Fundamental Studies for the Design of Tantalum Nitride Photoanodes for Solar Water Splitting a Dissertation Submitted to the De

FUNDAMENTAL STUDIES FOR THE DESIGN OF TANTALUM NITRIDE PHOTOANODES FOR SOLAR WATER SPLITTING

A DISSERTATION SUBMITTED TO THE DEPARTMENT OF CHEMICAL ENGINEERING AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Blaise A. Pinaud October 2013

This work is licensed under a Creative Commons Attribution- Noncommercial 3.0 United States License. http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/gp551gc7846

ii I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Thomas Jaramillo, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Stacey Bent

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Mark Brongersma

Approved for the Stanford University Committee on Graduate Studies. Patricia J. Gumport, Vice Provost for Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file in University Archives.

iii iv

Abstract

One of today’s greatest challenges is meeting the increasing global energy demand, projected to rise from our current consumption of 18 TW to nearly 26 TW by 2035. A growing environmental awareness of the impact of fossil fuels on many aspects of life, from climate change to personal health, is driving the development of clean, renewable energy sources which must be both technologically and economically viable. The synthesis of chemical fuels such as hydrogen from sustainable energy sources such as solar or wind is an attractive option. Photoelectrochemical (PEC) water splitting is one promising route for hydrogen production. In PEC devices, semiconductor absorbers harvest solar energy to generate excited electrons and holes to drive the hydrogen and oxygen evolution reactions in aqueous electrolytes.

The first part of this dissertation focuses on a technoeconomic evaluation of conceptual water splitting plants based on four different reactor designs with the aim of identifying key research needs in the field. A significant finding is that more efficient semiconductor photoelectrodes must be developed to make this technology cost-competitive with existing fossil fuel energy sources.

Tantalum nitride (Ta3N5) is a promising photoanode candidate due to its nearly ideal band structure for solar water splitting. The remainder of the dissertation focuses on understanding which properties may limit its performance in order to ultimately design a higher efficiency oxygen-evolving photoanode of this material. An emphasis is placed on developing well-defined sample types and accurate measurement tools to systematically study the fundamental structural, optical, electronic, and photoelectrochemical properties of Ta3N5 photoanodes.

Photoelectrochemical measurements on Ta3N5 thin films grown via thermal oxidation and nitridation of Ta foils reveal that their photoactivity is strongly correlated with increased surface area, suggesting poor hole transport. While the thermal conversion of Ta foils is facile, it is difficult to control the surface morphology which hinders the systematic study of material properties. Work then shifts to the development of an improved sample architecture for the synthesis of flat, crack-free films of tightly controlled thickness. It is

v achieved by converting evaporated Ta layers with or without a Pt contact supported on fused silica to form Ta3N5/Pt/fused silica or Ta3N5/fused silica. These samples enable quantitative assessment of the electronic conductivity and optical absorption of Ta3N5.

Results from the literature suggest the presence of impurity phases in some Ta3N5 photoelectrodes. We therefore seek to control the synthesis of phase-pure materials through an understanding of the effect of nitridation temperature and the underlying substrate on film quality. We discover that temperature has little consequence on the crystallinity and absorption properties of Ta3N5 synthesized on fused silica but that the presence of mobile Ta atoms in Ta foil substrates can result in the formation of reduced nitride phases (e.g. Ta2N, Ta5N6) at temperatures at or above 1000°C. The localization of the Ta2N phase at the Ta3N5/Ta foil interface is demonstrated with grazing incidence x- ray scattering measurements.

We next turn our attention to a well-known issue with Ta3N5, its degradation under illumination to Ta2O5 which blocks transport of the photogenerated holes to the surface.

Several catalysts for the oxygen evolution reaction (OER) are deposited on the Ta3N5 surface with the aim of stabilizing the material and improving the water oxidation kinetics. While the precious metal-based catalysts Pt, RuO2, and IrO2 have higher inherent catalytic activity than the novel CoTiOx catalyst developed in our lab, the latter outperforms the three others both in terms of increasing the photocurrent of and stabilizing Ta3N5. This result highlights the important role of the catalyst/semiconductor interface.

Lastly, the knowledge of the optimal synthesis conditions, hole and electron transport lengths, and absorption depth is combined to design a core-shell Ta-Ta3N5 photoanode. Several approaches are explored for the nanostructuring of the Ta scaffold on which the

Ta3N5 shell can be grown thermally or through anodization. Nanosphere lithography to form a Cr mask followed by reactive ion etching transfers a porous pattern to the Ta foil. A short oxidation time is crucial for limiting the thickness of the shell and preserving the

Ta core. We demonstrate a proof-of-principle Ta-Ta3N5 core-shell and discuss means by which to increase the overall aspect ratio.

In summary, this dissertation covers fundamental studies of the properties of Ta3N5 to design and develop a high performance photoanode which will hopefully enable more efficient solar water splitting devices for the production of hydrogen fuel.

vii

viii

Acknowledgements

Graduate school is not always an easy undertaking but the wonderful support of a wide network of people – colleagues, professors, friends, and family – have made my time at Stanford infinitely more productive and enjoyable.

I would like to start by acknowledging my terrific adviser, Professor Tom Jaramillo. I vividly remember debating which research group at Stanford would be the right fit for me and I was ultimately drawn to Tom’s enthusiasm and drive to not simply run experiments, but to rigorously design the right tests to draw meaningful conclusions. I truly appreciated his willingness to allow students to get hands on experience with every aspect of research in the group, from synthesis to characterization to testing. I am a significantly more well-rounded researcher for my time spent under Tom’s guidance.

While your choice of adviser and research direction play a major role in your graduate experience, it is with your fellow group members that you spend most of your waking hours. I could not have asked for a better group of people than my lab mates in the Jaramillo group! Our once small group grew significantly over the years, with each new member bringing something to the mix, but I would like to acknowledge those who particularly impacted my time at Stanford.

First and foremost, I must acknowledge the incredible mentorship I received from Dr. Zhebo Chen. I joined the group with only a basic knowledge of electrochemistry and absolutely no understanding of semiconductor physics. He patiently taught me the important concepts and showed me how to run the experiments correctly. Despite little overlap in our research projects, Dr. Kendra Kuhl has also been an excellent mentor and an incredible friend. Dr. Yelena Gorlin’s keen insights into electrochemical phenomena and attention to detail pushed me to think deeply about my data. Dr. Jakob Kibsgaard’s superior graphic design skills helped elevate the quality of many of my presentations and figures. Perhaps even more importantly, his positive attitude made it a pleasure to come into the office each morning. Dr. Peter Vesborg was an invaluable mentor for work on tantalum nitride and oxynitrides, always providing excellent guidance when I did not know what to try next. I was fortunate enough to be a member of a very talented

ix photoelectrochemistry sub-group. Jesse Benck, Linsey Seitz, and Pong Chakthranont are not only incredibly intelligent people, they are fantastic friends. If I was having a tough day in lab where nothing seemed to work, Jesse, Linsey, and Pong were always willing to help troubleshoot and remind me of the greater goal of our work. I would like to also thank David Abram, my fellow classmate who joined the Jaramillo group. Together we learned the ropes in lab and shared many great adventures!

Whether lounging with a beer at a group social out on the lawn or competing for the almond during Danish Christmas lunch, group members Dr. Benjamin Reinecke, Etosha Cave, Ariel Jackson, Desmond Ng, Toru Hatsukade, Ieva Narkeviciute, Tommy Hellstern, Jeremy Feaster, Dr. Arnold Forman, Dr. Chris Hahn, Dr. Maureen Tang, and Dr. Samuel Fleischman have made my daily life all that much more fun. I would also like to acknowledge all of the other visiting professors, post-doctoral scholars, master’s students, and undergraduates who worked in the lab. Lastly, I was fortunate enough to have the opportunity to mentor several undergraduate and graduate students. Desmond Ng and Ieva Narkeviciute were both eager rotation students who, happily, chose to join our group. Jared O’Leary was an undergraduate student who worked diligently on the very difficult problem of synthesizing phase pure tantalum oxynitride; his unfailing enthusiasm and work ethic were much appreciated.

Of course, none of the research presented in this dissertation would have been possible without funding via fellowships and grants to support everything from purchasing laboratory supplies to paying my stipend. I was fortunate enough to come to Stanford with a two-year Natural Sciences and Engineering Research Council of Canada PGS-M fellowship and was subsequently granted another two years of funding through a PGS-D award. At the start of my third year, I was selected as a recipient of the United Technologies Research Center (UTRC) fellowship in Renewable Energy. Not only did this award provide funding for my final three years, I was given the opportunity to visit UTRC in East Hartford, Connecticut. This visit was truly an eye-opening experience; first class research applicable to real-world applications was being carried out in a well- equipped, immaculate, modern facility by incredibly talented individuals. I hope to one day work in this type of phenomenal industrial research center.

Graduate students are rather expensive; even after paying for tuition and a stipend there are a myriad of other costs including supplies, new equipment, instrument time, conference fees, and even manuscript publication costs. Several grants from the National Science Foundation (NSF) covered these expenses over the course of my PhD. Early work on the photoelectrochemistry of birnessite-type MnO2 was supported by NSF Grant No. 0824484. All work on the tantalum nitride system was supported by the NSF Chemical Center of Innovation (CCI) for Solar Fuels based at Caltech but also involving institutions from across the continent. Dr. Siddharth Dasgupta worked tirelessly to encourage open lines of communication between the research groups. The CCI was about more than just funding – it was an opportunity to collaborate with and learn from students and professors with a wide variety of backgrounds and areas of expertise under the solar fuels umbrella. The annual retreat in Huntington Beach was an excellent chance to talk science and I received many helpful suggestions which directly impacted my research direction. The CCI led directly to two particularly close collaborations. The first was with Dr. Shane Ardo who was very instrumental in the technoeconomic evaluation of solar water splitting. The second was with Professor Giulia Galli and Dr. Juliana Morbec at University of California Davis who provided valuable insight into the electronic band structure of Ta3N5 through density functional theory calculations.

One of the reasons I chose to come to Stanford was the availability of an amazing suite of characterization tools. Not only is there an instrument on campus for every three letter acronym technique imaginable, students are trained to use these tools themselves and analyze the data. The instruments I utilized on a nearly daily basis fall under the auspices of three different facilities, the Stanford Nanocharacterization Laboratory (SNL), Stanford Nanofabrication Facility (SNF), and Soft and Hybrid Materials Facility (SMF). I would like to acknowledge all the support staff, especially Dr. Arturas Vailionis, Richard Chin, and Chuck Hitzman at SNL, Dr. Nancy Latta, Dr. Jim McVittie, Dr. Jim Kruger, and Dr. Ed Myers at SNF, and Dr. Jeffrey Tok at SMF. Grazing incidence x-ray scattering experiments to study the localization of tantalum nitride phases were carried out at the Stanford Synchrotron Radiation Lightsources, a Directorate of SLAC National Accelerator Laboratory and an Office of Science User Facility operated for the U.S. Department of Energy Office of Science by Stanford University. The expertise of Dr.

Arturas Vailionis and Dr. Chad Miller was invaluable in enabling the very rapid collection of this data.

I am grateful to Professors Stacey Bent and Mark Brongersma for taking the time to be on my reading committee and provide valuable feedback on this work. I also thoroughly enjoyed my many interactions with members of their research groups, including constructive discussions of electrochemical phenomena with Katie Pickrahn of the Bent group and working with Dr. Isabell Thomann of the Brongersma group to understand plasmonic enhanced photoelectrochemical water splitting on Fe2O3. I have also appreciated the time I spent collaborating with students of Professor Alfred Spormann, another member of my committee. Dr. Svenja Lohner, Holly Sewell, and Ann Lesnefsky took the time to explain the biology and reactions so that I understood their needs in developing a silicon photocathode for methyl viologen reduction. In return, I was happy to act as an occasional electrochemistry consultant. I also enjoyed a fruitful collaboration on the study of III-V semiconductor materials for water splitting with Dr. Xinyu Bao of Professor Philip Wong’s group.

There have been highs and lows throughout my five years in graduate school. There is absolutely no way that I would have been able to make it through this time without the support of my incredible family just a phone call away. My brother Marc has ensured my math skills stayed sharp and my sister Claire has sent me countless little treats that never fail to brighten my day. I would like to thank my parents, Carol and Louis, for keeping me grounded while always encouraging me to reach for the stars. Though my educational path has taken me progressively further afield, they will always make Sumac Street home. And last but certainly not least, I am forever indebted for the incredible support of my fiancé Ash Charles. He is the best dining companion, the greatest travel buddy, and most understanding friend that one could ever ask for. Merci!

xii

Publications

1. P. Chakthranont, A. J. Forman, B. A. Pinaud, L. C. Seitz, and T. F. Jaramillo, “Nanoporous Hematite Photoabsorber on High Surface Area Transparent Conductive Oxide Electrodes for Solar Water Splitting”, In Preparation.

2. L. C. Seitz, B. A. Pinaud, D. Nordlund, and T. F. Jaramillo, “Effect of Temperature

Treatment on CoTiOx Catalyst for the Oxygen Evolution Reaction”, In Preparation.

3. J. D. Benck, B. A. Pinaud, Y. Gorlin, and T. F. Jaramillo, “Substrate Selection for Fundamental Studies of Electrocatalysts and Photoelectrodes: Inert Potential Windows in Acidic, Neutral, and Basic Electrolyte”, Submitted.

4. B. A. Pinaud, A. Vailionis, and T. F. Jaramillo, “Controlling the structural and

optical properties of Ta3N5 films through nitridation temperature and the nature of the Ta metal”, Submitted.

5. L. C. Seitz, Z. Chen, A. J. Forman, B. A. Pinaud, J. D. Benck, and T. F. Jaramillo, “Modeling the Practical Performance Limits of Photoelectrochemical Water Splitting Based on the Current State of Materials Research”, Submitted.

6. B. A. Pinaud, J. D. Benck, L. C. Seitz, A. J. Forman, Z. Chen, T. G. Deutsch, B. D. James, K. N. Baum, G. N. Baum, S. Ardo, H. Wang, E. Miller, and T. F. Jaramillo, “Technical and economic feasibility of centralized facilities for solar hydrogen production via photocatalysis and photoelectrochemistry”, Energy & Environmental Science 6, 1983-2002 (2013).

7. B. A. Pinaud, P. C. K. Vesborg, and T. F. Jaramillo, “Effect of Film Morphology and

Thickness on Charge Transport in Ta3N5/Ta Photoanodes for Solar Water Splitting”, Journal of Physical Chemistry C 116 (30), 15918-15924 (2012).

xiii

8. I. Thomann, B. A. Pinaud, Z. Chen, B. M. Clemens, T. F. Jaramillo, and M. L. Brongersma, “Plasmon Enhance Solar-to-Fuel Energy Conversion”, Nano Letters 11 (8), 3440-3446 (2011).

9. B. A. Pinaud, Z. Chen, D. N. Abram, and T. F. Jaramillo, “Films of Sodium

Birnessite-Type MnO2: Optical Properties, Electronic Band Structure, and Solar Photoelectrochemistry”, Journal of Physical Chemistry C 115 (23), 11830-11838 (2011).

10. X.-Y. Bao, B. A. Pinaud, J. Parker, S. Aloni, T. F. Jaramillo, and H.-S. P. Wong, “Monolithic III-V nanowires PV for photoelectrochemical hydrogen generation”, 35th IEEE Photovoltaic Specialists Conference, 001793-001796 (2010).

xiv

Table of Contents

Chapter 1: Introduction ...... 1 1.1 Introduction ...... 1 1.2 Solar Water Splitting Fundamentals ...... 3 1.3 Tantalum Nitride Background ...... 5 1.4 Dissertation Overview ...... 6 1.5 Copyright ...... 8 Chapter 2: Technical and Economic Feasibility of Centralized Facilities for Solar Hydrogen Production ...... 11 2.1 Abstract ...... 11 2.2 Introduction ...... 12 2.3 Theoretical and Demonstrated Water Splitting Efficiencies ...... 13 2.3.1 Calculation of Practical System Efficiencies ...... 13 2.3.2 Demonstrated Research Efficiencies ...... 17 2.4 Conceptual Reactor Designs ...... 21 2.4.1 Type 1 Reactor: Single Bed Particle Suspension ...... 22 2.4.2 Type 2 Reactor: Dual Bed Particle Suspension ...... 24 2.4.3 Type 3 Reactor: Fixed Panel Array ...... 25 2.4.4 Type 4 Reactor: Tracking Concentrator Array ...... 26 2.4.5 Reactor Design Limitations ...... 27 2.5 Plant Design and Operation...... 29 2.5.1 Solar Insolation ...... 29 2.5.2 Type 1 Plant Design: Single Bed Particle Suspension...... 30 2.5.3 Type 2 Plant Design: Dual Bed Particle Suspension ...... 31 2.5.4 Type 3 Plant Design: Fixed Panel Array ...... 33 2.5.5 Type 4 Plant Design: Tracking Concentrator Array ...... 33 2.5.6 Gas Processing and Control System Subassemblies ...... 34 2.6 Costs ...... 36 2.6.1 System-specific Costs ...... 37 2.6.2 System Capital Costs ...... 38 2.6.3 Levelized Hydrogen Cost & Sensitivity Analysis ...... 39 2.6.4 Cost Comparison ...... 44 2.7 Conclusions ...... 45

2.8 Author Contributions...... 46 2.9 Copyright ...... 46

Chapter 3: Effect of Film Morphology and Thickness on Charge Transport in Ta3N5/Ta Photoanodes for Solar Water Splitting ...... 49 3.1 Abstract ...... 49 3.2 Introduction ...... 49 3.3 Experimental Methods ...... 51 3.3.1 Sample Preparation ...... 51 3.3.2 Film Characterization...... 52 3.3.3 Electrochemical Testing...... 53 3.4 Results and Discussion ...... 54 3.4.1 Film Thickness and Morphology ...... 54 3.4.2 Film Crystallinity ...... 57 3.4.3 Photoelectrochemistry...... 58 3.4.4 Implications for Charge Transport ...... 63 3.5 Conclusions ...... 66 3.6 Author Contributions...... 66 3.7 Copyright ...... 66 Chapter 4: Development of a Flexible Platform for Studying Electrical, Optical, and Photoelectrochemical Properties of Ta3N5 ...... 67 4.1 Abstract ...... 67 4.2 Introduction ...... 67 4.3 Experimental Methods ...... 68 4.3.1 Synthesis and Physical Characterization ...... 68 4.3.2 Electrical Characterization ...... 69 4.3.3 Electrochemical Characterization ...... 69 4.4 Design of Sample Architecture ...... 71 4.4.1 Potential Synthetic Routes ...... 71 4.4.2 Substrate Selection ...... 71 4.5 Results and Discussion ...... 74 4.5.1 Physical Characterization...... 74 4.5.2 Electrical Characterization ...... 76 4.5.3 Photoelectrochemical Characterization ...... 78 4.5.4 Drawbacks of Architecture ...... 80 4.6 Conclusions ...... 82

xvi

Chapter 5: Co-catalyst Functionalization of Ta3N5 Photoanodes ...... 83 5.1 Abstract ...... 83 5.2 Introduction ...... 83 5.3 Experimental Methods ...... 86

5.3.1 Ta3N5 Photoanode Synthesis...... 86 5.3.2 Electrochemical and Photoelectrochemical Characterization ...... 86 5.3.3 Pt Synthesis ...... 87

5.3.4 IrO2 Synthesis ...... 88

5.3.5 RuO2 Synthesis ...... 89

5.3.6 CoTiOx Synthesis ...... 90 5.4 Results and Discussion ...... 92 5.4.1 Catalyst Coverage and Morphology ...... 92 5.4.2 Photoelectrochemical Activity of Functionalized Photoanodes ...... 93 5.4.3 Stability ...... 95 5.5 Conclusions ...... 97

Chapter 6: Controlling the Structural and Optical Properties of Ta3N5 through Nitridation Temperature and the Nature of the Ta Metal ...... 99 6.1 Abstract ...... 99 6.2 Introduction ...... 99 6.3 Experimental Methods ...... 100 6.3.1 Sample Preparation ...... 100 6.3.2 X-ray Analysis of Crystallinity ...... 103 6.3.3 Optical Absorption Measurements ...... 103 6.4 Results and Discussion ...... 103

6.4.1 Effect of Temperature on Ta3N5 Crystallinity ...... 103

6.4.2 Effect of Substrate on Ta3N5 Growth ...... 104 6.4.3 Mechanism of Phase Transformations ...... 106

6.4.4 Effect of Temperature on Ta3N5 Absorption Properties ...... 109

6.4.5 Optical Features of Ta3N5 ...... 111 6.5 Conclusions ...... 114 6.6 Author Contributions...... 115

Chapter 7: Design of Core-shell Ta-Ta3N5 Architecture for High Efficiency Solar Water Splitting ...... 117 7.1 Abstract ...... 117 7.2 Introduction ...... 117

xvii

7.3 Nanostructuring Ta Metal ...... 119 7.3.1 Templating via Porous Alumina ...... 120 7.3.2 Templating via Nanosphere Lithography ...... 123

7.4 Formation of Core-shell Ta-Ta3N5 ...... 128

7.4.1 TaOx Formation by Anodization ...... 129 7.4.2 Thermal Oxidation and Nitridation ...... 130 7.4.3 Co-catalyst Functionalization ...... 131

7.5 Characterization of Ta-Ta3N5 Photoanodes ...... 131 7.5.1 Morphology...... 131 7.6 Conclusions ...... 132 Chapter 8: Conclusions and Future Directions ...... 135 8.1 Conclusions ...... 135 8.2 Future Directions ...... 136 Appendix A: Substrate Selection for Electrocatalysts and Photoelectrodes: Inert Potential Windows in Acidic, Neutral, and Basic Electrolyte ...... 139 A.1 Abstract ...... 139 A.2 Introduction ...... 139 A.3 Experimental Methods ...... 141 A.3.1 Substrate Preparation ...... 141 A.3.2 Electrochemical Characterization ...... 142 A.3.3 Testing Methodology ...... 143 A.4 Results and Discussion ...... 143 A.4.1 Testing Methodology ...... 143 A.4.2 Transparent Substrates ...... 145 A.4.3 Indium Tin Oxide ...... 145 A.4.4 Fluorine-doped Tin Oxide ...... 148 A.4.5 Aluminum-Doped Zinc Oxide ...... 149 A.4.6 Opaque Substrates ...... 151 A.4.7 Gold...... 151 A.4.8 Stainless Steel ...... 154 A.4.9 Glassy Carbon ...... 157 A.4.10 Highly Oriented Pyrolytic Graphite ...... 159 A.4.11 Summary of Inert Potential Windows ...... 161 A.5 Conclusion ...... 163 A.6 Author Contributions...... 163

xviii

Appendix B: Compression Cell and Integrating Sphere Design and Implementation ... 165 B.1 Introduction ...... 165 B.2 Compression Cell Design ...... 165 B.3 Illumination Calibration ...... 167 B.4 Integrating Sphere Design ...... 170 B.5 Conclusions ...... 173

Appendix C: Thin Films of Na Birnessite-type MnO2: Optical Properties, Electronic Band Structure, and Solar Photoelectrochemistry ...... 175 C.1 Abstract ...... 175 C.2 Introduction ...... 175 C.3 Experimental Section ...... 179 C.3.1 Sample Preparation ...... 179 C.3.2 Film Characterization...... 179 C.3.3 Photoelectrochemical Testing ...... 180 C.4 Results and Discussion ...... 181 C.4.1 Film Morphology ...... 181

C.4.2 MnO2 Crystal Structure by XRD ...... 183 C.4.3 Mn Oxidation State by XPS ...... 186 C.4.4 Band Gap by UV-Vis ...... 188 C.4.5 Photocurrent Generation with Applied Bias and Film Stability ...... 189

C.4.6 Na Birnessite-type MnO2 Band Structure ...... 194 C.5 Conclusions ...... 199 C.6 Author Contributions...... 200 C.7 Copyright ...... 200 Appendix D: Photoreduction of Methyl Viologen on Silicon Photocathodes for Bio- electrochemical Reactors ...... 201 D.1 Introduction ...... 201 D.2 Semiconductor Selection ...... 202 D.3 Photoelectrode and Test Apparatus Design ...... 203 D.4 Photoelectrochemical Characterization ...... 204 D.5 Conclusion ...... 209 Appendix E: Optimization of the Synthesis of Pure, Crystalline TaON ...... 211 E.1 Introduction ...... 211 E.2 Synthetic Parameters ...... 212 E.3 Physical Characterization ...... 214

xix

E.4 Thermal Stability ...... 216 E.5 Conclusions ...... 217 References ...... 219

List of Tables

Table 2.1: Summary of hydrogen output of a net 1 TPD plant module for each of the four conceptual PEC hydrogen production plants with values for the solar input, system efficiencies, reactor dimensions, and emplacement area...... 31

Table 2.2: Summary of liquid and gas handling systems required for each reactor type. 35

Table 2.3: Selected H2A default and assumed input parameters for the H2A costing analysis...... 37

Table 2.4: Utility usage for unit operations and feedstocks for 1 TPD H2 production plant modules of each of the four reactor types...... 38

Table 2.5: Summary of all direct capital expenditures and installation costs for the four

different 1 TPD net H2 production plant modules...... 39

Table 4.1: Change in sheet resistance of the Pt back contact with adhesion layer after undergoing oxidation and nitridation treatments...... 74

Table 5.1: Elemental composition (in atomic %) of the surface of the functionalized

Ta3N5 samples. Only the CoTiOx completely covers the surface of the Ta3N5...... 93

Table C.1: Crystallite size for Na birnessite-type MnO2 thin films ...... 186

xxi

xxii

List of Figures

Figure 2.1: Maximum theoretical solar-to-hydrogen efficiency for a single absorber material plotted as a function of the photoabsorber band gap...... 15

half-size baggie each for O2 and H2 evolution, (c) Type 3 reactor design showing the encased composite panel oriented toward the sun with buoyant separation of gases, and (d) Type 4 reactor design with an offset parabolic cylinder receiver concentrating light on a linear PEC cell. Drawing not to scale...... 23

xxiii

Figure 2.6: Plant module layout of reactor arrays for 1 TPD H2 production for reactors (a) Type 1, (b) Type 2, (c) Type 3, and (d) Type 4. A make-up water subassembly provides water flow to the reactors, the gas processing subassembly purifies and compresses (if needed) the product gas while the control room is used for monitoring. These unit operations are centralized and driveways provide access to the individual beds or reactor arrays. Panel array emplacement is designed to minimize shadowing...... 32

Figure 2.7: Distribution of cost contributions to the levelized price of hydrogen...... 40

Figure 3.1: SEM images showing the surface morphology of (a) as-received unpolished, (b) mechanically polished, and (c) electropolished Ta foils. Optical photographs of the samples are shown in the insets of (a) and (b)...... 51

Figure 3.2: Side view of the compression cell used for three-electrode electrochemical and photoelectrochemical measurements...... 53

Ta3N5 film is colorized to facilitate distinction from Ta metal...... 55

Figure 3.4: Cross-sectional SEM images for (a) Ta metal, (b) Ta2O5, and (c) Ta3N5 on Si substrates. All initial metal films were 140 nm and clear changes in thickness are evident due to differences in the density of the phases...... 56

Figure 3.5: The thickness of Ta, Ta2O5, and Ta3N5 films measured by profilometry (confirmed by SEM) is plotted versus the nominal target thickness measured on the quartz crystal microbalance in the electron beam evaporator...... 56

xxiv

Figure 3.7: X-ray diffractograms for the films of four different thicknesses showing only

peaks attributed to crystalline Ta3N5 or the underlying cubic Ta metal substrate. .... 58

Figure 3.9: Incident photon-to-current efficiency for the Ta3N5/Ta films of four different thicknesses. The two order of magnitude higher activity of the 780 nm and 630 nm films compared to the 260 nm and 60 nm films is clearly evident...... 60

Figure 3.10: Incident photon-to-current efficiency for the Ta3N5/Ta films of four different thicknesses at 1.27 V vs RHE. The trends in activity are the same as at a higher potential...... 60

Figure 3.11: (a) Absorption of Ta3N5 thin films of three different thicknesses grown on quartz substrates with Fabry-Pérot interference fringes clearly visible at longer

wavelengths. (b) Calculated absorption coefficient for Ta3N5. Note that the large error at long wavelengths is due to the interference fringes evident in (a)...... 61

Figure 3.13: Schematics showing the effect of film geometry and structure on electron

and hole transport for supported Ta3N5 in the form of (a) powder films, (b) thick crystalline films, (c) thin crystalline films, and (d) thick, rough crystalline films. ... 64

xxv

Figure 4.2: Ta3N5 samples grown on Ta foils, Pt/fused silica, or fused silica have different degrees of uniformity. Samples with a conductive back contact can be used for electrochemical characterization while samples on a non-conductive, transparent substrate can be used to study optical and electronic properties...... 74

Figure 4.4: X-ray diffractograms contain only peaks assigned to crystalline Ta3N5 and the underlying Pt substrate. Note that the shoulder on the large Pt peak near an angle of 39° arises from Pt due to the Cu K radiation which was not filtered out rather than unconverted Ta metal...... 75

Figure 4.5: Solid state current-voltage curves show that that In-Ta3N5 contact exhibits ohmic behavior...... 78

Figure 4.7: Chopped cyclic voltammograms for all five Ta3N5/Pt/fused silica films of different thicknesses were collected and the trend in photocurrent as a function of thickness at a given potential is shown in each inset. The first cycle is shown in (a) and there does not appear to be a clear trend; the photocurrent initially increases but reaches a plateau for the three thicker films. On the second cycle shown in (b), there is a nearly linear trend between photocurrent and film thickness...... 80

xxvi

Figure 4.8: Cyclic voltammograms collected in the dark on a Pt wire and in the dark and

under illumination for a Ta3N5/Pt/fused silica and a Ta3N5/Ta foil sample. While the

photocurrent on the Ta3N5/Ta foil sample is easily measured, the significant dark

current arising from pinholes on the Ta3N5/Pt/fused silica convolutes the

photocurrent measurement. The expression jphoto and accompanying arrows designate the true photocurrent...... 81

Figure 5.1: Upon illumination, photogenerated holes (shown in the first panel) reach the

surface of Ta3N5 but photooxidize the material to Ta2O5 in addition to driving water oxidation. The band structure alignment (shown in the second panel) is such that further transport of holes to the surface is blocked. One solution (shown in the third panel) is to add a co-catalyst to stabilize the surface and improve the kinetics of water oxidation...... 84

Figure 5.2: (a) SEM images showing the morphology of Pt catalysts electrodeposited for 30 s – 2 min. The catalyst nucleates as particles which grow in size as the potential is held at 0 V vs. Ag/AgCl. (b) Optical images reveal pale gray films on FTO but bubbles (left image) must be tapped off after the pre-polarization step to achieve uniform coverage (right image)...... 88

Figure 5.4: SEM images showing the morphology of CoTiOx films dip coated from diluted (50% v/v, 25% v/v, and 12.5% v/v) sol gel solutions. Stresses in thicker films result in many cracks while thin films conformally coat the FTO grains...... 91

xxvii

color from a nearly transparent film to a deep yellow/brown as shown in the circular tested area in the optical image on the right...... 91

Figure 5.6: Morphology of the Pt, IrO2, RuO2, thin CoTiOx, and thick CoTiOx catalysts

deposited on Ta3N5. An image of bare Ta3N5 is shown as a reference...... 92

Figure 5.7: Chopped cyclic voltammograms revealing the improvement in activity upon

functionalization of the Ta3N5 surface with Pt, IrO2, RuO2, or CoTiOx catalyst. The highest increase in photocurrent and best improvement in stability were achieved

with the CoTiOx catalysts...... 94

Figure 5.8: Chronoamperometry at 1.26 V vs. RHE reveals that the current decays almost

immediately for the samples with no catalyst or Pt and IrO2 catalysts. The

photocurrent on the sample with RuO2 is fairly low but both the thin and thick 2 CoTiOx layers lead to a sustained photocurrent > 0.2 mA/cm ...... 96

Figure 6.1: The Ta3N5 film thickness was measured by cross-sectional SEM imaging on sister samples deposited on Si or by profilometry on a masked fused silica sample.

All samples were nitrided at 900°C in NH3. Thickness values in this chapter are reported based on the cross-sectional SEM values. The Ta metal thickness plotted on the horizontal axis was measured by profilometry...... 101

Figure 6.2: Schematics showing the temperature ramp profile and gas flows during heat treatment. (a) Evaporated Ta on fused silica is oxidized and nitrided in a single run.

(b) To prepare Ta3N5/Ta, all films are first oxidized at 550°C for 15 minutes to form

a film of well-defined thickness. (c) Ta3N5/Ta films are all nitrided in a second step. In each case, the point at which the gas flow is changed is denoted by an arrow. .. 102

Figure 6.3: X-ray diffractograms in (a) of Ta3N5 supported on fused silica prepared at

850°C, 900°C, 950°C, and 1000°C in NH3 reveal nearly identical crystal structures with no impurity phases, regardless of the temperature. Grain sizes in (b) calculated via the Scherrer equation reveal a consistent crystallite size of approximately 17.5 nm...... 104

xxviii

Figure 6.4: Comparison of x-ray diffractograms for films grown on Ta foils at 900°C,

950°C, and 1000°C in NH3. Peaks in the spectrum for the sample prepared at 1000°C

are assigned to Ta5N6 rather than Ta3N5. Both spectra exhibit peaks assigned to a

Ta2N impurity phase formed at the film-substrate interface...... 105

Figure 6.5: Grazing incidence x-ray scattering of Ta3N5/Ta films prepared at (a) 900°C,

(b) 950°C, and (c) 1000°C. At ω = 0.5°, only the surface is probed and either Ta3N5

(900°C and 950°C) or Ta5N6 (1000°C) is detected. Only when the bulk is probed

with ω = 6.0° do the textured features corresponding to Ta and Ta2N appear. For convenience, the reciprocal lattice spacing Q (in Å-1) is converted to a 2θ position assuming Cu Kα radiation shown on the upper horizontal axis...... 107

stoichiometric Ta3N5. This process has been ruled out as no Ta2N or Ta5N6 is detected when fused silica is used as the substrate. The nitridation of sub-oxides (ii)

at the Ta2O5/Ta interface could result in lower valence nitrides. During heat treatment, the Ta atoms from the substrate could diffuse into the film (iii). At higher

temperatures, this process would become more facile which could lead to bulk Ta5N6 formation...... 108

Figure 6.7: UV-vis absorption spectra of Ta3N5 supported on fused silica prepared at

850°C, 900°C, 950°C, and 1000°C in NH3 collected (a) in transmission mode, and (b) with an integrating sphere . Optical images inset in (b) show the gray tint in samples heated to 1000°C. Note that the interference fringes in (a) arise from specular and diffuse reflectance off the flat, uniform samples...... 110

Figure 6.8: Total light balance for Ta3N5 films of different thicknesses supported on fused

silica prepared at 900°C in NH3. The light is broken down into the portion that is (a) absorbed, (b) specularly and diffusely reflected, and (c) transmitted and scattered.112

Figure 6.9: The films of five different thicknesses were synthesized multiple times to verify the reproducibility of the optical measurements. Shown above is data for two

xxix

batches synthesized and measured several months apart; there is excellent agreement in the absorption features at each thickness. Data with colored traces is shown and discussed in the main text of this chapter...... 113

Figure 7.1: Schematic of a Ta-Ta3N5 core-shell architecture consisting of a catalyst-

coated thin film of Ta3N5 grown on a Ta nanostructure. The structure dimensions, catalyst loading, and synthetic conditions are dictated by the knowledge gained from previous studies...... 118

Figure 7.2: Schematic illustrating the process of nanostructuring Ta using an AAO template. The sputtered Al is pre-anodized and the poorly ordered oxide etched away. The final anodization is followed by a pore widening etch and ion milling to transfer the porous structure to the underlying Ta metal foil...... 120

Figure 7.3: SEM images of the AAO template after anodization and following pore widening etches of 45 min, 1.5 hrs, and 2.5 hrs...... 121

Figure 7.7: SEM images of a self-assembled monolayer of latex spheres on Ta foil produced at different spin coating speeds...... 124

xxx

Figure 7.8: SEM images of the latex spheres after reduction of the diameter after

exposure to an O2 plasma for 3, 4.5, 6, and 7.5 min...... 125

Figure 7.11: Cross-sectional SEM of pores etched into a Ta foil using a Cr mask patterned by nanosphere lithography...... 127

Figure 7.14: X-ray diffractograms for Ta foils anodized at 60, 90, and 120 V and nitrided at 900°C...... 130

(b) Top-down and cross-sectional views of a true Ta-Ta3N5 core-shell nanostructure without catalyst synthesized by thermal oxidation (3 min at 550°C using insert at temperature method) and nitridation (4 hrs at 900°C)...... 132

xxxi

Figure A.1: Schematic showing the progressive scan testing methodology...... 144

Figure A.2: Area-normalized circuit resistances for each substrate in each electrolyte. 144

Figure A.3: Electrochemical activity and inert potential range for indium tin oxide. .... 146

Figure A.4: Electrochemical activity and inert potential range for fluorine-doped tin oxide...... 149

Figure A.5: Electrochemical activity and inert potential range for aluminum-doped zinc oxide...... 150

Figure A.6: Electrochemical activity and inert potential range for gold...... 152

Figure A.7: Electrochemical activity and inert potential range for stainless steel 304. .. 155

Figure A.8: Electrochemical activity and inert potential range for glassy carbon...... 158

Figure A.9: Electrochemical activity and inert potential range for highly oriented pyrolytic graphite...... 160

Figure A.10: Potential range in which each substrate is inert for all electrolytes. Chemical stability is indicated by the color of the trace...... 162

The latter is intended to capture only the photons which are absorbed by a Ta3N5 photoanode with a 2.1 eV band gap...... 169

xxxii

Figure B.4: Cumulative photon flux for energies higher than a given threshold wavelength in 50 nm increments. The expected photocurrent for a given flux is shown on the right axis...... 170

(b). The symbols represent the following fractions of light: Io is incident intensity, Rs

is specular reflectance, Rd is diffuse reflectance, S is scattered light, T is transmitted

light, and Ameas is measured absorption. The fractions of light collected by the detector in each method are boxed...... 171

Figure B.6: (a) Image of the integrating sphere used for absorption measurements. (b) Photo of the interior of the integrating sphere showing the light input, baffles, and sample holder...... 171

Figure B.7: Absorption of a ~ 200 nm layer of Ta3N5 due to interiorly reflected light in the integrating sphere...... 172

2 Figure C.2: (a) SEM image of Pt layer deposited on a thin film of MnO2 (50 mC/cm ) with a trench milled away to reveal a cross-section, and (b) SEM image at 45° angle

of Pt/MnO2/FTO interface at cross-section used to measure film thickness...... 182

Figure C.3: SEM image of electrodeposited MnO2 on FTO substrate (bare surface shown in inset, same scale) before electrochemical testing...... 183

Figure C.4: Schematic showing the structure of layered Na birnessite-type MnO2 + consisting of edge-sharing MnO6 octahedra intercalated with Na cations and H2O. A

xxxiii

negative charge within the MnO2 layers is believed to arise from the substitution of Mn(IV) by Mn(III), and is counterbalanced by Na+ cations intercalated between the layers...... 184

● indicates peaks assigned to the birnessite phase of MnO2 and * indicates a peak arising from the FTO substrate...... 185

Figure C.7: XPS spectra from a 100 mC/cm2 film for the (a) Mn 2p region revealing a Mn 2p satellite separation of 11.8 eV, and (b) Mn 3s region revealing a ΔE3s

splitting of 4.5 eV. These measurements are consistent with MnO2...... 187

Figure C.9: (a) Current-potential curve (anodic sweep at 10 mV/s shown) for Na

birnessite-type MnO2 (solid trace) with 1 Hz chopped illumination showing n-type photocurrent and reasonable catalytic activity for the oxygen evolution reaction. Data taken under the same conditions on a bare FTO substrate (dashed trace) is shown as a reference. (b) Photocurrent as a function of time for a film with Q = 25 mC/cm2 at 1.65 V vs. RHE showing only a slight decay after 75 min. Data points are taken every 5 min from a 0.1 Hz chopped chronoamperometry experiment and the dark current has been subtracted...... 190

xxxiv

Figure C.11: Incident photon-to-current efficiency spectrum at 1.65 V vs. RHE for the film with highest photocurrent (Q = 25 mC/cm2)...... 193

Figure C.12: Plots of photocurrent density squared as a function of electrode potential used to extrapolate the onset. The potential of photocurrent onset for films of three different thicknesses corresponding to Q = 25 mC/cm2 (circles), Q = 50 mC/cm2 (squares), and Q = 100 mC/cm2 (triangles) are 1.56 V, 1.54 V, and 1.53 V vs. RHE, respectively. Photocurrents are as measured at 20 mV intervals during cyclic voltammetry with 1 Hz chopped illumination...... 195

Figure C.13: Mott-Schottky measurements in the dark at 85 Hz (circles), 56 Hz (squares),

and 36 Hz (triangles) yielding Efb ~ 1.63 V vs. RHE for Na birnessite-type MnO2. 196

Figure C.14: Position of Ecb (red), Efb (gray), and Evb (black) for Na birnessite-type MnO2 as calculated from the optical band gap and the illuminated open circuit potential, potential for photocurrent onset, and Mott-Schottky plots. The absolute potential of the standard hydrogen electrode is taken to be – 4.44 V.300 ...... 196

Figure D.1: Photo of the assembled H-cell showing the cathode compartment with a Si photoelectrode and the anode compartment with the graphite counter electrode. The cell is clearly anaerobic as evidenced by the reduced MV+ still present in the cathode compartment...... 203

xxxv

Figure D.5: (a) Current as a function of time during five consecutive 1 hour holds at -0.3 V vs. Ag/AgCl. (b) CVs collected after each hold reveal a slow decay in the photoactivity...... 207

Figure D.6: Stability test over 5 hours with inset images of the H-cell. The solution turns deep blue as MV+ is generated but, initially, returns to clear when the solution is

saturated with O2. There is however a gradual yellowing of the solution over time...... 208

Figure E.2: SEM image showing a thick TaON film grown on a Ta metal foil. The inset optical image highlights the bright yellow color of TaON. The underlying red is

likely due to a Ta3N5 impurity...... 214

Figure E.3: (a) SEM image of the textured morphology of the TaON film grown on a Ta

foil. (b) XRD of crystalline TaON/Ta foil samples with and without detectable Ta2O5

or Ta3N5 impurity phases...... 215

Figure E.4: XPS spectra of the O 1s and Ta 4f regions as a function of sputtering time for

a Ta2O5 commercial powder...... 215

xxxvi

Figure E.5: In situ annealing in air of a TaON + Ta3N5/Ta foil sample. While the TaON features remain unchanged up to a temperature of 700°C, the nitride is converted to

Ta2O5...... 216

xxxvii

xxxviii

Chapter 1 Introduction

1.1 Introduction Meeting the current and future global energy needs is one of the greatest challenges facing society today. World energy consumption was 530 quadrillion BTU (17.7 TW) in 2010 and is expected to grow to over 770 quadrillion BTU (25.7 TW) by 2035, with growth led by emerging economies.1 Increased environmental awareness has motivated the search for sustainable sources of energy but these must also be cost effective. The largest source of renewable energy is by far the sun; 1.76 x 105 TW of solar energy reaches the earth’s surface though the fraction of land allocated to power generation limits the amount which can be utilized. A practical number which has been proposed is 600 TW which means a 10% efficient device could provide 60 TW of energy – more than enough to meet growing demand.2

Intensive research and development efforts over the past few decades on non-fossil fuel- based energy solutions have led to a steady increase in incorporation of these technologies into the electrical grid. Examples include electricity generation by wind turbines and solar photovoltaics, technologies that are seeing increased market penetration. The issue of intermittency, however, will hinder large-scale deployment of a number of these carbon-free energy sources by creating problems in maintaining overall grid stability. The synthesis of chemical fuels by using intermittent renewable energy is one pathway to circumvent this instability; chemical fuels can be generated while the wind is blowing or the sun is shining and can be consumed at another time to meet changing power demand. Furthermore, the energy density of chemical fuels far exceeds that of capacitors and batteries. For renewable chemical fuels to make up a significant fraction of the world’s ever-increasing energy needs, production must match the sizeable global demand.

Molecular hydrogen is one of the many chemical fuels being explored. As a commodity, hydrogen is already produced on a large scale (50 million tonnes per year worldwide3), used mostly for petroleum refining as well as the synthesis of ammonia for fertilizer. The majority (> 95%) of global hydrogen is currently produced from fossil fuels, primarily via steam methane reforming.3 Its envisioned use as a clean energy carrier on a large scale is hindered by the need for a cost-competitive, renewable production route and lack of storage and transportation infrastructure. A key advantage of renewable, solar hydrogen over fossil-based chemical fuels is that its use in fuel cells or combustion engines to power vehicles leads to no CO2 emissions. Many renewable hydrogen production technologies exist in various stages of development and these can be broken down into the following three main categories: thermal processes, electrolytic processes, and photolytic processes. The first category includes reforming of bio-derived fuels and thermal cycles with metal oxides (MxOy such as ZnO/Zn) or lower temperature cycles with S-I or Cu-Cl chemistries.4 The second consists of coupling a renewable electricity source, such as wind or solar, with an electrolyzer.5 Photolytic processes can be biological, making use of molecular complexes, hydrogen–evolving enzymes, and natural organisms,6 or photoelectrochemical/photocatalytic involving molecular chromophores or semiconductor absorbers. This last technology is the focus of this thesis.

Photoelectrochemical (PEC) water splitting, a process in which solar energy is used to evolve H2 and O2 from water, is a promising technology because it offers a potentially affordable, carbon-free route to the synthesis of hydrogen. Another key benefit of this process is the high purity of the output hydrogen gas stream, an important requirement for its use in fuel cells. Since the seminal paper from Fujishima and Honda of 1972 7 describing the PEC water splitting process on a TiO2 photoelectrode , significant technical advances have resulted in functional bench-scale systems. PEC cells composed of III-V group semiconductors have demonstrated solar-to-hydrogen (STH) efficiencies as high as 12.4%8 and 18.3%9 depending on the exact device configuration, while multijunction silicon PEC cells have yielded efficiencies in the range of 4.7%10 to 7.8%11 depending on the type of co-catalysts used. Each system faces technical scale-up challenges which range from improving durability to further increasing efficiency to lowering materials and manufacturing costs.

1.2 Solar Water Splitting Fundamentals The process of photoelectrochemical water splitting begins with the absorption of a solar photon in a semiconductor material to form an excited electron-hole pair. If the semiconductor is immersed in an aqueous electrolyte, band bending at the semiconductor/electrolyte interface provides a driving force for the separation of the photogenerated charge carriers. Band bending can also be generated through the use of a p/n junction or other solid state junction analogous to a solar cell. The excited holes must reach one surface to drive the oxygen evolution reaction (OER) while the electrons are consumed by the hydrogen evolution reaction (HER) at a separate electrode or surface. PEC hydrogen production systems can incorporate a single photoanode with dark cathode, a single photocathode with dark anode, or multiple absorbers to make up a tandem device. Note that for a tandem system, the absorbers can be stacked one on top of another or placed side-by-side. In a single absorber system consisting of an n-type semiconductor photoanode, the OER occurs on the photoelectrode surface and electrons, the majority charge carriers, flow to the cathode. In a system consisting of a p-type semiconductor photocathode, the HER will occur on the photoelectrode surface and holes, the majority charge carriers, flow to the anode. In a tandem cell device, the overall water splitting process consists of the same reactions, but both the photoanode and photocathode absorb photons and create excited charge carriers. A tandem structure with two photoelectrodes is illustrated in Figure 1.1. PEC water splitting is not limited to systems with panel electrodes. Suspensions of photocatalyst particles on which either one or both water splitting half-reactions occur have been studied. Several comprehensive reviews12 of the physics and chemistry13, materials requirements14, 15 and candidate semiconductors16 have been published recently and the reader is referred to these for an in depth review of the field. The key requirements for the semiconducting material(s) are a suitable band gap for light absorption, proper band edge alignment for the redox reactions of interest, long term stability in an aqueous environment, as well as cost and material availability. The first material constraint of importance is the band gap. Thermodynamics dictate a minimum voltage requirement of 1.23 V to split water at standard temperature, pressure, and concentrations, thus necessitating at least a 1.23 eV band gap semiconductor.

Figure 1.1: Schematic band diagram showing the phenomena of photon absorption, band bending, charge separation, as well as hydrogen and oxygen evolution on semiconductor photoanode and photocathode surfaces. The metallic back contacts could also be replaced by a redox mediator to shuttle charges between the two photoelectrodes. In practice, however, entropic losses, reaction overpotentials, and other parasitic losses raise the overall band gap requirement. Simply put, the band gap must be large enough to provide the necessary photovoltage to split water but must be as small as possible to absorb the greatest portion of the solar spectrum. Solar photon flux utilization can be maximized by employing multiple smaller band gap absorber layers connected in series to yield a combined voltage large enough to split water at relevant reaction rates. This approach has proven successful in the photovoltaic industry17-19 as well as in laboratory PEC water splitting devices8, 9, 11. The generation of the requisite photovoltage is a necessary but insufficient condition to split water. The energy levels at which the electrons and holes are injected to solution must exceed the electrochemical redox potentials for the HER and the OER, respectively. Thus, at the very least, the potential of the conduction band of the semiconductor at the semiconductor/liquid junction must be more negative than 0.0 V vs. RHE while the valence band must be more positive than 1.23 V vs. RHE. Catalysis also plays an important role in PEC water splitting. Reducing the overpotential for each redox reaction lowers the total voltage required to obtain a rapid rate of water splitting. If the surface of the optimal absorber material is not

4 inherently a good catalyst (which is oftentimes the case), it can be decorated with an HER or OER co-catalyst but care must be taken to ensure additional losses are not introduced at the semiconductor/catalyst interface due to shadowing or the formation of interfacial defect states. Charge transport within the absorber material and across the electrode/electrolyte interface must be fast to reduce recombination. Stability in an aqueous environment is essential for long term operation and plant durability. The electrode must not corrode or undergo any changes detrimental to performance either in the dark (nighttime conditions) or under illumination (daytime conditions). Lastly, the market cost and accessibility (e.g. earth abundance) of the constituent materials are key considerations if solar hydrogen from water splitting is to be viable on a large scale.20

1.3 Tantalum Nitride Background

Tantalum nitride (Ta3N5) was first identified as a promising photoanode candidate roughly a decade ago. While first studied as a photocatalyst21, it was very rapidly adopted as a thin film photoanode.22, 23 This material has several advantages over more well 24 25 studied photoelectrodes such as TiO2 or Fe2O3. With a band gap of 2.1 eV, this deep red material can absorb a large portion of the solar spectrum.26 In fact, the optical limit (i.e. if all above band gap photons are absorbed and generate current) is 12 mA/cm2.27 Furthermore, the band structure is nearly ideal with both the conduction and valence bands straddling the HER and OER redox potentials, respectively. Electrochemical and spectroscopic techniques which place the conduction band at – 0.52 V vs. RHE and the valence band at 1.58 V vs. RHE are in good agreement.26 DFT calculations suggest that the conduction band is primarily composed of Ta 5d states while N 2p orbitals constitute the top of the valence band.28 The higher energy of the N 2p states compared to O 2p states explains the smaller band gap of Ta3N5 compared to many oxide semiconductors.

Based on the band structure of Ta3N5, photocurrent onset can be expected as early as ~ 0 V vs. RHE. Paired with an appropriate photocathode, such as Si, a tandem system with

Ta3N5 as the photoanode could yield a STH efficiency of over 15% even when practical losses are taken into account.29

There exists a wide variety of synthetic routes for preparing Ta3N5 powders, thin films, or 30-32 nanostructures. These include, but are not limited to, nitridation of Ta2O5 powders ,

5 thermal oxidation and nitridation of Ta foils33, sputtered thin films34, electrochemical anodization and nitridation to form nanotubes35, 36, templated nanorod growth37, vapor- 38 39 phase hydrothermal nanorod growth , drop casting doped films from TaCl5 , atomic layer deposition40, 41, and molten salt synthesis followed by nitridation.42 The wide diversity of preparation conditions also leads to a large spread in the reported photoactivity of these materials. It is very difficult to quantitatively compare any photoelectrode, Ta3N5 or otherwise, based on a single metric; the potential of photocurrent onset, shape of the illuminated cyclic voltammogram, and maximum photocurrent achieved are all important. The highest reported water splitting photocurrent 2 (i.e. oxygen is evolved) for Ta3N5 regardless of applied bias is near 6 mA/cm , still far 2 37 short of the 12 mA/cm optical limit. Efforts to increase the performance of Ta3N5 photoanodes have focused primarily on adding co-catalysts31, 36 or nanostructuring.37, 38, 43 The natural n-type doping in these materials is generally believed to arise from anion defects such as N3- vacancies. However, some groups have attempted to further increase the carrier concentration through doping with alkali metal ions.44 Interestingly, there are essentially no studies which have directly probed the doping density and mobility of carriers in Ta3N5.

Despite its great promise, several studies have pointed to potential charge transport 31, 44 limitations in Ta3N5. The large discrepancy between the maximum predicted photocurrents and the best achieved performance to date motivates studies aimed at gaining a better understanding of Ta3N5 photoanodes in order to identify and overcome material deficiencies.

1.4 Dissertation Overview This dissertation focuses on the development of tantalum nitride photoanodes for solar water splitting. The emphasis is on studying the fundamental material properties such as crystal structure, optical absorption, and charge transport to design a higher performance photoelectrode. This new information is also leveraged to explain trends in the photoelectrochemical activity of supported Ta3N5 thin films.

Chapter 1: Increasing world energy demand motivates the search for a cost effective, carbon-free, renewable source of energy. The fundamentals of photoelectrochemical water splitting are discussed followed by a review of Ta3N5 as a photoanode candidate.

Chapter 2: The work begins in Chapter 1 with a technical and economic feasibility study of four conceptual centralized hydrogen production facilities. Calculations of practical efficiency limits are used to select the solar-to-hydrogen efficiency of particle and panel- based systems. The final cost of hydrogen from each is reported and a sensitivity analysis identifies key research areas for driving costs down. Note that this work summarizes and expands upon a report commissioned by the US Department of Energy.

Chapter 3: Ta3N5 films were first grown by thermal oxidation and nitridation of Ta foils. The photoactivity of the films is found to scale primarily with the electrochemical surface area rather than thickness. The effect of the film morphology and thickness on charge transport of electrons and holes in Ta3N5 is discussed.

Chapter 4: In light of the limitations of the Ta3N5/Ta foil films discussed in Chapter 3, an improved sample architecture is developed. Ta3N5 films grown on a Pt contact supported on fused silica are found to be crack-free and highly uniform. The Pt contact is also omitted to directly probe the optical and electrical properties of Ta3N5.

Chapter 5: A major limitation of Ta3N5 is its susceptibility to photodegradation. Several different catalysts (Pt, RuO2, IrO2, CoTiOx) are explored to stabilize the surface and improve the kinetics of oxygen evolution.

Chapter 6: Results from the literature suggest synthesis conditions strongly influence the performance of Ta3N5. A systematic study of the effect of nitridation temperature on the structural and optical properties is used to show that impurity phases are formed in films grown on Ta foils. Grazing incidence x-ray scattering is used to generate a depth profile of the nitride phases. Mechanisms for formation of Ta-rich impurity phases and bulk defects are proposed.

Chapter 7: Finally, the knowledge gained from the previous fundamental studies is leveraged to design a Ta-Ta3N5 core-shell nanostructure. The synthetic challenges are discussed and a proof-of-principle device is presented.

Appendix A: The selection of an appropriate substrate for catalyst and photoelectrode materials is non-trivial due to the wide variety of requirements (conductivity, transparency, stability, inertness, etc.). The baseline electrochemical testing of common transparent conducting oxide and opaque metallic or carbon-based electrodes is detailed with the aim of guiding substrate selection for electrochemical studies.

Appendix B: The design of a custom compression cell for photoelectrochemical studies as well as the proper technique for calibrating irradiation intensity are described. Additional details are provided for the integrating sphere used to measure optical properties.

Appendix C: Early work focused on the development of a novel MnO2-based photoanode. The band structure was probed via photoelectrochemical and optical measurements and it was found that the conduction band of birnessite-type MnO2 was too low and the band gap too wide for efficient water splitting.

Appendix D: This appendix describes a collaboration with the group of Prof. Alfred Spormann in which the goal was to photoreduce a methyl viologen redox shuttle for use in a bio-electrochemical reactor. The optimization and performance of the p-Si photoelectrode are presented.

Appendix E: A close relative of Ta3N5, TaON was also explored as a photoanode candidate. The challenges involved in synthesizing phase pure materials are examined in the context of determining their suitability for fundamental studies.

1.5 Copyright Portions of this chapter from

B. A. Pinaud, J. D. Benck, L. C. Seitz, A. J. Forman, Z. Chen, T. G. Deutsch, B. D. James, K. N. Baum, G. N. Baum, S. Ardo, H. Wang, E. Miller, and T. F. Jaramillo, “Technical and economic feasibility of centralized facilities for solar hydrogen

8 production via photocatalysis and photoelectrochemistry”, Energy & Environmental Science 6, 1983-2002 (2013).

Reproduced by permission of The Royal Society of Chemistry.

Chapter 2 Technical and Economic Feasibility of Centralized Facilities for Solar Hydrogen Production

2.1 Abstract Photoelectrochemical water splitting is a promising route for the renewable production of hydrogen fuel. This chapter presents the results of a technical and economic feasibility analysis conducted for four hypothetical, centralized, large-scale hydrogen production plants based on this technology. The four reactor types considered were a single bed particle suspension system, a dual bed particle suspension system, a fixed panel array, and a tracking concentrator array. The current performance of semiconductor absorbers and electrocatalysts were considered to compute reasonable solar-to-hydrogen conversion efficiencies for each of the four systems. The U.S. Department of Energy H2A model was employed to calculate the levelized cost of hydrogen output at the plant gate at 300 psi for a 10 tonne per day production scale. All capital expenditures and operating costs for the reactors and auxiliaries (compressors, control systems, etc.) were considered. The final cost varied from $1.60 - $10.40/kg H2 with the particle bed systems having lower costs than the panel-based systems. However, safety concerns due to the cogeneration of

O2 and H2 in a single bed system and long molecular transport lengths in the dual bed system lead to greater uncertainty in their operation. A sensitivity analysis revealed that improvement in the solar-to-hydrogen efficiency of the panel-based systems could substantially drive down their costs. A key finding is that the production costs are consistent with the Department of Energy’s targeted threshold cost of $2.00 - $4.00/kg H2 for dispensed hydrogen, demonstrating that photoelectrochemical water splitting could be a viable route for hydrogen production in the future if material performance targets can be met.

2.2 Introduction As global energy consumption continues to rise, it is imperative that we develop renewable alternatives to the fossil fuel energy sources that currently power our civilization, curb CO2 emissions, and secure a permanent energy supply for the future. Although the solutions to these global challenges are likely to consist of many different energy storage and conversion technologies, sustainably produced chemical fuels will likely play an important role due to their high energy density. Hydrogen gas is an especially promising energy carrier, but current hydrogen production processes such as steam methane reforming are unsustainable. Photoelectrochemical (PEC) water splitting is an alternative process that enables sustainable hydrogen production from water using the energy from sunlight. PEC water splitting has been demonstrated on the laboratory scale, but it has never been implemented on a large scale relevant to the global energy demand, so the prospects for scaling up this process have remained controversial. Fundamentally, economics are the driving force in our energy landscape so there is one key question which all researchers in the field should be asking: If the technical barriers to implementation of photoelectrochemical water splitting on a large scale are overcome, can hydrogen be produced at a cost which is competitive with that of fossil fuels?

The United States Department of Energy (DOE) contracted Directed Technologies Inc. (DTI) (now Strategic Analysis Inc.) to carry out a detailed technoeconomic evaluation of PEC hydrogen generation based on conceptual systems formulated by the DOE PEC Working Group.45 This team brings together university researchers, scientists from national laboratories, and industry leaders with experience in PEC water splitting systems. The major findings of their cost analysis are presented here in conjunction with an evaluation of the technical feasibility of the assumptions pertaining to material properties and system efficiencies. The practical operating efficiencies for PEC reactors are established, four potential reactor and centralized plant designs are detailed, and the projected cost of the hydrogen produced using each design is discussed. PEC hydrogen production has a very low technology readiness level (TRL 1 – 2)46 but there is a need for an objective, unbiased technoeconomic analysis to determine where research dollars are best spent to lead to commercially viable solutions. This study emphasizes large-scale

12 operations based on realistic material performance targets to calculate a cost for H2. Results of this study place the levelized cost of hydrogen for these systems between

$1.60 - $10.40/kg H2, indicating that commercial-scale PEC water splitting could be cost- competitive with fossil-based fuels. To help guide continuing research in this field, key challenges that must be overcome to drive down the cost of large-scale hydrogen production by PEC water splitting are identified.

2.3 Theoretical and Demonstrated Water Splitting Efficiencies

2.3.1 Calculation of Practical System Efficiencies

The performance of PEC water splitting devices is best quantified by their solar-to- hydrogen efficiency, which is defined as the amount of chemical energy produced in the form of hydrogen divided by the solar energy input without the use of any external bias. There is also the additional requirement that the other coupled half-reaction must specifically be oxygen evolution in order to maintain a sustainable overall reaction in which sunlight and water are the only inputs.27 STH efficiency is a metric by which device performances can be quantifiably compared on an equivalent basis, which is not possible with the inclusion of unsustainable inputs such as sacrificial reagents or with incomplete half-cell designs that drive only one of the two half-reactions (HER or OER).

In designing the conceptual water splitting systems described later in this work, it was very important to select realistic efficiencies due to the impact on cost. Reasonable STH efficiencies for various PEC water splitting device configurations were calculated by taking into account total solar irradiance, entropic losses due to blackbody radiation and recombination, and kinetic overpotentials needed to drive the two half-reactions. A thermodynamic minimum of 1.23 V is required to split water, but in practice the voltage required to drive this reaction increases due to these unavoidable losses. The calculations and corresponding assumptions included here both review and expand upon past work47- 49 developed to determine upper limits of achievable efficiencies for the best available materials at this time. Other losses such as those due to nonideal band edge alignment and series resistances from the solution or wiring can further decrease the amount of usable voltage, but a full analysis of these effects is beyond the current scope of this work. One

13 of the seminal derivations of solar conversion efficiencies was developed by Shockley and Queisser50 and later expanded upon by Ross51, though the work was framed in the context of photovoltaic devices and thus did not aim to include losses unique to PEC systems such as the kinetic overpotentials required to drive electrochemical reactions. Weber and Dignam47, 48, Miller and Rocheleau52, as well as Bolton et al.53 further elaborated on solar conversion efficiencies specifically addressing PEC systems. More recent work has focused on calculating efficiencies for tandem systems with various geometries.54 Prior to introducing the PEC-specific losses, it is instructive to consider the thermodynamic limits as an upper bound. Previous calculations have shown that a single absorber PEC system can reach 29%55 - 31%49 while a tandem system with two absorbers could reach 40%49 - 41%55. Taking into account multiple exciton generation or solar concentration raises these numbers further.49 Results presented in this work are produced using the Air Mass 1.5 Global (AM 1.5 G) spectrum (ASTM G173-03) and a few updated assumptions, described briefly below to reasonably reflect the current state of technology for PEC materials. A more detailed description of the calculations will be made available elsewhere.29

Equation 1 is one definition for STH efficiency, using the product of voltage and short- circuit current to calculate the chemical power output of the PEC water splitting cell under standard-state conditions relative to the power input to the cell by 1 sun AM 1.5 G illumination, assuming 100% Faradaic conversion of water to H2 and O2.

| ( )| ( ) [ ] (1) ( )

The first step taken to calculate reasonably achievable STH efficiency values was deriving the maximum photocurrent under illumination for a given band gap by integrating the AM 1.5 G spectrum. Here, there is no applied bias and ideal band edge alignment is assumed. Kinetic overpotentials as well as energy and entropic losses arising from material defects and nonradiative recombination were then calculated and subtracted to produce current-voltage relationships for each band gap from which the STH efficiency value was extracted. The maximum photocurrents were calculated by summing the absorbable photons over the solar spectrum for materials of varying band

14 gaps assuming all photons with energy greater than the band gap are absorbed. The open- circuit voltages, used to estimate the useable photovoltage, were calculated for each band gap using the procedure outlined by Ross51 and taking into account entropic losses modeled after single crystal silicon. Kinetic overpotentials were calculated for the range of possible currents drawn from the system assuming Butler-Volmer kinetics fitted to the hydrogen evolution and oxygen evolution activities of platinum and ruthenium oxide, respectively.56, 57 Shunt losses were neglected in this treatment because their value is largely device dependent and not an intrinsic material property.

Results of these calculations are shown here for three different device configurations that reflect those employed in several reactor designs that will be presented later. The three configurations are a single photoabsorber system, a dual stacked photoabsorber system, and a dual side-by-side photoabsorber system. These configurations differ in the number and geometry of photoabsorber materials used which has a direct impact on the spectrum and number of photons absorbed in each material and thus the maximum possible STH efficiency. Figure 2.1 shows the maximum practical theoretical limits for a single absorber system where the maximum STH efficiency is 11.2% for a band gap of 2.26 eV.

Figure 2.1: Maximum theoretical solar-to-hydrogen efficiency for a single absorber material plotted as a function of the photoabsorber band gap.

This calculation assumes that an area equal to that of the photoabsorber was available for each water splitting half-reaction; this restriction is not always necessary as a higher area electrode driving a reaction in the dark could be orthogonalized. Note that this value of 11.2% falls far short of the 31% thermodynamic limit, highlighting the large losses associated with reaction overpotentials and the need for better catalysts. Figure 2.2 and Figure 2.3 show the maximum practical STH efficiencies for dual stacked and side-by- side photoabsorber systems, respectively, as a function of the two photoabsorber band gaps.

Figure 2.2: Maximum theoretical solar-to-hydrogen efficiency for a dual stacked absorber configuration plotted as a function of the top and bottom photoabsorber band gaps. The top photoabsorber is assumed to be placed above the bottom photoabsorber, thus only photons with energy less than the band gap of the former make it to the latter. The maximum STH efficiency for a dual stacked absorber system is 22.8% with bottom and top photoabsorber band gaps of 1.23 eV and 1.84 eV, respectively. Once again, this value is considerably lower than the thermodynamic limit of 41%. In comparison, the maximum STH efficiency is 15.5% for a dual side-by-side absorber system with photoabsorbers of the same band gap of 1.59 eV. These values provide a baseline from which reasonable device efficiencies can be projected for materials within this range of band gaps.

Figure 2.3: Maximum theoretical solar-to-hydrogen efficiency for a dual side-by-side absorber configuration plotted as a function of the two photoabsorber (denoted ‘A’ and ‘B’) band gaps. The two electrodes are assumed to be placed next to each other and can each access the full solar spectrum. A very important finding is that to achieve efficiencies > 10%, dual absorber systems are likely required. Note that the dual stacked absorber system itself can be configured in more than one way, for instance as either a stacked planar thin film reactor or a particle with two absorber materials. Particles can also be made with only one absorber material, represented by the single absorber system, but that would result in lower attainable efficiencies compared to the dual absorber design. Lastly, the dual side-by-side absorber system is representative of a reactor that uses the added photovoltage of two absorbers, covering twice the area, to achieve the required water splitting voltage. This set-up has some advantages over a single absorber system in that less photovoltage is required from each absorber, allowing for use of smaller band gap materials, but it is still not as efficient as a dual stacked absorber system.

2.3.2 Demonstrated Research Efficiencies

The theoretical calculations above provide realistic limits on PEC water splitting efficiencies based on the performance of modern materials. It is also instructive to consider the best bench scale systems reported in the literature. Bearing both sets of values in mind, efficiencies around which the conceptual solar hydrogen plants will be

17 designed can be selected. Several excellent materials reviews15, 16, 58 cover a wide range of published systems while the focus here is on only the highest reported efficiencies for both electrode and particulate based systems. Demonstrated panel electrode solar-to- hydrogen systems generally fit into one of three categories. First are the pure photoelectrochemical systems that are minority carrier devices and have a single or dual PEC junction. The second type is a single PEC junction coupled with either an integrated or external photovoltaic (PV) device. The third category does not contain semiconductor/electrolyte (PEC) junctions and instead involves devices that have separated components, namely a photovoltaic joined with an electrolyzer, potentially in an integrated structure that is immersed into the aqueous electrolyte.

The voltage requirement for solar water splitting using a single absorber in a pure PEC system necessitates the use of a reasonably wide band gap (> 2.1 eV for an STH efficiency > 5%) semiconductor. Many known materials with band edge potentials that encompass both half-reactions have a valence band edge that is significantly more 59 positive than the water oxidation potential as is the case for the oxides SrTiO3 and 60 KTaO4 . Recently a non-oxide, GaN, has demonstrated spontaneous (unbiased) water splitting.61, 62 All known single junction systems capable of full water splitting absorb only ultraviolet photons, severely limiting their attainable efficiencies in terrestrial PEC systems. Dual PEC electrodes have been demonstrated that are capable of unbiased solar water splitting. Efficiencies are still low when a wide band gap photoanode is employed63 but up to 8% has been measured when two lower band gap III-V materials were coupled together64. This p-InP/n-GaAs system experienced a 10% drop in relative efficiency over the first hour but maintained constant short-circuit photocurrent over the next nine hours of operation that were interrupted by extended periods where the electrodes sat in solution in the absence of illumination.

The second category of devices uses a photovoltaic to overcome energetic barriers at a

PEC interface. One particular device utilizes a p-GaInP2 (photocathode) that makes a PEC junction with the electrolyte and a dark anode. It is synthesized on a p/n-GaAs photovoltaic and connected via a solid-state tunnel junction to yield a single crystal monolithic device. This system achieved 12.4% unbiased solar-to-hydrogen efficiency

18 under 12 suns of illumination.8 However, over the course of 20 hours the photocurrent declined from 120 mA/cm2 to 105 mA/cm2. Other PEC/PV systems have combined a metal oxide (photoanode) with a separate photovoltaic, typically a dye-sensitized solar cell65-67 and achieved modest efficiencies (2 - 3%).68 Although the efficiencies were lower, the allure of metal oxide-based systems is that they could be less costly and potentially more stable as photoanodes, though the durability of these systems has not been reported. A WO3 photoanode mechanically stacked on an a-Ge/a-Si tandem PV achieved 0.6% efficiency under Hawaiian sunlight, but showed signs of degradation after 10 hours of operation.69 A hybrid photocathode based on a multijunction a-Si PV capped with an a-SiC PEC layer has demonstrated a 1.6% STH efficiency at zero bias, but the short-circuit photocurrent declined over a short period of testing.70

The last category of solar water splitting cells is majority carrier devices that use coupled PV and electrolysis components, where the PV device is either immersed in the electrolyte or separated. In the case of the immersed PV, it is protected from coupling with the electrolyte by an ohmic contact and catalyst, so the electric field that separates photogenerated charges occurs in a buried solid-state p/n junction, not at a semiconductor/electrolyte junction. III-V-based systems have achieved efficiencies upwards of 18%.71-73 In the highest reported efficiency systems, the electrode areas responsible for driving the water splitting reactions were much larger than the semiconductor light-absorbing component, which further reduced overpotential losses since those losses are directly related to current density. PEC systems that employ light concentration might be able to accommodate a counter electrode area up to an order of magnitude greater than the absorber. Low-cost systems based on multi-junction a-Si have shown efficiencies up to about 8%.10, 11, 72 The PV/electrolysis devices address the instability of a semiconductor/electrolyte interface by eliminating this type of junction but these systems should be compared with completely separated commercial PV and electrolyzers that are independently optimized. Given the significant differences in plant design for this case versus the others, a technoeconomic analysis for this particular scheme was outside the scope of this work. Overall, it is clear that STH efficiencies of approximately 15% are achievable for panel-based systems, though durability and cost questions remain.

Consider now particle based PEC systems found in the literature which generally fall into one of the three following categories: i) single particle/single photon water splitting, ii) two particle/two photon water splitting, and iii) half-reaction water splitting plus sacrificial reagent and/or nonwater splitting photocatalysts. The efficiencies of these systems are considerably lower than panel devices and are often not directly comparable; literature in this area often reports the external quantum yield (EQY) at a given wavelength rather than the STH efficiency. Here, the definition of the EQY is usually (but not always) reported as shown in Equation 2:

(2)

where n is the number of electrons transferred per product molecule. Note that there is no accounting for the voltage of a particular reaction or product and thus, overall energy efficiency is not accounted for. The reader is cautioned to make the distinction between the EQY which is more or less analogous to a diagnostic current conversion efficiency and an STH efficiency which is a true power conversion efficiency.

The first category consists of materials capable of driving both the HER and the OER directly with a single absorber particle and without the need for any additional redox reactions. These systems directly drive true water splitting. Barring the employment of multiple exciton generation schemes in these particle systems, quantum efficiencies of photogenerated products are based on absorption of a single photon. The highest EQYs 74 are obtained by using the wider band gap systems, where EQY > 50% with Eg = 4.1 eV. Unfortunately, these and other UV–based systems provide little utility under terrestrial solar insolation due to limited absorption and therefore more recent efforts have been devoted toward developing smaller band gap materials. Domen and co-workers have demonstrated true water splitting with an EQY of 2.5% at 420 nm using a Rh2-yCryO3- 75 loaded (Ga1-xZnx)(N1-xOx), and continue to refine their system.

The second category, often referred to as a ‘Z-scheme’ system, also drives stoichiometric water splitting but utilizes two separate absorber particles, tuned to drive either the HER or the OER individually. The two half-reactions occurring on separate particles are connected via a reversible, charge transfer redox couple in solution. This system requires

20 twice as many photons to drive water splitting, but offers a higher voltage and broader range of materials choices. Abe and co-workers report true water splitting with a Z- - - scheme using Pt-SrTiO3:Cr/Ta and Pt-WO3 particles with an IO3 /I redox mediator. 76 Stoichiometric H2 and O2 were evolved with an EQY of ~ 1% at 420 nm. Kudo et al. 3+/2+ used a similar Pt-SrTiO3:Rh particle but coupled it with BiVO4 particles and an Fe redox mediator and reported an EQY of 0.3% at 440 nm. Here, both particles had a band gap of 2.4 eV. Fujihara devised a system using separate compartments loaded with TiO2 - particles, one side driving the HER and the oxidation of Br to Br2, the other compartment driving the OER with concurrent reduction of Fe3+ to Fe2+ to yield overall water splitting.77

The third category described in the literature encompasses a very broad range of particulate PEC design types but differs in that true, stoichiometric water splitting via the HER and the OER is not driven. Fundamentally, the optoelectronics can be the same as the first and/or second categories described above, but the significant difference is found in the thermodynamics of the redox reactions which are catalyzed. In this category, the free energy (ΔG) of reaction differs from that required for true water splitting. In practice, the ΔG of the reactions chosen is typically smaller than that required for water splitting74, 78 or, in some reported PEC systems, negative and therefore thermodynamically favored.74 These systems may find application in photocatalytic decomposition of organic pollutants to clean air or water79 and could offer fundamental mechanistic insight into the dynamics of complex PEC systems, but they will not be discussed here as the primary goal of this work is to address the feasibility of converting and storing photon energy in the chemical bonds of H2 and O2.

2.4 Conceptual Reactor Designs The centralized plant designs based on four reactor types, named Type 1, 2, 3, and 4, were conceived by the DOE PEC Working Group. These four configurations do not represent all possible types of systems but are meant to represent a range of complete, albeit basic, designs of potentially realizable systems for the purpose of preliminary technical and economic evaluation. Each system incorporates all components required to absorb solar photons, generate adequate voltage to evolve H2 and O2 from water, and

21 collect and compress only the H2 product. The reactors can be grouped into one of two general classes, namely particle suspensions or planar arrays. The Type 1 and Type 2 systems are enclosed aqueous reactor beds of suspended photoactive particles while the Type 3 and Type 4 systems consist of multilayer absorber planar arrays immersed in an aqueous electrolyte and oriented toward the sun. This section describes the four reactor configurations and outlines key assumptions about their performance in light of the calculations and state of the art described previously.

2.4.1 Type 1 Reactor: Single Bed Particle Suspension

The Type 1 reactor is the simplest of the four and consists of a low-lying horizontal plastic bag, termed henceforth a ‘baggie’, containing a slurry of photoactive particles in a 0.1M potassium hydroxide (KOH) electrolyte. The plastic baggie is designed to retain the electrolyte, photoactive particles, and evolved gases while allowing light to penetrate. A schematic of the design for a Type 1 reactor is shown in Figure 2.4(a). High density polyethylene (HDPE) is selected for both the transparent upper layer and opaque bottom liner due to its high optical transmission (90%), low hydrogen permeability (156 cm3·mm/m2·atm·day)80, resistance to degradation in an alkaline electrolyte, and low cost. The size of each baggie is 323.1 m long by 12.2 m wide. These dimensions were chosen because they are as large as possible while enabling the baggie segments to be produced as single sheets of plastic using existing manufacturing technology and easily transported on large rolls using a standard 16.8 m truck. Two plastic sheets are laminated together to construct the baggie. Upon filling the baggie with the aqueous particle slurry, the upper portion of the envelope initially rests on the liquid surface. This upper portion will rise and fall as gas is generated and drawn off. Gas accumulates in the extra volume and is stored in the headspace during the daylight H2 production phase to level the load on the purification and compression equipment required for this design. Since the baggie is completely sealed, water vapor is not vented to the atmosphere. Preventing evaporative cooling in this way allows the beds to operate at temperatures above 60°C in summer which slightly reduces the thermodynamic voltage for water splitting but this effect may be offset by a decrease in efficiency due to greater entropic losses in the absorber.

Figure 2.4: Schematic of the four reactor types including (a) Type 1 reactor cross-section showing the particle slurry contained within baggies separated by an access driveway, (b) Type 2 reactor cross-section showing the particle slurries contained within baggie assemblies consisting of an alternating arrangement of a full size and half-size baggie each for O2 and H2 evolution, (c) Type 3 reactor design showing the encased composite panel oriented toward the sun with buoyant separation of gases, and (d) Type 4 reactor design with an offset parabolic cylinder receiver concentrating light on a linear PEC cell. Drawing not to scale.

The photoactive particles are modeled as conductive spherical cores (40 nm diameter) coated with photoanodic and photocathodic materials (~ 5 nm layers) as islands, particles, or thin film shells. The primary motivation for selecting this particle design is to consider the cost of a relatively complex composition and morphology. One could also envision a photoactive particle at the core coated with only one additional absorber layer. The particle geometry could also take the form of a photoabsorbing core with co-catalysts for both the HER and the OER. In this case, H2 and O2 evolution occur simultaneously on the surface of a single particle which produces a mixed gas stream which must be purified. The required unit operations for this purification step are described later. The cogeneration of H2 and O2 in stoichiometric quantities results in a combustible mixture and engineering controls will be needed to mitigate this serious safety risk. Preliminary analysis conducted at the University of California at Santa Barbara suggests a 10 cm deep particle suspension is appropriate for full light absorption when considering light scattering and typical values for semiconductors with indirect band gaps or low absorption coefficients. The layered structure of the photocatalyst particles is akin to a dual stacked configuration which has a calculated maximum theoretical efficiency of 22.8%, assuming high quality semiconductors and highly active catalysts, but so far experimentally demonstrated particle systems have fallen far short of this limit. Thus, a more conservative baseline STH efficiency of 10% was used for the technoeconomic analysis.

2.4.2 Type 2 Reactor: Dual Bed Particle Suspension

The Type 2 reactor (Figure 2.4(b)) shares many characteristics of the Type 1 reactor. The primary difference is the use of separate beds for O2 and H2 production which affords two general benefits: (i) intrinsic separation of the two gases, which improves plant safety while reducing needs for gas separation in the processing subassembly, and (ii) greater flexibility in the semiconductor properties needed for effective water splitting. This design, however, requires the use of a redox mediator (A/A-) and porous bridges to transport it from one compartment to the other. The equations for water splitting now become as follows:

+ - O2 Evolution Bed: 4hν + 2H2O + 4A → O2 + 4H + 4A + - H2 Evolution Bed: 4hν + 4H + 4A → 2H2 + 4A

The redox mediator could be any species exhibiting rapid, reversible reactivity and large diffusivity in either redox state, such as iodine, bromine, or iron complexes. The absorber in this system is modeled as spherical substrate particles (40 nm diameter) with a single photoactive layer deposited over the surface. Once again, this geometry is selected for costing purposes but many other viable options exist such as a photoactive core decorated with a co-catalyst. Continuous slurry circulation through perforated pipes running the length of the baggies facilitates mixing and transport of the redox mediator. Feed-through bridges located below the gas/liquid interface are provided between the H2 and O2 production beds with a porous membrane running the entire length of the baggies. These allow transport of the redox mediator but prevent gas migration. Each bed assembly comprises a full and a half-size baggie each for O2 and H2 evolution, as shown in Figure 2.4(b), to minimize the distance between complementary reaction sites. The total assembly dimensions are 61.0 m long by 6.1 m wide; the decrease in width from the Type 1 baggies further reduces transport distances and the decrease in length facilitates attaching porous bridges. Decoupling of the HER and the OER onto separate particles requires twice the photon capture area of the Type 1 system and is analogous to the side- by-side configuration for which the upper efficiency limit computed was 15.5%. Again, a more conservative value is selected for the baseline technoeconomic analysis; the assumed net STH efficiency for this system is 5%. Better solar utilization and thus a higher water splitting efficiency could be achieved by stacking the baggies one on top of the other rather than positioning them side-by-side. While potentially interesting, this embodiment was not considered in this analysis.

2.4.3 Type 3 Reactor: Fixed Panel Array

A great deal of experimental research to date has focused on planar photoelectrodes immersed in solution, either with an integrated cathode and anode or with spatially separated electrodes. The Type 3 and 4 reactors reflect this design archetype with photocells closely resembling commercial photovoltaics. The Type 3 reactors (Figure 2.4(c)) feature an integral planar electrode with multiple photoactive layers sandwiched

25 between two electrodes. The entire assembly is encased in a transparent plastic electrolyte reservoir containing 0.1M KOH. Poly(methyl methacrylate) (PMMA) is selected for its high optical transmission, mechanical strength, and resistance to chemical degradation in basic electrolyte and serves as a good starting point for costing. Individual cells, as large as can be readily manufactured, are assembled to form panels 1 m wide and 2 m long. The panels are tilted toward the Earth’s equator and are oriented at an angle from the horizontal equal to the local latitude for optimal solar photon collection over the course of the entire year. Two layers of photoactive material are used to maximize solar spectrum utilization and provide the requisite voltage to split water. O2 evolution occurs on a transparent conducting (TC) anode which allows photons to pass through and be absorbed in the two underlying photoactive layers. The bottom of the stack is composed of a metal cathode where electrons are collected to drive H2 evolution. Gas separation is achieved by the physical partitioning of the O2 and H2 reaction sites and separate buoyant collection of the gases can be exploited due to the inclination of the panel. A depiction of the cell stack composition and orientation is shown in Figure 2.4(c). In terms of maximum theoretical efficiencies, this system closely resembles the stacked configuration of Type 1 so a conservative STH efficiency of 10% is used for the baseline cost calculations.

2.4.4 Type 4 Reactor: Tracking Concentrator Array

A tracking concentrator system maximizes the direct radiation capture and enables the use of higher efficiency, higher cost materials because the photocell area is greatly reduced. The Type 4 reactor (Figure 2.4d)) uses an offset parabolic cylinder array to focus sunlight on a linear PEC cell receiver and has 2-axis steering to track the daily movement of the sun. This style of concentrator was selected to keep costs low, reduce weight, and because it allows the photoreactor, water feed, and H2 collection piping to be located in the reflector base assembly. Each concentrator array is 6 m wide and 3 m in height. A laboratory-demonstrated solar concentration ratio of 10:1 is used for the analysis, though higher ratios would further reduce reactor costs. Multijunction photovoltaics can operate at solar concentration ratios upwards of 400:181 but PEC cells are limited by the ability of a catalyst to drive current densities above 1 A/cm2, light

26 scattering due to bubble formation, and temperature constraints for moderate cost materials. A ratio of 100:1 is likely an upper limit8 and a more conservative value of 10:1 is adopted for this analysis. The smaller volume electrolyte reservoir allows direct pressurization of the electrolyte and gases to 300 psi using the inlet water pump, precluding the need for a separate compressor for the H2 gas product. Added benefits of pressurization include minimizing water vapor loss and reducing gas bubble size and the associated detrimental photon scattering. The PMMA window is made cylindrical to reduce stress arising from the increased operation pressure and to focus light on the PEC cells. An additional benefit of the 10:1 concentration is an increased efficiency due to the larger maximum photovoltage. The photocell stacks have the same general composition as in the Type 3 reactor but allow for more expensive, high quality materials to be used since the photon capture area is reduced. Higher quality materials will likely result in better performance so the assumed STH efficiency is raised to 15%, consistent with calculated efficiencies for a dual stacked absorber system (Figure 2.2) and well within the range of previously reported high efficiency PV-PEC8 and integrated PV-electrolysis71, 72 devices that operate between 12 - 18% STH efficiency.

2.4.5 Reactor Design Limitations

Each reactor design described above has its limitations. The Type 1 and Type 2 reactors are relatively simple. However, compared to the planar electrode designs there is greater uncertainty associated with the fabrication of the photoactive particles as well as much lower demonstrated bench-scale efficiencies. There is also a lack of understanding of the effective photon capture area per particle and particle density required for total solar flux utilization. Partial shading of particles deeper in the bed is also problematic as it will result in a lower effective incident light intensity reaching these deep particles and thus lower performance due to a decrease in their maximum photovoltage. Further testing and modeling of these systems is required. The large size of the Type 1 baggies may present practical limitations in that a failure in the mechanical integrity (due to weather, bird damage, etc.) of a single baggie would result in a significant release of electrolyte and plant capacity loss. Mitigating this concern by decreasing the size of the baggies, which would require a greater number of baggies, is likely to increase the costs associated with

27 this system. The Type 2 system has additional limitations in that mediator transport rates across the bridges and associated voltage losses are uncertain. More bridges may be needed to overcome these challenges, which would increase system costs. It is also not trivial to guarantee the presence of solely O2 or H2 in each baggie if there is undesired gas transport across the bridges. Also, the simultaneous evolution of both gases on the particles in a single baggie could be a problem if the band positions of the absorbers and catalyst selectivities are not ideal.

In contrast to the particulate systems, the Type 3 and Type 4 systems are considerably more complex. The higher current densities drawn on the much smaller photocell of the Type 4 system will require high performing HER and OER catalysts. This more stringent catalyst requirement could be mitigated by increasing the surface area for catalysis while preserving the smaller absorber area; however, decoupling of the absorber and catalyst presents additional design challenges and costs. Given the large scale of all the reactors, there may be a significant voltage loss due to the voltage drop across the solution.82 The size of the baggies for particulate systems may need to be reduced or additional transport channels introduced in planar electrodes to minimize these losses.

Note that the planar array PEC reactors described here can just as easily be applied to integrated PV-electrolysis units. Both of these options have advantages over a system composed of a commercially available photovoltaic cell connected to a conventional water electrolyzer. The costs associated with the contacts for current collection, charge conditioning and current transmission losses in the latter are eliminated. Photon capture over a large area in the case of the PEC reactors or PV-electrolysis units also relaxes the catalyst requirements, enabling the use of inexpensive nonprecious metal catalysts. The particles or electrodes have large areas where current densities on the order of mA/cm2 rather than A/cm2 are drawn, requiring a lower voltage to drive the OER and the HER than in a conventional electrolyzer. In addition, unknown complications (e.g. product crossover) may arise in electrolyzers operating at these decreased current densities because electrolyzers are traditionally operated at larger current densities. Nevertheless, there exists the opportunity to design improved nonintegrated PV-driven electrolysis. As solar absorption and catalysis can be decoupled with such a system, semiconductors

28 unstable in aqueous environments could be considered. As the analysis presented in this work is limited to systems in which the semiconductor is directly immersed in the electrolyte, the technoeconomics of PV-driven electrolysis is outside of the scope. A similar analysis for such an approach would be welcomed by the community.

2.5 Plant Design and Operation Complete PEC hydrogen production plants were designed based on the dimensions, operating conditions, and performance characteristics of the four reactor types. To enable accurate cost analysis, these plant designs incorporated all the unit operations necessary to deliver a purified and compressed hydrogen gas product. The plant designs include the reactor layout and spacing, gas processing components, control systems, and support piping and wiring. Each plant module consists of reactor arrays and gas separation/compression equipment sized to deliver 1 tonne per day (TPD) of H2 at 300 psi (20.4 atm). The pipeline pressure was selected to be directly comparable with other

DOE H2A Production Plants. Note that in the Type 1 system where there are H2 losses associated with the gas separation process, the reactor arrays are sized to deliver a net production of 1 TPD H2 after separation. For the purposes of the cost analysis, an entire

H2 production plant will consist of 10 of such systems for a total net production of 10 TPD.

2.5.1 Solar Insolation

An accurate measure of photon flux is necessary to size the plants and determine the reactor layouts. A hypothetical plant site of Daggett, CA, USA, at 35° North latitude, was assumed for the purposes of determining the insolation. This location was chosen because it is the most favorable of the 239 National Solar Radiation Database sites due to its high terrestrial insolation and minimal cloud cover. The solar insolation model considered both direct and diffuse radiation as a function of time of the day and day of the year. The different reactor types utilize these sources of radiation with different efficiencies. The NREL SOLPOS model was used to calculate the extraterrestrial radiation (ETR) intensity and solar position. The clearness index, defined as the average loss due to atmospheric absorption, scattering, and cloud cover, was derived from the NREL Solar Radiation Data

Manual. To determine the direct insolation intensity, the ETR was multiplied by the clearness index. The total insolation on the Type 1, Type 2, and Type 3 systems was the sum of the direct and diffuse components whereas the Type 4 system used only the direct solar component. The amount of incident radiation that is captured depends on the reactor design, specifically the orientation and type of covering (HDPE vs. PMMA). Figure 2.5 shows the average monthly insolation calculated for the different reactor types.

Figure 2.5: Average monthly refracted insolation for each reactor type. Refracted insolation refers to the light intensity incident upon the photoabsorbers, as opposed to the incident intensity (not shown), which is incident upon the HDPE or PMMA reactor covering. The insolation and the STH efficiency assumptions outlined previously were used to determine the reactor array sizes required to achieve an average annual production of 1

TPD H2. All reactors and plants were designed to accommodate variations in the hydrogen production rate due to changes in irradiance over the course of a day and throughout the year. However, the daily output of hydrogen varies substantially over a year, which may not match market demand for the hydrogen product. A summary of the pertinent parameters for each system design and plant layout is shown in Table 2.1 while additional details follow below.

2.5.2 Type 1 Plant Design: Single Bed Particle Suspension

A horizontal plate field geometry with a solar input including direct and diffuse radiation was used to model the suspension bed and the average yearly insolation calculated for this reactor type was 5.77 kWh/m2/day.

323 m x 12.2 m each are required for a net production of 1 TPD H2. The layout of the plant was designed to minimize the spacing between the baggies for optimal land use. Some space was left between the reactors for maintenance access and a total of 30% additional land area was allocated for auxiliaries. Because these reactors are very wide and short, solar shadowing was not a consideration. A top-view diagram illustrating the plant layout is shown in Figure 2.6(a). The reactors are integrated with the gas processing subassembly via ports in each HDPE baggie that allow for water input and gas output. This embodiment is the only one of the four selected for evaluation which results in a combustive mixture of stoichiometric H2 and O2. Compression of this explosive mixture is considered a very important design concern but not an insurmountable problem.

2.5.3 Type 2 Plant Design: Dual Bed Particle Suspension

The Type 2 plant design is very similar to the Type 1 plant design. Pure O2 and H2 are now produced in separate compartments, so the need for gas separation equipment is obviated. This configuration also significantly reduces the risks associated with inadvertent H2/O2 gas combustion. The same set of assumptions was used to calculate the average yearly insolation, 5.77 kWh/m2/day. Accounting for the STH efficiency of 5%,

347 assemblies of 61.0 m x 6.1 m each are required to produce 1 TPD H2. The layout of this plant is also designed to maximize land usage while allowing some space between the assemblies for maintenance access and auxiliaries, estimated to be 30% additional area. As with the previous reactor type, solar shadowing is not a concern.

A diagram illustrating the plant layout is shown in Figure 2.6(b). Once again, water input and gas output are accomplished by ports in the HDPE baggies.

2.5.4 Type 3 Plant Design: Fixed Panel Array

As with the Type 2 system, because the O2 and H2 are produced in separate compartments, no gas separation process is required and the safety concerns associated with inadvertent gas combustion are minimized. To maximize the solar radiation captured, the fixed planar panels are tilted toward the South at an angle equal to the plant latitude. Using a 35° tilted array geometry for Daggett, California and assuming a solar input including direct and some diffuse radiation, the average yearly insolation calculated for this reactor type is 6.19 kWh/m2/day. With an STH efficiency of 10%, a total of 26,923 panels each 1 m wide and 2 m long are required to achieve a target production of

1 TPD H2. The layout of this plant is designed to minimize the effects of solar shadowing by adjacent panels. The North-South panel spacing is set such that there is no shadowing at sun angles greater than 10° above the horizon. Based on the panel dimensions, this requires a separation distance of 8.1 m, resulting in an emplacement area ratio of 4.07 m2 land per m2 panel. A diagram illustrating the plant layout is shown in Figure 2.6(c). The water input and gas outputs are connected to manifolds that lead to large central collection pipes.

2.5.5 Type 4 Plant Design: Tracking Concentrator Array

In this system, the water input is pressurized and the O2 and H2 are produced in separate compartments. Therefore, neither gas separation nor compression is required. The concentrator array angle in the Type 4 system is controlled so that the panels point directly toward the sun throughout the day. This tracking system captures the direct radiation very effectively, but due to the shape of the concentrators, collection of diffuse radiation is minimal. These assumptions lead to an average yearly insolation of 6.55 kWh/m2/day for this reactor type. With an STH efficiency of 15%, a total of 1,885 concentrator arrays each 6 m wide and 3 m in height are required to produce 1 TPD H2. The spacing between the concentrator arrays was again determined to minimize the effects of shading by adjacent concentrators. In this case, because the arrays track the

33 sun, the spacing along both the North-South and East-West axes was considered. The arrays are spaced such that there is no shadowing when the sun is more than 10° above the horizon in the East-West direction or 26° above the horizon to the South. The necessary spacing was 6.71 m in the North-South direction and 17.3 m in the East-West direction, resulting in an emplacement area ratio of 6.57 m2 of land per m2 concentrator. A diagram illustrating the plant layout is shown in Figure 2.6(d). The water input and gas outputs are connected via manifolds to central collection pipes.

2.5.6 Gas Processing and Control System Subassemblies

In addition to the reactors, the plants also include subassemblies for separating, purifying, and compressing the product gas streams and controlling the reactor operation. The basic components are similar for Types 1 – 4, but due to the specifics of each design, not all elements are required for each reactor type. These subassemblies are supported by extensive piping and wiring. All piping in the designs consists of polyvinyl chloride to minimize costs. The effects of hydrogen embrittlement and gas diffusion are assumed to be negligible given the moderate temperature and pressure operating conditions.

The hydrogen delivered from each plant is purified and compressed to 300 psi. The gas processing subassembly, which includes compression, separation, and purification unit operations, is used to condition the hydrogen. Prior to compression, the gas stream for the Type 1, Type 2, and Type 3 systems is cooled to 40°C with a cooler/condenser to remove water vapor and reduce the volumetric flow to the compressor. A two-stage oil free piston compressor with intercooling is selected for the compression unit operation. For the Type 1 system, the outlet O2/H2 gas is compressed to 305 psi to allow for the 5 psi pressure loss associated with separation of the product gases. The yearly average plant size is 1 TPD and without interim compressed H2 storage facilities, the compressors are sized to handle the peak H2 output and will operate at reduced capacity most of the year. For Types 1 and 2, this output is the average daily production on June 21 because the baggies have the capacity to expand to accumulate excess gas in the bed headspace over the 24 hour day. For the Type 3 system, the compressor is sized for the instantaneous peak output at noon on June 21 because there is no space for accumulation. Given the additional volume of the O2 in the product stream, the Type 1 compressor must handle

1.5 times the volume flow when compared to the Type 2 and 3 compressors. The Type 4 reactor does not require an additional compression stage because the H2 exits the reactor at 300 psi so only a condenser is needed at the gas outlet to separate water vapor.

Commercial methods considered for separating out the H2 in the product stream of the Type 1 reactor include pressure swing adsorption (PSA), temperature swing adsorption (TSA), palladium membrane separation, nano-porous membrane separation, and electrochemical pumps. PSA, frequently employed for the separation of H2 from steam methane reforming product gases, was selected as the best option to separate a mixture of

H2, O2, and H2O for the following reasons: lower cost, higher technological maturity, reduced cycle time, lower temperature and pressure requirements, and no need for additional gas cleanup processes. PSA purification proceeds by flowing the pressurized mixed stream over an absorbent bed designed to capture the undesired gas while H2 continues flowing through the system; a carbon sorbent is used for O2 and silica gel for

H2O. Once the adsorbent is saturated, the bed is vented by decreasing the pressure and introducing a small amount of pure H2 to drive out residual O2 and H2O. Hydrogen recovery is less than unity because a small quantity of H2 is trapped in the adsorption bed and product gas is lost during the purge cycle. A summary of the auxiliary components needed for each system is shown in Table 2.2.

Table 2.2: Summary of liquid and gas handling systems required for each reactor type. A control system subassembly is also required for local and remote monitoring, alarming for hazardous conditions, and controlling of plant equipment. To minimize control costs, the lowest degree of control sophistication which allows full functionality and safe operation is assumed. Water level controllers are incorporated to assure adequate flow to

35 the reactors, flow meters at the gas processing outlet are used to measure product flow, pressure sensors are used to monitor for pressure build-up or loss, and gas sensors are needed to sample for contaminants in the output stream. Support components such as programmable logic controllers, control room computers and software, power and instrumentation wiring, wiring conduits, alarms, and electrical power are all included in the costs. The large areas covered by the plants increase the costs of the control system and impose the need for remote control capabilities. Each reactor type requires a different number of sensors, alarms, etc. due to vastly different plant layouts. Lastly, a makeup water assembly is required to continuously feed water to the reactors as it is consumed to evolve gaseous products.

2.6 Costs The levelized cost of hydrogen in $/kg for each system type was computed using the H2A model, version 2.0 in 2005 U.S. dollars. The DOE H2A analysis tool is primarily intended to compare different hydrogen production pathways or, as is the case here, different embodiments of the same approach. The structured format of the H2A model allows a user to enter the cash inflows and outflows associated both with construction and operation of a hydrogen production plant. A discounted cash flow analysis was performed within the H2A framework to evaluate an appropriate return on investment used to assess yearly costs for capital equipment investments. All plant-specific parameters related to location, operation, and reactor design are determined or assumed for each system design while the H2A default values for many general parameters are retained. A complete list of default values may be found in the original report45 while an abbreviated list of the most important is shown in Table 2.3 along with some assumed parameters common to all systems. Note that while high purity oxygen is produced as a by-product, it is vented to atmosphere and no cost credit is taken.

This type of costing analysis is not without limitations, especially given the number of performance and cost projections required. For example, the semiconductor materials selected for pricing the functional materials in each system embodiment, such as Fe2O3 and TiO2, do not have the required functionality. While serving as a good starting point for a cost surrogate, serviceable materials may be considerably more expensive.

Table 2.3: Selected H2A default and assumed input parameters for the H2A costing analysis. It is also very difficult to engineer a single semiconductor to act both as an excellent absorber and efficient catalyst but the cost of adding a co-catalyst was not directly considered. However, one of the coated layers already incorporated in the costs could be a catalyst layer. It is important to bear in mind that there is currently no infrastructure for a hydrogen distribution network and the cost of delivery outside of the plant gate is not assessed. This analysis targeted only the production cost of the hydrogen in getting it to the plant gate at the desired purity and pressure as dictated by the H2A analysis. The primary purpose of this work is to report the main findings of the cost analysis and as such, many of the costing details incorporated in the report issued by DTI45 are not discussed here. We focus on identifying the major contributors to cost for each type of PEC system. Some system-specific costs are discussed below with a summary of all capital and operating costs for each system which lead to a levelized cost of hydrogen in

$/kg H2.

2.6.1 System-specific Costs

The particularities of each cell and plant design result in unique costs associated with a specific system type. For the Type 1 and 2 systems, the reactor baggies cover a considerable land area. Thus, care was taken to accurately determine excavation costs based on local labor rates, equipment required, and estimated time to level the area for one baggie. The estimated excavation costs exclusively for leveling the land under the baggies were $46,259 and $82,237 for the Type 1 and Type 2 systems, respectively.

Costing of the particles was based on a slurry coating process similar to that used in the pharmaceutical industry; the particles are modeled as 40 nm Fe2O3 ($188/kg) substrates coated with two 5 nm layers of TiO2 ($278/kg). Assuming a production volume of 41,600 kg/year (sufficient to evolve 500 TPD H2), a short Design for Manufacture and Assembly (DFMA) analysis yields a total cost of $304/kg of which $209/kg is materials, $17/kg is the coating process, and $79/kg is the markup to account for scrap, R&D, and profit. While cheaper production technologies may be developed and the demand volume may be vastly different, it will be shown that these developments would have a minimal impact on the final H2 cost, which is fairly insensitive to particle cost due to the low mass of particles required. For the Type 3 system, the cost of panel electrodes is a key factor. Given the present lack of industrial production of such electrodes, costs are based on the solar cell open literature cost reports, NREL cost projections for solar cells83, and a DFMA style cost analysis. A cost of $150 - $200/m2 is assumed based on using low cost PV printing techniques. For the Type 4 system, the processing subassembly, along with pumps and the control system, will require electrical power. Furthermore, water is required as a feedstock for electrolysis. The utility usage was computed for each system, as shown in Table 2.4, and the H2A model rates were used to find the total utilities cost. Plant staffing requirements were based on the assumption that one worker could oversee 100 acres of baggies or panels and a plant supervisor is required for large operations.

Table 2.4: Utility usage for unit operations and feedstocks for 1 TPD H2 production plant modules of each of the four reactor types. 2.6.2 System Capital Costs

All components except for the HDPE baggies, photoactive particles, and PEC cells were assumed to have a 20 year lifetime and not contribute to yearly replacement costs. Both

38 the baggies and particles were assumed to have a five year service lifetime for the Type 1 and Type 2 systems while the PEC panels for the Type 3 and Type 4 systems had an assumed lifetime of 10 years. Lifetime refers to the length of time the component is in use prior to being replaced but does not consider any deterioration in performance. For all system types, the bulk of the cost of hydrogen stems from capital expenditures. The capital cost of each 1 TPD system was calculated and the results are summarized in Table 2.5.

Table 2.5: Summary of all direct capital expenditures and installation costs for the four different 1 TPD net H2 production plant modules. 2.6.3 Levelized Hydrogen Cost & Sensitivity Analysis

Each system was designed for 1 TPD H2 production but the overall levelized cost of hydrogen calculation assumes a 10 TPD H2 demand and thus, 10 plant modules each producing 1 TPD. The increase in scale primarily results in a decrease in labor costs per kilogram of hydrogen produced. All the systems considered are centralized hydrogen production facilities; the cost represents the price of hydrogen at the plant gate and does not take into account delivery or dispensing costs. The calculated levelized cost of

39 hydrogen for the Type 1, Type 2, Type 3, and Type 4 systems is $1.60/kg H2, $3.20/kg

H2, $10.40/kg H2, and $4.10/kg H2, respectively. There is some uncertainty in the absolute output values ($/kg H2) emerging from this model. However, these results show that PEC hydrogen can potentially meet the DOE cost goal of $2.00 - $4.00/kg H2 discussed later in the text. The numbers can also be used to make instructive comparisons of system costs. It is useful to breakdown the cost into capital costs, fixed operation and maintenance costs, variable costs, and decommissioning costs as shown in Figure 2.7.

Figure 2.7: Distribution of cost contributions to the levelized price of hydrogen. For every system, the output varies significantly over the course of the year, with the December output being much lower than the June production. If adequate storage options are not available and winter demand is high, the systems would need to be scaled up and the costs would be elevated. We once again emphasize that these are conceptual systems for which there is a large degree of uncertainty in the system performance, H2 demand schedule, durability, and cost. To help illustrate the effects of these uncertainties on the cost of the H2 output at the plant gate, a sensitivity analysis was carried out to gauge the relative effects of system efficiency and component lifetime. Sensitivity to the cost of the photocatalytic particles was also considered for the Type 1 and 2 systems while the cost of the PEC cells was considered for the Type 3 and 4 systems. The sensitivity analysis presented here is an attempt to identify the most impactful parameters

40 on the final cost of H2 but other assumed costs (e.g. land costs, labor rates, pumps/compressors, etc.) will also vary to some degree; a more extensive sensitivity analysis of all parameters to determine the full range of error is beyond the scope of the current work but will likely be pursued as the technology matures. The results of the sensitivity analysis are shown in Figure 2.8 along with the range of evaluation parameters explored.

Figure 2.8: Effect of efficiency, particle or panel cost, and component lifetime on the cost of hydrogen from each reactor design. Each calculation represents the variation of a single parameter from the base case to a higher and lower value as indicated on the left axis. For all four systems, the capital costs make up a significant fraction of the overall cost. We consider the major contributions in each case to identify where design uncertainty could lead to increases in cost and where research progress could drive down costs significantly. The cost of hydrogen produced from the Type 1 and Type 2 systems is very low at $1.60/kg and $3.20/kg, respectively. However, the performance of the particulate systems on a large scale is not well established given incomplete demonstration of the effective performance of particles as a function of depth in the baggies, photovoltage generated by multilayer particles, voltage losses across porous bridges, particle lifetime,

41 and scalable particle fabrication methods. For the Type 1 system, gas compression equipment accounts for over half of the direct capital costs. Given the safety concerns associated with compression of a combustible mixture of stoichiometric H2 and O2, costs could rise as suitable engineering controls are incorporated to mitigate the risk. Another possible source of rising costs could come from increasing the number of baggies to diminish the capacity loss in case of failure, likely bringing the cost of hydrogen closer to the Type 2 system. Among particle efficiency, cost, and lifetime, the sensitivity analysis for the Type 1 system revealed that efficiency has the biggest impact; however, all cost differentials were small. The keys to bringing this type of system to market appear to be more related to technological feasibility and could be: (i) to develop particles that provide the photovoltage required to split water but still utilize a large portion of the solar spectrum, and (ii) to find innovative solutions for the compression of the H2/O2 mixture.

The Type 2 system cost breakdown is similar to the Type 1 system, with the capital costs and associated installation expenses for the greater number of baggies accounting for the higher levelized cost of $3.20/kg H2. Electrolyte voltage losses in this system could prove problematic as the redox shuttle must travel between the separate O2 and H2 generating baggies. The baggie width has been reduced to alleviate these losses but molecular transport on the order of meters rather than centimeters, typical of research cells, has not been demonstrated. A priority for developing a Type 2 system is designing the membrane bridges and slurry circulation system to minimize voltage losses while preventing O2 and

H2 diffusion. In addition to demonstrating the effective use of a redox mediator, there is still the principal challenge of developing the separate particles needed for O2 and H2 generation. An advantage of this system over its Type 1 counterpart is the potential for researchers to independently optimize the O2 and H2 generating particles, allowing for much greater flexibility in designing materials. In summary, particulate systems have lower predicted cost but significant technical risk given the state of development of both the reactor systems and the photocatalytic particles.

The baseline costs of the Type 3 and Type 4 systems are higher than that of the particulate systems but leveraging the knowledge from the PV industry increases the cost certainty for components such as the concentrating system and thin film panels. Capital

42 costs dominate the price of $10.40/kg H2 for a Type 3 system, the highest of the four embodiments considered. More than 80% of the cost originates in the materials, construction, and installation of the photoactive cells. Given the rigid encapsulation framework, as opposed to the flexible baggies of the Type 1 and 2 reactors, compression costs for the Type 3 reactor are also high since the auxiliary units must be sized for the peak hourly output and not the average output over the day. While the accuracy of the cost analysis for the Type 3 and 4 systems benefitted from the pricing information available from the PV industry, it also relies on the projected development of lower cost thin film materials. It is clear from the sensitivity analysis that all parameters, i.e. efficiency, cell cost, and durability of planar electrodes, affect the costs; research and development must focus on all three aspects to ultimately produce high-efficiency, stable materials that can be manufactured cheaply from earth abundant elements.

The Type 4 system, yielding a cost of $4.10/kg H2, is the more attractive of the two panel options based on this analysis. The reduced photoelectrode area required with 10:1 solar concentration significantly reduces the capital costs associated with panel production and brings the levelized cost of hydrogen much closer to that of the slurry systems. This fact exemplifies the utility of an upfront technoeconomic analysis for energy technologies that do not yet exist at the commercial scale. Researchers can recognize that any PEC system developed in the laboratory will likely be made more cost effective at the commercial scale if 10:1 solar concentration is employed, thus motivating research into the effects of increased light intensity on photoabsorber materials. In the Type 4 reactor, the solar collector structure is now the primary expense driving the price of hydrogen. Progress in the PV industry has already brought these costs down but modest improvements may still be possible when considering a system tailored for PEC water splitting. Increasing the concentration ratio to 20:1 would further lower the costs by an estimated 10%, but there may be catalysis and bubble formation issues which limit practical implementation. Efficiency is a key parameter driving costs for this system and thus materials discovery and development should focus on high efficiency systems. A PEC cell efficiency reaching 25% would reduce the levelized cost of hydrogen to a value of $2.90/kg. Figure 2.3 illustrates that a 22.8% STH efficiency is potentially realizable in a tandem structure

43 even when practical losses are considered; future improvements in both catalysts and semiconductors or the use of triple junction cells should allow the small step to 25%.

2.6.4 Cost Comparison

It is important to establish what the target cost of hydrogen should be to compete with other liquid fuels or hydrogen production technologies. In the DOE Fuel Cell Technologies Program Multi-Year Research, Development and Demonstration Plan, the objective was set to reduce the cost of hydrogen to $2.00 - $4.00/gasoline gallon equivalent (gge) delivered at the pump. The basis for this target is that the cost of hydrogen should be roughly the same as that of untaxed gasoline. Based on the lower heating values, one gallon of gasoline is approximately equivalent to one kilogram of hydrogen resulting in a target cost of $2.00 - $4.00/kg H2.

The least expensive current H2 production process is steam methane reforming, which 84 provides hydrogen at a cost of $1.00 - $5.00/kg H2 but of course, direct consumption of fossil fuels to produce hydrogen is not a sustainable process. The low price of natural gas, likely to persist for the foreseeable future due to the recent advent of hydraulic fracturing 85 methods, means the cost of H2 is likely to be on the lower end of this range. Using the linear relationship between natural gas prices and hydrogen production cost established by Lemus et al., the centralized cost (i.e. not including delivery) of steam methane 86 reforming is roughly $1.25/kg H2 based on recent natural gas prices around $3.50/GJ. Large-scale electrolysis, using the industrial coal-powered electricity price of $0.05/kWh, 84 has a cost of $4.09/kg H2 computed using an H2A type analysis. Once again, fossil fuels appear in the chemical balance sheet unless all the grid electricity comes from renewable sources. A completely clean route to hydrogen would be connecting an electrolyzer to a PV array, both commercial technologies today which can be coupled87. While the cost of PV is coming down, it is still not as cheap as coal-powered electricity.88 As a result, if the grid electricity supplying the electrolyzer was replaced with PV electricity, the cost would still be > $4.09/kg H2. We emphasize that the full technoeconomics of PV-electrolysis is outside of the scope of this work but such a project would be interesting. Clearly there is room for optimization of the coupling of PV and electrolysis, such as considering load leveling, which would presumably drive costs

44 down. A technoeconomic analysis of PV-electrolysis could serve to identify pathways to achieve the DOE goal as the analysis presented here has done for the case of direct water photolysis.

A previous report that evaluated the cost of hydrogen produced from several distributed and centralized technologies, without consideration of costs of compression, storage and delivery, revealed that the price of production alone was $1.61/kg for centralized biomass gasification, $1.33/kg for natural gas reforming, $4.50/kg for wind electrolysis, and $2.05/kg for coal gasification with carbon capture.89 These costs would be slightly higher if compression to 300 psi were included to directly compare to the PEC hydrogen production costs. Given the estimated costs for H2 at 300 psi from the four conceptual water splitting systems described in this chapter, as low as $1.60/kg H2 for particle-based systems and $4.10/kg H2 for concentrated panel systems, it is clear that PEC hydrogen production is a viable option among the carbon-free processes. Interestingly, a previous analysis carried out nearly two decades ago, using a slightly different methodology and assumptions, came to the same conclusion that a dual bed particle based system would be more economical than a commercial PV coupled to an electrolyzer, a multilayer panel PEC system, or a concentrated multilayer panel PEC water splitting system.90

2.7 Conclusions The levelized cost of hydrogen was computed using standard H2A methodology to assess the viability of photoelectrochemical water splitting as a carbon-free means to produce hydrogen. The four conceptual systems evaluated were a Type 1 single bed particle suspension, a Type 2 dual bed particle suspension, a Type 3 fixed panel array, and a Type 4 tracking concentrator array. For each photoabsorber arrangement, theoretical efficiency calculations were carried out and compared to actual laboratory-scale materials benchmarks in order to determine reasonable target system efficiencies. The baseline levelized production cost of hydrogen was computed to be $1.60, $3.20, $10.40, and

$4.10/kg H2 for the Type 1, Type 2, Type 3 and Type 4 systems, respectively. The particle slurry systems have significantly lower capital costs but there is greater uncertainty associated with their operation, such as safety concerns over the cogeneration of H2 and O2 for the Type 1 system or the long molecular transport lengths of redox

45 shuttles in the Type 2 system. The panel array systems were more expensive due to their significant capital costs. Panel fabrication and encapsulation costs dominate for the Type 3 system while the solar concentrator and tracking components drive the cost of the Type 4 system. However, the sensitivity analysis reveals that there is a significant opportunity to reduce the cost of the panel based systems by improving materials efficiency and by employing solar concentration. PEC cell cost and durability are also secondary drivers for the cost of the output hydrogen. This work clearly demonstrates that if technical progress is made to meet material performance targets and with appropriate plant-scale engineering, direct solar hydrogen produced by photoelectrochemical water splitting can be produced at a cost which meets the DOE target of $2.00 - $4.00/kg H2.

2.8 Author Contributions The involvement of many authors spanning academia, industry, and government was necessary to ensure the analysis was comprehensive, technically sound, and clearly presented. Blaise A. Pinaud wrote all text unless otherwise attributed here and produced the figures. Jesse D. Benck wrote the section on Plant Design and Operation and assisted in generating the graphics. Linsey C. Seitz carried out the efficiency calculations and wrote the section entitled Calculation of Practical System Efficiencies. Arnold J. Forman wrote the portion detailing Demonstrated Research Efficiencies of photocatalytic systems. Zhebo Chen assisted with the efficiency calculations and editing of the entire text. Todd G. Deutsch authored the text on Demonstrated Research Efficiencies for photoelectrochemical systems. Brian D. James, Kevin N. Baum, and George N. Baum conducted the H2A analysis and wrote the original report which inspired this work as well as provided feedback on the manuscript. Shane Ardo, Heli Wang, Eric Miller, and Thomas F. Jaramillo were all involved in the conception of this work and provided significant edits.

2.9 Copyright Portions of this chapter from

B. A. Pinaud, J. D. Benck, L. C. Seitz, A. J. Forman, Z. Chen, T. G. Deutsch, B. D. James, K. N. Baum, G. N. Baum, S. Ardo, H. Wang, E. Miller, and T. F. Jaramillo,

“Technical and economic feasibility of centralized facilities for solar hydrogen production via photocatalysis and photoelectrochemistry”, Energy & Environmental Science 6, 1983-2002 (2013).

Reproduced by permission of The Royal Society of Chemistry.

Chapter 3 Effect of Film Morphology and Thickness on

Charge Transport in Ta3N5/Ta Photoanodes for Solar Water Splitting

3.1 Abstract

Ta3N5 is a promising photoanode material due to its nearly ideal band structure for solar water splitting. Previous results indicate Ta3N5 may be hindered by charge transport limitations attributed to poor bulk charge transport, charge transport across grain boundaries, and/or charge transfer across the interface at the back contact. The primary goal of this chapter was to study these mechanisms, especially bulk hole and electron transport, to determine which processes limit device efficiency. Crystalline thin films (60 nm – 780 nm) of Ta3N5 (Eg = 2.1 eV) on Ta foils were synthesized by oxidation of Ta metal in air at 550°C and subsequent nitridation in NH3 at 900°C. Scanning electron microscopy revealed that thermal stresses and differences in the density of the phases resulted in the formation of porous, textured films with high surface area. Films were characterized by their photon absorption, crystal grain size, and electrochemically active surface area. Trends in photoactivity as a function of film thickness under broadband illumination as well as in the incident photon-to-current efficiency revealed that minority charge carrier (hole) and majority charge carrier (electron) transport both play important roles in dictating photoconversion efficiency in Ta3N5 films.

3.2 Introduction

As discussed in the first chapter, tantalum nitride (Ta3N5) is an excellent photoanode candidate as the band positions are well matched to the oxygen evolution reaction (OER) 26 and hydrogen evolution reaction (HER) redox potentials. Furthermore, red Ta3N5 has a band gap of 2.1 eV and thus absorbs in the visible.26, 33 Based on the optical limit (complete absorption and 100% utilization of above band gap photons), a photocurrent of

2 27 12.5 mA/cm is theoretically possible. Previous work to synthesize Ta3N5 includes 26, 30, 32 nitridation of Ta2O5 powder , thermal oxidation and nitridation of Ta foils to yield thin films22, 33, sputtered thin films34, anodization of Ta foils to make nanotube arrays35, 39 40, 41 drop casting doped films from TaCl5 , atomic layer deposition , and fabrication of nanoparticles91, 92. The photoactivity of many of these samples has been measured (under various conditions); even with a substantial applied bias, the highest reported photocurrents under visible light (intensity not specified) have been limited to ~ 4 2 33 3- 4- mA/cm . The addition of a highly reversible redox couple such as Fe(CN)6 / Fe(CN)6 to circumvent issues with surface catalysis still yields photocurrents < 7 mA/cm2.33 In order to engineer this material to approach its optical limit, its fundamental limitations must be identified through careful study of its properties; previous work31, 35 has pointed to possible issues with charge transport which we seek to investigate in the study herein.

Photoanode performance can potentially be hindered by several charge transport limitations, including poor (i) charge transport of both electrons (e-) and holes (h+) through the bulk of the material, (ii) charge transport of the two carriers across grain boundaries, (iii) charge transfer of electrons across the interface at the back contact, and (iv) charge transfer of holes across the semiconductor/liquid interface, i.e. catalytic water oxidation. The first limitation might arise in Ta3N5 due to many defect sites which act as recombination centers. The second limitation depends significantly on the synthesis of

Ta3N5 which is usually accomplished by nitridation of tantalum oxides, resulting in a multicrystalline material with relatively small domains and thus many grain boundaries. The third issue is predicated on the fact that the optimal back contact must have a work function which closely matches the Fermi level of the semiconductor material and depends greatly on the atomic-scale structure of the interface. The fourth limitation involves the efficiency of water oxidation at the solid/liquid interface which can be significantly improved by adding an OER co-catalyst.31, 93 The goal of this work is to gain insight into bulk hole and electron transport by studying the photoactivity of thin films of Ta3N5 as a function of thickness synthesized directly on Ta metal. This back contact is selected to help mitigate issue (iii); the work function of polycrystalline Ta 94 metal (4.25 eV) is nearly ideal for matching up with the band structure of n-type Ta3N5 for which the conduction band (CB) is at 4.0 eV (– 0.4 V vs. NHE).26 To our knowledge,

50 this work is the first to focus on studying these charge transport effects to explicitly identify limiting processes.

3.3 Experimental Methods

3.3.1 Sample Preparation

Ta foils (10 mm x 10 mm x 0.5 mm, Alfa Aesar, 99.95% metals basis excluding Nb) were mechanically polished to yield a flat surface. The as-received foils were rough with features on the micron scale which had to be smoothed to facilitate the synthesis of 95 uniform, thin films of Ta3N5. Electrochemical polishing is an alternate route that was explored but the mechanically polished samples were more uniform and the process more scalable. Micrographs of the as-received, electropolished, and mechanically polished bare Ta foils are shown in Figure 3.1.

Electropolishing in a bath of 3 M H2SO4 in methanol at 7.3 V vs Hg/Hg2SO4 for 5 – 10 min resulted in a relatively smooth surface but with a large amount of debris as shown in Figure 3.1(c). It was also difficult to ensure uniform polishing across the entire length and width of the sample. In addition, the throughput of this method was low and there was a high degree of sample to sample variability. Mechanical polishing, for which the morphology is shown in Figure 3.1(b), was very reproducible and several samples could be prepared in tandem; all Ta foils were polished mechanically prior to oxidation and

51 nitridation. Polished foils were cleaned by sequential sonication in acetone, isopropanol and Milli-Q water. Ta3N5 thin films grown on Ta (denoted Ta3N5/Ta henceforth) were synthesized in a two step process involving separate oxidation and nitridation steps. The thermal oxidation of Ta metal is well documented in the literature.96-98 Briefly, Ta foils were inserted into a hot tube furnace (single zone, Mellen Company) held at 550°C and open to air. The oxidation time, which dictates oxide film thickness since we found that

Ta metal does not readily nitride in NH3, was varied between 5 min – 30 min. Nitridation was carried out by placing the samples in a sealed, continuous-flow tube furnace. A flow of 100 sccm Ar (Praxair, 99.999%) was maintained while the temperature was ramped to 900°C at a rate of 10°C/min. The sample was then held at 900°C for 1 hr while flowing

50 sccm NH3 (Praxair, 99.995%) and subsequently cooled to room temperature in an Ar atmosphere. The nitridation conditions were identical regardless of initial oxide thickness and the synthesis of oxide-free Ta3N5 was confirmed by x-ray diffraction (XRD) analysis of film crystallinity. Samples were produced in triplicate for imaging and so that fresh samples were available for photoelectrochemical measurements with broadband and monochromatic illumination.

3.3.2 Film Characterization

The morphology of the Ta3N5 films was imaged using scanning electron microscopy (SEM) (FEI XL30 Sirion, 5 kV and Phi 700 AES, 10 kV). The film thickness was also evaluated by SEM; the films supported on Ta foils were sheared to reveal a cross-section which was mounted at approximately 90°, with some off-axis variation leading to a small error in the measurements. Each reported film thickness is the average of five separate measurements each on three different areas of the cross-section. Ambient temperature XRD (Phillips PANanalytical X’Pert Pro) using Cu Kα radiation (λ = 1.54184 Å) was employed to study the crystallinity of the films. Crystallite size was computed using the Scherrer equation99, assuming a shape factor of 0.9. Efforts were made to determine the elemental composition of the Ta3N5 films through x-ray photoelectron and Auger spectroscopy (results not shown). This approach has a severe limitation in that the surface is typically oxygen rich so sputtering is required to evaluate the bulk composition. The bulk rather than surface composition is of interest as it will dictate the absorption and

52 charge transport properties of the material. Reduction of the Ta and preferential sputtering of the various elements during sputtering render this method inaccurate.

3.3.3 Electrochemical Testing

All electrochemical and photoelectrochemical testing was carried out in a custom designed Teflon compression cell (see Appendix B for details) with an exposed working electrode area of 0.5 cm2 as shown in Figure 3.2.

Figure 3.2: Side view of the compression cell used for three-electrode electrochemical and photoelectrochemical measurements.

Contact was made to the back of the Ta3N5/Ta working electrode using conductive copper tape. The measured ohmic drop in the 0.1 M KOH (pH 13.3) solution purged with

N2 was < 67 Ω and no correction was applied to the data. The remainder of the conventional three-electrode configuration consisted of a Pt mesh counter electrode and a Ag/AgCl (4 M KCl) reference electrode (+ 0.965 V vs. RHE). Data was collected using a potentiostat (Bio-Logic VSP). Relative electrochemical surface area was determined from capacitance measurements by cyclic voltammetry in the region between – 0.25 V and 0 V vs. Ag/AgCl in 0.1 M KOH carried out in the dark. Under these conditions, all current is attributed to capacitive charging in this potential window due to the absence of any redox features. The scan rate was varied between 25 mV/s and 300 mV/s with a 30 s hold at each vertex to allow the current to decay to zero. The average value of the anodic current at – 0.05 V vs. Ag/AgCl and cathodic current at – 0.20 V vs. Ag/AgCl is reported. As expected for a charging process not limited by mass transfer, the relationship between scan rate and capacitive current was found to be linear. The illumination source for the study of photoactivity was a 1000 W Xenon lamp (Newport) fitted with a water filter to

53 reduce output in the IR region. When necessary, a neutral density filter (optical density of 0.3 at 632.8 nm, Newport) was used to uniformly reduce the output over all wavelengths. The light intensity and spectral distribution incident on the sample was measured prior to each experiment using an integrating sphere (International Light INS150) and spectroradiometer (International Light RPS900-W). The measurement was carried out by assembling the cell in full without a sample or electrolyte and placing the integrating sphere directly beneath the opening in the base. A broadband illumination of 154 mW/cm2 (integrated over λ = 250 nm – 950 nm) was measured, corresponding to ~ 2 suns at λ = 500 nm. Monochromatic light (fwhm = 10.1 nm ± σ = 0.3) was obtained using a monochromator (74100 Oriel Cornerstone) and suitable cut-off filters were used to eliminate light from higher order diffractions. To determine the incident photon-to- current efficiency, each sample was held at constant potential while the wavelength of light was varied in 20 nm increments between 440 nm – 800 nm. The light was chopped at a frequency of 0.2 Hz and the dark current subtracted from the total current to yield actual photocurrent. An average value over 5 chopping cycles is reported.

3.4 Results and Discussion

3.4.1 Film Thickness and Morphology

Cross-sectional images for the Ta3N5/Ta films of four different thicknesses are shown in Figure 3.3(a)-(d). Despite an initially polished surface, the resulting films are textured and somewhat porous. The porosity is due to changes in density during the 3 30 3 100 transformation from Ta (ρ = 16.6 g/cm ) to Ta2O5 (ρ = 8.20 g/cm ) , and finally to 3 101 Ta3N5 (ρ = 9.85 g/cm ) . The formation of the porous structure was studied in more detail by synthesizing films on single crystal Si and quartz substrates which are flatter and better defined than the Ta foils, allowing for more facile observation of changes in thickness and morphology. Electron beam evaporation was used to deposit a well-defined thickness of Ta metal. These samples were then oxidized at 750°C for 30 min in air and nitrided at 900°C for 60 min under identical conditions to the Ta3N5/Ta samples and formation of crystalline Ta3N5 was confirmed by XRD (not shown).

Figure 3.3: Cross-sectional SEM images of Ta3N5/Ta films for average thicknesses of (a) 60 nm, (b) 260 nm, (c) 630 nm, and (d) 780 nm. The top-down images show the textured morphology for both (e) thin films (60 nm) and (f) thick films (780 nm). Ta3N5 film is colorized to facilitate distinction from Ta metal. The film thickness of the resulting nitride was measured via profilometry on samples which were masked prior to Ta evaporation. SEM images for Ta, Ta2O5, and Ta3N5 films on single crystal Si are shown in Figure 3.4 while the changes in thickness are recorded in Figure 3.5. Note that the initial thickness of Ta metal deposited for all samples shown in Figure 3.4 was 140 nm. It was found that while the thickness of the film expands by a factor of 2 during oxidation, the higher density of the Ta3N5 phase compared to the Ta2O5 phase results in the formation of void space rather than a decrease in film thickness upon nitridation. Unfortunately, the only way to avoid forming this porosity, which may have a - negative impact on e transport, is to directly synthesize Ta3N5 without passing through the oxide phase. While not explored in the course of this work, it should be possible to accomplish this using atomic layer deposition41 or reactive sputtering.34

The rough surface texture of the Ta3N5/Ta samples is clearly evident in the top-down micrographs in Figure 3.3(e) and (f) which show differential growth patterns on the Ta grains. The surface texturing effect is more pronounced in thicker (Figure 3.3(f)) than in thinner (Figure 3.3(e)) films.

We found that varying the oxidation temperature and nitridation conditions (NH3 flow, time, and temperature) as well as attempting a simultaneous oxidation and nitridation procedure with a mixed stream of O2 and NH3 did not produce morphologies any more uniform or dense than those shown in Figure 3.3. Note as well in the cross-sectional views that there is a very sharp interface between that Ta3N5 and Ta but that there is no delamination indicating good physical and electrical contact between the two phases. The resulting average film thicknesses (60 nm, 260 nm, 630 nm, and 780 nm) are shown in Figure 3.6 as a function of oxidation time (5 min, 10 min, 15 min, and 30 min) where the large standard deviation can be attributed to the non-uniformity of the film thickness. The observed film growth rate is highly non-linear; the initial slower growth is believed to be due to thermal conductivity, i.e. the heating time required for the sample to reach the temperature of the furnace (550°C), while growth at longer times may be limited by diffusion of oxygen.

Figure 3.6: The measured Ta3N5 film thickness based on cross-sectional SEM images as a function of the nominal oxidation time. Thin films of tens to hundreds of nanometers have successfully been synthesized. It is clear, however, that the average thickness of the films spans the range of tens to hundreds of nanometers which should yield substantial differences in the charge carrier transport lengths.

3.4.2 Film Crystallinity

X-ray diffractograms for a set of four films are shown in Figure 3.7 and confirm the synthesis of phase-pure Ta3N5 with no TaON or Ta2O5 impurities based on reference spectra (PDF cards 00-004-0788 for Ta, 01-071-0639 for Ta2O5, 01-071-0178 for TaON, and 01-079-1533 for Ta3N5). The peak at 38.6° corresponds to the underlying cubic Ta foil crystal structure. While the features at 17.3°, 24.3°, and 31.3° align with the reference spectra, the peaks indexed to the (130) and (113) diffractions which should be at 35.0° and 36.0° are shifted to lower angles for the two thinnest films. The origin of this shift is currently unknown. The Ta3N5 grain size for each film thickness was computed from the two most intense, unshifted peaks at 24.3° and 31.3°, corresponding to the (110) and (023) diffractions, using the Scherrer equation and the results are shown in Table 3.1. The calculated crystallite size varies between 12 nm – 19.6 nm with the thicker films exhibiting only slightly more long range order.

Figure 3.7: X-ray diffractograms for the films of four different thicknesses showing only peaks attributed to crystalline Ta3N5 or the underlying cubic Ta metal substrate.

Table 3.1: The average crystal grain size in the films was calculated by applying the Scherrer equation to the peaks at 24.3° and 31.3°. There was only a small difference in the crystallite sizes for all films. While the degree of crystallinity of a semiconductor material can potentially impact charge transport properties by means of trap states at the grain boundaries that act as recombination centers, the small differences in crystallinity found amongst these samples are unlikely to factor significantly in any trends in electrode performance or charge transport.

3.4.3 Photoelectrochemistry

The first metric for film performance evaluated was the photocurrent generated under broadband illumination. Chopped (f = 0.2 Hz) cyclic voltammograms (scan rate = 10 mV/s) for 60 nm, 260 nm, 630 nm, and 780 nm Ta3N5/Ta films as well as a bare Ta foil are shown in Figure 3.8. A very important detail to note at this point is that photocurrent is not necessarily the result of water oxidation. In fact, Ta3N5 is known to photodecompose through a process in which photogenerated holes oxidize the film itself 33 resulting in the release of N2, rather than oxidizing water.

Figure 3.8: Chopped cyclic voltammograms for Ta3N5/Ta films as well as a bare Ta foil under white light illumination. All films were photoactive with the two thickest films outperforming the two thinner films significantly (note the different current scales). Photocurrent may be the result of both water oxidation and oxidative photodegradation. However, a co-catalyst was intentionally omitted in this study because even slight differences in coverage due to variation in the synthesis could result in effects which would convolute the study of charge transport. For example, a higher coverage on one film compared to another could provide a greater number of active sites for water oxidation or could result in decreased absorption in the Ta3N5 since the catalysts are typically not optically transparent. Thus we studied all films without any intentional modification of their surfaces. The photodegradation results in hysteresis of the photocurrent with progressive cycling; near-stable operation was achieved after five cycles (5th cycle shown in Figure 3.8).

The onset of dark current for O2 evolution and/or degradation is at 1.6 V vs. RHE for both the nitride films and for the bare Ta foil. The additional oxidative feature in the dark, centered at 1.7 V vs. RHE, is attributed to electrochemical oxidation of exposed Ta metal, which may still be present in the nitrides due to cracks in the films. The onset of

59 photocurrent close to 0 V vs. RHE for all films is consistent with the previously reported 26 conduction band position at – 0.4 V vs. NHE for Ta3N5. Of great interest was the finding that the photocurrents for the 630 nm and 780 nm films were both over an order of magnitude higher than those of the 60 nm and 260 nm films.

To investigate this result further, monochromatic illumination was used to directly calculate the incident photon-to-current efficiency (IPCE), analogous to an external quantum efficiency, which is shown in Figure 3.9 for an applied potential of 1.57 V vs. RHE. A second IPCE was also collected at a lower potential of 1.27 V vs. RHE, shown in Figure 3.10, which reveals similar trends.

Figure 3.10: Incident photon-to-current efficiency for the Ta3N5/Ta films of four different thicknesses at 1.27 V vs RHE. The trends in activity are the same as at a higher potential.

The onset of significant photocurrent at ~ 620 nm (2 eV) is consistent with the published band gap of 2.1 eV.26 Consistent with the findings from broadband illumination, the IPCE at λ = 540 nm for the 780 nm and 630 nm films was significantly greater (approximately 170 and 210 times greater, respectively) than the IPCE of the 260 nm and 60 nm films. It is also interesting to note that in comparing the two thicker films, the IPCE of the 630 nm film was 1.2 times higher than that of the thicker 780 nm film.

These trends in photoactivity and IPCE contain key information regarding the material properties of these Ta3N5 thin films. The following two possible explanations for these trends could be postulated: (i) differences in film thickness could impact photon absorption due to differences in optical path lengths for each sample, and (ii) differences in the film morphology and roughness could impact charge transport through the film. To investigate the first effect, we aimed to calculate an absorbed photon-to-current efficiency (APCE) for each sample. To do so, the value of the absorption coefficient as a function of wavelength is required for the Ta3N5 films. The manner in which they were grown, however, was not conducive to such a measurement as the opaque Ta substrates posed experimental challenges in studying the optical properties of the films. As an alternative, thin Ta3N5 films were synthesized on transparent quartz substrates and the optical properties were studied by UV-vis absorption spectroscopy in transmission mode as shown in Figure 3.11.

Figure 3.11: (a) Absorption of Ta3N5 thin films of three different thicknesses grown on quartz substrates with Fabry-Pérot interference fringes clearly visible at longer wavelengths. (b) Calculated absorption coefficient for Ta3N5. Note that the large error at long wavelengths is due to the interference fringes evident in (a).

Note that the curves for each set of three samples of the same thickness nearly overlay with each other which is evidence of good reproducibility in the synthesis. This data was then used to extract an average absorption coefficient which is plotted as a function of wavelength in Figure 3.11(b). Subtraction of the quartz baseline is the only data processing carried out. Note the Fabry-Pérot interference fringes which are especially prominent at longer wavelengths and complicate analysis. These arise because the Ta3N5 films prepared on quartz have a very uniform thickness. The absorption coefficient for 4 -1 3 -1 Ta3N5 was found to be 6.2 x 10 cm (σ ~ 8 x 10 cm ) at λ = 540 nm. This value for the absorption coefficient is taken only as a first approximation for two reasons. Firstly, x-ray diffraction confirms these films are also crystalline Ta3N5 but slight differences in the spectra indicate that the growth orientation may be different from the Ta3N5 material grown on Ta foils. Secondly, the Fabry-Pérot interference fringes result in large error bars on the absorption coefficient at longer wavelengths though this error is substantially less close to the band edge.

Therefore, though the synthetic route for Ta3N5 films on quartz very closely resembled the synthesis on Ta foils, precise values for APCE should not be extracted as the syntheses are not (and cannot be) exactly the same. Nevertheless, the measured optical absorption in the transmissive samples did illustrate that absorption differences cannot be a dominant effect. An absorption coefficient of 6.2 x 104 cm-1 (σ ~ 8 x 103 cm-1) at λ =

540 nm was estimated for Ta3N5 based on UV-Vis absorption spectroscopy. Applying Beer’s Law, absorption accounts for a mere factor of 3.2 for the 780 nm film and 3.1 for the 630 nm film compared to the 60 nm film, and substantially less for the 260 nm film. Thus differences in optical absorption cannot be expected to give rise to the photocurrent/IPCE trends described earlier.

There is, however, strong evidence that differences in surface area, a result of the morphological changes as a function of thickness, play a major role. The relative surface area of the Ta3N5 films was determined from the capacitive region of cyclic voltammograms as shown in Figure 3.12 (representative sweeps for the 780 nm films are shown in Figure 3.12(a)).

Figure 3.12: (a) Example cyclic voltammograms showing the capacitive current for the 780 nm film at six different scan rates from 25 mV/s to 300 mV/s. (b) Plot showing the linear relationship between the capacitive current and scan rate for all films. The relative electrochemically active surface areas (using the 60 nm as the baseline) are compiled in the inset table. It is not possible to extract a value of the true electrochemical surface area per projected geometric area without an atomically flat Ta3N5 reference standard and detailed knowledge of the electronically accessible surface sites. However, assuming that the intrinsic specific surface capacitance of all Ta3N5 films is approximately the same, the relative surface area of the films can be determined. The 630 nm and 780 nm films are 17.9 and 27.1 times higher in surface area, respectively, than the 60 nm film. The increased area arises not only from higher surface roughness but also from cracks which form in the thicker films. The differences in surface area have important ramifications for charge transport, as discussed below.

3.4.4 Implications for Charge Transport

To put these results in context, it is instructive to consider diagrams showing the movement of photogenerated e- and h+ in the material as shown in Figure 3.13. The first image in Figure 3.13(a) is representative of many of the previous Ta3N5 films made by depositing a powder on an electrode. The high surface area results in very short distances the photogenerated h+ must travel to reach the reactive sites at the semiconductor/electrolyte interface. However, the lack of long range crystallinity and very high porosity leads to a tortuous path for e- transport. A thick, highly crystalline film as shown in Figure 3.13(b) can remedy the majority carrier transport issue by providing a continuous, less-defected path for e- movement to the back contact but it does so to the

63 detriment of the minority carrier transport as only a small fraction of the photogenerated charge carriers will be produced near the solid/liquid interface. Collection of photogenerated h+ near the surface is still facile but carriers generated deeper in the material have to travel much further than the average hole diffusion length. A thin crystalline film geometry as illustrated in Figure 3.13(c) solves this problem; the distances that the photogenerated carriers must travel are now small for both h+ and e- leading in theory to a high APCE. Presumably, photon absorption in such a film will likely be limited, resulting in a low IPCE, so methods to increase the path length of light through the material would be required.

Figure 3.13: Schematics showing the effect of film geometry and structure on electron and hole transport for supported Ta3N5 in the form of (a) powder films, (b) thick crystalline films, (c) thin crystalline films, and (d) thick, rough crystalline films. Based on the illustrative examples of Figure 3.13(a)-(c), we are in a better position to describe the phenomena observed in the Ta3N5/Ta films synthesized in this work. The thinner films (60 nm and 260 nm) are in the regime of thin, conformal films best represented by Figure 3.13(b). The data supports this point of view as (i) both films have similar, low surface area as shown in Figure 3.13(b), and (ii) both films exhibit nearly

64 identical photoactivity; the extra photon absorption in the thicker 260 nm film has little effect on overall efficiency, leading us to conclude that hole transport is likely a limiting process in both of these films.

The surface area measurements of Figure 3.13(b) indicate that the thicker films likely have a different film morphology, akin to that shown in Figure 3.13(d). The highly textured surface and porosity of the thicker films means that a much larger fraction of the photogenerated e-/h+ pairs is produced near the solid/liquid interface, significantly improving h+ collection. This translates to a one to two orders of magnitude increase in photoactivity of the two thicker films (780 nm and 630 nm) compared to the thinner films (60 nm and 260 nm). The fact that photocurrents appear to scale primarily with surface area, rather than with thickness and therefore with e- transport lengths, points to h+ transport as a fundamental limiting process in n-type Ta3N5. Further evidence in support of this conclusion is the steep, continuous rise in the IPCE (Figure 3.9) at wavelengths above the band gap; this suggests that charge carriers generated by higher energy photons, which are absorbed closer to the top surface of the films as we employed front- side illumination, are collected more efficiently even if the e- must travel further to reach the Ta back contact.

Hole transport, however, does not account for all of the observed trends in photoactivity. As noted earlier, in comparing the two thicker films (630 nm and 780 nm) to one another, the 780 nm film has a slightly lower activity than the 630 nm film. Based on the discussion above relating higher surface area to improved hole transport, one would expect the 780 nm film to perform better; another limiting mechanism must be at play. The data suggest that e- transport begins to play a more important role for thicker films in which most of the photons are absorbed in the region closest to the light source (at the top of the film). Our experimentally-determined absorption coefficients described earlier suggest that for above band gap photons, 90% should be absorbed within the top-most 370 nm of the film. Any additional thickness (irrespective of the morphology) simply adds resistance for majority carriers to traverse between the site of photon absorption and the back contact. Thus for the 780 nm sample, even though the higher surface area allows for improved hole transport, the increased path length for majority carriers (electrons)

65 counteracts those benefits and leads to decreased performance compared to the 630 nm film. Both hole transport and electron transport thus play important roles in Ta3N5 thin films, processes that depend greatly on film thickness and morphology. In order to approach the 12.5 mA/cm2 theoretical limit for this material, an emphasis needs to be placed on designing electrodes which minimize the distance for h+ transport or employ strategies to improve the minority charge carrier diffusion length without adversely affecting e- transport.

3.5 Conclusions

Ta3N5 films of different thicknesses were synthesized via thermal oxidation and nitridation of Ta foils. All photoanodes were active under both broadband and monochromatic illumination but thicker films exhibited much higher photocurrents than would be expected based solely on enhanced absorption. Their morphological differences, specifically large changes in surface area, played a significant role in governing photoelectrode activity, leading to the conclusion that both h+ and e- transport can be fundamental limiting processes in these films. Control of the morphology and thickness of the films can allow for the development of devices with improved photoconversion efficiency.

3.6 Author Contributions Blaise A. Pinaud performed all measurements and data analysis. Blaise A. Pinaud, Peter C. K. Vesborg, and Thomas F. Jaramillo participated in writing the manuscript based on this work.

3.7 Copyright Reprinted with permission from

B. A. Pinaud, P. C. K. Vesborg, and T. F. Jaramillo, “Effect of Film Morphology and

Thickness on Charge Transport in Ta3N5/Ta Photoanodes for Solar Water Splitting”, Journal of Physical Chemistry C 116 (30), 15918-15924 (2012).

Chapter 4 Development of a Flexible Platform for Studying Electrical, Optical, and

Photoelectrochemical Properties of Ta3N5

4.1 Abstract In this chapter, we explore various substrates and synthetic methods to develop a sample architecture that yields a uniform Ta3N5 film morphology that allows for fundamental studies of this material. More specifically, the goal is to utilize such a platform to study the optical, electrical, and photoelectrochemical properties of Ta3N5. Initial tests revealed that the ideal stack composition was one composed of a fused silica substrate with an e- beam evaporated Pt back contact and Ta adhesion layer. Ta metal of controlled thickness was then evaporated and converted to the nitride phase, without affecting the conductivity of the Pt contact. Omitting the Pt contact allowed for optical and electrical properties to be measured directly. Supported Ta3N5 films ranging from 25 – 193 nm thick were synthesized and imaged, revealing very uniform, crack-free films. This result was confirmed by assessing the relative electrochemical surface area by measuring the capacitive current. While there were clear trends in the photoactivity as a function of thickness, the rapid photodegradation of the Ta3N5 made these measurements very time- sensitive. It was also discovered that the underlying Pt is electrochemically accessible leading to significant dark current. These limitations are discussed in the context of finding a way to mitigate them in future development efforts.

4.2 Introduction As clearly demonstrated in the work detailed in the previous chapter, drawing trends in photoelectrode performance as a function of a given parameter (e.g. thickness) is difficult if other factors (e.g. non-uniform morphology) convolute the results. Thermal oxidation and nitridation of Ta foils is the simplest method for synthesizing photoanodes of this

67 material but this synthetic route has several drawbacks. It is very difficult to tightly control the film thickness; tuning the oxidation conditions is somewhat effective but there is considerable spatial variation across the film. Use of oxidation time and temperature to control film thickness also restricts the ability to optimize the heat treatment conditions to minimize thermal stresses and produce smooth, crack-free films. For example, the

Ta3N5/Ta foil samples discussed in the previous chapter were synthesized by inserting the Ta foil at temperature but it is hypothesized that less cracking would result if the sample were slowly ramped up to temperature and back down. While electrophoretic deposition of a Ta3N5 powder enables slightly better control of film thickness by controlling the deposition time, the resulting films are not dense and require several post-treatment steps 31 to improve charge transport. Furthermore, the structure of Ta3N5 grown on Ta foils does not lend itself to certain types of measurements. It is highly desirable to have a transparent substrate for carrying out absorption measurements in transmission mode and a non-conductive support is required for facile characterization of the electrical properties by four point probe or via measurement of the Hall effect. The ability to synthesize uniform films of different sizes is also key; while an area of 1 cm2 may be sufficient for photoelectrochemical testing, larger samples are often necessary for resistivity measurements via four point probe to exclude edge effects. The vast majority of studies of Ta3N5 have focused on developing higher efficiency electrodes rather than fundamental studies of the material properties, likely due to the limitations imposed by conventional sample architectures. Consequently, the next goal of this research was to develop a sample architecture that would facilitate systematic studies of the physical and photoelectrochemical properties of Ta3N5.

4.3 Experimental Methods

4.3.1 Synthesis and Physical Characterization

Thin films of Ta3N5 were synthesized on several different metal (Ti, Ta, stainless steel) and glass (fused silica with Cr or Ta adhesion layer) substrates with a Pt contact. Metal foils were all high purity (> 99.95% metals basis) and used without polishing or surface modification. Fused silica (Chemglass, 75 x 25 x 1 mm slides) were diced to a size of 1

68 cm x 1.2 cm. All substrates were cleaned by sequential sonication for 30 min in acetone, isopropanol, and Milli-Q water and dried under Ar. The thickness of e-beam evaporated (Innotec, ES26C) layers was monitored in-situ via a quartz crystal microbalance. The actual thickness of each layer was measured post-deposition by profilometry (Veeco Dektak 150) on samples of fused silica masked with aluminum foil strips during deposition. Both oxidation and nitridation heat treatments were carried out in a tube furnace (Mellen Company SC13R, one zone) and the specifics are given in detail for each sample as they appear in the text. Scanning electron microscopy (SEM, FEI Magellan 400 XHR, 5 kV) was used to assess the surface morphology of the resulting nitride films. X-ray diffraction (XRD, Phillips PANanalytical X’Pert Pro) scans between 2 = 15° – 45° with Cu Kα radiation (λ = 1.54184 Å) were collected to confirm that crystalline

Ta3N5 was formed.

4.3.2 Electrical Characterization

The sheet resistance of the Pt back contact before and after heat treatment was assessed by four point probe (Signatone, S-302-4). The potential was swept between -0.2 to 0.2 V at 10 mV/s and the sheet resistance was calculated from the average slope of the anodic and cathodic regions. A minimum of three measurements was taken on each sample and no geometric correction factor was applied. The electrical resistance of Ta3N5 was evaluated by using the same four point probe method described above. Hall effect measurements (Ecopia, HMS-3000) were also carried out using the Van der Pauw calculation method. Contacts were made on the four corners of the sample by pressing small pieces of In shot (Alfa Aesar, 99.99% metals basis) into the film.

4.3.3 Electrochemical Characterization

All electrochemical measurements were carried out in a standard three electrode configuration in the custom made Teflon compression cell described in Appendix B. The 2 working electrode was a Ta3N5 film with an exposed area of 0.5 cm , the counter electrode was a clean Pt wire, and the reference electrode was Ag/AgCl (4 M KCl) calibrated to the reversible hydrogen electrode (+0.958 V vs. RHE). The electrolyte was 0.1 M KOH (pH 12.9) purged with Ar throughout testing. All voltammetry experiments

69 were controlled with a potentiostat (Bio-Logic VSP). The relative electrochemical surface area was assessed by measuring the capacitive current in the dark in the region from -0.05 – 0.25 V vs. Ag/AgCl at scan rates of 25, 50, 75, 100, 200, and 300 mV/s. This region was selected as there are no Faradaic redox features present. The current stabilized by the fourth cycle and the average value of the anodic and cathodic currents at potentials of 0.2 and 0 V vs. Ag/AgCl, respectively, was used in the calculation. The relative area was determined with respect to the thinnest sample in all cases as the intrinsic capacitance of Ta3N5 has not been measured.

Photoelectrochemical experiments under illumination were carried out using a 1000 W Xenon lamp (Newport) equipped with a water filter and a neutral density filter (Newport, optical density of 0.3 at 632.8 nm). The intensity of the illumination was measured by placing a cosine corrector (Ocean Optics, 3900 m diameter) directly beneath the assembled compression cell without the sample or electrolyte. The broadband illumination from  = 280 – 980 nm was measured to be 72.4 mW/cm2 (AM 1.5G = 72.5 mW/cm2) using a spectrometer (Ocean Optics, Jaz). Given the spectral differences between solar and simulated irradiation, it can be more useful to calibrate the intensity for only the above-band gap photons which will be absorbed as described in Appendix B.

For Ta3N5, which has an absorption edge near 600 nm, the intensity was measured to be 34.2 mW/cm2 (AM 1.5G = 32.2 mW/cm2) between  = 280 – 590 nm. For cyclic voltammograms, the illumination was chopped at a frequency of one chop every 5 s.

Sacrificial reagents can be used as a diagnostic tool in photoelectrochemistry but care should be taken to never report a solar efficiency based on a reaction which does not 4- result in the splitting of water into hydrogen and oxygen. In this work, the [Fe(CN)6] 3- /[Fe(CN)6] couple was used to attempt to stabilize the films to facilitate measuring the photocurrent. The concentrations in solution (pH 7) were 0.1 mM K3[Fe(CN)6] (Sigma

Aldrich, ≥ 99.0%) and 0.1 M K4[Fe(CN)6] (Sigma Aldrich, 98.5 – 102%). A small concentration of K3[Fe(CN)6] is necessary to minimize the absorption above λ = 460 nm from this yellow compound. The irradiation intensity for experiments using the ferri/ferrocyanide couple was measured to be 75 mW/cm2 for  = 280 – 980 nm.

4.4 Design of Sample Architecture

4.4.1 Potential Synthetic Routes

A wide variety of methods for producing Ta3N5 have been explored in the literature. The major requirement for the flexible platform under development was consistent synthesis of phase-pure, crystalline Ta3N5. Atomic layer deposition has been established as a 40, 41 method of synthesizing dense films of Ta3N5. The slow growth rate (~ 0.25 Å/cycle), however, makes the synthesis of thick films on the order of hundreds of nanometers time- prohibitive.40 Reactive sputtering has also been explored and makes possible the synthesis of thick, dense films of controlled thickness.34, 102 Reports vary widely on the optimal conditions and significant optimization is required to tune all parameters (e.g.

Ar/N2/O2 gas mixture) in order to obtain crystalline Ta3N5. Thermal oxidation and nitridation was selected as it remains the most flexible and straightforward synthetic method. E-beam evaporation can be used to form a very uniform Ta metal film of controlled thickness from several nm up to nearly a micron. A major benefit of this approach is the flexibility with respect to the substrate; Ta can be evaporated on and adheres well to nearly any material. Sputtering the Ta would also be possible though it can be more difficult to monitor the deposited thickness in situ.103

4.4.2 Substrate Selection

The next step in developing the new sample architecture was to select an appropriate substrate. A key requirement for the substrate is that it must be able to withstand the harsh oxidation (> 550°C in air) and nitridation (> 800°C in NH3) conditions required to synthesize high quality, crystalline Ta3N5. Not only must the material maintain its structural integrity, but the substrate must also retain sufficient conductivity after heating to transport the majority charge carriers (electrons) along the length of the sample to be collected at the back contact. Ideally, the substrate should also be transparent in the region of interest for studying the optical properties of Ta3N5 (λ < 700 nm). It is also desirable to have a very flat substrate to avoid forming rough films with vastly different surface areas. A review of the baseline activity of substrates suitable for electrochemical

71 and photoelectrochemical studies is found in Appendix A and the discussion below is limited to the most promising candidates for this particular application.

Transparent conducting oxides such as fluorine-doped tin oxide, indium tin oxide, or aluminum-doped zinc oxide meet the transparency criteria initially104 but are susceptible to nitridation which significantly reduces both their optical transmission and electrical conductivity. Pt was selected as the material for the back contact due to its exceptional inertness.105 Use of pure Pt foils would be cost-prohibitive so a secondary substrate is required to provide mechanical support. The range of substrates tested included metal foils (Ti, Ta, stainless steel) and glass substrates (fused silica) with conductive metal layers (200 nm Pt with and without a 100 nm Cr sticking layer, all e-beam evaporated). Soda lime and borosilicate glasses are not appropriate as their softening temperatures of 700°C and 820°C, respectively, are too low.106 Boron-doped diamond was also considered but was ruled out as a result of currently being cost-prohibitive for producing a large number of samples. To test the suitability of these substrates, a 200 nm Pt contact was first evaporated onto a roughly 1 cm x 1 cm piece of each. Next, a 70 nm layer of Ta was evaporated, taking care to mask off the top 2 – 3 mm to preserve an area on which to make contact for photoelectrochemical measurements. The samples were heated at 550°C in air for 1 hr to oxidize the Ta to Ta2O5. All samples were then nitrided at 900°C for 5 hrs in a pure NH3 atmosphere with a flow of 50 sccm NH3. Optical images of the substrates with the deposited Pt and Ta, after oxidation, and after nitridation are shown in Figure 4.1. All substrates appeared to retain their mechanical integrity during the oxidation step. Note that the films appear orange or yellow as a result of optical interference; Ta2O5 is a wide band gap semiconductor which is typically transparent or slightly blue in color. It is clear however that both the Ti and stainless steel were severally warped during the nitridation step. Furthermore, the Ti metal was attacked by the NH3, likely forming a titanium nitride. While the Ta substrate held up well, there is no way of preventing oxygen from attacking the substrate and more importantly, the Ta foils would need to be polished to form a flat film. The films on fused silica fared much better than the Ti or stainless steel. However, the Pt appeared to migrate or delaminate at the temperature of 900°C when no adhesion layer was present, as evidenced by the complete disappearance of any material on the top contact portion as shown in Figure 4.1.

Figure 4.1: Fused silica, Ti foil, Ta foil, stainless steel foil, and fused silica with a Cr adhesion layer were tested as candidate substrates with a Pt contact layer. The substrates after deposition of 70 nm Ta are shown in the first row. Sister samples in the second and third rows are shown after oxidation and nitridation treatments, respectively. Based on these results, the best option identified is a conductive Pt layer deposited on a very thin sticking layer on a fused silica substrate. The Cr sticking layer was eventually replaced with Ta simply to reduce the complexity (and remove an element) from the system but this modification had no effect on the adhesion of the Pt to the underlying fused silica. The thickness of the Pt contact was also reduced from 200 nm to 25 nm. While even 25 nm of Pt is too thick to permit absorption experiments in a transmission configuration, fused silica is highly transparent throughout the visible. A major benefit to this new approach for synthesizing Ta3N5 films is that samples can be synthesized on fused silica following the same method but omitting the Pt contact. These samples can then be used for optical measurements or to study the electronic properties of the material.

It is critical for the thin Pt layer to retain sufficient conductivity to conduct electrons to the exposed portion of the back contact for collection. A 25 nm Pt layer with 5 nm Ta adhesion layer on fused silica was heat treated in air at 700°C for 1 hr followed by nitridation at 850°C for 2 hrs in NH3. The resistance of the electrode was measured ex situ by four point probe after each processing step. The initial contact had a sheet resistance of 8.2 Ω/□ and there was a slight decrease (an improvement) to 6.5 Ω/□ after

73 the annealing step in air as shown in Table 4.1. While the treatment in NH3 resulted in a slight increase in resistance to 22.1 Ω/□, this value is still adequate for charge collection. The synthesis and types of measurements which can be made on each sample type discussed to this point are summarized in Figure 4.2.

Table 4.1: Change in sheet resistance of the Pt back contact with adhesion layer after undergoing oxidation and nitridation treatments.

4.5.1 Physical Characterization

Having selected the appropriate synthetic route and substrate platform, the next step was to confirm that crystalline Ta3N5 films of controllable thickness and uniform morphology can be synthesized. E-beam evaporation was used to deposit a nominal thickness (actual thickness in parenthesis) of 10 nm (16 nm), 25 nm (30 nm), 50 nm (49 nm), 75 nm (72 nm), and 100 nm (92 nm) of Ta metal on a 25 nm Pt layer with 5 nm Ta adhesion layer (total actual thickness of contact layer was 32 nm) on fused silica substrates. All samples were oxidized by ramping to 700°C in 20 sccm O2 + 80 sccm Ar at 10°C/min and holding for 1 hr before purging with 100 sccm Ar. Next, the samples were ramped at the same

74 rate to 850°C in 100 sccm Ar before nitridation for 8 hrs in 50 sccm NH3. Finally, samples were cooled under Ar atmosphere following the same ramping procedure. The final nitrides had thicknesses of 25, 50, 101, 147, and 193 nm as measured by profilometry. SEM micrographs of all samples are shown in Figure 4.3.

Figure 4.3: SEM images of Ta3N5 showing the uniform, crack-free morphology for films ranging from 25 – 193 nm thick. The inset optical images highlight the high spatial uniformity across the substrate. The films are very flat and uniform with clearly visible grains. The use of slow temperature ramps for the oxidation and nitridation steps significantly reduced the number of cracks. Only the thinnest sample appears different; the bright spots in the image may be due to some exposed Pt and there appear to be discontinuities in the film. The inset optical images reveal the high spatial uniformity of the films across the whole sample area. All films were confirmed to be crystalline Ta3N5 with no impurity phases based on the x-ray diffractograms shown in Figure 4.4.

Using the full-width half maximum from the peaks at 17.4°, 31.6° and 35.1° and a shape factor of 0.9, the Scherrer equation99 was used to compute the grain size which is shown in the inset of Figure 4.4. For the four thicker films, the grain size was relatively constant between 16.4 – 18.6 nm. Once again the thinnest film was an outlier with a less than 10 nm grain size. The initial metal thickness for this film was only 16 nm and it is possible that the layer was not continuous which led to the smaller grains and different morphology.

4.5.2 Electrical Characterization

There are very few reports in the literature on the electrical properties of Ta3N5 materials. A major benefit to the synthetic approach described in this chapter is that it is easily modified to produce films on a non-conductive, transparent support simply by omitting the Pt contact layer. The majority carrier (electron) transport can thus be probed directly.

While minority carrier (hole) transport may be limiting in Ta3N5, electron transport is still very important. Carriers photogenerated at the surface of the film must travel the entire thickness (often hundreds of nm or even microns) to the back contact. Two different methods were employed to measure the electrical properties. The first is a standard four point probe measurement which is straightforward but yields only a value for bulk resistivity. Conductivity, the inverse of resistivity, is actually the product of the carrier concentration and mobility and it is useful to know which quantity may be limiting.107 Hall effect measurements are a more powerful tool which can give explicit information about both the carrier concentration and mobility. However, the following conditions are imposed on the sample for an accurate measurement: a uniform, flat film with no isolated holes, a homogeneous and isotropic sample, and four contacts at the edges with at least an order of magnitude smaller area than the sample.108 It is very difficult to meet all of these criteria simultaneously. In fact, the two thinner (25 nm and 50 nm) samples could not be measured by this technique, likely due to small pinholes in the film.

To minimize edge effects for the four point probe measurement, larger (2 cm x 2.3 cm)

Ta3N5/Ta samples were synthesized and the resistivity measurements are shown below in Table 4.2.

Table 4.2: Resistivity, carrier concentration, and mobility values for Ta3N5 as measured by 4 point probe and the Hall effect. The error on each value is reported as ± σ on an average of at least 3 measurements. These measurements are complemented by the Hall effect data on smaller samples (1 cm x 1.2 cm) which is also shown in Table 4.2. An average value of resistivity, carrier concentration, and mobility is reported as these are bulk material properties which should be independent of thickness. The resistivity values of 0.70 and 4.6 Ω∙cm are in relatively good agreement. In analyzing the carrier concentration and mobilities, it is useful to have a point of reference. Silicon is an excellent, high quality semiconductor which has carrier concentrations ranging from 1012 – 1020 cm-3 and mobilities on the order of ≤ 1400 cm2/V∙s, depending on doping, yielding a resistivity between 103 – 10-3 Ω∙cm.109 Near the other end of the spectrum is Al-doped Fe2O3, a semiconductor known for poor electron transport, which has a carrier concentration near 1017 cm-3 and a mobility around 1 2 -1 110 cm /V∙s which leads to a resistivity around 10 Ω∙cm. In comparison, Ta3N5 appears to have a carrier concentration on the order of 1019 cm-3 which could signal the presence of many defects. However, the mobility is very low near 10-1 cm2/V∙s. The net result is a moderate resistivity on the order of 101 Ω∙cm. It is therefore clear that to improve electron transport in Ta3N5, it is necessary to increase the carrier mobility which can be done by improving the crystallinity and decreasing the number of defects. While defects may increase the doping density, the mobility is also correlated to the doping density because scattering by the ionized defect centers decreases mobility.107 The mechanism by which defects are formed and their concentration controlled in Ta3N5 is discussed in more detail in Chapter 6.

Note that in the above discussion, the order of magnitude rather than the absolute value of each property was discussed. The standard deviation (shown in Table 4.2) of these measurements was exceptionally high and the results should therefore be considered with caution. A common source of error is a non-ohmic contact. In this case, solid state

77 current-voltage curves on two different samples, shown in Figure 4.5, confirmed that the contact formed between Ta3N5 and In exhibited the linear behavior expected of an ohmic contact.

Figure 4.5: Solid state current-voltage curves show that that In-Ta3N5 contact exhibits ohmic behavior. The high degree of deviation could be due to the slightly asymmetric sample geometry and relatively large contacts, which may require corrections to the van der Pauw method.111 Due to the relatively high resistance of the samples, the instrument was also operated near the voltage compliance limit. Further tests on a greater number of samples will be necessary to improve accuracy.

4.5.3 Photoelectrochemical Characterization

As shown in the previous chapter, the principal challenge with Ta3N5 films grown on Ta foils was the vastly different electrochemically accessible surface area for films of different thicknesses. It was therefore not possible to conclusively assess the effect of thickness on the photoactivity of Ta3N5. The electrochemical surface area was once again assessed using measurements of the capacitive current to ensure that this new

Ta3N5/Pt/fused silica platform does provide a more uniform surface area. The results for two separate sets of samples are shown below in Figure 4.6. All surface areas are reported relative to the 25 nm sample since the intrinsic capacitance of Ta3N5 is not known.

Figure 4.6: Cyclic voltammetry in the region between 0.91 – 1.21 V vs. RHE was utilized to extract the capacitive current for each sample. The relative area of all samples varies by less than a factor of three. The measurements were repeated on a set of sister samples for confirmation. There is still a slight increase in surface area as a function of thickness but rather than being over an order of magnitude, it is limited to less than a factor of three which is a significant improvement. It is not entirely unexpected that the surface area still scales at least to some extent with the thickness given the porous nature of thermally grown films.112

Having developed this platform, it is now possible to study the photoelectrochemical activity as a function of the Ta3N5 film thickness. Due to the instability of Ta3N5 discussed later, chopped cyclic voltammograms rather than continuous light and dark sweeps were carried out. The results are shown in Figure 4.7 for both the 1st and 2nd cycles. The photocurrent at three different potentials (1.03, 1.23, and 1.43 V vs. RHE) was measured for each film by subtracting the dark current from the light current and is plotted in the inset. Note that the capacitance scans shown in Figure 4.6 were collected between these two cycles. Interestingly, the trend in photoactivity is vastly different from the 1st to the 2nd cycle. For the 1st cycle, the photocurrent increases with thickness for films from 25 – 101 nm but then reaches a plateau for thicker films. As it has been established that the crystallinity, film morphology, and surface area do not vary substantially for the five films, this trend could potentially explained by differences in absorption and transport lengths.

As revealed by the optical measurements, Ta3N5 is a fairly good absorber and previous work has shown that hole transport is possibly limiting.112 These two factors combined suggest that the majority of electron-hole pairs which generate photocurrent will originate in the top portion of the film; the additional thickness of the 147 nm and 193 nm simply acts as a resistor. Given the moderate resistivity of Ta3N5, this resistance does not significantly impede current collection so the photocurrent is nearly constant. The photocurrent decreases significantly for all samples from the 1st to the 2nd cycle and the trend is now nearly linear with respect to the film thickness. Most importantly, these results highlight the need to stabilize the Ta3N5 as photodegradation while cycling is a major concern. The photogenerated holes oxidize the nitride surface, thereby converting it to Ta2O5 and releasing N2. The mechanism of photodegradation and the effect on electrode performance is discussed in more detail in the next chapter in the context of adding co-catalysts to the surface. The true effect of thickness and thus the optimal film thickness remains elusive without the ability to stabilize the photoelectrode surface.

4.5.4 Drawbacks of Architecture

The Ta3N5/Pt/fused silica architecture is a very promising approach to study the fundamental properties of Ta3N5 photoanodes. Unfortunately, this platform does exhibit

80 several limitations. The most important is the presence of exposed Pt. While no Pt is detected by x-ray photoelectron spectroscopy (XPS), the Pt back contact is clearly accessible electrochemically. Its presence was first detected when some photoelectrodes were driven to negative potentials past the hydrogen evolution potential and significant current was observed at low overpotentials which could only be attributed to Pt. XPS has a detection limit on the order of about 1% and can be further limited by the fact that the Pt is likely exposed only via pinholes; an ejected electron is more likely to interact with the adjacent Ta3N5 than exit through the pinhole to the detector. The electrolyte can however fill these holes, even if they are on the order of only a few nm wide. As a result, electrochemical reactions are driven directly on the back contact. The problem is exacerbated by the fact that Pt is one of the best known catalysts for many reactions. This 4- 3- effect can best be illustrated using a facile redox couple such as [Fe(CN)6] /[Fe(CN)6] .

The photoactivity of a Ta3N5/Pt/fused silica electrode was compared to that of a Ta3N5/Ta 4- 3- foil electrode in a solution containing the [Fe(CN)6] /[Fe(CN)6] redox couple. Both dark and light scans were collected and are shown in Figure 4.8.

Figure 4.8: Cyclic voltammograms collected in the dark on a Pt wire and in the dark and under illumination for a Ta3N5/Pt/fused silica and a Ta3N5/Ta foil sample. While the photocurrent on the Ta3N5/Ta foil sample is easily measured, the significant dark current arising from pinholes on the Ta3N5/Pt/fused silica convolutes the photocurrent measurement. The expression jphoto and accompanying arrows designate the true photocurrent. The test was repeated in the dark on a bare Pt wire to evaluate the reversible potential for the reaction. For a high quality semiconductor, it should not be possible to draw a significant current in the dark even for a facile reaction due to the typical band structure

81 of photoanodes. As expected, there is very little dark current on the Ta3N5/Ta foil sample and the current upon illumination increases significantly; the difference is an easily quantified photocurrent. There is however a significant dark current on the 4- Ta3N5/Pt/fused silica which is attributed to oxidation of [Fe(CN)6] on the exposed Pt surface. An increase in current is still observed upon illumination but it is much more difficult to identify the onset potential and quantify the photocurrent. The same type of behavior was also observed with both H2O2 and methanol as sacrificial reagents. A corollary to these observations is that sacrificial reagents cannot be used with this platform to stabilize the Ta3N5 against photodegradation for fundamental studies; the high dark activity of the very small amount of exposed Pt overwhelms the photocurrent.

4.6 Conclusions Several metal foil (Ti, Ta, stainless steel) and glass (fused silica with Cr or Ta sticking layers) substrates with Pt contacts were screened as candidates for a flexible platform for studying the fundamental material properties of Ta3N5. The Ta3N5/Pt/fused silica architecture was selected for the excellent resistance of Pt to high temperature oxidation and nitridation treatments. Another benefit to this sample type was that simply by omitting the Pt back contact, Ta3N5 films on a non-conductive, transparent support could be synthesized using the same methods, producing ideal samples for electrical and optical property measurements. The optical property measurements on such samples are discussed in detail in Chapter 6. The high uniformity and crack-free morphology of the samples was confirmed by SEM. Photoelectrochemical measurements revealed very different trends as a function of cycle number, highlighting the need to stabilize the films. Unfortunately, there is some electrochemically exposed Pt which prevents the use of sacrificial reagents. Functionalizing the surface with co-catalysts is therefore the next step in this research.

Chapter 5

Co-catalyst Functionalization of Ta3N5 Photoanodes

5.1 Abstract

Ta3N5 is a promising semiconducting material for photoelectrochemical water splitting.

However, photooxidation of Ta3N5 leads to the formation of Ta2O5 which forms a hole blocking layer on the surface. The result is a sharp drop in performance after only a short period of illumination. In this chapter, four co-catalysts (Pt, IrO2, RuO2, CoTiOx) were explored in an effort to stabilize the surface of Ta3N5 while also improving the kinetics of water oxidation. The synthetic routes for preparing the catalysts on a Ta3N5/Pt/fused silica electrode are detailed and the photoelectrochemical activity of each is assessed. It was found that the ability of a co-catalyst to prevent photodegradation is strongly correlated to the extent of coverage. A thin, conformal layer of CoTiOx yielded the highest photocurrent while a thicker layer of CoTiOx was more effective in stabilizing the photoelectrode during a 30 min test. Transients in the chopped cyclic voltammograms point to a potential barrier for hole transfer at either the semiconductor/catalyst or catalyst/solution interface.

5.2 Introduction

A major limitation of Ta3N5 as a photoanode is its susceptibility to photodegradation. It has been previously reported that photogenerated holes can react with lattice nitrogen in 33, 34, 38 the Ta3N5 film, resulting in the formation of Ta2O5 on the surface. The equation by which this process proceeds is as follows:

+ - 3 5 15 Ta3N5 + 15 h + 15 OH → /2 Ta2O5 + /2 N2 + /2 H2O

Ta2O5 is actually a highly stable, wide band gap compound and one might expect it to act as a passivating layer, preventing further photooxidation. However, the band alignment between Ta3N5 and Ta2O5 actually results in an interface with a large barrier for hole

83 transport to the photoelectrode surface. The bottom of the conduction bands of both materials are very similar in energy as both are composed primarily of Ta 5d states.28 The top of the valence band of Ta2O5 is dictated by the O 2p levels which are significantly 26 lower than the N 2p states which make up the valence band of Ta3N5. As a result, there is a significant barrier for photogenerated holes to transfer from the Ta3N5 to the Ta2O5 which is necessary in order for them to reach the surface and drive water oxidation. The diagram in Figure 5.1 illustrates a simplified picture of this process where band bending is neglected.

Figure 5.1: Upon illumination, photogenerated holes (shown in the first panel) reach the surface of Ta3N5 but photooxidize the material to Ta2O5 in addition to driving water oxidation. The band structure alignment (shown in the second panel) is such that further transport of holes to the surface is blocked. One solution (shown in the third panel) is to add a co-catalyst to stabilize the surface and improve the kinetics of water oxidation. Evidently, if the holes cannot reach the surface, they will eventually recombine and the net current being driven decreases significantly. When the Ta2O5 layer is very thin, it may be possible for holes to tunnel through but all photocurrent is eventually quenched as the

Ta2O5 grows. As O2 readily diffuses in Ta2O5, the surface layer will continue to grow as long as there is oxygen present at the electrode surface to transport.97 Even an effective purge with an inert gas (e.g. N2 or Ar) cannot inhibit the formation of Ta2O5 as oxygen is generated at the surface of the photoanode which can readily react. The photodegradation process is especially favored when the kinetics for water oxidation are slow, which is the 38 case with a bare Ta3N5 surface.

There is clearly a need to inhibit this photodegradation mechanism, both for fundamental studies and future possible widespread use of this material. Supplying a facile redox couple as discussed in the previous chapter can be useful for diagnostic studies but is not a viable solution for true water splitting. The most common approach is to functionalize the surface with a co-catalyst.31, 34, 36, 43 If such an interfacial catalyst can be developed,

84 the benefit is actually two-fold. The catalyst covers the surface and prevents reaction of the nitride with oxygen and can also significantly reduce the overpotential required for driving the oxygen evolution reaction (OER). This effect should manifest as a shift in the photocurrent onset to a more negative potential, a desired characteristic for photoanodes.

The catalyst must meet several requirements in order to be effective as shown in Figure

5.1. It must conformally coat the entire surface of the Ta3N5 and be highly transparent for λ < 600 nm to allow photons to reach the absorber layer. A high activity for the OER will reduce the overpotential and improve the efficiency of the photoanode when matched with an appropriate photocathode. Lastly, the interface formed between the catalyst and the semiconductor must promote rather than inhibit hole transfer.113 Such interfaces are not well understood but there is increasing evidence that it plays a major role in device performance.114 The effect of the catalyst on electron-hole separation can be influenced by the formation of interfacial states at the junction or changes in the band bending in the light absorber.115 Furthermore, changes in the surface dipole induced by the presence of the catalyst can shift the band edge positions of the semiconductor.116 As a result, is important to cast a wide net and consider not only the most active catalysts, but also materials which might interface well with Ta3N5.

A total of four different catalysts were considered in this study. Pt was selected for its moderate OER activity but excellent stability.57, 117 For a front-side illuminated photoanode, the Pt layer would have to be made very thin as the metal is reflective when thick. The second material of interest is IrO2 which has been previously used in 31, 34 conjunction with Ta3N5 photoanodes. Its high OER activity in both basic and acidic 118 media make it an attractive candidate. Previous work also indicates that IrO2 materials 119 57, 118 are fairly transparent. RuO2 was also selected for its high activity for the OER. The three catalysts mentioned so far all contain precious metals which are neither abundant nor inexpensive. As shown in Chapter 2 when considering the technoeconomics of water splitting, it is highly desirable to utilize low cost, earth-abundant materials. Several different cobalt oxide-based catalysts have been used with varying degrees of 36, 38, 43 success when applied to Ta3N5 photoanodes. A promising CoTiOx catalyst developed in our group was therefore the last catalyst considered. The synthesis of the

85 catalysts and deposition on Ta3N5/Pt/fused silica are described. Photoelectrochemical measurements revealed that the CoTiOx yielded the best improvement in both photoactivity and stability. This trend is in part explained by the high coverage achieved with this catalyst.

5.3 Experimental Methods

5.3.1 Ta3N5 Photoanode Synthesis

All functionalized photoanodes were synthesized identically using the Ta3N5/Pt/fused silica platform discussed in Chapter 4. Briefly, an adhesion layer of 5 nm of Ta metal followed by a 25 nm Pt back contact was e-beam evaporated (Innotec, ES26C) on clean fused silica substrates (1 cm x 1.2 cm). The top 2 – 3 mm of the sample were masked off prior to depositing 50 nm of Ta metal to leave a contact area for electrochemical measurements. The Ta metal was first converted to an oxide by heating to 700°C in 20 sccm O2 + 80 sccm Ar for 1 hr in a three zone furnace (Mellen Company, SC12.5R). The furnace was then heated to 850°C and the oxide was converted to a nitride by flowing

100 sccm NH3 for 4 hrs. Lastly, the samples were cooled in Ar. The formation of crystalline Ta3N5 was confirmed by x-ray diffraction (data not shown here but results for similar films are shown in Chapter 4) and the approximate final Ta3N5 thickness was 100 nm. Scanning electron microscopy (SEM, FEI Magellan 400 XHR, 5 kV) was used to image the morphology and coverage of all co-catalysts on the Ta3N5 compared to the bare

Ta3N5 film. The coverage and composition of the co-catalysts was assessed by x-ray photoelectron spectroscopy (XPS, PHI VersaProbe Scanning XPS) using Al Kα radiation (hν = 1486.6 eV).

5.3.2 Electrochemical and Photoelectrochemical Characterization

Electrochemical characterization of the OER activity of the catalysts was carried out in a three electrode configuration in either a compression cell or beaker. All photoelectrochemical experiments were carried out in the custom compression cell described in Appendix B. The working electrode was the supported catalyst on fluorine- doped tin oxide (FTO) or complete functionalized Ta3N5 photoelectrode, the counter electrode was a Pt wire, and the reference electrode was Ag/AgCl (4 M KCl) calibrated to

86 the reversible hydrogen electrode (+0.963 V vs. RHE). The electrolyte for all experiments was 0.1 M KOH pre-purged with N2. Scan parameters were controlled and data was collected with a potentiostat (Bio-Logic VSP). For electrochemical evaluation of baseline catalyst activities on FTO, the cyclic voltammetry data was voltage compensated for 85% of the solution resistance which was measured by impedance spectroscopy at the start of the experiment. The onset for oxygen evolution was determined by extrapolating the anodic current to a value of zero and recording the potential. The illumination source was a 1000 W Xenon lamp (Newport) equipped with a water filter and a neutral density filter (Newport, optical density of 0.3 at 632.8 nm). The irradiance was measured to be 96 mW/cm2 (λ = 250 nm – 950 nm) using an integrating sphere and spectroradiometer (International Light) calibrated against NIST standard F- 420. The light was chopped every 5 s when collecting cyclic voltammograms.

5.3.3 Pt Synthesis

The Pt was electrodeposited from a solution of 0.005 M H2PtCl6 in 0.1 M KCl following a modified procedure from the literature.120 The deposition was first optimized on fluorine-doped tin oxide (FTO, Hartford Glass Co., sheet resistance 8 Ω/□) substrates. Pulsed deposition (50 ms pulses at -2 V vs. Ag/AgCl followed by 950 ms at open circuit) yielded large cubic crystals while a constant potential deposition led to more desirable small particles with high coverage. It was found that a pre-polarization of the substrate at –1 V vs. Ag/AgCl for 1 min improved the film uniformity. The substrate was gently tapped to remove any adsorbed H2 bubbles before stepping the potential to 0 V vs. Ag/AgCl and depositing for 30 s. Deposition times between 30 s – 2 min were also tested but longer times led to very opaque films as the particles grew as shown in Figure 5.2(a). The resulting film on FTO was a pale gray color as shown in Figure 5.2(b). The deposition process was repeated on the Ta3N5 photoanode following the same methodology with a 30 s deposition time. The resulting film on the photoelectrode also had a pale gray color.

5.3.4 IrO2 Synthesis

A colloidal suspension of IrO2 particles (2 - 3 nm) was prepared following a procedure 121 adapted from the literature. The pH of a 2 mM Na2IrCl6 (Acros Organics, 99.9+% trace metals basis) aqueous solution was adjusted to ~ 12 with 400 μL of 1 M NaOH. The solution was heated at 90°C while stirring for 20 min and turned from brown to light yellow to clear over the first few minutes and finally light blue. The solution was then placed in an ice bath and cooled to 4°C prior to acidifying with 0.5 mL of 3 M HNO3. The solution rapidly turned green then dark blue. Still in the ice bath, the solution was stirred for 60 minutes and the blue/purple color deepened. The final solution was stored in the refrigerator and was stable on the order of months.

Several methods were explored for depositing the IrO2 particles. Once again, optimization of the process was carried out on FTO substrates. Previous reports suggest that simply immersing the substrate in the colloidal suspension results in the deposition 31 of catalyst. However, no IrO2 was deposited on FTO even after 16 hrs of immersion. Drop casting is a convenient method because the catalyst loading can be tightly controlled. The long drying time necessary to drive off the aqueous solvent led to very nonhomogeneous films due to the coffee stain effect which worsened when the sample was heated to speed the process. Electrodeposition was found to be an effective method of depositing a thin catalyst film of the particles. The mechanism by which the material

88 deposits on the surface is different in acidic, neutral, and basic media. Electrodeposition at high pH proceeds via electro-flocculation which leads to randomly distributed clusters of catalyst.121, 122 In contrast, if the films are prepared in acidic solution, a uniform, 123 conformal film is obtained. Ta3N5 appears to be stable even at very low pH so electrodeposition was carried out in the as-prepared acidic solution (pH ~ 1.5) diluted 1:1 with Millipore water. The substrate (FTO or Ta3N5/Pt/fused silica) was held at 1.6 V vs.

Ag/AgCl for 30 min; the current slowly increases with time as the IrO2 deposits, leading to oxygen evolution. The resulting film on FTO is mostly transparent with a blue tint as shown in Figure 5.3(a).

Figure 5.3: (a) The acidic colloidal IrO2 solution is a deep purple color while the deposited film on FTO is light blue. (b) SEM imaging reveals a conformal film is formed after electrodeposition. (c) The activity of the catalyst for the OER is excellent. Note that the tested area was not measured but was approximately 1 – 2 cm2. SEM imaging shown in Figure 5.3(b) reveals a very conformal textured film covering the grains of FTO. The catalytic activity of the film for the OER was tested and is shown in Figure 5.3(c). The overpotential for onset of the OER was approximately 270 mV on

FTO. The deposition was then repeated on the Ta3N5 photoelectrode.

5.3.5 RuO2 Synthesis

124 A vapor phase synthetic route was used to deposit a very thin, conformal RuO2 layer.

The first step was to functionalize the Ta3N5 surface by heating the sample at 95°C in a solution of 20 mM dopamine hydrochloride (Sigma Aldrich) in 30% v/v methanol in

Millipore water. This reaction proceeds on some surfaces (e.g. TiO2, Fe2O3, GaN:ZnO) but not others (e.g. FTO, SiO2). Therefore, TiO2 grown on a Ti foil was used as the test

89 substrate rather than FTO. Following the grafting step, samples were rinsed and dried under a stream of N2. Aqueous solutions of 0.5 mg/mL RuCl3 (Sigma Aldrich, Ru content

45-55%) and 1 mg/mL of KMnO4 (Sigma Aldrich, ≥ 99.9%) were prepared separately. Each sample was taped to the inside of the cover of a small glass Petri dish. Working in the fume hood due to the high toxicity of RuO4 vapor, 3 mL of each solution was dispensed into the bottom of the Petri dish and quickly enclosed with the cover and affixed sample. The solution did not come in contact with the sample at any point. The

RuCl3 reacts with the KMnO4 to form gaseous RuO4 which in turn reacts with the dopamine functional groups on the surface to yield RuO2. The loading and coverage is controlled by the overall amount of reactants supplied for the reaction. The reaction was allowed to proceed for 2 hrs. Samples were rinsed and dried in N2 before being calcined at 170°C for 2 hrs in air.

5.3.6 CoTiOx Synthesis

The CoTiOx is synthesized via a sol gel and dip coating route reported previously and expanded upon in the group.125 Precursor solutions of 1.25 M titanium isopropoxide (Sigma Aldrich, 99.999% trace metals basis) and 1.25 M cobalt acetate tetrahydrate (Sigma Aldrich, 99.999% trace metals basis) in 2-methoxyethanol (Sigma Aldrich, anhydrous, 99.8%) are prepared separately then combined and stirred overnight at room temperature. The resulting solution is deep purple with no precipitate. The sol gel solution, either as prepared or diluted with 2-methoxyethanol, was dip coated (Thor Labs, MTS50/M-Z8 translational stage) on substrates at a speed of 0.5 mm/s with a 60 s hold in solution prior to withdrawal. Initial optimization on FTO revealed that the CoTiOx film thickness was strongly influenced by the sol gel concentration as shown in Figure 5.4. In the case of the 50% v/v dilution, the film is very thick and cracked. There is a significant improvement moving to a 25% v/v dilution where the film now coats the FTO grains. As the dilution is increased to 12.5% v/v, there is no discernible film and the features in the grains of FTO are clearly visible. Later XPS analysis confirms that there is actually complete coverage of the surface by CoTiOx. The catalytic activity of the CoTiOx films was tested and is shown in Figure 5.5. There was only a slight decrease in activity as the sol gel was diluted from 100% v/v (i.e. no dilution) to 12.5% v/v.

Figure 5.5: Electrochemical activity of the CoTiOx catalyst films on FTO. A redox feature associated with a change in oxidation state of the Co is present on the first cycle but disappears on subsequent cycles. Associated with this feature is a change in color from a nearly transparent film to a deep yellow/brown as shown in the circular tested area in the optical image on the right. The overpotential for the onset of the OER was around 325 mV which is higher than that of the IrO2 catalyst but still places it among the better non-precious metal catalysts for this reaction. The redox feature which appears immediately before oxygen evolution is associated with a change in oxidation state of the Co and disappears on cycling. There is an accompanying color change in which the initially transparent film turns a deep yellow/brown. Given the highly absorbing nature of the conditioned catalyst, it will be necessary to use the thinnest layer possible to limit optical losses. For deposition on

Ta3N5/Pt/fused silica, the as prepared sol gel was diluted by 25% v/v (referred to henceforth as the thick CoTiOx) and 12.5% v/v (referred to henceforth as the thin

CoTiOx) with 2-methoxyethanol. The functionalized samples were annealed in air at 150°C for 1 hr.

5.4 Results and Discussion

5.4.1 Catalyst Coverage and Morphology

The catalysts were all deposited on identical Ta3N5/Pt/fused silica samples as described above. The morphology of the catalysts was imaged by SEM and is shown in Figure 5.6.

Figure 5.6: Morphology of the Pt, IrO2, RuO2, thin CoTiOx, and thick CoTiOx catalysts deposited on Ta3N5. An image of bare Ta3N5 is shown as a reference. There was a notable change in the morphology of the Pt when the substrate was changed from the test FTO samples to the semiconducting Ta3N5. Fewer particles nucleated on the surface and the final particles were significantly larger. The low coverage may be detrimental as the majority of the Ta3N5 surface is unprotected. The IrO2 and RuO2 thin film catalysts appeared to evenly coat the surface with just a few gaps in coverage in the latter as shown in Figure 5.6. Both CoTiOx films are very thin (the grains of Ta3N5 are still visible beneath the catalyst) but appear to cover the entire surface of the sample. The coverage of a select group of catalyst layers was quantified by XPS and the results are shown in Table 1.

Table 5.1: Elemental composition (in atomic %) of the surface of the functionalized Ta3N5 samples. Only the CoTiOx completely covers the surface of the Ta3N5. The Pt sample was not analyzed by XPS as it is clear from the SEM imaging that the majority of the Ta3N5 remains exposed. Data was not collected on the thick CoTiOx sample because it was revealed that even the thin CoTiOx completely covered the semiconductor photoanode. The information depth of XPS is on the order of several nm and while the catalyst layer thickness was not measured, all catalysts are expected to be at least 10 nm or thicker based on previous experience. Consequently, the detection of Ta or N signals incomplete coverage. Based on the XPS results, the catalyst coverage from highest to lowest is CoTiOx > RuO2 > IrO2. The ratio of Co:Ti was close to 1:1 as expected based on the stoichiometry in the sol gel solution. Both the RuO2 and IrO2 films appeared to be oxygen rich, though it is commonly accepted that there is also oxygen present at the surface of the Ta3N5 films which may explain the discrepancy.

5.4.2 Photoelectrochemical Activity of Functionalized Photoanodes

Chopped cyclic voltammograms were used to assess the improvement in photoactivity upon application of the co-catalysts to Ta3N5. The results of these measurements are shown in Figure 5.7. The baseline activity of the bare Ta3N5 is very low and decays rapidly. In fact, there is virtually no photoactivity after the first sweep to anodic 2 potentials. The maximum photocurrent achieved for the bare Ta3N5 is 0.41 mA/cm . The Pt catalyst shifts the onset to earlier potentials but the photocurrent is even lower (< 0.38 mA/cm2), possibly due to shadowing effects, and decays rapidly. This result is an indication that the Pt particles are improving the kinetics for water oxidation but since the coverage is so low, there is no overall increase in photocurrent and the degradation of

Ta3N5 to Ta2O5 is not suppressed.

Figure 5.7: Chopped cyclic voltammograms revealing the improvement in activity upon functionalization of the Ta3N5 surface with Pt, IrO2, RuO2, or CoTiOx catalyst. The highest increase in photocurrent and best improvement in stability were achieved with the CoTiOx catalysts. There is also a shift in the onset of photocurrent to more negative potentials in the case of 2 the IrO2 catalyst and the maximum photocurrent increases to 0.73 mA/cm . As shown by XPS, the catalyst does not completely cover the surface and as a result, the current once again decreases significantly upon cycling. Furthermore, there are some reports which 118, 121 suggest that IrO2 catalysts are not completely stable at high pH.

The RuO2 and CoTiOx were much more successful in protecting the Ta3N5. There is a substantial increase in the photocurrent of the sample up to a value of 1.73 mA/cm2 with

RuO2 though the onset is not quite as early as in the case of IrO2. The potential for photocurrent onset is nearly the same for the thin and thick CoTiOx though the thin film actually results in a higher achieved photocurrent of 3.23 mA/cm2 compared to the 2.84 2 mA/cm reached on the sample with a thick layer of CoTiOx.

The lower photocurrent on the Pt, IrO2, and RuO2 functionalized samples compared to those with CoTiOx could be due to the higher optical density of these catalysts. It was not possible to synthesize the RuO2 on a transparent substrate, but this catalyst is typically a deep brown color while the CoTiOx was nearly transparent (see optical image in Figure 5.5). As noted previously, the quality of the interface between the semiconductor and catalyst also plays a major role. Large transient photocurrents often point to increased recombination from interfacial effects. Consider the transient photocurrent behavior of

94 the IrO2, thin CoTiOx, and thick CoTiOx and to a lesser extent the RuO2. At a low applied bias, there is a significant transient when the light is chopped both on and off. This short- lived current spike signals that holes are accumulating when the light is on and then recombining with free electrons in the conduction band, replenished from the external circuit, when the light is turned off.126 Potential causes are a large barrier to charge transfer into the catalyst from the absorber or sluggish reaction kinetics for water oxidation. In either case, when the applied bias is increased, the driving force for both processes increases and the transients disappear.

One method to discern whether the increased recombination is due to the semiconductor/catalyst interface or slow kinetics of water oxidation is to introduce a facile redox couple in solution.126 If the latter is the case, transients should decline significantly when the barrier for hole transfer to solution is reduced. Note that the transients are much larger for the thick CoTiOx compared to the thin CoTiOx catalyst layer. The extra length over which the hole must be transported likely increases the accumulation of holes and also explains why the photocurrent is higher on the thin

CoTiOx/Ta3N5 sample.

5.4.3 Stability

Stability is an extremely important metric when evaluating different co-catalysts applied to Ta3N5. Each sample was held at 1.26 V vs. RHE under constant illumination for 30 min to track the change in photocurrent with time. The results of this stability test are shown in Figure 5.8. The photoactivities of the bare Ta3N5 and Pt/Ta3N5 decay to zero nearly immediately and the current drawn from the IrO2/Ta3N5 lasts only marginally longer. There is a sustained photocurrent from the RuO2/Ta3N5 over the course of the 2 entire experiment though it is < 0.1 mA/cm . While the CoTiOx catalysts exhibit the highest photocurrents throughout, there is still a gradual decrease over the course of 30 min. The thin CoTiOx/Ta3N5 sample exhibits a higher initial photocurrent but the rate of decrease is faster than that of the thick CoTiOx/Ta3N5. As a result, the thick

CoTiOx/Ta3N5 has the highest photocurrent after the 2.5 min mark.

Figure 5.8: Chronoamperometry at 1.26 V vs. RHE reveals that the current decays almost immediately for the samples with no catalyst or Pt and IrO2 catalysts. The photocurrent on the sample with RuO2 is fairly 2 low but both the thin and thick CoTiOx layers lead to a sustained photocurrent > 0.2 mA/cm . While current is a measure of the rate of reaction at the electrode surface, neither the cyclic voltammograms nor chronoamperometry scans give any indication as to what reaction is occurring. In the case of Ta3N5, it is very important that the reaction being driven is water oxidation rather than photodegradation of the semiconductor or catalyst.

Ideally, real time detection of O2 is necessary to confirm 100% Faradaic current efficiency for oxygen evolution. A quick calculation involving the thick CoTiOx catalyst does however provide qualitative evidence for water oxidation. The total charge passed over the course of the 30 min stability test was 492 mC. This amount of charge is one order of magnitude larger than what would be passed if the entire CoTiOx had undergone a one electron/hole reduction/oxidation.

Further direct evidence of water oxidation are the bubbles which form on the electrode surface throughout the experiment. The small jumps in the photocurrent visible in Figure 5.8 result when these bubbles detach from the surface, exposing additional surface area for reaction. Some of the decrease in photocurrent over the entire experiment may be due to bubbles which stay absorbed, masking surface area. These are clearly visible on the electrode throughout testing. In a second, shorter constant potential experiment on the thick CoTiOx/Ta3N5 sample (data not shown), the experiment was paused to flush bubbles off the electrode surface and it was found that nearly 100% of the photocurrent was recovered.

5.5 Conclusions

Four different catalysts (Pt, RuO2, IrO2, and CoTiOx) were deposited on Ta3N5/Pt/fused silica photoanodes to stabilize the surface against photooxidation. Low coverage of the Pt catalyst led to only a marginal improvement in the photoactivity and no improvement in stability. Both the IrO2 and RuO2 catalysts nearly covered the Ta3N5 surface and led to a shift in the onset of photocurrent to lower potentials but resulted in only a moderate increase in the photocurrent. However, the RuO2/Ta3N5 drew a small but sustained photocurrent over the course of a 30 min stability test indicating that this catalyst is at least partially effective in preventing the formation of Ta2O5 on the Ta3N5 surface. The 2 highest photocurrent (3.23 mA/cm ) was achieved with a thin layer of CoTiOx but a thicker layer yielded a larger sustained photocurrent during stability testing. Large transients point to potential barriers for hole transfer to the catalyst or injection into solution. Future work should focus on deposition techniques yielding conformal coverage, quantification of the Faradaic efficiency for oxygen evolution, and studying the photocurrent transient behavior to better understand the catalyst/semiconductor interface.

Chapter 6 Controlling the Structural and Optical

Properties of Ta3N5 through Nitridation Temperature and the Nature of the Ta Metal

6.1 Abstract

The development of a reliable synthetic route to produce high performance Ta3N5 photoanodes has been complicated by the large number of synthetic parameters, notably nitridation conditions. In this chapter, we focus on identifying the conditions to produce phase-pure Ta3N5 for optical measurements. A systematic study of nitridation from 850°C – 1000°C reveals that, contrary to common knowledge, nitridation temperature has little effect on the quality of the Ta3N5 produced. Rather, it is the nature of the tantalum starting material and substrate that play a key role. Ta3N5 films synthesized by thermal oxidation and subsequent nitridation of Ta thin films on inert fused silica substrates exhibit identical structural and optical properties, regardless of preparation temperature. The optical spectra collected on these samples reveal clear, distinct features that give insight into the electronic band structure. Films grown in the same manner on Ta foils, however, reveal that textured Ta2N is formed at the Ta3N5/Ta interface even at low temperature, as shown by grazing incidence x-ray scattering. Ta3N5 on Ta foils is converted to bulk Ta5N6 at 1000°C and the possible mechanisms for these phase transitions are discussed.

6.2 Introduction

Tantalum nitride (Ta3N5) is an n-type semiconductor that has received increasing attention recently as a photoanode for solar water splitting.123 Its band gap of 2.1 eV yields absorption of a large fraction of the solar spectrum and its band edge positions straddle the hydrogen and oxygen evolution redox potentials.26 Despite the near-optimal band structure, several challenges remain including stabilizing the surface against

99 photooxidation, improving surface reaction kinetics by the addition of co-catalysts, as well as optimizing charge transport and light absorption via nanostructuring.36-38, 43

Interestingly, there exists no widespread standard method to prepare Ta3N5. The range of synthetic routes explored to date includes thermal nitridation of tantalum oxide films33, 112 or powders30, 31, electrochemical anodization followed by thermal nitridation to form nanorods or nanotubes35-37, reactive sputtering34, vapor phase hydrothermal processes38, reverse homogeneous precipitation to form an oxide precursor127, and atomic layer deposition40, 41. A key step in nearly all of the above methods is a high temperature heat treatment in ammonia (NH3) to convert a precursor oxide to the nitride. The goal of the work presented in this chapter is to use well-defined sample architectures, first presented in Chapter 4, to systematically study this key synthetic step in order to gain deeper insights as to how the structural and optical properties of Ta3N5 films are affected by the nitridation temperature. We find that while temperature has little effect on the crystal grain size or absorption properties of Ta3N5, the presence of metallic Ta in the underlying substrate results in the formation of interfacial TaxNy phases even at low temperatures which extend into the bulk as the temperature increases. At a temperature of 1000°C, the reducing atmosphere triggers the formation of reduced Ta species leading to sub-band gap optical absorption. Thus the nature of the Ta starting material and substrate play key roles in Ta3N5 film quality, more so than nitridation temperature.

6.3 Experimental Methods

6.3.1 Sample Preparation

Two types of samples were utilized to probe the properties of interest as a function of nitridation temperature. The first type consists of Ta3N5 supported on fused silica

(referred to as Ta3N5/fused silica) and the second more conventional type was Ta3N5 grown on Ta foils (referred to as Ta3N5/Ta). Photoanode devices must be synthesized on a conductive back contact such as Ta for effective collection of the majority charge carriers. However, samples on the transparent fused silica substrates were used for accurate measurement of the absorption properties. Both fused silica and polished Ta foil substrates were cleaned by sequential sonication for 30 minutes in acetone, isopropanol,

100 and Milli-Q water then dried under a stream of Ar. All heat treatments were carried out in a tube furnace (Mellen Company SC12.5R, three zones) with a ramp rate of 10°C/min.

Ta3N5/fused silica samples were synthesized by e-beam evaporation (Innotec ES26C) of 10 – 100 nm Ta metal onto clean fused silica substrates (Chemglass, 75 x 25 x 1 mm slides diced into 12.5 x 12.5 mm pieces). The thickness of the as-deposited metal and the resulting nitride films were measured by two methods: (i) profilometry (Veeco Dektak 150) on fused silica samples masked with thin strips of aluminum foil, and (ii) cross- sectional scanning electron microscopy (FEI Magellan 400 XHR, 5 kV) of sister samples deposited on silicon and cleaved. Figure 6.1 reveals excellent agreement between the two methods. The measured thicknesses of the as-deposited Ta metal, as determined by profilometry, were 16, 30, 49, 72, and 92 nm.

Figure 6.1: The Ta3N5 film thickness was measured by cross-sectional SEM imaging on sister samples deposited on Si or by profilometry on a masked fused silica sample. All samples were nitrided at 900°C in NH3. Thickness values in this chapter are reported based on the cross-sectional SEM values. The Ta metal thickness plotted on the horizontal axis was measured by profilometry. The temperature in the furnace has been carefully calibrated by inserting a thermocouple in the center of the tube during a simulated run with a flow of 50 – 100 sccm Ar. The oxidation and nitridation of the Ta metal were carried out in a single run as shown graphically in Figure 6.2(a). First, the temperature was ramped to 700°C in 20 sccm O2 +

80 sccm Ar and held for 1 hr to oxidize the Ta metal to Ta2O5. The furnace was then purged with 100 sccm Ar and ramped to 850°C, 900°C, 950°C, or 1000°C and held for 8 hrs in 50 sccm NH3 before cooling to room temperature in 100 sccm Ar. The resulting nitrides prepared at 900°C had thicknesses of 17, 54, 96, 150, and 199 nm (Figure 6.1).

101

Figure 6.2: Schematics showing the temperature ramp profile and gas flows during heat treatment. (a) Evaporated Ta on fused silica is oxidized and nitrided in a single run. (b) To prepare Ta3N5/Ta, all films are first oxidized at 550°C for 15 minutes to form a film of well-defined thickness. (c) Ta3N5/Ta films are all nitrided in a second step. In each case, the point at which the gas flow is changed is denoted by an arrow.

The samples used in the temperature study all had a Ta3N5 thickness of approximately 200 nm.

Ta3N5/Ta samples were synthesized from Ta foils (20 x 10 x 0.5 mm, Alfa Aesar, 99.95% metals basis excluding Nb) mechanically polished to a mirror finish. To produce Ta3N5 films of similar thickness (< 1000 nm) as the aforementioned Ta3N5/fused silica samples, a slightly different thermal treatment was used, consisting of separate oxidation and nitridation steps as shown in Figure 6.2(b)-(c). The temperature was ramped to 550°C in

100 sccm Ar followed by a 15 min oxidation in 20 sccm O2 + 80 sccm Ar prior to cooling to room temperature in 100 sccm Ar. By using a lower oxidation temperature than with the Ta3N5/fused silica samples, the thickness of the Ta3N5/Ta sample could be controlled by kinetically limiting the oxidation rate. X-ray diffraction (not shown) shows that crystalline Ta2O5 is formed at both 550°C and 700°C, thus we do not expect this

102 difference in oxidation temperature to have a profound effect on the quality of the Ta3N5 film ultimately produced. Next, the temperature was ramped to 900°C, 950°C, or 1000°C in 100 sccm Ar and held for 8 hrs in 50 sccm NH3 before cooling to room temperature in 100 sccm Ar.

6.3.2 X-ray Analysis of Crystallinity

A PANanalytical X’Pert Pro Materials Research Diffractometer was used to carry out x- ray diffraction (XRD) scans under specular reflection conditions with Cu Kα radiation (λ = 1.5419 Å). Depth-dependent structural information was obtained employing grazing incidence x-ray scattering (GIXS) at the SLAC National Accelerator Laboratory on beamline 1-5 with an x-ray energy of 20 kV (λ = 0.620 Å). The MAR345 image plate detector distance was calibrated using a NIST LaB6 standard 660b. Data analysis was carried out using the SSRL’s in-house WxDiff software package developed by Dr. Stefan Mansfeld. The following International Centre for Diffraction Data (ICDD) reference spectra (based on incident Cu Kα radiation) are shown as references: powder diffraction file (PDF) cards 00-004-0788 for Ta, 01-071-0178 for TaON, 01-079-1533 for Ta3N5,

00-050-1176 for Ta4N5, 01-075-0628 for Ta5N6, 00-049-1283 for cubic TaN, 00-039-

1485 for hexagonal TaN, and 00-026-0985 for Ta2N.

6.3.3 Optical Absorption Measurements

Absorption data were collected using a UV-vis-NIR spectrophotometer (Varian Cary 6000i) in transmission mode or inside a custom-fitted integrating sphere (8” AdaptaSphere, Labsphere Inc.) with illumination from a 1000 W Xe lamp (Newport).

6.4 Results and Discussion

6.4.1 Effect of Temperature on Ta3N5 Crystallinity

Variations in crystallinity can be manifested in numerous ways, for example as different crystal phases, grain sizes, or fraction of crystalline material. The influence of these parameters on the photoelectrochemical performance of a semiconductor is wide-ranging. For example, impurity phases or crystallite grain boundaries can introduce trap states which act as recombination centers. A low crystalline Ta3N5 fraction also has

103 implications for light absorption. To examine the effect of crystallinity in these films, as a first step we compare Ta3N5/fused silica films synthesized at temperatures of 850°C, 900°C, 950°C, and 1000°C. The x-ray diffractograms shown in Figure 6.3(a) reveal that all films are phase-pure orthorhombic Ta3N5.

Figure 6.3: X-ray diffractograms in (a) of Ta3N5 supported on fused silica prepared at 850°C, 900°C, 950°C, and 1000°C in NH3 reveal nearly identical crystal structures with no impurity phases, regardless of the temperature. Grain sizes in (b) estimated from the Scherrer equation reveal a consistent crystallite size of approximately 17.5 nm. The grain sizes in Figure 6.3(b) were estimated from the Scherrer equation99 with a shape factor of 0.9 after fitting the five most intense peaks (2θ = 17.4°, 24.5°, 31.5°, 35.0°, 36.0°) with a Gaussian profile. Contrary to what might be expected, increasing the nitridation temperature did not lead to larger grain sizes. The grain size remains constant at 17.3 – 17.7 nm within the studied temperature range. Furthermore, the intensity of all peaks is identical for all samples. While establishing the crystalline fraction by gauging the intensity of the x-ray diffractions is qualitative at best, it is yet another piece of evidence which suggests that the materials synthesized are indistinguishable in terms of phase present, grain size, and extent of crystallinity.

6.4.2 Effect of Substrate on Ta3N5 Growth

The previous result might be somewhat surprising given the numerous reports of Ta3N5 photoanode performance varying with nitridation temperature.37, 43 The discrepancy can be explained by the effect of the supporting substrate on the Ta3N5 growth and resulting film crystallinity. While there was no discernible difference among tantalum nitride films grown on an inert fused silica substrate by a thermal nitridation process, the same is not

104 true for tantalum nitride films grown on Ta foils which are typical for the fabrication of photoelectrodes. Shown in Figure 6.4 are the x-ray diffractograms for Ta3N5/Ta samples nitrided at 900°C, 950°C, and 1000°C.

Figure 6.4: Comparison of x-ray diffractograms for films grown on Ta foils at 900°C, 950°C, and 1000°C in NH3. Peaks in the spectrum for the sample prepared at 1000°C are assigned to Ta5N6 rather than Ta3N5. Both spectra exhibit peaks assigned to a Ta2N impurity phase formed at the film-substrate interface. The sample prepared at 900°C exhibits features corresponding to crystalline orthorhombic Ta3N5 and the underlying Ta cubic metal as expected. The grain size is also plotted in Figure 6.3(b) and is the same as that of the Ta3N5/fused silica samples.

However, this Ta3N5/Ta sample shows additional features corresponding to Ta2N, unlike the case of the Ta3N5/fused silica sample nitrided in exactly the same fashion. Other researchers have previously reported the presence of this or other impurity phases.37, 43

For samples nitrided at 950°C, the features corresponding to Ta2N grow in intensity though the film retains the signatures of Ta3N5. Surprisingly, when the nitridation temperature is further increased to 1000°C the Ta3N5 features disappear entirely; the diffractogram reveals the presence of Ta5N6 and Ta2N. Note that the conversion to bulk 128 Ta5N6 is highly detrimental as this phase is metallic rather than semiconducting , hindering photoabsorption. However, a thin layer of metallic Ta2N or Ta5N6 present at the Ta interface could potentially benefit charge transport by providing a smooth, conductive transition from Ta3N5 to Ta. It is not surprising that Ta3N5, Ta5N6, and Ta2N are the primary nitride phases observed as these are the only three structures predicted to be thermodynamically stable.128

105

Clearly, significant differences are observed in forming tantalum nitride phases by thermal treatment of thin Ta films on fused silica vs. Ta foils. The Ta3N5/fused silica samples show only the Ta3N5 phase while the Ta3N5/Ta samples can contain multiple phases, particularly Ta2N. An important question is whether the Ta2N is distributed throughout the film or whether it is localized at the Ta3N5/Ta interface; previous work suggests the latter.37 Using grazing incidence x-ray scattering to profile the crystal structure as a function of depth we provide the first direct evidence of the Ta2N phase formation at the Ta3N5/Ta interface. Our results indicate that the Ta2N phase is textured, i.e. exhibiting a preferential orientation with respect to the Ta substrate which is also textured.

GIXS spectra were collected at two different depths using a 2-dimensional image plate detector to capture all angular positions of the diffracted photons (denoted χ). The results of these experiments for samples nitrided at 900°C, 950°C, and 1000°C are shown in Figure 6.5 at incidence angles of ω = 0.5° and ω = 6.0°, corresponding to approximate information depths in Ta3N5 of 350 nm and 4.4 μm, respectively. Textured phases exhibit spots only at certain values of χ while features for non-textured phases are observed as continuous lines. When only the top portion of the film is probed (ω = 0.5°), the only phases detected are Ta3N5 (900°C and 950°C) or Ta5N6 (1000°C) and neither exhibit preferential orientation as continuous lines are observed. When the entire sample (film and substrate) is probed (ω = 6.0°), additional spots emerge at numerous angles of χ signaling the presence of oriented phases. From the overlayed reference spectra it is clear that these new features correspond entirely to the Ta foil substrate and Ta2N. This data provides direct evidence that the Ta2N is present at the Ta3N5/Ta interface (900°C and

950°C) or the Ta5N6/Ta interface (1000°C).

6.4.3 Mechanism of Phase Transformations

Early work suggested successive phase transformations of Ta3N5 → Ta4N5 → Ta5N6 → hexagonal TaN → Ta2N upon heating Ta3N5 but the mechanism of these phase transformations was not probed.129

106

Figure 6.5: Grazing incidence x-ray scattering of Ta3N5/Ta films prepared at (a) 900°C, (b) 950°C, and (c) 1000°C. At ω = 0.5°, only the surface is probed and either Ta3N5 (900°C and 950°C) or Ta5N6 (1000°C) is detected. Only when the bulk is probed with ω = 6.0° do the textured features corresponding to Ta and -1 Ta2N appear. For convenience, the reciprocal lattice spacing Q (in Å ) is converted to a 2θ position assuming Cu Kα radiation shown on the upper horizontal axis.

107

We offer the following three possible mechanisms for the formation of the Ta-rich phases, illustrated schematically in Figure 6.6: (i) reduction of the Ta5+ by a reducing atmosphere during nitridation, (ii) conversion of sub-stoichiometric oxides, or (iii) diffusion of metallic Ta atoms from the substrate into the Ta3N5 films.

Figure 6.6: Schematic depicting the three proposed mechanisms for the formation of Ta-rich phases during the nitridation process. Hydrogen formed from the decomposition of ammonia at high temperature (i) could reduce or prevent the formation of stoichiometric Ta3N5. This process has been ruled out as no Ta2N or Ta5N6 is detected when fused silica is used as the substrate. The nitridation of sub-oxides (ii) at the Ta2O5/Ta interface could result in lower valence nitrides. During heat treatment, the Ta atoms from the substrate could diffuse into the film (iii). At higher temperatures, this process would become more facile which could lead to bulk Ta5N6 formation.

43, 130 The decomposition of NH3 at high temperature into N2 and H2 has been proposed in 5+ which case the resulting H2 could then reduce the Ta . As discussed below, this mechanism is plausible for the formation of defects in the film but our data suggest that it is not the process by which crystalline Ta2N and Ta5N6 phases are formed. Otherwise, all

Ta3N5 films, regardless of substrate, should be transformed at a given temperature.

However, for Ta3N5 films on fused silica there is no trace of Ta2N or Ta5N6 in the x-ray diffractograms, even at 1000°C.

We next consider the second proposed mechanism, the conversion of sub-stoichiometric 2- oxides. Conversion of Ta2O5 to Ta3N5 proceeds by substitution of three O with two 3- 127 N . XRD analysis of the oxides prior to nitridation indicates that a bulk Ta2O5 is indeed formed. However, lower Ta valence oxides may exist at the Ta/Ta2O5 interface due to incomplete oxidation97, 131 which could lead to the formation of the interfacial

Ta2N layer even at low temperatures. While this mechanism cannot be conclusively ruled

108 out for the formation of Ta2N (though no crystalline sub-oxides were detected by XRD), it cannot explain the bulk reduced phases. Our results suggest that the third and final mechanism for the formation of Ta-rich bulk phases such as Ta5N6 is the most probable.

As the temperature increases, the Ta atoms directly underneath the Ta3N5 phase become more mobile and can diffuse further into the nitride film towards the Ta3N5 surface. This process could also exhibit a time-dependence, particularly with certain morphologies (e.g. nanowires vs. film), with longer times leading to phases that are more reduced and/or thicker; investigating time-dependence was outside the scope of the studies presented herein and will be investigated in the future.

The ability for metallic Ta at the bottom interface to diffuse into the Ta3N5 structure 37 explains why the Ta2N phase has been observed here and elsewhere when Ta foils are 32 132 utilized but not with Ta3N5 powders or when more inert substrates such as Pt or fused silica are used. A metallic source of Ta is needed. Ta transport can depend significantly on a number of parameters such as sample morphology (e.g. nanowires vs. film), temperature, and nitridation time (2 hrs vs. 8 hrs). These factors could explain why researchers have reported a wide range of optimal (i.e. higher photocurrent) processing conditions, e.g. a temperature of 800°C43 vs. 1000°C37, even though our work on

Ta3N5/fused silica suggests that this range of temperatures has a negligible effect on the resulting material.

6.4.4 Effect of Temperature on Ta3N5 Absorption Properties

The next step was to measure the absorption properties of Ta3N5. Accurate measurement of absorption requires analysis of all light interacting with the sample. For samples that reflect or scatter a significant fraction of light, traditional absorption spectroscopy measurements in transmission mode might not capture the true absorption characteristics. Light that is reflected or scattered is not appropriately accounted for, thus overestimating the fraction of photons truly absorbed. Placing the sample inside an integrating sphere to collect all light not absorbed by the semiconductor yields a truer measurement. Schematics shown in Appendix B highlight the differences in each measurement technique.

109

For both methods, it is imperative to collect an accurate baseline. A clean, bare fused silica substrate was used in the case of the transmission experiment. To collect the baseline with the integrating sphere, the sample being measured was rotated off-axis from the beam to avoid the primary beam of photons; this method helps account for any absorption of light reflected off the inner surface of the sphere before exiting through the detector (i.e. second and subsequent passes through the film). Absorption spectra collected by both methods are shown in Figure 6.7 for Ta3N5/fused silica films prepared at 850°C, 900°C, 950°C, and 1000°C.

Figure 6.7(a) shows the UV-vis data collected in a conventional transmission mode. Note the interference fringes at longer wavelengths which convolute the data and could be misattributed to sub-band gap absorption from defects. These are completely absent in the much cleaner integrating sphere measurement in Figure 6.7(b), facilitating identification of true optical absorption features. The onset of absorption is at 600 nm which is in agreement with the reported value in literature of 2.1 eV.26 The spectra for the samples prepared at 850°C, 900°C, and 950°C are all identical which is consistent with the results from the structure analysis that suggested the nitridation temperature does not affect the characteristics of Ta3N5. However, the sample prepared at 1000°C exhibited some sub-band gap absorption.

Figure 6.7: UV-vis absorption spectra of Ta3N5 supported on fused silica prepared at 850°C, 900°C, 950°C, and 1000°C in NH3 collected (a) in transmission mode, and (b) with an integrating sphere . Optical images inset in (b) show the gray tint in samples heated to 1000°C. Note that the interference fringes in (a) arise from specular and diffuse reflectance off the flat, uniform samples.

110

The optical images in the inset of Figure 6.7(b) show that the 1000°C sample has a gray tint unlike the sample at 950°C. It is not possible from the absorption data or the optical images to determine whether this darker color is associated with the sample surface or the bulk Ta3N5. The possibility of a carbonaceous deposit due to the high temperature processing was ruled out by using an O2 plasma to clean the sample surface and no change in absorption was observed. Previous work suggests that reducing conditions at high temperature generate reduced Ta species (Ta4+, Ta3+, etc.) and that these defects are the source of the absorption tail at longer wavelengths.132, 133 It is well-known that sputtering with Ar+ ions results in a similar reduction of transition metal centers.134 In fact, upon sputtering with Ar+ ions at 5 keV, the sample surface darkened further supporting the hypothesis that the gray color can be attributed to reduced Ta centers. While an increase in the number of defects could be detrimental, it is not necessarily the case. Doping and electron conduction in n-type nitrides and oxynitrides is due to reduced Ta species from anion defects such as N3- vacancies. Recent work has shown that increasing the number of defects in BaTaO2N by annealing in hydrogen reduces its electroresistance, thereby improving its photoactivity.135 It follows that a high quality

Ta3N5 synthesized at high temperature without impurity phases but with sufficient defects to enable charge mobility could improve performance. In fact, one of the highest reported photocurrents for Ta3N5 is one such structure synthesized at 1000°C; the long nanowire 37 geometry likely limited Ta diffusion, preserving the crystalline Ta3N5.

6.4.5 Optical Features of Ta3N5

Five Ta3N5/fused silica films of different thicknesses were prepared at 900°C in NH3 to study the absorption in more detail using a total light balance approach. Light absorption, diffuse/specular reflectance, and transmission/scattering were all directly measured and accounted for quantitatively. In this context, diffuse reflectance is defined as light scattered off the front surface while scattering refers to light scattered off the back surface. Reflectance measurements were made by placing a light sink behind the sample such that the only light reaching the detector could arise from reflection (specular or diffuse). The transmission/scattering spectrum can then be calculated as it follows that any light not reflected or absorbed must be transmitted or scattered. This data is plotted in

111

Figure 6.8(a)-(c) as a percentage of the incident photons absorbed, reflected, or transmitted. Note that the reproducibility of these measurements is excellent, even on samples synthesized in separate batches as shown in Figure 6.9.

Figure 6.8: Total light balance for Ta3N5 films of different thicknesses supported on fused silica prepared at 900°C in NH3. The light is broken down into the portion that is (a) absorbed, (b) specularly and diffusely reflected, and (c) transmitted and scattered.

112

Figure 6.9: The films of five different thicknesses were synthesized multiple times to verify the reproducibility of the optical measurements. Shown above is data for two batches synthesized and measured several months apart; there is excellent agreement in the absorption features at each thickness. Data with colored traces is shown and discussed in the main text of this chapter. Once again, the absorption edge occurs near 600 nm (2.1 eV) for all samples, consistent with the known band gap of Ta3N5. A second clear feature is observed at a wavelength of 500 nm (2.5 eV). While this feature is barely discernible in some previously reported spectra, our use of thin, uniform samples enables clear observation of this feature and how it trends as a function of thickness.

There are three possibilities for the origin of the absorption feature at 500 nm. The first is the presence of a wider band gap tantalum oxynitride impurity (such as TaON). However, this possibility is unlikely given that all features in the x-ray diffractograms are assignable to Ta3N5 and a substantial impurity would be required to generate such a strong absorption signal. The second is that it is an optical effect resulting from interference at the air/film, film/substrate, or substrate/air interfaces. If this phenomenon were the case, a shift in the wavelength of the feature would be expected as a function of film thickness; however, the feature remains at 500 nm for all five films. The last possibility is an electronic transition at this particular energy. While it is well known that optical absorption in Ta3N5 proceeds by excitation of an electron from a N 2p to Ta 5d orbital,28 there has been very little discussion of the dispersion of the energy bands and transition type (indirect vs. direct).

Density functional theory calculations for Ta3N5 from over a decade ago predicted an indirect band gap of 1.1 – 1.2 eV and a direct band gap of 1.4 eV.28 Electronic structure

113 calculations using this method generally underestimate the band gap136 but the qualitative assessment of two band gaps, one indirect and one direct with a gap ~ 0.3 eV larger, is consistent with our data. Computational methods for predicting band gaps have improved significantly since this time137, 138 and work is currently underway to perform calculations using modern theoretical methods to better understand the observed features. In any case, the clear experimental data shown in Figure 6.7(b) reveals the true absorption properties of Ta3N5 and yields crucial information necessary for the development of this semiconductor for photoelectrochemical water splitting and other fields where optoelectronic properties are important.

6.5 Conclusions

In this chapter, we have synthesized Ta3N5 thin films and examined their structure as a function of different underlying substrates and under a range of nitridation temperatures, 850°C – 1000°C, subsequently measuring optical properties for the most well-defined

Ta3N5 thin films. We have shown that varying the nitridation temperature from 850°C –

950°C does not affect the structural and optical properties of Ta3N5 synthesized from

Ta2O5 thin films on fused silica. Increasing the nitridation temperature to 1000°C results in the generation of reduced Ta species which explains the sub-band gap absorption observed in the optical spectrum for this sample. The nature of the tantalum starting material, however, plays a major role; Ta-rich phases including Ta2N and/or Ta5N6 can form at temperatures of 850°C – 1000°C when the films are grown on Ta foils. Using grazing incidence x-ray scattering methods, we have provided the first direct evidence that the textured impurity phase Ta2N is located at the Ta3N5/Ta interface and proposed that bulk Ta5N6 is formed via diffusion of Ta atoms from the underlying substrate.

Tantalum diffusion can explain the different reports of optimal Ta3N5 preparation conditions reported in the literature, as Ta diffusion will depend greatly on a number of factors including the sample morphology, temperature, and nitridation time.

As the Ta3N5 thin films on fused silica at 900°C showed no other phases present nor evidence of reduced Ta species, we synthesized a number of such phase-pure samples with controlled thicknesses ranging from 17 – 199 nm. We then measured their optical properties using a total light balance approach, which quantitatively accounts for

114 absorption, diffuse/specular reflectance, and transmission/scattering. These measurements produced the clearest optical absorption spectra for Ta3N5 reported to date, allowing for a more accurate view of the electronic band structure and photon absorption properties of this material, crucial information for developing Ta3N5 for photoelectrochemical and optoelectronic applications.

6.6 Author Contributions Blaise A. Pinaud performed all measurements and data analysis. Arturas Vailionis provided guidance and data analysis assistance for the GIXS experiments. Blaise A. Pinaud, Arturas Vailionis, and Thomas F. Jaramillo participated in writing the manuscript based on this work.

115

116

Chapter 7

Design of Core-shell Ta-Ta3N5 Architecture for High Efficiency Solar Water Splitting

7.1 Abstract

The work detailed in previous chapters has identified the major limitations of Ta3N5 as a photoanode material, notably poor charge transport. One way to overcome this limitation is to design high aspect ratio Ta-Ta3N5 core-shell photoelectrodes. In these structures, light absorption and charge transport length scales are decoupled; thin Ta3N5 films grown on a conductive Ta core enable efficient carrier collection while the high aspect ratio ensures sufficient material to absorb all incident light. The new knowledge gained regarding this material’s optical, structural, and electronic properties is now applied in this chapter to engineer this optimized Ta-Ta3N5 core-shell photoelectrode. The use of porous anodic aluminum oxide and nanosphere lithography are explored as methods to template a Ta foil with a Cr mask. Subsequent etching results in the formation of a nanostructured, porous Ta scaffold. The nitridation of amorphous oxides formed by electrochemical anodization yields poor quality Ta3N5 so thermal oxidation is employed.

Lastly, a proof-of-principle core-shell Ta-Ta3N5 photoanode is demonstrated.

7.2 Introduction

The band structure of tantalum nitride (Ta3N5) makes it a nearly ideal photoanode 26 candidate. However, the performance of even the best reported Ta3N5 photoelectrodes does not approach the theoretical limits for maximum photocurrent and onset potential.37, 139 The fundamental studies described in the previous chapters have directly probed several important material properties to elucidate the main limitations. This knowledge is now leveraged to design an advanced Ta-Ta3N5 core-shell architecture which is expected to have high efficiency. A schematic of the envisioned structure is shown in Figure 7.1 with all of the key pieces of information required for its design. There are a few reports of 35, 36 Ta3N5 nanostructures in the literature but these consist of either nanotubes or

117 nanowires.37, 42 While these are very high surface area, their main shortcoming is electron transport. The micron length scale of these structures is not consistent with the only moderate electron conductivity of Ta3N5.

Figure 7.1: Schematic of a Ta-Ta3N5 core-shell architecture consisting of a catalyst-coated thin film of Ta3N5 grown on a Ta nanostructure. The structure dimensions, catalyst loading, and synthetic conditions are dictated by the knowledge gained from previous studies. The chief advantage of the core-shell architecture is that it offers a highly conductive, metallic core for transporting electrons to the back contact. Ta metal is selected as the conductive scaffold because its work function of 4.25 eV94 should form a good ohmic contact with Ta3N5 and it allows direct growth of Ta3N5 through the facile thermal route.

It is instructive to analyze the results of the previous studies to determine the optimal structure dimensions, catalyst type and loading, and synthetic conditions. Consider first the film morphology which was discussed in Chapter 2. It was shown that thermal oxidation and nitridation lead to a high surface area, textured film which is actually beneficial for hole transport. The electrical property measurements in Chapter 4 established that while the doping density was high, the mobility of the electrons was low, leading to moderate conductivity. The hole diffusion length was not directly probed but is expected to be on the order of 10's of nm at most considering similar materials.140 A very thin film of approximately 10 – 50 nm thick is therefore deemed optimal to maximize hole collection and electrons should easily be transported over this length based on our previous measurements. A further benefit to a thinner film is that the space charge region, in which there is a driving force for the separation of electrons and holes, will extend

118 through most of the film.141-143 This property is key for ensuring a high collection efficiency of the photogenerated carriers. It is of course desired to avoid the formation of reduced nitride phases such as Ta5N6 so the nitridation temperature should be kept below 950°C, as shown in Chapter 6.

The absorption measurements detailed in Chapter 4 revealed that Ta3N5 is a relatively good absorber. Based on the calculated absorption coefficient, a depth between 240 nm (λ = 425 nm) and 540 nm (λ = 525 nm) is required to absorb 90% of the incident photons. If the thickness of the structure is on the order of 10 – 50 nm in order to optimize electron and hole transport, then a roughness factor (RF) of ~ 10 – 20 is likely needed. In order to stabilize the surface against photodegradation, a co-catalyst should also be added to the surface. Catalyst integration with a nanostructure is more complex as some deposition techniques cannot be used to conformally coat a non-flat sample. For example, the

CoTiOx catalyst detailed in Chapter 5 was found to have the highest activity and confer the best stability but the dip coating procedure is not amenable to the textured samples.

While the vapor phase deposition of RuO2 and electrodeposition of either Pt or IrO2 would be possible, these catalysts did not significantly improve the photoactivity or stability of Ta3N5. Knowing that a cobalt oxide-based catalyst which conformally coats the surface is expected to yield good results, a very simple CoOx catalyst deposited from solution is selected for use on the core-shell structure.43

Several different methods for patterning and etching the Ta metal scaffold were explored before selecting nanosphere lithography followed by plasma etching. The growth of thin layers of tantalum oxide, the precursor phase to Ta3N5, is also discussed given the challenge of controllably synthesizing thin, high quality Ta2O5 films. Finally, a proof-of- principle core-shell structure is synthesized and characterized demonstrating the potential of this type of advanced electrode architecture for photoelectrochemical water splitting.

7.3 Nanostructuring Ta Metal The fabrication of the nanostructure which will form the core requires two steps. The substrate is first patterned and then this pattern is transferred to the substrate by an etch process. While the use of a Ta metal scaffold facilitates later processing steps,

119 specifically the Ta3N5 growth, it does present some interesting fabrication challenges as Ta is not a common material processed in nanofabrication facilities.

All syntheses started with clean, mechanically polished Ta foils (10 mm x 20 mm x 0.5 mm, Alfa Aesar, 99.95% metals basis excluding Nb). The two patterning options explored were templating with a porous anodic alumina layer and later, use of nanosphere lithography. Note that the final structure could take the form of nanorods extending up from the surface or pores drilled down into the Ta foil; for a given feature diameter, both structures have the same roughness factor as they are simply the inverse of one another. The porous (rather than rod) structure was targeted only because it was easier to synthesize. Ta metal is remarkably resistant to most acids and bases which precludes the use of wet chemistry for the etching step. Reactive ion etching or ion milling is therefore necessary to etch Ta.

7.3.1 Templating via Porous Alumina

The formation of a well-ordered network of pores during anodization of alumina has been extensively studied.144, 145 The applied voltage and electrolyte dictate the pore diameter and the interpore spacing while the anodization time controls the pore depth.146 The pore size can further be tuned by a pore widening etch in acid.147 Typically, the ordering of the pores is improved by a pre-anodization and oxide etch step. The use of an anodic aluminum oxide (AAO) template for patterning a Ta foil is illustrated in Figure 7.2. Approximately 700 nm Al was sputtered (Metalica, custom-built) onto clean Ta substrates.

120

Initial experiments using an oxalic acid electrolyte resulted in very small pores so it was replaced with phosphoric acid.148 The anodizations were carried out in a solution containing 1% phosphoric acid, 10% methanol, and 89% Millipore water. The methanol is necessary to lower the freezing point of the electrolyte as experiments are run in a beaker set in a chiller bath (Thermo Fisher, Isotemp 3016S) at -4°C. Contact to the substrate was made with a wire which was subsequently protected with a chemical maskant (Caswell, Mask-It Lacquer). The two electrode electrochemical cell consisted of a Al/Ta substrate as the working electrode and a stainless steel foil as the counter electrode. Power was sourced from a high voltage power supply (Thermo Fisher, EC- 1000XL). To prevent overheating which can lead to current spikes and ‘burning’,149 the voltage was slowly stepped to the target anodization voltage of 195 V. For the pre- anodization step, the voltage profile was as follows: 1 min @ 100 V → 1 min @ 150 V → 1 min @ 185 V → 2 min @ 195 V. The oxide formed during pre-anodization was etched away in a solution of 0.2 M chromic acid and 0.4 M phosphoric acid at 60°C for 5 min. The voltage profile for the final anodization step was as follows: 1 min @ 100 V → 1 min @ 150 V → 1 min @ 185 V → x min @ 195 V, where x was varied between 12 – 22 min. Pore widening etches were carried out at room temperature in a 5 wt% solution of phosphoric acid for times varying between 0 – 150 min. Scanning electron microscopy (SEM, FEI Magellan 400 XHR, 5 kV) images in Figure 7.3 show that the pores are ordered but there is a large variation in their diameter.

Figure 7.3: SEM images of the AAO template after anodization and following pore widening etches of 45 min, 1.5 hrs, and 2.5 hrs. The evolution of the pore diameter as a function of pore widening etch time is linear as shown in Figure 7.4(a). The sample was also cleaved and imaged in cross-section which is shown in Figure 7.4(b). There are two interesting features in the pore shape. The first is that the pore tapers markedly at the top suggesting the pore diameter, measured from a

121 top-down view, overestimates the actually pore diameter deeper in the AAO. The second is that there appears to be a growth in the bottom of each pore. This growth could be composed of alumina or possibly anodized Ta from the underlying support.150

Figure 7.4: (a) Increase in pore diameter as a function of etch time for two different samples. (b) Cross- sectional view of the porous alumina template on the Ta foil. Note the conical shape of the top portion of the pore and the growth at the bottom of the pore. The next step was to etch the template and transfer the porous structure to the Ta. Milling with Ar ions (IntlVac Nanoquest) was selected for its high degree of anisotropy and ability to etch nearly any metal, including Ta. The sample was mounted on a silicon carrier wafer with silver paste. The substrate was held at 5°C and rotated at 30 rpm during milling with a beam voltage of 400 V, a beam current of 50 mA, and an accelerating voltage of 80 V. The samples were milled in 30 min increments with imaging between milling steps to determine the etch end point. The diameter of the pores was tracked and actually decreased as the template was milled as shown in Figure 7.5. In fact, the diameter decreased until it eventually corresponded with the initial diameter prior to the pore widening etch. This result is consistent with the tapered shape of the top portion of the pore shown in Figure 7.4(b). The etch rate appears to be higher near the surface; the rate at the bottom of the pore channel could be limited by inhomogeneous dissolution of alumina.147 There were some small features etched in to the Ta as shown in Figure 7.5(b). However, their diameter is much smaller than that of the original template. While the Ar ions impinge perpendicular to the template surface, it is clear from the SEM image in Figure 7.4(b) that the pore walls are not straight. Coupled with the tapered shape of the pores, this uneven milling results in ill-defined features.

122

Figure 7.5: (a) The pore diameter decreases as the template is milled and eventually coincides with the initial diameter prior to pore widening. (b) Cross-sectional SEM of milled template with only a thin layer of AAO remaining and very small pores in the Ta. Further optimization of anodization parameters may improve the template geometry but it was not possible to achieve high aspect ratio Ta nanostructures using the current method. As a result of these findings, a different patterning approach, described below, was adopted.

7.3.2 Templating via Nanosphere Lithography

Nanosphere lithography is an inexpensive, faster alternative to electron beam lithography for producing periodic arrays of particles, rods, or pores across large areas.151-153 A schematic depicting the process as applied for patterning Ta foils is shown in Figure 7.6.

Figure 7.6: Schematic illustrating the use of nanosphere lithography to pattern Ta foils. A plasma etch is used to reduce the diameter of self-assembled polymer spheres. A metal mask is then deposited over the structure and the spheres dissolved in a solvent. The metal mask is then used to etch pores into the underlying Ta foil. Either latex or silica spheres can be used but the former were selected for this application because they are readily available from commercial suppliers in sizes ranging from 30 nm to > 1 μm in diameter. All initial work was carried out with 500 nm latex spheres

123

(Duke Scientific, 5050A, 10% solids by weight) though smaller spheres will be necessary to achieve higher aspect ratios in the range of 10 – 20. There are several approaches for forming a monolayer of the spheres including self-assembly of the spheres at an air/water interface,154-156 drop casting,157 and spin coating.158 All of these techniques were tested and spin coating was the most facile, repeatable, high throughput method for producing a single monolayer. Several parameters had to be optimized including the surface preparation of the substrate, the weight % solids in the spin cast solution, as well as the spin coating speed and time. Wetting of the substrate was significantly improved by first making the Ta surface hydrophilic using a 5 min O2 plasma (Drytek Model 100, 85 W,

200 mT, 100 sccm O2). The square samples were spin coated using an inverted rotating disk electrode (Pine Research Instrumentation) or spin coater (Brewer Science, CEE 200X). The optimal loading was 100 μL/cm2 of a solution composed of 350 μL of the colloidal sphere solution, 350 μL Millipore water, and 100 μL 400:1 Methanol:Triton X. The final spin coating parameters were as follows: the speed was first ramped to 400 rpm for 10 s to spread the solution, then spun at 800 rpm – 1000 rpm for 1 min and finally, the speed was increased to 1400 rpm for 10 s to remove any excess liquid accumulated in the corners. SEM images of samples spin coated at 800 rpm, 900 rpm, and 1000 rpm shown in Figure 7.7 reveal slightly better coverage with a slower spin speed.

Figure 7.7: SEM images of a self-assembled monolayer of latex spheres on Ta foil produced at different spin coating speeds. The domain size of the ordered self-assembled monolayer is relatively small but it is not of great importance in this case. Complete coverage is key for making high roughness factor structures whereas ordering may be critical for photonics applications.

The next step was to reduce the sphere diameter. An isotropic dry etch in an O2 plasma

(85 W, 200 mT, 100 sccm O2) was carried out for times ranging from 3 – 7.5 min. The

124 actual initial diameter of the spheres was measured to be 491 ± 26.6 nm while the sphere diameters for etch times of 3, 4.5, 6, and 7.5 min were 429 ± 5.7, 380 ± 11.5, 283 ± 8.8, and 198 ± 8.8 nm, respectively. SEM images shown in Figure 7.8 reveal that the ordering is preserved for short etch times but that the positions of smaller spheres shift.

Figure 7.8: SEM images of the latex spheres after reduction of the diameter after exposure to an O2 plasma for 3, 4.5, 6, and 7.5 min. This phenomenon is a result of the pedestal formed at the point of contact with the substrate which eventually detaches.159 As will be shown later, the optimal sphere diameter largely depends on the metal mask thickness. Cr was selected for masking due to its excellent resistance to most etch chemistries. A 195 nm layer of Cr was e-beam evaporated (Innotec, ES26C) on the samples with dry etched sphere diameters of 342, 290, and 161 nm as shown in Figure 7.9(a). The Cr forms a continuous film on the larger spheres resulting in complete lift-off of the metal mask when the spheres are dissolved by sonication in toluene for 5 min as shown in Figure 7.9(b). In the case of the smaller spheres, there is deposition of Cr between the spheres to form a porous mask but the Cr also forms a complete protective shell around some spheres, preventing their dissolution in toluene. An intermediary sphere diameter is ideal for forming a reasonable template with circular pores. There is essentially a trade-off between sphere diameter and Cr thickness. A thicker mask allows longer etch times to form deeper structures but also requires the use of larger spheres, which reduces the number of pores per unit area. Based on this and other data collected during development of this process, the thickness of the Cr should be limited to roughly one half the sphere diameter to ensure complete sphere dissolution and the formation of a good template.

125

Figure 7.9: (a) SEM imaging reveals that a continuous Cr film completely covers all larger spheres while Cr is clearly deposited between the smaller spheres. (b) SEM imaging after dissolving the spheres shows that the entire Cr film is lifted off if the spheres are too large while not all small spheres are removed. An intermediary diameter of 290 nm results in a well ordered mask with circular pores. The last step is to etch the masked sample to form pores in the Ta foil. As in the case of the AAO template, a first attempt was made using Ar ion milling. A major drawback of this technique is poor selectivity. The etch ratio was quantified by evaporating Cr and Ta on Si and using cross-sectional SEM to measure the thickness as a function of etch time, as shown in Figure 7.10(a).

Figure 7.10: (a) The Cr and Ta metal thickness on Si was tracked as a function of milling time. The etch ratio is very near 1:1. (b) Given the low selectivity for Ta, the etched pattern in the Ta foil is very shallow. The relative etch rate for Cr and Ta is very close to 1, limiting the depth of the pores to approximately the thickness of the Cr mask (i.e. < 200 nm). Since much deeper features are required to achieve a roughness factor of 10 – 20, reactive ion etching was

126

160, 161 161 investigated. While plasma etching in SF6 + O2 or CF4 + O2 yielded excellent selectivity for Ta etching, the etch was highly isotropic resulting in significant undercutting of the Cr mask and the formation of pyramids rather than pores. A much more anistropic etch was achieved using a high density inductively coupled plasma (ICP) etcher with an electrode bias (Plasma-Therm, Versaline ICP). An established chlorine etch chemistry was used and the parameters were as follows: 1000 W ICP forward power, 150 W electrode forward bias, 20 sccm BCl3 + 70 sccm Cl2 + 10 sccm N2 gas flows, 7 mTorr chamber pressure, 20°C, and 4 Torr He pressure. A typical etch profile for 60 s processing is shown in Figure 7.11. The etch rate for Ta was found to be close to 740 nm/min while that of Cr is only 44 nm/min, yielding a selectivity of 16:1 for Ta. Any remaining Cr (visible on top of the Ta in Figure 7.11) was removed by a 1 min wet etch in a nitric acid and diammonium hexanitratocerate-based commercial Cr etchant (Sigma Aldrich, 651826).

Figure 7.11: Cross-sectional SEM of pores etched into a Ta foil using a Cr mask patterned by nanosphere lithography. The sidewalls are not strictly vertical but the variation in diameter from 170 – 320 nm is acceptable. In fact, the slight angle may improve light trapping within the structure. Several identical templates were fabricated and etched for times ranging between 30 – 90 s. Top-down SEM images of these three samples are shown in Figure 7.12(a).

The relative surface area of the final Ta structures compared to a flat Ta foil was probed by measuring the capacitance in the region between -0.5 and -0.1 V vs. Ag/AgCl in 0.1 M KOH at scan rates of 25, 50, 75, 100, 200, and 300 mV/s. There are no Faradaic redox features in this potential window. The average of the anodic and cathodic capacitive currents at -0.2 and -0.4 V vs. Ag/AgCl, respectively, are plotted in Figure 7.12(b).

127

Figure 7.12: (a) Top-down SEM image of three patterned Ta foils etched in a chlorine chemistry for 30, 60, and 90s. (b) The capacitive current was measured as a function of scan rate to determine the relative roughness factor of the nanostructures compared to a polished Ta foil. The structure etched for 30 s had a roughness factor of 1.9 while the 60 s and 90 s etches resulted in similar roughness factors of 2.7 and 2.9, respectively. The minimal difference between the 60 s and 90 s etch times could be caused by slowing of the etch rate as the pores become deeper and diffusion of etch products more difficult. While these roughness factors appear quite low, they are in line which what would be expected based on the initial sphere diameter and spacing as well as the etch depth. Now that the entire process is well established, it should be possible to move to smaller spheres to reach the target RF of 10 – 20.

7.4 Formation of Core-shell Ta-Ta3N5

The next step in the fabrication of Ta-Ta3N5 core-shell structures is oxidation of a thin layer of the Ta metal nanostructure. Thermal oxidation yields a high quality Ta2O5 layer but it can be difficult to precisely control the thickness and uniformity of the film. The investigation of anodization as an alternative is also discussed in this section.

128

7.4.1 TaOx Formation by Anodization

The anodization of valve metals such as tantalum in acid and salt-based electrolytes has been previously reported in the literature.162, 163 Compact oxide films can be formed under the right conditions, with the thickness tightly controlled by the applied voltage. This method of forming an oxide was first tested on flat, polished Ta foils. Initial wicking and thermal runaway issues were resolved by masking (Mask-It Lacquer) the top portion of the sample and contact and reducing the ionic strength of the electrolyte. The anodizations were carried out in a two electrode configuration with a graphite foil counter electrode in 1 mM KOH at room temperature. The applied voltage between 60 – 140 V was held constant for 20 min with a high voltage power supply. The anodized foils are a bright color which changes dramatically with thickness as a result of Bragg’s law which dictates that the optical path length is a function of the index of refraction and thickness.162 The formation of a compact oxide was confirmed by SEM as shown in Figure 7.13(a) and cross-sectional images were used to measure the thickness plotted in Figure 7.13(b).

Figure 7.13: (a) SEM image showing the compact TaOx layer grown by anodization of a Ta foil at 120 V for 20 min. (b) The oxide thickness of the uniform, brilliantly colored films increases linearly with voltage. Note that while the breakdown voltage has been previously reported as 120 V,162 films were successfully synthesized even at 140 V. All films were highly uniform over the entire foil surface. X-ray diffraction (XRD, Phillips PANanalytical X’Pert Pro) with Cu Kα radiation (λ = 1.54184 Å) was used to study the crystallinity of the oxides. The only

129 features in the spectra (not shown) of the as-synthesized oxides were ascribed to the Ta substrate. Annealing in air at temperatures of 300°C, 400°C, or 500°C for 2 hrs did not convert the amorphous oxide to crystalline Ta2O5. Higher annealing temperatures would have resulted in extensive oxidation of the bulk metal.97 The oxides were nitrided in a three zone tube furnace (Mellen Company, SC12.5R) for 8 hrs at 900°C while flowing 50 sccm NH3. The temperature was ramped and cooled at a rate of 10°C/min in 100 sccm Ar. The x-ray diffractograms for three of these films, anodized at voltages of 60, 90, and 120 V, are shown in Figure 7.14.

Figure 7.14: X-ray diffractograms for Ta foils anodized at 60, 90, and 120 V and nitrided at 900°C.

While there are peaks that are assigned to orthorhombic Ta3N5, all are shifted to lower angles indicating a lattice expansion. For the pair of peaks around 2θ = 35° there are actually two sets with one pair aligned to the reference and the other shifted to substantially lower angles. The lack of a high quality crystalline Ta3N5 was also confirmed by photoelectrochemical measurements in which these samples generated exceptionally small photocurrents. It is clear that an amorphous oxide is not readily converted to a crystalline nitride. Therefore, while electrochemical anodization allows tight control of film thickness, thermal oxidation was revisited as a means to form the thin oxide layer on the Ta nanostructures.

7.4.2 Thermal Oxidation and Nitridation

Heat treatments were carried out in a one zone (Mellen Company) or three zone (Mellen Company, SC12.5R) tube furnace. Two separate oxidation treatments were explored. In

130 the first, the nanostructured Ta samples were oxidized under a flow of 20 sccm O2 + 80 sccm Ar for 1.5 - 5 min at 550°C. The samples were heated to the target temperature in an Ar atmosphere at a rate of 10°C/min and also cooled in inert atmosphere. This process resulted in extensive oxidation of the nanostructures, likely due to the extra time required for complete removal of O2 from the tube by the Ar purge. In the second procedure, the furnace was ramped to 550°C and left open to air. The sample was then inserted at temperature for 1.5 – 3 min before being removed. While the latter results in more thermal stresses which can lead to cracking as shown in Chapter 3, the oxidation time is more tightly controlled. All samples were nitrided for 4 hrs at 900°C under a flow of 50 sccm NH3. Both the ramping and cooling were also carried out in NH3 to ensure absolutely no O2 was introduced which could further oxidize the nanostructures.

7.4.3 Co-catalyst Functionalization

43 A CoOx catalyst is loaded on the surface following a method from the literature. As explained earlier, this catalyst is selected over others explored in Chapter 5 due to the ease of application and ability to coat a nanostructured surface. The sample is immersed in 0.1 M Co(NO3)2 for 10 min then rinsed with Millipore water. This process is repeated twice to yield a thicker catalyst. While the catalyst is not visible upon imaging of the surface, the significant improvement in activity and stability (data not shown) confirms that the catalyst is present and covers most of the surface.

7.5 Characterization of Ta-Ta3N5 Photoanodes The primary technique used to verify the synthesis of a core-shell structure was cross- sectional SEM. The porous nature of Ta3N5 makes it easily distinguishable from the dense Ta metal.

7.5.1 Morphology

The key to making a core-shell structure is fine tuning of the oxidation conditions to yield a thin, uniform layer. The first samples synthesized from Ta nanostructures oxidized using the ramping to 550°C method were very thick and an example of such a film is shown in Figure 7.15(a). While these produced bright red nanostructured Ta3N5 samples, there is no conductive core for electron transport.

131

Figure 7.15: (a) Cross-sectional SEM image of a nanostructured sample with CoOx co-catalyst which has been oxidized (5 min at 550°C using ramping method) and nitrided (4 hrs at 900°C). The oxidation time was too long resulting in an oxidation and this nitridation depth larger than the structure width and even the structure depth. (b) Top-down and cross-sectional views of a true Ta-Ta3N5 core-shell nanostructure without catalyst synthesized by thermal oxidation (3 min at 550°C using insert at temperature method) and nitridation (4 hrs at 900°C). The oxidation time was subsequently shortened and the sample inserted into the furnace pre-heated to 550°C and immediately removed at the end of the oxidation time. The resulting films form a 50 – 100 nm Ta3N5 shell around a Ta core as shown in Figure 7.15(b). Note that the shell is thicker near the top of the nanostructure but tapers towards the bottom of the pore. This geometry may be caused by slower O2 transport deeper into the structure. Another concern in the synthesis of core-shell structures is the preservation of the pores (i.e. avoiding filling them during formation of the shell). Due to the difference in density of the metal, oxide, and nitride phases, there is an expansion into the pore as the metal is first oxidized. The top-down image in Figure 7.15(b) clearly shows that the pores are preserved and are roughly 200 – 300 nm in diameter, definitely wide enough for transport of reactants and gaseous products in and out of the structure. The maximum depth achieved to date is 750 μm. It is expected that the pore depth, and thus overall roughness factor, could be increased by switching to a more selective reactive ion etch utilizing fluorine chemistry. The challenge will be to optimize the etch to achieve a high degree of anisotropy.

7.6 Conclusions

The previous studies on the optical, structural, and electronic properties of Ta3N5 were leveraged to design a Ta-Ta3N5 core-shell photoelectrode. A method of templating and etching Ta foils was developed using nanosphere lithography and reactive ion etching with a Cr mask. A roughness factor of just under 3 was measured with the expectation that higher values, in the desired range of 10 – 20, can be achieved with the use of

132 smaller spheres and more selective etch chemistries. While anodization of Ta foils yields a dense, uniform oxide of tightly controlled thickness, its amorphous nature resulted in the formation of a poor quality Ta3N5. Thermal oxidation and nitridation were used to successfully synthesize a core-shell structure. Future work will focus on evaluating the photoelectrochemical performance of these structures and the fabrication of Ta scaffolds with higher roughness factors to unlock the full potential of Ta3N5 as a photoanode for solar water splitting.

133

134

Chapter 8 Conclusions and Future Directions

8.1 Conclusions There is a clear need to develop large-scale, sustainable alternatives to fossil fuels to meet increasing global energy demand. Photoelectrochemical (PEC) water splitting is a promising route to make hydrogen fuel using sunlight, the most abundant renewable resource. A technoeconomic analysis has shown that centralized hydrogen production using either particle or panel-based systems could be viable if stable, high efficiency (> 10%) PEC devices are developed.

The excellent band structure and visible light absorption of tantalum nitride (Ta3N5) have driven its study as a potential photoanode for solar water splitting. This dissertation has detailed research to probe its fundamental material properties such as crystal structure, optical absorption, and electronic conductivity in order to identify key limitations, with the ultimate goal of engineering a higher performance photoelectrode. Initial experiments with Ta3N5 films grown on Ta foils revealed that photoactivity scaled primarily with surface area rather than thickness, pointing to potential issues with hole transport in the semiconductor. To facilitate quantitative evaluation of material properties, a new device architecture composed of Ta3N5 grown on Pt/fused silica or fused silica substrates was developed. The absorption coefficient of Ta3N5 was determined for the first time and this material was found to have moderate electronic conductivity.

A systematic study of the effect of the nitridation temperature on the structural and optical properties revealed the presence of interfacial reduced nitride phases such as Ta2N when Ta3N5 is grown on Ta foils. It was proposed that bulk Ta5N6 was formed at temperatures above 1000°C as a result of tantalum migration. Lastly, the new knowledge gained from the studies covered in this thesis was leveraged to develop a Ta-Ta3N5 core- shell photoelectrode.

135

Photoelectrochemical measurements next revealed that photodegradation of Ta3N5 to

Ta2O5 impeded deeper fundamental understanding of the photoelectrochemistry. To this end, four co-catalysts (Pt, IrO2, RuO2, and CoTiOx) were explored to stabilize the surface and improve the kinetics of water oxidation, with the CoTiOx catalyst leading to the highest increase in photoactivity and stability.

8.2 Future Directions

Despite significant progress on understanding Ta3N5 photoanodes and developing improvements for this system, there remains much work to be done. A particular area of interest, with relevance not only to Ta3N5 but also other photoelectrodes, is the integration of catalysts with semiconductor absorbers. The work described in this thesis covered a small set of catalysts and deposition methods and large current transients suggested a hole transfer barrier. Other techniques, such as spray coating and photodeposition, and catalysts, such as Co-Pi and Ni or Fe oxyhydroxides, could be explored to improve the interface. There is also a need for novel characterization approaches to probe these interfaces in order to better understand how this junction affects overall performance.

It is widely accepted in the community that a tandem device composed of at least one photocathode and one photoanode will likely be needed to achieve high efficiency water splitting. While it is easier for research groups to optimize each component separately, it is nonetheless important to consider how these systems will one day be integrated. For example, most photocathodes (e.g. Si) being developed operate in acid while oxide photoanodes (e.g. Fe2O3, BiVO4) tend to be more stable in basic electrolytes. The development of an acid stable photoanode would represent a significant contribution to the field. Interestingly, Ta3N5 is fairly stable at low pH. The addition of IrO2, the only acid stable oxygen evolution catalyst discovered to date, is a very promising research avenue.

Finally, nitrides and oxynitrides make up a very broad class of materials and there are countless possible compositions and stoichiometries when binary or ternary alloys are considered. The synthetic approaches, device platforms, and rigorous measurement

136 techniques described in this thesis can be applied for the study of a host of new materials. Specifically, calculations have shown that perovskite oxynitrides of some Ta alloys (e.g.

MgTaO2N) have a suitable band structure for water splitting.

The continued pursuit of a deeper understanding of materials on the microscopic scale will enable the development of more efficient solar water splitting lab-scale devices. It is clear that we must also find a means to scale such systems to address the serious energy challenges of the future. While the challenges may be great, the enthusiasm, ingenuity, and perseverance of the people working on this problem provide hope that such technical obstacles can be overcome.

137

138

Appendix A Substrate Selection for Electrocatalysts and Photoelectrodes: Inert Potential Windows in Acidic, Neutral, and Basic Electrolyte

A.1 Abstract The selection of an appropriate substrate is an important initial step for many studies of electrochemically active materials, including Ta3N5 photoanodes. In order to help researchers with the substrate selection process, we employ a consistent experimental methodology to evaluate the electrochemical reactivity and stability of seven potential substrate materials for electrocatalyst and photoelectrocatalyst evaluation. Using cyclic voltammetry with a progressively increased scan range, we characterize three transparent conducting oxides (indium tin oxide, fluorine-doped tin oxide, and aluminum-doped zinc oxide) and four opaque conductors (gold, stainless steel 304, glassy carbon, and highly oriented pyrolytic graphite) in three different electrolytes (sulfuric acid, sodium acetate, and sodium hydroxide). We determine the inert potential window, defined as the range in which there are no redox features or background activity, for each substrate/electrolyte combination and make recommendations about which materials may be most suitable for application under different experimental conditions. Furthermore, the testing methodology provides a framework for other researchers to evaluate and report the baseline activity of other substrates of interest to the broader community.

A.2 Introduction The selection of an appropriate substrate is an important preliminary step in accurately evaluating electrochemically active materials including electrocatalysts, photoelectrodes, and electrochemical capacitors.164 The substrate is typically defined as an inert, electrically conductive support onto which a material of interest can be deposited,165 but

139 the substrate may also need to fulfill a variety of additional requirements for specialized studies. Key properties of the substrate may include optical transparency, thermal stability, mechanical strength, and chemical stability, among others. Thus, the selection of an appropriate substrate can be challenging, as an experimentalist must consider many different requirements for the substrate material, and the relevant properties will vary depending on the testing parameters.

The electrochemical reactivity of the substrate is a key criterion which is particularly important when choosing a substrate for electrochemical applications. In most cases, an inert substrate that exhibits no electrochemical activity under the testing conditions is preferred. Such a substrate facilitates a straightforward analysis because all electrochemical features can be attributed to the active material. In practice, this ideal is never perfectly attained, as the substrate almost always contributes some electrochemical features through capacitance, surface phase changes, or background electrocatalysis. In some cases, the substrate can also modify the properties of the electrocatalyst or photoelectrode material.166, 167 This type of interaction can be either beneficial or detrimental to the performance of the system, and as these interactions can be difficult to predict and control, they are not routinely desired for evaluating electrocatalysts or photoelectrodes. Thus, for the majority of electrocatalyst or photoelectrode evaluations, the best strategy is to choose a substrate which approximates an ideal inert support as closely as possible under the given testing conditions. Assessing the electrochemical reactivity of a substrate, however, can be a challenge in its own right because the observed behavior depends not only on the properties of the substrate, but also on the electrolyte, voltage range, temperature, gas purge, and other testing conditions.168

The difficulties associated with selecting an appropriate substrate are confounded by the wide array of potential substrate materials and the lack of systematic published data aimed at aiding in the selection. The electrochemical reactivity of many individual candidate substrate materials such as indium tin oxide and gold has been studied extensively, 169-177 but these studies have been performed under widely different conditions, and applying these data with the aim of selecting an appropriate substrate is not straightforward. In contrast, there are few reports about the electrochemical reactivity

140 of many other substrate materials such as aluminum-doped zinc oxide. While there have been a few efforts to address this issue over the past several decades,178 to the best of our knowledge, there has been no comprehensive and systematic experimental study of electrochemical substrate materials with the aim of developing guidelines for appropriate substrate selection.

In this work, we employ a consistent experimental methodology to examine the electrochemical reactivity and stability of several transparent and opaque conductive materials that are frequently used as substrates in the evaluation of electrocatalyst and photoelectrode materials. We evaluate three transparent conducting oxide substrates (indium tin oxide, fluorine-doped tin oxide, and aluminum-doped zinc oxide) and four opaque substrates (gold, stainless steel 304, glassy carbon, and highly oriented pyrolytic graphite). We use testing parameters that approximate the conditions commonly employed in the evaluation of electrocatalyst and photoelectrode materials. Using cyclic voltammetry with a progressively increased scan range, we evaluate the electrochemical reactivity of each substrate in acidic, neutral, and basic aqueous electrolyte. These data reveal the potential window over which each substrate exhibits minimal electrochemical features. These results provide useful insights into the behavior of these materials and serve as a starting point for the selection of appropriate substrate materials for evaluating novel electrocatalysts and photoelectrodes.

A.3 Experimental Methods A.3.1 Substrate Preparation

All substrates were rigorously cleaned prior to testing by following the standard practice for each particular material type. An excellent review of substrates and the appropriate preparation conditions is available elsewhere.164 Indium tin oxide (ITO, Delta Technologies, 150 - 200 nm on aluminosilicate glass, 4 – 10 Ω/□), fluorine-doped tin oxide (FTO, Hartford Glass, ~ 600 nm on soda lime glass, 6 Ω/□), and aluminum-doped zinc oxide (AZO, Advanced Film Services, 1.3 μm on soda lime glass, 6 Ω/□) were cleaned by sequential sonication for 30 min each in the following solvents: soapy water, acetone, isopropanol, and Millipore water. The substrates were then dried in ambient air.

141

Gold foils (Alfa Aesar, 0.127 mm, 99.99% metals basis) were hydrogen flame annealed then soaked in 30% nitric acid overnight. After rinsing in Millipore water, the hydrogen flame annealing process was repeated. Stainless steel foils (SS, Alfa Aesar, 0.1 mm, SS 304) were soaked in 0.5 M sulfuric acid for cleaning followed by rinsing with Millipore water. The glassy carbon (GC) disk electrodes were prepared from 200 mm long glassy carbon rods (SigradurG, HTW Hochtemperatur-Werkstoffe GmbH, 5 mm diameter). These rods were processed by the Stanford University crystal shop to prepare 4 mm long pieces with the top side polished to a surface RMS roughness of less than 50 nm. The glassy carbon pieces were sonicated in Millipore water for 15 minutes prior to electrochemical characterization. The highly oriented pyrolytic graphite (HOPG, SPI Supplies, 1 mm x 1 mm) surface was prepared by freshly cleaving the crystal followed by a 3 min anodization in 0.2 M phosphate buffer (pH 7.2) at 1.65 V vs. Ag/AgCl. Such a pre-treatment is common when using HOPG to roughen the surface and provide edge sites for electrocatalyst deposition.179, 180

A.3.2 Electrochemical Characterization

Electrochemical testing of all substrates except for glassy carbon was carried out in a polytetrafluoroethylene (PTFE) compression cell setup in a standard three electrode configuration. The diameter of the exposed substrate was 8 mm corresponding to an area of 0.503 cm2. Glassy carbon testing was performed in a rotating disk electrode configuration (Pine Instruments) but without rotation. The electrode diameter was 5 mm resulting in an area of 0.196 cm2. For all testing, a Ag/AgCl (4 M KCl) reference electrode and a Pt wire or mesh counter electrode were used. The reference electrode was regularly calibrated to the reversible hydrogen electrode (RHE) in each electrolyte and the data shifted accordingly. A potentiostat (Bio-Logic VMP3 or VSP) was used for potential control and data acquisition. The pHs of the three freshly prepared electrolyte solutions were measured as follows: 0.1 M H2SO4 (pH 1.0), 0.1 M NaAc (pH 7.2 – 7.8), and 0.1 M NaOH (pH 13.0). All solutions were prepared from reagent grade chemicals without further purification. The electrolyte was purged with N2 throughout testing via a glass dispersion frit. Potentio-electrochemical impedance spectroscopy (PEIS) was employed to measure the series resistance at open circuit and compensate for 85% of the

142 iR-drop in situ. A mathematical correction for the remaining 15% was applied in post- processing of the data.

A.3.3 Testing Methodology

A progressive scan methodology was applied to examine the cathodic and anodic inert potential windows of the seven substrates in each of the three electrolytes (acid, neutral, and base). An intermediary starting potential close to +0.35 V vs. RHE (+0.10 V vs. Ag/AgCl in acid, -0.25 V vs. Ag/AgCl in neutral, and -0.60 V vs. Ag/AgCl in base) was selected to delimit the cathodic and anodic testing windows. For anodic scans, the potential was swept 100 mV positive of the starting potential at a rate of 25 mV/s and then swept back. This cyclic scan was repeated 1 to 3 times, depending on whether any redox features were observed. If a new feature emerged, the scan was repeated within the same window to observe if it changed over time or was stable. The anodic vertex potential was increased in 100 mV increments in this manner until a current density greater than 2 mA/cm2 was achieved. Using a fresh substrate, the process was repeated in the cathodic direction, again increasing the scan range in increments of 100 mV until a current density of -2 mA/cm2 was reached. Anodic and cathodic scans were performed in each of the three electrolytes such that six individual samples of each substrate material make up a complete set of data. While up to 2 mA/cm2 of current was drawn, 50 A/cm2 was defined as the cut-off at which a substrate is no longer considered inert for the purposes of this study. Inertness refers here to whether there are any redox features or background activity whereas stability refers to whether or not there are changes in the composition or properties of the electrode over time, either from being immersed in the electrolyte or from applying a potential.

A.4 Results and Discussion A.4.1 Testing Methodology

The testing procedure outlined above is shown graphically in Figure A.1. The progressive scanning technique employed in this study has many benefits. Most importantly, it facilitates the correlation of oxidative features with the corresponding reductive process and vice versa.

143

Figure A.1: Schematic showing the progressive scan testing methodology. Each substrate has a finite inert potential window under a given set of conditions, and progressive scanning allows accurate determination of this window as the substrate is not irreversibly degraded at the outset from scanning to very positive or negative potentials.

The measured circuit resistances for six of the substrates are shown in Figure A.2 as area- normalized values.

Figure A.2: Area-normalized circuit resistances for each substrate in each electrolyte. These values show a strong correlation to the type of electrolyte. While the FTO consistently exhibits a slightly higher circuit resistance, the measured value is primarily dictated by the mobility and concentration of current-carrying ions in the solution.165 The circuit resistance also depends on the distance between the working electrode and reference electrode. While this distance was approximately constant for all the electrodes tested in a compression cell setup, it was larger for glassy carbon, which was tested in a rotating disk electrode configuration. This resulted in higher area-normalized circuit

144 resistance values for GC. Since the substrates themselves represent a minimal contribution to the circuit resistance, the data was voltage-compensated for 100% of the resistance (85% in situ and 15% in post-processing). It is important to note that the conductivity of some substrates, such as the transparent conductive oxides FTO and ITO, is heavily influenced by heat treatment conditions. Both the temperature and annealing atmosphere can influence the doping density and thus resulting conductivity.181 The substrates in this study were all used as-received (except for the annealed gold foil and anodized HOPG) and no major changes were expected. When studying heat treated supported electrocatalysts, care must be taken to ensure any drop or rise in performance is due to intrinsic activity of the catalyst rather than a change in substrate conductivity.

A.4.2 Transparent Substrates

Transparent conducting oxides (TCOs) are typically degenerately-doped wide band gap (> 3 eV) semiconductors.182-184 For many common n-type substrates, free electrons in the conduction band are generated by oxygen vacancies or substitution of the host metal by 185 higher valency metal atoms (e.g. Sn in Sn:In2O3 or Al in Al:ZnO). TCOs are employed commercially in a number of solid-state devices and are also used extensively as substrates for electrochemical studies.183 Their high degree of transparency facilitates optical absorption measurements of catalysts, enables spectroelectrochemical studies, and permits both front and back-side illumination for the study of photoelectrodes. The exact synthetic route can affect their transmittance and conductivity significantly, so substrates should be selected with the appropriate specifications for a given application.186

A.4.3 Indium Tin Oxide

Indium tin oxide, the most prevalent TCO substrate, consists of a solid solution of In2O3 184 and SnO2 with 5 – 10 atomic % Sn. A thin layer is typically sputter-coated on an inert glass support.187 It is employed commercially in a number of products including displays, coatings, and solar cells.104 Further widespread use may be limited by the cost and scarcity of indium.20, 182

The results of the progressive electrochemical cycling of ITO are shown in Figure A.3.

145

Figure A.3: Electrochemical activity and inert potential range for indium tin oxide. The sweeps in the cathodic region show several significant redox features apart from the hydrogen evolution reaction (HER) in all three electrolytes. The oxidative features (denoted a in Figure A.3) appear only when sweeping towards positive potentials after a reductive current is drawn upon scanning to potentials of -0.55, -0.48, and -0.45 V vs.

RHE in H2SO4, NaAc, and NaOH, respectively. On subsequent cycles, a corresponding reduction peak (denoted b in Figure A.3) appears prior to the onset of hydrogen evolution. All features grow as the progressive scanning is extended to more negative potentials. These redox features are attributed to reduction of Sn and In to lower valence or metallic states and subsequent partial reoxidation/rereduction.164, 176, 188 Previous work has shown by chemical analysis that these changes extend at least several nm deep.188 It is also possible that some oxidized ions dissolve into solution and can plate back onto the electrode on subsequent cycles. This cycling of the redox states leads to irreversible

146 changes to the electrode which degrade its electrical and optical properties. During cycling or when held at potentials more negative than -0.55 V vs. RHE, the ITO electrode is observed to turn gray in color due to the metal cation reduction, resulting in a significant decrease in its transparency.

In contrast to the cathodic region, sweeps in the anodic region are relatively featureless. The only significant current arises from water oxidation through the oxygen evolution reaction (OER), with the earliest onset at 1.92 V vs. RHE in NaOH and the latest onset at

2.15 V vs. RHE in H2SO4. While these potential limits may appear to make ITO an ideal substrate for testing catalysts at positive potentials, there is a slow degradation of the substrate. There is a progressive decrease in water oxidation current in the NaAc electrolyte upon cycling. A similar trend, though less pronounced, is also observed in both H2SO4 and NaOH. During anodic polarization, In-O and Sn-O bonds are broken as 2- 189 lattice oxygen (O ) is oxidized, resulting in O2 evolution and dissolution of In and Sn. Elemental analysis of the electrolyte confirms the presence of both In and Sn but with a higher than expected In:Sn ratio. Under certain conditions, stable SnO2 crystallites can reform on the surface.189 The net result is an increase in the surface roughness but also an increase in the resistance of the substrate due to the SnO2 passivating layer. The rate of dissolution is strongly influenced by the nature of the ions present in the electrolyte, with ions capable of better solvating In and Sn, such as Cl-, accelerating the process.

After sweeps in both the cathodic and anodic regions in H2SO4, a clear color change of the ITO film in the tested area is visually observed. According to the Pourbaix diagram, 190 In2O3 is unstable at all potentials at pH = 1 so it is likely dissolving into solution. ITO is therefore an unsuitable substrate in this electrolyte.

In summary, while ITO has an electrochemically inert potential window extending from

-0.46 to 2.15 V vs. RHE in H2SO4, it is chemically unstable in this electrolyte, and therefore not recommended. ITO is an appropriate substrate for electrochemical studies in NaAc and NaOH across a wide region extending between -0.62 to 1.96 V vs. RHE and - 0.45 to 1.92 V vs. RHE, respectively, where the reduction of metals atoms and significant water oxidation current can be avoided. However, it should not be employed for extended stability tests if the supported film does not completely cover the ITO surface. The length

147 of time (i.e. minutes, hours, or days) over which the ITO will be sufficiently stable depends on the electrolyte concentration, the potential range, and the extent of coverage of the catalyst. Otherwise, slow leaching of Sn and In degrades the electrical properties and can lead to complete failure of the electrode.

A.4.4 Fluorine-doped Tin Oxide

Fluorine-doped tin oxide is a SnO2-based wide band gap semiconductor with fluorine doping on the order of 5 x 1020 - 1021 cm-3.186 FTO also has commercial applications, primarily in energy efficient windows.184 While it can be more challenging to synthesize, it has better mechanical and chemical durability than other TCOs and is less expensive.104, 191

The progressive cycling of FTO is shown in Figure A.4. The features associated with the cycling of FTO in the cathodic region are very similar to those of ITO. In H2SO4, the sweep is featureless until a small reductive current is observed at a potential of -0.39 V vs. RHE. Upon sweeping back to positive potentials, an oxidative peak (denoted a in Figure A.4) appears which also scales with the amount of reductive current drawn during progressive cycling to more negative potentials. The reductive feature (denoted b in Figure A.4) only grows in after the oxidative peak is observed. These redox features are once again attributed to changes in the oxidation state of Sn. A similar evolution of peaks is observed in the case of the NaAc and NaOH electrolytes but at more negative potentials. Hydrogen evolution currents of nearly 1 mA/cm2 at potentials of -1.19 and - 0.85 V vs. RHE in NaAc and NaOH, respectively, are drawn before any Sn redox features appear.

There are no significant features in the progressive cycling in the anodic region other than catalytic water oxidation, which onsets at potentials of 2.22, 2.09, and 1.73 V vs. RHE in

H2SO4, NaAc, and NaOH, respectively. There is a slight decrease, most pronounced in NaAc, in the water oxidation current with cycling. The origin of this decrease is currently unknown. The leaching of Sn from SnO2 has not been reported on long time scales.

Unlike the In2O3 in ITO, the Sn is already present as a stable SnO2 phase at all pHs and electrode potentials tested here.190

148

Figure A.4: Electrochemical activity and inert potential range for fluorine-doped tin oxide. While there is very little literature on the long-term stability of FTO at different pHs, the experiments reported herein can confirm that FTO substrates are stable on the order of at least hours.

In summary, FTO is a suitable substrate for use in all three electrolytes over a wide range of potentials. The inert region, bounded mainly by the onsets of hydrogen and oxygen evolution, corresponds to potentials between -0.39 and 2.22 V vs. RHE, -0.72 and 2.09 V vs. RHE, and -0.51 and 1.73 V vs. RHE in H2SO4, NaAc, and NaOH, respectively.

A.4.5 Aluminum-Doped Zinc Oxide

Aluminum-doped zinc oxide is another low-cost TCO option which exhibits good optical transmission. The level of Al doping is typically less than 5%.192 Properties which

149 distinguish AZO from other TCOs are its resistance to hydrogen-rich plasmas and a lower work function which makes it more suitable than ITO or FTO for use as a cathode support.193 It is however highly unstable in acid. We were unable to test the AZO in

H2SO4 because the material dissolved immediately upon immersion in the electrolyte.

The results of cycling in the other two electrolytes are shown in Figure A.5.

Figure A.5: Electrochemical activity and inert potential range for aluminum-doped zinc oxide. The AZO exhibits a similar pattern to both ITO and FTO in terms of redox features appearing in the cathodic region. The baseline is flat until a potential of -0.77 or -0.50 V vs. RHE in NaAc and NaOH, respectively, when a small reductive current begins. The oxidation feature (denoted a in Figure A.5) then appears on the sweep to more positive potentials followed by the reduction feature (denoted b in Figure A.5) on the subsequent sweep in the cathodic direction. These features are attributed to the cycling of the Zn oxidation state. Previous work has shown that Zn2+ is reduced to metallic Zn and subsequent oxidation/reduction occurs via various soluble zincate complex ions (e.g. - 2- 194, 195 Zn(OH)2, Zn(OH)3 , Zn(OH)4 ).

The primary feature of the progressive cycling in the anodic region is catalytic water oxidation, though there is a small (< 100 μA/cm2) oxidative pre-feature in the NaOH at

150

2.00 V vs. RHE which decays rapidly after the first few cycles (shown in Figure A.5 inset). The activity for water oxidation is highly unstable in both electrolytes, requiring a progressively more positive potential to draw any current. In fact, the target current of 2 mA/cm2 was not attained in NaOH due to the current falloff. ZnO is not stable in aqueous solutions at any pH and the rate of dissolution is rapid for a pH ≤ 5 or pH ≥ 11.196 Even at 2+ + - 2- neutral pH, the electrode slowly corrodes to form Zn , ZnOH , Zn(OH)3 , and Zn(OH)4 species which eventually leads to catastrophic failure.

In summary, the inert potential region corresponds to potentials between -0.77 and 2.69 V vs. RHE in NaAc and -0.50 and 1.99 V vs. RHE in NaOH. However, AZO is not suitable as a substrate in any electrolyte on any time scale relevant to electrochemical cycling, except possibly if the catalyst or photoelectrocatalyst forms a truly conformal, pin-hole free layer on the substrate to protect it. If this were the case, its primary advantages would be a low sheet resistance and a low work function.

A.4.6 Opaque Substrates

Opaque substrate materials are appropriate for evaluating electrocatalysts and photoelectrodes when back-side illumination or transmission experiments are not necessary. While there are many metallic conductors that might be appropriate choices for electrochemical substrate materials, we have chosen to analyze several in particular that may be appropriate under different experimental conditions.

A.4.7 Gold

Gold is an appropriate electrode material to consider because of its very low chemical reactivity. Gold has been called “the noblest of all the metals” 197 because of its chemical inertness. The galvanic potential of gold is very high,198 which means that it is not susceptible to corrosion. Additionally, the electrochemical behavior of gold has been studied extensively,169-171, 174, 199-206 and thus it may be easier to predict and understand the behavior of a gold electrode. A key drawback of gold is its price of around $40 - $50 per gram,207 which may make gold an impractical choice when a large amount of substrate is required.

151

The results of the electrochemical reactivity tests on gold are presented in Figure A.6.

Figure A.6: Electrochemical activity and inert potential range for gold. The sweeps in the cathodic region show few features. In each electrolyte, the only reaction observed is hydrogen evolution. The gold surface is most active for the HER in

H2SO4. In this electrolyte, the electrochemically inert potential range extends to -0.10 V vs. RHE (shown in Figure A.6 inset), and the HER activity of the gold surface remains constant over progressive cathodic cycles. The HER activity of the gold surface is lower in both NaAc and NaOH solutions, so the region with no electrochemical features is larger. In these solutions, the HER onset shifts to slightly more negative potentials with continued cycling, showing that the HER activity of the gold surface decreases slightly, possibly due to surface restructuring or poisoning adsorption of Na+ or other ions.

152

The sweeps in the anodic region on the gold surface are more complex. In each electrolyte, anodic oxidation of the gold surface precedes the oxygen evolution reaction. Detailed explanations of the gold oxidation features are provided elsewhere.169-171, 174, 199- 203, 205, 206 A series of oxidative features (denoted a in Figure A.6) corresponding to the adsorption of OH- groups and the initial oxidation of the gold surface is observed between 1.30 and 1.60 V vs. RHE in each electrolyte. These features may also arise from 206 the adsorption of other anions such as the sulfate ions in the H2SO4 electrolyte. The two peaks observed near 1.40 and 1.50 V vs. RHE in H2SO4 grow in size with continued cycling, corresponding to increasingly rapid surface oxidation on each subsequent cycle, reaching a maximum current density of around 0.4 mA/cm2, while these oxidation features only grow to around 0.15 mA/cm2 in NaAc and NaOH. In each electrolyte, the oxidative features are accompanied by corresponding reductive peaks (denoted b in Figure A.6) appearing between 1.00 and 1.20 V vs. RHE on the reverse cycle. These - 2- features correspond to the desorption of OH or SO4 ions or reduction of the gold oxide surface.205, 206 At more anodic potentials, the gold surface catalyzes the OER. After oxygen is evolved, current (denoted c in Figure A.6) corresponding to oxygen reduction is observed on the reverse sweep due to incomplete removal of O2 from the surface by the

N2 purging. This feature is especially apparent in NaOH electrolyte, with an oxygen reduction feature appearing prominently at about 0.70 V vs. RHE.

In summary, the inert potential region corresponds to potentials between -0.10 and 1.33 V vs. RHE, -0.38 and 1.44 V vs. RHE, and -0.14 and 1.29 V vs. RHE in H2SO4, NaAc, and NaOH, respectively. Gold electrodes may be appropriate for evaluating some materials at potentials less than 0.00 V vs. RHE due to the lack of any substantial electrochemical features other than the HER, though this reaction could interfere with measurements at large negative potentials, especially in acidic electrolyte. Gold may not be an ideal substrate material for studies requiring potentials higher than ca. 1.30 V vs. RHE due to the features corresponding to gold oxidation that are observed in each electrolyte and the gold’s reasonably high activity for oxygen evolution, especially in alkaline electrolyte.

153

A.4.8 Stainless Steel

Similar to gold, stainless steel is known to be chemically inert and resistant to corrosion in many types of electrolytes.168 Unlike gold, which is inert due to its high galvanic potential, the corrosion resistance of stainless steel is conferred by its passivating, chromium-rich native oxide.208, 209 Stainless steel is much less expensive than gold, an attractive feature for studies that require very large electrodes or many samples. Stainless steel also has excellent mechanical properties.210 However, since stainless steel is an alloy that contains many elements such as iron, chromium, nickel, and carbon,210 the complexity of this material increases the risk of contamination or undesirable side reactions. This difficulty is amplified by the large number of available stainless steel varieties, which may have quite different electrochemical properties.208, 209 We chose to evaluate SS 304 because it is the most widely used type of stainless steel and is considered to exhibit excellent corrosion resistance.210, 211 SS 304 is composed of iron alloyed with 18-20% chromium, 8-12% nickel, up to 2% manganese, and small amounts of carbon, phosphorus, sulfur, silicon, and nitrogen.210

The electrochemical reactivity data for our stainless steel 304 samples are presented in Figure A.7. In the sweeps in the cathodic region, the stainless steel exhibits several redox features before the onset of the HER. In NaOH, we observe an oxidation/reduction couple with peaks at 0.26 and 0.00 V vs. RHE and very small currents of less than 10 μA/cm2. This couple likely corresponds to nickel oxidation and reduction.212 The inert potential range extends to -0.43 V vs. RHE, where the onset of the HER is initially observed. The HER activity increases slightly with repeated cycling, possibly due to surface restructuring and/or the reduction of the surface oxide. In NaAc, we observe a small reductive feature at -0.04 V vs RHE that likely corresponds to native oxide reduction. This feature decreases in size with repeated cycling, but limits the cathodic inert range to 0.05 V vs RHE. The only other reductive feature corresponds to the HER, which is first observed at approximately -0.73 V vs. RHE. The HER activity again increases slightly with cycling. In H2SO4, no features are observed until the onset of the HER at -0.30 V vs. RHE. The HER activity increases substantially with cycling, and an oxidative peak (denoted a in Figure A.7) appears in the final several cycles. This feature

154 has previously been attributed to the oxidation of hydrogen atoms absorbed within the stainless steel during the HER.213

Figure A.7: Electrochemical activity and inert potential range for stainless steel 304. In the sweeps in the anodic region, the observed features generally correspond to either the OER or metal oxidation/dissolution.168, 208 In NaOH, an oxidative feature (denoted b in Figure A.7) is observed with no corresponding reductive peak, and the size of this feature decreases with cycling. This feature may be associated with either surface oxidation or the dissolution of the native oxide. Due to the size of this peak, the anodic inert potential limit occurs at 1.16 V vs. RHE, but it may be possible to use SS 304 at more positive potentials if this initial oxidative feature is unimportant for a given application. The onset of the OER is observed at 1.55 V vs. RHE. Some oxidative dissolution of iron, nickel, or other metals may also occur in this regime.168, 208 In NaAc,

155 we observe a couple with small oxidation and reduction peaks at 1.45 and 0.80 V vs. RHE, respectively (shown in Figure A.7 inset). This couple has peak currents less than 100 μA/cm2, and likely corresponds to a reversible change in the oxidation state of the native chromium oxide. The inert potential range extends to 1.34 V vs. RHE, and the onset of OER is observed at around 1.65 V vs. RHE. Some oxidative metal dissolution may also occur in the highly anodic potential regime. On the final sweeps in the negative direction after oxygen has been evolved, a reductive feature is observed near 0.35 V vs.

RHE arising from the reduction of oxygen that remains near the electrode. In H2SO4, a large oxidative peak (denoted c in Figure A.7) is observed. Prior studies of stainless steel corrosion have suggested that this feature arises from the dissolution of the chromium- rich native oxide in the transpassive (i.e. highly anodic) potential regime.214, 215 The OER wave begins around 1.85 V vs. RHE and may be accompanied by oxidative metal dissolution. A small current (denoted d in Figure A.7) corresponding to oxygen reduction is also observed on the negative sweeps after the OER in each electrolyte.

In summary, the inert potential region corresponds to potentials between -0.30 and 1.23 V vs. RHE, 0.05 and 1.34 V vs. RHE, and -0.43 and 1.16 V vs. RHE in H2SO4, NaAc, and NaOH, respectively. Stainless steel may be an appropriate substrate for use in alkaline electrolytes, where few features aside from the HER and OER are observed. It may also be a good choice for specialty applications that require high mechanical strength or a large number of metal substrates. Otherwise, the relatively small inert potential range of this material in acidic and basic electrolyte may make this substrate a less than ideal choice for many studies. We consider it especially important for researchers who wish to use stainless steel electrodes to conduct their own experiments to determine the inert potential range, because the electrochemical behavior of stainless steel may change substantially depending on the details of the exact starting material and the experimental procedures.168 For example, electrolytes containing chloride ions may result in increased 208 corrosion of stainless steel. Additionally, as observed in the H2SO4 electrolyte, cathodic polarization and hydrogen evolution can lead to hydrogen absorption, which could change the oxidative behavior of the electrode if a broad potential range is required.213, 216

156

A.4.9 Glassy Carbon

Glassy carbon has been widely used as an electrode material since its discovery in 1962.217, 218 This material consists of tangled graphite nanoribbons and possesses no long- range atomic ordering.178, 219 Glassy carbon is an ideal substrate for many electrochemical studies because it is chemically stable and electrochemically inert in a large potential window.217, 219-221 Unlike many other conductive carbon materials, glassy carbon is generally impermeable to gases and can be polished to a mirror finish.178, 220 Additionally, glassy carbon can be readily obtained in disk form for use with a rotating disk apparatus. Finally, glassy carbon may be useful for studies involving spectroscopic characterization techniques because its single element composition typically produces a clean background signal. These features have made glassy carbon a convenient choice for many studies of electrocatalyst materials.

The main electrochemical reactions expected on glassy carbon are the electrolyte decomposition reactions (i.e. the HER and OER), ion adsorption/desorption, and oxidation/reduction of the glassy carbon surface.178, 217, 219, 222 As shown in Figure A.8, the HER is the main electrochemical feature observed in the cathodic region and limits the inert potential range for each electrolyte. In NaOH and H2SO4, the HER activity of the glassy carbon increases slightly with potential cycling, possibly due to a reduction of any oxidized surface species or removal of surface impurities. In contrast, the HER activity decreases slightly with potential cycling in the NaAc.

In the sweeps in the anodic region, the main feature observed is the OER. This may be accompanied by some oxidation of the glassy carbon surface to produce carbon dioxide or oxidized surface species.178, 220, 223 This is especially true in the NaOH electrolyte, where the onset of oxidative current is initially observed at 1.66 V vs. RHE. With subsequent potential cycles, the onset of the oxidative reaction shifts to a potential of around 1.30 V vs. RHE where a small reductive peak appears as well. The combination of this reductive peak and the large hysteresis in the oxidative potential sweeps suggests that the oxidative features correspond to a combination of oxygen evolution and glassy carbon oxidation.

157

Figure A.8: Electrochemical activity and inert potential range for glassy carbon. The magnitude of the oxidative peak might be increasing due to an increase in the rates of these oxidative reactions as the surface is cleaned or roughened by the repeated formation and reduction of an oxide layer. It is also possible that as the glassy carbon surface is oxidized, the non-Faradaic capacitive current observed in this region also increases. In

NaAc, the only feature is the OER wave. In H2SO4, the OER onset limits the anodic stability window, and while the OER activity increases slightly with cycling, no reductive features are observed.

In summary, the inert potential region corresponds to potentials between -0.42 and 1.76 V vs. RHE, -0.76 and 1.96 V vs. RHE, and -0.64 and 1.66 V vs. RHE in H2SO4, NaAc, and NaOH, respectively. Glassy carbon’s large inert potential window makes it an ideal substrate for studying many electrocatalyst and photoelectrode materials. However, we

158 note that previous studies have shown that glassy carbon surfaces may possess a variety of functionalities, and that the nature of this surface can affect its electrochemical performance,178, 220, 224-237 thus researchers should pay careful attention to the GC surface preparation.

A.4.10 Highly Oriented Pyrolytic Graphite

Highly oriented pyrolytic graphite (also called highly ordered pyrolytic graphite) is another carbon allotrope that has proven useful for many studies in electrochemistry.164, 238, 239 It is a form of graphite made up of lamellar crystallites with a very high degree of crystallographic orientation (less than 1° angular spread in the c-axis directions).238 Thus, HOPG is an anisotropic material, and HOPG electrodes with either edge planes or basal planes exposed at the surface may be obtained.240 In this study, we used basal plane HOPG (sometimes called basal plane pyrolytic graphite).241 A key advantage of basal plane HOPG is its very smooth surface, which typically consists of atomically-flat terraces of several hundred nanometers between step edges.164, 242, 243 This makes HOPG a convenient support when scanning probe microscopy techniques are required.242, 244 Similar to glassy carbon, HOPG may also be advantageous when spectroscopic techniques are necessary because of its relatively clean background signal.239 HOPG electrodes can also be easily reused because the HOPG surface can be renewed by cleaving the electrode with a piece of tape to reveal a pristine top surface.164, 220, 239, 242

Prior to our electrochemical analysis, we performed a pre-anodization of the freshly cleaved HOPG surface. This anodization procedure introduces edge-site defects and surface oxygen into the HOPG basal planes.164, 245 These sites are more reactive than the basal plane sites and may improve adhesion of supported materials.179, 180 We performed a pre-anodization in this study because this type of pre-treatment is common in other studies where HOPG was used as a substrate for the study of electrocatalyst or photoelectrode materials.179, 180

The polarization curves collected on HOPG electrodes are displayed in Figure A.9.

159

Figure A.9: Electrochemical activity and inert potential range for highly oriented pyrolytic graphite.

The main reaction observed in the cathodic region is the HER. In H2SO4, the HER onset is initially observed at -0.55 V vs. RHE, but after repeated cycling, the HER activity increases and the onset shifts to a more positive potential of -0.45 V vs. RHE. A reductive shoulder (denoted a in Figure A.9) also appears at around -0.60 V vs. RHE. The increase in HER activity with repeated cycling may be attributed to surface roughening to expose more edge sites or the reduction of surface functionalities such as ethers and hydroxyl groups.246 The origin of the reductive shoulder is not clear, but it may also correspond to the reduction of oxidized surface groups or proton intercalation.

In the sweeps in the anodic region, the main electrochemical features correspond to anion intercalation, surface oxidation, and the OER. 247 In NaOH, an oxidation feature (denoted b in Figure A.9) with large hysteresis is followed by the onset of the OER. The oxidative

160 feature likely corresponds to surface oxidation. In NaAc, the OER wave is the only important feature. In H2SO4, significant oxidative current is first observed at around 2.06 V vs. RHE. Prior studies have shown that graphite may undergo oxidation and anion intercalation in sulfuric acid in this potential regime.220, 245 The oxidative current at the highly anodic potentials likely corresponds to a combination of these two processes along with oxygen evolution. The OER activity of the HOPG increases with potential cycling, probably due to surface cleaning or roughening. After the potential scan range is increased beyond the OER onset, a small reductive feature (shown in Figure A.9 inset) is observed on the sweeps in the negative direction. This feature likely corresponds to surface oxide reduction or de-intercalation, processes that can occur within the OER potential window.

In summary, the inert potential region corresponds to potentials between -0.55 and 2.06 V vs. RHE, -1.03 and 2.20 V vs. RHE, and -0.72 and 1.94 V vs. RHE in H2SO4, NaAc, and NaOH, respectively. HOPG’s excellent inert potential range in all three electrolytes makes it an ideal candidate substrate material for many studies. Like most of the other electrodes studied herein, the properties of the particular HOPG electrodes and the details of the experimental parameters used can influence the electrochemical behavior. Most notably, graphite step edge sites may have different reactivity than the basal plane sites, 220, 248 so special care should be taken to assess the step edge density for applications where this parameter may be significant.

A.4.11 Summary of Inert Potential Windows

We employed a threshold current density of 50 μA/cm2 to determine the potential boundaries at which each substrate could no longer be considered electrochemically inert. Each substrate has a different value of capacitance so this 50 μA/cm2 was measured above the baseline capacitive current. Any initial transients were ignored. The actual potential at which this threshold is first reached was taken and the window of inertness of each substrate is shown in Figure A.10.

161

Figure A.10: Potential range in which each substrate is inert for all electrolytes. Chemical stability is indicated by the color of the trace. The chemical stability of the substrate in each electrolyte, relevant for longer term testing (> 1 hr), is also indicated. In general, the TCOs have wide windows of inertness but are less stable than the opaque substrates. The GC and HOPG also draw very little current over a wide potential range and are very stable in all electrolytes. The results in Figure A.10 should provide an excellent starting point for researchers in the selection of substrate materials for electrochemical studies. For example, FTO and ITO are suitable substrates for the study of thick, semiconducting photoelectrocatalysts while GC and HOPG are more appropriate for evaluating the activity of nanoparticulate or other low coverage catalysts.

162

While these points were already known among researchers in the field, here we have quantified the useable potential windows for these important substrates to facilitate the substrate process for researchers in the future.

A.5 Conclusion The electrochemical activity and stability of several transparent conducting oxides (ITO, FTO, AZO) and opaque substrates (gold, SS, GC, HOPG) commonly used for evaluation of electrocatalysts and photoelectrodes have been evaluated. For each substrate, we identify the potential window in which the substrate is inert. While factors other than electrochemical inertness and stability, such as work function or surface termination, are also important to determine the appropriate substrate for a given application, the electrochemical properties of the substrate are almost always critical to consider for electrochemical applications. We therefore emphasize that each electrochemist should perform this type of baseline testing prior to electrocatalyst or photoelectrode evaluation. Due to the specific nature of the interactions between the substrate and electrolyte, some of the characteristic features may depend on the particular materials or experimental conditions employed. The results in this work provide a consistent basis for identifying viable substrates while the testing methodology reported herein provides a framework that can be used to make fair comparisons between potential substrates for their own studies.

A.6 Author Contributions Blaise A. Pinaud and Jesse D. Benck performed all measurements and data analysis and contributed equally to this work. Blaise A. Pinaud, Jesse D. Benck, Yelena Gorlin, and Thomas F. Jaramillo participated in writing the manuscript based on this work.

163

164

Appendix B Compression Cell and Integrating Sphere Design and Implementation

B.1 Introduction Thorough characterization of a material for photoelectrochemical (PEC) water splitting encompasses a wide variety of electrochemical methods (e.g. cyclic voltammetry, chronoamperometry, impedance spectroscopy) as well as solid-state techniques (e.g. UV- vis absorption spectroscopy, current-voltage curves, Hall effect measurement).27 There are several documents detailing appropriate experimental procedures to help guide researchers but there is still considerable variation in implementation of these methods in the literature.27, 164 While improper methodology is an unfortunately common problem, discrepancies also arise from inherent limitations such as differences in light sources, illumination configuration, geometric sample area, and product detection instruments. These factors make it difficult to establish a single standard protocol for materials testing in the field. Therefore, it is the responsibility of each researcher to design their experiments in the best way to report their data as fairly and accurately as possible. This section discusses the development of a compression cell adapted for PEC testing with both front and back illumination as well as the proper adjustment of the illumination intensity. A custom integrating sphere and sample holder were also designed to enable accurate measurement of the absorption properties of supported thin film semiconductors.

B.2 Compression Cell Design Vessels for electrochemical testing can range from a simple beaker to complex heart- shaped cells with multiple ports to more elaborate custom designs. Illumination requires either a transparent window or top-down illumination through the electrolyte. When samples are fully immersed in the electrolyte for testing, it is often necessary to mask the back contact and/or define an exact illumination area with an epoxy or resin. Such compounds can be difficult to remove, precluding further processing and

165 characterization, e.g. due to incompatibilities with some vacuum systems. A simple way of overcoming this challenge is the use of a compression cell. Electrical contact is made to the back or top portion of the sample. The sample is then sandwiched between two solid pieces with an o-ring which forms a seal. The size of the o-ring will dictate the area exposed to solution and gives a clear, consistent geometric area for converting current to current density which is a more useful metric for comparing materials. A compression cell was designed with a polytetrafluoroethylene (PTFE) base and borosilicate glass body. A photograph of the cell in use is shown in Figure B.1(a) while a schematic of the cell is illustrated in Figure B.1(b).

Figure B.1: (a) Photograph of the PEC compression cell under illumination with an active gas purge. (b) Schematic illustrating the key components of the cell (image credit: Dr. Jakob Kibsgaard). PTFE was selected for the base due to its excellent chemical resistance to both strong acid and base.249 It should be noted that since PTFE is somewhat porous and can retain small quantities of electrolyte even after rinsing, the cell is stored in Millipore water to leach out any acid or base and prevent cross-contamination. The body was not made of PTFE because it is highly reflective and scattering; light which reflects off the base could reflect or scatter off the side walls and reach the sample on a second pass which would overestimate the photocurrent. Either a dark material which absorbs the light or a transparent material which allows it to exit the cell is necessary. Borosilicate was selected for its high degree of transmission, low cost, and ease of glass blowing. Four equidistant screws spaced around the diameter allow the application of even pressure to make a seal on the sample with an o-ring. The o-ring is PTFE-coated silicone (McMaster Carr); pure PTFE o-rings are not very compressible and more difficult to seal while silicone is not sufficiently chemically resistant on its own. Typically, the bottom plate is a solid piece of

166

PTFE but holes were cut into several pieces to allow back-side illumination of photoelectrodes if needed.

A large silica bundle with a collimating lens is used to direct the light at the sample. This fiber is encased in a fused silica (i.e. quartz) tube with a flat window at the end to protect the lens from the electrolyte. It is important that fused silica be used rather than borosilicate or soda lime glass to guarantee high transmission including the deep ultraviolet region of the spectrum. Not shown in the schematic in Figure B.1(b) but seen in the photograph in Figure B.1(a) is the cap which conveniently holds the electrodes, glass frit for purging, and quartz tube in place. Fixing these components in place also ensures excellent reproducibility. For example, the measured solution resistance will vary with the distance between the working electrode and reference electrode but in this design, the distance is constant for all samples.

B.3 Illumination Calibration One of the most important parameters in photoelectrochemical measurements is the illumination intensity. While it is typically reported as a power per unit area (e.g. mW/cm2), this value is not sufficient for describing the light source. The spectral distribution and ultimately, the number of photons, is the key metric as each photon can lead to the generation of an electron-hole pair which might be harvested by the semiconductor to drive current. The spectral distributions of the most common broadband sources, xenon arc lamps and tungsten-halogen filament lamps, vary substantially.250 The former has a high output in the UV while the latter emits quite a bit of infrared light which must be filtered out. Experiments can be carried out using either but it is important to detail the type of source, the illumination intensity, and preferably the complete spectral distribution.

It is also imperative that the intensity and spectral distribution are quantified exactly where the sample is irradiated. With the compression cell described in the previous section, the entire cell can be assembled with only the bottom plate omitted. A cosine corrector (Ocean Optics, 3900 m diameter), connected to a spectroradiometer (Ocean Optics, Jaz), is used to capture the light. It is positioned nearly flush with the opening in

167 the cell base, exactly where the sample will be located as shown in Figure B.2(a) and Figure B.2(b).

Figure B.2: (a) The cosine corrector is positioned directly beneath the fully assembled compression cell. (b) The detector is placed at the same distance from the source at which the sample will be located. (c) The uniform beam of light is larger than the circular sample area. As shown in Figure B.2(c), it is desirable for the illuminated area to overfill rather than underfill the sample. Underfilling can set up potential gradients across the film and leads to different active areas for light and dark current. The one disadvantage to this calibration method is that the illumination intensity will be slightly overestimated because it does not account for absorption of the aqueous electrolyte.

The commonly accepted standard for solar illumination is the equivalent of 1 sun of air mass (AM) 1.5 G illumination.251 This spectrum is shown in Figure B.3 and has an overall intensity of 100 mW/cm2 when integrated across all wavelengths. A sample spectrum from a 1000 W Xenon lamp (Newport) is also shown in Figure B.3 to illustrate the importance of selecting appropriate integration bounds. Many spectroradiometers can only measure over a limited range of wavelengths, often from about 250 nm – 1000 nm. For example, in the region from 280 nm – 980 nm, the integration of the AM 1.5 G spectrum yields 72.5 mW/cm2, nearly 30% less than the commonly cited 100 mW/cm2. The Xe arc lamp output should then be scaled to this reduced value as shown in Figure B.3 where the intensity is 72.4 mW/cm2 from 280 – 980 nm.

However, there is yet another complication. The spectral matching is actually most important in the region where a semiconductor absorbs the incident photons.

168

Figure B.3: Spectral distribution of AM 1.5 G solar illumination and simulated solar light from a 1000 W Xe source. The spectrum can be integrated over the shaded region (λ = 280 – 980 nm) or the smaller hashed region (λ = 280 – 590 nm) for comparison. The latter is intended to capture only the photons which are absorbed by a Ta3N5 photoanode with a 2.1 eV band gap. It can therefore be even more accurate to measure the intensity for only the above-band gap energies. In the case of Ta3N5 with a band gap of 2.1 eV, the region of interest corresponds to 280 nm – 590 nm.26 The intensity is now found to be 34.2 mW/cm2 which is only slightly higher than that of AM 1.5 G in this region (32.3 mW/cm2). The small difference highlights the close spectral match of the Xe source with the solar spectrum but it is certainly not always the case with simulated light.

As mentioned earlier, the best comparison is actually one based on the number of photons. To illustrate this fact, the cumulative photon flux for three different sources tuned for 1 sun total illumination is shown in Figure B.4. The number of photons was calculated at 50 nm intervals by integrating all photons with higher energy than the set threshold energy. For example, the data point at 600 nm represents the total number of photons in the light spectrum from 280 nm – 600 nm. Figure B.4 also highlights that even for a given type of lamp, the output can vary substantially based on the type of filters and lenses used. The AAA solar simulator (Newport) and 1000 W Xe lamp both have Xe arc lamps but the spectral distributions of the photons are different. The AAA solar simulator in this case has been optimized for testing small band gap solar cells and the number of photons matches the AM 1.5 G spectrum very well at long wavelengths.

169

Figure B.4: Cumulative photon flux for energies higher than a given threshold wavelength in 50 nm increments. The expected photocurrent for a given flux is shown on the right axis. On the other hand, the 1000 W Xe lamp shown here is mostly used to study larger band gap oxide or nitride semiconductors and has been optimized so that the photon count matches AM 1.5 G illumination at shorter wavelengths. Inherent spectral differences between true solar light and simulated irradiation make it nearly impossible to match the photon count simultaneously at all wavelengths; each researcher needs to take care to ensure their source is optimized for the types of materials under study. This data shows that testing a material with a band gap of between 550 nm – 750 nm on the AAA solar simulator could result in a photocurrent of several mA/cm2 higher than would be expected under true solar light. Admittedly, calibrating light sources using photon counts can be cumbersome as some measurement tools such as silicon diodes only output total power, but a periodic check is useful to ensure photocurrents and reported efficiencies can be accurately compared and reflect their expected performance under actual solar light.

B.4 Integrating Sphere Design It is very important when measuring the absorption of thin semiconductor films that all light interacting with the sample be properly accounted for. While transmission measurements are fairly straightforward, they do not typically capture the reflected or scattered light resulting in overestimation of the absorption. Placing the sample directly inside an integrating sphere to capture each segment of light can mitigate this issue. Note that here, diffuse reflectance is defined as light scattered off the front surface while

170 scattering refers to light scattered off the back surface. Both experimental configurations are shown in Figure B.5 to contrast the differences.

Figure B.5: Schematic illustrating the measurement of absorption in an integrating sphere versus in a standard transmission experiment. As shown in (a), the specular and diffuse reflectance is not collected in a transmission experiment. The true absorption can be measured by placing the sample directly in an integrating sphere as shown in (b). The symbols represent the following fractions of light: Io is incident intensity, Rs is specular reflectance, Rd is diffuse reflectance, S is scattered light, T is transmitted light, and Ameas is measured absorption. The fractions of light collected by the detector in each method are boxed. An integrating sphere with 5 ports was purchased for these types of measurements and is shown in Figure B.6(a).

Figure B.6: (a) Image of the integrating sphere used for absorption measurements. (b) Photo of the interior of the integrating sphere showing the light input, baffles, and sample holder. One port is fitted with a collimating lens connected to an optical fiber (Ocean Optics, 600 m) for light input. A second port, not directly in line with first, has a larger diameter fiber (Ocean Optics, 1000 m) for collecting the light which is not absorbed by the sample. The sample is held in place with a custom-designed holder shown in Figure B.6(b). Note that the sample must be tilted slightly off axis in order to direct the specularly reflected light into the sphere. If this angle is more than a few degrees, then the

171 actual path length of light through the sample is much longer than the thickness and should be calculated by multiplying by a factor of cos where  is the angle off- normal. A major benefit of the integrating sphere is the ability to measure each component of light separately. The diffuse and specular reflectance can be isolated by placing a light sink directly behind the sample to block transmitted light. The transmission can in turn be directly measured by placing the detecting fiber directly behind the sample. Nearly all photoelectrode materials are supported on a substrate which is not entirely transparent. An important experimental detail in using such systems is thus to select an appropriate baselining method. While the temptation may be to simply use a bare substrate in place of the sample, this approach has a drawback. An integrating sphere is designed to evenly distribute light across the entire interior surface of the sphere which is a near-perfect Lambertian scatterer. With the light repeatedly bouncing off the surface, some photons will hit the sample again. The result is an overestimation of the absorption.

This additional absorption was quantified for several Ta3N5 samples supported on fused silica and is shown in Figure B.7.

Figure B.7: Absorption of a ~ 200 nm layer of Ta3N5 due to interiorly reflected light in the integrating sphere. This effect in part motivated the selection of a relatively large (8” diameter) sphere to mitigate second, third, etc. passes. It also means that for highly absorbing samples supported on substrates with negligible absorption, a better way of baselining involves doing so with the sample actually in the sphere but rotated out of the beam. If the sample

172 is very thin or a poor absorber and a significant amount of light will reach the substrate, it is still useful to quantify and subtract the absorption of the substrate material.

B.5 Conclusions The development of a PTFE compression cell with glass body has greatly facilitated photoelectrochemical measurements on a host of materials. The primary benefits over conventional glass cells are the well-defined sample area, closed environment for purging, and ability to measure the illumination intensity at the exact location of the sample. It was also shown that the varying spectral distribution of light sources and limited measurement range of some spectroradiometers necessitates very careful calibration of simulated light. One way of ensuring neither over or underestimation of the photocurrent is to match the power to the AM 1.5 G spectrum specifically in the range of wavelengths absorbed by the semiconductor. Since the number of photons rather than power is the important metric, it is good practice to calculate and compare the flux of the simulated light to true solar illumination. Lastly, an integrating sphere was designed to improve the accuracy of absorption measurements for thin film samples, detailed in Chapter 6, and an appropriate baselining method was established.

173

174

Appendix C

Thin Films of Na Birnessite-type MnO2: Optical Properties, Electronic Band Structure, and Solar Photoelectrochemistry

C.1 Abstract

Inexpensive and earth-abundant, manganese oxides (MnxOy) have attracted considerable attention for catalysis, but fewer efforts have focused on their semiconducting properties.

In the study detailed in this chapter, crystalline Na birnessite-type MnO2 thin films were investigated for their surface and bulk chemistry in the context of hydrogen production by photoelectrochemical water splitting. Thin films were synthesized by electrodeposition onto fluorine-doped tin oxide substrates, imaged by scanning electron microscopy, and characterized by x-ray diffraction as well as x-ray photoelectron spectroscopy and UV-Vis spectroscopy. Three different electrochemical methods (illuminated open circuit potential, potential of photocurrent onset and Mott-Schottky plots) were used to probe the flat band potential required to construct a band diagram for the material. Photostability, conductivity, and band structure are discussed as potential causes of the low external quantum efficiency (< 1%) for the birnessite-type MnO2 photoanode. The position of the conduction band well below the hydrogen evolution potential likely mitigates this material’s potential use in a single absorber configuration, but its chemical, optical, and electronic characteristics as shown in this work may be well-suited for a photoanode in a tandem device.

C.2 Introduction Semiconducting materials play an important technological role in a number of industries, from electronics to medicine to energy and the environment. Fundamental understanding of the surface and bulk chemistry of semiconductors is essential to the development of improved technological devices. One particular area of importance is that of solar

175 photoelectrochemistry, in which a semiconductor under irradiation drives redox chemistry at its surface to produce a fuel or chemical of interest in a potentially renewable and sustainable manner.14 We study this photon-driven chemistry in the work described herein, as we explore the bulk and surface chemistry of Na birnessite-type

MnO2 within an aqueous electrochemical environment. We synthesize crystalline thin films of this material, characterize its bulk and surface structure, and investigate its electronic band structure with in situ methods in order to construct a band diagram under operating conditions. The focus of this work is on connecting these physical and chemical insights to the material’s ability to split water into H2 and O2 under solar irradiation; a sustainable energy system that could be cost-effective on a global scale if appropriate materials can be developed.45 The work presented herein is also pertinent to the development of manganese oxide based materials for other applications in an aqueous, electrochemical environment, e.g. batteries, supercapacitors, and electrolyzers.

Hydrogen is a tremendously important feedstock in several industrial processes. In fact, 252 yearly production of H2 is roughly 45 billion kg. Its use is split nearly evenly between two primary applications: synthesis of ammonia via the Haber process for use in fertilizer and hydrocracking to transform heavy petroleum to lighter fractions for use as fuel.253 Virtually all of this hydrogen is produced at the point of use and is derived almost exclusively from the processing of fossil fuels (natural gas, oil, and coal). H2 production alone consumes 2% of the world’s energy demand.252 Hence, there are enormous benefits to developing a means to derive hydrogen sustainably by shifting production to an abundant, low cost feedstock such as water while using a renewable source of energy such as solar, wind, or hydroelectric.14 The clean production of hydrogen is also imperative if its envisioned expansion as a fuel for transportation is to become a reality.

Photoelectrochemical (PEC) water splitting encompasses both solar energy capture and conversion of water directly to pure hydrogen and oxygen in a carbon-free process. However, the development of a highly efficient PEC system is severely hindered by the large number of material requirements which must be simultaneously satisfied. Thermodynamics dictate a minimum energy requirement of 1.23 eV for each of the four photons in equation (1), but losses due to recombination, electrode and electrolyte

176 resistances, entropic losses of the photogenerated carriers, and catalytic overpotentials increase the minimum photon energy required to drive the reaction.14

4hν + 2H2O(liquid)  2H2(gas) + O2(gas) (1)

For a single absorber system, a band gap of 1.8 eV – 2.2 eV is considered optimal254, 255 for balancing the need to have a semiconductor photoelectrode which absorbs strongly in the visible portion of the solar spectrum while still absorbing sufficient energy to drive both the hydrogen evolution reaction (HER) and oxygen evolution reaction (OER). A good catalytic surface is needed to reduce the energy requirement by minimizing the overpotential of the reaction occurring at the photoelectrode. Another critical constraint on the material is that the valence and conduction bands must straddle the HER and OER + redox potentials, denoted as E(H /H2) and E(H2O/O2), respectively. This is a necessary but insufficient condition as the actual energy available for water splitting is less than the difference between the conduction and valence bands due to entropic losses associated with the photogenerated carriers.256 Lastly, the material should be inexpensive, abundant, and stable under the operating conditions over long periods to be industrially relevant.

Despite significant research efforts dating back to the first demonstration of water photolysis in 1972,7 even the most promising PEC materials to date are fraught with performance limitations. The band positions in TiO2 and its excellent aqueous stability are well suited for water splitting but a large band gap (Eg ~ 3.2 eV) limits adsorption to UV irradiation257 thereby restricting the solar-to-hydrogen (STH) efficiency to below 27 16, 257 1%. WO3 (Eg ~ 2.6 eV – 2.8 eV) also suffers from poor visible light absorption.

While α-Fe2O3 (Eg ~ 2.2 eV) has been studied extensively due to its stability and near optimal band gap, photocurrent densities are limited by poor conductivity requiring 258 sophisticated doping. Other attractive materials such as CdS (Eg ~ 2.4 eV) and Cu2O 16, 259 (Eg ~ 2.0 eV) are susceptible to photooxidation or photoreduction, respectively. Tandem cell configurations incorporating both a photocathode and photoanode are an interesting alternative to single absorber systems. Over a decade ago, Khasalev and 8 Turner reported an STH efficiency of 12.4% for a p/n-GaInP2(Pt)/GaAs device, but significant cost reductions in the materials and/or synthesis as well as improved stability are needed before this device can be considered viable. For all PEC cell configurations –

177 whether single or dual band gap – there is clearly a need to engineer high efficiency water splitting photocatalysts from low cost materials consisting of earth-abundant elements.

The study of new semiconductor materials is also hindered by the lack of knowledge of their electronic band structure which is crucial in evaluating their suitability for a PEC water splitting system. While vacuum-based spectroscopic techniques can shed light on the positions of the conduction and valence bands as well as the Fermi level, these do not necessarily reflect the state of the photoelectrode under operation in an electrolyte. Several electrochemical techniques do allow researchers to access this information in an aqueous media – illuminated open circuit potential, Mott-Schottky plots or the potential of photocurrent onset – though there are challenges with each method.27 A careful comparison between techniques with consideration of all contributing effects is required to determine an accurate electronic band diagram. In this paper we explore these methods as applied to an oxide photocatalyst consisting of Na birnessite-type MnO2, for which the material selection rationale is described below.

In recent years, manganese oxides (MnxOy) have drawn increasing attention for battery applications,260 supercapacitors,261 and even visible light driven catalysis.262, 263 Manganese is the 10th most abundant element in the Earth’s crust,264 and is several orders of magnitude cheaper than rare earth metal catalysts such as Pt, Ir, Pd, or Ru.265 While

IrO2 and RuO2 still exhibit the highest OER activity reported to date, MnO2 has been shown to be a good oxygen evolution catalyst.117, 266-268 Furthermore, evidence of the photoreductive dissolution of manganese minerals has led researchers to study the semiconducting nature of manganese oxides.269-271 The photocatalytic ability of both crystalline and amorphous manganese oxides was first demonstrated nearly two decades ago through oxidation of 2-propanol.272 More recent research into the structure of the oxygen evolving center of photosystem II has shed light on the role of manganese in nature’s own water oxidation photocatalyst273 and fueled the development of biomimetic OER catalysts.274, 275 The promising semiconducting and catalytic properties coupled with inexpensive, abundant and non-toxic elements make manganese oxides compelling photoanode candidates for a PEC cell. While there is a first report of photocurrent generated by MnO2 nanosheets under visible light, the incident photon-to-current

178 efficiency (IPCE) at all wavelengths is very low (< 0.2%).276 Otherwise, research in the field is limited and there is no shortage of phases to be explored; manganese exists in the (II), (III), or (IV) states forming over 30 oxide/hydroxide single or multivalent 264 minerals. Edge- or corner-sharing MnO6 octahedra form the basic unit for tunnel or layered structures and additional complexity is achieved through incorporation of cations.

In this chapter, we evaluate the photocatalytic activity of a specific phase of manganese oxide, Na birnessite-type MnO2. The direct water splitting reaction (without sacrificial reagents) is studied with an emphasis on determining the band structure of the material to evaluate its suitability for a PEC system either as a single absorber or in a tandem configuration. We combine knowledge gained from both spectroscopic methods and electrochemical methods in order to establish a fundamental understanding of this material’s surface and bulk chemistry.

C.3 Experimental Section C.3.1 Sample Preparation

Fluorine-doped tin oxide (FTO) substrates on glass supports (Hartford Glass Co., sheet resistance of 8 Ω/) were ultrasonically cleaned using the following sequence of solvents: soapy water, acetone, isopropanol, and Milli-Q water. Na birnessite-type MnO2 films were prepared by potentiostatic electrodeposition at 1.0 V vs. Ag/AgCl in a bath containing 2 mM MnSO4 (Sigma-Aldrich, ≥ 99%) and 50 mM NaClO4 (Aldrich, 99.99% trace metals basis) as described previously in the literature.277 The type of intercalated cation in the birnessite structure is dictated by the composition of the electrolyte. Film area was approximately 1 cm2 and film thickness was controlled by varying the charge deposited from 25 mC/cm2 to 100 mC/cm2.

C.3.2 Film Characterization

Film morphology was investigated by scanning electron microscopy (SEM) (FEI XL30 Sirion) with a beam voltage of 5 kV. Film thickness was evaluated by depositing a 1 μm Pt layer and milling away a 1 μm × 4 μm trench using a focused ion beam (FIB) (FEI Strata 235DB Dual-Beam FIB/SEM) to reveal a cross-section. Crystallinity was studied in ambient conditions by x-ray diffraction (XRD) (Phillips PANanalytical X’Pert Pro)

179 using Cu Kα radiation (λ = 1.54184 Å). The optical band gap of the semiconducting thin films was determined by UV-Vis absorption spectroscopy (Varian Cary 6000i) in transmission mode with baseline correction for the underlying FTO substrate. The oxidation state of the manganese was determined by x-ray photoelectron spectroscopy (XPS) (PHI VersaProbe Scanning XPS) using Al Kα radiation (hν = 1486.6 eV). The position of the Mn 2p peaks alone is typically insufficient for distinguishing between Mn oxidation states; the previously reported metrics of ΔE3s splitting and Mn 2p1/2 satellite 278-281 position were used to distinguish between MnO, Mn2O3, Mn3O4, and MnO2. Reference spectra were also collected using powders purchased from Sigma Aldrich and tested as received except in the case of MnO; the surface of the MnO sample was sputtered to remove any higher oxide on the surface. All spectra were calibrated to the C 1s line of adventitious carbon at 284.6 eV.282

C.3.3 Photoelectrochemical Testing

A glass test cell was set up in a conventional three-electrode configuration with 0.1 M

NaOH as the electrolyte, the supported MnO2 film as the working electrode, a Pt mesh counter electrode, and a Ag/AgCl (sat. KCl) reference electrode (+ 0.954 V vs. RHE). Photoelectrochemical measurements were collected using a potentiostat (Bio-Logic VMP3) and a 1000 W Xenon light source (Newport) fitted with a water filter to reduce intensity in the IR region and an AM1.5 filter (Newport) to more closely match the solar spectrum. An integrating sphere and spectroradiometer (International Light RPS900-W) were used to measure the intensity and spectral distribution of the lamp output. A broadband illumination of 634 mW/cm2 (integrated over λ = 250 nm – 950 nm) was measured, corresponding to an intensity of ~ 10 suns at λ = 500 nm. The optical reflection and scattering of the borosilicate glass window and electrolyte were not taken into account resulting in this value being an upper limit for the true incident intensity at the electrode, thus all reported efficiencies represent lower limits. For stability testing, a neutral density filter (optical density of 0.5 at 632.8 nm, Newport) was also used. To determine the incident photon-to-current efficiency, monochromatic light (fwhm = 10 nm) was obtained using a monochromator (74100 Oriel Cornerstone) and suitable cut-off filters were used to eliminate light from higher order diffractions.

180

C.4 Results and Discussion C.4.1 Film Morphology

A steady-state deposition current of 0.06 mA/cm2 was recorded at 1.0 V vs. Ag/AgCl. The amount of material deposited on each sample was varied by controlling the charge (in mC) deposited over an area of 1 cm2. Films were deposited at charge densities (Q) of 25 mC/cm2, 50 mC/cm2, and 100 mC/cm2 with a corresponding film thickness of approximately 28 nm per 25 mC/cm2 deposited. The current profiles for electrodeposition of these films are shown in Figure C.1 along with representative images of films of these thicknesses. The color of the films ranged from light brown/yellow to dark brown with increasing thickness.

Figure C.1: (a) Electrodeposition current profiles for films with deposited charge densities of Q = 25 mC/cm2, Q = 50 mC/cm2, and Q = 100 mC/cm2 at sequentially applied potentials of 0.8 V vs. Ag/AgCl and 1.0 V vs. Ag/AgCl.(b) Representative image of films of three different thicknesses showing their yellow/brown coloring. The correlation between charge density deposited and film thickness was evaluated by 2 imaging a cross-section of a MnO2 film with a charge deposited of 50 mC/cm . In the FIB instrument, a rectangular 1 μm thick Pt layer was first deposited to protect the film from local damage during the ion milling process. Next, a 1 μm × 4 μm trench was milled through the Pt, MnO2 and FTO right down to the glass support beneath as shown in Figure C.2(a).

181

2 Figure C.2: (a) SEM image of Pt layer deposited on a thin film of MnO2 (50 mC/cm ) with a trench milled away to reveal a cross-section, and (b) SEM image at 45° angle of Pt/MnO2/FTO interface at cross-section used to measure film thickness. The sample was transferred to the SEM to obtain high resolution images of the interface for determining the film thickness as shown in Figure C.2(b). As the cross-sectional image was taken at an angle of 45°, the measured thickness on the image of 40 nm was converted to an actual film thickness per charge density deposited via equation (2).

( ) ( )

Note that the relationship between charge deposited and film thickness is based on the analysis of a single sample; it is presented to provide an order of magnitude approximation of film thickness and is not intended as an exact measurement.

Discussion of the electrodeposition mechanism from Mn(II) to MnO2 can be found elsewhere in the literature.283-285 During preliminary electrochemical testing, some films peeled off the FTO substrate. To improve the oxide-FTO interface, a very thin layer (< 2

182 nm) was deposited at a much lower current density (~ 0.008 mA/cm2) by applying a lower potential of 0.8 V vs. Ag/AgCl. After 5 min, the potential was raised to 1.0 V vs. Ag/AgCl to obtain the desired birnessite structure. Films deposited using this method demonstrated significantly improved adhesion. It is believed that the nucleation rate for an electrodeposition process is typically slower at a lower overpotential leading to the formation of larger grains with fewer boundaries; defects of this type can weaken the overall cohesiveness of the film.286 Unless otherwise specified, all data shown is for films deposited in this manner. The SEM image in Figure C.3 is representative of the sheet-like structure consistent with previously reported films synthesized using this method.277 The FTO grains are distinguishable beneath the thin conformal film.

Figure C.3: SEM image of electrodeposited MnO2 on FTO substrate (bare surface shown in inset, same scale) before electrochemical testing.

C.4.2 MnO2 Crystal Structure by XRD

The general structure of birnessite-type MnO2 consists of layers of edge-sharing MnO6 octahedra intercalated with cations and water molecules. However, the imperfect crystallinity of the natural and synthetic variations of the birnessite phase and the different types of defects/vacancies lead to difficulty in determining the exact crystal and chemical structure. While natural birnessite minerals exhibit hexagonal symmetry in which a negative charge in the MnO2 layer results from Mn(IV) vacancies, synthetic birnessites tend to have a triclinic structure with defects arising from the substitution of Mn(IV) by Mn(III).270 There also exists the possibility of substitution of Mn(IV) by Mn(II) or displacement of O by OH groups.287 The reader is referred to many excellent

183 discussions in the literature regarding the crystal structure of this phase and the proportions of Mn(IV), Mn(III), and Mn(II).270, 287-289 The electrodeposited birnessite reported here is expected to have a crystal structure (shown in Figure C.4) similar to that described in the work of those who pioneered the electrodeposition synthetic route.277

Figure C.4: Schematic showing the structure of layered Na birnessite-type MnO2 consisting of edge-sharing + MnO6 octahedra intercalated with Na cations and H2O. A negative charge within the MnO2 layers is believed to arise from the substitution of Mn(IV) by Mn(III), and is counterbalanced by Na+ cations intercalated between the layers. Diffraction patterns for films dried in air of three different thicknesses as-deposited and after photoelectrochemical testing are shown in Figure C.5. XRD revealed distinct yet broad peaks at 2θ = 12.2° and 24.5° which can be assigned to the (0 0 1) and (0 0 2) planes of a crystalline birnessite-type layered structure with small grain size.277 Data for 2θ = 35° – 90° is not shown as any and all peaks in this range were attributable to the FTO substrate. No significant changes in the XRD patterns were observed after PEC testing, indicating that the structure was preserved. In an effort to increase long-range film crystallinity, three additional films (Q = 50 mC/cm2) were heated at 100°C, 200°C, and 300°C in ambient air for 1 hr. The XRD spectra collected around the major peak at 12° for these heated samples are shown in Figure C.6 and compared to an unannealed sample as a reference. Note that all of the films had the same deposited charge density of 50 mC/cm2.

184

2 Figure C.5: Diffraction patterns for electrodeposited MnO2 films with Q = 100 mC/cm (a) as-deposited and (b) after PEC testing, Q = 50 mC/cm2 (c) as-deposited and (d) after PEC testing, and Q = 25 mC/cm2 (e) as-deposited and (f) after PEC testing. A ● indicates peaks assigned to the birnessite phase of MnO2 and * indicates a peak arising from the FTO substrate.

Figure C.6: XRD spectra from 2θ = 10° to 14° showing the shift in peak position to higher diffraction angles after annealing to 100°C and 200°C for 1 hr each, and the complete disappearance of the major peak at 12° upon annealing to 300°C for 1 hr. All films had a charge deposited of 50 mC/cm2 and identical spectra prior to annealing. Heating to higher temperatures resulted in a progressive shift of the peak at 12.2° to higher diffraction angles indicating a decrease in the d-spacing. The intensity of the peak also steadily decreased. These observations are consistent with dehydration of the interlayer spacing resulting in the collapse of the ordered layers and a decrease in crystallinity. This phenomenon has been previously reported in the literature.288, 290, 291 Importantly, peaks corresponding to the FTO substrate did not shift or exhibit a change in intensity upon annealing and no new peaks were observed. A similar dehydration effect

185 was also observed for films exposed to ultra high vacuum (UHV) during characterization (data not shown). To preserve the original birnessite structure, films were tested without heat treatment and prior to any exposure to UHV.

The crystallite size can be computed by extracting the full-width at half maximum (FWHM) for the major birnessite peak at 12.2° and applying the Scherrer equation given in equation (3) where L is the crystallite size, λ is the probe light wavelength, and k is the shape factor.

( ) ( )

Assuming a shape factor (k) of 0.9 and using λ = 1.54184 Å, the corresponding crystallite sizes for all films are shown in Table C.1.

Label on Figure C.5 Film Description Crystallite Size / (nm) (a) Q = 100 mC/cm2 as-deposited 6.8 (b) Q = 100 mC/cm2 after PEC testing 7.1 (c) Q = 50 mC/cm2 as-deposited 7.4 (d) Q = 50 mC/cm2 after PEC testing 8.2 (e) Q = 25 mC/cm2 as-deposited 9.2 (f) Q = 25 mC/cm2 after PEC testing 8.4

Table C.1: Crystallite size for Na birnessite-type MnO2 thin films The crystallite size computed from the broadening in the diffractograms yielded a consistent grain size in the range 6.8 – 9.2 nm for films of all three thicknesses, both as- deposited and after PEC testing. Grain features observed in the SEM shown in Figure C.3 which appear larger are likely comprised of many smaller crystallites of less than 10 nm.

C.4.3 Mn Oxidation State by XPS

The correlation between the Mn nuclear charge and peak position of the Mn 2p3/2 line has previously been investigated, revealing a progressive shift to higher binding energy with increasing Mn oxidation state.278 However, the breadth of the Mn 2p doublet peaks makes it difficult to positively determine oxide constitution based solely on this metric, especially in the case of multivalent compounds. Consideration of the following two parameters is also warranted: the Mn 2p1/2 satellite structure as well as the Mn ΔE3s splitting. The former arises from shake-up processes while the latter results from

186 multiplet splitting with the Mn 3d electrons.281 A key benefit of considering peak separations rather than absolute binding energies is that these metrics eliminate error inherent to the instrument or calibration procedure and facilitate comparing data within the literature.

XPS spectra for the Mn 2p and Mn 3s regions of the as-deposited material are shown in Figure C.7.

Figure C.7: XPS spectra from a 100 mC/cm2 film for the (a) Mn 2p region revealing a Mn 2p satellite separation of 11.8 eV, and (b) Mn 3s region revealing a ΔE3s splitting of 4.5 eV. These measurements are consistent with MnO2.

The Mn ΔE3s splitting was found to be 4.5 eV while the splitting between the Mn 2p1/2 peak and its satellite was 11.8 eV. Both of these results are consistent with Mn(IV)

(MnO2 stoichiometry) for which the reported values are 4.5 eV and 11.9 eV, 279 respectively. Our internal MnO2 standard yielded values of 4.5 eV and 11.8 eV, 57 respectively. Small shoulders on the Mn 2p1/2 and Mn 2p3/2 peaks are due to Mn(II) at

187 the surface which is likely from residue from the electrodeposition bath; these features disappear after sputtering with Ar (5 kV, 1 μA) for 2 min (total sputter depth of approximately 15 – 20 nm). As described earlier, conductivity in the structure is the result of Mn(IV) vacancies or substitution of Mn(IV) by Mn(III) for hexagonal and triclinic birnessite respectively. In either case, these result in regions of negative charge which are balanced by the intercalated Na+ ions to maintain overall charge neutrality. Peaks which could be assigned to Na were not detected in either the survey scan or a high resolution scan around 1072 eV where the strong Na 1s line is expected. This result may indicate a very low concentration of intercalated Na+ and thus a low defect concentration. Though likely present in small quantities to preserve the structure, the Na+ content is below the detection limits of XPS.

C.4.4 Band Gap by UV-Vis

Sherman et al. reported a band gap of 1.8 eV for Na birnessite-type MnO2 based on oxygen K-edge absorption and emission spectroscopy.271 The electronic transition between the valence and conduction bands consists primarily of an excitation from an O 2p to Mn 3d state but with some Mn 3d to Mn 3d character.270, 271 Sakai et al. synthesized nanosheets of protonic (rather than sodium) MnO2 birnessite, and identified a band gap of 2.23 eV based on a plot of the square root of the IPCE times the photon energy versus photon energy.276 In this study, the band gap was probed optically via UV- Vis absorption spectroscopy. Note that in the case of polycrystalline materials, UV-Vis alone is not always sufficient to conclusively determine the band gap because mid-gap states can result in absorption at longer wavelengths; it is important to also consider the wavelength-dependent photoresponse. The absorption spectrum for the electrodeposited

MnO2 is shown in Figure C.8 with Tauc plots used to extrapolate an allowed indirect band gap of 2.1 eV and an allowed direct band gap of 2.7 eV.

It is worthy to note that the interpretation of such plots is not trivial.27 The analysis of the direct Tauc plot is more straightforward than with the indirect Tauc plot as the former has a clear baseline signal. The indirect Tauc plot presents considerable difficulties.

188

Figure C.8: (a) UV-visible absorption spectra for as-deposited Na birnessite-type MnO2 films with Q = 25 mC/cm2 (solid line), Q = 50 mC/cm2 (dashed line) and Q = 100 mC/cm2 (dotted line). (b) Tauc plots from thinnest film for determining the indirect (2.1 eV) and direct (2.7 eV) band gaps. For example, it is believed that the absorption tail at low photon energies arises from changes in the reflection and scattering off the FTO substrate when it is covered by a film, effects which are not properly taken into account by the initial baselining procedure using bare FTO. The tangent line drawn to the baseline in the indirect Tauc plot in Figure C.8 assumes this effect scales linearly with photon energy which is only an approximation. Despite such limitations, there exists evidence in the raw UV-Vis spectrum that supports the presence of a band gap that is smaller than the allowed direct band gap of 2.7 eV. The onset of absorption occurs just above 700 nm (1.8 eV) which points to a spectroscopic transition around this energy.

C.4.5 Photocurrent Generation with Applied Bias and Film Stability

The current-potential curve under chopped broadband illumination for the Q = 25 mC/cm2 film is shown in Figure C.9(a).

189

Figure C.9: (a) Current-potential curve (anodic sweep at 10 mV/s shown) for Na birnessite-type MnO2 (solid trace) with 1 Hz chopped illumination showing n-type photocurrent and reasonable catalytic activity for the oxygen evolution reaction. Data taken under the same conditions on a bare FTO substrate (dashed trace) is shown as a reference. (b) Photocurrent as a function of time for a film with Q = 25 mC/cm2 at 1.65 V vs. RHE showing only a slight decay after 75 min. Data points are taken every 5 min from a 0.1 Hz chopped chronoamperometry experiment and the dark current has been subtracted. The immersed film area was 1 cm2 but the illumination area was only 0.5 cm2. Current rather than current density is reported because of this difference in geometric area for dark versus light activity. The anodic photocurrent supports the n-type semiconducting nature of the films. At potentials greater than 1.6 V vs. RHE there is significant dark current arising from purely catalytic oxidation of water which confirms that the surface is a reasonable OER catalyst (onset at ~ 400 mV overpotential). Even without online detection of O2 or H2, there is sufficient evidence to suggest that photocurrent arises primarily from water oxidation on the MnO2 surface. No sacrificial reagents were added and the electrolyte was purged with N2 throughout. If photooxidative dissolution of the film were to occur, it would likely follow equation (4).276

190

+ 4+ MnO2 + 4h  Mn + O2↑ (4)

Photoreductive dissolution of MnO2 to Mn(II) occurs in nature and has also been demonstrated experimentally.292 While unlikely given the use of the film as a photoanode, it may be possible for photoexcited electrons to reduce MnO2 while holes oxidize water at the same surface, producing no net measurable current through the external circuit.

In order to assess stability, a long-term photocurrent test was conducted over 75 min (0.1 Hz chopped broadband illumination at a bias of 1.65 V vs. RHE for a film with Q = 25 mC/cm2). As shown in Figure C.9(b), a slight decrease in photocurrent was observed gradually over the course of the experiment, from an initial value of 28 μA/cm2 to a final value of 21 μA/cm2 at the end of the experiment. However, the total photocurrent over 75 min corresponded to a charge passed of 28 mC which is larger than the total charge deposited of 25 mC during film synthesis, making significant degradation unlikely. The total photo-induced charge passed is calculated by integrating the photocurrent during the illuminated periods (10 s/chop) and taking into account the illuminated area of 0.5 cm2. There was also no visible change in the film color or its structural integrity (no flaking or delamination). As described earlier, no change in the XRD spectra (Figure C.5) was observed for the as-deposited and PEC tested films. These quantitative and qualitative results support the hypothesis that water oxidation rather than photodegradation of MnO2 is the main reaction.

The photostability of MnO2 in a PEC cell is not without precedent as Sakai et al. also provided similar evidence to show that their MnO2 nanosheets did not photodecompose.276 It has previously been proposed that the concentration of Mn vacancies plays an important role in the photoreactivity and thus stability of the birnessite structure.270 The lack of Na signal in our XPS data could be indicative of fewer intercalated cations and thus fewer Mn vacancies which would in turn result in greater stability. Under-coordination of atoms at a defect can change the site reactivity293 and thus the susceptibility to corrosion, while Kwon et al. proposed effects of defects on the 270 band gap and recombination rates. A logical next step is to quantify H2 and O2

191 production and determine the Faradaic efficiency for water splitting to conclusively rule out photodecomposition of MnO2.

The current-potential curves for two samples of greater thickness (Q = 50 mC/cm2 and 100 mC/cm2) were measured under the same conditions as the thinnest film (Q = 25 mC/cm2). Despite increased absorption for thicker films, photoactivity decreases with increasing thickness as shown in Figure C.10.

Figure C.10: Photocurrent as a function of electrode potential for films of three different thicknesses corresponding to Q = 25 mC/cm2 (circles), Q = 50 mC/cm2 (squares), and Q = 100 mC/cm2 (triangles). Photocurrents were measured potentiostatically under 0.2 Hz chopped illumination in which the potential was stepped in 100 mV increments from 1.25 V to 1.85 V vs. RHE. Photons absorbed throughout the film generate electrons and holes which must travel to either the solid-liquid interface or the back contact to be collected. Shortening these distances by reducing the film thickness results in a lower rate of recombination and can improve carrier collection significantly for a weakly conductive material. Poor electron conductivity in the Na birnessite-type MnO2 could be the result of a low doping density which is consistent with the hypothesized small defect concentration discussed earlier. Confirming this potential performance limitation requires knowledge of the electrical properties of the films. Direct measurement of our MnO2 film conductivity is not trivial for the following two reasons: i) the highly conductive FTO support beneath provides an effective electron highway for shuttling charges between instrument probes, and ii) all films are very thin (< 100 nm) making it difficult to deposit a top contact that will not create a short to the underlying FTO. However, four-point probe measurements on pressed pellets of Na birnessite-type MnO2 carried out by other researchers indicate a

192 very low conductivity of 1.92 x 10-6 S/cm at 25°C.294 A significant improvement in carrier collection could potentially be achieved by depositing a very thin layer of the

MnO2 on a conductive nanostructure. Orthogonalizing the direction of charge carrier collection and light absorption reduces the distance photogenerated holes and electrons must travel while maintaining high absorption.295

The incident photon-to-current efficiency spectrum for the thinnest film of birnessite-type

MnO2 is shown in Figure C.11.

Figure C.11: Incident photon-to-current efficiency spectrum at 1.65 V vs. RHE for the film with highest photocurrent (Q = 25 mC/cm2). The onset of photocurrent at λ = 600 nm is consistent with an indirect band gap of around 2.1 eV, supporting the Tauc plot analysis described earlier. Note the kink in the spectra at a wavelength of 425 nm after which the IPCE increases significantly faster than at longer wavelengths. This energy falls very near to the direct band gap of 2.7 eV (459 nm) as measured by UV-Vis absorption. The direct band gap transition results in higher photon absorption which leads to a greater number of extracted electron-hole pairs for incident photons of higher energy. The IPCE curve shown was collected at a relatively high bias (1.65 V vs. RHE) and yet the overall external quantum efficiency for the photoanode remains well below 1%. The IPCE data was not collected at the potential of photocurrent saturation because by this point, significant dark current arises from purely electrocatalytic water oxidation which interferes with the measurement of the photocatalytic activity.

193

Ideally, an n-type photoanode should exhibit photocurrent at a potential negative of + E(H /H2) if the material is to function as a single absorber with high efficiency for water splitting. The lack of photocurrent from the Na birnessite-type MnO2 electrode at potentials negative of even E(H2O/O2) points to a fundamental material limitation, motivating our investigation of its electronic band structure.

C.4.6 Na Birnessite-type MnO2 Band Structure

All Na birnessite-type MnO2 films, regardless of thickness, required a significant bias to split water. Given our measurements of relatively wide band gap values of 2.1 eV (indirect) and 2.7 eV (direct), this suggests a potential misalignment of the conduction band or the valence band to the HER potential (0 V vs. RHE) or the OER potential (1.23 V vs. RHE), respectively. We thus focus our attention on determining the band structure for Na birnessite-type MnO2; we employ three different photoelectrochemical measurements for this purpose.

There is one previous report on the experimental determination of the band structure of a synthetic Na birnessite-type MnO2 mineral; this study was based on x-ray absorption spectroscopy (XAS) and x-ray emission spectroscopy (XES) of the oxygen K-edge, and the resulting band diagram placed the conduction and valence bands at – 0.16 V vs. SHE + 271 and 1.64 V vs. SHE, respectively, at pH 8.3 (E(H /H2) = – 0.490 V vs. SHE). This band structure, however, was determined based on work at low pH and thus is not necessarily relevant to the high pH conditions of photoelectrochemical water splitting; it has been shown that the structure of birnessite MnO2 can change significantly across the pH scale.296 Also, the position of the O 1s orbital by photoelectron spectroscopy was required to shift the data to an absolute binding energy scale. Due to a lack of data for the birnessite-type MnO2 structure, the authors assumed that it would be the same as pyrolusite MnO2 for which this data was available.

Hence, additional work examining the electronic band structure of Na birnessite-type

MnO2 is needed, particularly under the high pH operating conditions relevant to PEC and water electrolysis, and with in-situ methods. Sakai et al. reported a band diagram for their protonic birnessite-type MnO2 based on the results of their photoelectrochemical

194

+ + testing (E(H /H2) = – 0.49 V vs. Ag/Ag ) showing the conduction and valence bands at – 0.35 V vs. Ag/Ag+ and 1.85 V vs. Ag/Ag+, respectively.276 However, in that work details were not provided to describe the specific metrics used and what assumptions were made in order to place these energy levels.

Accurate determination of the band edges is difficult given inherent limitations for all techniques, including the various electrochemical techniques. Our approach to determine the positions of the band edges consists of first employing electrochemical methods to determine the flat band potential (Efb) of the semiconductor, which for n-type 297 semiconductors is located near the conduction band (Ecb). If one can measure the doping density of the semiconductor, the difference Ecb – Efb can be calculated accurately; the difference in energies is typically between 0.1 eV and 0.4 eV for oxides.298, 299 Without knowledge of the doping concentration in the electrodeposited Na birnessite-type MnO2, Ecb – Efb is approximated as 0.3 eV. The position of the valence band (Evb) relative to the conduction band could then be determined by the optical band gap value. An inherent limitation to this approach is that the bulk band gap may differ from the surface electronic band gap due to the presence of mid-gap surface states. Determining an accurate band structure can be difficult with very thin films if the space charge region extends deep into the film, which may be the case if weakly doped.

195

The three methods used to measure Efb are illuminated open circuit potential (OCP), potential of photocurrent onset (Figure C.12), and Mott-Schottky plots (Figure C.13).

Figure C.13: Mott-Schottky measurements in the dark at 85 Hz (circles), 56 Hz (squares), and 36 Hz (triangles) yielding Efb ~ 1.63 V vs. RHE for Na birnessite-type MnO2.

These three methods yielded Efb values of 0.98 V, 1.54 V, and 1.63 V vs. RHE, respectively. An average over samples of three different thicknesses was taken for the first two values while Mott-Schottky plots for only the film with the highest photocurrent (Q = 25 mC/cm2) are shown. A comparison of the band diagrams based on each different measurement is shown pictorially in Figure C.14.

196

While Efb is a material property which should be independent of the measurement technique, there is a 650 mV spread in the values measured using the three electrochemical methods and a discussion of the limitations of each follows.

The open circuit potential measurement was taken under fairly intense illumination (~ 10 suns). However, the photoresponse of the OCP (the change in potential in the dark to light) was less than 5 mV for all films and is characterized by a slowly drifting potential which may result from thermal heating effects caused by sustained illumination. This finding can be explained if the Fermi level (and thus Efb) of the Na birnessite-type MnO2 semiconductor lies very near the solution Fermi level. As a result, equilibration of the Fermi levels results in only slight band bending in the absence of an applied bias. This is an early indication that this material will not be suitable to split water in a 2-electrode zero-bias system since the relatively small band bending at the interface (i.e. a small electric field) will provide minimal driving force to separate charges at the interface. This would suggest band misalignment because the Fermi level of a single absorber, n- + type photoelectrode should ideally lie well above E(H /H2). The small photoresponse could also be explained if a large number of defect states are present in the MnO2; surface trap states act as recombination centers and prevent the accumulation of photogenerated electrons and holes. Without the build-up of charges on either side of the depletion region, no opposing electric field is created to effectively ‘unbend’ the bands. In this case, more intense illumination is required to elicit an accurate OCP measurement.

The onset of photocurrent, as extrapolated from the linear region in the plots of 2 photocurrent density squared (jphoto ) vs. electrode potential in Figure C.12, varied within a 30 mV range from 1.53 V to 1.56 V vs. RHE with an average value of 1.54 V vs. RHE.

This method typically yields a potential which is more positive of the true Efb for an n- type material. A key assumption is that photocurrent will flow after the transition from forward to reverse bias generates a small electric field to separate photogenerated electron-hole pairs. In reality, other interfacial effects come into play and can delay onset 27 significantly. The findings for the MnO2 reported here are consistent with this rationale as the Efb measured by photocurrent onset lies over 650 mV more positive than that of the illuminated OCP as shown in Figure C.14. While plots such as the one in Figure C.12 are

197 common for determining the potential of photocurrent onset27, this point could also be interpreted as the potential at which the first sign of photocurrent is recorded on a chopped cyclic voltammogram. This metric yields a lower value of Efb of 1.35 V vs. RHE for all films tested in this study.

Finally, electrochemical impedance spectroscopy was used to measure capacitance- voltage behavior and ultimately produce Mott-Schottky plots for these materials. Mott- Schottky analysis typically works best for highly crystalline samples, and the Na birnessite-type MnO2 samples exhibited a well-behaved linear region with positive slope further indicating an n-type material. To extract a capacitance at each frequency, the photoelectrode was modeled as a resistor in series with a capacitor, representing the solution/bulk semiconductor resistance and space charge capacitance, respectively. The space charge capacitance is assumed to be much smaller than that of the Helmholtz double layer and thus dominates the response. The plots exhibited a frequency dependence as shown in Figure C.13 which is attributed to surface state capacitance. Despite different slopes, extrapolation of the linear region of the Mott-Schottky plots to the potential along the x-axis yields values in the narrow potential region of 1.62 V to 1.63 V vs. RHE.

To probe the sensitivity of the result on the circuit model used, a second circuit was employed in a separate analysis. The circuit is similar to the previous one but with an additional resistor, representing the charge transfer resistance, placed in parallel with the previously described space charge capacitance. Modeling of the impedance data over the entire frequency range with this second equivalent circuit to extract the capacitance at each potential yielded a similar flat band potential as the first.

As viewed graphically in Figure C.14, the Mott-Schottky plots reveal values of Efb which are more positive than those found by the other two methods, though within only 0.1 eV of the photocurrent onset method. Exact reasons for discrepancies among the three methods are unclear, though discrepancies are generally more often the case than not – particularly with imperfect samples that are not single crystalline nor defect-free.27, 301 A comparison of the electrochemically determined band structure with one generated using complementary electronic spectroscopies such as ultraviolet photoelectron spectroscopy

198

(probes occupied states) and inverse photoelectron spectroscopy (probes unoccupied states) could shed light on the accuracy of each method; we emphasize though that one should expect differences for samples measured in vacuum versus in the operating environment of an aqueous electrochemical cell.

While a precise value of Efb remains elusive, the three measurements presented herein are sufficiently accurate to establish the following key finding: much like other more extensively studied transition metal oxides such as Fe2O3 and WO3, in a single absorber configuration the conduction band of Na birnessite-type MnO2 is far too low (more positive of the reversible potential for the HER) to effect water splitting without an externally applied bias. If a higher IPCE could be achieved, perhaps by improving film conductivity, then this material could potentially be interesting as a photoanode in a tandem photoelectrochemical cell where the photocathode provides the additional bias required to split water. Other phases of manganese oxide which have yet to be thoroughly explored may possess better semiconducting properties for use as a photoelectrode.

C.5 Conclusions

We successfully synthesized crystalline Na birnessite-type MnO2 thin films via electrodeposition, probed their physical, chemical, and electronic nature, and investigated their photoactivity for the water splitting reaction. The low external quantum efficiency + (< 1%) and onset of photocurrent at a potential much more positive than E(H /H2) led us to focus on the band structure of this material. While the photoactive electrode exhibited good stability, we found that the conduction band of Na birnessite-type MnO2 lies well below the hydrogen evolution potential. As a result, this material is not suitable for a single absorber water splitting photoelectrode. However, the deep valence band, moderate band gap (2.1 eV), and good catalytic activity for the oxygen evolution reaction make this material potentially interesting for a tandem cell configuration. The experimental study of manganese oxides for photoelectrochemical water splitting is still in its infancy and there is a large amount of materials phase space which remains to be explored.

199

C.6 Author Contributions Blaise A. Pinaud performed all measurements and data analysis. Blaise A. Pinaud, Zhebo Chen, David N. Abram, and Thomas F. Jaramillo participated in writing the manuscript based on this work.

C.7 Copyright Reprinted with permission from

B. A. Pinaud, Z. Chen, D. N. Abram, and T. F. Jaramillo, “Films of Sodium Birnessite-

Type MnO2: Optical Properties, Electronic Band Structure, and Solar Photoelectrochemistry”, Journal of Physical Chemistry C 115 (23), 11830-11838 (2011).

200

Appendix D Photoreduction of Methyl Viologen on Silicon Photocathodes for Bio-electrochemical Reactors

D.1 Introduction Water splitting to produce hydrogen fuel is only one of many alternative energy strategies being explored. Another promising route is CO2 fixation by microbes to produce organic molecules such as methane.302 Despite increased awareness of the environmental impact of greenhouse gas emissions, petroleum remains the cornerstone of the transportation energy market. The production of methane from CO2 could help close the carbon cycle; natural gas, of which methane is the primary component, provides nearly 30% of the energy consumed globally.1 There are still however significant challenges for effective, scalable carbon capture.303, 304 Perhaps more practically, microbial electrosynthesis could enable the storage of surplus electricity from intermittent renewables such as solar cells and wind turbines.305 Energy can then be produced as needed by consuming the organics in a microbial fuel cell.306 It may also be possible to directly couple electrosynthesis with solar energy. To this end, a collaboration was initiated between the Spormann and Jaramillo groups to combine their expertise in engineering metabolic pathways with our experience in preparing photoelectrodes.

302 It has been shown that methanobacterium palustre can convert CO2 to methane but another interesting reaction is the conversion of fumarate to succinate by shewanella oneidensis. The latter conversion can proceed in an anaerobic reactor supplied with CO2 and a reduced redox shuttle such as neutral red or methyl viologen.307 Methyl viologen in the fully oxidized state (MV++) can undergo two sequential reductions, first to blue MV+ and then to the colorless, insoluble compound MV0.308 The reversible potentials for the first and second reductions are -0.446 and -0.88 V vs. NHE, respectively.308 The goal of

201 this work was to develop a photoelectrode to drive the reduction of MV++ to MV+ with as much underpotential as possible. Ideally, the photoelectrode would provide sufficient voltage to also drive water oxidation, however, as a proof-of-principle we focused solely on the cathodic reaction. The design and electrochemical evaluation of a Si-based photoelectrode for this purpose is discussed in this appendix.

D.2 Semiconductor Selection The primary requirements for the p-type semiconductor needed for this application are a conduction band above the redox potential for the one electron reduction of MV++ and the ability to drive several mA/cm2 in order to facilitate eventual product detection in the bio- reactor. Since the microbial bio-reactors are run for hours or days, the photoelectrode must also be very stable in neutral electrolyte. All experiments were run in the same media used to grow and sustain the cells which was composed of 50 mM KH2PO4, 40 mM NH4Cl, 0.5 mM nitriloacetate, 0.2 mM MgCl2, 50 μM FeCl2, 1 μM CoCl2, 1 μM 309 Na2MoO4, and 5 μM NiCl2 (referred to henceforth as MTM media). TiO2 is one of the most studied photoelectrodes for water splitting but its band gap of 3.1 eV limits absorption to ultraviolet photons.24 Si has been employed successfully as a photocathode specifically for MV++ reduction and was selected for its ready availability and established processing procedures.310-312 With a band gap of 1.1 eV, it should be possible to draw over 30 mA/cm2 of photocurrent.312 All p-type Si (referred to as p-Si) used in this study was boron-doped <100> with a single side polished and a resistivity from 0.001 – 0.005 Ω∙cm. Contact to the electrodes was made by scratching the back surface to remove the native oxide and applying a gallium-indium eutectic mixture (Sigma Aldrich, ≥99.99% trace metals basis). Adhesive copper tape (3M) was then used to cover the liquid contact. As will be shown later, the best photoelectrodes actually had a buried p-n junction. The + top portion of the Si was phosphorus doped (referred to as n p-Si) for 10 min in POCl3 at 900°C. An 8 nm layer of Ti was then sputtered (Metalica, custom-built) to protect the Si from oxidizing and forming an insulating layer (referred to as Ti:n+p-Si). Electrodes varied in area but were approximately 6 cm2.

202

D.3 Photoelectrode and Test Apparatus Design Experiments were initially carried out using an electrolysis cell modified for back-side illumination through a polycarbonate window.313 The benefits of this cell were a large electrode area (6 cm2), small liquid volume, and the ability to separate the two compartments with a Nafion membrane. This last criteria is important for ensuring the microorganisms which will eventually grow in the cathode compartment are not exposed to oxygen produced in the anode compartment. However, this configuration had two major disadvantages. The first is that Si photoelectrodes typically perform better with front illumination. The second is that back-side illumination requires the contact to be limited to the periphery of the photoelectrode given that neither the gallium-indium nor copper tape is transparent. This geometry can result in a substantial voltage gradient across the sample and large spatial variation in the photocurrent. The electrolysis cell was consequently replaced with a standard H-cell bio-reactor with a Nafion membrane (Sigma Aldrich, Nafion 117, 0.007” thickness) separating the cathode and anode compartments. The back contact to the Si was masked with epoxy (Loctite Hysol 9462) so that the entire electrode could be submerged. The electrode was illuminated through the glass and while there are some optical losses due to the radial curvature of the cylindrical vessel, these were found to be minor. The solution was actively stirred with a magnetic stir bar throughout all experiments. An image of the assembled cell is shown in Figure D.1.

203

The counter electrode was a graphite rod and the reference electrode was Ag/AgCl. Since the microbes must be maintained in an oxygen-free environment, the cell was continuously purged with N2 unless otherwise specified. No transfer of the methyl viologen across the membrane was visually observed.

D.4 Photoelectrochemical Characterization The first photoelectrochemical measurements were carried out with a p-Si photoelectrode in the MTM media with and without 5 mM MV++ (Acros Organics, 98%). The results of testing in the dark and under constant illumination are shown in Figure D.2 and the reversible potential for MV++ reduction to MV+ is clearly denoted.

Figure D.2: No current is observed on a p-Si photoelectrode in the dark or in the absence of MV++. Under illumination, features associated with both the first (MV++ → MV+) and second reduction (MV+ → MV0) are observed. The chemical structure for methyl viologen is shown in the inset. Based on the onset of photocurrent, the p-Si is generating a photovoltage of approximately 400 mV which is reasonable, though not as high as the best reported photoelectrodes of this material.311 Reductive waves corresponding to both the first and second reductions of MV++ are clearly visible. The oxidative feature which peaks near - 0.15 V vs. Ag/AgCl is attributed to reoxidation of MV+. It is also evident from this cyclic voltammogram (CV) that the potential should not be swept more cathodic than -0.55 V vs. Ag/AgCl under illumination to avoid the irreversible reduction of MV+ to MV0. Another observation from this data is that the photocurrent was substantially lower than what had been reported previously.312 The initial hypothesis was that the current might be mass transport limited. To verify this proposition, reduction experiments in the dark on

204 highly conductive, degenerately-doped p-Si were carried out. The two variables explored were the effect of stirring as well as the concentration of MV++ and the results are shown in Figure D.3.

Figure D.3: Effect of stirring and methyl viologen concentration on the reduction of MV++ to MV+ on a degenerately-doped p-Si electrode in the dark. The dashed lines indicate a quiescent solution while the solid lines represent experiments in which the solution was stirred. There is only a slight reduction in the current when the solution is quiescent as opposed to stirred. The most pronounced effect of stirring is actually in the oxidation wave; when the solution is not stirred, MV+ remains near the electrode surface and is reoxidized. When the solution is stirred, the reduced product is rapidly swept away from the surface and there is no oxidation peak. These CVs show that it is the concentration of MV++ which limits the current. It also requires a substantial overpotential to drive large (> 10 mA/cm2) currents. For the experiments with the bio-reactor, the concentration must be limited to 5 mM as methyl viologen is toxic to shewanella oneidensis organisms at higher concentrations.

Another consideration was the type of illumination. A 1000 W Xe lamp (Newport) was used for all diagnostic studies but the actual bio-reactor will be run inside an incubator, requiring a portable source of illumination. Initial tests with a tungsten-halogen lamp revealed that the heat generated by the source raised the temperature inside the incubator significantly. A 20 W white LED (Bridgelux, Digi-Key) with a metal heat sink was then identified as a potential source. As shown in Figure D.4(a), the output differs significantly from the Xe source but there are still many above-band gap photons for the

205

Si to absorb. It was ultimately shown that the irradiation intensity did not have a significant impact on the photoactivity of the electrode for MV++ reduction due to the mass transport limitations described earlier. Broadband white illumination at two different intensities (achieved with no filter and a neutral density filter) and light above only 700 nm (achieved with a 700 nm cut-on filter) were tested as shown in Figure D.4(a).

Figure D.4: (a) Illumination intensity and spectral distribution for white LED, Xe lamp with 700 nm cut-on, Xe lamp with neutral density (ND) filter, and Xe lamp with no filter. (b) The type of illumination does not change the photoreduction behavior of the p-Si photoelectrode. The motivation for testing the light with a 700 nm cut-on filter was that the MV+ species absorbs strongly in the visible from 300 nm – 700 nm while the MV++ species absorbs at wavelengths shorter than 300 nm.314 As the dark blue MV+ accumulates in solution, it may block visible light from reaching the photoelectrode. The resulting photocurrent on a p-Si electrode did not vary substantially as shown in Figure D.4(b) indicating the spectral distribution and illumination intensity are not important factors.

While the 400 mV photovoltage from the p-Si photoelectrode was acceptable, it was desired to push the onset of reduction to even more positive potentials using a p-n junction. A slow deactivation of the p-Si, attributed to surface oxidation, was also observed which motivated the addition of a Ti stabilization layer. The stability of the Ti:n+p-Si photoelectrode was verified using a combination of cyclic voltammetry and chronoamperometry techniques as shown in Figure D.5(a) and Figure D.5(b).

206

Figure D.5: (a) Current as a function of time during five consecutive 1 hour holds at -0.3 V vs. Ag/AgCl. (b) CVs collected after each hold reveal a slow decay in the photoactivity. The photoelectrode was poised at a potential of -0.3 V vs. Ag/AgCl in 1 hour increments. Between each 1 hr potentiostatic measurement, a CV was measured between -0.6 and 0 V vs. Ag/AgCl. Note the earlier onset of MV++ photoreduction on the Ti:n+p-Si which is now supplying a photovoltage of 550 mV, equivalent to the highest reported values.311

The electrolyte was actively purged with N2 throughout the experiments with the following exception: after 30 min of each 1 hour hold, the experiment was paused for 5 ++ 315 min and O2 was introduced in order to regenerate the MV , as observed when the solution became clear. The O2 in solution was then purged out with N2 before restarting. This procedure was also carried out at the end of each hour prior to collecting the CV.

The regeneration of the MV++ is necessary as there is a slow decrease in current over time due to depletion of MV++; previous work ruled out absorption by MV+ as a probable cause. After 30 min, the O2 purge recovers the current almost entirely to its initial value for a given hold. Despite this step, there appears to be a slow decay in activity at longer time scales, especially during the last two hours of testing. Interestingly, the CVs reveal that the onset of MV++ reduction is shifting to more negative potentials but the current in the mass transfer limited regime remains constant. This behavior could indicate an issue near the photoelectrode surface and it is suspected that this might be caused by changes in the electrolyte. The same stability data is plotted in Figure D.6 but with images of the cell at various times.

207

As expected, the solution turns deep blue after ~ 10 min upon reducing MV++ to MV+.

After the first regeneration with O2, the solution is completely clear. However, a yellow color, visible after regeneration, develops in the solution as time elapses. By the end of the 5 hr experiment, the solution is deep yellow as shown in Figure D.6. While the applied potential of -0.3 V vs. Ag/AgCl is well before the onset of MV+ reduction, it is possible that some MV+ is fully reduced to MV0 when the potential is extended to -0.6 V vs. Ag/AgCl during the CV. However, the MV0 compound would precipitate out due to its low solubility in aqueous solution.

Figure D.6: Stability test over 5 hours with inset images of the H-cell. The solution turns deep blue as MV+ is generated but, initially, returns to clear when the solution is saturated with O2. There is however a gradual yellowing of the solution over time. While not confirmed, it is believed that the yellow color may arise from a contaminant in the methyl viologen powder or the reduction of a metal ion from one of the salts in the solution. These products may be adsorbing to the surface and blocking reaction sites. In either case, it was confirmed that the decrease in current was not due to a change in the photoelectrode. When the electrolyte was refreshed and the cell thoroughly cleaned, the CV on the same Ti:n+p-Si photoelectrode exactly matched the initial CV before the 5 hour stability test (data not shown).

208

D.5 Conclusion A stable Ti:n+p-Si electrode was developed for the photoreduction of MV++ to MV+. The photoelectrode generated a 550 mV photovoltage and yielded a current of ~ 4 mA/cm2 when held at a potential of -0.3 V vs. Ag/AgCl. The slight decrease in current over time is attributed to changes in the electrolyte. This photoelectrode is now suitable for coupling with shewanella oneidensis in a bio-reactor for electrosynthesis.

209

210

Appendix E Optimization of the Synthesis of Pure, Crystalline TaON

E.1 Introduction

While tantalum nitride (Ta3N5) has received significant attention over the past decade as a photoanode material, tantalum oxynitride (TaON) is another interesting candidate from the same class of materials. The goal of this section was to synthesize phase pure, crystalline TaON semiconductor photoanodes without any of the TaOx or TaNx impurity phases often observed. Similar to the Ta3N5 system, it was expected that the synthesis of dense TaON films on Ta foils could reduce the number of grain boundaries and mitigate charge transport issues though the photoelectrochemical activity was not explored here. The optimization of the synthetic route to TaON is discussed in detail in this appendix as well as the most appropriate tools for identifying the phase.

The band gap of TaON is slightly larger (2.4 eV) than that of Ta3N5 (2.1 eV) as the 26 presence of the O 2p orbitals lowers the valence band. Unlike Ta3N5, the electronic structure of TaON has been studied extensively by theoreticians.28, 316, 317 The typical color of TaON films is yellow, though green films have also been reported which is interesting as this color cannot be observed as the result of electronic interband transitions.318 It has been proposed that the green hue is caused by Nb contamination.318 The incorporation of oxygen in the lattice makes TaON less susceptible to oxidative photodegradation, though self-deactivation is still observed unless an oxygen evolving catalyst covers the surface.22, 93 Photocurrents as high as 4 mA/cm2 with an onset as early as -0.2 V vs. RHE have been reported on TaON powders electrophoretically deposited on fluorine-doped tin oxide supports.31, 319 There exist several methods for synthesizing TaON including reactive sputtering,102, 320-322 atomic layer deposition,40 humid nitridation 323, 324 22 325 of Ta2O5, nitridation of a thin Ta2O5 film, nitridation of Ta2O5 nanotubes, and 326 ammonolysis of macroporous Ta2O5. Many of these materials do not actually contain

211 an oxygen to nitrogen ratio of unity and are best described by the formula TaOxNy. The properties of the oxynitrides are heavily influenced by differences in stoichiometry. For example, the conductivity of TaON increases by 3½ orders of magnitude as the nitrogen content is increased.30

E.2 Synthetic Parameters The synthesis of crystalline TaON is not trivial. While some researchers report making 22, 327 TaON by flowing dry NH3 over Ta2O5, the majority agree that the introduction of an 101, 318, 323, 324 oxygen source (e.g. H2O, O2) is necessary to achieve the TaON phase. The use of wet ammonia maintains an equilibrium between the competing nitridation and hydrolysis reactions. One possible cause for the conflicting reports is small air leaks in gas lines or fittings which introduce O2. Our experiments confirmed that heating a Ta2O5 film in dry NH3, regardless of the flow rate, leads to Ta3N5 rather than TaON.

TaON thin films were prepared from mechanically polished Ta foils (10 mm x 10 mm x 0.127 mm, Alfa Aesar, 99.95% metals basis excluding Nb) cleaned by sequential sonication for 30 min in acetone, isopropanol, and Millipore. A tube furnace (Mellen Company) was used for heating. The oxynitridation was carried out with a combination of dry Ar, wet Ar, and dry NH3 (Praxair, 99.95%, anhydrous). The wet Ar is obtained by flowing dry Ar through a bubbler containing H2O. This approach is preferred to directly humidifying the NH3 in order to be able to control the NH3 flow and water vapor content separately. Also, flowing NH3 through water requires a pre-saturation step given the high solubility of this gas in aqueous solution.

Initially, the gas lines for this process consisted of polyethylene tubing. The gas delivery manifold was eventually switched entirely to ¼” stainless steel piping to minimize the possibility of leaks. A schematic of the gas flows required for oxynitridation is shown in Figure E.1(a) along with photos of the updated furnace and gas delivery system in Figure E.1(b). The key to obtaining pure TaON is to optimize and tightly control the flow rates of the gases. The optimal combination was found to be 20 sccm dry NH3, 10 sccm wet Ar, and 50 sccm dry Ar. The temperature profile for the humid nitridation process is

212 shown in Figure E.1(c). The samples were first ramped to a temperature of 800°C in dry Ar then oxynitrided for 45 min at this temperature.

Figure E.1: (a) Diagram illustrating the gas flows to the furnace. (b) Photos of the gas delivery manifold and the tube furnace (image credit: Jared O’Leary). (c) Temperature profile for ramping and holding at 800°C during humid nitridation. Cooling to room temperature was carried out under an inert flow of dry Ar. Importantly, a deviation by as little as 10 sccm in a flow rate or 25°C in temperature leads to an entirely different result (i.e. phase other than TaON). It is common to first oxidize the Ta foil prior to the humid nitridation step but it was found that this step is not necessary. In fact, carrying out the exact same process on a film pre-oxidized at 550°C in air for 15 min resulted in a TaON film with a slight Ta2O5 impurity. While Ta metal is not attacked by pure NH3, it is clear that even the very low O2 content in the gas stream is sufficient to initiate the conversion of Ta to TaOx and subsequently TaON. Unfortunately, this phenomenon makes it difficult to control the TaON film thickness. An example of a thick TaON film (> 20 μm) is shown in the scanning electron microscope (SEM, FEI XL30 Sirion, 5 kV) image in Figure E.2 with an inset optical image showing the yellow color.

As the O2 will continue to react with the Ta as it flows through the furnace, the only method to tune the thickness is by controlling the humid nitridation time. To produce thin films on the order of hundreds of nm would require very short times (< 1 min); it was

213 found that it takes at least several minutes to establish the proper equilibrium for forming TaON. Therefore, it was not possible to synthesize thin films of TaON using this method.

Figure E.2: SEM image showing a thick TaON film grown on a Ta metal foil. The inset optical image highlights the bright yellow color of TaON. The underlying red is likely due to a Ta3N5 impurity. E.3 Physical Characterization There are three reported polymorphs of TaON, a well-characterized monoclinic β-TaON, a controversial hexagonal α-TaON, and a recently discovered γ-TaON.317, 328 X-ray diffraction (XRD, Phillips PANanalytical X’Pert Pro) using Cu Kα radiation (λ = 1.54184 Å) revealed that the films synthesized by the method described above were β- TaON. An SEM image of a TaON/Ta foil sample surface is shown in Figure E.3(a) while the diffractogram is shown in Figure E.3(b) with reference data (PDF cards 00-004-0788 for Ta, 01-071-0639 for Ta2O5, 01-071-0178 for TaON, and 01-079-1533 for Ta3N5).

Diffractograms for TaON/Ta foil films with Ta3N5 and Ta2O5 impurity phases are also shown in Figure E.3(b). The former results from insufficient humid Ar flow or too high an NH3 flow while the latter results from too high a humid Ar flow.

A common method for confirming the formation of TaON and determining the thickness through depth profiling is by x-ray photoelectron spectroscopy (XPS). The Ta 4f7/2 binding energy shifts to higher binding energy as the oxidation state increases. This peak 22 is located at 25.0 eV for Ta3N5, 26.0 eV for TaON, and 26.6 eV for Ta2O5. However, the presence of an oxygen-rich layer at the surface requires sputtering to evaluate the bulk composition. While surface properties are important for the oxygen evolution catalysis, it is the bulk properties which are important for the absorption of light.

214

Figure E.3: (a) SEM image of the textured morphology of the TaON film grown on a Ta foil. (b) XRD of crystalline TaON/Ta foil samples with and without detectable Ta2O5 or Ta3N5 impurity phases. Unfortunately, sputtering results in a reduction of the Ta, precluding the use of binding energy to determine the phase. This effect can be demonstrated by depth profiling a pure

Ta2O5 in which all Ta should be in the 5+ oxidation state. A commercial powder was pressed into an In foil to prevent charging. This sample was analyzed by XPS (PHI VersaProbe Scanning XPS) with Al Kα radiation (hν = 1486.6 eV) and progressively sputtered with Ar+ ions (5 kV, 1 μA, 2 mm x 2 mm). All binding energies were corrected based on the C 1s peak from adventitious carbon at 285.0 eV. The Ta 4f and O 1s regions are shown below in Figure E.4.

Figure E.4: XPS spectra of the O 1s and Ta 4f regions as a function of sputtering time for a Ta2O5 commercial powder.

215

While the oxygen signal remains mostly unchanged, there is a gradual shift of the Ta 4f7/2 and Ta5/2 peaks to lower binding energies. There is clearly a superposition of several doublets corresponding to multiple oxidation states. The reduction of transition metal centers by Ar+ sputtering has been reported previously.134 This result highlights the need to supplement or replace XPS with other techniques such as XRD for identifying the bulk composition of tantalum nitrides and oxynitrides.

E.4 Thermal Stability As discussed earlier, some authors have proposed that TaON can be synthesized directly by nitridation of Ta2O5 without any oxygen source in the gas stream. The question then follows, can TaON be obtained by the reverse reaction of heating Ta3N5 in air? A mixed

TaON + Ta3N5/Ta foil sample was synthesized by pre-oxidizing a Ta foil for 25 min at

550°C in air and then heating at 850°C for 45 min in a flow of 20 sccm NH3. The sample was annealed in air in situ during XRD measurements on a heated stage. The temperature was ramped from 400 – 700°C in 10°C increments. Diffractograms between 2θ = 20° – 40° were collected at each temperature; there was no hold time other than the 3.5 minutes required for each scan. The results are shown in Figure E.5 with reference spectra.

Figure E.5: In situ annealing in air of a TaON + Ta3N5/Ta foil sample. While the TaON features remain unchanged up to a temperature of 700°C, the nitride is converted to Ta2O5.

216

Initially, there are strong features corresponding to Ta3N5 but the intensity decreases progressively with heating and the crystal structure is destroyed by a temperature of

540°C. It appears that this Ta3N5 is converted to crystalline Ta2O5 as peaks for this phase begin to appear at 580°C and grow in intensity as the sample is heated further. There is one anomalous peak in the data at a position of 38°. This feature is not present in the initial scan but grows in starting at a temperature of 430°C. It then disappears once the sample is heated to 690°C. While it could not be assigned to any of the commonly expected phases, it may arise from a partial oxidation of Ta which is ultimately converted to stoichiometric Ta2O5 at higher temperature. There are small peaks in the room temperature spectrum which correspond to TaON. Throughout the entire experiment, these features neither increase nor decrease in intensity. The implication of this finding is that TaON is stable in air up to a temperature of at least 700°C. In fact, it has been reported that TaON is stable up to a temperature of 1000°C.30 However, TaON does not appear to be an intermediary phase in the conversion of Ta3N5 to Ta2O5.

E.5 Conclusions The direct growth of phase pure TaON on Ta foils has been demonstrated through the use of a humid nitridation procedure. The flow rates of dry Ar, wet Ar, and dry NH3 as well as the temperature are all critical experimental parameters for ensuring that no Ta2O5 or

Ta3N5 impurities are formed. XRD is the preferred method for confirming the presence of

TaON as the reduction of Ta during sputtering in XPS shifts the Ta 4f7/2 peak away from the expected value of 26.0 eV. Lastly, XRD data collected during in situ annealing of a

TaON + Ta3N5/Ta foil sample showed that TaON is stable at high temperature but is not an intermediate crystalline phase in the oxidation of Ta3N5.

217

218

References

1. J. J. Conti, P. D. Holtberg, J. A. Beamon, S. A. Napolitano, A. M. Schaal and J. T. Turnure, Annual Energy Outlook 2012: With Projections to 2035, Energy Information Administration, Washington, 2012. 2. J. Stringer and L. Horton, Basic Research Needs to Assure a Secure Energy Future: A Report, Oak Ridge National Laboratory, Oak Ridge, 2003. 3. N. Armaroli and V. Balzani, ChemSusChem, 2011, 4, 21-36. 4. Z. Wang, R. R. Roberts, G. F. Naterer and K. S. Gabriel, Int J Hydrogen Energ, 2012, 37, 16287-16301. 5. A. Ursua, L. M. Gandia and P. Sanchis, Proc. IEEE, 2012, 100, 410-426. 6. S. K. Ngoh and D. Njomo, Renew. Sust. Energ. Rev., 2012, 16, 6782-6792. 7. A. Fujishima and K. Honda, Nature, 1972, 238, 37. 8. O. Khaselev and J. A. Turner, Science, 1998, 280, 425-427. 9. S. Licht, B. Wang, S. Mukerji, T. Soga, M. Umeno and H. Tributsch, J. Phys. Chem. B, 2000, 104, 8920-8924. 10. S. Y. Reece, J. A. Hamel, K. Sung, T. D. Jarvi, A. J. Esswein, J. J. H. Pijpers and D. G. Nocera, Science, 2011, 334, 645-648. 11. R. E. Rocheleau, E. L. Miller and A. Misra, Energy & Fuels, 1998, 12, 3-10. 12. M. G. Walter, E. L. Warren, J. R. McKone, S. W. Boettcher, Q. X. Mi, E. A. Santori and N. S. Lewis, Chemical Reviews, 2010, 110, 6446-6473. 13. A. J. Nozik and R. Memming, J. Phys. Chem., 1996, 100, 13061-13078. 14. T. Bak, J. Nowotny, M. Rekas and C. C. Sorrell, Int J Hydrogen Energ, 2002, 27, 991-1022. 15. M. Kitano and M. Hara, Journal of Materials Chemistry, 2010, 20, 627-641. 16. A. Kudo and Y. Miseki, Chemical Society Reviews, 2009, 38, 253-278. 17. S. van Riesen, A. Gombert, E. Gerster, T. Gerstmaier, J. Jaus, F. Eltermann and A. W. Bett, AIP Conference Proceedings, 2011, 1407, 235-238. 18. G. S. Kinsey, K. Stone, J. Brown and V. Garboushian, Progress in Photovoltaics, 2011, 19, 794-796. 19. R. K. Jones, P. Hebert, P. Pien, R. R. King, D. Bhusari, R. Brandt, O. Al Taher, C. Fetzer, J. Ermer, A. Boca, D. Larrabee, X. Q. Liu and N. Karam, 2010 35th IEEE Photovoltaic Specialists Conference (PVSC), 2010, 000189-000195. 20. P. C. K. Vesborg and T. F. Jaramillo, RSC Adv., 2012, 2, 7933-7947. 21. G. Hitoki, A. Ishikawa, T. Takata, J. N. Kondo, M. Hara and K. Domen, Chemistry Letters, 2002, 736-737. 22. M. Hara, E. Chiba, A. Ishikawa, T. Takata, J. N. Kondo and K. Domen, J. Phys. Chem. B, 2003, 107, 13441-13445. 23. G. Hitoki, T. Takata, J. N. Kondo, M. Hara, H. Kobayashi and K. Domen, Electrochemistry, 2002, 70, 463-465. 24. A. Fujishima, X. Zhang and D. A. Tryk, Surface Science Reports, 2008, 63, 515- 582. 25. K. Sivula, F. Le Formal and M. Gratzel, ChemSusChem, 2011, 4, 432-449. 26. W. J. Chun, A. Ishikawa, H. Fujisawa, T. Takata, J. N. Kondo, M. Hara, M. Kawai, Y. Matsumoto and K. Domen, J. Phys. Chem. B, 2003, 107, 1798-1803.

219

27. Z. Chen, T. F. Jaramillo, T. G. Deutsch, A. Kleiman-Shwarsctein, A. J. Forman, N. Gaillard, R. Garland, K. Takanabe, C. Heske, M. Sunkara, E. W. McFarland, K. Domen, E. L. Miller, J. A. Turner and H. N. Dinh, Journal of Materials Research, 2010, 25, 3-16. 28. C. M. Fang, E. Orhan, G. A. de Wijs, H. T. Hintzen, R. A. de Groot, R. Marchand, J. Y. Saillard and G. de With, Journal of Materials Chemistry, 2001, 11, 1248-1252. 29. L. C. Seitz, Z. Chen, A. J. Forman, B. A. Pinaud, J. D. Benck and T. F. Jaramillo, unpublished work. 30. J. H. Swisher and M. H. Read, Metallurgical Transactions, 1972, 3, 489-494. 31. M. Higashi, K. Domen and R. Abe, Energy Environ. Sci., 2011, 4, 4138-4147. 32. T. Takata, G. Hitoki, J. N. Kondo, M. Hara, H. Kobayashi and K. Domen, Research on Chemical Intermediates, 2007, 33, 13-25. 33. A. Ishikawa, T. Takata, J. N. Kondo, M. Hara and K. Domen, J. Phys. Chem. B, 2004, 108, 11049-11053. 34. D. Yokoyama, H. Hashiguchi, K. Maeda, T. Minegishi, T. Takata, R. Abe, J. Kubota and K. Domen, Thin Solid Films, 2011, 519, 2087-2092. 35. X. J. Feng, T. J. LaTempa, J. I. Basham, G. K. Mor, O. K. Varghese and C. A. Grimes, Nano Letters, 2010, 10, 948-952. 36. Y. Cong, H. S. Park, S. Wang, H. X. Dang, F.-R. F. Fan, C. B. Mullins and A. J. Bard, The Journal of Physical Chemistry C, 2012, 116, 14541-14550. 37. Y. Li, T. Takata, D. Cha, K. Takanabe, T. Minegishi, J. Kubota and K. Domen, Advanced Materials, 2013, 25, 125-131. 38. C. Zhen, L. Wang, G. Liu, G. Q. Lu and H.-M. Cheng, Chemical Communications, 2013, 49, 3019-3021. 39. Y. Q. Cong, H. S. Park, H. X. Dang, F. R. F. Fan, A. J. Bard and C. B. Mullins, Chemistry of Materials, 2012, 24, 579-586. 40. M. Ritala, P. Kalsi, D. Riihela, K. Kukli, M. Leskela and J. Jokinen, Chemistry of Materials, 1999, 11, 1712-1718. 41. Z. W. Fang, H. C. Aspinall, R. Odedra and R. J. Potter, Journal of Crystal Growth, 2011, 331, 33-39. 42. C. H. Wu, C. Hahn, S. B. Khan, A. M. Asiri, S. M. Bawaked and P. Yang, Chemistry – An Asian Journal, 2013, n/a-n/a. 43. H. X. Dang, N. T. Hahn, H. S. Park, A. J. Bard and C. B. Mullins, The Journal of Physical Chemistry C, 2012, 116, 19225-19232. 44. Y. Kado, C.-Y. Lee, K. Lee, J. Muller, M. Moll, E. Spiecker and P. Schmuki, Chemical Communications, 2012, 48, 8685-8687. 45. B. D. James, G. N. Baum, J. Perez and K. N. Baum, Technoeconomic Analysis of Photoelectrochemical (PEC) Hydrogen Production, Directed Technologies Inc. (US DOE Contract No. GS-10F-009J), Arlington, VA, 2009. 46. E. Miller, Advanced Materials for Water Photolysis, US DOE Task 26 Annual Report, Washington, 2011. 47. M. F. Weber and M. J. Dignam, J Electrochem Soc, 1984, 131, 1258-1265. 48. M. F. Weber and M. J. Dignam, Int J Hydrogen Energ, 1986, 11, 225-232. 49. M. C. Hanna and A. J. Nozik, J Appl Phys, 2006, 100, 074510-074511-074510- 074518.

220

50. W. Shockley and H. J. Queisser, J Appl Phys, 1961, 32, 510-519. 51. R. T. Ross, J Chem Phys, 1967, 46, 4590-4593. 52. R. E. Rocheleau and E. L. Miller, Int J Hydrogen Energ, 1997, 22, 771-782. 53. J. R. Bolton, S. J. Strickler and J. S. Connolly, Nature, 1985, 316, 495-500. 54. S. X. Hu, C.; Haussener, S.; Berger, A.; Lewis, N. S., Energy Environ. Sci., 2013, DOI: 10.1039/c3ee40453f. 55. R. T. Ross and T. L. Hsiao, J Appl Phys, 1977, 48, 4783-4785. 56. S. Schuldiner, J Electrochem Soc, 1959, 106, 891-895. 57. Y. Gorlin and T. F. Jaramillo, Journal of the American Chemical Society, 2010, 132, 13612-13614. 58. X. B. Chen, S. H. Shen, L. J. Guo and S. S. Mao, Chemical Reviews, 2010, 110, 6503-6570. 59. J. G. Mavroides, J. A. Kafalas and D. F. Kolesar, Applied Physics Letters, 1976, 28, 241-243. 60. A. B. Ellis, S. W. Kaiser and M. S. Wrighton, J. Phys. Chem., 1976, 80, 1325- 1328. 61. N. Kobayashi, T. Narumi and R. Morita, Japanese Journal of Applied Physics Part 2-Letters & Express Letters, 2005, 44, L784-L786. 62. I. Waki, D. Cohen, R. Lal, U. Mishra, S. P. DenBaars and S. Nakamura, Applied Physics Letters, 2007, 91, 093519-093511-093519-093513. 63. A. J. Nozik, Applied Physics Letters, 1976, 29, 150-153. 64. R. C. Kainthla, B. Zelenay and J. O. Bockris, J Electrochem Soc, 1987, 134, 841- 845. 65. J. H. Park and A. J. Bard, Electrochemical and Solid State Letters, 2006, 9, E5- E8. 66. H. Arakawa, C. Shiraishi, M. Tatemoto, H. Kishida, D. Usui, A. Suma, A. Takamisawa and T. Yamaguchi, Proceedings of the SPIE - The International Society for Optical Engineering, 2007, 6650, 665003. 67. J. Brillet, M. Cornuz, F. Le Formal, J. H. Yum, M. Gratzel and K. Sivula, Journal of Materials Research, 2010, 25, 17-24. 68. J. Brillet, J. H. Yum, M. Cornuz, T. Hisatomi, R. Solarska, J. Augustynski, M. Graetzel and K. Sivula, Nat. Photonics, 2012, 6, 823-827. 69. E. L. Miller, B. Marsen, D. Paluselli and R. Rocheleau, Electrochemical and Solid State Letters, 2005, 8, A247-A249. 70. A. Madan, Photoelectrochemical hydrogen production, http://www.hydrogen.energy.gov/pdfs/review11/pd053_madan_2011_o.pdf, Accessed November 29, 2012. 71. S. Licht, B. Wang, S. Mukerji, T. Soga, M. Umeno and H. Tributsch, Int J Hydrogen Energ, 2001, 26, 653-659. 72. O. Khaselev, A. Bansal and J. A. Turner, Int J Hydrogen Energ, 2001, 26, 127- 132. 73. G. Peharz, F. Dimroth and U. Wittstadt, Int J Hydrogen Energ, 2007, 32, 3248- 3252. 74. A. Kudo, H. Kato and I. Tsuji, Chemistry Letters, 2004, 33, 1534-1539. 75. K. Maeda, K. Teramura, L. Daling, T. Takata, N. Saito, Y. Inoue and K. Domen, Nature, 2006, 440, 295-295.

221

76. R. Abe, K. Sayama and H. Sugihara, The Journal of Physical Chemistry B, 2005, 109, 16052-16061. 77. K. Fujihara, T. Ohno and M. Matsumura, Journal of the Chemical Society, Faraday Transactions, 1998, 94, 3705-3709. 78. C. A. Linkous, N. Z. Muradov and S. N. Ramser, Int J Hydrogen Energ, 1995, 20, 701-709. 79. A. Mills and S. Le Hunte, Journal of Photochemistry and Photobiology A: Chemistry, 1997, 108, 1-35. 80. D. M. Blake and C. Kennedy, Hydrogen reactor development and design for photofermentation and photolytic processes, Report CH-560-38480, NREL, Golden, CO, 2005. 81. A. Luque, J Appl Phys, 2011, 110, 031301-031301-031301-031319. 82. S. Haussener, C. X. Xiang, J. M. Spurgeon, S. Ardo, N. S. Lewis and A. Z. Weber, Energy Environ. Sci., 2012, 5, 9922-9935. 83. K. Zweibel, Sol. Energy Mater. Sol. Cells, 2000, 63, 375-386. 84. B. Kroposki, J. Levene, K. Harrison, P. K. Sen and F. Novachek, Electrolysis: Information and Opportunities for Electric Power Utilities, Report NREL/TP- 581-40605, NREL, Golden, CO, 2006. 85. J. J. Conti, P. D. Holtberg, J. A. Beamon, A. M. Schaal, J. C. Ayoub and J. T. Turnure, Annual Energy Outlook 2011, Report DOE/EIA-0383(2011), Washington, DC, 2011. 86. R. G. Lemus and J. M. Martínez Duart, Int J Hydrogen Energ, 2010, 35, 3929- 3936. 87. N. A. Kelly, T. L. Gibson and D. B. Ouwerkerk, Int J Hydrogen Energ, 2011, 36, 15803-15825. 88. D. B. Feldman, G.; Margolis, R.; Wiser, R.; Darghouth, N; Goodrich, A, Photovoltaic (PV) Pricing Trends: Historical, Recent, and Near-Term Projections, Report DOE/GO-102012-3839, Golden, CO, 2012. 89. M. Ruth, M. Laffen and T. A. Timbario, Hydrogen pathways: Cost, well-to- wheels energy use, and emissions for the current technology status of seven hydrogen production, delivery and distribution scenarios, Report TP-6A1-46612, NREL, Golden, CO, 2009. 90. D. L. Block, in Proceedings of the 12th World Hydrogen Energy Conference, Buenos Aires, Argentina, 1998, vol. 1, pp. 185-194. 91. Y. Fukasawa, K. Takanabe, A. Shimojima, M. Antonietti, K. Domen and T. Okubo, Chemistry-an Asian Journal, 2011, 6, 103-109. 92. C. T. Ho, K. B. Low, R. F. Klie, K. Maeda, K. Domen, R. J. Meyer and P. T. Snee, J. Phys. Chem. C, 2011, 115, 647-652. 93. M. Higashi, K. Domen and R. Abe, Journal of the American Chemical Society, 2012, 134, 6968-6971. 94. H. B. Michaelson, J Appl Phys, 1977, 48, 4729-4733. 95. O. Piotrowski, C. Madore and D. Landolt, Electrochimica Acta, 1999, 44, 3389- 3399. 96. D. A. Vermilyea, Acta Metallurgica, 1958, 6, 166-171. 97. R. Chandrasekharan, I. Park, R. I. Masel and M. A. Shannon, J Appl Phys, 2005, 98.

222

98. P. Kofstad, Journal of the Institute of Metals, 1962, 90, 253. 99. A. L. Patterson, Physical Review, 1939, 56, 978-982. 100. P. Patnaik, McGraw-Hill, 2003. 101. G. Brauer, J. Weidlein and J. Strahle, Zeitschrift Fur Anorganische Und Allgemeine Chemie, 1966, 348, 298-308. 102. M. Stavrev, D. Fischer, C. Wenzel, K. Drescher and N. Mattern, Thin Solid Films, 1997, 307, 79-88. 103. D. M. Mattox, Handbook of Physical Vapor Deposition (PVD) Processing, William Andrew Publishing/Noyes, 1998. 104. R. G. Gordon, MRS Bull., 2000, 25, 52-57. 105. G. C. Bond, Platinum Metals Rev., 1979, 23, 46-53. 106. N. P. Bansal and R. H. Doremus, Handbook of glass properties, Academic Press Inc., United States, 1986. 107. R. F. Pierret, Semiconductor Device Fundamentals, Addison-Wesley Publishing Company, 1996. 108. B. H. Michael, in The Measurement, Instrumentation and Sensors Handbook on CD-ROM, CRC Press, 1999. 109. I. P. T. Institute, Silicon Electrical Properties, http://www.ioffe.rssi.ru/SVA/NSM/Semicond/Si/electric.html, Accessed July 23, 2013. 110. S. S. Shinde, R. A. Bansode, C. H. Bhosale and K. Y. Rajpure, Journal of Semiconductors, 2011, 32, 0130011 - 0130018. 111. R. Chwang, B. J. Smith and C. R. Crowell, Solid-State Electronics, 1974, 17, 1217-1227. 112. B. A. Pinaud, P. C. K. Vesborg and T. F. Jaramillo, The Journal of Physical Chemistry C, 2012, 116, 15918-15924. 113. J. Yang, D. Wang, H. Han and C. Li, Accounts of Chemical Research, 2013. 114. J. A. Seabold and K.-S. Choi, Journal of the American Chemical Society, 2012, 134, 2186-2192. 115. J. A. Seabold and K.-S. Choi, Chemistry of Materials, 2011, 23, 1105-1112. 116. R. L. Grimm, M. J. Bierman, L. E. O’Leary, N. C. Strandwitz, B. S. Brunschwig and N. S. Lewis, The Journal of Physical Chemistry C, 2012, 116, 23569-23576. 117. S. Trasatti, Electrochimica Acta, 1984, 29, 1503-1512. 118. Y. Lee, J. Suntivich, K. J. May, E. E. Perry and Y. Shao-Horn, The Journal of Physical Chemistry Letters, 2012, 3, 399-404. 119. L. Trotochaud, T. J. Mills and S. W. Boettcher, The Journal of Physical Chemistry Letters, 2013, 4, 931-935. 120. K. Fukami, K. Kobayashi, T. Matsumoto, Y. L. Kawamura, T. Sakka and Y. H. Ogata, J Electrochem Soc, 2008, 155, D443-D448. 121. Y. Zhao, E. A. Hernandez-Pagan, N. M. Vargas-Barbosa, J. L. Dysart and T. E. Mallouk, The Journal of Physical Chemistry Letters, 2011, 2, 402-406. 122. T. Nakagawa, C. A. Beasley and R. W. Murray, The Journal of Physical Chemistry C, 2009, 113, 12958-12961. 123. Y. Moriya, T. Takata and K. Domen, Coordination Chemistry Reviews, 2013, 257, 1957-1969.

223

124. A. Kleiman-Shwarsctein, A. B. Laursen, F. Cavalca, W. Tang, S. Dahl and I. Chorkendorff, Chemical Communications, 2012, 48, 967-969. 125. S. H. Chuang, R. H. Gao, D. Y. Wang, H. P. Liu, L. M. Chen and M. Y. Chiang, J. Chin. Chem. Soc., 2010, 57, 932-937. 126. F. F. Abdi and R. van de Krol, The Journal of Physical Chemistry C, 2012, 116, 9398-9404. 127. K. Maeda, N. Nishimura and K. Domen, Applied Catalysis A: General, 2009, 370, 88-92. 128. C. Stampfl and A. J. Freeman, Physical Review B, 2005, 71, 024111. 129. N. Terao, Japanese Journal of Applied Physics, 1971, 10, 248-&. 130. Y. Lee, K. Nukumizu, T. Watanabe, T. Takata, M. Hara, M. Yoshimura and K. Domen, Chemistry Letters, 2006, 35, 352-353. 131. S. Lecuyer, A. Quemerais and G. Jezequel, Surface and Interface Analysis, 1992, 18, 257-261. 132. A. Dabirian and R. van de Krol, Applied Physics Letters, 2013, 102, 033905- 033904. 133. L. Yuliati, J.-H. Yang, X. Wang, K. Maeda, T. Takata, M. Antonietti and K. Domen, Journal of Materials Chemistry, 2010, 20, 4295-4298. 134. S. F. Ho, S. Contarini and J. W. Rabalais, J. Phys. Chem., 1987, 91, 4779-4788. 135. M. Higashi, K. Domen and R. Abe, Journal of the American Chemical Society, 2013. 136. R. O. Jones and O. Gunnarsson, Reviews of Modern Physics, 1989, 61, 689-746. 137. I. E. Castelli, T. Olsen, S. Datta, D. D. Landis, S. Dahl, K. S. Thygesen and K. W. Jacobsen, Energy Environ. Sci., 2012, 5, 5814-5819. 138. Y. Ping, D. Rocca and G. Galli, Physical Review B, 2013, 87, 165203. 139. M. Li, W. Luo, D. Cao, X. Zhao, Z. Li, T. Yu and Z. Zou, Angewandte Chemie International Edition, 2013, n/a-n/a. 140. F. L. Souza, K. P. Lopes, E. Longo and E. R. Leite, Physical Chemistry Chemical Physics, 2009, 11, 1215-1219. 141. R. R. Sawant, S. S. Shinde, C. H. Bhosale and K. Y. Rajpure, Solar Energy, 2010, 84, 1208-1215. 142. R. v. d. Krol and M. Grätzel, eds., Photoelectrochemical hydrogen production, Springer, New York, 2012. 143. K. Rajeshwar, in Encyclopedia of Electrochemistry, Wiley-VCH Verlag GmbH & Co. KGaA, 2007. 144. H. Masuda and K. Fukuda, Science, 1995, 268, 1466-1468. 145. F. Y. Li, L. Zhang and R. M. Metzger, Chemistry of Materials, 1998, 10, 2470- 2480. 146. J. Martin, C. V. Manzano and M. Martin-Gonzalez, Microporous Mesoporous Mat., 2012, 151, 311-316. 147. L. Zaraska, G. D. Sulka and M. Jaskula, J. Solid State Electrochem., 2011, 15, 2427-2436. 148. H. Masuda, K. Yada and A. Osaka, Jpn. J. Appl. Phys. Part 2 - Lett., 1998, 37, L1340-L1342. 149. S. Ono, M. Saito, M. Ishiguro and H. Asoh, J Electrochem Soc, 2004, 151, B473- B478.

224

150. A. Mozalev, G. Gorokh, M. Sakairi and H. Takahashi, Journal of Materials Science, 2005, 40, 6399-6407. 151. J. C. Hulteen and R. P. Vanduyne, J. Vac. Sci. Technol. A-Vac. Surf. Films, 1995, 13, 1553-1558. 152. J. C. Hulteen, D. A. Treichel, M. T. Smith, M. L. Duval, T. R. Jensen and R. P. Van Duyne, J. Phys. Chem. B, 1999, 103, 3854-3863. 153. C. L. Haynes and R. P. Van Duyne, J. Phys. Chem. B, 2001, 105, 5599-5611. 154. J. Yu, Q. F. Yan and D. Z. Shen, ACS Applied Materials & Interfaces, 2010, 2, 1922-1926. 155. S. J. Barcelo, S.-T. Lam, G. A. Gibson, X. Sheng and D. Henze, Nanosphere lithography based technique for fabrication of large area, well ordered metal particle arrays, HP Laboratories, 2012. 156. C. Geng, L. Zheng, J. Yu, Q. F. Yan, T. B. Wei, X. Q. Wang and D. Z. Shen, Journal of Materials Chemistry, 2012, 22, 22678-22685. 157. S. Astilean, Romanian Reports in Physics, 2004, 56, 340-345. 158. C. L. Cheung, R. J. Nikolic, C. E. Reinhardt and T. F. Wang, Nanotechnology, 2006, 17, 1339-1343. 159. A. Plettl, F. Enderle, M. Saitner, A. Manzke, C. Pfahler, S. Wiedemann and P. Ziemann, Advanced Functional Materials, 2009, 19, 3279-3284. 160. A. Picard and G. Turban, Plasma Chem. Plasma Process., 1985, 5, 333-351. 161. R. Hsiao and D. Miller, J Electrochem Soc, 1996, 143, 3266-3270. 162. C. C. Sheng, Y. Y. Cai, E. M. Dai and C. H. Liang, Chin. Phys. B, 2012, 21. 163. S. N. Wosu, Journal of Materials Science, 2007, 42, 4087-4097. 164. C. G. Zoski, ed., Handbook of electrochemistry, Elsevier, Amsterdam, 2007. 165. A. J. Bard and L. R. Faulkner, Electrochemical methods: Fundamentals and applications, Wiley, New York, 1980. 166. J. L. Zhang, M. B. Vukmirovic, Y. Xu, M. Mavrikakis and R. R. Adzic, Angewandte Chemie-International Edition, 2005, 44, 2132-2135. 167. L. Timperman, Y. J. Feng, W. Vogel and N. Alonso-Vante, Electrochimica Acta, 2010, 55, 7558-7563. 168. R. W. Revie and H. H. Uhlig, in Corrosion and corrosion control: an introduction to corrosion science and engineering, Wiley-Interscience, Hoboken, 4th edn., 2008. 169. S. Barnartt, J Electrochem Soc, 1959, 106, 991-994. 170. L. D. Burke, V. J. Cunnane and B. H. Lee, J Electrochem Soc, 1992, 139, 399- 406. 171. L. D. Burke and J. F. Osullivan, Electrochimica Acta, 1992, 37, 2087-2094. 172. G. Guenther, G. Schierning, R. Theissmann, R. Kruk, R. Schmechel, C. Baehtz and A. Prodi-Schwab, J Appl Phys, 2008, 104. 173. C. A. Huang, K. C. Li, G. C. Tu and W. S. Wang, Electrochimica Acta, 2003, 48, 3599-3605. 174. M. M. Jaksic, B. Johansen and R. Tunold, Int J Hydrogen Energ, 1993, 18, 91- 110. 175. M. A. Martínez, J. Herrero and M. T. Gutiérrez, Electrochimica Acta, 1992, 37, 2565-2571. 176. E. Matveeva, J Electrochem Soc, 2005, 152, H138-H145.

225

177. H. Wang, C. Zhong, C. J. Jiang, X. Gu, J. Li and Y. M. Jiang, Acta Physico- Chimica Sinica, 2009, 25, 835-839. 178. A. M. Bond and F. Scholz, Zeitschrift Fur Chemie, 1990, 30, 117-129. 179. T. F. Jaramillo, J. Bonde, J. D. Zhang, B. L. Ooi, K. Andersson, J. Ulstrup and I. Chorkendorff, J. Phys. Chem. C, 2008, 112, 17492-17498. 180. J. D. Zhang, Q. J. Chi, S. J. Dong and E. K. Wang, Bioelectrochemistry and Bioenergetics, 1996, 39, 267-274. 181. Z. Chen, D. Cummins, B. N. Reinecke, E. Clark, M. K. Sunkara and T. F. Jaramillo, Nano Letters, 2011, 11, 4168-4175. 182. T. Minami, Semiconductor Science and Technology, 2005, 20, S35-S44. 183. A. Klein, J. Am. Ceram. Soc., 2013, 96, 331-345. 184. Handbook of transparent conductors, Springer, New York ;, 2010. 185. J. R. Bellingham, W. A. Phillips and C. J. Adkins, J Mater Sci Lett, 1992, 11, 263-265. 186. K. L. Chopra, S. Major and D. K. Pandya, Thin Solid Films, 1983, 102, 1-46. 187. M. Senthilkumar, J. Mathiyarasu, J. Joseph, K. L. N. Phani and V. Yegnaraman, Materials Chemistry and Physics, 2008, 108, 403-407. 188. N. R. Armstrong, A. W. C. Lin, M. Fujihira and T. Kuwana, Anal. Chem., 1976, 48, 741-750. 189. A. Kraft, H. Hennig, A. Herbst and K. H. Heckner, Journal of Electroanalytical Chemistry, 1994, 365, 191-196. 190. M. Pourbaix, Atlas of electrochemical equilibria in aqueous solutions, Pergamon Press, Oxford, 1966. 191. Q. Qiao, J. Beck, R. Lumpkin, J. Pretko and J. T. McLeskey Jr, Sol. Energy Mater. Sol. Cells, 2006, 90, 1034-1040. 192. A. Stadler, Materials, 2012, 5, 661-683. 193. H. Liu, V. Avrutin, N. Izyumskaya, Ü. Özgür and H. Morkoç, Superlattices and Microstructures, 2010, 48, 458-484. 194. I. E. Paulauskas, G. E. Jellison, L. A. Boatner and G. M. Brown, International Journal of Electrochemistry, 2011, 2011. 195. M. Cai and S. M. Park, J Electrochem Soc, 1996, 143, 2125-2131. 196. J. Han, W. Qiu and W. Gao, Journal of Hazardous Materials, 2010, 178, 115- 122. 197. B. Hammer and J. K. Norskov, Nature, 1995, 376, 238-240. 198. P. Vanýsek, in CRC handbook of chemistry and physics, ed. W. M. Haynes, CRC ; Taylor & Francis [distributor], Boca Raton, Fla.; London, 2012. 199. B. S. Yeo, S. L. Klaus, P. N. Ross, R. A. Mathies and A. T. Bell, ChemPhysChem, 2010, 11, 1854-1857. 200. L. D. Burke and J. F. Osullivan, Electrochimica Acta, 1992, 37, 585-594. 201. K. Ogura, S. Haruyama and K. Nagasaki, J Electrochem Soc, 1971, 118, 531-535. 202. T. Izumi, I. Watanabe and Y. Yokoyama, Journal of Electroanalytical Chemistry and Interfacial Electrochemistry, 1991, 303, 151-160. 203. K. Juodkazis, J. Juodkazytė, V. Jasulaitienė, A. Lukinskas and B. Šebeka, Electrochemistry Communications, 2000, 2, 503-507. 204. Y. Ling, J. C. Elkenbracht, W. F. Flanagan and B. D. Lichter, J Electrochem Soc, 1997, 144, 2689-2697.

226

205. J. P. Hoare, J Electrochem Soc, 1984, 131, 1808-1815. 206. M. Orlik and Z. Galus, in Encyclopedia of Electrochemistry, Wiley-VCH Verlag GmbH & Co. KGaA, 2007. 207. in The Economist,, 2013, vol. 2013. 208. C. O. A. Olsson and D. Landolt, Electrochimica Acta, 2003, 48, 1093-1104. 209. R. W. Revie and H. H. Uhlig, in Corrosion and corrosion control: an introduction to corrosion science and engineering, Wiley-Interscience, Hoboken, 4th edn., 2008. 210. AK Steel, 304/304L Stainless Steel Product Data Sheet, AK Steel, West Chester, OH, 2007. 211. Azom, Stainless Steel - Grade 304 (UNS S30400), http://www.azom.com/article.aspx?ArticleID=965, Accessed July 17, 2013. 212. J. M. Olivares-Ramirez, M. L. Campos-Cornelio, J. U. Godinez, E. Borja-Arco and R. H. Castellanos, Int J Hydrogen Energ, 2007, 32, 3170-3173. 213. B. Jegdic, D. M. Drazic and J. P. Popic, Journal of the Serbian Chemical Society, 2006, 71, 543-551. 214. S. Haupt and H. H. Strehblow, Corrosion Science, 1995, 37, 43-54. 215. J. Olsson and R. Qvarfort, Transpassive Corrosion of High Alloy Stainless Steels and Nickel Base Alloys, 2002. 216. L. J. Qiao and J. L. Luo, Corrosion, 1998, 54, 281-288. 217. H. E. Zittel and F. J. Miller, Anal. Chem., 1965, 37, 200-203. 218. S. Yamada and H. Sato, Nature, 1962, 193, 261-262. 219. W. E. Van der Linden and J. W. Dieker, Anal. Chim. Acta, 1980, 119, 1-24. 220. K. Kinoshita, Carbon: electrochemical and physicochemical properties, 1988. 221. G. Li and P. Miao, in Electrochemical Analysis of Proteins and Cells, Springer Berlin Heidelberg, 2013, pp. 5-18. 222. S. Michalkiewicz and M. Kaczor, Chemia Analityczna, 2004, 49, 121-128. 223. L. J. Kepley and A. J. Bard, Anal. Chem., 1988, 60, 1459-1467. 224. D. Jürgen and E. Steckhan, Journal of Electroanalytical Chemistry, 1992, 333, 177-193. 225. Q.-L. Zhao, L. Bao, Q.-Y. Luo, M. Zhang, Y. Lin, D.-W. Pang and Z.-L. Zhang, Biosensors and Bioelectronics, 2009, 24, 3003-3007. 226. Q.-L. Zhao, Z.-L. Zhang, L. Bao and D.-W. Pang, Electrochemistry Communications, 2008, 10, 181-185. 227. G. K. Kiema, M. Aktay and M. T. McDermott, Journal of Electroanalytical Chemistry, 2003, 540, 7-15. 228. A. Dekanski, J. Stevanović, R. Stevanović, B. Å. Nikolić and V. M. Jovanović, Carbon, 2001, 39, 1195-1205. 229. M. L. Bowers, J. Hefter, D. L. Dugger and R. Wilson, Anal. Chim. Acta, 1991, 248, 127-142. 230. M. L. Bowers and B. A. Yenser, Anal. Chim. Acta, 1991, 243, 43-53. 231. A. L. Beilby, T. A. Sasaki and H. M. Stern, Anal. Chem., 1995, 67, 976-980. 232. G. N. Kamau, Anal. Chim. Acta, 1988, 207, 1-16. 233. R. C. Engstrom and V. A. Strasser, Anal. Chem., 1984, 56, 136-141. 234. I.-F. Hu, D. H. Karweik and T. Kuwana, Journal of Electroanalytical Chemistry and Interfacial Electrochemistry, 1985, 188, 59-72.

227

235. G. N. Kamau, W. S. Willis and J. F. Rusling, Anal. Chem., 1985, 57, 545-551. 236. D. C. Thornton, K. T. Corby, V. A. Spendel, J. Jordan, A. Robbat, D. J. Rutstrom, M. Gross and G. Ritzler, Anal. Chem., 1985, 57, 150-155. 237. R. C. Engstrom, Anal. Chem., 1982, 54, 2310-2314. 238. IUPAC, in Compendium of Chemical Terminology, eds. A. D. McNaught and A. Wilkinson, Blackwell Scientific Publications, Oxford, 2006. 239. SPI Supplies, Highly Ordered Pyrolytic Graphite, http://www.2spi.com/catalog/new/hopgsub.php, Accessed July 21, 2013. 240. R. N. Goyal, S. Chatterjee and A. R. S. Rana, Talanta, 2010, 83, 149-155. 241. R. R. Moore, C. E. Banks and R. G. Compton, Anal. Chem., 2004, 76, 2677-2682. 242. J. Kibsgaard, J. V. Lauritsen, E. Lægsgaard, B. S. Clausen, H. Topsøe and F. Besenbacher, Journal of the American Chemical Society, 2006, 128, 13950- 13958. 243. H. B. R. Lee, S. H. Baeck, T. F. Jaramillo and S. F. Bent, Nano Letters, 2013, 13, 457-463. 244. B. Zhang and E. Wang, Electrochimica Acta, 1995, 40, 2627-2633. 245. R. J. Bowling, R. T. Packard and R. L. McCreery, Journal of the American Chemical Society, 1989, 111, 1217-1223. 246. F. Y. Chen, J. G. Liu, H. Chen and C. W. Yan, International Journal of Electrochemical Science, 2012, 7, 3750-3764. 247. C. A. Goss, J. C. Brumfield, E. A. Irene and R. W. Murray, Anal. Chem., 1993, 65, 1378-1389. 248. X. B. Ji, C. E. Banks, W. Xi, S. J. Wilkins and R. G. Compton, J. Phys. Chem. B, 2006, 110, 22306-22309. 249. Cole-Parmer, Chemical Compatibility Database, http://www.coleparmer.com/Chemical-Resistance, Accessed August 6, 2013. 250. E. J. G. Beeson, Light. Res. Technol., 1978, 10, 164-166. 251. A. S. T. M. International, in Standard Tables for Reference Solar Spectral Irradiances: Direct Normal and Hemispherical on 37 degree tilted surface, West Coshohocken, PA, 2003. 252. A. Midilli and I. Dincer, Int J Hydrogen Energ, 2007, 32, 511-524. 253. R. Ramachandran and R. K. Menon, Int J Hydrogen Energ, 1998, 23, 593-598. 254. R. Memming, Electrochimica Acta, 1980, 25, 77-88. 255. J. Nowotny, C. C. Sorrell, L. R. Sheppard and T. Bak, Int J Hydrogen Energ, 2005, 30, 521-544. 256. H. Gerischer, Solar photoelectrolysis with semiconductor electrodes, Springer- Verlag, 1979. 257. M. Gratzel, Nature, 2001, 414, 338-344. 258. R. Shinar and J. H. Kennedy, Solar Energy Materials, 1982, 6, 323-335. 259. B. P. Rai, Solar Cells, 1988, 25, 265-272. 260. A. R. Armstrong and P. G. Bruce, Nature, 1996, 381, 499-500. 261. C. C. Hu and T. W. Tsou, Electrochemistry Communications, 2002, 4, 105-109. 262. F. Jiao and H. Frei, Energy Environ. Sci., 2010, 3, 1018-1027. 263. S. P. Yu, M. J. Xi, X. Jin, K. F. Han, Z. M. Wang and H. Zhu, Catalysis Communications, 2010, 11, 1125-1128.

228

264. J. E. Post, Proceedings of the National Academy of Sciences of the United States of America, 1999, 96, 3447-3454. 265. P. A. Plunkert and T. S. Jones, Metal prices in the United States through 1998, Interior Department, U.S. Geological Survey, Washington, DC, 1999. 266. M. Morita, C. Iwakura and H. Tamura, Electrochimica Acta, 1977, 22, 325-328. 267. M. Morita, C. Iwakura and H. Tamura, Electrochimica Acta, 1979, 24, 357-362. 268. S. Trasatti, Journal of Electroanalytical Chemistry, 1980, 111, 125-131. 269. Y. Xu and M. A. A. Schoonen, American Mineralogist, 2000, 85, 543-556. 270. K. D. Kwon, K. Refson and G. Sposito, Geochimica Et Cosmochimica Acta, 2009, 73, 4142-4150. 271. D. M. Sherman, Geochimica Et Cosmochimica Acta, 2005, 69, 3249-3255. 272. H. Cao and S. L. Suib, Journal of the American Chemical Society, 1994, 116, 5334-5342. 273. K. N. Ferreira, T. M. Iverson, K. Maghlaoui, J. Barber and S. Iwata, Science, 2004, 303, 1831-1838. 274. G. C. Dismukes, R. Brimblecombe, G. A. N. Felton, R. S. Pryadun, J. E. Sheats, L. Spiccia and G. F. Swiegers, Accounts of Chemical Research, 2009, 42, 1935- 1943. 275. R. Brimblecombe, A. Koo, G. C. Dismukes, G. F. Swiegers and L. Spiccia, Journal of the American Chemical Society, 2010, 132, 2892-+. 276. N. Sakai, Y. Ebina, K. Takada and T. Sasaki, J. Phys. Chem. B, 2005, 109, 9651- 9655. 277. M. Nakayama, T. Kanaya, J. W. Lee and B. N. Popov, Journal of Power Sources, 2008, 179, 361-366. 278. J. S. Foord, R. B. Jackman and G. C. Allen, Philosophical Magazine a-Physics of Condensed Matter Structure Defects and Mechanical Properties, 1984, 49, 657- 663. 279. V. Dicastro and G. Polzonetti, Journal of Electron Spectroscopy and Related Phenomena, 1989, 48, 117-123. 280. A. A. Audi and P. M. A. Sherwood, Surface and Interface Analysis, 2002, 33, 274-282. 281. M. Oku, K. Hirokawa and S. Ikeda, Journal of Electron Spectroscopy and Related Phenomena, 1975, 7, 465-473. 282. B. V. Crist, XPS International, 2007. 283. M. Nakayama, S. Konishi, A. Tanaka and K. Ogura, Chemistry Letters, 2004, 33, 670-671. 284. N. Larabi-Gruet, S. Peulon, A. Lacroix and A. Chausse, Electrochimica Acta, 2008, 53, 7281-7287. 285. M. Nakayama, M. Fukuda, S. Konishi and T. Tonosaki, Journal of Materials Research, 2006, 21, 3152-3160. 286. F. Hu, K. C. Chan and T. M. Yue, Thin Solid Films, 2009, 518, 120-125. 287. V. A. Drits, E. Silvester, A. I. Gorshkov and A. Manceau, American Mineralogist, 1997, 82, 946-961. 288. J. E. Post and D. R. Veblen, American Mineralogist, 1990, 75, 477-489. 289. J. E. Post, P. J. Heaney and J. Hanson, Powder Diffraction, 2002, 17, 218-221. 290. E. A. Johnson and J. E. Post, American Mineralogist, 2006, 91, 609-618.

229

291. H. Matsui, J. Ju, T. Odaira and N. Toyota, Journal of the Physical Society of Japan, 2009, 78. 292. G. Yang, X. Zhao and J. Qi, Journal of Ocean University of China (English Edition), 2003, 2, 79-84. 293. I. Chorkendorff and J. W. Niemantsverdriet, Concepts of modern catalysis and kinetics, Wiley-VCH, Weinheim, 2007. 294. R. N. De Guzman, A. Awaluddin, Y. F. Shen, Z. R. Tian, S. L. Suib, S. Ching and C. L. Oyoung, Chemistry of Materials, 1995, 7, 1286-1292. 295. B. M. Kayes, H. A. Atwater and N. S. Lewis, J Appl Phys, 2005, 97. 296. B. Lanson, V. A. Drits, E. Silvester and A. Manceau, American Mineralogist, 2000, 85, 826-838. 297. L. Weinhardt, M. Blum, M. Bar, C. Heske, B. Cole, B. Marsen and E. L. Miller, J. Phys. Chem. C, 2008, 112, 3078-3082. 298. Y. Matsumoto, M. Omae, I. Watanabe and E. Sato, J Electrochem Soc, 1986, 133, 711-716. 299. Y. Matsumoto, Journal of Solid State Chemistry, 1996, 126, 227-234. 300. S. Trasatti, Pure and Applied Chemistry, 1986, 58, 955-966. 301. Y. V. Pleskov, V. M. Mazin, Y. E. Evstefeeva, V. P. Varnin, I. G. Teremetskaya and V. A. Laptev, Electrochemical and Solid State Letters, 2000, 3, 141-143. 302. S. A. Cheng, D. F. Xing, D. F. Call and B. E. Logan, Environ. Sci. Technol., 2009, 43, 3953-3958. 303. R. S. Haszeldine, Science, 2009, 325, 1647-1652. 304. C. Azar, K. Lindgren, E. Larson and K. Möllersten, Climatic Change, 2006, 74, 47-79. 305. K. Rabaey and R. A. Rozendal, Nat. Rev. Microbiol., 2010, 8, 706-716. 306. B. E. Logan, Microbial Fuel Cells, John Wiley & Sons, Inc., Hoboken, NJ, 2008. 307. D. H. Park, M. Laivenieks, M. V. Guettler, M. K. Jain and J. G. Zeikus, Appl. Environ. Microbiol., 1999, 65, 2912-2917. 308. C. L. Bird and A. T. Kuhn, Chemical Society Reviews, 1981, 10, 49-82. 309. P. Schönheit, J. Moll and R. Thauer, Arch. Microbiol., 1980, 127, 59-65. 310. E. S. Kooij, R. W. Despo, F. P. J. Mulders and J. J. Kelly, Journal of Electroanalytical Chemistry, 1996, 406, 139-146. 311. E. L. Warren, S. W. Boettcher, M. G. Walter, H. A. Atwater and N. S. Lewis, J. Phys. Chem. C, 2011, 115, 594-598. 312. S. W. Boettcher, J. M. Spurgeon, M. C. Putnam, E. L. Warren, D. B. Turner- Evans, M. D. Kelzenberg, J. R. Maiolo, H. A. Atwater and N. S. Lewis, Science, 2010, 327, 185-187. 313. K. P. Kuhl, E. R. Cave, D. N. Abram and T. F. Jaramillo, Energy Environ. Sci., 2012, 5, 7050-7059. 314. J. F. Stargardt and F. M. Hawkridge, Anal. Chim. Acta, 1983, 146, 1-8. 315. G. Levey and T. W. Ebbesen, J. Phys. Chem., 1983, 87, 829-832. 316. P. Li, W. L. Fan, Y. L. Li, H. G. Sun, X. F. Cheng, X. A. Zhao and M. H. Jiang, Inorganic Chemistry, 2010, 49, 6917-6924. 317. M. W. Lumey and R. Dronskowski, Zeitschrift Fur Anorganische Und Allgemeine Chemie, 2003, 629, 2173-2179. 318. E. Orhan, F. Tessier and R. Marchand, Solid State Sciences, 2002, 4, 1071-1076.

230

319. R. Abe, M. Higashi and K. Domen, Journal of the American Chemical Society, 2010, 132, 11828-11829. 320. J. H. Hsieh, C. C. Chang, J. S. Cherng and F. Y. Hsu, Thin Solid Films, 2009, 517, 4711-4714. 321. J. H. Hsieh, C. Li and H. C. Liang, Thin Solid Films, 2011, 519, 4699-4704. 322. C. M. Leroy, R. Sanjines, K. Sivula, M. Cornuz, N. Xanthopoulos, V. Laporte and M. Gratzel, in Emrs Symposium T: Materials for Solar Hydrogen Via Photo- Electrochemical Production, ed. M. Gratzel, 2012, vol. 22, pp. 119-126. 323. M. Hara, T. Takata, J. N. Kondo and K. Domen, Catalysis Today, 2004, 90, 313- 317. 324. S. Ito, K. R. Thampi, P. Comte, P. Liska and M. Gratzel, Chemical Communications, 2005, 268-270. 325. S. Banerjee, S. K. Mohapatra and M. Misra, Chemical Communications, 2009, 7137-7139. 326. M. Y. Tsang, N. E. Pridmore, L. J. Gillie, Y. H. Chou, R. Brydson and R. E. Douthwaite, Advanced Materials, 2012, 24, 3406-3409. 327. R. Nakamura, T. Tanaka and Y. Nakato, J. Phys. Chem. B, 2005, 109, 8920-8927. 328. Z. Wang, J. G. Hou, C. Yang, S. Q. Jiao, K. Huang and H. M. Zhu, Energy Environ. Sci., 2013, 6, 2134-2144.

231