Application of the X-Ray Photoelectron Spectroscopy for Development of the Niobium Chemical Mechanical Process, Photomodificatio

Application of the X-ray Photoelectron Spectroscopy for Development of the Niobium Chemical Mechanical Process, Photomodification of Silicon for the Field Release Mass Spectrometer, and Analysis of the Multifunctional Oxide Heterostructures

A Thesis Presented

Natalia Maximova

The Department of Chemical Engineering

in partial fulfillment of the requirements for the degree of

Master of Science

Chemical Engineering

Northeastern University Boston, Massachusetts

April 17, 2008 ABSTRACT

X-ray photoelectron spectroscopy (XPS) provides information about the

elemental chemical composition of a surface and the bonding states of those elements.

This information is critical to diverse applications where the surface of the material

determines its functionality such as tuned catalysts, engineered polymer coatings, and nanoelectronic heterostructures. For this thesis, XPS has been applied in both traditional and novel ways to niobium surface polishing, silicon surface modification, and electronic structure measurement.

The current method for fabricating niobium superconducting cavities produces rough and defective surfaces. A proof-of-concept project to develop a niobium Chemical

Mechanical Polishing (CMP) process used XPS to monitor surface composition and structure under varying CMP parameters. XPS confirmed rapid oxidation of the niobium with a self-limiting surface oxide of 5.0±0.8 nm. CMP surface effects were explored and

a smooth (24 nm average roughness) niobium wafer with an ordered surface was

produced.

Current methods of detecting complex airborne toxins such as anthrax are time

consuming and often give false positives [1]. A modified field release mass spectrometer

(FRMS) will enable specific and selective real-time detection of air toxins through

utilization of the modified silicon surfaces for the capture and release of these analytes.

Photomodification of the silicon surface with undecylenic acid resulted in 42±1 %

monolayer coverage as determined from the XPS and angle resolved XPS (ARXPS) data

ii using an adapted method from Haber et al. in [2]. Carbon contamination was shown to be detrimental to the formation of the monolayers.

Determination of the valence band offsets in multifunctional oxide heterostructures provided a tool for insight in the electronic properties of these materials.

The valence band offset for magnesium oxide grown epitaxially on silicon carbide was found to be 1.13±0.12 eV which is consistent with expected offsets based on the band gaps of the two materials. Future work will focus on determining repeatability and accuracy of the valence band offset measurements in various heterostructures.

iii

ACKNOWLEDGEMENT

A great number of people assisted me in completion of this work. I would not be writing this if it was not for my family, my friends, fellow students at Northeastern, and last but not least faculty. Let me start by acknowledging the Northeastern crowd.

I would like to thank my advisor Dr. Katherine S. Ziemer who let me work in her laboratory, guided me through these three years at Northeastern, and secured financial aid to allow me to concentrate on studies and research. My gratitude goes to my committee members, Dr. Shashi K. Murthy, Dr. Daniel Burkey, and Dr. Sinan Muftu for help with my thesis. In addition to being on my committee, Dr. Shashi K. Murthy allowed me to use a chemical hood in his lab for silicon cleaning experiments. His students helped me with set up my equipment, especially Brian Plouffe. It was a pleasure working with Dr.

Sinan Muftu and George Calota who performed CMP experiments on the niobium wafers.

Throughout three years at Northeastern, I worked as a Connections lab manager under leadership of Rachelle Reisberg. I would like to thank Rachelle for giving me an opportunity to work at Connections and helping me through tough times.

Dr. Ronald Willey allowed me to use his lab space for purging of the reagent for photomodification in the silicon project. Between Christmas and New Year’s of 2006 when most people are on a break, he came in to change the pressure gauge to make sure that nothing blew up during my purging procedure. I look forward to meeting him at

AIChE meetings and while geocaching. His graduate students, James Minicucci and

Edward Viveiros helped me with setup of my experiment in their lab.

My boyfriend, Jonathan Meade, deserved acknowledgement with the University

folks as he helped me build a photoreactor. He put all pieces together and wired it. He

has been a great support since the time I joined Northeastern. Thanks to the in-network

minutes we stayed close and grew our relationship despite living in different states.

My fellow students deserve a special thank you as well. Kathleen McCarthy was my comrade in research and job search. During long discussions about research, economy, job search, politics, classes, family, relationships, etc, I feel that we have built a strong friendship. I hope to continue my relationship with Kathleen in my professional life.

Trevor Goodrich and Zhuhua Cai taught me how to use XPS and AES, and turned off the instruments on multiple occasions so that I could catch my train. Trevor did most of the SEM analysis of niobium samples, and Zhuhua cleaned silicon carbide and grew magnesium oxide for the valence band offset experiment. Bing Sun, thank you for your encouragement and support.

Dr. Albert Sacco Jr. allowed me to move into CAMMP office. Dr. Juliusz

Warzywoda and CAMMP students welcomed me and assisted me throughout my stay.

Mariam Ismail, Dennis Callahan, and Julo helped me with SEM imaging. Jonathan

Leong captured AFM images of the polished niobium surfaces.

Dr. Elizabeth Podlaha and her graduate students assisted with determination of the

OCP potential for the niobium CMP project.

Undergraduate students, Chris McLaughlin, Katie Passino, and Abby Deleault, contributed to the silicon project.

Robert Eagan has built an excellent support for the Barnstead filtering system, so

that I could wheel it around. He also assisted with some other projects which were

always completed in no time.

Now I would like to thank my family in U.S. and in Russia who kept my spirit up

throughout my academic career. My mom, her husband, and my sister encouraged and

supported me. My mom deserves special thanks for raising me and teaching me

discipline and determination. I would have not been here if it was not for her. My aunts in Saint Petersburg and Moscow, and my cousins listened to me complain and provided great support that would have been enough for two degrees.

My dear friend TO, who has been my friend since my studies at Saint Petersburg

University of Refrigeration and Food Technologies, has listened to me for hours, entertained me with her crazy adventures, and was just the best friend one could ever dream of.

My friends from Connecticut believed in me and helped me as well. I especially want to acknowledge my boyfriend’s family, especially his mom, who supplied me with prepared food and Costco products to help me save money and stay healthy.

TABLE OF CONTENTS

LIST OF FIGURES…………………………………………………………………… ix

LIST OF TABLES…………………………………………………………………….. xiii

1.0 INTRODUCTION………………………………………………………………1

2.0 BACKGROUND INFORMATION ON X-RAY PHOTOELECTRON SPECTROSCOPY………………………………………………………...... 6

2.1 Description of the XPS Instrument…………………………………………. 6 2.2 Physics behind XPS………………………………………………………….. 8 2.3 Spectral Interpretation………………………………………………………. 13 2.3.1 Survey Spectra…………………………………………………………. 14 2.3.2 Elemental Spectra………………………………………………………15 2.4 Quantitative XPS Analysis…………………………………………………... 16 2.4.1 Inelastic Mean Free Path and Sampling Depth……………………… 17 2.4.2 Area under the Curve, Sensitivity Factors for Elemental Composition, and Composition of Bonding States……………………………………... 18 2.4.3 Peak Fitting……………………………………………………………. 23 2.4.4 Thickness Calculations………………………………………………... 26 2.4.5 Error Analysis…………………………………………………………. 27 2.5 Beyond Composition…………………………………………………………. 29 2.5.1 Angle Resolved XPS for Depth Profiling and Increased Surface Sensitivity………………………………………………………………… 29 2.5.2 Valence Band Offset Measurements………………………………….. 31 2.6 Experimental Apparatus and Procedures in the Interface Engineering Laboratory…………………………………………………………………… 33 2.7 Applications of XPS………………………………………………………….. 36

3.0 NIOBIUM CHEMICAL MECHANICAL POLISHING……………………….37

3.1 Critical Literature Review: Niobium CMP ………………………………... 40 3.1.1 Surface Requirements for SRF Cavities……………………………… 41 3.1.2 Niobium: Surface Studies……………………………………………... 47 3.1.3 Chemical Mechanical Polishing: Applications to Semiconductors and Metals…………………………………………………………………….. 57 3.1.4 Niobium Chemistry; Pourbaix Diagrams…………………………….. 61 3.1.5 Summary………………………………………………………………. 65 3.2 Experimental: Niobium CMP………………………………………………. 66 3.2.1 Materials……………………………………………………………….. 66 3.2.2 Experimental Procedure: Degrease, BCP, Oxidizing and Etch Treatments, CMP………………………………………………………… 68 3.2.3 XPS: Data Acquisition and Manipulation……………………………. 71

vii

3.2.4 SEM and AFM Analysis………………………………………………. 75 3.3 Results and Discussion: Niobium CMP…………………………………….. 76 3.3.1 Chemical Stability of Niobium in Oxidizing and Etching Aqueous Solutions………………………………………………………………….. 76 3.3.2 Niobium Surface Condition after CMP Process……………………… 87 3.3.2.1. Effect of Starting Surface Condition…………………………88 3.3.2.2. Effect of Slurry Parameters: Type of Abrasive Particle, pH…………………………………………………………………..97 3.3.2.3. Multi-Slurry Process …………………………………………106 3.4 Summary and Conclusions for Nb CMP…………………………………....111 3.5 Recommendation Nb CMP…………………………………………………..112

4.0 SILICON SURFACE MODIFICATION WITH ORGANIC MONOLAYERS………………………………………………………………114

4.1 Description of the Project …………………………………………………... 114 4.2 Critical Literature Review: Silicon Surface Modification………………… 119 4.2.1 Si Modification via Wet Chemical Approach………………………… 120 4.2.2 Application of XPS for Analysis of Modified Si Surfaces……………. 128 4.2.3 Summary………………………………………………………………. 135 4.3 Experimental Section: Silicon Surface Modification……………………… 136 4.3.1 Materials……………………………………………………………….. 136 4.3.2 Experimental Procedure: Degrease, RCA Cleaning, Fenner Etch, Ar Purge, Photomodification, Stability Protocol…………………………… 137 4.3.3 XPS: Data Acquisition and Manipulation……………………………. 139 4.4 Results and Discussion: Silicon Surface Modification………………………142 4.4.1 Starting Surface: RCA Cleaning and Fenner Etch…………………... 142 4.4.2 Effect of Initial Carbon Contamination on Monolayer Coverage…… 152 4.5 Summary and Conclusions for Silicon Surface Modification ……………..159 4.6 Recommendations for Silicon Surface Modification……………………….161

5.0 SUMMARY…………………………………………………………………….163

6.0 NOMENCLATURE……………………………………………………………166

7.0 REFERENCES………………………………………………………………....169

8.0 APPENDICES………………………………………………………………….177

8.1 Appendix A ………………………………………………………………..178 8.2 Appendix B ………………………………………………………………..188 8.3 Appendix C ………………………………………………………………..198

viii

LIST OF FIGURES

Figure 1: Schematic of main instrumental components in XPS...... 8

Figure 2: Schematic of the photoelectron process in C1s atom during XPS analysis..... 9

Figure 3: Schematic diagram illustrating the take-off angle in XPS...... 12

Figure 4: Survey spectrum of the etched silicon sample (step size 1 eV/S, pass energy 89.45 eV, 5 consecutive spectra)...... 15

Figure 5: C1s spectrum of silicon functionalized with undecylenic acid. Blue squares represent experimental data, orange lines are individual fit for four bonding environments, and the red line shows the overall fit...... 16

Figure 6: Effect of the angle on the sampling depth [22]...... 30

Figure 7: Depth profile of niobium sample oxidized in air [21]...... 31

Figure 8: Energy band structure of the interface layers of the semiconductor substrate [23]...... 32

Figure 9: Valence band maximum for GaAs was determined to be 0 eV [24]...... 33

Figure 10: Schematic representation of the operation of the niobium cavities for electron acceleration [30]...... 38

Figure 11: EP results in smoother surface than BCP ………………………………….. 46

Figure 12: Surface roughness values of up to 2 µm are desirable to prevent RF field enhancement with 1300 MHz SRF cavities ………………………………. 46

Figure 13: Schematic of the serration of niobium surface due to oxidation in air for over one week …………………………………………………………………... 49

Figure 14: Niobium oxides are hypothesized to serrate niobium surface …………….. 55

Figure 15: Values for Nb and its oxides found in the literature. A= [20], B= [18], C= [19], D= [17], E= [16], F= [3], G= [3] ………………………………………… 56

Figure 16: Schematic representation of CMP process…………………………………. 58

Figure 17: Pourbaix diagram for Nb-H2O system at 25, 75, and 95°C [61]. Lines a and b represent hydrogen evolution and oxygen reduction, respectively...... 64

Figure 18: Nb3d spectrum……………………………………………………………... 74

Figure 19: Nb3d spectra of as-received niobium wafer exhibits presence of two oxides: NbO and Nb2O5. Blue squares represent experimental data and brown line is the overall fit. All of the peaks are fitted with Gaussian-Lorentzian curves (80% Gaussian)……………………………………………………………. 80

Figure 20: Nb3d peak for samples treated with hydrogen peroxide, HF, and HF and hydrogen peroxide...... 82

Figure 21: Oxidative treatment results in rapid oxide growth...... 86

Figure 22: Optical images of the starting surface condition for the wafers that were rolled, BCP treated, and mechanically abraded...... 89

Figure 23: Optical images for the final surface in Experiment 3 (left) and Experiment 6 (right)……………………………………………………………. 91

Figure 24: Nb3d spectra of the sample obtained from the wafer polished in Experiment 3...... 92

Figure 25: Nb3d5 spectrum of the samples obtained in Experiment 6...... 93

Figure 26: Optical microscope image of the wafer obtained in Experiment 11 after 21 min of polishing...... 94

Figure 27: Nb3d5 peak of the wafer obtained in Experiment 11. SEM image shows presence of particles on the surface. XPS revealed alumina...... 95

Figure 28: Optical image of the final surface condition obtained in Experiment 8...... 95

Figure 29: Nb3d5 spectrum (Experiment 8) shows pentoxide as a predominant oxide, oxides thickness 4.9±0.7 nm. SEM image insert proves particle contamination...... 96

Figure 30: Optical image post-CMP for Experiment 1………………………………… 99

Figure 31: XPS spectrum of the sample obtained in Experiment 1. SEM image (insert) shows presence of particles which were determined with XPS to be silica particles...... 100

Figure 32: Optical microscope images of the surface treated with alumina slurry for 16 and 31 min in Experiment 12...... 101

Figure 33: Nb3d spectrum shows that pentoxide species are predominant on Nb surface after CMP in Experiment 12. SEM image (insert) shows high particulate contamination on the surface...... 102

Figure 34: Graph showing evolution of the niobium wafers treated with silica and alumina containing slurries...... 103

Figure 35: Pourbaix diagram of niobium with slurries plotted [61]...... 104

Figure 36: Graphs representing evolution of the roughness values over time for the samples polished in Experiments 11 and 12 do not show significant trends, indicating that removal rates are not affected by change in the pH from 7 to 10...... 106

Figure 37: Optical microscope image of the wafer obtained in Experiment 10...... 108

Figure 38: SEM images obtained from the samples polished in Experiment 10 show presence of particles on the surface...... 108

Figure 39: The AFM images of the surfaces obtained in Experiment 10 allow to determine that surface between defects and particles is significantly smoother than the values determined by the optical microscope...... 109

Figure 40: Nb3d5 spectrum for Experiment 10………………………………………... 110

Figure 41: Conceptual representation of FRMS...... 115

Figure 42: Release of negatively charged analyte after electric field if applied...... 118

Figure 43: Schematic representation of H-Si (100) and (111) surfaces ……………….. 121

Figure 44: Schematic of the reaction between unsaturated organic molecule and hydrogen terminated silicon (111) surface [78]...... 122

Figure 45: Mole fraction of reactants in the reaction mixture was found to be the same as in the monolayer, which was determined by ATR-FTIR [93]...... 126

Figure 46: Survey XPS spectra show (a) H-Si, (b) chlorinated with saturated PCl5, and (c) alkylated with C8H17MgCl [83]...... 129

Figure 47: Oxidation of H-terminated Si(111) in air was monitored by observing Si2p peak. Curves represent (a) H-Si, (b) 4 h in air, (c) 72 h in air, and (d) 216 h in air [83]...... 131

Figure 48: XPS allows monitoring of chemical reaction such as deprotection of carboxylic acid ester. Plane A shows sample with trifluoroethyl ester of undecylenic acid on the surface, and plane B shows surface with undecylenic acid [79]...... 133

Figure 49: C1s spectra of (a) the carboxylic acid-terminated surface and (b) NHS-ester terminated surface [75]...... 134

Figure 50: Survey spectra at 90° take-off angle of (a) degreased Si(100) , (b) unsuccessful cleaning and etching, and (c) successful cleaning and etching...... 143

Figure 51: C1s spectra of a sample (a) the H-terminated Si (100) with (b) undecyletnic acid, and (c) after exposure to air for 8 days...... 153

Figure 52: C/Si at 90° take-off angle…………………………………………………... 155

Figure 53: Illustration of the effect of contamination on the undecylenic acid monolayer coverage...... 155

Figure 54: C1s peak for samples with high initial carbon concentration: (a) H-Si(100), (b) after modification with undecylenic acid, and (c) after 8 days of air exposure in the dark...... 156

Figure 55: O/Si ratio at 90° take-off angle…………………………………………….. 157

Figure 56: XPS analysis at 30° take-off angle showed similar evolution of C/Si (A) and O/Si (B) ratios as at 90°...... 158

xii

LIST OF TABLES

Table 1: IMFP and sampling depth values for the elements that were studied during the course of this work (Mg Kα anode)...... 17

Table 2: Sensitivity factors that were used to calculate atomic percent for the surface composition in this work...... 21

Table 3: Primary and secondary shifts for C1s [4]...... 25

Table 4: Chemical shift values for niobium [3], [16], [17], [18], [19], [20], [21]...... 26

Table 5: Repeatability of the calibration for gold and copper over consecutive measurements...... 35

Table 6: Common causes of the SRF cavity failure and current solutions …………… 41

Table 7: Role and examples of the main slurry components...... 59

Table 8 Density of niobium and its oxides...... 63

Table 9: Properties of the slurries used...... 67

Table 10: Slurries used during CMP experiments...... 70

Table 11: Typical energy ranges and number of sweeps obtained for niobium samples after static and CMP experiments...... 72

Table 12: Area under curves indicated in Figure 18...... 74

Table 13: Summary of oxide thicknesses and the FWHM values for the oxidation time study...... 85

Table 14: Experiments that allow comparison of the starting surface effect on CMP process...... 89

Table 15: Experiments performed to determine the effect of slurry on CMP of Nb: particle type and pH...... 98

Table 16: pH and OCP values of the slurries...... 104

Table 17: Experiments allowing comparison of the multi-slurry processes...... 107

Table 18: Typical bond energies for Si with Si, H, C, O, and F atoms, kJ mol-1 …….. 118

xiii

Table 19: Methods of initiation, reaction parameters, and organic molecules that were attached to silicon substrates...... 123

Table 20: Typical energy ranges and number of sweeps that were used to obtain silicon surface composition...... 140

Table 21: Selected experimental results directed to determine source of carbon contamination………………………………………………………………… 146

Table 22: Carbon gain for high and low carbon contamination samples due to photomodification. Error = standard deviation.……………………………… 159

xiv

1.0 INTRODUCTION

X-ray photoelectron spectroscopy (XPS) has been a valuable surface characterization technique since the 1950s with the first commercial instruments being available in 1960-1970 [4]. Some of the industrial applications of XPS are in metallurgy, polymer science, semiconductor industry, adhesive science, and ceramics [5]. For example, XPS is extensively used in the polymer technology for quality control purpose.

XPS is able to detect low levels (0.3%) of contaminants or additives in the product [6].

For some application such as oxidation prevention and fire retardancy, surface segregation of the additive is unwanted; in others, such as low friction polymers, the additives must migrate to the surface [6]. Thus, use of XPS allows verification of the additive distribution in the material. Overall in the polymer industry, due to minimal surface preparation and high surface sensitivity, XPS has contributed to production of the quality polymeric materials and development of the new ones with novel or improved properties.

XPS detects electrons from the core levels of the atom that were excited by the X- ray photons. Kinetic energy of these electrons is measured by the detector and can be related to the binding energy of the orbital from which the electron was ejected. This binding energy is characteristic of the element and the bonding environment of the element in the material. Thus, by scanning through the range of kinetic energies and calculating of binding energies, the instrument plots a graph of the intensity (counts) versus binding energy. The binding energies of all elements are known, so identification of the elements present in the sampling depth of the material, except for hydrogen and

1 helium which cannot be analyzed with XPS, is easily accomplished. Thus, qualitative information identifying the elements and their bonding state in the sampling depth is obtained. The intensity at a particular binding energy is directly proportional to the number of atoms of that element in that binding state. Therefore, through the use of the relative sensitivity factors, relative atomic composition in the sampling volume can be determined.

One of the advantages of XPS is its surface sensitivity. Due to inelastic scattering of electrons in the material, 95% of the electrons detected by the spectrometer originate from the sampling depth of less than 10 nm [4]. Sampling depth is determined by the inelastic mean free path (IMFP), which is an average distance that electron can travel between successive inelastic collisions [4], and the take-off angle, the angle between the surface and the detector. IMFP is dependent on the properties of the material as well as the kinetic energy of the emitted electrons; kinetic energy is dependent on the photon excitation energy. Sampling depth can be changed by either changing the incident energy, which is done by using synchrotron radiation, or by varying the take-off angle between the sample and the sample surface, which is utilized in angle resolved XPS

(ARXPS). XPS using synchrotron radiation is performed at the accelerator facilities such as Brookhaven National Laboratory. In ARXPS, a standard X-ray source can be used, making it more accessible alternative.

In a layered structure, XPS allows determination of the thickness by relating signals from the substrate and the overlayer. So evaluation of the thickness can be done for thin films, such as protective oxide films on metals and self-assembled monolayers on various substrates, within the sampling depth of the instrument. ARXPS allows one to

perform more surface sensitive analysis and aid in determination of structure of the

material and layer thickness.

XPS detects valence electrons from the material as well. Analysis of the valence

band spectra can help identify and distinguish between compounds having the same

relative elemental composition, but different bonding structures, such as poly(ethylene)

and poly(1-butene) [6], and assist in determination of the electronic structure of the

material. The valence band electrons are detected in the range between -5 and 30 eV.

Valence band offset gives an insight in the alignment of the energy bands in a composite

material. Energy band bending occurs at the interface between a semiconductor and

vacuum or another material. Valence band analysis of the surface and interface allows

determination of the valence band offset, and was first shown by Kraut et al. in 1980 [7].

The goal of this thesis was to demonstrate the versatility of XPS in three different projects. A proof-of-concept project to develop a niobium chemical mechanical polishing (CMP) process to produce smooth ordered niobium surfaces was pursued by

H.C. Starck Company and Chemical Engineering and Mechanical Engineering

Departments at Northeastern University. In this project, XPS was applied to gain understanding of niobium surface chemistry for slurry development and determine effect of the CMP process on the atomic order of the niobium wafers. These objectives utilized the surface sensitivity of XPS and its ability to distinguish both bonding states and near surface atomic structure. In this project, the resolution limits of the XPS were tested in resolving the various oxide states, and the sensitivity of the tool in accurate measurement of changes in both relative intensity and peak width were demonstrated. We also believe that this is the first demonstration of CMP of niobium.

The second project, in which knowledge of the surface composition and chemistry

was critical, was modification of a silicon surface for the field release mass spectrometer

(FRMS). A photoreactor was built to produce functionalized silicon surfaces. Due to

high surface sensitivity, ARXPS was used to monitor surface modification and to determine the effect of the surface contamination on the monolayer formation. The difficulties of both developing an effective photoreactor and using the XPS to distinguish the organic monolayer coverage of the surface were explored. The method for monolayer thickness determination used by Haber et al. in [2] was adapted for the undecylenic acid and contamination monolayer coverage calculations of the silicon surfaces.

Determination of the coverage is important to work involving self-assembled

monolayers, as monolayers can not easily be seen by scanning electron microscopy or

atomic force microscopy alone.

Finally, preliminary work has been started on determination of the valence band

offsets in multifunctional oxide heterostructures. In the Interface Engineering

Laboratory, various oxides are grown using molecular beam epitaxy. The ability to “see” band bending in the material is important to predicting electrical performance of these

functional films for possible applications. This project explores the limits of the valence

band measurement capability of XPS.

This thesis is organized thus: the XPS background information is given in section

2.0, followed by separate sections on the niobium project and the silicon surface

modification project, and finally overall conclusions regarding the used of XPS. The

preliminary valence band study is provided in the appendices along with supporting

experimental development efforts. The niobium CMP section (3.0) includes a critical

4 literature review of the niobium surface science and processing, an experimental section, results, conclusions, and recommendations. The silicon surface modification for the development of field release mass spectroscopy is described in section 4.0, and again includes a project-specific literature review, an experimental description, results, conclusions, and recommendations. The overall thesis conclusion is given in section 5.0.

Preliminary results and background information on the valence band analysis can be found in Appendix A. Additional information on niobium buffered chemical polishing of various samples is given in Appendix B. Description of the custom-built photoreactor can be found in Appendix C.

2.0 Background Information on X-ray Photoelectron Spectroscopy

X-ray photoelectron spectroscopy (XPS) has been a valuable surface

characterization technique since the 1950s with the first commercial instruments

available in 1960-1970 [4]. From the analysis of the core level photoelectrons, XPS

allows one to obtain information about the elemental chemical composition of the sample, and the bonding states of those elements. XPS is characterized by high surface

sensitivity, sufficient energy resolution to distinguish between different bonding states,

and ability to perform quantitative analysis. Physical background related to XPS

analysis, spectral interpretation, quantitative XPS analysis, advanced applications of

XPS, experimental apparatus, and applications of XPS in industry will be discussed in

turn in the following sections.

2.1 Description of the XPS Equipment

In XPS, X-ray source excites the electrons in the near surface layers of the

material. One of the processes that occurs due to X-ray excitation is ejections of the core

level electrons from the atoms. The XPS instrument is tuned to effectively detect these

core level electrons. If gas molecules are present between the sample surface and the

detector, the electrons will be scattered and lost from the analysis. Thus the vacuum

requirement arises. Based on the sticking probabilities of the gas molecules to the sample

surface, the ultra high vacuum (UHV) is required. Typically, pressure below 10-9 Torr is used, and the system should not be operated at the base pressure above 10-8 Torr.

To have sample and the XPS components (source, analyzer, and detector) in the

UHV environment, stainless steel vessels with mu-metal cladding, which provides screening from the earth’s magnetic field, are used. Figure 1 shows a schematic representation of the UHV vessel and the main instrument’s components that are used in

XPS. The X-ray source consists of a filament and an anode. The electrons from the filament bombard the anode material producing X-ray photons. Aluminum and magnesium are the most common materials used as anodes for the X-ray generation [4].

The energies of the X-ray photons generated are 1486.6 eV and 1253.6 eV for aluminum and magnesium, respectively. The linewidth, which is an energy distribution of the X-ray photons, is an important characteristic that determines the resolution of the instrument.

For aluminum and magnesium anodes, the linewidth values are 0.85 eV and 0.70 eV, respectively. The linewidth of aluminum can be decreased by eliminating of the

Bremsstrahlung, satellite, and ghost radiation by means of a quartz crystal monochromator. As a result, the linewidth of the monochromated Al anode decreases to

0.3 eV.

When X-ray beam generated by the X-ray source excites and ejects electrons from the surface of the sample, these electrons are detected with electron energy analyzer as shown in Figure 1. The concentric hemispherical analyzer (CHA) has been used in XPS since the development of the technique [6]. This analyzer consists of the two concentrically positioned hemispheres. A potential bias, which is also called a deflection voltage, is applied to the two hemispheres, so that the median equipotential will be at the radius of the entrance and exit slits of the analyzer. Only electrons with the “correct” energy will be able to pass through the analyzer to the detector. By ramping up the

deflection voltage the electrons of progressively higher energy will be allowed to reach

the detector [4].

Electron Energy Analyzer

Electron X-Ray Source Detector

UHV Vessel

Sample

Figure 1: Schematic of main instrumental components in XPS.

Finally, the detector is used for detection of the electrons that passed through the

analyzer. The simplest type of the detector is a single channel electron multiplier which

allows electron amplification to about 108 [4]. It allows single electron counting. Thus,

the analyzer controls which energies are allowed to pass and the detector counts the

electrons with that energy.

2.2 Physics behind XPS

When a solid substance is irradiated with X-rays of a known energy, several

electron emission transitions can occur: photoelectrons from the core and valence energy

levels, secondary electrons, and Auger electrons. The hemispherical electron energy

analyzer with an electron multiplier is typically used for the photoelectron detection.

This analyzer is tuned to accurately detect the electrons of the higher kinetic energy such

as photoelectrons. The photoelectron transition is the basis of X-ray photoelectron spectroscopy, or XPS.

Figure 2 shows the excitation of a carbon core electrons from the 1s level with X-

ray photons. Not only electrons from primary ionization (photoelectrons) are detected,

but also electrons from the Auger relaxation process (Auger electrons) can be detected.

The C1s electrons which are excited by the X-ray beam overcome binding energy of the

carbon 1s orbital. The binding energy is characteristic of an atom and its bonding state.

The energy of the X-ray photon will be transferred to the carbon 1s electron and will

allow it to leave 1s orbital and, once at the interface of the material and vacuum, to

overcome surface interaction and travel through vacuum and through the analyzer to the

detector. The energy which is associated with the electron leaving the surface of the

material and traveling to the analyzer is called a work function. Thus, the photon energy

will be transferred to the electron and used to overcome the binding energy in the atom

and the work function, the remaining energy will be in a form of a kinetic energy. The

work function and the X-ray photon energy are known, and the kinetic energy of the

photoelectrons is measured by the detector.

Photon Photoelectron

Figure 2: Schematic of the photoelectron process in C1s atom during XPS analysis.

The law of the conservation of energy dictates that for the isolated system the

energy must be constant. Thus the energy of the X-ray photons must be equal to the

binding energy, work function, and kinetic energy of the photoelectrons. Equation relating the kinetic energy of the electrons with excitation energy and binding energy was

first proposed by Rutherford in 1914 [4]. The work function term was later determined.

The form of the equation which is currently used to calculate the kinetic energy is shown

in Equation 1. Due to the fact that an excitation energy and a work function are known

and the kinetic energy is measured, the binding energy can be easily calculated from

Equation 1. The instrument calculates binding energy and counts how many electrons

which a particular binding energy are detected, thus providing a spectra of counts versus

binding energy. XPS spectra are discussed in more detail in section 2.2.

EK = hν − EB −ϕ 1

Where: hν = excitation energy (Mg Kα = 1486.6 eV and Al Kα = 1253.6 eV)

EK = kinetic energy (eV)

EB = binding energy (eV)

φ = work function is defined as energy required to overcome surface

attraction and travel from sample surface to the detector (eV)

X-rays penetrate many micrometers into the material [8]; however, only electrons

from less than 10 nm depths are detected due to inelastic scattering in the material.

Inelastic scattering leads to the electron attenuation which is described by the Beer-

Lambert law, as seen in Equation 2 [4].

I z = I 0 exp(−z / λ sinθ ) 2

Where: Iz = intensity from the atoms at depth z (nA)

I0 = intensity from the surface atoms (nA)

z = depth (nm)

λ = inelastic mean free path, IMFP, (nm)

θ = electron take-off angle

The detector, an electron multiplier, measures the photoelectron current, which is

a current generated by the photoelectrons. The photoelectron current from a layer of

atoms at depth z will be attenuated by the thickness of the material above it, and so it will

be a fraction of the current from the surface atoms, even if the number of atoms in the

layer is the same as the number of atoms on the surface. The inelastic mean free path

(IMFP) is defined as the distance an electron can travel in a material between successive

inelastic collisions [4]. IMFP is specific for the element and the substrate, or solid

material matrix of which the element is a part. The IMFP determines the escape and

sampling depths. Escape depth is defined as “the distance normal to the surface, at which

the possibility of an electron escaping without significant energy loss due to inelastic

scattering process dropped to e-1(36.8%) of its original value” [4]. And sampling depth is the depth from which 95% of the electrons originate and is equal to three times the length of the escape depth. Commonly, escape and sampling depths are calculated using

Equation 3 and Equation 4, respectively. The take-off angle is illustrated in Figure 3.

Escape depth = λ sinθ 3

Sampling depth = 3λ sinθ 4

Where: λ = inelastic mean free path (nm)

θ = electron take-off angle

Analyzer X-ray Beam

Figure 3: Schematic diagram illustrating the take-off angle in XPS.

The inelastic mean free path values can be calculated using NIST Electron

Effective-Attenuation-Length Database (SRD 82). NIST SRD 82 database is actually a

computer program which is available for free. User is required to input type of element,

material, kinetic energy of the electrons, as well as some physical properties (density and

band gap). Calculation of the IMFP is done using TPP-2M equation which was derived from the analysis of the IMFP values for the groups of elemental solids and organic compounds [9]. Equation 5 shows the TPP-2M equation [9].

E λ = 2 2 E p [β ln(γE) − (C / E)+)D / E )]

2 2 0.5 0.1 β = −0.10 + 0.94 /(E p + Eg ) + 0.069ρ γ = 0.191ρ −0.5 5 C = 1.97 − 0.91U D = 53.4 − 20.8U

2 U = Nυ ρ / M = E p /829.4

Where λ = the IMFP (Ǻ)

E = the electron energy (eV)

Ep = the free-electron Plasmon energy (eV)

Nv = the number of valence electrons per atom or molecule

ρ = the bulk density (kg/m3)

M = the atomic or molecular weight (kg/mol)

Eg = the band-gap energy for non-conductors (eV)

Tanuma et al. found that deviation (RMS) between calculated values determined using Equation 5 and calculated from optical data were 10.2% and 8.5% for the groups of elements and organic compounds, respectively [10]. However, RMS for diamond, graphite, and cesium were greater than 39% for the IMFP values calculated from

Equation 5 and optical data [11]. Due to the fact that the error for the IMFP calculation for the C, Al, Si, and Nb was reported to be approximately 13% [12]. We will use 15% for the error analysis in this thesis.

2.3 Spectral Interpretation

The XPS instrument presents data as counts versus binding energy. Due to the fact that binding energy is a function of the element and bonding states, XPS provides information of what elements are present in the sample and in what bonding state.

Usually two types of spectra are obtained: survey and elemental. Survey spectra give a snap shot of what elements are present in the sample; however, information about the bonding states and the relative atomic percent should be obtained from the elemental spectra in order to have a higher accuracy. Survey and elemental spectra will be discussed in the following sections.

2.3.1 Survey Spectra

The purpose of the survey spectra is to give overall information about the

elements present in the sample and their relative abundance. The maximum survey spectra range is given the excitation energy minus work function, because binding energy of the photoelectrons cannot exceed the excitation energy of the X-ray photons. Thus, if the work function is equal to zero, spectra above 1253.6 eV and 1486.6 eV for magnesium and aluminum anodes, respectively, will not be meaningful. For most of the samples analyzed in this work spectra of 1-1000 eV were taken, because no peaks were present at energies higher than 1000 eV. For example, Figure 4 shows a survey spectrum of a silicon sample taken with the Mg anode with the step size of 1 eV/S (eV per step) and pass energy of 89.45 eV. Step size indicates how often data point were obtained, in this case a data point at every 1 eV was recorded. The pass energy determines the

resolution. When the instrument scans the kinetic energies of the photoelectrons and calculates corresponding binding energy values, it count electrons which have the binding

energy of a certain range around any particular binding energy value, this range is the

pass energy. In XPS, the full width at half maximum (FWHM) is a measure of the

spectral resolution. FWHM is a function of a material and an instrument; it is determined

by the following instrumental parameters: geometry of the system, linewidth of the X-ray source, pass energy, and the analyzer resolution. For the instrument used in this work, the FWHM was equal to 0.015 of the pass energy [13]. The higher the pass energy the lower resolution but higher signal-to-noise ratio will be. So survey spectra are usually run with high step size and pass energy to obtain the low resolution spectra at reasonably

short period of time. Usually 5 to 10 survey spectra, or sweeps, are taken and averaged to

obtain acceptable signal-to-noise ratio. Oxygen and silicon were detected in the sample

shown in Figure 4.

3500

3000 Si1s Si2p3

2500 Auger O O1s

2000

1500 Count per second Count

1000

500

0 1000 900 800 700 600 500 400 300 200 100 0 Binding Energy, eV

Figure 4: Survey spectrum of the etched silicon sample (step size 1 eV/S, pass energy

89.45 eV, 5 consecutive spectra).

2.3.2 Elemental Spectra

From the survey spectra, one determines what elements are present in the sample.

In the sample shown in Figure 4, silicon and oxygen are detected, so elemental spectra for these and any other elements can be obtained. Elemental spectra are obtained at the lower step size (we use 0.05 eV per step) and lower pass energy of 35.75 eV. These

settings allow obtaining of a data point every 0.05 eV and the minimum FWHM of 0.54

eV as oppose to 1.34 eV which is used for the survey scans. Thus, instrument resolution

is enhanced from that of the survey scan settings. In order to obtain reasonable signal-to- noise ratio and resolution, 30 to 40 spectra are averaged. Figure 5 shows a C1s spectrum

of a silicon sample modified with undecylenic acid. The experimental data, which is

shown with blue squares, falls into an asymmetric peak which was fitted with four

separate peaks corresponding to C-C, C-Si, C-O, and C-(O)O bonding states of carbon.

Thus, elemental spectra contain vast amount of information on the bonding environment

of atoms in the interface. In addition, elemental spectra can be used to determine atomic composition at the surface with higher accuracy than from survey spectra due to improved resolution and signal-to-noise ratio. Fitting of the data is discussed in greater detail in section 2.3.2.

C-C

C-O

Count/s COOH C-Si

291 289 287 285 283 281 Binding Energy, eV

Figure 5: C1s spectrum of silicon functionalized with undecylenic acid. Blue

squares represent experimental data, orange lines are individual fit for

four bonding environments, and the red line shows the overall fit.

2.4 Quantitative XPS Analysis

It was shown that few nanometers of the surface are analyzed using XPS and that

qualitative information can be obtained from survey and elemental spectra. In addition to

identifying elements present in the sampling volume of the samples, XPS allows one to

perform quantitative analysis. Relative concentrations can be calculated from the areas

under the curves in XPS spectra; these areas are proportional to measured photoelectron

current intensities. The following sections will describe the peak fitting procedure and

determination of the sampling volume, relative concentrations, and overlayer thickness,

from the fitted peak information.

2.4.1 Inelastic Mean Free Path and Sampling Depth

Sampling depth in XPS analysis is on the order of few nanometers which is a consequence of the inelastic scattering in the material. It was shown in Equation 4 that sampling depth is determined by the IMFP and take-off angle. Table 1 shows IMFP and sampling depth values for a few elements at the take-off angles of 90° and 30°. IMFP

values were calculated using NIST SRD 82 database. The sampling depth at the 90°

take-off angle is less than 10 nm for all elements, as seen in Table 1. If the take-off angle

is decreased to 30°, the sampling depth is decreased by half. Decrease in the sampling

depth with decrease in the take-off angle is utilized in angle resolved XPS (ARXPS)

which is discussed in section 2.5.1.

Table 1: IMFP and sampling depth values for the elements that were studied during

the course of this work (Mg Kα anode).

IMFP, nm Sampling depth = 3λ sinθ , nm

Take-off angle 90° 30°

2.7 Silicon, Si2p 8.2 4.1

2.4 Carbon, C1s 7.2 3.6

1.6 Fluorine, F1s 4.8 2.4

1.9 Oxygen, O1s 5.7 2.9

2.7 Niobium, Nb3d 7.0 3.5

2.3 Aluminum, Al 6.7 3.4

2.4.2 Area under the Curve, Sensitivity Factors for Elemental Composition, and

Composition of Bonding States

Quantitative analysis using XPS is possible due to the fact that intensity of the

photoelectron current for a certain binding energy detected is directly proportional to the

number of atoms that have that binding energy in the sample. Equation 6 shows all of the

variables that affect photoelectron current for the core level X of the element A.

Instrument (J, G(EX), D(EX)) and material (p, σ(hν,EX), L((hν,X), Na(z), λX(EX)) parameters determine the photoelectron current measured by the instrument.

x I (X ) = Jpσ (hυ, E )L(hυ, X )G(E )D(E ) N (z)exp[−z / λ (E )sinθ ]dz 6 A X X X ∫ a m x 0 Where: J = X-ray flux

EX = kinetic energy of the photoelectrons from the core level X (eV)

p = surface roughness

σ(hν,EX) = the photoionization cross-section for ionization of X by photon

hν

L((hν,X) = the angular asymmetry factor for emission of X by photon hν

(cm-2s-1)

G(EX) = the spectrometer étendue, which is a product of the transmission

efficiency and the area from which the electrons are emitted at

kinetic energy EX

D(EX) = the detector efficiency at energy EX

Na(z) = the distribution of atoms A with depth z

λX(EX) = the inelastic mean free path of electrons EX in matrix M

θ = is the angle of emission [4]

For the constant X-ray flux and uniform sample composition the photoelectron

current in Equation 6 can be simplified to Equation 7 [4]. Again photoelectron current

intensity is directly proportional to number of electrons in the sample.

I A (X ) = KσLN Aλm sin(θ )GD 7

Where: K = X-ray flux of the instrument

σ = the photoionization cross-section

L = the angular asymmetry factor for emission of X by photon hν (cm-2s-1)

NA = the distribution of atoms A with depth x

λm = the IMFP (nm)

G = the spectrometer étendue at kinetic energy

D = the detector efficiency λm

θ = is the angle of emission [4]

As seen from Equation 7, intensity is a function of a material, thus intensities of elements needs to be referenced to some material. Fluorine F1s peak in polytetrafluoroethylene (PTFE) is often used as a reference peak for determining the normalizing factor which is called relative sensitivity factor in XPS. Equation 8 shows that the ratio of intensities is taken and the few terms that correspond to the instrument

(G, D, K) will cancel out if the same instrument is used in the experiment. Thus, material properties can be put together as a sensitivity factor SA. Intensity from fluorine atoms can

be taken as 1, so that intensity of the signal from atom A is expressed as the ratio of the

number of atoms A and F times the relative sensitivity factor.

I A Kσ A LN AλAc sin(θ )GA DA N A = = S A 8 I F Kσ F LN F λF sin(θ )GF DF N F

Where: K = X-ray flux of the instrument

σ A = the photoionization cross-section from ionization of A by photon hν

σ F = the photoionization cross-section from ionization of F by photon hν

L = the angular asymmetry factor for emission from X by photon hν

(cm-2s-1)

NA = distribution of atoms A

NF = distribution of atoms F

GA = the spectrometer étendue at kinetic energy of A

GF = the spectrometer étendue at kinetic energy of F

DA = detector efficiency at kinetic energy of A

DF = detector efficiency at kinetic energy of F

θ = is the angle of emission

SA = sensitivity factor for A [4]

Sensitivity factors can be experimentally determined in a lab for a particular geometry of the system. Measuring sensitivity factors in-house results in a more accurate determination of the sensitivity factors as slight modifications in geometry of the system may have a serious impact on instrumental factors that are part of the Equation 8.

However, in order to determine sensitivity factors in the lab one will have to have pure standards, which may be hard to obtain or may require generation in-house. In this work, relative sensitivity factors were obtained from the manufacturer (Physical Electronics).

These sensitivity factors are listed in Table 2.

Table 2: Sensitivity factors that were used to calculate atomic percent for the

surface composition in this work.

Core electron Binding Energy, eV Height Area

F1s 688.5 1.000 1.000

O1s 533.4 0.711 0.711

C1s 284.4 0.296 0.296

Si2p 99.3 0.314 0.339

Nb3d 202.2 1.911 2.921

Nb3d5 202.2 1.911 1.753

Al2p 74.0 0.228 0.234

Use of relative sensitivity factors allows one to relate photoelectron current intensities to the number of atoms and, ultimately, calculate relative atomic percent of elements in the sample. Concentration of a certain element A can be determined if number of atoms A and the total number of atoms in the sampling volume are known.

Because number of atoms is proportional to photoelectron current intensity measured by

XPS, concentration can be expressed through intensities and sensitivity factors as shown in Equation 9.

I A S C = A 9 A I ( n ) ∑ S n n

Where: CA = atomic percent of element A (%)

IA = intensity of element A

SA = sensitivity factor for A

In = intensity of element n

Sn = sensitivity factor for n

The absolute error that is associated with calculating relative atomic concentrations using Equation 9 is between 5 and 10 % depending on properties of a material and relative sensitivity factors that were used. However, this error can be significantly reduced (to ~1%) if relative changes in composition are used. Thus, relative changes in composition will be often used throughout this work.

Relationship of photoelectric current intensity to number of atoms in the samples was established; however, it was not explained how one determines photoelectric current intensity from XPS spectra discussed in section 2.2. The area under the cure for a characteristic range of the binding energies is a photoelectric current. As seen in Figure

5, the carbon peak is asymmetric peak of certain width; the area under the curve can be determined by a computer program. Thus, area under the curve is used in the Equation 9 for calculating relative atomic composition.

In order to determine area under the curve, satellite peaks, which are lower intensity lines produced by magnesium and aluminum due to less probable transitions

[14], and background must be subtracted for each element present in the sample.

Background is defined as any signal below the baseline. Few background subtraction methods are used for XPS data: the straight line, the Shirley method, and the Tougaard method [4]. In Shirley method the background is determined by a right-to-left integration between the two end points [15]. After subtracting the background, a computer program calculates area under the curve. Additionally, the spectrum can be fitted, as seen in

Figure 5, with components. The instrument fits individual components and calculates

22 overall goodness-of-fit. Goodness-of-fit calculated using AugerScan software used in this thesis is defined as an error-mean-squared values calculate by summing the squares of the difference between the experimental and curve fit data for each data point, which is divided by the number of data points minus the number of parameters [15]. So the smaller the number the better, with zero being a perfect fit [15]. Area under the curve for each individual component and the percent that area represents of the total peak is calculated by the software as well. Thus from the areas of individual peaks concentration of an element in any bonding state in the sampling volume can be determined using

Equation 9.

2.4.3 Peak Fitting

Curve fitting allows detailed interpretation of the X-ray spectra. It must be noted that curve fitting is not the same as deconvolution which is done to enhance energy resolution by removing instrumental broadening [4]. Second derivatives of the original scan can be taken to determine maxima of the components. Parameters such as FWHM, position, intensity, Gaussian-Lorentzian ratio can be specified and locked. The solution will incorporate “the maximum amount of operator understanding about the system” [4].

Thus, even though the perfect fit is at the goodness-of-fit value of zero, this fit may not be meaningful, so the operator needs to have an understanding of the system and the material.

Peak fitting is performed on the elemental spectra. Before fitting peaks, the satellite peaks and background need to be subtracted. In this work the Shirley background function was used to define background. After these transformations, the

23 spectrum can be fitted with the individual components. The curves are fitted using

Gaussian-Lorentzian sum function. The Gaussian-Lorentzian ratio which is used for fitting peaks is a function of the source lineshape and contribution from the spectrometer

[4]. The lineshape of the source is mostly Lorentzian, while instrumental contribution is

Gaussian [4]. For the peaks measured in the Interface Engineering Laboratory, Gaussian-

Lorentzian ratio of 80:20 was used.

The full width at half maximum (FWHM) is an important characteristic of the peaks. During fitting procedure, the FWHM can be determined based on the peak shape and the slopes. The FWHM is a function of the excitation source, analyzer resolution and material. The linewidth of the monochromated Al source is below 0.3 eV as oppose to the linewidth of 0.85 eV for the nonmonochromated source [4]. Thus, resulting peaks will have higher FWHM for the nonmonochromated source. The crystalline samples with few defects are expected to have low FWHM, while powders and polymeric films will have a high FWHM. Therefore, comparison of the XPS spectra that were obtained on the same instrument for the samples that were processed differently will give information on the material atomic order.

Position of the peaks during fitting procedure is often based on the expected chemical shifts that were observed in the past. Chemical shift is a shift in binding energy of an element caused by the existence of the different bonding states, i.e. carbon bonded to carbon appears at 285 eV or carbon bonded to fluorine at 293 eV, so the chemical shift for C-F is approximately +8 eV. Fluorine, as the most electronegative atom, exhibits the highest chemical shift for carbon. So the higher the difference between the two elements the higher the chemical shift will be. Fluorine even exerts secondary substituent effect

24 which is referred as a β shift [4]. Examples of primary and secondary chemical shifts relative to the saturated hydrocarbon are showed in Table 3. However, while assigning peaks caution should be taken, as sample charging results in peak shift to the higher binding energy. Often internal referencing is used to compensate for the sample charging.

One of the peaks which can be unambiguously assigned is chosen as a reference.

Saturated hydrocarbon C1s peak at 285.0 eV is a common correction for polymers which contain these chains [4].

Table 3: Primary and secondary shifts for C1s [4].

Functional Chemical Shift Number of

Group Min. Max. Mean examples

C-O-C 1.13 1.75 1.45 18

C-OH 1-47 1.73 1.55 5

C=O 2.81 2.97 2.90 3

OCC 3.64 4.23 3.99 21

HCO 4.18 4.33 4.26 2

C=C -0.24 -0.31 -0.27 4

C-Si -0.61 -0.78 -0.67 3

C-F - - 2.91 1

-CF2 - - 5.90 1

-CF3 7.65 7.72 7.69 2

-O-C-C ~0.2 6

F-C-C ~0.4 11

OCC ~0.4 7

OCC CH3 ~0.7 8 O

Similar to carbon, niobium was detected to be bonded to a number of species.

Values for the niobium bonded to oxygen are shown in Table 1. Unlike carbon which has been studied with XPS extensively and chemical shifts have been established and widely accepted, niobium is still being studied, and there is a range of values that are used for the same bonding state. Thus, assignment of the niobium peaks is more complicated requiring understanding of the XPS system and more extensive work with niobium itself to get a feel for what is reasonable to see in the system.

Table 4: Chemical shift values for niobium [3], [16], [17], [18], [19], [20], [21].

Bonding State Chemical Shift Number of

Min. Max. Mean examples

Nb2-O 0.8 1.0 0.9 3

Nb-O 0.9 3.0 1.6 6

Nb-O2 3.6 4.1 3.9 5

Nb2O5 5.2 5.6 5.4 7

2.4.4 Thickness Calculations

Samples containing a layer on a substrate can be analyzed with XPS to determine thickness of the overlayers from measuring attenuation of the substrate signal. For these

26 calculations it is necessary that the sample has a uniform layer as oppose to a mixture of materials and the layer is thin enough that the substrate is detected. For two chemical states that have the same element, such as silicon dioxide on silicon surface, Equation 11 can be used to calculate thickness of the overlayer. In order to be able to use Equation

11, intensity of the pure substrate and pure overlayer must be known. Thickness of niobium oxide and carbon overlayer were calculated from the XPS data, details on calculations for each case are described in section 3.2.3 for niobium and section 4.2.3 for carbon.

⎡⎛ I ⎞⎛ I 0 ⎞ ⎤ d = ln ⎜ ov ⎟⎜ sub ⎟ +1 λ sinθ 10 ov ⎢⎜ ⎟⎜ 0 ⎟ ⎥ ⎣⎢⎝ I sub ⎠⎝ I ov ⎠ ⎦⎥ Where: Iov = intensity of the overlayer element peak area

ISi = intensity of the substrate

0 Isub = intensity of the pure substrate

0 I ov = intensity of the pure overlayer

λ = inelastic mean free path (nm)

θ = take-off angle

2.4.5 Error Analysis

For the data presented in this work, the following errors were used. As described previously, the FWHM values are a function of a material and an instrument. Only elemental spectra were used in this work for the peak fitting and data analysis. All elemental spectra were obtained with the pass energy set to 35.75 eV, thus the minimum

FWHM that the analyzer could reliably determine was 0.54 eV. However, the linewidth of the Mg anode was 0.70 eV. The lowest FWHM value used in this work was 1.05 eV

27 for high RRR niobium. From experience with the instrument and the systems analyzed, we feel confident in citing the FWHM values with the accuracy of ±0.05 eV as an absolute error.

As discussed in section 2.2, calculations that allow approximation of the IMFP contain error from 8.5% up to 39%; however, most of the elements that were studied in this work were cited to agree with the measured IMFP within approximately ±13%.

Thus, we will use a slightly more conservative value, ±15% to account for the errors associated with approximation of the IMFP. This error is used in presenting thickness calculations and sampling depth values.

The relative atomic composition contains from 5 to 10% of error because of the properties of a material, instrument, and the relative sensitivity factors that are used. Due to the fact that we used sensitivity factor derived by a manufacturer, we give the relative atomic composition values with the absolute error of 7%. As mentioned above, when relative changes in composition are used, the absolute error for these measurements decreases to about 1% as the instrument contributions cancel out. Thus, the area ratios will be cited with the absolute error of 3%, which is a very conservative value.

Finally, when multiple runs of a certain experiment were done, standard deviation values are given. It is clearly specified when standard deviation for the repeatable samples is given.

2.5 Beyond Composition

So far, the physical phenomena behind XPS and its more traditional applications have been discussed. XPS analysis for determination of the relative atomic composition, bonding states, and thicknesses of the layers have been traditionally used; however, over the years this technique has been enhanced to provide additional information such as structural information from the valance band and nondestructive depth profile. Angle resolved XPS (ARXPS) takes advantage of the dependence of the sampling depth on the take-off angle. ARXPS offers a destruction-free depth profile, more sensitive surface analysis, and surface structure analysis. The valence band analysis using XPS can be used to gain structural information as well as insight of the electronic properties of the materials.

2.5.1 Angle Resolved XPS for Depth Profiling and Increased Surface Sensitivity

ARXPS takes advantage of the decrease in the sampling depth with decrease in the take-off angle. Figure 6 illustrates difference in the sampling depth for the 90° and

10° take-off angles. When the take-off angle is 90°, the electrons will travel distance z which is the same as the sampling depth (d) in the material towards the analyzer.

However, when the sample is turned so that the take-off angle is 10°, the travel distance

(z) will no be equal to the sampling depth (d). The photoelectrons for the sample turned to 10° will be generated from a considerably thinner slice of a material.

z z

Figure 6: Effect of the angle on the sampling depth [22].

Variation of the sampling depth with the take-off angle allows one to obtain nondestructive depth profiles, verify thickness of the overlayers, and information on the structure of the materials using ARXPS. Depth profile using ARXPS allows one to measure relative atomic composition at different take-off angles, thus varying the sampling depth. Figure 7 shows an example of the depth profile of the oxidized niobium sample, thicknesses of the overlayers were calculated using equation similar to Equation

10. Depth profile obtained with ARXPS can aid in determination whether the sample contains a homogeneous mixture of the elements or a layered structure. If the sample contains layers of a material, then by changing the take-off angle the different atomic composition will be determined. While for a sample with a homogeneous mixture of elements constant composition at different take-off angles will be found. For example,

Darlinski et al. in [21] detected shadowing of the oxide with metal at more grazing angles which they concluded was the effect of nonuniform oxide layers, which suggested serration of the niobium surface.

Figure 7: Depth profile of niobium sample oxidized in air [21].

2.5.2 Valence Band Offset Measurements

The valence band spectra are measured in the range between -5 and 30 eV. All of the valence electrons from the sample will be detected in this region, so resolution of the components is nearly impossible. Measuring of the valence band offsets using XPS was shown by Kraut et al. in [7]; they determining the valence-band discontinuity for

Ge/GaAs(100). Due to a lattice disruption crystal substrates will have charge deviation from that of the bulk of the material as seen in Figure 8. The energy layers at the interface will bend due to difference in the local charge densities [23]. Bend bending occurs at the interface between a material and vacuum, metal, or semiconductor, because

31 the Fermi energy level must align between the materials in contact, the other energy levels have to bend to allow flow of electrons.

Figure 8: Energy band structure of the interface layers of the semiconductor

substrate [23].

To determine valence and conduction band offsets using Kraut et al. method, binding-energy values for the core and valence levels must be known with certainty of

±0.025 eV [23]. One needs to know the difference between the core and the valence binding energy values for the bulk substrate, the bulk film, and a difference between the core levels in a thin film, according to Equation 11.

Y Y X X Xfilm Yfilm ∆Ev = (ECL − Ev ) − (ECL − Ev ) − (ECL − ECL ) 11 Y Where: ECL = binding energy of the core level of the bulk substrate Y (eV)

Y Ev = binding energy of the valence band of the bulk substrate Y (eV)

X ECL = binding energy of the core level of the bulk material X (eV)

X Ev =binding energy of the valence band of the bulk material X (eV)

Yfilm ECL = binding energy of the core level in material Y after thin film X was

deposited (eV)

Xfilm ECL = binding energy of the core level in material X after thin film of X

was deposited (eV)

The valence band energy is determined by finding the valence band maximum

(VBM). VBM can be found using a linear extrapolation method where one needs to determine intercept of the valence band peak slope with the baseline, as shown in Figure

9. Determination of the VBM is complicated by the fact that intensity of the valence peak is very low, so even after performing over 100 scans signal-to-noise is low.

Figure 9: Valence band maximum for GaAs was determined to be 0 eV [24].

2.6 Experimental Apparatus and Procedures in the Interface Engineering

Laboratory

In the Interface Engineering Laboratory, a sample is introduced into the ultrahigh vacuum (UHV) chamber through a loading dock where reduced pressure is achieved, and then it is placed using a precision manipulator in a required position with respect to the

33 electron optics entrance aperture. Dual (magnesium/aluminum) nonmonochromated source (Physical Electronics 04-548) was used for excitation in this work. These are most common anodes because of their narrow linewidth, sufficient excitation energy, ease of fabrication, and thermal conductivity [4].

Emitted electrons were detected by electron energy analyzer and a detector; concentric hemispherical analyzer (Physical Electronics 10-360) with the single channel detector is installed in Interface laboratory. Concentric hemispherical analyzer (CHA), also referred to as the spherical sector analyzer, is constructed of the two concentric hemispheres to which potentials are applied [4]. Only electrons with correct energy will be allowed to pass between the spheres to the focus towards the detector. By increasing the deflection voltage instrument scans electron energies from low to high [4]. Electrons focused in CMA enter the electron detector.

XPS was calibrated according to the international standard procedure VII ISO

15472 for calibration of the energy scales for XPS [25]. In short, copper foil, from which the oxide layer was removed by hydrochloric acid, and gold foil samples were transferred into the analysis chamber. Each anode was calibrated by first adjusting the scale factor which controls distance between two peaks. The scale factor was adjusted to obtain difference between Au4f7/2 and Cu2p3/2 of 848.68 eV and 848.67 eV for aluminum and magnesium anodes, respectively. After the correct range between peaks was achieved, work function was adjusted to obtain correct peak position. Work function affects peaks linearly, so gold was adjusted to 83.95 eV for both anodes. Copper peak was moved to

932.63 eV for aluminum anode and 932.62 for magnesium anode. CuL3VV Auger peak was used as a third point in calibration with the values of 567.93 eV for Al Kα and

334.90 eV for Mg Kα. Standard deviation on the difference between the reference values listed above and obtained values for consecutive measurements are given in Table 5.

Acceptable deviation from the reference values is less than 0.2 eV [25], when this value is reached, instrument must be recalibrated. For valence band offset analysis, calibration was performed before each experiment.

Table 5: Repeatability of the calibration for gold and copper over consecutive

measurements.

Offset Au4f7/2, eV Offset Cu2p3/2, eV Offset CuL3VV, eV

Al Kα 0.00±0.00 0.02±0.01 -0.02±0.00

Mg Kα -0.01±0.00 0.01±0.01 -0.08±0.00

Data was acquired and manipulated using AugerScan software (version 3.2 Beta) provided by RBD Enterprises, Inc. [15]. All survey spectra were obtained with pass energy of 89.45 eV and step size of 1 eV/S. An average of 5 survey spectra between 0 and 1000 eV was obtained. Elemental spectra for the individual elements were obtained using pass energy of 35.75 eV and step size of 0.05 eV/S. No data exporting is required to fit curves using AugerScan software. Once scan acquisition is competed, satellite peaks are subtracted and end points are chosen. End points are required for background subtraction. Integrated (Shirley) background subtraction was used for fitting all of the data. Background is determined by a right-to-left integration between the two end points

[15].

2.7 Applications of XPS

XPS has been used for quantitative and qualitative surface analysis for over three decades. It allows surface sensitive analysis of less than ≈ 10 nm of the sample. Relative atomic composition, overlayer thickness, and bonding states can be determined using

XPS data. XPS is currently used by researchers in the industry and in academia. For example, in metallurgy, XPS is used quality assurance; in ceramics, XPS has been used to study catalysis and natural minerals; and organic polymers are successfully identified and studied using XPS [5].

Application of XPS to three projects will be discussed in this thesis. Surface composition and oxide thicknesses were obtained of the niobium samples as part of the work in developing niobium chemical mechanical polishing process. ARXPS was used in characterizing monolayers formed on the silicon surface due to photoreaction for the field release mass spectrometer. Finally, preliminary work in determining valence band offsets of multifunctional oxide heterostructures was performed using XPS to measure the valence band maximum.

3.0 Niobium Chemical Mechanical Polishing

Superconducting radio frequency (SRF) technology is currently used in several applications, such as free electron laser, and is recommended for the International Linear

Collider [26] which will give new opportunities for fundamental research in physics.

SRF cavities for the particle acceleration applications have been studied since the 1970s because they have lower power losses and higher accelerating gradients than conducting cavities [27]. Pure niobium becomes superconducting at 9.25 K [27] which is the highest transition temperature for a pure metal [28]. Cavities made out of niobium dissipate 105 to 106 less power at liquid helium temperature than copper cavities at room temperature, so even with low efficiency of the refrigeration system niobium cavities provide savings

[27]. In addition, niobium has the highest superheating field, Hsh, for a pure metal, it is a type II superconductor i.e. gradual transition from normal to superconducting state, and it is soft which allows forming [29].

Cavities can be used for acceleration of the electrons. Figure 10 shows schematic drawing of electron acceleration. Because electrons are negatively charged, they will be attracted to the positive charge in a cavity. An electromagnetic field is applied to the cavity to induce periodic change in charge that the electron sees. Cavities are operating frequencies of about 1500 MHz and temperature of liquid helium (4.2 K).

Figure 10: Schematic representation of the operation of the niobium cavities for

electron acceleration [30].

The most common SRF cavity failures are caused by a drop in the Q-value, which is a ratio of the stored energy in the cavity to the power lost in the cavity walls per RF cycle [27], Quenches, and high residual resistance. A decrease in the Q-values can be caused by multipacting and the field emission. Multipacting or resonant electron loading is an event that occurs when secondary electrons from the cavity wall are emitted, accelerated, and redirected into the walls by the RF field due to striking primary electrons. The field emission or non-resonant electron loading occurs when electrons from the surface are accelerated and gain sufficient energy to produce heat and

Bremsstrahlung . Multipacting and field emission are both caused by contaminated surfaces [27]. Quenches and high residual resistance are results of the material defects, such as cracks and contamination [27]. Some groups were able to eliminate these issues by better understanding and addressing the interior surface quality of the cavities [26].

Saito from KEK reported achieving an accelerating gradient of 40 MV/m in a single cell cavity which is close to the fundamental limit for pure superconducting niobium cavities

[31]. The accelerating gradient is the maximum gain that a charge particle can gain in the

RF field [27]. Because of the importance of surface condition, buffered chemical polishing (BCP) and electropolishing (EP) became the two main processes used to remove damaged and contaminated surface layers. EP results in better surface condition

38 than BCP due to smoothing that occurs at the grain boundaries making surfaces an order of magnitude smoother (~0.1 µm) [32]. Currently, BCP treated cavities attain accelerating fields of about 30 MV/m [32], while application of EP results in an accelerating gradient of up to 40 MV/m [31]; however reproducibility is still an issue.

Introduction of the chemical mechanical polishing (CMP) to semiconductor devices over 20 years ago resulted in an improvement of the multi-level metallization, decreasing of processing time by substituting ineffective process such as reactive ion etching [33], increasing of the transistor packing density, and reducing metal waste [34].

The main objective of the CMP process is to planarize topography from the previous processing steps in interlayer dielectric (IDL) or to remove the overburden of one material stopping at the other material leaving a planar surface in shallow trench isolation

(STI) [33]. In both IDL and STI processes, CMP results in smooth planar surfaces. For example, a silicon wafer can be polished to an average surface roughness of 8 nm as determined for the 0.7 x 0.5 mm area [35], while roughness of the SRF cavities can be as high as 30 µm at the electron beam weld [36]. The standard CMP process is performed by pressing and rotating a flat wafer against a rotating polishing pad in the presence of a slurry containing abrasive particles and aqueous mixture of oxidizers, inhibitors, buffers, and stabilizers. The most aggressive component of a typical slurry is an oxidizer that is generally used in a very low concentration (e.g. hydrogen peroxide at 1% in Cabot EP-

B6678 slurry). The absence of the harsh chemicals is an attractive characteristic of the

CMP, especially compared to concentrated hydrofluoric and nitric acids used in BCP and sulfuric and hydrofluoric in EP.

To the best of our knowledge, niobium has not been polished by CMP. We hypothesize that based on the results with silicon and other materials used in the semiconductor processing field, the CMP process will produce a better surface finish for

SRF cavities than either BCP or EP. As CMP is currently used for flat surfaces only, designing of a different machinery will be necessary for the curved surfaces of the cavities. However, the first step in testing this hypothesis is to demonstrate the effectiveness of the CMP process to produce an average surface roughness of approximately 10 nm on a flat niobium wafer.

Chemical mechanical polishing (CMP) is so named as both the surface chemistry of the material being polished and the abrasive (or mechanical) removal of surface material is optimized to produce the smoothest surface of the desired material or materials. Thus, the chemical interactions between the Nb and the CMP slurry will need to be understood and optimized as necessary for the achievement of the smoothest and most contaminant free surface of Nb. XPS is ideal for studying the surface chemistry of

Nb with the slurry due to its surface sensitivity to detect surface contaminants in very low levels, and as the multiple oxidation states of Nb are easily differentiated.

3.1 Critical Literature Review: Niobium as a Substrate for SRF Cavities

The following sections will outline the current state of SRF cavity fabrication and the problems associated with current processing techniques. The surface of niobium has been studied with X-ray photoelectron spectroscopy; these studies will be highlighted.

CMP process and its application to semiconductors and metals will be summarized. And

properties of niobium especially in aqueous solutions will be discussed.

3.1.1. Surface requirements for the SRF cavities

Niobium has been a material for manufacturing of the superconducting cavities

since the 1960s [36]. The fundamental limit of niobium for magnetic field and

accelerating gradient are 2400 Oe and 50 MV/m, respectively; however, obtaining this

level was shown to be difficult due to low Q-values[27]. In Jefferson Lab, Free Electron

Laser cavities are operated at 15 MV/m [37]. Over the last few decades the major causes

of the cavity breakdown were identified, and they are shown in Table 6. Table 6 also

provides the current solutions for these problems that have enhanced the accelerating

field to greater than 15 MV/m [26].

Table 6: Common causes of the SRF cavity failure and current solutions [27].

Cause Problem Cavity performance Current solution

Secondary electron Resonant electron Low Q-value, Modification of SRF emission, can be loading which means that cavity shape, enhanced by surface (“Multipacting”) cavities lose more Buffered chemical contamination power in the cavity polishing (BCP),

(particulate and walls Electropolishing (EP) chemical)

Surface contamination Loss of Loss of Eddy current and

(chemical residue, superconductivity superconductivity SQUID scanning, particles), surface (“Quenching”) defect-free electron

41 roughness, weld beam welds, pure splatter chemicals for BCP

and EP, prolonged

rinsing , assembly in

clean rooms

Conducting and Residual surface Low Q-value Buffered chemical dielectric defects, resistance polishing (BCP), particulate surface Electropolishing (EP), contamination, assembly in a clean adsorbates room environment,

(hydrocarbons, residual annealing to remove gas condensation), dissolved gasses macroscopic surface imperfections, precipitation of hydrogen [27]

Micron-size particles Non-resonant Low Q-value High Peak Power

(dust), adsorbates electron loading Processing , high

(hydrocarbons), grain (“Field emission”) pressure ultra pure boundaries [38] water sensing,

efficient

contamination control

Some cavity breakdown problems such as multipacting and high residual resistance have been addressed and successfully resolved; however, field emission remains the major obstacle [26]. It is understood that field emission is caused by particulate contamination, adsorbates, and grain boundaries, so elimination of these conditions is critical. Surface imperfections and grain boundaries were observed to harbor various types of contamination [28], [39].

Importance of the surface condition can be explained by the fact that electromagnetic field penetration depth for niobium is approximately 60 nm at a frequency of 1500 mHz below niobium’s critical temperature of 9.25 K [27]. Classical surface analysis techniques such as XPS and SIMS have been used to study correlations between surface characteristics and cavity performance. Residual resistance is linked to

Q-value drop. Dissolved hydrogen and various oxides were observed to increase residual resistance [27]. Annealing process is used to degas niobium off hydrogen, and disperse material inhomogeneities [27]. It is not surprising that surface studies were performed to determine the species of niobium during the annealing process. Niobium has four known oxide states that differ in properties from each other and from the metal. For example,

NbOx (x≈1) is a metallic oxide with critical temperature of 1.38 K [40], while Nb2O5 is dielectric [41], so identification of the oxide states is important for SRF application.

Few methods, such as buffered chemical polishing (BCP), electropolishing (EP), and high pressure ultrapure water rinsing, are currently used to remove both particulate and adsorbed chemical contamination. Chemical polishing was one of the first methods used to remove mechanically damages surface layers and chemical contaminants. Nitric

43 acid is used to oxidize niobium surface, and hydrofluoric acid to remove niobium pentoxide according to the following equations [42]:

6Nb + 10 HNO3 → 3Nb2O5 + 10 NO↑ +5 H2O 12

Nb2O5 + 10HF → 2NbF5 + 5H2O 13

Nb2O5 + 10HF → 2H2NbOF5 + 3H2O 14

Overall: 6Nb + 10HNO3 + 30 HF → 6NbF5 + 10 NO↑ + 20H2O 15

Nitric acid oxidizes Nb surface producing NO gas as seen in Equation 12, while hydrofluoric acid dissolves niobium pentoxide forming soluble niobium pentafluoride or pentafluoxyniobium acid, which is a hydrated from of the niobium pentafluoride as seen in Equation 13 and 14. A mixture of hydrofluoric and nitric acids (1:1 by volume) was used to remove approximately 150 µm of the surface [36], [43]. However, this acid mixture was observed to etch preferentially at the grain boundary leading to higher than desirable surface roughness in the range of a micron [32]. In addition, grain boundaries were discovered to be responsible for the localized heating and superconductivity breakdown [36], so chemical polishing that enhanced the grain boundaries was not desirable. Thus, modifications to the standard chemical polishing were made: a buffer substance was added to reduce etching rate and cooling of the mixture below 15°C was done [43]. Phosphoric acid is most often used as a buffer in a niobium cavity cleaning.

The typical ratio used in industry is 1:1:2 (by volume) of hydrofluoric acid (40%) : nitric acid (65%) : phosphoric acid (85%) [32], [43]. This new process, called buffered chemical polish (BCP), resulted in a higher surface quality, leading to an average gradient of 26.1±2.3 MV/m [43]. However, the BCP still resulted in a strong grain

44 boundary etching and cavities that did not usually exceed accelerations higher than 30

MV/m [32].

Further improvement to Nb cavity fabrication was achieved by H. Diepers et al. at

Siemens AG in 1971 through the development of an electropolishing process

[36],[26],[31]. The electropolishing process (EP) for niobium utilizes a mixture of sulfuric (96%) and hydrofluoric (48%) acids in a ratio of 9:1 by volume [3]. During the

EP process, an electric field is applied which is enhanced at sharp points or protrusions from the surface and thus etches hills and edges at a higher rate [32], thus producing a smoother overall surface. Surface roughness of post-EP surfaces is 50 to 100 times smoother, than that of the post-BCP [42]. Figure 11 demonstrates that surface roughness increases for BCP process and decreases for EP process. Saito et al. called a buffered chemical polishing as a chemical polishing. The values used in the graph shown in

Figure 11 came from a number of publications cited by the authors. Most of these publications were the conference proceedings, so very limited experimental details were given. Despite these facts, the results presented here are important, because they allow us to relate surface roughness to material removal for BCP and EP processes. It was observed that for EP surface roughness decreased as more niobium was removed; however for the BCP, the opposite trend was true. Surface roughness of niobium samples treated with BCP was greater than that of EP for all samples after 100 µm was removed.

Preliminary results by Saito et al. in [31] of studying relationship between surface roughness and the RF field enhancement showed that surface roughness needs to be below 2 µm, as shown in Figure 12 to prevent RF field enhancement [31]. Field

45 enhancement increases 8 fold when surface roughness increases to 4 µm. Unfortunately, no further published results by Saito et al. were found.

Figure 11: EP results in smoother surface than BCP [31]

Figure 12: Surface roughness values of up to 2 µm are desirable to prevent RF field

enhancement with 1300 MHz SRF cavities [31]

In general, EP process produces cavities that perform better (up to 40 MV/m)

[31], than cavities finished by BCP (rarely above 30 MV/m) [32]. However, the reproducibility of the EP process is not satisfactory [32], and the combination of the harsh chemicals and an electric field makes it an desirable process for cavity manufacturers [44]. Eliminating surface roughness may be a key to obtaining high

46 quality cavities not only because of the work of Saito et al. described earlier, but also because of work by Septier et al. [39] who observed that contamination, such as carbon impurities, accumulates in surface defects such as scratches, holes, and grain boundaries contributing to electron emission, increase in resistance, and, ultimately, to Q-value drop.

To conclude, removal of the mechanically damaged layers of niobium and absorbed contamination can be obtained by using BCP and EP. Overall, the EP process produces a higher quality surface, which is hypothesized to contribute to higher accelerating fields (40 MV/m) [31]. However, BCP and EP processes require the use of concentrated acids, produce hazardous gases (NOx, H2, and O2). In addition, BCP is known to roughen the surface by etching at the grain boundaries.

3.1.2. Niobium: Surface Studies

At the time when niobium cavities were first designed and manufactured, XPS was already an established procedure. Thus, when cavities failed to perform as anticipated and surface condition was suspected to be the cause of failures, scientists turned to XPS to gain an understanding of the niobium surface chemical characteristics.

It is important to note that inelastic mean free path (IMFP) of the Nb electrons is 2.3 nm for Mg Kα anode, making the sampling depth of the XPS on the order of 7 nm.

One of the first thorough studies on niobium for SRF applications was performed by M. Grundner and J. Halbritter in 1980, where the effects of standard processing procedures, such as annealing in ultra high vacuum (UHV), electropolishing, oxipolishing, and exposure to air, water, and hydrogen peroxide were studied [16].

Using a Mg anode, they determined that the predominant oxide shifted 5.2 eV to the high

47 energy side of the Nb metal peak was niobium pentoxide (Nb2O5). Pentoxide was present in all samples prior to annealing. These authors were first to observe a difference in pentoxide shift from the metal peak depending on preparation conditions: samples prepared in aqueous solutions were shifted 0.3 eV to the higher energy compared to UHV prepared samples. This was attributed to the enhanced positive charging of the oxide formed in aqueous solutions due to growth conditions or by x-ray irradiation.

M. Grundner and J. Halbritter calculated an oxide thickness of ~6 nm with an angle resolved XPS (ARXPS) using the escape depth of 3 nm for all samples subjected to oxidative treatments. The first step of niobium oxidation in the UHV environment was the formation of NbO0.02 clusters of approximately 50 nm; however Grundner and

Halbritter did not provide any data to support this statement. They cited work by

Schwarz and Halbritter in [45] where comparison of the conductivity data allowed the authors to infer that NbO0.02 as clusters was present in a sample.

Below 1700°C, Grundner and Halbritter observed presence of the intermediate oxides, NbO and Nb2O, located ~2 eV and ~1 eV to the high energy side of the metal peak, respectively. Nb2O was observed to be present only after the sample was sputtered with Ar, indicating that only metallic NbO and insulating Nb2O5 were stable on the oxidized surfaces under the experimental conditions used. This does not agree, however, with some of the papers that followed, for example Ma et al. in [46] found Nb2O present on samples that were not etched with argon.

A few years later, A. Darlinski and J. Halbritter published another ARXPS study on oxidation of niobium [21]. By studying dependence of the surface composition with respect to the take-off angle, they determined that on a defect free niobium single crystal,

48 the pentoxide grew homogeneously up to thickness of about 1 nm, and then it caused serration of the metal. In one of the subsequent papers, a diagram depicting serration of niobium surface was published [47]; this diagram is shown in Figure 13. Niobium pentoxide was determined to grow on the metal surface forming irregular layers which cut into the metal surface when the sample was exposed to air for 1 week as seen in

Figure 13. In addition to determining serration of the niobium surface due to oxidation,

Darlinski and Halbritter found regions of NbO0.2 with ARXPS on niobium surface. The authors stated that due to enhanced resolution they were able to detect NbO0.2; however, they used Mg Kα anode, an analyzer with a pass energy of 65 eV and 1 eV energy resolution, and they took average of 2-8 scans for each sample [21]. Parameters that the authors used to run their XPS analysis do not allow high resolution spectra to be obtained. In addition, the chemical shift between Nb and NbO0.2 was determined to be

0.7 eV which overlaps with the Nb2O region as found in [16], [18], [20], [19].

Figure 13: Schematic of the serration of niobium surface due to oxidation in air for

over one week [47]

Halbritter in his subsequent paper showed dependence between residual resistivity ratio (RRR) and surface chemistry [47]. RRR is a measure of a purity of a material and is

49 defined as a ratio of resistivity values at room temperature and 0 K [29]. According to

Bonin, RRR = 300 corresponds to the “state-of-the-art material” with C, N, and O present at about 100 at. ppm, and tantalum at 1000 at. ppm [29]. Post purification of niobium allows RRR improvement from 30 - 300 (reactor grade) to 200 – 2000 (state-of-the-art)

[29]. Halbritter determined that in the Nb samples with RRR of over 100 an incomplete monolayer of NbO pas present [47]. NbO is thought to be cause increase in resistance of

SRF cavity due to the fact that it is a metallic oxide with a low critical temperature.

A study of annealing at lower temperatures (30 - 1000°C) by Dacca et al. showed that Nb samples (RRR~200) with either BCP or anodization surface treatments resulted in qualitatively same surface behavior during and after annealing. First, in the temperature range between 280° C and 380° C the Nb2O5 doublet disappeared and the metal (202.3 eV) and NbO2 (206.1 eV) appeared. Between 380° C and 1000 ° C NbO

(203.2 eV) was detected [17]. Using ARXPS, the authors determined Nb2O5 to be approximately 60 Å thick for an anodized sample at room temperature before heat treatments [17]. Dacca et al. reported to have removed all oxygen from the surface when they annealed above 1600° C, unlike in previous studies, for example [16],[47] where

Nb2O and NbO were on surfaces during annealing below 1700° C. This indicates that the niobium samples do not lose oxide at one set temperature which may be due to different crystal structure and purity of the samples used.

Q. Ma and R.A. Rosenberg published a number of studies on the annealing of niobium. In their work, which was presented at the 10th Workshop on RF

Superconductivity in 2001, they showed that electron beam irradiation resulted in the conversion of Nb2O5 to NbO2 thus agreeing with Grundner et al. work on annealing of

50 niobium [16]. However, unlike Grundner et al. in [16] and Dacca et al. in [17] they did not identify NbO2 or NbO in the oxidized samples saying that the existence of these peaks was inconclusive [46]. Their subsequent ARXPS study was focused on atmospheric oxidation of niobium surfaces treated with BCP, annealing at 250°C in

UHV, and then re-oxidation in air [18]. The cavities with BCP treated surfaces oxidized rapidly at the rate of 0.5 Å/min for the first 100 min, thus supporting hypothesis that Nb oxidation follows the field-assisted mechanism (i.e. oxidation is promoted by an electric field between the adsorbed negative oxygen ions and the metal) [18]. Ma and Rosenberg deconvoluted the Nb3d peak into Nb (202.2 eV), Nb2O (203.0 eV), NbO (203.6 eV),

NbO2 (206.3 eV), and Nb2O5 (207.6 eV) [18] which agrees with both Grundner [16] and

Dacca [17]. However, their average total oxide thickness of approximately 4 nm [18] was lower than thicknesses previously reported by Grundner et al. in [16] and Dacca et al. in [17]. Prior to Nb2O5 disappearance, NbC was formed on the surface. After annealing the BCP treated surfaces at 250°C, the Nb2O5, NbO2, and NbO disappeared.

The thickness of the remaining Nb2O was 3.7±0.2 nm [18]. Re-oxidation of the annealed samples started at the rate of 0.8 Å/min, leading to a decrease in Nb2O thickness and the formation of Nb2O5; NbC disappeared as well indicating that Nb2O5 is more stable at atmospheric conditions [18]. Carbon content remained unchanged during annealing and re-oxidation. However, the position of the Nb2O5 formed in air was 0.4 eV lower than the

Nb2O5 observed before annealing [18], which does not agree with Grundner et al. who observe a shift of 0.3 eV for samples oxidized in aqueous environment [16]. This indicates that other than aqueous method of oxidation may result in the shift of the

51 pentoxide peak, and that hypothesis of a positive charging build up due to the growth conditions is probably not correct [16].

Ma and Rosenberg also found that Nb2O peak shifted to the higher energy by ~

0.4 eV during re-oxidation, so the authors stated that assignment of Nb2O based on the binding energy was arbitrary [18]. This indicates that identification of niobium 3d peak components based on their position is not necessarily correct as it would be with other elements such as carbon. Finally, and similarly with Grundner [16], Ma and Rosenberg concluded that the re-oxidation occurs by oxygen from the bulk diffusing to the surface thus forming higher oxidation states.

At the Argonne National Laboratory in 2004, Ma et al. also studied the effect of the annealing on the properties of Nb (100) sample [20]. Glancing-incidence XPS

(GIXPS) and ARXPS were used to characterize the pre- and post-annealed surfaces. The

Advanced Photon Source (undulator beam line) at the Argonne National Laboratory was used for these experiments. The BCP treated Nb samples were oxidized in the air and then annealed to 430 K and 540 K under ultra-high vacuum (UHV) conditions [20]. The oxide thickness for the sample before annealing was determined to be approximately 3.3 nm using ARXPS. During annealing, some of the niobium pentoxide converted into

NbO and NbO2 forming a continuous layer under the pentoxide as determined by GIXPS i.e. the amount of NbO2 increased with incidence angle suggesting a conformal layer.

Ma et al. concentrated on the annealing of the BCP treated Nb surfaces, and

Kowalski and co-workers studied effect of annealing on both the BCP and the EP treated

Nb surfaces [3]. In the study by Kowalski et al. the Nb surface was ground off prior to

BCP treatment resulting in the average surface roughness (Ra) for the ground + BCP

52 treated samples of 0.1 µm. The Nb surface was not ground off prior to the EP process, and the Ra for EP-only treated samples 0.2 µm [3]. Higher roughness values for the EP samples is surprising, because generally the EP process results in a smoother surface than one produced by the BCP process [32],[31]. After exposure to air for a few months, all samples used by Kowalski et al. were annealed and then analyzed by XPS. The researchers detected the following species post anneal on the surfaces: Nb (201.6 eV),

NbO (203.1 eV), NbO2 (205.2 eV), and Nb2O5 (207.2 eV) [3]. NbC was found on all surfaces after annealing. After argon sputtering, NbC remained on the surfaces while all oxides were removed. At high temperature (above 140° C) and long time of annealing

(duration was not specified by the authors), the NbO peak shifted to 203.5 eV, and the

NbO2 peak to 205.0 eV. Peak shifting of the NbO and NbO2 was attributed to the lower stability of these phases [3]; however, it is important to note that these researchers did not identify Nb2O peaks (3d3/2 and 3d5/2). The missing peaks would have overlapped with one of the doublets of the NbO and NbO2 peaks, thus possibly removing the shift altogether. In addition, the full width at half maximum (FWHM) for the pentoxide peak increased from 1.8 to 2.4 eV. These results indicate an increase in crystallographic defects of pentoxide due to annealing [3].

Another synchrotron study was performed by Tian et al. Their more recent (2006) work investigated the effect of a solution flow rate during BCP treatment on the surface, topography, and repeatability [19]. During BCP treatment, the BCP solution is flowing through the cavity resulting in different surface flow rates at different locations within the cavity. Different flow rate may lead to different surface characteristics throughout the cavity, which may be an important factor in SRF cavity performance. In their

53 experiments, Tian et al. attached small samples of niobium to the Teflon holder that was rotated creating desired flow rates. After the two consecutive BCP treatments the average roughness of the niobium samples was determined with stylus profilometry of the 2 mm x 2 mm fields to be approximately 0.60 – 0.70 µm, with a significant 30% error based on the standard deviation of the samples obtained from different batches [19].

While the general grain orientation was determined to be (100) by the electron backscatter diffraction in the scanning electron microscope; other grain orientations were present as well [19].

From varying the photon energy during XPS analysis, Tian et al. determined that the pentoxide thickness was between 3.0 and 4.4 nm. Variation of the photon energy resulted in greater surface sensitivity than the variation of the take-off angle as is available in the standard XPS systems. As for the composition of the films, the authors followed peak assignment used by Ma et al. in [18] who did not use synchrotron source.

Interestingly, the variation of the solution flow rate did not affect the surface roughness after BCP treatment. However, the oxide thickness was determined to be higher for the samples that were processed at a higher flow rates, with up to 40% thicker oxide when processed at a solution flow rate of 1.95 in/s than when processed in a stagnant solution [19]. This suggests that surface composition may be significantly different at the equator or iris of the niobium cavities.

Niobium has been used for superconducting cavities for over 40 years, over this period of time multiple surface studies were performed which resulted in understanding of niobium properties and its surface composition. Figure 14 shows current understanding of the structure of the niobium surface. It is believed that niobium is

54 covered with a thin (0.5 nm [41]) layer of NbOx (x≤1) and by a thicker layer from of

Nb2O5-y (y<1). Thickness of Nb2O5 varies depending on surface treatment that niobium sample received, for example Ma et al. determine pentoxide thickness to be 3.3 nm [20] for BCP treated samples that were oxidized in air, while Dacca et al. found 6.0 nm of niobium pentoxide on anodized surfaces [17]. During oxidation, niobium was believed to crack forming irregular layers or oxides. In addition, oxygen was determined to dissolve in niobium bulk forming localized regions of NbO0.02 [16],[47],[41]. The outer surface on niobium is covered with hydrogen-bonded carbon and oxygen containing species [41].

Many researchers determine intermediate oxides (Nb2O and NbO2) on the surface as well

[20],[3],[16],[19],[48].

Figure 14: Niobium oxides are hypothesized to serrate niobium surface [41]

In literature published, interpretation of the niobium 3d spectra does not agree.

Peak positioning for most of the niobium oxidation states fall in ranges and the ranges

0 often overlap, as it can be seen for Nb , Nb2O, and NbO in Figure 15. Binding energy values for various species of niobium were published in a few articles, in Figure 15 these values are grouped by oxidation state. For example, NbO was found in a range between

203.1 and 204.0 eV, and Nb2O5 between 206.2 and 207.7 eV. Ma et al. assigned Nb2O

55 and NbO 0.4 eV apart [18]. In addition, few researchers found that peak position shifted due to heat treatment [3],[18]. Differences in values may be due to different processing conditions, purity and crystal structure of niobium, and XPS specification. These facts are important for anybody who is trying to use XPS for niobium characterization, because caution should be taken in assigning peaks based on the literature published values. And only the researcher who knows his system (material and XPS equipment) will be able to identify the peaks reasonably and correctly.

208

207

206 A 205 B, C D 204 E 203 F G 202 Binding Energy, eV

201

200 Nb Nb2O NbO NbO2 Nb2O5

Figure 15: Values for Nb and its oxides found in the literature. A= [20], B= [18], C= [19],

D= [17], E= [16], F= [3], G= [21]

To conclude, XPS was successfully applied for niobium surface characterization.

Niobium oxidizes via electric field assisted, Cabrera-Mott mechanism [18], [49] forming thick (~4 nm) niobium pentoxide layer. It was found that up to five oxidations states of niobium can be detected; however, some uncertainty exists regarding peak assignment.

Overall, XPS was shown to be an invaluable technique that allowed monitoring of the

56 niobium surface composition during annealing and surface treatments such as BCP, EP, and oxidation in air.

3.1.3. Chemical Mechanical Polishing

Chemical mechanical polishing (CMP) has been used in the semiconductor industry since the 1980s for the polishing thermally produced silicon dioxide layer prior to metal deposition. CMP is currently used in multiple steps such as interconnect polishing, shallow trench isolation (STI), and damascene. In a semiconductor manufacturing process CMP is used up to ten times [33]. CMP was originally developed to replace difficult to control (etch rate, uniformity, and selectivity) reactive ion etching process for the removal of the excess tungsten [33], [34]. The CMP process removes material by pressing a rotating wafer face down on a slurry-covered polishing pad which rotates in the opposite direction as seen in Figure 16. The pad has grooves which allow delivery of a slurry under the wafer. Abrasive particles from a slurry are trapped in the pad asperities and result in a mechanical material removal. A pad needs to be conditioned i.e. roughened to provide enough asperities for trapping of a slurry. Slurry provides lubrication and temperature control during the process, so that wafers can be polished to a desired smoothness. Many variables affect polishing rate during CMP process, so it is hard to generalize what conditions result in increase in material removal.

Wafer carrier with wafer facing Slurry down Polishing pad

Figure 16: Schematic representation of CMP process

Because the key objective of CMP is material removal, many models were proposed to describe CMP process [33]. One of the simplest modeling equations used is

Preston’s Law, as seen in Equation 16 [33]. Here, pressure and velocity give information on mechanical processes that take place during CMP. All of the chemical components, such as surface properties of the wafer, abrasive particle characteristics, slurry and polishing pad properties, are combined in a Preston’s coefficient. This is an empirical law which can be used only as approximation [33].

RR = K p ⋅ P ⋅V 16 Where: RR = removal rate (m/s)

2 Kp = Preston’s coefficient (m /N)

P = local pressure on the wafer surface (N/m2)

V = relative velocity of the point on the surface of the wafer versus the pad

(m/s)

In order to continue making more powerful computing systems at smaller scales, integrating novel materials in semiconductor chips is needed. Research on the CMP process has resulted in the state-of-the-art processes for obtaining rapid and selective material removal of variety of compounds by tuning both the chemical and mechanical aspect of CMP [34]. For example, current tungsten removal rates in STI process exceed

58 silicon dioxide removal by 150 to 1 [34]. Such specificity of the tungsten CMP is a result of carefully designed CMP process, especially the slurry chemistry and particle size. In order to design a slurry for a specific material task, one needs to have an understanding of surface and colloidal sciences, rheology, and electrochemistry [50]. Particles, oxidizers

(especially for metal CMP), buffers, stabilizing agents, inhibitors, and surfactants are the common ingredients in CMP slurries. It is important to point out that slurries used in

CMP are not hazardous. For example, the copper slurry iCue®5001 from Cabot

Mictroelectronics has the HMIS rating of 1 – Health, 0 – Flammability, 0 – Reactivity, and B – Personal protective equipment [51]. The role of each ingredient is outlined in

Table 7.

Table 7: Role and examples of the main slurry components.

Ingredient Role Example

Abrasive particle To remove chemically Silica, alumina, ceria,

treated surface exposing manganese oxide [52]

untreated material for

chemical treatment [50]

Surfactants To increase solubility of Alkyl sulfates, poly(acrylic

compounds with low acid) [52], alkyl amines

solubility [50] [52], carboxylic acid [52],

quaternary ammonium salt

[52], fluorinated carboxylic

acids [53]

Buffers To provide electrostatic Nitric acid [52], citric acid

stabilization of the slurry [52]

[50]

Inhibitors To prevent negative side Benzotriazole [52],

reactions during CMP hydrogen phthalate [53, 54],

process, prevent metal alkyl amines [52]

corrosion [50]

Oxidizers To passivate or oxidize Oxalic acid [55],[52],

surface [50] hydroxylamine, hydrogen

peroxide[52], nitric acid

[52], urea hydrogen

peroxide [56]

Stabilizers To inhibit reverse reduction, EDTA[52], ethanol amines

to bind to charged particles [52], citric acid [52]

thus decrease double layer

thickness on the particles

[50]

Metal CMP is different than dielectric CMP in a way that the metal surfaces require an oxidizing agent in the slurry to form surface oxides that protect the metal from unwanted corrosion in aqueous environment and are softer than the unoxidized metal surface. So oxidizer on one hand produces softer surface for increased material removal, and on the other it protect the surface from corroding. Therefore, we would expect an oxidation of niobium to be an important aspect in developing a niobium CMP process.

Unfortunately, no slurries are available for niobium, moreover to the best of our knowledge there is no other group trying to develop niobium CMP. Thurs, as shown in pervious CMP development for metal [34], slurry design is anticipated to be necessary in order to successfully polish niobium and obtain acceptable removal rates and quality.

3.1.4. Niobium chemical properties and Pourbaix diagrams

The CMP process requires understanding of the chemical and mechanical properties of the material to be polished. Therefore, one needs to understand chemical and mechanical properties of niobium in order to develop successful CMP process.

Corrosion of niobium in various solutions and its bulk mechanical properties has been known for a few decades [57], [58]. Surface chemistry of niobium has been studied in the context of SRF cavities after annealing and oxidation, for example [17], [18], [27].

However, surface properties and condition of niobium in aqueous solution during mechanical abrasion are not known.

Niobium has an interesting history. Up until 1950, niobium was called columbium in the U.S.A. while niobium in Europe. Originally niobium had been discovered in a mineral called columbite in 1801 by Charles Hatchett who named the mineral columbium [59]. Until 1846 niobium and tantalum were thought to be one metal

[59]. In Greek mythology Niobe was a daughter of the God Tantalus. Because of the similarity to tantalum, Heinrich Rose who separated two metals in 1846, named this different metal, niobium [59]. Pure niobium was obtained in 1866 by the Swiss chemist

Jean-Chaerles Galissard de Marignac [59]. Finally after a century of calling the same element by a different name, in 1950 the name columbium was changed to niobium at the

61 meeting of International Union of Pure and Applied Chemistry (IUPAC). However, even now niobium is called columbium in metallurgy.

Niobium and tantalum have very similar properties; their high corrosion resistance properties make them an attractive material or additive for products exposed to harsh environments. Niobium is more abundant than tantalum, so it is used as a substitute for tantalum in many applications. For example, it is currently used in high- strength low-alloy steel products for high stress structures such as bridges, in special alloys for jet aircraft engines and power generation turbine blades, and for applications requiring a superconducting metal such as in SRF cavities [60].

SRF cavities have been manufactured from niobium since 1960s due to niobium’s high critical temperature of 9.25 K [27]. Niobium is a corrosion resistant metal and is only affected by concentrated hydrofluoric, sulfuric, and nitric acids. Thus these are the acids that are used to chemically remove damaged and contaminated surface layers from niobium surface. Hydrofluoric and nitric acids are used in CP and BCP, while sulfuric and hydrofluoric acids are used in EP. Nitric and sulfuric acids oxidize the niobium surface to form various oxides, and hydrofluoric acid etches the oxide. CP, BCP, and EP utilize the competing processes properties of these acid mixtures to form and dissolve the oxides in order to both remove surface layers which are damaged or contaminated, and to provide continuous surface renewal.

It was stated earlier that to the best of our knowledge there is no published literature on the CMP of niobium. It was also shown that in the standard CMP process, metals require oxidation to protect their surface from unwanted corrosion and allow for the increased material removal due to lower the hardness of metal oxides. The hardness

62 of niobium and its oxides can be deduced from their density values. Table 8 shows the density for niobium and three of its most abundant surface oxides. It is thought that a niobium metal is covered with oxides of decreasing density in order to relieve crystallographic mismatch [20], [18]. The density of the pentoxide is approximately half of that of the metal; therefore the pentoxide should be on the surface and softer than the metal or other oxides and hence the formation of niobium pentoxide could be important for the niobium CMP.

Table 8 Density of niobium and its oxides.

Nb NbO NbO2 Nb2O5

Density, g/cm3 8.57 [18] 7.26[18] 5.90 [58]. 4.60[18]

Marcel Pourbaix studied the thermodynamic equilibrium of different elements in aqueous solutions and constructed diagrams, which are now called Pourbaix diagrams.

Figure 17 shows a Pourbaix diagram of the niobium. This diagram plots boundaries of predominant species with respect to pH and potential of the solution. Transitions across horizontal lines involve exchange of electrons, while transitions across vertical lines require addition or removal of hydrogen ions. Lines at the angle involve exchange of electrons and hydrogen ions. One can think of the Pourbaix diagram as a phase diagram with pH and potential as axis. Thus, the Pourbaix diagram gives information on multiple reactions and thermodynamically favored states with respect to potential and pH of a solution. Several regions of stable niobium species can be seen in Figure 17: Nb, NbO,

- + NbO2, NbO3 , Nb2O5, and Nb(OH)4 .

Figure 17: Pourbaix diagram for Nb-H2O system at 25, 75, and 95°C [61]. Lines a

and b represent hydrogen evolution and oxygen reduction, respectively.

How niobium samples behave in an equilibrium solution of a given potential and pH can be seen in the Pourbaix diagram in Figure 17. For example, at potential of -1.5 V and any pH, niobium is in a metal form. At this potential, niobium is immune to oxidation and corrosion. Most metals are immune to corrosion at negative potential, and cathodic protection uses this principle to protect metals which are exposed to corrosive environments such as bridge supports in salt water. If the potential is increased from -1.5

V, niobium will first oxidize to form niobium (II) oxide and then niobium (I) oxide.

NbO2 will undergo oxidation to a +5 state, but different species will be formed depending on the pH of the solution. At a high potential and in a highly acidic (pH < 0.5) environment, a protonated form of a niobium hydroxide will exist; while in a basic (pH >

6.5) environment, a deprotonated form of a niobium acid is thermodynamically favored.

The solution potential can be measured as an open circuit potential using a potentiostat

64 where a piece of niobium is used as the working electrode. Therefore, by measuring the pH and potential of the aqueous CMP slurry solution, one can predict the thermodynamically stable Nb surface state. In addition, one can predict the impact of adjusting the pH and potential of the CMP slurry on the thermodynamically favored surface states and the potential change in the material removal rates. Thus, the chemical composition of the slurry may be the key to the successful niobium polishing. Analysis of the niobium surface using XPS will be a primary method of determining which slurries and what treatments to use for the CMP experiments.

3.1.5. Summary

Current methods for SRF cavity manufacturing seldom produce cavities with close to theoretical performance they require hazardous chemicals and they lack repeatability. A safer process to obtain contamination-free smooth surfaces is needed.

CMP, an established process which results in repeatable surfaces in the semiconductor industry, is proposed to be an alternative method of obtaining smooth, contamination-free surfaces. To the best of our knowledge, niobium has not been polished with CMP. We hypothesize that niobium can be polished by CMP to a high surface order and smoothness that will improve performance of the SRF cavities. The first objective in achieving this goal is to demonstrate the feasibility of the CMP process on a flat niobium wafer to achieve an average roughness on the order of 10 nm. As the impact of slurry chemistry on the removal rates and surface finish is historically important in metal CMP processes, the effect of various treatments on niobium surface will be studied using XPS to determine oxidation states of niobium, oxide thickness, and atomic order of the surface.

3.2 Experimental Section: Niobium CMP

Static and CMP experiments have been performed. For the static experiments, solutions containing hydrogen peroxide and hydrofluoric acid have been used to study oxide growth and dissolution. In the CMP experiments, various slurries were tested to determine their effect on the material removal rates. The resulting surfaces were characterized with XPS, scanning electron microscopy (SEM), and atomic force microscopy (AFM).

3.2.1 Materials

Methanol (electronic grade), trichloroethylene (TCE) (electronic grade), acetone

(OPTIMA or electronic grade), hydrochloric acid (37 %, trace metal grade), hydrofluoric acid (49%, trace metal grade), ammonium hydroxide (30%, OPTIMA grade), and hydrogen peroxide (30%, certified ACS grade) were purchased from Fisher Scientific.

Reverse osmosis deionized water (RODI) was obtained from the central RODI water line in Egan Research Center. Ultrahigh purity argon was purchased from Med-Tech Gas

Company. Polycrystalline, monocrystalline, and high residual resistivity ratio (RRR) niobium wafers were provided by H.C. Starck Company.

Copper slurry, EP-C678, was purchased from Cabot Microelectronics, Aurora, IL.

Oxide slurry, Microplanar CMP115, was purchased from EKC division of DuPont,

Hayward, CA. Oxide slurry, Nalco 2398, was purchased from Nalco Company,

Naperville, IL. Alumina slurry NA-1005-128, and alumina suspensions NANO-1010-

128 and NANO-1105-128 were purchased from Pace Technologies, Tuscon, AZ.

Properties of the slurries used in the CMP experiments are shown in Table 9.

Table 9: Properties of the slurries used.

Slurry Supplier Type of particle and Concentration Hardness of name its size and pH the particle

Cu-slurry Cabot Colloidal silica pH=9.8 6-7[62]

Microelectronics 90 -140 nm

EP-C6678

Oxide- EKC/DuPont, Colloidal silica pH=10.5 6-7[62] slurry Microplanar 50-100 nm

CMP1150

Oxide- Nalco Company, Colloidal Silica 28% weight 6-7[62] slurry Nalco 2398 70-100 nm pH=10.5

Alumina Pace Technologies Polycrystalline 20% weight 9 [63] slurry NA-1005-128 Alumina pH≈7

50 nm

Alumina Pace Technologies Calcinated Alumina 20% weight 9 [63] suspension NANO-1010-128 1 µm pH≈7

Alumina Pace Technologies Calcinated Alumina 20% weight 9 [63] suspension NANO-1105-128 0.5 µm pH≈7

3.2.2 Experimental Procedure: Degrease, Oxidizing and Etch Treatments, BCP,

CMP

Degrease Procedure

All beakers with solvents were placed in the sonicator. First, niobium pieces were placed into a Teflon basket which was submerged into a beaker with TCE at 70°C. After

5 min, the basket was transferred into a second beaker with TCE at 70°C for 5 min. The sample was not allowed to dry between the steps by rinsing with acetone or methanol during transfer. The basket was then transferred into a beaker with acetone at 55°C for 2 min. The sample in the basket was then placed into the first beaker with methanol at

70°C for 5 min. Upon completion, the sample was placed into a second beaker with methanol 70°C for another 5 min. The sample in the basket was rinsed with methanol for

5 s and blow-dried with UHP argon.

Oxidizing Treatment

Degreased dry sample was placed into a Teflon basket. Then the basket with sample was placed into the hot (80°C) modified standard clean 1 (SC1) solution (30%

NH4OH: 30% H2O2: DI H2O = 1: 1: 5) for desired amount of time. The sample was then rinsed with RODI water for 15-30s and blow-dried with UHP argon.

Etch Treatment

Extreme caution was taken when performing BCP procedure due to use of concentrated acids (nitric and hydrofluoric) and evolution of gasses (H2 and NO) during

BCP treatment.

Degreased dry piece of niobium was placed into a Teflon basket which was placed into a Teflon beaker containing 5 mL of hydrofluoric acid, 17 mL of nitric acid,

68 and 51 mL of methanol. Upon completion, basket was placed into a beaker containing

RODI water for 15s. Basket was removed from water and sample was dried with UHP

Ar.

BCP Procedure

Extreme caution was taken when performing BCP procedure due to use of concentrated acids (nitric and hydrofluoric) and evolution of gasses (H2 and NO) during

BCP treatment.

Degreased dry piece of niobium was placed into a Teflon basket which was placed into a Teflon beaker containing 5 mL of hydrofluoric aicd, 5 mL of nitric acid, and

10 mL of phosphoric acid (1:1:2 by volume). Beaker with sample was placed into a sonicator, where it remained for the desired amount of time. Upon completion, basket was placed into a beaker containing RODI water which was placed into a sonicator for

30s. Basket was removed from water and sample was dried with UHP Ar.

CMP Process Procedure

George Calota, a graduate student from Mechanical Engineering department at

Northeastern University, performed CMP polishing on a G&P Technologies Poli-500

(Soeul, Korea). A summary of all CMP experiment is given in Table 10. George obtained surface profiles using a Zygo NewView 6000 (Hartford, CT) optical profiler every 1 to 4 minutes during polishing. Before each surface profile imager and mass measurements, he rinsed the wafer with DI water and dried it with nitrogen gas. Once the surface roughness stopped changing between polishing intervals, 1 x 1 cm samples were cut from three regions of the wafer (center, edge and halfway between) using a Dremel and a cutoff wheel.

Table 10: Slurries used during CMP experiments.

Experiment Slurry Type of particle, size pH Time, min

1 Microplanar CMP1150 Silica, 50-100 nm 10 16

2 EP-C6678 Silica, 90-140 nm 9.8 16

3 EP-C6678 Silica, 90-140 nm 9.8 16

4 EP-C6678 Silica, 90-140 nm 9.8 16

5 NANO-1010-128 Alumina, 1 µm 7 10

NANO-1105-128 Alumina, 0.5 µm 7 47

Microplanar CMP1150 Silica, 50-100 nm 10 14

6 EP-C6678 Silica, 90-140 nm 9.8 16

7 NANO-1010-128 Alumina, 1 µm 7 10

NANO-1105-128 Alumina, 0.5 µm 7 20

Nalco 2398 Silica, 70-100 nm 10.5 40

8 NANO-1010-128 Alumina, 1 µm 7 12

NANO-1005-128 Alumina, 0.05 µm 7 18

9 NANO-1010-128 Alumina, 1 µm 10 12

NANO-1105-128 Alumina, 0.5 µm 10 14

NANO-1005-128 Alumina, 0.05 µm 10 18

10 NANO-1010-128 Alumina, 1 µm 7 12

Nalco 2398 Silica, 70-100 nm 10.5 10

11 NANO-1010-128 Alumina, 1 µm 7 12

NANO-1005-128 Alumina, 0.05 µm 7 28

12 NANO-1010-128 Alumina, 1 µm 10 12

NANO-1005-128 Alumina, 0.05 µm 10 19

3.2.3 XPS Analysis: Data Acquisition and Manipulation

Samples after treatments explained in Section 3.2.2 were analyzed using an XPS

(X-ray Photoelectron Spectroscopy) manufactured by PHI (model 04-548 Mg/Al dual anode, non-monochromatic x-ray source and a 10-360 hemispherical analyzer). All post-

CMP samples were degreased according to the degrease procedure explained in Section

3.2.2 prior to XPS analysis.

Samples were loaded onto a sample holder which was placed into a loading dock where pump down to 9.0·10-7 Torr was performed prior to transfer into the analysis chamber. During XPS analysis, the background pressure in 10-9 Torr was routinely attained. Analysis was performed using Mg anode (1253.6 eV). Data acquisition and manipulation were performed using AugerScan software (Version 3.2 Beta), RBD

Enterprises, Inc.

All survey spectra were obtained with pass energy of 89.45 eV and step size of 1 eV/S. An average of 5 survey spectra of 0 to 1000 eV was obtained for the as-received samples and samples in static experiments. An average of 5 survey spectra of 0 to 1200 eV was obtained for the samples after CMP polishing. Elemental spectra for the individual elements were obtained using pass energy of 35.75 eV and step size of 0.05 eV/S. Silica and alumina were detected on some of the samples after CMP polishing, while fluorine and phosphorus on some of the samples after BCP treatment. Table 11

71 shows elements that were detected on various niobium samples and setting used to obtained elemental spectra of these elements.

Table 11: Typical energy ranges and number of sweeps obtained for niobium

samples after static and CMP experiments.

Element, peak Energy Range, eV Number of sweeps

Niobium, Nb3d 198-217 35

Oxygen, O1s 525-540 15

Carbon, C1s 281-293 30

Silicon, Si2p 96.5-110 25

Aluminum, Al2p 68-88 25

Fluorine, F1s 680-700 10

Phosphorus, P2p3 127-144 10

After elemental spectra were obtained, the background was subtracted using

Shirley (integrated) background function. Peak fitting was performed using Gaussian peak function with the Gaussian component of 80% and Lorentzian component of 20%.

Relative atomic percent for the samples was calculated using sensitivity factors listed in

Table 2.

Nb3d spectra were first fitted with metal and pentoxide peaks. Doublets were set

2.70 eV apart and ratio of the areas under the Nb3d3/2 peak to Nb3d5/2 was set to 0.66:1.

Two additional pairs of peaks were then added with doublet split of 2.70 eV and the peak area ratios of the doublet components if 0.66:1. Peak position for intermediate oxides was corrected to give chemical shift to the higher energy from the metal peak of 1 eV for

NbOx (1≤x≤2) and 3.6 eV for NbO2. Then the full width at half maximum (FWHM) for the metal and pentoxide peaks were determined. The FWHM for the metal peak was found by matching the overall fit to the slope of the experimental data by changing

FWHM and intensity of the Nb metal 3d5/2 peak. Similar, procedure was performed to fit niobium pentoxide. FWHM was changed to obtain the best fit for Nb2O5 3d5/2 and 3d3/2 peaks.

Thickness of niobium oxides was obtained by converting the atomic percent to a volumetric percent using the density values of metal and different oxides. Then volumetric percent can be related to the thickness. It is assumed that the oxide forms a uniform layer on the surface. For example, Figure 18 shows a niobium spectrum which was curve fitted using AugerScan software. Nb3d5/2 peak is the main component of the niobium doublet, so only Nb3d5/2 peaks are used to calculate thickness of the overlayer.

Four niobium binding states were identified. The areas under the curve for the 3d5/2 peaks (assigned on the graph) are shown in Table 12.

Nb5+

Count Nb0

Nbx+

Nb4+

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 18: Nb3d spectrum

Table 12: Area under curves indicated in Figure 18.

Peak Density, kg/m3 Area under the curve, arbitrary units

Nb0 8570 1228

Nbx+ 7260 (use density for NbO) 877

Nb4+ 5900 370

Nb5+ 4600 3828

If the whole sampling volume was composed of the metal, then every niobium electron would have to come from the metal. By knowing density of the metal, one can calculate how many niobium atoms are in the sampling volume (6.68·10-7 cm3) –

3.71·1014 atoms. The same calculations are done assuming that the sampling volume

74 contained only oxides in the same ratios as it is in my sample. Weighted average density of this film at this composition is 5154.4 kg/m3. This density can be related to the number of moles, and the number of niobium atoms is calculated to be 1.59·1014 atoms.

Note: weighted average of molar masses is used.

In this example, 80.5% of electrons detected were ejected from different oxides.

By assuming some oxide thickness as a starting point, I can calculate how many atoms would be in the sampling volume with this thickness and % of oxides. Letting % of oxides converge with the actual number, 80.5% in this case, I obtain oxide thickness of

4.6±0.7 nm.

3.2.4 SEM and AFM Analysis

Surface morphology after BCP and CMP treatment was determined using field emission scanning electron microscopy (FE-SEM) performed on a Hitachi S-4700 FE-

SEM with accelerating voltage 2 kV, beam current 10 µA, and 12 mm working distance in secondary electron imaging mode. Surfaces obtained in Experiments 5, 7, and 10 were analyzed with atomic force microscopy (AFM). AFM analysis was performed on a

Dimension 3000 Nanoscope IIIa AFM (Veeco Metrology Group, Digital Instruments) using silicon pyramidal tips in tapping mode.

3.3 Results and Discussion: Niobium CMP

Both static and dynamic experiments were performed to develop the chemical mechanical polishing process of niobium. As described in the pervious section, the static experiments did not involve the CMP machine. Data from the static tests allowed us to choose slurries to be used in the dynamic tests. Buffered chemical polishing (BCP) was also tested statically to determine its effect on the surface composition and morphology.

Results of the BCP experiments can be found in Appendix B. These static experiments allowed us to narrow the number of experiments run dynamically on the CMP machine.

There were two stages of dynamic tests; the first stage helped to determine the relative impact of changes in slurry composition and these results were used to formulate a controlled CMP study to determine the effects of slurry pH, hardness of the particle, starting surface condition, and multi-step processing.

3.3.1 Static Tests: Chemical stability of niobium in oxidizing and etching aqueous

solutions

The purpose of the static tests was to understand the niobium oxide chemistry in terms of the impact of different pretreatments and slurries on oxide type and thickness, and hence its impact on the polishing rate. As it was discussed previously, XPS is a powerful technique for surface analysis. During static tests, surface composition and oxidation states of niobium were determined using XPS in the Interface Engineering

Laboratory. Niobium samples were tested statically in oxidizing and etching solutions to gain an understanding of the Nb surface chemistry possible in the slurry conditions.

Hydrogen peroxide (H2O2) is a common oxidizing agent in the CMP slurries, for example 1% of hydrogen peroxide is present in the copper slurry EP-C6678 from Cabot

Microelectronics. Hydrofluoric acid is used in BCP and electropolishing (EP) for removal of the damaged surface layers. Thus, etching solution containing 5 mL of hydrofluoric acid (49%), 17 mL of nitric acid (65%), and 51 mL of methanol was used to observe surface oxide chemical removal, and a solution containing 10 mL of hydrogen peroxide (30%), 5 mL of ammonium hydroxide (30%), and 50 mL of deionized water was used to study niobium oxide formation. The effect of these solutions on the surface chemistry of niobium was studied with XPS. It was also important to understand the slurry chemistry and its impact on the oxide solubility through the Pourbaix diagrams.

Thus pH and electrostatic potential of various slurry compositions were measured.

Determination of the optimal conditions for the oxide formation and a study of the oxide layers were two of our initial goals, because in metal CMP formation of surface oxide is necessary to protect metal from corrosion and increase material removal rates due to lower hardness of metal oxides. The effects of the oxidizer, hydrogen peroxide, and the etchant, hydrofluoric acid, were established through a series of static chemistry experiments. 1 cm2 samples cut from of a rolled niobium wafer were submerged in different chemical environments for various time periods.

All experiments started with a degreased sample as described above. Figure 19 shows a Nb3d spectrum of the as-received niobium sample which was degreased. The

Nb3d peak contains two peaks corresponding to the two spin states: 3d5/2 and 3d3/2. 3d3/2 peak is shifted 2.70 eV to the higher binding energy with respect to 3d5/2 peak. As identified in Figure 19 three detected oxides, NbxO (1≤x≤2), NbO2, and Nb2O5, are

77 shifted by ~ 1.0 eV, 3.6 eV and ~ 5.1 eV to the higher binding energy than the metal, respectively. We choose to assign NbxO peak instead of separate NbO and Nb2O, because of some discrepancies in peak identification in the literature [64],[18],[19],[3], instrumental resolution with a nonmonochromated source which was used for this analysis, and the fact that we feel that having these oxides in one peak does not limit our analysis or our understanding of the surface chemistry processes.

The most prominent peaks are for the niobium pentoxide and metal, while the intermediate oxides have relatively small peaks. In Figure 19, the blue curve represents the combination of the intermediate oxides (i.e. NbxO (1≤x≤2) and NbO2) of niobium.

Due to the fact that NbxO (1≤x≤2) and NbO2 peaks overlap and nonmonochromated X- ray source was used for XPS analysis, the XPS fitted data presented throughout this thesis will have the combined NbxOy (1≤x,y≤2) oxide peak instead of the individual oxides. Combining of the lower oxidation states was done by other researchers in the past, for example see Grundner et al. [16]. Confidence of the fit for the pentoxide and the metal components of the Nb3d spectrum is high because there is not overlapping of the metal 3d5/2 and the pentoxide 3d3/2 components, making determination of the peak position and the full width at half maximum relatively straight forward.

The total oxide thickness for this and all as-received samples, was determined to be 4.5±0.7 nm as determined from the attenuation of the metal peak through the oxides overlayer per the standard procedure described earlier and using the NIST database (SRD

82) for determination of the inelastic mean free path. The error in the thickness calculation is the absolute error associated with calculation of IMFP, and the experimental error from repeated runs of as-received samples falls within this error. In

78 general, the error shown is the absolute error in the calculation and experimental variation is expressed separately as appropriate. The oxide thickness value agrees with other researchers [17],[19],[16], [3]. Disagreement with other groups may be explained by the error associated with calculation of IMFP and the fact that processing conditions affect oxide thickness. For example, Tian et al. observed formation of oxides of different thickness depending on a flow rate of the BCP solution [19]. Kowalski et al. found that electopolished and BCP-treated niobium had different oxidation kinetics [3]. Note that the sampling depth of the XPS is 7.0 nm based on inelastic mean free path (IMFP) of Nb electrons ionized with Mg anode; IMFP was calculated using NIST SRD82 database.

Thus the sampling volume of the as-received samples is 83.9% oxide and 16.1% metal, assuming an oxide on metal layered structure as described in Section 2.3.4.

The full width at half maximum (FWHM) of the XPS peaks is measure of the atomic order in the samples. For example, Kowalski et al. observed increase in FWHM for pentoxide peak from 1.8 to 2.4 eV during annealing, they attributed this observation to increase in crystallographic defects due to thermal treatment [3]. The metal peak of the as-received sample shown in Figure 19, is more narrow (FWHMmetal = 1.30±0.05 eV) than the oxide peaks (FWHMNb+5 = 1.55±0.05 eV and FWHMNbxOy = 1.90±0.05 eV ) which means that the oxides are less ordered structures than the metal. When we perform the static chemical and the CMP experiments on the Nb wafers, the FWHM will be used to evaluate the subsurface ordering of the niobium bonding. The FWHM for NbxOy will not be compared due to less certainty in peak assignment as discussed above. Relating the

FWHM values after processing to the as-received sample will enable us to conclude if the processing treatment distorted the surface or removed disorganized layers.

Nb2O5 Nb NbxO(1≤x≤2) 3d3/2 3d5/2 NbO2 Nb2O5 Nb NbxOy (1≤x,y≤2)

3d3/2 3d5/2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 19: Nb3d spectra of as-received niobium wafer exhibits presence of two

oxides: NbO and Nb2O5. Blue squares represent experimental data and

brown line is the overall fit. All of the peaks are fitted with Gaussian-

Lorentzian curves (80% Gaussian)

The first static chemistry tests performed on the niobium samples were to determine if the as-received sample could be further oxidized by using a hydrogen peroxide solution, if all oxide could be removed by hydrofluoric acid etch, and could the etched sample then be reoxidized in a more ordered layer. Figure 20 shows a comparison of the spectra that were obtained from three separate experiments: oxidation of the as- received sample with hydrogen peroxide, etch of the as-received sample with a solution

80 containing hydrofluoric acid, and the hydrofluoric acid etch followed by oxidative treatment.

All spectra shown in Figure 20 have metal, pentoxide, and a combination of the intermediate oxides. Peak intensities for the as-received sample and the sample oxidized in H2O2 solution are very similar, and the total oxide thicknesses were determined to be

4.5±0.7 nm and 4.6±0.7 nm, respectively. In addition, the FWHM for the pentoxide peak decreased to 1.50±0.05 eV from 1.55±0.05 eV after exposure to H2O2, which suggests that hydrogen peroxide may have promoted formation of a more conformal and ordered pentoxide layer; however, the differences in FWHM are just beyond the measurement error. FWHM for the metal and intermediate oxides, however, did not change.

Hydrofluoric acid etching of the as-received niobium sample resulted in an increase of the metal to pentoxide ratio from 0.26±0.03 for as-received sample to

0.84±0.03; the intermediate oxides level reached 38.3% from 23.1% in the as-received sample as seen in Figure 20. Pentoxide was the predominant component in the as- received sample totaling 60.9% of Nb3d peak area, and that changed after hydrofluoric acid etch resulting in higher intensity for the combination of metal and NbxOy peaks,

33.6% for pentoxide and 66.4% for the metal and intermediate oxides. The total oxide thickness decreased from 4.5±0.7 nm to 4.1±0.6 nm. The 44.8% decrease in the area of the pentoxide peak is countered by a 65.8% increase in the total peak area for the intermediate oxides. It was noted by other researchers that they could not move the sample into the UHV environment for the XPS analysis fast enough to prevent oxidation in air [46]. The etched sample was loaded for analysis within less than 10 minutes after the acid treatment.

Nb Nb2O5 NbxOy 1≤x,y≤2

3 min HF etch followed by 10 min H2O2

3 min HF etch Count

10 min H2O2

As received

214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 20: Nb3d peak for samples treated with hydrogen peroxide, HF, and HF and hydrogen peroxide.

In addition to the removal of the Nb2O5, the FWHM for the metal decreased from

1.30±0.05 eV to 1.20±0.05 eV, indicating more ordered metal peaks in the XPS sampling volume. We hypothesize that the disordered Nb layers were etched away, revealing deeper layers and hence more ordered niobium surface. The FWHM for all oxide peaks did not change from the as-received sample values, implying a consistent oxidation in air mechanism.

To test if the oxide can be grown on the etched surface and the gain in the atomic order due to hydrofluoric acid can be retained; a sample was first etched in the HF solution and then oxidized in the hydrogen peroxide solution. This sample had the Nb3d spectrum similar to that of the oxidized sample (in air or in solution), and is shown in orange in Figure 20. Again, the pentoxide was a predominant component of the 7-nm sampling depth. Visually, the intensity of the pentoxide and intermediate oxides was similar to the as-received and H2O2 oxidized sample. The total oxide thickness was determined to be 4.6±0.7 nm, the same as the as-received sample and the sample only exposed to the hydrogen peroxide solution. This indicates that during oxidative treatment etched niobium sample developed the full oxide, which was encouraging in light that oxidation is important for CMP; in addition, it implied that niobium surface oxidation might be self-limiting.

The FWHM for the sample, which was etched and then oxidized, were the same as for the as-received sample i.e. 1.30±0.05 eV. Hydrogen peroxide treatment resulted in increase of the FWHM value to 1.30±0.05 eV from 1.20±0.05 eV for the etched sample, indicating that the atomic order of metal was reduced possibly due to nonuniform oxide layer formation as suggested by Darlinski and Halbritter in [21] and serration by

Halbritter in [49]. At the same time, FWHM of the pentoxide remained the same as for the as-received sample.

The fact that we were able to regrow oxide layer during 10-min oxidative treatment was encouraging, because the formation of an oxide is typically necessary for reasonable metal removal rates in the CMP process [34]. During CMP process, however, the pad and wafer rotational velocities are 60-100 rpm (typical velocities used with the

G&P Technologies Poli-500 CMP machine) and the removal rates are 400 nm/min

(tungsten CMP) [34]. Therefore, the surface oxidation needs to be sufficiently rapid to produce softer oxides for enhanced material removal and to prevent niobium from corrosion during CMP process.

Information about the oxidation rate of niobium was obtained by treating the samples, which were etched with HF containing solution (~3.4% HF) for 3 minutes, in solutions containing H2O2 (~4.6% H2O2) for 10 s, 30 s, 1 min, and 10 min. After oxidative treatment samples were transferred into the UHV environment for the XPS analysis within 10 min; the obtained Nb3d spectra are shown in Figure 21. Visually, the metal and intermediate oxides are more pronounced for the samples that were treated with hydrogen peroxide solution for less than 10 min. However, the total oxide thickness was found to be within the calculation error for all samples: 4.5±0.7 nm for the 10-s sample, and 4.6±0.7 nm for the other ones as seen in Table 13. This indicates that oxide formation was very rapid, and suggests a self-limiting oxide thickness. Interestingly, the

FWHM of metal and pentoxide peaks for the 10-s, 30-s, and 1-min samples was lower

1.20±0.05 eV and 1.50±0.05 eV, respectively, than that of 10-min treated and as-received sample. As-received sample and the etched sample that was oxidized in hydrogen

peroxide solution for 10 min were 1.30±0.05 eV and 1.55±0.05 eV for metal and

pentoxide peaks, respectively. This finding supports the idea of formation of more

ordered oxide layers which become more disordered as time progresses, possible

serration of the metal surface may occur which decreases its atomic order.

Table 13: Summary of oxide thicknesses and the FWHM values for the oxidation

time study.

Treatment Oxide thickness, nm FWHM Nb metal, FWHM Nb2O5,

eV eV

HF 3 min 4.1±0.6 1.20±0.05 1.55±0.05

HF 3 min, H2O2 10 s 4.5±0.6 1.20±0.05 1.55±0.05

HF 3 min, H2O2 30 s 4.6±0.7 1.20±0.05 1.50±0.05

HF 3 min, H2O2 1 min 4.6±0.7 1.20±0.05 1.50±0.05

HF 3 min, H2O2 10 min 4.6±0.7 1.30±0.05 1.55±0.05

BCP treatments on the niobium samples were performed, and it was found that BCP removed the disordered metal layers associated with the rolling process, but it increased surface roughness likely due to preferential etching of the grain boundaries. Therefore, BCP treatment alone would not produce both a smooth and ordered niobium surface on the polycrystalline material. See Appendix B for more details.

Nb Nb2O5 NbxOy 1≤x,y≤2

10 s in H2O2

Count 30 s in H2O2

1 min in H2O2

10 min in H2O2

214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 21: Oxidative treatment results in rapid oxide growth.

In summary, the static etch and oxidation experiments with hydrofluoric acid and hydrogen peroxide, respectively, concur with literature that niobium oxidizes very rapidly in all oxidizing environments tested, producing mostly Nb2O5 oxide. The total oxide thickness was determined to be on the order of 5 nm. Concurrently, CMP experiments showed that the material removal rates for slurries containing hydrogen peroxide and silica particles was low. Thus, design of a special slurry for niobium was found to be unnecessary. It was hypothesized that niobium CMP may be mechanically driven, so testing of the commercially available slurries with different particles was proposed.

3.3.2 Niobium Surface Condition after CMP Process

Effect of a few process variables was determined: the starting surface condition, type of an abrasive particle, and the pH of a slurry. BCP and barrel polishing, which is a mechanical abrasion with plastic stones, an abrasive powder, and water, of the niobium cavities are common processes in SRF cavity manufacturing [26]. Both processes are performed to remove damage and contamination from previous processing steps [26].

Mechanical abrasion of the rolled as-received wafers was carried out by H.C. Starck to imitate barrel polishing. The effect of BCP and mechanical abrasion on CMP process was tested.

Silica and alumina particles are most commonly used particles in CMP [50]. Due to difference in hardness values of these particles the material removal rates are expected to be different as well. Typically, chemistry of the slurry plays a very important role in semiconductor CMP processing [50], and Pourbaix diagrams are used as a guide in the

87 slurry design. One of the parameters that can be relatively easy to adjust is pH, so the effect of a pH on the material removal rates was studied.

Finally, from the studies of the effect of the starting surface, abrasive particles, and pH of the slurry, a two-slurry process for niobium polishing was developed. The first

CMP treatment involved the use of large (1 µm) alumina particles, followed by either smaller alumina (0.05 µm) or silica (≤0.1 µm) particles.

3.3.2.1 Effect of starting surface condition

We were interested in investigating different starting surface conditions to determine if it would affect material removal rates or surface chemistry. Three starting surface finishes were available: rolled, BCP treated, and mechanically abraded. Most of the BCP and all of the mechanically abraded wafers were supplied by H.C. Starck

Company. XPS of the as-received rolled niobium was performed. Due to the fact that limited number of pretreated (BCP and mechanical abrasion) wafers was available, the starting surface of the BCP treated (by H.C. Starck) and mechanically abraded wafers was not studied with XPS, because for XPS analysis small samples are required which would have required sacrificing of full wafers. BCP of the rolled wafers was performed at

Northeastern University according to procedure in Section 3.2.2; XPS analysis of these samples was performed and can be found in Appendix B.

Figure 22 shows optical images that were taken from the wafers that were rolled,

BCP treated, and mechanically abraded. As seen from Figure 22, morphology of the surfaces is different. The surface of the rolled wafers (Ra≈0.82 µm) exhibits deep scratches that were produced by the rolling process. The surface of the BCP treated

(Ra≈0.85 µm) wafer shows multiple grains with deep boundaries. Finally, mechanically abraded (Ra≈0.31 µm) wafer has a great number of small randomly positioned defects.

Experiments designed to compare effect of the starting surface on CMP process were designed and performed. Table 14 gives a list of experiments that can be compared to determine effect of the starting surface on the CMP process.

a) rolled b) BCP treated

c) mechanically abraded Figure 22: Optical images of the starting surface condition for the wafers that were

rolled, BCP treated, and mechanically abraded.

Table 14: Experiments that allow comparison of the starting surface effect on CMP

process.

Experi Type of starting Slurry Particle type and Starting/Final ment surface size Roughness (Ra), µm

3 Rolled 16 min EP-C6678 Silica, 90-140 nm 1.024 / 0.568

6 Rolled, then 6 16 min EP-C6678 Silica, 90-140 nm 0.670 / 0.359

min BCP (NU)

11 abraded 12 min NANO- Alumina 1 µm and 0.364 / 0.093

1010-128, pH≈7 alumina 0.05 µm

19 min NANO-

1005-128, pH≈7

8 BCP 12 min NANO- Alumina 1 µm and 0.775 / 0.075

1010-128, pH≈7 alumina 0.05 µm

18 min NANO-

1005-128, pH≈7

In Experiments 3 and 6, a copper slurry was used for 16 min. Starting roughness of the wafers was different. Rolled wafer in Experiment 3 had starting roughness (Ra)

1.024 µm, and the wafer in Experiment 6 was treated with BCP at Northeastern

University for 6 min, its starting roughness was 0.670 µm. The optical microscope images are shown in Figure 23. It can be seen that the rolled wafer after CMP process retained parallel hills and valleys seen on the starting surface. Likewise, the wafer pretreated with BCP and polished with a copper slurry for 16 min, shows isolated pits similar to those observed on the starting surface. By the end of the 16 min, the roughness

(PV, RMS, and Ra) of the wafers was observed to be leveling off, indicating that the slurry used in these CMP experiments was ineffective in removing material. Final roughness of the BCP sample was slightly better than that of the rolled wafer, 0.359 µm versus 0.568 µm, which is expected because of the lower starting roughness of the starting wafer.

Figure 23: Optical images for the final surface in Experiment 3 (left) and

Experiment 6 (right)

After CMP, the wafers in Experiments 3 and 6 were cut to obtain samples from the center, middle, and the edge of the wafer. All samples were degreased with sonication according to procedure in Section 3.2.2. Silicon contamination (6.4 – 9.8 %) in a form of silica was detected on all samples from Experiments 3 and 6. It is possible that particles may have been trapped in the surface crevices of the post-CMP samples.

Figure 24 shows Nb3d spectra of the rolled sample that was polished with the copper slurry in Experiment 3. The total oxide thickness for the Experiment 3 samples was determined to be 4.5±0.7 nm. Metal to pentoxide ratio for this sample increased to

0.45±0.01 from 0.26±0.01 as determined for the as-received sample, indicating that due to CMP treatment surface chemistry of niobium changed toward thinner pentoxide layer.

In addition, it is important to note that the FWHM for the pentoxide peak decreased to

1.45±0.05 eV from 1.55±0.05 eV for the as-received wafer. This indicates the improved atomic order of the pentoxide as a result of the CMP treatment. The FWHM for the metal did not change, which is not surprising as low material removal was observed during the CMP polishing with the copper slurry.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV Figure 24: Nb3d spectra of the sample obtained from the wafer polished in

Experiment 3.

In Experiment 6, the BCP treated wafer was polished for 16 min with the copper slurry. The optical image of the final surface is shown in Figure 23. Similar to the wafer obtained in Experiment 3, surface was contaminated with silica, but level of contamination was lower (2.7 – 5.1 %) than that of the rolled wafer polished in

Experiment 3. Lower particle contamination may be due to lower roughness and/or different morphology which allowed easier removal of the particles with rinses and sonication. Figure 25 show Nb3d spectra obtained for one of the samples from the wafer polished in Experiment 6. The total oxide thickness for the wafer obtained in Experiment

6 was calculated to be 4.7±0.7 nm. Consistent with the total oxide thickness calculations, the metal to pentoxide ratio was 0.28±0.01, which was very close to that of the as- received rolled wafer 0.26±0.01, but significantly lower than that of the BCP treated wafer, 0.86±0.03. This indicates that niobium surface was oxidized during CMP process as expected from the static experiments discussed in Section 3.3.1. The FWHM values for this wafer were the same as for the sample obtained in Experiment 3, 1.45±0.05 eV

92 and 1.30±0.05 eV for the pentoxide and the metal peaks, respectively. However, the starting surface after BCP treatment had the FWHM of 1.60±0.05 eV and 1.10±0.05 eV for the pentoxide and the metal respectively. Thus, the atomic order of the metal decreased during CMP, but increased for the pentoxide. These observations are consistent with static tests results, when the FWHM for the metal peak of the HF solution treated samples decreased due to hydrogen peroxide treatment. However, unlike static tests when the FWHM of the pentoxide remained the same during the H2O2 treatment, the

FWHM decreased for the samples due to CMP treatment. This may be explained by continuous removal and regrown of oxide on the niobium wafer during CMP, while static tests show a snap shot of what happens after the oxide is first chemically etched and then regrown.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 25: Nb3d5 spectrum of the samples obtained in Experiment 6.

The CMP Experiments 11 and 8 were performed on the mechanically abraded and

BCP treated wafers, respectively, with alumina slurry for 16 min. After CMP process, wafers were cut for XPS analysis. All samples were degreased with sonication before

93 loading for the XPS analysis. Starting surface condition for Experiment 11 was mechanical abrasion with roughness (Ra) of 0.364 µm as seen in Table 14. The CMP process with 1 µm (first slurry 12 min) and 0.05 µm (second slurry 18 min) was performed resulting in reduction of the roughness (Ra) value to 0.093 µm. However, the lowest roughness (Ra), 0.058 µm, for this wafer was obtained after 21 min of polishing

(12 min with 1 µm and 9 min with 0.05 µm alumina particles) as seen in Figure 26.

Morphology of the surface changed dramatically from that of the mechanically abraded wafer as seen in Figure 22c. Localized spikes were removed, while pits developed.

Figure 26: Optical microscope image of the wafer obtained in Experiment 11 after

21 min of polishing.

The XPS analysis of the final surface of the wafer in Experiment 11 determined alumina contamination (9.7-14.5%) which was supported by SEM analysis, as can be seen from the insert in Figure 27. The niobium pentoxide component is greater than that for the sample polished with the silica containing slurry (for example in Experiment 3) or as-received sample. The oxide thickness was approximately 4.8±0.7 nm. The FWHM for the metal did not change from that of the FWHM of the as-received rolled wafer,

1.30±0.05 eV. The FWHM for the pentoxide peak decreased slightly from 1.55±0.05 eV to 1.50±0.05 eV. This indicates that despite lower bulk roughness values, the atomic order was not affected by CMP treatment as determined from the FWHM values.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 27: Nb3d5 peak of the wafer obtained in Experiment 11. SEM image shows presence of particles on the surface. XPS revealed alumina.

BCP treatment for the wafer used in Experiment 8 was performed by H.C. Starck

Company. Starting surface roughness (Ra) was higher than that of the mechanically abraded wafer in Experiment 11, 0.775 µm. However the final surface roughness for the wafer polished with the same slurry combination (alumina 1 µm for 10 min and alumina

0.05 µm for 22 min) reduced surface roughness (Ra) to 0.075 µm (Figure 28). However, similar to the wafer in Experiment 11, surface was contaminated with alumina particles as it can be seen from the SEM image insert in Figure 29.

Figure 28: Optical image of the final surface condition obtained in Experiment 8.

XPS analysis determined that alumina contamination in the range from 11.7% to

13.5% on the surface of the wafers obtained in Experiment 8. The Nb3d XPS spectra of the wafer obtained in Experiment 8 is shown in Figure 29. The total oxide thickness was determined to be 4.9±0.7 nm. Despite improvement in overall roughness, XPS analysis showed that FWHM for the metal peak increased from 1.10±0.05 eV to 1.30±0.05 eV, indicating that BCP surface after CMP with alumina particles has more defects than the starting BCP surface. The FWHM for the niobium pentoxide peak decreased to

1.50±0.05 eV from 1.60±0.05 eV for BCP treated surface. This indicates that the atomic order of the pentoxide improved due to CMP polishing, similar to results in Experiment

11. Total oxide thickness for this sample was similar to the sample polished in

Experiment 11, 4.9±0.7 nm.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding energy, eV

Figure 29: Nb3d5 spectrum (Experiment 8) shows pentoxide as a predominant

oxide, oxides thickness 4.9±0.7 nm. SEM image insert proves particle

contamination.

Comparison of the results obtained in Experiments 11 and 8 allows us to conclude, that starting surface condition does not have significant effect on the bulk material removal for CMP process with the alumina slurry. Reduction in a surface roughness was greater for the BCP treated wafer. However, the atomic order of the BCP treated and mechanically abraded wafers were very similar, indicating that starting surface condition did not affect atomic order of the surfaces polished with alumina slurries.

To conclude, three surface conditions were compared: rolled and BCP

(Experiments 3 and 6), and abraded and BCP (Experiment 11 and 8). Starting surface roughness had no or little effect on the final roughness post-CMP. When copper slurry

(silica particles) was used in Experiments 3 and 6, morphology of the surface resembled that of the starting surface which indicated that material removal was not sufficient. In addition the atomic order, as determined from XPS analysis, was the same for the wafers in Experiments 3 and 6. Similarly, in Experiments 11 and 8 when a slurry with alumina particles was used, the atomic order of the surfaces after BCP treatment and mechanical abrasion was equivalent. BCP treated wafer was shown to produce lower average roughness after polishing with alumina slurry despite the fact that the starting roughness of this wafer was greater.

3.3.2.2 Effect of Slurry: Abrasive Particle, pH

The static experiments indicated that niobium oxidizes rapidly air and oxidizing aqueous solution, thus it was concluded that no special slurry was necessary. It was hypothesized that the CMP process of niobium might be dominated by a mechanical

97 material removal as oppose to a chemical reaction on a surface. From the experiments described in Section 3.3.2.1, it was concluded that the starting surface condition did not have a significant effect on the atomic order of the surfaces during CMP process. Thus, it was decided to test commercially available slurries containing different abrasive particles to determine effect of the particle hardness on the material removal rates. Due to the fact that the pH of aqueous solutions was shown to be a factor in determining surface species of niobium, as seen from Pourbaix diagram of niobium shown in Figure 17, the effect of the slurry pH was tested. Table 15 shows experiments that allow determination of the effect of the abrasive particle type and the pH.

Table 15: Experiments performed to determine the effect of slurry on CMP of Nb:

particle type and pH.

Experi Slurry Particle type and size Starting/Final ment Roughness (Ra), µm

1 16 min Microplanar Silica, 50-100 nm 0.865 / 0.286

CMP1150, pH≈10

12 12 min NANO-1010- Alumina 1 µm and 0.316 / 0.061

128, pH≈10 alumina 0.05 µm

19 min NANO-1005-

128, pH≈10

11 12 min NANO-1010- Alumina 1 µm and 0.364 / 0.093

128, pH≈7 alumina 0.05 µm

19 min NANO-1005-

128, pH≈7

12 12 min NANO-1010- Alumina 1 µm and 0.316 / 0.061

128, pH≈10 alumina 0.05 µm

19 min NANO-1005-

128, pH≈10

Our first experience (Experiment 1) with polishing of a niobium wafer was with the Microplanar CMP1150 slurry. The wafer was polished for 16 min. Final roughness

(Ra) achieved was 0.435 µm, as seen in Figure 30. All of the three samples were from the center of the wafer. Samples exhibited uniform elemental distribution as would be expected for samples from the same location. High silicon contamination (approximately

8 atomic percent) in a form of silica was found on all samples using XPS and SEM.

SEM image shown in Figure 31, exhibited multiple particles on the surface of the sample.

It is possible that due to the high surface roughness of this wafer, the 50 to 100 nm particles were trapped in the grooves and crevices as opposed to being imbedded or attracted to the surface.

Figure 30: Optical image post-CMP for Experiment 1

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 31: XPS spectrum of the sample obtained in Experiment 1. SEM image

(insert) shows presence of particles which were determined with XPS to

be silica particles.

The Nb3d spectrum for one of the samples is presented in Figure 31. The total oxide thickness was determined to be 4.6±0.7 nm which indicates no change in oxide thickness from that of the as-received sample occurred. The FWHM for the metal peak was 1.30±0.05 eV, the same value as the degreased surface. The oxide FWHM decreased to 1.50±0.05 eV from 1.55±0.05 eV for the as-received sample. Insignificant decrease of the FWHM and high roughness of this sample indicate that the CMP process with the

Microplanar slurry containing silica particles was not successful.

The alumina slurry is received with the pH of approximately 7. The pH was adjusted to 10 to match that of the Microplanar slurry and eliminate the additional variables. The final roughness for the wafer in the Experiment 12 was 0.061 µm after 31 min of polishing. Figure 32 shows optical images and roughness values for 16 min and

100

31 min of polishing for the wafer in Experiment 12. Both values obtained in this experiment are significantly lower than those obtained in Experiment 1 when the wafer was polished with the Microplanar slurry containing silica particles. Thus, we conclude that alumina slurry is more effective in removing material i.e. reducing roughness.

a) after 16 min, PV=0.668 µm b) after 31 min, PV=0.354 µm Ra=0.061 µm Ra=0.044 µm

Figure 32: Optical microscope images of the surface treated with alumina slurry for

16 and 31 min in Experiment 12.

XPS analysis for the wafer in Experiment 12 was done after 31 min of polishing due to the fact that the wafer must be cut for the XPS analysis, thus making further polishing impossible. Figure 33 shows Nb3d5 spectrum from which total oxide thickness was calculated to be 4.9±0.7 nm. The FWHM for metal peak did not change; however the FWHM decreased to 1.45±0.05 eV for the pentoxide peak. This indicates that pentoxide layer had more atomic order post-CMP process. However, comparison of the

XPS results should be done with caution due to different polishing times.

The roughness values were lower for the sample obtained in Experiment 12; however, it is important to notice that the starting surface roughness and duration of the

CMP treatments were different. So in order to see the effect more clearly, a graph relating polishing time with roughness was made. Because starting roughness was different, all of the roughness values were normalized to the initial value. Figure 34

101 clearly shows that alumina particles are more effective in reducing roughness (i.e. removing material). Roughness values (peak-to-valley - PV, root mean square - RMS, and average roughness - Ra) were greater at all times for the CMP of niobium with the slurry containing silica particles. Thus, we conclude that alumina particles are more effective in reducing surface roughness and removing material. After CMP, wafers in

Experiment 1 and 12 had considerable particle contamination.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 33: Nb3d spectrum shows that pentoxide species are predominant on Nb

surface after CMP in Experiment 12. SEM image (insert) shows high

particulate contamination on the surface.

102

1.2

0.8 Silica PV Silica RMS Silica Ra 0.6 Alumina PV Alumina RMS 0.4 Alumina Ra Normalized roughness 0.2

0 0246810121416 T i m e, min

Figure 34: Graph showing evolution of the niobium wafers treated with silica and

alumina containing slurries.

The importance of the slurry composition was emphasized in section 3.2.3. One of the slurry components is a buffer for adjustment of the pH. The Pourbaix diagram of niobium is shown in Figure 17. Because the pH and the potential of the aqueous solutions determine the equilibrium surface species, the pH and the OCP values of the slurries were measured, and results can be seen in Table 16. The pH and OCP values fall

- in the same region in the Pourbaix diagram – NbO3 as seen in Figure 35. Our original plan was to adjust the pH of the alumina slurries to shift niobium to a different equilibrium state. Due to the fact that as-received alumina slurries operate close to the border line, we wanted to shift pH of the aluminum slurries to more acidic region in order to operate in Nb2O5 region, and more basic to operate in the region of the oxide slurry

(Microplanar CMP1150). However, coagulation of the alumina slurries was observed at the pH 6 and below. Thus it was decided to use the alumina slurries at pH≈7 and pH≈10.

103

Table 16: pH and OCP values of the slurries.

Slurry Particle type OCP pH

Microplanar CMP1150 Silica -0.5142 10.54

EP-C6678 Silica -0.3750 9.82

NANO-1010-128 Alumina -0.2379 6.73

NANO-1105-128 Alumina 0.2141 6.02

Oxide Al2O3 Slurry Slurry

Al2O3 Slurry +NaOH

Figure 35: Pourbaix diagram of niobium with slurries plotted [61].

The wafer in the Experiment 11 was polished first with alumina slurry containing

1 µm alumina particles followed by a slurry containing 0.5 µm particles. These slurries were used as-received with the pH of approximately 7. The wafer obtained in the experiment 11 was discussed in Section 3.3.2.2. SEM (Figure 29) and XPS analysis of this wafer showed significant concentration of alumina particles on the surface. The aluminum concentration on the surface of the wafer varied between 9.7 and 14.5, with the

104 highest value in the middle. Visually, the wafer in Experiment 11 looked opaque. Figure

29 shows niobium spectrum from one of the samples polished with the alumina slurry in the Experiment 11. The oxide thickness was approximately 4.8±0.7 nm. The FWHM for the metal peak was 1.30±0.05 eV and 1.50±0.05 eV for the pentoxide peak. Thus, the atomic order was not affected by the CMP treatment.

As for the wafer polished with the alumina slurry with pH of 10 in Experiment

12, it had considerable particulate contamination of the surface as determined by SEM and XPS. Figure 33 shows Nb3d peak and SEM image obtained after the CMP treatment. As mentioned above, the FWHM for the metal peak did not change; however it decreased to 1.45±0.05 eV for the pentoxide peak, indicating that some improvement in the surface order was observed due to CMP treatment. Thus, XPS analysis showed that niobium surfaces obtained in Experiments 11 (pH≈7) and 12 (pH≈7) were very similar.

Some improvement in the FWHM of the pentoxide peak was observed for wafers in

Experiments 11 and 12, indicating that the CMP process improved atomic order of the outer most oxide layer.

In order to compare removal rates, a graph similar to Figure 34 was constructed.

Figure 36 shows evolution of roughness with time for wafers polished with alumina slurry at pH≈7 (Experiment 11) and pH≈10 (Experiment 12). No significant difference between the two experiments is seen. Therefore, we conclude that pH adjustment of alumina slurry to 10 does not affect removal rates of niobium.

105

0.9

0.8

0.7 Exp. 11 PV 0.6 Exp. 11 RMS Exp. 11 Ra 0.5 Exp. 12 PV 0.4 Exp. 12 RMS Exp. 12 Ra 0.3 Normalized roughness 0.2

0.1

0 0 5 10 15 20 25 30 35 Time

Figure 36: Graphs representing evolution of the roughness values over time for the

samples polished in Experiments 11 and 12 do not show significant

trends, indicating that removal rates are not affected by change in the

pH from 7 to 10.

3.3.2.3 Multi-step process

Experiments 3 and 6, and 11 and 8 showed that the starting surface condition had little impact on material removal and surface characteristics (chemical and crystallographic). Experiments 1 and 12, and 11 and 12 allowed us to conclude that the alumina particles were more effective in reducing surface roughness, but the pH adjustment of the alumina slurry were not necessary as very similar surfaces were obtained after treatment with a slurry of the pH 7 and the pH 10. Although the alumina particles reduced surface roughness considerably, we could not obtain surface with roughness less than roughly the size of the particles used. For example, in Experiment 11 after polishing for 12 min with 1-µm aluminum particles, roughness (PV) leveled off at

106

1.122 µm. After additional 10 min of polishing with 0.05-µm particles, average roughness further decreased leveling off at the value of 0.468 µm. Similar situation occurred in Experiment 12 and other ones when we used alumina slurry. These observations, lead us to conclude that alumina particles may be useful in removing

“micro” roughness, but ineffective in removing “nano” roughness due to their hardness.

So polishing of the niobium wafers was proposed to do with the alumina slurry first, followed by oxide slurry containing silica particles.

Table 17 shows two experiments that allow comparison between multi-step process using alumina particles of different size or the larger alumina particles followed by smaller silica particles. It is important to note that selection of the polishing time was based on the roughness values. If roughness values stopped changing, polishing process was either stopped to change the slurry or terminated altogether.

Table 17: Experiments allowing comparison of the multi-slurry processes.

Experiment Slurry 1, time, Slurry 2 Starting/Final

particle size Roughness (Ra), µm

10 NANO-1010-128 Nalco 2398 0.306 / 0.024

12 min, 10 min,

Alumina 1 µm Silica 70-100 nm

11 NANO-1010-128 NANO-1005-128 0.670 / 0.093

12 min, 28 min,

Alumina 1 µm Alumina 0.05 µm

107

In Experiment 10, the mechanically abraded wafer was polished with the alumina slurry containing 1 µm particles for 12 min, then with Nalco 2398 slurry for additional 10 min. The final roughness (Ra) for this sample was 0.024 µm as it can be seen from Figure 37. This wafer had a mirror-like finish. SEM and AFM images (Figure

37 and Figure 38) show some particulate contamination, but overall surface of the wafer obtained in Experiment 10 was very smooth and had significantly lower particulate contamination than wafers polished with alumina slurries as seen in inserts in Figure 27,

Figure 29, and Figure 33. This indicates that if a silica containing slurry is used after alumina slurry, then particulate contamination can be significantly decreased.

Figure 37: Optical microscope image of the wafer obtained in Experiment 10.

Figure 38: SEM images obtained from the samples polished in Experiment 10 show

presence of particles on the surface.

108

+70 nm

-60 nm

±10 nm ±6 nm

4.0µm 1.0µm

Figure 39: The AFM images of the surfaces obtained in Experiment 10 allow to

determine that surface between defects and particles is significantly

smoother than the values determined by the optical microscope.

Figure 40 shows XPS spectrum of the wafer after polishing with alumina/silica slurry combination in Experiment 10. From Nb3d spectra the total oxide thickness was calculated to be 4.9±0.7 nm. The FWHM for the metal was determined to be 1.35±0.05 eV and for the pentoxide 1.50±0.05 eV. This indicates that the FWHM for the metal peak slightly increased while the FWHM for the pentoxide slightly decreased due to

CMP process; however both values are within error of the as-received surface. No alumina or silica contamination was determined with XPS.

The wafer obtained in the Experiment 11 was discussed in more detail in section

3.3.2.1. This wafer had an opaque residue on the surface of the wafer, which by XPS and

SEM analysis was determined to be alumina particles. The SEM image is shown in

Figure 27. The total oxide thickness for this sample was calculated to be 4.8±0.7 nm.

The FWHM for the metal peak was 1.30±0.05 eV which was the same as for the as-

109 received sample. The FWHM for the pentoxide peak decreased from 1.55±0.05 eV to

1.50±0.05 eV. This indicates that despite lower bulk roughness values, the atomic order was not affected by CMP treatment as determined from the FWHM values.

Nb Nb2O5 NbxOy 1≤x,y≤2 Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure 40: Nb3d5 spectrum for Experiment 10

Overall, from the comparison of the wafers obtained in Experiments 10 and 12, we conclude that alumina containing slurry followed by silica containing slurry is an effective combination to produce smooth niobium surfaces with low particulate contamination. The SEM and AFM images showed presence of particles on the surface which match the size of silica particles used for the final CMP step. The XPS analysis showed that the surface of the wafers in the Experiments 10 and 12 were comparable.

These surfaces did not exhibit significant improvement is surface order after CMP process.

110

3.4 Summary and Conclusions of Nb CMP

The static tests performed on pieces of the rolled wafers indicated that the niobium surface forms an oxide of approximately 4.5±0.7 nm after air exposure.

Hydrofluoric acid treatment was observed to remove oxide from the surface, reducing the total oxide to 4.1±0.6 nm; however the oxide could be grown to its original thickness within 10 s in an oxidizing environment. Thus, we concluded that oxide formation on the niobium surface is very rapid therefore a special slurry is not required to maintain a soft surface oxide for the niobium CMP. Thus, controlling the mechanical abrasion of niobium during the CMP process is probably more important than tuning chemical surface reactions.

It was demonstrated that the starting surface (rolled, BCP-treated, or mechanically abraded) did not affect the results of the CMP process. The alumina slurry of pH 7 showed no observable differences in reduction of the surface roughness and improvement of the surface order from the alumina slurry of pH 10. The lack of impact of pH on CMP results supports the previous conclusion that the niobium CMP is dominated by mechanical processes.

The higher material removal rates occurred during CMP with alumina particles than with silica particles. The XPS analysis of the surfaces after CMP showed that the atomic order of the wafers polished with slurries containing alumna particles was higher than the wafers treated with silica containing slurries. This is likely due to the removal of more surface layers damaged by previous processing steps with the harder alumina particles. Therefore, the hardness of the particles seems to play a significant role in a

111 final niobium surface finish. However, it was also observed that the alumina particles were more likely than the silica particles to remain as contamination on the final niobium surface. In addition, the surface roughness could not be reduced to the desired values.

Therefore, a multi-slurry process utilizing alumina particles of 1 µm diameter, which was used to reduce the bulk of surface roughness, then followed by silica particle

(70-100 nm) slurry to achieve surface roughness of 0.024 µm. However, XPS analysis did not show significant improvement of the atomic order of these surfaces. Thus, smooth surfaces were obtained, but defect-free surfaces with high atomic order were not achieved at the two-step processing conditions tested.

3.5 Recommendations for Nb CMP

The collaboration between H.C. Starck Company, and the Chemical and

Mechanical Engineering Departments of Northeastern University successfully demonstrated the proof-of-concept for niobium CMP by producing a smooth niobium surface with a roughness on the order of 10 nm and low defects, as measured by high surface order. Through the static and the CMP experiments it was demonstrated that the niobium CMP is dominated by mechanical abrasion mechanisms as opposed to chemical surface reactions.

The success of the proof-of-concept study suggests that both careful optimization of the CMP process and the development of a polishing process based on CMP for the curved surfaces should be pursued. Based on the preliminary results of this work, parameters such as downward pressure and rotational velocity of the wafer holder, and

112 slurry particle concentration should be explored in a more controlled manner for optimization of the CMP process.

ARXPS should be considered for the future experiments to evaluate niobium oxides at various thicknesses and to determine whether or not serration occurs in the near surface layers. In addition, ARXPS can contribute to a more accurate niobium thickness determination. Care should also be taken to understand the impact of air exposure and time between polishing and surface analysis. Particles on the niobium surface post-CMP can be a concern for the SRF community; thus an effort should be undertaken to understand particle adhesion and methods for post-CMP cleaning of the niobium wafers should be developed.

113

4.0 Monolayer Functionalization of Silicon Surface

Functionalization of the silicon surface via silicon – carbon bonding is desired for the development of the field release mass spectrometer (FRMS). Modified FRMS will enable specific and selective real-time detection of air contaminants. A radical reaction between unsaturated hydrocarbons and a hydrogen terminated surface has been previously shown to produce covalent silicon – carbon bond. This work uses ultra-violet photons (UV) initiation of covalent silicon – carbon bond formation to modify the silicon surface for FRMS. The opening sections of this chapter will give an overview of the field release mass spectrometer, methods for silicon surface modification, and current applications of XPS to monolayer characterization. The experimental details presented include the method that was used to determine monolayer coverage. The results of both silicon surface preparation and the impact of different silicon surfaces on monolayer coverage, conclusions, and recommendations will complete this section.

4.1 Project Description

A new type of Field Release Mass Spectrometer (FRMS) was proposed by Dr.

Roger Giese, Dr. Demetrios Papageorgiou, Dr. Poguang Wang, Dr. Tom Whitaker, and

Dr. Katherine S. Ziemer in order to enable real-time selective monitoring of airborne toxins. This FRMS (Figure 41) will consist of a moving chassis with silicon pillars

(micro- and nanometer scale). The surfaces of these pillars will be modified so that they will have a specific affinity for a desired analyte. This analyte will be attracted to the

114 modified pillar surface, bind non-covalently, and be transported to the section of the

FRMS where polarization of the pillar will occur by an applied voltage. The released analyte will then be directed to the mass spectrometer which will detect it. It is hypothesized that this modified FRMS will allow specific and sensitive detection of substances in real-time. One of the applications of this instrument can be detection of the bioterrorism agents such as anthrax.

Mass Spectrometer

Figure 41: Conceptual representation of FRMS.

Design of the modified FRMS is an interdisciplinary project. Chemical engineers can contribute by engineering the pillar surface and interface for the desired analyte. The semiconductor pillars possess necessary electrical properties and shape to concentrate the applied field to remove the analyte. The surface modification of these pillars with the organic molecules which will provide specificity and selectivity without interfering with the release of the analyte is the goal of this project. Thus, the silicon pillars need to be modified with an organic molecule engineered to selectively and electrostatically attach

115 to the analyte, and to covalently bond to the silicon so that the monolayer will be stable under the applied electric field which will release the analyte.

Instruments for the effective and real-time detection of the biological warfare agents, such as anthrax, plague, smallpox, and viral encephalitedes, are needed for homeland security purposes. Bacillum anthracis (anthrax) is a Category A bioterrorism agent [65]. Terrorist attacks in September and October of 2001 caused eleven inhalation anthrax cases five of which were fatal, according to the Center for the Disease Control and Prevention of the Department of Health and Human Services [65]. In a 2000 review article, Walt and Franz concluded that systems that were available at that time had low sensitivity, high false positive rate, and lengthy response times [1]. Bravada et al. prepared a report reviewing the detection and diagnostics systems for bioterrorism agents in 2004. They found that all 74 detection systems that they examined had serious deficiencies in their evaluation; for example many lacked sensitivity and selectivity information, and specific response time [66]. This indicated that potentially all 74 of the systems were not adequate for the use of protecting the public from the bioterrorism agents.

Current detection strategies of the airborne contaminants include particle [66], absorption [1], fluorescence [66], flow cytometry [66], immunoassays [66], and scattering [1] detectors. Researchers from MIT developed a sensor, called CANARY, utilizing the B cell line to detect various pathogens including anthrax [67]. The B lymphocytes used in this sensor were genetically engineered to emit light within seconds after detection of a pathogen. The commercial PANTHER sensor, a 37 pound unit, uses this CANARY sensor and is, since January 2008, being sold through Innovative

116

Biosensors, Inc. [68]. The CANARY sensor requires a special line of cells which can be easily produced, according to authors. The prepared cells can be stored at room temperature for 2 days, refrigerated for 2 weeks, or frozen indefinitely [67]. Thus, just like sensors utilizing antibodies, this sensor has constraints over its shelf life.

In the current conception, FRMS will not be as portable a unit as PANTHER, because the mass spectrometer is an instrument which is typically used in a stationary environment. However, we do not expect storage conditions to have a strong effect on

FRMS due to the fact that no live cells will be used in this instrument. In addition, a wider range of substances can be detected with FRMS due to the fact that mass spectrometer instead of specific cells will be performing the identification of analyte.

Moreover, the interchangeability of the surface modification on the silicon pillar surface will allow selected targeting of a number of analytes.

Silicon was chosen as the material for a pillar fabrication, because silicon is relatively inexpensive, has high electrical conductivity (when doped), is atomically flat, has ordered surface structure due to crystalline nature, can easily be fabricated into the silicon pillar shape, and is shown to bind organic monolayers covalently [69]. A covalently bonded monolayer on a silicon surface is defined as a maximum number of the molecules that are directly bonded to the surface. Due to steric hindrance a fraction of the silicon surface atoms is expected to be covalently bonded [70].

Figure 42 shows a schematic of how silicon pillar may be modified with compound that has negative charge on the surface, thus attracting a positive analyte that will be anchored on the pillar. When an electric field is applied, the charge of the pillar is

117 either neutralized or reversed, thus ejecting analyte towards mass spectrometer. The covalently bonded compound is then available to detect another analyte.

+ + - +

+ + Electric field Due to tip is applied polarization, analyte is released

Figure 42: Release of negatively charged analyte after electric field if applied.

The decision to attach the monolayers to the silicon surface via a covalent Si-C bond is dictated by stability of the Si-C bond. Table 18 gives typical bond energies for silicon with a few other elements. Here Si-C bond is more stable than Si-H bond; however, one needs to be careful about making decisions about stability from bond energies. For example, Si-F is very stable, but its high polarization toward fluorine makes it very reactive toward nucleophilic attack [71]. The Si-C bond was shown to protect the silicon surface from oxidation over prolong periods of time (days or weeks)

[72]. Thus, by binding surface via stable bonds such as Si-C a more stable surface is produced.

Table 18: Typical bond energies for Si with Si, H, C, O, and F atoms, kJ mol-1 [69]

Si H C O F

Si 210-250 323 369 368 582

118

4.2 Critical Literature Review: Silicon Surface Modification

Covalent attachment of an alkyl group to the silicon surface using a radical initiator was achieved by Linford et al. in 1993 [73]. Since then, some alternative methods emerged for the attachment of organic molecules to silicon surfaces: thermal initiation, electrochemical deposition, photochemical reaction, gas phase reaction in

UHV, and catalytic reaction with metal complexes [69]. In the following section, the most relevant work to this project on creation of the covalently bonded monolayers on silicon is reviewed.

Verification of the covalent attachment to the surface has been done using a number of techniques. Information about the chemical composition, form of attachment, and structure of the resulted layers is most often obtained from the attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy and XPS. Atomic force microscopy (AFM) is used to determine morphology. Even though morphology is important, chemical structure and composition of the attached layers are the most critical pieces of information, and cannot be determined by AFM. Advantage of XPS over ATR-

FTIR, is that XPS allows determination of the elemental concentrations in addition to identification of the bonding states. ATR-FTIR is useful in identifying organic molecules and H-Si bonds as XPS cannot detect hydrogen, but ATR-FTIR is not capable in differentiating between covalently attached and physisorbed carbon species on silicon surface. In addition, the oxidation of silicon can be easily detected with XPS as the Si-O bond interferes with other peaks at 1000 - 1100 cm-1 in FTIR [74]. Most researchers do not scan below 1500 cm-1, as seen in several works such as Voicu et al. in [75] and Liu et

119 al. in [76]. Thus, XPS has been shown to be very useful in obtaining bonding information about monolayers, so silicon modification and monolayer characterization using XPS will also be highlighted below.

4.2.1 Silicon Surface Modification via Wet Chemical Approach

Silicon has been used for decades in the semiconductor industry. It is a relatively inexpensive and well studied material. Silicon’s more common orientations are (100) and (111), which can be commercially obtained as p- or n- doped. Modification of the silicon surfaces with ordered monolayers has been performed to passivate the surface as an alternative to the amorphous oxide obtained through thermal oxidation [77] and to add chemical functionality for biosensing [78]. Organic monolayers on glass and oxidized silicon surfaces have been used for DNA attachment, but amorphous structure of silicon dioxide does not allow desired control over the density of the bonding sites for DNA attachment and poor electrical properties of the silicon dioxide interfere with the electrical signals from semiconductor bulk and organic molecules [77], [75, 79].

Utilization of crystalline silicon which has well-defined crystal structure and silicon- carbon (Si-C) bonds which are not insulating solves these problems.

Formation of Si-C bonds requires hydrogen termination of the silicon surface

[71], [73], [80]. Treatments with 40% aqueous ammonium fluoride for Si(111), and 1-

2% aqueous hydrofluoric acid for Si(100) produce hydrogen termination on these surfaces. The H-terminated surfaces are stable for up to few hours [71]. Figure 43 shows structure of the Si(100) and Si (111) surfaces. In Si(111) silicon monohydrates are present on the surface and they are positioned equal distance from other H-atoms, while

120 on Si(100) each silicon atom has two hydrogen atoms and H-atoms are positioned in distinct rows. From molecular modeling studies, it was determined that due to steric

hindrance approximately 61% of the = Si − H 2 sites on Si(100) and 53% of ≡ Si − H on

Si(111) can terminated with organic group [70].

Figure 43: Schematic representation of H-Si (100) and (111) surfaces [70]

The mechanism of a covalent attachment of the unsaturated organic compounds has been determined. Figure 44 shows a step-by-step mechanism of the olefin attachment. Hydrogen terminated surface is exposed to some kind of initiator which creates radical by removing a hydrogen atom from the silicon surface (reaction a). This radical reacts with the double bond of the olefin creating a surface-bonded alkyl radical

(reaction b) which abstracts hydrogen from a neighboring silicon atom (reaction c). Thus reaction proceeds.

Unsaturated hydrocarbons have been attached to the silicon surface producing alkyl monolayers. However, for the FRMS alkyl monolayers will not be adequate, because not only stability, but polarizability of the monolayer is needed for attachment of the analyte and its release, as shown in Figure 42. Other than alkyl monolayers have been attached to the silicon surface. For example, DNA has been immobilized on the silicon surface which had active groups such as carboxyl (-COOH), amine (-NH2), or hydroxyl (-OH) [77].

121

Figure 44: Schematic of the reaction between unsaturated organic molecule and

hydrogen terminated silicon (111) surface [78].

Few approaches have been used to attach organic molecules: gas phase (UHV) and wet chemical. UHV approach requires expensive equipment and is less frequently used, thus wet chemical reactions were considered for monolayer formation. Wet chemical methods utilized various initiation methods: radical initiators [72, 73], alkyllithium reagents [81, 82], alkyl Grignards [78, 83], [84], Lewis acid catalysts [83],

[85], [78], mechano-chemical (scribing) [86], electrochemical [87], [83], thermal [88,

89], and photochemical[77, 80]. Table 19 shows parameters that a number of researchers used to produce monolayers with specified compounds. It can be seen that most of the methods require degassing of the reagents to prevent oxidation reaction on the silicon surface, and heating to over 100° C for prolonged periods of time is often used. Thermal and photochemical approaches require lower temperatures and shorter reaction times making them more attractive than others. Thus, thermal and photochemical initiation methods will be discussed in more detail below.

122

Table 19: Methods of initiation, reaction parameters, and organic molecules that

were attached to silicon substrates.

Initiation method Parameters Compounds attached

Radical Heat to 100° C for 1 h in presence of Alkyl from neat diacyl

degassed reagents [72, 73] peroxides or in mixture

with olefins [72, 73]

Alkyl lithium 12-24 h with alkyl lithium with Si-Cl Alkyl [81, 82] reagents surface in presence of degassed

reagents [81, 82]

Alkyl Grignards 1.5–16 h at 70-80° C depending on Alkyl [83], [84], [78]

length of the chain [83]; 16 h at 90° C

[84]; 16 h at 85° C in presence of

degassed reagent [78]

Lewis acid Room temperature 12 h in presence of Alkyl [83], [78] catalysts degassed reagents [83]; 18 h at 100° C

in presence of degassed reagent [78]

Mechano- Room temperature 24 h [86] Alkyl chlorides [86] chemical

Electrochemical 0.1 mA cm-2 for 5 min in presence of Alkyl Grignard [83];

Grignard reagent [83]; [87] did not bromophenyl and

specify conditions bromobenzyl [87]

Thermal Heat up to 200° C for 1 h in presence of Alkyl from olefin [72],

degassed reagents [72]; [88], [89], [90]; protected

123

Heat up to 210° C for 2 h in presence of amine [89]

degassed reagents [88], [89], [90]

Photochemical 447 nm light for 15 h in presence of Sacharides [91], alkyl

degassed reagents [91], 350 nm light from olefin[77], [80], acid

for 3-4 h in presence of degassed [77], ester [77], alkynes

reagents [77], 297 – 450 nm light for [80]

30-54 min in presence of degassed

reagents [80]

In 1993, Linford and Chidsey from Stanford University reported first successful preparation of dense long alkyl monolayers covalently bonded to Si(111) and Si(100)

[73]. H-terminated Si(111) and Si(100) were pyrolized in the presence of diacyl peroxides which produced alkyl radical initiators. They hypothesized that alkyl radical absorbs hydrogen from H-Si creating a surface radical which combined with another alkyl radical forming Si-C bond between alkyl chain and Si surface. Full alkyl monolayer coverage was not obtained due to steric hindrance and formation of bonds with carbonyl groups. Comparison with Langmuir-Blodgett films allowed them to determine that approximately 30% of the monolayer covalently attached to silicon was formed. Boiling in water removed approximately 30% of the formed monolayer which was attributed to the presence of hydrolysable bonds. In their subsequent work, Linford and Chidsey tried heating hydrogen terminated silicon samples up to 200°C for 1 h in the presence of 1-octadecene producing a covalently bonded monolayer by thermal initiation

124

[72]. This work showed that covalent attachment using heat as an initiator was possible; however, thermal initiation resulted in a lower quality monolayer formation.

Sieval et al. modified Si(100) samples by heating them to 200°C in presence of alkenes, 10-undecylenic acid, 10-undecen-1-ol, stearic acid, and 10-undecylenic acid esters for 2 h [88]. As expected, acid and alcohol termination resulted in hydrophilic surfaces; contact angle measurements showed high contact angles for alkenes and low contact angles for acid and alcohol terminated surfaces. Contact angle increased with increase in the length of an ester group. Reaction with the unprotected acid, alcohol, and allyl esters resulted in the disordered monolayer as deduced from broadened methylene stretching vibrations in ATR-FTIR spectrum. One of the disadvantages of the thermal hydrosilylation was that a large excess of alkene was required which lead to unnecessary expense [92], so feasibility of using solutions instead of neat reagents for thermal initiation was successfully tested.

UV light has been shown to be successful in attaching unsaturated hydrocarbons

[81]. Terry et al. attached 1-pentene and detected the Si-C bond using XPS. From the scanned-energy photoelectron diffraction, Si-C bond length was found to be 1.85 ± 0.05

Ǻ [81]. Photochemical reaction of ethyl undecylenate with Si(111) was performed by

Boukherroub and Wayner in [93]. They used various ratios of ethyl undecylenate in 1- decene to react with the H-Si(111) in 300-nm light. The ratio of the ester carbonyl stretch to methylene stretch as detected by ATR-IR was found to be the same as the molar ratio of the ethyl undecylenate in the reaction mixture, as seen in Figure 45. Thus, potential of formation of multifunctional surfaces by adjusting molar ratio of the mixture was demonstrated. Alkyl group coverage was determined to be 30-40% of the Si(111)

125 atom. Ester group was successfully removed using NaBH4 leaving alcohol terminated surface which was esterified. The terminal ester group was shown to be versatile with conversion greater than 90% as determined by FTIR and no oxidation of silicon [93].

Figure 45: Mole fraction of reactants in the reaction mixture was found to be the

same as in the monolayer, which was determined by ATR-FTIR [93].

Cicero et al. determined that the hydrogen terminated silicon surface is activated by UV light by breaking the H-Si bond [80]. H-Si bond strength is approximately 79-84 kcal/mol [94], [95], [96] which corresponds to 350-nm photon [80]. The reaction between silicon surface and alkenes is a radical reaction, where UV light produces free radicals by cleaving Si-H bond [80], see Figure 44.

Hamers and his group pioneered in attaching DNA molecules to a hydrogen- terminated silicon surface [79]. First, H-Si(111) was modified with ω-undecylenic acid ester by UV light. This surface was hydrolyzed in the presence of potassium tert- butoxide producing carboxylic acid terminated surface. Carboxylic group was utilized to obtain linkage with the cross-linker which later formed a bond with DNA. Thus, Hamers et al. obtained carboxylic acid terminated surface after deprotection of the ester.

Voicu et al. successfully modified Si (111) surface with terminal carboxylic acid

(undecylenic acid) in one step using 300-nm UV light for 3 h [75]. N- hydroxysuccinimide (NHS) was allowed to react with the surface modified with the

126 undecylenic acid producing NHS-activated ester layer [75]. The single-strand DNA was immobilized on this surface [75] proving that silicon surfaces can be rather easily modified for various biological applications. Use of one-step process not only simplifies

DNA attachment, but it results in higher quality monolayers, because hydrolysis was shown to damage attached monolayer [75] and not remove all of the protecting groups

[97]. In addition to producing carboxylic acid monolayers in one step, Voicu et al. showed that silicon surface can be photopatterned to have regions resistant to nonspecific adsorption or chemical reaction.

Visible light can be used to attach organic molecules to the silicon surface as well.

Buriak and co-workers observed attachment of the organic molecules to porous H-Si

[98], [99]. These researchers concluded that the mechanism for the visible light attachment involves an attack of the hydrogen-terminated surface by the unsaturated alkene or alkyne [98]. Zuihof studied dependence of wavelength and substrate doping [ p-, n- Si (100) and p-, n- Si (111)] on monolayer coverage using ARXPS, IR, and contact angle measurements [100], [101]. Visible light was observed to result in monolayer formation, but reaction times had to be extended to 14 h from usual 3-4 h for UV light induced reactions. UV light caused change in color and composition, while no change in precursor using visible light was observed, indicating that UV light alters precursor.

Monolayers with undecylenic acid resulted in contact angle of 66-68° which was lower than that obtained by [77] indicating presence of the low density monolayers; no IR or

XPS data for undecylenic acid modified surfaces was provided.

Photochemical reaction utilizing UV light is attractive because of the mild reaction conditions (room temperature), relatively short reaction times, and ability to

127 pattern the surface if necessary. In addition, for the photochemical modification continuous purging of the solvent with inert gas is not requires as is for many methods including thermal. Continuous purging requires the use of the Schlenk lines which utilize dual manifold supplying inert gas and vacuum. The Schlenk lines are costly and must be installed in the hood to contain chemicals and fumes. For UV method, sample needs to be placed into a degassed reagent and purged, and then the tube with silicon and reagent can be isolated from atmosphere and placed in a photoreactor. So only purging and degassing needs to be done using vacuum and inert gas lines.

4.2.2 Application of XPS for Analysis of Modified Si Surfaces

High surface sensitivity and ability to differentiate between different carbon and silicon species of XPS have been successfully utilized in characterization of the covalently bonded monolayers on silicon surface. Many researchers rely on XPS to provide them with information necessary to evaluate quality of the obtained films. Few of the papers that illustrate utilization of XPS for analysis of covalently bonded monolayers are reviewed.

Research group of Nathan S. Lewis from California Institute of Technology performed very thorough XPS analysis of modified surfaces. In one of their publications, they evaluated covalently bonded monolayers which were attached via chlorination/alkylation, Lewis acid-mediated reduction, and electrochemical oxidation with Grignard reagent [83]. The authors characterized their films with XPS to determine chemical composition and silicon oxidation, and time-resolved radio frequency photoconductivity method to determine the recombination velocities. The

128 nonmonochromated Al anode and the 35° take-off angle between surface and detector were used. Figure 46 demonstrates that chlorination and alkylation of H-terminated silicon surface could be detected with XPS. Spectrum (a) shows a survey spectrum of H- terminated silicon; silicon and carbon (red circle) were observed. The authors stated that only silicon peaks can be seen; however, carbon peak is clearly present. Carbon on this spectrum indicates contamination that remained on the surface after cleaning and etching of the silicon sample. Spectrum (b) represents the silicon surface after chlorination reaction. Peaks for chlorine, oxygen, and carbon appeared on this spectrum. Carbon and oxygen are common contaminants on the silicon surface. Finally, spectrum (c) shows an increase in carbon and oxygen concentrations due to alkylation which substituted silicon- chlorine bond (absent) for silicon-carbon bond.

Figure 46: Survey XPS spectra show (a) H-Si, (b) chlorinated with saturated PCl5,

and (c) alkylated with C8H17MgCl [83].

129

The authors gave area values normalized to Si2p peak instead of the relative atomic percent. Comparison of these values for different samples leads to conclusion that some carbon was on surfaces prior to alkylation. For example for one of the samples, after chlorination with Cl2(g) the normalized peak area for carbon was 0.3±0.2, while after alkylation with C2H5MgX it was 0.36±0.06 [83], unfortunately no values were given for H-Si. Carbon contamination on the surface prior to modification raises doubts about actual attachment of the organic groups with Si – C covalent bond.

Any peaks in region between 100 and 104 eV Webb et al. attributed to Si+ - Si4+ which was consistent with [102]. In addition, ratio of peak areas of oxides (SiOx) to total

Si 2p peak was used to determine the overlayer thickness (Equation 17).

⎪⎧ ⎡ I 0 I ⎤⎪⎫ Si ov d ov = λov sinθ ⎨ln⎢1+ 0 ⋅ ⎥⎬ 17 ⎩⎪ ⎣ I Si I Si ⎦⎭⎪ Where: dov = overlayer thickness (nm)

λov = attenuation factor for the overlayer, nm (for SiO2 = 2.6 nm [83])

θ = electron take-off angle relative to the sample surface

0 I Si 0 = instrument normalization factor related to the sensitivity and I Si

abundance for the substrate and overlayer species (1.3 for the instrument

used by Webb and Lewis in [83]),

I ov = measured peak area ratio of silicon dioxide to silicon. I Si

Because one monolayer of SiO2 is 0.35 nm [83], number of monolayers could be obtained from dov. For example, when H-terminated silicon is exposed to air it oxidizes.

Webb et al. determined that after 216 h in air H-Si formed 1.4 monolayers of silicon

130 dioxide [83]. Figure 47 shows the peaks that were used to determine monolayer coverage. The Si2p spectrum of the freshly etched H-terminated silicon (a) has only one peak centered around 99.4 eV, while spectra corresponding to the samples exposed to air for 4 - 216 h exhibit an additional peak around 103.4 eV. This peak corresponds to Si-O bonds on silicon surface. Silicon in different oxidation states was shown to have peaks in the region from 100 to 104 eV [102]. Comparing the oxide monolayer coverage values, researchers determined that during the first ≈30 h oxidation was negligible, and after 200 h approximately 0.9 of SiO2 monolayers was formed [83]. Despite oxidation of alkylated samples over time, these surfaces had long charge carrier lifetimes and low surface recombination velocities indicating that alkylation was a viable passivation method [83].

Figure 47: Oxidation of H-terminated Si(111) in air was monitored by observing

Si2p peak. Curves represent (a) H-Si, (b) 4 h in air, (c) 72 h in air, and

(d) 216 h in air [83].

In one of their subsequent works, Webb and Lewis used synchrotron XPS at the

National Synchrotron Light Source at Brookhaven National Laboratory [103]. The

131 incident energy of 140 eV and the take-off angle of 45° were used for this analysis. In this experiment, chlorination followed by a reaction with the alkylmagnesium compound was studied. Use of the synchrotron radiation allowed to Webb et al. to differentiate between two spin states of silicon (Si2p3/2 and Si2p1/2) and subtract the higher energy component of the lower intensity (Si2p1/2). The analysis of the Si2p spectrum with the synchrotron source revealed that the Si-H bonds were detected 0.15 eV to the higher binding energy than the main Si2p3/2 peak, while the Si-C bond was found to be 0.34 eV higher in the binding energy than Si2p3/2 peak [83]. Differentiation between these two states (Si-H, Si-C, and some lower oxidation states of Si) will be questionable with the use of non-synchrotron source. Many researchers have access to monochromated X-ray source, which allows one to resolve between two spin states of silicon due to better resolution than nonmonochromated sources; however, they do not claim to be able to determine Si-C bonds in the Si2p spectrum.

Unlike Webb and Lewis who concentrated on studying of the Si2p peak, Strother et al. monitored the C1s peak in addition to Si2p [79, 97]. An attachment of a protected form of the unsaturated carboxylic acid was obtained using UV light irradiation. These surfaces were later deprotected and allowed to react with a linker for further attachment of DNA. Figure 48A demonstrates that trifluoroethyl of undecylenic acid has four peaks corresponding to C-C in alkyl chain, CH2-O of ester, C(O)O of carboxylic group, and C-

F3. Trifluoroethyl ester was especially synthesized to give very distinct shift which was easy to monitor. Fluorine, as the most electronegative atom, depletes carbon atom of the electron density shifting its binding energy by +8.4 eV which can be easily distinguished from other shifts. After hydrolysis with potassium tert-butoxide, as seen in Figure 48B,

132 peak associated with fluorine bonded to carbon disappeared, fluorine F1s spectrum showed no fluorine present as well, not shown here. Thus, Strother et al. monitored deprotection of the carboxylic acid ester and demonstrated that XPS is an extremely useful technique in studying the covalently attached monolayers on a silicon surface. In their other publication, Strother et al. attached protected amine to carboxylic surface [97] and were able to monitor its deprotection as well.

A B

Figure 48: XPS allows monitoring of chemical reaction such as deprotection of

carboxylic acid ester. Plane A shows sample with trifluoroethyl ester of

undecylenic acid on the surface, and plane B shows surface with

undecylenic acid [79].

Voicu et al. monitored attachment of the carboxylic acid to Si(111) and a reaction with N-hydroxysuccinimide (NHS) to produce succinimidyl ester which could be used as a specific anchor for DNA strands [75]. Figure 49 shows the two C1s spectra for the samples before and after conversion to the succinimidyl ester. Before the reaction,

Figure 49a, C1s peak was fitted with three peaks corresponding to the carbon – carbon in the alkyl group, C-O and C(O)OH. C-O bond was not present in the molecule, so it probably was due to contamination. In addition, there was a low energy component

133 which was not fitted at all. Other researchers found that the Si-C bond is observed ~0.9 eV to the lower energy from alkyl carbon on the C1s spectrum [104], [81], so it is possible that the unassigned peak was produced by the Si-C bond. Succinimidyl ester was detected after modification, Figure 49b, by an additional peak 286.6 eV [75].

Similar to previous spectrum, C-O peak was observed and the Si-C peak was not fitted.

The ratios of carbon to oxygen, which was determined from XPS peak intensities, for both acid and NHS-ester modified surfaces were lower than expected indicating that oxidation of silicon was competing with formation of Si-C bonds.

COOH Si

COONHS Si

Figure 49: C1s spectra of (a) the carboxylic acid-terminated surface and (b) NHS-

ester terminated surface [75].

The XPS analysis with various X-ray sources was used to successfully characterize the attached monolayers on the silicon surfaces. Synchrotron and monochromated sources provide superior energy scale resolution allowing determination of silicon spin states and low energy shifts; however nonmonochromated sources were

134 shown to be successful in determining chemical composition, bonding states and monolayer coverage.

4.2.4 Summary

The modified field release mass spectrometer when completed will allow real- time detection of the bioterrorism agents. The silicon pillars will carry covalently and non-covalently bonded organic layers which will allow selective real-time capture of analyte. Application of an electrical field to the pillar will create polarization between the organic layer on the pillar and analyte, resulting in the ejection of the analyte toward a mass spectrometer. Attachment of the organic layers to the silicon surface via a silicon – carbon bond is desired due to Si-C bond stability, electrical properties, and predictable bond density. The photochemical method using UV light for attachment of monolayers to the surface is preferred because of the mild reaction conditions and short reaction times. Methods requiring thermal energy input may be harmful for our application with the micro- and nanoscale features. In addition, photopatterning was shown to be possible with the UV light initiation.

X-ray photoelectron spectroscopy was shown to be a powerful technique allowing one to monitor modification progress, unwanted oxidation of silicon, changes in the chemical structure of the monolayers and their thickness. XPS which utilizes synchrotron and monochromated sources has advantage of superior energy scale resolution; however regular nonmonochromated XPS instruments were shown to provide essential information about surface composition and structure of the attached monolayers.

135

4.3 Experimental Section for Silicon Modification

The surface of the silicon wafers is contaminated with organic residue (waxes, oils, etc.), particles, and metal containing compounds [65]. Degrease and RCA cleaning provide contamination removal, while Fenner etch removes native oxide producing hydrogen termination. Degrease with trichloroethylene (TCE), acetone, and methanol removes gross organic contamination such as finger prints, waxes, and dust [65]. RCA cleaning procedure was developed in 1970 by Werner Kern to remove particle and metallic contamination from the surface of the silicon wafer [105]. The wafer is oxidized and etched a number of times during this procedure which provides contamination removal. Fenner etch removes native silicon oxide leaving hydrogen terminated surface

[106]. Photomodification of the H-Si surfaces with the UV light produces the monolayers on silicon surface which are monitored with XPS.

4.3.1 Materials

Methanol (electronic grade), trichloroethylene (TCE) (electronic grade), acetone

(OPTIMA or electronic grade), hydrochloric acid (37 %, trace metal grade), hydrofluoric acid (49%, trace metal grade), undecylenic acid, ammonium hydroxide (30%, OPTIMA grade), 1-dodecene (95%), absolute reagent alcohol (ARA) (HPLC grade), sulfuric acid

(98%, reagent grade) tetrahydrofuran (THF) (anhydrous, stabilized 99.9%), and hydrogen peroxide (30%, certified ACS grade) were purchased from Fisher Scientific. Ethyl undecylenate (97%) and 1,1,1-trichloroethane (ACS reagent grade) were purchased from

136

Sigma-Aldrich. Ultrahigh purity water was obtained from Barnstead filtering system,

Model D4741, using filter kit D4801.

Si(100) p-type was generously supplied by Dr. Demetrios Papageorgiou.

Ultrahigh purity and high purity argon was purchased from Med-Tech Gas company.

4.3.2 Experimental Procedure: Degrease, RCA Cleaning, Fenner Etch, Ar Purge,

Photomodification, Stability Protocol

Degrease Procedure

Si (100) chip (0.5 x 1.5 cm) was placed into a Teflon basket which was submerged into a beaker with TCE at 70°C. After 5 min, the basket was transferred into a second beaker with TCE at 70°C for 5 min. The sample was not allowed to dry between the steps by rinsing with acetone or methanol during transfer. The basket was then transferred into a beaker with acetone at 55°C for 2 min. The sample in the basket was then placed into the first beaker with methanol at 70°C for 5 min. Upon completion, the sample was placed into a second beaker with methanol 70°C for another 5 min. The sample in the basket was rinsed with methanol for 5 s and blow-dried with ultrahigh purity (UHP) argon and rinsed with DI water for 5 min.

RCA Cleaning and Fenner Etch Procedures

After rinsing with water for 5 min the sample was dried with UHP the basket with sample was placed into the hot (80°C) standard clean 1 (SC1) solution (30% NH4OH:

30% H2O2: DI H2O = 0.5: 1: 5) for 10 min. The sample was then rinsed with DI water for 5 min and placed into a Teflon beaker with HF:H2O solution (1: 50) for 15 s. The

137 sample was rinsed for 30 s and transferred into a standard clean 2 solution (37% HCl:

30% H2O2: DI H2O = 1: 1: 6) for 10 min. The basket with sample was rinsed with DI water for 20 min and placed into a Teflon beaker with HF: ARA: DI H2O for 2.5 min; the sample was then rinsed with ARA and blow-dried with UHP Ar.

Photomodification Procedure

Reagent was placed into a quartz Schlenk tube. An adapter with a stopcock was used to isolate the inside of the tube from the ambient air. The tube was purged with Ar three times using a vacuum line. The tube was then removed and opened to air to place the sample in. Clean and etched silicon sample was placed into a slot in the sample holder and lowered into a tube with the purged reagent. The tube was purged again to remove any oxygen that entered during sample placement.

Sample in the tube was placed into a photoreactor. The sample was located so that it was facing the central lamp. The fan was turned on and the lamps were powered.

Sample stayed undisturbed in the photoreactor for 6 h. Upon completion, the lights and the fan were turned off. The adapter was removed from the tube, and she sample holder was removed from the tube. The sample was then rinsed with 30 mL of 1,1,1- trichloroethane and 30 mL of THF, and blow-dried with UHP Ar.

Air Stability Protocol

A sample after photomodification and subsequent analysis was placed into a clean tin can and was kept in the dark for 8 days. After which it was analyzed with XPS.

138

4.3.3 XPS: Data Acquisition and Manipulation

Samples after cleaning and etching, photomodification, and air stability were analyzed using an XPS (X-ray Photoelectron Spectroscopy) manufactured by PHI (model

04-548 Mg/Al dual anode, non-monochromatic x-ray source and a 10-360 hemispherical analyzer). Samples were loaded onto a sample holder which was placed into a loading dock where pump down to 9.0·10-7 Torr was performed prior to transfer into the analysis chamber. During XPS analysis, the background pressure in 10-9 Torr was routinely attained. Analysis was performed using Mg anode (1253.6 eV). Data acquisition and manipulation were performed using AugerScan software (Version 3.2 Beta), RBD

Enterprises, Inc.

Spectra were obtained at 90° and 30° take-off angles. All survey spectra were obtained with pass energy of 89.45 eV and step size of 1 eV/S. An average of 5 survey spectra of 0 to 1000 eV was obtained for etched samples, modified, and after stability in air. Elemental spectra for the individual elements were obtained using pass energy of

35.75 eV and step size of 0.05 eV/S. Table 11 shows energy range and number of sweeps used to determine composition and bonding states on the silicon surface before and after modification, and after exposure to air for 8 days.

After the elemental spectra were obtained, the background was subtracted using

Shirley (integrated) background function. The relative atomic percent for the samples was calculated using sensitivity factors listed in Table 2. Peak fitting was performed using the Gaussian peak function with the Gaussian component of 80% and Lorentzian component of 20%. The FWHM for all carbon peak components was kept at 1.50 eV.

Peak positions were assigned according to [6], relevant peak shifts are show in Table 3.

139

Oxygen peak was fitted with one or two peaks with the FWHM of 1.80 eV, no peak assignment was performed due to broadness of the peak and the fact that oxygen is a common contaminant. For fitting of the silicon peak, slope of the experimental peak was used as a guide to determine the FWHM, which was typically around 1.30 eV. Low

FWHM for Si peak is due to high atomic order of the crystalline silicon wafers.

Table 20: Typical energy ranges and number of sweeps that were used to obtain

silicon surface composition.

Element, peak Energy Range, eV Number of sweeps

Carbon, C1s 281-292 30-35

Oxygen, O1s 525-540 20

Silicon, Si2p 94-107 20

Fluorine, F1s 680-700 10

Thickness of the overlayers was determined using Equation 18 from [2]. Area under the curve of silicon was used as the intensity of the substrate. Area of the carbon peak was used as intensity of the overlayer. To calculate monolayer thickness after modification, starting carbon area was subtracted from that of the carbon area after photomodification. Sensitivity factors of silicon and carbon were used for sensitivity factors of substrate and overlayer, respectively. Sensitivity factors used can be found in

Table 2. Molar density of the substrate was 0.083 mol/cm3 [2], of carbon contamination was 0.033 mol/cm3 [107], and of undecylenic acid was 0.052 mol/cm3. Molar density of undecylenic acid was calculated from estimation of the chain length (1.3 nm using approach from [88]), molar mass (184.28 g/mol), and assumption of 50% termination at

140 best, i.e. one monolayer will have 3.4·1014 number of the undecylenic molecules attached on 1 cm2.

⎡⎛ I ⎞⎛ S ⎞⎛ ρ m ⎞ ⎤ d = ln ⎜ ov ⎟⎜ sub ⎟⎜ sub ⎟ +1 λ sinθ 18 ov ⎢⎜ ⎟⎜ ⎟⎜ m ⎟ ⎥ ⎣⎢⎝ I sub ⎠⎝ Sov ⎠⎝ ρov ⎠ ⎦⎥ Where: Iov = intensity of the overlayer element peak area

ISi = intensity of the substrate

Ssub = sensitivity factor of the substrate

Sov = sensitivity factor of the overlayer

m 3 ρsub = molar density of the substrate (mol/cm )

m 3 ρov = molar density of the overlayer (mol/cm )

λ = inelastic mean free path (nm)

α = take-off angle

After the thickness of the overlayer was determined, the number of atoms of carbon could be determined. Because one molecule of undecylenic acid contains 11 carbon atoms, from the total number of carbon atoms, the percent of the monolayers of the undecylenic acid is calculated.

141

4.4 Results and Discussion: Silicon Surface Modification

Our goal was to obtain and study covalently bonded monolayers on silicon surface using UV light initiation. A hydrogen terminated silicon surface is necessary to obtain a silicon-carbon covalent bond. RCA cleaning procedure is a standard procedure in the semiconductor manufacturing process [108]. Hydrogen termination of Si(100) is achieved by exposure of the surface to the solution containing hydrofluoric acid.

Typically hydrofluoric acid solution is water based [108], however, Fenner et al. determined that alcohol based solutions produced higher quality surfaces with lower contamination levels due to better wetting properties of alcohol [106]. In this work, RCA cleaning and Fenner etched were used to obtain clean hydrogen terminated surfaces.

Results from the experiments targeted on obtaining of reproducible clean H-terminated

Si(100) are described in section 4.3.1.

Hydrogen terminated surfaces were modified with undecylenic acid in a custom- built photoreactor. The effect of a starting carbon contamination on the monolayer coverage was determined using ARXPS. Results from the photomodification experiments are discussed in section 4.3.2.

4.4.1 Starting Surface: RCA Cleaning and Fenner Etch

Hydrogen termination of the Si(100) surface is obtained removal of the native oxide layer in a dilute solution of hydrofluoric acid. However, not only oxide needs to be removed. Carbon is a typical contaminant on the silicon surface due to processing, storage, and wafer handling. Figure 50 shows comparison of the survey spectra from (a)

142 the degreased sample, (b) a sample after unsuccessful cleaning and etching, and (c) a sample after successful cleaning and etching. Spectrum (a) shows significant peaks for oxygen and carbon in addition to silicon peaks. After the unsuccessful cleaning and etching, oxygen concentration decreased, but carbon content increased (b). Successful cleaning (c) shows very low signal from carbon and oxygen, and strong peaks for silicon.

Auger Auger C (KLL) O(KVV) F1s O1s C1s Si2s Si2p

Count b

1000 900 800 700 600 500 400 300 200 100 0 Binding Energy, eV

Figure 50: Survey spectra at 90° take-off angle of (a) degreased Si(100) , (b)

unsuccessful cleaning and etching, and (c) successful cleaning and

etching.

Inability to repeatedly clean the silicon surface to a low carbon and oxygen content motivated us to look into each of the procedures. RCA cleaning has been used in the semiconductor industry for a number of years, thus information on the purpose of each step and chemical reactions associated with them could be found. The first cleaning

143 step, SC-1 which contains ammonium hydroxide and hydrogen peroxide is used to remove organics by an oxidative dissolution [109]. Particles and some metals are removed as well. This solution slowly dissolves oxide (0.09-0.4 nm/min at 80° C for

1:1:5 ratio [109]) and forms new one by oxidation with hydrogen peroxide. SC-1 solution must heated and kept at 70° C, but temperature should not exceed 80° C to prevent hydrogen peroxide decomposition and excessive loss of ammonia [109]. Thus, temperature should be controlled during this step.

Typically, after SC-1 solution, the sample is transferred into a beaker containing a solution of hydrofluoric acid (1:50 ratio by volume of HF: DI water). This treatment is done to remove the oxide layer which may contain entrapped contamination. After oxide stripping, sample is oxidized in a solution containing hydrochloric acid and hydrogen peroxide (6:1:1 vol. of DI water:H2O2:HCl) at 70° C for 5-10 min [109]. This solution is supposed to remove metal ions such as copper, gold, aluminum, iron, magnesium, and zinc [109].

Finally, in order to obtain hydrogen termination, Si(100) needs to be treated with a solution of hydrofluoric acid. Fenner et al. showed in [106] that if alcohols were used as a solvent in the oxide stripping solution quality (lower contamination) of the samples was higher, it was hypothesized that better results were due to better wetting properties of alcohol [106]. In Fenner’s work, HPLC grade reagent alcohol was used as a solvent for the final hydrofluoric acid treatment (1:1:10 vol. of HF: DI H2O: Reagent alcohol).

Since development of the RCA cleaning process a considerable research has been done to understand reasons for the cleaning failures. The following list gives some of the most critical points that impact quality of the wafers after the cleaning process:

144

1. Even ultrapure water can contaminate silicon surface with ions, bacteria, organic and

inorganic particles [109].

2. Relative humidity is critical for particle removal, as it can condense on a wafer and

fill in space between a particle and the surface [109].

3. All chemicals must have low particulate and metal contamination.

4. Hydrofluoric acid solutions should have extremely low metal levels, because metals

like copper and gold can plate on the wafer surface [109].

5. HF solutions should be free of organics as subsequent cleaning and rinsing solutions,

because hydrophobic surface is very prone to hydrocarbon adsorption [109]. And

hydrophobic surfaces are very susceptible to particulate contamination especially

when exposed to gas-liquid interface [110].

6. Unless high-purity and point-of-use ultra-filtered water and particle-free HF solution

is used under controlled conditions, recontamination of the samples will occur [109].

7. SC-1 solution was observed to roughen the surface due to dissolution of silica [110],

[109]. Lowering of the ammonium hydroxide component is recommended [109].

8. Ultrapure water should not contact any contaminated or metallic surfaces before

touching a sample, as it will degrade its quality even after a brief exposure [109].

From the list above, one can see that air, water, chemicals, mixing ratios, temperature, and technique are important. The total of 142 cleanings was performed some of which were done to check repeatability and some to check the effect of a certain parameters. Our goal was to obtain samples with less than 0.08±0.00 of carbon to silicon ratio which is equivalent to less than 7.0±0.5 % of carbon on the surface. Table 21 shows changes to the procedure that were done and values for carbon contamination that were

145 obtained. The values are given as a ratio of the adjusted area of C1s peak to Si2p peak

(adjusted area = area under the curve divided by a sensitivity factor).

Table 21: Selected experimental results directed to determine source of carbon

contamination

Sample Change in procedure Purpose C/Si, Conclusion

±3%

NM 2-04b Degrease, no RCA, Starting surface 0.47 n/a

no FE

NM 1-98 RCA, FE Typical FE 0.19 n/a

NM 1-63 Full RCA cleaning, Check if FE results 0.29 RCA failed

no FE (final HF etch in C gain

in RA)

NM 2-06 No RCA, just FE Check if RCA results 0.21 FE removes

in C gain contamination with

oxide

NM 2-08 No RCA, no FE, Check if RA 0.45 Water based solution of

only 30 s HF:H2O contributes to C HF results in high C

(1:50) followed by contamination contamination

30 s H2O rinse

NM 2-10b No rinse after Test if minimization 0.20 Modification in

degrease, SC-1, rinse of H2O rinses help, procedure did not result

30 s, HF:H2O, 30 s remove SC-2 in lower contamination

H2O rinse, FE

146

NM 2-11b Repeat 2-10b but Check if NH4OH 0.15 NH4OH may be

lower NH4OH to 5 roughens the surface roughening the surface

ml (0.5:1:5) leading to higher C

accumulation

NM 2-12 Repeat 2-11b, but Test if higher 0.22 Increase in H2O2 did

increase H2O2 hydrogen peroxide not help

results in better

carbon removal

NM 2-12b Improve NM 2-11b Test if hydrophobic 0.11 Removal of HF: H2O

by removing HF: surface after HF:H2O improved C level,

H2O attracts unable to repeat

contamination

NM 2-55 Full RCA, FE Experiment in Dr. 0.08 Water possibly made a

Murthy’s lab with DI difference, but unable

water feed into to repeat

filtering system

NM 2-72 SC-1 (15 mL H2O2: 5 Try clean water with 0.05 Proved that water is

mL NH4OH: 50 mL improved SC-1 important and removal

H2O), 5 min rinse, or certain RCA steps is

FE, water feed for justifiable, repeated

filters RODI in Egan once

147

The first two samples that are shown in Table 21 are for the degreased sample and typical RCA and FE procedures which are used as controls to determine effect of the changes that were made in the later studies. It can be seen that after a degrease procedure considerable carbon contamination was on the surface of the sample (0.47±0.01). Carbon content was decreased by cleaning and etching procedures to 0.19±0.01 (14.7±1.0 %); however, this values is greater that our goal.

Experiments to detect source of carbon contamination were performed. Sample

NM 1-63 was performed to determine if FE step resulted in high carbon concentration.

Only RCA cleaning steps (SC-1, HF:H2O, SC-2) according to a procedure described in a section 4.2.2 were performed. From C/Si it can be seen that cleaning steps either did not work or the sample was contaminated between cleaning and analysis, transfer time was typically 10-15 min. The next experiment, NM 2-06, showed that FE alone resulted in less carbon contamination than the RCA cleaning procedure. Reduction in carbon may have occurred due to removal of the surface oxide layer which may have contained carbon contamination. In addition, results from NM 2-06 indicated that the RCA procedure was either not removing any contamination or recontamination of the surface occurred.

FE uses reagent alcohol as a solvent, so there was a concern that during a drying process after FE some alcohol was dried on the surface. Experiment, NM 2-08, where aqueous solution of hydrofluoric acid was used, and carbon content was almost as high as for the degreased sample, 0.45±0.01, which indicated that 1) reagent alcohol was not contributing to the carbon contamination, 2) DI water from Barnstead system that was used for rinsing and solutions might be contaminated. Therefore, from experiment NM

148

1-63, NM 2-06, and NM 2-08 we can conclude that FE step was not responsible for the carbon uptake and that either RCA cleaning steps (SC-1 specifically) failed to remove carbon contamination or the sample gained carbon during sample handling.

The next few experiments were done with modified RCA cleaning procedure.

Due to the fact that metal contamination on our samples was bellow detection limit of

XPS, and the metallic contamination was not critical for our application, SC-2 step was removed to simplify procedure and eliminate opportunities for mistakes. Suspicion that water may be a source of the carbon contamination was tested on sample NM 2-10b when water rinsing was reduced to 30 s. C/Si ratio value obtained was 0.20±0.01 for

C/Si ratio, which was lower than that for the as-received degreased and full RCA/FE samples, indicating that prolonged water rinsing (up to 20 min in original procedure) might have contributed to the carbon contamination gain. Removal of SC-2 might have affected outcome of this experiment by simplifying process and reducing exposure to water and aqueous solutions. However, these changes did not lead to the target value of

≤0.08.

The effect of the ammonium hydroxide concentration on the carbon contamination was studied, because ammonium hydroxide solutions are known to roughen silicon surface and rougher surfaces have higher surface area for carbon contamination. In experiment NM 2-11b, the ammonium hydroxide volume used in SC-1 solution was reduced in half (NH4OH:H2O2:H2O = 0.5:1:5) resulting in a reduction of carbon concentration and C/Si ratio to 0.15±0.00. This result indicated that ammonium hydroxide may have been roughening the surface producing higher surface area for contamination.

149

Increase in hydrogen peroxide to 0.5:1.5:5 (NH4OH: H2O2: H2O) concentration to test if more of the oxidizing agent would remove more carbon contamination was tested in the experiment NM 2-11b. Carbon to silicon ratio for this sample did not improve

C/Si (0.22±0.01), indicating that failure to remove carbon might be originating in other than SC-1 step. As mentioned above, wafer quality is especially important for the hydrophobic surfaces such as surfaces produced after hydrofluoric acid etch, and if no high purity water is available, generation of the hydrophobic surface should be avoided.

Thus the aqueous solution of hydrofluoric acid was not performed in experiment NM 2-

12b. Carbon contamination decreased to C/Si value of 0.11±0.00, indicating that the hydrophobic surface that was created during HF treatment might have attracted considerable contamination possibly due to rinsing with water.

Suspicion that DI water used in the experiments was contaminated motivated us to sent water samples to Barnstead International for testing. Results for total organic carbon (TOC), 0.26 and 0.29 ppm, indicated that filters might have been exhausted.

Through communications with Barnstead technical staff, few solutions were found: to replace filters with Organic-free filter set which should give TOC less than 10ppb, and to use DI water as feed into the filtering system. Thus, experiment NM 2-55 was performed in Dr. Murthy’s lab when full RCA and FE procedures were performed. This experiment resulted in C/Si value of 0.08±0.00; four attempts to repeat this result in Dr. Murthy’s lab were unsuccessful. System was moved to Egan Research Center and was attached to

RODI water feed on the 2nd floor. Two consecutive trial of modified procedure as in experiment NM 2-12b (higher hydrogen peroxide concentration) were performed reducing C/Si ratio to 0.05±0.00 (NM 2-72). Attempts to reproduce this result were

150 unsuccessful. Suspicion that the filters were not producing low TOC arose. Thus, the water sample before and after filters were sent to Barnstead; results indicated that before water entered filtering system with TOC of 0.10 ppm and exited with TOC of 0.35 ppm.

This indicated that carbon addition occurred when water passed through the filtering system. Thus water was found to be one of the sources of carbon.

Additionally, inconsistent results were observed for the samples that were obtained using the same procedure as in experiment NM 2-06, i.e. degreased and FE only. This procedure resulted in C/Si values from 0.24±0.01 to 0.25±0.01 (three out of three trials NM 2-67, 68, 69) in September 2007. However, when the same procedure was repeated in November 2007, four consecutive experiments resulted in C/Si ratio of

0.04±0.00, 0.05±0.00, 0.08±0.00, and 0.04 ±0.00 (NM 2-81, 82, 82b, 83, where NM 2-83 was cut from a new Si(100) wafer). The samples for the experiments NM 2-55, 72, 74,

67, 68, 69, 81, 82, and 82b were cut from the same Si(100) wafer. Nine subsequent trials after NM 2-83 resulted in C/Si ratio consistent with September results (0.18±0.01 –

0.30±0.01). We cannot explain this discrepancy.

Correct drying process is critical in obtaining contamination-free wafers. All wafers in this work were dried with ultra-high purity argon. A stream of argon from a nozzle was directed to the sample surface at the angle, so that instead of drying liquid on the sample it was pushed off the sample leaving a very thin layer of liquid that evaporated. This method is similar in principle to centrifugal method used in the industry, where the wafer is spun so that liquid is pushed to the sides by centrifugal force and the thin layer of liquid that is left on the surface dries leaving minimal residue [109].

Even though an effort was made to keep drying process as consistent as possible, some

151 variation in argon flow rate, impact angle, and duration of drying was inevitable. In addition, the nozzle may have had carbon contamination on it despite sonication and rinsing with organic solvent. However, a problem with drying process does not explain why short term results were consistent. Humidity and air quality were shown to be critical for wafer cleaning [109], but no significant correlations were noticed over the time that wafers had low and high contamination.

4.4.2 Effect of Initial Carbon Contamination on Monolayer Coverage

A study of the effect of the starting carbon contamination on the silicon surface modification was motivated by problems in obtaining sufficiently clean silicon samples consistently. In addition a number of studies had carbon contamination on their H-Si surfaces or no information on the H-Si surface condition, for example [2], [78], [83]. In these experiments the H-terminated silicon samples were allowed to react with undecylenic acid for 6 h in a photoreactor. Undecylenic acid is an organic acid containing two terminal groups: double bond and a carboxylic group. Samples were analyzed pre and post UV modification, and 8 days of storage in air.

Figure 51 shows C1s spectra of the hydrogen terminated silicon surface (a) that was exposed to undecylenic acid and UV light for 6 h (b) and then exposed to air for 8 days (c). H-Si spectrum shows insignificant carbon contamination. After modification

(b) three additional peaks appeared corresponding to Si-C, C-O, and C(O)OH. C-C bond is the main component as this type of bonding represents most of the atoms in undecylenic acid. C(O)OH corresponds to carboxylic group present in undecylenic acid.

Because no silicon oxidation was determined, we conclude that acid molecules reacted

152 using the double bond forming Si-C bond instead of the carboxylic bond and forming Si-

O bond.

C-C C-O C(O)OH Si-C

c Count

291 289 287 285 283 281 Binding energy, eV

Figure 51: C1s spectra of a sample (a) the H-terminated Si (100) with (b)

undecyletnic acid, and (c) after exposure to air for 8 days.

Figure 52 shows change in carbon to silicon ratio for samples with low and high initial carbon concentration, error bars reflect standard deviation over the three samples that had similar starting composition. UV exposure of samples that had low initial carbon concentration (C/Si=0.05±0.01), which is equivalent to 0.8±0.2 (standard deviation) monolayers of carbon on the surface, resulted in the increase of the carbon concentration as determined by the carbon to silicon ratio 0.3±0.0 (standard deviation).

Despite significant increase in the carbon concentration on the surface, the monolayer coverage for the samples with low starting carbon contamination, coverage was determine to be only 42±1% (standard deviation) of the full monolayer of the

153 undecylenic acid. Due to the fact that undecylenic acid is a rather bulky molecule, we would not expect it to attach to more that 50% of silicon atoms. Thus, less than 25% of all silicon atoms were modified. This indicates that even low carbon contamination may be an adverse factor in formation of the dense organic monolayers on the silicon surface.

Figure 53 shows illustration of the effect of the starting contamination on monolayer coverage. If the surface is not pristinely clean, then initiation may not occur or may occur, but the reactant will be isolated by contamination as illustrated in Figure 53.

The low UV light intensity may play a role in lower than expected monolayer coverage. The lamps used in these experiments had intensity of approximately 5 mW/cm2. From preliminary experiments, low UV light intensity was suspected, so the lamps were placed as close as possible to the quartz tube. In addition, longer exposure than used in the literature [77] was used in these experiments (i.e. 6 h) to ensure that the sample received enough initiation.

Excess of oxygen after modification was observed on the surfaces with low starting carbon contamination. C/O ratio for the undecylenic acid should be 5.5, but for all samples it was 2.0 – 2.3 indicating significant oxygen uptake. However, despite the fact that we did not obtain full monolayer and oxygen enrichment was observed, the monolayer coverage values were almost identical for the three samples (as shown by low standard deviation) indicating that samples that had low carbon contamination could be repeatedly modified with the same monolayer coverage.

154

0.60

0.50

0.40 low C 0.30

O/Si high C 0.20

0.10

0.00 pre post 8d stability

Figure 52: C/Si at 90° take-off angle

HHHHHH RHRHRH HHHHHH HHRHHH

Figure 53: Illustration of the effect of contamination on the undecylenic acid

monolayer coverage.

XPS spectra of surface with high initial carbon content are shown in Figure 54.

Carbon contamination on H-Si(100) was 0.29±0.07 (standard deviation), which was equivalent to 3.9±0.1 (standard deviation) monolayers of carbon. The C-C peak increased insignificantly during photomodification, as seen from spectra (a) and (b)

Figure 54. The peak corresponding to carboxylic acid was observed. Surfaces were modified with 20±14% (standard deviation) of the monolayer of the undecylenic acid.

Variation in the monolayer coverage was significantly greater for the samples with high carbon contamination, as seen in Figure 52. This indicates that surface coverage was uneven and possibly the acid molecules were not attached to the surface, but rather physisorbed. Lower surface coverage is most likely due to the fact that surfaces had

155 significant starting carbon concentration preventing a reaction between H-Si and the undecylenic acid molecules.

C-O C-C C(O)OH Si-C (c)

Count/s (b)

(a)

291 289 287 285 283 281 Binding Energy, eV

Figure 54: C1s peak for samples with high initial carbon concentration: (a) H-

Si(100), (b) after modification with undecylenic acid, and (c) after 8 days

of air exposure in the dark.

Stability study of the low carbon contamination samples showed that C/Si ratio did not change significantly; however, the highly contaminated samples showed greater instability. Reduction in carbon signal was observed which possibly due to carbon loss during exposure to air, as seen from Figure 52 and Figure 54. Surfaces gained oxygen during stability, as seen from Figure 55. High carbon contaminated samples exhibited higher O/Si ratio and greater SiO2 peaks, indicating that the monolayer did not provide protection for silicon from oxidation. Thus, this supports a hypothesis that monolayers on the sample which had high starting carbon concentration were not covalently bonded.

In addition, the carbon to oxygen ratio for the high carbon contaminated samples was

156 significantly lower than that for low carbon contamination sample after exposure to air for 8 days, 0.31 – 0.89 versus 1.96 – 2.27, respectively. This indicates that the samples with a high starting carbon contamination were more susceptible to oxidation than the samples with lower contamination.

0.60

0.50

0.40 low C 0.30

O/Si high C 0.20

0.10

0.00 pre post 8d stability

Figure 55: O/Si ratio at 90° take-off angle

ARXPS analysis of these samples was performed as well. Figure 56 shows C/Si and O/Si ratios for the spectra obtained at 30° take-off angle. Carbon and oxygen content are higher at 30° take-off angle, which is expected because ARXPS is more sensitive than

XPS analysis at the angle normal to the surface. The trends, however, are similar to those obtained at 90° take-off angle, as seen in Figure 52 and Figure 55. Lower variation at the

30° angle may be due to the fact that more of the surface area was analyzed than that at

90°, as depicted in Figure 6.

157

0.80 0.60 0.70 A 0.50 B 0.60 0.40 0.50 low C 0.40 0.30 C/Si O/Si high C 0.30 0.20 0.20 0.10 0.10 0.00 0.00 pre post 8d stability pre post 8d stability

Figure 56: XPS analysis at 30° take-off angle showed similar evolution of C/Si (A)

and O/Si (B) ratios as at 90°.

The carbon increase was calculated for low and high carbon concentrations. Table

22 shows that low carbon contaminated samples increased carbon concentration by approximately 6 fold, which supports observations from inspecting Figure 51, Figure 52,

Figure 54, and Figure 55. Difference between carbon increase for the low carbon samples at different take-off angles may be due to the fact that different surface area was analyzed at the two angles. Thus, the surface modification may be nonuniform. Low surface uniformity may be due to spotty contamination of the starting surface. For the samples with the high starting carbon contamination, we observed that this surface had consistent carbon gain for the two take-off angles used. Low carbon gain values obtained through ARXPS analysis confirm our hypothesis, that carbon contamination on the surface of the silicon blocks the sites for photomodification.

158

Table 22: Carbon gain for high and low carbon contamination samples due to

photomodification. Error = standard deviation.

Take-off angle High C Low C

90° 0.4±0.2 5.6±0.7

30° 0.4±0.4 7.7±0.7

To conclude, a study of the effect of the starting carbon contamination on the monolayer coverage via UV initiated radical reaction revealed that low carbon contamination resulted in a more consistent and complete undecylenic acid monolayer coverage and a greater stability during oxidation in air for 8 days. Our experiments show that initial level of contamination is indeed important to obtain a stable silicon surface modified with organic molecules. Thus, some of the previous results, for example [78],

[83], [2], should be used with caution.

4.5 Summary and Conclusions for Modification of Silicon Surface

Modification of the silicon surface with undecylenic acid using UV-light initiation was demonstrated. The formation of a Si-C covalent bond was shown by XPS analysis and stability studies. An algorithm for determining the thickness and coverage of the monolayer was determined based on the geometry of the molecules and the XPS data interpretation similar to Haber et al. in [2]. Despite multiple cleaning modifications and trials, a high purity clean and hydrogen terminated Si(100) surface could not be obtained consistently. Due to problems in producing the clean H-Si surfaces, the focus of the work

159 was redirected to the determination of the effect of the starting carbon contamination on the monolayer coverage with undecylenic acid. Effect of oxygen contamination on monolayer formation and stability was not studied at this time.

The effect of surface contamination on photomodification with undecylenic acid was measured by monolayer coverage and stability experiments using ARXPS. Surfaces that had low (less than 5 %) starting carbon contamination resulted in a formation of the covalently bonded monolayers of undecylenic acid with a monolayer coverage of 42±1 %

(standard deviation). A full monolayer of undecylenic acid coverage is defined as covalent termination of 50% of the silicon surface atoms. We hypothesize that a full monolayer was not obtained either due to some remaining carbon contamination present on the wafers or insufficient UV light intensity in the photoreactor. During exposure to air for 8 days, the samples with low (less than 5%) carbon contamination showed gain in oxygen which resulted in 2.0±0.7 % (standard deviation) silicon oxide layer formation and change of the carboxylic peak position to the lower binding energy as detected by

ARXPS.

Surfaces that had high (over 18%) starting carbon contamination formed disordered monolayers on the surface as determined by ARXPS with the monolayer coverage of 20±14 % (standard deviation). After a stability study, the silicon surface of these samples had 3.4±1.0 % (standard deviation) of oxide formed, in addition the peak corresponding to a carboxylic acid practically disappeared indicating that the disordered monolayer did not protect silicon surface as a covalently bonded monolayer should.

From stability studies we can conclude that surface contamination was a direct cause of the nonuniform low density monolayers.

160

4.6 Recommendations for Silicon Surface Modification

A study focused on determination of the effects of the starting carbon contamination on the silicon surface modification resulted in the conclusion that pristinely clean silicon surfaces are necessary to obtain high density (near 100% coverage) monolayers. Even a contamination level of 5% was shown to be a possible cause of low monolayer coverage of a silicon surface. Developing a technique, such as the Fenner Etch, for obtaining consistently clean silicon surfaces is recommended. In this work starting surfaces had minimal oxygen contamination; however its impact was not studies. Thus determination of the effect of oxygen on monolayer formation and stability is recommended.

Insufficient UV light intensity is another possible cause of low monolayer coverage, as moving the lamps closer to the substrate did not measurably impact monolayer coverage. Therefore, photoreactor efficiency should be determined by performing time studies. However, only sufficiently clean silicon samples should be used, as any inconsistency in starting carbon contamination level will mask differences in photomodification based on additional time in the photoreactor. If necessary, further modification to the photoreactor may be made, such as higher intensity lamps to increase the photon flux to the sample. However, caution should be taken if purchasing fluorescent lamps, as their power increases proportionally with length, so that intensity per area remains the same.

After photoreactor effectiveness is determined, surface modification should be extended beyond undecylenic acid; possibly attachment of the protected amines can be

161 tested. For these second-generation organic monolayers attenuated total reflectance

Fourier transform infrared spectroscopy (ATR-FTIR) should be performed to complement information obtained from the XPS analysis. ATR-FTIR will enable monitoring of the carbon – hydrogen bond vibrations, and will allow easier determination of some of the carbon containing groups which may be a small fraction of the C1s XPS, such as 1 carboxylic carbon for 10 aliphatic carbons in undecylenic acid.

162

5.0 Summary

The XPS analysis contributed to the three different projects. The niobium surface has been studied using XPS to determine the surface composition, oxide thickness, and atomic order. Functionalization of a silicon (100) surface in a custom-built photoreactor was achieved, and the impact of the starting surface contamination on monolayer formation was studied. Finally, the valence band region of silicon carbide and the silicon carbide/magnesium oxide interface was analyzed to determine the valence band offset at the interface of the magnesium oxide film on the silicon carbide substrate.

XPS was used to determine that niobium readily oxidizes, forming a number of different oxides. The sampling depth (7 nm) of the as-received wafer contained metal,

NbxO (1≤x≤2), NbO2, and Nb2O5. The oxide states were verified with the chemical shifts published in the literature. Due to the fact that the X-ray source used for the analysis was not monochromated, NbxO (1≤x≤2) could not be further resolved with great certainty.

However, the accuracy of assignment of the two main components – niobium metal and pentoxide – was high. Using areas of the metal and oxides peaks, the total oxide thickness was determined by relating the metal peak attenuation to the oxide overlayers.

Typically, the total oxide thickness of niobium of less than 5 nm was detected.

The atomic order of the niobium surface was detected by monitoring changes in the full width at half maximum (FWHM) of the metal peak in the Nb3d spectrum.

Treatments containing hydrofluoric acid led to improvement of the atomic order of the metal as seen by a decrease in the FWHM. Oxidation of the hydrofluoric acid treated surface resulted in the increase of the FWHM indicating that the atomic order of the

163 metal phase decreased. Thus, this finding supports the hypothesis that the bulk metal is serrated with the oxides which was proposed by Darlinski et al. in [21]. However, more investigations, possibly with ARXPS, are needed in order to understand if serration occurs or not. Serration may be considered as a defect on the niobium surface, thus its minimization would be beneficial for the SRF applications. XPS analysis of the CMP surfaces showed that different polishing conditions resulted in different surface composition and order; thus substantiating that XPS should be used in the future studies on optimization of the niobium CMP process.

Modification of the silicon surface using UV initiation was studied using angle resolved XPS in order to verify photomodification and characterize the obtained monolayers. The ARXPS analysis revealed four carbon bonding environments after photomodification with undecylenic acid: C-C, C-Si, C-O, and C(O)O. Formation of Si-

C bonds and presence of C(O)O indicated successful attachment of undecylenic acid.

A study aimed to determine effect of starting carbon contamination on the silicon surface modification was possible due to the high surface sensitivity of XPS and the ability of ARXPS to further differentiate the top layer from the near surface layers. XPS analysis enabled determination of the monolayer coverage by relating silicon signal attenuation to the carbon content. Using both XPS and ARXPS in a stability study helped to determine that surfaces with high carbon contamination resulted in low monolayer coverage and reduced stability. The blockage of the reactive sites on the silicon surface may have resulted in a physisorption of the undecylenic acid molecules.

Thus, with the added surface sensitivity of ARXPS, XPS can effectively monitor organic monolayer coverages under different experimental conditions.

164

Finally, preliminary experiments showed that the valence band offset calculations using XPS valence band analysis was possible using the equipment in the Interface

Engineering Laboratory. The valence band offset for magnesium oxide grown using molecular beam epitaxy on silicon carbide, was found to be 1.13±0.12 eV. It was shown in the literature that calibration of the instrument is critical for the accurate determination of the core and valence band binding energies [23]. Thus, a great effort was directed on calibration of the XPS and understanding of the accuracy of the valence band measurements. However, more work needs to be done on calibrating of the binding energy scale to be able to use XPS not only for the offset analysis, but for determination of the absolute values for the valence band maximum position.

In conclusion, this thesis has demonstrated the capability and limits of XPS in distinguishing oxide bonding states, has shown the utility of the surface sensitivity with

ARXPS to determine organic monolayer coverage on an inorganic substrate, and has demonstrated the potential of XPS in measuring electronic band bending at a film/substrate interface. Each of the individual projects has unique conclusions and recommendations relating to the project goals. Various aspects of the XPS technique have proved valuable for achieving those project goals.

165

6.0 Nomenclature

CA atomic percent of element A (%)

D(EX) the detector efficiency at energy EX

E the electron energy (eV)

EB binding energy (eV)

Eg the band-gap energy for non-conductors (eV)

EK kinetic energy (eV)

Ep the free-electron Plasmon energy (eV)

EX kinetic energy of photoelectrons from the core level X (eV)

E0 center of the peak

Y ECL binding energy of the core level of the bulk substrate Y (eV)

Y Ev binding energy of the valence band of the bulk substrate Y (eV)

Yfilm ECL binding energy of the core level in material Y after thin film X was deposited (eV)

F(E) intensity at energy E

FWHM full width at half maximum (eV)

G(EX) the spectrometer étendue at kinetic energy EX hν excitation energy (Mg Kα = 1486.6 eV and Al Kα = 1253.6 eV)

H peak height

I0 intensity from the surface atoms (nA)

0 I sub intensity of the pure substrate

Iz intensity from the atoms at depth z (nA)

J X-ray flux

166

K constant X-ray flux of the instrument

2 KP Preston’s coefficient (m /N)

L((hν,X) the angular asymmetry factor for emission from the core level X by photon hν (cm-2s-1) m a mixing ratio (1 for pure Gaussian and 0 for pure Lorentzian)

M the atomic or molecular weight (kg/mol)

NA distribution of atoms A

Nv the number of valence electrons per atom or molecule p surface roughness

P local pressure on the wafer surface (N/m2)

RR removal rate (m/s)

SA sensitivity factor for A

V relative velocity of the point on the surface of the wafer versus the pad (m/s)

X the core level of element A z depth (nm)

Greek

θ electron take-off angle

λ inelastic mean free path, IMFP, (nm)

ρ the bulk density (kg/m3)

m 3 ρ sub molar density of the substrate (mol/cm )

σ(hν,EX) the photoionization cross-section for ionization from the core level X by photon hν

φ work function is defined as energy required to overcome surface attraction and travel from sample surface to the analyzer (eV)

167

Superscripts m associates principle symbol with the molar property

X associates principle symbol with overlayer material in VB analysis

Y associates principle symbol with substrate in VB analysis

Yfilm associates principle symbol with the substrate after thin film deposition in VB analysis

0 associates principle symbol with pure substance

Subscripts

A designates element

CL designates core level ov designates overlayer sub designates substrate v designates valence band level

X designates core level from which electron originated z designates thickness from which electrons originate

0 designates center of the peak

168

7.0 References

1. Walt, D.R. and D.R. Franz, Biological warfare detection. Analytical Chemistry, 2000. 72(23): p. 738A-746A. 2. Haber, J.A. and N.S. Lewis, Infrared and X-ray Photoelectron Spectroscopic Studies of the Reactions of Hydrogen-Terminated Crystalline Si(111) and Si(100) Surfaces with Br2, I2, and Ferrocenium in Alcohol Solvents. Journal of Physical Chemistry B, 2002. 106(14): p. 3639-3656. 3. K. Kowalski, A.B., W. Singer, X. Singer, J Camra. In Situ XPS Investigation of the Baking Effect on the Surface Oxide Structure Formed on Niobium Sheets Used in Superconducting RF Cavity Production. in 11th Workshop on RF- Superconductivity 2003. Lubeck, Germany. 4. Briggs, D., Surface Analysis of Polymers by XPS and Static SIMS. 1998, Cambridge, UK: Cambridge University Press. 198. 5. Watts, J.F., An Introduction to Surface Analysis by Electron Spectroscopy. 1990, New York: Oxford University Press. 85. 6. Briggs, G.B.a.D., High resoltuion XPS of organic polymers: the Scienta ESCA300 database. 1992, New York: John Wiley & Sons Ltd. 295. 7. Kraut, E.A., et al., Precise determination of the valence-band edge in x-ray photoemission spectra: application to measurement of semiconductor interface potentials. Physical Review Letters, 1980. 44(24): p. 1620-3. 8. Wagner, C.D., et al., Handbook of X-ray Photoelectron Spectroscopy. 1979: Perkin-Elmer Corporation, Physical Electronics Devision. 190. 9. Powell, C.J., et al., New developments in data for Auger electron spectroscopy and X-ray photoelectron spectroscopy. Journal of Surface Analysis, 2005. 12(2): p. 88-96. 10. Tanuma, S., C.J. Powell, and D.R. Penn, Calculations of electron inelastic mean free paths. V. Data for 14 organic compounds over the 50-2000 eV range. Surface and Interface Analysis, 1994. 21(3): p. 165-76. 11. Tanuma, S., C.J. Powell, and D.R. Penn, Calculations of electron inelastic mean free paths. VIII. Data for 15 elemental solids over the 50-2000 eV range. Surface and Interface Analysis, 2005. 37(1): p. 1-14. 12. Tanuma, S., C.J. Powell, and D.R. Penn, Calculations of electron inelastic mean free paths. II. Data for 27 elements over the 50-2000 eV range. Surface and Interface Analysis, 1991. 17(13): p. 911-26. 13. Technician's Model 10-360 Precision Energy analyzer Component Manual. 2000, Physical Electronics. 14. Auger and X-ray Photoelectron Spectroscopy. 2 ed. Practical Survace Analysis, ed. D. Briggs, and M.P. Seah. Vol. 1. 1990, New York: John Willey and Sons. 657. 15. AugerScan in In-House Release. 2005, RBD Enterprises, Inc. 16. Grundner, M. and J. Halbritter, XPS and AES studies on oxide growth and oxide coatings on niobium. Journal of Applied Physics, 1980. 51(1): p. 397-405.

169

17. Dacca, A., et al., XPS analysis of the surface composition of niobium for superconducting RF cavities. Applied Surface Science, 1998. 126(3/4): p. 219- 230. 18. Ma, Q. and R.A. Rosenberg, Angle-resolved X-ray photoelectron spectroscopy study of the oxides on Nb surfaces for superconducting r.f. cavity applications. Applied Surface Science, 2003. 206(1-4): p. 209-217. 19. Tian, H., et al., Surface studies of niobium chemically polished under conditions for superconducting radio frequency (SRF) cavity production. Applied Surface Science, 2006. 253(3): p. 1236-1242. 20. Ma, Q., et al., Thermal effect on the oxides on Nb(100) studied by synchrotron- radiation x-ray photoelectron spectroscopy. Journal of Applied Physics, 2004. 96(12): p. 7675-7680. 21. Darlinski, A. and J. Halbritter, Angle-resolved XPS studies on oxides at niobium mononitride, niobium monocarbide, and niobium surfaces. Surface and Interface Analysis, 1987. 10(5): p. 223-37. 22. X-ray Photoelectron Spectroscopy - XPS. . [cited 2006 March 30, 2006]; Available from: http://www.kjemi.uio.no/kj414/xps/pensum.pdf. 23. Kraut, E.A., et al., Semiconductor core-level to valence-band maximum binding- energy differences: precise determination by x-ray photoelectron spectroscopy. Physical Review B: Condensed Matter and Materials Physics, 1983. 28(4): p. 1965-77. 24. Lu, Y., et al., Measurement of the valence-band offset at the epitaxial MgO- GaAs(001) heterojunction by x-ray photoelectron spectroscopy. Applied Physics Letters, 2006. 88(4): p. 042108/1-042108/3. 25. Seah, M.P., Summary of ISO/TC 201 Standard: VII ISO 15472: 2001 - surface chemical analysis - x-ray photoelectron spectrometers - calibration of energy scales. Surface and Interface Analysis, 2001. 31: p. 721-723. 26. Kneisel, P., Cavity preparation/assembly techniques and impact on Q, realistic Q- factors in a module, review of modules. Nuclear Instruments & Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors, and Associated Equipment, 2006. 557(1): p. 250-258. 27. Kneisel, P., High gradient superconducting niobium cavities. A review of the present status. Applied Superconductivity, IEEE Transactions on, 1999. 9(2): p. 1023-1029. 28. Antoine, C., et al., Evidence of preferential diffusion of impurities along grain boundaries in very pure niobium used for radio-frequency cavities. Journal of Applied Physics, 1997. 81(4): p. 1677-1682. 29. Bonin, B. Materials for superconducting cavities. 1996 28 April 1999 [cited 2008 16 March]; Available from: http://preprints.cern.ch/cgi- bin/setlink?base=cernrep&categ=Yellow_Report&id=96-03. 30. How does the accelerator work? [cited 2008 1 February]; Available from: http://education.jlab.org/sitetour/guidedtour05.l.alt.html. 31. Saito, K., Development of electropolishing technology for superconducting cavities. Proceedings of the Particle Accelerator Conference, 20th, Portland, OR, United States, May 12-16, 2003, 2003: p. 462-466.

170

32. Lilje, L., et al., Improved surface treatment of the superconducting TESLA cavities. Nuclear Instruments & Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors, and Associated Equipment, 2004. 516(2- 3): p. 213-227. 33. Oliver, M.R., CMP Technology, in Chemical-Mechanical Planarization of Semiconductor Materials, M.R. Oliver, Editor. 2004, Springer: New York. p. 7- 40. 34. Evans, D.R., Metal Polishing Process in Chemical-Mechanical Planarization of Semiconductor Materials, M.R. Oliver, Editor. 2004, Springer: New York. p. 42- 83. 35. Calota, G. and S. Muftu, Material removal from a rough silicon wafer using CMP. 2006, Northeastern University. 36. Kneisel, P. Surface Characterization: What Has Been Done, What Has Been Learnt? in 11th Workshop on RF-Superconductivity 2003. Lubeck, Germany. 37. Valente, A.-M., Daly, E., Delayen, J., Drury, M., Hicks, R., Hovater, C., Mammosser, J., Phillips, L., Powers, T., Preble, J., Reece, C.E., Rimmer, R.A., and H. Wang. Production and Performance of the CEBAF Upgrade Cryomodule Intermediate Prototypes. in European Particle Acceleration Conference 2004. Lucerne, Switzeland. 38. Wang, T., C.E. Reece, and R.M. Sundelin, Enhanced field emission from chemically etched and electropolished broad-area niobium. Journal of Vacuum Science & Technology, B: Microelectronics and Nanometer Structures-- Processing, Measurement, and Phenomena, 2003. 21(4): p. 1230-1239. 39. Septier, A. Surfaces studies and electron emissions. in Workshop on RF superconductivity. 1980. Kernforschungszent, Karlsruhe: Conservatoire Natl. Ants Metiers,Paris,Fr. 40. Hulm, J.K., et al., Superconductivity in the titanium oxide and niobium oxide systems. Journal of Low Temperature Physics, 1972. 7(3-4): p. 291-307. 41. Halbritter, J. Material Science of Nb RF Accelerator Cavities: Where Do We Stand 2001? in 10th Workshop on RF Superconductivity. 2001. Tsukuba, Japan. 42. Antoine, C.Z., et al. Alternative Approaches for Surface Treatment of Nb Superconducting Cavities. in 9th Workshop On RF Superconductivity. 1999. Santa Fe, New Mexico, USA: Los Alamos National Laboratory. 43. Casalbuoni, S., et al., Surface superconductivity in niobium for superconducting RF cavities. Nuclear Instruments & Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors, and Associated Equipment, 2005. 538(1- 3): p. 45-64. 44. Jepson, P., Personal Communication. 2007. 45. Schwarz, W. and J. Halbritter, On oxygen enrichments in Nb surface layers and their apparent conductivity as observed by the superconducting penetration depth Delta lambda (T,f,B[sub ac]). Journal of Applied Physics, 1977. 48(11): p. 4618- 4626. 46. Ma, Q., Rosenberg, Richard A. . Thermal and Electron-Beam Irradiation Effects on the Surfaces of Niobium for RF Cavity Production. in 10th Workshop on RF Superconductivity. 2001. Tsukuba, Japan.

171

47. Halbritter, J., ARXPS analysis and oxidation of niobium compounds. Electrochimica Acta, 1989. 34(8): p. 1153-55. 48. Tian, H., Anne Marie Valente, Michael J. Kelley, Cormac McGuinness, Per Anders Glans, Kevin E. Smith. Near-Surface Composition of Electropolished Niobium by Variable Photon Energy XPS. in 11th Workshop on RF- Superconductivity 2003. Lubeck, Germany. 49. Halbritter, J., On the oxidation and on the superconductivity of niobium. Applied Physics A: Materials Science & Processing, 1987. 43(1): p. 1-28. 50. Robinson, K., Fundamentals of CMP Slurry, in Chemical-Mechanical Planarization of Semiconductor Materials, M.R. Oliver, Editor. 2004, Springer: New York. p. 215-249. 51. Material Safery Data Sheet iCue®5001 2003, Cabot Microelectronics. 52. Golden, J.H., et al., Evaluating and treating CMP wastewater. Semiconductor International, 2000. 23(12): p. 85-86,88,90,92,94,98. 53. Hosali, S.D., et al., Composition and method for polishing a composite of silica and silicon nitride. 2000, (Rodel Holdings, Inc., USA). Application: US. p. 4 pp , Cont -in-part of U S 6,042,741. 54. Avanzino, S.C., et al., Chemical-mechanical polishing slurry formulation and method for tungsten and titanium thin films. 1998, (Advanced Micro Devices, Inc., USA). Application: WO. p. 31 pp. 55. Yang, K., S. Avanzino, and C.M.-C. Woo, Slurry for chemical mechanical polishing of copper. 2000, (Advanced Micro Devices, Inc., USA). Application: US. p. 7 pp. 56. Kaufman, V.B., R.C. Kistler, and S. Wang, Chemical mechanical polishing slurry useful for copper substrates. 2001, (Cabot Microelectronics Corp., USA). Application: US. p. 13 pp , Cont -in-part of U S 6,126,853. 57. Sisco, F.R. and E. Epremian, eds. Columbium and Tantalum. 1963, John Wiley & Sons, Inc.: New York. 635. 58. F. Fairbrother, D.S., F.R.I.C., The Chemistry of Niobium and Tantalum. 1967, New York: Elsevier Publishing Company. 59. J.R. De Laeuter, J.K.B., P. De Bievre, H. Hidaka, H.S. Peiser, K.J.R. Rosman, and P.D.P. Taylor, Atomic Weights of the Elements Review 2000. Pure Applied Chemistry, 2003. 75(6): p. 683-800. 60. Niobium Metallurgical Products. 2008, Cabot Supermetals 61. Asselin, E., T.M. Ahmed, and A. Alfantazi, Corrosion of niobium in sulfuric and hydrochloric acid solutions at 75 and 95 DegC. Corrosion Science, 2007. 49(2): p. 694-710. 62. Oliver, M.R., ed. Chemical-Mechanical Planarization of Semiconductor Materials. 1st ed. Materials Science, ed. R.M.O. R. Hull, Jr., J. Parisi, H. Warlimont. 2004, Springer: New York. 425. 63. PACE Technologies: Data Sheet. 2007: Tuscon, AZ. 64. Dacca, A., et al., Thermal evolution of anodized Nb by XPS. Surface Science Spectra, 1998. 5(4): p. 332-338. 65. Samuel L. Groseclose, D.V.M., M.P.H., et al. Summary of Notifiable Diseases - United States, 2001 2003 [cited 2006 April 7]; Available from: http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5053a1.htm.

172

66. Bravata D.M., S.V., McDonald K.M., Smith W.M., Szeto H., Schleinitz M.D., et al., Detection and diagnostic decision support systems for bioterrorism response. . Emerg Infect Dis 2004. 10(1). 67. Rider, T.H., Petrovick, Martha S., Nargi, Frances E., Harper, James D., Schwoebel, Eric D., Mathews, Richard H. , Blanchard, David J., Bortolin, Laura T., Young, Albert M., Chen, Jianzhu , and Mark A. Hollis, A B Cell-Based Sensor for Rapid Identification of Pathogens. Science, 2003. 301: p. 213-215. 68. Reuters, Bio-Sensor Quickly Detects Anthrax, Smallpox and Other Pathogens, in ScienceDaily. 2008. 69. Buriak, J.M., Organometallic chemistry on silicon surfaces: formation of functional monolayers bound through Si-C bonds. Chemical Communications (Cambridge), 1999(12): p. 1051-1060. 70. Sieval, A.B., et al., High-quality alkyl monolayers on silicon surfaces. Advanced Materials (Weinheim, Germany), 2000. 12(19): p. 1457-1460. 71. Buriak, J.M., Organometallic Chemistry on Silicon and Germanium Surfaces. Chemical Reviews (Washington, D. C.), 2002. 102(5): p. 1271-1308. 72. Linford, M.R., et al., Alkyl Monolayers on Silicon Prepared from 1-Alkenes and Hydrogen-Terminated Silicon. Journal of the American Chemical Society, 1995. 117(11): p. 3145-55. 73. Linford, M.R. and C.E.D. Chidsey, Alkyl monolayers covalently bonded to silicon surfaces. Journal of the American Chemical Society, 1993. 115(26): p. 12631-2. 74. Saito, N., et al., Exploration of the chemical bonding forms of alkoxy-type organic monolayers directly attached to silicon. Journal of Vacuum Science & Technology, A: Vacuum, Surfaces, and Films, 2004. 22(4): p. 1425-1427. 75. Voicu, R., et al., Formation, Characterization, and Chemistry of Undecanoic Acid-Terminated Silicon Surfaces: Patterning and Immobilization of DNA. Langmuir, 2004. 20(26): p. 11713-11720. 76. Liu, Y.-J., N.M. Navasero, and H.-Z. Yu, Structure and Reactivity of Mixed w- Carboxyalkyl/ Alkyl Monolayers on Silicon: ATR-FTIR Spectroscopy and Contact Angle Titration. Langmuir, 2004. 20(10): p. 4039-4050. 77. Asanuma, H., G.P. Lopinski, and H.Z. Yu, Kinetic Control of the Photochemical Reactivity of Hydrogen-Terminated Silicon with Bifunctional Molecules. Langmuir, 2005. 21(11): p. 5013-5018. 78. Boukherroub, R., et al., New Synthetic Routes to Alkyl Monolayers on the Si(111) Surface. Langmuir, 1999. 15(11): p. 3831-3835. 79. Strother, T., et al., Synthesis and Characterization of DNA-Modified Silicon (111) Surfaces. Journal of the American Chemical Society, 2000. 122(6): p. 1205-1209. 80. Cicero, R.L., M.R. Linford, and C.E.D. Chidsey, Photoreactivity of Unsaturated Compounds with Hydrogen-Terminated Silicon(111). Langmuir, 2000. 16(13): p. 5688-5695. 81. Terry, J., et al., Determination of the bonding of alkyl monolayers to the Si(111) surface using chemical-shift, scanned-energy photoelectron diffraction. Applied Physics Letters, 1997. 71(8): p. 1056-1058. 82. Terry, J., et al., Alkyl-terminated Si(111) surfaces: A high-resolution, core level photoelectron spectroscopy study. Journal of Applied Physics, 1999. 85(1): p. 213-221.

173

83. Webb, L.J. and N.S. Lewis, Comparison of the Electrical Properties and Chemical Stability of Crystalline Silicon(111) Surfaces Alkylated Using Grignard Reagents or Olefins with Lewis Acid Catalysts. Journal of Physical Chemistry B, 2003. 107(23): p. 5404-5412. 84. Fellah, S., et al., Hidden Electrochemistry in the Thermal Grafting of Silicon Surfaces from Grignard Reagents. Langmuir, 2004. 20(15): p. 6359-6364. 85. Buriak, J.M. and M.J. Allen, Lewis Acid Mediated Functionalization of Porous Silicon with Substituted Alkenes and Alkynes. Journal of the American Chemical Society, 1998. 120(6): p. 1339-1340. 86. Lua, Y.Y., et al., First Reaction of a Bare Silicon Surface with Acid Chlorides and a One-Step Preparation of Acid Chloride Terminated Monolayers on Scribed Silicon. Langmuir, 2005. 21(6): p. 2093-2097. 87. Allongue, P., et al., Phenyl layers on H-Si(111) by electrochemical reduction of diazonium salts: monolayer versus multilayer formation. Journal of Electroanalytical Chemistry, 2003. 550-551: p. 161-174. 88. Sieval, A.B., et al., Highly Stable Si-C Linked Functionalized Monolayers on the Silicon(100) Surface. Langmuir, 1998. 14(7): p. 1759-1768. 89. Sieval, A.B., et al., Amino-Terminated Organic Monolayers on Hydrogen- Terminated Silicon Surfaces. Langmuir, 2001. 17(24): p. 7554-7559. 90. Sieval, A.B., et al., Monolayers of 1-alkynes on the H-terminated Si(100) surface. Langmuir, 2000. 16(26): p. 10359-10368. 91. De Smet, L.C.P.M., et al., Covalently attached saccharides on silicon surfaces. Journal of the American Chemical Society, 2003. 125(46): p. 13916-13917. 92. Sieval, A.B., et al., An Improved Method for the Preparation of Organic Monolayers of 1-Alkenes on Hydrogen-Terminated Silicon Surfaces. Langmuir, 1999. 15(23): p. 8288-8291. 93. Boukherroub, R. and D.D.M. Wayner, Controlled Functionalization and Multistep Chemical Manipulation of Covalently Modified Si(111) Surfaces. Journal of the American Chemical Society, 1999. 121(49): p. 11513-11515. 94. Kanabus-Kaminska, J.M., et al., Reduction of silicon-hydrogen bond strengths. Journal of the American Chemical Society, 1987. 109(17): p. 5267-8. 95. Laarhoven, L.J.J., P. Mulder, and D.D.M. Wayner, Determination of Bond Dissociation Enthalpies in Solution by Photoacoustic Calorimetry. Accounts of Chemical Research, 1999. 32(4): p. 342-349. 96. Pai, S. and D. Doren, Substituent effects in silicon hydrides: implications for models of surface sites. Journal of Physical Chemistry, 1994. 98(16): p. 4422-7. 97. Strother, T., R.J. Hamers, and L.M. Smith, Covalent attachment of oligodeoxyribonucleotides to amine-modified Si (001) surfaces. Nucl. Acids Res., 2000. 28(18): p. 3535-3541. 98. Stewart, M.P. and J.M. Buriak, Exciton-Mediated Hydrosilylation on Photoluminescent Nanocrystalline Silicon. Journal of the American Chemical Society, 2001. 123(32): p. 7821-7830. 99. Schmeltzer, J.M., et al., Hydride Abstraction Initiated Hydrosilylation of Terminal Alkenes and Alkynes on Porous Silicon. Langmuir, 2002. 18(8): p. 2971-2974.

174

100. Sun, Q.Y., et al., Covalently Attached Monolayers on Hydrogen-Terminated Si(100): Extremely Mild Attachment by Visible Light. Angewandte Chemie International Edition, 2004. 43(11): p. 1352-1355. 101. Sun, Q.Y., et al., Covalently Attached Monolayers on Crystalline Hydrogen- Terminated Silicon: Extremely Mild Attachment by Visible Light. J. Am. Chem. Soc., 2005. 127(8): p. 2514-2523. 102. Grunthaner, P.J., et al., The localization and crystallographic dependence of silicon suboxide species at the silica/silicon interface. Journal of Applied Physics, 1987. 61(2): p. 629-38. 103. Webb, L.J., et al., High-Resolution X-ray Photoelectron Spectroscopic Studies of Alkylated Silicon(111) Surfaces. Journal of Physical Chemistry B, 2005. 109(9): p. 3930-3937. 104. Wallart, X., C.H. de Villeneuve, and P. Allongue, Truly Quantitative XPS Characterization of Organic Monolayers on Silicon: Study of Alkyl and Alkoxy Monolayers on H-Si(111). Journal of the American Chemical Society, 2005. 127(21): p. 7871-7878. 105. Kern, W. and D.A. Puotinen, Cleaning solution based on hydrogen peroxide for use in silicon semiconductor technology. RCA Rev., 1970. 31(2): p. 187-206. 106. Fenner, D.B., D.K. Biegelsen, and R.D. Bringans, Silicon surface passivation by hydrogen termination: a comparative study of preparation methods. Journal of Applied Physics, 1989. 66(1): p. 419-24. 107. Tufts, B.J., et al., X-ray photoelectron spectroscopic studies of interfacial chemistry at n-type silicon/liquid junctions. Journal of Physical Chemistry, 1992. 96(11): p. 4581-92. 108. Handbook of Semiconductor Wafer Cleaning Technology: Science, Technology, and Applications, ed. W. Kern. 1993, Park Ridge, New Jersey, U.S.A.: Noyes Publications. 623. 109. Burkman, D.C., Deal, D., Grant, D.C., Peterson, C.A., Aqueous Cleaning Processes, in Handbook of Semiconductor Wafer Cleaning Technology: Science, Technology, and Applications, W. Kern, Editor. 1993, Noyes Publications: Park Ridge, New Jersey, U.S.A. p. 111-151. 110. Higashi, G.S., Chabal, Yves J. , Silicon Surface Chemical Composition and Morphology, in Handbook of Semiconductor Wafer Cleaning Technology: Science, Technology, and Applications, W. Kern, Editor. 1993, Noyes Publications: Park Ridge, New Jersey, U.S.A. p. 433-496. 111. Sherwood, P.M.A., Valence Band Studied by XPS, in Surface Analysis by Auger and X-ray Photoelectron Spectroscopy, D. Briggs and J.T. Grant, Editors. 2003, IM Publications. p. 531-555. 112. Chen, J.J., et al., Determination of MgO/GaN heterojunction band offsets by x-ray photoelectron spectroscopy. Applied Physics Letters, 2006. 88(4): p. 042113/1- 042113/3. 113. Chambers, S.A., et al., Accurate valence band maximum determination for SrTiO3(001). Surface Science, 2004. 554(2-3): p. 81-89. 114. Liu, C., et al., Band offset measurements of the pulsed-laser-deposition-grown Sc2O3 (111)/GaN (0001) heterostructure by X-ray photoelectron spectroscopy.

175

Physica Status Solidi C: Current Topics in Solid State Physics, 2007. 4(7): p. 2330-2333. 115. Mi, Y.Y., et al., Epitaxial LaAlO3 thin film on silicon: Structure and electronic properties. Applied Physics Letters, 2007. 90(18): p. 181925/1-181925/3. 116. Wang, S.J., et al., Energy-band alignments at ZrO2/Si, SiGe, and Ge interfaces. Applied Physics Letters, 2004. 85(19): p. 4418-4420. 117. Gassenbauer, Y. and A. Klein, Electronic and Chemical Properties of Tin-Doped Indium Oxide (ITO) Surfaces and ITO/ZnPc Interfaces Studied In-situ by Photoelectron Spectroscopy. Journal of Physical Chemistry B, 2006. 110(10): p. 4793-4801. 118. Konovalov, I., et al., Intermixing, band alignment and charge transport in AgIn5S8/CuI heterojunctions. Thin Solid Films, 2005. 493(1-2): p. 282-287. 119. Spectral Distribution of Irradiance Density for UV Lamps. 2002 [cited 2008 14 March]; Available from: http://www.rayonet.org/spectral-graphs.htm.

176

8.0 APPENDICES

177

Appendix A

Valence Band Offset Determination

178

Introduction

A valence band (VB) region has been analyzed using XPS since the early 1970s with the purpose of comparing the valence band spectra with the density of states calculations [111]. The VB spectra arise from the photoelectron emission from the molecular orbitals involved in a chemical bonding [6] between -5 and 30 eV. The VB spectra obtained with XPS reflect the initial state density of states, unlike ultraviolet photoelectrons spectroscopy where valence spectra depend on initial and final density of states [111]. Thus XPS VB spectra are easier to interpret. Because VB electrons are directly involved in formation of chemical bonds, they should have more information on bonding states in the sampling volume. Aliphatic hydrocarbon polymers such as poly(ethylene) and poly(1-butene) have identical C1s spectra; however they can be distinguished by their valence band spectra [6].

Position of some elements in the VB region have been determined as shown in

Table A-1; however, without calculating the valence band spectra based on the density of states more detailed peak assignment is not reliable [111]. Complexity of the VB spectra is due to the fact that all valence electrons in the material show up in the same region

(~30 eV) unlike core electrons that are spread over a much larger range. For polymers, the VB spectra are 20-100 times weaker than the core spectra [6], so some researchers recommend obtaining over 500 spectra to gain acceptable signal-to-noise ratio [23].

However, over longer analysis times, radiation damage to some samples may occur.

179

Table A-1: Binding energy values for selected elements in the valence band region

Peak Br 4p Cl 3p Cl 3s N 2s O 2s Na 2p F 2s

Binding Energy, eV ~5 3-6 14-15 21-25 26-30 ~31 32-36

In addition to confirmation of the density of states calculations, VB analysis has been used to monitor surface potential and to determine band bending, Schottky-barrier height, and heterojunction band discontinuities [7]. Kraut et al. in [7] developed a method for determination of the valence band offset which occurs to allow the flow of electrons between two materials.

To determine the valence and conduction band offsets using Kraut et al. method, binding energy values for the core and valence levels must be known with certainty of

±0.025 eV [23]. In order to calculate VB offset using this method, one needs to know the difference between the core and valence binding energy values for both substrates and a thin film, according to Equation A-1.

Y Y X X Xfilm Yfilm ∆Ev = (ECL − Ev ) − (ECL − Ev ) − (ECL − ECL ) A-1

Y Where: ECL = binding energy of the core level of the bulk substrate Y (eV)

Y Ev = binding energy of the valence band of the bulk substrate Y (eV)

X ECL = binding energy of the core level of the bulk material X (eV)

X Ev =binding energy of the valence band of the bulk material X (eV)

Yfilm ECL = binding energy of the core level in material Y after thin film X was

deposited (eV)

180

Xfilm ECL = binding energy of the core level in material X after thin film of X

was deposited (eV)

Determination of the valence band maximum (VBM) has been done using different methods. Linear extrapolation method to determines the VBM by finding a projection on the binding energy scale of the straight lines passing through the baseline and the VB peak, as shown in Figure A-1. While a number of researchers used this method to determine VBM [112], [24], Kraut et al. stated in their work that this method introduced “major uncertainty” in determination of the VMB and VB offset for semiconductors [23].

Figure A-1: Valence band maximum for GaAs was determined to be 0 eV [24].

Kraut et al. fitted instrumentally broadened theoretical valence-band density of states (VB DOS) into the experimental data using the method of least squares [7, 23].

This method was found to be accurate in determining VBM in semiconductors, but not in oxides such as n-SrTiO3(001), as seen in Figure A-2 [113]. Figure A-2 shows various

DOS curves and the VBM of 3.7 eV. However, the linear extrapolation method resulted in VBM values of 3.27 eV. Chambers et al. showed that by changing the DOS function they were able to achieve agreement with experimental data; however they concluded

181 that, in general, linear extrapolation method was more accurate than the VB DOS method for the band offsets of oxide interfaces [113].

Figure A-2: Comparison of the various DOS methods and linear extrapolation.

Experimental VB XPS spectrum of SrTiO3(001) (circles), generalized

gradient corrections DOS (dotted curve), and DOS after broadenting with

Gaussians of FWHM=0.46 eV (dashed curve) and FWHM=1.00 eV (solid

curve). Insert is VBM determined by the linear extrapolation method.

Various oxides are grown on silicon carbide substrate in Interface laboratory at

Northeastern University, thus interest in VB analysis and VB offset determination is not a surprise. Magnesium oxide is used as an interlayer to alleviate the lattice mismatch between the substrate, silicon carbide, and the functional film. Two studies by Chen et al. in [112] and Lu et al. in [24] have been published on determination of the VB offset of

MgO on gallium nitride and gallium arsenide, respectively.

Chen et al. grew magnesium oxide on GaN/sapphire templates using molecular beam epitaxy (MBE) at 100° C [112]. The samples were analyzed with nonmonochromated aluminum anode. The VB offset was determined to be 1.06±0.15 eV

182 using Ga3d peak and Mg2p for the core level binding energy. VBM was determined using a linear extrapolation method. The authors indicated that the instrument was calibrated using polycrystalline gold foil to 84.00±0.02 and 0.00±0.02 for Au 4f7/2 and

Fermi-edge inflection point, respectively, and that peak positions were corrected to C1s peak. Chen et al. stated that their energy scale had an absolute scale error of 0.02-0.03 eV over the binding energy range of 0 to 100 eV; however C1s peak that they references their values was at 284.5 eV [112]. When the instrument in Interface laboratory was calibrated using Au 4f7/2 and Fermi-edge to the values used by this group, C1s peak shifted to ~288 eV. This indicates values obtained by this group may have undetermined error associated with them.

Lu et al. conducting a study to determine VB offset of the MgO grown using

MBE at 580° C on n-doped GaAs [24]. Magnesium anode was used for the sample analysis. Similar to Chen et al. in [112] the binding energy scale was calibrated using gold foil to 84.00±0.05 and 0.00±0.05 for Au 4f7/2 and Fermi-edge inflection point, respectively. The VB offset was calculated to be 4.2±0.1 eV using Ga 3d5/2 and O 2s as core levels [24]. VBM for GaAs was determined using linear extrapolation method.

Theoretical valence band density of states was used for determination of the VBM of

MgO. The use of oxygen for analysis is somewhat unusual because oxygen is a common contaminant is characterized by a broad peak due to presence of many bonding states.

183

Experiment

Calibration of the binding energy scale was performed utilizing procedure described in section 2.6. The binding energy scale was found to be accurate within ±0.03 eV. Silicon carbide sample was cleaned in a custom-built hydrogen furnace. This sample was analyzed to determine surface composition and used as a substrate for MgO deposition. A thin (~18 Ǻ) MgO was deposited on the substrate which temperature was held constant at 150° C. Magnesium metal from the effusion cell and oxygen from the oxygen plasma (100 W) were used as precursors for MgO film. The chamber pressure of

5·10-6 Torr was maintained. After a thin oxide growth, the sample was transferred to the analysis chamber without being exposed to air. The thin MgO layer was analyzed for

Mg, O, C, Si, and valence band region. After analysis, the sample was transferred into the growth chamber for the growth of the thick oxide film. The final oxide thickness was determined to be ~108 Ǻ. This sample was transferred and analyzed in the analysis chamber for Mg, O, C, Si, and valence band region.

Results

Preliminary study of the VB offset of MgO grown by MBE on SiC was conducted. XPS analysis on the samples was performed after SiC cleaning, after thin (~

18 Ǻ) of MgO, and after thick (~ 108 Ǻ) MgO. It was determined that SiC surface after cleaning in hydrogen furnace had oxygen contamination of 15.9±0.8 %, indicating that

SiC surface had an oxide layer on the surface. The core level of Si 2p and Mg 2p, and the

VBM determined using linear extrapolation method values were used in calculating of the valence band offset. Values that were used in calculations are shown in Table A-2.

184

Table A-2: Core and VBM values for clean SiC substrate, thin MgO, and thick MgO

samples.

Sample Si 2p, eV Mg 2p, eV VBM, eV

SiC 101.51±0.03 - 2.75±0.03

Thin MgO 100.84±0.03 49.60±0.03 2.65±0.03

Thick MgO - 49.49±0.03 3.10±0.03

From the values shown in Table A-2 using Equation A-1, the VB offset was

calculated. For the MgO grown using MBE on SiC we obtained a VB offset of

1.13±0.12 eV. This value is similar to the value of MgO/GaN obtained by Chen et al. in

[112], but different from the value of MgO/GaAs obtained by Lu et al. in [24]. However,

more experiments need to be conducted in order to determine if this measurement was

correct.

Conclusion

Feasibility of conducting valence band offset analysis in the Interface Engineering

Laboratory was demonstrated. For the MgO grown using MBE on SiC a VB offset of

1.13±0.12 eV was obtained. This value is similar to that of Chen et al.; however, more

experiments are necessary to establish the valence band offset of MgO on SiC and be

able to pursue more complex structures grown via MBE on SiC.

185

Recommendations

Calibration of the binding energy scale is crucial for accurate determination of the

VBM and VB offset [23]; however, no standard calibration procedure is used by researchers performing VB analysis. For example, Au, Ag, Cu, and Fermi edge were used by Liu et al. [114], Au, Ag, and Cu were used by Mi et al. in [115] and Wang et al. in [116], Ag with no Fermi edge by Gassenbauer et al. in [117], while Konovalov et al. in

[118] stated that “no absolute calibration of the energy axis” was necessary because of calculation of the differences. We feel that calibration is extremely important, so we used international standard VII ISO 15472 from [25] to calibrate our instrument. However, the VII ISO 15472 standard is a calibration standard for XPS in general. Thus, the effect of the different calibration methods on the valence band offset measurements should be determined. A modified calibration procedure for the VB measurements may be necessary.

Additionally, until effect of oxygen or other contamination is determined on the

VB offset through a controlled study, pristinely clean SiC samples should be used to eliminate any uncontrollable factors. Oxygen contamination level used in this study

(>15%) should be considered unacceptable. However, in the future effect of oxygen can be a focus of an interesting investigation.

186

Appendix B

Buffered Chemical Polishing of Niobium

187

BCP treatment was performed on three types of niobium: polycrystalline, monocrystalline, and high RRR (an RRR value is unknown). Samples were degreased prior to BCP treatment. BCP was performed according to procedure in section 3.2.2.

Samples were placed into the loading dock of the analysis chamber for XPS analysis within 10 min after the treatment was performed. After XPS analysis mass loss of the samples was determined. Figure B-1 shows mass loss for polycrystalline, monocrystalline, and high RRR niobium samples after 1 and 6 min of BCP. The values are averages, and the error bars represent the range between the values that were obtained for the same type processed at identical conditions.

3.5% 12% 3.0% A B 10% 2.5% 8% 2.0%

1.5% 6% Mass loss, % 1.0% Mass % loss, 4%

0.5% 2%

0.0% 0% poly mono high RRR poly mono high RRR

Figure B-1: Mass loss due to BCP treatment of polycrystalline, monocrystalline, and

high RRR niobium samples after 1 min (A) and 6 min (B).

Polycrystalline samples show significantly higher material removal in BCP that monocrystalline and high RRR samples. It was shown that BCP etches at the grain boundaries [32]. Monocrystalline samples were cut from a slab of niobium which had very few grain boundaries. Exact characteristics of polycrystalline and high RRR samples are not known; however, from BCP experiments it is clear that high RRR samples behave in BCP more like monocrystalline than polycrystalline samples.

188

The XPS analysis of the samples showed that oxide was removed due to BCP treatment. Figure B-2 shows evolution of monocrystalline sample with BCP treatment.

Figure B-2A is Nb3d spectrum of as-received monocrystalline sample. Very small amount of niobium metal is detected. Niobium pentoxide is the predominant oxide.

After 1 min of BCP, Figure B-2B, metal peak dramatically increased, while pentoxide peak decreased. Higher fraction of intermediate oxides was detected. Sample treated with BCP for 6 min (Figure B-2C) has a similar spectrum as the sample treated with BCP for 1 min.

C Count

216 214 212 210 208 206 204 202 200 198 Binding Energy, eV

Figure B-2: BCP treatment on monocrystalline removes niobium pentoxide and

creates intermediate oxide on the surface of niobium. A = as-received

sample, B = 1 min BCP, C = 6 min BCP.

The FWHM changed for all samples due to BCP treatment. Figure B-3 shows the

FWHM values for polycrystalline, monocrystalline, and high RRR samples. The FWHM

189 for the pentoxide peak increased due to BCP treatment, while it decreased for the metal peak. For high RRR sample, the FWHM remained the same through the BCP treatment.

Increase in the FWHM for the oxide peak indicates that crystallographic disorder of pentoxide increased due to the BCP treatment, despite improvement in order of the metallic niobium. We hypothesize that oxidation during short period of time between treatment and analysis in UHV environment forms less uniform oxide layer than long exposure to air.

1.9 1.8 1.7 1.6 Metal Poly Nb2O5 Poly 1.5 Metal Mono 1.4 Nb2O5 Mono Metal High RRR FWHM, eV FWHM, 1.3 Nb2O5 High RRR 1.2 1.1 1.0 Initial 1 min BCP 6 min BCP

Figure B-3: FWHM changed for polycrystalline, monocrystalline, and high RRR

samples.

Comparison of the ratio of metal peak to the total niobium peak shows that more metallic niobium is detected after the BCP treatment. Results for samples BCPed for 1 and 6 min are compared with as-received surfaces in Figure B-4. Two fold increase in

Nb/Total for all samples, polycrystalline, monocrystalline, and high RRR, was

190 determined. Therefore, it can be concluded that material removal, as seen in Figure B-4

(bar graph) was most likely due to niobium pentoxide dissolution. Total oxide thickness for all samples decreased due to BCP treatment, as seen in a line graph in Figure B-4.

Oxide thickness for the polycrystalline and monocrystalline samples showed slight increase in the total oxide thickness between 1 min and 6 min BCP treatments. For high

RRR samples, oxide thickness decrease monotonously. This may indicate an effect of a crystallographic structure, but more thorough investigation and more information about the starting surface condition is needed to conclude decisively.

0.6 5.0 4.8 0.5 4.6 4.4 0.4 4.2 0.3 4.0

Nb/Total 3.8 0.2 3.6 Oxide thickness, nmOxide thickness, 0.1 3.4 3.2 0.0 3.0 Initial 1 min 6 min

Polycrystalline Monocrystalline High RRR

Figure B-4: Niobium/Total ratio determined from Nb3d spectra obtained with XPS,

shows increase in metal signal in sampling volume (bar graph) and

decrease in total oxide thickness (line graph).

191

The SEM analysis of the sample after the BCP treatment showed removal of the surface features. Obtaining of the SEM images on all samples that were treated with 6 min of BCP was difficult due to a low contrast caused by a removal of the surface defects. Figure B-5, Figure B-6, and Figure B-7 show surface evolution of the polycrystalline niobium samples. Figure B-5 shows numerous defects on the surface before BCP. These defects are partially removed by BCP as seen in Figure B-6 and

Figure B-7. Higher magnification images in Figure B-6 and Figure B-7 show that some point defects appear on the surface. It is not clear if these defects were flaws in the material that were revealed during BCP etch or they were caused by BCP treatment.

50.0 µm 5.00 µm

Figure B-5: SEM images of polycrystalline niobium samples before BCP.

50.0 µm 5.00 µm

Figure B-6: SEM images of the polycrystalline sample show defects on the surface

after 1 min BCP.

192

50.0 µm 5.00 µm

Figure 7: SEM images of the polycrystalline samples treated with 6 min BCP exhibit

presence of the grain boundaries.

Monocrystalline samples were treated with BCP and showed less mass loss as seen in Figure B-1. Figure B-8 and Figure B-9 show monocrystalline surface before and after BCP treatment. A crystal grain boundary was not preferentially etched during BCP when comparing images from Figure B-8 and Figure B-9. BCP produced smooth surfaces which lacked contrast, so only high magnification images were obtained.

200 µm 5.00 µm

Figure B-8: SEM images of monocrystalline surface before BCP

193

A B

200 µm 50.0 µm

Figure B-9: SEM images for monocrystalline samples after (A) 1 min and (B) 6 min of BCP.

Finally, SEM images of the high RRR samples prior to the BCP treatment showed highly irregular and defected surface as seen in Figure B-10. These defects were significantly smoothed out by 1 min BCP treatment as seen in B-11A. 6 min BCP treatment produced smooth surface with low contrast in Figure B-11B. No point-like defects which were seen on polycrystalline samples treated with BCP were observed on high RRR BCP treated sample.

50.0 µm 5.00 µm

Figure B-10: High RRR surface prior to BCP treatment

194

A B

50.0 µm 50.0 µm

Figure B-11: SEM images of the high RRR samples treated with BCP for (A) 1 and

(B) 6 min.

To conclude, niobium samples behave differently in BCP solution depending on their crystal structure. The polycrystalline niobium samples exhibited the highest material removal over 1 and 6 min treatment in BCP. For the polycrystalline samples, the FWHM of the metal peak decreased from 1.30±0.05 eV to 1.15±0.05 eV, and for the pentoxide peak it increased from 1.55±0.05 eV to 1.80±0.05 eV for 1 min and 1.70±0.05 eV for 6 min BCP. The SEM images of the polycrystalline samples showed removal of the damaged layers; however, undetermined defects appeared on the surface after the

BCP treatment. Monocrystalline and high RRR samples had significantly lower amount of the material removed by BCP. For the monocrystalline samples, the FWHM decreased for the metal peak from 1.25±0.05 eV to 1.05±0.05 eV, while the FWHM of pentoxide increased from 1.45±0.05 eV to 1.60±0.05 eV. For the high RRR samples, the

FWHM of metal remained the same, but increased for the pentoxide from 1.05±0.05 eV to 1.70±0.05 eV.

In general, BCP was found to smooth the surface and improve atomic order of the metal, but decrease the atomic order of the pentoxide layer. The BCP pre-treatment of the monocrystalline and the high RRR samples may be beneficial for CMP process;

195 however, BCP on the polycrystalline samples may be creating defects. The roughness measurements of these samples may be useful to conclude if using BCP may be beneficial to remove bulk of the material prior to the CMP polishing.

196

Appendix C

Photoreactor Specifications

197

A photoreactor for silicon surface modification was built, Figure C-1. Reactor lamps and quartz vessel (tube) were purchased for Southern Connecticut Ultra Violet

Company (Table ). Spectral characteristics of the 300-nm lamps purchased are shown in

Figure C-1. In order to operate these lamps, starters and ballasts had to be installed.

Photoreactor housing and supports were constructed from various materials that were purchased from Home Depot, Wal-Mart, and Home Goods.

Table C-1: Main photoreactor parts.

Part number Purchased from

Quartz vessel (18 mL) 14/35 Joint RQV-7 Southern CT UV Company Reactor lamps RMR-3000A

Ballast LPL-5-9-TP

Light bulb holders 3ZL03 Granger

Starter 2F974

A B

Figure C-1: Photoreactor (A) front, (B) back.

198

Originally the three lamps were positioned around the opening for the tube, see

Figure C-1B. However, intensity measurements showed that intensity dropped rapidly with distance. From Figure C-3 it can be seen that each lamp produced 0.0052±0.0005

W/cm2 at 0.00 in away from the lamp, but intensity decreased to zero at 0.75 in. This finding, in addition to some preliminary experiments targeted to determine photoreactor effectiveness, indicated that lamps needed to be as close to the sample as possible. In addition, it was realized that illumination of the tube from all sides was not necessary because a flat surface needed to be modified. Therefore lamps were repositioned to be close to each other and the tube. Intensity measurements in this configuration could not be performed due to inability to fit radiometer into the center of the photoreactor.

Figure C-2: Spectral distribution of irradiance density for UV lamps, 300-nm light

source is in blue [119].

199

During the preliminary experiments, the sample was placed directly into the reagent in the tube; however its placement and angle between the sample and the lamps was not controlled. Thus, a special Teflon sample holder was machined. It had a slot for the sample to be put in vertically. This holder was used for the study described in section

4.3.2.

0.007

0.006

0.005

0.004

0.003 Intensity (W/cm2) Intensity 0.002

0.001

0.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Distance, in

Figure C-3: Intensity drops to zero at 0.75 in.

200