Growth and Characterization of Gallium Nitride Nanowire LEDs for Application as On-Chip Optical Interconnects

Transcription

1 University of Colorado, Boulder CU Scholar Mechanical Engineering Graduate Theses & Dissertations Mechanical Engineering Spring Growth and Characterization of Gallium Nitride Nanowire LEDs for Application as On-Chip Optical Interconnects Matt Brubaker University of Colorado at Boulder, Follow this and additional works at: Part of the Materials Science and Engineering Commons, and the Nanoscience and Nanotechnology Commons Recommended Citation Brubaker, Matt, "Growth and Characterization of Gallium Nitride Nanowire LEDs for Application as On-Chip Optical Interconnects" (2012). Mechanical Engineering Graduate Theses & Dissertations This Dissertation is brought to you for free and open access by Mechanical Engineering at CU Scholar. It has been accepted for inclusion in Mechanical Engineering Graduate Theses & Dissertations by an authorized administrator of CU Scholar. For more information, please contact

2 GROWTH AND CHARACTERIZATION OF GALLIUM NITRIDE NANOWIRE LEDS FOR APPLICATION AS ON-CHIP OPTICAL INTERCONNECTS by Matt Brubaker B.S., The Pennsylvania State University 1995 M.S., The Pennsylvania State University 1998 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirement for the degree of Doctor of Philosophy Department of Mechanical Engineering 2012

3 This thesis entitled: Growth and Characterization of Gallium Nitride Nanowire LEDs for Application as On-Chip Optical Interconnects written by Matt Brubaker has been approved for the Department of Mechanical Engineering Victor M. Bright Committee Chairman, Dept. of Mechanical Engineering Kris Bertness Thesis Advisor, National Institute of Standards and Technology Norman Sanford National Institute of Standards and Technology Scott Bunch Dept. of Mechanical Engineering Bart Van Zeghbroeck Dept. of Electrical, Computer, and Energy Engineering Date The final copy of this thesis has been examined by the signatories, and we find that both the content and the form meet acceptable presentation standards of scholarly work in the above mentioned discipline.

4 Brubaker, Matt D. (Ph.D., Department of Mechanical Engineering) iii Growth and Characterization of Gallium Nitride Nanowire LEDs for Application as On-Chip Optical Interconnects Thesis directed by Dr. Kris Bertness, Dr. Norman Sanford, and Professor Victor M. Bright ABSTRACT Gallium nitride (GaN) nanowires have potential as nanoscale optoelectronic building blocks that can be functionally integrated with silicon MEMS and IC devices. This dissertation presents an overview of the synthesis, characterization, and application of GaN nanowire lightemitting-diodes (LEDs) grown by plasma-assisted molecular beam epitaxy (MBE). Specifically, this research demonstrates discrete axial p-n junction nanowires that produce ultra-violet (UV) electroluminescence at ~40 nw optical power. It further demonstrates that a two-nanowire optical interconnect device can be fabricated from axial p-n junction nanowires with lightemitting and photoconductive capabilities. The nanowire structures obtained from MBE growth were found to depend sensitively on the morphology and crystallographic polarity of the underlying Aluminum Nitride (AlN) nucleation layer. These observations were enabled by piezoresponse force microscopy, which was developed and validated against polarity sensitive etching using uniform and mixed polarity AlN layers. The polarity and overall morphology of the AlN layers could be controlled by the V/III flux ratio and substrate temperature during MBE growth. GaN nanowires were observed to propagate the structural characteristics and crystallographic polarity of the underlying AlN layer, and in some cases a differential growth rate with respect to polarity was observed.

5 Band-edge electroluminescence was obtained in axial p-n junction nanowires that iv incorporated a thin AlGaN electron blocking layer in the p-region of the device. Electroluminescence was below detection limits for p-n junction nanowires with no blocking layer, despite diode-like I-V characteristics and optically measured internal quantum efficiencies (IQEs) of ~1 %. I-V measurements of the p-regions in p-n junction nanowires, as well as nanowires doped with Mg only, indicate low p-type conductivity and asymmetric Schottky-like p-contacts. These observations, in conjunction with device models, suggest that imbalanced carrier injection from the junction and p-contact can produce significant non-radiative losses. The role of the blocking layer in these devices is attributed to a reduction in the electron overflow current, which permits larger biasing and increased hole injection at the p-contact. This dissertation also discusses the fabrication, device characteristics, and optical coupling of a two-nanowire device comprising GaN nanowires with light-emitting and photoconductive capabilities. Axial p-n junctions were transferred to a non-native substrate, and selectively contacted to form discrete optical source or detector nanowire components. The performance of the individual nanowires was characterized by electroluminescence measurements and spectrally-resolved photoconductivity measurements, which were then compared with the coupled behavior.

6 v This dissertation is dedicated to my children, Henry and Maya, and to my hope that they enjoy the same freedom and opportunity to pursue their dreams as I did.

7 vi The sun is but a morning star. Walden Henry David Thoreau Thou shalt not partake of decaf Milo Aukerman The Descendents

8 ACKNOWLEDGEMENTS vii First and foremost, I would like to thank my NIST advisor, Kris Bertness, an untiring mentor that trained me in the intricacies of MBE, who kept an even keel through all my UHV mishaps, and taught me everything from melting an aluminum cell to hyphenating compound adjectives (well, I m still working on the hyphenation). I would also like to gratefully thank my CU advisor, Victor Bright, who gave me the chance to return from the comforts of commerce to the rigors of academia, who facilitated key opportunities, and knew when to say when are we going to finally make some devices? Special thanks are also due to Norm Sanford, for impressing upon me the importance of carrier confinement, training me on photoconductivity measurements, and for waxing poetic about the finer points of textbooks written in the 50s. Many thanks to the Committee members; Victor Bright, Kris Bertness, Norman Sanford, Scott Bunch, and Bart Van Zeghbroeck, who willingly signed up for the herculean task of reading this dissertation and administering the final exam. I am indebted to the members of the GaN nanowire research group at NIST, past and present. Sincere thanks go out to Paul Blanchard, who fabricated most of the devices in this study and measured many of the I-V curves for the p-type nanowires. Thank you to John Schlager, who provided advice and performed numerous optical measurements that were critical to this work. I apologize for not including the first nanowire EL spectra that we worked so hard to obtain. Aric Sanders, many thanks for teaching me about all forms of analytical measurements, for the numerous brainstorming sessions on the way to Taco Bell (oops, I mean Whole Foods), and for sharing insight on how timely vacuuming sessions during naptime will teach one s kids to sleep through the night. Thanks are also due to Alexana Roshko, for help

9 with SEM imaging, providing advice on AFM measurements, and for keeping a straight face viii while I expounded on crackpot theories about plasma sources. I am grateful to Todd Harvey, for helping me with MBE growth runs and maintenance, for coming into the lab to bail me out on more UHV mishaps in the middle of the night, and for showing me how to single handedly decommission an MBE system with a sawzall. Thank you Shannon Duff, for helping me with lift-off processing and providing data on the free hole concentrations in the p-gan films. Thanks are also due to Andrew Herrero, for showing us what the bottom of an annealed p-type contact looks like. Sincere appreciation goes to Lorelle Mansfield, for building the MBE2 growth system, training me on its operation, and providing a starting point for my research. I would also like to thank Devin Rourke for helping me with the initial polarity sensitive etch tests. My appreciation goes out to Joel Weber, whose intro to research project was the precursor for AlN polarity measurements by PFM. I would like to acknowledge assistance and advice from many people at CU Boulder, including the members of Victor Bright s research group. In particular, Jason Gray provided several days of e-beam lithography and related processing for the optical interconnect devices. Thanks to Sharon Anderson, for helping me to navigate the graduate school protocols. The Nanomaterials Characterization Facility and staff provided measurement capabilities and assistantship at the outset of this research project. Thanks are due to colleagues at NIST Gaithersburg for key contributions, technical assistance, and fruitful discussions. Igor Levin performed the CBED polarity analysis on GaN nanowires using a TEM sample that was prepared by Sandra Claggett. Albert Davydov provided XRD measurements of the AlN films and helped me to solidify my understanding of the growth

10 ix mechanism through extensive discussions. I would also like to acknowledge helpful discussions on device modeling with Ben Klein of Georgia Tech. I am grateful for the financial support provided for this research by the National Institute of Standards and Technology under the professional research experience program (PREP) at CU Boulder. Partial support was also provided by the DARPA Center on Nanoscale Science and Technology for Integrated Micro/Nano-Electromechanical Transducers (imint). An assistantship for the first year of my studies was also provided by the CU Mechanical Engineering department. Last and certainly not least, I am deeply grateful for the ongoing role of my family over the course of my academic development. Mom and Dad, thank you for nurturing that initial curiosity and encouraging it to grow throughout more than 20 years of education. My wife, Maggie, deserves recognition on many different levels; most notably for shouldering a hugely disproportionate share of the parental responsibilities during the writing of this manuscript, and for raising two smart, loving, and well-adjusted children despite my long hours in the lab and library. This dissertation would not have been started, let alone completed, had it not been for your willingness to cast security aside and explore new possibilities. Henry and Maya, thank you for running across the courtyard at the day s end to give me a hug; this dispelled many frustrations that I brought home. You also taught me that we don t just pursue academic ambitions for personal accomplishment, but to improve the world you inherit.

11 TABLE OF CONTENTS x Abstract... iii Acknowledgements... vii Table of Contents... x List of Figures... xiii List of Tables... xvi 1.Introduction Unique Characteristics of GaN Nanowires Objectives, Scope, and Organization of this Research Primary Accomplishments Background and Literature Review Gallium Nitride Plasma-Assisted Molecular Beam Epitaxy Planar Film Growth by PAMBE GaN Nanowire Growth by PAMBE GaN Nanowire Devices Surface depletion effects GaN Nanowire LEDs GaN Nanowire Photodetectors Experimental Details MBE Growth Systems MBE MBE

12 xi 3.2 Substrates and Preparation Ex-situ Analytical Characterization FESEM Imaging Atomic Force Microscopy (AFM) and Piezoresponse Force Microscopy (PFM) Polarity Sensitive Etching Secondary Ion Mass Spectroscopy (SIMS) Transmission Electron Microscopy (TEM)/ Convergent Beam Electron Diffraction (CBED) Photoluminescence (PL) Measurements Device Fabrication and Characterization Electrical Characterization Optical Characterization MBE Growth of GaN Nanowires Growth Process Diagnostics Group III Fluxes Active Nitrogen Composition and Flux Doping Fluxes MBE3 Growth Process Diagnostics Nanowire Nucleation AlN Polarity Measurements AlN Polarity for Various Growth Conditions Effect of AlN Polarity on MBE2 Nanowire Morphology MBE3 Nanowire Nucleation Process Post-Nucleation Nanowire Propagation

13 xii 5.Axial p-n Junction Nanowires for Nanoscale LEDs and Optical Interconnects GaN Nanowire LEDs Mg-Doped Nanowires Axial p-n Junction Nanowires Electron Blocking Layers Model of GaN Nanowire LEDs Nanowire Optical Interconnects Conclusions and Suggestions for Future Work References Appendix 1 Calculation of TFE Current Appendix 2 List of Abbreviations Appendix 3 List of Symbols

14 LIST OF FIGURES xiii 2.1. Bandgap energy and lattice constant for III-nitride materials [Schubert 2006] Crystallographic properties of III-nitride materials Free hole and magnesium concentrations in Mg:GaN films[kaufmann 2000] Growth diagram for planar GaN films using PAMBE [Heying 2000] Vapor pressures of elements and dopants in PAMBE growth of III-nitrides [Honig 1969] FESEM images of GaN nanowires grown by PAMBE [Bertness 2008] Growth diagram for GaN nanowires using PAMBE [Fernandez-Garrido 2009] Schematic diagram of cylindrical surface depletion model for p-type nanowire Axial heterostructure junction nanowire LED array grown by PAMBE [Kikuchi 2004] Discrete axial p-n junction nanowire LED grown by HVPE [Kim 2003] Discrete core-sleeve heterostructure nanowire LED grown by MOCVD [Qian 2005] Photoconductivity in GaN nanowires [Calarco 2005] Schematic diagram and photo of MBE2 growth system Photo of MBE3 growth system Schematic diagram of idealized axial p-n junction nanowire growth FESEM images of example nanowire morphologies Reflectance measurements of Ga sticking coefficient in MBE Reflectance measurements of Ga cell flux in MBE AlN growth rate curve for MBE Identification of active nitrogen species observed in OES and QMS measurements QMS measurements of atomic and molecular nitrogen species OES and QMS instrumentation for characterizing active nitrogen flux in MBE

15 xiv 4.9. Comprehensive results from QMS and OES nitrogen plasma source measurements QMS measurements of atomic and molecular nitrogen ions QMS measurements of background atomic nitrogen for MBE Plasma startup transients for MBE Long-term drift in plasma source output characteristics QMS measurements of Mg flux for MBE AlN growth rate curve for MBE OES plasma source measurements for MBE Schematic diagram of PFM polarity measurements for III-nitrides PFM and polarity-sensitive etch measurements of unipolar AlN films PFM measurement of mixed polarity AlN film Correlation of PFM and polarity-sensitive etching for a mixed polarity AlN film AlN films grown under various V/III ratios for polarity measurements Effect of V/III ratio on MBE3 AlN polarity at constant N flux Effect of V/III ratio on MBE2 AlN polarity at constant N flux Effect of V/III ratio on MBE2 AlN polarity at constant Al flux Polarity measurement of MBE1 AlN film Effect of high nucleation temperature on MBE3 AlN films nucleated at 780 C Effect of AlN layer on GaN nanowire structure for MBE TEM image and CBED polarity measurements for GaN nanowires grown in MBE Polarity of GaN matrix layer in MBE GaN nanowire morphology for optimized growth process in MBE Effect of Ga flux on axial and lateral nanowire growth rates in MBE Low-temperature Ga-rich nanowire diameter expansion process for MBE

16 xv 5.1. FESEM images of as-grown nanowires before and after SIMS SIMS scans for N150 and N151 Mg:GaN nanowires SIMS scans for as-grown and dispersed N176 Mg:GaN nanowires Average Mg concentration versus flux for Mg:GaN nanowires grown in MBE FESEM images and PL measurements of N274 Mg:GaN nanowires Electrical characteristics of N274 Mg:GaN nanowires Electrical characteristics of N216 Mg:GaN nanowires I-V characteristics of N140 and N272 axial p-n junction nanowires I-V characteristics of n- and p- sections of N140 and N272 axial p-n junction nanowires Schematic band diagrams of axial p-n junction nanowires illustrating effect of EBL I-V characteristics of D048 and D049 p-n junction nanowires and related structures EL spectra, EL images, and FESEM images of D048 nanowires Schematic band diagram and equivalent circuit used to model injection efficiency Calculated p-contact Schottky barrier characteristics Calculated injection efficiencies for axial p-n junction nanowires Ideality factors for p-type, n-p/ebl, and p-n junction nanowires Schematic diagram and images of optical interconnect device LED nanowire characteristics of optical interconnect device PC nanowire characteristics of optical interconnect device Coupled LED/PC measurements of optical interconnect device Potential causes of low LED/PC NW coupling in optical interconnect device Suggested improvement to NW structure for optical interconnect device

17 LIST OF TABLES xvi 3.1. MBE growth systems used for experiments Active nitrogen threshold ionization energies from QMS measurements Effect of V/III flux ratio on polarity of MBE3 grown AlN films Effect of V/III flux ratio on polarity of MBE2 grown AlN films and GaN nanowires Growth conditions and device details for Mg:GaN nanowire experiments Growth conditions and device details for axial p-n junction nanowire experiments Parameters used for modeling injection efficiency in axial p-n junction nanowires

18 1 1. Introduction 1.1 Unique Characteristics of GaN Nanowires Semiconductor nanowires have emerged as an increasingly important subset of nanoscience research in the last decade [1]. Nanowires are roughly cylindrical in form, with dimensions on the order of tens or hundreds of nanometers in diameter, and can be synthesized in excess of 10 micrometers long. While the minimum dimensions are large by nanoscience standards, the diameter is sufficiently small that the surface proximity significantly influences the nanowire properties. Consequently, semiconductor nanowires can exhibit characteristics that differ substantially from their bulk counterparts for a given material system. In particular, III- Nitride (Ga,Al,In-N) nanowires and their related alloys are particularly compelling for the scientific and technological reasons outlined below: Synthesis of Defect-Free Material High-quality bulk GaN substrates are not commonly available for epitaxial film growth. Hence, GaN films are typically produced by heteroepitaxial growth techniques on highly lattice-mismatched substrates, and consequently have high defect densities. In contrast, free-standing GaN nanowires can be grown as defect-free single crystals that coherently accommodate lattice mismatch strain. Discrete Nanoscale Components GaN nanowires can be removed from the original growth substrate and placed onto nonnative substrates to produce discrete nanowire devices. This facilitates integration of the optoelectronic capabilities of nitride materials with non-photonic silicon IC and MEMS devices.

19 Surface Sensitivity The nanowire conductivity is partially determined by free carrier depletion effects 2 resulting from surface charge. Variation in surface charge resulting from photogenerated carriers or chemical adsorbates can be detected through changes in the nanowire conductivity, enabling photodetectors and chemical sensors. Novel Device Geometries The cylindrical structure provides unique advantages for certain device applications. In particular, GaN nanowires support optical waveguiding modes, providing larger light extraction efficiency in nanowire LEDs, or the optical cavity for nanowire laser diodes. 1.2 Objectives, Scope, and Organization of this Research The primary objective of this thesis is to demonstrate electroluminescence in a discrete axial p-n junction GaN nanowire grown by plasma-assisted molecular beam epitaxy (MBE). The secondary objectives are to demonstrate that axial p-n junction nanowires are functional nanoscale components, capable of optical emission and detection; and can be integrated onto non-native substrates as optical interconnects. Multiple sub-studies are implicit to these objectives, and define the overall scope of this study. An overview of the content and organization of the chapters in this thesis are described in the following paragraphs. A primer on gallium nitride is presented in Chapter 2; as relevant to the synthesis, characterization, and application of axial p-n junction nanowires discussed in the experimental results. The basic crystallography and extrinsic dopants of GaN are presented, followed by discussion of planar film and nanowire growth by plasma-assisted molecular beam epitaxy. A literature review of previous work on GaN nanowire LEDs is included, and is limited to studies

20 3 that produced single nanowire LEDs. Surface depletion effects are also discussed, as pertaining to conductivity and photodetection in GaN nanowires. The equipment and procedures used for experimental studies are discussed in Chapter 3. Several MBE growth systems and their analytical instruments are described, in conjunction with typical procedures and growth substrates. Equipment and procedures for ex-situ analytical measurements are also described; and include imaging, polarity, composition, and optical measurements. Last, fabrication and characterization of single nanowire devices is described. Chapter 4 presents results on controlling the nanowire structural characteristics, as pertaining to synthesis of discrete nanowires suitable for re-integration onto non-native substrates. Methodologies developed for characterizing key MBE growth parameters are discussed, including group III and active nitrogen fluxes, by use of in-situ and ex-situ process diagnostic tools. A specialized scanning probe technique for measuring crystallographic polarity was developed, validated, and applied to AlN layers used for nucleating GaN nanowires. By use of these newly developed techniques, the underlying growth mechanics were investigated and optimized. Experimental results obtained during the development of axial p-n junction nanowire LEDs are presented in Chapter 5. Specific attention is dedicated to p-doping, and its effects on nanowire conductivity and contact resistance, by use of experimental data and contact resistance modeling. Axial p-n junction nanowires with an electron blocking layer are shown to produce diode-like electrical characteristics and band-edge electroluminescence; while nanowires with p- n junctions only, are shown to exhibit diode-like electrical characteristics and below-detectionlimit electroluminescence. These observations are explained as electron overflow in the p-

21 4 region; and are modeled as a forward biased n + -p junction emitting electrons into a reversebiased Schottky barrier p-contact. Chapter 5 also presents the final phase of this research, an application-level device consisting of optically coupled nanowires with LED and photoconductive functionality. Axial p- n junction nanowires incorporating an electron blocking layer and lightly doped n-region, are experimentally demonstrated capable of electroluminescence or photoconductive response depending on the contact metallization layout. Optical coupling is demonstrated in a twonanowire device, illustrating proof of concept that nanowires may provide optical interconnects that can be integrated with non-photonic materials. Chapter 6 presents the overall conclusions of this work, in addition to recommendations for further study. 1.3 Primary Accomplishments This research resulted in several conference presentations and first-author publications which are outlined below. Two of the publications are currently under review. A more general assessment of the technical contributions resulting from this work is also included. Publications: Effect of AlN Buffer Layer Properties on the Morphology and Polarity of GaN Nanowires by Molecular Beam Epitaxy, Journal of Applied Physics 110, 2011 (selected to appear in the Virtual Journal of Nanoscience and Technology volume 24, issue 12). Towards Discrete Axial p-n Junction Nanowire Light-Emitting-Diodes Grown by Plasma-Assisted Molecular Beam Epitaxy, accepted for publication in the Journal of Electronic Materials, On-Chip Optical Interconnects Made With Gallium Nitride Nanowires, submitted for publication, 2012.

22 Conference Presentations: Towards Discrete Axial p-n Junction Nanowire Light-Emitting-Diodes Grown by Plasma-Assisted Molecular Beam Epitaxy, oral presentation at the 54 th Electronics Materials Conference, Penn State University Effect of AlN Buffer Layer and Nitrogen Plasma Conditions on the Growth of GaN Nanowires by Plasma Assisted Molecular Beam Epitaxy, poster presentation at the Materials Research Society Spring Meeting, San Francisco Decoupling of Growth Temperature and Nanowire Density for Magnesium Doped GaN Nanowires Grown by Plasma Assisted MBE, poster presentation at the 16 th International Conference on Molecular Beam Epitaxy, Berlin Evidence of Magnesium Enhanced Desorption in Gallium Nitride Nanowires, oral presentation at the 51 st Electronics Materials Conference, Penn State University Technical Contributions Demonstration of axial p-n junction nanowires with useful characteristics: o Band-edge electroluminescence in discrete nanowires o Multi-functional nanowires with source and detector capability o Optical coupling between source and detector nanowires Established methods and validation in determining crystallographic polarity in AlN layers, including polarity mapping by use of piezoresponse force microscopy Provided insight on role of AlN layers in nanowire nucleation Developed methods for quantifying group III, nitrogen, and doping fluxes using in-situ diagnostic tools, including composition of active nitrogen

23 2. Background and Literature Review Gallium Nitride Gallium Nitride (GaN) constitutes the core material in the III-nitride family and has become a material of significant technological importance, largely due to breakthroughs in p- doping and improvements in crystalline quality in the 1990s. As shown in Fig. 2.1, GaN forms a continuous alloy system with Al and In, enabling tunable and direct bandgaps that span from deep ultra-violet (UV), through visible, and into the infra-red (IR) regions of the spectrum [2]. High quality GaN has good thermal stability and excellent mechanical properties, making it suitable for a diverse array of applications. InGaN based devices have been intensively developed over the past several decades, primarily for visible-wavelength LEDs and blue laser diodes [3]. GaN photodetectors have found application as solar-blind detectors, which are sensitive in the UV spectral range, but at wavelengths shorter than background solar radiation. GaN most commonly takes on the wurtzite crystal structure, which is composed of two interpenetrating hcp lattices displaced along the c-axis [4], as shown in Fig The bonding in GaN is partially ionic with Ga cations tetrahedrally coordinated to nitrogen anions. Crystallographic planes parallel to the (0001) basal plane of the unit cell are composed entirely of either cations or anions, and are correspondingly called polar planes or c-planes. The (1-100) sides of the unit cell are termed m-planes and have a non-polar character as the surface normal is perpendicular to the [0001] polar axis. It should be noted that the top and bottom polar faces of a GaN crystal are not equivalent, due to the lack of inversion symmetry in the wurtzite structure. By convention, the crystal is Ga-polar if the [0001] direction points in the same direction as a vector from the Ga atom to N atom along the c-axis, and N-polarity corresponds to the antiparallel orientation. Similarly, the [0001] direction points towards the Ga-face of the

24 Figure 2.1. Bandgap energy and lattice constant for III-nitride semiconductors [5]. The lattice constants of typical substrate materials are indicated at the bottom of the plot. The hexagonal lattice constant for Si (111) is 3.84Å (not shown). 7

25 Figure 2.2. Crystallographic properties of III-nitride materials. Diagrams illustrate (a) wurtzite unit cell, (b) important crystal planes [6], (c) crystallographic polarity [2], and (d) polarization conventions and induced surface charges. 8

26 9 crystal and the [000-1] direction points toward the N-face. The physical and chemical properties of the crystal differ depending on the polarity. The Ga-N bond lengths are not identical for a unit tetrahedral, with the bond parallel to the c-axis being slightly longer [7]. This elongation displaces the center of charge in the unit cell and gives rise to a dipole moment along the c-axis. This built-in dipole results in a spontaneous polarization, which points in the [000-1] direction for Ga-polarity, as shown in Fig Also present is a strain-induced piezoelectric polarization along the c-axis that results from lattice mismatch in heterostructures. The sum of the spontaneous and piezoelectric polarizations leads to macroscopic sheet charge densities at interfaces and surfaces, and enables an avenue of band engineering that is not available in non-polar semiconductors. This effect has been exploited for formation of 2D electron gas (2DEG) channel regions in high electron mobility transistors (HEMTs). It also produces a parasitic effect in LED devices, the quantum confined Stark effect, which spatially separates electron and hole carrier populations in thicker quantum wells and reduces radiative recombination rates. Intentional doping in GaN is typically accomplished by introduction of substitutional impurities on the Ga site. Silicon provides effective n-type doping with a shallow donor ionization energy of mev [8]. The activation energy for silicon dopants is less than the thermal energy at room temperature, resulting in a high ionization fraction and free-electrondensity similar to the doping density. Defects and impurities in GaN often act as n-type dopants, with unintentionally doped material typically producing n-type behavior. Nitrogen vacancies and/or oxygen impurities incorporated during growth can result in a large n-type background concentration.

27 P-doping in GaN is obtained using magnesium impurities, and presented one of the 10 primary hurdles in the realization of GaN based electronic devices [3]. Mg dopants exhibit deep acceptor levels, with ionization energies of approximately mev [5, 9]. Consequently, the fraction of ionized acceptors is roughly 1% of the concentration of Mg impurities at room temperature, leading to low free-hole concentrations in p-gan material. Increasing the Mg concentration leads to a self-compensation effect (see Fig. 2.3), that results from nitrogen vacancies incurred at high impurity concentrations [9]. As with unintentional dopants, these defects contribute to the background n-type carrier concentration, and eventually outstrip any gains associated with increased Mg concentration. There are some reports that degenerate Mg doping can create an impurity band with reduced activation energy [10, 11], as inferred by measured ionization fractions of ~10 % at low cm -3 Mg concentrations. Last, Mg acceptors can become bound to atomic hydrogen during growth, an effect that deactivates the dopant and decreases the free-hole concentration. Deactivated dopants can be reactivated using a high temperature anneal to drive out the hydrogen. GaN epilayers are most commonly grown on sapphire, silicon carbide, and silicon substrates, which incur large mismatches in lattice constant and coefficient of thermal expansion. The resulting high dislocation densities introduce mid-gap states and defect conduction paths, and can decrease the carrier mobility. These defects decrease the internal quantum efficiency of LED devices through non-radiative recombination processes; however, GaN based optoelectronic devices work surprisingly well despite these extended defects. GaN epilayers and nanostructures can be grown by a variety of techniques including Molecular Beam Epitaxy (MBE), Metal-Organic Vapor Phase Epitaxy/ Metal-Organic Chemical Vapor Deposition (MOVPE/MOCVD), and Hydride Vapor Phase Epitaxy (HVPE). MBE

28 Figure 2.3. Free hole and magnesium concentrations in Mg:GaN films [9]. Plot illustrates low acceptor ionization efficiency, Mg self-compensation effects, and background donor concentration effects. 11

29 12 growth will be described in detail in the following section. For MOCVD processes, the group III elements are transported to the growth surface in metallorganic form and nitrogen is supplied as ammonia. The precursors react at the growth surface, yielding GaN and evolved gaseous byproducts. MOCVD is performed at high pressures relative to MBE, which serves to stabilize the GaN against decomposition that would otherwise be observed for similar MBE growth temperatures. HVPE is a furnace tube based growth process, capable of very high GaN growth rates of several hundred nm per minute. Group III chlorides are produced by flowing HCl vapor over molten source metals, and then transported to the substrate in conjunction with group VI hydrides (ammonia). GaN epilayers and nanostructures can be grown by each of these techniques; however, the scope of this work will be limited to GaN growth by Plasma-Assisted MBE. 2.2 Plasma-Assisted Molecular Beam Epitaxy MBE is a ultra-high-vacuum (UHV) based growth technique, in which tightly controlled fluxes of evaporated group III elements and chemically activated nitrogen species impinge a heated growth substrate. Adsorbed species have sufficient thermal energy to diffuse to available lattice sites or desorb from the surface, resulting in epitaxial and stoichiometric crystal growth. The growth rates are consequently slow, and require low base pressures to prevent contamination from background fluxes. Ga does not react with ground state diatomic nitrogen; hence, plasma sources are used to generate excited molecular and atomic states of nitrogen, referred to collectively as active nitrogen. The pressure during growth is typically on the order of 10-5 Torr, and is composed primarily of non-reacting inert nitrogen. Alternatively, ammonia can be used as a nitrogen source for MBE GaN growth.

30 2.2.1 Planar Film Growth by PAMBE 13 The primary MBE growth parameters for planar GaN films are the V/III flux ratio and the substrate temperature [12]. The effect of these parameters on film morphology is illustrated by the growth diagram shown in Fig. 2.4 [13]. For N-rich (N-stable) conditions the growth rate is limited by the Ga flux, with excess active nitrogen desorbing from the surface. The growth surface under N-rich conditions is typically rough and exhibits columnar structure. For low substrate temperatures and metal-rich (Ga-stable) conditions, a smooth and homogenous GaN film is formed with excess Ga accumulating on the surface as droplets. The growth rate is limited by the active nitrogen flux. At higher growth temperatures and Ga-rich conditions, the excess Ga is desorbed from the surface yielding smooth surfaces with no droplets. This is the intermediate Ga-rich regime, and represents the typical growth conditions used for planar film growth. The smooth surfaces obtained under Ga-rich conditions are due to the surface diffusivity of Ga during growth. Under intermediate Ga-rich conditions, a thin liquid Ga bilayer forms and increases the Ga diffusivity [14], resulting in a self-surfactant effect responsible for the improved surface quality. In contrast, the diffusion of nitrogen has been calculated to be negligible under normal growth conditions [15]. Impurities like hydrogen (from ammonia) and In can also increase the surface diffusivity of Ga. The GaN growth rate drops substantially at higher growth temperatures ( C), and can even decompose pre-existing film material. This high temperature growth limitation has been attributed to several factors including the thermodynamic stability of GaN [16], excessive Ga desorption [17], and decomposition by higher reactivity active nitrogen species [18, 19]. GaN is not predicted to be thermodynamically stable under typical MBE growth conditions and forms as a metastable phase, in contrast to GaAs MBE growth [20]. The low pressure utilized

31 Figure 2.4. Growth diagram for planar GaN films using PAMBE [13]. TEM cross-sectional images and AFM surface topography scans are shown for (a) N-rich, (b) intermediate Ga-rich, and (c) Ga-rich growth conditions. 14

32 15 for MBE growth can favor decomposition of GaN to gaseous products at high temperatures [17]. In contrast, MOCVD and HVPE growth is stabilized against decomposition by higher growth pressures. The GaN decomposition rate for PAMBE growth can be suppressed by increasing active nitrogen flux, which serves to drive the reaction equilibrium away from decomposition [21]. The susceptibility of the GaN surface to decomposition has also been observed to depend on the composition of active nitrogen and on the crystallographic polarity, with N-polar faces typically decomposing under atomic N fluxes [19, 22]. As described above, the chemical composition of the nitrogen flux influences Ga surface diffusion, decomposition reactions, and ultimately GaN growth quality. Thus, the active nitrogen composition and the plasma sources used to generate it merit some discussion. Excited states of nitrogen are generated through gas-phase collisions with accelerated nitrogen ions in the plasma; producing excited neutral molecules (N * 2 ), ionized molecules (N + 2 ), atomic nitrogen (N), and ionized atomic nitrogen (N + ) [23]. The ionization fraction and relative concentrations of these species depends on the particular source configuration [22], and operational parameters such as nitrogen flow rate and radio-frequency (rf) power [24]. In many cases these excited states are short lived, and relax to metastable or ground states during transport to the growth substrate. At low growth temperatures (600 C), where Ga desorption and GaN decomposition can be neglected, the nitrogen incorporation rate in Ga-polar films was proportional to the flux of atomic nitrogen [25] over a wide range of nitrogen flow and rf power conditions. From this observation, it was concluded that atomic nitrogen is the dominant nitrogen species for GaN growth. However, as growth temperatures approach 750 C, atomic nitrogen is observed to increase the decomposition rate. Decomposition is minor in comparison when using primarily metastable molecular nitrogen [18, 26].

33 Doping in PAMBE grown GaN is accomplished by sublimation of solid Si and Mg 16 sources. The vapor pressures of Si and Mg vary substantially from the group III elements (see Fig. 2.5), resulting in large differences in sticking coefficients. Silicon has a very low vapor pressure, which results in minimal re-evaporation and negligible surface diffusion. In contrast, Mg has an extremely high vapor pressure, and requires considerable fluxes to offset reevaporation losses at high growth temperatures. The Mg incorporation efficiency varies depending on the polarity of the growth surface, with N-polarity exhibiting lower overall incorporation. Moreover, high Mg fluxes and surface concentrations can cause a Ga-polar surface to invert to N-polarity, creating defective regions called polarity inversion domains. One advantage for Mg:GaN grown by PAMBE is the absence of hydrogen during crystal growth and the attendant dopant deactivation phenomena GaN Nanowire Growth by PAMBE GaN nanowires can be grown by PAMBE using catalyst-based and catalyst-free techniques. This work will primarily focus on spontaneously grown nanowires using catalystfree methods, which frequently produce faceted single crystals with very low unintentional impurity levels. The nanowires grow with the c-axis parallel to the growth surface normal, and have {1-100} m-plane sidewalls, as shown in Fig The facetted character of the nanowires implies that there is a thermodynamic driving force that promotes incorporation on the c-plane tip, and inhibits growth on the non-polar m-plane sidewalls [27]. The growth rate anisotropy is due to facet surface energies and surface adatom kinetics [28], which are ultimately manifested as differing sticking coefficients between the nanowire tip and sidewall surfaces [27]. The nanowire structure allows for in-plane relaxation of strain, and encourages threading dislocations at the root to be excluded to the surface [29]. As a result, single nanowires are typically

34 Figure 2.5. Vapor pressures of elements and dopants in PAMBE growth of III-nitrides. Vapor pressure and melting point data adopted from reference [30]. 17

35 Figure 2.6. FESEM images of GaN nanowires grown by PAMBE [27]. Images show nanowires with tapered (a,b) and uniform (c,d) diameters. 18

36 observed to be free of strain and extended defects, as observed by TEM images and 19 photoluminescence spectra [31]. The most commonly used substrates have been silicon (111) and sapphire, which present a hexagonal close packed surface arrangement similar to the basal plane of the wurtzite unit cell. Nanowires grown on Si (111) are epitaxially oriented according to the relationship (0001) GaN {111} Si, <10-10> GaN <11-2> Si. Thin AlN buffer layers have been used to prevent nitridation of the silicon surface at the start of growth, and to prevent interdiffusion reactions between Ga and Si [32-34]. The AlN buffer layers also present some reduction in mismatch strain through a 5:4 lattice coincidence with silicon [35]. Alternatively, intentional nitridation of the silicon surface has been utilized by others to reduce interdiffusion and mismatch strain. For nitrided interfaces, the nanowire growth is still c-axis, but exhibits some misalignment in the outof-plane orientation and has random in-plane orientation [36]. GaN nanowire growth has been widely acknowledged to occur spontaneously under N-rich conditions, and at growth temperatures close to the high temperature limit. A nanowire growth diagram analogous to the planar film growth diagram has been reported [37], as shown in Fig. 2.7, and illustrates several of the key growth issues. Nanowire growth occurs over a relatively narrow temperature window between planar film growth conditions and no-growth conditions. In many reports, a defective planar film layer grows concurrently with the nanowires, and is referred to as a matrix or compact layer. A substantial fraction of the incident Ga flux is lost to re-evaporation or decomposition at nanowire growth temperatures, and results in an actual V/III ratio that is higher than the nominal flux V/III ratio. Thus, nanowires can be grown under nominally Ga rich conditions, as indicated on the growth diagram, even though the local conditions are likely to be N-rich. Evaporative Ga loss at the growth surface also results in a

37 Figure 2.7. Growth diagram for GaN nanowires using PAMBE [37]. Flux and temperature conditions under which GaN nanowires spontaneously grow are identified as NC, while planar film growth is identified as compact layer. The stoichiometric flux condition is identified by (Ga) = (N). 20

38 21 film/nanowire transition temperature that increases with Ga flux, and results in the collapse of the nanowire process window at high Ga fluxes. One of the challenges in catalyst-free PAMBE nanowire growth stems from control of this narrow process window. However, conditions for optimal nucleation of nanowires can be different from conditions for optimal late-stage growth propagation, allowing some relaxation of process constraints for multi-step growth schedules. Hence, it is illustrative to consider both nucleation and propagation growth stages separately in the following discussion. A wide variety of nanowire structures have been reported in the literature, spanning isolated nanowires with ideal hexagonal cross sections, to highly dense and coalesced nanowires with irregular cross sections. Based on this diversity in morphology, it is likely that nanowire nucleation can occur through a variety of channels, depending on specific conditions of the growth process, reactor geometry, and substrate surface condition. As a result, many contradictory reports exist on process-related nucleation phenomena, and optimization of the nucleation process conditions is usually experimentally determined for a given reactor. Nonetheless, it is illustrative to review several previous studies on nanowire nucleation, and its effect on nanowire density and diameter. The primary stage of the nucleation process involves adsorption, diffusion, and agglomeration of Ga to create discrete GaN islands [34, 38, 39]. The density and size of the islands increases with growth time, accumulated Ga, and Ga surface diffusion length. The GaN islands are initially strained spherical caps that are coherent with the underlying substrate [40]. Eventually the stored strain energy in the spherical cap exceeds the energy required to form a misfit dislocation at the interface; and plastic shape transitions occur from spherical caps, to pyramidal structures, and finally giving way to nanowire nuclei. The nanowire nuclei have been

39 22 shown to be fully relaxed [41]. The lattice mismatch at growth conditions determines the critical radius at which shape transformation occurs, thereby determining the typical nanowire diameter. In some reports, a fraction of the islands or nanowire nuclei coalesce to form the surrounding matrix layer, which decreases the overall nanowire density. After nuclei formation, the nanowires increase in length and propagate the initial crosssectional geometry along the nanowire axis. Individual nanowires can be grown arbitrarily long, provided they do not coalesce. Ga flux impinges the tip and sides of the nanowire, with impinging tip fluxes incorporating directly, and sidewall fluxes diffusing to the tip for incorporation or desorbing from the sidewall. In some reports the nanowire growth rate exhibited an inverse relationship to nanowire radius, resulting from the ratio of tip incorporation area to sidewall collection area [42]. However, this 1/r dependence with the axial growth rate is not universally observed, and is possibly related to shadowing effects, short Ga diffusion lengths, Ga evaporative or decomposition losses, and Ga- or N-rich growth conditions. While catalyst-based techniques lie outside of the scope of this study, it is worth mentioning a few key points. Much of the pioneering work done on nanowire growth utilized the catalyst-based Vapor-Liquid-Solid (VLS) technique, originally published in 1964 [43]. In the VLS technique a metal collector, which is typically an Au or Ni droplet, is deposited on the growth surface, providing a reservoir of reactants at the nanowire tip. Growth conditions are selected so that super-saturation conditions exist in the collector. Nanowire growth proceeds at the nanowire/droplet interface, with the interfacial contact area defining the nanowire diameter, and the droplet persisting at the tip of the nanowire. In some cases, the collector droplet is consumed by the nanowire, creating conically tapered structures. In practice, this process allows straightforward control of nanowire diameter and position. However, VLS grown GaN

40 23 nanowires were shown to be of lower crystallographic quality than those grown by catalyst-free techniques, with TEM and PL observations of many basal plane stacking faults [44]. There is also concern that the metal catalyst is a background source of impurities. Self-catalysis by Ga droplets was once believed to be the sole explanation for catalyst-free nanowire growth; however, this theory has lost favor in recent years [27, 39]. 2.3 GaN Nanowire Devices The following section provides background discussion on the electrical characteristics of GaN nanowires, including a general discussion of surface depletion effects. A literature review of GaN nanowire LEDs is presented, limited in scope to discrete or single nanowire architectures. Last, photoconductivity in GaN nanowires is discussed Surface depletion effects The semiconductor surface is an abrupt discontinuity to the crystal structure, and is widely recognized to influence electrical properties in its vicinity. These effects are particularly important for GaN nanowires, where the surface-to-volume ratio is large, and only a small fraction of the total volume may exhibit bulk properties (or not at all in the case of thin or lightly doped nanowires) [45]. GaN nanowire surfaces are depleted of free carriers under equilibrium conditions, due to surface charges associated with trapped carriers and surface adsorbates [46]. Experimentally, the surface band bending has been observed to depend on the conductivity type in planar GaN films[47], with both n-type and p-type material exhibiting depleted surfaces. The nanowire conductivity depends in part on the diameter of the neutral core [6], which can be estimated from the Poisson equation in cylindrical coordinates with no axial or azimuthal dependence as

41 2 1 d 2 r r dr, 24 (2.1) where is the potential, is the volume charge density, and is the permittivity. For p-type conductivity the charge density can be approximated using the depletion approximation as shown in Fig. 2.8: N a q, (2.2) where N a is the ionized acceptor density and q is the electron charge. The total surface charge (Q s ) is related to the total depletion charge (Q depl ) by the charge neutrality condition Q s +Q depl =0. It is convenient to introduce a reduced surface charge parameter that normalizes the total surface charge by the total depletion charge total surface charge total depletion charge 2N RN s a, (2.3) where R is the nanowire radius and N s is the areal surface charge density. The bands are flat for a reduced surface charge parameter value of 0 (N s =0), and the nanowire is fully depleted for a value of 1(N s =R N a /2). Using boundary conditions that both the electric potential and the field are zero at the edge of the depletion region [48], the potential can be expressed as a function of radius qn r r 1 (2.4) 2 2 (1 ) 2, 2 a 2 ( r) R (1 ) ln 2 r 0 R 1 R

42 Figure 2.8. Schematic diagram of cylindrical surface depletion model for p-type nanowire. Diagram illustrates coordinate system, radial charge density, and calculated band diagram. 25

43 which shows the general form of the surface band bending potential. Note the sign of the 26 potential for n-type material is opposite from the above result. The radius of the conductive neutral core (r 0 ) can be determined from charge neutrality conditions to be r0 R 1. (2.5) In some references the surface depletion has been quantified using the surface band bending potential ( s ), which is defined as the potential difference a majority carrier experiences when being moved from the conductive core to the surface and has the form qn R r R s a 0 2 r0 ln (2.6) r0. It should be noted that this parameter has a unique relation to the surface charge and doping concentration only in the case that the nanowire is not fully depleted. The boundary condition of zero potential at the depletion layer edge is not observed when > 1. The nanowire resistance can now be expressed as R ohm 1 L L qn A h 2 A R (1 ). (2.7) From this result it can be established that the nanowire conductivity depends not only on the doping concentration and carrier mobility, but also the surface charge as it affects the extent of surface depletion. In practice, it is difficult to decouple these parameters using electrical measurements alone, hence analytical measurements are used to independently determine doping concentration, carrier mobility, surface potential [49]. It should be noted that the model

44 described above best approximates nanowires that are partially depleted ( < 1). For full 27 depletion conditions ( 1), it has been shown that free carrier contributions to the volume charge density should be considered, and that the overall conductivity is better approximated by the radially integrated free charge density [49]. Nonetheless, the approach described above illustrates the basic principles of surface depletion, within the constraints of partially depleted nanowires. From a practical viewpoint, GaN nanowire depletion effects can result in unwanted parasitic resistances in LED devices, particularly for thin or lightly doped nanowires. This necessitates the ability to synthesize larger diameter nanowires and achieve higher doping concentrations, which are central issues for growth experiments in this study. However, nanowire surface depletion also enables sensing capability, in which the nanowire conductivity can be used to monitor changes in surface charge. Here nanowire surface depletion provides functionality, and will be discussed in relation to photoconductivity in section GaN Nanowire LEDs Substantial research efforts have been underway for the past several years to develop GaN nanowires for LED applications, motivated largely by applications in solid-state lighting, where the waveguiding properties of nanowires are exploited to enhance light extraction efficiencies. Hence, most studies have focused on nanowire-array-based device architectures [50-66], that utilize an ensemble of nanowires in their as-grown format, as shown in Fig The array architecture is based on an axial p-n junction structure in which the p-region is grown last, and often under low temperature growth conditions, to encourage lateral growth and Mg incorporation [55, 56]. This allows a transparent planar top metallization to contact the p-type nanowire tips, with n-contacts being made through the growth substrate. The expanding

45 Figure 2.9. Axial heterostructure junction nanowire LED array grown by PAMBE [55]. Figure shows (a) SEM image of device morphology, (b) schematic diagram of device structure including axial p-n junction, InGaN multiple quantum wells, and global top p-contact, (c) L-I curve, and (d) EL emission spectra. 28

46 diameter of the p-region likely helps to mitigate surface depletion effects. A variety of axial 29 nanowire structures have been fabricated including p-n junctions [50, 54], InGaN multiple quantum wells [50, 55-58, 61, 62], InGaN/GaN double heterostructures [51, 52, 59, 60], AlGaN/GaN double heterostructures [63, 64], and AlGaN electron blocking layers [50]. Nanowire LED arrays have also been synthesized from planar films, that have been patterned using lithography and etch-back processes to reveal nanowire-like structures [67-71]. N-type nanowire arrays grown on p-gan substrates, where the growth interface serves as the junction, have also been investigated [53]. Less common are reports of single nanowire LEDs [72-76], removed from the growth substrate, and metallized with contact pads on a non-native substrate. Single p-n junction nanowires grown by HVPE [72] were shown to exhibit electroluminescence near the junction, with a peak emission near 390 nm, as shown in Fig The sub-bandgap emission was attributed to donor-acceptor-pair (DAP) recombination. The authors also reported ohmic contacts, and confirmed p- and n-type doping of separate nanowire samples by use of backgating experiments [77]. Single nanowire LEDs with a core-shell structure have been grown using MOCVD [74, 75]. N-type nanowires were overcoated with an InGaN quantum well, AlGaN electron blocking layer, and p-gan shell, as shown in Fig The AlGaN electron blocking layer provides confinement for injected electrons, and serves as a lower index of refraction material for promoting internal waveguiding of electroluminescence. Emission was observed to emanate from the nanowire tip, with peak emission wavelengths ranging from nm, depending on InGaN quantum well thickness. Single axial p-n junction nanowires using InGaN/GaN double heterostructure grown by PAMBE were also reported [73]. These nanowires exhibited peak intensities between nm, depending on the In concentration. Single GaN

47 Figure Discrete axial p-n junction nanowire LED grown by HVPE [72, 77]. Figure shows (a) SEM image of LED nanowire device, (b) I-V characteristics of p-n junction, p-type, and n-type nanowires, (c) optical image of EL, and (d) EL emission spectrum. 30

48 Figure Discrete core-sleeve heterostructure nanowire LED grown by MOCVD [74]. Figure shows (a) schematic diagram of device structure, (b) schematic band diagram, (c) EL emission spectra, and (d) optical image of EL. 31

49 nanowire LEDs with non-epitaxial junctions have also been demonstrated, using n-type 32 nanowires dispersed onto p-gan substrates [78-81], which circumvents difficulties in producing p-type nanowires GaN Nanowire Photodetectors GaN nanowires exhibit large increases in electrical conductivity under optical excitation [45, 46, 82], due to modulation the surface depletion layer. The surface depletion field causes photogenerated minority carriers to drift to the nanowire surface, where they partially screen the surface charge, resulting in increased diameter of the conductive core region. This contrasts with conventional photodetectors, where photocurrent results directly from field separation of optically generated carriers by a junction. Restated slightly differently, optically generated carriers drift orthogonal to the photocurrent in nanowire photodetectors, and parallel to the photocurrent in conventional photodetectors. This type of photosensitivity occurs under lowinjection conditions; at high-injection conditions the increased majority carrier concentrations also affect the overall conductivity. The increase in nanowire current with respect to optical excitation intensity is termed the photocurrent gain, and depends on the nanowire diameter and the extent to which it is depleted [82]. As shown in Fig. 2.12, nanowires close to full depletion can exhibit several decades of photocurrent gain, due to the highly resistive dark state. The photoconductive gain is smaller for larger diameter nanowires, which are only partially depleted, and support significantly higher dark current levels. Hence, nanowires near full depletion typically provide greater sensitivity as photodetectors.

50 Figure Photoconductivity in GaN nanowires [82]. Figure shows (a) light and dark I-V curves for various nanowire diameters, (b) persistent photoconductivity after a UV pulse, (c) schematic band diagrams of surface depletion for various nanowire diameters, and (d) photocurrent versus nanowire diameter. 33

51 34 After illumination, a significant period of time may be required for the nanowire current to return to the original dark level, and this phenomenon is generally referred to as persistent photoconductivity. Relaxation to equilibrium conditions occurs by recombination of photogenerated minority carriers at the surface with majority carriers from the nanowire core. The kinetics are limited by the activation energy associated with majority carriers surmounting the surface band bending potential [46]. The surface band bending potential ( s ) is smaller for fully depleted nanowires, as shown in Fig. 2.12, which results in a decreased time constant for relaxation to equilibrium. For photoconductive nanowire devices, the persistent photoconductivity serves to reduce the photoconductive gain and/or the device speed. Nanowires close to full depletion conditions provide the best combination of gain, reset speed, and low off-state current. On a final note, it is apparent that above-bandgap illumination will induce the largest photocurrent gain, however sub-gap wavelengths can also induce photoconductivity by coupling with acceptor or defect states in the material [46].

52 3. Experimental Details MBE Growth Systems Several MBE growth systems were utilized over the course of this study, as indicated in Table 3.1. The systems are numerically indexed in the order that they were commissioned in the lab. Much of the GaN nanowire research preceding this study was conducted in MBE1, which was configured for both nitride and arsenide materials. Use of MBE1 in this particular study was limited to the growth of AlN buffer layers, which were occasionally used in MBE2. MBE1 was not configured with an Mg source for p-doping in GaN materials, due to cross-contamination concerns with arsenide materials, and will not be discussed in detail here. The primary growth system used in this study, MBE2, was built specifically for GaN nanowire growth as described in detail below. Towards the end of this study a third MBE system was installed (MBE3), and was utilized for similar purposes as MBE2. Sample numbers for each growth were designated by an alphabetic prefix corresponding to the growth system used (see Table 3.1), followed by the run number MBE2 MBE2 is a home-built system based on a spherical UHV chamber, as shown in Fig The growth chamber is pumped with a turbomolecular pump, backed by a mechanical dry pump, and routinely reached base pressures in the low Torr range after baking. A loadlock chamber allowed introduction of substrates into the growth chamber without breaking vacuum. The substrate manipulator allows heating of growth substrates to a maximum temperature of approximately 950 C (for silicon) and provided rotation during growth. The growth chamber, manipulator, and sources were all water cooled to reduce excessive heating to the system; however no cryo-panels were utilized to suppress residual background fluxes. System

53 36 System ID Description Sample ID Purpose in This Study prefix MBE1 EPI/ Veeco 930 * B,C AlN buffer layers MBE2 Home-built MBE NGS N AlN polarity, NW growth studies, NW LED devices, Optical interconnect devices MBE3 Veeco GEN10 * D AlN polarity, NW growth studies, NW LED devices Table 3.1. MBE growth systems used for experiments. * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

54 Figure 3.1. Schematic diagram and photo of MBE2 growth system [6]. 37

55 38 components were controlled via machine control software written in LabVIEW *. EPIC * Recipe control software (commercially available, also based in LabVIEW) was able to provide additional scheduled control of substrate and cell temperatures, shutter states, and growth durations during runs. Conventional Knudsen evaporation cells were used to supply group III (Ga, Al, In) and dopant fluxes (Si, Mg). The sources are mounted such that the source-substrate distance is approximately 25 cm and make an angle of approximately 50 with the substrate normal. The Ga cell uses a graphite crucible and dual zone hot lip design to prevent condensation on the lip of the crucible. The crucible lip was typically held at a temperature of 65 C hotter than the primary temperature. The Al crucible was made from pyrolytic boron nitride (PBN) crucible and incorporated a cold lip design to prevent creep of molten Al out of the crucible. Exposure to active nitrogen during MBE nitride growth is known to cause Al creep, which was observed to cause uncontrolled variation in flux and in some cases catastrophic failure of the cell. Both Si and Mg cells used PBN crucibles. Group III solid source materials had a minimum of 7N purity for Ga and In and 6N purity for Al. The cells were powered by DC power supplies and controlled using Eurotherm 2408 controllers. Cell temperatures were routinely stable within 1 C of setpoint and fluxes could be modulated on or off using an integral rotary shutter. Approximately half way through the course of this research a dual-zone valved source was installed for Mg (between N287 and N288). This source used a Ta crucible which provided better thermal coupling to the Mg source material. The heated valve provides better shutoff control for medium vapor pressure materials like Mg. * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

56 39 An Oxford HD25 * commercial rf plasma source supplied active nitrogen for GaN growth. This source utilizes an rf coil to inductively generate a nitrogen plasma from which excited molecular and atomic nitrogen species are extracted. The plasma is confined within a PBN tube and extraction takes place through a PBN aperture plate. Several aperture plates with different hole patterns and conductances were investigated during this work. Ion deflection plates positioned near the exit apertures were biased at 500V to remove ionic nitrogen species from the active nitrogen flux. A load-matching network maximizes transmission of rf power to the plasma. This matching network was upgraded from manual tuning to automatic tuning towards the latter part of the study (between N343 and N344). Nitrogen gas with 5N purity was further purified with Nanochem filters and delivered to the plasma source via a mass flow controller with typical flow rates of sccm. The nitrogen plasma source generated a variety of excited atomic and molecular nitrogen species, which were optically characterized by use of an Ocean Optics USB4000 * spectrometer with a wavelength range of nm. Optical emission from the plasma was collected and transmitted to the spectrometer via an optical port and fiber coupling. The substrate temperature was measured by a pyrometer with optical access to the backside of the growth substrate. This backside pyrometer was periodically calibrated against a separate pyrometer, which measured the front surface temperature of a bare silicon wafer. The uncertainty of this approach was previously estimated to be 8 C [27]; however, the emissivity of the nanowire growth surface may introduce additional uncertainties in the actual surface temperature. The thermal uniformity was measured using an IR camera, and was found to have a * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

57 10-15 C range over a 2 cm radius. The temperature outside of 2 cm radius was found to be 40 approximately C colder than the wafer center. Results presented in this thesis represent areas from the central 2 cm radius, unless indicated otherwise. An ORS EpiEYE * reflectance system was used for in-situ film thickness measurements. This instrument generates an incident laser beam and measures the intensity of the reflected beam at normal incidence to the growth substrate. Modulation of the reflected intensity results from interferometric changes at the growth surface, with the period between maxima and minima yielding measurement of the film thickness. This instrument uses dual wavelength lasers (635 nm and 950 nm) to provide sensitivity over a greater thickness range. A tilted optical access window is used to eliminate artifacts resulting from interference effects in the window. This system was used for endpoint detection for AlN buffer layer growth and for measuring GaN growth rates as a function of process conditions. The growth chamber was equipped with a Hiden V 201/3F * triple-filter quadrupole mass spectrometer (QMS) for leak checking and measurement of residual gases. Gases impinging the QMS are ionized by electron impact, mass filtered by the triple filter quadrupole, and counted at the detector. This instrument is equipped with a faraday cup detector and an electron multiplier detector, the latter of which has high sensitivity for low pressure measurements. This detector was calibrated against the faraday cup detector using a small flow rate of nitrogen from the plasma source. The mass spectrometer is mounted on a movable bellows assembly so that it can be lowered into the beam path for measurement of source fluxes. Specifically, the beam path of the spectrometer was aligned with the beam path of the Mg source so that higher resolution flux * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

58 measurements could be obtained than with ionization gauge measurements alone. It was also 41 possible to measure flux characteristics of the nitrogen plasma source using Appearance Potential Mass Spectroscopy (APMS), which will be described in the results section. Reflection High Energy Electron Diffraction (RHEED) is a common analytical technique for in-situ characterization of MBE epilayer growth, in which a glancing angle electron beam is diffracted at the growth surface and imaged by a phosphor screen. The resulting diffraction patterns can yield information on crystallographic structure, lattice parameter, surface reconstructions, surface flatness, and growth rate. A Staib * RHEED electron source was used to generate the electron beam and could provide beam rocking for adjusting the angle of incidence. The e-beam was typically operated at 10kV acceleration potential and RHEED patterns were collected using a phosphor screen MBE3 MBE3 is a commercially available growth system with two reactors, separately configured for III-nitride or III-arsenide materials. The growth reactors, prep module, and loadlock module are linked by a central UHV cluster tool (see Fig. 3.2), capable of transferring substrates between modules without breaking vacuum. The III-nitride reactor will be described here briefly, with primary attention to its differences from MBE2. MBE3 is equipped with cryo-panels, which enshroud the substrate manipulator and sources, to minimize background or stray fluxes that may otherwise be incorporated into the nanowires during growth. The source configurations are similar to MBE2, with dual-zone effusion cells fitted with PBN crucibles for group III elements * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

59 Figure 3.2. Photo of MBE3 growth system. 42

60 and a Unibulb rf plasma source for active nitrogen. A three-zone valved cracker supplies Mg 43 doping fluxes and a conventional single-zone effusion cell supplies Si doping fluxes. MBE3 and MBE 2 have similar in-situ diagnostic capabilities, with a few notable exceptions. A beam flux monitor, which is essentially an ionization gauge that is inserted into the beam path, was used to measure source fluxes. Because this instrument has no mass filter to subtract partial pressures from background gasses, the lower detection limit is determined by the base pressure of the system. Consequently, measurements of the Mg doping fluxes were close to the noise floor for the beam flux monitor. A fixed position mass spectrometer is available for this system; however, it is located far from the beam path and is primarily intended for leak checking and measurement of background fluxes. The growth surface temperature was measured using several techniques, including front-side and back-side IR pyrometry and bandedge thermometry. The transition between silicon (1x1) and (7x7) reconstructions was observed at the expected temperature of 830 C [83], when measured by the frontside pyrometer. In this thesis, growth temperatures will be reported as the backside pyrometer temperature, as calibrated against the frontside pyrometer measurement for a bare silicon wafer, similar to the procedure used for MBE2. The band-edge thermometer measures the absorption spectrum of the growth surface, from which the temperature can be inferred using the temperature dependence of the absorption edge. As this instrument measures the spectrum of reflected illumination, film thickness and real-time growth rates can also be determined from the measured interference fringes. These measurements were primarily used when growing Mg:GaN films on transparent sapphire substrates, which complicates interpretation of IR pyrometry measurements.

61 3.2 Substrates and Preparation 44 Three inch silicon (111) substrates were used for most growth runs in this study. The wafers were doped n-type with antimony and had a resistivity of Ωcm. Silicon substrates were cleaned with a 2 minute immersion in dilute HF solution (1:10 HF:H 2 O), DI water rinsed for 2 minutes, and then blown dry with nitrogen. Silicon substrates with AlN buffer layers grown in MBE1 were cleaned using the same procedure, prior to loading into MBE2. After loading into the MBE system, the substrates were outgassed in successive stages to remove adsorbates and residual oxide from the growth surface. For MBE 2, the substrates were outgassed in the loadlock for 1 hour by use of a quartz lamp located at the wafer back surface. The wafer stage temperature during the loadlock outgas would reach approximately 390 C, although the wafer temperature will be lower. After transfer to the growth chamber, the wafer was outgassed for an additional 15 minutes at approximately 910 C. For MBE3, the substrates were outgassed in the loadlock for 1 hour at a setpoint of 250 C, after which a secondary outgas was performed in the prep module for 45 minutes at a setpoint of 250 C. A final outgas was applied in the growth chamber for 20 minutes at a temperature of ~940 C as measured by the front-side pyrometer. RHEED images of silicon wafers outgassed using the MBE3 procedure reproducibly yielded the 7x7 reconstruction pattern, which is indicative of an atomically clean silicon surface. Sapphire substrates with GaN films grown by HVPE were occasionally used for p-doping growth studies. These substrates were loaded into the growth systems with no additional wet cleaning procedures. Outgassing was carried out at lower temperatures, typically C higher than growth temperature, to prevent thermal decomposition of the GaN films. The substrates also received a 30-minute N 2 plasma clean prior to growth.

62 3.3 Ex-situ Analytical Characterization 45 A suite of post-growth analytical measurements were used to characterize as-grown nanowire samples, dispersed nanowire samples, AlN buffer layers, and Mg:GaN planar films. Many of these measurements were taken in-house, although some results were obtained from measurements performed by collaborators FESEM Imaging An FEI Sirion 4000 * SEM was used for imaging all samples, and could routinely provide resolution on order of a few nanometers. An in-lens detector was used to image secondary electrons, and imaging was typically performed at 5KV acceleration voltage to minimize charging effects. In some cases, the acceleration voltage was modified to highlight contrast between doped regions. For as-grown samples, high magnifications would frequently cause the nanowires to snap together Atomic Force Microscopy (AFM) and Piezoresponse Force Microscopy (PFM) An Agilent 5500 * Scanning Probe Microscope with a closed loop scanner was used for all AFM based measurements. The scanner was periodically calibrated in x, y, and z using a Nanosurf * calibration grid. Topography measurements were typically obtained in tapping mode, and used tips with a radius of ~10 nm and resonant frequencies between khz. Image analysis was performed using Gwyddion * software. PFM measurements were taken in contact mode using Veeco * conductive diamond coated tips. The 100-nm-thick doped diamond tips had a significantly larger tip radius (35-50 nm), which frequently resulted in tip imaging artifacts. Tips with 1-5 N/m and N/m force * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

63 46 constants were used, which produced similar results. A voltage sine wave with DC bias offset was applied to the tip using a Stanford Research Systems DS335 * function generator, and the substrate was held at ground potential through a backside contact. An external Stanford Research System SR830 * lock-in amplifier measured the amplitude and phase (or X and Y components) of the deflection signal, which was mapped back to the AFM system. External electronics were used to measure the deflection signal waveforms, as the signal-to-noise ratio for the built-in circuitry was too low for accurate measurements Polarity Sensitive Etching Polarity sensitive etching was performed using a short immersion in hot phosphoric acid. The immersion time for AlN films was 20 s, using a range of bath temperatures from C, and was followed by a 60 s DI water rinse. Etched samples were then imaged using AFM and/or FESEM to evaluate etch figures. N-polar AlN reveals pyramidal etch figures with a hexagonal basal plane. The orientation of the basal plane and sidewall angle measurements indicate that the slow etching surfaces are {10-11} planes, consistent with literature reports for N-polar material [84]. In-contrast, Al-polar material develops hexagonal pits and etches at an overall slower rate than N-polar AlN. After polarity sensitive etching, Al-polar inversion domains in N-polar AlN are often observed to remain as tall hexagonal inclusions. GaN samples required more aggressive etch conditions, typically 60s immersion at C bath temperatures Secondary Ion Mass Spectroscopy (SIMS) Mg concentrations in GaN nanowire and planar GaN film samples were obtained by use of SIMS measurements and were performed by QSpec Technology *. An O 2 + ion beam with 8keV * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

64 47 energy and na beam current was used to sputter profile the samples. The raster area for the beam was 125 m, and the collection area for secondary ions was 30 m. 71 Ga + and 24 Mg + secondary ion species were monitored during the scan. Measurement of a planar GaN film calibration standard was used to correlate raw counts to concentration, using an estimate of the total volume of sputtered material derived from stylus profilometry of the etch crater. The lower detection limit of Mg was estimated to be 1x10 17 cm -3 by this approach Transmission Electron Microscopy (TEM)/ Convergent Beam Electron Diffraction (CBED) CBED measurements are widely cited in determination of crystallographic polarity of IIInitride materials. The technique is similar to Selected Area Electron Diffraction (SAED) used in TEM analysis, except that the electron beam wave vectors are conically convergent instead of plane waves. This adds structural features to the diffraction patterns, which are related to the crystallographic polarity and measurement parameters, including convergence angle, accelerating voltage, and sample thickness. Fitting algorithms are then used to generate simulated patterns, which are compared against measured patterns for identification of polarity. TEM and CBED measurements for this study were performed by colleagues at the Material Metrology Laboratory at NIST Gaithersburg. A GaN nanowire cross-sectional TEM sample was prepared by mechanical polishing and ion milling. CBED measurements were obtained in a Phillips CM-30 TEM * operated at 100 kv and a sample temperature of 100 K. CBED simulations were performed using Bloch wave calculations implemented in the Javascript version of the EMS * code. * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

65 3.3.6 Photoluminescence (PL) Measurements 48 Room temperature PL measurements were performed on as-grown nanowire samples using a HeCd laser with emission at 325 nm for optical pumping. The measurement spot was focused to the minimum diameter, although no estimate was made of the excitation intensity. Luminescence was detected using a fiber coupled Ocean Optics UBS 2000 * detector, which provided a pixel wavelength resolution of 0.36 nm. The scan integration time was adjusted depending on the luminescence intensity for a given sample, thus normalized emission intensities are reported for these PL measurements. 3.4 Device Fabrication and Characterization Discrete NW devices were fabricated by releasing the NWs from the growth substrate via ultrasonic agitation in isopropanol, and dispersing the resulting suspension onto a separate oxidized silicon substrate. In most cases an oxidized ( nm thick) silicon substrate was utilized, although in some cases fused silica was used to facilitate optical measurements. Optical photolithography, e-beam evaporation, and lift-off processing were used to create an array of two-terminal or four-terminal electrodes over the randomly dispersed NWs, producing devices with electrodes variously registered along the NW length. Several metallization types were used, including Ni 50 nm/au 170 nm, Ti 20 nm/al 200 nm, and Pd 20 nm/pt 50 nm/au 120 nm. All p-n junction nanowire devices were metallized with nominally identical contacts on both sides of the junction. Several different post-contact anneals were utilized, designated as n-type (1 minute, 500 C, 5% H 2 /Ar), p-type (10 minute, 450 C, 5% N 2 /O 2 ), or none. In many cases the devices would be tested before and after an RTA-based contact anneal. It should be mentioned that various aspects of contact process were continually

66 optimized over the course of this research, including surface pre-contact surface treatment, 49 metallization type, metallization thickness, and post-contact anneal Electrical Characterization Electrical measurements were made using a Kiethley 4200-SCS * system and probe station equipped with tungsten probe tips. The source-measuring units (SMUs) were typically calibrated before each measurement session. Two-terminal I-V sweeps were typically performed from 0-25 V in both bias polarities. EL measurements of nanowire LEDs were often performed at constant current bias, which produced voltage biases up to 37 V for some devices. Four-terminal measurements for optical interconnect devices were performed using separate two-terminal test setups for LED and PC nanowires. The PC nanowires were biased at constant potential using a Kiethley 487pA * voltage source, and the current was monitored on the opposing device terminal using a Stanford Research Systems SR570 * low-noise preamplifier. The LED nanowires were subjected to constant current pulses generated by the Kiethley SCS *. An oscilloscope logged the timing of the pulses and the PC NW current measured at the SR570 * preamp Optical Characterization Electroluminescence measurements were typically made using the probe station microscope, which could be alternately fitted with a spectrometer or a CCD camera. EL spectra were collected over a wavelength range of nm, by use of an Ocean Optics * peltiercooled CCD/grating spectrometer using a 15 s collection period and 1 nm boxcar average. EL images were collected as video during measurements, which were then converted into a single * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

67 image average to increase resolution and clarity. In some cases, optical bandpass filters were 50 inserted between the microscope and camera to produce EL images at a specific wavelength range. In order to make quantitative EL measurements, the spectrometer was benchmarked against an independently characterized UV light source. A spectrometer calibration factor (C p ) was defined as the ratio of optical power at the spectrometer fiber to the peak count rate (S cr ), and was measured with a calibrated photodiode and commercial UV LED with peak emission at 362 nm. Optical transmittance between the microscope objective lens and the spectrometer (T mic ) was measured in a similar manner, and found to be 14.4 %. The total optical emission power of the NW LED (P NW ) was estimated by considering the EL as a point source originating from the NW center with uniform emission into all 4π steradians. The fraction of P NW collected by the microscope (F col ) was limited by the numerical aperture of the objective lens (NA=0.3). By approximating the EL as a plane wave passing through a surface parallel to the substrate (e.g. a GaN nanowire m-plane facet), F col was calculated as F col NA 2 n GaN 2. (3.1) A value of 2.5 was used for the GaN index of refraction (n GaN ). This calculation may underestimate F col by up to a factor of 5, as it neglects the contributions of EL refracted from additional NW facets and EL reflected from the substrate interfaces. The total optical emission from the active region of the NW was then calculated as P NW C F p col S T cr mic, (3.2)

68 with a lower detection limit of approximately 1 nw. For external quantum efficiency (EQE) 51 calculations it was assumed that 60 % of P NW is lost to the contacts due to internal waveguiding [85], with an additional 20 % being lost to the substrate. Thus the EQE was estimated as q p P EQE 0.2 h c I NW NW, (3.3) where p is the peak wavelength, I NW is the NW electrical current, q is the electron charge, h is Planck s constant, and c is the speed of light. Photoconductivity measurements were conducted at room temperature and ambient conditions by use of a xenon arc-lamp, a tunable monochrometer, and a probe station with environmental chamber. The intensity at 365 nm was approximately 1 mw/cm 2, and could be attenuated by neutral density filters inserted into the optical beam path. The applied bias potential and current measurements were both provided by a Kiethley 487pA *. * Reference to specific products or trade names does not constitute an endorsement by the National Institute of Standards and Technology. Other vendors may provide products of comparable or superior value.

69 4. MBE Growth of GaN Nanowires 52 GaN nanowires with various structures were produced throughout the course of this study, depending on the specific growth reactor and process conditions utilized. Initially, many of the nanowire growths were unsuitable for producing discrete nanowire devices, which prompted efforts to measure, understand, and optimize the key growth parameters. This chapter first provides an overview of a general growth process and typical nanowire structures; with subsequent sections discussing the development of process diagnostic capabilities, characterization of effects controlling nucleation behavior, and identification of strategies for optimizing the overall growth morphology. A schematic diagram illustrating an idealized axial p-n junction sample structure for spontaneously grown GaN nanowires is shown in Fig First, an AlN nucleation layer is grown, initiated by a few monolayers of Al to suppress nitridation of the silicon surface by the active nitrogen flux. The AlN layer is then grown to ~40 nm total thickness, and is epitaxial to the underlying silicon such that the c-axis is normal to the substrate surface. This orientation propagates to the GaN nanowires, which are bounded by hexagonal m-planes sidewalls and a c- plane tip surface. The nanowires increase in length with growth time, during which doping impurities are incorporated into the crystal. By modulating the doping flux from silicon to magnesium, GaN nanowires with axial p-n junctions can be synthesized. In practice, ideal nanowire structures are not always produced, as shown by the experimentally observed growth morphologies in Figure 4.2. Sample B895 exhibits near-ideal morphology, with coalescence-free nanowires that possess symmetrical hexagonal cross sections and uniform lengths. This growth type, routinely produced in MBE1 prior to this study, illustrates the target morphology to be obtained in MBE2 and MBE3. In contrast to the near-

70 53 Figure 4.1. Schematic diagram of idealized axial p-n junction nanowire growth. Figure 4.2 FESEM images of example nanowire morphologies. Images compare morphology for MBE2 samples [N211, N257, N274] and MBE1 samples [B985].

71 ideal structure of sample B895, sample N211 exhibits tilted nanowires and significant lateral 54 growth, resulting in a film-like surface and spindly eroded roots at the interface. Alternatively, some growths resulted in thick nanowires with highly irregular cross sections, as typified by sample N257. After extensive experimentation, isolated single nanowires (sample N274) were grown with similar characteristics to those in sample B895. The following sections describe the course of study by which isolated nanowires were grown in MBE2 and MBE3, as determined using in-situ growth process diagnostics and ex-situ materials characterization techniques. 4.1 Growth Process Diagnostics As described in the literature review, the characteristics of GaN films and nanowires grown by MBE are largely determined by the V/III flux ratio and the growth temperature. This section describes the procedures developed to quantify group III fluxes, active nitrogen fluxes, and doping fluxes for MBE2. An additional section is included that describes a more limited series of process-related measurements for MBE Group III Fluxes Procedures for measurement of Group III fluxes were developed over the course of this research, based on ionization gauge measurements, growth rates obtained from in-situ reflectance measurements, and ex-situ FESEM imaging. Beam Equivalent Pressure (BEP) measurement of source fluxes are conventionally obtained using an ionization gauge that is temporarily positioned in the beam path. The primary growth system used in this study, MBE2, has a fixed position ion gauge which measures the off-normal beam pressure. In either case, ionization gauge measurements are convenient and quantitative, but they must be calibrated against film growth rate measurements to obtain the actual flux. For GaN the growth rate could be determined in a single growth run using in-situ reflectance measurements to obtain a series of

72 growth rates for several Ga cell temperatures [86]. Growth rates are determined under 55 conditions in which all incident group III fluxes are incorporated, or unity sticking coefficient conditions. Several experiments were conducted to determine the sticking coefficient and surface film quality as a function of growth substrate temperature, as shown in Fig The reflectance shows multiple oscillations as film thickness increases through multiple constructive and destructive interference conditions. The growth rate was determined directly from the oscillation period, from which the sticking coefficient was calculated assuming unity sticking coefficient at the lowest temperature. Interestingly, the sticking coefficient was unity over the entire range investigated, and likely reflects the N-rich conditions the experiment was performed under. The total estimated film thickness by these reflectance measurements was within about 10% of the thickness measured by SEM. The amplitude of the oscillations decreases with increasing film thickness and subsequent surface roughness, particularly for the 635 nm wavelength. The reflectance was measured for a planar film grown with successively decreasing Ga cell temperatures as indicated in Fig The flat regions in the reflectance indicate periods where the shutter was closed and the Ga cell temperature was adjusted, providing convenient data markers. The measured growth rate is linear with respect to the ionization gauge measurements, indicating that the growth was N-rich over the entire experiment. Moreover, an Arrhenius plot based on the growth rate and Ga cell temperature yield an enthalpy of vaporization of 2.5 ev, which is close to the reported value of 2.6 ev for gallium evaporation. The Ga flux ( Ga ) is calculated as dgan Navogadro Ga t M M Ga N GaN, (4.1)

73 Figure 4.3. Reflectance measurements of Ga sticking coefficient in MBE2. Measurements were obtained for planar GaN films [N259, N261] grown at various substrate temperatures. Figures show (a) reflectance vs. growth time at 635 nm and 950 nm wavelengths, (b) calculated Ga sticking coefficients for two separate experiments, and (c) FESEM cross-section image of resulting films. Data collected 56

74 Figure 4.4. Reflectance measurements of Ga cell flux in MBE2. Measurements were obtained for planar GaN film [N245] grown at various Ga cell temperatures. Figures show (a) reflectance vs. growth time at 635 nm wavelength, (b) GaN growth rate vs Ga ionization gauge pressure, and (c) Arrhenius plot of growth rate and Ga cell temperature. 57

75 58 where d GaN is the density of GaN, N avogadro is Avogadro s number, t GaN is the film growth rate, and M Ga and M N are the molar masses of Ga and N, respectively. The ionization gauge measurements are correlated to absolute flux according to C P Ga Ga Ga, (4.2) where C Ga is the ionization gauge tooling factor, and is measured to be 4.8 x cm -2 s -1 Torr -1. Al flux measurements were attempted using the procedure outlined above, but were largely unsuccessful due to the difficulty in growing thick AlN films with low porosity. Instead, a series of thinner AlN films (~ 100 nm thick) were grown at varying Al cell temperatures and 630 C substrate temperature for direct measurement of thickness and growth rate. Fig. 4.5 shows the AlN growth rate as a function of Al flux, which exhibits the expected linear N-rich regime and the plateau-like Al-rich regime [33]. As before the Al flux could be calculated from the AlN growth rate in the N-rich region using equation 8 and appropriate parameters for AlN. The ionization gauge tooling factor for Al (C Al ) was calculated to be 4.1 x cm -2 s -1 Torr -1. An abrupt transition was observed between N-rich and Al-rich conditions, which likely reflects the relatively low substrate temperature and negligible re-evaporation of Al [87]. The transition between these regimes delineates the stoichiometric flux condition (V/III ratio = 1), and indicates an active nitrogen flux of approximately x10 14 cm -2 s -1. It is uncertain whether the active nitrogen flux measured for AlN growth is representative of the active nitrogen flux for GaN growth. In summary, group III fluxes were measured via in-situ reflectance measurements for GaN and ex-situ FESEM imaging for AlN with an accuracy of ± 15 %, depending on variation of

76 Figure 4.5. AlN growth rate curve for MBE2. Al flux is calculated from the linear N-rich section of the curve and the active nitrogen flux is determined using the stoichiometric flux condition ( Al = N ). Sample ID numbers are N271, N269, N270, N276, and N275 for increasing Al flux. 59

77 surface topography. AlN V/III flux ratios were estimated within ± 20% and were calculated 60 using an active nitrogen flux determined from the transition between N-rich and Al-rich conditions Active Nitrogen Composition and Flux Measurement of the active nitrogen flux presents a substantially larger challenge. Unlike group III fluxes which are elemental and controlled directly by cell temperature; the active nitrogen flux is made up of a variety of chemical species, which may or may not contribute to film growth, and are indirectly coupled with plasma process controls. Nonetheless, even relative measurements of plasma characteristics can yield insight into the growth process. In this study the plasma source was characterized under a variety of operating conditions and equipment configurations, by use of Optical Emission Spectroscopy (OES) and Appearance Potential Mass Spectroscopy (APMS). In OES, the plasma composition inside the plasma tube is evaluated by analyzing the wavelength and intensity of the spectral lines in the emission spectrum [24, 88]. As excited species in the plasma relax to lower energy levels, photons at specific wavelengths are spontaneously emitted, providing a means for identifying particular plasma species, as shown in Fig The concentration of a particular plasma constituent is proportional to its integrated line intensities. As seen in Fig. 4.6(a), a series of sharp atomic lines (747 nm, 822 nm, and 868 nm) are super-imposed against a number of band-like emissions related to excited molecular transitions. The bands at approximately nm correspond to the first positive system of N 2, and those from nm correspond to the second positive system of N 2 ; which represent that cascade sequence of excited molecular species to the long-lived A 3 + u metastable state, which is believed to be the primary molecular species for GaN growth. The first negative series

78 Figure 4.6. Identification of active nitrogen species observed in OES and QMS measurements. Plots illustrate (a) typical OES spectra with identification of spectral lines [88] and (b) potential energy diagram illustrating energy levels and transitions for excited nitrogen states [89]. 1 st positive (1 + ), 2 nd positive (2 + ), and 1 st negative (1 - ) series transitions are shown as hollow green arrows. Transitions associated with QMS measurements are indicated by red arrows for production of ionic nitrogen, and by the blue arrow for production of molecular nitrogen species. 61

79 62 of N 2 + produces several lines overlapping the second positive series for N 2, and corresponds to relaxation of ionic molecular nitrogen. In this study, the emission intensities were integrated for atomic lines, 1 st positive series molecular N 2 lines, and 2 nd positive series molecular N 2 lines separately [24]. The integrated intensities were then normalized and used to assess qualitative changes over a variety of process conditions including rf power and N 2 flow, as well as to monitor stability of the plasma source. APMS was used to characterize the active nitrogen flux at the growth substrate, in contrast to OES which primarily yields the ionization fraction in the plasma tube. APMS is a mass spectroscopy technique in which the electron energy of the ionizer is scanned, allowing discrimination between excited states of atomic and molecular nitrogen [90]. Ionized atomic nitrogen (N + ) is formed through electron impact with atomic nitrogen (N) at a threshold of 14.5 ev, while dissociative ionization of background diatomic N 2 occurs at higher electron energies (24.3 ev), as shown in Fig. 4.6(b). This presents a window in electron energy where the 14 amu signal can be attributed solely to atomic nitrogen. Fig. 4.7(a) shows APMS scans under vacuum conditions, 1 sccm N2 flow, and 1sccm N2 flow at various rf plasma powers. No 14 amu signal is observed at any electron energy for vacuum conditions. Upon introduction of 1 sccm of nitrogen (N 2 ), a signal is observed at electron energy > 27 ev, corresponding to dissociative ionization of the background nitrogen pressure. When the nitrogen plasma is ignited, a lower energy threshold is observed at ~17 ev, corresponding to direct ionization of atomic nitrogen (N) generated by the plasma source. The measured data shows a 3 ev offset from reported values for both direct and dissociative ionization, as summarized in Table 4.1, and is possibly due to a bias offset in the ionizer potentials. However, the energy difference between the direct and dissociative threshold ionization energies agree well with the expected value. Based on these

80 Figure 4.7. QMS measurements of atomic and molecular nitrogen species. Figures show (a) electron energy scans at 14 amu for various nitrogen plasma source conditions, (b) electron energy scans at 28 amu for various nitrogen plasma source conditions, and (c) detector signal with electron energy fixed at 25 ev illustrating linear dependence of atomic N flux on rf power. All measurements made with showerhead aperture. 63

81 64 Ionization Process Reported Threshold Measured Threshold Delta N + e - N + + 2e N 2 + e - N + + N + 2e N 2 + e - N e Table 4.1. Active nitrogen threshold ionization energies from QMS measurements. Measurements, reported values, and difference values (delta) are indicated for direct ionization of atomic nitrogen, dissociative ionization of molecular nitrogen, and ionization of molecular nitrogen.

82 results, the QMS signal at 25 ev electron energy and 14 amu was used as a metric for atomic 65 nitrogen flux at the substrate, and is observed to vary linearly with plasma rf power, as shown in Fig 4.7(b). As shown in Fig 4.7(c), molecular ionic nitrogen (N + 2 ) was observed at 28 amu with a threshold ionization energy of 18 ev, exhibiting a 2-3 ev offset from reported values, similar to the threshold offsets measured for atomic nitrogen. Some reports exist of lower threshold energies for excited molecular states of nitrogen, but these were not observed under any conditions in this study. The output characteristics of the plasma source were characterized using the setup shown in Fig These measurements were performed with the plasma source temporarily installed on the Mg source port, which has a direct line of sight to the QMS. The rf power and nitrogen flow rate were varied, during which OES and QMS measurements were collected to produce the data sets shown in Fig 4.9. For the purposes of this analysis, the integrated emission intensities from OES were normalized with respect to their emission series (i.e. atomic, 1 st positive series, 2 nd positive series). Two different exit apertures for the plasma tube were investigated, one with a 255 hole showerhead pattern and another with a single hole with approximately 40% higher conductance. The showerhead aperture presents a larger surface for reflection of plasma emission to the spectrometer; hence comparisons between aperture types using OES data should be avoided. As measured by QMS and shown in Fig. 4.9(a) and Fig. 4.9(c), the overall flux of atomic nitrogen to the substrate is significantly larger when using the single-hole aperture, and is observed to increase monotonically with rf power and nitrogen flow rate. In contrast, the showerhead aperture produces smaller fluxes of atomic nitrogen at the substrate, which saturate at higher rf power and N2 flow rate. Interestingly, the concentration of atomic nitrogen in the

83 Figure 4.8. OES and QMS instrumentation for characterizing active nitrogen flux in MBE2. 66

84 Figure 4.9. Comprehensive results from nitrogen plasma source measurements. Figures show (a,b) QMS data and (c,d) normalized OES data; while varying (a,c) rf power and (b,d) nitrogen flow rate. Measurements were made using both plasma source apertures, as indicated. 67

85 68 plasma tube does not saturate, but increases linearly with rf power, as shown by OES data in Fig. 4.9(c). This is possibly explained by recombination of atomic nitrogen at the small showerhead aperture holes, similar to observations reported by others for N + ions [22]. Atomic and molecular nitrogen concentrations in the plasma tube are observed to decrease with nitrogen flow rate for the showerhead aperture [Fig 4.9(d)]; presumably due to lower plasma ion energies resulting from increased gas-phase collisions and shorter mean free paths incurred at higher pressure. However, the decrease in dissociation fraction is offset by the increase in flow, producing a nearly constant atomic nitrogen flux at the substrate [Fig. 4.9(b)]. The N + and N + 2 ion fluxes from the plasma source could also be measured by turning off the QMS ionizer and the plasma source ion deflection plates. As shown in Fig. 4.10, the ion fluxes for the single-hole aperture increase with rf power and decrease with N 2 flow, as expected based on ion energy and mean-free path considerations. These ion fluxes were completely eliminated upon activation of the plasma source bias deflection plates. An unexpected observation from these measurements was that the measured atomic nitrogen flux was insensitive to direct line of sight between the source and QMS, as shown in Fig The flux was invariant to the N shutter state or the position of the QMS with respect to the beam path, indicating that atomic nitrogen is able to survive wall collisions without recombining to neutral molecular nitrogen. The recombination probability of atomic nitrogen is defined as the ratio of recombining N atoms to the number of wall collisions, and has been reported elsewhere to be 0.07 for stainless steel [91]. This suggests that atomic nitrogen may survive upwards of 10 collisions, depending on the chamber surface condition. Consequently, a background flux of atomic nitrogen is to be expected during growth, irrespective to the N 2 shutter state.

86 69 Figure QMS measurements of atomic and molecular nitrogen ions. Atomic (14 amu) and molecular (28 amu) ion fluxes generated by nitrogen plasma source (single-hole aperture) were measured while varying (a) rf power and (b) nitrogen flow rate. The QMS ionizer and plasma source ion deflection plates were disabled during these measurements. Upon activation of the ion deflection plate bias, the QMS signal is extinguished; and results in an ion current to the deflection plates, which is shown on the right hand axes of the plots above. Figure QMS measurements of background atomic nitrogen for MBE2. Measurement were obtained at 14 amu with open and closed shutters states, and with the QMS instrument in the beam path and retracted. The black curve is the nitrogen baseline with plasma source off.

87 70 Plasma startup transients were measured for both apertures, with the showerhead aperture exhibiting long and complex behavior, as shown in Fig An initial transient, characterized by an increasing atomic N concentration and plasma pressure, may be related to higher dissociation or gas heating in the plasma tube. In comparison to the showerhead aperture, the initial transient for the single-hole aperture is relatively minor and largely complete within ten minutes. A secondary transient is observed for the showerhead aperture after approximately 15 minutes; characterized by an abrupt decrease in pressure and molecular line intensities, and an increase in atomic line intensities. It is possible that the small aperture holes increase in diameter as the aperture plate heats up, thereby decreasing the plasma pressure. As discussed in section 4.2.2, these transients occur during AlN buffer layer growth and may introduce non-repeatability during the nucleation process. A cumulative and long-term evolution in plasma characteristics was also observed when using the showerhead aperture. As shown in Fig. 4.13(b), the plasma pressure at 1 sccm flow rate increased substantially over the course of the growth campaign, signifying decreasing aperture conductance. Upon removal, the aperture holes were found to be clogged, potentially from deterioration of the boron nitride or from migration of re-evaporated Ga from the substrate. In either case, the plasma output changed substantially during the growth campaign resulting in increasingly non-uniform growths. It should be noted that these observations were made over the duration of the study with frequent experimentation to determine best known conditions for the plasma source. As a result the growth studies described in subsequent sections were performed using a variety of operational conditions and apertures. Efforts were made, however, to produce sample sets grown close in time, and under nominally identical plasma conditions.

88 Figure Plasma startup transients for MBE2. Figures indicate (a) showerhead and (b) single-hole aperture plates; as measured by OES, QMS, and plasma pressure measurements. Plasma conditions were varied in (a), as indicated by red labels for rf power and blue labels for N2 flow. 71

89 Figure Long-term drift in plasma source output characteristics. Measurements were obtained under identical operating conditions, as measured by (a) OES and (b) plasma tube pressure. Plasma source conditions were 400 W rf power, 1.0 sccm nitrogen flow, and showerhead aperture plate. 72

90 4.1.3 Doping Fluxes 73 In most cases Mg fluxes were below detection limits for the ionization gauge and were measured using the QMS, which was in direct line of sight to the Mg cell. No attempt was made to correlate the relative fluxes measured by the QMS to absolute fluxes. Instead, the primary metric for doping is the Mg concentration and overall conductivity of the Mg:GaN nanowires, which will be discussed in section Nonetheless, the relative flux measured by the QMS was used to assess the stability and reproducibility of the Mg effusion cell. For the first half of this study, Mg was evaporated using a conventional effusion cell fitted with a PBN crucible. As shown in Fig. 4.14(a), the Mg flux required approximately 25 minutes to reach steady state. The flux was measured at various cell temperatures, yielding a straight line on an Arrhenius plot, as shown in Fig. 4.14(b,c). The enthalpy of vaporization was measured to be 135 kj/mole using the slope of the Arrhenius plot, which is close to the value determined from thermodynamic tables (144 kj/mole). This provides some level of confidence in the measurement approach. It was found that the Mg flux typically dropped over the course of several runs, with subsequent recalibrations indicating a change in flux output [Fig. 4.14(d)] that was consistent with an increasing temperature offset. Although the exact cause of the offset is not known, it seems likely that the coupling between the source material and the heaters changed with time. This source was replaced with the dual zone valved source to provide better shutoff control and a potentially more stable heating configuration. The silicon source was not in direct line of sight to the QMS and was not characterized with respect to flux output. Moreover, the atomic mass of silicon coincides with that of diatomic nitrogen, coupling the QMS signals for silicon and background nitrogen. The primary metric is

91 Figure QMS measurements of Mg flux for MBE2. Figures illustrate (a) initial flux transient, (b) typical cell temperature staircase ramp for flux calibration, (c) enthalpy of sublimation derived from Arrhenius plot, and (d) long-term drift in flux calibration. 74

92 free-electron concentration and nanowire conductivity, which were evaluated using electrical 75 measurements MBE3 Growth Process Diagnostics The MBE3 fluxes were characterized by measurements similar, but less detailed, than those described above. The absolute Al and nitrogen fluxes were estimated from AlN films grown at various Al cell temperatures, as shown in Fig and described in greater detail in section The conversion factor between the measured aluminum beam equivalent pressure (BEP) and absolute flux was 1.75 x cm -2 s -1 Torr -1. The active nitrogen flux was ~3.3 x cm -2 s -1, as determined from the Al flux at the stoichiometric condition. Interestingly, the active nitrogen flux is approximately double that of MBE2. Several other differences became apparent from OES measurements, as shown in Fig 4.16(a). Spectral lines pertaining to atomic and molecular 1 st positive series are observed similar to MBE2 optical emission; however, spectral lines from the 2 nd positive series are completely absent. The C 3 e molecular species from which 2 nd positive system transitions originate is quite high in potential energy, as shown in Fig. 4.6(b). In fact, this excited molecular species is even higher in energy than some atomic species, which have been reported to induce GaN decomposition at higher growth temperatures. Hence, it might be speculated that GaN nanowire growth in MBE3 would be less susceptible to decomposition effects than in MBE2, due to the absence of 2 nd positive series emission lines. The plasma source for MBE3 was also measured to be stable, as shown in Fig. 4.16(b), over the entire duration required for a typical 40 nm AlN buffer layer growth. 4.2 Nanowire Nucleation While it is generally recognized that GaN nanowires grow under N-rich flux conditions and high substrate temperatures relative to film growth, it is unlikely that nanowire nucleation

93 76 Figure AlN growth rate curve for MBE3. Al flux is calculated from the linear N-rich section of the curve and the active nitrogen flux is determined using the stoichiometric flux condition ( Al = N ). Figure OES Plasma source measurements for MBE3. Figures show (a) OES spectra and (b) stability of integrated atomic and molecular OES counts.

94 77 occurs by a single and universally observed mechanism. More likely, nucleation occurs through various mechanisms that may or may not be active, depending on the specific characteristics of the initial growth surface and process conditions. In this section, the thin AlN buffer layer is examined with respect to its effect on nanowire nucleation and morphology. Most of the results originating from MBE2 have been published previously [92], with excerpts from the published manuscript appearing below. Results from samples more recently grown in MBE3 are included, when appropriate. AlN buffer layers are commonly employed in growing high quality GaN films on nonnative substrates. While it is generally observed that a smooth AlN buffer layer grown by MBE near stoichiometric flux conditions is beneficial for GaN epilayers, it is less clear whether a smooth AlN buffer layer is preferred for GaN nanowire growth. Low-temperature AlN buffer layers grown on silicon frequently exhibit a rough columnar microstructure [93] with nanowirelike features that may act as nucleation sites for subsequent GaN nanowires. In fact, several groups have reported a strong dependence of GaN nanowire morphology on the characteristics of the AlN buffer layer [32, 39, 41, 94], which at least potentially enables some level of control over shape, diameter, and density of nanowires. The sensitivity of the nanowire morphology to the AlN growth surface underscores the role of the nanowire nucleation process, and implies that a seeded nucleation process may be active in addition to the spontaneous shape change/plastic deformation nucleation process described by others [40, 41]. AlN buffer layers are also utilized to impart a preferred crystallographic polarity onto subsequent GaN layers, most notably for GaN grown on sapphire [95]. For heteroepitaxial growth of III-nitrides on non-polar Si(111), however, there is no a priori expectation for a preferred AlN polarity. The AlN polarity instead has been reported to depend on growth

95 78 conditions including V/III ratio [96, 97], growth temperature and surface reconstructions [98], and initial dose of Al or N [93, 99]. Thus, it is important to experimentally establish the AlN polarity and reveal its effects on GaN nanowire morphology. Mixed-polarity GaN nanowires growing from a film-like N-polar matrix or compact GaN layer have been reported elsewhere and were speculated to result from the underlying AlN polarity [100]. A polaritydependent growth rate was also observed [101] and attributed to a high-temperature growth limitation for N-polar GaN [19], suggesting a potential mechanism for suppressing growth of N- polar material. This variation in growth kinetics could in theory be used to preferentially grow Ga-polar GaN nanowires, which are expected to exhibit a more favorable Mg incorporation efficiency similar to that of Ga-polar Mg:GaN epilayers [102]. The following sections present a description of techniques for measuring crystallographic polarity in III-nitrides and include cross-correlation measurements of AlN buffer layer polarity. Once these techniques have been validated, the AlN polarity is evaluated against several growth parameters including Al/N flux ratio and growth temperature using samples from MBE2 and MBE3. Last, the morphology and polarity of GaN nanowires grown on AlN buffer layers with various polarity configurations is presented AlN Polarity Measurements High defect densities associated with low-temperature AlN growth can complicate measurement of film polarity, particularly for diffraction-based techniques like Convergent Beam Electron Diffraction (CBED). Polarity-sensitive etching is also dependent on defect densities; hence Piezoresponse Force Microscopy (PFM) measurements were developed to provide supplemental information about AlN buffer layer polarity and surface topography. PFM allows for direct measurement of the actual growth surface, in contrast to polarity-sensitive

96 79 etching which may lift-off features of interest. As shown in Fig. 4.17(a), PFM is an AFM-based imaging technique in which the tip is electrically biased with an AC signal to induce piezoelectric deformation in the sample, which is then detected through the cantilever deflection via a lock-in amplifier. The phase of the deflection signal with respect to the bias signal determines the polarity in nitride materials [ ], with metal-polarity exhibiting in-phase behavior and N-polarity exhibiting out-of-phase behavior (see Fig 4.17 (b)). PFM measurements were carried out by use of a commercial AFM system and external lock-in amplifier coupled to the AFM deflection signal, as shown in Fig. 4.17(c). A conductive diamond AFM tip was used as the top electrode and the silicon wafer served as the bottom electrode for the PFM setup. The tips had a force constant of 5 N/m and were biased at 5 V rms and a frequency of 10 khz, far from the cantilever resonant frequency of 118 khz. Additional measurements were made with higher force constant tips (20-80 N/m) and found to yield similar results. Several measurement artifacts have been reported to introduce 180 phase offsets, complicating direct interpretation of phase data, including electrostatic interactions [106, 107], cantilever buckling effects [108], and laser position on the cantilever [109]. Whether these artifacts are operative or not depends largely on specific measurement conditions, including cantilever spring constant and tip-sample capacitance gradient, which in turn depends on film thickness and surface depletion layers. To eliminate these bias-dependent ambiguities, PFM phase measurements were benchmarked against polarity-sensitive etch tests by use of films with uniform Al-polarity or N- polarity (Fig. 4.18). Polarity sensitive etching for Al-polar material clearly reveals hexagonal etch pits. The corresponding PFM images shows uniform out-of-phase behavior, indicating a 180 phase offset from the theoretically expected result. Conversely, for N-polar material the

97 Figure Schematic diagram of PFM polarity measurement for III-nitrides. Figures show (a) theory of operation, (b) application to III-nitrides, and (c) measurement apparatus. 80

98 Figure PFM and polarity-sensitive etch measurements of unipolar AlN films. Measurements were obtained using AlN films that were predominantly (a) Al-polar [D015] or (b) N-polar [D011]. PFM phase measurements include topography, PFM amplitude, and PFM phase; with corresponding average roughness (R a ), modal value out-of-plane piezoelectric coefficient (d 33 ), and modal value PFM phase, as indicated. 81

99 polarity sensitive etch produces obvious pyramidal structures, with the corresponding PFM 82 images indicating uniform in-phase behavior. From these measurements, it was empirically determined that PFM yields the correct polarity, as validated by polarity sensitive etch tests, at V dc =0 V providing a 180 phase offset is applied. It was also found that the PFM phase could be modulated between 0 and 180 depending on the DC bias (V dc ) applied to the tip, indicating non-negligible electrostatic contribution to the piezoresponse signal. While the overall purpose of this study was to ascertain the polarity from PFM phase information, the d 33 piezoelectric coefficient can also be derived from the PFM magnitude [103]. Assuming that the electric field (E z ) and stress (S z ) are normal to the surface and uniform over the AlN layer thickness; d 33 is calculated as d S h h h z 33, (4.3) E V z AC VAC h where h is the film thickness, h is the magnitude of the tip displacement oscillation, and V AC is the magnitude of the applied sinusoidal potential. This calculation neglects the clamping effect of the surrounding film, a typical assumption identified in reported values as an effective d 33 coefficient. The measured PFM magnitude, which is more specifically the magnitude of the AFM photodetector bias, is converted to tip displacement magnitude by correlation with measured force-distance curves. By this approach, the d 33 coefficient for the AlN films shown in Fig 4.18 are calculated to be pm/v, which are within a factor of 2-3 lower than those reported elsewhere [103]. The relative correspondence between the coefficients measured in this study and those reported elsewhere provide an increased confidence level for these PFM measurements.

100 The PFM polarity measurement was also validated for a mixed polarity AlN layer. As 83 shown in Fig. 4.19, the PFM phase image exhibits out-of-phase domains embedded within an inphase background. According to the 180 phase offset convention, this sample is predominantly N-polar with isolated Al-polar domains. The PFM amplitude is observed to approach zero at the polarity inversion boundaries, due to net extension and contraction in adjacent regions [106]. An experiment correlating the as-grown, PFM, and post-etch images was conducted for the same location on this sample and is shown in Fig The tip radius was found to be significant in proportion to domain size, thus spatial resolution in topography and phase images were limited by the tip-imaging artifact. For domains larger than the tip radius, PFM yields the correct polarity by use of the V dc =0 V, 180 phase offset convention; however, erroneous PFM phase information was observed in regions where tip imaging dominates, such as pinholes or areas proximal to protruding columns. Tip shape analysis and certainty mapping algorithms were used in this cross-correlation experiment to improve data quality, by identification of regions subject to tip imaging and removal of associated data from the PFM phase image [Fig. 4.20(b)]. Interestingly, some Al-polar regions identified by PFM did not survive the polarity-sensitive etch, but left a visible crater behind indicating that domain may have been undercut and lifted off during etch. It was also observed that protruding columns exhibit both polarity types, which precludes the possibility that the phase contrast is solely due to a topographic artifact AlN Polarity for Various Growth Conditions A series of AlN samples were grown in MBE2 and MBE3 under various V/III flux ratios for characterization of morphology and polarity. The V/III flux ratio was modulated by varying the Al flux, while keeping the nitrogen plasma source conditions constant. The growth time for each sample was chosen so that the overall AlN film thickness was approximately 100 nm,

101 Figure PFM measurement of mixed polarity AlN film [N270]. Figure shows (a) topography, (b) PFM amplitude, and (c) PFM phase for the same sample location using a single scan. Also show is a (d) histogram of the PFM phase data. 84

102 Figure Correlation of PFM and polarity-sensitive etching for a mixed polarity AlN film [N270]. Images show (a) PFM measurement of as-grown surface, (b) FESEM image of as-grown surface, and (c) FESEM image of post-polarity-sensitive etch surface for same location on AlN buffer layer. PFM image shows topography (brown) with phase overlay (green = Al-polar, unmarked = N-polar). As grown FESEM image has PFM phase overlay with tip image artifacts removed (green = Al-polar, blue=npolar, unmarked = data removed due to tip imaging). 85

103 86 allowing greater microstructural development than for the 40 nm thick layers typically utilized for nanowire growths. Samples grown in MBE2 received a 13 s dose of Al, corresponding to approximately 1-3 monolayers, at a substrate temperature of 680 C prior to AlN film growth at 630 C. The plasma source was fitted with a showerhead aperture, and operated at 350W power and 1.0 sccm flow conditions. Samples grown in MBE3 were deposited with approximately 3 monolayers of Al at ~ 740 C for all samples. The nitrogen plasma source was then started and allowed to stabilize for approximately 5 minutes, in contrast to the MBE2 procedure where no stabilization period was used. AlN growth and ramp to the final growth temperature of ~720 C were then initiated simultaneously. RHEED images were collected throughout the process and during cooldown. The V/III ratio spanned from N-rich to Al-rich growth conditions for both MBE2 and MBE3 samples sets, as indicated by the growth rate curves and as-grown structure shown in Fig Under N-rich conditions, the growth rate is linear with Al flux, with films exhibiting a rough and columnar morphology. Under Al-rich conditions, the growth rate is plateau-like and Al droplets are observed on the film surface. Films grown near the stoichiometric flux condition are relatively smooth and planar, by comparison. RHEED images collected for MBE3 samples are similarly illustrative of this transition; exhibiting spotty 3D growth patterns for N-rich conditions, streaky 2D planar growth patterns for stoichiometric conditions, and low brightness patterns when attenuated by Al droplets under Al-rich conditions. The polarity of the MBE3 grown samples will be discussed first, as these samples are predominantly unipolar with clear correlation between PFM and polarity sensitive etch results. These samples are also smooth, with an average roughness of ~ 1 nm or less (excepting Al droplets), which eliminates the tip-imaging artifacts discussed above. As shown in Fig. 4.22, the

104 Figure AlN films grown under various V/III ratios for polarity measurements. Plan and cross-section FESEM images are shown for as-grown (a) MBE3 and (b) MBE2 samples. V/III ratios are referenced by N-rich, slightly N-rich, and Al-rich conditions, as indicated on (c) AlN growth curves. RHEED images obtained during AlN film growth are also shown for MBE3 samples. Sample ID numbers are D010, D011, D009 in (a) and N271, N269, N270, N276, and N275 in (b) for increasing Al flux. 87

105 Figure Effect of V/III ratio on MBE3 AlN polarity at constant N flux. PFM and polaritysensitive etch measurements obtained for AlN films grown under (a) N-rich [D010], (b) stoichiometric [D011], and (c) Al-rich [D009] V/III ratios. PFM phase measurements include topography, PFM amplitude, and PFM phase; with corresponding average roughness (R a ), modal value out-of-plane piezoelectric coefficient (d 33 ), and modal value PFM phase, as indicated. The PFM topography z- height scale is 15 nm for (a,b) and 35 nm for (c). FESEM image of etched Al-rich sample showing suspected (d) Al droplet is indicated by arrow. 88

106 89 film grown under N-rich conditions exhibits uniform Al-polarity in both PFM and polaritysensitive etch measurements. The films grown under stoichiometric and Al-rich conditions are predominantly N-polar, indicated by the hexagonal etch figures and in-phase PFM measurements. It is not currently understood why the polarity inverts in the proximity of the stoichiometric flux condition; however, similar tendencies was observed in separate MBE2 samples, which will be discussed later. For descriptive purposes, the flux at which the transition between Al- and N- polarity occurs is designated the polarity transition flux. A smaller fraction of Al-polar material is also observed in the stoichiometric and Al-rich samples, identified by the larger mesa-like etch figures or by the out-of-phase areas in the PFM phase images. These Alpolar domains appear to coincide with Al-droplets in the PFM phase images; however, the PFM amplitude for these regions is low and may incur noise in the phase measurement. Nonetheless, the observed mesa-like etch figures have a similar size and density to the Al droplets, indicating a potential connection between the droplets and the Al-polar domains. It has been reported elsewhere that Al-rich conditions have a tendency to produce Al-polarity films, speculated to result from Al-Al bonding and a subsequent inversion in the polarity [96]. These results are summarized in Table 4.2, from which the following primary conclusions can be drawn: the polarity changes from Al- to N-polar with increasing Al flux, occurring near the stoichiometric flux condition; and the Al-droplets accumulated on the surface induce Al-polarity domains against the otherwise N-polar film. The MBE2 samples present trends similar to those of MBE3 samples, with respect to the effect of V/III ratio on the dominant polarity type. However, mixed polarity and highly featured morphology are generally observed for these samples, as shown in Fig and summarized in Table 4.3. The polarity transition flux occurs under N-rich conditions in MBE2 samples, rather

107 90 V/III Ratio Al Cell Temperature (Tip/Base C) PFM Polarity Sensitive Etch N-rich 1025/1075 Al-polar Al-polar Stoichiometric 1035/1085 N-polar N-polar Al-rich 1040/1090 N-polar N-polar Table 4.2. Effect of V/III flux ratio on polarity of MBE3 grown AlN films. Layer AlN GaN matrix GaN nanowires Growth conditions V/III flux ratio Metallization Anneal CBED N-rich 1.7 Film: Al-polar Film: Al-polar - Columns: ~30% Al-polar Columns: ~20% Al-polar Slightly Film: N-polar Film: N-polar N-rich Columns: ~20% Al-polar Columns: ~20% Al-polar Al-rich Film: N-polar Film: N-polar - Columns: ~70% Al-polar Columns: ~50% Al-polar Slightly N-polar - N-rich Ga-polar Table 4.3. Effect of V/III flux ratio on polarity of MBE2 grown AlN films and GaN nanowires. Summarized polarity results obtained from PFM, polarity-sensitive etching, and CBED measurements.

108 Figure Effect of V/III ratio on MBE2 AlN polarity at constant N flux. Plan and crosssectional FESEM images of as-grown and polarity-sensitive etched AlN films, representative of (a) N-rich [N271], (b) slightly N-rich [N269], and (c) Al-rich [N275] V/III ratios. Also shown are PFM images consisting of topography (brown) and PFM phase overlay (green = Al-polar, unmarked = N-polar). The z-height scale is 150 nm for (a), 105 nm for (b), and 45 nm for (c). 91

109 92 than near the stoichiometric flux condition observed for MBE3 samples. This offset between the polarity transition flux and the stoichiometric flux is potentially explained by plasma transients, as discussed in section If the initial AlN growth stage occurs at lower nitrogen flux, the stoichiometric flux condition would be temporarily offset to lower Al fluxes. Thus, growth may proceed under Al-rich conditions for the initial growth stage, and evolve to N-rich conditions as the plasma source stabilizes during the late growth stage. In any case, V/III ratios between the polarity transition and the stoichiometric flux conditions are designated as slightly N-rich conditions for the description of MBE2 sample morphology and polarity in the following paragraphs. It should be mentioned that the V/III ratio was non-uniform for the samples described above; in some cases N-rich conditions were observed at the wafer center, while Alrich conditions, indicated by Al-droplets, were observed at the wafer periphery. To account for this non-uniformity, the description of AlN characteristics with respect to varying process conditions will be confined to a circular region in the wafer center with a radius of 1 cm. Mixed-polarity columns were observed to protrude from a finer grained AlN film layer, as shown in Fig The columns exhibit nanowire-like structure with approximately m-plane sidewalls and a hexagonal top face of varying irregularity. These nanowire-like structures may provide a template, from which GaN nanowires propagate during subsequent nanowire growth, and are described below to provide some sense of their structural characteristics. For N-rich conditions, the columns are tall with respect to the surrounding film and are tilted by up to 15 relative to the substrate normal. Others have also observed similar misorientation under N-rich conditions, manifested by high values of FWHM in XRD rocking-curve measurements [33, 110]. Approximately 30% of the columns are Al-polar with a mean diameter of 35 nm. With increasing Al flux into slightly N-rich conditions, the columns align with the substrate normal,

110 decrease in height relative to the film surface, and ultimately become smooth near the 93 stoichiometric flux condition. XRD measurements indicate these films possess the expected wurtzite structure and expected epitaxial relationship of AlN (0001) AlN {111} Si, <10-10> AlN <11-2> Si. Under Al-rich conditions, the overall grain size increases and larger platelet structures emerge. The density and size of the Al-polar columns (~80 nm diameter) increases significantly under Al-rich conditions, as summarized in Table 4.3, and is likely induced by Al droplets as discussed above. A fraction of the taller grains in the Al-rich region are observed to elongate along the <11-20> directions and have inclined facets on the top surface. The morphology of GaN nanowires grown on a companion set of AlN buffers, produced under similar growth conditions, will be presented in section One final set of experiments illustrating the effect of V/III ratio on AlN polarity is presented here, using MBE2 samples grown at 800 C under constant Al and varying N fluxes. These results were sought for practical reasons, as the high Al cell temperatures used to produce slightly N-rich and Al-rich conditions previously also produced significant Al creep from the crucible. This eventually resulted in catastrophic failure of the Al cell, incurring expensive repairs and extensive down time. The Al cell temperature was fixed at 1125 C, providing a constant and relatively low Al flux. To obtain V/III ratios near the stoichiometric flux condition, the N shutter was modulated using various duty cycles in 20 s intervals, producing the samples shown in Fig In this experiment the plasma source was fitted with the single-hole aperture, which exhibited less transient response, albeit with higher overall atomic N flux as discussed in section To further minimize the N flux, the plasma source was operated at 150 W, 1.5 sccm conditions. It was not possible to determine the stoichiometric flux condition based on growth rate curves or observation of Al droplets; however, the transition between

111 Figure Effect of V/III ratio on MBE2 AlN polarity at constant Al flux. FESEM and PFM images of as-grown MBE2 AlN films, grown at constant Al and varying nitrogen fluxes. The nitrogen shutter was modulated according to (a) fully open [N349], (b) 15 s On/5 s Off [N350], (c) 10 s On/ 10 s Off [N348] recipe cycles. PFM images indicate topography (brown) and PFM phase overlay (green = Al-polar, unmarked = N-polar). 94

112 95 columnar and smooth morphology is indicative of the stoichiometric flux condition, which takes place between 50-75% N for these samples. The stoichiometric flux condition appears to coincide with the polarity transition flux for these samples, as illustrated by the associated PFM phase images. It is possible that the slightly N-rich condition is not observed in this experiment, due to the stable N-flux provided by the single-hole aperture. As described above, N-rich growth conditions were observed in some cases to produce misoriented AlN columns, while Al-rich growth conditions produced granular surfaces with Al droplets. As will be discussed in the next section, these surfaces are poorly suited to be templates for subsequent GaN nanowire growth. Slightly N-rich conditions, which are presumed to be a brief excursion to Al-rich conditions at the beginning of the growth, provide a more favorable template with a surface that is reasonably smooth, droplet-free, and N-polar. However, the AlN films grown by MBE1were measured to be smooth, droplet-free, and Al-polar, as shown in Fig While it is not known what specific growth parameters produced this result, GaN nanowires grown on this surface often exhibited nearly ideal morphology. Hence, additional experiments were conducted to identify growth parameters, other than the V/III ratio, for producing higher quality Al-polar films. The approach used for growing Al-polar AlN layers in MBE3 largely follows the results presented in reference [98]. In that work, a two-stage growth temperature consisting of an initial nucleation temperature and subsequent primary growth temperature were utilized. The polarity was observed to depend on the initial nucleation temperature, with 700 C producing uniform N- polar films, and C producing uniform Al-polar films. The primary growth temperature was then increased to ~870 C for the remaining growth duration. The transition temperature between the Si (1x1) and (7x7) reconstructed surface was reported to take place at 830 C, which

113 Figure Polarity measurement of MBE1 AlN film [C255]. Figure shows (a) PFM images (b) as-grown FESEM image, and (c) polarity-sensitive etch FESEM images. PFM phase measurements include topography, PFM amplitude, and PFM phase; with corresponding average roughness (R a ), modal value out-of-plane piezoelectric coefficient (d 33 ), and modal value PFM phase, as indicated. 96

114 97 coincides with the transition temperature observed for MBE3. Hence, temperatures similar to those reported in reference [98] were used for this experiment. AlN films were grown using the stoichiometric V/III ratio, which was shown previously to produce uniform N-polar films when nucleated at 740 C. In contrast, the films in this experiment were nucleated at ~790 C and then ramped to the final growth temperature of 860 C; which, resulted in a uniform Al-polar film, as shown for sample D015 in Fig While this approach initially produced the desired uniform Al-polar result, repeated growths under nearly identical conditions produced an increasing fraction of isolated N-polar domains, as shown for samples D017 and D018 in Fig It should be mentioned that D017 and D018 were terminated with an extra ½ and 3/2 monolayers of Al, respectively, as part of a separate RHEED study; however, this additional surface coverage was not expected to affect the AlN polarity or measurements. Additional experiments probing the process window with respect to V/III ratio may provide some insight on this instability Effect of AlN Polarity on MBE2 Nanowire Morphology This section presents results on nanowire morphology observed in MBE2, as influenced by the structure and polarity of the underlying AlN buffer layer. A companion set of GaN nanowire samples grown on buffer layers similar to those shown in Fig. 4.27, provided correlation between the AlN characteristics and the GaN nanowire morphology. The AlN thickness was fixed at 40 nm by reducing the total growth time and V/III flux ratios nominally identical to those in Fig were used. Mg-doped GaN nanowires were then grown under nominally N-rich conditions with a Ga flux of x10 13 cm -2 s -1 and an Mg cell temperature of 370 C. The plasma source was fitted with the showerhead aperture and operated at 400 W rf power, 1.0 sccm flow rate. A

115 Figure Effect of high nucleation temperature on MBE3 AlN films nucleated at 790 C. PFM and polarity-sensitive etch measurements for several MBE3 AlN films [D015, D017, D018] nucleated ~50 C below the Si (1x1) to (7x7) transition temperature. PFM measurements include topography, PFM amplitude, and PFM phase; with corresponding average roughness (R a ) and modal value PFM phase, as indicated. As these films are predominantly Al-polar, a blue overlay is applied to topography and PFM amplitude images to indicate N-polarity. 98

116 Figure Effect of AlN layer on GaN nanowire structure for MBE2. Plan and tilt-view FESEM images showing morphology of GaN nanowires grown on AlN buffer layers with Al/N flux ratios corresponding to (a,d) N-rich conditions [N260], (b,e) slightly N-rich conditions [N267], and (c,f) near stoichiometric flux conditions [N257]. 99

117 two-stage temperature schedule was used for nanowire growth, comprised of 2 hours at 820 C followed by 12 hours at 800 C. 100 AlN layers that were grown under N-rich conditions yield thin nanowires, as shown in Fig. 4.27(a,d), with a diameter comparable to that of the Al-polar columns remaining after polaritysensitive etching (~30 nm). The nanowire density (1-2 m -2 ) is significantly less than the total AlN column density (50 m -2 ), indicating that nanowire growth does not occur solely due to continuation of growth at the AlN column tips. Also observable are a high density of short nanowire nuclei, usually located at the center of a hexagonal pit. These nuclei and the surrounding matrix layer do not grow thicker than a few tens of nanometers at typical nanowire growth temperatures, regardless of the growth duration. The density of the shorter nanowire nuclei is much higher (50 m -2 ) and is roughly equivalent to the AlN column density. These results suggest that the AlN columns act as nucleation sites for nanowire growth, as mentioned earlier, but will only grow to appreciable length on the small fraction of Al polar columns. Considering that nanowire growth occurs just 20 C under no growth conditions, it is possible that the high-temperature limitation for N-polar GaN [19] is suppressing growth on the N-polar fraction of AlN columns, similar to the dependence of nanowire growth rate on polarity reported elsewhere [101]. The high concentration of atomic or highly excited molecular nitrogen, discussed in section 4.1.2, may contribute to the discrepancy in growth rate on Al- and N-polar surfaces. Lastly, many nanowires can be observed to be tilted (up to 20 ) with respect to the substrate normal, similar to the misorientation observed for columns in AlN layers grown under similar conditions.

118 101 Increasing the Al flux to slightly N-rich conditions causes the nanowires to align with the substrate normal, as shown in Fig. 4.27(b,e) and similar to the AlN columns for similar conditions described above. TEM images corroborate this observation and further illustrate that there is no tapering of nanowire sidewalls [Fig. 4.28(a)]. Nanowires fabricated under these conditions were grown coalescence-free up to 9 m and were limited only by the growth time. In many cases, the cross-sectional shape of the nanowires deviates from ideal hexagonal structure, and is preserved along the full length of the nanowire. This result implies that the asymmetric cross-sectional shape of the nanowires is determined at the nucleation stage, likely from the irregular shape of AlN columns in the buffer layer. The length is observed to vary between nanowires and does not follow the inverse radius relationship observed by others [42], or the near homogenous lengths observed from the other growth system in our lab [27]. While shadowing effects could account for a random variation in growth length, it seems unlikely that this is the case for such low nanowire densities. It is not presently clear whether this dispersion in length is due to delayed nucleation or differential growth rates between nanowires. It is also unclear why many but not all nanowires have a facetted tip surface, as shown in Fig 4.28(a). Similar to nanowire growths on N-rich AlN buffers, there is a high density of short (less than 50 nm in length) nanowire nuclei growing in hexagonal pits, in addition to the low density longer nanowires. It should be noted that the associated buffer layer shows a similarly proportionate density of N-polar and Al-polar columns, supporting the hypothesis that the bimodal nanowire morphology results from the AlN buffer polarity. GaN nanowires on AlN buffers grown near the stoichiometric flux condition show a marked change in morphology, with a substantial increase in the overall nanowire size and surface fill factor, as shown in Fig. 4.27(c,f). The nanowire cross sections are very irregular yet

119 Figure TEM image and CBED polarity measurements for GaN nanowires grown in MBE2. Figures show (a) cross-section TEM image of GaN nanowires grown on slightly N-rich AlN buffer layer [N267], (b) experimental CBED diffraction pattern and (c) simulated CBED diffraction pattern showing Ga-polarity of long nanowire. The CBED pattern was calculated for a sample thickness of 38 nm. 102

120 still retain the sidewall faceting and axial homogeneity observed in N-rich AlN samples, 103 illustrating the templating effect of the underlying AlN buffer layer. Unlike N-rich AlN samples, the nanowire diameter exceeds the Al-polar column diameter, possibly resulting from excess Al accumulated on the surface under Al-rich conditions. This excess Al may convert to AlN during ramp to nanowire growth temperatures, creating a thin layer of AlN overgrowth. During this growth interruption the N shutter remained open, exposing the growth surface to an equivalent N flux approximately five times higher than the excess Al flux. It is uncertain whether this AlN overgrowth was resolved in the AlN polarity measurements described above. The abrupt transition between N-rich and Al-rich conditions at low AlN growth temperatures causes the morphology to be hypersensitive at the stoichiometric flux condition. This instability is further compounded by variations in active nitrogen flux resulting from plasma startup transients and long term clogging of plasma exit apertures. CBED analysis was carried out on the nanowire sample grown on slightly N-rich AlN (V/III flux ratio = 1.3), to experimentally determine the polarity of the long nanowires. The CBED pattern and simulation [Figs. 4.28(b,c)] for a long nanowire indicate Ga-polarity, supporting the hypothesis that the longer nanowires are Ga polar. The AlN layer was too defective to obtain a well-defined diffraction pattern, prohibiting direct correlation of AlN and GaN polarity for a given nanowire. Likewise, the quality of CBED patterns recorded from the short nanowire nuclei was insufficient for unambiguous determination of polarity. Polaritysensitive etching of GaN nanowires produced similarly inconclusive results, with tapering observed at both the root and tip of the nanowire. It is likely that the tip faceting in as-grown nanowires obscures reliable examination of etch effects, compounded by the limited resolution of FESEM imaging.

121 104 To resolve the matrix layer polarity, a low-temperature nanowire growth ( C) on slightly N-rich AlN was carried out to encourage thick matrix layer growth for polarity-sensitive etching. While these conditions were more typical for planar film growth, there was also a low density of nanowires observed (not shown). Polarity-sensitive etching reveals pyramidal structures on the top surface [Fig. 4.29(b)], indicating that the GaN matrix layer is in fact N- polar. We performed a similar low-temperature nanowire growth using a predominantly Alpolar AlN buffer film grown in the other growth system. More aggressive etch conditions (125 C, 60 s) were required for GaN and etching would preferentially attack the AlN buffer layer, allowing the etchant to access the interface side of the GaN film. Consequently, pyramidal etch figures emerge at the interface surface underside of the GaN film, indicating Ga-polarity [Fig. 4.29(d)]. These experiments illustrate that the GaN matrix layer propagates the polarity of the underlying AlN film layer. In conclusion, the morphology of GaN nanowires was found to depend on the V/III flux ratio of the underlying low-temperature AlN buffer layer. AlN buffers grown under slightly N- rich conditions provided a favorable growth surface for straight, low-density, and coalescencefree nanowires that could be grown arbitrarily long. Diverging from this condition led to tilted nanowire growth for N-rich AlN conditions or to large and highly irregular nanowires for Al-rich conditions, similar to the surface topography observed in associated AlN buffer layers. Al-polar columns are speculated to act as nanowire nucleation sites, based on the similar size, shape, density, and polarity of GaN nanowires. The morphology and polarity of the resulting GaN nanowires suggests that Ga-polar growth rates are much higher than N-polar growth rates under typical nanowire growth conditions, leading to sparse nanowires for AlN buffer layers that are mostly N-polar with isolated Al-polar columns.

122 Figure Polarity of GaN matrix layer in MBE2. FESEM images of low temperature GaN nanowire growth on (a,b) N-polar [N261] and (c,d) Al-polar [N147] AlN buffer layers, showing as-grown (a,c) and post polarity-sensitive etched morphology (b,d). The arrows indicate the N-face as determined from the emergence of pyramidal etch structures. 105

123 4.2.4 MBE3 Nanowire Nucleation Process 106 A final series of growth experiments was conducted to optimize nanowire morphology in MBE3. The detailed record of these experiments will be omitted here, and will instead focus on describing the process by which the nanowires shown in Fig were grown. An AlN buffer layer was grown under Al-rich conditions, using the high temperature nucleation process described above. Next a series GaN and AlN layers were growth at 30 s durations, separated by 15 s growth interruptions under nitrogen flux. An additional Ga-rich nucleation layer was grown, consisting of alternating Ga-only and GaN steps. These nucleation layers were grown at approximately 710 C, under a reduced plasma output 300W and 0.5 sccm, and at more than double the normal Ga flux. Silicon doping, which was shown elsewhere to act as a surfactant for Al during AlN growth [111], was also used to increase diffusivity and improve the crystalline quality in the nucleation layers. Last, a 1 hour temperature ramp to primary nanowire growth conditions was implemented. The RHEED signal was nearly extinguished during the nucleation layers, suggesting that Ga or Al droplets may play a role during the nucleation process. The results presented here were reproducible over several growth runs, using the complex sequence of nucleation layers discussed above. While further experiments deconstructing the effects of each step should be eventually be conducted, it will suffice for the remainder of this study to utilize this process for growth of device length nanowires, as discussed in the next chapter. 4.3 Post-Nucleation Nanowire Propagation In late-stage growth the nanowires elongate as GaN is incorporated at the c-plane tip, allowing synthesis of nanowires with virtually unlimited length and uniform diameter. However, under some growth conditions there is a finite lateral growth rate, which causes the nanowire diameter to increase slowly with growth duration, and ultimately results in nanowire coalescence

124 Figure GaN nanowire morphology for optimized growth process in MBE3. Nanowires were grown [D043] by use of Al-polar AlN and low-temperature GaN nucleation layers. 107

125 108 for long growths. This section provides a brief background on the nature of and conditions for lateral growth in GaN nanowires. Experimental data is then presented on identifying conditions in MBE2 that result in lateral growth, in addition to preliminary results on utilizing this phenomenon for controlling nanowire diameter. Lateral growth results in a tapered nanowire structure, characterized by consecutive groups of atomic steps on the nanowire m-planes [112]. Some investigations have linked this effect to local Ga-rich conditions at the nanowire tip [113]; or more specifically to a Ga droplet on the nanowire tip, with radius slightly larger than that of the nanowire [39]. Si and Mg doping fluxes were observed to increase the lateral growth rate [112], possibly due to a surfactant effect for enhanced diffusion of Ga on the nanowire sidewalls. Increased lateral growth is also observed at lower temperatures, where decreased re-evaporative Ga losses bias the growth into Ga-rich conditions. The incidence angle of Ga and N fluxes was also observed to affect the lateral growth rate [114]. These effects have been variously utilized to produce quasi-planar surfaces for nanowires devices with a global top contact [55], or for free-standing films with low-defect densities [100] [115]. For MBE2, the lateral growth rate is observed to depend on the Ga flux, as shown in Fig These samples were grown at a substrate temperature of 783 C using the showerhead plasma aperture. Note that the AlN buffers used for these experiments were grown under N-rich conditions, which results in the observed off-normal tilt. The axial growth is linear up to a Ga flux of approximately 5 x cm -2 s -1 indicating N-rich growth conditions. For N-rich conditions the lateral growth rate is negligible, resulting in nanowires with a tip to base diameter ratio of unity. The axial growth rate plateaus for higher Ga fluxes, indicating that growth is locally Ga-rich at the nanowire tip. It should be noted that local Ga-rich conditions can be

126 Figure Effect of Ga flux on axial and lateral nanowire growth rates in MBE2. Figures show (a) FESEM images of nanowire tapering and (b) overall growth rate curves. Nanowire tapering is quantified by the tip/base diameter ratio. Sample IDs are N244, N243, and N242 for increasing Ga flux. 109

127 110 observed, even under nominally N-rich flux conditions, due to Ga diffusing from the nanowire sidewalls to the tip [113]. A significant lateral growth rate is observed for Ga rich conditions, producing nanowires with an average tip to base ratio of 5. Brief excursions to low temperature and Ga-rich conditions during nanowire growth were explored as a potential route for increasing the nanowire diameter, and are hereafter referred to as expansion processes. In some instances, increased nanowire diameters are desired to minimize surface depletion effects, as discussed in section After an initial nucleation period under N-rich conditions, the growth temperature and overall V/III ratio were decreased, producing the samples shown in Fig Diameter expansion was only observed at the outer perimeter of the wafer, where a lower V/III ratio and substrate temperature was observed in previous studies. The plasma source was fitted with the single-hole aperture for these experiments, which likely provided higher nitrogen fluxes and introduced difficulty in obtaining local Ga-rich conditions at the wafer center. Nonetheless, the samples shown in Fig exhibit increased diameters that coincide with the region grown during the expansion step. Further experiments using more optimal plasma conditions or surfactants may ultimately provide greater control over nanowire diameter.

128 Figure Low-temperature Ga-rich nanowire diameter expansion process for MBE2 [N297, N327]. 111

129 5. Axial p-n Junction Nanowires for Nanoscale LEDs and Optical 112 Interconnects This chapter presents results obtained during the development axial p-n junction GaN nanowires, which ultimately resulted in discrete nanowires capable of producing band-edge electroluminescence. The characteristics and efficacy of the nanowire LEDs are discussed with respect to p-doping challenges, and were improved substantially by use of an AlGaN electron blocking layer (EBL). Finally, an application-level device for providing on-chip optical interconnects in non-photonic devices is demonstrated by use of axial p-n junction nanowires with LED and photoconductive capabilities. 5.1 GaN Nanowire LEDs In principle, the cornerstone of an efficient GaN NW LED is a p-n junction with reasonably high free-carrier densities and low-resistance contacts. However, p-type doping of NWs presents additional challenges beyond those encountered for planar GaN films, including incorporation of Mg dopants and control of NW morphology. GaN NWs are grown at substrate temperatures higher than those of planar films, which limits the incorporation of medium-vaporpressure elements such as Mg [116]. High Mg fluxes can also increase the lateral growth rate [112] and nucleation density[117, 118], leading to NW coalescence and the possibility of defectconduction pathways. Ni/Au contacts were observed to make poor contacts to discrete p-type NWs, with a significant void fraction aggregating at the metal/gan interface during contact anneals [119]. Lastly, the near-surface region in NW structures is often depleted of free carriers, which can significantly reduce conductivity in thin or lightly-doped NWs [46]. For example, a highly doped p-type NW (p = 3 x cm -3 ) with 170 nm diameter would be expected to be substantially depleted for a surface band bending of 1.1 ev [47]. These issues can introduce

130 parasitic effects that dramatically affect the overall performance of the device, and will be 113 discussed in the following sections Mg-Doped Nanowires The overall objective for this part of the study was to establish the effect of growth conditions on Mg incorporation efficiency and to measure the electrical characteristics of Mgdoped nanowires. Mg concentrations were quantified by SIMS for samples grown under a variety of conditions. Photoluminescence measurements and electrical characteristics are then presented for a subset of samples, selected based on suitable growth morphologies and Mg doping concentrations. SIMS is commonly used to measure Mg concentrations in GaN planar films, but its application to nanowire samples has not been well established. Mg concentrations in planar films are typically determined by calculating the ratio of Mg counts (Y Mg ) to Ga counts (Y Ga ), which are correlated to actual concentrations using an instrument specific correlation factor (RSF) derived from standard samples. Depth profiles are then calculated on a point-by-point basis with sputter etch depth as C Mg, p Y Mg RSF, (5.1) Y Ga where C Mg,p is the concentration of magnesium in GaN at a specific depth. In contrast to planar films, GaN nanowires are often melted during SIMS measurements, as shown in Fig The variations in Mg and Ga counts with sputter etch depth were likely influenced by elemental mixing and incongruent evaporation in the molten droplets. It is doubtful that the point-by-point Mg concentrations are representative of the actual compositional variation along the nanowires.

131 Figure 5.1. FESEM images of GaN nanowires (a) before and (b) after SIMS illustrating melting induced by measurement [N102]. 114

132 115 Instead, an average Mg concentration (C Mg ) was used as a metric for nanowires, and is calculated using the sum of the detector counts as C Mg Y Y Mg Ga RSF. (5.2) Several measurements are compared below to assess consistency in the SIMS data. First, measurements were made on two nominally identical Mg-doped nanowires growths, differing only by overall growth duration. As shown in Fig. 5.2, both measurements were carried out on as-grown samples until the Ga signal dropped off. The point-by-point Mg concentration exhibits a stretched v-shaped profile that was characteristic of most nanowire samples. The average Mg concentration calculated for both samples was 1.7±0.2 x cm -3. SIMS measurements were also compared for as-grown and dispersed nanowire samples prepared using the same growth run, as shown in Fig As before, both scans were carried out until the Ga counts dropped off. The point-by-point Mg concentration yielded a similar v-shaped profile, irrespective of the cross-section that the nanowires present to the ion beam, providing further confirmation that the shape of the depth profile is unrelated to the actual variation in Mg concentration. The average Mg concentration for the dispersed nanowire sample is larger than the as-grown sample by more than a factor of 3. Measurements across the full radius of the wafer were also observed to vary by up to an order of magnitude (not shown). The average Mg concentration was compared against the measured Mg flux for 11 samples grown in MBE2, as shown in Fig Measurements were made on various sample structures, including dispersed and as-grown nanowires. The nanowire growth temperature is indicated by

133 Figure 5.2. SIMS scans for N150 and N151 Mg:GaN nanowires. Figures show (a) SIMS scan data and FESEM images in (b) cross-section and (c) plan views for samples grown under nominally identical conditions, differing only in duration and overall nanowire length. The Mg concentration plotted in (a) is calculated from the point-by-point ratio of Mg to Ga counts. The average Mg concentration is indicated by the value in parenthesis in the figure annotation. 116

134 Figure 5.3. SIMS scans for as-grown and dispersed N176 Mg:GaN nanowires. Figures show (a) SIMS scan data and FESEM images for (b) as-grown and (c) dispersed nanowire samples. The images in (c) were obtained after SIMS measurement. The Mg concentration plotted in (a) is calculated by the point-by-point ratio of Mg to Ga counts. The average Mg concentration is indicated by the value in parenthesis in the figure annotation. 117

135 Figure 5.4. Average Mg concentration versus flux for Mg:GaN nanowires grown in MBE2. Average Mg concentration was calculated by use of SIMS scan data and Mg flux was quantified by use of QMS measurements. The growth temperature, as measured by the backside pyrometer, is indicated by the color of the markers. Data is shown for as-grown nanowire samples (solid circles), dispersed nanowire samples (open circles), as-grown samples measured at locations other than the wafer center (crossed circles), and planar film samples (squares). The gray bracket on the right side of the plot indicates Mg concentrations reported to yield free hole concentrations of 1-6x10 17 cm -3 in planar films [9], with an optimal value indicated by the gray dotted line. 118

136 119 the color of the data points. There is a large degree of scatter in the log-log data plot, which may partially reflect the variations in SIMS measurement described above and the Mg flux calibration drift discussed in section Nonetheless, there appears to be an overall trend showing increasing Mg concentration for high Mg fluxes. This trend is evident even though the growth temperature varies among samples, and includes two samples grown at low temperatures that produced low Mg concentrations where higher Mg incorporation would be expected. The optimal Mg concentration reported for planar films [9] is indicated by the horizontal dotted line on Fig. 5.4, and the range of Mg concentrations leading to free-hole concentrations greater than 1x10 17 cm -3 is indicated by the bracket on the right sidebar. For comparison, a 1.3 m thick Mg:GaN film grown in MBE2 on a GaN/sapphire template (N368) is also plotted on Fig This film was grown near nanowire growth temperatures, and had an Mg concentration ranging from x10 19 cm -3. Hall measurements on this film yielded a free-hole concentration of 7.6x10 16 cm -3 and a hole mobility of 5.6 cm 2 /V s, which are within a factor of 2 of typical values for p-type GaN films. There has been some irreproducibility in the Hall measurements for this and similar films, which may be related to punch-through of the contacts to the unintentionally-doped GaN underlayer [120]. However, assuming that the Hall measurements are correct, this experiment demonstrates that p-gan material with reasonable free-hole concentrations can be produced in MBE2 at nanowire growth temperatures. Considering the measured data for the planar films and the data reported in reference [9], the Mg concentration might be expected to be near optimal for several of the nanowire samples, assuming that the optimal concentration translates between planar and nanowire materials. Several samples with good morphology and near optimal Mg concentrations were selected as case studies for the following discussion and are summarized in Table 5.1. As shown in Fig.

137 120 Growth Run Mg cell Temp Base Diameter Expansion Low Temp Mg Tip Growth NW Length (mm) Contact Metal Anneal N C none 650 C 15 min Mg only 9 Ni/Au None N C 640 C 60 min undoped 640 C 60 min Mg doped 10 Ni/Au None p-anneal 450 C p-anneal 650 C Table 5.1. Growth conditions and device details for Mg:GaN nanowire experiments. Figure 5.5. FESEM images and PL measurements of N274 Mg:GaN nanowires. Figures show (a) FESEM images and (b) room temperature PL data for as-grown nanowire samples.

138 5.5, the N274 growth produced low-density single nanowires with an Mg concentration of 121 3x10 19 cm -3. Photoluminescence (PL) measurements for an as-grown sample indicate the expected near-band-edge emission at 3.4 ev, and an additional broad peak centered at 3.2 ev that may correspond to the donor-acceptor-pair transition (DAP), which is often observed in Mg doped material. It is possible that quasi-cubic defects related to stacking faults in the matrix layer [27] may also produce PL emission at 3.2 ev [121]. To explore this possibility, PL measurements were also made on a separate section of the same sample, in which only matrix layer and very few nanowires grew. No PL was observed for this section, suggesting that the matrix layer did not contribute to the observed DAP peak or the near-band-edge emission. While these measurements are only qualitative and do not resolve ionized vs. unionized donors, they do corroborate the SIMS data and further indicate that Mg has been incorporated into the nanowires. The I-V characteristics of N274 nanowires with Ni/Au (50 nm/170 nm) contacts are shown in Fig The nanowires were highly resistive and supported currents less than 10 na at 25 V applied bias. Minor improvements in the I-V characteristics were observed after a 10 minute 450 C contact anneal in air, however the characteristics deteriorated substantially after an additional anneal at 650 C under similar conditions. The minimum nanowire diameter ranged from nm with a mean average of 162 nm, as defined using FESEM diameter measurements on either side of the electrode gap. The minimum diameter is indicated in Fig. 5.6(a-c) by the color of the I-V curves. For nanowires annealed at 450 C, an abrupt transition in the maximum current at 25 V bias was observed for diameters near 150 nm, as shown in Fig. 5.6(e). Most nanowires thicker than 150 nm were coalesced, which raises the possibility that the higher conductivity may be related to defect-related conduction, however it should be emphasized that not all coalesced nanowires necessarily have defects.

139 Figure 5.6. Electrical characteristics of N274 Mg:GaN nanowires. I-V curves were obtained for single nanowires metallized with Ni/Au contacts. Measurements were taken with (a) no post-contact anneal, (b) after 10 min anneal at 450 C in air, and (c) after 10 min anneal at 650 C in air. The minimum nanowire diameter is indicated by the color of the traces. Also shown is the (d) calculated nanowire diameter corresponding to a surface band bending potential of 1.1 ev versus the concentration of ionized acceptors. The vertical bars indicate the calculated ionized acceptor concentration for an average Mg concentration of 3x10 19 cm -3, at acceptor ionization energies of 160 mev and 200 mev. The maximum current at 25 V bias for samples shown in (b) is plotted against the minimum nanowire diameter in (e). A typical N274 twoterminal device is shown in (f). 122

140 In general, contact resistances and surface depletion effects contribute to the overall 123 nanowire resistance. The asymmetry observed for many of the N274 I-V curves would not be expected if the overall resistance were to be dominated by surface depletion, suggesting that the contact resistances are asymmetric and non-negligible. The low conductivity and asymmetry was also observed under UV illumination [122], further indicating the role of the contacts. The extent of surface depletion can be estimated from equation 2.4, using the doping concentration measured by SIMS and a surface band bending value of 1.1 ev reported for Mg:GaN planar films [47]. Neglecting background donor or Mg auto-compensation effects, the concentration of ionized acceptors ( N ) can be estimated from the average Mg concentration (C a Mg ) as N a E C N a Mg v 2 e kt, (5.3) g where N v is the effective density of states in the conduction band, g is the degeneracy of the band states (g = 4 for acceptor states), E a is the acceptor activation energy, and kt is the thermal energy at room temperature [5, 123]. The reported activation energy for Mg acceptors ranges from mev, and results in a calculated ionized acceptor concentration of 2-5x10 17 cm -3, as indicated by the vertical color bars on Fig. 5.6(d). The nanowire diameter that produces a surface band bending of 1.1±0.1 ev is calculated using equation 2.4, and is plotted for two values of the reduced surface charge parameter ( ). These calculations indicate that nanowires with diameters between nm would be significantly depleted, and exhibit resistances that are increased by a factor of over the resistance at flatband conditions. The maximum nanowire current at 25 V shows an abrupt transition in conductivity near these diameters, as shown in Fig. 5.6(e); however, this effect may also be related to the coalesced nature of the thicker nanowires.

141 Mg-doped nanowires from the N216 growth run had an average Mg concentration of x10 20 cm -3, and were primarily larger coalesced structures with multiple roots. I-V characteristics were obtained for two-terminal devices with Ni/Au contacts and no post-contact anneal, as shown in Fig The I-V characteristics were typically asymmetric and non-ohmic, similar to the N274 devices, but did support significantly higher currents. However, single nanowires produced lower overall current levels, as shown in Fig 5.7(c). This suggests that the larger currents for coalesced nanowires may be related to defect conduction pathways or trapassisted tunneling at the contacts. In summary, a large fraction of the Mg-doped nanowires in this study were characterized by low conductivity and non-ohmic contacts, even when Mg is present at near optimal concentrations and standard contact metallization processes are used. Experimental results suggest that high contact resistance limits the overall conductivity, and calculations show that surface depletion becomes significant for thinner nanowires. Coalesced nanowires appear to support larger currents, although the conduction mechanism is not clear. Interestingly, while p- doping and p-contacts are widely recognized as essential for efficient nanowire LEDs, the only reports of confirmed ohmic p-type conductivity in single nanowire devices are for HVPE [77] and MOCVD [124] grown nanowires Axial p-n Junction Nanowires Axial p-n junction nanowires were grown by switching the Mg and Si doping fluxes midway through the growth, and used p-doping conditions expected to produce Mg concentrations similar to the p-type nanowires discussed above. The electrical characteristics of the nanowires and the individual p- and n-sections will be discussed, in relation to EL emission

142 Figure 5.7. Electrical characteristics of N216 Mg:GaN nanowires. FESEM images show (a) asgrown morphology and (b) typical two-terminal device structure metallized with Ni/Au contacts. I-V Measurements are shown in (c) for devices with no post-contact anneal, where the minimum nanowire diameter is indicated by the color of the traces. The maximum current at 25 V bias is plotted against the total minimum nanowire diameter in (d). 125

143 126 measurements. Issues related to low p-type conductivity will be shown to suppress EL in axial p-n junction NWs, despite otherwise diode-like electrical characteristics. Several axial p-n junction NW samples were grown by use of n-first (n-p, N140) and p- first (p-n, N272) growth sequences, as indicated in Table 5.2. Rectifying I-V characteristics were typically observed for N140 and N272 devices (Fig. 5.8) when electrodes bracketed the p-n junction. As expected, forward-bias current flow was observed in both samples when positive voltage was applied to the p-type section, with respect to negative bias on the n-type section of the NW. Interestingly, most of the p-n junction NWs supported substantial forward bias currents, unlike the p-type nanowires with similar doping discussed above. The ideality factors for conductive devices shown in Fig. 5.8 were in the range of 1-6, similar to ideality factors reported for planar GaN LEDs [125]. However, despite nearly ideal diode-like I-V characteristics, EL measurements were below detection limits for these devices even at current injection levels of 10 5 A/cm 2. In some of these samples, very low-intensity visible-wavelength EL was observed through the probe station microscope. An upper limit on the EQE was estimated at ~10-4 %, based on the detection limits of the spectrometer. However, the photoluminescence (PL) intensity ratio for single nanowires at 4 K and 290 K was measured to be 1.6 % for N140 and 1.0 % for N272. While the PL intensity ratio is only indicative of the relative IQE value [126], these PL measurements are similar to those reported previously [127], in which the absolute IQE was determined to be ~15 %. Thus, the estimated IQE for N140 and N272 exceeds the upper EQE estimate by approximately 5 orders of magnitude, suggesting that the nanowires are of reasonably high material quality, and that factors other than defects must account for the large discrepancy between the IQE and EQE.

144 Growth Run NW Structure NW Length ( m) Metallization Anneal N140 n-p 12 Pd/Pt/Au None N272 p-n 5 Ni/Au p-anneal D043 n 11 Ti/Al n-anneal D046 p 11 Ti/Al None D048 n-p/ebl 12 Ti/Al None D049 n-p 9 Ti/Al None N352 n-p/ebl 12 Ti/Al None 127 Table 5.2. Growth conditions and device details for axial p-n junction nanowire experiments.

145 Fig I-V characteristics of N140 and N272 axial p-n junction nanowires. I-V characteristics are plotted on linear and logarithmic scale for (a,c) N140 (n-p) and (b,d) N272 (p-n) devices with the p-n junction registered in the electrode gap and voltage bias applied to the NW tip. The flat sections of the curves near the top axis correspond to the current-limiting level imposed by the test equipment. 128

146 129 The individual sections of the axial p-n junction NWs were characterized by use of devices in which the electrodes were registered exclusively over the n- or p-regions. The n-region was observed to be highly conductive in both bias polarities for the N140 and N272 devices, as shown in Fig. 5.9(a) and Fig. 5.9(c). In contrast, the p-region was found to be insulating for the majority of N140 devices [Fig. 5.9(b)], with a smaller fraction of devices exhibiting asymmetric or rectifying I-V characteristics. The p-region in N272 devices was insufficiently long to characterize it exclusively; however, devices with electrodes registered mostly on the p-region exhibited a similar low conductivity under forward-bias conditions (see Fig. 5.8(b)). The p- region for the N272 devices was located on the thinner NW root with an average diameter of approximately 150 nm, indicating substantial surface depletion even for ideal doping concentrations. However, insulating behavior was also observed for N140 devices with large p- region diameters (600 nm-1100 nm), suggesting that surface-depletion effects are not solely responsible for the low conductivity and likely include high p-contact resistance effects. The Mg doping concentrations for both samples was estimated in the low cm -3 range, based on comparison to nanowires grown under similar conditions and discussed in section It should be noted that the non-ohmic behavior of the p-region in N140 is similar to the separate p- type NW samples discussed above. A possible explanation for the discrepancy between EQE and IQE is carrier injection imbalance in the p-region, where holes supplied by the p-contact are vastly outnumbered by electrons injected from the conductive n-region. The recombination rate for injected electrons decreases, due to the low hole concentration, and allows an electron current to flow directly into the p-contact. While the overall effect is similar to electron overflow from quantum wells [128, 129], the root cause in this study is attributed to the low conductivity observed for the p-region,

147 Fig I-V characteristics of n- and p- sections of N140 and N272 axial p-n junction nanowires. I-V characteristics are plotted for NW devices with electrode gap registered over (a) N140 n-region, (b) N140 p-region, and (c) N272 n-region with voltage bias applied to the NW tip. Representative diode-like and asymmetric I-V curves are shown as dashed and dotted lines, respectively for N140 p-region devices (some curves removed for clarity). The top axis corresponds to the current-limiting level imposed by the test equipment. Also shown are (d) SEM images of typical NW and device structures for these samples, with the expected junction location indicated by the vertical white dashed line and electrodes indicated by the white brackets. 130

148 131 and results in a more dramatic reduction in the EQE. As illustrated by the band diagram shown in Fig. 5.10(a), the n-region is considered to be ohmic and quasi-neutral, while the p-region is partially depleted and incorporates a Schottky-like p-contact that blocks injection of holes [4]. Under forward bias, electrons are injected from the junction, causing the p-region to become further depleted. As very little potential develops on the p-contact, increased forward bias simply results in larger carrier injection imbalance, which supports large current flow with little emission of EL Electron Blocking Layers An AlGaN electron blocking layer (EBL) was incorporated into the axial p-n junction nanowire device structure to mitigate the current imbalance caused by the high p-contact resistance. As shown in Fig. 5.10(b), the EBL presents a barrier in the conduction band of the p- region, which blocks the flow of electrons to the p-contact. In theory, this should enable a larger reverse bias to develop on the p-contact, increasing hole injection through thermionic field emission or other reverse-breakdown current-flow mechanisms [130, 131]. Additional samples were grown in MBE3 for this study, including n-p/ebl NWs (D048), n-p NWs (D049), n-type NWs (D043), and p-type NWs (D046), as indicated in Table 5.2. The EBL was nominally 20 nm thick with an alloy composition of Al 0.15 Ga 0.85 N and was positioned 1 m from the junction in the p-region. As shown in Fig. 5.11(a), the I-V characteristics of the n- type NWs were highly conductive, while the p-type NWs exhibited rectifying behavior similar to that of the p-region of some of the N140 devices discussed above. The average Mg concentration of the D046 p-type NWs was measured by SIMS to be 1x10 19 cm -3. The asymmetry of current flow for the p-type NWs was such that higher current flowed when positive bias was applied to the NW tip. There are several plausible explanations for this

149 Fig Schematic band diagrams of axial p-n junction nanowires illustrating effect of EBL. Diagrams show (a) axial p-n junction with blocking p-contact and resulting electron overflow, and (b) axial p-n junction with EBL for reducing electron overflow and increasing bias on p-contact for increased hole injection. Inactive current-flow mechanisms are indicated by a large X. 132

150 Fig I-V characteristics of D048 and D049 p-n junction nanowires and related structures. Measurements were obtained for (a) D043 (n-type) and D046 (p-type) NWs; (b) D048 (n-p/ebl) NWs; and (c) D049 (n-p) NWs with voltage bias applied to the NW tip. Also shown are (d) SEM images of typical NW and device structures for these samples with the expected junction location indicated by the vertical white dashed line and electrodes indicated by the white brackets. 133

151 observation, including a gradient in the Mg doping concentration, asymmetry in the NW 134 diameter, or polarization-induced charges at the NW ends. The n-p/ebl NWs exhibited I-V characteristics [Fig. 5.11(b)] and ideality factors (n~30) similar to those of the p-type NWs, which is consistent with total device current being limited by hole injection from the p-contact, rather than by electron overflow to the p-contact. Approximately half of the n-p/ebl devices produced band-edge EL with peak emission wavelength between 363 nm-365 nm and narrow line-widths (FWHM of 7 nm-12 nm), as shown in Fig A longer wavelength shoulder peak near 380 nm was sometimes observed and may be related to donor-acceptor-pair transitions [132]. The peak emission intensity increased monotonically with current and, in some cases, exhibited above-linear behavior, similar to reports elsewhere [54]. The EL was observed to originate from the junction (Fig. 5.12), suggesting that it results from p-n carrier injection and not from minority carrier injection by the contacts [133]. However, impact ionization may also play a role, as relatively high forward biases (~35 V) were required to produce EL in these devices [134]. The EL was observed at the p-n junction for NWs coalesced along the full NW length, coalesced only at the p-type tip, and with no obvious points of coalescence (Fig. 5.12). While coalescence can affect the NW conductivity, it does not appear to determine the location of EL or to short-circuit the junction in this limited number of devices. The total optical power for the NW LEDs shown in Fig was estimated at approximately 9 nw-26 nw with external quantum efficiencies ranging from 0.01 %-0.07 %. For comparison, the EQE of the first planar GaN p-n junction LEDs was reported in the range of 0.1 %-1 % [135, 136]; limited in part by p-type doping activation, similar to observations in this study. Reports of absolute EQE measurements for single GaN

152 Fig EL spectra, EL images, and FESEM images of D048 nanowires. The EL is observed at the expected location of the p-n junction for D048 (n-p/ebl) NWs, which is indicated by the dashed line. EL spectra were measured at various current injection levels, as shown. 135

153 136 NW LEDs are less common, but include 5.8 % for MOCVD grown core-sleeve heterostructures [74] and % for MBE grown axial heterostructures [137]. The EQE in the latter reference was obtained by measuring individual NWs in an array, where it was found that a significant fraction of the NWs did not exhibit EL, and was speculated to result from variation in the forward-bias voltage of the individual NWs. Current planar LEDs using optimized and high efficiency heterostructures have efficiencies on the order of 40% [135]. EL was also observed for a small fraction (3 of 13) of thick and coalesced D049 n-p NWs. In contrast to the D048 n-p/ebl NWs, the EL emission was non-localized along the NW length, and with lower EQEs ranging from %. The remainder of the D049 n-p nanowires were very thin (average diameter ~110 nm), with uniformly insulating electrical characteristics similar to some of the N272 p-n NWs. The overall low conductivity observed for these structures may have allowed larger biases to develop on the p-contact, similar to the n-p/ebl samples Model of GaN Nanowire LEDs This section provides a theoretical framework to the experimental observations discussed in the preceding sections, specifically the low EQE of the axial p-n junctions and the improvements when an EBL is used. The model calculates the 1-D axial characteristics of the p- n junctions and neglects radial effects related to the surface band bending. This approximation is valid for thick nanowires, although it may also be applicable to thinner nanowires as discussed later. As shown in Fig 5.13(a), the p-contact is assumed to be a reverse-biased Schottky barrier with variable barrier height, which is used to parameterize the inadequacy of the p-contact. Ideal p-doping levels are assumed for the p-region, and n-doping levels are chosen such that the

154 Fig Schematic (a) band diagram and (b) equivalent circuit of axial p-n junction structure used used to model injection efficiency. 137

155 138 junctions have n + -p doping asymmetry. For this scenario, the analysis can be restricted to the p- region of the device, producing the equivalent circuit shown in Fig. 5.13(b). The p-contact Schottky barrier and p-n junction are indicated as diodes, and the overflow of unrecombined electrons is represented by a current source linked to the p-n injection current. This approach is similar to the Ebers-Moll model for bipolar junction transistors with an open base configuration. The p-contact hole injection characteristics are determined by the reverse saturation current of the Schottky barrier, which results from electrons emitted over or transmitted through the barrier. Thermionic field emission (TFE) is the dominant current flow mechanism for the p- doping levels considered in this analysis, and corresponds to holes tunneling at energies just below the barrier. The barrier thins with increasing reverse bias, resulting in higher tunneling probabilities for emission of holes into the p-region. The tunnel injection current is also sensitive to other factors, including gap states and image force barrier lowering. These effects were neglected here as the purpose of this analysis is to demonstrate relative variations in the barrier height, rather than absolute values. The Schottky barrier band-bending and current-injection profiles are shown for several bias levels in Fig. 5.14(a), as calculated using the parabolic barrier described in appendix 1 and using the parameters listed in Table 5.3. It should be noted that the contact-depletion region only extends several hundred nm into p-region, even for large biases. The I-V curves for a 200 nm diameter nanowire are calculated for barrier heights ranging from V, as shown in Fig 5.14(b). The barrier heights and bias conditions shown in the plot cover the typical current range observed for p-type nanowires. The ideality factor for these curves is , as determined from the slope of the log-scale I-V plots. The hole injection current is plotted against the barrier height for two bias levels in Fig. 5.14(c), which correspond

156 Fig Calculated p-contact Schottky barrier characteristics. Figures show (a) band bending under varying bias, (b) I-V characteristics for various barrier heights, and (c) hole current injection levels at 3.25 V and 30 V bias conditions. All calculations were made using an ionized acceptor concentration of 3x10 17 cm -3, with the exception of the error bars in c) which indicate the hole injection current for an ionized acceptor concentration of 1x10 16 cm

157 Parameter Name Ionized acceptor concentration Ionized donor concentration Schottky barrier height Parameter Symbol Value Units N a 3x10 17 cm -3 N d 1x10 19 cm -3 b 0.2, 0.3, 0.4, 0.5 V 140 Nanowire diameter d NW 200 nm Width of p-region w p 2 m p-contact length w c 1 m Temperature T 300 K Intrinsic concentration n i 1.9x10-10 cm -3 Hole effective mass m h Electron effective mass Minority Hole Lifetime Minority Electron Lifetime Hole diffusion coefficient Electron diffusion coefficient Effective DOS in valence band Effective DOS in conduction band Spontaneous recombination coefficienc m e h 0.5 ref[138] ns e 0.01 ns D p 0.75 cm 2 V -1 s -1 D n 39 cm 2 V -1 s -1 N v 1.8x10 19 cm -3 N c 2.3x10 18 cm -3 B cm 3 s -1 Relative permittivity Richardson constant A * ref[139] Acm -2 K -2 Table 5.3. Parameters used for modeling injection efficiency in axial p-n junction nanowires.

158 141 to the approximate potentials applied to the p-n and p-n/ebl nanowire structures discussed in the sections above. The lower error bars indicate the hole injection current for a much lower p- doping concentration (1x10 16 cm -3 ), which corresponds to a decrease in the hole injection current of about 1 decade. The active region in the axial n + -p junction is in the p-region, defined by the edges of the p-contact and p-n junction depletion regions, from which holes and electrons are injected respectively. In contrast to the usual approach, recombination in the p-region is not determined solely by the electron minority carrier lifetime and excess electron concentration, due to the nonequilibrium hole concentrations arising from the Schottky barrier p-contact. Instead, the continuity equations for holes and electrons must be solved using the overall recombination rate (R). Assuming dark and steady-state conditions, the continuity equations can be written as D D 2 dn n 2 dx 2 d p p 2 dx R 0 R 0, (5.4) where n is the electron concentration, p is the hole concentration, D n is the electron diffusion coefficient, and D h is the hole diffusion coefficient. The overall recombination rate includes spontaneous radiative recombination (R rad ) and non-radiative Shockley-Read-Hall (R SRH ) recombination and is given by R Rrad RSRH Bnp pn n 2 i n n p n p i n i, (5.5) where B is the spontaneous emission coefficient, p is the minority hole lifetime, n is the minority electron lifetime, and n i is the intrinsic carrier concentration. The SRH recombination

159 142 term assumes a mid-gap energy trap level. Auger recombination was found to be small using the Auger coefficient reported for defect-free nanowires [52] and was neglected in this analysis. The coupled 2 nd order continuity equations were solved numerically, using boundary conditions derived from the carrier injection levels as dp J ( x 0) dx q D h, pc p dp ( x w p ) 0 dx dn Je, j ( x w p ) dx q D nx ( 0) 0 n, (5.6) where w p is the width of the active region, J h,pc is the hole current density injected by the p- contact, J e,j is the electron current density injected by the p-n junction, and q is the electron charge. The values of J h,pc used for p-contact injection current are those shown in Fig. 5.14(c). The hole current is prescribed to be zero at the p-n junction, consistent with the assumption of n + -p doping. The value for J e,j is set at a constant value and calculated as a 5 A current injected into a 200 nm diameter nanowire. This corresponds to a p-n junction bias of 3.25 V, as calculated using the ideal diode equation. The vanishing electron concentration at x=0 is commonly used in modeling short p-n junctions. An active region width of 2 m is used for all calculations. The solutions to the continuity equations are shown in Fig. 5.15; yielding carrier concentrations, current densities, and emission profiles in the active region of the device. The electron concentration in the active region far exceeds the hole concentration, as the injected hole current is insufficient to support recombination of the injected electrons. Consequently, the

160 Fig Calculated injection efficiencies for axial p-n junction nanowires. Solutions to continuity equations show (a) carrier concentrations, (b) electron and hole current densities, (c) radiative emission rate, and (d) injection efficiency and optical power. The colors correspond to hole injection currents calculated using a barrier height of 0.2 ev (red), 0.3 ev (green), 0.4 ev (blue), 0.5 ev (gray). 143

161 144 electron current density is nearly constant across the active region of the device, and results in significant electron current into to the p-contact. This indicates that the n-region effectively extends to the p-contact and that the device does not operate as a conventional p-n junction. Instead, the radiative recombination rate is limited by injected holes with a radiative emission profile that is skewed towards the p-contact, as shown in Fig. 5.15(c). It can also be noted that the electron concentration is potentially high enough to provide screening of the surface charges in the p-region, which may flatten the radial band bending profiles and mitigate surface depletion in thinner nanowires. The injection efficiency and optical power of the axial p-n junction nanowires were calculated from carrier concentration profiles. The total number of recombining carriers (N rec ) in the nanowire is given by w p 2 pn n i Nrec ANW Bnp dx, (5.7) 0 p n ni n p ni where A NW is the nanowire cross-section area. The injection efficiency ( inj ) is defined as the total carrier recombination rate to the total carrier injection rate and is calculated as inj Nrec Jtot A q NW. (5.8) Finally, the total optical emission power (P NW ) is calculated from the spontaneous recombination rate as w p PNW Eph ANW B n p dx, (5.9) 0

162 where E ph is the photon energy at 365 nm. As shown in Fig 5.15(d), the injection efficiency 145 accounts for over 3 decades of loss even at b =0.2 ev, and drops precipitously for higher barrier p-contacts. The total optical power calculated for the axial p-n junctions is less than 1 nw over the full range of p-contact barrier heights studied, which is consistent with the below-detectionlimit EL measurements obtained experimentally. In summary, this model corroborates the experimental data showing that a strong loss mechanism related to carrier overflow exists when the electron and hole injection levels are grossly mismatched. By incorporating an EBL in the active region, the flow of excess electrons through the p- region can be blocked, allowing accumulation of electrons between the EBL and the p-n junction. The height of the EBL ( EBL ) is approximately 0.3 ev for the AlGaN alloy composition used in the experiments, and can be used to estimate the electron overflow current parameter ( ) as the fraction of electrons with energy sufficient to surmount the EBL. The concentration of electrons injected into the active region is related to the position of the quasi-fermi level for electrons (E fn ) according to n inj ( Ec Efn) Nc exp kt, (5.10) where N c is the effective density of states in the conduction band and E c is the conduction band energy level. Likewise, the concentration of electrons with energies greater than the EBL is given by n E EBL ( Efn EEBL ) Nc exp kt. (5.11) The ratio of these concentrations gives the overflow parameter

163 146 ne EBL EBL exp n kt, (5.12) inj which is calculated to be % for an EBL with 0.3 ev barrier height, indicating negligible carrier overflow loss and inj near unity. In this case, the overall EQE = IQE ext and is calculated at 0.2 %, which is within a factor of 3 of the best EQE reported above for a p-n/ebl nanowire. The forward voltage of the p-n/ebl structure can be estimated, knowing that electron overflow current can be neglected. As shown in Fig 5.13(b), the hole current in the p-contact and electron current in the p-n junction are equal with 0, allowing the total bias to be calculated from the individual junction characteristics. The forward-bias J-V characteristics are expressed as kt J V n q J V e, j j j ln e, js kt J h, pc pc npc ln q J h, pcs, (5.13) where n j /n pc and J e,js /J h,pcs are the ideality factors and saturation current densities of the p-n junction and p-contact Schottky barrier, respectively. Assuming the saturation current densities are equal, the J-V characteristics reduce to qv qv J Jh exp exp, pcs Jh, pcs n n pc n j kt pckt. (5.14) using the observation that the ideality factor of the p-contact is much greater than the 1-2 usually observed for a p-n junction. This shows that that the I-V characteristics of the p-n/ebl

164 147 nanowires are determined primarily by the large ideality factor of the p-contact [140], as shown in Fig. 5.14(b), and that biases on the order of tens of volts may be required to produce currents in the A range. This relationship between p-contact and p-n/ebl ideality factors was observed experimentally for p-type (D046) and p-n/ebl (D048) nanowires as shown in Fig Fig Ideality factors for p-type, n-p/ebl, and p-n junction nanowires. Histogram shows ideality factors for p-type [D046], n-p/ebl [D048], and p-n [N272] nanowires. The arrows indicate the mode values of the distributions.

165 5.2 Nanowire Optical Interconnects 148 The final section of this chapter presents an application-level nanowire device, based on the functional LED nanowires demonstrated in the previous section. Gallium nitride nanowires present novel opportunities for integrating optically active nanoscale components with nonphotonic devices. They can be grown on epitaxially-convenient substrates and transferred to separate device substrates, bypassing process integration constraints that would otherwise be encountered for direct GaN growth on the device substrate. This approach has been utilized previously to fabricate discrete light-emitting diode nanowire (LED NW) devices [72-74] and photoconductive nanowire (PC NW) devices [45, 46, 82] on non-native substrates. Until now, a GaN NW device comprising both LED and PC NWs has yet to be demonstrated, and is the objective of this section. This type of device would be attractive for providing on-chip optical interconnects and isolation between MEMS and CMOS subsystems, which may operate at significantly different bias conditions. For instance, optically-coupled nanowires could be used as a nano-scale solid state relay for turning on higher power MEMS actuators or for transmission of sensor signals which may have high or floating potentials back to CMOS circuitry. Some of the salient features of this device include a low off-state current and high sensitivity for the photodetector nanowire; as well as highly divergent emission for the LED nanowire, which would minimize cross-talk to adjacent CMOS components. Optical coupling in an LED/PC NW pair occurs when a forward-biased LED NW illuminates a PC NW, inducing a photocurrent, as shown in Figure 5.17(a). Unlike conventional planar p-n junction photodetectors, GaN NW photoconductivity results when the surface depletion layer shrinks as photogenerated minority carriers accumulate at the NW surface

166 Figure Schematic diagram and images of optical interconnect device. Figures show (a) schematic illustration of device concept and operation, (b) layout and testing conventions for fabricated device, and (c) SEM images for coupled LED/PC NW device [N352]. 149

167 150 [46, 82]. For an axial p-n junction GaN NW that is sufficiently long to accommodate multiple contact locations, either LED or PC functionality can be selected, depending on the contact layout with respect to the junction. Contacts positioned on opposite sides of the p-n junction produce an LED NW, while a PC NW is obtained for contacts positioned on the n-side only. This affords considerable simplification to the device fabrication process by allowing NWs from a single growth run to serve as both source and detector. In this section, the approach described above is employed to fabricate an LED/PC NW pair, investigate the characteristics of the individual LED and PC NW components, and measure the overall response of the coupled system. Axial p-n junction nanowires were grown n-side first, with the p-n junction located approximately at the middle of the NW, and with an overall length of approximately 12 m (N352, see Table 5.2). The nanowires incorporate an AlGaN EBL that is located 1 m from the junction on the p-side. The overall nanowire structure and EBL were nominally identical to that of the D048 nanowires discussed above. It is noteworthy for that the LED NWs based on the n- p/ebl structure produce an EL spectrum at the GaN band-edge for efficient re-absorption by PC NWs, which would not be the case for GaN NW LEDs with quantum wells. The p-region conductivity in these nanowires was similar to the N140 devices discussed above; however, the n-region was also measured to be of low conductivity, the cause of which is not presently understood. Nonetheless, this provides functionality for this particular application, as lightly doped and highly depleted nanowires produce a large photoconductive response. The NWs were released from the growth substrate into suspension via ultrasonic agitation in isopropanol, and dispersed onto an oxidized silicon substrate. An array of two-terminal LED NW electrodes and PC NW contact pads was created by use of optical photolithography, e-beam

168 151 evaporation, and lift-off processing. The devices were then pre-screened for proper LED NW contact registration and availability of a second proximal PC NW. Local electrodes to the n- region of the PC NW, identifiable as the smaller-diameter section of the NW, were then created by use of e-beam lithography and a second metallization to produce the device shown in Figures 17(b) and 17(c). Both contact metallizations consisted of a 20 nm Ti/ 200 nm Al stack, with no post-contact anneal. The electrical and optical characteristics of the LED NW are shown in Figure 5.18, by use of the testing conventions illustrated in Figure 5.17(b). Rectifying I-V characteristics were obtained with forward-bias conditions corresponding to the expected bias polarity. The large ideality factor observed in the I-V characteristics is similar to that of the D048 nanowires discussed previously. The electroluminescence spectra exhibit significant content in the 365 nm 370 nm range, with intensity increasing monotonically with current injection level. The total optical power is estimated to be approximately 40 nw for 20 A injection current. The electroluminescence is observed to emanate from the junction, approximately 35 m from the PC NW. Some diffuse scattering from the contact or minor electroluminescence is also observed near the left side of the electrode gap. The PC NW response was characterized with a wavelength-tunable UV light source based on a xenon arc lamp and a monochromator. I-V measurements of the initial dark level were less than 10 pa for +/- 10 V bias, indicating that the PC NW is substantially depleted. As shown in Figure 5.19(a), illumination with above-bandgap light induces a photocurrent several decades above the initial dark level. The wavelength dependence of the photocurrent is characteristic of GaN, decreasing abruptly for wavelengths longer than 365 nm. The post-uv illumination dark

169 Figure LED NW characteristics of optical interconnect device. Figures show (a) I-V characteristics, (b) electroluminescence spectra, and (c) image of electroluminescence (EL) emission from LED NW [N352]. 152

170 Figure PC NW characteristics of optical interconnect device. Figures show (a) photocurrent as function of optical excitation wavelength (inset shows logarithm scale plot), (b) photoconductive decay measurements, and (c) photocurrent as function of optical excitation intensity for PC NW [N352]. 153

171 level was increased significantly from the initial dark level, indicated by the persistent 154 photoconductivity level on the inset of Figure 5.19(a). Because the switching speed and lower detection limit of the PC NW is related in part to the persistent photoconductivity level, its time dependence was studied in more detail. As shown in Figure 5.19(b), asymmetry with respect to the PC NW bias polarity was observed in the steady-state photocurrent and the photoconductive decay rate. The photoconductive gain is higher for the +10 V bias condition; however, a period of several minutes is required for decay back to the persistent photoconductivity level, which was then stable for a duration in excess of 12 hours. In contrast, the -10 V bias condition exhibited a relatively fast decay rate, but at the expense of lower photoconductive gain. Interestingly, a brief bias duration at -10 V was able to restore the initial dark level of the PC NW device. It is possible that the bias-dependent asymmetry in the photoconductive decay rate results from the proximity of the p-n junction or from surface or trap states, although further study would be required to clarify this phenomenon. The PC NW sensitivity was established by correlating the steady-state photocurrent to the illumination intensity for several wavelengths. The more sensitive +10 V PC NW bias condition was used for these measurements, starting from the persistent photoconductivity level (represented by the bottom axis in Figure 5.19(c)). These measurements demonstrate that the PC NW should be sensitive to irradiance levels of approximately 10-7 W/cm 2 for a spectral output similar to that of the LED NW (370 nm peak intensity). Minor sensitivity to sub-gap illumination is also observed, similar to the wavelength scan shown in Figure 5.19(a). The coupled response between the LED NW and the PC NW was characterized by fourterminal measurements, according to the test configuration shown in Figure 5.17(b). Current

172 155 pulses were applied to the LED NW, with an on-state duration of 5 s followed by an off-state duration of 10 s. The resulting PC NW photocurrent is shown in Figure 5.20(a), at a fixed PC NW bias of +10 V. These waveforms show the initial three pulses starting from the persistent photoconductivity level. The PC NW photocurrent tracks the output state of the LED NW, indicating optical coupling between the NWs. Increasing the on-state current level and optical emission from the LED NW induces a larger photocurrent in the PC NW, as expected. The magnitude of the induced photocurrent pulses increases with iteration, as the off-state duration is shorter than the period required for full decay to the initial persistent photoconductivity level. To eliminate the possibility of coupling through parasitic leakage pathways, measurements were taken with the LED NW ground pad disconnected. In this scenario, bias pulses were applied to the device, but without optical emission from the LED NW. No pulsed response was observed in the PC NW photocurrent, indicating that the coupling was not due to parasitic leakage pathways. Steady-state measurements indicate that the induced photocurrent stabilizes at ~3 na for an LED current injection level of 20 A. This corresponds to an optical intensity at the PC NW in the range of mid 10-7 W/cm 2, as correlated to the experimental data in Figure 5.19(c). This is well below the estimated intensity supplied by the LED NW (mid 10-4 W/cm 2 ), assuming the LED emission pattern is uniform in all directions. More likely, the far-field emission pattern is non-uniform [141], with a line of sight that is partially occluded by the contact pads, as shown in Figure Further attenuation may result from transmission losses through off-normal interfaces along the line of sight between the NWs. A key feature of this device is the ability to maintain optical coupling while floating the LED and PC NW bias levels relative to one another, provided that the potential drop on the individual components remains constant. Figure 5.20(b) shows pulsed operation under

173 Figure Coupled LED/PC measurements of optical interconnect device. Four-terminal measurements were obtained for 5 s I LED pulses and show I PC for (a) increasing LED NW current injection levels and (b) offset LED/PC NW bias conditions for coupled LED/PC NWs [N352]. 156