:H:Si:O.

Figure 4.3 shows a comparison between a-C, a-C:H, and a-C:H:Si:O samples produced by

ReaxFF using the Si/C/H/O force field from Newsome et al.[156]. The samples were quenched from 6000 K over 20 ps and relaxed to 1 GPa compressive stress, with the excep- tion of the 3.2 g/cm3 a-C, which required 8000 K to melt due to the higher density. As can be seen from the figure, ReaxFF simulations using the Si/C/H/O force field produce reasonable matches to experimental trends. For a-C there is a slight underestimation of the sp3 fraction vs. density relationship, as is typically seen in other simulations. For a-C:H

(35% H), there is excellent agreement between the simulations and experimental data. Two sets of a-C:H:Si:O simulations, A and B, are also shown in Figure 4.3. For a-C:H:Si:O A the composition was matched to the experimental DLN 180 samples ([C] = 57%; [H] =

34%; [Si] = 6%; [O] = 3%), while for a-C:H:Si:O B the composition was matched to the experimental NCD Tech samples ([C] = 28%; [H] = 47%; [Si] = 10%; [O] = 15%, see Ta- ble 2.2). For the simulated DLN 180 composition the sp3-C fraction was similar to a-C:H,

while the NCD Tech composition had a noticeably higher sp3-C fraction than other samples

at similar densities.

To check the effect of quenching rate, for two different initial configurations, asampleof

a-C:H:Si:O was produced using quenching times of 1, 5 and 10 ps. The results are shown in

Figure 4.4. It is observed that the final 3sp carbon fraction does not change significantly with

increased quenching time. In addition, the average potential energy per atom shows only a

slight decrease as the quenching time is increased, suggesting that any highly energetically

unfavorable structures are not present in the 1 ps quench rate structures. The changes in sp3

95 100 a-C a-C:H a-C:H:Si:O A 80 a-C:H:Si:O B a-C (Exp.)

) a-C:H (Exp.)

( 60

3 40

0 1.5 2.0 2.5 3.0 3.5 3 Density (g/cm )

Figure 4.3: Comparison between DLC structures produced with the ReaxFF Si/C/H/O force field and experimental data. ReaxFF slightly underestimates3 thesp fraction at a given density for a-C, but generally reproduces the experimental trends. The addition of H (30-40% in experiments vs. 35% in simulations) decreases the density while increasing the sp3 fraction, and the simulated a-C:H closely matches the experimental measurements. Two different a-C:H:Si:O samples are also shown. In (A) the composition is 33.5% H,6.4% Si, and 2.5% O, matching the DLN 180 samples, while in (B) the composition is 47% H, 10% Si, and 15% O, matching the NCD Tech samples. Experimental data for a-C is from Refs ◦ [148], [149], and △ [150], and a-C:H data is from Ref [171]. fraction and potential energy shown in Figure 4.4 are not statistically significant since they are within the range of variation seen when replicating the simulations with randomized initializations.

4.4. Atomic structure of a-C:H:Si:O

A typical a-C:H:Si:O structure ([H] = 34%; [Si] = 6%; [O] = 3%) resulting from the liquid quenching procedure is shown in Figure 4.5, and the frequency of bonds between different elements is shown in Table 4.1. The structure produced is amorphous, with no long-range order. The a-C:H and Si:O are fully intermixed with the Si and O well distributed throughout the a-C:H. Hydrogen is primarily bonded to carbon, forming a dense

96 40 -5.0 Sample 1 Sample 1 Sample 2 Sample 2 35 -5.5

30 -6.0 Fraction (%) 3

sp 25 -6.5 Average PE (eV/atom)

20 -7.0 1 5 10 1 5 10 Quench Time (ps) Quench Time (ps) (a) (b)

Figure 4.4: Effect of quenching time on resulting a-C:H:Si:O structures. (a) sp3 fraction vs. quenching time for two different initial configurations. The change insp3 fraction with quenching time is small compared to the effect of the randomization of the initial configuration. (b) Average potential energy (eV) per atom vs. quenching time. The change in energy with increased quenching time is minor. a-C:H network, and bonds to both sp3 and sp2 carbon (Figure 4.5b). Silicon is dispersed throughout the a-C:H matrix with significant C-Si bonding. Oxygen preferentially bonds to silicon, (50 ± 4% of oxygen bonds are to silicon) forming small local clusters of a-Si:O

(Figure 4.5c).

Figure 4.6 shows the distribution of bond lengths and angles for sp3-carbon in the simulated structure as formed (before annealing) based on the average of three replications.

The average bond lengths observed in the simulation agree with experimentally reported

Table 4.1: Bonds between elements in simulated a-C:H:Si:O before annealing.

Element (X) # of bonds Percent of bonds to each element (%) X-C X-H X-Si X-O Carbon 3192 ± 26 56.7 ± 1.5 26.7 ± 1.4 15.3 ± 0.4 1.3 ± 0.2 Hydrogen 933 ± 34 91.5 ± 0.8 0.3 ± 0.3 5.4 ± 0.3 2.8 ± 0.6 Silicon 626 ± 8 77.8 ± 2.0 8.0 ± 0.7 3.2 ± 1.1 11.0 ± 1.7 Oxygen 136 ± 24 30.5 ± 3.1 19.0 ± 0.5 50.4 ± 3.6 0.0 ± 0.0

97 (a) (b) (c)

Figure 4.5: (a) Simulated structure of a-C:H:Si:O. The liquid quenching procedure results in a fully amorphous simulation cell. The silicon and oxygen are intermixed in the a- C:H network. (b) Snapshot of a-C:H:Si:O showing only the carbon and hydrogen bonded to carbon. The structure is composed of a dense a-C:H network. The majority of the hydrogen (91.5 ± 1.0%) is bonded to carbon. (c) Snapshot of a-C:H:Si:O showing the silicon and oxygen. The silicon is dispersed throughout the a-C:H matrix. Oxygen typically bonds to silicon (50.4 ± 4.0% of oxygen bonds are to silicon), forming small regions of amorphous silica. There is a small amount of hydrogen bonded to silicon, which is also shown in (c). values [172], and the mean sp3-carbon bond angle of 105.9◦ is close to the expected value of 109.5◦ for perfectly ordered tetrahedral bonding. Here, the sp3-carbon-oxygen bonds are

not shown, since they represent less than 2% of all carbon bonds. The carbon hybridization

is 14.7 ± 0.6% sp1, 54.7 ± 0.6% sp2 and 30.6 ± 1.3% sp3. There is a strong correlation

between Si and sp3 carbon. Of the carbon not bonded to any silicon, 28% is sp3 hybridized, while 55% of the carbon bonded to at least one silicon is sp3 hybridized. The relatively high fraction of sp1 hybridized carbon is typical of MD simulations of amorphous carbon, particularly at lower densities [146]. The total sp3 fraction of 30.6 ± 1.3% is significantly lower than the value of 52 ± 2% computed from NEXAFS measurements of the experimental a-C:H:Si:O samples, however, as discussed previously this is typical for MD simulations of amorphous carbon.

The tail in the bond angle distribution towards smaller angles and the small peak near

60◦ is due to the presence of 3 and 4 member rings, which have been seen in other experimental and simulation studies of amorphous carbon [18, 67, 143, 173].

The overall conclusion is that the ReaxFF potential using the chosen parameterization

98 15 3 sp -C:H 20 3 sp -C:C 3 sp -C:Si 15 10

5 Probability (%) Probability (%) 5

0 0 1.0 1.5 2.0 2.5 40 80 120 160 Bond Length (Å) Bond Angle (°) (a) (b)

Figure 4.6: Distribution of bond lengths (a) and angles (b) for sp3-carbon before annealing. The average bond lengths are Csp3-H: 1.09 A,˚ Csp3-C: 1.54 A,˚ and Csp3-Si: 1.98 A.˚ The average bond angle is 108.5◦.

can accurately model the structure of a-C:H and a-C:H:Si:O. In the next chapter, the results

of annealing simulation will be presented, which provide insight into the mechanisms by

which the addition of Si and O can increase thermal stability.

99 CHAPTER 5: Atomistic origins of enhanced thermal stability from Si and O

As shown in experiments in Chapter3, a-C:H:Si:O has significantly enhanced thermal stability compared to a prototypical a-C:H. However, experimental spectroscopic techniques are not able to provide a complete picture of the reaction processes leading to thermal decomposition, or to show how the addition of Si and O increases thermal stability. Therefore, in order to gain atomistic insight into the mechanisms by which a-C:H:Si:O achieves enhanced thermal stability MD annealing simulations were performed.

Simulated samples of a-C:H and a-C:H:Si:O produced following the liquid quenching procedure described in Section 4.2 were heated over a range of temperatures. Two types of simulations were considered. First, single temperature annealing was performed at 275, 525, and 825 ◦C. The simulations were performed in the NVT ensemble where the temperature was controlled with a Nose-Hoover thermostat with the damping constant set to 100 time steps (10 fs). The temperature was ramped to the target over 5 ps and maintained at that temperature for 100 ps. The results of these simulations were used to explore the mechanisms of thermal decomposition and the effect of Si and O. Second, linear heating ramps were performed up to 1525 ◦C. These heating ramp simulations were used to compute activation energy distributions for thermally induced rehybridization from sp3 to sp2 and

were compared to quantitative experimental results.

5.1. Constant temperature annealing of a-C:H and a-C:H:Si:O

Figure 5.1 shows results for the thermal annealing of a-C:H:Si:O. The sp3 carbon fraction

initially decreases with annealing time, consistent with thermally-activated rehybridization,

followed by apparent stabilization at a final value [67, 68, 119]. The amount of rehybridiza-

tion increases with the annealing temperature. As can be seen in Figure 5.1, the sp3 fraction

stabilizes after approximately 40 ps of annealing, suggesting that the simulation timescale

was long enough to capture the primary mechanisms for thermally activated rehybridiza-

tion. It is clearly possible that processes with slower kinetics due to larger energy barriers

will occur on longer timescales; such processes are beyond the capabilities of these sim-

100 35 275°C 35 525°C ) 525°C ) 825°C

% %

( 825°C (

n n

o o

ti ti

c 30 c 30

a a

r r

F F

n n

o o

b b

r r 25 25

Ca Ca

3 3

p p

s s

20 20 0 20 40 60 80 100 0 20 40 60 80 100 Time (ps) Time (ps) (a) (b)

Figure 5.1: Sample of simulated annealing data from MD for a-C:H:Si:O (a) and a-C:H (b). The temperature was ramped to the value indicated in the legend over 5 ps. There is an initial drop in sp3-bonded C, followed by stabilization. The amount of rehybridization increases with temperature. ulations. However, the agreements between these results and experimental observations support the contention that they have captured essential processes. Specifically, previous work has shown rehybridization from sp3 to sp2-carbon occurs with annealing for both a-C:H

and a-C:H:Si:O [76, 119, 120], and a comparison of the current simulation results to XPS

measurements shows good qualitative agreement in terms of sp3 to sp2 conversion [121]. In

addition, short timescale heating is technologically relevant for applications such as HAMR

disk drives, where the heating time is on the order of 1 ns [77]. In the simulations, the pri-

mary mechanism for rehybridization was the breaking of C-C bonds between sp3 hybridized

atoms, leading to the creation of sp2-hybridized C atoms. For comparison, a-C:H was also

annealed in simulations, and the results are shown in Figure 5.1b. Qualitatively, the pri-

mary mechanism for rehybridization is the same (the breaking of C-C bonds). However,

there is a greater fraction of rehybridization for a-C:H compared to a-C:H:Si:O.

In experiments, dehydrogenation of a-C:H is observed during annealing [69, 73, 75], and

similar behavior has been measured for a-C:H:Si:O [120]. This has been attributed to the

thermally activated breaking of sp3 C-H bonds and the release of trapped free hydrogen

101 in the film [73, 120]. However, this process is not observed in the current simulations for either material. In the simulations, rehybridization is due to the breaking of C-C bonds

(and some C-Si bonds for a-C:H:Si:O), while the breaking of C-H bonds is relatively rare.

Similar behavior has been observed previously in simulations of a-C:H using the REBO potential. In Ref [74] the thermal degradation of a-C:H was investigated as a function of hydrogen content, and it was observed that while C-C bond breaking began between 500–

600 ◦C, there was minimal C-H bond breaking at temperatures less than 2000 ◦C. Similarly, in Ref [76] the REBO potential was used to simulate annealing at 1135 ◦C, and observed a progressive increase in the sp2 carbon fraction but no decrease in the number of C-H bonds.

In terms of bond strength, the relative strength of C-C vs. C-H bonds depends on the compound in question [174], with possibilities for either to be stronger. It can be hypothesized that due to the structural disorder there is a large degree of C-C bond length and angle strain which does not significantly affect the C-H bonds, and these strained C-

C bonds are weaker than the C-H bonds. One additional explanation for the apparent discrepancy between the simulations and experiments regarding the loss of hydrogen is that the hydrogen loss is not from the breaking of C-H bonds, but primarily the results of the release of free H2 that was trapped in the film (which is not present in the MD simulations). Experimental support for this theory can be found in work by Peng et al. [120]. There, a large decrease in the hydrogen content was observed for a-C:H:Si:O annealed to 300 ◦C.

However, this decrease was attributed primarily to the desorption of water and surface hydrocarbons and the effusion of trapped hydrogen, rather than C-H bond breaking and rehybridization, since the change in hydrogen content did not correspond to a change in sp3

fraction. In addition, Rose et al. [76] performed rapid thermal annealing of a-C:H to 650 ◦C

and observed only minor loss of hydrogen, while Raman spectra from the annealed samples

showed an increase in the PL background, which is correlated with increased C-H bonding.

102 5.2. Si doping increases thermal stability through a reduction in bond length disorder

To study how the inclusion of Si and O into a-C:H affects thermal stability, simulated annealing results for a-C:H and a-C:H:Si:O were compared in detail. Re-hybridization was primarily the result of tensile strained bonds breaking at high temperatures, as observed in the bond length histograms for both materials, as shown in Figure 5.2. This is expected, as the bond strength is related to bond length, so the highly strained bonds have the lowest strength. The change in sp2 fraction vs. annealing temperature (for 100 ps) is shown in

Figure 5.3a. a-C:H is observed to have a greater rate of re-hybridization compared to a-C:H:Si:O. Similar to the case for a-C:H:Si:O, it is observed that the primary cause of

0.15 a-C:H Broken Bonds Unchanged 0.10

0.05

0.00 C-C bonds 3 0.15 a-C:H:Si:O Broken Bonds Unchanged 0.10 Fraction of sp

0.05

0.00 1.2 1.4 1.6 1.8 2.0 Bond Length (Å)

Figure 5.2: Distribution of sp3-C:C bond lengths at 300 K for a-C:H:Si:O (top) and a-C:H (bottom). The light colored region shows the bonds which broke resulting in rehybridization from sp3 to sp2-C when annealed at 1100 K for 100 ps. Rehybridization was primarily due to tensile strained C-C bonds.

103 re-hybridization is the breaking of tensile strained C-C bonds. However, when compared to a-C:H:Si:O, there are substantially more strained bonds in the unheated a-C:H sample

(see Figure 5.3b). For the simulated a-C:H, 25% of the sp3 C-C bonds had a length greater than 1.6 A˚ (compared to 20% for a-C:H:Si:O). In addition, highly strained C-C bonds in a-C:H represent 20% of the total sp3 C-X bonds, compared to only 10% of the total sp3 C-X bonds in a-C:H:Si:O. This larger number of highly strained bonds results in the greater rehybridization observed for a-C:H compared to a-C:H:Si:O for the same amount of annealing

(as seen in Figure 5.3a). Thus, a hypothesis for explaining the increased thermal stability of a-C:H:Si:O is that it arises from the presence of significant carbon-silicon bonds with longer equilibrium bond length (154 pm for C-C and 134 pm for C=C, vs. 185 pm for C-Si) [172].

This longer bond length allows a-C:H:Si:O to accommodate more structural disorder without the inclusion of many strained C-C bonds, increasing thermal stability. This is supported by the fact that a-C:H:Si:O is observed experimentally to have less residual stress than

75 a-C:H a-C:H:Si:O a-C:H 70 a-C:H:Si:O 0.10

65 bonds 3 Carbon 2

60 0.05

Percent sp 55 Fraction of sp

50 0.00 0 200 400 600 800 1.2 1.4 1.6 1.8 2.0 Temperature (°C) Bond Length (Å) (a) (b)

Figure 5.3: (a) Increase in sp2 carbon fraction as a function of annealing temperature for a-C:H and a-C:H:Si:O. Annealing was for 100 ps. The lines are guides for the eye. There is a greater degree of rehybridization for a-C:H. Error bars show the max and min values observed over three replications with randomized initial configurations. (b) C-C bond length histograms for a-C:H and a-C:H:Si:O. The increase in rehybridization is attributed to the greater fraction of highly strained bonds in a-C:H compared to a-C:H:Si:O as seen by comparing the histograms. This is due to the presence of longer C-Si bonds in a-C:H:Si:O.

104 a-C:H [14–16, 88, 89]. Due to the low atomic concentration of oxygen in the a-C:H:Si:O simulated here, and the low amount of C-O bonding in particular (less than 2% of all carbon bonds), it is not possible to draw reliable conclusions about the role of oxygen from the present simulations. However, recent density functional theory simulations have shown that carbon-silica bonds are stronger than carbon-silicon bonds, suggesting a possible mechanism by which oxygen also contributes to thermal stability [175]. In addition, experimental work has shown that a-C:H:Si:O resists oxidation at high temperatures in aerobic environments by the formation of a silica surface layer, an effect that would not be captured in these simulations [42]. Further experimental and simulation studies, including ones that examine increased amounts of oxygen in these films, would be needed to fully explore oxygen’s role.

5.3. Changes in Si bonding at high temperature

There were also changes observed in the bonding character of silicon when the sample was annealed, which are shown in Figure 5.4. Specifically, Si-H bonding decreases and Si-O bonding increases with temperature. The liberated hydrogen from the reduction in Si-H bonding re-bonded to carbon, while being replaced by oxygen, resulting in the increase in Si-O bonds. The Si-C bonding decreases slightly at higher temperatures, resulting in some of the observed rehybridization. However, the primary mechanism for rehybridization remains the breaking of C-C bonds. Although sp3 C-C bonds are typically stronger than

Si-C bonds, the disorder present in the amorphous structure gives rise to a large fraction of tensile strained C-C bonds in the unheated structure (as shown in Figure 5.2). For the simulated a-C:H:Si:O 20% of the sp3 C-C bonds have a bond length greater than 1.6 A.˚

These weakened C-C bonds can have lower strength than Si-C bonds on average. This observation is in agreement with density functional theory simulations of silica and diamond that have shown Si-C bonds can wear weakened diamond C-C bonds at the interface [175].

105 Si:O 1.2 Si:C Si:H 1.1

1.0

0.9

0.8 Fractional Change in Si:X Bonds

0 200 400 600 800 Temperature (°C)

Figure 5.4: Change in Si bonding for a-C:H:Si:O with annealing temperature. Error bars show the max and min values observed over three replications with randomized initial configurations.

5.4. Clustering and ordering of sp2 carbon

It is known that thermal energy can not only rehybridize bonds from sp3 to sp2 but, as reported for a-C:H films, can also cause existing sp2-hybridized bonds to cluster and eventually order [119]. Clustering of sp2 carbon was measured in the simulations by counting the average number of sp2-bonded neighbors around each sp2 carbon atom. Figure 5.5 shows the relative fraction of sp2-hybridized neighbors increased for both a-C:H and a-C:H:Si:O.

This indicates a progressive clustering of the sp2 carbon with increasing temperature. Con- sistent with the expected increase in thermal stability attributed to the inclusion of Si and

O, there is less clustering with annealing temperature for a-C:H:Si:O.

However, this is an imperfect metric for measuring clustering and ordering of the sp2 carbon. First, the sp2 fraction is increasing with annealing, which will increase the number of sp2-bonded neighbors. Second, this provides no information about ordering of the sp2 phase. One alternative is to look at the formation of graphitic clusters by considering the amount of sp2 carbon arranged in six-member rings. This is shown in Figure 5.6. The rings were identified using an algorithm from Yuan and Cormack [176], which was implemented

106 2.2 a-C:H a-C:H:Si:O 2.0

1.8 -bonded neighbors 2

1.6 Avg. # of sp 0 200 400 600 800 Temperature (°C)

Figure 5.5: Relative frequency of sp2-bonded neighbors to sp2 carbon, for a-C:H and a- C:H:Si:O. When annealed for 100 ps there is an increase in the frequency of sp2 carbon neighbors indicating a growth of sp2 clusters. Lines are guides for the eye. in MATLAB. As seen in the figure, the unheated structures for both a-C:H and a-C:H:Si:O had almost no sp2 carbon arranged in ordered six-member rings. However, when annealed at 825 ◦C, there are clear differences between the two materials. For a-C:H, the number

of six-member sp2 carbon rings increased rapidly, suggesting clustering and ordering of the

sp2 phase. In contrast, only a small increase was observed for a-C:H:Si:O. This indicates

the inclusion of silicon and oxygen significantly inhibits the clustering and ordering2 ofsp

carbon in a-C:H:Si:O.

5.5. Effect of varying Si content on the structure and thermal stability

To check the hypothesis that the presence of Si reduces the average C-C bond length, a

set of simulations was run with the ratio of elements (C:H and Si:O) and density matched to

the experimental samples, but with varying Si content. As can be seen in Figure 5.7a, there

is a strong correlation between Si content and the overall fraction of sp3-C in a-C:H:Si:O.

This is in good agreement with experimental measurements of films with varying Si:O

content [70, 80, 83, 87], which have shown an increase in sp3-C bonding with Si fraction.

107 20 a-C:H a-C:H:Si:O 15

10 6-member rings 2

5 Number of sp 0 0 20 40 60 80 100 Time (ps)

Figure 5.6: Increase in the number of six-member sp2 carbon rings with annealing time for a-C:H and a-C:H:Si:O at 825 ◦C. In both cases there was only one six-member sp2 carbon ring in the unheated structure. When annealed at 825 ◦C, the number of six-member sp2 carbon rings increased rapidly for a-C:H, suggesting clustering and ordering of the sp2 phase. In contrast, only a small increase was observed for a-C:H:Si:O.

The average C-C bond length decreases as the Si fraction increases, eventually reaching the

expected average value based on equilibrium carbon bond lengths. This results in a decrease

in the fraction of highly strained sp3 C-C bonds, as shown in Figure 5.7b, confirming that Si doping significantly relieves strain by preventing elongated C-C bonds from forming. This in turn suggest that film deposition recipes that increase the silicon content can leadto increased thermal stability.

5.6. Comparison between simulations and experimental results

Although quantitative agreement between the experiments and simulations is not expected due to the vastly different timescales, the simulation results presented here show good qualitative agreement with experiments. In both cases the addition of small concentrations of silicon and oxygen results in significant improvements in thermal stability. Both

Raman and NEXAFS measurements in Chapter3 show thermal degradation in a-C:H:Si:O through rehybridization from sp3 to sp2 carbon, and the rate of rehybridization decreases

108 40 20

35 15 -C:C bonds

30 3

Carbon (%) 25 10 3

20 5 15 Percent sp 10 Percent of strained sp 0 0 5 10 15 20 25 0 5 10 15 20 25 Si Fraction (%) Si Fraction (%) (a) (b)

Figure 5.7: (a) sp3-carbon fraction as a function of Si content in simulated a-C:H:Si:O. At a given density, the sp3 fraction significantly increases with Si content. (b) Percent of sp3 C-C bonds that are highly strained (bond length greater than 1.6 A)˚ as a function of Si content. As the Si content increases, the fraction of highly tensile strained C-C bonds decreases, from 18% at 0% Si to 2% at 25% Si.

with the addition of silicon and oxygen. As shown in Section 5.1, the same is true in simu-

lations as well. Furthermore, the structural details predicted by MD, in particular the low

fraction of C=O bonds (< 2% of all bonds), the increase in silicon oxidation state with annealing shown in Section 5.3, and the clustering and ordering of sp2 carbon in Section 5.4 are in agreement with analysis of XPS spectra by [121] discussed in Section 3.3. Finally, the increase in six-member sp2 carbon rings is in good agreement with the Raman spectra shown in Figure 3.3. For DLC, an increase in the Raman D peak intensity is expected to be related to an increase in sp2 carbon rings. The much larger increase in D peak intensity for a-C:H is in agreement with the determination of six-member rings from the MD data shown in Figure 5.6. These good qualitative agreements between the simulations and experimental spectroscopy provides confidence that the ReaxFF potential is able to accurately modelthe thermal degradation pathways in a-C:H:Si:O, and the conclusions drawn for the simulations are relevant to explaining the experimental observations.

109 5.7. Modeling the distribution of activation energies from heating ramps

As seen in the preceding work, simulations with the ReaxFF potential have accurately reproduced experimental trends, and shed insight into the thermal degradation processes in a-C:H:Si:O. In order to extend this work further, it would be of interest to compute the activation energy for carbon rehybridization to facilitate a quantitative comparison with experimental results. As discussed previously, the amorphous structure in DLCs gives rise to a large number of different bonding environments, which suggests a wide range of variation in activation energy. To handle this fact, one could choose to treat the activation energy as a probability distribution (often assumed to be Gaussian) and compute the mean and standard deviation. This approach has often been taken in literature dealing with modeling complex reactions, such as coal or biomass pyrolysis, and is referred to as a distributed activation energy model (DAEM) [177, 178]. Previously this model has been used by Sullivan et al. [67] to model rehybridization leading to stress relaxation in a-C, and by Mangolini et al. [119] to model thermally induced rehybridization in a-C:H.

In this approach it is assumed the rehybridization is governed by a first order reaction with a rate constant given by the Arrhenius equation:

k = νe−Ea/(kB T )

where ν is the attempt frequency (1/s), Ea is the activation energy (eV), T is the temperature (K), and kB is the Boltzmann constant. Then for a first order reaction where α is the

3 3 fraction of a reactant decomposed at time t (in this case α = spLost/spInitial), the reaction is described by the equation:

dα E = ν exp − a · (1 − α) dt kBT

110 which after integrating gives:

Z t E α = 1 − exp − ν exp − a dt 0 kBT where T can be a function of time. If N(Ea) represents the probability distribution of energy barriers, then the equation becomes:

Z ∞ Z t Ea α = N(Ea) 1 − exp − ν exp − dt dEa 0 0 kBT

Here N(Ea) can be any probability distribution and T any function of time. However, it is convenient to let N(Ea) be a normal distribution characterized by mean activation energy

E0 and standard deviation σ and vary the temperature linearly with a heating rate β (K/s). Then after substitution and simplification and rewriting in terms of remaining sp3 fraction rather than the amount of sp3 lost, the final fit equation is:

Z ∞ Z T 2 1 −ν ˜ (Ea − E0) 3 3 √ −Ea/(kB T ) ˜ spT = spi exp e dT − 2 dEa (5.1) σ 2π 0 β Ti 2σ where ν, E0, and σ are fitting parameters. However, both ν and E0 cannot be fit, or the solution would not be unique, so the value of ν must be fixed. In this case it is assumed

ν = 1012/s, which is a typical value used in the literature for solid state reactions [67,

119]. Finally, in a-C:H:Si:O there could be multiple processes contributing to thermal decomposition with different kinetic behavior (for example the mean activation energy for carbon bonded to silicon vs. carbon not bonded to silicon could be different). This can be modeled by allowing the activation energy distribution to be a sum of Gaussian distributions with different means and standard deviations weighted by relative atomic concentrations of atoms in the different bonding environments [177]. This is the approach used by Mangolini et al. [119] to model the thermal decomposition of a-C:H.

For fitting activation energies using Equation 5.1, MD simulations were performed using linear temperature ramps from 300 to 1100 K at heating rates of 4 × 1012 K/s, 8 × 1012 K/s,

111 and 1.6 × 1013 K/s, which are typical for MD simulations, although they are fast compared with experiments. Although heating at different rates is not necessary for computing the fit parameters, it is useful for validating that the overall modeling procedure is giving meaningful results by checking that consistent values are obtained at each heating rate.

The simulated samples were produced following the liquid quenching procedure described in Section 4.2, with 1–1.5 GPa in plane stress. For comparison, a-C and a-C:H samples were also annealed in the same manner.

Figure 5.8a shows the results of fitting the a-C:H:Si:O data for the three different heating rates with the same single distribution fit. As can be seen from the figure, asingle fit with Ea = 0.96 ± 0.38 eV can model each set of data. However, this activation energy is significantly different from the value obtained from fitting the experimental results

(3.0 ± 1.1 eV) [121]. This discrepancy will be discussed in Section 5.8. In comparison, the activation energy distributions for a-C and a-C:H in the simulations were computed to be

0.74 ± 0.31 eV and 0.62 ± 0.24 eV respectively. These fits are shown in Figure 5.9. Thus, in order of thermal stability, a-C:H < a-C < a-C:H:Si:O.

In order to investigate the effect of local bonding environment on thermal stability, the simulation results for a heating rate of 8 × 1012 K/s were fit using multiple distributions, corresponding to carbon bonded to Si/H vs carbon not bonded to Si/H. Although it is possible to perform a fit using more than two distributions, in practice it is difficult with the small number of atoms in the simulation, and results in large uncertainty in the fit parameters. The results of this fitting for a-C:H:Si:O are shown in Figure 5.8b.

For the single distribution fit Ea = 0.96 ± 0.38 eV, as stated previously. For the multiple distribution fits, two cases were considered. In the first case, two distributions werefit corresponding to thermally activated rehybridization from sp3 to sp2 for those carbon atoms

3 2 bonded to at least one silicon atom (sp C-Si-to-sp ) vs. rehybridization for those carbon

3 2 atoms bonded to no silicon atoms (sp C-C-to-sp ). Note that this is not the same as the

3 2 activation energy for the breaking of C-Si vs. C-C bonds, and in most cases sp C-Si-to-sp conversion was by means of breaking a C-C bond. This yielded an activation energy of

112 24

Carbon 22 3

13 Simulation (1.6 x 10 K/s) Fit (E0 = 0.96 ± 0.38 eV) 12 20 Simulation (8 x 10 K/s)

Percent sp Fit (E0 = 0.96 ± 0.38 eV) 12 Simulation (4 x 10 K/s) Fit (E0 = 0.96 ± 0.38 eV) 18 400 600 800 1000 1200 Temperature (K) (a)

23 Carbon 3

Simulation Results Percent sp 21 Single Distribution Fit 2 Distribution Fit (C-C/C-Si) 2 Distribution Fit (C-C/C-H) 20 400 600 800 1000 Temperature (K) (b)

Figure 5.8: (a) Fit of temperature ramps at three different heating rates using a single setof 12 fitting parameters (Ea = 0.96 ± 0.38 eV). (b) Fit of a single heating ramp (at 8 × 10 K/s) with a single distribution fit compared with two distribution fits. The activation energies computed by each fit are given in the text.

113 50

Simulation (a-C:H) Carbon 3 30 Fit (E0 = 0.62 ± 0.24 eV) Simulation (a-C) Fit (E0 = 0.74 ± 0.31 eV)

20 Percent sp

10 400 600 800 1000 Temperature (K)

Figure 5.9: Single distribution fits for temperature ramps for a-C:H and a-C at a heating rate of 8 × 1012 K/s. The mean activation energy for rehybridization is lower for a-C:H than a-C, with both being lower than a-C:H:Si:O (Ea = 0.96 ± 0.38 eV).

3 2 3 2 0.79 ± 0.31 eV for sp C-C-to-sp conversion vs. 1.50 ± 0.60 eV for sp C-Si-to-sp conversion. The higher activation for rehybridization of carbon atoms bonded to silicon is in good

agreement with the hypothesis that Si-doping increases thermal stability by reducing the

number of tensile strained C-C bonds, increasing the average C-C bond strength.

In the second case, two distributions were fit corresponding to thermally activated re-

hybridization from sp3 to sp2 for those carbon atoms bonded to at least one hydrogen atom

3 2 (sp C-H-to-sp ) vs. rehybridization for those carbon atoms bonded to no hydrogen atoms

3 2 (sp C-C-to-sp ). Again, it is important to mention that this is not the same as the activation energy for the breaking of C-H vs. C-C bonds. In this case, an activation energy

3 2 3 of 0.68 ± 0.27 eV was found for sp C-C-to-sp conversion vs. 1.31 ± 0.52 eV for sp C-H-to- sp2 conversion. This higher activation energy for rehybridization of carbon atoms bonded

to hydrogen is surprising, since experimental results have shown a-C:H has significantly

lower thermal stability than a-C, and thermal stability decreases with increasing H con-

tent [69, 73, 75]. This is not just the case for a-C:H:Si:O, as performing the same analysis

3 2 for the a-C:H samples gives an activation energy of 0.72 ± 0.26 eV for sp C-H-to-sp vs.

3 2 0.56 ± 0.22 eV for sp C-C-to-sp .

114 This presents a rather unexpected result. Although a-C:H is less thermally stable compared to a-C (Ea = 0.62 ± 0.24 eV vs. 0.74 ± 0.31 eV for single distribution fits), carbon bonded to at least one hydrogen atom is more thermally stable than carbon bonded to no hydrogen atoms. These results suggest that the effect of hydrogen on thermal stability isto decrease the stability of C-C bonds, rather than C-H bonding being a weak link resulting in lower thermal stability. This description is consistent with the observed bond breaking discussed in Section 5.1, where relatively few C-H bonds break with annealing.

5.8. Effect of stress on activation energy distribution

As mentioned previously, by comparison to experiments, it is clear that the activation energies computed from the simulations are too low. For the single distribution fit for a-C:H:Si:O the value of Ea = 0.96 ± 0.38 eV is far lower than the experimental results (3.0 ± 1.1 eV) [121], and suggests that some rehybridization would occur even at room temperature (since a non-negligible fraction of sp3 carbon atoms are predicted to have

Ea ∼ 0 eV). However, it was hypothesized that the activation energy should increase with compressive stress, which was ∼ 0 GPa in the previous simulations. The reason for this is due to the difference in volume between3 sp and sp2 carbon (with sp3 carbon being denser).

Thermodynamically, sp2 carbon is known to have a lower cohesive energy than sp3 carbon at zero pressure [179]. However, due to the volume increase between sp3 and sp2 carbon, the energy difference decreases with pressure. This results in a phase diagram where, above a certain pressure, diamond, rather than graphite, is the more thermodynamically stable form of carbon [179].

Experimentally DLC films are known to possess significant amount of residual compressive stress. For example, values of compressive stress for ta-C can range from 4–

15 GPa [18, 82], with values higher than 6 GPa being typical [67]. a-C:H tends to have slightly lower compressive stress, but it still can as high as 10 GPa [18]. There tends to be good correlation between residual compressive stress and sp3 fraction for DLC [18], and it has been suggested that local compressive stresses produced during deposition are neces-

115 sary for the stabilization of the sp3 bonding [28, 31, 32]. The addition of silicon, however, is known to reduce the residual compressive stress. Studies of Si-doping have found the stress reduction varies non-monotonically with Si content, and values between 0.1 and 3 GPa have been reported [80–82]. For a-C:H:Si:O the residual stress is similar, with values between

0.7 and 1.5 GPa found in the literature, depending on deposition conditions [84, 89].

The effect of stress on activation energy was investigated in the simulations byapply- ing 1–1.5 GPa compressive stress in the X-Z plane during the heating ramps. Although

the amount of compressive stress in the experimental samples is not known, these values

are reasonable given the previously discussed stress measurements for Si and Si:O-doped

DLCs. The results are shown in Figure 5.10. As can be seen, the activation energy dis-

tribution is strongly dependent on stress, increasing from Ea = 0.96 ± 0.38 eV at 0 GPa to

Ea = 2.81 ± 1.09 eV at 1.5 GPa. This latter value agrees very well with the 3.0 ± 1.1 eV computed from experiments.

In addition, multiple distribution fits were computed, with two distributions corre-

Fraction (%) 22 3 sp 1.0 GPa 21 Fit (Ea = 1.55 ± 0.48 eV) 1.5 GPa Fit (Ea = 2.81 ± 1.09 eV) 20 500 1000 1500 Temperature (K)

Figure 5.10: Simulation data and single distribution activation energy fits for a-C:H:Si:O with in-plane compressive stress. The fitted activation energy (shown in the legend) is strongly dependent on stress.

116 3 2 3 2 sponding to sp C-C-to-sp vs. sp C-Si-to-sp . For the 1 GPa data Ea = 1.17 ± 0.30 eV

3 2 3 2 for sp C-C-to-sp compared to 3.38 ± 1.32 eV for sp C-Si-to-sp , while for the 1.5 GPa data

3 2 3 2 Ea = 2.01 ± 0.79 eV for sp C-C-to-sp compared to 4.93 ± 1.83 eV for sp C-Si-to-sp . The

3 2 large standard deviations for the sp C-Si-to-sp activation energies is due to the very small amount of total rehybridization that occurs, which decreases the ability to accurately fit

the data.

The effect of stress on activation energy is often modeled as a linear effect andgiven

by [180, 181]:

Etot = Ea ± σ∆V

where Etot is the total activation energy, Ea is the stress -free activation energy, σ is the stress, and ∆V is a quantity called the activation volume. It would take more simulations

at different amounts of stress to confidently determine the effect of stress on activation

energy, but from the simulations currently preformed, the behavior appears non-linear (see

Figure 5.11). One possibility is that the activation energy is linearly related to stress, but

the trend is not apparent from the limited number of simulations preformed and given the

uncertainty. In addition, due to the small volume and stiff bonds, stress is not well controlled

in the simulations and has large fluctuations over time (± 1 GPa). To better evaluate the

effect of stress on activation energy, it would be necessary to perform additional simulations

(ideally with larger numbers of atoms), which is computationally challenging for the ReaxFF

potential.

5.9. Conclusions

MD simulations using the ReaxFF potential have provided new insight into the mech-

anisms responsible for the enhanced thermal stability of a-C:H:Si:O compared to a-C:H.

MD results agree with experimental data that the incorporation of even modest amounts of

silicon and oxygen (e.g. 6 at.% and 3 at.% respectively) in a-C:H increases thermal stability

significantly. The primary mechanism for the rehybridization of sp3 to sp2 carbon is the

117 4

1 Activation Energy (eV)

0 0.0 0.5 1.0 1.5 Compressive Stress (GPa)

Figure 5.11: Activation energy for sp3 to sp2 conversion vs. stress for a-C:H:Si:O. The error bars represent the fit standard deviation for the activation energy distribution, not the confidence interval for the mean activation energy. The large increase in standard deviation at 1.5 GPa is due to the small amount of total rehybridization, which makes fitting the activation energy distribution difficult.

breaking of highly strained C-C bonds, while C-H bonds do not break until higher temper-

atures, and it is the reduction of these strained bonds by incorporation of Si that leads to

the increased thermal stability. Specifically, a comparison of C-C bond length distributions

for a-C:H:Si:O and a-C:H shows that the incorporation of silicon allows the structure to

accommodate more disorder by reducing the fraction of highly strained C-C bonds due to

the presence of Si-C bonding. This presents a mechanism by which the thermal stability

of a-C:H is enhanced by the addition of silicon. Simulations as a function of silicon con-

tent show a strong correlation between the atomic fraction of Si and the amount of sp3-C bonding. The fraction of high strained (bond length > 1.6 A)˚ C-C bonds decreases with silicon content, and approximately 25 atomic % silicon results in less than 2% of the C-C bonds being highly strained. This information could be used in film design where the silicon content can be controlled as it suggests the thermal stability would increase with silicon content up to around 25% Si.

Additionally, the activation energy for thermally induced carbon rehybridization was modeled as a first order reaction with a Gaussian distribution of energy barriers, matching

118 the approach used to fit the experimental results. The in-plane compressive stress, which is commonly observed for DLC films was found to have a significant impact on themean activation energy. This is attributed to the increase in volume occupied by sp2 carbon compared to sp3 carbon. Assuming a typical value of compressive stress for a-C:H:Si:O gives an activation energy of 2.81 ± 1.09 eV compared to 3.0 ± 1.1 eV for experiments. This strong impact of compressive stress on thermal stability is generally not considered in the literature.

For example, MD studies of the annealing behavior DLC formed by liquid quenching do not report or control the stress [145]. Generally, liquid quenching to form DLC is carried out at constant volume, where the density is fixed, but the resulting structures can have widely varying stress states [143, 145, 146]. A better approach is to produce structures with a specific amount of residual stress, while adjusting the initial volume to get approximately the desired density.

Multiple distribution fits were used to investigate the effect of local bonding on activation energy and it was found that carbon bonded to silicon has significantly higher mean activation energy for rehybridization compared to carbon not bonded to silicon, in agreement with the observed increase in thermal stability for a-C:H:Si:O. It was also observed that although a-C:H has lower thermal stability compared to a-C or a-C:H:Si:O, the mean activation energy for rehybridization was higher for carbon bonded to hydrogen compared to carbon not bonded to hydrogen. In the literature the thermal degradation of a-C:H is often discussed in terms of three processes: the rehybridization of sp3 to sp2 carbon, the clustering and ordering of sp2 carbon, and loss of hydrogen [10, 69, 119, 120]. It is often assumed that the loss of hydrogen is due to the breaking of sp3 C-H bonds at low temperatures, which results in the lower thermal stability of a-C:H compared to a-C [69, 73, 119].

However, these simulation results suggest a different possible mechanism; that the decrease in thermal stability for a-C:H is related to the weakening of C-C bonds with increasing hydrogen content, rather than rehybridization due to C-H bond breaking. In this case, the loss of hydrogen at low temperatures could be due to the release of trapped hydrogen or small hydrocarbon gas molecules and the desorption of surface hydrogen, rather than C-H

119 bond breaking [120]. Further investigation is required to elucidate the exact mechanisms by which hydrogen influences the thermal stability.

120 CHAPTER 6: Conclusions

Hydrogenated amorphous carbon films are an interesting and important technological material. However, a-C:H becomes unstable above 150 ◦C, preventing its use in high temperature applications, such as heat assisted magnetic recording (HAMR) disk drives. Also, low friction and wear is only maintained in dry and vacuum conditions. To understand and address these limitations, this thesis considered the effect of adding silicon and oxygen to a-C:H films.

6.1. Main results and implications for the design and use of a-C:H:Si:O coatings

In this work, the tribology of a-C:H:Si:O sliding against 52100 steel was investigated through ball on flat tribometer experiments over a range of loads and environments. The use of multisection wear tracks allowed the progression of wear to be monitored in addition the coefficient of friction. AFM measurements revealed detailed nanoscale topography which suggested key mechanisms controlling run-in, friction, and wear. In addition, a variety of experimental techniques were used to probe the thermal stability of a-C:H:Si:O in comparison to a-C:H and showed that the incorporation of even modest amounts of silicon and oxygen (e.g. 6 at.% and 3 at.% respectively) increases thermal stability significantly.

MD simulations using the ReaxFF potential have provided new insight into the mechanisms responsible for this enhanced thermal stability of a-C:H:Si:O compared to a-C:H. Beyond advancing fundamental understanding of the tribology and thermal stability of a-C:H:Si:O, the work in this thesis can help improve the design of tribological coatings, for example, through optimizing the quantity of Si and O.

6.1.1. Environmental dependence of the friction and wear for a-C:H:Si:O

The strong environmental dependence of the friction, in particular the large increase in friction and wear with humidity, is a major limitation for the use of a-C:H. High hydrogen content a-C:H (∼ 40% H) has superlow friction (µ ∼ 0.001–0.02) in vacuum and inert

121 environments, but this will increase to values between 0.02 and 0.5 in humid air [10, 12].

In Section 2.5, friction and wear testing was performed on two different a-C:H:Si:O samples with varying atomic composition in three different environments; laboratory air

(40% RH), dry air (< 5% RH), and dry nitrogen (< 5% RH). NCD Technologies a-C:H:Si:O displayed similar environmental trends to a-C:H, however, the increase in friction coefficient with humidity (0.03 in dry nitrogen and 0.05 in dry air vs. 0.13 in humid air) was more moderate. This behavior is consistent with prior literature on DLN 180 a-C:H:Si:O which also shows a lower coefficient of friction in dry environments (0.02–0.05) compared tohumid air (0.22) [17]. Wear is also environmentally dependent, and increases significantly compared to sliding in dry nitrogen when oxygen or water vapor are present. AFM images of wear track topography in Section 2.6 show wear is progressing through dominantly adhesive interactions (formation and breaking of adhesive junctions in the film). In more reactive environments, such as dry air compared to dry nitrogen, wear increases. This is consistent with the idea of reactive environmental species etching carbon bonds in the film. In AFM measurements this is evident due to an increase of adhesive wear events present in the topography images.

For DLN 360 a-C:H:Si:O, which has a higher silicon content than the NCD Technologies samples, different behavior was observed. In humid air, the steady state coefficient offriction was ∼ 0.15, similar to the NCD Technologies films. However, in dry air the coefficient of friction increased to over 0.6. Sliding in dry nitrogen was highly unstable, and could jump between two regimes. In some cases friction coefficients less than 0.05 were observed, while in other cases the friction would increase rapidly to values similar to dry air (0.6). In

Section 2.7 this was show to depend on tribofilm formation during the run-in process. For a-C:H:Si:O, the run-in process is proposed to consist of surface modification (possibly due to sp3 to sp2 conversion) and tribofilm formation. These run-in processes depend onfilm composition and structure. In the case of the high silicon DLN 360 a-C:H:Si:O, stable tribofilm formation was difficult in dry environments, preventing the typical low friction behavior. Further work is needed to determine the precise mechanisms behind the inhibited

122 tribofilm formation for DLN 360 films, however, one possible hypothesis is that the increased

Si-C bonding limited the formation of sp2 carbon during sliding. These observations are highly relevant to the design of a-C:H:Si:O coatings, since they show that the tribological behavior is strongly influenced by the atomic composition of the film.

6.1.2. Dependence of the friction and wear on load

Prior studies on DLN 180 a-C:H:Si:O films have reported large variations in coefficients of friction measured in humid environments. While some studies reported low coefficients of friction (0.04–0.08) in humid environments [16, 91, 99], others reported friction coefficients as high as 0.22 [17]. This difference was attributed to the different loads used (contact stresses between 0.3 and 0.8 GPa) [17].

In Section 2.8, friction vs. load measurements using a macroscopic reciprocation tribometer with a 6.35 mm 52100 steel ball sliding against DLN 360 a-C:H:Si:O in humid air showed non-Amontonian behavior (the coefficient of friction had a power law dependence on load). The friction coefficient decreases from 0.27 at a load of 84 mN to 0.13 at5000mN.

These loads correspond to contact stress between 0.15 and 0.6 GPa. The power law decrease in coefficient of friction with load is expected based on contact mechanics for aspherical contact, which would predict the coefficient of friction should scale as L−1/3. This is due to the dependence of real area of contact on load due to Hertizian elastic deformation of the ball, assuming smooth contact; roughness renders the relation between friction and load to be more linear. In the case of a-C:H:Si:O, the extremely smooth surfaces and soft tribofilms result in such scaling remaining true for this macroscale contacts. However, likely due to the growth of the contact area caused by tribofilm deformation, the measured scaling exponent is −0.179 instead of −1/3. Similar behavior is observed by Scharf et al. [17] on DLN 180 a-C:H:Si:O, where the coefficient of friction showed a power law decrease from 0.22 ata contact pressure of 0.3 GPa to 0.09 at a contact pressure of 0.6 GPa, although they only considered three different loads.

The coefficient of friction in humid air measured here is higher than what is reported

123 in many prior papers on the friction of a-C:H:Si:O in humid air, which report friction coefficients less than 0.1[16, 91, 99]. However, in those papers, a higher load range was used, resulting in contact stresses greater than 0.8 GPa. These results can be consistent with prior literature when extrapolated to a higher load range, which gives a coefficient of friction equal to 0.1 at a contact pressure of 0.95 GPa. This is still slightly higher than the previously reported values. However, those measurements were on a different composition a-C:H:Si:O.

Finally, in Section 2.10, wear rates for a-C:H:Si:O were considered in the context of

Archard’s wear equation [102, 115], which predicts that the worn volume should be proportional to the sliding distance times the load. The constant of proportionality is known as the wear coefficient. Although this equation is purely empirical and wear coefficients computed from it may not be transferable to different sliding conditions, the wear coefficient is frequently used in the literature to quantify a material’s wear resistance [10, 102]. However, this model does not adequately describe the wear data on a-C:H:Si:O since it was shown in this thesis that the wear coefficient measured for DLN 360 films depends strongly onthe load. This demonstrates that the wear coefficient is a poor method by which to quantify the wear resistance of a-C:H:Si:O, since the value is only meaningful in the context of a specific set of experimental conditions. A better approach, motivated by the correlation between friction and wear, is to consider the wear as a function of the work done by the friction force, which gives an approximately linear fit to the data. This same approach works for the NCD Technologies a-C:H:Si:O environmentally dependent friction as well, but not for the DLN 360 samples, likely due to the different energetics of the friction and wear process when sliding on bare steel. One possible application of this approach is to consider wear as occurring through tribochemical reactions, and determine an activation energy for wear.

Previously this method has been applied to ta-C coatings [118]. The energy wear coefficient was found to be temperature dependent, suggesting that wear in ta-C is a tribochemical process, and was used to compute an activation energy for wear. The wear activation energy from this analysis was the same for three different ta-C coatings, suggesting that in each

124 case the same tribochemical process was controlling the wear [118].

6.1.3. a-C:H:Si:O as a thermally stable alternative to a-C:H

The low thermal stability of a-C:H is a significant limitation for many technological applications. In Chapter3, a variety of experimental techniques were used to probe the thermal stability of a-C:H:Si:O in comparison to a-C:H. Annealing in laboratory air up to 350 ◦C for 1 hour resulted in complete film volatilization for a-C:H. Under the same conditions, Raman spectroscopy showed moderate changes for a-C:H:Si:O. Characterization of the Raman spectra based on fitting the D and G peaks suggests thermal degradation through rehybridization from sp3 to sp2 carbon and clustering of sp2 carbon, following a

similar thermal degradation pathway to a-C:H, but at higher temperatures.

It is difficult to quantify the Raman spectra. However, these results are in agreement

with quantitative XPS and NEXAFS spectroscopy performed by Mangolini et al. [121],

which are described in Section 3.3. NEXAFS spectra were used to quantify the sp3 carbon

fraction and observed an increase in sp2 carbon with annealing. This allowed for modeling of the activation energy for thermally induced rehybridization based on a Gaussian distribution of energy barriers. This resulted in an activation energy of 3.0 ± 1.1 eV for a-C:H:Si:O [121] compared to 2.6 ± 1.2 eV for a-C:H [119], showing that the addition of Si and O increases the mean activation energy for rehybridization. These quantitative measurements were used as the basis of comparison between the MD simulations in Chapter5 and the experimental results.

Finally, in Section 3.4, heated AFM probes were used to investigate thermal stability at time and length scales relevant to applications such as HARM disk drives. Significant amounts of film volatilization were observed for a-C:H, even at low temperatures andin- creased with heater temperature (although this is also influenced by the applied stress).

Under the same conditions no changes were observed for a-C:H:Si:O.

This combination of experimental techniques clearly shows the addition of small amounts of silicon and oxygen can substantially increase the thermal stability of a-C:H, overcoming

125 a major technological limitation for these films. However, it is not possible to determine the atomistic mechanisms responsible for this enhanced thermal stability from these spectroscopic measurements. This motivated the use MD simulations to understand the physical basis for the effect of silicon and oxygen on the thermal stability.

6.1.4. Atomistic origins of the enhanced thermal stability through the addition of silicon

MD simulations are a valuable tool to gain insight into the atomic interactions that influence the structure/property relations in a-C:H:Si:O. There are several approaches to modeling DLC including density functional theory (DFT) based methods and empirical potentials. The most common potential used in the literature is REBO, with several studies investigating the tribology and thermal stability of DLC [58, 59, 61, 74, 153, 154]. However, no versions of the REBO potential include all four elements needed for a-C:H:Si:O. In this thesis, the ReaxFF potential was used, with a Si/C/H/O force field developed by

Newsome et al. [156], which is the only reactive potential parameterized to include all of these elements.

The results of MD simulations using the ReaxFF potential to investigate the effect of silicon and oxygen on the thermal stability of a-C:H:Si:O are presented in Chapter5.

MD results agree with experimental data that the incorporation of even modest amounts of silicon and oxygen (e.g. 6 at.% and 3 at.% respectively) in a-C:H increases thermal stability significantly. The primary thermal degradation pathway in a-C:H / a-C:H:Si:O is the rehybridization of sp3 to sp2 carbon through the breaking of highly strained C-C bonds.

In Section 5.2 it is shown that the incorporation of Si leads to a reduction of these strained bonds. This presents a mechanism by which the thermal stability of a-C:H is enhanced by the addition of silicon. Simulations as a function of silicon content in Section 5.5 show a strong correlation between the atomic fraction of Si and the amount of sp3-C bonding. The fraction of high strained (bond length > 1.6 A)˚ C-C bonds decreases with silicon content, and approximately 25 atomic % silicon results in less than 2% of the C-C bonds being highly strained. This information could be used in film design where the silicon content can

126 be controlled as it suggests the thermal stability would increase with silicon content up to around 25% Si.

Following the same approach as the experimental work, in Section 5.7, the activation energy for thermally induced carbon rehybridization was modeled as a first order reaction with a Gaussian distribution of energy barriers. Fits using multiple energy barriers corresponding to different local bonding environments found that carbon bonded to silicon has significantly higher mean activation energy for rehybridization compared to carbon not bonded to silicon. This supports the increase in thermal stability for a-C:H:Si:O.

Interestingly, it was also observed that although a-C:H has lower thermal stability compared to a-C or a-C:H:Si:O, the mean activation energy for rehybridization was higher for carbon bonded to hydrogen compared to carbon not bonded to hydrogen. The thermal degradation of a-C:H is often discussed in terms of three processes: sp3 to sp2 carbon rehybridization, clustering and ordering of sp2 carbon, and loss of hydrogen [10, 69, 119, 120].

The loss of hydrogen, which is observed to begin at low temperatures, is often attributed to the breaking of sp3 C-H bonds, resulting in the lower thermal stability for a-C:H compared to a-C [69, 73, 119]. However, the loss of hydrogen at low temperatures could be due instead to the release of trapped hydrogen or small hydrocarbon gas molecules and the desorption of weakly bound surface hydrogen, rather than C-H bond breaking [76, 120]. These simulation results support the latter explanation. The simulations suggest that the decrease in thermal stability for a-C:H is caused by the weakening of C-C bonds with increasing hydrogen content, rather than rehybridization due to C-H bond breaking.

Finally, Section 5.8 shows that the mean activation energy depends strongly on the in- plane compressive stress. This is attributed to the increase in volume occupied by sp2 carbon compared to sp3 carbon. Assuming a value 1.5 GPa compressive stress for a-C:H:Si:O gives an activation energy of 2.81 ± 1.09 eV compared to a value of 0.96 ± 0.38 eV with ∼0 GPa compressive stress. This large effect of compressive stress is generally not considered inthe literature. For example, MD studies of the annealing behavior DLC do not report or control the stress [145]. Liquid quenching to form DLC is often carried out at constant volume which

127 results in a fixed density, but the resulting stress state can vary greatly [143, 145, 146]. A better approach is to produce structures with a specific amount of residual stress, while adjusting the initial volume to get approximately the desired density. The activation energy of 2.81 ± 1.09 eV at 1.5 GPa agrees well with the value of 3.0 ± 1.1 eV from experiments.

6.2. Future work

There are several promising directions to extend this thesis, both in terms of experiments and simulations, in order to address unanswered questions and further explore new ideas suggested by this work.

6.2.1. Tribometer testing in well controlled humidity

The tribological results shown in Chapter2 for both the NCD Technologies and DLN

360 a-C:H:Si:O were clearly sensitive the relative humidity, but the humidity could be much better controlled. Although dry environments with less than 5% RH could be produced using the gas blowing tribometer setup described in Section 2.2, the humidity in the humid air tests varied based on laboratory conditions (typically in the range of 25–45% RH). It would be useful and interesting to investigate the tribological behavior of these a-C:H:Si:O samples as a function of well controlled humidity that could be adjusted over a range of humid to dry air conditions in order to observe the transition in tribological behavior.

6.2.2. Tribometer testing with different balls

Two possible extensions to the tribometer experiments would be to vary size or material of the balls used. In terms of materials, all of the experiments performed in this thesis used steel ball bearings. Changing the counterface material could alter the strength of the adhesive interactions or change the direction of material transfer. For instance, previous work by Koshigan et al. [100] showed that in high vacuum the adhesive junctions broke in the steel side of the contact, changing the direction of material transfer, resulting in high friction.

128 By changing the ball material it would be possible to investigate the effect of different counterface material properties and surface forces on run-in and tribofilm formation.

Second, in the load dependent tribology results presented in Section 2.8, the ball diameter was constant, resulting in the both the contact area and contact pressure changing as the load was varied. The observed trends in wear track topography were due to a combination of these effects. By using balls with different diameters, it is possible to changethe contact pressure and area independently. In particular, contact pressure could be varied with a constant contact area or vice versa, allowing for a deconvolution of these two effects.

6.2.3. Localized mapping of mechanical and structural properties in wear tracks

The AFM friction images shown in Section 2.6 show clear contrast between different regions inside the wear tracks. These large changes in friction suggest local structural differences in the wear tracks. In order to understand the surface modifications occurringit would be interesting to map the local chemical composition and mechanical properties. This is challenging, since it requires highly surface sensitive analysis techniques with nanometer scale lateral resolution. However, a few promising techniques exist that could be used to probe this information.

First, localized mechanical property measurements could be made using a technique called PeakForce Quantitative Nanomechanical Mapping (QNM) [182]. This is an AFM- based technique available on the Dimension Icon AFM used in this thesis. For this work the AFM was operated in contact mode, where the tip slides over the surface at a fixed normal displacement / load. In addition to topography measurements, this allows lateral forces to be measured based on torsion of the cantilever. In a second mode of operation, known as tapping mode, the cantilever is oscillated near its resonant frequency (typically

100’s of kHz) as it scans the surface. In this case the tip is only in contact with the sample for a small fraction of the time, making friction forces negligible. However, the phase shift of the oscillating AFM tip is a function of the energy dissipated into the material, making it sensitive to local material properties [183]. Unfortunately, quantifying the phase image is

129 difficult and it can be challenging to interpret the data, since the phase image isafunction of several contributions, including mechanical properties, adhesion, capillary forces, and topography [183].

In PeakForce QNM the tip is oscillated over the surface, but at a much lower frequency than in tapping mode. This allows the cantilever normal force as a function of tip sample separation to be recorded for every cycle. These force curves can then be analyzed to determine the samples modulus, adhesion, and energy dissipation [182]. The main limitation of this technique is the specialized AFM tips need for use on hard materials. In order to accurately probe the mechanical properties, the cantilever needs to be able to indent the sample. This requires the use of extremely stiff single crystal diamond AFM tips[182].

In terms of localized spectroscopy, two options are tip-enhanced Raman spectroscopy

(TERS) and AFM infrared spectroscopy (AFM-IR). Optical spectroscopic techniques are diffraction limited, resulting in a lateral resolution on the order of 200–400 nmatbest[184].

However, in many cases these techniques actually have lateral resolution on the order of micrometers [185]. This makes them unsuitable for distinguishing nanoscale structural features in the wear tracks. TERS and AFM-IR are two techniques, however, that take advantage of an AFM tip in contact with the sample to achieve lateral resolution on the order of a few nanometers. In TERS, a metallic AFM tip (often silver coated silicon) is used to enhance the Raman signal within the nanoscale volume near the tip apex [184].

When the metallic tip is illuminated by the Raman laser surface plasmons are created at the tip apex which excite additional Raman scattering from the sample volume directly under the tip. This results in an intensity enhancement factor on the order of 103–106 from a nanometer scale area [184]. In AFM-IR, a tunable IR laser is used to excite the sample and produce photothermal expansion when tuned to an absorbing band. This expansion produces oscillations, with amplitude proportional to the absorption, in an AFM cantilever in contact with the surface. From this, an IR spectrum is produced from only the material near to the AFM tip [185]. Both of these techniques can simultaneously acquire AFM images, allowing the spectra to be correlated with topography. In addition, AFM-IR can

130 also map local sample stiffness based on shifts in the contact resonance of the cantilever oscillations [185].

6.2.4. Experimental investigation of the effect of stress on thermal stability

In Section 5.8, MD simulations showed a strong effect of stress on thermal stability, which had not been considered experimentally. Experiments investigating the influence of stress on thermal stability would be useful for validating simulation results and could have technological implications as another method to enhance thermal stability. Stress in thin films can be determined from the substrate curvature through Stoney’s Equation [186]:

2 Eshsκ σf = (6.1) 6hf (1 − νs) where σ is the in-plane residual stress, E is Young’s modulus, ν is Poisson’s ratio, h is

the thickness, and κ is the curvature. The subscripts f and s denote the film and sub-

strate. First, determining the residual stress in experimental samples with known atomic

composition and density would be useful for providing a well-defined basis for MD simu-

lations. Second, experimental investigation of the thermal stability for samples with the

same composition and density, but with different residual stress, would provide a method

to determine the effect of stress. Such samples could be produced by depositing filmswith

the same deposition parameters but vary the deposition time. This would produce films of

different thickness, and different stress, since the intrinsic stress in thin films isknownto

be thickness dependent [187].

6.2.5. MD simulations of thermal stability with oxygen or water atmosphere

All of the MD simulations presented in Chapter5 are based on vacuum annealing,

which was also true for the XPS and NEXAFS annealing performed by Mangolini et al.

[121]. However, in most applications, annealing would take place in an atmosphere. In this

case it is expected that oxygen or water vapor could have a significant effect on thermal

131 decomposition due to oxidation. XPS measurements by Mangolini et al. [188] on DLN 180 a-C:H:Si:O annealed in air up to 450 ◦C showed rapid oxidation and carbon volatilization in the near surface region. This resulted in the formation of a silica-like surface layer. In contrast a-C:H films rapidly erode under similar conditions. It is proposed by Mangolini et al. [188] that this surface layer acts as a barrier, preventing further reactions between oxygen / water and the underlying carbon phase.

In the simulations oxygen or water could be added at the free surfaces. Then annealing simulation could be performed to model the energetics of reactions between oxygen or water and a-C:H:Si:O or a-C:H. Such simulations could help to understand the reaction pathways by which film volatilization and the formation of this silica-like surface layer occurs.

6.2.6. Simulating the tribological behavior of a-C:H:Si:O

A final idea is to perform sliding simulations to probe the nanoscale tribology ofa-

C:H:Si:O. There are a number of papers using molecular dynamics simulations to study the tribology of DLCs, including some that make use of ReaxFF [58, 59, 155, 164]. These simulations have provided insight into atomistic mechanisms controlling the DLC tribology such as surface passivation and sp3 to sp2 conversion. MD simulations could be used to

investigate the nanotribological behavior a-C:H:Si:O. One possible approach is to simulate

sliding of a-C:H:Si:O against a silicon, diamond, or ta-C surface. These counterfaces are

all available AFM tip materials, which would allow for simulations and experiments of the

same single asperity contact (although on very different timescales). This could provide

insight into the atomistic mechanisms controlling the tribological behavior of a-C:H:Si:O at

the nano and macroscale. Another possibility is to study the behavior of a simpler system,

such as a-C:H:Si. These films are also available experimentally and have similar properties

to a-C:H:Si:O. One advantage of this system is that, by reducing the number of elements,

other potentials can be used. For example, the REBO potential has been parameterized for

C, H, and Si. In this case, REBO could be used if the speed of ReaxFF proves too limiting

for tribology simulations.

132 BIBLIOGRAPHY [1] Ali Erdemir and Christophe Donnet. Tribology of diamond-like carbon films: recent progress and future prospects. Journal of Physics D: Applied Physics, 39(18):R311– R327, September 2006. ISSN 0022-3727. doi: 10.1088/0022-3727/39/18/R01.

[2] Christophe Donnet and Ali Erdemir. Diamond-like carbon films: A historical overview. In Christophe Donnet and Ali Erdemir, editors, Tribology of Diamond- Like Carbon Films, chapter Intro., pages 1–12. Springer US, Boston, MA, 2008. ISBN 978-0-387-30264-5. doi: 10.1007/978-0-387-49891-1.

[3] S.V. Hainsworth and N.J. Uhure. Diamond-like carbon coatings for tribology: Produc- tion techniques, characterization methods and applications. International Materials Reviews, 52(3):153–174, 2007. doi: https://doi.org/10.1179/174328007X160272.

[4] R. Hauert. An overview on the tribological behavior of diamond-like carbon in technical and medical applications. Tribology International, 37(11-12 SPEC.ISS.):991–1003, 2004. ISSN 0301679X. doi: 10.1016/j.triboint.2004.07.017.

[5] A. Grill. Diamond-like carbon coatings as biocompatible materials—an overview. Diamond and Related Materials, 12(2):166–170, 2003. ISSN 09259635. doi: 10.1016/ S0925-9635(03)00018-9.

[6] Matthew Allen, Ben Myer, and Neil Rushton. In vitro and in vivo investigations into the biocompatibility of diamond-like carbon (dlc) coatings for orthopedic applications. Journal of Biomedical Materials Research, 58(3):319–328, 2001. ISSN 00219304. doi: 10.1002/1097-4636(2001)58:3⟨319::AID-JBM1024⟩3.0.CO;2-F.

[7] Andrea Carlo Ferrari. Diamond-like carbon for magnetic storage disks. Surface and Coatings Technology, 180-181:190–206, March 2004. ISSN 02578972. doi: 10.1016/j. surfcoat.2003.10.146.

[8] Cinzia Casiraghi, John Robertson, and Andrea C. Ferrari. Diamond-like carbon for data and beer storage. Materials Today, 10(1-2):44–53, 2007. ISSN 13697021. doi: 10.1016/S1369-7021(06)71791-6.

[9] JP Sullivan, T A Friedmann, and K Hjort. Diamond and amorphous carbon mems. MRS Bulletin, 26(April):309–311, 2001.

[10] H. Ronkainen and K. Holmberg. Environmental and thermal effects on the tribological performance of dlc coatings. In Christophe Donnet and Ali Erdemir, editors, Tribology of Diamond-Like Carbon Films, chapter 6, pages 155–200. Springer US, Boston, MA, 2008. ISBN 978-0-387-30264-5. doi: 10.1007/978-0-387-49891-1.

[11] N. Wang, K. Komvopoulos, Franck Rose, and B. Marchon. Structural stability of hydrogenated amorphous carbon overcoats used in heat-assisted magnetic recording investigated by rapid thermal annealing. Journal of Applied Physics, 113(8):083517, 2013. ISSN 00218979. doi: 10.1063/1.4792521.

133 [12] Julien Fontaine. Towards the use of diamond-like carbon solid lubricant coatings in vacuum and space environments. Proceedings of the Institution of Mechanical Engineers, Part J: Journal of Engineering Tribology, 222(8):1015–1029, December 2008. ISSN 1350-6501. doi: 10.1243/13506501JET323.

[13] M. Grischke, K. Bewilogua, K. Trojan, and H. Dimigen. Application-oriented modifications of deposition processes for diamond-like-carbon-based coatings. Surface and Coatings Technology, 74-75(PART 2):739–745, 1995. ISSN 02578972. doi: 10.1016/0257-8972(94)08201-4.

[14] C. Venkatraman, D. J. Kester, A. Goel, and D. Bray. Diamond-like nanocomposite coatings - a new class of materials. In Surface Modification Technologies ,IX 1996.

[15] C. Venkatraman, C. Brodbeck, and R. Lei. Tribological properties of diamond-like nanocomposite coatings at high temperatures. Surface and Coatings Technology, 115 (2-3):215–221, July 1999. ISSN 02578972. doi: 10.1016/S0257-8972(99)00241-8.

[16] D.J. Kester, C.L. Brodbeck, I.L. Singer, and A. Kyriakopoulos. Sliding wear behavior of diamond-like nanocomposite coatings. Surface and Coatings Technology, 113(3): 268–273, March 1999. ISSN 02578972. doi: 10.1016/S0257-8972(99)00003-1.

[17] T. W. Scharf, J. A. Ohlhausen, D. R. Tallant, and S. V. Prasad. Mechanisms of friction in diamondlike nanocomposite coatings. Journal of Applied Physics, 101(6): 063521, 2007. ISSN 00218979. doi: 10.1063/1.2711147.

[18] John Robertson. Diamond-like amorphous carbon. Materials Science and Engi- neering: R: Reports, 37(4-6):129–281, May 2002. ISSN 0927796X. doi: 10.1016/ S0927-796X(02)00005-0.

[19] John Robertson. Classification of diamond-like carbons. In Christophe Donnet and Ali Erdemir, editors, Tribology of Diamond-Like Carbon Films, chapter 1, pages 13–24. Springer US, Boston, MA, 2008. ISBN 978-0-387-30264-5. doi: 10.1007/978-0-387-49891-1.

[20] W. Jacob and W. M¨oller.On the structure of thin hydrocarbon films. Applied Physics Letters, 63(13):1771–1773, September 1993. ISSN 0003-6951. doi: 10.1063/1.110683.

[21] Sol Aisenberg and Ronald Chabot. Ion-beam deposition of thin films of diamondlike carbon. Journal of Applied Physics, 42(7):2953, 1971. ISSN 00218979. doi: 10.1063/ 1.1660654.

[22] Alfred Grill, V. Patel, and B. Meyerson. Tribological behavior of diamond-like carbon: effects of preparation conditions and annealing. Surface and Coatings Technology, 49 (1-3):530–536, December 1991. ISSN 02578972. doi: 10.1016/0257-8972(91)90112-A.

[23] Kazuhisa Miyoshi, Richard L.C. Wu, and Alan Garscadden. Friction and wear of diamond and diamondlike carbon coatings. Surface and Coatings Technology, 54-55: 428–434, January 1992. ISSN 02578972. doi: 10.1016/S0257-8972(07)80061-2.

134 [24] H. Ronkainen, J. Koskinen, A. Anttila, K. Holmberg, and J.-P. Hirvonen. Tribological characterisation of hard carbon films produced by the pulsed vacuum arc discharge method. Diamond and Related Materials, 1(5-6):639–643, April 1992. ISSN 09259635. doi: 10.1016/0925-9635(92)90182-N.

[25] Ali Erdemir, M. Switala, R. Wei, and P Wilbur. A tribological investigation of the graphite-to-diamond-like behavior of amorphous carbon films ion beam deposited on ceramic substrates. Surface and Coatings Technology, 50(1):17–23, January 1991. ISSN 02578972. doi: 10.1016/0257-8972(91)90187-2.

[26] Christophe Donnet, Michel Belin, JC Auge, JM Martin, Alfred Grill, and V. Patel. Tribochemistry of diamond-like carbon coatings in various environments. Surface and Coatings Technology, 69:626–631, 1994.

[27] Jason A. Bares, Anirudha V. Sumant, David S. Grierson, Robert W. Carpick, and Kumar Sridharan. Small amplitude reciprocating wear performance of diamond-like carbon films: dependence of film composition and counterface material. Tribology Letters, 27(1):79–88, April 2007. ISSN 1023-8883. doi: 10.1007/s11249-007-9209-x.

[28] D. R. McKenzie, D. Muller, and B. A. Pailthorpe. Compressive-stress-induced formation of thin-film tetrahedral amorphous carbon. Physical Review Letters, 67(6): 773–776, 1991. ISSN 00319007. doi: 10.1103/PhysRevLett.67.773.

[29] J. Robertson. Deposition mechanisms for promoting sp3 bonding in diamond-like carbon. Diamond and Related Materials, 2(5-7):984–989, 1993. ISSN 09259635. doi: 10.1016/0925-9635(93)90262-Z.

[30] Y. Lifshitz, S. R. Kasi, J. W. Rabalais, and W. Eckstein. Subplantation model for film growth from hyperthermal species. Physical Review B, 41(15):10468–10480, May 1990. ISSN 0163-1829. doi: 10.1103/PhysRevB.41.10468.

[31] D. R. McKenzie. Generation and applications of compressive stress induced by low energy ion beam bombardment. Journal of Vacuum Science & Technology B: Mi- croelectronics and Nanometer Structures, 11(5):1928, 1993. ISSN 0734211X. doi: 10.1116/1.586524.

[32] C. A. Davis. A simple model for the formation of compressive stress in thin films by ion bombardment. Thin Solid Films, 226(1):30–34, 1993. ISSN 00406090. doi: 10.1016/0040-6090(93)90201-Y.

[33] John Robertson. The deposition mechanism of diamond-like a-c and a-c: H. Dia- mond and Related Materials, 3(4-6):361–368, 1994. ISSN 09259635. doi: 10.1016/ 0925-9635(94)90186-4.

[34] Olivier Jarry, CédricJaoul, Pascal Tristant, ThérèseMerle-Méjean, Maggy Colas, Christelle Dublanche-Tixier, HélèneAgeorges, Claude Lory, and Jean-Marie Jacquet. Tribological behaviour of diamond-like carbon films used in automotive application: A comparison. Plasma Processes and Polymers, 6(S1):S478–S482, June 2009. ISSN 16128850. doi: 10.1002/ppap.200931007.

135 [35] N. V. Novikov, A. G. Gontar, S. I. Khandozhko, A. M. Kutsay, V. N. Tkach, V. Y. Gorokhov, G. M. Belitsky, and A. V. Vasin. Protective diamond-like coatings for optical materials and electronic devices. Diamond and Related Materials, 9(3-6):792– 795, 2000. [36] Bharat Bhushan and Vilas N. Koinkar. Microscale mechanical and tribological characterization of hard amorphous carbon coatings as thin as 5 nm for magnetic disks. Surface and Coatings Technology, 77:655–669, 1995. [37] T. W. Scharf, R. D. Ott, D. Yang, and J. A. Barnard. Structural and tribological characterization of protective amorphous diamond-like carbon and amorphous cn[sub x] overcoats for next generation hard disks. Journal of Applied Physics, 85(6):3142, 1999. ISSN 00218979. doi: 10.1063/1.369654. [38] Julien Fontaine, Christophe Donnet, and Ali Erdemir. Fundamentals of the tribology of dlc coatings. In Christophe Donnet and Ali Erdemir, editors, Tribology of Diamond- Like Carbon Films, chapter 5, pages 139–154. Springer US, Boston, MA, 2008. ISBN 978-0-387-30264-5. doi: 10.1007/978-0-387-49891-1. [39] Christophe Donnet and Alfred Grill. Friction control of diamond-like carbon coatings. Surface and Coatings Technology, 95:456–462, 1997. [40] Ali Erdemir. Genesis of superlow friction and wear in diamondlike carbon films. Tribology International, 37(11-12):1005–1012, November 2004. ISSN 0301679X. doi: 10.1016/j.triboint.2004.07.018. [41] Michael Moseler, Peter Gumbsch, Cinzia Casiraghi, Andrea Carlo Ferrari, and John Robertson. The ultrasmoothness of diamond-like carbon surfaces. Science (New York, N.Y.), 309(5740):1545–8, September 2005. ISSN 1095-9203. doi: 10.1126/science. 1114577. [42] Filippo Mangolini, Brandon A. Krick, Tevis D.B. Jacobs, Subarna R. Khanal, Frank Streller, J. Brandon McClimon, James Hilbert, Somuri V. Prasad, Thomas W. Scharf, James A. Ohlhausen, Jennifer R. Lukes, W. Gregory Sawyer, and Robert W. Carpick. Effect of silicon and oxygen dopants on the stability of hydrogenated amorphous carbon under harsh environmental conditions. Carbon, 130:127–136, April 2018. ISSN 00086223. doi: 10.1016/j.carbon.2017.12.096. [43] H. I. Kim, J. R. Lince, O. L. Eryilmaz, and Ali Erdemir. Environmental effects on the friction of hydrogenated dlc films. Tribology Letters, 21(1):51–56, January 2006. ISSN 1023-8883. doi: 10.1007/s11249-005-9008-1. [44] Christophe Donnet, Julien Fontaine, Alfred Grill, and T Le Mogne. The role of hydrogen on the friction mechanism of diamond-like carbon films. Tribology Letters, 9(3):137–142, 2001. [45] J. Andersson, R. A. Erck, and Ali Erdemir. Frictional behavior of diamond- like carbon films in vacuum and under varying water vapor pressure. Surface and Coatings Technology, 163-164:535–540, January 2003. ISSN 02578972. doi: 10.1016/S0257-8972(02)00617-5.

136 [46] Yue Qi, Erkan Konca, and Ahmet T. Alpas. Atmospheric effects on the adhesion and friction between non-hydrogenated diamond-like carbon (dlc) coating and aluminum – a first principles investigation. Surface Science, 600(15):2955–2965, August 2006. ISSN 00396028. doi: 10.1016/j.susc.2006.05.008. [47] Christophe Donnet, Thierry Le Mogne, L. Ponsonnet, Michel Belin, Alfred Grill, V. Patel, and C. Jahnes. The respective role of oxygen and water vapor on the tribology of hydrogenated diamond-like carbon coatings. Tribology Letters, 4:259– 265, 1998. ISSN 1573-2711. doi: 10.1023/A:1019140213257. [48] Liping Wang, Longchen Cui, Zhibin Lu, and Hui Zhou. Understanding the unusual friction behavior of hydrogen-free diamond-like carbon films in oxygen atmosphere by first-principles calculations. Carbon, 100:556–563, 2016. ISSN 00086223. doi: 10.1016/j.carbon.2016.01.060. [49] A. R. Konicek, D. S. Grierson, A. V. Sumant, T. A. Friedmann, J. P. Sullivan, P. U.P.A. Gilbert, W. G. Sawyer, and R. W. Carpick. Influence of surface passivation on the friction and wear behavior of ultrananocrystalline diamond and tetrahedral amorphous carbon thin films. Physical Review B - Condensed Matter and Materials Physics, 85(15):1–13, 2012. ISSN 10980121. doi: 10.1103/PhysRevB.85.155448. [50] Julien Fontaine, Thierry Le Mogne, J.L. Loubet, and Michel Belin. Achieving superlow friction with hydrogenated amorphous carbon: some key requirements. Thin Solid Films, 482(1-2):99–108, June 2005. ISSN 00406090. doi: 10.1016/j.tsf.2004.11.126. [51] Shu Wei Liu, Chen Hui Zhang, Eryilmaz Osman, Xin Chun Chen, Tian Bao Ma, Yan Zhong Hu, Jian Bin Luo, and Erdemir Ali. Influence of tribofilm on superlubricity of highly-hydrogenated amorphous carbon films in inert gaseous environments. Science China Technological Sciences, 59(12):1795–1803, 2016. ISSN 1862281X. doi: 10.1007/s11431-016-0078-6. [52] Juliene Fontaine, J. L. Loubet, T. Le Mogne, and A. Grill. Superlow friction of diamond-like carbon films: A relation to viscoplastic properties. Tribology Letters, 17 (4):709–714, 2004. ISSN 10238883. doi: 10.1007/s11249-004-8077-x. [53] Y. Liu, Ali Erdemir, and E.I. Meletis. An investigation of the relationship between graphitization and frictional behavior of dlc coatings. Surface and Coatings Technol- ogy, 86-87:564–568, December 1996. ISSN 02578972. doi: 10.1016/S0257-8972(96) 03057-5. [54] T Le Huu, H Zaidi, D Paulmier, and P Voumard. Transformation of sp 3 to sp 2 sites of diamond like carbon coatings during friction in vacuum and under water vapour environment. Thin Solid Films, 290-291:126–130, 1996. [55] Y. Liu, Ali Erdemir, and E.I. Meletis. A study of the wear mechanism of diamond- like carbon films. Surface and Coatings Technology, 82(1-2):48–56, July 1996. ISSN 02578972. doi: 10.1016/0257-8972(95)02623-1. [56] Y. Liu and E.I. Meletis. Evidence of graphitization of diamond-like carbon films during sliding wear. Journal of Materials Science, 32:3491–3495, 1997.

137 [57] F. Rabbani. Phenomenological evidence for the wear-induced graphitization model of amorphous hydrogenated carbon coatings. Surface and Coatings Technology, 184 (2-3):194–207, 2004. ISSN 02578972. doi: 10.1016/j.surfcoat.2003.11.006.

[58] Lars Pastewka, Stefan Moser, Michael Moseler, B Blug, S Meier, T Hollstein, and Peter Gumbsch. The running-in of amorphous hydrocarbon tribocoatings: a comparison between experiment and molecular dynamics simulations. International Jour- nal of Materials Research, 99(10):1136–1143, October 2008. ISSN 1862-5282. doi: 10.3139/146.101747.

[59] Lars Pastewka, Stefan Moser, and Michael Moseler. Atomistic insights into the running-in, lubrication, and failure of hydrogenated diamond-like carbon coatings. Tribology Letters, 39(1):49–61, January 2010. ISSN 1023-8883. doi: 10.1007/ s11249-009-9566-8.

[60] Pedro A. Romero, Lars Pastewka, Julian Von Lautz, and Michael Moseler. Surface passivation and boundary lubrication of self-mated tetrahedral amorphous carbon asperities under extreme tribological conditions. Friction, 2(2):193–208, 2014. ISSN 22237704. doi: 10.1007/s40544-014-0057-z.

[61] Tim Kunze, Matthias Posselt, Sibylle Gemming, Gotthard Seifert, Andrew R. Kon- icek, Robert W. Carpick, Lars Pastewka, and Michael Moseler. Wear, plasticity, and rehybridization in tetrahedral amorphous carbon. Tribology Letters, 53(1):119–126, 2014. ISSN 10238883. doi: 10.1007/s11249-013-0250-7.

[62] Praveena Manimunda, Ala’ Al-Azizi, Seong H. Kim, and Richard R. Chromik. Shear- induced structural changes and origin of ultralow friction of hydrogenated diamond- like carbon (dlc) in dry environment. ACS Applied Materials and Interfaces, 9(19): 16704–16714, 2017. ISSN 19448252. doi: 10.1021/acsami.7b03360.

[63] Robert H. Savage. Graphite lubrication. Journal of Applied Physics, 19(1):1–10, 1948. ISSN 00218979. doi: 10.1063/1.1697867.

[64] Bing K. Yen, Birgit E. Schwickert, and Michael F. Toney. Origin of low-friction behavior in graphite investigated by surface x-ray diffraction. Applied Physics Letters, 84(23):4702–4704, 2004. ISSN 00036951. doi: 10.1063/1.1760597.

[65] R. F. Deacon and J. F. Goodman. Lubrication by lamellar solids. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 243(1235): 464–482, February 1958. ISSN 1364-5021. doi: 10.1098/rspa.1958.0013.

[66] Vasilios Georgakilas, Jitendra N. Tiwari, K. Christian Kemp, Jason A. Perman, Athanasios B. Bourlinos, Kwang S. Kim, and Radek Zboril. Noncovalent functional- ization of graphene and graphene oxide for energy materials, biosensing, catalytic, and biomedical applications. Chemical Reviews, 116(9):5464–5519, 2016. ISSN 15206890. doi: 10.1021/acs.chemrev.5b00620.

[67] J P Sullivan, T A Friedmann, and AG Baca. Stress relaxation and thermal evolution of film properties in amorphous carbon. Journal of Electronic Materials, 26(9), 1997.

138 [68] D. S. Grierson, A. V. Sumant, A. R. Konicek, T. A. Friedmann, J. P. Sullivan, and R. W. Carpick. Thermal stability and rehybridization of carbon bonding in tetrahedral amorphous carbon. Journal of Applied Physics, 107(3), 2010. ISSN 00218979. doi: 10.1063/1.3284087. [69] Hongxuan Li, Tao Xu, Chengbing Wang, Jianmin Chen, Huidi Zhou, and Huiwen Liu. Annealing effect on the structure, mechanical and tribological properties ofhy- drogenated diamond-like carbon films. Thin Solid Films, 515(4):2153–2160, December 2006. ISSN 00406090. doi: 10.1016/j.tsf.2006.04.018. [70] Weng-Jin Wu and Min-Hsiung Hon. Thermal stability of diamond-like carbon films with added silicon. Surface and Coatings Technology, 111(April 1998):134–140, 1999. [71] Y. S. Zou, Y. F. Wu, R. F. Huang, C. Sun, and L. S. Wen. Mechanical properties and thermal stability of nitrogen incorporated diamond-like carbon films. Vacuum, 83(11):1406–1410, July 2009. ISSN 0042207X. doi: 10.1016/j.vacuum.2009.04.072. [72] Yong Seob Park and Byungyou Hong. The effect of thermal annealing on the structural and mechanical properties of a-c:h thin films prepared by the cfubm magnetron sputtering method. Journal of Non-Crystalline Solids, 354(33):3980–3983, September 2008. ISSN 00223093. doi: 10.1016/j.jnoncrysol.2008.05.043. [73] L.H. Zhang, J.P. Wang, and H. Gong. Laser and thermal induced micro-structural changes and decomposition of hydrogenated carbon films. Journal of Physics: Con- densed Matter, 13, 2001. [74] Wei Long Quan, Hong Xuan Li, Fei Zhao, Li Ji, Wen Du, Hui Di Zhou, and Jian Min Chen. Molecular dynamic simulation on graphitization and dehydrogenization of hydrogenated carbon films in vacuum. Chinese Physics Letters, 27(7):2–4, 2010. ISSN 0256307X. doi: 10.1088/0256-307X/27/7/078103. [75] Da-Yung Wang, Chi-Lung Chang, and Wei-Yu Ho. Oxidation behavior of diamond- like carbon films. Surface and Coatings Technology, 120-121:138–144, 1999. ISSN 02578972. doi: 10.1016/S0257-8972(99)00350-3. [76] Franck Rose, Na Wang, Robert Smith, Qi-Fan Xiao, Hiroshi Inaba, Toru Matsumura, Yoko Saito, Hiroyuki Matsumoto, Qing Dai, Bruno Marchon, Filippo Mangolini, and Robert W. Carpick. Complete characterization by raman spectroscopy of the structural properties of thin hydrogenated diamond-like carbon films exposed to rapid thermal annealing. Journal of Applied Physics, 116(12):123516, 2014. ISSN 0021- 8979. doi: 10.1063/1.4896838. [77] By Mark H Kryder, Edward C Gage, Terry W Mcdaniel, William A Challener, Robert E Rottmayer, Ganping Ju, Yiao-tee Hsia, and M Fatih Erden. Heat assisted magnetic recording. Proceedings of the IEEE, 96(11):1810–1835, 2008. [78] Wei Peng, Yiao-Tee Hsia, Kursat Sendur, and Terry McDaniel. Thermo-magneto- mechanical analysis of head–disk interface in heat assisted magnetic recording. Tri- bology International, 38(6-7):588–593, June 2005. ISSN 0301679X. doi: 10.1016/j. triboint.2005.01.040.

139 [79] K. J. Schoen, J. M. Woodall, and A. Goel. Electrical properties of metal-diamond- like-nanocomposite (me-dln) contacts to 6h sic. Journal of electronic, 25(3):193–197, 1997.

[80] Soon Eng Ong, Sam Zhang, Hejun Du, and Deen Sun. Relationship between bonding structure and mechanical properties of amorphous carbon containing silicon. Diamond and Related Materials, 16(8):1628–1635, 2007. ISSN 09259635. doi: 10.1016/j.diamond.2007.02.009.

[81] Junjun Wang, Jibin Pu, Guangan Zhang, and Liping Wang. Tailoring the structure and property of silicon-doped diamond-like carbon films by controlling the silicon content. Surface and Coatings Technology, 235:326–332, 2013. ISSN 02578972. doi: 10.1016/j.surfcoat.2013.07.061.

[82] Xiao-Wei Li, Min-Woong Joe, Ai-Ying Wang, and Kwang-Ryeol Lee. Stress reduction of diamond-like carbon by si incorporation: A molecular dynamics study. Surface and Coatings Technology, pages 4–7, June 2012. ISSN 02578972. doi: 10.1016/j.surfcoat. 2012.06.031.

[83] A. A. Ogwu, R. W. Lamberton, S. Morley, P. D. Maguire, and J. A. McLaughlin. Characterisation of thermally annealed diamond like carbon (dlc) and silicon modifed dlc films by raman spectroscopy. Film, 269:335–344, 1999.

[84] Xinchun Chen, Takahisa Kato, and Masataka Nosaka. Origin of superlubricity in a-c:h:si films: A relation to film bonding structure and environmental molecular characteristic. ACS Applied Materials and Interfaces, 6(16):13389–13405, 2014. ISSN 19448252. doi: 10.1021/am502416w.

[85] Seiji Kajita and M. C. Righi. A fundamental mechanism for carbon-film lubricity identified by means of ab initio molecular dynamics. Carbon, 103:193–199, 2016. ISSN 00086223. doi: 10.1016/j.carbon.2016.02.078.

[86] Seiji Kajita and M. C. Righi. Insigths into the tribochemistry of silicon-doped carbon- based films by ab initio analysis of water-surface interactions. Tribology Letters, 61 (2):1–7, 2016. ISSN 10238883. doi: 10.1007/s11249-015-0631-1.

[87] Hae Suk Jung and Hyung Ho Park. Determination of local bonding configuration and structural modification in amorphous carbon with silicon incorporation. Diamond and Related Materials, 12(8):1373–1377, 2003. ISSN 09259635. doi: 10.1016/S0925-9635(03)00110-9.

[88] Sar¯unasMeˇskinisandˇ Asta Tamuleviˇcien˙e. Structure , properties and applications of diamond like nanocomposite ( sio x containing dlc ) films : A review. Materials Science, 17(4):44–46, 2011.

[89] J.C. Damasceno and S.S. Camargo. Plasma deposition and characterization of silicon oxide-containing diamond-like carbon films obtained from ch4:sih4:o2 gas mixtures. Thin Solid Films, 516(8):1890–1897, February 2008. ISSN 00406090. doi: 10.1016/j. tsf.2007.10.119.

140 [90] V.F. Dorfman. Diamond-like nanocomposites (dln). Thin Solid Films, 212(1-2):267– 273, May 1992. ISSN 00406090. doi: 10.1016/0040-6090(92)90532-G.

[91] D Neerinck, P Persoone, M Sercu, A Goel, D Kester, and D Bray. Diamond-like nanocomposite coatings (a-c:h/a-si:o) for tribological applications. Diamond and Re- lated Materials, 7(2-5):468–471, 1998. ISSN 09259635. doi: 10.1016/S0925-9635(97) 00201-X.

[92] T. Das, D. Ghosh, T. K. Bhattacharyya, and T. K. Maiti. Biocompatibility of diamond-like nanocomposite thin films. Journal of Materials Science: Materials in Medicine, 18(3):493–500, 2007. ISSN 09574530. doi: 10.1007/s10856-007-2009-x.

[93] T S Santra, T. K. Bhattacharyya, F G Tseng, and T. K. Barik. Diamond-like nanocomposite (dln) films for microelectro-mechanical system (mems). In Interna- tional Symposium on Devices MEMS, Intelligent Systems & Communication, pages 6–9, 2011.

[94] L.K. Randeniya, a. Bendavid, P.J. Martin, M.S. Amin, and E.W. Preston. Molecular structure of siox-incorporated diamond-like carbon films; evidence for phase segregation. Diamond and Related Materials, 18(9):1167–1173, September 2009. ISSN 09259635. doi: 10.1016/j.diamond.2009.03.004.

[95] T. S. Santra, C. H. Liu, T. K. Bhattacharyya, P. Patel, and T. K. Barik. Characteri- zation of diamond-like nanocomposite thin films grown by plasma enhanced chemical vapor deposition. Journal of Applied Physics, 107(12):124320, 2010. ISSN 00218979. doi: 10.1063/1.3415548.

[96] Won Jae Yang, Tohru Sekino, Kwang Bo Shim, Koichi Niihara, and Keun Ho Auh. Microstructure and tribological properties of siox/dlc films grown by pecvd. Surface and Coatings Technology, 194(1):128–135, 2005. ISSN 02578972. doi: 10.1016/j. surfcoat.2004.05.023.

[97] Won Jae Yang, Yong-Ho Choa, Tohru Sekino, Kwang Bo Shim, Koichi Niihara, and Keun Ho Auh. Thermal stability evaluation of diamond-like nanocomposite coatings. Thin Solid Films, 434(1-2):49–54, June 2003. ISSN 00406090. doi: 10.1016/S0040-6090(03)00466-8.

[98] T. S. Santra, T. K. Bhattacharyya, P. Patel, F. G. Tseng, and T. K. Barik. Structural and tribological properties of diamond-like nanocomposite thin films. Surface and Coatings Technology, 206(2-3):228–233, 2011. ISSN 02578972. doi: 10.1016/j.surfcoat. 2011.06.057.

[99] D. Neerinck, P. Persoone, M. Sercu, A. Goel, C. Venkatraman, D. Kester, C. Halter, P. Swab, and D. Bray. Diamond-like nanocomposite coatings for low-wear and low- friction applications in humid environments. Thin Solid Films, 317(1-2):402–404, 1998. ISSN 00406090. doi: 10.1016/S0040-6090(97)00631-7.

[100] K. D. Koshigan, F. Mangolini, J. B. Mcclimon, B. Vacher, S. Bec, R. W. Carpick, and J. Fontaine. Understanding the hydrogen and oxygen gas pressure dependence

141 of the tribological properties of silicon oxide-doped hydrogenated amorphous carbon coatings. Carbon, 93(November):851–860, 2015. ISSN 00086223. doi: 10.1016/j. carbon.2015.06.004.

[101] Robert W. Carpick and Miquel Salmeron. Scratching the surface: Fundamental investigations of tribology with atomic force microscopy. Chemical reviews, 97(4):1163– 1194, June 1997. ISSN 1520-6890.

[102] Mathew Mate. Tribology on the Small Scale. Oxford University Press, New York, 2008.

[103] E. Meyer. Atomic force microscopy. Progress in Surface Science, 41(1):3–49, 1992.

[104] John E. Sader, James W.M. Chon, and Paul Mulvaney. Calibration of rectangular atomic force microscope cantilevers. Review of Scientific Instruments, 70(10):3967– 3969, 1999. ISSN 00346748. doi: 10.1063/1.1150021.

[105] Min-Seok Kim, Jae-Hyuk Choi, Jong-Ho Kim, and Yon-Kyu Park. Accurate determination of spring constant of atomic force microscope cantilevers and comparison with other methods. Measurement, 43(4):520–526, May 2010. ISSN 02632241. doi: 10.1016/j.measurement.2009.12.020.

[106] Erin E. Flater, George E. Zacharakis-Jutz, Braulio G. Dumba, Isaac a. White, and Charles a. Clifford. Towards easy and reliable afm tip shape determination using blind tip reconstruction. Ultramicroscopy, 146:130–143, 2014. ISSN 03043991. doi: 10.1016/j.ultramic.2013.06.022.

[107] P. Hariharan. Optical Interferometry. Academic Press, 2nd edition, 2003.

[108] H. Hertz. On the contact of elastic solids. In D.E. Jones and G.A. Schott, editors, Miscellaneous Papers by H. Hertz, chapter 5, pages 146–163. Macmillan, London, 1896.

[109] K.L. Johnson. Contact Mechanics. Cambridge University Press, Cambridge, 1985.

[110] Gwidon W. Stachowiak and Andrew W. Batchelor. Engineering Tribology. Elsevier, Oxford, 3rd edition, 2005.

[111] Daniel Maugis. Adhesion of spheres: The jkr-dmt transition using a dugdale model. Journal of Colloid And Interface Science, 150(1):243–269, 1992. ISSN 00219797. doi: 10.1016/0021-9797(92)90285-T.

[112] J. Brandon McClimon, A.C. Lang, Z. Milne, N. Garabedian, A.C. Moore, James Hilbert, Filippo Mangolini, Jennifer R. Lukes, D.L. Burris, M.L. Taheri, Julien Fontaine, and Robert W. Carpick. Mechanical and spectroscopic characterization of tribofilms formed during macroscale tribometry of silicon oxide-doped hydrogenated amorphous carbon in controlled environments (to be submitted). Tribology Letters, 2018.

142 [113] Andrea Carlo Ferrari and John Robertson. Interpretation of raman spectra of disordered and amorphous carbon. Physical Review B, 61(20):14095–14107, May 2000. ISSN 0163-1829. doi: 10.1103/PhysRevB.61.14095.

[114] D. Nesheva, C. Raptis, A. Perakis, I. Bineva, Z. Aneva, Z. Levi, S. Alexandrova, and H. Hofmeister. Raman scattering and photoluminescence from si nanoparticles in annealed sioxthin films. Journal of Applied Physics, 92(8):4678–4683, 2002. ISSN 00218979. doi: 10.1063/1.1504176.

[115] J. F. Archard. Contact and rubbing of flat surfaces. Journal of Applied Physics, 24 (8):981–988, 1953. ISSN 00218979. doi: 10.1063/1.1721448.

[116] A. Ramalho and J. C. Miranda. The relationship between wear and dissipated energy in sliding systems. Wear, 260(4-5):361–367, 2006. ISSN 00431648. doi: 10.1016/j. wear.2005.02.121.

[117] S. Fouvry and Ph Kapsa. An energy description of hard coating wear mechanisms. Surface and Coatings Technology, 138(2-3):141–148, 2001. ISSN 02578972. doi: 10. 1016/S0257-8972(00)01161-0.

[118] St´ephanieLafon-Placette, Julien Fontaine, Maria Isabel De Barros Bouchet, and Christophe Heau. Critical role of a metallic counterpart on the tribochemical wear of ta-c coatings in base oil. Wear, 402-403(October 2017):91–99, 2018. ISSN 00431648. doi: 10.1016/j.wear.2018.02.006.

[119] Filippo Mangolini, Franck Rose, James Hilbert, and Robert W. Carpick. Thermally induced evolution of hydrogenated amorphous carbon. Applied Physics Letters, 103 (16):161605, 2013. ISSN 00036951. doi: 10.1063/1.4826100.

[120] Jing Peng, Anastasiia Sergiienko, Filippo Mangolini, Phillip E. Stallworth, Steve Greenbaum, and Robert W. Carpick. Solid state magnetic resonance investigation of the thermally-induced structural evolution of silicon oxide-doped hydrogenated amorphous carbon. Carbon, 105:163–175, 2016. ISSN 00086223. doi: 10.1016/j.carbon.2016.04.021.

[121] Filippo Mangolini, James Hilbert, J. Brandon McClimon, Jennifer R Lukes, and Robert W Carpick. Thermally induced structural evolution of silicon- and oxygen- containing hydrogenated amorphous carbon: A combined spectroscopic and molecular dynamics simulation investigation. Langmuir, 34(9):2989–2995, March 2018. ISSN 0743-7463. doi: 10.1021/acs.langmuir.7b04266.

[122] G. Irmer and A. Dorner-Reisel. Micro-raman studies on dlc coatings. Advanced Engineering Materials, 7(8):694–705, August 2005. ISSN 1438-1656. doi: 10.1002/ adem.200500006.

[123] N. Khamnualthong, K. Siangchaew, and P. Limsuwan. Thermal stability evaluation of diamond-like carbon for magnetic recording head application using raman spectroscopy. Procedia Engineering, 32(Dlc):888–894, January 2012. ISSN 18777058. doi: 10.1016/j.proeng.2012.02.028.

143 [124] Andrea Carlo Ferrari and John Robertson. Raman spectroscopy of amorphous, nanos- tructured, diamond-like carbon, and nanodiamond. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences, 362(1824):2477–512, November 2004. ISSN 1364-503X. doi: 10.1098/rsta.2004.1452.

[125] Andrea Carlo Ferrari. Non-destructive characterisation of carbon films. In Christophe Donnet and Ali Erdemir, editors, Tribology of Diamond-Like Carbon Films, chapter 2, pages 25–82. Springer US, Boston, MA, 2008. ISBN 978-0-387-30264-5. doi: 10.1007/ 978-0-387-49891-1.

[126] Jack L. Koenig. Raman spectroscopy of polymers. In Spectroscopy of Polymers, chapter 5, pages 207–254. Elsevier, Amsterdam, 2 edition, 1999. doi: 10.1016/ B978-044410031-3/50005-0.

[127] W. G. Cui, Q. B. Lai, L. Zhang, and F. M. Wang. Quantitative measurements of sp3content in dlc films with raman spectroscopy. Surface and Coatings Technology, 205(7):1995–1999, 2010. ISSN 02578972. doi: 10.1016/j.surfcoat.2010.08.093.

[128] F. Mangolini, J. Ahlund,˚ G. E. Wabiszewski, V. P. Adiga, P. Egberts, F. Streller, K. Backlund, P. G. Karlsson, B. Wannberg, and R. W. Carpick. Angle-resolved environmental x-ray photoelectron spectroscopy: A new laboratory setup for photoe- mission studies at pressures up to 0.4 torr. Review of Scientific Instruments, 83(9): 0–10, 2012. ISSN 00346748. doi: 10.1063/1.4754127.

[129] Brent A. Nelson and William P. King. Thermal calibration of heated silicon atomic force microscope cantilevers. Sensors and Actuators A, 140:607–610, 2007. ISSN 09244247. doi: 10.1109/SENSOR.2007.4300203.

[130] Jungchul Lee, T Beechem, T Wright, Brent Nelson, S Graham, and William King. Electrical, thermal, and mechanical characterization of silicon microcantilever heaters. Journal of Microelectromechanical Systems, 15(6):1644–1655, 2006. ISSN 1057-7157. doi: 10.1109/JMEMS.2006.886020.

[131] Thomas Beechem, Samuel Graham, Sean P. Kearney, Leslie M. Phinney, and Justin R. Serrano. Invited article: Simultaneous mapping of temperature and stress in microde- vices using micro-raman spectroscopy. Review of Scientific Instruments, 78(6):061301, June 2007. ISSN 0034-6748. doi: 10.1063/1.2738946.

[132] William P King, Bikramjit Bhatia, Jonathan R Felts, Hoe Joon Kim, and Beomjin Kwon. Heated atomic force microscope cantilevers and their applications. In Gang Chen, Vish Prasad, and Yogesh Jaluria, editors, Annual Review of Heat Transfer, volume XVI, chapter 10, pages 287–326. Begell House, Inc., 2013.

[133] Brent A. Nelson and William P. King. Modeling and simulation of the interface temperature between a heated silicon tip and a substrate. Nanoscale and Microscale Thermophysical Engineering, 12(1):98–115, 2008. ISSN 1556-7265. doi: 10.1080/ 15567260701866769.

144 [134] M. Shamsa, W. L. Liu, A. A. Balandin, C. Casiraghi, W. I. Milne, and A. C. Ferrari. Thermal conductivity of diamond-like carbon films. Applied Physics Letters, 89(16): 138–140, 2006. ISSN 00036951. doi: 10.1063/1.2362601.

[135] Andrew J. Bullen, Keith E. O’Hara, David G. Cahill, Othon Monteiro, and Achim von Keudell. Thermal conductivity of amorphous carbon thin films. Journal of Applied Physics, 6317(2000):4, 2000. ISSN 00218979. doi: 10.1063/1.1314301.

[136] George Chen, Ping Hui, and Shi Xu. Thermal conduction in metalized tetrahedral amorphous carbon (ta-c) films on silicon. Thin Solid Films, 366(1-2):95–99, 2000. ISSN 00406090. doi: 10.1016/S0040-6090(99)01097-4.

[137] S. Callard, G. Tallarida, a. Borghesi, and L. Zanotti. Thermal conductivity of sio2 films by scanning thermal microscopy. Journal of Non-Crystalline Solids, 245:203–209, 1999. ISSN 00223093. doi: 10.1016/S0022-3093(98)00863-1.

[138] O. W. K¨ading,H. Skurk, and K. E. Goodson. Thermal conduction in metallized silicon-dioxide layers on silicon. Applied Physics Letters, 65(13):1629–1631, 1994. ISSN 00036951. doi: 10.1063/1.112933.

[139] Facundo Ruiz, W. D. Sun, and F. H. Pollak. Determination of the thermal conductivity of diamond-like nanocomposite films using a scanning thermal microscope. Applied physics letters, 73(13):1802–1804, 1998.

[140] A Majumdar. Scanning thermal microscopy. Annual Review of Materials Science, 29: 505–85, 1999.

[141] St´ephaneLef`evre, Sebastian Volz, Jean Bernard Saulnier, Catherine Fuentes, and Nathalie Trannoy. Thermal conductivity calibration for hot wire based dc scanning thermal microscopy. Review of Scientific Instruments, 74(4):2418–2423, 2003. ISSN 00346748. doi: 10.1063/1.1544078.

[142] H. Fischer. Quantitative determination of heat conductivities by scanning thermal microscopy. Thermochimica Acta, 425(1-2):69–74, 2005. ISSN 00406031. doi: 10. 1016/j.tca.2004.06.005.

[143] N. A. Marks, N. C. Cooper, D. R. McKenzie, D. G. McCulloch, P. Bath, and S. P. Russo. Comparison of density-functional, tight-binding, and empirical methods for the simulation of amorphous carbon. Physical Review B, 65(7):075411, 2002. ISSN 0163-1829. doi: 10.1103/PhysRevB.65.075411.

[144] Nigel Marks. Amorphous carbon and related materials. In Luciano Colombo and Annalisa Fasolino, editors, Computer-Based Modeling of Novel Carbon Systems and Their Properties: Beyond Nanotubes, pages 129–169. Springer, 2010. ISBN 9781402097171.

[145] Carla de Tomas, Irene Suarez-Martinez, and Nigel A. Marks. Graphitization of amorphous carbons: A comparative study of interatomic potentials. Carbon, 109:681–693, 2016. ISSN 00086223. doi: 10.1016/j.carbon.2016.08.024.

145 [146] Longqiu Li, Ming Xu, Wenping Song, Andrey Ovcharenko, Guangyu Zhang, and Ding Jia. The effect of empirical potential functions on modeling of amorphous carbon using molecular dynamics method. Applied Surface Science, 286:287–297, 2013. ISSN 01694332. doi: 10.1016/j.apsusc.2013.09.073.

[147] Lars Pastewka, Pablo Pou, Rub´enP´erez, Peter Gumbsch, and Michael Moseler. Describing bond-breaking processes by reactive potentials: Importance of an environment-dependent interaction range. Physical Review B, 78(16):161402, October 2008. ISSN 1098-0121. doi: 10.1103/PhysRevB.78.161402.

[148] P. J. Fallon, V. S. Veerasamy, C. A. Davis, J. Robertson, G. A J Amaratunga, W. I. Milne, and J. Koskinen. Properties of filtered-ion-beam-deposited diamondlike carbon as a function of ion energy. Physical Review B, 48(7):4777–4782, 1993. ISSN 01631829. doi: 10.1103/PhysRevB.48.4777.

[149] A. LiBassi, A.C. Ferrari, V. Stolojan, B.K. Tanner, John Robertson, and L.M. Brown. Density, sp 3 content and internal layering of dlc films by x-ray reflectivity and electron energy loss spectroscopy. Diamond and Related Materials, 9(3-6):771–776, April 2000. ISSN 09259635. doi: 10.1016/S0925-9635(99)00233-2.

[150] J. Schwan, S. Ulrich, H. Roth, H. Ehrhardt, S. R.P. Silva, J. Robertson, R. Samlenski, and R. Brenn. Tetrahedral amorphous carbon films prepared by magnetron sputtering and dc ion plating. Journal of Applied Physics, 79(3):1416–1422, 1996. ISSN 00218979. doi: 10.1063/1.360979.

[151] Jean-Michel Martin, Maria-Isabel De Barros Bouchet, Christine Matta, Qing Zhang, William A. Goddard, Sachiko Okuda, and Takumaru Sagawa. Gas-phase lubrication of ta-c by glycerol and hydrogen peroxide. experimental and computer modeling. The Journal of Physical Chemistry C, 114(11):5003–5011, March 2010. ISSN 1932-7447. doi: 10.1021/jp909940j.

[152] Qing Zhang, Adri C. T. Van Duin, Tahir Cagin, William A. Goddard, J. M. Martin, M. I. De Barros-Bouchet, T. Le Mogne, Takumaru Sagawa, and Sachiko Okuda. The structure and sliding friction of diamond-like carbon surfaces from molecular dynamics simulations. In STLE/ASME 2006 International Joint Tribology Conference Part B, pages 1043–1052. ASME, 2006. ISBN 0-7918-4259-2. doi: 10.1115/IJTC2006-12214.

[153] GT Gao, Paul T Mikulski, and Judith A Harrison. Molecular-scale tribology of amorphous carbon coatings: effects of film thickness, adhesion, and long-range interactions. Journal of the American Chemical Society, 124(24):7202–9, June 2002. ISSN 0002- 7863.

[154] GT Gao, Paul T Mikulski, GM Chateauneuf, and Judith A Harrison. The effects of film structure and surface hydrogen on the properties of amorphous carbon films. Journal of Physical Chemistry B, 107(40):11082–11090, 2003.

[155] J. David Schall, Guangtu Gao, and Judith A. Harrison. Effects of adhesion and transfer film formation on the tribology of self-mated dlc contacts. Journal of Physical Chemistry C, 114(12):5321–5330, 2010. ISSN 19327447. doi: 10.1021/jp904871t.

146 [156] David A Newsome, Debasis Sengupta, Hosein Foroutan, Michael F Russo, and Adri C.T. van Duin. Oxidation of silicon carbide by o 2 and h 2 o : A reaxff reactive molecular dynamics study , part i. The Journal of Physical Chemistry C, 18 (116):16111–16121, 2012.

[157] Adri C.T. van Duin, Siddharth Dasgupta, Francois Lorant, and William A. Goddard. Reaxff: A reactive force field for hydrocarbons. The Journal of Physical Chemistry A, 105(41):9396–9409, October 2001. ISSN 1089-5639. doi: 10.1021/jp004368u.

[158] Kimberly Chenoweth, Adri C.T. van Duin, and William a Goddard. Reaxff reactive force field for molecular dynamics simulations of hydrocarbon oxidation. The journal of physical chemistry. A, 112(5):1040–53, February 2008. ISSN 1520-5215. doi: 10. 1021/jp709896w.

[159] Yun Kyung Shin, Tzu-Ray Shan, Tao Liang, Mark J. Noordhoek, Susan B. Sinnott, Adri C.T. van Duin, and Simon R. Phillpot. Variable charge many-body interatomic potentials. MRS Bulletin, 37(05):504–512, May 2012. ISSN 0883-7694. doi: 10.1557/ mrs.2012.95.

[160] ACT Van Duin, Alejandro Strachan, Shannon Stewman, Qingsong Zhang, Xin Xu, and William A. Goddard. Reaxff sio reactive force field for silicon and silicon oxide systems. The Journal of Physical Chemistry A, 107:3803–3811, 2003.

[161] Mo Zheng, Xiaoxia Li, and Li Guo. Algorithms of gpu-enabled reactive force field (reaxff) molecular dynamics. Journal of molecular graphics and modelling, 41C:1–11, February 2013. ISSN 1873-4243. doi: 10.1016/j.jmgm.2013.02.001.

[162] Adri C.T. van Duin. Reactive force fields : concepts of reaxff. Technical report, California Institute of Technology, 2008.

[163] Steve Plimpton. Interatomic potential comparisons, 2012. URL https://lammps. sandia.gov/bench.html{#}potentials.

[164] M I De Barros Bouchet, C Matta, Th Le-Mogne, J Michel Martin, Q Zhang, W God- dard, M Kano, Y Mabuchi, and J Ye. Superlubricity mechanism of diamond-like carbon with glycerol. coupling of experimental and simulation studies. Journal of Physics: Conference Series, 89:012003, November 2007. ISSN 1742-6596. doi: 10.1088/1742-6596/89/1/012003.

[165] Giulia Galli, Richard M. Martin, Roberto Car, and Michele Parrinello. Ab initio calculation of properties of carbon in the amorphous and liquid states. Physical Review B, 42(12):7470–7482, 1990. ISSN 01631829. doi: 10.1103/PhysRevB.42.7470.

[166] NA Marks, MF Cover, and C Kocer. The importance of rare events in thin film deposition: a molecular dynamics study of tetrahedral amorphous carbon. Molecular Simulation, 32(15):1271–1277, 2006.

[167] TianBao Ma, Yuan-Zhong Hu, Hui Wang, and Xin Li. Microstructural and stress properties of ultrathin diamondlike carbon films during growth: Molecular dynamics

147 simulations. Physical Review B, 75(3):035425, January 2007. ISSN 1098-0121. doi: 10.1103/PhysRevB.75.035425.

[168] Steve Plimpton. Fast parallel algorithms for short – range molecular dynamics. Jour- nal of Computational Physics, 117(June 1994):1–19, 1995. ISSN 00219991. doi: 10.1006/jcph.1995.1039.

[169] H. M. Aktulga, J. C. Fogarty, S. A. Pandit, and A. Y. Grama. Parallel reactive molecular dynamics: Numerical methods and algorithmic techniques. Parallel Computing, 38(4-5):245–259, 2012. ISSN 01678191. doi: 10.1016/j.parco.2011.08.005.

[170] Kimberly Chenoweth, Sam Cheung, Adri C. T. van Duin, William a Goddard, and Ed- ward M Kober. Simulations on the thermal decomposition of a poly(dimethylsiloxane) polymer using the reaxff reactive force field. Journal of the American Chemical Soci- ety, 127(19):7192–7202, May 2005. ISSN 0002-7863. doi: 10.1021/ja050980t.

[171] M Weiler, S Sattel, T Giessen, K Jung, H Ehrhardt, V. Veerasamy, and J Robertson. Preparation and properties of highly tetrahedral hydrogenated amorphous carbon. Physical Review B, 53(3):1594–1608, 1996. ISSN 0163-1829. doi: 10.1103/PhysRevB. 53.1594.

[172] Frank H Allen, Olga Kennard, David G Watson, Lee Brammer, A Guy Orpen, and Robin Taylor. Tables of bond lengths determined by x-ray and neutron diffraction. part 1. bond lengths in organic compounds. Journal of the Chemical Society, Perkin Transactions 2, (12):S1, 1987. ISSN 0300-9580. doi: 10.1039/p298700000s1.

[173] Jiecai Han, Wei Gao, Jiaqi Zhu, Songhe Meng, and Weitao Zheng. Density-functional theory study of the microstructure, electronic structure, and optical properties of amorphous carbon. Physical Review B, 75(15):155418, April 2007. ISSN 1098-0121. doi: 10.1103/PhysRevB.75.155418.

[174] John Dean. Lange’s Handbook of Chemistry. McGraw-Hill, 15 edition, 1999.

[175] Anke Peguiron, Gianpietro Moras, Michael Walter, Hiroshi Uetsuka, Lars Pastewka, and Michael Moseler. Activation and mechanochemical breaking of c-c bonds initiate wear of diamond (110) surfaces in contact with silica. Carbon, 98:474–483, 2016. ISSN 00086223. doi: 10.1016/j.carbon.2015.10.098.

[176] Xianglong Yuan and A. N. Cormack. Efficient algorithm for primitive ring statistics in topological networks. Computational Materials Science, 24(3):343–360, 2002. ISSN 09270256. doi: 10.1016/S0927-0256(01)00256-7.

[177] Benedetta De Caprariis, Paolo De Filippis, Carlos Herce, and Nicola Verdone. Double- gaussian distributed activation energy model for coal devolatilization. Energy and Fuels, 26(10):6153–6159, 2012. ISSN 08870624. doi: 10.1021/ef301092r.

[178] Junmeng Cai, Weixuan Wu, and Ronghou Liu. An overview of distributed activation energy model and its application in the pyrolysis of lignocellulosic biomass. Renewable and Sustainable Energy Reviews, 36:236–246, 2014. ISSN 13640321. doi: 10.1016/j. rser.2014.04.052.

148 [179] R. Berman. The diamond-graphite equilibrium calculation: The influence of a recent determination of the gibbs energy difference. Solid State Communications, 99(1): 35–37, July 1996. ISSN 00381098. doi: 10.1016/0038-1098(96)00171-8.

[180] Tevis D.B. Jacobs, Bernd Gotsmann, Mark A. Lantz, and Robert W. Carpick. On the application of transition state theory to atomic-scale wear. Tribology Letters, 39 (3):257–271, 2010. ISSN 10238883. doi: 10.1007/s11249-010-9635-z.

[181] Tevis D.B. Jacobs and Robert W. Carpick. Nanoscale wear as a stress-assisted chemical reaction. Nature Nanotechnology, 8(2):108–112, 2013. ISSN 17483395. doi: 10.1038/nnano.2012.255.

[182] Bede Pittenger, Natalia Erina, and Chanmin Su. Quantitative mechanical property mapping at the nanoscale with peakforce qnm. Technical report, Bruker Nano Surfaces Division, 2012.

[183] Dalia G. Yablon and Greg Haugstad. Phase imaging. In Dalia G. Yablon, editor, Scan- ning Probe Microscopy in Industrial Applications: Nanomechanical Characterization, chapter 4, pages 95–114. Wiley, Hoboken, 1 edition, 2014.

[184] Prabhat Verma. Tip-enhanced raman spectroscopy: Technique and recent advances. Chemical Reviews, 117(9):6447–6466, 2017. ISSN 15206890. doi: 10.1021/acs. chemrev.6b00821.

[185] Alexandre Dazzi and Craig B. Prater. Afm-ir: Technology and applications in nanoscale infrared spectroscopy and chemical imaging. Chemical Reviews, 117(7): 5146–5173, 2017. ISSN 15206890. doi: 10.1021/acs.chemrev.6b00448.

[186] G. G. Stoney. The tension of metallic films deposited by electrolysis. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 82(553): 172–175, May 1909. ISSN 1364-5021. doi: 10.1098/rspa.1909.0021.

[187] Kenneth Holmberg, Helena Ronkainen, Anssi Laukkanen, Kim Wallin, Sture Hog- mark, Staffan Jacobson, Urban Wiklund, Roberto M. Souza, and Per St˚ahle.Resid- ual stresses in tin, dlc and mos2coated surfaces with regard to their tribological fracture behaviour. Wear, 267(12):2142–2156, 2009. ISSN 00431648. doi: 10.1016/j.wear.2009.01.004.

[188] Filippo Mangolini, Tevis D.B. Jacobs, Subarna R. Khanal, Frank Streller, J. Brandon McClimon, James Hilbert, Jennifer R. Lukes, and Robert W. Carpick. Origin of the thermo-oxidative stability of silicon- and oxygen-containing hydrogenated amorphous carbon (submitted). The Journal of Physical Chemistry Letters, 2018.

149