Studies on the Reliability of Ni-Gate Aluminum Gallium Nitride / Gallium Nitride High Electron Mobility Transistors Usin...


Material Information

Studies on the Reliability of Ni-Gate Aluminum Gallium Nitride / Gallium Nitride High Electron Mobility Transistors Using Chemical Deprocessing
Physical Description:
1 online resource (159 p.)
Whiting, Patrick G
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Materials Science and Engineering
Committee Chair:
Committee Co-Chair:
Committee Members:


Subjects / Keywords:
afm -- aluminum -- gallium -- hemt -- nitride -- sem
Materials Science and Engineering -- Dissertations, Academic -- UF
Materials Science and Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation


Aluminum Gallium Nitride / Gallium Nitride High Electron Mobility Transistors are becoming the technology of choice for applications where hundreds of volts need to be applied in a circuit at frequencies in the hundreds of gigahertz, such as microwave communications.  However,because these devices are very new, their reliability in the field is not well understood, partly because of the stochastic nature of the defects which formas a result of their operation.  Many analytical techniques are not well suited to the analysis of these defectsbecause they sample regions of the device which are either too small or too large for accurate observation.   The use of chemical deprocessing in addition to surface-sensitive analysis techniques such as Scanning Electron Microscopy and Scanning Probe Microscopy can be utilized in the analysis of defect formation in devices formed with nickel gates.  Hydrofluoric acid is used to etch the passivation nitride which covers the semiconducting layer of the transistor.  A metal etch utilizing FeCN/KI is used to etch the ohmic and gate contacts of the device and a long exposure in various solvent solutions is used to remove organic contaminants,exposing the surface of the semiconducting layer for analysis.   Deprocessing was used in conjunction with a variety of metrology techniques to analyze three different defects.  One of these defects is a nanoscale crack which emanates from metal inclusions formed during alloying of the ohmic contacts of the device prior to use in the field, could impact the yield of production-level manufacturing of these devices.  This defect also appears to grow, in some cases, during electrostatic stressing. Another defect, a native oxide at the surface of the semiconducting layer which appears to react in the presence of an electric field, has not been observed before during post-mortem analysis of degraded devices.  It could play a major part in the degredation of the gate contact during high-field, off-mode electrostatic stressing and could be the initiator of the pitting of the semiconducting layer of the gate contact, a defect which was also observed.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Patrick G Whiting.
Thesis (Ph.D.)--University of Florida, 2013.
Co-adviser: LAW,MARK E.

Record Information

Source Institution:
Rights Management:
Applicable rights reserved.
lcc - LD1780 2013
System ID:

This item is only available as the following downloads:

Full Text




2 2013 Patrick Guzek Whiting


3 For my family and friends


4 ACKNOWLEDGMEN TS I quite literally wouldn't be where I am if it weren't for my parents. I don't mean this in a superficial way, that they were the ones who brought me into this world, but that if it weren't for their constant support I would have never been able to ach ieve this. I learned to read and write because they were there to struggle with me every step of the way. So, it's no surprise that I love them. I also owe a great debt of gratitude to every single person I've ever gone to school with, from highschool, through college and on into graduate school. Without the intellectual stimulation of the people who I've had around me, I wouldn't have grown as a student and as a scientist. This work isn't j ust the product of years of experiments done in laboratories. It is the product of many nights spent working on so l id state physics with friends like Andrew Wagner and Ryan Aguinaldo. It is the product of struggling through fracture mechanics equations with Nicholas Vito. It is the product of going to conferences wi th Brad Yates, Blake Darby, Aaron Lind and Ray Holzworth. And it's the product of learning how to catch a fish with Dr. Kevin Jones. On that subject, I should mention that I owe a lot to the advisors I've had over the years as well as all of my teachers. Chronologically, this starts with Dr. Dale Ewbank and Dr. Karl Hirschman, of Rochester Institute of Technology, who encouraged me to do research when I was a freshman and who made sure that I never lacked for interesting intellectual problems to obsess o ver. It was they who saw me through the early years of my academic career and it was Kevin S. Jones, of University of Florida, who saw me through my doctoral research. I feel that I've grown a lot as an engineer (not the least of which being far more hum ble) as a result of being here and as a direct result of his guidance. I've learned that I know a lot, and that the amount that I don't know is incredibly vast in comparison to what I do know. I would also like to thank the other


5 members of my dissertati on committee. I've worked with all of them through the years and their guidance was integral in my research. I owe a lot to Dr. Gerald Bourne, Dr. Nick Rudawski and Eric Lambers, of MAIC, who were invaluable when I needed help with FIB/SEM, TEM and AFM. Dr. Brent Gila, of NRF and one of my comittee members, has also been immensely helpful since the earliest days of my work on this project. I'd also like to thank Dr. Yan Xin and the staff of the National High Magnetic Field Lab for the use of their ARM 2 00F TEM and peerless expertise with the instrument in analyzing several samples. Lastly, I would like to thank the men and women of the United States Air Force. Without their financial support, this project would not have been possible.


6 TABLE OF CONTEN TS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 LIST OF ABBREVIATIONS ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................................ ... 11 C HAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 13 1.1 Power Transistor Technologies ................................ ................................ ........................ 13 1.2 AlGaN/GaN HEMT Processing ................................ ................................ ....................... 17 1.3 AlGaN/GaN HEMT Operation ................................ ................................ ......................... 20 1.4 Factors Affecting AlGaN/GaN HEMT Electrical Characteristics ................................ .... 24 2 EXPERIMENTAL METHODS ................................ ................................ ............................. 31 2.1 Electrical Measurements ................................ ................................ ................................ ... 32 2.2 Transmission Electron Microscopy ................................ ................................ .................. 34 2.3 AlGaN/GaN HEMT Maskset and Chemical Deprocessing ................................ .............. 43 2.4 Scanning Electron Microscopy ................................ ................................ ......................... 50 2.5 Surface Probe Microscopy ................................ ................................ ................................ 56 3 OHMIC NANOCRACK FORMATION AND ITS IMPACT ON RELIABILITY .............. 68 3. 1 Progress in Ohmic Contact Annealing ................................ ................................ ............. 70 3.2 Metal Inclusions and Nanocrack Formation ................................ ................................ ..... 74 3.3 Nanocrack Morphology ................................ ................................ ................................ .... 76 4 PROGRESS IN THE ANALYSIS OF THE RELIABILITY OF THE GATE ELECTRODE ................................ ................................ ................................ ......................... 88 4.1 Inverse Piezoelectric Strain and the Reliability of Pt Gate HEMT s ................................ 89 4.2 Reliability of Ni Gate HEMTs ................................ ................................ ......................... 98 4.3 Efforts in Surface Characterization of AlGaN/GaN HEMTs ................................ ......... 104 5 RELIABILITY OF THE GATE CONTACT ................................ ................................ ....... 112 5.1 Device Quality Below the Critical Voltage ................................ ................................ .... 115 5.2 Defect Formation at the Critical Voltage ................................ ................................ ........ 116 5.3 Defect Formation Beyond the Critical Voltage ................................ .............................. 128 6 CONCLUSIONS ................................ ................................ ................................ .................. 145


7 LIST OF REFERENCES ................................ ................................ ................................ ............. 150 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 159


8 LIST OF FIGURES Figure page 1 1 Schematics of two different power device technologies.. ................................ .................. 27 1 2 A performance diagram for a variety of power transistor families.Performance is gauged in terms of breakdown voltage and operational frequency.. ................................ 27 1 3 A process flow for the cr eation of an AlGaN/GaN HEMT ................................ .............. 28 1 4 The electrical behavior of a schottky gate in forward and reversed bias.. ......................... 29 1 5 A schematic of the op erational modes of a transistor ................................ ....................... 30 2 1 A schematic view of the electron o ptics utilized in a typical TEM ................................ ... 61 2 2 The steps associated with Focused Ion Beam milling of a generic sample. Progression runs down t he left column and then down the right column. ......................... 62 2 3 A top down image of a typical maskset analyzed as part of this work. ............................. 63 2 4 The Deprocessing of an AlGaN/GaN HEMT. ................................ ................................ ... 63 2 5 AlGaN surfaces after exposure to HF. ................................ ................................ ............... 64 2 6 AlGaN surfaces after exposure to var ious organic solvents ................................ ............. 65 2 7 Incomplete Removal of SiN from a T gate. ................................ ................................ ....... 66 2 8 HAADF STEM images of deprocessing of the near gate region of a HEMT. ................. 66 2 9 A schematic diagram of a basic SEM system. ................................ ................................ ... 67 2 10 A schematic diagram of a generalized SPM system. ................................ ......................... 67 3 1 A HAADF STEM image of the cross section of a 100nm gatelength device. .................. 83 3 2 A metal inclusion formed after the annealing o f a Al/Ni/Ti/Au metal stack. .................... 83 3 3 Top down SEM analysis of the ohmic contact regions of an AlGaN/GaN HEMT. These ohmic contacts were formed via an anneal at 850C for 30s ................................ .. 84 3 4 FIB/TEM of a nanocrack observed in SEM. ................................ ................................ ..... 84 3 5 Histograms of crack lengths observed in 20 HEMT devices. ................................ ............ 85 3 6 Stepped stressing of an AlGaN/GaN HEMT and resulting crack formation. .................... 86


9 3 7 The resulting crack distribution and associated tensile stress for crack growth in the stressed HEMT. ................................ ................................ ................................ .................. 87 4 1 Piezoelectric strain (white arrows) resulting from a vertically applied electric field (black line) in an AlGaN/GaN HEMT. ................................ ................................ ............ 108 4 2 The dependance of gate leakage current with increasing electrostatic stress. ................. 108 4 3 Pulsed stress experiments on an AlGaN/GaN HEMT. ................................ .................... 109 4 4 A schematic image of a crack like defect formed at the drain side of the gate. .............. 109 4 5 The transient electrical measurement performed by Kubal and coworkers. .................... 110 4 6 An exemplary stressing experiment performed on a Ni gated AlGaN/GaN HEMT. ...... 111 4 7 BF TEM image of a defect un der the gate of an AlGaN/GaN HEMT. ........................... 111 5 1 A plot of gate current degradation and associated defect formation prior to V CRIT ........ 132 5 2 A plot of the gate degradation up to a voltage approximating V CRIT .............................. 133 5 3 A plot of the degradation of a circular HEMT, with the same "banding" defect. ............ 134 5 4 Morphological differences in banding induced by electrostatic stress and annealing. .... 135 5 5 The gate leakage current before and after annealing at 500C for 30 min. ...................... 137 5 6 SPM of an AlGaN/GaN HEMT annealed at 500C for 30 min. ................................ ...... 138 5 7 HAADF TEM of the interface before and after annealing. ................................ ............. 139 5 8 HRTEM and FFTs of the Ni/AlGaN interface before and after annealing. ..................... 140 5 9 HAADF STEM and EELS of the Ni/Al GaN interface before and after annealing. ........ 141 5 10 An EDS linescan of the Ni/AlGaN interface before and after annealing ....................... 142 5 1 1 Stepped stressing of a device to voltages well in excess of V CRIT ................................ .. 143 5 12 Percentage of gate contact area consumed by under gate defects. ................................ .. 144


10 LIST OF ABB REVIATIONS AlGaN/GaN Aluminum Gallium Nitride / Gallium Nitride HEMT High Electron Mobility Transistor RF Radio Frequency RC Resistance Capacitance BJT Bipolar Junction Transistor MOSFET Metal Oxide Semiconductor Field Effect Transistor HBT Heterojunct ion Bipolar Transistor TEM Transmission Electron Microscopy SEM Scanning Electron Microscopy UHR Ultra High Resolution TLD Through Lens Detector FIB Focused Ion Beam Miller DF Dark Field BF Bright Field HAADF High Angle Annular Dark Field EDS Energy Disper sive X Ray Spectroscopy EELS Electron Energy Loss Spectroscopy STEM Scanning Tunneling Electron Microscopy SPM Surface Probe Microscopy V DS Source to Drain Voltage V GS Source to Gate Voltage I G Gate Current V CRIT The critical voltage at which current throu gh the gate increases


11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy STUDIES ON THE RELIABILITY OF NI GATE ALUMINUM GALLIUM NITRIDE / GALLIUM NITRIDE HIGH ELECTRON M OBILITY TRANSISTORS USING CHEM IC AL DEPROCESSING By Patrick Guzek Whiting December 2013 Chair: Kevin S. Jones Major: Materials Science and Engineering Aluminum Gallium Nitride / Gallium Nitride High Elec tron Mobility Transistors are becoming the technology of choice for applications where hundreds of volts need to be applied in a circuit at frequencies in the hundreds of gigahertz, such as microwave communications. However, because these devices are very new, their reliability in the field is not well understood, partly because of the stochastic nature of the defects which form as a result of their operation. Many analytical techniques are not well suited to the analysis of these defects because they sam ple regions of the device which are either too small or too large for accurate observation. The use of chemical deprocessing in addition to surface sensitive analysis techniques such as Scanning Electron Microscopy and Scanning Probe Microscopy can be uti lized in the analysis of defect formation in devices formed with nickel gates. Hydrofluoric acid is used to etch the passivation nitride which covers the semiconducting layer of the transistor. A metal etch utilizing FeCN/KI is used to etch the ohmic and gate contacts of the device and a long exposure in various solvent solutions is used to remove organic contaminants, exposing the surface of the semiconducting layer for analysis.


12 Deprocessing was used in conjunction with a variety of metrology technique s to analyze three different defects. One of these defects is a nanoscale crack which emanates from metal inclusions formed during alloying of the ohmic contacts of the device prior to use in the field, could impact the yield of production level manufactu ring of these devices. This defect also appears to grow, in some cases, during electrostatic stressing. Another defect, a native oxide at the surface of the semiconducting layer which appears to react in the presence of an electric field, has not been ob served before during post mortem analysis of degraded devices. It could play a major part in the degredation of the gate contact during high field, off mode electrostatic stressing and could be the initiator of the pitting of the semiconducting layer of t he gate contact, a defect which was also observed.


13 CHAPTER 1 INTRODUCTION Aluminum Gallium Nitride / Gallium Nitride High Electron Mobility Transistors (AlGaN/GaN HEMTs) represent an emerging technology poised to greatly expand the application of power electronics, and specifically Monolithically Integrated Microwave Circuit technologies (MIMCs), in the field of Resonant Frequency (RF) and power electronics. Because of their status as an emerging technology in a manufacturing setting, demonstrating lon g term reliability of these devices remains a technical challenge. In this chapter, an overview of power and microwave transistor technologies will be presented. It will be followed by a section explaining, in some detail, the process flow associated with the fabrication of an AlGaN/GaN HEMT, with an exposotory figure detailing the fabrication of the HEMTs to be used in this study. A section will be devoted to an overview of the device operation and a final section will be devoted to materials effect s which influence operation. 1.1 Power Transistor Technologies As wireless technologies become more and more ubiquitous, demand will increase for semiconductor devices which can be used in RF electronics applications. For a transistor technology to be us eful in RF electronics, it must possess two separate figures of merit which compare favorably to other technologies. Firstly, a transistor used in RF appplications must be capable of achieving high power densities. The power density is related to maximum voltage current product achievable by the transistor. This power density is heavily influenced by the Breakdown Voltage, which is the voltage at which the transistor becomes incapable of carrying current while maintaining its ability to shut off current passing through the device. The Breakdown Voltage is, itself, heavily influenced by the difference in energy between the position


14 of the conduction band and the position of the valence band in the semiconducting material used in the fabrication of the tra nsistor. Larger bandgaps lead to an increased dielectric constant for a given material, which reduces the electric field required to sustain a given voltage applied to the device's electrodes and, correspondingly, increases the maximum achievable voltage before dielectric breakdown occurs. Larger energy gaps between the conduction and valence bands of a semiconducting material also increase the voltage required to induce thermionic effects such as avalanche and gate leakage in a semiconductor devices [1]. Secondly, a transistor technology must be capable of reaching switching speeds which allow it to operate in the frequency bands useful for RF communications. The maximum frequency a transistor is capable of achieving is measured by considering the swit ching speed at which the gain of the transistor reaches a value of unity, also known as the threshold frequency [2]. The performance of a transistor at this frequency is dictated by the device architecture being used as well as a variety of other physical characteristics. The fundamental physical limit to a transistor's operating frequency for a given architecture is determined by the mobility and, with it, the saturation velocity of free carriers within the semiconducting current carrying layers of the d evice. Higher free carrier mobilities and saturation velocities impart faster switching speeds. This is caused by an increase in the material conductivity and reductions in the associated critical Resistance Capacitance (RC) time constants associated wit h operation. GaN enjoys a substantial saturation velocity (25x10 6 cm/s) in comparison to a variety of other elemental and compound semiconductors (9x10 6 cm/s for Si, 9x10 6 cm/s for InP and 6x10 6 cm/s for GaAs) [1]. The implication of this is that, even though the electron mobility in the current conducting region of a typical device is comparable to the mobility other material systems, AlGaN/GaN HEMTs are still capable of very high operating frequencies [3]. GaN and


15 AlGaN are very wide bandgap semicondu ctors, making them more akin to a semi insulating material systems. The substantial bandgap observed in GaN and AlGaN is due to their crystal structure. When in thermal equilibrium, both materials form in a Wurtzite crystal structure characterized by alt ernating layers of Gallium (and Aluminum, when present) as well as Nitrogen. This forms a highly polar crystal structure where electrons are tightly bonded to their parent atoms [4]. The result are materials which are higly unreactive [5], with very larg e bandgaps [6]. Pure GaN has a bandgap of 3.1eV, while AlGaN has a bandgap which varies depending on stoichiometry from 3.1eV for GaN to 6.2eV for AlN. Because these materials have substantial bandgaps, they possess few intrinsic carriers, even at high t ermperatures, making them ideal for high temperature applications [7]. Two seaparate transistor architectures are commonly adopted for high frequency and high power applications. Bipolar Junction Transistors used in high power applications are three term inal devices in the "NPN" configuration. The emitter and collector of the BJT are doped with electron donors in order to create two separate electron rich semiconductor regions. The collector and emitter are separated by a thin region of electron accepto r doped, hole rich semiconductor material, called the Base. Because the emitter is doped with more electron donors than the collector, a diffusion current exists between the Emitter and the Collector. This current can be modulated by modulating the difff erential electron concentration between the emitter and collector. Larger differentials result in larger currents. The current in the transistor may also be modulated by modulating the current into the base and, as a result the hole concentration within t he base [8]. Field driven power devices function by forming a conducting channel between two electrodes: the Source (which supplies electrons) and Drain (which collects electrons). The resistance of the conducting channel between the Source and Drain of the device is modulated


16 electrostatically by the gate electrode, which is, ideally, electrically isolated from the channel. In the case of MOS FETs used in digital logic applications this electrical isolation is achieved through the deposition or growth o f an insulating layer; often SiO 2 or some metal oxide in silicon based devices. In the case of HEMTs, which are always formed from compound semiconductors grown via epitaxy, this isolation is achieved by forming a schottky contact [9]. Current can be mod ulated in a field driven transistor by varying the voltages on the source and drain electrodes because the conducting channel formed between the two electrodes acts as a resistor. Current can also be modulated by modulating the voltage on the gate to indu ce image charge in the conducting channel. Most conventional field driven devices utilize a conducting channel which is hole rich, inducing free carrier depletion between the source and the channel. As a result, a voltage must be applied to the gate in order to induce inversion of the channel region to an electron rich, conducting state. Because a voltage must be applied to the gate to induce a current, these are called Enhancement Mode devices. Some semiconductors can't be doped to be hole rich, howe ver. So, no free carrier depletion occurs between the Source and the channel. Field driven devices formed from these systems must be inverted to a hole rich state by the Gate in order to reduce current flow, making them Depletion Mode devices [10]. As s hown in Figure 1 2, modern power transistor technologies cover a wide range of power and frequency applications. Silicon has been leveraged to great effect in applications where high frequency operation is required. Alloying with germanium to form silico n germanium based Heterojunction Bipolar Transistors (HBT) has extended the frequency range accessable by silicon technology, but at the expense of breakdown voltage. This trade off


17 between stability at high input voltages and high frequency operation is a general trend in high power electronics. Indium phosphide (InP) and galium arsenide (GaAs) are compound semiconductors which posses larger bandgaps and higher electron mobilities than silicon [11,12]. As a result, HBTs formed from these compounds gen erally demonstrate larger Breakdown Voltages and higher operational frequencies than HBTs formed from silicon or silicon germanium alloys. However, the demand exists for devices with ever higher Breakdown Voltages [13]. The large bandgap of GaN and AlGaN has limited their implementation in transistor technologies. The materials suffer from very low thermal conductivities, which makes heat conduction in AlGaN/GaN transistor technologies a significant limiting factor in their operation. Additonally, form ation of hole rich AlGaN and GaN layers has been difficult due to a lack of a useful electron acceptor to be used as a dopant [14]. Because of this, field effect devices such as HEMTs must be used in AlGaN/GaN systems rather than bipolar technologies such as BJTs or HBTs, which are faster than HEMTs by virtue of using a much smaller pathway for conduction. Because of their large bandgaps, HEMTs are capable of sustaining very large electric fields without breaking down, making them very attractive for high power applications. Improvements in material quality and device architectures are driving these devices towards higher threshold frequencies, as well. 1.2 AlGaN/GaN HEMT Processing Epitaxy is required for the production of most GaN based devices. The pr oduction of an AlGaN/GaN HEMT begins with the surface preparation of some substrate for epitaxial growth of GaN on its surface. In general, this seed crystal needs to be electrically insulating to prevent leakage through the bottom of the device but therm ally conducting in order to mitigate self heating of the HEMT. Seed crystals must also posess an atomic density and spacing similar to


18 that present in the GaN or AlGaN basal plane of the deposited semiconductor in order to reduce defects associated with l attice mismatches. Three separate materials are typically employed for this growth. 6H Silicon Carbide (SiC) [15], Sapphire [16] and (111) Silicon [17] are the most common substrates used for these reasons. Sapphire was the first substrate used, because of its hexagonal crystal structure and relatively high availability, but was supplanted by SiC when thermal conductivity became a major factor in the operation of these devices. The following procedure for the creation of an AlGaN HEMT is outlined and il lustrated in Figure 1 2. This figure highlights key components of the HEMT and also includes the processing conditions used in the fabrication of HEMTs which will be used in this study [18]. The initial growth of GaN in modern technologies is generally p erformed on the surface treated substrate via Metallorganic Chemical Vapor Deposition (MOCVD). On SiC substrates, a nucleation layer of Aluminum Nitride (AlN) is employed to aid in epitaxial growth and to reduce conduction through the substrate. A buffer layer of GaN is grown on the substrate via Metal Organic Chemical Vapor Deposition (MOCVD) to a thickness of, generally, a few microns. This buffer layer is grown to a substantial thickness in order to trap defects and strain fields formed due to lattice and thermal expansion coefficient mismatches between the GaN and the SiC substrate. This layer is often intentionally doped with iron because it acts as a deep level electron donor making it a source of n type doping [19]. After the GaN buffer layer i s grown, an additional layer of AlGaN, with a 28% Al concentration in the case of this research is grown via MOCVD which is capable of providing a lattice matched layer with minimal epitaxial stress. The thickness of this layer is chosen to reduce tunne ling current through the AlGaN to acceptable levels while maximizing transconductance. It is made as thin as possible in this case, approximately 25 nm [18]


19 Even with the benefits of gettering and epitaxial growth, GaN/AlGaN layers are still rich in t hreading dislocations, which are present in state of the art material at concentrations of approximately 5x10 9 /cm 2 [20]. Both of these layers are left only unintentionally doped in order to reduce the effects of impurity scattering on mobility. Because o f the highly polar nature of their bonding, AlGaN and GaN are materials which spontaneously form a constant electric field within their matrices and, correspondingly, a spontaneous polarization induced charge at their interfaces. This field is further mod ified by the epitaxial strain on each layer due to the piezoelectric effect. When AlGaN is grown on top of GaN, the differential electric field results in a charged layer which is generally a few nanometers thick at the interface of the two materials. Th is layer of charge is called the Two Dimensional Electron Gas (2DEG) because of this tight confinement of charge along the c axis of the AlGaN/GaN crystal which manifests itself in conduction band bending below the Fermi Level [21]. The voltage at which t he transistor switches on and off can be modulated by modulating the thickness of the AlGaN layer. Af ter the deposition of AlGaN, a GaN based cap layer is grown onto the epitaxial stack to reduce its reactivity with the ambient Access to the 2DEG for ch arge conduction is achieved by the deposition of and patterning Ti/Al/Ni/Au contacts, which form the Source and Drain electrodes of the transistor. A Rapid Thermal Anneal is performed after deposition and patterning in order to improve charge conduction o ut of the semiconductor and through these contacts. Annealing is supposed to achieve this through a process of metal diffusion down threading dislocations through the AlGaN and into the GaN. This brings the Source and Drain into intimate contact with the 2DEG, improving the ohmic nature of conduction through the HEMT [22].


20 After annealing is performed on the ohmic contact, the gate contact is formed through a combination of photolitho graphic lift off patterning and metal d eposition. The metal used for c ontact to the AlGaN surface must posess a workfunction which makes it an effective rectifying contact in order to limit gate leakage. In this work, nickel is used as a 20 nm liner layer which forms the Schottky contact and is capped with gold. After the deposition of the gate metal, a capping layer formed from a dielectric oxide or nitride is deposited to reduce dangling bonds at the surface of the epitaxial stack, which tend to reduce the conductivity of the electron gas. At this point, the transistor i s ready to undergo the processing for whatever interconnect technology and packaging is required for utilization. AlGaN/GaN HEMTs are used as discrete packaged devices and, more frequently, in Monolithic Microwave Integrated Circuit (MMIC) designs, where multiple metal interconnect layers are utilized. 1.3 AlGaN/GaN HEMT Operation AlGaN/GaN HEMTs consist of a conducting two dimensional sheet of electrons (the 2DEG) and a gate electrode which is isolated from the 2DEG and modulates its resistivity via elec trostatics. The gate electrode is isolated from the 2DEG by the rectifying contact formed between the AlGaN and the gate. This rectifying contact is a schottky diode formed between the metal of the gate contact and the AlGaN semiconductor. When these t wo materials are brought into intimate electrical contact, they must necessarily share the same fermi energy. However, they must also share the same vacuum level. To satisfy both of these requirements, the valence and conduction bands of the semiconducto r must bend near the interface between the metal and the semiconductor. The direction in which the semiconductor's bands bend is dependant upon where the fermi levels of the semiconductor and metal sit in relation to the vacuum level. The difference in e nergy


21 between the fermi level of the metal or of the semiconductor and the vacuum level is called that material's workfunction. If the workfunction of the metal is larger than the workfunction of the semiconductor, the conduction and valence bands will bend up to allow the two materials to equilibrate when they are joined. If the workfunction of the metal is smaller than the workfunction of the semiconductor, the conduction and valence bands will bend down to allow the two materials to equilibrate. Thi s can be represented by the built in voltage, V BI (in volts), which represents the turn on voltage for the schottky diode and may be expressed as the difference between the metal M S (in eV), divided by a fundamental electronic charge [23]. Epi taxial AlGaN used in HEMT gate stacks is generally only unintentionally doped, which makes the difference between the conduction band energy and the fermi energy roughly equal to half the distance between the valence and conduction bands. In most cases, intrinsic AlGaN is slightly n type in the as deposited state. In n type materials, a rectifying contact will induce depletion at the surface between the metal and the semiconductor, resulting in hole accumulation near the interface. This occurs when th e metal workfunction is larger than that of the semiconductor and the conduction and valence bands bend down. Given that depletion occurs, it is not surprising that the current equation for a gate contact is similar to that for a pn junction diode. The current through the diode, I G (in A), is as follows, where A is the contact area (in cm 2 ), J RG is the recombination generation current density and A* is Richardson's Constant (in A/cm 2 K 2 ). The difference in the metal and semiconductor workfunctions may be expressed as a built in voltage, V BI equal to the difference beween the metal workfunction and the semiconductor workfunction divided by the charge of a single


22 electron. V G is the potential on the gate while V 2DEG is the potential applied to the 2DEG (in volts), which may vary along the length of the gate electrode [ 24 ]. (1 1) As shown in Figure 1 5, The gate contact electrostatically modulates the resistivity of the 2DEG which is situated directly below the gate contact and is, gener ally, well isolated by the AlGaN from the 2DEG as long as the gate is not forward biased. In this way, the conductive channel of the gate may be turned on and off by placing a potential on the gate electrode, inducing image charges within the 2DEG which i ncrease or decrease the carrier concentration. The threshold voltage of this device is induced by charges accumulated at the interfaces between the GaN and AlGaN as well as the AlGaN and the gate metal. It can be expressed as a sum of potentials generate PZ is the spontaneous polarization and inverse piezoelectric effect induced charge at the AlGaN/GaN interface (in cm 2 ), t AlGaN AlGaN is the permittivit y of the C is the conduction band offset (in eV) between the AlGaN and GaN. The assumption is made that the AlGaN layer is undoped [25]. (1 2) If V GS doesn't exceed the threshold voltage, the section of the 2DEG under the ga te electrode will be depleted of charge carriers, and the 2DEG will, ideally, not pass any current regardless of the potential applied to the drain. This is known as "off" mode. As the potential on the gate exceeds the threshold voltage, the transistor s witches into the linear regime, or "semi on" mode, as it is described by some authors. In this regime, the current flow through the device increases roughly linearly with increasing potential on the drain. The extent of this linear regime is defined by t he following condition, where V GS represnts the gate to source voltage and V TH


23 represents the threshold voltage of the device, and V DS represents the drain to source voltage [26]. (1 3) If the above condition is not met, the transist or enters the saturation regime, where additional voltage does not result in any additional current flow. This occurs because the part of the channel close to the drain side of the device remains depleted, or "pinched off" and the transistor self limits. In order to calculate the current flowing through the channel of the AlGaN/GaN HEMT, the elecrostatics of the device must be considered as the gate voltage is raised above V TH and the transistor is switched out of "off" mode in order to form a conducti ng channel. The current density within the device at any point along the channel is equal to the conductivity of the channel multiplied by the lateral electric field. The conductivity of the channel is dependant upon such variables as the fundamental ch carrier concentration The carier concentration is derived from the capacitance of the AlGaN AlGaN divided by t AlGaN ) as well as the potential difference between V GS a nd V TH The electric field is equal to the change in V 2DEG with respect to position along the channel as a consequence of potential applied to the drain electrode of the device. In order to determine the current passing through the full device, rather th an through a single infinitessimal slice thereof, this current density equation is integrated over the full width, W (in cm), and full length, L (in cm), of the device. This yields the relationship between the current flowing from the from the source elec trode to the drain electrode and the voltages applied to the various electrodes present within the HEMT, but only for the linear regime of operation


24 (i.e where the channel has not yet been "pinched off". The constant saturation current may be calculated b y applying the constraint of the condition listed in Equation 1 3 to Equation 1 4 [27]. ( 1 4) 1.4 Factors Affecting AlGaN/GaN HEMT Electrical Characteristics As is evidenced by the previous section, the conduction of charge carriers throug h an AlGaN /GaN HEMT is influenced by a variety of factors, some of which are dependant upon geometry (such as the length and width of the HEMT, the thickness of the AlGaN layer, etc.). However, a variety of materials parameters also affect ideal device o peration. Their variability both during processing and after processing, in the field, are of prime importance. It should also be noted that geometry also influences the characteristics of the HEMT, especially at short length scales, where short channel effects such as velocity saturation and hot electrons cause the device to perform differently than what would be expected given the equations outlined in the above section. To a first order, however, the equations above and their dependance upon given mat erials constants is accurate. Saturation current is affected by several material characteristics. The dielectric constant of the AlGaN, which is dependant upon the stoichiometric composition of the epitaxial layer, has a direct influence upon the curre nt observed passing through the device. Likewise, the carrier mobility within the 2DEG also directly influences the magnitude of saturation current observed passing through the channel. A variety of factors can influence the carrier mobility in the 2DE G, the most commonly encountered being scattering by impurities or trap centers. Trap centers can arise from a variety of sources in an AlGaN/GaN HEMT. Dangling bonds found at unpassivated or imperfectly passivated surfaces (such as the top surface of th e AlGaN) can form deep level traps which act as scattering centers which reduce the mobility of carriers in the 2DEG. This phenomenon is of


25 particular importance in the field of device reliability, where stressing involving high fields (and hot electrons) can greatly impact the concentration of trap centers present on passivated and unpassivated surfaces alike [28]. A variety of passivation layers have been explored for use in reducing surface states in AlGaN/GaN HEMTs, from SiN [29], to Al 2 O 3 [30], to Sc 2 O 3 [31]. Threading dislocations and misfit dislocations, which are a nearly unavoidable consequence of epitaxial growth of GaN beyond its pseudomorphic thickness, also act as scattering and non radiative recombination centers which influence carrier mobili ty and minority carrier lifetimes [32]. Even the iron doping used to improve the quality of MOCVD grown GaN can have a deleterious effect on the carrier dynamics associated with device operation because they act as deep level electron acceptor [33]. It bears mentioning that these ionized scattering centers can also compensate the charge levels within the 2DEG of a HEMT, further reducing the conductivity of the channel region. This effect is often observed as an increase in the "access" resistances asso ciated with the source drain contacts of the device. These access resistances are not captured in the physics of operation described in the previous section, but they can be considered as connected in series with the transisting element of the HEMT which was described in the previous section. Increases in access resistance also manifest themselves as a reduction in the observed saturation current in the HEMT device, despite the fact that their effect is not directly captured by the current equations which were outlined previously. The ideality of the Schottky diode utilized in the gate contact of the HEMT is also of critical importance to its proper operation. The ability of the contact to sustain low reverse biased currents, even at high applied fields, ensures that leakage from the 2DEG of the device and the resulting power dissipation is minimized in "off" mode. As before, trap centers present


26 at surfaces as well as within the bulk of the AlGaN play a large role in the quality of this contact in rever se biased mode. These trap centers can induce recombination generation current, which increases the current observed in the diode in the reverse biased mode of operation. These trap centers can also induce fermi level pinning in the device, which influen ces the observed energetic barrier to current conduction, and associated turn on voltage of the gate contact [26]. Reactions between the gate contact and the underlying AlGaN and GaN can also generate parallel paths of conduction from the 2DEG and the gat e electrode, effectively shorting out the schottky contact and greatly increasing the observed leakage current in the device. The reliability of AlGaN/GaN HEMTs studied as part of this work is influenced by all of these materials factors. Degradation of the electrical properties of the device can be linked back to changes in the structure of the device and associated changes in the materials constants which influence the ideal behavior of the HEMT. It will be the purpose of this work to outline the effec ts of structural abnormalities in HEMTs, observed with various microscopy techniques both before and after electrical stressing, on the electrical properties of these devices.


27 A B Figure 1 1. Schematics of two different power device technologies. A) A simple operational schematic describing the structure of a BJT, which operates on diffusion current, is shown. B) The structure of a High Electron Mobility Transistor, which is a field effect device, is shown. Figure 1 2. A performance diagram fo r a variety of power transistor families.Performance is gauged in terms of breakdown voltage and operational frequency. Included are Si and SiGe BJTs and HBTs, InP HBTs and HEMTs, GaAs HBTs and AlGaN/GaN HEMTs. The tendency towards reduced breakdown volt age at higher frequency is apparent, as is the trend of increasing breakdown voltage with increasing band gap [1].


28 Figure 1 3 A process flow for the creation of an AlGaN/GaN HEMT [19].


29 A B Figure 1 4 The electrical behavior of a schottky gate in fo rward and reversed bias. A) A diagram showing the band bending which occurs both at the instant when a metal and n type semiconductor are joined and when the fermi level (equal in this case to the mid gap energy) has a time to equilibrate. B) A diagram of the forward biased (positive voltage) and reverse biased (negative voltage) currents with respect to the potential applied to the gate.


30 Figure 1 5 A schematic of the operational modes of a transistor. When the potential supplied to the gate is smal ler than the threshold voltage, V TH the 2DEG is depleted and the transistor is in "Off" mode. As the gate voltage exceeds the threshold voltage, the transistor can either enter the linear or saturation regimes, depending upon if V DS is greater than the d ifference between V GS and V TH If V DS is larger than V GS V TH the channel pinches off near the drain side of the device and the transistor saturates.


31 CHAPTER 2 EXPERIMENTAL METHODS A variety of techniques are used in analysis of semiconductor device reliability. The techniques utilized in reliability studies cover a wide spectrum of electrical and structural materials properties. Metrology techniques can also be used to identify processing related defects within a device which do not directly correl ate to materials properties as well as to identify poor device operation (i.e. poor electrial behavior). This section of the dissertation will focus on the analytical methods used to characterize the AlGaN/GaN HEMTs used in these studies. It will begin with a discussion of two of the more commonly utilized techniques in the literature. The first of these are the electrical measurements made before, after and (sometimes) during electrical stressing. These electrical measurements are the first step in t he detection of defects induced by normal device operation and are generally used in conjunction with microscopic and spectroscopic techniques to correlate changes in electrical properties with changes in structural properties. Arguably the most popular t echnique, Transmission Electron Microscopy (TEM), will be discussed after the section on electrical measurements. As will be shown, this technique is of limited use in analysis of the stochastic processes which lead to electrical defect formation in these HEMTs. These sections will be followed by a discussion of the chemical deprocessing scheme developed as part of this work and its utility in analyzing stochastic processes such as defect formation in the HEMTs used in this study. Some special attention will be paid to variations in the deprocessing scheme and their effects on sample quality. This section will be followed by a discussion of Scanning Electron Microscopy (SEM) and Atomic Force Microscopy (AFM). These techniques proved to be of great utili ty in the characterization of the surface of the AlGaN


32 epitaxial layers after deprocessing. Special attention will be paid to the settings used in both of these technique to provide high quality imaging of the AlGaN surface. 2.1 Electrical Measurements E lectrical stressing and electrical measurements of AlGaN/GaN HEMTs were performed on the same instrumentation, as is commonly the case in reliability studies because of the convenience of not having to switch the sample from one system to another to make m easurements after stressing. This is especially true in cases where stepped stressing experiments are called for, where having to physically move the sample for each stressing step would greatly impede throughput. The system used for electrical analysis in all studies described in this work was an HP4146 parameter analyzer mated to a Tectronix 370A curve tracer. These systems are capable of multiple parallel measurements of various parameters both before, during, and after stressing. Generally, analysis of a device under stress begins and ends with a cursory examination of its electrical characteristics. This examination begins with an analysis of the characteristics of the gate contact, where the gate to source current (I GS ) is plotted against the gate to source voltage (V GS ). Generally, this measurement is carried out from V GS equal to 1V (the schottky diode being "on" in this case) out to V GS equal to some significant negative voltage (where the diode is "off"), in 0.1 V increments. The ability of t he gate to turn the 2DEG channel "on" and "off" (i.e. its transconductance) is also measured via a plot of the drain to source current (I DS ) versus the gate voltage (V GS ). This analysis generally occurs over the same voltage range as the measurement of th e gate current and with 0.1 V increments on V GS as described previously. The behavior of the device in saturation is also of critical importance to understanding the overall quality and degradation of a device before and after stressing. To this end, the traditional "family of curves" plot common to most FETs is employed. This measurement involves analysis


33 of I DS as the voltage between the source and drain (V DS ) is swept from a value of 0 V to some value which induces saturation in the device (usually around 4 V) in 0.1V increments. Generally, V GS is also stepped in order to generate a family of curves rather than a single curve and to observe the increase in V DS required to induce saturation in the device as V GS is increased. A typical range of valu es for V GS to form a family of curves is 3 V to 1 V stepped in 1 V increments. Stressing of devices under analysis for this study took place in an air ambient under standard temperature and pressure. Stressing can involve maintenance of a device at a single voltage or in a "stepped" configuration where the voltage is incremented at a specified rate over a given period of time. Currents flowing through any terminal of the HP4146C cannot exceed a level of 1 00 mA without causing damage to the testing system. Because of this, devices which were electrostatically stressed in this study were stressed in off mode, where the channel of each device was pinched off with a large negative voltage on the gate electrode. This limits the leakage current flowing through the device to a level which does not induce compliance within the testing system. Depending on the specifics of the study performed one voltage (either V DS or V GS ) would be held constant as another voltage (again, either V DS or V GS ) was stepped with an increment of 1 V/min. Generally, this stepped stressing continued until some maximum voltage was achieved or until the device under stress reached one of the preset current compliances applied to all electrodes. Stressing ended on devices which were stressed as part of this study when the gate contact reached compliance, which varied from study to study, ranging from 100 A to 1 mA on


34 the gate contact of the device. The gate electrode was, consistently, the first electrode to achieve the current dictated by the set compliance. During stressing, devices were subjected to a set of abbreviated measurements after each voltage increment. This measurement, generally, occurred over a 1min time period, resulting in a 50% duty cycle for stressing. These abbreviated measurements consisted of analysis of I GS at 1V V GS in order to determine the magnitude of leakage current through the gate electrode. The respective currents through both the gate electrode and the source/drains were recorded during stressi ng in order to determine leakage current in the high field case as well. It is important to point out that electrical measurements form the baseline for all studies into the reliability of the devices studied as part of this work. In order to successfull y "diagnose" and remedy degradation and failure of devices, structural changes must be linked to changes in electrical properties. Without electrical measurements which demonstrate that the overall quality of the device under test changed as a result of stressing, observations of physical abnormalities in the structure of a stressed device can only yield educated guesses as to how the physics and chemistry of degradation in the field adversely effects device characteristics. 2.2 Transmission Electron Micr oscopy Transmission Electron Microscopy (TEM) functions by forming a coherent source of electrons and accelerating these electrons through a large potential difference, through an electron transparent layer of sample material and into some electron sensit ive system which acts as a sensor. The image which is collected by this sensor can yield detailed qualitative and quantitative data about the representative sample material. This data can range from information about the density and structure of the ma terial under observation, to chemical information detailing atomic ratios and (in some cases) atomic bonding, and even to electric fields present within the sample (via techniques such as electron holography) [34]. Given the wealth of data


35 available from this technique, it should not be surprising that TEM represents one of the more oft employed analytical techniques in the field of semiconductor reliability. The two TEMs utilized in this study were a JEOL 2010F and a JEOL ARM200F. A basic TEM consists of several different modules which work to generate the electron beam, direct that beam through a cross sectioned sample, and collect the electrons which were transmitted through to the other side of that sample. The first of these modules is the electron source, which is used to extract the electron current used for imaging from a filamentary source by means of thermionic or field emission. Sources vary from TEM to TEM, but the most high performance of these sources is a field emission gun (FEG) source w hich can be used to extract electrons with minimal energetic variation and with a smaller starting probe size, making the entire microscope less prone to chromatic lens aberration. The extracted electrons exist at high potential and are accelerated throug h the microscope column, and towards lower potentials, as a result [35]. After the electrons are accelerated out of the source, they encounter a pair of electromagnetic quadrupole or hexapole lenses, known as the condenser lenses, which are used in conju nction with an aperture just above them in the column to form a coherent beam from the electron source image. In Bright Field TEM, the condenser lenses are adjusted to form a parallel beam to uniformly illuminate the sample. In the case of Scanning Trans mission Electron Microscopy (STEM), the lenses are adjusted so that the electron beam is convergent on the sample, rather than parallel to the optic axis [36]. In some modern systems, such as the JEOL ARM200F, additional lensing is installed in order to c orrect for spherical aberration in the beam optics [37]. After passing through these additional lens elements, the electrons encounter the cross sectioned specimen. At this point, the


36 electrons may either pass through without interference, scatter off of the material which comprises the sample itself, or they may be absorbed. The component of the incident beam which is scattered by atoms within the sample can scatter in two very different ways. If the region of the cross sectioned sample which the elect rons passed through is amorphous, there is no preferential direction associated with scattering, and the diffraction pattern formed assumes the appearance of a diffuse "cloud" (with a radius approximately equal to the nearest neighbor distance of the amorp hous material) if it is imaged directly. In the case of a crystalline sample, electron diffraction occurs and the diffraction pattern occupies very specific positions, when imaged directly. If the beam passes through a region with many different crystal directions represented, the resulting diffracted pattern will form rings rather than the singular spots associated with diffraction in a single crystal. After the electrons pass through the sample, they encounter the objective lens and, below it, its asso ciated aperture. The lens collects electrons which comprised the parallel beam incident to the sample and focuses this beam on its back focal plane. It is easy to observe from any ray diagram of the system described above that, because the electron beam incident to the sample is parallel to the optic axis, every radial position of the objective lens collects diffraction information about the entire image of the sample. The objective aperture is used to screen out information which enters the objective l ens at high angles, improving contrast and increasing the depth of field of the instrument. The intermediate aperture is placed after the objective lens and is used to collect only certain diffraction information about the specimen. If the aperture is pl aced on axis, the component of the electron beam which passes through the aperture transmits information only


37 about the mass thickness variation data contained by the direct beam. This is called bright field imaging If the aperture is placed off axis (or the beam tilted so that the axis shifts), the transmitted beam contains diffraction information relating to electrons scattered only in the direction corresponding to the position of the opening in the intermediate aperture. This is called dark field ima ging, Both bright field and dark field imaging were utilized in this work. In order to collect the image formed by the transmission of electrons through the transparent sample, additional electromagnetic optics, in the form of the intermediate lens, are required in order to form an image on the projector screen, film, or CCD used to capture the image. This is accomplished by adjusting the power of the intermediate lens such that it is focused on the image plane of the objective lens. This results in the formation of either a bright field or dark field image on the projection plane of the microscope, depending upon the orientation of the intermediate aperture. If diffraction data is desired, this lens is weakened until it is focused on the back focal pla ne of the objective lens, rather than the image plane. In the case of diffraction, the relationship between where the characteristic bright spots (from a single crystal) or rings (from a polycrystalline sample) form in the final projected image of the dif fraction data. These positions are governed by the camera length equation, wherein the distance between the projection screen and the sample under analysis is defined by the variable L, the distance between diffracting crystal planes is defined by the v ariable d, the deBroglie wavelength spot to the diffracted spot of interest is defined by the variable r [38]. ( 2 1 )


38 The deBroglie wavelen gth of the electron is defined by the equation below, where h represents the Planck constant, V is the accelerating voltage, m is the rest mass of the electron and q is the fundamental charge of an electron [39]. (2 2 ) Thus, d iffraction spots which are formed from small interplanar spacings have large displacements from the direct beam of the electron signal while diffractions spots which are formed from large interplanar spacing have small displacements from the direct beam. It is important to note that, while this equation governs the position of scattered electrons in selected area diffraction, it also provides the same information which is necessary in generating a dark field image using a given diffraction bright spot. It also has utility in the convergent beam imaging which is utilized as part of Scanning Transmission Electron Microscopy (STEM), which wil be detailed in the following part of this section. STEM is of particular utility in situations where high resolution imaging is required. These are generally situations where a feature must be imaged with dimensions approaching the deBroglie wavelength of the electrons acceleated from the source. STEM functions by increasing the power of the second condenser lens such that the beam crosses over, or converges, at the sample surface. This converged electron beams beam converts the direct spot and diffraction spots, typically generated during illumination of the sample with a parallel beam, into a group of cones correspo nding to each spot. The radii of the projected images formed from these cones follow from the convergence angle of the beam while their positioning relative to one another is determined by the camera length equation, described above. The image resulting from one of these resultant "cones" of information is called a Convergent Beam Electron Diffraction (CBED) image [40].


39 The high resolution image generated by the TEM arises from the interference between the direct spot CBED image and the CBED images which result from the diffracted electrons. A proper STEM image is formed by rastering the converged beam across the sample surface and by recording the overall intensity of the resultant image collected by the detector. Because this intensity is impacte d by the interference between the CBED images formed by diffraction of the convergent beam, it is not surprising that the image formed is dependant upon which CBED cones are collected by the detector in use and, by association, the type of detector which i s inserted into the beamline. Generally, STEM is performed with two different types of detectors. If the CBED image resulting from the direct beam is of primary importance, a centered CCD array is utilized in order to collect the transmitted CBED image and the interference pattern arising from the diffracting CBED images. The resulting information takes on the form of a bright field STEM image. If a dark field image is preferred, an annular detector is favored over the centered CCD array. This detect or collects the components CBED image displaced from the optical axis by a distance larger than the inner radius of the detector (i.e any CBED image formed by scattering to an angle beyond the minimum collection angle of the annular detector). The resulti ng interference patterns detected by the annular ring form a dark field STEM image. In general, high resolution images generated via TEM (such as STEM images) can be used to determine structural information about the material under analysis without the ne ed for an image of the the electron diffraction pattern generated by the incident beam. This is accomplished by means of a Fast Fourier Transform (FFT), which is a matrix operation that converts the data contained within the high resolution image into a m ap of the component


40 frequencies associated with the repeating patterns contained within that image. An FFT treats these repeating patterns as a sum of sinusoidal functions [41]. STEM is also favorable over traditional TEM imaging techniques in cases wher e quantitative microscopy is desired in additional to high resolution imaging. The small footprint of the rastered beam makes it ideal for analysis of features at the nanoscale utilizing spectroscopic techniques such as X Ray Energy Dispersive Spectroscop y (EDS) [42] or Electron Energy Loss Spectroscopy (EELS) [43]. While there are a variety of other techniques commonly used in conjunction with STEM (such as Cathodoluminescence [44] and Electron Holography, to name a couple more), EDS and EELS will be the techniques favored as part of this work. As electrons enter the sample from the converged beam generated at the condenser lens, they may interact and scatter off of atoms present within this sample. This scattering is the result of inelastic collisions with the electron clouds of the sample's constituent atoms, where some of the energy of the incident electron is imparted to the electron shells of the atoms which it interacted with. In EELS, the energy lost by the electron in transit through the sample is determined through the use of an EELS detector placed in line with the optical axis of the microscope. As electrons enter the EELS detector, they encounter a static electric field which bends them in such a way that the electrons separate in the fie ld by energy before colliding with a CCD array. Electrons with low momentum and low kinetic energy are bent more readily than electrons with a greater momentum. Thus the variable intensity of the resultant electron beam with respect to position on the CC D array can be used to determine the proportional energy losses experienced by electrons being transmitted through a sample. The characteristic energy losses associated with transmission can be used to determine the chemistry of the material under


41 analysi s. Atomic ratios and, in some cases, even bonding can be determined from a high quality EELS spectrum. Also, because only the electrons which undergo minimal scattering are utilized in this technique, the EELS detector can be combined with the annular ri ng detector described above to yeild complementary EELS and dark field STEM imaging in tandem on a sample. When incident electrons interact with the electron shells of the constituent atoms of the sample and lose energy, this imparted energy can result in a variety of interactions within those constituent atoms which the electrons collided with. If an incident electron knocks a core electron free, a valence electron will lose energy and take the place of the lost core electron. The energy lost by this va lence electron exits the sample in the form of an x ray, which can be collected by a photodiode chilled by liquid nitrogen. The energy of this x ray can be determined by the number of photoelectrons which are collected by the diode as a result of stimulat ion by the incident photon. Not surprisingly, the energy of the incident photon is characteristic of the interorbital transition which the valence electron went through as it fell from one shell to another within the atom. If the incident photons collect ed by the photodiode are collated with respect to their energies, and EDS spectrum results and this EDS spectrum corresponds (albiet not directly due to matrixing effects which cannot be separated without a standard for comparison) to the relative ratios o f different atoms present within a sample. In this way, EDS can yield substantial quantitative information regarding the chemistry of a material under analysis. TEM is particularly effective when coupled with equipment which can extract a material sample from a specific site in a semiconductor device. This technique, called Focused Ion Beam (FIB) milling, utilizes a Scanning Electron Microscope with a secondary column of electron optics linked to a liquid gallium source which acts as a supply of gallium ions for site selective ablation. FIB based sample preparation makes it possible for an electron transparent sample to


42 be cut from a HEMT, mounted to a copper grid suitable for insertion into the electron column of a TEM, and thinned all in situ and wit hout breaking a vacuum [45]. This process is shown in Figure 2 2. The first step in this process is to deposit a protective layer, generally over 100nm of thermally evaporated carbon, onto the surface of the sample. This occurs prior to insertion into t he FIB. In samples where the structure itself is protective to the region of interest, this process can be skipped. The sample is then inserted into the FIB and a vacuum of 5x10 5 mmHg achieved by means of a turbo pump. The sample is set at eucentric he ight by means of SEM imaging and is then tilted 52 into the gallium beamline. After insertion of a platinum GIS needle, site selective deposition occurs over the area of interest as the gallium beam is rastered across a beam of organomettalic platinum mo lecules. This results in the deposition of a dense platinum layer, which acts as a protective mask in the subsequent milling of two trenches to either side of the site of interest with depths large enough to expose the region of interest under the deposit ed metal. The sample is thinned down by progressive milling steps to a thickness of approximately 1m and a cut is made which frees the bottom of the cross sectioned region of interest from the larger sample. A second cut is made which frees one of the s ides. At this point, a second attachment known as an omniprobe needle is inserted into the beamline of the chamber and the sample is tilted parallel with the electron beamline, putting it at 52 to the gallium beamline. For the purposes of in situ lift out, this needle must be hard (so that it can withstand the physical abuse of making solid contact with a surface without deforming plastically), but also thin (to facilitate bonding the cross section to the omniprobe needle). A tungsen needle thinned to a radius of curvature of 1m is perfect for this application. The needle


43 is brought into intimate contact with the cross sectioned area of interest and the Pt source is used, again, to weld the top of the cross sectioned sample to the omniprobe needle. At this time, the remaining attached side of the cross section is cut free and the cross section transported to the TEM grid. At this point, it is welded to the grid and the omniprobe needle is cut free. At this point, both the Pt GIS attachment and th e omniprobe attachment are retracted and they remain in this position. The mounted sample is tilted into the gallium beamline. The sample is now ready for final thinning. The final thinning steps begin with a group of successive milling steps which thin the sample down to a few hundred nanometers. At this point, the sample is often wedge cut, so as to taper the long side of the cross section, from one side to the other, with the intent of generating an ideal sample for analysis close to the tip of the w edge. The sample is then tilted off axis from the gallium beam by four degrees and a mill is performed at a low accelerating voltage (7 kV) to clean the surface of the sample prior to analysis with TEM. 2.3 AlGaN/GaN HEMT Maskset and Chemical Deprocessing It is important to note that, while TEM is a powerful analytical tool for site specific high resolution imaging of nanoscale features, it suffers from a tremendous disadvantage in the analysis of stochastic processes, which comprise the majority of defec ts observed in AlGaN/GaN HEMTs. This disadvantage is the small total volume of any sample generated for TEM. In order to form an electron transparent sample, a TEM cross sectioned must be thinned to a thickness of, at maximum, 100nm. In fact, some estim ates place the total volume of all TEM cross sections created between 1955 and 1970 as being less than one cubic millimeter [46]. For this reason, caution must be used whenever a study generated by TEM cross sectioning is being employed to understand the fundamental properties of a larger material set.


44 This has been found to be especially true for cross sectioning of samples formed from AlGaN/GaN HEMTs, where defect formation is highly stochastic in nature. The morphology of a defect can vary dramatic ally depending upon where a cross section is taken from a device. In many cases, a defect may be present in one area of a device, but not in another. In fact, some of the stochastic defects observed in this work only occur over a very small percentage (o ften less than 5%) of the total gate area and there have been many cases where a device demonstrated degradation electrically, but no TEM cross sections yielded a sample with an observable defect present. This null result, which opposes observed electrica l data, makes it difficult to derive a conclusion about the mechanism of defect formation from TEM data alone. Because TEM cross sectioning is a destructive process and because the defects formed under the gate during stressing of these devices cannot be directly imaged non destructively prior to cross sectioning, no means exists of determing the nature of a defect in a degraded HEMT prior to its being cross sectioned and all cross sectioning must be performed "blind", without any prior knowledge of whethe r or not a defect exists at all in a cross sectioned volume and also without any prior knowledge of what that defects morphology is if it is actually present in a given cross sectioned area. Because of this, TEM is not aptly suited for analyzing and qu antifying the stochastic nature of defect formation in AlGaN/GaN HEMTs and other analytical techniques are required in order to analyze the progression of defect formation quantitatively. The methodology utilized in this study was deprocessing of the AlGa N/GaN HEMT such that the passivation nitride and metal layers are removed from the device structure. This exposes the AlGaN epitaxial layer of the device to direct analysis using top down imaging techniques such as SEM and SPM, which will be described in greater detail later in this chapter. This deprocessing method has been


45 shown to be highly effective for use in analysis of defects formed at the surface of this AlGaN layer as well as for use in analysis of volumetric inclusions within the layer. The de processing strategy employed varies slightly with the device under analysis. With this in mind, it seems beneficial to discuss, briefly, the maskset used as part of this study. As shown in Figure 2 3, the maskset used to generate the HEMT structures cont ains a variety of structures, including a group of spaced ohmic contact pads used for Transmission Line Measurements (TLM), a lithographic alignment vernier, and five separate HEMT structures. Of these HEMT structures, three are of particular interest bec ause of their use in the studies detailed in this work. These are the device with a submicron scale "T gate" structure, a device with a 1m gatelength, and a device with a much larger, 60m gatelength (called the FatFET). All of these HEMT structures uti lized nickel as the gate metal of choice. Deprocessing begins with an exposure to hydrofluoric acid in the form of buffered oxide etch with a 6:1 stoichiometric ratio of HF to NH 3 F. HF etching occurs for a total of 15 minutes, during which time the PECVD silicon nitride which passivates the AlGaN surface of the HEMT is dissolved. It is expected that that the etch rate of the PECVD silicon ntride layer in BOE is, highly variable depending upon the processing conditions at which this nitride layer was depo sited (a common issue in wet etching of PECVD layers) as well as depending upon processing which the device was exposed to after the deposition of the passivation nitride. Variations in required etch time have even been observed between samples formed f rom SiC versus Si(111) substrates. The etch time required to clear the surface of the sample is also highly dependant upon the architecture of the device being deprocessed, with submicron "T gate" devices being observably more difficult to clear of all si licon nitride than devices formed with a gate length of 1m. The cause of this will be explained in more detail later in the section.


46 It bears noting that HF is not perfectly selective to silicon nitride over AlGaN. Surface features can develop on the A lGaN surface when a sample is over etched in BOE. As shown in Figure 2 5, the morphology of these surface features is highly dependant upon the substrate which the the GaN/AlGaN layers were grown on. Epitaxial layers grown on SiC will develop a wavy, "ti depool" morphology if they are significantly overetched. This morphology may follow the terracing of the underlying epitaxial layer. Epitaxial layers grown on Si(111) develop the more traditionally recognized etch pits when overetched in BOE. This morph ology results from BOE preferentially attacking the epitaxial material surrounding the threading dislocations which permeate the epitaxial layers of the device. This is a well documented phenomenon in silicon etching, where BOE has been observed to attack the high energy surfaces of dislocations [47]. Because of the inherent variability in silicon nitride etch rates, it is difficult to eliminate the surface features which develop as a result of this over etching. However, they can be minimized if the etch of the device under investigation is minimized. A 15 minute exposure in BOE generally accomplishes the task of etching the passivation layer without attacking the surface dramatically. With the passivation nitride removed from the surface of the HE MT and with the gate and ohmic contacts exposed as a result, the metal layers can removed to fully expose the surface of the AlGaN epitaxial layer. This is accomplished using a combination ferric cyanide (FeCN) and potassium iodide (KI) etch solution, kno wn as TFAC, which is commercially available from Transcene Company, Incorporated. TFAC is marketed as a gold etchant, specifically designed for use in the liftoff processing of metal layers on compound semiconductors, where an etch is required which can r emove deposited metal without attacking the exposed semiconductor layers present underneath the metal. The KI component of this etch solution actively attacks and


47 dissolves any metal, excepting platinum, while maintaining the quality of the AlGaN layer be low. No etch related defects have been observed in the AlGaN as a result of etching with TFAC, which compares favorably from previous results using Aqua Regia to remove the metal layers. The FeCN/KI mixture is received after purchase as a dry powder and must be dissolved in water to form a proper etch solution. This is accomplished by mixing a volume of 3.75L of water for every 225g of FeCN/KI and stirring at room temperature using a magnetic stir rod. The FeCN/KI mixture does not dissolve easiy in wate r and stirring may take up to 6hrs in order to fully dissolve the solid TFAC mixture. The rate of dissolution does seem to be be improved if the FeCN/KI solid mixture is ground down via a mortar and pestle prior to mixing with water, as dissolution appear s to be aided by smaller particle sizes and an associated larger surface area per volume of solid TFAC powder. Generally, the majority of the contact metal present within the channel of the device can be removed with an etch lasting 24hrs. In order to fu lly clear the ohmic contact regions of the device, a longer etch (sometimes as long as 96hrs), may be required. An etch which requires a substantially longer time period than this is generally indicative of incomplete removal of the silicon nitride passiv ation layer. An SiN layer which has not been completely removed will not interfere with the etching of the gate metal (though it may interfere with imaging of the region under the gate once the metal has been removed), but does interfere with the etching of the ohmic metal layers. In such an event, the SiN passivation layer must be re exposed to BOE and removed. Generally, 5 additional minutes of exposure in BOE will fully remove an SiN layer which was not completely eteched by the first 15 minute BOE ex posure.


48 After metal etching is completed, the two layers which obstruct the AlGaN epitaxial layer from direct viewing have been, ideally, removed. All that remains is final cleaning of the surface of the device. This consists of one or two steps, depe nding upon the device being analyzed. In the case of a micron scale device, the only additional deprocessing which is required is cleaning in a progression of organic solvents. This exposure is designed to eliminate scumming due to organic contamination on the surface of the sample. It is important to note that, while the methodology described here for eliminating organic scumming is effective, the best deprocessing results consistently come from devices which have been manipulated in a clean environment and with clean lab equipment. As shown in Figure 2 6 AlGaN surfaces which have been exposed to the BOE and TFAC solutions are, by no means, clean after etching. Organic scumming (seen here as the dark features which are scattered throughout the sample surface, is ubiquitous on most devices. This scumming is enhanced by dirty processing conditions, but it cannot be eliminated even by the cleanest handling procedures. The only solution to this scumming is exposure in organic solvents to loosen and diss olve the scumming films. It should be noted that the effectiveness of this descumming process is greatly enhanced by the use of ultrasonication as part of the cleaning regimen. All solvent exposure occurred in a Fischer Scientific FS60D ultrasonicator. Organic solvent exposure begins with a 2 hrs exposure in a 1:1 mixture of n heptane and acetone, which are used to dissolve and break up the bulk of the organic film in preparation for an additional 2 hrs exposure in methanol, which leaves a pristine, but hydrophobic AlGaN surface. An exposure in an equal parts solution of heptane and acetone is preferable to acetone alone, resulting in a much cleaner sample surface. This hydrophobic surface is, unfortunately, incredibly effective at attracting dust present in the processing environment. With this in mind,


49 the last step in the solvent cleaning process is a 2 hr ultrasonication in water, which further cleans the sample and also generates a hydrophillic AlGaN epitaxial surface, which attracts much less dust f rom the ambient and which is ideal for imaging with TEM/SPM. In addition to exposure in organic solvents, submicron devices also require a secondary BOE exposure. In a submicron device, the overhang of the T gate structure over the conformally deposited silicon nitride allows it to mask this region of the deposited nitride from exposure to BOE. As shown in Figure 2 7, this masking results in the formation of an silicon nitride "stringer", which runs parallel to the gate width of the device. Stringer fo rmation is, generally, not an issue for larger gate lengths and, as a result, no secondary BOE exposure is required for larger gate length devices. Because this "stringer" is so well masked by the submicron gate necessitate a second exposure in BOE after removal of the gate using a metal etch solution. This secondary BOE exposure is timed at 5min for submicron devices which are under inspection. After this secondary BOE exposure, the devices are cleaned in the same progression of organic solvents as was previously described for the case of the 1 m gatelength devices. This method of chemically deprocessing a device of interest would be of little real use if the top layer of the AlGaN was eroded away as a result of etching. With that in mind, the cross s ectional thickness of the AlGaN layer was analyzed with TEM, using the University of Florida's JEOL 2010F Transmission Electron Microscope in dark field mode. A TEM cross section was cut from a deprocessed device, which was previously coated with thermall y evaporated carbon as a protective layer, and compared to a control sample cut from the gate region of a pristine control device. As shown in Figure 2 8, the thickness of the AlGaN layer does not change as a result of exposure to the HF/TFAC chemistries or from organic solvent exposure. The deprocessing chemistry used is perfectly selective.


50 2.4 Scanning Electron Microscopy Scanning Electron Microscopy (SEM) is an electron imaging technique which has been employed in materials studies for a variety of applicatons since the techniques development in the mid 1930's. The technique is particularly popular in the semiconductor industry, where it is used in both non destructive analysis of devices at key stages during the manufacturing process (such as dry e tching and lithography), as well as in destructive post mortem analysis of defective devices [48]. SEM actually encompasses a variety of analytical techniques utilized heavily in material analysis. All of these techniques involve scanning an electron bea m over the surface of the material of interest and collecting some observable signal which results from the interaction of incident electrons with the material being analyzed. This technique is performed in a vacuum, but the vacuum level employed often va ries depending upon the specific system being utilized as well as the application of interest to the user. An SEM system consists of four modules which act in conjunction to provide an electron image of the material being analyzed [49]. The sample chambe r and electron column represent the module where all important interactions between the electron beam and the material under study occur and are measured. However, it would be impossible to induce these interactions, or to measure them, without the vacuum system which is used to pump atmosphere out of the sample chamber and increase the mean free path of electrons to a point where they can be collected by the detectors within the sample chamber. Likewise, the intricate electromagnetic lenses of the electr on column would not yield a usable image without the scanning electronics used to raster the beam across the sample surface and to correlate the changing signal from the detectors being used to the changing position of this rastered beam. Material samples meant for study with SEM generally yield higher quality images when they are conductive and connected to some ground electrode. Without this grounding


51 connection, sample charging can occur, resulting in distortion of the output image during analysis, whi ch generates an electric field that interferes with the electrons' path to the sample Grounding can be accomplished by a variety of methods. Carbon tape or silver paste are both commonly used to affix samples to the sample stub placed within the SEM syst em prior to analysis. These two adhesion methods are desirable because they also form an electrical path from the back surface of the sample to the stub and, through it, to the grounded clamping mechanism located in the sample chamber. This method is eff ective in the case of an AlGaN/GaN HEMT sample, but the presence of a semi insulating layer on the near surface of the sample (i.e the AlGaN layer) still predisposes the sample to charging during analysis. In order to eliminate this issue, carbon tape is affixed to the stub such that it bridges the top surface of the AlGaN and the carbon tape holding the back surface of the sample to the stub. This carbon tape contacts the n GaN layer present on top of the AlGaN, shorting it and ensuring that charge injec ted into the sample has some pathway to ground. Utilizing the n GaN capping layer is desirable over coating with a conducting film, which would impede direct imaging of the semiconductor surface. Once the sample is properly grounded, it is mounted within the SEM and the system is pumped down to its ideal operating pressure. Depending upon the SEM, this pressure can vary from near atmosphere to Ultra High Vacuum. In the case of the SEMs used for analysis in this study, this was accomplished by means of a combination of a rough pump and turbo pump, designed to bring the system down to a pressure of 5x10 5 mmHg. Once the sample chamber is pumped down to a suitable vacuum, the electron source can be turned on. An aperture opens between the sample chamber and the electron column, which is maintained at a pressure lower even than that of the sample chamber. Electrons with an energy


52 defined by the instrument's operator are emitted from a source at the top of the column and are accelerated through a potential difference before encountering a series of electromagnetic lenses designed to condense the image of the electron source into a beam of uniform illumination and demagnify this beam as much as possible. It is in this region that the steering electrodes can be found, which raster the electron beam back and forth over the sample surface. The overall beam current passing through this column is controlled by an aperture placed between the end of the electron source and the beginning of the lens optics of the e lectron column. It bears noting that the lens optics located in the electron column must, generally, be tuned by the user to induce an image with the maximum resolution. The incident beam may be adjusted to yield a perfectly circular spot, as opposed to an ellipsoid, by "wobbling" the electron beam optics in and out of focus and varying the x and y stigmation until the image produced defocuses evenly in both directions. The alignment of the electromagnetic lenses present within the electron column is also critical for right resolution imaging. This alignment is accomplished, again, by wobbling the beam optics in and out of focus and adjusting the lens alignment in x and y such that the resultant image does not shift while the beam is being "wobbled". A variety of electron sources have been utilized in SEMs since their inception. The most popular of these sources utilized in state of the art systems is the thermal field emission source, which is capable of producing an electron beam which is smaller, more coherent and energetically uniform, and with three orders of magnitude higher current density than older sources such as sources which use thermionic emission. Modern SEMs are capable of a wide range of accelerating voltages and generally operate in the 500V to 30kV regime and are capable of demagnifying the source to an electron beam which approaches 1nm spot sizes. The SEMs used in this study utilized a thermal field emission source and the extracted electron beam was


53 accelerated through a voltage of 5kV. In each system, an aperture was selected which maximized current without hindering high resolution imaging. The FEI NovaSEM, utilized for some imaging as part of this study, operated in a regime capable of generating a 1.0 1.5nm spot size elect ron beam, while the FEI DB235 FIB, which was utilized for most of the imaging performed in this work, operated in a regime capable of generating approximately a 5nm spot size electron beam. The mean free path of the electrons which leave the column is lar ge enough that they do not interact with any other matter en route to the sample. When they encounter the material under study, the electrons are slowed by interactions with the nuclei and electron clouds of the atoms which make up the sample and their ki netic energy is deposited into the material itself as the electrons are decelerated. The result is a characteristic "teardrop" which represents the interaction volume that a given injected electron has with a material sample as it follows a brownian path of travel influenced by interactions with the sample atoms. Different resultant information, in the form of three separate characteristic electron signals and two characteristic light signals, are emitted from different regions of the interaction volume depending upon the specifics of the interaction which occurs. Of these, only three separate signals (backscattered electrons, secondary electrons and cathodolumenescent light) are germane to data represented in this work. The reader is directed to othe r texts if signals other than the three listed above (characteristic and continuum x rays and auger electrons) are of interest [50]. The size of the teardrop, itself, increases with increasing incident electron energy as well as with decreasing atomic num ber. As these two criteria are met and the interaction volume grows in size, it does so in both the vertical direction (penetration depth) and in the lateral direction (spatial resolution). It is for this reason that all imaging performed on the


54 material s studied as part of this work was performed at a moderate accelerating voltage (5kV) rather than at larger accelerating voltages. As electrons pass through the sample material, they encounter and interact with the atoms which comprise that material. Wh en an electron encounters an atom, it can glance off of the electron cloud of that atom, imparting some of its kinetic energy in an inelastic collision and slowing down as a result. The other option available to the electron is to interact with the bulk o f the electron cloud. If this occurs, the incident electron recoils violently in an inelastic collission which ejects it from the sample material. These electrons are called backscattered electrons and, not surprisingly, they are highly sensitive to the number of electrons present in the cloud of each atom. As atomic number goes up and the density of electrons goes up with it, the yield of backscattered electrons increases. Because they are the result of the collision of a material interaction with an i ncident electron which lost little energy, backscattered electrons can be found throughout most of the interaction volume. In the case of an inelastic scattering event, the interaction between the incident electron and the electron cloud of the atoms co mprising the sample can result in an electron being kicked free of its parent atom with a very low energy (<50eV). Secondary electrons located near the sample surface can be emitted from the material with high probability, yielding substantial information about the topography of the sample. If a secondary electron isn't emitted, it can be captured by an atom and fill an unoccupied orbital. In semiconducting and semi insulating materials, this process occurs with high frequency, as a large number of onocc upied orbitals exist in the form of holes within the material. This electron hole recombination event is known as cathodoluminescence [51]. Secondary electrons are are generated close to the surface of the sample (within 5nm to 50nm), where the electrons incident to the sample surface have not yet


55 lost substantial kinetic energy and may still undergo a large number of inelastic scattering events. The position of the secondary electrons in the thinner, necked region of the teardrop near the surface result s in a small lateral straggle for the signal, enhancing the spatial resolution of the measurement being made. For this reason, all imaging performed in this study was performed using detector settings meant to enhance the collection of secondary over back scattered electrons. A standard secondary electron detector (SED) sits off axis from the electron column in order to detect the electron signal from the sample without interfering considerably with the demagnified image of the source as the incident elect ron beam approaches the sample from the source. The SED consists of a phosphor based scintillator covered with a faraday gride and in line with a light guide and photomultiplier tube. As secondary electrons are ejected from the sample surface, they are p ulled towards the secondary electron detector by the faraday grid, which is held at a positive bias of, at maximum, a few hundred volts. These electrons are bent towards and accelerated through the faraday grid and impact the scintillator. The scintillat or fluoresces and the escaping light is channeled down a light guide towards the photomultiplier tube, which amplifies the signal from the scintillator so that it is more readily detected by the photodiode mounted at the end of the tube [52]. This is an e ffective method for detecting the secondary electron signal emanating from the sample material. However, imaging a secondary electron signal at high magnifications is far more challenging because the probe size, and resultant number of secondary eleectron s output from the sample, is small. To that end, a different method of collection is required for high resolution SEM with secondary electrons. The solution is the use of a through lens detector (TLD), which is a scintillating electrode mounted directly to the pole piece of the microscope,


56 where the electron beam generated in the source and manipulated by the column enters the sample chamber. This positively biased electrode sits much closer to the sample than a conventional SED, and is more readily able to detect a secondary electron signal [20]. In modern systems, such as the FEI Nova SEM utilized in this study, the TLD is coupled with an immersion lens, so called because it physically envelops the sample surface in an electromagnetic field which induc es a helical path of travel for secondary electrons leaving the sample, increasing the efficiency with which the TLD located within the pole piece of the microscope collects secondary electrons and screening out the backscattered electron signal. This imm ersion lense also screens out ambient magnetic fields present within the sample chamber of the microscope and the environment in the area where the microscope itself is installed, eliminating their deleterious effects on imaging. For this reason all imagi ng performed on samples studied with SEM is performed using the TLD, when available. 2.5 Surface Probe Microscopy Surface Probe Microscopy (SPM), also known in the literature as Atomic Force Microscopy (AFM), utilizes a nanoscale probe located at the tip of a micromachined cantilever beam to measure the properties of surfaces on scales ranging from the micrometer to the nanometer scale [53]. An SPM system consists of three major modules which work together to form an image of a materials surface [54]. The first of these modules is the "head" of the SPM system, which is an assembly into which the SPM cantilever which is used for imaging is attached. The tip itself is moved about the sample surface using piezoelectric actuators located in the SPM head wh ich are capable of shifting the cantilever in the lateral and horizontal directions. Another piezoelectric actuator is used to control the vertical deflection of the cantilever as it travels over the sample surface. This deflection is measured by bouncin g the beam emitted by a laser diode off of the tip of the


57 cantilever. This bounced beam strikes a four cell photodiode and the resulting difference in induced photovoltage between each cell is used to measure the total deflection of the cantilever beam du ring its travel. The deflection measured by means of the four cell photodiode is fed into the second major module of the SPM system, which are the computer controlled electronics used to control the cantilever tip and to collect deflection data resulting from the measurement being taken. In addition to a Personal Computer, which acts as the operator interface for the AFM, a lock in amplifier is used to measure the signal output from the four cell photodiode and a frequency generator is used for input sign als to the piezoelectric height actuator used to induce probe deflection. The function of the control electronics varies depending upon the mode of imaging selected for the experiment. There are a variety of methods which can be used to image a surface with SPM, but they all fall, more or less, into two separate imaging modes. In contact mode, the probe tip of the SPM cantilever is put in direct contact with the sample surface, such that a repulsive force is generated between the sample and the probe, resulting in cantilever deflection. The control electronics work to maintain this deflection, and the magnitude of the repulsive force which generates the deflection, at a constant value. In tapping mode, the SPM cantilever is vibrated above the sample s urface a a frequency just higher or just lower than its resonant frequency. As the tip approaches the sample surface, interactions betwen the probe at the end of the SPM cantilever and the sample surface induce a change in the resonant frequency of the AF M tip, which is detected by the control electronics. In tapping mode, the control electronics maintain the deflection of the AFM tip in such a state that the resonant frequency of vibration does not change as the tip is scanned over the material of intere st. This changing deflection can be


58 converted into a map of the interactions at the sample surface. In the case of a surface which does not yield and which does not cause the SPM tip to "stick", as is the case for the AlGaN surface of a HEMT, this map of interactions can be assumed to be equivalent to the varying height of the surface of the sample relative to the scanning SPM tip. The third, and arguably the most important, module of the SPM system is the SPM cantilever itself, which is generally a stri p of thick silicon or silicon nitride material several millimeters long by several millimeters wide. A "T" structure is typically etched out of the top surface of this cantilever to mark the desired position of the structure for mechanical clamping to the SPM head while the probe is located at the tip of the cantilever on its "bottom" side. Sometimes, the top surface of this cantilever will be coated with a reflective thin film in order to maximize the reflected optical signal travelling from the cantilev er to the photodiode based detector of the SPM. The material which this tip is formed from and its general dimensions can vary dramatically based the on application desired. Nonconductive probes with a high aspect ratio and a small radius of curvature are ideal for the measurements of nanometer scale features on smooth surfaces, as is the case for defects present within the AlGaN/GaN material system. These tips are generally formed from the cantilever itself, which can be reactive ion etched to form a n ideal tip geometry, or from the direct bonding of nanotubes or nanowires to the underside of the cantilever beam. A variety of imaging artifacts can influence the measurement of surfaces using SPM and lead to spurious interpretation of data. Some of th ese artifacts can be eliminated by properly adjusting the various settings of the control electronics, chiefly the proportional and integral gains, so as to ensure the proper travel of the SPM tip across the sample surface as associated


59 recording of the pr operties of that surface. Generally, the overall quality of the image formed by SPM analysis can also be improved if the scan rate of the SPM tip across the sample surface is reduced, allowing the cantilever more time to equilibrate and yield a low noise signal for a given pixel before transitioning to the next point in the image. The SPM tip used for analysis can also have a large bearing on the quality of the image formed as the tip is rastered across the sample surface. The two chief metrics which h ave a bearing on overall image quality for a given SPM tip are the aspect ratio of the tip itself and the nominal tip diameter. SPM cantilevers which have large probe tip diameters are prone to imaging artifacts because a large tip area interacts with the surface at any given point in time. Not surprisingly, his effectively limits the resolution of a given micrograph to the tip diameter and results in small surface features being represented as having larger dimensions than is physically the case. Small features which are inset from the sample surface, such as pits or trenches, may be rendered effectively "invisible" in this case, as the SPM cantilever tip may travel directly over the recessed feature without detecting it due to a large tip diameter. Tip s with a low aspect ratio can also lead to imaging artifacts when imaging densely packed features possessing a high aspect ratio. In this case, the tip may interact with the sidewalls of a recessed or feature rather than its bottom surface, resulting in q uestionable topographic data. As might be surmised from the discussion above, it is of critical importance to utilize an SPM cantilever tip which possesses a nominal tip diameter and aspect ratio which matches the dimensions of the features under observat ion. All SPM imaging performed as part of this study occurred on a Bruker Dimension 3100 Atomic Force Microscope in tapping mode, with a deflection voltage ranging from approximately 300mV to 400mV, depending upon the sample under observation, and with th e


60 integral gain and proportional gain for the feedback electronics set at values of 5.0. Prior to measurement, the system was calibrated so that the observed difference in four cell photovoltage was below 100mV. The scan rate for image capture was set at 0.5Hz and 512x512 pixel arrays were generated, corresponding to a height map of the AlGaN surface. A Bruker TESP HAR tip with an 5:1 aspect ratio, 40 N/m stiffness, and 10nm nominal tip diameter, was utilized for all measurements.


61 Figure 2 1 A sc hematic view of the electron optics utilized in a typical TEM. Visible are the electron source and condensing optics, used in forming a coherent source for imaging, as well as the objective and intermediate optics, which determine the information about th e sample which is transmitted to the projector lenes and screen.


62 Figure 2 2 The steps associated with Focused Ion Beam milling of a generic sample. Progression runs down the left column and then down the right column. The configuration of the sample relative to the ion and electron beams on the left and a top down image of the sample is on the right. After deposition of a protective mask via an ionized organomettallic beam and trench milling to either side of the sample, the sample is tilted relative to the gallium beam in order to perform an undercut and release one side of the lamella. An omniprobe needle is inserted, and welded to the sample. After welding, the last side of the lamella is cut free and the sample is retracted along with the omnipr obe needle.


63 Figure 2 3 A top down image of a typical maskset analyzed as part of this work. Three separate device architectures are utilized for combined electrical and microscopy analysis. These are the "FatFET" (a large device with a 75m gatelen gth), a submicron device (with a gatelength ranging from 100nm to 160nm) and a micron device (with a gatelength of 1m). Figure 2 4 The Deprocessing of an AlGaN/GaN HEMT. Top down SEM micrographs with corresponding schematic cross sectional respresent ations (not to scale) show the etching strategy employed for deprocessing AlGaN/GaN HEMTs to allow quantitative under gate defect analysis prior to deprocessing (left), etched for 15 min with BOE (middle), and subsequently etched for 18 h with TFAC, etched for 5 min in BOE for a second time, and cleaned ultrasonically for 2hrs in a volumetrically equal mixture of acetone and n heptane, for 2hrs in methanol and 2hrs in water (right). The locations of the source (S), gate (G), and drain (D) contacts for each image are indicated.


64 A B C Figure 2 5 AlGaN surfaces after exposure to HF. A) Plan view SEM images of a sample etche with BOE until full removal of SiN with minimal over etching. B) Plan view SEM of a sample formed from a SiC seed crystal whic h was over etched and to reveal a "tidepool" mophology. C) Plan view SEM of a sample formed from a Si seed crystal which was over etched and which formed more characteristic etch pits.


65 A B C Figure 2 6 AlGaN surfaces after exposure to various organic solvents A) Plan view SEM images of a sample as deprocessed, with no futher exposure to organic solvents. B) Plan view SEM of a sample exposed for 2hrs in methanol C) Plan view SEM of a sample exposed for 2hrs in 1:1 acetone and n heptane an d for 2 hrs in methanol.


66 Figure 2 7 Incomplete Removal of SiN from a T gate. This is due to masking by the metal gate, resulting in the formation of a "stringer" which must be removed with a secondary etch in BOE after TFAC exposure. A B Figure 2 8 HAADF STEM images of deprocessing of the near gate region of a HEMT. A) The HEMT structure is shown prior to deprocessing, with Au and Ni metal layers. B) After deprocessing, thorough removal of the metalization layers with the AlGaN and GaN epilayer s left completely intact. The protective C layer was deposited prior to sample preparation using focused ion beam milling to prevent surface damage.


67 Figure 2 9 A schematic diagram of a basic SEM system. The sample chamber, control and display electron ics, and electron column are all present. The vacuum system (not pictured) maintains pressure in the electron column as well as in the sample chamber. Figure 2 10 A schematic diagram of a generalized SPM system. The SPM head produces a deflection sig nal which is controlled by an set of negative feedback loops, adjusted via a user controlled PC. In this schematic, the tip is contained within the SPM head, though it is a modular component.


68 CHAPTER 3 OHMIC NANOCRACK FORMATION AND ITS IMPACT ON RELIABIL ITY Ohmic contacting schemes to compound semiconductors are quite varied in their nature, ranging from traditional methods including doped wells, to a variety of alloyed contacting schemes, and even to nanostructured contacts which utilize thin interfaci al layers to alter the energetic barrier to charge transfer. It is not surprising, therefore, that different contacting styles utilized for making an ohmic connection to different semiconductors should suffer from different degradation mechanisms. Figure 3 1 demonstrates the contacting scheme utilized in the AlGaN/GaN HEMT structures studied as part of this work. The ohmic contacts utilized in this work are separated from each other by approximately 4.0m and from the gate electrode of the device by a ma ximum of 2m and sometimes as little as 1.5m, depending upon the geometry of the gate contact of the device. The contacts, themselves, are formed from a traditional metal stack utilized in contacting: Ti/Al/Ni/Au. These metals mix together to form a met allic alloy during Rapid Thermal Annealing (RTA) at 850C for 30s. The result is the formation of an ohmic contact with a contact resistance ranging from approximately .2 [19]. This RTA process is known to greatly impact the overall qual ity of the ohmic contact made to the AlGaN surface and has been linked to changes in the surface roughness of the contacting metal pad [55]. Deprocessing of ohmic contacts via metal etching has also demonstrated that this contacting scheme appears to rely on the diffusion of some component of the ohmic metal into the underlying AlGaN epitaxial layer, presumably down the threading dislocations which extend up from the growth substrate, through the MOCVD grown GaN and into the AlGaN [6]. This diffusion even t results in the formation of a metal inclusion, which


69 shorts out the ohmic contact, resting on top of the AlGaN epitaxial layer, to the 2DEG which is extant at the interface between the AlGaN and GaN. The first section of this chapter will focus on previ ous studies into the annealing of various metal stacks formed on AlGaN/GaN in order to form an ohmic contact to the buried 2DEG layer. Special attention will be paid to intermettalic reactions which both aid in conduction between the 2DEG and the ohmic co ntacts and which hinder this conduction through the formation of resistive phases and the degradation of contact morphology. The second section of this chapter will focus on the structural nature of this metal inclusion and will discuss the conditions whi ch it induces the surrounding epitaxial materials during its formation. These conditions, specifically the large compressive stresses induced in the epitaxial layer as the result of the formation of this metal inclusion as well as the tensile stresses ind uced at the edge of these inclusions, influence defects formed during processing which hinder device performance. The second section of the chapter will focus on the defects which form during processing. These defects are nanocracks which originate at th e ohmic contact and which can penetrate into the channel material of the HEMT, adversely effecting charge transfer characteristics within the device. Evidence also exists which suggests that these defects can be influenced by electrostatic stressing of th e device. Devices under analysis as part of this study were formed using the same processing conditions and device architectures specified in Chapter 1. AlGaN/GaN HEMTs with a 100nm gatelength were used. In order to observe the formation of these defec ts, two separate analytical tools were employed. Lamellae for cross sectional TEM analysis were formed via FIB/SEM on a FEI DB235, where in situ sample milling, liftout, mounting, and thinning were accomplished via the use of an Omniprobe needle. The ion source used for milling was a gallium beam. All


70 samples were coated with a 200 nm carbon layer, deposited via thermal evaporation and an organnometallic Pt source was used to form the 2m protective mask utilized during milling. Analysis of lamellae for med via FIB/SEM was performed on a JEOL2010F TEM, with an EDS attachment. Analysis of the sample was done using HAADF STEM because of its utility in quantitative analysis using EDS. Samples under analysis were also studied using top down SEM after the d eprocessing methodology described in Chapter 2 was performed. Imaging of deprocessed samples was performed on the same FEI DB235 in Ultra High Resolution mode with the Through Lens Detector selected. As mentioned previously, all stressing and measurement of device characteristics was accomplished via the use of an HP4146C device tester, capable of independant measurements of current and voltage on all three device contacts, and a Techtronix 370A curve tracer. The devices used in this study were formed us ing the processing described in the first chapter of this dissertation on a SiC substrate. Stepped stressing was performed on one device analyzed as part of this study, where the device was taken from a value of V GS = 10V to V GS = 42V at 1V/min with V D S = 15V. Measurements were made after every voltage step with a duration of 1ms, making the duty cycle of the measurement effectively equal to 100%. 3.1 Progress in Ohmic Contact Annealing In addition to their characteristic low thermal conductivity [57 ] and semi insulating behavior [1], GaN based systems are highly unreactive [58]. This stems from the highly ionic nature of bonding in GaN and AlGaN, which contributes to all of the qualities listed above. Because of the structure of AlGaN/GaN HEMTs, wh ere the conductive channel is buried under a thin layer of semi insulating AlGaN, the unreactive nature of this material system makes ohmic contacting a challenge [59].


71 Current contacting schemes to AlGaN/GaN HEMTs rely on a multi layer stack of deposited metals which are annealed in order to induce alloying within the contact and between the contact and the AlGaN/GaN epitaxial layers. This metal stack initially was formed from a layer of titanium placed in contact on the AlGaN surface, followed by a laye r of aluminum, both deposited by electron beam evaporation or sputtering [60]. Upon annealing at temperatures ranging from 750C to 950C, the titanium undergoes a reaction with the Al layer above it as well as the AlGaN below it, forming TiN as well as A lTi 2 N, which are formed via a reaction where nitrogen is depleted from the AlGaN epitaxial layer. This reaction results in the formation of an electron rich surface layer in the AlGaN which results in enough band bending within the AlGaN to form an energe tic barrier which permits the tunneling of electrons [61]. The TiN formed as part of the annealing process also results in enhanced conduction. In some cases, it expands into the AlGaN layer down threading screw dislocations to form an inclusion [61]. T hese inclusions reduce the distance between the ohmic contact and the 2DEG present at the interface between the AlGaN and the GaN and reduce the resulting tunneling distance for electrons. In some cases the metal inclusion completely bridges the distance and forms a direct link between the two charge conducting layers. This alloying scheme is not without its flaws, however. Titanium and Aluminum are highly reactive metals and oxidation results from contact annealing, even during rapid thermal annealing i n a nitrogen purged environment [62]. Contact degradation due to oxidation can be averted by coating the ohmic contacts with a noble metal. Gold is the contacting metal of choice for this application. However, gold intermixes with the Aluminum and Titan ium in the ohmic contact during annealing, resulting in the formation of a variety of intermetallic phases, observed via EDS and STEM, including Al 2 Au [63], AlAu 4 AlAu, and AlAu 2 Ti [64]. None of these


72 phases appear aid in conduction out of the ohmic cont act and their formation often degrades the morphology of the thin film stack, resulting in substantial roughening which also degrades the contact resistance of the device In order to avert the formation of these intermetallic phases, which greatly degr ade the overall quality of the device, a barrier layer is often inserted into the device structure. This barrier layer is used to act as a means of preventing Au diffusion into the Al and Ti layers of the ohmic contact. A variety of metals have been util ized for this application, including Pd [65], Ti [66], Mo [67], and the barrier layer used in the devices studied as part of this work: Ni [68]. Needless to say, the metals used in barrier layers have variable efficacy in addressing the problem of Au diff usion and reaction with the underlying Al and Ni layers and in doing so without suffering from some other adverse reaction which degrades the character of the electrical contact. Ni, which has been the barrier metal of choice up until recently, is a per fect case of a mm) but which suffers from a deleterious secondary reaction which greatly impacts the surface morphology of the contact, making soldering during later pa ckaging processes difficult [69]. multiple XTEM studies have show that, upon annealing, Ni reacts with Al and, in doing so, ceases to behave as a contiguous layer, allowing Au to diffuse down into the underlying Al to form intermettallic Au Al phases. Th e film stack roughens substantially upon annealing as the Ni Al and Au Al phases separate from one another [70]. Mo, which is becoming the barrier metal of choice for contacting, does not react with Al and does not disintegrate. However, Au is capable of diffusing through the grain boundaries of the metal thin film to interact with the Al below. The action of this diffusion degrades this barrier layer at high temperatures (950C).


73 A variety of methods exist to improve the efficacy of these layers, throu gh optimization of different metal layer thicknesses, to optimization of annealing conditions and even to addition of additional materials to form a more complex alloyed contact. It stands to reason that, all other things being equal, a thicker barrier me tal layer should reduce the amount of Au which intermixes with Al. Ideally, a barrier layer should be thicker than the diffusion length of gold within that layer. Studies performed on a Ti/Al/Ni/Au stack demonstrated that the contact resistance and obser ved line edge roughness of contacts were reduced for Ni layers which were two times as thick as the deposited Ti, Al and Au contacting layers, with a contact resistance of mm. The morphology of the contact was found to be dominated largely by the t hickness of Au. Increasing Au thicknesses resulted in ever growing surface roughnesses, presumably due to Au balling. Contact resistance goes through a local minimum at a Ni thickness of 1.8 times the Ti and Al thicknesses, likely because Ni acts as an e fficient diffusion barrier to Au without consuming substantial Al volume [71]. Annealing of the ohmic contact is critical to the formation of TiN inclusions which short the AlGaN by joining, via tunneling or direct metallurgical connection, the 2DEG to th e alloyed contact. Alloyed contacts formed from Ti/Al based metallurgy go through a local minimum of resistance, as measured by TLM pads, in the range of 800C to 850C. A sharp rise in contact resistance occurs at temperatures above this range. This l ocal minimum in contact resistance is generally believed to be induced by the reaction of Ti and Al with the AlGaN or GaN, while the sharp rise in resistance which occurs above 850C is due to intermetallic reactions between gold and the rest of the alloye d contact. A multi step annealing process performed on a Ti/Al/Ni/Au metal stack, consisting of three 45s RTA cycles at 400C followed by 700C and finished at 830C, has been shown to greatly improve the surface morphology of alloyed contacts as well as


74 their contact resistance. This improvement has been attributed to enhanced reactivity of Al with Ti and GaN at lower annealing temperatures, which reduces the Al available for reaction with Au [72]. Recent research has also shown that the addition of a s puttered Si layer to the alloyed metal stack seems to enhance conductivity of the ohmic contacts such that the contact resistance mm after annealing at 750C. Analysis with XTEM demonstrated that this reaction was likely caused by intermetallic reactions between the silicon and the rest of the alloyed contact which resulted in a reduced reaction between Al and Au [73]. 3.2 Metal Inclusions and Nanocrack Formation Figure 3 2 is a representative image of a metal inc lusion formed in a HEMT as a result of the ohmic contact annealing process which was previously described. The top of the HAADF STEM image is representative of the metal pad utilized in forming the source drain ohmic contacts. Visible within this metal l ayer are columnar grains of TiN, which phase separates from the other components of the ohmic contacts upon annealing. Below the deposited metal layer and these TiN grains is the surface of the AlGaN and, below the AlGaN, the MOCVD grown layer of GaN upon which the AlGaN sits. A metal inclusion is visible in the image, extending from the AlGaN surface down into the underlying GaN. A threading dislocation is visible extending from the bottom surface of this metal inclusion into the Ga. The presence of th is threading dislocation under the metal inclusion is consistent with previous reports regarding the annealing of metal contacts on AlGaN/GaN epitaxial structures. Metal inclusions form by the diffusion of metal atoms down high energy threading dislocatio ns, where the activation energy for diffusion is reduced. The AlGaN and GaN are not capable of supporting the high metal concentrations within these threading dislocations as metal diffusion


75 proceeds, resulting in the formation of an inclusion rather than a metal rich "pipe" of AlGaN or GaN. It is worth noting that several distinct layers of material are observable w ithin this metal inclusion, as was the case in previous research on Ti based ohmic contacts Special attention should be paid to the signals of the metal inclusion. These signals indicate that the inclusion is comprised largely of titanium likely in the form of TiN, given the observations made by previous researchers. Galliu m is conspicuously absent from this inclusion and has likely been displaced. An aluminum layer, either formed from the aluminum in the metal contact itself or (more likely) by the aluminum which made up the AlGaN layer displaced by the inclusion itself, s urrounds the TiN inclusion. Figure 3 3 demonstrates the result of the formation of these metal inclusions. After deprocessing, top down SEM revealed three distinct topographical regions on each of the mesas upon which the transistors under analysis were formed. Two of these are regions where ohmic contacts were formed: the Source and the Drain. Fully 10% of the total aerial density of the source and drain regions is comprised of areas which were once metal inclusions. The average size of these inclusio ns is 62 nm in diameter. The source and drain are separated by a smooth AlGaN surface which once comprised the channel of the device. These inclusions are hexagonally faceted, not rounded which would reduce the surface to volume ratio and the free ener gy in the case of a material with equal surface energies for all orientations and appear to be bisected by nanocracks observed in the deprocessed contact regions. A larger volume inclusion, where the prism planes of the AlGaN are eroded preferentially is a lower energy feature than a cylindrical inclusion where all planes of the AlGaN are eroded with equal certainty. This fits well with the substantial bonding energy inherent to the wurtzite GaN


76 system, where surface energies vary by many eV per angstr om for different crystal orientations thanks to the highly polar nature of their bonding [74]. A consequence of the formation of these inclusions are nanoscaled cracks which appear to radiate from the corners of the TiN metal into the surrounding volume o f the AlGaN. The sharp faceted corners of these metal inclusions appear to be very effective in inducing the initial formation of a surface defect. The average length of one of these nanocracks is 52nm. If an inclusion exists at the edge of the source or drain, the crack will extend into the channel. In general, such a crack does not extend into the channel much farther than the cracks present within the ohmic contact regions. Cracks present in the channel region are, on average, no more than a few h undred nanometers longer than cracks present within the ohmic contact regions, with an average length of 273nm. 3.3 Nanocrack Morphology Cross sectional TEM was performed on a nanocrack which was observed in UHR SEM. This was accomplished by, first, ma rking off the crack with fiduciaries which allowed for proper orientation of the ion beam and associated lamella formed via FIB after the surface of the sample was coated in carbon. This allowed for a cross section to be formed which runs perpendicular to the direction in which the crack was propagating. An image of this nanocrack and the corresponding orientation of the nanocrack is shown in Figure 3 4. The resulting cross section of the extended nanocrack was imaged using HAADF STEM as well as conventi onal electron diffraction utilizing a parallel beam. The bulk of this nanocrack is present in the AlGaN epitaxial layer and the total crack depth can be estimated to be approximately equal to 20 nm. It does not extend deeply into the GaN. Thus, the feat ures observed in SEM appear to be channel cracks forming, and more or less constrained, to the thin film of epitaxial AlGaN present on top of the GaN. Electron diffraction


77 taken from the AlGaN crystal region of the TEM cross section demonstrates that the crack itself propagates along the [11 20] prism planes of the lattice, which fits well with cracks being present both parallel and at 60 angles to the channel direction, as the channels of these transisting devices are oriented normal to the prism planes of the hexagonal lattice. Crack propagation has been observed along these prism planes by previous studies as a result of film growth well beyond the pseudomorphic limit [75]. The variable length of nanocracks inside and outside of the ohmic contacts c an be readily explained by considering the nature of the cracking which is occuring in the epitaxial layer under the channel and under the ohmic contacts. The assumption that AlGaN and GaN share a very similar Young's Modulus and Poisson's Ratio has been made by researchers in the past, when analyzing channeling cracks [76] and the assumption that the fracture toughness of bulk AlGaN is similar to that of GaN also seems reasonable, given this knowledge. The AlGaN layer in the channel region can be consi dered to be stress free with the exception of the residual stresses induced in the layer as a result of epitaxy. A crack which forms at the edge of an ohmic contact, in the channel region, will respond to this residual stress and will propagate until the length of the crack is such that the residual tensile stress cannot support additional crack formation at the tip. This effect can be modeled by a modified form of Griffith's Equation [77], where K is the fracture toughness of the film (in Pa m 1/2 ), c is the crack length (in m), and where the tensile stress acting on the film required for additional crack growth state of the AlGaN under the channel, provide d that the geometry of the epitaxial film is taken into account. In this case, a modifier (Z) is used to account for the geometry of a crack forming in an epitaxial film on top of a much thicker substrate. This constant approaches a value of 1.976


78 for a 2D channeling crack as the film becomes thinner, becoming nearly indistinguishable from this value roughly after a ratio of 1:10 is achieved between the film thickness and the thickness of the relaxed layer below it, which is certainly the approximate case for the AlGaN epitaxial layer on top of GaN [78]. ( 3 1 ) It is noted that the scenario detailed above is not exactly the physical case in the device under stress, which is coated with a passivating nitride film. However, because the co nstant resulting from any channeling crack in a thin film varies only between 1 and 2, this approximation is assumed accurate to a first order. The author suspects that the solution for the channel cracking of a thin epitaxial layer with a passivating thi n film layer (almost certainly under some unknown plane stress and with an unknown compliance at the nanoscale), is a task meant for simulation. However, the accuracy of this result would be highly dependant on a variety of materials factors which almost certainly vary substantially for a PECVD SiN process [79]. In order to calculate the stress present in the film from a given crack length, the fracture toughness must be estimated. This is accomplished utilizing the following equation, which calculates of the crack (equal to twice the surface energy of a [11 20] prism plane, or 157 eV/ 2 assumed by linear interpolation between the theoretically calculated surfa ce energies of the GaN [11 20] and the AlN [11 20] planes, for an estimate) [74,80]. The Young's Modulus (E) is assumed to be equal to 309GPa [81] and the The Poisson Ratio (v) is assumed to be equal to 0.51 [76] (3 2 )


79 The average str ess calculated from these coupled equations for an average crack present in the channel is 584 MPa, which is roughly a fourth of the value calculated from the difference in latice constant between relaxed and pseudomorphic Al 0.28 Ga0 .72 N (1.94 GPa), and not a very good fit [82]. Because the crack has been observed in the AlGaN layer away from the metal inclusions, this value of tensile stress can be assumed to be just larger than the average stress encountered by cracks which move through the channel, howev er, this assumption is made without any knowledge about the distribution of crack lengths inside and outside the ohmic contact regions. The histogram of crack lengths and corresponding tensile stresses for additional growth associated with these lengths is shown in Figure 3 4. This histogram is representative of approximately 600 cracks present under the ohmic contact and channel regions, each. It originated from 20 devices recieved as fabricated. Cracks present within an ohmic contact occupy a very n arrrow distribution centered around a maximum of 40 60 nm, which is approximately equal to the calculated average. Nevertheless, the histogram demonstrates that this distribution is not normal, but skewed towards larger crack lengths. This skew intensifi es dramatically in the case of cracks observed in the channel region, where the maximum is located around 140 nm, which is much smaller than the calculated average of 273 nm. This leads to an estimate of 780 MPa for the stress in the AlGaN film, which is about a forty percent of the tensile stress in the film produced with theoretical calculations a better fit in comparison to the value derived from the average value of the crack lengths in the channel. Using the same equation, the stress in the ohmi c contact region can be estimated as equal to 1.87 GPa, which is significantly larger. This indicates that the epitaxial AlGaN is under


80 compression as a result of the formation of metal inclusions. It is this compressive axial stress which first causes the formation of the flaws from which the cracks observed in the channel grow. This flaw formation is due to a tensile hoop stress, oriented radially with respect to the inclusion, induced by the force of this inclusion pushing on the AlGaN epitaxial laye r. The compressive stress which results in this flaw forma tion can be calculated using Equation 3 3 [83], where R is the approximate radius of the metal inclusion itself and set equal to 26 nm. (3 3 ) In this case, H is a constant associated with the comparative dimensions of the crack initiated by the inclusion and the size of the inclusion, itself, and never exceeds a value of 0.24, provided that the crack's size exceeds half the radius of the inclusion. This assumption is accura te for the case of cracking in the ohmic contact regions. So, the compressive stress induced in the AlGaN by the metal inclusions, and acting on the inclusions to generate tensile hoop stress, may be calculated in this fashion. This results in an estimat ion of the compressive stress present in the ohmic contact region due to the formation of a metal inclusion as being equal to 38.1 GPa, which is a substantial compressive stress and actually about 23% higher than the compressive stress derived from the str ain based purely on the aerial density of pits in the ohmic contact regions (30.8 GPa). The stresses present in the epitaxial AlGaN in the ohmic contacting region can be expected to be larger than this value, as the effects of neighboring inclusions was no t taken into account as part of this equation. It bears noting that only the most infinitessimal amount of additional tensile strain is required within the channel in order to induce additional growth of one of these cracks. It seems plausible, therefo re, that electrostatic stressing could induce additional crack growth by means of biaxial tensile strain induced by the inverse piezoelectric effect, which is described in more detail


81 in Chapter 4. If this occurs, it likely occurs as particularly long cr acks approach the gate contact, where the electric field is substantial. This must occur infrequently, as lengthened cracks have only been observed once during the course of the electrostatic stressing studies associated with this work. Figure 3 6A and Figure 3 6B represent the V DS I DS family of curves for the transistor as well as the I GS V GS relationship for the shottky contact which comprises the gate. They are presented both before and after stressing for a device which underwent stepped stressing f rom V GS = 10V to V GS = 42V at 1V/min and with V DS = 15V. An increase in gate current and a decrease in saturation current can be observed. This could be due to shorting of the gate contact to the source and drain contacts, possibly through the 2DEG. The nanocracks present within this stressed device were analyzed after deprocessing, as evidenced by the historgram in Figure 3 6C. It should be noted that the implicit assumption is made throughout this text that the formation of nanocracks occurs in t he device prior to deprocessing. This may not be the case, but the assumption is made regardless because nanocracks cannot be observed on these devices without the aid of deprocessing. As is shown on this historgram, many of the crack lengths in this str essed device were much larger than crack lengths observed in other devices which were recieved as f abricated Some of these cracks have grown to a length three times greater than even the longest cracks observed in as f abricated devices. It is possible t hat cracks which formed in this device as a result of processing were long enough to encounter strain in the AlGaN layer induced by electrostatic stress which allowed them to grow. This stress is highest under the gate contact of the device, and several c racks extended to this contact or beyond it.


82 Given that some of these cracks have bridged nearly the full length of the channel region, it seems plausible that those channel cracks which were present under the gatee electrode allowed it to short out the AlGaN locally when it refilled the void formed by each channeling crack. Such a situation would fit with an increase in the gate current of the device, which has been observed previously by other authors studying microcracks with Raman spectroscopy [84], and given that no pitting defects were observed in the channel of the device.


83 Figure 3 1 A HAADF STEM image of the cross section of a 100nm gatelength device. The Gate, Source an Drain electrodes are prominently labeled as well as the various mater ial layers present within the device. A B Figure 3 2 A metal inclusion formed after the annealing of a Al/Ni/Ti/Au metal stack. These ohmic contacts were formed via an anneal at 850C for 30s A) A HAADF STEM micrograph of the interface reveals the presence of a metal inclusion. B) An EDS of the inclusion suggests that it is formed from TiN.


84 A B Figure 3 3 Top down SEM analysis of the ohmic contact regions of an AlGaN/GaN HEMT. These ohmic contacts were formed via an anneal at 850C for 30s A) The wholesale formation of metal inclusions throughout the contacted surface. Cracks can be seen nucleating on the faceted corners of the metal inclusions. B) Cracks which nucleate at the edges of the ohmic contact regions can extend into the channel for much longer distances. A B Figure 3 4 FIB/TEM of a nanocrack observed in SEM. A) A lamella for TEM analysis was formed perpendicular to the propagation direction of this crack. B) HAADF STEM analysis coupled with electron diffraction reveal s that the crack extends through the AlGaN layer and is roughly perpendicular to the prism directions


85 A B Figure 3 5 Histograms of crack lengths observed in 20 HEMT devices. A) The crack length distributions associated with cracks found under the ohmic contacts as well as in the channel of the device are shown. B) The tensile stress required to induce additional crack growth for given crack lengths is also shown.


86 A B Figure 3 6 Stepped stressing of an AlGaN/GaN HEMT and resulting crack fo rmation. A device was step stressed from V GS = 10V to V GS = 42V at 1V/min and with V DS = 15V resulted in degradation of the device. A) A reduction in the saturation current was observed as a result of this stressing and is represented by the V DS I DS family of curves. B) An increase in the gate leakage current is also observed and is represented by the V GS I GS characteristics of the gate contact.


87 A B Figure 3 7 The resulting crack distribution and associated tensile stress for crack growth in the stressed HEMT. After stressing, this device was deprocessed and the distribution of nanocracks observed on its surface was determined. A) This distribution of cracks can be used to estimate the biaxial tensile stress in the AlGaN which used to grow these longer features. B)


88 CHAPTER 4 PROGRESS IN THE ANALYSIS OF THE RELIABILITY OF THE GATE ELECTRODE A common factor affecting the overall reliability of AlGaN/GaN HEMTs during DC stressing in the field is the degradation of the gate electrode of t hese devices. This degradation generally manifests itself as a dramatic increase in the reverse biased leakage current through the gate contact observed as devices are stressed at progressively higher electric fields. The structural change which is gen eraly accepted as the underlying cause of this increase in current is some reaction which induces shorting of the gate contact to the 2DEG present below the AlGaN epitaxial layer. This shorting can be caused by the formation and refilling of cracks under the gate electrode, metal diffusion down threading dislocations leading to an observed increase in trap centers or even complex electrochemical reactions between the gate contact, the operating ambient, and the semiconducting epitaxial layers of the HEMT d evice. This chapter will detail the progress made in understanding the mechanisms of formation associated with this defect in several different HEMT devices. It will begin with an analysis of early work which was performed on Pt gate devices, where the r elatively unreactive metal which forms the gate contact leads to degradation only at very high fields. In these systems, the large inverse piezoelectric strain can lead to the formation of mechanical defects under the gate contact which short out the 2DEG This section will be followed by a discussion of the degradation of HEMTs which possess a nickel gate electrode. Because of nickel's enhanced reactivity in comparison to platinum, these devices fail at lower fields than their noble metal contacted coun terparts, but this failure is much more gradual and progresses by a completely dissimilar mechanism. This chapter will be concluded with a section detailing the efforts made in surface characterization of various HEMT devices with specific attention paid to the observation of defect formation under the gate and its stochastic nature.


89 4.1 Inverse Piezoelectric Strain and the Reliability of Pt Gate HEMTs When an electric field is applied to a highly polar crystal, the bonds of this crystal orient themselves in the direction of the electric field, resulting in the presence of uncancelled charge at the interfaces of the crystal. This reorientation is what ultimately gives rise to the 2DEG in an AlGaN/GaN HEMT [74]. When mechanical strain is applied to a po lar crystal, this pushes the alternating atomic planes which give rise to polar bonding closer to one another, increasing the induced electric field within the crystal and the number of charges at the interfaces of the crystal. This electronic response to strain is called the Piezoelectric Effect. A less common phenomenon is the application of strain to a crystal as the result of an applied bias which changes the electric field. This is called the Inverse Piezoelectric Effect. [85]. The strain experi enced by a crystal may be expressed with the following pair of coupled equations, where S is the vector representing the strain in all directions applied to the crystal. The symbol Y is the tensor representing the elastic modulus in all directions, while vector representing the stress in all directions (in cm/cm) within the crystal (in Pa). The symbol d I is the tensor which represents the inverse piezoelectric coefficients (in pm/V) and F is the vector form of the electric field (in V/cm) [86]. (4 1) Because the majority of the field is present between the gate electrode and the 2DEG, the AlGaN layer of the HEMT tends to experience the most stress. For a relaxed system, the first term in the equation can be ignored, as no ext ernal stresses are applied to the crystal which would induce strain. This is not the case for AlGaN/GaN epitaxial structures. Any lattice mismatch associated with epitaxial growth manifests itself in the first term in the above equation as the AlGaN la yer


90 is not thick enough to induce relaxation through defect formation This tensile stress may be calculated by comparing the "a" spacing of relaxed Wurtzite Al 0.28 Ga 0.72 N (3.17) to that of pseudomorphic Al 0.28 Ga 0.72 N (3.19) and multiplying the normalize d spacing difference between these two materials by the Young's Modulus. Current literature suggests that the strain resulting from this mismatch should be approximately equal to 1.94GPa [82]. Stress in the metal and passivation layers caused by therma l expansion coefficient mismatches or non ideal deposition conditions could also result in stress on the AlGaN layer. In order to calculate the stress resulting from the mismatches associated with thermal expansion, a first order model of the gate stack m ay be employed where a film stack comprised of the 2um GaN layer, the 14 nm AlGaN epitaxial thin film, and the deposited metal gate (20nm of Ni followed by approximately 500nm of Au) are considered. In this model, the Young's Modulus and thermal expansion coefficients of the Au and Ni (79 and 200 GPa for Young's Modulus, respectively;1.4x10 5 and 1.3x10 5 for the thermal expansion coefficient, respectively) are taken from standard texts [87], while the Young's Modulus of GaN and AlGaN are assumed to be rou ghly equal to 309GPa [81]. The thermal expansion coefficient for GaN is taken from literature (5.6x10 6 ), while the thermal expansion coefficient of AlGaN is derived from linear interpolation between GaN and AlN (5.6x10 6 ) [88]. The solution for the stre ss in this thin film is derived by taking the difference between the strain in the AlGaN due to thermal expansion and the "average strain" (a 0 ) of the multistack system, computed as a weighted average, as follows (where T 1 is the starting temperature, T 2 i s the ending temperature, t i is the thickness of a given film, E i is the Young's Modulus of that film and i is the unitless thermal expansion coefficient of that film) [89]. This "average strain" is shown below.


91 (4 2) The resulting stress due to thermal expansion coefficient mismatches derived from this "average stress" is minimal. For an o perating temperature of 500C, which is large compared to literature results, the computed stress due to mismatch is equal to only 60 MPa of biaxial tensile stress. Given this result, the effects of thermal expansion can be neglected. Because of the rela tive magnitudes of the inverse piezoelectric coefficients, AlGaN responds significantly to vertical electric fields between the gate and 2DEG. This response far outpaces stress due to thermal expansion coefficient mismatches, but is not as pronounced as t he stress induced by lattice mismatch in the AlGaN. The basal plane is stretched along the prism directions with increasing vertical electric field, putting the AlGaN layer in tension. Particularly high electric fields could, concievably, cause mechanica l damage. In order to calculate the biaxial tensile stress induced under the gate due to the inverse piezoelectric effect to a first order, the following equation is used, where d 31 is the biaxial component of the inverse piezoelectric tensor which intera cts with the vertical field (assumed to be equal to approximately 1.57pm/V in AlGaN, based off of linear interpolation between GaN and AlN) [90], and c 31 is the elastic stiffness constant (assumed to be equal to 103GPa by assuming that the elastic stiffnes s constant for AlGaN is similar to that of GaN, which seems reasonable given that they share very similar Young's Moduli) [91]. (4 3 ) The resulting stress from this interaction is 179MPa at 20V, which is 9% of the stress present in the Al GaN film as a result of the formation of a pseudomorphic layer of AlGaN on top of GaN. This is a substantial amount of biaxial stress and it stands to reason that piezoelectric


92 stress could induce a fracture event within the AlGaN layer if the conditions were right (perhaps at a sharp corner, like the edge of the gate). Joh, delAlamo and coworkers have posited that the inverse piezoelectric effect and the strains associated with this effect might lead to the formation of crystalline defects within the AlG aN layer. As these defects form and agglomerate, they induce deep level traps which allow for current conduction between the gate contact and 2DEG through the AlGaN. As these defects continue to agglomerate and as the stress increases, plastic deformatio n and crack formation may occur, resulting in more current conduction. They further posited that this defect formation may be the cause of substantial degradation observed in stressed HEMT devices which are stressed in conditions where a substantial later al field is not present to aid in hot electron based degradation [92]. If the the inverse piezoelectric effect is the cause of defect formation during off state stressing, the degradation should depend upon the vertical field present in the AlGaN semicond uctor underlying the gate electrode.. There should also exist some critical voltage at which a substantial change in the gate leakage current occurs, as mechanical defects form. This critical voltage should also change with any mechanical stress applied to the HEMT. Increasing the tensile stress present within the HEMT should reduce the critical voltage while increasing compressive stress should increase the critical voltage required for defect formation. Finally, because the inverse piezoelectric effec t should be highly dependant upon the field present within the device, defects should form preferentially on the drain side of the gate. Joh and coworkers stressed a group of devices at V DS =0V with V GS stepped from 10V to 50V at a rate of 1V/s on an AlG aN/GaN HEMT with a Pt gate at room temperature and in an N 2


93 ambient. The results of this off state stressing was a substantial increase in reverse bias leakage current accompanied by a moderate increase in the forward biased current as well. This measu rement was repeated at variable temperatures both before and after stressing in order to determine an activation energy for the increase in the forward and reverse currents characteristic to these rectifying contacts. The activation energy was determined by fitting the changing diode current density, J(T), to the following equation, where J 0 is a pre exponential constant, k is Boltzmann's constant, T is the temperature (in Kelvins), and E A is an activation energy (in eV/K). (4 4 ) Th e activation energy associated with forward biasing did not change from its starting value of 0.26eV as a result of stressing, indicating that the degradation observed does not result from an alteration of the built in voltage of the device. However, the value of the activation energy of the reverse current changed from a value of 0.45eV to a value of 0.003eV, indicating that some event occured during stressing which fundamentally altered the mechanism of leakage through the gate contact. The stressed HEMT device also demonstrated substantial reductions in I DS after degradation as well as a reduction in transconductance. Because stressing at V DS =0 results in minimal current through the channel in comparsion to "on" state stressing, the collapse in I DS and reduction of transconductance cannot be ascribed to hot electrons. This degradation was attributed in some part to an increase in the drain resistance, which fits well with a field effect being the cause of degradation [93]. The leakage current rema ined low but increased exponentially with V GS below a critical voltage (V CRIT ). This was followed by a range over which this exponential increase in gate


94 current occurred at a much higher rate and resulted in an increase in current of over two orders of m agnitude. After this region of rapidly increasing gate leakage current, the exponential rate of increase returned to its original value. This experiment was repeated in the "on" state, with a variable I DS The same critical voltage is observed as V GS w as stepped from low to high values, but the critical voltage did not appear to vary in any rational manner as I DS changed. When stressing occurs in "off" mode (V DS no longer equal to zero) V CRIT decreases as V GS and the associated maximum vertical electri c field density under the gate increases. This was demonstrated by Joh and coworkers in an experiment where they applied varying potentials between the gate and source electrodes and proceeded to step the potential at 1V/min between the gate and drain fro m values corresponding to V DS =0 up to 50V [95]. Another experiment with V DS =0 was performed on HEMTs with gatelengths that varied from 1.50 GS was stepped at a rate of 1V/s from 0V up to a value of 50V. As was predicted by Joh and coworkers, the value of V CRIT decreases with decreasing gate length and increasing maximum electric field. This decrease appears to be particularl y substantial at aggressively scaled gatelengths [96]. As shown schematically in Figure 4 3, In order to validate the cause of the degradation observed in the studies described previously, Joh and coworkers performed pulsed stressing on the gate electr ode of a Pt gated HEMT with V DS =0. This pulsed stressing consisted of three cycles. In each cycle, there is a two hour stressing period, where the gate of the device is subjected to a potential of 40V and degradation of the HEMT should occur, followed b y a one hour recovery period where the gate of the device is set to a value of 0V. At each minute of the measurement, the gate current in "off" mode, as well as the gate and drain currents in "off" mode were measured. Because the stress voltage was of a higher magnitude than V CRIT off mode I G


95 rapidly increased over the first stressing period and stabilized at its maximum value for each stressing period. On mode I D however, degraded exponentially over time during each stressing period. When the device cycled into recovery, current collapse was reduced as I D increased at a rate corresponding to a sum of inverse exponentials. It returned to its pre recovery value whenever a new stressing cycle began and the maximum current achieved at the end of the reco very period decreased during each cycle. This behavior is consistent with trap formation in the AlGaN layer which decreases the conductivity of the 2DEG and causes a recoverable current collapse as a result [97]. As shown in Figure 4 4, Joh and coworkers have documented the formation of pitting along the drain edge of the gate electrode during on mode stressing (V DS =40V, I G =250mA/mm): a phenomenon consistent with a degradation process caused by the inverse piezoelectric effect. These pitting defects were evaluated using cross sectional transmission electron microscopy (XTEM) [98] as well as scanning electron microscopy (SEM) and AFM of a surface deprocessed with a combination of wet chemical treatments, which will be described in greater detail later in t his work. With this method, Joh and coworers were able to observe the formation of metal filled pits on the drain edge of the gate electrodes of Pt gated AlGaN/GaN HEMTs stressed in "off" mode (V GS = 7V) with V DG stepped from 8V to 50V at a rate of 1V/s. These pits grow with increasing gate to drain voltage when the HEMT is biased in "off" mode, which is consistent with some event driven by the electric field. The leakage through the gate appears to follow the density of pitting defects [99]. The results above indicate that the degradation of devices in "off" mode, "on" mode and when V DS =0 is driven by the vertical field present in the AlGaN layer and not by current density


96 through the 2DEG, as is the case for hot electron effects induced in the "semi on" mode. The results do not rule out, however, that the gate current density, rather than the density of current in the 2DEG might influence the reaction, so a positive correlation with the inverse piezoelectric effect cannot necessarily be assumed for Pt g ate HEMTs. A recent detrapping study performed by Kuball and coworkers on a Pt gate HEMT posessing a Al 0.26 Ga 0.76 N/GaN epitaxial stack with no capping n GaN layer does suggest that the defect which forms is related to metal diffusion. This HEMT was s tressed in "off" mode (V DS =30V,V GS = 5V) from room temperature to 150C and for times ranging from 0h to 40h. Detrapping analysis consisted of a voltage driven measurement where the sample was put in a "filled" state (V GS = 10V, V DS =0V) for 1s (in order to induce trap filling) and an "transient" state (V GS =1V, V DS =.5V) for 1000s (to observe trap emptying) [100,101]. This methodology is demonstrated, schematically, in Figure 4 5. Kuball and coworkers observed the presence of a band of traps which increased dramatically over the course of stressing and progressed from a peak trap energy of 0.45eV to 0.65eV. The magnitude of this trapping signature, C (in cm 3 ), (which should, itself, be linearly related to the trap density) indicates that the trap concentrat ion varies with respect to the following equation, representing one dimensional diffusion from a pseudo infinite source [102]. (4 5) In this equation, z is the characteristic tunneling length from the 2DEG into the AlGaN layer (3nm), D is the diffusivity (in cm 2 /s), of diffusing, defect forming species (likely Pt) down into the AlGaN layer, t AlGaN is the time in seconds and S is the surface concentration (in cm 3 ) of defect forming species. Variation of the observed diffusivity with tempe rature was fitted with an Arrhenius Function, as seen in the equation below, where D 0 is the diffusivity at an infinite


97 temperature (in cm 2 /s) and E A is the activation energy (in eV) for diffusion and k is Boltzmann's constant. (4 6) The activation energy for diffusion was found to be roughly equal to 0.23eV. This corresponds well with previous studies of oxygen diffusion down dislocation lines during annealing by Pearton and coworkers [103]. Kuball and coworkers suggest that this indic ates that pitting related defects form via metal diffusion down pre existing threading dislocation lines under the gate. This diffusion is enhanced by piezoelectric strain induced by the vertical field. It bears noting that this process is similar to deg radation processes observed in GaAs HEMTs formed in the early years of research into the material, when dislocation densities were large [104,105]. A few general trends detailed above for Pt gate HEMTs also hold true for other gate technologies and bear s pecial emphasis. Firstly, the sharp increases in gate current observed during stepped stressing of V GS are commonly observed in other gate stack technologies. The voltage at which this transistion from low leakage current to high leakage current occurs i s variable from one technology to the other. Furthermore, the manner in which gate current increases, or whether it increases at all, after the onset of the critical voltage varies from one architecture to the next [106]. The increases in observed drain resistance as a result of gate shorting is also common among various gate stack technologies. This effect is induced by the formation of an alternative path for current to travel out of the device, that being through the gate rather than through the drain This path for leakage current reduces the observed current through the drain of the device, for a given voltage, resulting in a lower observed resistance via Ohm's Law.


98 Lastly, it bears noting that Pt gated HEMTs are not the only architecture where the gate leakage current has been correlated with the formation of some form of defect bridging the gate and the underlying semiconductor material. In Pt gated devices, this sort of defect is induced by mechanical failure, specifically cracking, and refil ling with gate or passivation material. In other devices, however, electrochemistry or diffusion may also be involved. It is generally agreed upon, however, that the electrostatic field present between the gate electrode and the 2DEG drives any reactions resonsible for the degradation of the gate electrode. 4.2 Reliability of Ni Gate HEMTs Initial studies of the failure of the Ni gate electrode of an AlGaN/GaN HEMT was performed by Chang and coworkers on devices similar to those which will be used in the se studies. A device with a 140nm gatelength AlGaN/GaN HEMT with V GS step stressed from 10V to 42V at 1V/s and V DS held at 5V. The failed device exhibited the same sharp increase in gate leakage current in off mode as was the case for Pt gated devices studied previously by delAlamo and co workers. This non recoverable spike in gate leakage current is also evident in the plot of I GS versus V GS for the HEMT device, which demonstrates that the reverse biased leakage current in the gate contact of the dev ice is dramatically increased after stressing [107]. This is also similar to results observed by delAlamo and coworkers. Curiously, the observed gate leakage did not stabilize after this initial increase after V CRIT as was the case in Pt gated HEMTs. I nstead, the observed off state leakage current continued to increase. It also bears noting that as a general trend in these devices, depending upon the stressing conditions, the relative increase in gate current observed after exceeding V CRIT is highly va riable. Generally, the relative increase in gate current at V CRIT increases with increasing voltage between the source and drain electrodes of a Ni gated HEMT [108]. Both of these observations suggest that the mechanism of degradation for a HEMT device p ossessing a


99 nickel gate may be fundamentally different from the mechanism of degradation of Pt gated HEMTs. An example of a stressing experiment similar to the one described in Chang's work is demonstrated in Figure 4 6. Chang and coworkers also observed a decrease in the intensity of the peak in the photoluminescent spectrum correlating to the band to band radiative recombination event for GaN after stressing. This decrease in the PL intensity observed after stressing suggests the formation of non radia tive recombination centers as a result of off state stressing, which would quench the band to band, radiative recombination peak. The relative decrease in peak intensity was maximized near the gate and drain electrodes, which correlates well with the expe cted position of the degradation present within the device. It is often the case that a native oxide is formed on the surface of the AlGaN layer of a HEMT after all epitaxial layers have been deposited and vacuum is broken to move the device from epitaxia l processing equipment to lithography and metallization. In the devices studied as part of Chang's work as well as in this work, this interfacial native oxide layer is approximately 1.5nm thick. XTEM of the gate electrode of Chang's Ni gated device did not demonstrate the existence of a crack at the drain side of the gate electrode, as was observed previously in Pt gate devices. Instead, intermixing and dissolution of the interfacial layer was observed. Douglas and coworkers observed decreases in V CRIT in HEMTs which used a Ni gate rather than the Pt gates studied by Joh and coworkers. The gatelength of the Ni electrode was scaled from 100nm to 170nm and the devices were tested at V DS =0 with V GS stepped at a rate of 1V/min from 5V to 45V. While the critical voltage does change dramatically with HEMT gatelength, as was demonstrated previously by delAlamo and coworkers, the simulated critical vertical electric field within the device did not appear to increase, remaining fixed at

PAGE 100

100 approximately 2.5 MV/c m 2 further suggesting that degradation is induced by the vertical electric field within the HEMT. In additional studies, where stepped stressing of V GS from 10V to 42V was performed on devices at a variable V DS ranging from 0V to 15V, Douglas and cowor kers observed that the critical voltage at which a Ni gated HEMT degrades is related to the voltage difference between the drain and source electrodes as well as the voltage difference between the gate and source electrodes, with devices stressed at a high er V DS having a lower critical V GS However, if the critical voltage for failure was analyzed in terms of the voltage between the gate and the drain rather than the gate and the source, V CRIT remained fixed at a similar number as was observed in the case of stepped stressing at V DS =0V [109]. The field which resulted between the 2DEG and the gate of the HEMT was comprable to the field calculated in the case of stepped stressing with V DS =0V, suggesting that the field present between these two high conductiv ity layers is the driving force for degradation in Ni gated HEMTs. Using Equation 13, in a similar fashion to how it was described previously, a field resulting from a voltage of 22V (the approximate critical voltage) results in a value of 197MPa. This value can be related to the fracture toughness of a 2D channeling crack and, through it, the surface energy of AlGaN via another equation, similar to Griffith's Equation [110]. It shares all the same variables (even Z is the same 1.976), except one, whi ch is h: the depth of the channeling crack. (4 7 ) This is not enough stress to induce 14 nm of crack growth, which is, at minimum, 5GPa. It stands to reason that fracture is not a direct driving mechanism for defect formation in these dev ices. It also bears noting that Douglas and coworkers performed additional studies of Ni gated HEMT failure at voltages lower than V CRIT in an unpublished study. Surprsingly, they

PAGE 101

101 discovered that the spike in leakage current which occurs at the observed critial voltage during stepped stressing can also occur at voltages much less than V CRIT This time dependant critical voltage suggests, firstly, that degradation in stepped stressing is not dependant only on the field, but is also dependant upon time. T his implies that the two factors may be convolved in stepped stressing experiments, making analysis difficult. The range of voltages at which this spike in leakage current is observed is so broad that shorting of the AlGaN due purely to cracking seems unl ikely, as well. Overall, the preponderance of data suggests that some mechanism other than strain induced material failure at the gate electrode is the primary driving force for degradation at V CRIT and beyond. Evidence exists that degradation kinetics are different in HEMTs which utilize nickel gate materials. Lo and coworkers compared the degradation of Ni gated and Pt gated Al 0.25 GaN 0.75 /GaN 1 DS =5V with V GS varying from 10V to 100V. HEMTs possessing a Ni gate appear to degrade at a much faster rate, at a lower critical voltage and to higher absolute leakage currents than the Pt gated HEMTs. Also, current collapse is more pronounced in these devices than in Pt gate HEMTs. Lo and coworkers also evaluated the X Ray Photoemission Spectroscopy of as formed Ni and Pt gate contacts on GaN as well as similar contacts which had been annealed for 30 min at 300C. XPS indicated no change in bonding configuration for the Pt gate electrodes. However, it indicated that, during annealing, the Ni appears to diffuse into the GaN semiconductor. The oxygen present in the interfacial GaO layer reacts with t he Ni preferentially to form nickel oxygen bonds [111]. A similar effect was noted by Burnham and coworkers, who stressed Ni gate AlGaN/GaN HEMTs in "on" mode with V DS =15V and V GS set in order to achieve the I DS

PAGE 102

102 corresponding to a channel temperature whi ch varied from 136C to 320C and which was stepped at a rate of 23C/day. Both samples were stressed until an 11% collapse in I DS was observed. HEMTs which were stressed in an air ambient failed much more rapidly and at a lower temperature (205C and 10 0h) than HEMTs which were stressed in an N 2 ambient, which failed at a temperature of 320C after over 350h of stressing. They observed with XTEM that a defect had manifested itself over the full length of the gate contact in both devices rather than only at the drain electrode. They observed nickel interdiffusion into the semiconductor along the full gate lengths of devices stressed in both air and N 2 not just at the drain side of the gate edge as had been noted earlier in the pitting studies by Joh and coworkers. A substantial oxygen content was observed in devices stressed in an air ambient, suggesting oxygen diffusion and reaction with the interdiffusing nickel [112]. Holzworth and coworkers performed XTEM on a Ni gated devices failed in an oxygen ambient at room temperature. The devices were stressed at V DS of 5V while V GS was stepped from 5V and 10V to 42V at 1V/min increments. They observed that samples degraded under these conditions possessed the characteristic decreases in I DMAX and incr eases in I GS which corresponded to defect formation in previous studies of defects Ni and Pt gated electrodes. They further observed that devices which did not degrade rapidly and which did not experience significant drops in I DMAX also did not show evid ence of defect formation in XTEM. Devices which did degrade at an accelerated rate and which manifested dramatic decreases in observed I DMAX however, did yield evidence of defect formation during analysis with XTEM. These defects were of a profoundly di fferent nature than the defects previously observed in XTEM of Pt gated devices by delAlamo and coworkers [113]. As shown in Figure 4 7, the defect which formed as a result of off state stressing is not confined to the drain side of

PAGE 103

103 the gate contact of th e device, as was previously observed. The gate contact of the degraded device was also pushed out of contact with the AlGaN epitaxial layer, as well as with the nitride passivation layer which had covered its surface, by well over 20nm. This defect also spanned the full thickness of the AlGaN, resulting in direct shorting of the gate electrode of the device to the 2DEG. The defect, itself, was determined to be amorphous via analysis with in situ electron diffraction. EELS mapping of the defect observed in cross section demonstrated that the defect, itself, was comprised of a mixture of nickel and oxygen. The shape of the defect itself was commensurate with the electric field lines present in the HEMT device and modeled with the Florida Object Oriented Device Simulator (FLOODS). A variety of conclusions about the mechanism of formation for pitting defects in Ni gated HEMTs can be drawn from the observations made in this study. Firstly, it is apparent that, unlike degradation in Pt gated devices, th e formation of the defect under the gate is not driven purely by cracking due to piezoelectric strain and that some electrochemistry must be involved which gives rise to the formation of a nickel oxide inclusion. This inclusion is, volumetrically, larger than the nickel liner layer which it formed from and it pushes the gate electrode out of contact with the AlGaN layer. This defect still appears dependant upon electric field in some capacity, as its morphology matches closely with the magnitude of vertic al field present between the gate electrode and the 2DEG of the device. This correlates well with other studies of Ni gated AlGaN/GaN HEMTs, which have demonstrated that changes in device architecture which can reduce the residual electric field around th e gate, such as the introduction of a field plate, can greatly enhance the performance of devi c es during stressing, resulting in a much higher V CRIT than is achievable without the aid of a field plate [114].

PAGE 104

104 4.3 Efforts in Surface Characterization of AlGaN /GaN HEMTs While efforts to characterize device degradation with cross sectional TEM have met with success in both Pt gated and Ni gated HEMT architectures, efforts to quantify defect formation with these techniques have not been as effective because TEM methods are not as sensitive to stochastic effects, as has been described previously. DelAlamo and coworkers developed experimental methods to better analyze the stochastic effects which play a part in the degradation process by utilizing a combination of novel etch chemistries as well as top down SEM as well as AFM measurements. Their method begins with deprocessing of the device structure in order to expose the epitaxial AlGaN layer to inspection with SEM or AFM. This deprocessing begins with an expo sure in HF, meant to eliminate the passivation nitride which sits atop of the metal layers present within a device as well as the AlGaN which comprises the channel. This exposure in HF is briefly followed by an exposure in Aqua Regia, meant to eliminate the Au alloyed and Pt alloyed ohmic and gate metal contacts, respectively. Aqua regia was chosen over a FeCN/KI based chemistry, described earlier, likely because of its effectiveness in etching platinum metal layers, which FeCN/KI is incapable of accomp lishing. This aggressive metal etchant does have its draw backs, however. Aqua regia does not exhibit the high selectivity of FeCN/KI between metal layers and the underlying AlGaN. Rather, it tends to attack the AlGaN, resulting in the formation of etch pits and other surface topographies which can result in false positives when pits are being counted, measured and analyzed. Despite this drawback, this Aqua Regia based deprocessing solution has been used to great effect in studies of Pt gated as well as Ni gated HEMTs [115]. Makaram and coworkers used this methodology in conjunction with SEM to identify the formation of crack based pitting in Pt gated devices and to find a direct relationship between the pit densities with the leakage

PAGE 105

105 current observed i n these devices. They also made use of AFM to demonstrate that the size of crack induced pits increases with increasing applied reverse bias to the gate electrode of these Pt gated devices, indicating that degradation of the device continued despite the f act that large increases in the gate leakage current is generally not observed in Pt gated devices. The deprocessing techniques used by Makaram and coworkers have found application in HEMTs with nickel gates. Gao and coworkers used the same deprocessing method as was described previously to analyze pit formation in Ni gated HEMT devices with a 1m gatelength in an air ambient [116]. These devices were biased with V DS equal to 5V and V GS stepped from 5V to 40V. This biasing scheme resulted in an asymme tric electric field. In these devices, the field was highest between the gate electrode and the 2DEG, specifically at the drain side of the device, where the difference in potential between the gate and 2DEG was maximized. Not surprisingly, this device m anifested the same gate leakage current dependance on stepped gate to source voltage as had been observed in devices previously, with a sharp increase in current at V CRIT Deprocessing by Gao and coworkers revealed that the increase in gate leakage curr ent during stepped stressing was induced by pitting of the AlGaN layer under the gate electrode. Furthermore, this pitting was closely confined to the drain side of the gate electrode, where the field was maximized. They further demonstrated that the phy sical location of the pitting under the gate could be switched from one edge of the gate to the other if the biasing scheme was changed such that the source of the device was biased at the higher voltage than the drain (effectively making the source the dr ain and the drain the source). This further illustrated that electric field has a major impact on the reactivity of the gate metal with the underlying AlGaN/GaN as well as the ambient.

PAGE 106

106 The chemical deprocessing methodology described above has also been u sed in conjunction with non destructive techniques such as Electroluminescence spectroscopy, to reveal the impact of stepped stressing and associated defect formation under off state conditions on the current transport through a HEMT. Excitons in AlGaN Ga N, like many other compound semiconductors, undergo direct, radiative, band to band recombination, where the momentum of the constituent electron and hole of the exciton is conserved and all energy is transferred to a photon. This photon is free to exit t he material system and be collected by a sensing instrument like, in the case of experiments done by Bajo and coworkers, an astronomy grade CCD camera in line with a 50X magnification microscope objective [117]. Bajo and coworkers stressed a single Ni g ated device with 600nm gatelength in "off" mode, with V DS equal to 30V and with V GS equal to 15V for a 760s time period and viewed it through the CCD equipped microscope detailed above. They observed the formation of a group of bright spots of EL intensi ty between the gate and drain of the device under stress. These bright spots matched closely with the location of pits present under the drain side of the gate of this device which were very shallow (being 2 4nm deep) and very broad (being 100 200nm wide) In order to understand the correlation between the bright spots which Bajo and coworkers observed with EL and the pitting which they observed with AFM, it is important to consider what leads to variation in EL intensity. The EL intensity is directly proportional to the number of direct recombination events which occur in a given region of a material. Not surprisingly, the number of direct recombination events which occurs in a given time period and in a given material volume is related to the product of the concentrations of electrons and holes in the

PAGE 107

107 material as well as a general "attempt frequency". More EL intensity arises from areas where more electrons are present in the material and, therefore, more recombination events occur. In the device observed by Bajo and coworkers, the path of least resistance for leakage current passing through the device was from the drain of the device, through the drain side of the channel and, then, through the areas of the gate where pitting had occured. So, ele ctron current in the 2DEG crowded around these pitted regions. The random recombination events where these crowded electrons recombined with holes present in the GaN resulted in the formation of luminescent "spots" in the EL image. Bajo's obvservation of these "spots" in EL suggests that leakage current through the gate electrode is related to the presence of pitting defects and, as is the case with Pt gated devices, that the leakage current increases with increasing pit density. All of the research abov e paints a very clear picture of the degradation which appears to occur in a Ni gate HEMT after long periods of stressing or when steppe stressing occurs up to potentials well in excess of the critical voltage. Very little materials characterization has o ccurred at the critical voltage, itself, however. This might warrant further study.

PAGE 108

108 Figure 4 1 Piezoelectric strain (white arrows) resulting from a vertically applied electric field (black line) in an AlGaN/GaN HEMT. Figure 4 2 The dependance of g ate leakage current with increasing electrostatic stress. The electrostatic stress was applied between the gate and the source electrodes at a constant rate (V GS = 10V to V GS = 42V at 1V/min). The regions of potential lower than V CRIT equal to V CRIT and greater than V CRIT are highlighted.

PAGE 109

109 Figure 4 3 Pulsed stress experiments on an AlGaN/GaN HEMT. 3h stressing periods at V DS =0V and V GS = 40V were followed by a 1h detrapping periods to determine the effect of traps at the gate AlGaN interface on the cur rent collapse and gate leakage current [92]. Figure 4 4 A schematic image of a crack like defect formed at the drain side of the gate. This sort of defect could form of an AlGaN/GaN HEMT stressed at V DS =40V with V GS set to yield I DS = 250mA/mm [98]. The crack like defect formed on the drain side of the gate and refilled with passivation dielectric, silicon dioxide or gate metal, depending upon the defect imaged.

PAGE 110

1 10 Figure 4 5 The transient electrical measurement performed by Kubal and coworkers. Thi s measurement was performed on AlGaN/GaN HEMTs prior to and during stressing to measure traps present in the device. This measurement beging with a UV light pulse, meant to empty all traps present within the device, followed by a 1s trap filling pulse on the gate electrode while the drain and source are shorted to ground. An emission transient is observed when the device is switched into "on" mode.

PAGE 111

111 Figure 4 6 An exemplary stressing experiment performed on a Ni gated AlGaN/GaN HEMT. In this case, V D S = 5V and V GS was stepped from at 1V/min. The increase in gate current at the critical voltage, as well as the continuing increase in gate current beyond this voltage, is pictured. Figure 4 7 BF TEM image of a defect under the gate of an AlGaN/GaN HEMT. The electrochemical reaction of nickel and the subsequent consumption and pitting of the AlGaN surface results in an enhanced leakage current through the gate.

PAGE 112

112 CHAPTER 5 RELIABILITY OF THE GATE CONTACT As was discussed in the previous chapter, st epped stressing has become one of the most popular techniques utilized in understanding the nature of degradation in AlGaN/GaN HEMTs. It is utilized by many different organizations and has been used to examine failure mechanisms on many different device a rchitectures. While the technique has been used to great effect in understanding the operation and degradation of Pt gated HEMT structures in the field, stepped stressing is only just beginning to be utilized extensively in the study of failure in Ni gate d devices. Research performed previously by various authors has demonstrated that the failure of these Ni gated devices is unlike what has been observed previously in HEMTs formed with a Pt gate, where cracking induced by the inverse piezoelectric effect seems to be the driving force for defect formation as high fields on the drain side of the gate induce the formation of grooves and cracks which can refill with passivation material or gate metal itself, shorting out the gate to the 2DEG. In Ni gated dev ices, this same shorting event occurs (and appears to be heralded by a spike in gate leakage current at V CRIT ), but it does not appear to be induced by cracking between the 2DEG and gate contact of the device. Rather, a new defect forms, comprised of some mixture of nickel and oxygen as well as trace concentrations of aluminum and gallium. The formation of this defect appears to be mediated by electric field, given that the critical voltage required to induce a large increase in gate leakage current inc reases for larger gate lengths, maintaining the critical field required for defect formation (as is the case for Pt gated devices). However, defect formation also appears to be mediated by the ambient in which the stressing occurs, with ambients capable o f supplying a steady flow of oxygen to the sample inducing failure of the device much more rapidly than inert ambient. Studies performed by

PAGE 113

113 various authors also suggest that the interface present between the metal gate and the epitaxial AlGaN layer of the device plays a major part in defect formation and failure of the device. Mixing at this interface has been posited as a possible cause of increasing leakage current at the critical voltage and the thickness of the interfacial native oxide layer separatin g the gate metal from the AlGaN has been shown to impact the formation of pitting defects by diffusion. In order to fully understand the various mechanisms which may play a part in defect formation in these Ni gated devices, the behavior of example device s during various stages of stepped stressing must be studied. The performance of an AlGaN/GaN HEMT device during stepped stressing can be effectively separated into three separate regimes. The first of these regimes is operation at voltages below V CRIT where no permanent increase in leakage current is observed. Of the three regimes, defect formation at voltages well above V CRIT is, probably, the most well documented and well understood. It is degradation within this regime which led to the formation of the nickel oxide based defects previously observed during stressing of various Ni gated HEMTs. No direct studies of Ni gated HEMT devices have been performed where these devices were stressed only up to V CRIT This regime is, probably, the least well un derstood for the case of defect formation in AlGaN/GaN HEMTs. In order to better understand the degradation and overall change in behavior of AlGaN/GaN HEMTs in these three regimes, three separate devices were stressed in off mode, with V DS =5V and with V GS stepped at a rate of 1V/min from a starting voltage of V GS = 5V to a variable ending voltage with a maximum of V GS = 42V. The first of these devices was stressed to a maximum potential between the gate and source of V GS = 16V. This ensured that this de vice did not exceed the critical voltage, maintaining the device under test in the first low voltage

PAGE 114

114 regime of stepped stressing. The second device was stressed up to a voltage of V GS = 16V, achieving V CRIT and ensuring that the second device had entered t he second regime of stepped stressing. A final device was stressed up to a voltage of V GS = 42V, well above the critical voltage and well into the third regime of operation for stepped stressing. A measurement of the gate leakage current passing through the gate electrode of the device as well as the currents passing through the source and drain was made after every stepped increase in voltage. This measurement occurred over a 1ms timespan, so the duty cycle of stressing may be assumed to be roughly equ al to 100%. As mentioned previously, all stressing and measurement of device characteristics was accomplished via the use of an HP4146C device tester, capable of independant measurements of current and voltage on all three device contacts, and a Techtroni x 370A curve tracer. The devices used in this study were formed using the processing described in the first chapter of this dissertation on a SiC substrate. All devices stressed as part of this study possessed a gatelength of 100nm. Analysis as part of this initial investigation involved the deprocessing of these three devices after stressing, using the method detailed in the middle section of the experimental chapter of this work. After deprocessing, these stressed devices were analyzed with plan view SEM (performed on an FEI Strata DB235 Focused Ion Beam Miller in electron control mode using the TLD in UHR imaging mode), which was utilized to quantify the aerial density of defects observed after stepped stressing. What proceeds from this point onwar d is a discussion of the behavior of AlGaN/GaN HEMTs under stepped stressing conditions in all three of these regimes, with the experiment described above acting as a means of reference and comparison from which a group of more detailed experiments studyin g behavior in each of these regimes is derived. The sections of this

PAGE 115

115 chapter will follow directly from the three separate regimes of stepped stressing discussed previously, starting at low potentials between the gate contact and the 2DEG and proceeding to ever increasing potential. 5.1 Device Quality Below the Critical Voltage As demonstrated in Figure 5 1, if the potential between the gate and 2DEG of the device is not increased above the critical voltage, substantial increases in gate leakage current d o not occur. This suggests that defect formation does not play a critical role in the characteristics of the device under stress at these low voltages, and the results from plan view SEM analysis of these devices appears to corroborate this supposition. No defects were observed after deprocessing and SEM analysis of this device. As is evident upon inspection of the representative image in Figure 5 1, the deprocessed channel is devoid of any features at all except for some dark spotting indicative of res idual organic contamination on the surface of the sample. It should be noted that this result is not an indication that defect formation is impossible at voltages below the critical voltage defined by a 1V/min stepped stressing experiment. As has been previously observed by Douglas and coworkers, enhanced leakage currents (and, presumably, the formation of defects under the gate electrode of the device) can be achieved at voltages below the critical voltage defined by the experiment above, indicating t hat the rate of reaction which leads to defect formation is non zero below V CRIT as it is defined in this experiment. The implication of this information in conjunction with the first result of the experiment described below is that, while the reaction which appears to induce defect formation is still active below the critical voltage as it is defined by this experiment, it is definitely mediated by field. Furthermore, the results of this part of the experiment also indicate that leakage current may be a

PAGE 116

116 good indicator of defect formation under the gate electrode of the device. No defects exist in the absence of enhanced leakage current from the gate electrode of the device. 5.2 Defect Formation at the Critical Voltage Figure 5 2A demonstrates that t he critical voltage, that point where the off state leakage current passing through the gate electrode of a HEMT device increases dramatically as potential is applied between the gate and the source, is the initial point where defect formation begins. Obs ervation of the leakage current graphed in the first part of the figure demonstrates that as the voltage approaches V CRIT for the device, the gate leakage current begins to increase. In this case, the leakage current passing through the device increased b y fully two orders of magnitude. The image of the deprocessed device after stressing, shown in Figure 5 2B, indicates that defect formation begins as this leakage current begins to increase. Interestingly, the defect formed in the device in this stressin g regime bears little to no resemblance to defects observed previously by delAlamo, Palacios, Kubal, and others. While previously observed pitting defects appeared at random sections along the gate stochastically, this defect appears to have formed along the majority of the entire gatelength during stressing. In contrast to typical pitting defect formation, which might have an aerial density of only a few percent in comparison to the total gate area, this defect was present along fully 66% of the full gat e width [118]. The morphology of this defect also appears strikingly different from the morphology of previously observed pitting defects. This defect manifests itself as a dark band of contrast in the secondary electron image, extending along the enti re gate width. This dark band of contrast is two toned for the 100nm gate width device. The darker toned portion of the band has a width which matches well with the width of that section of the gate which directly contacts the surface of the AlGaN epitax ial layer (100nm). The extent of the lighter banding to either side of this

PAGE 117

117 dark region matches well with the total width of the top structure of the gate electrode, which does not sit in direct contact with the AlGaN epitaxial region (500nm). The dark contrast of the band relative to the rest of the channel region of the AlGaN/GaN HEMT may be explained by considering how the secondary electron yield is influenced by different materials. Secondary electrons only have tens of eV of energy because they a re the result of inelastic collisions between the bombarding electrons of the SEM and the electron clouds of the atoms which comprise the materials which the bombarding electrons pass through. The yield of these secondary electrons is exponentially depend ant upon their depth in the film, because larger depths imply more material which must be travelled through in order to reach the top surface and have a chance of ejection into the vacuum and interaction with the secondary electron detector. Secondary ele ctron yield is also dependant upon the workfunction of the interface between the sample material and the vacuum. Materials with larger workfunctions (such as insulators) will have a dramatically lower secondary electron yield than materials with small wor kfunctions (such as metals). Thus, it seems plausible that the contrast observed in SEM is due to a thin insulating layer present on top of the AlGaN surface which absorbs electrons on their way to the interface between the sample and the vacuum and which also has a large workfunction which prevents the ejection of electrons out of the sample [119]. It bears noting that this defect could not be removed with an additional round of the HF/TFAC based deprocessing solution. After one hour of exposure in HF, this feature was finally eliminated. The appearance of this banded structure serves as further indication that the mechanism of defect formation during stepped stressing is not the same for Ni gated devices as for Pt gated devices studied previously by v arious authors. Indeed, defect formation in the HEMTs which were analyzed as part of this study appears to be comprised of two separate

PAGE 118

118 processes rather than a single process, as is the case in Pt gated devices: those being pit formation at high voltages and (at low voltages approaching V CRIT ) banding. The nature of this new banding defect is largely unknown, however. In order to understand the mechanism of its formation and the manner in which it degrades the performance of the device, further studies o f the composition and formation of this defect are required. Firstly, it is important to understand the electrical conditions, besides the sudden increase in gate leakage current at V CRIT which are associated with this banding defect. As shown in Figure 5 3A, the potential applied across the 2DEG betwen the source and drain electrodes also appears to impact the formation of the banding defect under the gate contact of the device. The devices under study in this case were processed in a similar fashion t o the processing scheme described previously, in Chapter 1, with the exception that the architecture was designed to accomodate a circular gate with a field plate attached, rather than a linear gate with no field plate attachment, as has been described pr eviously. The incorporation of this field plate results in higher V CRIT values (due to reduction in peak electric field) than those reported on non field plated devices, but the mechanics of electrical degradation appear to be similar in both cases. It i s also imporant to note that these devices possess a 2m gatelength, rather than the 100nm gatelengths of the devices mentioned just previously. So, the banding defect observed here extends over the entire region where the gate contact of the devices cont acts the surface of the AlGaN layer. In this case, the devices were stressed on the same HP4146C testing station as the devices studied in other sections of this dissertation and were step stressed with V GS = 6V and V DS stepped from 10V to 100V in 1V/m in increments. Interestingly, a similar device which was stressed at V DS = 0V, rather than V DS = 5V did not manifest banding as the critical voltage was attained. This stands in stark contrast to the

PAGE 119

119 device stressed with a 15V potential applied across th e source and drain electrodes, which manifested banding over a small region (10% percent of the total gate area) of the device footprint, as shown in Figure 5 3B. The device which manifested banding also manifested a much larger increase in total gate lea kage current as it achieved V CRIT The behavior noted here might be the general trend for banding: devices stressed with a potential applied between the source and drain electrodes of the device manifest banding while devices stressed with no potential ap plied between the source and drain do not manifest banding. It bears noting that both of these devices were stressed to potentials much larger than the critical voltage, but no pitting was observed in either device and the gate leakage current did not inc rease dramatically past V CRIT The implication of this result is that banding requires potential across the 2DEG in order to occur. It may be that the mechanism which influences banding defect formation requires the presence of free electrons within the 2DEG to aid in an electrochemical reaction which degrades the quality of the gate electrode. The formation of a banding defect can also be induced by means of thermal annealing of a Ni gated device. Figure 5 4B demonstrates that a banding defect, with mo rphology similar to that of the banding defects described previously was observed after deprocessing of a typical device annealed in a Lindberg Furnace at 500C for 30 minutes in an air ambient. This defect was present in devices with both 1m gate length s and 100nm gate lengths and the morphology of this defect was the same between both devices. In contrast to the results demonstrated previously during stressing of the 100nm gate devices, the banding defect observed on these devices only extends over the region where the gate electrode directly makes contact with the AlGaN epitaxial layer. The wider and lighter banding region which flanks the darkly banded

PAGE 120

120 region on the T gate device is absent. This seems to imply that banding induced by thermal anneali ng does not proceed with to the same equilibrium state, or with the same rapidity, or both as banding induced by electrostatic stressing. It stands to reason that a lighter feature which exhibits less contrast in SEM is induced by material which is a poor er insulator (and which, therefore, has a smaller workfunction and corresponding secondary electron yield), or which may be thinner, or both. In order to understand why this difference in morphology arises, it is important to consider the field present b etween the gate electrode and the 2DEG of the device in the case of electrical stressing and in the case of thermal annealing. The vertical electric field present 1nm below the surface of the AlGaN in these devices was simulated using the Florida Object O riented Device Simulator (FLOODS) and is shown both in the case of stepped stressing up to V CRIT and annealing at 500C as part of Figure 5 4C and Figure 5 4D, respectively. The device which was simulated had a gatelength of 100nm and had a total lengt h of 500nm at the top of the T gate. The crossbar of the T gate was raised 90nm above the surface of the AlGaN. The AlGaN layer, itself, was 15nm thick and formed on top of a 2m GaN buffer layer. In the case of stepped stressing up to V CRIT potentials of V DS = 5 V and V GS = 22 V were applied to the electrodes of the devices. In the case of annealing at 500C, potentials of 0 V were applied to all electrodes. During electrical stressing, the gate contact is placed under a high negative potential. Elec tric field permiates the device underneath the entire gatelength, but is maximized where the gate makes direct contact with the AlGaN. In this region, the vertical electric field achieves a value of 4.3x10 6 V/cm A lower field exists in the region of th e AlGaN where the gate overhangs the surface, which is unsurprising given that the electric field between two conducting

PAGE 121

121 layers is inversely proportional to the distance between those layers. The field under the edge of the T gate is equal to 4.8x10 5 V/cm In the case of thermal annealing, a null potential is applied to the gate electrode. As a result, the only field present between the gate electrode and the 2DEG arises from the built in field (and associated dipole) generated by the difference in wo rkfunction between the gate electrode and the AlGaN layer beneath it. So, field is only present where the gate electrode makes direct physical contact with the AlGaN epitaxial layer. The vertical field directly under the gate in this case is significantl y reduced, only achieving a value of 4.8x10 5 V/cm, and is more or less confined to the region directly under the gate. It bears noting that the field under the gate, in this case, has a value which is of the same order of magnitude as the vertical elect ric field present at the furthest extent of the T gate in the previous simulation. If vertical electric field is the driving factor for banding, this would suggest that the band which forms under the gate foot in an annealed device should have reacted les s than the bands which form under the raised portion of the T gate during annealing at 500C. The SEM micrographs in Figure 5 4A and Figure 5 4B suggest that this is, indeed, the case. The implication of the differences in morphology observed between def ect formation by means of electrical stressing and defect formation by means of thermal annealing is that both field and temperature are required to induce defect formation. Thus, banding is some sort of electrochemical reaction, similar to phenomena in o ther semiconductors such as time dependant dielectric breakdown [120]. It is dependant upon the field and ambient temperature present under the gate electrode of the device and it also appears to require a steady supply of free electrons within the 2DEG.

PAGE 122

122 In order to demonstrate causality between banding and the increase in gate leakage current rather than mere correlation, the I GS V GS current characteristics of the gate contact of an AlGaN / GaN HEMT were analyzed both before and after annealing at 500C f or 30 minutes. The results of this study can be seen in Figure 5 5. A slight (2X) increase in the gate leakage current is observed at large voltages after thermal annealing at 500C. Leakage current increased to a much lesser degree in the annealed devi ce than in the device which was electrostatically stressed, further suggesting that thermal annealing does not initiate degradation which is as severe as degradation which occurs when the vertical electric field is much larger. This fits well with observa tions of banding in SEM. SPM analysis was performed on the surface of a sample which formed the observed "banding" defect after annealing at 500C for 30min. This SPM was performed with a Veeco Dimension 3100 SPM in tapping mode. The integral gain was s et at 5.0 and proportional gain was also set at 5.0 for all testing while the photodiode amplitude setpoint was held at 350mV in order to ensure good coupling between the AFM tip and the surface. A TESP HAR AFM tip, available from Bruker Nanosurfaces, wit h a 10nm tip diameter, a 40 N/m spring constant, and a 5:1 aspect ratio was used for this study. The region studied as part of SPM analysis was the channel region of the device, represented as a 512x512 pixel topographic map, as measured by AFM. This 512 x512 array is representative of a 6.75m x 6.75m section of the channel region of the device. As shown in Figure 5 6A, the dark band of contrast which appeared in SEM and which matched closely with the dimensions of the 1m Ni gate is still visible durin g AFM analysis. A 3.5m wide region of the channel was integrated in order to analyze the thickness of this banded region. According to thie averaged linescan resulting from this analysis, shown in Figure 5 6B,

PAGE 123

123 this dark band of contrast corresponds with a raised region on the surface of the AlGaN epitaxial layer possessing a height of approximately 1.0 nm in height. It is interesting that this banded region appears to possess peaked edges, which correspond to the peaked nature of the vertical electric f ield present at the edges of the gate contacts of the simulated devices. The roughness of this region differs from the roughness of the surrounding deprocessed AlGaN. The RMS roughness of the channel region was determined by means of a second measureme nt with AFM, this time of a 1.5 x 1.5 m section inside and outside of the "banded" defect region within the channel. All other settings, except for the dimensions of the scanned area, were maintained between the two measurements. Analysis of the Root Me an Square (RMS) Surface Roughness was performed on both regions. The channel region of the device, in absence of banding, was found to have a surface roughness approximately equal to 3.7 with a standard deviation of 1.6, while the surface roughness of t he banded region was determined to be approximately equal to 3.4 with a standard deviation of 1.1. This would suggest that the material being formed as a result of thermal annealing is likely the product of a reaction at the epitaxial interface between the AlGaN and the Ni. It is interesting to note that the integrated AFM linescan of the band present under the gate manifests a pair of "peaks" where the edges of the contact were present. This agrees with simulation of the vertical electric field, where a pair of peaks was observed at the edges of the gate contact. This AFM analysis does not explain why this banded region is present in annealed and stressed samples in the first place, however. In order to explain the cause of band formation as well as the mechanism of its formation, information about the composition of this defective region is required. Arguably, the best means of extracting this information is direct observation

PAGE 124

124 of the banded defect with TEM. To this end, cross sectional TEM analysis was performed on two separate samples the interface between the AlGaN and the Ni gate of a HEMT device. A control sample was formed from a device which was received as formed from fab processing and a second sample was formed from a device which was an nealed at 500C for 30 min in order to induce the formation of a banding defect. This device was marked with fiduciaries which set off the gate region of the device, deprocessed using the etch chemistry described previously and coated with thermally evapo rated carbon to ensure that the gallium ion beam of the FIB did not interact with the banding defect present on the surface of the AlGaN. As shown in Figure 5 7A, HAADF STEM analysis of the control device, performed with a JEOL ARM200F, demonstrates the c onvenient presence of an interfacial layer between the AlGaN surface of the HEMT device and the nickel gate of this device. The thickness of this interfacial layer is approimately 1.5nm prior to annealing and approximately 2nm post anneal. This increase in the observed thickness of the interfacial region indicates that it may have reacted under the gate electrode of the device. The difference in observed thickness between as fabricated and annealed devices could be due to consumption of the AlGaN, which appears to be occuring after annealing, as shown in Figure 7 B. It could also be due to stochastic variations in the thickness of the interfacial layer. The author notes that the band height extracted from many linescanes of the banded region are much la rger than the "averaged" linescan presented in Figure 5 6B, often in excess of 2 nm. Stressing by means of thermal annealing or electrostatic biasing appears to induce some sort of chemical change within this interfacial layer which makes it much less per vious to the etch chemistries used for sample preparation in plan view SEM analysis and which induces

PAGE 125

125 higher leakage currents within the gate contact of the HEMT. This interfacial layer remains after deprocessing, as shown in Figure 5 7B. In order to ana lyze the order present with this interfacial layer, HRTEM was employed, as shown in Figure 5 8. Analysis of the Fast Fourier Transform (FFT) of the interfacial layer (processed with Image J software, which is open source and available at http://rsbweb.n suggests that this region is amorphous in nature both before and after the formation of the "banded' defect present under the gate. The FFT of this region is dominated by a bright central disk, indicative of the nearest neighbor distance assoc iated with the amorphous region. The bright array of spots associated with a single crystal sample are present within this FFT, but are dim and match closely with the positioning of spots in the FFT of the underlying AlGaN layer. This indicates that the observed dim diffraction spots in the FFT of the interface do not arise unique crystalline phases within this layer, but are more likely induced by thickness variation within the cross sectioned sample. In order to determine the chemical composition of t he interfacial layer, EELS and EDS were performed in conjunction with HAADF STEM of the interfacial region. As shown in Figure 5 9A, EELS analysis reveals that this interfacial layer is oxygen rich in comparison to the surrounding AlGaN regions and that t his remains the case both before and after annealing and deprocessing of the device structure. The EELS signal arising from N is present along with Ga this interfacial layer. However, it is noted that the brightness of the N and Ga signal is not as subst antial as the brightness in the AlGaN layer. Upon annealing, the signal arising from N in this layer is greatly diminished, as is shown in Figure 5 9B. The signal arising from Ga is also diminished and diffuse. The signal from O also appears to be diffu se. It appears that, after

PAGE 126

126 annealing, the native oxide which is present on the surface of the AlGaN reforms, rejecting N, and grows. Ga outdiffusion also appears to occur. The presence of an oxide interface between the AlGaN and Ni gate fits well with previous studies (via Atom Probe Tomography) of the interfacial region of the Ni gated HEMT gate stack. These studies suggested that a native oxide forms between these two regions if a sample is taken out of a high vacuum environment in between deposition of the AlGaN epitaxial layer of the HEMT via MBE and the deposition of the nickel gate of the device [113]. The native oxide present within the commerial device analyzed as part of the APT studies was thinner than the native oxide present in this study, however, and it is difficut to draw any conclusions from this previous research (other than that a native oxide interface should, likely, exist between the AlGaN and nickel) because of the different processing conditions inherent in manufacturing of these two different HEMT architectures. In order to ascertain more information about the chemical structure of this native oxide, EDS was performed in conjunction with HAADF STEM. EDS was chosen as a complementary technique to EELS because of the large overlap between the Gallium signal in the EELS spectra and the peaks associated with Aluminum and Nickel. The superposition of these two peaks made even qualitative analysis of the atomic concentrations of aluminum and nickel within the interfacial layer much mo re difficult. Figure 5 10 shows the EDS spectra associated with a STEM linescan across the interfacial layer in samples formed both before and after thermal annealing at 500C for 30 min of the HEMT device structure. The regions over which these linescan s f ormed are depicted in the inset figures contained within each linescan. Special attention should be paid to the x ray

PAGE 127

127 and the annealed device, and its relation to Three general regions of interest are contained within each of these EDS linescans. The deepest region of interest is the AlGaN region itself, which is dominated by EDS signals from The shallowest region of interest in this linescan is the gate both dem onstrate large signals in the EDS spectrum in the annealed sample. This is due to intermixing between the nickel liner layer of the gate electrode, which is approximately 20nm thick prior to annealing, with the gold layer that acts as the contact pad for probing of the device. Between these two regions is the interfacial oxide present between the AlGaN and the gate This interface is formed when vacuum is broken after the deposition of the AlGaN epitaxial layers, prior to the formation of the gate electrode via lithography and metal deposition and it is the changes which occur in this interface during annealing which are of interest in this study. The EDS lin escan shown in Figure 5 10B demonstrates that, upon annealing at 500C for layer. This result is consistent over a range of different regions and is consistent from one annealed sample to another. What appears to be occuring is the segregation of the aluminum present within the AlGaN into the native oxide layer formed between the AlGaN epitaxial layer oxide. As aluminum segregates into the interfacial oxide from the AlGaN just below, a region of

PAGE 128

128 aluminum deficient AlGaN forms, lea from the AlGaN epitaxial layer just below the native oxide interface. This segregation during annealing may be due to the restructuring of the interfacial oxide into the most thermodynamically stab le phase. It is well known that gallium oxides do not possess the characteristic thermal stability enjoyed by aluminum oxides at elevated temperatures. In fact, Holzworth and coworkers demonstrated previously on a commercial sample that it is common for the oxide present at this interface to be formed from alumina rather than some gallium oxide [113]. It is possible that, upon annealing or electrostatic stressing, the interfacial oxide undergoes a phase change, converting from some disordered oxidized st ate to some form of an aluminum oxide. This thermally stable aluminum oxide would be more resistant to BOE etching and could give rise to the feature observed in SEM and AFM. The formation of this alumina interfacial layer would also change the leakage observed at the gate contact, as the dipole formation associated with fermi level pinning of the semiconductor at the gate electrode is highly dependant upon interfacial quality. It bears noting that the thickness associated with maximum conduction of an Al contact to Si with an Al 2 O 3 interfacial layer is approximately 1.5nm, which fits well with observations made on these devices [121]. 5.3 Defect Formation Beyond the Critical Voltage The initial study described at the start of this chapter is revisite d in Figure 5 11, which demonstrates the quality of the surface of the AlGaN epitaxial layer as a device is step stressed at 1V/min well beyond the critical voltage. As shown in this figure, the banding characteristic to device degradation at V CRIT is no longer visible. It is understood that further electrostatic stressing above the critical voltage induces substantial reactivity and mobility of nickel atoms present within the liner layer of the gate

PAGE 129

129 electrode. This layer may getter the oxygen from th e interface, effectively consuming it, causing the strange "disappearance" of the banding defect. It is also possible that, in cases where the voltage is driven up above V CRIT rapidly, banding only forms in very isolated regions of the gate contact. The increase in leakage current density appears to be substantial when banding occurs via electrostatic stressing. After all, the total gate current in the device increased by two orders of magnitude as a result of banding over 66% of the surface of the devic e. This implies that the power dissipated through the interfacial oxide per unit area as a result of banding should increase by approximately two orders of magnitude, which is a substantial increase. It seems plausible that this increase in gate curren t due to banding could result in substantial joule heating in isolated regions of the gate contact. The temperature rise as a result of joule heating could enhance the reaction which forms the pitting defect which is so commonly observed at voltages well above V CRIT This could even be the case in situations where the voltage applied between the source and drain is zero, thanks to trap assisted tunneling of electrons from the 2DEG or local conduction through threading dislocations. The banding defect is replaced by the well documented pitting defect observed by various authors in Ni gated devices. This pitting defect covers approximately 5.5% of the total aerial density of the 100nm gate studied as part of the experiment. While the pitting defect has be en observed previously by various experimentalists, the nature of its formation and how this defect affects the current characteristics of the HEMT is not well understood for pitting defects induced in a Ni gated device. To that end, an additional experim ent was performed to ascertain the dependance of gate leakage current on the formation of pitting defects under the gate electrode.

PAGE 130

130 As before, devices were stressed in an air ambient using a Tektronix 370A curve tracer and an HP 4156 parameter analyzer The devices were placed in a deeply pinched off mode(with gate voltage = 8 V) to ensure minimal current flow between the source and drain electrodes as well as to maximize the field present between the gate and the 2DEG. The voltage between the gate elec trode and the drain, V DG was increased at a rate of 1V/min from a value of 5V to as much as 70V or until the device achieved current compliance, which varied from 500A to 1mA, depending on the device under stress. The gate leakage current during stress as well as in "off" mode, with VGS = 1V, was measured during the experiment. As before, this measurement occured over a time frame approximately equal to 1ms, meaning that the relative duty cycle associated with stressing was, effectively, 100%. The le akage at the end of this stressing period for seven separate devices was correlated with the density of pitting defects which were observed after standard deprocessing utilizing FeCN/KI combined with HF. This measurement was performed, as before, on an FE I SEM in UHR mode utilizing a TLD in secondary electron mode and the image analysis associated with estimating the aerial density of pitting defects was performed with Image J . This aerial density was normalized to the area of each device under analysis in order to yield a density in terms of the percent of the gate area which reacted. The result is shown in Figure 5 12. A direct correlation can be observed between the leakage current through the gate electrode during off mode stressing and the observe d aerial density of pitting defects. This is true for devices with large gatelengths and for devices with aggressively scaled gatelengths. However, it bears noting that defects which form under larger gates appear to induce larger leakage densities throu gh these pitting defects than those which form in aggressively scaled devices.

PAGE 131

131 A metallic Ni phase would be a more effective conductor than an oxidized Ni phase. Thus, the magnitude of the leakage current density should vary depending upon whether the defects present under the gate are comprised of metallic or oxidized Ni. In devices with 100 nm gate lengths, the diffusion length required for O 2 to permeate the entire gate length after diffusing through the SiN x passivation layer is as little as ~100 n m, while the diffusion length required to device with 100 nm gate length, O 2 may diffuse enough during stressing such that the majority of defects generated ar e oxidized Ni based rather than metallic Ni based, resulting in lower leakage current densities for those devices over the 1m devices.

PAGE 132

132 A B Figure 5 1 A plot of gate current degradation and associated defect formation prior to V CRIT A) Gate curre nt is compared to the source and drain currents to generate leakage current components passing through the gate electrode from the two contacting electrodes. B) An exemplary scanning electron micrograph of the deprocessed surface of this stressed device. No evidence exists of defect formation during this regime of stressing.

PAGE 133

133 A B Figure 5 2 A plot of the gate degradation up to a voltage approximating V CRIT A) At V CRIT the leakage current through the gate electrode both during stressing and in of f mode, increases by two orders of magnitude. B) An exemplary scanning electron micrograph of the deprocessed surface of this stressed device. A pair of inset dark bands in the same aerial position as the t gate are observed.

PAGE 134

134 A B Figure 5 3 A plo t of the degradation of a circular HEMT, with the same "banding" defect. A) Induced gate leakage current resulting from stepped stressing, at a rate of V GS = 1V/min and starting at V GS = 10V of a circular HEMT with V DS =15V, indicates that V CRIT is achieved and that no additional leakage current is induced past V CRIT B) A scanning electron micrograph formed by an FEI NovaSEM operating with the TLD detector and the immersion lens, reveals that this stepped stressing resulted in the formation of a band lik e defect under 10% of the total gate area.

PAGE 135

135 A B Figure 5 4 Morphological differences in banding induced by electrostatic stress and annealing. A) Stepped stressing results in the formation of a pair of inset dark bands on the AlGaN surface where th e nickel gate was once present. B) A similar single band defect can be observed on the AlGaN surface in a device which was annealed in a Lindberg Furnace at 500C for 30 min.

PAGE 136

136 C D Figure 5 4 (Continued) C) FLOOPS was used to simulate the vertical e lectric field present 1nm below the AlGaN surface in the stressed device at V GS = 22 V and V GS = 5 V. D) It was also use to simulate the vertical electric field present in the device with all electrodes of the device set to a potential of 0 V when tha t device is held a temperature of 500C.

PAGE 137

137 Figure 5 5 The gate leakage current before and after annealing at 500C for 30 min. A small increase in gate leakage current is observed, which is commensurate with themally induced banding under the gate electr ode of the device.

PAGE 138

138 A B Figure 5 6 SPM of an AlGaN/GaN HEMT annealed at 500C for 30 min. A) A raised band of roughness, comparable to the band observed in SEM, was detected. B) An average linescan of this band over 5m of the deivce width revea ls that it is approximately 1.0 nm in thickness.

PAGE 139

139 A B Figure 5 7 HAADF TEM of the interface before and after annealing. A) Analysis of lamella formed from an as recieved device indicates the presence of a thin interfacial region which is amorphous i n nature and ~1.5nm. B) Cross sectional BF TEM of a lemella formed from a device which was annealed at 500C for 30min and then deprocessed indicates that this interfacial region is preserved, if some intermixing does appear to occur.

PAGE 140

140 A B Figure 5 8 HRTEM and FFTs of the Ni/AlGaN interface before and after annealing. A) Analysis of lamella formed from an as recieved device indicates the presence of a thin interfacial region which is amorphous in nature and roughly equal in thickness (~1.5nm). B ) Cross sectional BF TEM of a lemella formed from a device which was annealed at 500C for 30min and then deprocessed indicates that this interfacial region is preserved, if some intermixing does appear to occur.

PAGE 141

141 A B Figure 5 9 HAADF STEM and EELS of the Ni/AlGaN interface before and after annealing. A) Analysis of a lamella formed from an as recieved sample indicates that the thin interfacial region separating the AlGaN and Ni gate is comprised of oxygen, indicating that a native oxide may have f ormed. B) Cross sectional DF STEM, in conjunction with EELS, of a lamella formed from a sample annealed at 500C for 30min and deprocessed, indicates that this layer remains oxygen rich, if somewhat more diffuse, after annealing.

PAGE 142

142 A B Figure 5 10 An EDS linescan of the Ni/AlGaN interface before and after annealing. A) Analysis of the linescan from a sample as fabricated indicates that the EDS signals arising from Al and Ga are constant up through the interfacial region until they hit the Ni of the gate electrode. B) An EDS linescan of this same interfacial region, after annealing at 500C for 30s and deprocessing in an etch solution, indicates a segregation of aluminum into this interfacial layer. The EDS signal arising from the ppears to be enhanced.

PAGE 143

143 A B Figure 5 11 Stepped stressing of a device to voltages well in excess of V CRIT This stressing eliminates the banding observed earlier at the critical voltage. Pitting of the AlGaN layer, as was observed previously by Holz worth et al., is observed in the channel region previously occupied by the gate electrode.

PAGE 144

144 Figure 5 12 Percentage of gate contact area consumed by under gate defects. This metric is a function of gate leakage current density (prior to catastrophic fail ure) for both nm and scale devices.

PAGE 145

145 CHAPTER 6 CONCLUSIONS Because of the highly stochastic nature of defect formation in AlGaN/GaN HEMTs, techniques which only sample small volumes samples from a much larger device (such as TEM) are poorly suited in analytical studies. Deprocessing and analysis utilizing top down SEM or AFM is often preferable to TEM for many studies. A deprocessing scheme was devised which utilizes a 15 minute exposure in BOE followed by a 28hr exposure in a mixture of FeCN and Ki otherwise known as TFAC. After degreasing via ultrasonic cleaning in a 1:1 mixture of n heptane and acetone for 2hr, a 2hr ultrasonication in methanol and a 30 minute exposure in DI water, a clean AlGaN surface can be imaged utilizing UHR SEM. This dep rocessing technique has been used with great effect in studies of defects in as formed samples as well as in samples which were stressed electrostatically. Investigations of AlGaN/GaN HEMTs formed with Ti/Al/Ni/Au ohmic contacts via annealing at 850C for 30s have revealed the formation of TiN inclusions which form at the AlGaN interface. These inclusions generally form in the vicinity of a threading dislocation, which is consistent with a model which involves inclusion formation via metal diffusion down a dislocation core. A layer of Al rich material was also observed at the interface between the TiN and the AlGaN surrounding it. Analysis with HAADF STEM and EDS indicate that the TiN inclusion is devoid of Ga. The Ga which was present in the epitaxial layer in place of the TiN may be insoluable in these metal inclusions and is pushed out into the remainder of the AlGaN in a front around the metal inclusion. The inclusions themselves appear to have a faceted morphology, which fits with the substantial energetic differences between different wurtzite GaN surfaces. XTEM demonstrates that, as these inclusions form and push Ga out into the surrounding AlGaN layer, cracks nucleate

PAGE 146

146 at the corners of their faceted sidewalls along the [112 0], or prism directi ons. It also demonstrates that the cracks themselves attain depths approximately equal to the AlGaN thickness. Nanocracks grow differently after their nucleation depending upon their location in relation to the edges of the source and drain contacts. Cracks which nucleate in the interior of the ohmic contact encounter substantial compressive stresses due to the presence of a substantial aerial density of metal inclusions. Because of this, cracks which nucleate in the ohmic contact, due to tensile hoop stress brought on by compressive stresses oriented radially around the metal inclusion, are short with the maximum of the distribution of their crack lengths being approximately equal to 40nm 60nm. Cracks which nucleate at the edges of the ohmic contac t propagate into a region of the AlGaN/GaN epitaxial layer which is not under substantial compressive stress, as no metal inclusions are present. The cracks encounter the tensile epitaxial stresses of the AlGaN layer and grow to much longer lengths, resul ting in a distribution of cracks lengths with a maximum at 140 nm and with a much larger variance than the distribution associated with cracks present in the interior of the ohmic contacts. It should be noted that channel cracks at the long end of the d istribution require a tensile stress for growth on the order of the those associated with the extreme voltages attained during electrostatic off mode stressing. This is of interest given results from electrostatic stressing of an AlGaN/GaN HEMT stressed a t V DS = 15V and from V GS = 10V to V GS = 42V at a rate of 1V/min. Deprocessing followed by Plan View SEM analysis revealed that the lengths of nanocracks present in the channel of this device were much longer than cracks observed in as formed devices, sugge sting that stressing may have induced further crack growth. The degradation of I DS and V GS in the device might be commensurate with this phenomenon, as

PAGE 147

147 cracks in AlGaN/GaN epitaxial structures can induce a shorting of the gate contact to the source or dra in if they extend under the gate contact itself. This result illustrates a fundamental flaw of ohmic contact formation via TiN metal inclusions. It stands to reason that, in order to further reduce the contact resistance of an alloyed metallurgical con tact to AlGaN/GaN, the total area over which the TiN directly contacts the 2DEG must be maximized via wholesale reaction of Ti with AlGaN. It also stands to reason that, as the density of inclusions increases, so too will the density of nucleated cracks w ithin the ohmic contacts and within the channel region. Cracks extending into the channel region will degrade the device by means of gate shorting. An alternative method of contact formation or some adjustment of architecture to trap cracks which form on TiN inclusions may be required in order to further improve the reliability of devices under stress. Analysis of devices stressed from V GS = 10V up to values of V GS < V CRIT V GS = V CRIT V GS > V CRIT at a rate of 1V/min with V DS = 5V revealed the relatio nship between the increase in I G at V CRIT and a structural change in the AlGaN surface of a HEMT stressed with a bias applied between the source and drain electrodes. This new defect was a "band" of dark contrast which was observed under a 100nm Lg gate e lectrode in UHR SEM utilizing a TLD. A similar dark band of contast can be induced by annealing a device at 500C for 30min. The differences in morphology of these two bands can be explained by an field induced reaction mechanism where applied electric f ield impacts the activation energy of a reaction, similar to time dependant dielectric breakdown. It should be noted that this defect only appears in devices which were stressed with a potential applied across the source and drain. It may be that a stead y supply of electrons from the 2DEG to the interface of the device is necessary for the formation of this banding defect.

PAGE 148

148 The I G V GS characteristics of an AlGaN/GaN HEMT after annealing at 500C for 30min reveals an increase in the reverse biased gate l eakage current. The increase is not as dramatic as the increase which was induced in the AlGaN/GaN HEMT after electrostatic stressing, but it should be noted that the contrast of the band observed in the annealed sample is also less pronounced than the s ample which was stressed electrostatically. Presumably, thermal annealing with only the built in field of the schottky contact to reduce the activation energy associated with the reaction does not induce as high a rate of reaction as stressing with a high applied field. AFM analysis of the banded surface of an AlGaN/GaN HEMT with a 1m gatelength reveals that the banding observed in SEM corresponds to a raised feature with a thickness of approximately 1.75nm. This defect has approximately the same surfac e roughness as the channel, suggesting that it may be formed by some reaction at the interface between the epitaxially deposited AlGaN and the gate contact. Interestingly, the thickness of the band as it was observed in AFM matches the thickness of the na tive oxide layer formed at the interface of the AlGaN prior to gate metal depoition. STEM combined with EELS reveals that this oxide layer is amorphous in nature and is present both before and after annealing at 500C. EDS analysis of the interface revea led that, upon annealing, aluminum appears to segreagate into the native oxide layer from the AlGaN. As stressing proceeds past the critical voltage, the band defect is no longer present. This strange case of disappearance may be due to the tendency of nickel present within the gate to react with oxygen present in the ambient and in the device. Given that pitting defect formation appears to occur after V CRIT in this stressing regime, It may be that banding is a process with a lower inherent activation energy than the reaction by which Ni reacts with O and pits the surface.

PAGE 149

149 Analysis of other devices stressed above the critical voltage via deprocessing and top down SEM indicate that the pitting reaction has a direct impact on the reverse biased gate le akage observed in an AlGaN/GaN HEMT after stressing well beyond V CRIT The aerial density of pits observed in SEM appears to influence directly the reverse biased leakage current, which fits with pitting forming an ohmic contact to the 2DEG and effectivel y shorting out the gate contact of the device. The formation of banding and pitting based defects has a substantial impact on the overall reliability of the gate contact of AlGaN/GaN HEMT technologies. It may be that these reliability issues cannot be engineered out without substantial changes to device architecture. Introduction of a thermally stable oxide layer as well as a change to a more inert gate metal, such as Pt, might alleviate banding and pitting by fundamentally changing the nature of the c ontact of the device as well as its reactivity. A MOS architecture which also makes use of a field plate to reduce the peak electric field under the gate contact seems like a very promising candidate for eliminating the reliability issues inherent to the contacting of a Ni schottky gate directly to AlGaN.

PAGE 150

150 LIST OF REFERENCES [1]: M. Feng, "Device Technologies for RF Front End Circuits in Next Generation Wireless Communications", Proc. of the IEEE, vol. 92, no. 2, pp. 354 375, Feb., 2004. [2]: P. H. 1776 1781, Aug., 2001. [3]: E. O. Johnson, "Physical limitation on frequency and power parameters of transistors", RCA Rev., pp. 163 176, Jun.,1965. [4]: F. Bemardini, taneous polarization and piezoelectric constants of III V Phys. Rev. B, vol. 56, no. 16, pp. R10024 10027, May, 1997. [5]: N.M. Stanton, Photoenhanced wet chemical etching of MBE grown gallium nitride", Mat. Sci. Eng. B, vol. 68, no. 1, pp. 5 2 55, Dec., 1999. [6]: S.T. Sheppard, "High power microwave GaN/AlGaN HEMTs on semi insulating silicon carbide substrates", IEEE Elec. Dev. Lett., vol. 20, no. 2, pp. 161 163, Apr.,1999. [7]: W.S. Tan, High temperature performance of AlGaN/GaN HEMTs o n Si substrates", Solid State Elect., vol. 50, no. 2, pp. 511 513, Mar., 2006. [8]: C.C. Hu, "Bipolar Transistor" in Modern Semiconductor Devices for Integrated Cirucits, New York, NY: Pearson Education, 2010, pp. 309. [9]: R.F. Pierret,"MS Contacts a nd Schottky Diodes" in Semiconductor Device Fundamentals, K. Harutunian, Ed. Boston, Ma: Addison Wesley Publishing Company, Inc., 1998, pp. 477 484. [10]: C.C. Hu,"MOS Transistor", in Modern Semiconductor Devices for Integrated Cirucits, New York, NY: Pe arson Education, 2010, pp. 234. [11]: E.W. Williams, "Indium Phosphide 1. A Photoluminescence Materials Study", J. Electrochemical Soc.: Solid State Sci. and Tech., vol 120, no. 2, pp. 1741 1749, Dec., 1973. [12]: S. Firoz "Consideration of AlGaN/G aN and AlGaAs/GaAs based HEMT device under doping consideration", Int. Journ. of Adv. in Eng. and Tech., vol 1, pp. 12 19, May, 2011. [13]: D. Pavlidis, "III V Nitride Electronics", IEEE Indium Phosphide and Rel. Mat. Conf. Proc., pp. 443 446, 2002. [14 ]: M.K. Kwon, Gradient Doping of Mg in p Type GaN for High Efficiency InGaN GaN Ultraviolet Light Emitting Diode", IEEE Photonics Terchnology Letters, vol 19, no. 23, pp. 1880 1882, Dec., 2007. [15]: E. Reese, "Wideband Power Amplifier MMICs Utilizing GaN on SiC", IEEE 2010 Intern. Microwave Symp. Proc., pp. 1230 1233, Oct., 2010.

PAGE 151

151 [16]: J. Sun, Thermal Management of AlGaN GaN HFETs on Sapphire Using Flip Chip Bonding With Epoxy Underfill", IEEE Elec. Dev. Lett., vol 24, no. 6, pp. 375 377, Jun., 200 3. [17]: P. Javorka, "AlGaN/GaN HEMTs on (111) Silicon Substrates", IEEE Elect. Dev. Lett. Vol 23, no. 1, pp. 4 6, Jan., 2002. [18]: E.A. Douglas, Degradation of Sub micron Gate AlGaN/GaN HEMTs due to Reverse Gate Bias", IIRW Fin. Rep., pp. 125 129, Oct., 2010. [19]: Y.C. Choi, The Effect of an Fe doped GaN Buffer on OFF State Breakdown Characteristics in AlGaN/GaN HEMTs on Si Substrate", IEEE Trans. on Elect. Devi. 53, no. 12, pp. 2926 2931, Dec., 2006. [20]: D. Li, Influence of threading disl ocations on GaN based metal semiconductor metal ultraviolet photodetectors", Appl. Phys. Lett., vol. 98, no. 1, pp. 11108, Dec., 2011. [21]: J.P. Colinge, "Heterojunction Devices", in Physics of Semiconductor Devices New York, NY: Springer Verlag, 2003 pp. 323. [22]: B. Luo, Comparison of ohmic contact properties on n GaN/p SiC and n AlGaN/p SiC heterojunctions Solid State Elec., vol. 46, no. 9, 1345 1349, Sept., 2002. [23]: B.G Streetman "Junctions", in Semiconductor Device Fundamentals, M. Ho rton and A. Dworkin, Eds., Boston, MA: Addison Wesley Publishing Company, 2000, pp. 142 227. [24]: R.F. Pierret, "Field Effect Introduction The J FET and MESFET", in Semiconductor Device Fundamentals Boston, MA: Addison Wesley Publishing Company, 1996 pp. 525 557. [25]: M.A. Huque, Temperature dependent analytical model for current voltage characteristics of AlGaN/GaN power HEMT", Solid State Elec., vol 53, no. 2, pp. 341 348, Mar., 2009. [26]: B.G Streetman "FIed Effect Transistors", in Semicon ductor Device Fundamentals, M. Horton and A. Dworkin, Eds., Boston, MA: Addison Wesley Publishing Company, 2000, pp. 241 315. [27]: R.F. Pierret, "Field Effect Introduction The J FET and MESFET", in Semiconductor Device Fundamentals Boston, MA: Addiso n Wesley Publishing Company, 1996, pp. 525 557. [28]: G. Menghesso, "Reliability of GaN High Electron Mobility Transistors: State of the Art and Perspectives", IEEE Trans. Dev. Mat. Rel., vol. 8, no. 2, pp. 332 344, Jun. 2008. [29]: G.E. Jellison, S pectroscopic ellipsometry characterization of thin film silicon nitride Thin Solid Films, vol. 313 314 pp. 193 197, Feb., 1998

PAGE 152

152 [30]: D.H. Kim, "ALD Al2O3 passivated MBE grown AlGaN/GaN HEMTs on 6H SiC", Elec. Lett., vol 18, no. 2, Mar., 2007. [31]: B.P. Gila, "Gadolinium oxide and scandium oxide: Gate dielectrics for GaN MOSFETs", Phys. Stat. Sol. A, vol 188, no. 1, pp. 239 242, Nov. 2001. [32]: J.S. Speck "The role of threading dislocations in the physical properties of GaN and its alloys" Phy sica B: Condensed Matter, vol. 273 274, pp. 24 32, Dec. 1999. [33]: H. Witte, Deep Defects in Fe Doped GaN Layers Analysed by Electrical and Photoelectrical Spectroscopic Methods", MRS Proc., vol. 798, Dec., 2003. [34]: W.D Rau, "Two Dimensional Mappi ng of the Electrostatic Potential in Transistors by Electron Holography", Phys. Rev. Lett., vol. 82, no. 12, pp. 2614 2617, Mar., 1999. [35]: D.B Williams, "Electron Sources",in Transmission Electron Microscopy: A Textbook for Materials Science, New York NY: Springer Sceince+Business Media, 2009, pp. 73 87. [36]: D.B Williams, "Using Convergent Beam Techniques",in Transmission Electron Microscopy: A Textbook for Materials Science, New York, NY: Springer Sceince+Business Media, 2009, pp. 347 366. [37]: M. Haider, "A spherical aberration corrected 200kV transmission electron microscope", Ultramicroscopy, vol. 75, no. 1, pp. 53 60 ,Oct., 1998. [38]: J.J. Bozola, "Analytical Electron Microscopy", in Electron Microscopy: Principles and Techniques for Biol ogists, R.S. Marks, Ed., Burlington, MA: Jones and Bartlett Learning, 1998, pp. 368 395. [39]: M.F. L'Annuziat, "Electromagnetic Radiation Photons", in Handbook of Radioactivity Analysis, Amsterdam, Netherlands: Elsevier, 2012, pp. 94. [40]: Springe r Handbook of Materials Measurement Methods, H. Czichos, T. Saiato and L.R. Smith, Eds., Springer Science+Business Media, Berlin, Ger., 2006, pp.183. [41]: L. Wojnar, "Main Tools for Image Treatment", in Image Analysis: Applications in Materials Engineer ing B. Ralph Ed., New York, NY: CRC Press LLC, 1998, pp. 7 53. [42]: D.B Williams, "X Ray Spectrometry",in Transmission Electron Microscopy: A Textbook for Materials Science, New York, NY: Springer Sceince+Business Media, 2009, pp. 581 601. [43]: P. Kohler Redlich "Quantitative Analytical Transmission Electron Microscopy", High Resolution Imaging and Spectroscopy of Materials R.M Osgood, R. Hull and J. Parisi, eds New York, NY: Springer Verlag, 2003, pp. 119 187.

PAGE 153

153 [44]: D. Nakaji "Contrast analys is of dislocation images in TEM cathodoluminescence technique", Journal of Elec. Mic., vol. 54, no. 3, pp. 223 230, Jun., 2005. [45]: L.A. Gianuzzi, "A review of focused beam ion milling techniques for TEM sample preparation", Micron, vol. 30, no. 3, pp 197 204, Jun., 1999. [46]: D.B Williams, "The Transmission Electron Microscope",in Transmission Electron Microscopy: A Textbook for Materials Science, New York, NY: Springer Sceince and Business Media, 2009, pp. 3 36. [47]: W.N. Borle, "Dislocation etch pits on various crystal planes of silicon", Journal of Crystal Growth, vol. 36, no. 2, pp. 259 262, Dec., 1976. [48]: D.B Holt, "An Introduction to Multimode Scanning Electron Microscopy",in SEM Microcharacterization of Semiconductors Techniques in Physics Vol. 12, D.B. Holt and S.C. Joy, Eds., London, UK: Academic Press, 1989, pp. 24. [49]: K.D. Vernon Perry, "Scanning Electron Microscopy: an introduction", III Vs Review, vol. 13, no. 4, pp. 40 44, Jul., 2000. [50]: I.M. Watt, "Electron specime n interactions: processes and detectors", in The Principles and Practice of Electron Microscopy Cambridge, England: Cambridge University Press, 1997, pp. 30 57. [51]: B.G. Yakobi, "Introduction", in Cathodoluminescence Microscopy of Inorganic Solids N ew York, NY: Plenum Press, 1990, pp. 1 6. [52]: J.J. Bozola, "The Scanning Electron Microscope", in Electron Microscopy: Principles and Techniques for Biologists, R.S. Marks, Ed., Burlington, MA: Jones and Bartlett Learning, 1998, pp. 202 239. [53]: Ha ndbook of Instrumentation and Techniques for Semiconductor Nanostructure Characterization R. Haight, F.M. Ross and J.B. Hannon, Eds., Singapore: World Scientific Publishing, 2011, pp. 9. [54]: G. Haugstadt, "Overview of AFM", in Atomic Force Microscopy: Understanding Basic Modes and Advanced Applications Hoboken, NJ: John Wiley and Sons, 2012, pp.1 31. [55]: Q. Feng, "The improvement of ohmic contact of Ti/Al/Ni/Au to AlGaN/GaN HEMT by multi step annealing method", Solid State Elec., vol. 53, pp. 955 958, Jun., 2009. [56]: A. Vertiatchikh, "Structural properties of alloyed Ti/Al/Ti/Au and Ti/Al/Mo/Au ohmic contacts to AlGaN/GaN", Solid State Elec., vol. 50, no. 7, pp. 1425 1429 Jul., 2006. [57]: M.D. Kamatagi, "Thermal Conductivity in GaN", Diamond Rel. Mat., vol. 16, no. 1, pp. 98 106 Jan., 2006.

PAGE 154

154 [58]: D. Zhuang "Wet Etching of GaN/AlN and SiC: A Review", Mat. Sci. Eng. R, vol. 48, no.1 pp. 1 46, Jan., 2005. [59]: J.C. Gerbedoen, "Study of ohmic contact on AlGaN/GaN HEMT with AlN spacer on Si Microwave Int. Cir. Conf., pp. 136 139, Sept., 2009. [60]: M.E. Lin, "Low resistance ohmic contacts on wide band gap GaN", Appl. Phys. Lett., vol. 63, no. 8, pp. 1003 1005, Oct., 1993. [61]: M.W. Fay, "Structural and electrical characterization of A uTiAlTi/AlGaN/GaN ohmic contacts", J. Appl. Phys., vol. 92, no. 1, pp. 94 100, Apr., 2002. [62]: A.N. Bright, Correlation of contact resistance with microstructure for Au/Ni/Al/Ti/AlGaN/GaN ohmic contacts using transmission electron microscopy ", J. Appl Phys., vol. 89, no. 6 pp. 3143, Dec., 2001. [63]: F. Roccaforte, Nanoscale carrier transport in Ti/Al/Ni/Au Ohmic contacts on AlGaN epilayers grown on Si(111) ", Appl. Phy. Lett., vol. 89, no. 2, pp. 022103, Jul., 2006. [64]: L. Wang, "Formation me chanism of ohmic contacts on AlGaN/GaN heterostructures: electrical and microstructure characterizations", J. Appl. Phys., vol 103, no. 9, pp. 093516, May, 2008. [65]: M.W. Fay, Structural and electrical characterization of AuPdAlTi ohmic contacts to Al GaN/GaN with varying Ti content", J. Appl. Phys., vol 96, no. 10 pp. 5588, Aug., 2004 [66]: J. Chen, "Microstructural analysis of Ti/Al/Ti/Au ohmic contacts to n AlGaN/GaN", J. Vac. Sci. Tech. A, vol. 20, no. 3 pp. 1004, Jun., 2002 [67]: L. Zhou, Ti/Al/Ni/Au contacts for AlInN/AlN/GaN based heterojunction field effect transistors", J. Appl. Phys., vol. 107, no. 1 pp. 014508, Jan., 2010. [68]: R. Gong, "Analysis of surface roughness in Ti/Al/Ni/Au ohmic contacts to AlGaN/GaN high electron mobilit y transistors", Appl. Phys. Lett., vol. 97, no. 6, pp. 062115, Aug., 2010. [69]: B. Jacobs, "Optimization of a Ti/Al/Ni/Au ohmic contact on AlGaN/GaN FET structures", J. Crys. Growth, vol. 241, no. 1 2, pp. 15 18, May, 2002. [70]: N Chaturvedi, "Mechani sm of formation of ohmic contact formation in AlGaN/GaN high electron mobility transistors", Semi. Sci. Tech., vol. 21, no. 2, pp. 175 179, Jan., 2006. [71]: H.P. Xin, "Optimization of AlGaN/GaN HEMT Ohmic Contacts for Improved Surface Morphology with Lo w Contact Resistance", CS MANTECH Conference, May, 2010.

PAGE 155

155 [72]: Q. Feng, "The improvement of ohmic contact of Ti/Al/Ni/Au to AlGaN/GaN HEMT by multi step annealing", Sol. State Elec., vol. 53, pp. 955 958, Jun., 2009. [73]: V. Desmaris, "Transmission el ectron microscopy assessment of the Si enhancement of Ti/Al/Ni/Au Ohmic contacts to undoped AlGaN/GaN heterostructures", J. Appl. Phys., vol 100, no. 3, pp. 034904, Aug., 2006. [74]: J.E. Northrup "Theory of GaN [101 0] and [112 0] surfaces", Phys. Rev. B, vol. 53, no.16, pp. R10477 R10480, Apr., 1996. [75]: P.J. Parbrook, "Crack formation and development in AlGaN/GaN structures", Physica Status Solidi C, vol. 0, no. 7, pp. 2055 2058, Sept., 2003. [76]: S. Einfeldt, "Strain relaxation in AlGaN under tensile plane stress", J. App. Phys. ,vol. 88, no. 12, Dec., 2000. [77]: T.H. Courtney, "Fracture Mechanics", in Mechanical Behavior of Materials Long Grove, IL: Waveland Press, 2000, pp. 404 448. [78]: R. Huang, "Channel cracking of thin films with the extended finite element method", Eng. Frac. Mech., vol. 70, pp. 2513 2526, Feb., 2003. [79]: K.D. Mackenzie, "Stress control of Si based PECVD dielectrics", in 207 th Electrochemical Society Meeting, Quebec, Can., 2005, pp. 148 159. [80]: D. Holec "Surface energies of AlN allotropes from first principles", Scripta Materialia, vol. 67, no. 9, pp. 760 762, Nov., 2012. [81]: G. Steude, "Strain modification of GaN in AlGaN/GaN epitaxial films", Jpn. J. App. Phys., vol. 38, no. 2, pp. L498 L500, May, 1 999. [82]: O. Ambacher, "Two dimensional electron gas induced by spontaneous and piezoelectric polarization charges in N face and Ga face AlGaN/GaN heterostructures", J. App. Phys., vol. 85, no. 6, pp. 3222 2234, Mar., 1999. [83]: J. Suo, "Reliability of Interconnect Structures", in Comprehensive Structural Failue vol. 8, W. Gerberich, W. Yang, I. Milne, R.O. Ritchie and B. Karihaloo, Eds., Amsterdam, Netherlands: Elsevier, 2003, pp. 265 324. [84]: J.W. Pomeroy, "Thermal mapping of defects in AlGaN/G aN heterostructure field effect transistors using micro Raman spectroscopy" Appl. Phys. Lett., vol. 87, no. 10, Sept., 2005. [85]: G. Gautschi, Piezoelectric Sensorics: Force, Strain, Pressure, Acceleration and Acoustic Emission Sensors, Materials and A mplifiers New York, NY: Springer Science, pp. 32, 2002.

PAGE 156

156 [86]: D. Damjanovic, Ferroelectric, dielectric and piezoelectric properties of ferroelectric thin films and ceramics ", Rep. Prog. Phys., vol. 61 pp. 1267 1324, Sept. 1998. [87]: CRC Handbook of Chemistry and Physics D.R. Lide, Ed., CRC Press, Boca Raton, FL, 2008 pp. 201. [88]: L.C. Xu, "Thermal expansions in AlN, GaN and InN: first principle phonon calculations", J. Appl. Phys., vol. 10, no. 4, pp. 043528, Aug., 2011. [89]: C. A. Klein, Strains and stresses in multilayered elastic structures: The case of chemically vapor deposited ZnS/ZnSe laminates", J. Appl. Phys., vol. 8, no. 5, pp. 2265 2272, Mar., 2000. [90]: I.L. Guy, "Extensional piezoelectric coefficients of gallium nitride and aluminum nitride", Appl. Phys. Lett., vol. 75, no. 26, pp. 4133 4135, Dec., 1999. [91]: A. F. Wright, "Elastic properties of zinc blende and wurtzite AlN, GaN and InN", J. Appl. Phys., vol. 82, no. 6, pp. 2833, Jun., 1997. [92]: J. delAlamo, "GaN HEMT Reliability", Microelectronics Reliability, vol. 49, no. 9 11, pp. 1200 1206, Jul., 2009. [93]: J. Joh "Gate current degradation mechanisms of GaN high electron mobility transistors", IEEE IEDM Tech. Dig., pp. 385 388, Dec., 2007. [94]: J. Joh "Critic al voltage for electrical degradation of GaN highelectron mobility transistors", IEEE Elec. Dev. Lett., vol. 29, no. 4, 287 289, Apr., 2008. [95]: M.H Somerville, "Off state breakdown in power pHEMTs: the impact of the source", IEEE Trans. on Elec. Dev., vol. 45, no. 4, pp. 1883 1889, Sept., 1998. [96]: J. Joh, "Impact of electrical degradation on trapping characteristics of GaN high electron mobility transistors", IEEE IEDM Tech. Dig., pp. 461 464, Dec., 2008. [97]: J. Joh, "A simple current collap se measurement technique for GaN high electron mobility transistors", IEEE Elec. Dev. Lett., vol. 29, no. 7, pp. 665 667, Jul., 2008. [98]: U. Chowdhury, "TEM observation of crack and pit shaped defects in electrically degraded GaN HEMTs", IEEE Elec. De v. Lett., vol. 29, no. 10, pp. 1098 1100, Oct., 2008. [99]: A. Sarua International Conference on Compound Semiconductor Manufacturing Technology, unpublished, May, 2009. [100]: M. Kuball, "AlGaN/GaN HEMT device reliability and degradation evolution: Importance of diffusion processes", Mic. Rel., vol. 5, no 2, pp. 195 200, Feb., 2010.

PAGE 157

157 [101]: M.E. Glicksman, Diffusion in Solids: Field Theory, Solid State Principls and Applications New York, NY: John Wiley and Sons, pp. 35, 2000. [102]: S.J. Pearto n, "Oxygen diffusion into SiO 2 capped GaN during annealing", Appl. Phys. Lett., vol. 75, no. 19, pp. 2939, Sept., 1999. [103]: Wang ACM, "Leakage and hFE degradation in microwave bipolar transistors", IEEE Trans. Elec. Dev., vol. 21, no. 7, pp. 667 674, Nov. 1974. [104]: Huff H.R., "An electronics division retrospective 1952 2002 and future opportunities in the twenty first century", J. Electrochem. Soc., vol. 149 152, S35 S59, 2002. [105]: V. Palankovski, "Field plate optimization of AlGaN/GaN HEMTs" Compound Semiconductor Integrated Circuit Symposium, pp. 107 110, Nov., 2006. [106]: C. induced degradation of AlGaN/GaN high electron [10 7]: L.Liu, Comparison of DC Performance of Pt/Ti/Au and Ni/Au Gated AlGaN/GaN [108]: and off state str [109]: F. Bemardini, V Phys. Rev. B, vol. 56, no. 16, Oct., 1997. [110]: M Guziewicz, Comparative study on stress in AlGaN /GaN HEMT structures grown on 6H SiC, Si and on composite substrates of the 6H SiC/poly SiC and Si/poly SiC", J.Phys. Conf. Ser., vol. 100, no. 4, pp. 042035, 2008. [111]: C.F. Lo, Improvement of Off State Stress Critical Voltage by Using Pt Gated AlG aN/GaN High Electron Mobility Transistors ", Electrochem. Sol. State Lett. vol. 14, no. 7 pp. H264 H267, Apr., 2011 [112]: S. D. Burnham, "Reliability of T Gate AlGaN/GaN HEMTs", P hysica Status Solidi C, vol 8, no. 7 8 pp. 2399 2403, Jun., 2011. [1 13]: M.Holzworth, Phys. Lett. Vol. 98, no. 12, pp. 122103, Mar., 2011. [114]: L. Liu, Effect of source field plate on the characteristics of off state, step stressed AlGaN/GaN high elect ron mobility transistors", JVSTB: Microelec. and Nanometer Struct., vol. 29, no. 3, pp. 32204, Apr., 2011.

PAGE 158

158 [115]: P. Makaram, "Evolution of structural defects associated with electrical degradation in AlGaN/GaN high electron mobility transistors", Appl. Phys. Lett., vol. 96, no. 23, pp. 233591 Jun., 2010. [116]: F. Gao, Role of oxygen in the OFF state degradation of AlGaN/GaN high electron mobility transistors ", Appl. Phys. Lett., vol. 99, no. 22 pp. 223506, Nov., 2011 [117]: M. M. Bajo, On the l ink between electroluminescence, gate current leakage, and surface defects in AlGaN/GaN high electron mobility transistors upon off state stress", Appl. Phys. Lett., vol. 101, no. 3 pp. 033508, Jul., 2012. [118]: P.G. Whiting, "Under gate defect formati on in Ni gate AlGaN/GaN high electron moblity transistors", Mic. Rel. vol. 52, no. 1, pp. 2542 2546, Nov., 2012. [119]: J. Goldstein, "Electron Beam Specimen Interactions" in Scanning Electron Microscopy and X Ray Microanalysis, New York, NY: Kulwer Academic / Plenum Publishers, 2003, pp. 61 97. [120]: M. Kimura, "Field and temperature acceleration model for time dependant dielectric breakdown", IEEE Trans. Elec. Dev., vol. 46, no. 1, pp.220 229, Jan., 1999. [121]: J. Hu, "Metal/III V schottky barrier height tuning for the design of field effect transistor source/drain contacts", J. Appl. Phys., vol. 107, no. 6, pp.06712, Mar. 2010.

PAGE 159

159 BIOGRAPHICAL SKETCH Patrick Whiting was born in Chicago, Illinois in the United States of America in 1986 an d he grew into adulthood in East Bloomfield, New York a small town outside of the city of Rochester, nestled snugly in the Finger Lakes region of the state. He graduated from high school in 2004, was the salutatorian of his class, and was voted "most li kely to succeed". His interest in electronics and the growing field of nanotechnology brought him to Rochester Institute of Technology, located in West Henrietta, New York. He began his carrer as a researcher in his freshman year of school, studying fr esnel microlens arrays, sputtered thin film dielectrics for MRI, high power silicon electronics and single crystal thin film transistors. In the wint er of 2009, he graduated Magna cum Laude with his BS in microelectronic e ngineering and his MS i n material s science and e ngineering. Faced with the horrible economic climate of a worldwide recession and with a ravenous hunger for more knowledge, Patrick decided that he would pursue a Ph.D in materials science and e ngineering from the University of Florida und er Dr. Kevin S. Jones. He studied AlGaN/GaN High Electron Mobility Transistors, a new technology which was becoming popular for a variety of applications, including the infamous "naked scanners" which can be found at most domestic airports. Patrick will graduate in December 2013 and move on to work for Intel Corporation in Hillsboro, Oregon where he will be an engineer working on process development for next generation microprocessors. Patrick expects that, when he isn't working, he will be found in the environs around the city of Portland, usually writing some crazy science fiction story or other.