"Applied Science, Faculty of"@en . "Engineering, School of (Okanagan)"@en . "DSpace"@en . "UBCO"@en . "Rajput, Harish C."@en . "2012-11-05T17:43:11Z"@en . "2011"@en . "Master of Applied Science - MASc"@en . "University of British Columbia"@en . "A steady state analytical solution for MOSFET (metal oxide semiconductor field effect transistor) channel temperature estimation has been derived and the analytical model has been used to develop a software tool called HeatMOS\u00A9. HeatMOS\u00A9 estimates the MOSFET channel temperature based on information from the device layout and an industry standard BSIM3 compact model. The steady state solution is an approximation for the channel temperature distribution along its length. The HeatMOS\u00A9 model has been designed to be integrated into a VLSI CAD flow to predict the steady state temperature of a full micro-chip. An equivalent M-network model for steady state temperature can be extended for each MOSFET device in a complete micro-chip. In future work, HeatMOS\u00A9 can be combined with the models of interconnect to develop a full micro-chip thermal analysis software tool."@en . "https://circle.library.ubc.ca/rest/handle/2429/43566?expand=metadata"@en . "A Steady-state Analytical Solution for MOSFET Channel Temperature Estimation HeatMOS\u00C2\u00A9 by Harish C. Rajput A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in The College of Graduate Studies (Electrical Engineering) THE UNIVERSITY OF BRITISH COLUMBIA (Okanagan) April 2011 \u00C2\u00A9 Harish C. Rajput 2011 Abstract A steady state analytical solution for MOSFET (metal oxide semiconductor field effect transistor) channel temperature estimation has been derived and the analyti- cal model has been used to develop a software tool called HeatMOS\u00C2\u00A9. HeatMOS\u00C2\u00A9 estimates the MOSFET channel temperature based on information from the device layout and an industry standard BSIM3 compact model. The steady state solution is an approximation for the channel temperature distribution along its length. The HeatMOS\u00C2\u00A9 model has been designed to be integrated into a VLSI CAD flow to predict the steady state temperature of a full micro-chip. An equivalent M-network model for steady state temperature can be extended for each MOSFET device in a complete micro-chip. In future work, HeatMOS\u00C2\u00A9 can be combined with the models of interconnect to develop a full micro-chip thermal analysis software tool. ii Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix List of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.3 Research Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4 Scope and Organization . . . . . . . . . . . . . . . . . . . . . . 14 iii Table of Contents 2 Heating in Integrated Circuits . . . . . . . . . . . . . . . . . . . . . 16 2.1 Electrostatic Analogy of Heat Conduction . . . . . . . . . . . . . 18 2.2 Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3 Heat Generation in MOSFETs . . . . . . . . . . . . . . . . . . . 22 2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 3 Thermal Model and Analytical Solution . . . . . . . . . . . . . . . 25 3.1 BSIM3 MOSFET Model . . . . . . . . . . . . . . . . . . . . . . 25 3.2 MOSFET Channel . . . . . . . . . . . . . . . . . . . . . . . . . 27 3.2.1 Channel Geometry for Thermal Model . . . . . . . . . . 32 3.2.2 Electrical Resistance . . . . . . . . . . . . . . . . . . . . 34 3.2.3 Channel Height Models . . . . . . . . . . . . . . . . . . 37 3.2.4 Channel Length Models . . . . . . . . . . . . . . . . . . 39 3.2.5 Thermal Conductance . . . . . . . . . . . . . . . . . . . 40 3.3 Solution of the Heat Equation . . . . . . . . . . . . . . . . . . . 42 3.3.1 Region 1: Non-uniform cross-sectional area . . . . . . . 42 3.3.2 Region 2: Uniform cross-sectional area . . . . . . . . . . 44 iv Table of Contents 3.3.3 Boundary Conditions and Complete Solution . . . . . . . 47 3.3.4 The Average Channel Temperature . . . . . . . . . . . . 51 3.3.5 The Maximum Channel Temperature . . . . . . . . . . . 51 3.3.6 The Channel Temperature at Pinch-off . . . . . . . . . . 53 3.4 M-Network Representation of the Channel . . . . . . . . . . . . 53 3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4 HeatMOS\u00C2\u00A9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.1 About HeatMOS\u00C2\u00A9 . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.2 Integration in the VLSI CAD Flow . . . . . . . . . . . . . . . . 61 5 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . 63 5.1 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 5.1.1 Test Scenario using Externally Supplied Drain Current . . 67 5.1.2 Test Scenario for a Smaller MOSFET Device . . . . . . . 69 5.1.3 Test Scenario for the Case When the Source, Drain, and the Reference Temperatures are Same . . . . . . . . . . . 69 5.2 Saturation Mode of Operation . . . . . . . . . . . . . . . . . . . 72 v Table of Contents 5.2.1 Test Scenario When Gate and Drain are at Same Voltage 72 5.2.2 Test Scenario for the Case of Different Gate and Drain Voltages . . . . . . . . . . . . . . . . . . . . . . . . . . 74 5.2.3 Test Scenario for the Case of High Gate and Drain Voltages 76 5.2.4 Test Scenario for a Large MOSFET Device with High Drain Current . . . . . . . . . . . . . . . . . . . . . . . . . . 76 5.2.5 Test Scenario for Different Source/Drain Temperatures and Voltages . . . . . . . . . . . . . . . . . . . . . . . . . . 79 5.3 Linear/Ohmic Mode of Operation . . . . . . . . . . . . . . . . . 79 5.3.1 Test Scenario for the Case of Different Temperatures at the Source and the Drain . . . . . . . . . . . . . . . . . . . 81 5.3.2 Test Scenario for a Larger Device in the Ohmic Operating Region . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 5.4 Average and Maximum Temperature Distributions . . . . . . . . 84 5.4.1 Channel Temperature Distribution with Drain Voltage and Drain Current . . . . . . . . . . . . . . . . . . . . . . . 85 5.4.2 Channel Temperature Distribution with Effective Gate Volt- age and Drain Current . . . . . . . . . . . . . . . . . . . 85 5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 vi Table of Contents 6 Conclusions and Future Directions . . . . . . . . . . . . . . . . . . 94 6.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 6.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 6.2.1 Fundamental Limitations . . . . . . . . . . . . . . . . . 95 6.2.2 Limitations of First Implementation . . . . . . . . . . . . 96 6.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Appendices A BSIM3 Model Parameters . . . . . . . . . . . . . . . . . . . . . . . 106 B SPICE Netlist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 C Numerical Solution of the Heat Equation . . . . . . . . . . . . . . . 112 D HeatMOS\u00C2\u00A9 Development . . . . . . . . . . . . . . . . . . . . . . . . 124 D.1 Development Environment . . . . . . . . . . . . . . . . . . . . . 124 D.2 A Sample HeatMOS\u00C2\u00A9 Execution . . . . . . . . . . . . . . . . . 124 vii List of Tables 2.1 Thermal and electrostatic analogies . . . . . . . . . . . . . . . . 19 2.2 Thermal and electrical conductivities (300 K) . . . . . . . . . . . 19 2.3 Thermal and electrical versions of heat transfer or capacitance . . 20 A.1 BSIM3 Parameters used in HeatMOS\u00C2\u00A9 . . . . . . . . . . . . . . 107 A.2 Other parameters used in HeatMOS\u00C2\u00A9 . . . . . . . . . . . . . . . 108 viii List of Figures 1.1 Transistor thermal resistance versus CMOS technology scaling . . 3 1.2 Impact of technology scaling on temperature increase . . . . . . . 5 1.3 SoC power consumption trends . . . . . . . . . . . . . . . . . . 11 1.4 A typical micro-chip layout . . . . . . . . . . . . . . . . . . . . . 12 1.5 Interconnect-transistor network . . . . . . . . . . . . . . . . . . . 13 2.1 Conservation of heat (energy) within a differential slice of a resis- tor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.2 Cross-sectional view of an nMOS transistor . . . . . . . . . . . . 23 3.1 A typical I-V characteristic of an nMOS transistor . . . . . . . . . 28 3.2 A comparison for I-V characteristics of an nMOS transistor . . . 30 3.3 Simplified 2-D cross-section of an nMOS transistor . . . . . . . . 31 3.4 3D view of the channel geometry for thermal model . . . . . . . . 34 ix List of Figures 3.5 3D view of region 1 of the channel . . . . . . . . . . . . . . . . . 42 3.6 3D view of the region 2 of the channel . . . . . . . . . . . . . . . 45 3.7 Resistor M-network used for EMR verification . . . . . . . . . . 55 4.1 VLSI CAD flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 5.1 An nMOS layout . . . . . . . . . . . . . . . . . . . . . . . . . . 64 5.2 I-V characteristics of nMOS device under test produced by Heat- MOS\u00C2\u00A9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 5.3 SourceTemp = 565K | DrainTemp = 567K | Re f Temp = 456K | Vgs = 1.0V |Vds = 1.5V | Ids = 2mA | v7\u00E2\u0088\u0092nmos.spice . . . . . . . 68 5.4 SourceTemp = 565K | DrainTemp = 567K | Re f Temp = 456K | Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 70 5.5 SourceTemp= 10K |DrainTemp= 10K | Re f Temp= 10K |Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . . . . 71 5.6 SourceTemp = 325K | DrainTemp = 325K | Re f Temp = 300K | Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 73 5.7 SourceTemp = 325K | DrainTemp = 325K | Re f Temp = 300K | Vgs = 1.8V |Vds = 2.2V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 75 5.8 SourceTemp = 325K | DrainTemp = 325K | Re f Temp = 300K | Vgs = 5.0V |Vds = 5.0V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 77 x List of Figures 5.9 SourceTemp = 325K | DrainTemp = 325K | Re f Temp = 300K | Vgs = 5V |Vds = 5V | v7\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . . . 78 5.10 SourceTemp = 318K | DrainTemp = 324K | Re f Temp = 298K | Vgs = 1.2V |Vds = 1.8V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 80 5.11 SourceTemp = 300K | DrainTemp = 325K | Re f Temp = 300K | Vgs = 1.0V |Vds = 0.1V | v8\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 82 5.12 SourceTemp = 318K | DrainTemp = 324K | Re f Temp = 300K | Vgs = 1.0V |Vds = 0.1V | v7\u00E2\u0088\u0092nmos.spice . . . . . . . . . . . . . 83 5.13 The average channel temperature versus the drain current and the drain voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5.14 The maximum channel temperature versus the drain current and the drain voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.15 The maximum channel temperature versus the drain current and the drain voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 88 5.16 The maximum channel temperature versus the drain current and the effective gate voltage . . . . . . . . . . . . . . . . . . . . . . 89 5.17 A typical cross-section of an HEMT device . . . . . . . . . . . . 91 5.18 Experimental channel temperature of HEMT device . . . . . . . 92 6.1 FinFET typical layout and schematic cross-sectional structures (Hisamoto et al., 2000) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 xi List of Figures D.1 A sample HeatMOS run part 1 . . . . . . . . . . . . . . . . . . . 125 D.2 A sample HeatMOS run part 2 . . . . . . . . . . . . . . . . . . . 126 xii List of Acronyms 1D One Dimensional 2D Two Dimensional 3D Three Dimensional \u00C2\u00B5m Micro-meter ASIC Application Specific Integrated Circuit CAD Computer-Aided Design CMOS Complementary Metal-oxide Semiconductor EMR Electro-migration Reliability GDSII Graphic Database Information Interchange HEMT High Electron Mobility Transistor IC Integrated Circuit K Kelvin mA Mili-ampere xiii List of Acronyms MOS Metal-oxide Semiconductor nm Nano-meter SoC System-on-a-Chip SOI Silicon on Insulator V Volt VLSI Very Large Scale Integration xiv Acknowledgements I would like to thank Prof. Andrew Labun for the supervising my research and his guidance. The concept of using thermal and electrostatic analogies shown in chapter 2 are based on an unpublished paper by Prof. Labun. I am very thankful for the support that Prof. Andrew has given me. I would like to thank Prof. Carolyn Labun for reviewing my english style and her suggestions to improve my presentation and writing skills. I would like to thank Prof. Thomas Johnson for reviewing the research work and providing further guidance to complete the thesis. I was a member of the Multiphysics research group led by Prof. Labun and I would like to thank other group members including Karan, Ska-hiish, Jason Liao, and Sadegh as well. I would like to thank Prof. Abbas Milani with whom I worked on a different project. I would like to thank my graduate supervising committee members, Prof. An- drew Labun, Prof. Thomas Johnson, Prof. Stephen O\u00E2\u0080\u0099 Leary, and Prof. Kian Mehravaran. xv Acknowledgements I would like to thank Prof. Deobrah Roberts, Associate Director Graduate Studies and Research. I would like to thank my external examiner Prof. Robert Sobot from University of Western Ontario for reviewing my thesis. Finally, I would like to thank everyone (including department of graduate studies staff, faculty of engineering staff, other faculty members, and fellow students) who have contributed towards this research work indirectly. xvi Chapter 1 Introduction 1.1 Background Verifying the reliability of complex integrated circuits has become an increasingly complex design challenge as device geometries in fabrication processes continue to shrink (Alam et al., 2007). The low cost of fabricating integrated circuits (ICs) us- ing CMOS (complementary metal-oxide semiconductor) continues to be the dom- inant fabrication technology. Micro-chips fabricated using CMOS technology are very complex and complete systems (SoC - system-on-a-chip) are common in the semiconductor industry. Predicting the operating temperature of these micro-chips should be done as early as possible to minimize the expense of resolving thermal issues late in the design phase. The ITRS (International Technology Roadmap for Semiconductors) predicts that electro-thermal reliability will continue to present significant challenges as device geometries continue to shrink below 22 nm (ITR, 2009). Due to the high tem- peratures that can arise in current CMOS technologies, it has become extremely important to accurately estimate the temperatures within the micro-chip for EMR 1 1.1. Background (electromigration reliability) verification. Electromigration is a term applied to the transport of mass in metal and is accelerated by the Joule heating, the heat gen- erated by current flowing through interconnects and devices (Black, 1969). Metal atoms are pushed by strong dc electron currents, sometimes resulting in voids in the interconnect (causing open circuits) or extruding through dielectric (causing short circuits). The Joule melting resulting from extremely high electrical current has been discussed in Fantini and Morandi (1985). There is a need to model the micro- chip temperature at a detailed level, during post-layout reliability verification, and to identify the specific metal interconnects most at risk for EMR. If certain parts of the micro-chip are heated beyond their reliability limit, it may cause a breakdown of the functionality and reliability in the whole system. The thermal trends for the CMOS technology are demonstrated by recent data pre- sented by the ITRS. As device sizes are reduced to nano-scale gate lengths the power density will continue to rise. Although the power density increases, the ma- terial properties of silicon, dielectric layers and metal layers remain constant and the effective thermal conductivity per unit of power will decrease. As a result the effective thermal resistance will increase per microwave of power (Semenov et al., 2003) (Figure 1.1). Several factors contribute to the increase in power density as the device geometries shrink. First, the geometry is shrinking faster than the scaling in the supply volt- age which directly contributes to an increase in power density. Second, the high frequency performance of CMOS devices improves as device geometry shrinks which means that the average clock rate in digital circuits is expected to increase. 2 1.1. Background 1 1.5 2 2.5 3 3.5 4 0.1 0.15 0.2 0.25 0.3 0.35 Th er m al re si st an ce (C /uW ) CMOS technology generation (um) Thermal resistance Figure 1.1: Transistor thermal resistance versus CMOS technology scaling 3 1.1. Background Dynamic switching results in power losses that are proportional to \u00CE\u009BCV 2 f , where C is the capacitance, f is the clock frequency, \u00CE\u009B is the activity factor (0\u00E2\u0089\u00A4 \u00CE\u009B\u00E2\u0089\u00A4 1), and V is the applied voltage (Sedra and Smith, 2007; Weste and Harris, 2004). Parallel efforts are made in reducing capacitance (i.e. device size), voltage, and ac- tivity factor (i.e. introduction of power management subsystem), at the same time dissipation is proportional to the square of the voltage, and voltage scaling has not kept pace with device scaling. These factors have been analyzed by the ITRS and projections on the relative increase in the operating temperature of devices are shown in Figure 1.2 (Semenov et al., 2003). The junction temperature rise will grow very rapidly and the importance of analyzing thermal profiles and the impact on EMR will become increasingly important. The major contribution to the rising micro-chip temperatures is due to device and interconnect heating. The heat generated due to electric current flow is called Joule heating. If there is a way to estimate the temperature of a micro-chip from its layout, before manufacturing, we would be able to change the design to rectify the design issues arising from the thermal hot spots. Temperature estimation in the design phase of the integrated circuits will improve the confidence of designing for reliability and reduce the overall design time of a production micro-chip. The influence of the temperature on failure mechanisms which occur in the temperature range of -55\u00C2\u00B0C to 125\u00C2\u00B0C has been discussed by Pecht et al. (1992). Computer modeling of metallization failure of the integrated circuits due to electromigration has been discussed in Kirchheim and Kaeber (1991) and Sukharav et al. (2009). There are some industrial solutions for estimating interconnects temperature. Soft- 4 1.1. Background 1 1.5 2 2.5 3 3.5 4 0.1 0.15 0.2 0.25 0.3 0.35 N or m al iz ed te m pe ra tu re in cr ea se o f C M O S ch ip CMOS technology generation (um) W/L ratio is constant, W and L are scaled Figure 1.2: Impact of technology scaling on temperature increase 5 1.1. Background ware tools to estimate the micro-chip temperature at the package level are dis- cussed by Melamed et al. (2009) and Oprins et al. (2009). Most of these tools use numerical approaches for temperature estimation which are slow and prohibitive for designs with millions of transistors. There are two main sources of heat in integrated circuits. These are: the dynamic power generated by digital switching signals and the Joule heating generated by the resistance in the interconnects and the devices. Most of the power dissipation in large micro-chips is generated by digital switching signals where the dynamic power is proportional to fV 2, where f is the switching frequency and V is the applied voltage. On the other hand, Joule heat is proportional to I2, where I is the electric current. In today\u00E2\u0080\u0099s complex integrated circuits, switching frequencies are in the Giga Hertz (GHz) range while typical interconnect current and device currents are in the milliampere range. Several approaches have been proposed to analyze the thermal performance of complex micro-chips. Analytical treatments of the 1-D steady state heat equation in IC interconnect have been presented previously in the context of interconnect temperature scaling analysis by Ajami et al. (2005), Im et al. (2002), and Chuan et al. (2009). An analytical solution of the interconnect heating and temperature estimation has been presented by Labun and Jagjitkumar (2008). The focus of this research work is to develop a model for estimating device tem- perature based on device layout and steady state bias voltages. 6 1.2. Motivation 1.2 Motivation This research work is motivated by the need to have design tools to predict the steady state spatial variation of temperature on the integrated circuit in the design phase before the micro-chip is fabricated. A standard integrated design flow pro- cess primarily focuses on verifying the electrical performance of the design over process, temperature and voltage changes. However, design tools to model and predict the temperature behavior are not readily available. If efficient temperature modeling tools can be developed, then designers will have the ability to improve the reliability of designs in the early design phase rather than waiting for experi- mental results from a failure analysis. Given the complexity of three-dimensional (3D) structures in complex integrated circuits, it is very important to have an efficient way to predict the temperature characteristics of the micro-chip. Brute force numerical solutions of the heat equa- tion are computationally prohibitive for large IC\u00E2\u0080\u0099s with millions of devices and multiple layers of interconnect. Therefore, a key focus of this research work is to investigate efficient solutions to estimate thermal characteristics that are scalable to very large and complex micro-chips. In this research project, analytic models are developed to construct a thermal net list for the complete micro-chip. The net list is analogous to an electrical net list for a circuit. In this approach, a thermal circuit model is developed which can be used to predict the temperature. 7 1.2. Motivation The Multiphysics research group at UBC has developed a thermal modeling soft- ware tool called Therminator3D. This tool can be used to simulate the overall tem- perature profile of a micro-chip providing thermal models are developed for the interconnects and devices. Thermal models for interconnect have already been de- veloped by other researchers (Labun and Jagjitkumar, 2008). On the other hand, thermal models for devices is much more limited and a tractable model that can be incorporated in a thermal simulation of the micro-chip is required. The goal of this research project is to address this limitation and develop an analytic model for CMOS devices. The thermal effects in the MOSFET (Metal-Oxide Semiconductor Field Effect Transistor) are complex and multidimensional because the generated heat has spa- tial as well as time dependent behavior. In order to accurately estimate the tem- perature due to the generated heat, one must solve the differential heat equation. The heat is generated due to electric current flow and can be estimated by Joule heating. There can be different ways to solve the differential equation (e.g. finite element, finite difference (Chuan et al., 2009) and (Kuball et al., 2007), analytical), but in order to estimate the temperature rapidly so that it can be feasible to esti- mate the temperature of the full micro-chip during the design, analytic solutions are preferred. There has been some work done in this direction and the analyti- cal approach has been successfully implemented in order to estimate interconnect temperature (Labun and Jagjitkumar, 2008). However the authors did not analyze the device temperature which is the primary focus of this research. Figure 1.1 shows that the device thermal resistance increases as the technology scales down (Semenov et al., 2003), which in turn leads to the higher temperatures. 8 1.2. Motivation The most effective way to meet the reliability target for each technology node is to incorporate design-for-reliability at the start of each technology generation. The most effective way to meet the reliability of each technology node is to have com- plete built-in-reliability and design-for-reliability solutions available at the start of development of each technology generation. This requires an in-depth understand- ing of the physics of each failure mechanism and the development of powerful and practical reliability engineering tools. High temperatures not only cause significant changes in the delay characteristics for both the transistors and the interconnects but also degrade micro-chip reliability. With technology scaling, micro-chip power densities are increasing that has led to high operating temperatures and the large thermal gradients, thereby resulting in serious reliability concerns (Tsai and Kang, 2000). Scaling the technology results in higher transistor density and higher clock frequencies. Higher clock frequencies will increases dynamic switching power be- cause the dissipation in switching driving a capacitive load is proportional to fV 2, where f is the clock frequency and V is the applied voltage (Weste and Harris, 2004). As device technology scales down, there is an exponential increase in the temper- ature 4T between the MOSFET junction and the ambient temperature (Semenov et al., 2003). For example, as shown in Figure 1.2, 4T has increased by a fac- tor of three as CMOS gate lengths have reduced from 0.35 um to 0.13 um. The rapid increase in the thermal density with a reduction in scaling has increased the importance of analyzing the temperature. High temperatures or uneven tempera- ture distributions may result not only in the reliability issues, but also in the timing failures of micro-chips. To resolve these issues, high-quality, accurate thermal 9 1.3. Research Goals analysis, and thermally oriented placement optimizations are essential prior to fab- rication (Tsai et al., 2006). Other predictions in terms of the thermal trends in the nano-scale CMOS devices includes the total dissipation of the complex system-on-a-chip (SoC) circuits. ITRS data include for SoC dissipation is shown in Figure 1.3. The data shows that in ten years micro-chips could dissipate as much as 400 watts of heat energy(ITR, 2009). These trends demonstrate that the significant advances in terms of thermal management will be essential to keep junction temperatures within a reasonable operating temperature range. EMR failures become very significant at junction temperatures over 125 C. 1.3 Research Goals Figure 1.4 shows a typical layout of a micro-chip. The devices are connected via interconnects. The interconnects can be represented by an equivalent resis- tor network for EMR verification. Each net in the interconnects network can be represented by equivalent pi-network (Labun and Jagjitkumar, 2008) for thermal analysis, and interconnect network models are required for an interconnect metal- lization and CMOS devices. Therefore an interconnect-transistor resistor network can be used to do a full micro-chip thermal analysis as shown in Figure 1.5. The transistor in this representation may be considered as a special type of resistor. One of the objectives of this research work is to find an equivalent resistor network rep- resentation of a transistor which can be integrated in interconnect-transistor resistor network for rapid full micro-chip thermal analysis. 10 1.3. Research Goals 50 100 150 200 250 300 350 400 450 500 2008 2010 2012 2014 2016 2018 2020 2022 2024 Po w er c on su m pt io n (W ) SoC power consumption trends SoC power consumption Figure 1.3: SoC power consumption trends 11 1.3. Research Goals Figure 1.4: A typical micro-chip layout 12 1.3. Research Goals Figure 1.5: Interconnect-transistor network 13 1.4. Scope and Organization The main objective of this research work presented in this thesis is to model the heating behavior of a transistor and to find an analytical solution for the tempera- ture estimation. The research includes the following: \u00E2\u0080\u00A2 To develop a mathematical model for the Joule heating due to switching drain current in a MOSFET channel. \u00E2\u0080\u00A2 To find an analytical equation which solves the heat equation. \u00E2\u0080\u00A2 To estimate the temperature profile of the channel. \u00E2\u0080\u00A2 To develop a software package for estimating the channel temperature from the extracted device layout information and using information from a com- pact model. In this work, the SPICE BSIM3 format was chosen because of its relative simplicity. \u00E2\u0080\u00A2 To develop an equivalent resistor network representation of the channel which can be integrated into the interconnect-transistor resistor network for full micro-chip thermal analysis. \u00E2\u0080\u00A2 To validate the analytical solution using a numerical solution. 1.4 Scope and Organization The scope of this thesis is limited to the research objectives, problem definition, so- lution and comparative studies with numerical solution of the heat equation using the finite difference methods. The thesis treats the heating problems in integrated 14 1.4. Scope and Organization circuits and its solution is based on an analytical approach. Chapter 2 provides the background of the heating in the integrated circuits along with the problem defini- tion of the MOSFET channel heating. Chapter 3 provides the mathematical model and the analytical solution for MOSFET channel temperature estimation. Chapter 4 describes HeatMOS\u00C2\u00A9, a software package that has been developed through this research, implementing the analytical solution to estimate the channel temperature. Chapter 5 shows the results of HeatMOS\u00C2\u00A9 which are compared with numerical so- lutions using a finite difference solver. The conclusions and the future directions for this research are discussed in Chapter 6. The appendices provide other relevant informations such as different device parameters, SPICE netlist, and a pseudo code for the numerical solver. 15 Chapter 2 Heating in Integrated Circuits The spatial variation of temperature across an integrated circuit (IC) is caused by charge carrier flow through the devices and the interconnects. The temperature variation in ICs is predominantly due to device heating and package level heat transport and is a subject of various chip-scale modeling approaches (Li et al., 2006). In the case of MOSFETs, the applied voltages at it its terminals create the longitudinal and transverse electric fields and charge carriers move from the source end to the drain end via drift and diffusion mechanisms (Park et al., 1991). Since the channel is a conducting path for the electric current, it can be considered as a resistor and the electric current flow in the channel leads to the Joule heating. The Joule heating causes the rise in the channel temperature. Heat conduction in solids is governed by two principle mechanisms: diffusive heat conduction by electrons and heat conduction by lattice waves whose quanta are represented by phonons (Flik et al., 1992). Heat conduction by phonons in semi- conductors is reviewed by Abeles(Abeles, 1963). In silicon the heat conduction is dominated by phonons (Sinha and Goodson, 2006). The hot electrons near the drain end transfer some energy to phonons and this creates a transient, non- 16 Chapter 2. Heating in Integrated Circuits equilibrium, high temperature \"hot spot\" that effectively leads to higher thermal resistance (Sinha and Goodson, 2006). Thermal conductivity analysis and conductivity measurements for silicon are pro- vided by (Morris and Hust, 1961). The detailed thermal analysis at the dimensions less than a phonon mean free path (~100 nm) requires the transient solution of the Boltzmann transport equation (BTE), that may not be feasible for large, netlist- based, full micro-chip thermal analysis. The phonon transport of energy can be described by the BTE , which can capture the sub-continuum effects particularly important for short impulsive events such as ESD (electrostatic discharge). These sub-continuum effects are not modeled by the diffusion model of energy transfer. These sub-continuum effects should be accounted for during the transient analysis but in a steady state analysis the time-average temperature given by the diffusion model are appropriate. The BTE for the phonons can be written as: \u00E2\u0088\u0082e\u00E2\u0080\u009D \u00E2\u0088\u0082 t =\u00E2\u0088\u0092\u00CE\u00BD \u00C2\u00B7\u00E2\u0088\u0087e\u00E2\u0080\u009D+ e\u00E2\u0080\u009Deq\u00E2\u0088\u0092 e\u00E2\u0080\u009D \u00CF\u0084phonon +qelectron\u00E2\u0088\u0092phonon (2.1) where e\u00E2\u0080\u009D is the phonon energy per unit volume per unit solid angle, \u00CE\u00BD is the phonon velocity, e\u00E2\u0080\u009Deq is the phonon equilibrium energy density, \u00CF\u0084phonon is the phonon scat- tering rate, qelectron\u00E2\u0088\u0092phonon is the phonon energy absorption by hot electron. The effective thermal conductivity of a complex silicon FET structure is technology- specific and could be obtained by the detailed modeling, but this is beyond the 17 2.1. Electrostatic Analogy of Heat Conduction scope of the thesis, which is aimed at a very compact model for netlist-based anal- ysis. Our research is therefore directly applicable for the technology nodes above 180 nm (using the BSIM3 device model). The extension of this approach toward nanoscale technologies is discussed briefly in the Future Work. 2.1 Electrostatic Analogy of Heat Conduction The analytical methods developed in this research to model heat conduction in the devices are based on analogous relationships between the form of equations for the heat conduction and the electrostatics. The fundamental analogies between the heat and the electrostatics are shown by comparison of the governing equations in Table 2.1. The temperature T in a conduction problem in the presence of a heat sources \u00CF\u0086 is governed by the analogy to Poisson\u00E2\u0080\u0099s equation (Guass\u00E2\u0080\u0099 Law) in the electrostatics. Similarly, the heat flow is analogous to the current flow in the electric circuits. In these equations, \u00CE\u00BB is the thermal conductivity in W/mK, \u00CE\u00B5 is the permittivity in F/m, \u00CF\u0086 is the heat energy density in W/m3, \u00CF\u0081 is the electric charge density in C/m3, T is the temperature in K, V is the electric potential in V, q is the heat flux in W/m2, J is the electric current density in C/m2s. Use of both analogies solves the heating problem (Labun, 2009). Table 2.2 continues the analogy between electrical and thermal models by compar- ing material properties of thermal conductivity and electrical conductivity (Lide, 2004). Metals are good conductors of both heat and electrical current and di- electrics are poor conductors, although the ratio of conductance\u00E2\u0080\u0099s is only two orders 18 2.1. Electrostatic Analogy of Heat Conduction Table 2.1: Thermal and electrostatic analogies \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087 \u00C2\u00B7 (\u00CE\u00BB\u00E2\u0088\u0092\u00E2\u0086\u0092\u00E2\u0088\u0087T ) =\u00E2\u0088\u0092\u00CF\u0086 \u00E2\u0088\u0092\u00E2\u0086\u0092\u00E2\u0088\u0087 \u00C2\u00B7 (\u00CE\u00B5\u00E2\u0088\u0092\u00E2\u0086\u0092\u00E2\u0088\u0087V ) =\u00E2\u0088\u0092\u00CF\u0081 \u00E2\u0088\u0092\u00E2\u0086\u0092q =\u00E2\u0088\u0092\u00CE\u00BB\u00E2\u0088\u0092\u00E2\u0086\u0092\u00E2\u0088\u0087T \u00E2\u0088\u0092\u00E2\u0086\u0092J =\u00E2\u0088\u0092\u00CF\u0083\u00E2\u0088\u0092\u00E2\u0086\u0092\u00E2\u0088\u0087V Table 2.2: Thermal and electrical conductivities (300 K) Material \u00CE\u00BB (Wm\u00E2\u0088\u00921K\u00E2\u0088\u00921) \u00CF\u0083 (\u00E2\u0084\u00A6\u00E2\u0088\u00921m\u00E2\u0088\u00921) Cu 398 5.7\u00C3\u0097107 Al 237 3.7 W 174 1\u00C3\u0097107 Si (intrinsic) 83.5 10\u00E2\u0088\u00923 Si (doped) 60.5 10\u00E2\u0088\u00923 SiO2 1.4 10\u00E2\u0088\u009213 of magnitude for heat. The tabulated values also show that the range in thermal conductivity properties is much smaller than the range in electrical conductivity. For example, the thermal conductivity of copper is four times of magnitude better than silicon, while the the electrical conductivity of copper is 10 orders of magni- tude greater than silicon (Lide, 2004). The analogy with Possion\u00E2\u0080\u0099s equation leads to a capacitance/heat transfer coefficient analogy through the application of the divergence theorem to Poisson\u00E2\u0080\u0099s equation. Table 2.3 Row 1 shows the identical construction of integral equations which in the 19 2.2. Heat Equation Table 2.3: Thermal and electrical versions of heat transfer or capacitance \u00C2\u00B4 Volume \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087 (\u00CE\u00BB \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087T )d\u00E2\u0084\u00A6= \u00C2\u00B8 Area (\u00CE\u00BB \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087T )d \u00E2\u0088\u0092\u00E2\u0086\u0092 A = \u00CF\u0086Volume \u00C2\u00B4 Volume \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087 (\u00CE\u00B5\u00E2\u0088\u0092\u00E2\u0086\u0092V )d\u00E2\u0084\u00A6= \u00C2\u00B8 Area (\u00CE\u00B5 \u00E2\u0088\u0092\u00E2\u0086\u0092 \u00E2\u0088\u0087V )d \u00E2\u0088\u0092\u00E2\u0086\u0092 A = QVolume 1 RT = \u00CF\u0086T C = Q V electrical case are solved to obtain capacitance. The analogous thermal model for heat transfer shows that the reciprocal of the thermal resistance (RT ) is equal to the ratio of the heat density divided by the temperature difference across the material. The relationship between the heat conduction and the electrostatics is completed using Ohm\u00E2\u0080\u0099s law to relate the current flow with the heat flux. The Joule heating in a wire gives I2R= \u00CF\u0086Volume. The heat is conducted through the dielectrics according to the Ohm\u00E2\u0080\u0099s Law analogy. Thus, capacitance tables (in homogeneous dielectrics) can be read as thermal resistance tables by multiplying by a constant factor. 2.2 Heat Equation Before developing a heat model for a MOSFET channel, a model for heating in a resistor is derived first. Consider a resistor of uniform width and composition as shown in Figure 2.1. The resistor temperature variation over its cross-section is neglected. For metal interconnects embedded in a dielectric, this can be readily justified, since (by Table 2.2) the thermal conductivity within the resistor is much greater than the thermal conductivity of the surrounding insulator. In MOSFETs 20 2.2. Heat Equation Figure 2.1: Conservation of heat (energy) within a differential slice of a resistor the electric current is also confined to a well-defined region (the channel). Heat flow is confined by a poor thermal conductor above (the SiO2 gate dielectric) and also below, in SOI devices. However, even in bulk MOSFETs, the useful approx- imation of a uniform temperature over the channel cross-section is not too far off the mark because of the localization of the heat source. With reference to Figure 2.1, the heat per unit length, mCpT (x)4x in a differential slice of a resistor segment of length 4x, where T (x) is the temperature along the resistor\u00E2\u0080\u0099s length, m is the mass per unit length, and Cp is the heat capacity. The time rate of change of the heat is then given by (Labun and Jagjitkumar, 2008) mCp \u00E2\u0088\u0082T \u00E2\u0088\u0082 t 4x= F(x+4x)\u00E2\u0088\u0092F(x)+\u00CF\u0086(x)4x\u00E2\u0088\u0092 f (x)4x (2.2) 21 2.3. Heat Generation in MOSFETs Where F(x) is the diffusive heat flux. The heat flux is oriented along the resistor\u00E2\u0080\u0099s length and is related to the temperature gradient by Fourier\u00E2\u0080\u0099s law of heat conduc- tion, including the \u00E2\u0080\u009Clongitudinal\u00E2\u0080\u009D thermal conductance: F(x) = Glong \u00E2\u0088\u0082T (x) \u00E2\u0088\u0082x (2.3) The Joule heat per unit length is given by \u00CF\u0086 = I2rmsR ; R is the electrical resistance per unit length, assumed constant within a resistor; Irms is the time-averaged rms current. The function f (x) in equation (2.2) models the heat flux through the sides of the resistor to the surrounding environment at a temperature T re f . If there is diffusive heat flux through the sides of the resistor to the local environment at some uniform, constant temperature T re f (established, for example, by a full-chip temperature simulation including device heating), then one may define a \u00E2\u0080\u009Clateral\u00E2\u0080\u009D thermal conductance Glat per unit length f (x) = GlatT (x)\u00E2\u0088\u0092T re f (2.4) 2.3 Heat Generation in MOSFETs When the transistor is on and is in operation, the charge carriers move from the source to the drain and in this process these loose energy due to physical rea- sons such as collisions, scattering, phonon-electron interaction etc. The dissipated 22 2.3. Heat Generation in MOSFETs Figure 2.2: Cross-sectional view of an nMOS transistor energy is released in the form of heat and leads to an increase in the device tem- perature. A cross-sectional view of a MOS transistor has been shown in Figure 2.2. The transistor is said to be in operation when there is a channel formed by applying appropriate voltages across substrate, source, drain, and gate terminals. The channel can be considered as a resistor of variable cross-sectional area, and the heat generated in the channel can be approximated by Joule heating similar to a resistor. Therefore, heat generation is concentrated in the region of the highest resistance, the pinched-off region of the channel near the drain end. 23 2.4. Summary 2.4 Summary The heat generation in transistors has been presented with a focus on transistor channel heating. The basic equation for Joule heating has been explained, and it was shown that the heat equation is analogous to the solution of the Poisson equation in electrostatics. The basic heat equation for the Joule heating has been explained. 24 Chapter 3 Thermal Model and Analytical Solution The one-dimensional heat equation (2.2) is applied to the thermal model of the MOSFET and solved analytically to give a closed-form expression for the channel temperature from the source to the drain. The thermal model uses the BSIM3 parameters. 3.1 BSIM3 MOSFET Model BSIM3 is a physics-based, accurate, scalable, robust and predictive industry stan- dard MOSFET SPICE model for circuit simulation and CMOS technology devel- opment. It was developed by the BSIM Research Group in the Department of Electrical Engineering and Computer Sciences (EECS) at the University of Cal- ifornia, Berkeley. BSIM3 is the industry-standard MOSFET model. The third iteration of BSIM3 Version 3 (commonly abbreviated as BSIM3v3), was estab- lished by SEMATECH as the first industry-wide standard model for MOSFETs in 25 3.1. BSIM3 MOSFET Model December of 1996. The development of BSIM3 is based on Poisson\u00E2\u0080\u0099s equation us- ing the gradual channel approximation and coherent quasi 2D analysis, taking into account the effects of device geometry and process parameters. BSIM3 is appro- priate for MOSFETs for 180 nm MOSFET technology and larger. BSIM3 models the following physical characteristics in MOSFET devices (Liu et al., 1999): \u00E2\u0080\u00A2 Short and narrow channel effects on threshold voltage. \u00E2\u0080\u00A2 Non uniform doping effects. \u00E2\u0080\u00A2 Mobility reduction due to vertical field. \u00E2\u0080\u00A2 Bulk charge effect. \u00E2\u0080\u00A2 Velocity Saturation. \u00E2\u0080\u00A2 Drain-induced barrier lowering (DIBL). \u00E2\u0080\u00A2 Channel length modulation (CLM). \u00E2\u0080\u00A2 Substrate current induced body effect (SCBE). \u00E2\u0080\u00A2 Sub-threshold conduction. \u00E2\u0080\u00A2 Source/Drain parasitic resistances. BSIM3 uses a group device extraction strategy for parameter extraction. This re- quires measured data from devices with different geometries. All devices are mea- sured under the same bias conditions. The BSIM3 device parameters for a 180 nm CMOS technology have been used in this research. The model parameters are given in appendix A. 26 3.2. MOSFET Channel 3.2 MOSFET Channel A simplified model of the MOSFET channel accounts for many characteristics of MOSFET devices and is sufficient for the compact thermal model developed in this thesis. The channel is a cloud of free charge carriers beneath the gate surface which conducts electric current. In the case of nMOS, the electron cloud forms the channel. The electron density is greater at the source end compared to the drain end and decreases towards the drain end (Masu and Tsubouchi, 1994). An example of the typical IV characteristics of an nMOS transistor is shown in Figure 3.1 (Weste and Harris, 2004). The electric current increases as we increase the drain voltage and after a certain voltage, the current becomes constant and the channel is said to be pinched off. Initially when there is no drain voltage, the electrons are attracted towards the positive gate voltage (enhancement mode n- channel MOSFET) and it is clear that the Joule heat constitutes only a small part of the micro-chip power consumption as compared to the power consumption due to switching activity. When the gate voltage is greater than the threshold voltage and the drain voltage is less than the gate voltage, the transistor is turned on, and a channel has been created which allows current to flow between the drain and the source. The MOSFET operates like a resistor controlled by the gate voltage relative to both the source and the drain voltages. Now, if the drain voltage is increased and becomes greater than the gate voltage, the transistor is said to be in the saturation or the active mode of operation. Since the drain voltage is higher than the gate voltage, the electrons spread out, and conduction is through a broad, three-dimensional channel and the current extends away from the interface. The onset of this region is also known as pinch-off to indicate the lack of channel region 27 3.2. MOSFET Channel Figure 3.1: A typical I-V characteristic of an nMOS transistor near the drain. The drain current is now weakly dependent upon drain voltage and controlled primarily by the gate\u00E2\u0080\u0093source voltage. Since the channel provides a conducting path for electric current, it can be consid- ered as a resistor. The cross-sectional area of this resistor is higher at the source end and continuously decreases towards the drain end. When the gate voltage is greater than the threshold voltage, Vgs > Vth and Vds \u00E2\u0089\u00A5 (Vgs\u00E2\u0088\u0092Vth), where Vds is the drain voltage, the channel depth at drain end decreases to almost zero, and the channel is said to be pinched-off (Weste and Harris, 2004). Beyond the pinch-off, the deple- tion region around the drain end provides a path for the drain current to flow. In the channel thermal model the region between the pinch-off point and the drain end 28 3.2. MOSFET Channel can be considered as a region of uniform cross-section. The conductivity of this channel resistor depends on complex physical parameters including area, doping concentration, mobility of charge carriers etc. The channel is controlled between the oxide layer and the substrate layer in vertical direction, and controlled between the source well and the drain well in horizontal direction as shown in Figure 3.3. Let us consider the case of an nMOS transistor. Initially, when the gate voltage Vgs is less than the threshold voltage Vth, and the drain voltage is at the same voltage as the source, an inversion layer is formed beneath the gate, and the channel is said to exist between the source and the drain. The inversion layer can be assumed to have a uniform cross-sectional area from the source end to the drain end of the transistor. Now, if the gate voltage is increased beyond the threshold voltage of the transistor, a sufficient number of electrons accumulate in the inversion layer to conduct electric current. The electric current increases with the increase in the gate voltage. Since the drain and the source are at same voltage at this point, the height of the inversion layer, and hence the channel remain uniform (Weste and Harris, 2004). Now, if the drain voltage is increased, the voltage causes an electric current to flow between the source and the drain. The electric current flowing in the channel Ids will be proportional to the effective gate voltage Vgs\u00E2\u0088\u0092 Vth, and to the voltage Vds which causes Ids to flow. Under these conditions the channel is said to be in the linear or ohmic region. AsVDS continues to increase, the channel height near the drain end starts decreasing. However, in the linear mode of operation, the channel height at the drain end is approximately the same as that at the source end (HP w 0.99HO). IfVds > (Vgs\u00E2\u0088\u0092Vth), Ids no longer increases linearly and becomes constant. The transistor is said to operate in the saturation region or in 29 3.2. MOSFET Channel Figure 3.2: A comparison for I-V characteristics of an nMOS transistor 30 3.2. MOSFET Channel Figure 3.3: Simplified 2-D cross-section of an nMOS transistor the on state. Other second order effects in the BSIM3 model such as drain-induced barrier lowering (DIBL), substrate current induced body effects (SCBE) have not been incorporated in HeatMOS\u00C2\u00A9 to calculate effective drain current, which are typically modeled in industrial quality SPICE simulators. A comparison of the the I-V characteristics of simplified device model used in HeatMOS\u00C2\u00A9 with the full BSIM3 model (ngspice simulator) is shown in Figure 3.2. The saturation current calculation equation will be explained in section 5.2. 31 3.2. MOSFET Channel 3.2.1 Channel Geometry for Thermal Model Figure 3.3 shows the channel region in a MOSFET. In the saturation region of operation the channel shrinks near the drain end and the drain current becomes almost constant. The drain current flows through the depletion region at the drain end. For the thermal model in saturation mode the channel is modeled as two separate regions. \u00E2\u0080\u00A2 Region 1: Region 1 is a channel of constant width We f f and non-uniform height which tapers from the source end to the pinch-off point (Figure 3.4). This region is approximated by an exponential function of the form y = Ae\u00E2\u0088\u0092\u00CE\u00B1x where where y is the vertical height and x is the horizontal length of the channel. An exponential function is selected for the model to lead to tractable ana- lytic results and differs from the linear model used for the gradual channel approximation used to develop the electrical model for MOSFETs. How- ever, the parameters A and \u00CE\u00B1 in the exponential model are determined from the BSIM3 model parameters and are matched for the operating point of the device. If L1 is the length of the channel between the source end and the pinch-off point, HO is the height of the channel at the source end and HP is the height of the channel at the pinch-off point then x= 0\u00E2\u0087\u0092 y= HO 32 3.2. MOSFET Channel x= L1\u00E2\u0087\u0092 y= HP HP = HOe\u00E2\u0088\u0092\u00CE\u00B1L1 \u00E2\u0087\u0092 \u00CE\u00B1 = 1L1 ln ( HO HP ) y= HOe \u00E2\u0088\u0092 { 1 L1 ln ( HO HP )} x (3.1) \u00E2\u0080\u00A2 Region 2: Region 2 models the thermal conduction through the depletion region of the pinched off channel. In this region, the channel cross-section is uniform and has width We f f and height HP. The channel length in region 2 is denoted as L2. If L is the total length of the channel then L= L1 +L2 Figure 3.4 shows the 3D view of channel formed beneath the gate. 33 3.2. MOSFET Channel Figure 3.4: 3D view of the channel geometry for thermal model 3.2.2 Electrical Resistance The conduction channel formed beneath the oxide layer is of non-uniform cross- sectional area in nature from the source end to the pinch-off point and has uniform cross-sectional area from the pinch-off point to the drain end. The resistance of non-uniform cross-sectional area of region 1 can be approximated by integrating the resistance of a differential slice over the length from 0 to L1: R1 = L1\u00CB\u0086 0 \u00CF\u0081e\u00CE\u00B1x We f fHO dx= \u00CF\u0081 We f fHO\u00CE\u00B1 ( e\u00CE\u00B1L1\u00E2\u0088\u00921) (3.2) \u00CE\u00B1 = 1 L1 ln ( HO HP ) (3.3) 34 3.2. MOSFET Channel Le f f = L1 +L2 (3.4) where, \u00CF\u0081 = Electrical resistivity HO = Height of channel at source end HP = Height of channel at pinch-off point We f f = Effective channel width L1 = Length of channel between source end and pinch-off point L2 = Length of channel between pinch-off point and drain end Le f f = Effective channel length The effective channel length and width can be approximated by (Liu et al., 1999) Le f f = LDrawn\u00E2\u0088\u00922dL (3.5) We f f =WDrawn\u00E2\u0088\u00922dW (3.6) dW =Wint + Wl LWln + Ww WWwn + Wwl LWlnWWwn (3.7) 35 3.2. MOSFET Channel dL= Lint + Ll LLln + Lw W Lwn + Lwl LLlnW Lwn (3.8) where, LDrawn = Layout drawn length of transistor. WDrawn = Layout drawn width of transistor. Wint ,Lint ,Wl,Ww,Wwl,Wwn,Ll,Lw,Lwn,Lwl = BSIM3 parameters. The electrical resistivity of the channel can be approximated as \u00CF\u0081 = 1 We f fCox\u00C2\u00B5e f f (3.9) where, Cox = Capacitance of oxide layer. \u00C2\u00B5e f f = Effective mobility of charge carriers. Region 2 has a uniform cross-sectional area. This can be considered as a uniform bar of length L2, width We f f , and height HP. Its resistance is given by R2 = \u00CF\u0081L2 We f fHP (3.10) 36 3.2. MOSFET Channel 3.2.3 Channel Height Models Initially when the channel starts taking its shape an inversion charge layer is formed beneath the oxide layer. The charge accumulates beneath the gate and forms an approximately uniform layer of a certain height and width from the source end to the drain end. It can also be approximated that when the channel takes on its non- uniform shape in the saturation region of operation, the drain end of the channel shrinks, while the source end of the channel remains same. The drain end of the channel shrinks to the height, HP, the height at the pinch-off point. The height of the channel at the source end can be approximated by the thickness of the inversion layer (Masu and Tsubouchi, 1994). The height of the channel at the source end is HO = tinv = kT\u00E2\u0088\u009A \u00CE\u00B5\u00CE\u00B5o 2qNA\u00CF\u0086s (3.11) \u00CF\u0086s = 2 kT q ln ( NA ni ) (3.12) where, k = Boltzmann\u00E2\u0080\u0099s constant T = Temperature \u00CE\u00B5 = Dielectric constant 37 3.2. MOSFET Channel \u00CE\u00B5o = Permittivity of free space q = Electron charge NA= Substrate doping concentration ni = Intrinsic silicon concentration \u00CF\u0086s= Surface potential We can approximate the height of the channel at the pinch off point HP by the following method. R1 +R2 = VDS Isat (3.13) \u00CF\u0081 ( e\u00CE\u00B1L1\u00E2\u0088\u00921) We f fHO\u00CE\u00B1 + \u00CF\u0081L2 We f fHP = VDS Isat Under the approximation HP <<< HO, HP = \u00CF\u0081L2Isat VDSWe f f (3.14) where Isat = Electric current in saturation region of operation VDS = Drain-Source voltage 38 3.2. MOSFET Channel The height of the channel at the drain end is the same as at the pinch-off point, HP, as it is assumed that the channel becomes uniform in nature from the pinch-off point to the drain end. 3.2.4 Channel Length Models The channel length between the pinch-off point and the drain can be calculated using the channel length modulation approximation (Park et al., 1991). The length of the channel between the pinch-off point to the drain is L2 = VDS\u00E2\u0088\u0092VDS\u00E2\u0080\u00B2 EL (3.15) where EL = \u00E2\u0088\u009A qNa(VDS\u00E2\u0088\u0092VDS\u00E2\u0080\u00B2) 2\u00CE\u00B5\u00CE\u00B5O VDS\u00E2\u0080\u00B2 w\u00E2\u0088\u0092vsatLe f f /\u00C2\u00B5e f f vsat = Saturation velocity of charge carriers So, the channel length between the source end and the pinch-off point is L1 = Le f f \u00E2\u0088\u0092L2 (3.16) 39 3.2. MOSFET Channel 3.2.5 Thermal Conductance There are two kinds of thermal conductances: lateral thermal conductance and longitudinal thermal conductance. The conduction channel is formed beneath the oxide layer. The substrate can be approximated as having a uniform background temperature T re f . The lateral thermal conductance must be taken to both the gate electrode above and a thermal reference somewhere in the substrate below. These may be combined into a single lateral conductance to a single, averaged reference temperature. Given the relative short channel length, when compared to the dis- tant reference temperature (e.g substrate), the small variation in the height of the channel compared to that great distance is negligible. Therefore, the variation of thermal conductance along the height of the channel can be considered negligible. The lateral thermal conductance (Glat) can be considered uniform along the length of the channel. For complex device structures a uniform, effective lateral thermal conductivity could be calibrated from detailed 2D or even 3D thermal conduction simulations. The longitudinal thermal conductance (Glong) vary along the length of the channel and is of concern. Since the height of the channel is not uniform, a tractable so- lution to the heat equation arises if we can drive an expression of average thermal conductance in a similar manner to the electrical resistance of the channel. This weights the thermal conductance according to the localization of the heat source. The average longitudinal thermal conductance is 40 3.2. MOSFET Channel Glongavg = 1 Le f f Le f f\u00CB\u0086 0 G(x)dx (3.17) where Le f f is the effective channel length. The longitudinal thermal conductance along the length is Glong(x) = GOe\u00CE\u00B3x (3.18) where \u00CE\u00B3 =\u00E2\u0088\u0092 1Le f f ln [ GO GL ] GO = Thermal conductance per unit length at source end (i.e. when x = 0 \u00E2\u0087\u0092 Glong(x= 0) = GO). GL = Thermal conductance per unit length at drain end (i.e. when x = L \u00E2\u0087\u0092 Glong(x= L) = GL). The solution to the equation 3.17 is Glongavg = GO \u00CE\u00B3Le f f [ e\u00CE\u00B3Le f f \u00E2\u0088\u00921] (3.19) If \u00CE\u00BB is the thermal conductivity of channel, then GO = HOWe f f \u00CE\u00BB and GL = HPWe f f \u00CE\u00BB . 41 3.3. Solution of the Heat Equation 2.2 Figure 3.5: 3D view of region 1 of the channel 3.3 Solution of the Heat Equation The basic heat equation 2.2 is solved for both regions in the following sections. 3.3.1 Region 1: Non-uniform cross-sectional area Figure 3.5 shows region 1 of the channel. For a differential length4x of the channel, \u00CF\u0086(x)4x= I2DSR(x)4x 42 3.3. Solution of the Heat Equation R(x)4x= \u00CF\u00814x We f fHO e\u00CE\u00B1x where IDS is the drain current. The heat flux can be written as F(x) = Glongavg \u00E2\u0088\u0082T (x) \u00E2\u0088\u0082x F(x+4x) = Glongavg \u00E2\u0088\u0082T (x+4x) \u00E2\u0088\u0082x f (x) = Glat ( T (x)\u00E2\u0088\u0092T re f ) In the limit when 4x\u00E2\u0086\u0092 0, the basic heat equation 2.2 takes the form of the 2nd order non-homogeneous differential equation. Glongavg d2T dx2 \u00E2\u0088\u0092Glat (T (x)\u00E2\u0088\u0092T re f )+ I2DS\u00CF\u0081 We f fHO e\u00CE\u00B1x = 0 (3.20) If the temperature function for region 1 is T1(x), equation 3.20 can be written as Glongavg d2T1 dx21 \u00E2\u0088\u0092Glat (T1(x1)\u00E2\u0088\u0092T re f )+ I2DS\u00CF\u0081We f fHO e\u00CE\u00B1x1 = 0 (3.21) 43 3.3. Solution of the Heat Equation The solution of non-homogeneous differential equation 3.21 can be written as T1(x1) =C1e\u00CE\u00BE1x1 +C2e\u00E2\u0088\u0092\u00CE\u00BE1x1 +T re f \u00E2\u0088\u0092 \u00CE\u00A8 2 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 e\u00CE\u00B1x1 (3.22) where \u00CE\u00BE1 = \u00E2\u0088\u009A Glat Glongavg \u00CE\u00A8= \u00E2\u0088\u009A I2sat\u00CF\u0081 We f fHOG long avg K1 = \u00CE\u00A82 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 C1 and C2 are constants whose values will be determined. 3.3.2 Region 2: Uniform cross-sectional area Figure 3.6 shows the 3D view of region 2 of the channel. The total resistance in region 2 is denoted as R2. For a differential length4x of the channel 44 3.3. Solution of the Heat Equation Figure 3.6: 3D view of the region 2 of the channel \u00CF\u0086(x)4x= I2DS ( R2 L2 ) 4x The heat flux can be written as F(x) = Glongavg \u00E2\u0088\u0082T (x) \u00E2\u0088\u0082x F(x+4x) = Glongavg \u00E2\u0088\u0082T (x+4x) \u00E2\u0088\u0082x f (x) = Glat ( T (x)\u00E2\u0088\u0092T re f ) 45 3.3. Solution of the Heat Equation In the limit when4x\u00E2\u0086\u0092 0, the basic heat equation 2.2 takes the form of the second- order non-homogeneous differential equation 3.23. Glongavg d2T dx2 \u00E2\u0088\u0092Glat (T (x)\u00E2\u0088\u0092T re f )+ I2DSR2 = 0 (3.23) If the temperature function of region 2 is T2(x), the equation 3.23 can be written as Glongavg d2T2 dx22 \u00E2\u0088\u0092Glat (T2(x2)\u00E2\u0088\u0092T re f )+ I2DSR2 = 0 (3.24) The solution of differential equation 3.24 can be written as T2(x2) = D1e\u00CE\u00BE2x2 +D2e\u00E2\u0088\u0092\u00CE\u00BE2x2 + \u00CE\u00B2 \u00CE\u00BE 22 (3.25) where \u00CE\u00BE2 = \u00E2\u0088\u009A Glat2 Glongavg \u00CE\u00B2 =\u00E2\u0088\u0092 Glat2 Glongavg T re f \u00E2\u0088\u0092 I 2 DS Glongavg R2 K2 = \u00CE\u00B2 \u00CE\u00BE 22 46 3.3. Solution of the Heat Equation D1 and D2 are constants whose values will be determined using the boundary con- ditions. 3.3.3 Boundary Conditions and Complete Solution The channel is between the source end and the drain end. The end point tempera- tures are determined by the boundary conditions. If the temperature of the source end is TO and the drain end is TL, then the two boundary conditions can be written as T1(x1) |x1=0= TO (3.26) T2(x2) |x2=L2= TL (3.27) Since the channel is one entity, the pinch-off point is a virtual point. The temper- ature at the pinch-off point as determined by equation 3.22 should be the same as it is determined by equation 3.25. Further the heat flux leaving region 1 enters into region 2. These facts give us two more conditions: T1(x1) |x1=L1= T2(x2) |x2=0 (3.28) Glongavg dT1 dx1 |x1=L1= Glongavg dT2 dx2 |x2=0 (3.29) 47 3.3. Solution of the Heat Equation Using equations 3.22, 3.25, 3.26, 3.27, 3.28, and 3.29, the unknown coefficients, C1, C2, D1, and D2 can be calculated. The values of these coefficients are given below. TO =C1 +C2 +T re f \u00E2\u0088\u0092 \u00CE\u00A8 2 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 (3.30) TL = D1e\u00CE\u00BE2L2 +D2e\u00E2\u0088\u0092\u00CE\u00BE2L2 + \u00CE\u00B2 \u00CE\u00BE 22 (3.31) C1e\u00CE\u00BE1L1 +C2e\u00E2\u0088\u0092\u00CE\u00BE1L1 +T re f \u00E2\u0088\u0092 \u00CE\u00A8 2 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 = D1 +D2 + \u00CE\u00B2 \u00CE\u00BE 22 (3.32) Glongavg ( C1\u00CE\u00BE1e\u00CE\u00BE1L1\u00E2\u0088\u0092C2\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092\u00CE\u00B1 \u00CE\u00A8 2 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 e\u00CE\u00B1L1 ) = Glongavg (D1\u00CE\u00BE2\u00E2\u0088\u0092D2\u00CE\u00BE1) (3.33) C1 =C \u00E2\u0080\u00B2 1\u00E2\u0088\u0092C \u00E2\u0080\u00B2\u00E2\u0080\u00B2 1 (3.34) D1 = D \u00E2\u0080\u00B2 1\u00E2\u0088\u0092D \u00E2\u0080\u00B2\u00E2\u0080\u00B2 1 (3.35) where 48 3.3. Solution of the Heat Equation C \u00E2\u0080\u00B2 1 = ( \u00CE\u00BE2 +\u00CE\u00BE2e2\u00CE\u00BE2L2 ) ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1+ e2\u00CE\u00BE2L2)\u00CE\u00BE2\u00E2\u0088\u0092 (e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2) \u00C3\u0097{ TLe\u00CE\u00BE2L2\u00E2\u0088\u0092 (TO\u00E2\u0088\u0092T re f )e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092T re f + \u00CE\u00A82 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 (e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)+ \u00CE\u00B2 \u00CE\u00BE 22 (1\u00E2\u0088\u0092 e\u00CE\u00BE2L2) } D \u00E2\u0080\u00B2 1 = ( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1+ e2\u00CE\u00BE2L2)\u00CE\u00BE2\u00E2\u0088\u0092 (e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2) \u00C3\u0097{ TLe\u00CE\u00BE2L2\u00E2\u0088\u0092 (TO\u00E2\u0088\u0092T re f )e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092T re f + \u00CE\u00A82 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 (e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)+ \u00CE\u00B2 \u00CE\u00BE 22 (1\u00E2\u0088\u0092 e\u00CE\u00BE2L2) } C \u00E2\u0080\u00B2\u00E2\u0080\u00B2 1 = ( 1\u00E2\u0088\u0092 e2\u00CE\u00BE1L1 ) ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1+ e2\u00CE\u00BE2L2)\u00CE\u00BE2\u00E2\u0088\u0092 (e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2) \u00C3\u0097{ (TO\u00E2\u0088\u0092T re f )\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092TLe\u00CE\u00BE2L2\u00CE\u00BE2 + \u00CE\u00A82 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 (\u00CE\u00B1e\u00CE\u00B1L1 +\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00B1L1)+ \u00CE\u00B2 \u00CE\u00BE 22 \u00CE\u00BE2e\u00CE\u00BE2L2 } 49 3.3. Solution of the Heat Equation D \u00E2\u0080\u00B2\u00E2\u0080\u00B2 1 = ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1+ e2\u00CE\u00BE2L2)\u00CE\u00BE2\u00E2\u0088\u0092 (e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1)(1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2) \u00C3\u0097{ (TO\u00E2\u0088\u0092T re f )\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092TLe\u00CE\u00BE2L2\u00CE\u00BE2 + \u00CE\u00A82 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 (\u00CE\u00B1e\u00CE\u00B1L1 +\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00B1L1)+ \u00CE\u00B2 \u00CE\u00BE 22 \u00CE\u00BE2e\u00CE\u00BE2L2 } C2 = TO\u00E2\u0088\u0092C1\u00E2\u0088\u0092T re f + \u00CE\u00A8 2 \u00CE\u00B12\u00E2\u0088\u0092\u00CE\u00BE 21 D2 = TLe\u00CE\u00BE2L2\u00E2\u0088\u0092D1e2\u00CE\u00BE2L2\u00E2\u0088\u0092 \u00CE\u00B2\u00CE\u00BE 22 e\u00CE\u00BE2L2 The temperature of the channel along its length from the source end to the drain end can be written as T (x) = \u00EF\u00A3\u00B1\u00EF\u00A3\u00B4\u00EF\u00A3\u00B4\u00EF\u00A3\u00B2\u00EF\u00A3\u00B4\u00EF\u00A3\u00B4\u00EF\u00A3\u00B3 T1(x1), 0\u00E2\u0089\u00A4 x1 \u00E2\u0089\u00A4 L1 f or0\u00E2\u0089\u00A4 x\u00E2\u0089\u00A4 L1 T2(x2), 0\u00E2\u0089\u00A4 x2 \u00E2\u0089\u00A4 L2 f orL1 \u00E2\u0089\u00A4 x\u00E2\u0089\u00A4 L (3.36) where L= L1 +L2 (3.37) 50 3.3. Solution of the Heat Equation 3.3.4 The Average Channel Temperature The average temperature of the channel can be found by integrating the temperature function over the length and by dividing the total length. Tavg = 1 2 \u00EF\u00A3\u00AE\u00EF\u00A3\u00B0 1 L1 L1\u00CB\u0086 0 T1(x1)dx1 + 1 L2 L2\u00CB\u0086 0 T2(x2)dx2 \u00EF\u00A3\u00B9\u00EF\u00A3\u00BB Tavg = 1 2 [ 1 L1 L1\u00CB\u0086 0 { C1e\u00CE\u00BE1x1 +C2e\u00E2\u0088\u0092\u00CE\u00BE1x1 +T re f \u00E2\u0088\u0092K1e\u00CE\u00B11x1 } dx1 + 1 L2 L2\u00CB\u0086 0 { D1e\u00CE\u00BE2x2 +D2e\u00E2\u0088\u0092\u00CE\u00BE2x2 +K2 } dx2 ] Tavg = 1 2L1 { C1 \u00CE\u00BE1 ( e\u00CE\u00BE1L1\u00E2\u0088\u00921 ) \u00E2\u0088\u0092C2 \u00CE\u00BE1 ( e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u00921 ) +L1T re f \u00E2\u0088\u0092 K1\u00CE\u00B11 ( e\u00CE\u00B11L1\u00E2\u0088\u00921)}+ 1 2L2 { D1 \u00CE\u00BE2 ( e\u00CE\u00BE2L2\u00E2\u0088\u00921 ) \u00E2\u0088\u0092 D2 \u00CE\u00BE2 ( e\u00E2\u0088\u0092\u00CE\u00BE2L2\u00E2\u0088\u00921 ) +K2L2 } 3.3.5 The Maximum Channel Temperature If Tmax1 is the maximum temperature given by the temperature function T1(x) and Tmax2 is the maximum temperature given by the temperature function T2(x), dT1 dx1 = 0\u00E2\u0087\u0092 x1 = xmax1 51 3.3. Solution of the Heat Equation Tmax1 = T1(x1) |x1=xmax1 dT2 dx2 = 0\u00E2\u0087\u0092 x2 = xmax2 Tmax2 = T2(x2) |x2=xmax2 where 0\u00E2\u0089\u00A4 xmax1 \u00E2\u0089\u00A4 L1 and 0\u00E2\u0089\u00A4 xmax2 \u00E2\u0089\u00A4 L2. The maximum channel temperature Tmax at a distance xmax along the length of the channel from the source end to the drain end is the greatest of Tmax1 and T max 2 . If Tmax1 \u00E2\u0089\u00A5 Tmax2 \u00E2\u0087\u0092 Tmax = Tmax1 and xmax = xmax1 . If Tmax2 > T max 1 \u00E2\u0087\u0092 Tmax = Tmax2 and xmax = L1 + xmax2 . For xmax2 : D1\u00CE\u00BE2e\u00CE\u00BE2x max 2 \u00E2\u0088\u0092D2\u00CE\u00BE2e\u00E2\u0088\u0092\u00CE\u00BE2xmax2 = 0 xmax2 = 1 2\u00CE\u00BE2 ln [ D2 D1 ] For xmax1 : 52 3.4. M-Network Representation of the Channel C1\u00CE\u00BE1e\u00CE\u00BE1x max 1 \u00E2\u0088\u0092C2\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1xmax1 \u00E2\u0088\u0092K1\u00CE\u00B11e\u00CE\u00B11xmax1 = 0 The above equation for xmax1 does not have a closed form solution and a numerical solution is needed. It is not feasible to calculate xmax1 under these conditions. How- ever, the maximum temperature along the length can be calculated via a simple technique. The channel is subdivided into small lengths and at each point T (x). After calculating all the values at each point, the maximum channel temperature can be found. 3.3.6 The Channel Temperature at Pinch-off The channel temperature at pinch-off point Tpinch is given as Tpinch = T1(x1) |x1=L1 3.4 M-Network Representation of the Channel The analytical solution gives promising results. HeatMOS\u00C2\u00A9 is able to do a thermal analysis of the transistor channel and estimates temperature along its length. The approach can be used for a complete micro-chip electro-thermal reliability verifica- tion. In a complete micro-chip EMR verification methodology, each of the resistors that forms a net is represented by an equivalent interconnect network (Labun and 53 3.4. M-Network Representation of the Channel Jagjitkumar, 2008) and (Alam et al., 2007). Based on this research work an equiva- lent M-network model is proposed to model channel temperature. The M-network model is shown in Figure 3.7 and can be used to integrate the channel temperature into a full micro-chip thermal analysis. The channel has two regions with conduc- tance \u00CE\u00B71 and \u00CE\u00B72 corresponding to region 1 and 2 of the channel respectively. The current sources S1 and S2 correspond to the source end and the drain end of the channel, respectively. The currents at the nodes of the M-network are given by FO (at the source end) and FL (at the drain end). The temperatures at the source and the drain end are TO and TL respectively. The temperature at the pinch-off point Tpinch (the junction of region 1 and region 2 in the channel) corresponds to a voltage source at pinch-off point. The source end and drain end are the ports of a M-network. The additional conductances \u00CE\u00B81 and \u00CE\u00B82, correspond to the source end and the drain end respectively. These conductances represent a conduction path to ground that is not usually important electrically (and so for electrical analysis, \u00CE\u00B81,\u00CE\u00B82 = 0) but will be important thermally (and so for thermal analysis, \u00CE\u00B81,\u00CE\u00B82 > 0). The heat currents, FO and FL, can be given by FO = Glongavg dT1 dx | x= 0 (3.38) FL = Glongavg dT2 dx | x= L2 (3.39) Kirchhoff\u00E2\u0080\u0099s current law can be applied to the both sides of the M-network of Figure 54 3.4. M-Network Representation of the Channel Figure 3.7: Resistor M-network used for EMR verification 3.7. FO = \u00CE\u00B71Tpinch\u00E2\u0088\u0092 (\u00CE\u00B71 +\u00CE\u00B81)TO\u00E2\u0088\u0092S1 (3.40) FL = (\u00CE\u00B72 +\u00CE\u00B82)TL\u00E2\u0088\u0092\u00CE\u00B72Tpinch+S2 (3.41) where Tpinch =C1e\u00CE\u00BE1L1 +C2e\u00E2\u0088\u0092\u00CE\u00BE1L1 +T re f \u00E2\u0088\u0092K1e\u00CE\u00B11L1 Equations 3.38, 3.39 can be solved and compared with equations 3.40, 3.41 to find 55 3.4. M-Network Representation of the Channel out the values of \u00CE\u00B71,\u00CE\u00B72, \u00CE\u00B81, \u00CE\u00B82, S1, and S2. \u00CE\u00B71 = Glongavg 2\u00CE\u00BE1 \u00CE\u00B81 = Glongavg 4\u00CE\u00BE1\u00CE\u00BE2 M e\u00E2\u0088\u0092\u00CE\u00BE1L1 +Glongavg \u00CE\u00BE1\u00E2\u0088\u0092\u00CE\u00B71 { 2\u00CE\u00BE2e\u00E2\u0088\u0092\u00CE\u00BE1L1 M +1 } + e\u00E2\u0088\u0092\u00CE\u00BE1L1 S1 = Glongavg { 2\u00CE\u00BE1 M \u00CE\u00B31\u00E2\u0088\u0092K1 (\u00CE\u00B11 +\u00CE\u00BE1) } \u00E2\u0088\u0092T re f \u00CE\u00BE1\u00E2\u0088\u0092 \u00CE\u00B71M { e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 } \u00CE\u00B31\u00E2\u0088\u0092 T re f e\u00E2\u0088\u0092\u00CE\u00BE1L1 +T re f +K1 { e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00CE\u00B11L1 } \u00CE\u00B72 =\u00E2\u0088\u0092 e \u00CE\u00BE1L1( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1) { Me\u00E2\u0088\u0092\u00CE\u00BE1L1 2\u00CE\u00BE2 +Glongavg e \u00CE\u00BE2L2e\u00E2\u0088\u0092\u00CE\u00BE1L1 [( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +\u00CE\u00BE1 ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )]} \u00CE\u00B82 = \u00CE\u00B72 \u00EF\u00A3\u00B1\u00EF\u00A3\u00B2\u00EF\u00A3\u00B32\u00CE\u00BE2e \u00CE\u00BE2L2 ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) M \u00E2\u0088\u00921 \u00EF\u00A3\u00BC\u00EF\u00A3\u00BD\u00EF\u00A3\u00BE\u00E2\u0088\u0092Glongavg { 2\u00CE\u00BE2 M e2\u00CE\u00BE2L2 [( e\u00CE\u00BE1L1+ e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +\u00CE\u00BE2 ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )]\u00E2\u0088\u0092\u00CE\u00BE2} 56 3.4. M-Network Representation of the Channel S2 = Glongavg { \u00CE\u00BE2K2 + 2\u00CE\u00BE2e\u00CE\u00BE2L2 M \u00CE\u00B32 } +T re f { e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u00921 } \u00E2\u0088\u0092K1 { e\u00E2\u0088\u0092\u00CE\u00BE1L1 \u00E2\u0088\u0092e\u00CE\u00B11L1}\u00E2\u0088\u0092 \u00CE\u00B72 M { e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 } \u00CE\u00B31 where M = ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )( 1+ e2\u00CE\u00BE2L2 ) \u00CE\u00BE2\u00E2\u0088\u0092 ( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 )( 1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2 ) \u00CE\u00B31 = \u00CE\u00BE2 ( 1+ e2\u00CE\u00BE2L2 ){ \u00E2\u0088\u0092T re f ( e\u00E2\u0088\u0092\u00CE\u00BE1L1 +1 ) +K1 ( e\u00CE\u00B11L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +K2 (1\u00E2\u0088\u0092 e\u00CE\u00BE2L2 )} + ( 1\u00E2\u0088\u0092 e2\u00CE\u00BE2L2 ){ \u00E2\u0088\u0092T re f \u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1 +K1 ( \u00CE\u00B11e\u00CE\u00B11L1 +\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +K2\u00CE\u00BE2e\u00CE\u00BE2L2 } \u00CE\u00B32 = T re f {( e\u00E2\u0088\u0092\u00CE\u00BE1L1\u00E2\u0088\u00921 )( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1 ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )} + K1 {( e\u00CE\u00B11L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) \u00E2\u0088\u0092 ( \u00CE\u00B11e\u00CE\u00B11L1 +\u00CE\u00BE1e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) +( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )} +K2 {( 1\u00E2\u0088\u0092 e\u00CE\u00BE2L2 )( e\u00CE\u00BE1L1 + e\u00E2\u0088\u0092\u00CE\u00BE1L1 ) \u00E2\u0088\u0092 \u00CE\u00BE2e\u00CE\u00BE2L2 ( e\u00CE\u00BE1L1\u00E2\u0088\u0092 e\u00E2\u0088\u0092\u00CE\u00BE1L1 )} 57 3.5. Summary 3.5 Summary An analytic solution for estimating channel temperature in a MOSFET device has been derived. The solution is composed of two sets of equations corresponding to the tapered channel region and the pinch-off region. Equations were also derived for estimating the average channel temperature and the temperature at the pinch- off point. The analytic equations are used to calculate equivalent model values in a M-network that can be incorporated in top level micro-chip simulations to predict the spatial variation of temperature across the micro-chip and assist designers in identifying potential EMR failure points. The foundation for the development of the software to estimate temperature of MOSFET channel has been set. 58 Chapter 4 HeatMOS\u00C2\u00A9 4.1 About HeatMOS\u00C2\u00A9 HeatMOS\u00C2\u00A9 is a software tool which solves the heat equation for MOSFET devices analytically and estimates the channel temperature profile. The HeatMOS\u00C2\u00A9 soft- ware tool is developed as part of this research project. The program uses physical device parameters extracted from BSIM3 models (Liu et al., 1999). HeatMOS\u00C2\u00A9 reads the device geometry information from an extracted layout in the SPICE for- mat. It calculates saturation electric current and other required variables for the analytical model. HeatMOS\u00C2\u00A9 then calculates the channel temperature using the analytical model and the outputs temperature distribution along the channel length. The input variables required for HeatMOS are listed below: 1. SourceTemp: Source end temperature. 2. DrainTemp: Drain end Temperature. 3. RefTemp: Reference temperature of the medium of the channel. 59 4.1. About HeatMOS\u00C2\u00A9 4. Vgs: Gate voltage. 5. Vds: Drain voltage. 6. Vbs: Substrate bias voltage. 7. Ids: Drain current; this is optional parameter, if not given will be calculated (device is in saturation region). 8. Netlist: Extracted spice netlist from layout. HeatMOS\u00C2\u00A9 outputs: 1. The channel temperature distribution along the length of the channel and a text file containing the temperatures at different points along the channel length. 2. The drain current. 3. The average thermal conductance. 4. The average channel temperature. 5. The maximum channel temperature. 6. The temperature at pinch-off point. 7. The effective length and width of the channel. A sample output of of HeatMOS has been provided in Appendix D. 60 4.2. Integration in the VLSI CAD Flow 4.2 Integration in the VLSI CAD Flow Figure 4.1 shows a typical post-logic synthesis VLSI (Very-large-scale Integration) CAD (Computer-aided Design) flow. Reliability verification is necessary for nano- scale technologies to ensure the reliability of integrated circuit. As can be seen from the figure, thermal verification is needed after physical verification. Heat- MOS\u00C2\u00A9 can be integrated with interconnect temperature estimation CAD tools (e.g. Therminator (Labun and Jagjitkumar, 2008)) to do a complete micro-chip thermal verification. If the integrated circuit design does not pass reliability verification, there is still a chance to modify the layout of the circuit in its design phase. 61 4.2. Integration in the VLSI CAD Flow Figure 4.1: VLSI CAD flow 62 Chapter 5 Results and Discussions HeatMOS solves the heat equation analytically. For a comparative analysis of the results of HeatMOS, the heat equation for both regions of the channel has been solved using the finite difference approach (numerical approach). The finite dif- ference approach is based on Taylor\u00E2\u0080\u0099s approximation (Strikwerda, 2004) for 2nd order derivatives given by equation 5.1 f (x\u00E2\u0088\u0092h)\u00E2\u0088\u00922 f (x)+ f (x+h) h2 = f \u00E2\u0080\u00B2\u00E2\u0080\u00B2(x)+ f (4)(v) 12 h2 (5.1) where, h is a small interval and f \u00E2\u0080\u00B2\u00E2\u0080\u00B2(x) is the second derivative of function f (x). The second term on the right hand side of equation 5.1 is the remainder term and can be considered as an approximation error. The detailed finite-difference method and its pseudo code is given in appendix C. 63 Chapter 5. Results and Discussions Figure 5.1: An nMOS layout 64 5.1. Verification If we apply voltages to the terminals of this nMOS transistor, an electric current will flow in the channel. In a complex micro-chip, the source and the drain of this transistor will be connected to other nets via vias and contacts. The temperatures of these vias can be estimated by interconnect estimations tools such as Thermi- nator3D (Labun and Jagjitkumar, 2008). If we neglect the distance between the source and the drain ends of the channel and the contacts, the source and the drain ends will be approximately at the same temperature as the contacts connecting these vias. The thermal channel model is verified by running a number of different test sce- narios with a typical nMOS device. The layout of the device is shown in Figure 5.1 and has a drawn length of 0.18 \u00C2\u00B5m and a drawn width of 0.40 \u00C2\u00B5m. The full SPICE BSIM3 model for the device is given in Appendix A. HeatMOS includes a utility to approximate the IV characteristics of the device using a simplified de- vice model and the results are shown in Figure 5.2. The IV characteristic at the pinch-off point is not smooth because the curve-fitting spline functions have not be modeled in HeatMOS. The program also support a direct input of the operating device current which can be extracted from full BSIM3 SPICE simulations. This feature is included in the test scenarios described below. 5.1 Verification Let us consider a few test scenarios to verify that HeatMOS\u00C2\u00A9 compiles and pro- vides the results that are consistent with numerical simulations. Te objective of 65 5.1. Verification 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0 0.5 1 1.5 2 2.5 Id s (m A) Vds (V) Vgs = 1.5 V Vgs = 1.0 V Figure 5.2: I-V characteristics of nMOS device under test produced by Heat- MOS\u00C2\u00A9 66 5.1. Verification these tests is to verify that HeatMOS provides accurate results under different test scenarios and free from run-time errors. 5.1.1 Test Scenario using Externally Supplied Drain Current The objective of this test is to verify that HeatMOS\u00C2\u00A9 can analyze channel tem- perature using an externally supplied current. Sometimes designer\u00E2\u0080\u0099s may want to analyze the temperature behavior for higher current than calculated from circuit simulations. In this test, a current of 2 mA has been supplied externally by de- signer, the layout drawn length of device under test is 0.28 \u00C2\u00B5m and the width is 0.40 \u00C2\u00B5m, the applied gate voltage is 1.0 V and the drain voltage is 1.5 V, the source end temperature is set to 565 K, the drain end is set at 567 K, and the refer- ence temperature has been set to 456 K. The results of this test scenario are shown in Figure 5.3. HeatMOS\u00C2\u00A9 calculates an average channel temperature of 586.19 K and the maximum channel temperature is 588.68 K at a distance of 0.2115 \u00C2\u00B5m from from the source end. The temperature at the pinch-off point is 588.46 K at a distance 0.2093 \u00C2\u00B5m from the source end. As we can see from the Figure 5.3 the difference between the analytic solution and numerical solution is significant and the peak temperature difference is approxi- mately 4 K. The difference is due to the round-off error in the finite difference approximation and the chosen step size. In this test scenario the gate voltage and drain voltage are different and electric current has been provided externally and not been calculated by HeatMOS\u00C2\u00A9. The accuracy of the numerical solution can be improved by reducing the step size as is shown by the blue line in Figure 5.6. 67 5.1. Verification 565 570 575 580 585 590 0 0.05 0.1 0.15 0.2 0.25 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Finite Difference with smaller step size Figure 5.3: SourceTemp= 565K |DrainTemp= 567K | Re f Temp= 456K |Vgs = 1.0V |Vds = 1.5V | Ids = 2mA | v7\u00E2\u0088\u0092nmos.spice 68 5.1. Verification 5.1.2 Test Scenario for a Smaller MOSFET Device In this test, the layout drawn length and width of the device are 0.18 \u00C2\u00B5m and 0.40 \u00C2\u00B5m, and gate and drain voltages are both 1.5 V. The source end temperature is 565 K, the drain end temperature is 567 K, and the reference temperature has been set at 456 K. The results of this test scenario are shown in Figure 5.4. Results from HeatMOS\u00C2\u00A9 show the average channel temperature is 575.87 K and the maximum channel temperature is 570.52 K at a distance of 0.114 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 570.36 K at a distance 0.109 \u00C2\u00B5m along the length of the channel from the source end. HeatMOS\u00C2\u00A9 calculates a saturation current of 0.269 mA. As we can see from the results of this test the channel becomes hotter towards drain end. It would be interesting to verify HeatMOS\u00C2\u00A9 keeping the source, the drain and the reference temperature all at the same time. 5.1.3 Test Scenario for the Case When the Source, Drain, and the Reference Temperatures are Same In this test, the layout drawn length of the device under test is 0.18 \u00C2\u00B5m and the width is 0.40 \u00C2\u00B5m, the applied gate voltage is 1.5 V, the drain voltage is 1.5 V, and the source, drain, and reference temperatures all are set to 10 K. The results of this test scenario are shown in Figure 5.5. The drain current calculated by HeatMOS\u00C2\u00A9 is 0.2699 mA, the average channel temperature is 5.05 K, and the maximum chan- nel temperature is 8.1 K at a distance of 0.1124 \u00C2\u00B5m along the length of the channel 69 5.1. Verification 565 566 567 568 569 570 571 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.4: SourceTemp= 565K |DrainTemp= 567K | Re f Temp= 456K |Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice 70 5.1. Verification 10 20 30 40 50 60 70 80 90 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.5: SourceTemp = 10K | DrainTemp = 10K | Re f Temp = 10K | Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice from the source end. The temperature at the pinch-off point is 7.95 K at a distance of 0.109 \u00C2\u00B5m from the source end. The above verification tests shows that HeatMOS\u00C2\u00A9 is free from runtime errors, bugs, and provides meaningful results. There is a possibility to write other corner test scenarios to find out hidden bugs in the software. Some of the following test scenarios are focused to verify the HeatMOS\u00C2\u00A9 results when the device under test operates in different modes of operation. 71 5.2. Saturation Mode of Operation 5.2 Saturation Mode of Operation In the saturation mode, the MOSFET is on and it conducts current between the source and the drain. The charge carriers flow in the channel and the current be- comes constant (IDS = Isat) and can be approximated by (Weste and Harris, 2004) Isat = 1 2 We f f Le f f \u00C2\u00B5nCox (VGS\u00E2\u0088\u0092VTh)2 {1+\u00CE\u00BB (VDS\u00E2\u0088\u0092VDS)} (5.2) VDS= (VGS\u00E2\u0088\u0092VTh)\u00E2\u0088\u0092 12\u00CE\u00BB (VGS\u00E2\u0088\u0092VTh) 2 where We f f is the effective channel width, Le f f is the effective channel length, \u00C2\u00B5n is the electron mobility,Cox is the gate oxide layer capacitance per unit area, VGS is the applied gate voltage,VTh is the threshold voltage of the nMOS transistor, and \u00CE\u00BB is the fitting parameters for channel length modulation. In this mode of operation VDS > VDS. The nMOS threshold voltage for the BSIM3 technology file used in this work is 0.39V . The value of \u00CE\u00BB is 0.20. Let us consider a few test cases in this mode of operation. 5.2.1 Test Scenario When Gate and Drain are at Same Voltage In this test, the source and drain temperatures have been set to 325 K respectively, and the reference temperature has been set to 300 K. The applied gate voltage is 1.5 V and the drain voltage is 1.5 V. The layout drawn width of the nMOS is 72 5.2. Saturation Mode of Operation 325 326 327 328 329 330 331 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.6: SourceTemp= 325K |DrainTemp= 325K | Re f Temp= 300K |Vgs = 1.5V |Vds = 1.5V | v8\u00E2\u0088\u0092nmos.spice 0.40 \u00C2\u00B5m and the the length is 0.18 \u00C2\u00B5m. HeatMOS\u00C2\u00A9 calculates a saturation drain current of 0.2699 mA. The results of this test scenario are shown in Figure 5.6. The average channel temperature for this case is 332.50 K and the maximum channel temperature is 330.57 K at a distance of 0.113 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 330.46 K at a distance 0.109 \u00C2\u00B5m along the length of the channel from the source end. As we can see from this test that the temperature varies between 325 K to 330.57 K. It would also be interesting to note that in this case the source and the drain has 73 5.2. Saturation Mode of Operation been at the same voltage level. It would be interesting to check the temperature along the length of channel if the gate and the drain voltages are different. 5.2.2 Test Scenario for the Case of Different Gate and Drain Voltages In this test, the source and the drain end temperatures are set to 325 K respectively, and the reference temperature is set to 300 K. The applied gate voltage is 1.8 V and the drain voltage is 2.2 V. The layout drawn width of nMOS is 0.40 \u00C2\u00B5m and the length is 0.18 \u00C2\u00B5m. HeatMOS\u00C2\u00A9 calculates a saturation drain current of 0.4322 mA. The results of this test scenario are shown in Figure 5.7. The average channel temperature for this case is 336.78 K and the maximum channel temperature is 338.12 K at a distance 0.1137 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 337.81 K at a distance 0.108 \u00C2\u00B5m from the source end. The temperature difference between the minimum and the maximum temperature in this case is 13.12 K. We can see that upon increasing the gate and drain volt- age the maximum channel temperature rises. The channel temperature variation is complex and is not based on one factor (e.g. voltage change), therefore the corre- lation between voltage and temperature is not explicit. However in simple terms it indicate that upon increasing the gate and drain voltage, the drain current increases and it leads to higher temperature in the channel as the more heat is generated in the channel due to electric current flow. Let\u00E2\u0080\u0099s consider another test scenario with very high voltage on the gate and drain ends. 74 5.2. Saturation Mode of Operation 324 326 328 330 332 334 336 338 340 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.7: SourceTemp= 325K |DrainTemp= 325K | Re f Temp= 300K |Vgs = 1.8V |Vds = 2.2V | v8\u00E2\u0088\u0092nmos.spice 75 5.2. Saturation Mode of Operation 5.2.3 Test Scenario for the Case of High Gate and Drain Voltages In this test, source and drain temperatures are set to 325 K respectively, the refer- ence temperature has been set to 300 K. The applied gate and drain voltages are both 5.0 V. The layout drawn width of the nMOS is 0.18 \u00C2\u00B5m and the the length is 0.40 \u00C2\u00B5m. HeatMOS\u00C2\u00A9 calculates a saturation drain current of 4.52 mA. This is a high value for the drain current. The results of this test scenario are shown in Figure 5.6. The average channel temperature for this case is 460.66 K and the maximum channel temperature is 556.00 K at a distance 0.1102 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 550.26 K at a distance 0.10631 \u00C2\u00B5m from the source end. As we can see form this test that the maximum temperature difference is 231 K which is a very high value. This indicates that the higher drain current leads to higher temperature which is expected. It would be interesting to see the tempera- ture distribution for the same test on a larger device. 5.2.4 Test Scenario for a Large MOSFET Device with High Drain Current In this test, the source and the drain end temperatures have been set to 325 K respectively, and the reference temperature has been set to 300 K. The applied gate and drain voltages are both 5.0 V. The layout drawn width of the nMOS is 0.40 \u00C2\u00B5m and the length is 0.28 \u00C2\u00B5m. HeatMOS\u00C2\u00A9 calculates the saturation drain current of 2.689 mA. The results of this test scenario are shown in Figure 5.9. The average 76 5.2. Saturation Mode of Operation 300 350 400 450 500 550 600 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.8: SourceTemp= 325K |DrainTemp= 325K | Re f Temp= 300K |Vgs = 5.0V |Vds = 5.0V | v8\u00E2\u0088\u0092nmos.spice 77 5.2. Saturation Mode of Operation 320 340 360 380 400 420 440 460 480 500 520 0 0.05 0.1 0.15 0.2 0.25 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.9: SourceTemp= 325K |DrainTemp= 325K | Re f Temp= 300K |Vgs = 5V |Vds = 5V | v7\u00E2\u0088\u0092nmos.spice channel temperature for this case is 429.03 K and the maximum temperature is 500.50 K at a distance 0.208 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 499.17 K at a distance 0.206 \u00C2\u00B5m from the source end. The results show that the maximum channel temperature peaks at 175.5 K above the source and drain terminal temperatures which is about 50 K cooler than the previous case. Clearly the larger device has more area to dissipate heat in the channel. Let\u00E2\u0080\u0099s consider another arbitrary test scenario. 78 5.3. Linear/Ohmic Mode of Operation 5.2.5 Test Scenario for Different Source/Drain Temperatures and Voltages In this test, the source end temperature is set to 318 K, the drain end temperature is set to 324 K, and the reference temperature is set to 298 K. The applied gate voltage is 1.2 V and the drain voltage is 1.8 V. The layout drawn width of the nMOS is 0.40 \u00C2\u00B5m and the length is 0.18 \u00C2\u00B5m. HeatMOS\u00C2\u00A9 calculates a saturation drain current of 0.145 mA. The results of this test scenario are in Figure 5.9. The average channel temperature for this case is 327.91 K and the maximum temperature is 326.97 K at a distance of 0.118 \u00C2\u00B5m from the source end. The temperature at the pinch-off point is 326.58 K at a distance of 0.109 \u00C2\u00B5m from the source end. 5.3 Linear/Ohmic Mode of Operation When the MOSFET operates in the linear mode, the drain current increases linearly with the drain voltage. Under these conditions the MOSFET behaves as a voltage dependent resistor where the resistance is determined by the gate voltage. The current and the voltage follow the ohmic relationship in this mode of operation. The drain current in this mode of operation is given by (Sedra and Smith, 2007) IDS = \u00C2\u00B5nCox We f f Le f f [ 2.0(VGS\u00E2\u0088\u0092VTh)VDS\u00E2\u0088\u0092V 2DS ] (5.3) where, We f f is the effective channel width, Le f f is the effective channel length, \u00C2\u00B5n is the electron mobility, Cox is the gate oxide layer capacitance per unit area, VGS 79 5.3. Linear/Ohmic Mode of Operation 318 319 320 321 322 323 324 325 326 327 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ch an ne l t em pe ra tu re T (x) (K ) Effective channel length x (um) HeatMOS Finite Difference Figure 5.10: SourceTemp = 318K | DrainTemp = 324K | Re f Temp = 298K | Vgs = 1.2V |Vds = 1.8V | v8\u00E2\u0088\u0092nmos.spice 80 5.3. Linear/Ohmic Mode of Operation is the applied gate voltage, VDS is applied drain voltage, and VTh is the threshold voltage of the nMOS. In this mode of operation VDS J1 121 Appendix C. Numerical Solution of the Heat Equation b(i) = -(dx2^2)*beta2; else b(i) = 0.0; endif endfor A=sparse(J1+J2, J1+J2); A(1,1) = 1.0; A(J1+J2, J1+J2) = 1.0; for i=2:(J1+J2-1) if i < J1 A(i, [i-1, i, i+1]) = [1.0, -(2.0-(xie1^2)*(dx1^2)), 1.0]; elseif i > J1 A(i, [i-1, i, i+1]) = [1.0, -(2-(dx2^2)*(xie2^2)), 1.0]; else A(i, [i-1, i, i+1]) = [1.0/dx1, -(1.0/dx1 + 1.0/dx2), 1.0/dx2]; endif endfor 122 Appendix C. Numerical Solution of the Heat Equation Y=A\b; for i=1:J1 x(i)=x1(i); endfor for j=1:J2 x(J1+j)=len1.+x2(j); endfor myfile = \"chanTempFD.out\" ; FILE = fopen(myfile, \"w\"); for i=1:(J1+J2) fprintf (FILE, \"%e\t\", x(i)); fprintf (FILE, \"%e\n\", Y(i)); endfor fclose(FILE); %\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094\u00E2\u0080\u0094 123 Appendix D HeatMOS\u00C2\u00A9 Development D.1 Development Environment HeatMOS\u00C2\u00A9 has been developed in \u00E2\u0080\u009CC\u00E2\u0080\u009D programming language. It has been com- piled using gnu c-compiler, g++4.0.1 for i686-apple-darwin9 machine. HeatMOS\u00C2\u00A9 has been executed on Apple MAC OS X version 10.5.8 machine with Intel Core 2 Duo 1 GHz processor and 1 GB of DDR2 SDRAM. To plot graphs Gnuplot version 4.4 software has been used. Octave version 3.2.3 software has been used for the numerical solution of the channel heat equation. D.2 A Sample HeatMOS\u00C2\u00A9 Execution A screen snapshot of sample HeatMOS\u00C2\u00A9 execution is split in two parts and shown in Figure D.1 and D.2. The top line in Figure D.1 shows the command for Heat- MOS\u00C2\u00A9 execution. The temperature distribution along the length of the channel in being output in a text file, which can be used to plot graphs using a graph plotting software (e.g. gnuplot). 124 D.2. A Sample HeatMOS\u00C2\u00A9 Execution Figure D.1: A sample HeatMOS run part 1 125 D.2. A Sample HeatMOS\u00C2\u00A9 Execution Figure D.2: A sample HeatMOS run part 2 126"@en . "Thesis/Dissertation"@en . "2011-05"@en . "10.14288/1.0073362"@en . "eng"@en . "Electrical Engineering"@en . "Vancouver : University of British Columbia Library"@en . "University of British Columbia"@en . "Attribution-NonCommercial-NoDerivatives 4.0 International"@en . "http://creativecommons.org/licenses/by-nc-nd/4.0/"@en . "Graduate"@en . "A steady-state analytical solution for MOSFET channel temperature estimation"@en . "Text"@en . "http://hdl.handle.net/2429/43566"@en .