Design and Practical Implementation of Advanced Reconfigurable Digital Controllers for Low-Power Multi-Phase DC-DC Converters

Transcription

1 Design and Practical Implementation of Advanced Reconfigurable Digital Controllers for Low-Power Multi-Phase DC-DC Converters by Zdravko Lukic A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto Copyright by Zdravko Lukic 2011

2 Design and Practical Implementation of Advanced Reconfigurable Digital Abstract Controllers for Low-Power Multi-Phase DC-DC Converters Zdravko Lukic Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto 2011 The main goal of this thesis is to develop practical digital controller architectures for multi-phase dc-dc converters utilized in low power (up to few hundred watts) and cost-sensitive applications. The proposed controllers are suitable for on-chip integration while being capable of providing advanced features, such as dynamic efficiency optimization, inductor current estimation, converter component identification, as well as combined dynamic current sharing and fast transient response. The first part of this thesis addresses challenges related to the practical implementation of digital controllers for low-power multi-phase dc-dc converters. As a possible solution, a multiuse high-frequency digital PWM controller IC that can regulate up to four switching converters (either interleaved or standalone) is presented. Due to its configurability, low current consumption (90.25 μa/mhz per phase), fault-tolerant work, and ability to operate at high switching frequencies (programmable, up to 10 MHz), the IC is suitable to control various dc-dc converters. The applications range from dc-dc converters used in miniature battery-powered electronic devices consuming a fraction of watt to multi-phase dedicated supplies for communication systems, consuming hundreds of watts. ii

3 A controller for multi-phase converters with unequal current sharing is introduced and an efficiency optimization method based on logarithmic current sharing is proposed in the second part. By forcing converters to operate at their peak efficiencies and dynamically adjusting the number of active converter phases based on the output load current, a significant improvement in efficiency over the full range of operation is obtained (up to 25%). The stability and inductor current transition problems related to this mode of operation are also resolved. At last, two reconfigurable digital controller architectures with multi-parameter estimation are introduced. Both controllers eliminate the need for external analog current/temperature sensing circuits by accurately estimating phase inductor currents and identifying critical phase parameters such as equivalent resistances, inductances and output capacitance. A sensorless nonlinear, average current-mode controller is introduced to provide fast transient response (under 5 μs), small voltage deviation and dynamic current sharing with multi-phase converters. To equalize the thermal stress of phase components, a conduction loss-based current sharing scheme is proposed and implemented. iii

4 Acknowledgments I wish to thank my supervisor, Professor Aleksandar Prodić, for his great support during my doctoral studies. Professor Prodić s dedication, scientific vision, and research passion have served as inspiration and motivation for me to achieve more and perform better than I could have ever imagined. I am also thankful for his valuable advices that often extended beyond my thesis work. Professor Prodić was always able to give me honest constructive feedback that helped me to improve. His wisdom will stay with me, following me in my future career path. During my doctoral studies, I had the great honor and pleasure to work with my dear colleagues and friends: Zhenyu Zhao ( Frank ), Amir Parayandeh, S M Ahsanuzzaman, Andrija Stupar, Aleksandar Radić, Jason Weinstein, Massimo Tarulli, and Jurgen Alico. We always acted as a terrific team and helped each other to succeed. I will always remember those missionimpossible chip tape-outs with Amir, Aleksandar, and Jason, and conference paper deadlines with Ahsan and Frank. I would like to also thank our industry partners Lee Clevelend, Dimitry Goder, and Rob de Nie for sharing with us their faith and excitement for digitally-controlled SMPS. It was a remarkable experience to work and learn from these leading industry professionals. I am grateful for their invaluable support to our group during numerous research projects. I would like to mention family Milenov from Toronto and family Huerta-Calva from Celaya, Mexico. They shared the warmth of their home with me. I thank them for their generous help and for all those unforgettable moments that always kept me positive even through hard times. I am deeply thankful to my mother Slavica, father Rajko, and sister Milijana for their support and understanding of my passion for scientific work. The hard work of my parents and persistence always inspired me to be a better student. Finally, I would like to dedicate this thesis to my wife Santa Concepción Huerta Olivares. I thank her for unlimited understanding, profound love, and all the sacrifices she made during our doctoral studies. I am thrilled about our joined future. Forever and ever! iv

5 Table of Contents Acknowledgments... iv Table of Contents... v List of Tables... ix List of Figures... x List of Appendices... xiv Chapter 1 Introduction Design Requirements for Modern Power Management Systems in Low-Power Applications Multi-Phase Digital Controller Design Challenges Thesis Objectives and Organization... 4 References... 5 Chapter 2 Multi-Use and Fault-Tolerant Digital Controller IC Introduction System Operation and Controller Architecture Multi-Phase Digital Pulse-Width Modulator Operation with 3 Phases and Number Conversion Logic Dual-Biased Delay Cell for Wide-Frequency Range Segmented Bias Circuit Digital Delay-Lock Logic Programmable Dead-Time Circuit ADC Implementation IC Implementation Experimental Results Open Loop Tests v

6 2.7.2 Closed Loop Tests Fault-Tolerant Operation Conclusion References Chapter 3 Dynamic Efficiency Optimization for Multi-Phase Converters Introduction System Architecture Practical Implementation Accurate Current Regulation Jitter Elimination Transient Operation Design of Converter Phases Experimental System and Results Steady-State Operation Transient Operation Measured Converter Efficiency Conclusion References Chapter 4 Reconfigurable Digital Controllers with Multi-Parameter Estimation Introduction Principle of Current-Estimator Operation Practical Implementation of the Current Estimator Filter Architecture and Self-Calibration Filter Gain Calibration Procedure Filter Time Constant Calibration Procedure Offset Calibration Procedure vi

7 4.3.2 Steady-State Current Estimator Architecture Principle of Temperature Monitoring Operation Inductor and Output Capacitor Identification Inductor Identification Output Capacitor Identification Oversampled Non-Linear Average Current-Mode Controller for Multi-Phase Converters Steady-State Mode Multi-Phase Current Estimator Tuning and Phase Identification Conduction Loss-Based Current Sharing Scheme Transient Mode Hardware Optimization and Controller On-Chip Complexity Experimental Systems and Results Filter Gain Calibration Filter Time Constant Calibration Offset Calibration Calibrated Operation Current Estimation with the Steady-State Current Estimator Temperature Monitoring and Protection Overload Protection Multi-Phase Operation with the Non-Linear Average Current-Mode Controller Accuracy of Current and Temperature Estimation Conclusion References vii

8 Chapter 5 Conclusions and Future Work Thesis Contributions Future Work Appendix A Filter Time Constant Calculation Appendix B Quantization Error Effects on Current Estimator Accuracy viii

9 List of Tables Table 2.1: Summary of the IC parameters Table 3.1: Important parameters of the 4-phase logarithmic buck converter 40 Table 4.1: The silicon area and gate count of the multi-phase controller when implemented in TSMC 0.18-µm CMOS process.70 Table B.1: Current estimator relative error versus quantization step of the output voltage ADC.. 92 ix

10 List of Figures Figure 1.1: Block diagram of a power management system implemented in notebooks (source: Texas Instruments) Figure 2.1: A block diagram of the multi-use and fault-tolerant digital controller IC regulating operation of a 4 phase interleaved buck converter Figure 2.2: A block diagram of the proposed multi-phase digital pulse-width modulator Figure 2.3: Phase synchronization: a) start signals during 4-phase interleaved operation; b) 3- phase interleaved operation Figure 2.4: The 3-phase MDPWM operation: a) the absolute duty cycle error vs. 8-bit duty ratio command d c [n]; b) actual duty cycle versus d c [n] Figure 2.5: A delay cell with adjustable propagation time t pd : a) conventional current-starved delay cell b) proposed dual-biased current-starved delay cell Figure 2.6: Proposed adjustment of the current-starved delay cell current i m (t) in the delay line versus the desired switching frequency f sw Figure 2.7: Digitally controlled segmented-bias circuit Figure 2.8: a) A block diagram of digital delay-locking logic; b) MDPWM delay line signals when 1, 2, or 4 phases are active Figure 2.9: Programmable dead-time circuit Figure 2.10: a) Fabricated chip die of the multi-use digital controller IC b) Current consumption breakdown Figure 2.11: MDPWM linearity test at 100 khz Ch1: output converter voltage (1 V/div) Figure 2.12: MDPWM operation at 10 MHz: a) the duty ratio command is changed between 4 consecutive values b) linearly distributed falling edges of the pulse-width modulated signals (8- bit resolution at 10 MHz) x

11 Figure 2.13: Closed-loop operation of the controller IC: a) four 1-MHz buck converters are regulated to produce four supply rails between 1.2 and 3.3 V; b) transient response of the controller when the voltage reference is dynamically changed; c) two-phase interleaved steadystate operation; d) four-phase interleaved steady-state operation; Figure 2.14: Fault-tolerant operation in the closed loop Figure 3.1: Block diagram of a digitally controlled multi-phase converter with logarithmic current sharing Figure 3.2: Detailed system architecture of the proposed controller with a logarithmic current sharing Figure 3.3: The output voltage regulation is affected by the jitter (shown in the red circle) of i tot [n] Figure 3.4: Operation of the hysteretic logic output phase enable signal versus input control value i tot [n] Figure 3.5: The effect of finite rising Δ 1 and falling current slope Δ 2 on i tot (t) for a step of the digital current command i tot [n] from 5.4I nom A to 3.6I nom Figure 3.6: Simulated transient operation of the controller during a heavy-to-light load transient between 5.4I nom and 3.6I nom Figure 3.7: a) Simulated efficiency characteristics of the converter phases operating at different switching frequencies b) Simulated efficiency of the whole system Figure 3.8: Steady-state operation with two different load currents: a) 39 A b) 12 A Figure 3.9: Transient response for a load step from 21 A to 19 A when the phase enable corrector and transient current estimator are: a) disabled b) enabled Figure 3.10: Transient response: a) from 9 A to 38 A; b) from 38 A to 9 A Figure 3.11: Measured efficiency characteristics of the converter based on: a) logarithmic current sharing and uniform current sharing b) logarithmic current sharing and phase shadowing xi

12 Figure 4.1: a) A digitally-controlled buck converter with the self-tuning current estimator and remote temperature monitoring system b) A dual-phase buck converter regulated by the nonlinear multi-phase digital average current-mode controller based on the multi-phase self-tuning current estimator Figure 4.2: Current sensing techniques: a) conventional analog implementation b) implementation with a self-tuning digital filter Figure 4.3: Test current sink circuit used for the filter calibration Figure 4.4: Signals of a filter implemented with an over-sampling ADC Figure 4.5: The architecture of the tunable IIR filter used in the current estimator Figure 4.6: Filter gain calibration procedure: simulated response of the current estimator during output load change for two cases (bottom curve) the initial value of R L is twice the actual value (top curve) after the filter adjustment Figure 4.7: The buck converter (a) and its steady-state dc equivalent circuit R eq (b) used for current estimation Figure 4.8: Calibration of the time constant: simulated response of the current estimator during output load change between 2 A and 2.5A; (bottom) for τ f =0.5τ L ; (top) for τ f =τ L Figure 4.9: Offset in the estimated current i sense [n] caused by the operation of the dead-time circuit and converter Figure 4.10: The implementation of the steady-state current estimator Figure 4.11: Change of R L_calibrated due to the ambient temperature for different output load current conditions Figure 4.12: Differential output capacitor and phase inductance identification Figure 4.13: Mismatch in phase equivalent resistances generates unequal thermal stress on power switches xii

13 Figure 4.14: Transient mode operation of the non-linear average current-mode controller Figure 4.15: A time-multiplexed N-phase digital current estimator Figure 4.16: System operation Figure 4.17: The filter gain (1/R L_calibrated ) calibration procedure Figure 4.18: The filter time calibration procedure: a) τ f =0.5τ L ; b) τ f τ L Figure 4.19: The filter offset calibration procedure Figure 4.20: The estimated current during load steps between 1.5 A and 4.5 A Figure 4.21: The estimated current with a steady-state current estimator during the load step between 0.8 A and 5.1 A Figure 4.22: Thermal monitoring: a) at 45 C; b) at 105 C Figure 4.23: Overload protection implemented with the current estimator Figure 4.24: Current estimator tuning and component identification with the multi-phase converter Figure 4.25: Transient response of the nonlinear average current-mode controller Figure 4.26: The effectivness of the conduction-loss based current sharing scheme Figure 4.27: a) Estimated inductor current versus the output load current with a 1 A sink step size b) Relative error of the estimated current for two sink step sizes: 1 A and 2 A Figure 4.28: Estimated temperature versus sensor temperature xiii

14 List of Appendices Appendix A Filter Time Constant Calculation...90 Appendix B Quantization Error Effects on Current Estimator Accuracy..92 xiv

15 Chapter 1 Introduction This thesis presents the design and practical implementation of advanced digital controllers for multi-phase dc-dc converters utilized in low-power applications such as consumer electronics, computer servers, and telecommunication equipment. In the following sections, common design requirements in such applications are examined, multi-phase controller design challenges are identified, and thesis objectives are formulated accordingly. 1.1 Design Requirements for Modern Power Management Systems in Low-Power Applications Figure 1.1 shows a complex power management system found in a typical notebook computer. The power is delivered to internal devices (i.e. processor, graphic card, and memory) by several high-efficiency, power processing units (PPU). All PPUs are connected to a shared dc-bus voltage generated from either a lithium-ion battery source or an ac adapter. The PPUs usually employ multi-phase switching converters regulated by a dedicated controller circuit. Figure 1.1: Block diagram of a power management system implemented in notebooks (source: Texas Instruments). 1

16 CHAPTER 1. INTRODUCTION 2 The multi-phase converters, shown in Fig. 1.1, utilize between two and four dc-dc converter phases connected in parallel. The primary role of these phases is to step down the dc-bus voltage, usually between 0.5 V and 3.3 V, and supply and share the output load current. The number of parallel converter phases depends on the current consumption ratings of the specific load device. The common design requirements for the controller circuit and multi-phase converters, used in power management systems for low-power applications (up to 100 W), are: small size of required components due the small form factor of electronic devices, minimized component count and reduced cost obtained through on-chip integration of controller circuitry and regulation of a larger number of converter phases with a single controller integrated circuit, high efficiency of the converter and low power consumption of the controller circuitry required to reduce amount of generated heat, improve the battery life, and energy efficiency ratings of the device, fast transient response required to minimize the size of energy storage components and provide reliable operation of supplied devices during load current transients, over-current and over-temperature protection of the converter components and load devices, proper current sharing between paralleled converters to avoid unequal stress of converter components, provide equal aging of all converters, and uniform heat distribution. Digital controllers [1]-[23] that are capable of satisfying some of these stringent requirements have emerged just recently. To reduce the size of converter components, they usually operate at high switching frequencies [2]-[5], while having small power consumption and low hardware complexity suitable for on-chip integration. More advanced digital controllers are also able to achieve fast transient response [6]-[9] limited by the converter components, identify single-phase converter parameters, self-compensate [10]-[13], and optimize the efficiency of single-phase converters by adjusting the size of switching devices [4], [14] and changing modes of the controller operation [3], [15].

17 CHAPTER 1. INTRODUCTION 3 Previous controller designs were mainly focused on single-phase converters and practical controller solutions for multi-phase converters are still missing. The multi-phase converters are currently used in many applications that could greatly benefit with utilization of advanced control and power management techniques from single-phase converters. However, applying these techniques to multi-phase converters is not straightforward due mainly to various design challenges and implementation problems that will be discussed in the following section. 1.2 Multi-Phase Digital Controller Design Challenges The utilization of the state-of-the-art digital controller ICs [19], [21], [23] for multi-phase and multiple standalone dc-dc converters, used in the systems similar to the one shown in Fig. 1.1, is still restricted to high-end products such as computer servers and data communication equipment. There are two primary reasons for this limitation: 1) high power consumption of multi-phase digital controllers (several tens of milliwatts), and 2) higher development/production costs. The power consumption makes the multi-phase digital controllers impractical for utilization in battery-powered electronic devices, such as cellphones, ebook readers, and computer tablets. There are also several major technical problems that remain unsolved. The first problem is related to dynamic efficiency optimization of multi-phase converters used in power management systems of Fig To improve efficiency, reduce generated heat, and increase the battery life, digital controllers commonly utilize efficiency optimization schemes [24]-[26] based on the phase shedding method. However, controller designs with such schemes are quite challenging because the controlled plant dynamically changes every time the active number of phases is adjusted. To avoid potential stability problems, conservative controller designs that sacrifice transient response of the converter for improved efficiency are utilized in practice. The second major problem is associated with the current sensing for single-phase and multiphase converters and temperature monitoring. The current information is essential for overcurrent protection [27], efficiency optimization [22], [26], [28], [29] and current sharing purposes [30]-[36]. The converter inductor currents are still measured using analog circuitry [37]. Implementation of external analog current sensing circuits and their integration with integrated digital controllers is usually costly and difficult due to the requirement for multiple

18 CHAPTER 1. INTRODUCTION 4 analog-to-digital converters (ADCs). The third problem is related to parameter estimation of multi-phase converters. Existing identification methods [10]-[13] are designed to determine inductor-capacitor products of singlephase converters. However, they cannot be directly applied to multi-phase converters where multiple inductor branches are connected to a common output capacitor. If phase components (i.e. phase inductances and output capacitor) could be identified separately, their values could be used to improve the transient response of the converter. Combined fast transient response and dynamic current sharing is also an important design issue with multi-phase converters. Recent multi-phase digital controllers [38]-[40] provide proper sharing of the load current between multiple converter phases but they suffer from the limited controller bandwidth and slow converter response under large load steps. 1.3 Thesis Objectives and Organization In this thesis solutions for the above listed problems are presented. Mainly, new concepts are introduced and tested on single phase systems, then adapted for multi-phase systems. In other words, even designs for single-phase converters are implemented with the final goal of multiphase application in mind. The thesis goals and content of remaining chapters are briefly summarized below. A low-cost flexible multi-phase digital controller IC that can be easily configured and employed in various applications will be developed in the first part of this thesis. The IC will operate between 100 khz and 10 MHz and it will exhibit low power consumption for use in portable applications. The IC will be utilized for interleaved converters having between 2 and 4 converter phases and/or it will simultaneously regulate up to 4 standalone converters providing different output voltages. All major controller parameters (i.e. switching frequency, number of active phases, etc.) of the IC will be externally programmable in order to achieve optimum performance for a specific application. The design, implementation and experimental operation of the system will be presented in Chapter 2. The problem of dynamic efficiency optimization for multi-phase converters will be addressed and resolved in Chapter 3. This chapter will present a current-program mode digital controller architecture and a multiphase dc-dc converter with non-uniform current sharing that dynamically

19 CHAPTER 1. INTRODUCTION 5 optimize converter efficiency over the full load range of operation. Chapter 4 will introduce advanced reconfigurable digital controllers with multi-parameter estimation for single-phase and multi-phase converters. They can accurately estimate inductor currents, perform identification of key converter parameters, and provide remote temperature monitoring of power switches. In the case of the multi-phase converters, the proposed sensorless non-linear average current-mode controller will utilize the identified power stage parameters and estimated inductor currents to provide fast transient response, while maintaining dynamic current sharing and equalized thermal stress between converter phases. Chapter 5 concludes this thesis and suggests potential new research directions with author s anticipation that presented technical solutions in this thesis will fuel development of digitallycontrolled multi-phase converters that are inexpensive, energy efficient, and environmentally friendly. REFERENCES [1] A. Dancy, R. Amirtharajah, A. Chandrakasan, High-efficiency multiple-output dc-dc conversion for low-voltage systems, IEEE Transactions on VLSI Systems, vol. 8, no. 3, pp , [2] B. Patella, A. Prodić, A. Zirger, D. Maksimović, High-frequency digital controller PWM controller IC for DC-DC converters, IEEE Trans. Power Electronics, Special Issue on Digital Control, vol. 18, pp , Jan [3] J.Xiao, A.V. Peterchev, J. Zhang, S.R. Sanders, A 4-μA quiescent-current dual-mode digitally controlled buck converter IC for cellular phone applications, IEEE Journal of Solid-State Circuits, vol. 39, pp , Dec [4] O. Trescases, Integrated power-supplies for portable applications, Ph.D. thesis, University of Toronto, [5] Z. Lukić, N. Rahman, A. Prodić, "Multi-bit sigma-delta PWM digital controller IC for DC- DC converters operating at switching frequencies beyond 10 MHz," IEEE Transactions on Power Electronics, vol.22, pp , Sept [6] G. Feng, E. Meyer, and Y.-F. Liu, A new digital control algorithm to achieve optimal dynamic performance in DC-to-DC converters, IEEE Trans. Power Electron., vol. 22, pp , July [7] Zhenyu Zhao, A. Prodić, Continuous-time digital controller for high-frequency DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , Mar

20 CHAPTER 1. INTRODUCTION 6 [8] L. Corradini, P. Mattavelli, E. Tedeschi, and D. Trevisan, High-bandwidth multisampled digitally controlled DCDC converters using ripple compensation, IEEE Trans. Ind. Electron.,vol. 55, pp , Apr [9] A. Babazadeh, D. Maksimović, Hybrid digital adaptive control for fast transient response in synchronous buck DC DC converters, IEEE Trans. Power Electron., vol. 24, pp , Nov [10] Zhenyu Zhao, A. Prodić, Limit-cycle oscillations based auto-tuning system for digitally controlled DC-DC power supplies, IEEE Trans. Power Electron., vol. 22, pp , Nov [11] L. Corradini, P. Mattavelli, W. Stefanutti, S. Saggini, Simplified model reference-based autotuning for digitally controlled smps, IEEE Trans. Power Electron., vol. 23, pp , July [12] J. Morroni, R. Zane, D. Maksimović, Design and implementation of an adaptive tuning system based on desired phase margin for digitally controlled DC-DC converters, IEEE Trans. Power Electron., vol. 24, pp , Feb [13] M. Shirazi, R. Zane, D. Maksimović, An Autotuning digital controller for DC DC power converters based on online frequency-response measurement, IEEE Trans. Power Electron., vol. 24, pp , Nov [14] O. Trescases, Wai Tung Ng, H. Nishio, M. Edo, T. Kawashima, A digitally controlled DC-DC converter module with a segmented output stage for optimized efficiency, in Proc. International Symposium on Power Semiconductor Devices and IC's, 2006, pp [15] N. Rahman, A. Parayandeh, K. Wang, A.Prodić, Multimode digital SMPS controller IC for low-power management, in Proc. IEEE International Symposium on Circuits and Systems, 2006, 2006, pp [16] A.V. Peterchev, J. Xiao, S.R. Sanders, Architecture and implementation of a digital VRM controller, IEEE Transactions on Power Electronics, vol.18, issue 1, pp , Jan [17] Xu Zhang, Yang Zhang, R. Zane, D. Maksimović, Design and implementation of a widebandwidth digitally controlled 16-phase converter, in Proc. IEEE Workshops on, Computers in Power Electronics, 2006, pp [18] Z. Lukic, C. Blake, Santa C. Huerta, A. Prodić, Universal and fault-tolerant multiphase Digital PWM Controller IC for high-frequency DC-DC converters, in Proc. IEEE Applied Power Electronics Conf., 2007, pp [19] J. Zhang, S.R. Sanders, A digital multi-mode multi-phase IC controller for voltage regulator application, in Proc. IEEE Applied Power Electronics Conf., 2007, pp

21 CHAPTER 1. INTRODUCTION 7 [20] Data Sheet, ZL2004, Step-down DC-DC controller with power management, Zilker Labs. [21] Data Sheet, UCD9240, Digital Power Point of Load System Controller, Texas Instruments Inc. [22] Data Sheet, ZL2005, Step-down DC-DC controller with power management, Zilker Labs. [23] Data Sheet, UCD9220, Digital pwm system controller, Texas Instruments Inc. [24] P. Zumel, C. Fernandez, A. de Castro, O. Garcia, Efficiency improvement in multiphase converter by changing dynamically the number of phases, in Proc. IEEE Power Electronics Specialists Conf., 2006, pp [25] Jia Wei, F.C. Lee, Two-stage voltage regulator for laptop computer CPUs and the corresponding advanced control schemes to improve light-load performance, in Proc. IEEE Applied Power Electronics Conf., 2004, pp [26] J.A. Abu Qahoug, L. Huang, Highly efficient VRM for wide load range with dynamic non-uniform current sharing, in Proc. IEEE Applied Power Electronics Conf., 2007, pp [27] H. Peng, D. Maksimović, Overload protection in digitally controlled DC-DC converters," in Proc. IEEE Power Electronics Specialist Conf., Jun. 2006, pp [28] Z. Lukić, Zhenyu Zhao, A. Prodić, D. Goder, Digital controller for multi-phase DC-DC converters with logarithmic current sharing, in Proc. IEEE Power Electronics Specialist Conf., 2007, pp [29] A. Parayandeh, Chu Pang, A. Prodić, Digitally controlled low-power DC-DC converter with instantaneous on-line efficiency optimization, in Proc. IEEE Applied Power Electronics Conf., 2009, pp [30] B. Tomescu, H.F. Vanlandingham, Improved large-signal performance of paralleled DC- DC converters current sharing, IEEE Trans. Power Electron., vol. 14, pp , May [31] Peng Li, B. Lehman, A design method for paralleling current mode controlled DC-DC converters, IEEE Trans. Power Electron., vol. 19, pp , May [32] J. Abu-Qahouq, Mao Hong, I. Batarseh, Multiphase voltage-mode hysteretic controlled DC-DC converter with novel current sharing, IEEE Trans. Power Electron., vol. 19, pp , Nov [33] Jaber A. Abu Qahouq, Lilly Huang, Doug Huard and Allan Hallberg, Novel current sharing schemes for multiphase converters with digital controller implementation, in Proc. IEEE Applied Power Electronics Conf., Feb. 2007, pp

22 CHAPTER 1. INTRODUCTION 8 [34] Y. Zhang, R. Zane, D. Maksimović, Current sharing in digitally controlled masterless multi-phase DC-DC Converters, in Proc. IEEE Power Electronics Specialist Conf., Jun. 2005, pp [35] Mao Hong, Liangbin Yao, Caisheng Wang, I. Batarseh, Analysis of inductor current sharing in nonisolated and isolated multiphase DC-DC converters, IEEE Trans. Power Electron., vol. 54, pp , Dec [36] V. Choudhary, E. Ledezma, R. Ayyanar, R.M. Button, Fault tolerant circuit topology and control method for input-series and output-parallel modular DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , Jan [37] H. Forghani-zadeh, G. Rincón-Mora, Current sensing techniques for DC-DC converters, in Proc. IEEE MWSCAS, Aug. 2002, pp [38] Jaber A. Abu Qahouq, Lilly Huang, N-phase sensor-less current sharing digital controller, in Proc. IEEE Power Electronics Specialist Conf., 2008, pp [39] A. Kelly, Current Share in Multiphase DC DC Converters Using Digital Filtering Techniques, IEEE Trans. Power Electron., vol. 24, pp , Jan [40] Xu Zhang, L. Corradini, D. Maksimovic, Sensorless current sharing in digitally controlled two-phase buck DC-DC converters, in Proc. IEEE Applied Power Electronics Conf., 2009, pp

23 Chapter 2 Multi-Use and Fault-Tolerant Digital Controller IC 2.1 Introduction In multi-phase dc-dc converters, potential advantages of digital controllers over traditional analog solutions are evident. Digital systems allow implementation of complex power management techniques and advanced control laws [1]-[13]. It has also been shown that digital controllers are able to closely match multiple pulse-width modulated signals [14], which is important in implementing accurate current sharing [14,15]. Despite these advantages, there has been limited utilization of digital controllers in multiphase systems, largely due to unsolved practical implementation problems. Compared to analog solutions [16,17], most existing digital architectures require significantly larger silicon area and have a relatively high power consumption (usually in the range of tens of mw/mhz) [18,19], both of which are making them impractical for on-chip implementation required in targeted highvolume systems. High power consumption presents a very serious limitation in upcoming converters, which are expected to switch much faster than the existing converters, in order to reduce the size of energy storage components. The main goal of this chapter is to introduce a versatile multi-phase digital controller IC architecture, shown in Fig. 2.1, which has low power consumption and takes small silicon area, comparable to that of analog systems. The new IC also exploits flexibility of digital implementation. It can be set to operate as a controller for interleaved converters having 2 to 4 phases and/or simultaneously regulate up to 4 standalone converters providing different output voltages. Furthermore, in multi-phase mode, this architecture features fault-tolerant operation providing uninterrupted output voltage in the case of a failure in one of the phases. The key block of the new IC is a reconfigurable multi-phase digital pulse-width modulator (MDPWM). The programmable parameters are switching frequency (between 100 khz and 10 MHz), phase shifts between PWM signals, as well as dead-times. 9

24 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 10 In this IC, compensator coefficients and voltage references can be also dynamically adjusted through a digital communication interface based on either a parallel or serial I 2 C [20] communication protocol. All of the features allow the use of this flexible controller in various applications, including modern voltage regulator modules (VRMs) [21]-[23], multiple output converters for communication devices and television sets as well as in portable electronics. In interleaved mode the controller tolerates failures in up to three phases and it automatically switches to operation with reduced numbers of phase (for example, from 4 to 3). Potentially, the dynamic reduction of active converter phases can be exploited to improve efficiency at light and medium loads [24] when up to three phases can be disabled. Although very valuable, these features have not been presented in other MDPWM architectures [14], [25], [26], mostly due to the lack of solutions for the operation with an odd number of phases and overly large hardware complexity. In the following section, the operation of the multi-use controller IC is explained. Figure 2.1: A block diagram of the multi-use and fault-tolerant digital controller IC regulating operation of a 4 phase interleaved buck converter.

25 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC System Operation and Controller Architecture Figure 2.1 shows the multi-use controller regulating operation of a 4-phase interleaved buck converter. The controller consists of four analog-to-digital converters (ADCs) [27], four programmable look-up table PID compensators [28,29], MDPWM, dead-time circuits, and master management unit (MMU). The number of active ADCs and PID compensators is regulated by the MMU, through external mode control and over-current protection signals (mode and OCP). In interleaved mode, only one ADC and a compensator are used to regulate the operation of all the phases. When the regulation of multiple outputs is required, a larger number of these blocks are activated. The MMU generates clock signals for ADCs and PID compensators, adjusts phase shifts of control signals, and shuts down critical phases if the OCP signal, activated by a separate measurement circuits, is received. Based on the difference between the output voltage references V ref, set by a sigma-delta digital-to-analog converter [27], the ADCs produce error signals e[n] for PID compensators. The PID calculates duty ratio commands for MDPWM. Depending on the mode of operation, the MDPWM creates up to 4 pulse-width modulated signals, whose duty ratios can be independently adjusted. 2.3 Multi-Phase Digital Pulse-Width Modulator The architecture of the 4-phase MDPWM, shown in Fig. 2.2, is based on a modification of a hybrid DPWM [29,30], in which a low-resolution counter and delay line are combined to create a high-resolution pulse-width modulated signal. In this case, a programmable counter and a synchronization block are introduced, both of which are shared by all phases to minimize hardware complexity and improve matching between the phases. Each phase contains a first-order Σ- modulator [31]-[35], a programmable delay line, a segmented delay matching circuit and a number conversion block. The system uses an external clock signal whose frequency never exceeds nine times the switching frequency making MDPWM power consumption relatively low. At the beginning of each switching cycle, in each of the phases, a set pulse for the SR latch is created and the rising edge of the corresponding dpwm signal c pmwi (t), where i=1,,4, is created. The on-time duration is varied using the counter and delay lines, which reset the latch. The core steps (3 most significant bits - MSBs) of the

26 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 12 Figure 2.2: A block diagram of the proposed multi-phase digital pulse-width modulator. Figure 2.3: Phase synchronization: a) start signals during 4-phase interleaved operation; b) 3-phase interleaved operation. desired 11-bit duty ratio value d i [n] (i =1,..,4) are set by the counter, the fine adjustments (middle 5 bits) are performed through delay lines, and even finer ones (3 least significant bits - LSBs) are obtained through Σ- modulation. The mode of MDPWM operation depends on phase enable and phase angle signals, which select the combination of active phases and the angles between them, respectively. When the number of selected phases is 1, 2 or 4, the counter output changes from 0 to 7. While operating with 3 phases, it counts from 0 to 8. Based on the phase angle

27 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 13 signal, the synchronization block creates set pulses sp i that set SR latches. As an example, Fig. 2.3 illustrates set and counter s output signals during interleaved operation with 4 and 3 phases when phase shifts of 90 and 120 respectively are required Operation with 3 Phases and Number Conversion Logic When operating in the single-phase mode or with an even number of phases, the creation of the duty ratio value proportional to the 8-bit control input d c [n] (see Fig. 2.2) is quite simple. The counter output goes through eight cycles and its ramping output r[n] is compared with the 3- MSBs of d c [n] increased by phase angle value. When the two compared numbers are equal, pulse δ(t) for the delay line is created. The propagation time of the pulse through the 32-cell-long delay line is defined with the 5-MSBs of d c [n]. In the 3-phase mode, the situation is more complex. Now, in each switching cycle the counter goes through 9 steps, which combined with 32 delays, result in 288 different output duty ratio values. This number is higher than the number of possible values for d c [n], creating a problem of assigning a duty ratio value to appropriate input. If wrongly interpreted, the input values can cause non-linear or even non-monotonic input-tooutput DPWM characteristic and consequent closed-loop stability problems. In order to generate a monotonic characteristic, for each input value d c [n], a portion of the duty ratio increment is created by the counter and delay line. To determine this mapping, a minimum-error criterion is used. Here, the difference between the desired increment and sum of increments of the delay line D cn = N cn [n]/9 and the counter D dl = N dl [n]/256 is compared. Where N cn [n] is a 4-bit value controlling the number of counter steps before the delay line is triggered and N dl [n] is a 5-bit value defining the number of delay cells as shown in Fig More precisely, we look for the minimum of the following function representing the absolute error in d c [n] representation: [ n] N [ ] dc[ n] N cn dl n d = +. (2.1) The solution of this equation gives a set of 256 values of N cn [n] and N dl [n] that result in the error distribution shown in Fig. 2.4.a) limited to less than ±0.5 LSB. As a result, monotonic operation is obtained as illustrated in Fig. 2.4.b). The values of N cn [n] and N dl [n] are stored in two look-up

28 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 14 tables used for generating proper increment portions during 3-phase operation. Figure 2.4: The 3-phase MDPWM operation: a) the absolute duty cycle error vs. 8-bit duty ratio command d c [n]; b) actual duty cycle versus d c [n] Dual-Biased Delay Cell for Wide-Frequency Range One of the main challenges of DPWM architectures involving a delay line and a counter [36,37] or segmented delay lines [38,39] is linearity. When good matching between the counter time increment and the total propagation time of all delay cells is not achieved, a non-monotonic characteristic can occur [36]. As a result, at certain duty-ratio operating points, local stability problems [40] occur. In practice this undesirable behavior is caused by fabrication defects and temperature variations. The delay-locked loop (DLL) based structures are utilized [25], [37], [39], [41] to adjust the delay cells and allow linear operation at programmable switching frequencies [41]. Except for the segmented delay-line DPWM [41], existing DLL-based DPWM implementations are not designed for operation over a wide frequency range of the multi-use controller from Fig To achieve the 8-bit resolution of the core DPWM (without sigma-delta) in the frequency range between 100 khz and 10 MHz, the delay cell propagation time t pd needs to be varied significantly.

29 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 15 To change the propagation time, current-starved delay cells [42], from Fig. 2.5.a) are often employed. Their propagation time t pd is regulated by a programmable bias current source I bias. The finite steps ΔI bias of the current source are then further reduced through the mirroring stage 1/K 1 (K 1 >> 100), to obtain very precise control over t pd. The main problem of this circuit is that it requires a large biasing current (~ 1 ma) to reduce t pd from 39 ns to 390 ps, corresponding switching frequencies of 100 khz to 10 MHz, respectively. Such a high bias current value is not acceptable for low-power applications. Figure 2.5: A delay cell with adjustable propagation time t pd : a) conventional current-starved delay cell b) proposed dual-biased current-starved delay cell. To solve for this problem, a power efficient dual-bias delay cell is developed. The proposed circuit is shown in Fig. 2.5.b). It consists of two CMOS inverters and a dual current mirroring input stage that discharges the output equivalent capacitance seen at the node A. The propagation time of a signal entering the cell is inversely proportional to the instantaneous current of the mirroring stage i m (t). This current is formed as a scaled sum of currents produced by two current sources I coarse, I fine and during the delay cell transition period its value is ( t) I I =, (2.2) fine coarse m + K1 K 2 i where current ratio K 1 is larger than K 2. In this way, the need for a single current source having a wide current range and high power consumption, at high switching frequencies, is eliminated as shown in Fig Still, a relatively high current i m (t) ensuring short propagation time of delay

30 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 16 cells can be achieved by I coarse. When long propagation times are required, I coarse can be reduced and a precise delay regulation can be achieved through I fine adjustments. It should be noted that, in this application, i m (t) has a relatively small influence on the delay line s power consumption. This is because i m (t) occurs only during short state transients, and in the targeted range of switching frequencies its average value is small. This structure also provides more accurate regulation of delay times. For large delays, conventional current starved delay cells have poor regulation of delay times, due to inaccurate adjustment of low bias currents. In this case, this problem is minimized. Figure 2.6: Proposed adjustment of the current-starved delay cell current i m (t) in the delay line versus the desired switching frequency f sw Segmented Bias Circuit Current sources I coarse and I fine, introduced in Fig. 2.5, are designed to be digitally programmable with fixed current increments ΔI fine and ΔI coarse. The implementation of the bias circuit is shown in Fig It can be integrated in deep sub-micron CMOS technologies, due to a requirement for only two stacked transistors between supply rails. The current source I fine, consists of 5 binarysized PMOS transistors and 20 equally-sized PMOS transistors. This arrangement in the transistor sizing simplifies the digital logic that controls their operation while providing very accurate and monotonic change of i m (t) provided by I fine. The precise adjustment of I fine and t pd is then achieved by turning on/off the binary-sized PMOS transistors using a 5-bit binary pmos select input. Even finer current increments are obtained by using the 20-bit point pmos select

31 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 17 Figure 2.7: Digitally controlled segmented-bias circuit. input tied to the equally-sized transistor. The coarse change of t pd is achieved by adjusting the 8- bit range pmos select input. The transistors controlled with this input are sized in a thermometer code fashion. This was implemented in order to guarantee monotonic behavior of t pd during coarse bias current change. Additionally, to generate a smooth change of t pd when changing I coarse, the current ratio K 2 of the right-side mirror has to be sized according to: I max fine K 1 I coarse =. (2.3) K 2 Due to the smaller current mirror ratio K 2, the maximum value of I coarse required to obtain the smallest t pd (390 ps) is an order of magnitude smaller than that of the conventional cell, shown in Fig. 2.5.a. The propagation delay t pd can be also accurately controlled. In this case, over the proposed switching frequency range, the number of frequencies where propagation delay t pd is accurately adjusted is 5120 (2 5 x 20 x 8). The effective frequency resolution of the delay-lock loop is less than 2 khz. If further improvement is desired, one can change the number and sizing of biasing transistors and adjust the current mirror ratios K 1 and K 2.

32 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC Digital Delay-Lock Logic The block diagram of the digital delay-lock logic [37], [41] that adjusts the segmented bias circuit based on the input external clock is shown in Fig. 2.8.a). During each switching period, the reset pulse rp i (t) from the MDPWM is injected into the delay line. Depending on the delay cell bias current i m (t), the rising edge of rp i (t) travels through the delay line with a certain propagation speed. If i m (t) is set properly, when the falling edge of rp i (t) occurs 1/8 T sw after the pulse was created, the rising edge of rp i (t) should reach the last delay line node n 32. This situation is illustrated in Fig. 2.8.b). Similarly, when the MDPWM operates in the three-phase mode, the output of the last small cell (0.44 t pd ) at node n 28a exhibits the rising edge 1/9 T sw after the pulse was generated. However, if i m (t) is larger than required, the pulse propagates faster through the delay line. Consequently, when the falling edge of rp i (t) occurs, all delay cell outputs are set high including the last two nodes n 31 and n 32. Similarly, if the bias current is lower, the pulse propagates slower than desired and logic levels at both node n 31 and n 32 remain low. Therefore, by observing only n 31 and n 32 during the falling edge of rp i (t), an appropriate decision about increasing/decreasing the bias current can be made. Logic levels at n 31 and n 32 (1, 2, or 4 phases are active) or n 28 and n 28a (3 phases are active) are sampled each switching cycle at the falling edge of rp(t). Depending on the mode of operation (3-phase/4-phase), the sampled levels are fed to a current adjust finite-state machine (FSM) in Fig. 2.8.a). The FSM then decides the adjustment of the bias current for delay cells. For instance, if the total propagation time for the delay line is not matched initially, the sampled logic levels at n 31 and n 32 are 00 ( rp i (t) Figure 2.8: a) A block diagram of digital delay-locking logic; b) MDPWM delay line signals when 1, 2, or 4 phases are active.

33 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 19 propagates too slow) or 10 (to be determined). The bias current is then gradually increased by the FSM until 11 state (rp i (t) propagates too fast) is detected. The current level is then reduced until state 10 is reached. In this way, the rising edge of the pulse at node n 32 and falling edge of rp i (t) are tuned as close as possible as shown in Fig. 2.8.b). The bias current is then considered adjusted and further corrections are not performed unless state 00 or 11 are sensed again. In the opposite case, if the initial bias current is too large and state 11 is detected, the bias current is decreased until pulse slows down and state 10 is sampled for the first time. At that point, the bias current for the delay line is frozen and delay line is locked. 2.4 Programmable Dead-Time Circuit A simplified block diagram of the programmable dead-time circuit generating non-overlapping control signals c(t) and c (t) is shown in Fig The non-overlapping clock generator circuit [35] is modified to include two delay lines and two 16/1 multiplexers for dead time value adjustment. Two multiplexers MUX 16/1 allow selection of the desired falling and rising dead time values by adjusting the effective length of the delay lines. The maximum dead-time value that can be obtained with 16 delay cells is 6.25% of T sw. The dead-time increment is equal to 0.39% of the switching period. 2.5 ADC Implementation Figure 2.9: Programmable dead-time circuit. The ADCs from Fig. 2.1 are implemented, as described in [27]. This low-power differential delay-line ADC architecture is suitable for fabrication in deep sub-micron CMOS IC technologies. To provide tight output voltage regulation (error 1%) of the digital loop, the ADC quantization step scales with the analog voltage reference. Each of four analog ADC

34 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 20 references can be dynamically adjusted with four 10-bit Σ-Δ DACs interfaced with the MMU as shown in Fig This allows the use of this controller in modern power management systems with dynamic or adaptive voltage scaling [43]. 2.6 IC Implementation The digital controller of Fig. 2.1, was fabricated in TSMC 0.18-µm CMOS process. A photograph of the fabricated chip is shown in Fig a) and its important parameters are summarized in Table 2.1. The four-phase controller blocks occupy 1.72 mm 2 of the silicon area, which is comparable or even smaller than the area required for implementation of analog solutions [44]. The utilized area can be further reduced by using custom-designed memory blocks for storing controller parameters instead of flip-flop arrays and by time-sharing digital hardware blocks (PID compensators, delay-lock loops, etc.) between phases. The measured chip current consumption is 3.6 ma at the maximum switching frequency of 10 MHz when all 4 Table 2.1: Summary of the IC parameters. Technology TSMC 0.18 µm CMOS Supply voltage 1.8 V / 3.3 V Switching frequency 100 khz 10 MHz No. of phases 4 Max. current consumption MHz Normalized current consumption µa/mhz/per phase Controller silicon area 1.7 mm 2 Figure 2.10: a) Fabricated chip die of the multi-use digital controller IC; b) Current consumption breakdown.

35 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 21 phases are active and running separately. The current consumption breakdown is also presented in Fig b). The normalized current consumption figure of µa/mhz for each phase is equivalent to the state-of-art analog solutions [16,17] and much smaller than commercial digital controllers [18,19], that consume up to 50 ma when regulating four converter phases switching at 2 MHz. Consequently, this controller could be used in portable applications [41] as well. 2.7 Experimental Results To verify the operation of the controller IC, open-loop and closed-loop tests with four buck converters are performed. The buck converter phases used for the closed loop tests are designed to operate at the nominal switching frequency of 1 MHz and supply up to 5 A of the load current. All desired controller parameters and configuration settings for different tests are loaded to the on-chip memory via the built-in communication interface Open Loop Tests To test the monotonic behavior and check linearity of the MDPWM the pulse-width modulated output is first connected to a buck converter switching at 100 khz. The MDPWM is set in the three phase mode. The input 11-bit duty ratio command d[n] was gradually changed between 10% and 90% of its maximum value. The waveform shown in Fig demonstrates that the output voltage changes linearly. Large non-monotonic regions caused by the MDPWM cannot be observed due to the automatic delay adjustment performed by the digital delay-lock loop. Figure 2.11: MDPWM linearity test at 100 khz Ch1: output converter voltage (1 V/div).

36 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 22 In the next step, the external clock frequency is increased in order to create a 10-MHz switching signals. The input duty ratio command is now varied between four consecutive values to produce pulse-width modulated signals shown in Fig a). From the zoomed part of the waveform, shown in Fig b), it can be observed that the propagation time of delay cells is properly adjusted to 390 ps as required for 8-bit resolution at 10 MHz. The measured current consumption of one MDPWM phase operating at 10 MHz is 260 µa. Figure 2.12: MDPWM operation at 10 MHz: a) the duty ratio command is changed between 4 consecutive values b) linearly distributed falling edges of the pulse-width modulated signals (8-bit resolution at 10 MHz) Closed Loop Tests The closed loop operation of the controller IC is tested for both multiple-output and interleaved modes of operation. In the multiple-output mode, 4 buck converters operating at 1 MHz are used to provide the output voltages of 1.2 V, 1.8 V, 2.5 V, and 3.3 V. The desired voltage references are set digitally. Fig a) shows the captured waveforms. All output voltages are regulated/ at their designated values. Fig b) shows the response of the controller when a phase voltage reference is now dynamically modified through the communication interface. Once the data is received by the controller, the MMU quickly updates the reference value fed to the dedicated Σ- Δ DAC. As a result, the ADC reference voltage is increased from 1.5 V to 2.7 V and the controller regulates the converter output voltage to reach the new steady state within 30 μs, which is sufficiently fast for a most power saving techniques based on dynamic voltage adjustment [43], [45], [46]. Finally, the interleaved mode of operation is first verified with two phases enabled as shown in Fig c). The controller IC recognizes that only two phases are active, and it adjusts phase shifts between control signals to T sw /2 to reduce the output voltage

37 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 23 ripple. In the next step, two additional phases are enabled and phase shift is now again updated to T sw /4 by the MMU, as demonstrated in Fig d). Figure 2.13: Closed-loop operation of the controller IC: a) four 1-MHz buck converters are regulated to produce four supply rails between 1.2 and 3.3 V; b) transient response of the controller when the voltage reference is dynamically changed; c) two-phase interleaved steady-state operation; d) four-phase interleaved steady-state operation; Fault-Tolerant Operation The fault tolerant operation is tested by intentionally generating the over-current protection signal (OCP) (from Fig. 2.1) in one of the phases as shown in Fig Once the OCP is detected, the faulty phase is turned off by the MMU. The MMU then observes the remaining number of phases and starts updating phase angles for all active phases until all control signals are shifted by 1/3 of the switching period. Similarly, if one more phase is turned off, the MMU will adjust the phase shift to one half of the switching period. The automatic phase shift reconfiguration can be also utilized for optimizing the multi-phase converter efficiency when the

38 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 24 number of enabled phases varies with the load current. In other words, at light load currents up to three phases can be turned off [24] to reduce switching losses and improve the efficiency. Figure 2.14: Fault-tolerant operation in the closed loop Ch1: output converter voltage v out (t) at 3.3 V (500 mv/div); Ch2: Output converter voltage v out (t) AC component (50 mv/div); D1-D4: MDPWM control signals switching at 1 MHz; D5 OCP signal; Time scale is 2 μs/div. 2.8 Conclusion In this chapter a flexible multi-phase/multi-stage digital controller IC having low power consumption and occupying a small silicon area is presented. The controller can be used in various applications including portable electronic devices. To achieve these characteristics, novel architecture of a multi-phase DPWM is developed and combined with a newly designed current starved delay cell. The performances of the controller IC are experimentally verified on several experimental setups, including multiple-output and interleaved dc-dc converters. REFERENCES [1] A.V. Peterchev, S.R. Sanders, Digital loss-minimizing multimode synchronous buck converter, in Proc. IEEE Power Electronics Specialists Conf., 2004, pp [2] V. Yousefzadeh D. Maksimović, Sensorless optimization of dead times in DC-DC converters with synchronous rectifiers, in Proc. IEEE Applied Power Electronics Conf., 2005, pp [3] A. Prodić and D. Maksimović, Digital PWM controller and current estimator for a lowpower switching converter, in Proc. IEEE Computer in Power Electronics Workshop, 2000, pp

39 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 25 [4] W. Stefanutti, P.Mattavelli, S. Saggini, and M. Ghioni, Autotuning of Digitally Controlled Buck Converters Based on Relay Feedback, in Proc. IEEE Power Electronics Specialists Conf., 2005, pp [5] B. Miao, R. Zane, D. Maksimović, System identification of power converters with digital control through cross-correlation methods, IEEE Trans. Power Electron., vol. 20, pp , Sept [6] A.L. Kelly, K. Rinne, A self-compensating adaptive digital regulator for switching converters based on linear prediction, in Proc. IEEE Applied Power Electronics Conf., 2006, pp [7] Zhenyu Zhao, A. Prodić, Limit-cycle oscillations based auto-tuning system for digitally controlled DC-DC power supplies, IEEE Trans. Power Electron., vol. 22, pp , Nov [8] A. Prodić, D. Maksimović, and R.W Erickson, Digital controller chip set for isolated DC power supplies, in Proc. IEEE Applied Power Electronics Conf., 2003, pp [9] A. Soto, P. Alou, J.A. Cobos, Non-Linear Digital Control Breaks Bandwidth Limitations, in Proc. IEEE Applied Power Electronics Conf., 2006, pp [10] G. Feng, E. Meyer, and Y.-F. Liu, A new digital control algorithm to achieve optimal dynamic performance in DC-to-DC converters, IEEE Trans. Power Electron., vol. 22, pp , July [11] Zhenyu Zhao, A. Prodić, Continuous-time digital controller for high-frequency DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , Mar [12] L. Corradini, P. Mattavelli, E. Tedeschi, D. Trevisan, High-bandwidth multisampled digitally controlled DC-DC converters using ripple compensation, IEEE Trans. Ind. Electron.,vol. 55, pp , Apr [13] A. Babazadeh, D. Maksimović, Hybrid digital adaptive control for fast transient response in synchronous buck DC DC converters, IEEE Trans. Power Electron., vol. 24, pp , Nov [14] A.V. Peterchev, J. Xiao, S.R. Sanders, Architecture and implementation of a digital VRM controller, IEEE Transactions on Power Electronics, vol.18, issue 1, pp , Jan [15] J. A. A. Qahouq, L. Huang, D. Huard, A. Hallberg, Novel current sharing schemes for multiphase converters with digital controller implementation, in Proc. IEEE Applied Power Electronics Conf., 2007, pp [16] Data Sheet, ISL6264, Two-phase core controller for AMD mobile Turion CPUs, Intersil Corp.

40 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 26 [17] Data Sheet, TPS5210, Synchronous Buck Controller for Pentium Processors, Texas Instruments Inc. [18] Data Sheet, UCD9240, Digital power point of load system controller, Texas Instruments Inc. [19] Data Sheet, ZL2005, Step-down DC-DC Controller with power management, Zilker Labs. [20] Data Sheet, I 2 C bus specification, version 2.1, Philips Semiconductors. [21] Y. Panov, M. Jovanović, "Design considerations for 12-V/1.5-V, 50-A voltage regulator modules," IEEE Transactions on Power Electronics, vol. 16, issue 6, pp , Nov [22] Data Sheet, FDMF8700, Driver plus FET Multi-chip Module, Fairchild Semiconductors. [23] Data Sheet, PIP212-12M, DC-to-DC converter powertrain, NXP Semiconductors. [24] P. Zumel, C. Fernandez, A. de Castro, O. Garcia, Efficiency improvement in multiphase converter by changing dynamically the number of phases, in Proc. IEEE Power Electronics Specialists Conf., 2006, pp [25] R. Kelley, K. Rimme, R.C. Kavanagh, W.P. Marnane, M.G. Egan, Multiphase digital pulsewidth modulator, IEEE Transactions on Power Electronics, vol. 21, issue 3, pp , May [26] T. Carosa, R. Zane, D. Maksimović, Implementation of a 16 Phase Digital Modulator in a 0.35μm Process, in Proc. IEEE Computer in Power Electronics Workshop, 2006, pp [27] A. Parayandeh, A. Prodić, Programmable Analog-to-Digital Converter for Low-Power DC DC SMPS, IEEE Transactions on Power Electronics, vol. 23, issue 1, pp , Nov [28] A. Prodić, D. Maksimović, "Design of a digital PID regulator based on look-up tables for control of high-frequency DC-DC converters," in Proc. IEEE Computers in Power Electronics Conf., June 2002, pp [29] B. Patella, A. Prodić, A. Zirger, D. Maksimović, High-frequency digital controller PWM controller IC for DC-DC converters, IEEE Trans. Power Electronics, Special Issue on Digital Control, vol. 18, pp , Jan [30] A. P. Dancy, R. Amirtharajah, A.P. Chandrakasan, High-efficiency multiple-output DC- DC conversion for low-voltage systems, IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol.8, issue 3, pp , June 2000.

41 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 27 [31] Z. Lu, Z. Qian, Y.Zeng, W. Yao, G. Chen, Y. Wang, Reduction of digital PWM limit ring with novel control algorithm, in Proc. IEEE Applied Power Electronics Conf., 2001, pp [32] Z. Lukić, K. Wang, A. Prodić, High-frequency digital controller for dc-dc converters based on multibit sigma-delta pulse-width modulation, in Proc. IEEE Applied Power Electronics Conf., 2005, pp [33] A. Kelly, K. Rinne, High Resolution DPWM in a DC-DC converter application using digital sigma-delta techniques, in Proc. IEEE Power Electronics Specialists Conf., 2005, pp [34] Z. Lukić, Sigma-Delta DPWM Controllers for 12 MHz DC-DC Switch-Mode Power Supplies, M.Sc. thesis, University of Toronto, May [35] D.A. Johns and K. Martin, Analog Integrated Circuit Design, Wiley, [36] A. Syed, E. Ahmed. D. Maksimović, E. Alarcon, Digital Pulse-Width Modulator Architectures, in Proc. IEEE Power Electronics Specialists Conf., vol.6, pp June [37] V. Yousefzadeh, T. Takayama, D. Maksimović, Hybrid DPWM with Digital Delay- Locked Loop, Proc. IEEE Computer in Power Electronics Workshop, 2006, pp [38] O. Trescases, G. Wei, W.T. Ng, A Segmented Digital Pulse Width Modulator with Self- Calibration for Low-Power SMPS, in Proc. IEEE International Conference on Electron Devices and Solid-State Circuits, pp , Dec , 2005 [39] K. Wang, N. Rahman, Z. Lukić, A. Prodić, "All-digital DPWM/DPFM controller for lowpower DC-DC converters," in Proc. IEEE Applied Power Electronics Conf., 2006, pp [40] A. Peterchev, S. Sanders, Quantization resolution and limit cycling in digital controlled PWM converters, in Proc. IEEE Power Electronics Specialists Conference, vol. 2, 2001, pp [41] O. Trescases, Integrated power-supplies for portable applications, Ph.D. thesis, University of Toronto, [42] M. Maymandi-Nejad, M. Sachdev, A digitally programmable delay element: design and analysis, IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 11, issue 5, pp , Oct [43] T. Burd, T. Pering, A. Stratakos, and R. Brodersen, A dynamic voltage scaled microprocessor system, in IEEE Journal of Solid-State Circuits, vol. 35, no. 11, pp , Nov

42 CHAPTER 2. MULTI-USE AND FAULT-TOLERANT DIGITAL CONTROLLER IC 28 [44] Kuo-Hsing Cheng, Chia-Wei Su, and Hsin-Hsin Ko, A high-accuracy and high-efficiency on-chip current sensing for current-mode control CMOS DC-DC buck converter, IEEE 15 th International Electronics, Circuits and Systems Conference, 2008, pp [45] Design Guidelines, Voltage regulator module (VRM) and enterprise voltage regulatordown (EVRD) 11.1, Intel Corporation. [46] A. Soto, A. de Castro, P. Alou, J. A. Cobos, J. Uceda, A. Lotfi, Analysis of the buck converter for scaling the supply voltage of digital circuits, IEEE Trans. Power Electronics, vol. 22, pp , Nov

43 Chapter 3 Dynamic Efficiency Optimization for Multi-Phase Converters 3.1 Introduction The efficiency of a typical multi-phase dc-dc switch-mode power supply (SMPS) used in telecommunication devices, consumer electronics, and personal computers depends on its load [1]-[3]. Even though the supplied devices usually do not take constant power at all times, the converters are often designed to be most efficient at a certain load (usually maximum). As a result the efficiency at light and medium loads is degraded. In modern systems this causes significant losses, since the devices operate in suboptimal modes over large time intervals. In multi-phase systems, utilized for high-current applications [4], the load current is usually shared equally among several converter stages [5]-[8], to reduce the inductor current ripple [9], stress on components, and improve dynamic response. To optimize efficiency of multiphase converters over the full load range, several approaches have been proposed. The phase shadowing techniques [10] improve the efficiency of the converter with uniform current sharing by switching off phases at lighter loads. This technique provides a significant efficiency improvement over the full-range of operation but requires a relatively large number of phases to achieve a flat efficiency curve. The method based on non-uniform current sharing [11] utilizes multiple power stages with different current ratings. To improve the efficiency in this implementation, the phases are switched on and off in a predermined fashion similar to the phase shadowing method while the load current is distributed among converter phases by a current sharing loop according to their current ratings. This method results in a smaller number of phases but the shape of the efficiency curve is degraded. However, in both of the abovementioned methods the power stages are presented but the controller implementation issues have not been addressed. In particular, controller stability problems [12] during phase on/off transients, which will be addressed soon, cannot be solved in a simple manner. Hence, the authors demonstrate steady-state operation of the presented architectures only and suggest possible controller implementations. 29

44 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 30 The main goals of this chapter are to address and solve the problems of practical controller implementation in multi-phase converters and to show a new system that results in a small number of phases having both a flat and optimized efficiency curve. To achieve these characteristics the system, shown in Fig. 3.1, utilizes two key elements: a multi-phase digital current program-mode based controller, which does not require analog-to-digital converters for the inductor current measurements, and a power stage with logarithmic current sharing. The system employs N parallel converters operating as binary-weighted constant current sources based on a concept similar to the method previously used for designing high-frequency resonant converters [13]. Each of the N phases is optimized by selecting the size of semiconductor switches, the inductor value, and the switching frequency such that the maximum efficiency at its designated current I i = 2 i 1 I nom, is achieved, where i = 1, 2,, N. The instantaneous number of phases operating at any point of time is dynamically regulated by the digital current program mode controller (DCPM). Depending on the load current (I load ) requirements, the controller generates a digital control variable i tot [n] that enables or disables the phases, such that I load = ΣI i. The state of each phase is controlled through phase enable signals s 1, s 2,, s N. In this implementation, s 1 corresponds to the least significant bit (LSB) of i tot [n] and s N, controlling the current of 2 N 1 I nom, is the most significant bit (MSB). Since each of the converters individually operates at the most efficient point, in this way, the optimal efficiency over the full range of Figure 3.1: Block diagram of a digitally controlled multi-phase converter with logarithmic current sharing.

45 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 31 operation is dynamically achieved. Also, as discussed in the following section, the hardware requirements for a relatively complex current loop are minimized. 3.2 System Architecture Fig. 3.2 shows a 4-phase implementation of the proposed system. Its controller has a digital voltage loop containing an analog-to-digital converter (ADC) for the output voltage error e[n] measurement and a PI compensator producing the signal i tot [n], controlling the total current of the converter. The sampling of the output voltage and the update of i tot [n] is performed at the frequency equal to the rate of the fastest-switching converter phase. Compared to the traditional implementation of a fully-digital current programming controllers (DCPM) [14]-[16], which requires multiple ADCs for the phase current measurements, the system of Fig. 3.2 is much simpler. It eliminates the need for a large number of fast ADCs and Figure 3.2: Detailed system architecture of the proposed controller with a logarithmic current sharing.

46 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 32 instead, as shown in Fig. 3.2, uses a set of comparators with thresholds v nom, 2v nom, and 4v nom that are proportional to the binary weighted currents. As a result, the peak current-mode control [17] in all phases is achieved. It should be noted that the elimination of a large number of ADCs greatly simplifies the controller implementation and complexity. On the other hand, the comparators of Fig. 3.2 can be fairly simple. They need to be fast but not necessarily very accurate. Any offset in the comparator threshold is automatically compensated by the voltage loop and in most cases has only a negligible effect on the phase efficiency. To provide accurate current regulation and eliminate the stability problem related to the phase toggling, which will be addressed in the following sections, a digital-to-analog converter (DAC), an additional low-current phase, and a hysteretic logic block followed by a transient current estimator and a phase enable corrector blocks are added. Similar to conventional implementations [17], the controller also has current amplifiers and an SR latch, whose operation is either enabled or disabled with the phase enable signal s i. 3.3 Practical Implementation One of the advantages of the new controller architecture over the voltage-mode controllers [10,11] is the simplicity of the compensator implementation. This can be explained by looking at the control-to-output transfer function of voltage-mode controlled N-phase buck converter [17] G vd s 1+ vout ( s) ωesr ( s) = = Vin, (3.1) 2 d( s) s s 1+ + Q ω ω where the corner frequency ω 0 is 1 ω 0 =. (3.2) L C N It can be seen that the corner frequency ω 0 depends on phase inductance L, total output capacitance C, as well as on the number of active phases N. This dependence on the number of active phases dynamically changes transfer function and creates a serious control problem in

47 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 33 voltage mode controlled converters using efficiency optimization schemes [10,11] based on the phase shadowing method. To provide stability under all operating conditions, a slow compensator is often used, meaning that the dynamic performance of the system is sacrificed to improve power processing efficiency. The current-mode controller of Fig. 3.2 does not suffer from the phase-dependent transfer function variation problem. In this system the phases behave as control-signal dependent current sources [17], [21] connected to the parallel connection of the load resistance R and the output capacitance C. The small-signal first-order transfer function G vic (s) is shown [17] to be: G v ( s) s 1+ ω out esr vic ( s) = = R, (3.3) i s c ( s) 1+ ω p where the pole frequency ω p is inversely proportional to the RC product: 1 ω p =. (3.4) RC Accurate Current Regulation The main problem of the architecture shown in Fig. 3.1 is the accuracy of current regulation. Since the current can be changed in discrete steps only, related limit cycling problems affecting system stability and output current/voltage regulation can occur [22]. To eliminate the quantization effects without a significant increase in the number of phases, the originally proposed structure is modified as shown in Fig A phase with variable current that can be continuously changed from 0 to 1.5I nom is added and, for the 4-phase implementation, the remaining three phases are scaled as I nom, 2I nom, and 4I nom. As described in the following section, the rating of the variable phase is selected such that jittering of the system is eliminated. To achieve the desired current regulation, in this case with an 11-bit resolution, the 3 MSBs of the control signal are used as phase enable/disable variables and the 9 LSBs are fed into a singlebit Σ- DAC [23], which sets the reference for the phase with variable current. It should be noted that the implementation of the DAC is significantly simpler than that of an ADC [23,24] and its

48 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 34 introduction has a relatively small effect on the system complexity. Also, since the power rating of the variable-current phase is small, its wide-range operation does not significantly affect the system efficiency Jitter Elimination An additional difficulty in implementing the proposed controller is jittering (chattering) of the control signal i tot [n]. It occurs when the PI controller frequently changes its output between two approximate values resulting in a phase reconfiguration. For example, assume that a light load change requires the total inductor current, i.e. current reference i tot [n], to change between 3.99I nom and 4.01I nom. Initially, to provide a current of 3.99I nom, two constant current phases are switched on to produce I nom, 2I nom and the variable current phase produces 0.99I nom. To increase the current to 4.01I nom, the 4I nom can be turned on and the remaining part, of 0.01I nom, can be provided by the variable current phase. However, such an operation is undesirable since it requires inductor currents of all constant-current phases to change from rated values to zero or in opposite direction. This abrupt change of all currents can negatively affect the output voltage regulation, as shown in Fig Figure 3.3: The output voltage regulation is affected by the jitter (shown in the red circle) of i tot [n].

49 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 35 For this reason, a hysteretic logic block of Fig. 3.2 is introduced, and the maximum current of the variable-current phase is set to 1.5I nom. The operation of the hysteretic logic that creates the actual phase enabling signal is described using Fig Figure 3.4: Operation of the hysteretic logic output phase enable signal versus input control value i tot [n]. 3.4 Transient Operation Sudden load current changes ΔI load generate output voltage disturbances. To keep the output voltage tightly regulated, the PI compensator from Fig. 3.2 immediately reacts by adjusting the digital command i tot [n] until the output voltage returns to steady state (e[n]=0). To achieve a fast transient minimizing voltage deviation ΔV, it is necessary for the actual total current i tot (t)=σi Li (t) from all active phases to quickly follow the digital command i tot [n]. This condition is challenging to satisfy because the current slew rates are limited by the phase inductances and input-output voltage conditions. For example, in a buck converter, the rising slope Δ 1 of a phase current is significantly larger than the falling slope Δ 2 due to large difference between the input voltage v in and the output voltage v out. The slope Δ 1 and Δ 2 can be calculated as: v vout 1 =, (3.5) L in v out 2 =. (3.6) L

50 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 36 The effect of current slopes on i tot (t) when the digital current command i tot [n] is stepped down from 5.4I nom to 3.6I nom is illustrated in Fig From Fig. 3.5, such a change in the i tot [n] requires that the largest phase 4I nom is disabled while phases 2I nom and I nom are enabled and the variable phase is adjusted from 1.4I nom to 0.6I nom. The phases that are enabled quickly reach their nominal currents within a single switching cycle T sw since Δ 1 is large due to v in >>v out. On the other hand, for the largest phase it takes several switching cycles of the variable phase until the current in that phase reaches zero. As a result, i tot (t) initially goes in the opposite direction of i tot [n], creating an undesirable large voltage overshoot that will be demonstrated in Section 3.5. Figure 3.5: The effect of finite rising Δ 1 and falling current slope Δ 2 on i tot (t) for a step of the digital current command i tot [n] from 5.4I nom A to 3.6I nom.

51 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 37 To eliminate this problem, two new blocks: phase enable corrector and transient current estimator are added into the system as shown in Fig The role of the transient current estimator is to calculate phase currents i 4Inom [n], i 2Inom [n], i Inom [n] and i var [n] and their total sum i sum [n] during the load transient. This can be achieved by monitoring the digital current command i tot [n-1] and the 4-bit corrected phase enable [n-1] command calculated in the previous switching cycle. For example, if a corrected phase enable bit associated to a specific phase, changes from 1 to 0, the current in that phase starts to decrease from its nominal value at a rate equal Δ 2 T sw, in each switching cycle. Similarly, based on the rising slope Δ 1 the phase current can be estimated when the phase is enabled again. Finally, by summing up all estimated phase currents, i sum [n] is obtained. The value of i sum [n] is a digital equivalent of the actual total current i tot (t) fed to the load during the transient. To improve the accuracy of the estimation, when the converter reaches steady state (e[n]=0 for several switching cycles), the current values for each phase are calibrated using i tot [n] from the PI compensator. Simultaneously, based on the estimated currents, the phase enable corrector tries to match the estimated current sum i sum [n] to the digital current command i tot [n] as close as possible. To achieve this, it calculates necessary correction of the phase enable signal and the proper current reference i ref1 [n] for the variable phase. The decision process starts by calculating the difference δ max between the current command i tot [n] and estimated current sum i sum [n] less the estimated current in the variable phase i var [n]: δ = i n] ( i [ n] i [ n]). (3.7) max tot[ sum var The value of δ max indicates the value of the maximum current increment that can be handled by fixed-current phases 4I nom, 2I nom and I nom. Then by comparing the states of phase enable command and the previous corrected phase enable command and by using δ max, phases are enabled/disabled in such a way that the sum of all individual current increments in the fixed phases δ est = δ 4 + δ 2 + δ 1 (δ 4, δ 2, and δ 1 are obtained from Δ 1 and Δ 2 and current levels in the phases during transient) does not exceed δ max. In the final step, the reference i ref1 [n] for the variable phase applied in the next switching cycle is calculated as: i [( i [ n] i [ n + δ ] ref 1 [ n] = itot[ n] sum var ]) est. (3.8)

52 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 38 Note that i ref1 [n] cannot exceed 1.5I nom and therefore if i ref1 [n] becomes larger than 1.5I nom it is saturated to 1.5I nom. Fig. 3.6 shows the simulated operation of the controller during a heavy-to-light load transient between 5.4I nom and 3.6I nom. The PI compensator reacts to the increasing voltage deviation and decreases i tot [n] to the vicinity of the new load current level. Therefore the largest phase 4I nom needs to disabled while phases 2I nom and I nom enabled. To perform the phase reconfiguration and improve the converter efficiency without increasing the voltage deviation, the phase enable corrector and transient current estimator first discharge the inductor current in the largest phase, and then sequentially enable phases I nom and 2I nom as shown in Fig As a result, the multiphase converter is dynamically readjusted to operate at the most efficient phase configuration, while providing small voltage deviation and fast response. Figure 3.6: Simulated transient operation of the controller during a heavy-to-light load transient between 5.4I nom and 3.6I nom. 3.5 Design of Converter Phases Converter phases (i.e. power MOSFETs, gate driver circuit, the inductor and the phase switching frequency) for the proposed system from Figs. 3.1 and 3.2, are selected to obtain high efficiency

53 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 39 based on the practical converter design guidelines given in [25]. In particular, the power MOSFETs and gate driver circuits are chosen such that each phase operates at high efficiency around its nominal operating current as illustrated in Fig. 3.7.a). To improve the overall efficiency, the switching frequency of the largest phase is reduced to be twice lower than the phase operating at 2I nom. On the other hand, to provide fast transient response, the variable and I nom phase are configured to operate at the switching frequency 1.3 times faster than 2I nom phase. The resulting efficiency curve of the complete system is shown in Fig. 3.7.b). It can be seen that the resultant efficiency characteristic is improved compared to individual phase efficiency characteristics and it is nearly constant for most of the load current range allowing for significant energy loss reductions especially at light and medium load currents. The efficiency simulations, presented in Fig. 3.7 are performed using loss models, in which detailed switching and conduction losses of the power stages, as well as those of the gate drive circuits are included. The models are developed based on the practical guidelines provided in [25]. The non-identical switching frequencies and non-uniform rated currents of the converter phases impose a larger overall current ripple injected into the output capacitor. To provide small output voltage ripple, a larger output capacitor is required. However, in portable applications, the additional cost/size associated with the increased output capacitor can be compensated with the Figure 3.7: a) Simulated efficiency characteristics of the converter phases operating at different switching frequencies b) Simulated efficiency of the whole system.

54 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 40 reduced battery size due to improved power supply efficiency. In practice, the number of phases, their rated nominal currents, and switching frequencies can be further customized to suit a particular application and expected load behavior. 3.6 Experimental System and Results Based on the block diagram of Fig. 3.2 an experimental 4-phase 40-A, 12 V-to-1.8 V system was built. All digital blocks (PI compensator, hysteretic logic, phase enable corrector and transient current estimator) are implemented with an Altera FPGA DE2 board as well as the DAC that also uses an external reference and an RC filter. The main characteristics of the prototype buck converter are listed in Table 3.1. The nominal current I nom is set to be 5 A, and the switching frequencies of all stages but the two largest ones are the same. Table 3.1: Important parameters of the 4-phase logarithmic buck converter phase current [A] f sw [khz] L [μh] I var I nom I nom I nom C [μf] Steady-State Operation The steady-state converter operation for two different output load current is shown in Figs. 3.8.a) Figure 3.8: Steady-state operation with two different load currents: a) 39 A b) 12 A; Ch.1: Output voltage (500 mv/div); Ch.2, Ch.3 and Ch.4: inductor current of variable phase, phase 2I nom, phase 4 I nom respectively 6.66 A/V; D1-D4: Gate drive signals of all four phases. Time scale is 2 μs/div.

55 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 41 and 3.8.b). Fig. 3.8.a) shows the case when the output load current is 39 A, and all the phases are turned on (4I nom = 20 A, 2I nom = 10 A, I nom = 5 A, and the current of the variable phase is I var = 4 A). The other case shows the situation when the load current is reduced to 12 A. To improve the efficiency, the controller now turns off the phases 2I nom and 4I nom. A tight output voltage regulation in both cases can be observed due to the operation of the variable phase that dynamically changes to accommodate the difference between the load current and other phase currents Transient Operation To test the dynamic behavior of the proposed system and verify proper operation of the phase enable corrector and the transient current estimator, both blocks are intentionally disabled first. Then the output load is stepped down from 21 A to 19 A. The controller response is shown in Fig. 3.9.a). As it is predicted in Section 3.4, due to the inappropriate transition in the inductor currents, the output voltage exhibits a large overshoot of 150 mv although the load change is only 2 A. In the next step, the phase enable corrector and the transient current estimator are enabled and the controller response is tested again. The voltage deviation is now reduced by a factor of 6 to 25 mv due to the operation of these blocks as shown in Fig. 3.9.b). The subsequent phase reconfigurations are also prevented by the operation of hysteretic logic. Figure 3.9: Transient response for a load step from 21 A to 19 A when the phase enable corrector and transient current estimator are: a) disabled b) enabled; Ch.1: Output voltage (200 mv/div); Ch.2, Ch.3 and Ch.4: inductor current of variable phase, phase 2I nom, and phase 4 I nom respectively 6.66 A/V; D1-D4: Gate drive signals of all four phases. Time scale is 10 μs/div.

56 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 42 Figure 3.10: Transient response: a) from 9 A to 38 A; b) from 38 A to 9 A; Ch.1: Output voltage (200 mv/div); Ch.2, Ch.3 and Ch.4: inductor current of variable phase, phase 2I nom, and phase 4 I nom respectively 6.66 A/V. Finally, the converter operation is also verified with a large light-to-heavy (9 A 38 A) and heavy-to-light load step (38 A 9 A) as shown in Figs a) and 3.10.b) respectively. During the load transient, since the total current command i tot [n] is dynamically changing to accommodate new load conditions, the phases are turned on and turned off until the output voltage settles down and i tot [n] becomes constant. The response time is around 30 μs while the maximum voltage deviation is 200 mv. Fast response of variable and I nom phases due to the higher switching frequency helps also to reduce the peak of the voltage drop at the output capacitor during the load transient Measured Converter Efficiency Fig a) shows the measured efficiency of the converter with logarithmic current sharing and that of an equivalent 4-phase power stage with the uniform current distribution. It can be seen that the converter with logarithmic current sharing has a virtually flat characteristic over the analyzed range of operation and results in an efficiency improvement of up to 25% at light loads, and about 6% at medium loads. Compared to a 4-phase converter, switching at 500 khz and utilizing a phase shadowing method [10] for efficiency optimization, the converter with the logarithmic current sharing provides on average a 1.5% efficiency improvement as illustrated in Fig b) when the output load current is slowly varying. As the frequency of load disturbance changes increases, the efficiency improvement for the logarithmic current sharing method increases due to faster selection of the optimum number of the active converter phases.

57 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 43 Figure 3.11: Measured efficiency characteristics of the converter based on: a) logarithmic current sharing and uniform current sharing b) logarithmic current sharing and phase shadowing. 3.7 Conclusion In this chapter, a current-mode based digital controller and power stage with logarithmic current sharing are introduced. The controller dynamically configures the number of active phases to improve the efficiency of the converter. It is shown that by using a relatively simple controller and a small number of phases a virtually flat efficiency characteristic can be achieved. The digital controller requires no analog-to-digital converters for inductor current measurements and, as such, is suitable for systems operating at high switching frequencies. In addition, due to the use of the current-program mode, potential stability problems that exist for similar voltage mode configurations are eliminated. Quantization effects and phase toggling problems existing due to coarse current adjustments are addressed and resolved. The new system is implemented on a 4- phase prototype and efficiency improvements and good dynamic characteristics are verified. REFERENCES [1] Y. Panov, M. Jovanović, Design considerations for 12-V/1.5-V, 50-A voltage regulator modules, IEEE Trans. Power Electron., vol. 16, pp , Nov [2] Peng Xu, Jia Wei, F.C. Lee, Multiphase coupled-buck converter-a novel high efficient 12 V voltage regulator module, IEEE Trans. Power Electron., vol. 18, pp , Jan 2003.

58 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 44 [3] Z. Xunwei, M. Donati, L. Amoroso, F.C. Lee, Improved light-load efficiency for synchronous rectifier voltage regulator module, IEEE Trans. Power Electron., vol. 15, pp , Sept [4] X. Zhou, P.-L.Wong, P. Xu, F. C. Lee, A. Q. Huang, Investigation of candidate VRM topologies for future microprocessors, IEEE Trans. Power Electron., vol. 15, pp , Nov [5] S. Luo, Z. Ye, R. Lin, F. C. Lee, A classification and evaluation of paralleling methods for power supply modules, in Proc. IEEE Power Electronics Specialists Conf., vol. 2, 1999, pp [6] Jung-Won Kim, Hang-Seok Choi, Bo Hyung Cho, A novel droop method for converter parallel operation, IEEE Trans. Power Electron., vol. 19, pp , Jan [7] J. Abu-Qahouq, Hong Mao, I. Batarseh, Multiphase voltage-mode hysteretic controlled DC-DC converter with novel current sharing, IEEE Trans. Power Electron., vol. 19, pp , Nov [8] Hong Mao, Liangbin Yao, Caisheng Wang, I. Batarseh, Analysis of inductor current sharing in nonisolated and isolated multiphase DC DC converters, IEEE Trans. Power Electron., vol. 54, pp , Dec [9] J.A. Oliver, P. Zumel, O. Garcia, J.A. Cobos, J. Uceda, Passive component analysis in interleaved buck converters, in Proc. IEEE Applied Power Electronics Conf., 2004, pp [10] P. Zumel, C. Fernandez, A. de Castro, O. Garcia, Efficiency improvement in multiphase converter by changing dynamically the number of phases, in Proc. IEEE Power Electronics Specialists Conf., 2006, pp [11] J.A. Abu Qahoug, L. Huang, Highly efficient VRM for wide load range with dynamic non-uniform current sharing, in Proc. IEEE Applied Power Electronics Conf., 2007, pp [12] Wenkai Wu, Multiphase voltage regulator control and design Considerations, in Professional Education Seminars Workbook, IEEE Applied Power Electronics Conf., 2009, vol. 2, seminars 7-12, pp [13] J.M. Rivas, R.S. Wahby, J.S. Shafran, D.J. Perreault, New architectures for radiofrequency DC DC power conversion, IEEE Transactions on Power Electronics, vol. 21, pp , March [14] H. Peng, D. Maksimović, "Digital current-mode controller for DC-DC converters," in Proc. IEEE Applied Power Electronics Conf., 2005, pp [15] J. Chen, A. Prodić, R.W. Erickson, D. Maksimović, Predictive digital current programmed control, IEEE Transactions on Power Electronics, vol. 18, pp , Jan

59 CHAPTER 3. DYNAMIC EFFICIENCY OPTIMIZATION FOR MULTI-PHASE CONVERTERS 45 [16] S. Chattopadhyay, S. Das, A digital current-mode control technique for DC-DC converters, IEEE Transactions on Power Electronics, vol. 21, pp , Nov [17] R.W. Erickson and D. Maksimović, Fundamentals of Power Electronics, 2nd edition. Boston, MA: Kluwer [18] Y. Kaiwei, High-frequency and high-performance VRM design for the next generations of processors, Ph.D. thesis, Virginia Polytechnic Institute and State University, April [19] Y. Kaiwei, Q. Yang, X. Ming, F.C. Lee, F.C. Lee, A novel winding-coupled buck converter for high-frequency, high-step-down DC-DC conversion, IEEE Transactions on Power Electronics, vol. 21, pp , Nov [20] PIP212-12M data sheet, NXP Semiconductors, Eindhoven, Netherlands. [21] C. Deisch, Simple switching control method changes power converter into a current source, in Proc. IEEE Power Electronics Specialist Conf., 1978, pp [22] A. Peterchev, S. Sanders, Quantization resolution and limit cycling in digital controlled PWM converters, in Proc. IEEE Power Electronics Specialists Conference, vol. 2, 2001, pp [23] O. Trescases, Z. Lukić, W. T. Ng, and A. Prodić, A low-power mixed- signal currentmode DC-DC converter using a one-bit Delta-Sigma DAC, in Proc. IEEE Applied Power Electronics Conf., 2006, pp [24] A. Parayandeh, A. Prodić, Programmable analog-to-digital converter for low-power DC DC SMPS, IEEE Transactions on Power Electronics, vol. 23, pp , Jan [25] AN-6005 Synchronous buck MOSFET loss calculations with Excel model, Fairchild Semiconductors, CA, USA,

60 Chapter 4 Reconfigurable Digital Controllers with Multi-Parameter Estimation 4.1 Introduction Power management systems used for low-power electronic devices require dedicated controller circuits that provide protection from overload conditions [1], increase energy efficiency through multi-mode operation [2,3] and obtain fast dynamic response [4]-[18]. For multi-phase converters, controllers are also needed to ensure proper current sharing [19]-[25] between the converter phases. Even if all phases are comprised of the same components, mismatches in their actual values can result in serious problems. Some of the phases could take on significantly larger currents than others, resulting in current-stress related system failures [1] due to overheating and uneven component stress. To implement these critical features, actual converter parameters need to be identified [26]-[28] and inductor currents need to be measured and actively regulated to maintain safe operation. Generally, the current measurement methods for dc-dc converters can be categorized into voltage drop and observer based methods. In voltage drop based methods, a current passing through a sense-resistor or a MOSFET is extracted from the voltage drop it causes [29]-[33]. The observer-based systems usually estimate current from the voltage across the power stage inductor [34,35]. In most cases, the existing methods are not well-suited for integration with the rapidly emerging digital controllers of switch-mode power supplies (SMPS) for battery-powered electronic devices, where the overall size, the system cost, and the overall efficiency are among the main concerns [2,3]. The voltage drop methods either decrease efficiency of the converter [29] or require a wide-bandwidth amplifier, which is very challenging to implement in the latest CMOS digital processes. This is due to very limited supply voltages of standard digital circuits (often below 1 V), at which traditional analog architectures cannot be used. Hence, bulkier and less reliable multi-chip solutions are needed for such architectures, requiring a sensing circuit and controller implemented in different IC technologies. On the other hand, the observer based 46

61 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 47 methods suffer from limited accuracy [36]. The current estimation relies on prior knowledge of the inductance and equivalent series resistance values [35,36], which depend on operating conditions and change under external influences. In addition to the current sensing methods described above, temperature monitoring of switching components can be used to further improve system reliability. Dedicated temperature sensors [37,38] can be placed next to the power switches as parts of the over temperature protection system. They send information to the thermal protection circuit of the controller, which, in the case of overheating, shuts down the SMPS. Again, the implementation of such a system requires a multi-chip solution and is therefore unsuitable for low-power systems. Hence, though very useful, the temperature protection of switching components is seldom applied. The first goal of this chapter is to introduce a combined digital current and temperature estimator, shown in Fig. 4.1.a), which is suitable for single-chip integration with modern lowpower digital controllers regulating the operation of single-phase and multi-phase converters. The self-tuning current estimator, from Fig. 4.1.a), utilizes the flexibility of digital implementation to compensate for the changes in the inductor parameters and achieve accurate Figure 4.1: a) A digitally-controlled buck converter with the self-tuning current estimator and remote temperature monitoring system b) A dual-phase buck converter regulated by the non-linear multi-phase digital average currentmode controller based on the multi-phase self-tuning current estimator.

62 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 48 current estimation. In addition to a simple digital IIR filter and a load step circuit, the current estimator only requires a slow analog to-digital converter for the input voltage measurement. Therefore, a practical realization of the estimator results in a modest increase in digital controller complexity. Additionally, from the converter parameters identified during the current estimator calibration, such as the converter equivalent resistance R eq that inherently includes the resistance of the switching components, the temperature of the switches can be estimated and remotely monitored in real-time to prevent system failures due to overheating, as shown in Fig. 4.1.a). Both the current estimator and remote temperature monitoring system are also fully implementable in the latest digital CMOS technologies allowing for a simple integration with the upcoming digital controllers. The second goal of this chapter is to propose a non-linear average current-mode digital controller, shown in Fig. 4.1.b) that utilizes information about the estimated phase currents and identified converter parameters (phase inductances and output capacitance) to achieve fast transient response and dynamic current sharing. Recent digital and analog controllers [7], [12]- [16], based on a capacitor charge-balance principle [7], are designed to provide superior timeoptimal response of single-phase converters. Unfortunately, their control actions and output voltage response are sensitive to component tolerances and require both fast and accurate detection of the voltage valley point. To effectively resolve previous implementation challenges, the new two-step transient algorithm is introduced. The algorithm takes into account component mismatches, eliminates the need for fast valley-point detection, and provides a bumpless transition between two steady states by also considering the inductor current ripple. Uneven aging and thermal stress of converter phases due the parasitic parameter mismatches are prevented by dynamically adjusting the current distribution ratio based on the identified information about phase equivalent resistances R eq. As a result, dominant phase conduction losses at heavy load and heat dissipation are dynamically equalized between all phases. In the following section the basics of the inductor voltage based current estimation are briefly reviewed and the principle of the operation of the self-tuning digital current estimator is explained.

63 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Principle of Current-Estimator Operation Figures 4.2.a) and 4.2.b) explain the principle of operation of the conventional analog current estimator [35] and the self-tuning digital system introduced in this paper, respectively. In the analog implementation of Fig. 2.a), the inductor current i L (t) is extracted by placing an R-C filter in parallel with the power stage inductor and measuring the filter s capacitor voltage v sense (t). The relationship between the capacitor voltage and the inductor current is given by the following transfer function: V sense L 1+ s R 1+ s τ = L L L L (4.1) 1+ s R C 1+ s τ L L ( s) I ( s) R = I ( s) R, f f f where L and R L are the inductance and its equivalent series resistance values, respectively, and R f and C f are the values of the filter components. When the filter parameters are selected so that τ f = R f C f = L/R L = τ L, the capacitor voltage becomes an undistorted scaled version of the inductor current (the zero and pole cancel each other). This allows the inductor current to be reconstructed from the capacitor voltage measurements. As mentioned earlier, the main drawback of this method is that the inductor parameters are not exactly known and change over time, often causing large errors in the estimation [36]. To compensate for these variations, an analog filter Figure 4.2: Current sensing techniques: a) conventional analog implementation b) implementation with a self-tuning digital filter.

64 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 50 with programmable resistive networks is proposed in [39,40] where, in the latter publication, an on-chip implementation of the filter is shown. Even though the method significantly improves the estimator accuracy, its implementation still requires a relatively large number of analog components and the sensed current is still an analog signal, making it less suitable for integration with digital controllers in low-power SMPS. In the new estimator of Fig. 4.2.b) the analog filter is replaced with a fully-digital and tunable equivalent. In this implementation, the voltage across the inductor is converted into a digital value v L [n] and then processed in the digital domain, to extract the output value i sense [n], which is directly proportional to the inductor current. By manipulating (4.1) and applying a bilinear transformation, the simple difference equation suitable for the practical digital filter implementation can be derived: i vsense[ n] 1 n] = = { c1 RL isense[ n 1] + c ( vl[ n] + vl[ n 1]) }, R R sense [ 2 (4.2) L L where c 1 and c 2 are filter coefficients: c 1 L L = (2 1) /(1 + 2 ), (4.3) R T R T L s L L 1 c 2 = (1 + 2 R ), (4.4) L T s s and T s is the filter sampling rate. The estimator adjusts the filter gain factor 1/R L from (4.2) and coefficients c 1 and c 2 through a self-calibrating process. This is obtained with the help of a test current sink connected at the converter output, as shown in Figs. 4.1 and 4.3. Periodically, the sink implemented with a known resistor and a small switch connected parallel to the load, turns on for a short time. Then, based on the response of the filter, the Gain/τ Calibration Logic block adjusts the filter gain and coefficients so that the increase in i sense [n] corresponds to the exact increase in the load current. The calibration technique introduced here is similar to that in [40] where a known test current is injected into the inductor during the converter start-up phase for the purpose of calibrating an analog filter for current estimation. However, the limitation

65 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 51 Figure 4.3: Test current sink circuit used for the filter calibration. of this method is that it cannot be used during regular converter operation because it requires both converter switches to be turned off during the filter calibration phase. Since the series inductor resistance R L and inductance L dynamically change, due to variations of converter operating conditions (e.g. output load current or temperature), the accuracy of the current estimation might be compromised. Since the current sink of Fig. 4.3 does not require any change in converter operation, the method presented here can be used during normal converter operation and the calibration can be performed regularly. 4.3 Practical Implementation of the Current Estimator Even though the principle of operation of the self-tuning digital estimator is relatively simple, its practical implementation is not trivial. It potentially requires a very fast ADC, with a sampling rate significantly higher than the switching frequency, as well as an equally fast processor for the filter implementation. These components threaten to make the estimator impractical for cost-sensitive, low-power applications. The precision and speed of the estimator depend on the accuracy of the measurement of the average inductor voltage value. Even a small inaccuracy in this measurement can cause a large estimation error. To obtain fast estimation, accurate measurement of the inductor voltage over one switching cycle is required. This could be done with an ADC whose sampling rate is much higher than the switching frequency. The need for a very high sampling rate converter is illustrated in Fig In this case, the average value of the inductor voltage can be calculated by summing the sampled voltage values and dividing the result by the number of samples taken during one switching cycle. However, as seen in Fig. 4.4, the accuracy of this approach strongly

66 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 52 depends on the number of samples taken (i.e. the speed of the ADC), especially if the samples are not perfectly aligned with the inductor voltage transition point. Figure 4.4: Signals of a filter implemented with an over-sampling ADC. To alleviate this dependency, the input voltage of the power stage v g (t) is sampled at a rate lower than the switching frequency and the average value of the inductor voltage is calculated as: v [ n] = d[ n] v [ n] v [ n], (4.5) L ave where d[n] is the DPWM s control variable and v out [n] is the converter output voltage, both of which are readily available in the control loop of Fig The v out [n] value is provided by the already existing ADC of the voltage control loop and the duty ratio is provided by the digital PID compensator. A lower sampling rate is possible because in targeted battery-powered applications the input voltage often changes in a very slow fashion. It should be noted that the complete implementation of the new estimator, including ADC-s, is possible in standard CMOS processes. Recent publications [2], [3], [41], [42] show application-specific ADCs for SMPS that are implemented in the latest low-voltage CMOS technologies Filter Architecture and Self-Calibration in out The calculation of the average voltage described in the previous subsection reduces hardware requirements but concurrently affects the estimation accuracy. The actual average inductor voltage might differ from (4.5), due to the action of non-overlapping, i.e. dead-time circuit, and other parasitic effects. To compensate for this effect, as well as for the previously mentioned

67 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 53 variations in the inductor values, a current sink and calibration logic (Fig. 4.1) are used to tune the parameters of the infinite impulse response (IIR) digital filter shown in detail in Fig The calibration of the filter coefficients described with (4.2) and other parameters is performed in three phases. In the first phase, a known load current step is introduced by enabling the current sink and the accurate value of the filter gain 1/R L is found from the variation in the estimated inductor current. In the second phase, the current sink is disabled and the time constant τ L = L/R L, determining coefficients c 1 and c 2, is calculated from the estimator output overshoot/undershoot during the load step-down transient. Finally, in the third phase, any offset in the estimated current due to the action of the dead-time circuitry is removed by temporarily changing the switching frequency and monitoring the response of the estimator. A more detailed description of the calibration procedure is given in the following subsections. Figure 4.5: The architecture of the tunable IIR filter used in the current estimator Filter Gain Calibration Procedure The Calibration Logic block, shown in Fig. 4.1, periodically performs the calibration procedure. The correct value of filter gain 1/R L is determined from the difference between two steady state current values, estimated before and after a load step is applied as shown in Fig The initial steady state is detected by monitoring the error signal e[n] and at the time instant A (Fig. 4.6), the current before the transient is estimated as I 1 =i sense [n] and stored in a register. After a step ΔI is introduced and steady state is reached again, the new current I 2 = i sense [n] is found and the difference, ΔI m =I 2 I 1, (4.6) is calculated. This value is then compared to the actual value of the current step and the actual

68 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 54 Figure 4.6: Filter gain calibration procedure: simulated response of the current estimator during output load change for two cases (bottom curve) the initial value of R L is twice the actual value (top curve) after the filter adjustment. value of R L, named as R L_calibrated, is calculated as: R L _ calibrated I m = RL _ inital, (4.7) I where R L_inital is the initially set resistance value. The filter gain value is then adjusted as: 1 filter gain =. (4.8) R L _ calibrated The value of R L_calibrated includes not only the actual inductor DCR resistance, but also all parasitic resistances that are in the inductor current path between the input voltage and output voltage sampling instance: parasitic drain-source on-resistances of switching devices and PCB trace resistances. This is illustrated in Fig. 4.7.b) showing the steady-state dc equivalent circuit derived from the converter shown in Fig. 4.7.a). From Fig. 4.7.b), based on Ohm s law, the correct filter gain that, in steady state, multiplies the average inductor voltage v L- ave[n]=d[n] v in [n]-v out [n] to obtain the inductor current, is then: filter 1 gain =, R + d[ n] R + (1 d ) R + R (4.9) L [ n] ds1_ on ds2 _ on traces which is also calculated based on (4.8) as 1/R L_calibrated.

69 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 55 Figure 4.7: The buck converter (a) and its steady-state dc equivalent circuit R eq (b) used for current estimation Filter Time Constant Calibration Procedure The uncertainties of an actual inductor value L and its parasitic series resistance R L affect the time constant τ f (4.1) and therefore the time response of the filter. This effect is demonstrated in Fig. 4.8 showing the actual and estimated inductor currents during a load step for different time constants: the actual time constant τ L, and 50% of this value. It can be seen that the estimated current accurately follows i L (t) only when filter coefficients for the actual inductor value L and R L are properly set. In other cases, the estimated current exhibits either undershoot or overshoot as illustrated in Fig The calibration of τ f is performed during the transient, at the output voltage peak point (time instant B in Fig. 4.8), where the inductor current is equal to that of the load current [12]. At this time instant, the estimated current is compared with the expected value I 1 -ΔI and the difference, ΔI peak, is calculated: (4.10) Based on ΔI peak, the initially selected filter time constant τ f, and the measured time interval ΔT peak between the moment when the sink circuit is disabled and the peak point is detected, the actual inductor time constant τ L is obtained:

70 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 56 (4.11) For instance, for the simulation result shown in Fig. 4.8, τ L is selected to be equal to 2τ f and the filter response exhibits a large overshoot when the sink circuit is disabled. By using the parameters ΔI peak =0.68 A and ΔT peak =22 μs, dynamically obtained from the estimator response, for the τ f =32.6 μs and sink step size ΔI of 1 A, the inductor time constant τ L according to (4.11) is calculated to be equal to 2.02τ f. As a result, it is possible to adjust the new filter time constant τ f based on the calculated τ L from a single action of the sink calibration circuit. Once the correct τ L =L/R L is known, filter coefficients c 1 and c 2 are adjusted according to (4.3) and (4.4) respectively. The filter time constant tuning procedure presented here is significantly faster than previous iterative-based procedures demonstrated in [39], [40], [43]. This allows the current estimator to be utilized as a reliable solution for over-current protection and monitoring even if the power stage components have large tolerances. The detailed derivation of (4.11) is given in Appendix A. Figure 4.8: Calibration of the time constant: simulated response of the current estimator during output load change between 2 A and 2.5A; (bottom) for τ f =0.5τ L ; (top) for τ f =τ L.

71 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Offset Calibration Procedure Due to the operation of the dead-time circuit and unbalanced timing delays of gate driver circuitry driving parasitic capacitances of power MOSFETs, the duty-ratio value of nonoverlapping control signals is different than the duty ratio command d[n] calculated by the PID compensator from Fig Although the output voltage regulation is not affected since the compensator automatically adjusts d[n] to compensate for Δd introduced by the dead-time circuit, the current estimator produces an incorrect estimate of the inductor current as shown in Fig The difference between the current estimation and inductor current is caused by the incorrectly calculated average inductor voltage v L-ave [n] from d[n] according to (4.5). From Fig. 4.9 it can also be observed that the actual inductor voltage obtained as a difference of the voltage at the switching node v Lx (t) and the output voltage v out (t) does not have an ideal square-wave shape shown in Fig The presence of voltage ringing at v Lx (t), caused by converter parasitics, influences the actual average inductor voltage calculated by the estimator from (4.5). For instance, in steady state, with a duty ratio mismatch Δd added to the true duty ratio value, the estimated current becomes equal to: i sense v [ n] = L true ave[ n] ( d [ n] + d) vin[ n] vout[ n] =. (4.12) R R L L If we rearrange (4.12), an offset in the estimated current becomes proportional to the duty Figure 4.9: Offset in the estimated current i sense [n] caused by the operation of the dead-time circuit and converter.

72 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 58 ratio mismatch Δd: i sense true d [ n] vin[ n] vout[ n] d vin[ n] [ n] = +. (4.13) R R L L To eliminate the offset the following procedure is applied. By monitoring the error signal e[n], the converter steady state is first detected and current sample I 1 of the estimated current is taken. In the next step, the switching frequency f sw of the DPWM is increased by k-times ( f sw = k f sw ) and another sample I 2 is taken once the steady state is reached again. Since the duty ratio mismatch Δd is proportional to the switching frequency: t d = = t f sw, (4.14) T sw where Δt is a constant time interval caused by the dead-time circuit, from (4.13) and (4.14), I 2 becomes larger than I 1. Based on the difference between I 2 and I 1, the current offset from (4.13) and (4.14) can be calculated as: d vin [ n] I 2 I1 offset = =. (4.15) R k 1 L In applications, where the input voltage may change significantly between two offset calibrations, a sample of v in [n] (i.e. V in_offset ) can be also taken during the calibration and then used to additionally scale the offset value by a factor v in [n]/ V in_offset. This scaling dynamically compensates for the offset deviation proportional to v in [n] as predicted by (4.13) and (4.15). The estimated current is then corrected by subtracting the final offset value in the digital filter architecture as shown in Fig Steady-State Current Estimator Architecture Knowledge about the inductor current in steady state is sufficient for significantly improving system performance in many applications. For instance, to extend the battery life of portable systems, multimode operation can be performed by monitoring the steady-state inductor current. Examples of this include on-line switching between pulse-width and pulse-frequency modes of operation [2], [44], dynamic sizing of power switches [45], and gate-drive circuit adjustments

73 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 59 [45]. For these cases, several parts of the current estimator system architecture of Fig. 4.1 can be further simplified to obtain a steady-state current estimator [46,47] reducing system hardware requirements and cost. For example, the programmable digital IIR filter calculating current estimate i sense [n] from Fig. 4.5 can be optimized for operation where only the steady-state current is required. Consequently, the low-pass filtering logic shaping the response of the estimated current during load transients is now removed. The average inductor voltage v L-ave [n] is fed directly to the filter gain multiplication and offset adjustment. Since filter coefficients c 1 and c 2 are not used anymore, externally programmable look-up tables that keep their pre-calculated values are also eliminated. Also, the calibration logic can be simplified by removing the filter time-constant procedure that selects the proper set of filter coefficients. Figure 4.10: The implementation of the steady-state current estimator. 4.4 Principle of Temperature Monitoring Operation In the previous section, it is demonstrated that resistance R L_calibrated obtained during the gain calibration procedure for the current estimator is represented by a sum of all parasitic resistances in the inductor current path between the input voltage and output voltage sampling instance: [ n] ) R. R. R L _ calibrated = RL + d[ n] Rds 1_ on + (1 d ds2 _ on + traces (4.16) According to (4.16), the parasitic on-resistances R ds1_on and R ds2_on of the converter power switches are dominant parts of R L_calibrated. In addition to its original purpose of providing accurate current estimation, the extracted value of R L_calibrated can now be used for monitoring the temperature of the power switches and for overheating protection. As illustrated in Fig.11, R L_calibrated depends on the temperature, meaning that, for a given operating condition, the temperature of the switching components can be found from R L_calibrated and the load value.

74 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 60 As it will be shown later, a fairly accurate estimate of the temperature can be obtained by using a look up table with pre-stored temperatures for a range of R L_calibrated and the load current values. In Fig. 4.1, the look-up table is part of the temperature monitoring block. Overly high temperature in the components causes R L_calibrated to exceed a certain threshold value R thresh, at which point the block shuts down the switches and also sends an external overheat signal. Figure 4.11: Change of R L_calibrated due to the ambient temperature for different output load current conditions. 4.5 Inductor and Output Capacitor Identification The system shown in Fig. 4.1.a), with little additional hardware, is also capable of providing the controller with continuous information about key converter parameters: the sizes of the inductor and output capacitor. These parameters can be further utilized to optimize the controller response and implement power supply health monitoring [48]. To reduce the required hardware for the L and C parameter identification, parameters obtained during the filter calibration are simply reused for this purpose.

75 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Inductor Identification The inductor value L is estimated from the inductor time constant τ L and resistance R L_calibrated. Both τ L and R L_calibrated are already found during the filter gain and time constant calibration procedure described in Section 4.3. The inductance value L is then simply calculated from their product: L =τ L R L _ calibrated. (4.17) Output Capacitor Identification The actual size of the output capacitor is estimated during the filter time constant calibration when the output load current is stepped down by disabling the current sink as shown in Fig The size of the output voltage deviation ΔV peak caused by the sink action is obtained from the voltage error signal e[n] at the peak point: V = e V (4.18) peak peak q _ ADC. where ΔV q_adc is the quantization step size of the output ADC. Additional electric charge ΔQ injected into the output capacitor C from the current difference between the inductor and load current is proportional to the generated output voltage deviation ΔV peak : Q = C V peak. (4.19) Alternatively, if we approximate that the inductor current i L (t) linearly decreases with the slope ΔI/ΔT peak as shown in Fig. 4.8, the charge ΔQ during ΔT peak can be expressed in terms of the sink step size ΔI and time interval ΔT peak as: 1 Q = I T peak. (4.20) 2 By combining (4.19) and (4.20), the capacitor value C is estimated as: I Tpeak C =, (4.21) 2 V peak

76 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 62 under the assumption that its ESR value is small. For example, for the waveform shown in Fig. 4.8, the output capacitor used for simulation is 200 μf. According to the waveform parameters ΔT peak = 22 μs and ΔV peak = 54 mv for a sink step size ΔI of 1 A, the estimated capacitance is equal to μf. 4.6 Oversampled Non-Linear Average Current-Mode Controller for Multi-Phase Converters Unlike the voltage-mode controller from Fig. 4.1.a), the non-linear average current-mode controller regulates both the output voltage and phase inductor currents by generating references for the phase currents. As shown in Fig. 4.1.b), the new architecture is fully-digital, except for the analog-to-digital converters (ADC) and the current sink. The multiple current-sensing circuits are replaced with the multi-phase current estimator and current sink, allowing for significant cost and size reduction of the multi-phase converter. Tight voltage regulation, fast transient response and dynamic current sharing are obtained by switching between two modes of operation used for steady-state and during large load transients. The new system is comprised of a PI compensator, current sharing logic, current-compensator loops based on the multi-phase current estimator, and a transient compensator. To enable rapid detection of sudden load disturbances and produce a fast reaction of the controller, the output voltage ADC samples eight times per regular switching cycle. The operation of the controller and its key blocks are described in the following subsections Steady-State Mode For light-load changes and in steady state, the controller operates as follows. Based on the error between the output voltage and its reference, e[n], a PI compensator creates i tot [n], a digital value proportional to the total current of all phases, i.e. to the load current. The value i tot [n], is then passed to the current sharing logic that sets digital references i refi [n] for the currents of all phases (i=1,, N). To calculate the phase duty-ratio control signal d i [n] such that the average of the phase currents follows the reference i refi [n], the current compensators apply an average-current dead-beat algorithm presented in [4], [49]: d i ( ) [ n] D + k i [ n] i [ n] =, (4.22) i refi sensei

77 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 63 where D is the nominal steady-state duty ratio value, obtained as a ratio of the output voltage reference V ref and input voltage v in [n]. The gain coefficient k i is proportional to the estimated phase inductance L i and is dynamically adjusted as v in [n] changes according to: k i = L f i 0.5 sw. (4.23) vin[ n] Vref Multi-Phase Current Estimator Tuning and Phase Identification The calibration process used for the single-phase topologies cannot be directly applied for multiphase converters regulated by the voltage-mode controllers. For the single-phase converter, the load step introduced by the current sink must be equal to the inductor current. In multi-phase converters, this is not the case. The current sink step can be shared between the phases in many different ways, depending on the mismatch in power stage component values. To solve this problem, the average current-mode controller is used and a phase-by-phase calibration process, based on freezing the current of all phases but one, is implemented. As a result only the current in the active phase increases and the increment is equal to that of the test current sink allowing for proper current estimator calibration, as described in Section 4.3, for that specific phase and accurate identification of parameters such as phase equivalent resistance R eqi, where i =1,,N. The inherent capability of the current-mode controller from Fig. 4.1.b) to freeze the average phase currents presents the possibility for a simplified capacitor/inductor identification algorithm that does not require peak point detection. To eliminate errors in the identified parameters due to periodic noise introduced by interleaved switching of the converter phases, the output capacitor and inductances are also estimated based on a difference of two voltage samples. The samples are taken at identical time instances during two switching cycles as illustrated in Fig so that the common noise is subtracted. In the first step, the current references are frozen and the current sink is disabled causing the output voltage to increase at a rate inversely proportional to the output capacitor value. Shortly after the sink is disabled, the first voltage sample is taken. The introduced delay eliminates the effect of the capacitor ESR on the capacitor identification. After n switching cycles T sw (e.g. n=2), the second voltage sample is then obtained. From the

78 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 64 accumulated voltage difference ΔV 1 between two samples, the output capacitor is now estimated as: C I n Tsw =, (4.24) V 1 where ΔI is the current sink current. In the second step, the current reference is reduced in the phase where inductor value estimation is desired. This causes the inductor current to decrease. After a delay of one switching cycle, the third sample is taken, and from the difference ΔV 2 the phase inductance is identified as: L i 2 2 Tsw Vout = C V + I T 2 sw. (4.25) Figure 4.12: Differential output capacitor and phase inductance identification Conduction Loss-Based Current Sharing Scheme The output load current is often equally shared [19,20] between converter phases consisting of components with identical nominal values and current ratings. Due to the enforced equal current sharing, parasitic resistance mismatches of power switches and board traces between phases produce higher conduction power losses in one of the phases. For instance, for modern highcurrent integrated converters [50,51] supplying up to 35 A of current per phase, a small mismatch of 1 mω generates more than 1 W extra power loss, or between 5% and 10% of total

79 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 65 conduction loss per phase at full load. The additional power loss generates more heat and increases switch temperature [50], as shown in Fig. 4.13, leading to faster aging of components and premature failure of one of the phases. To eliminate the temperature mismatch, the controllers that adjust phase currents based on the feedback from external temperature sensors mounted on the top of the power switches, were introduced in [52,53]. However, in practice, such controllers are still not widely accepted in high-volume, cost-sensitive low-power applications due to the increased component count, size, and circuit complexity. To resolve this problem and eliminate the need for external temperature sensors, the controller from Fig. 4.1.b) balances phase conduction losses based on the estimated phase resistances R eqi : R I = R I = = R I, (4.27) eq1 L1 eq2 L2 eqn LN obtained during the calibration of the multi-phase current estimator. According to (4.27), for a two-phase converter, shown in Fig. 4.1.b), the desired current sharing ratio is equal to: iref 2 Req 1 ratio = =. (4.28) i R ref 1 eq2 Figure 4.13: Mismatch in phase equivalent resistances generates unequal thermal stress on power switches. From the resistance-based current ratio, the current-sharing logic from Fig. 4.1.b), splits total current command i tot [n] to produce current references i ref1 [n] and i ref2 [n] as:

80 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Transient Mode i i ref 1 1 [ n] = itot [ n], (4.29) 1 + ratio [ n] i n] i [ ] ref 2 tot[ ref 1 n =. (4.30) The transient compensator block, shown in Fig. 4.1.b), detects large load current changes ΔI load and dynamically overrides the operation of the conventional PI compensator to provide much faster transient response and smaller output voltage deviation. It processes error values e[n] eight times per switching cycle, which is four times faster than the PI compensator. From the error difference between two consecutive samples e[n-1] and e[n], it calculates the slope of the output voltage ΔV slope and accordingly estimates ΔI load : Vslope I load = C, (4.31) T sample where the actual value of the output capacitor C is obtained during the current estimator calibration. If ΔI load surpasses a threshold (i.e. 1/3 of the maximum load current I load ), the transient compensator temporarily disables the PI compensator, enters the transient mode and executes a two-step control algorithm illustrated in Fig for a two-phase buck converter. The load change threshold is set to prevent triggering of the transient compensator by the inductor current ripple, switching noise, and more importantly to avoid frequent switch-on/off actions of the power stage for light load changes that would decrease the power stage efficiency. In the first step, depending on the sign of ΔI load the transient compensator decides whether to turn on or turn off all phases in order to equalize the sum of average phase currents with the load current. For a light-to-heavy load step, +ΔI load, the turn-on time t on is proportional to ΔI load and an equivalent inductance L eq = L 1 L 2 L N of all phases now connected in parallel: t on Leq = I load, (4.32) v v in out For a heavy-to-light load step, ΔI load, the turn-off time is calculated as:

81 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 67 t off Leq = ( I load ). (4.33) v out At the moment when all phases are switched on/off, the DPWM carrier value, used for the synchronized generation of all pulse-width modulated signals, is frozen until either t on or t off are finished. Once unfrozen, the original phase switching sequence resumes exactly where it was interrupted as shown in Fig The inductor currents are brought to an average level, accommodating the load current and the output voltage deviation is rectified. Before the first step of the transient algorithm is fully completed, the transient compensator first updates i tot [n] of the PI compensator by adding an increment of ± ΔI load to its integral portion. The estimated average phase currents i sensei [n] are also modified by adding the transient increments Δv sensei to the registers holding the sensed voltage v sensei [n]: ± I L load eq vsensei = Reqi. (4.34) N Li Once the voltage deviation is blocked, the transient compensator performs the second step that quickly brings the output voltage back to steady state. The compensator now takes a voltage sample, intended for the PI compensator, and measures the deviation ΔV dev at that point as shown in Fig The total capacitor charge that needs to be recovered to reach the new steady state is equal to: Q = C dev V dev. (4.35) To compensate for ΔQ dev and achieve a bumpless transition when the PI compensator is reactivated, the transient compensator increases the duty ratio of all phases to D+Δd i during the first switching cycle. In the following cycle, the duty cycles are dropped to D Δd i. As a result, an additional charge, ΔQ i, is injected by each phase but the phase average currents are positioned properly for the PI compensator to resume as shown in Fig The injected charge by each phase is proportional to Δd i : v in 2 Qi = di Tsw. (4.36) Li

82 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 68 If ΔQ dev is now shared between N phases, the required Δd i for each phase is equal to: d i = f 2 sw Li C V N v dev in. (4.37) Figure 4.14: Transient mode operation of the non-linear average current-mode controller.

83 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 69 To generate increments ±Δd i with minimal additional hardware, the current-loop compensators are utilized. They are produced by changing the current references by ± Δi refi. The phase injected charge ΔQ i can be expressed in terms of Δi refi from (4.22) and (4.36) as: v in 2 Qi = k irefi Tsw. (4.38) Li Finally, by substituting (4.23), and (4.35) to (4.38), Δi refi for each phase is equal to: 2 irefi = f sw (1 D) C V N dev. (4.39) If necessary, to avoid current saturation of the inductors, the introduced current reference step Δi refi can be reduced by extending it across several switching cycles n cycles as: i ' refi 1 = n cycles i refi. (4.40) Once the charge ΔQ dev is fully compensated, the transient compensator reactivates the PI compensator which then continues to regulate the output voltage until the next large load transient occurs Hardware Optimization and Controller On-Chip Complexity To achieve a cost effective on-chip implementation, the controller hardware from Fig. 4.1.b) is carefully optimized for multi-phase operation. Since all phase control signals are interleaved, the current estimator is also time multiplexed to reduce computation resources (logic multipliers and adders). The block diagram of the multi-phase current estimator is shown in Fig The same approach is applied to the current compensators which are implemented as a single timemultiplexed current compensator calculating duty cycles intermittently. Finally, to reduce the number of internal registers used by both blocks and share partial computation results, the transient compensator and current estimator tuning logic are merged. Table 4.1 shows the occupied area and number of digital cells for each digital block when implemented in TSMC 0.18-µm CMOS technology. All digital blocks of the controller, shown in Fig. 4.1.b) occupy

84 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION mm 2 of silicon area or equivalently 15,610 digital logic cells. Figure 4.15: A time-multiplexed N-phase digital current estimator. Table 4.1: The silicon area and gate count of the multi-phase controller when implemented in TSMC 0.18-µm CMOS process. Design Block area [mm 2 ] logic cell count PI compensator Current sharing logic Two-phase current estimator Current compensators Transient compensator & current estimator tuning logic Two-phase DPWM (8-bits resolution) Total (digital blocks) Experimental Systems and Results An experimental single-phase system was first built based on the diagrams shown in Figs. 4.1.a), 4.3, and 4.5. The power stage is a 15-W, 1.5-V buck converter (nominal output filter component values are L=1.5 μh and C=200 μf), switching at f sw =500 khz, with an input voltage range between 2 V and 6.5 V. The controller, digital filter, calibration logic, temperature and current monitoring logic are realized with an Altera DE2 FPGA board by Verilog HDL. Two external ADCs sampling at f sw and f sw /8 are used for output and input voltage measurements respectively. The resolution of the ADC sampling the output voltage is 4 mv, while the resolution of the ADC for the input voltage is 16 mv. The test current sink was set to produce a 1 A pulse current,

85 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 71 which is only 10% of the maximum output current. To visualize the operation of the estimator, its digital output was sent to a flash digital-to-analog converter (DAC) and the resulting analog signal was observed. Figure 4.16 shows the closed loop operation of the controller during load transients between 1.5 A and 4.5 A (ΔI = 3 A) and demonstrates the self-tuning process of the current estimator. In the first (uncalibrated) phase due to the mismatch in the converter parameters, the filter gain, time constant τ f, and offset are not properly adjusted and an error in the current estimation occurs (the step is wrongly recognized as a 2.3 A to 4.3 A transition - ΔI est = 2 A). In the second Figure 4.16: System operation Ch1: Output converter voltage (200mV/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; D0-D1- load step and sink enable signals; Time scale is 500 μs/div. (calibration) phase, a 1-A test current sink step is introduced and the filter gain, the filter coefficients defining time constant τ f, and offset are tuned accordingly. The third (calibrated) phase shows repeated load transient, where the average current is estimated accurately both in steady state and during the transient, verifying the effectiveness of the self-tuning filter and the estimator operation Filter Gain Calibration To correct the filter gain (34% lower than its correct value) causing the actual inductor current step of 3 A to be recognized as a 2-A step in the estimated current during the uncalibrated phase

86 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 72 in Fig. 4.16, the filter gain calibration procedure based on a current sink is applied. Fig shows explicitly the operation of the current estimator during the filter gain calibration when the current sink is enabled. By monitoring the voltage error signal e[n], the calibration logic detects the steady state (e[n]=0 for 8 consecutive cycles) before and after the current sink is enabled and measures the difference ΔI m in the estimated steady-state currents to be 0.66 A. From ΔI m and the known value of the current sink step size ΔI=1 A, according to (4.7) and (4.8), R L_calibrated is calculated and the filter gain is therefore updated as shown in Fig As a result, the estimated current is increased by approximately ΔI/ ΔI m 1.51 times. If a real load transient occurs during the calibration, the obtained parameters are rejected and the procedure is repeated again. Figure 4.17: The filter gain (1/R L_calibrated ) calibration procedure Ch1: Output converter voltage (100 mv/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; D0-D1- load step and sink enable signals; Time scale is 20 μs/div Filter Time Constant Calibration Initially selected filter time constant τ f = 25.5 μs during the uncalibrated phase is approximately half of the inductor time constant τ L =L/R L_calibrated =1.5 μh/ 30 mω =50 μs resulting in a large overshoot of the estimated current as shown in Fig Shortly after the filter gain is calibrated by enabling the current sink, the calibration of τ f is performed by disabling the current sink causing the inductor current to drop by the value of the sink step size ΔI = 1 A. The actual tuning process of the filter time constant τ f is demonstrated in Fig a). The calibration logic first measures the estimated steady-state current I 1 before the current sink is disabled. In the next step,

87 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 73 the sink enable signal is set to zero and due to the closed-loop operation the inductor current and its estimate start to decrease. Once the calibration logic detects the output voltage peak recovery point, meaning that the inductor current is equal to the new steady-state inductor current I 1 -ΔI, the difference ΔI peak between the estimated current and I 1 -ΔI is calculated to be 0.65 A according to (4.10). At the same time, the time period ΔT peak between disabling the sink circuit and the detection of the peak point is captured to be ΔT peak = 20 μs. Based on the initially selected τ f, and the parameters obtained from the estimator response ΔI peak for the sink step size of ΔI, the calibration logic calculates the inductor time constant τ L according to (4.11) to be 52 μs. Once the actual inductor constant is identified, the filter time constant is matched to τ L and the coefficients c 1 and c 2 are properly adjusted. Figure 4.18: The filter time calibration procedure: a) τ f =0.5τ L ; b) τ f τ L Ch1: Output converter voltage (100 mv/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; D0-D1- load step and sink enable signals; Time scale is 20 μs/div. As a result, the estimator now exhibits proper response even during the transient as shown in Figs b) and 4.16 (during the calibrated phase). From the parameters obtained during the filter time constant calibration (ΔT peak = 20 μs and ΔV peak = 75 mv), the size of the output ceramic capacitor is estimated to be 133 μf. The estimated capacitor value is 34% lower than its nominal data sheet value at zero bias voltage. This is caused by the DC bias effect [52], caused by the output voltage, that reduces the actual capacitance of ceramic capacitors.

88 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Offset Calibration Even though the filter gain and time constant are now properly adjusted, the current estimator still exhibits an offset of 2 A between the estimated and the true inductor current as shown in Fig To compensate for this offset, the calibration logic samples the estimated steady-state currents before and after increasing the switching frequency by two times (k=2). Based on (4.15), the estimated offset is calculated by the calibration logic to be 2 A. Shortly after the switching frequency is reduced back to the nominal value, the offset value is subtracted in the digital IIR filter and good matching between the estimated and actual inductor currents is obtained as demonstrated in Fig Figure 4.19: The filter offset calibration procedure Ch1: Output converter voltage (100 mv/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; Time scale is 20 μs/div Calibrated Operation Figs a) and 4.20.b) demonstrate fast operation of the estimator when the gain, time constant and offset are properly tuned. They compare actual and estimated inductor current during both light-to-heavy and heavy-to-light load changes between 1.5 A and 4.5 A. As it can be seen, the average value of the current over one switching cycle is accurately estimated without any significant delay, allowing the estimator to be used for overload protection and power stage efficiency optimization.

89 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 75 Figure 4.20: The estimated current during load steps between 1.5 A and 4.5 A Ch1: Output converter voltage (200 mv/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; Time scale is 10 μs/div Current Estimation with the Steady-State Current Estimator The operation of the simple steady-state current estimator based on Fig is experimentally verified in Fig by varying the output load current between 0.8 A and 5.1 A. In this case, the estimated current is calculated from the average inductor voltage with a direct multiplication by the filter gain without filtering with coefficients c 1 and c 2. As a result, the estimator provides accurate information about the inductor current only in steady state but its hardware Figure 4.21: The estimated current with a steady-state current estimator during the load step between 0.8 A and 5.1 A Ch1: Output converter voltage (200 mv/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i L [n] 2 A/V; Time scale is 50 μs/div.

90 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 76 implementation is much simpler since it does not require time constant calibration Temperature Monitoring and Protection Figs a) and 4.22.b) verify proper operation of the temperature estimator/protection block from Fig. 4.1.a). To test its operation, the converter power MOSFETs are intentionally heated up. As it can be seen from Fig a), this results in an increase of R L_calibrated (see Fig b) above the threshold resistance R thresh, and the triggering of the overheating protection signal that shuts down the converter. Figure 4.22: Thermal monitoring: a) at 45 C; b) at 105 C Ch1: Output converter voltage (1 V/div); Ch2: estimated steady-state current i sense [n] 2 A/V; Ch3: estimated R L_calibrated 1 mω/32.5 mv; Ch4: MOSFET temperature 25 ºC/V; D1-D2- sink enable and overheat signal. Time scale is 200 μs/div Overload Protection Simple overload protection of the converter circuitry can be obtained by comparing the output of the current estimator with a predefined digital current threshold. Once the estimated current exceeds the threshold value, to prevent converter damage, it is immediately turned off and the estimator stops its operation as shown in Fig The output load current is intentionally increased from 2 A to 7.5 A above the threshold of 7 A. Therefore the overload protection signal is activated and the converter is rapidly turned off.

91 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 77 Figure 4.23: Overload protection implemented with the current estimator Ch1: Output converter voltage (1 V/div); Ch2: actual inductor current i L (t) 2 A/V; Ch3: estimated average current i sense [n] 2 A/V; Time scale is 20 μs/div Multi-Phase Operation with the Non-Linear Average Current-Mode Controller To verify the operation of the non-linear average current-mode controller, from Fig. 4.1.b), the single-phase buck is now replaced with a point-of-load, two-phase, 80-W, 12-V-to-1.8-V buck converter switching at f sw =500 khz. The nominal phase inductances are µh (±10% tolerances) while the total nominal output capacitance is 800 µf (-20/+50% tolerance). To accommodate the increased input voltage range (up to 16 V), the effective input ADC resolution is decreased to 32 mv by a voltage divider (1/8). On the other hand, the sampling rate of the output voltage ADC is increased to 8 times per switching cycle. The current sink used for the calibration of the multi-phase current estimator is increased to 4 A (less than 10% of max. load current) due to the increased output capacitance. Shortly after the system start-up, when the output voltage reaches steady state, the controller initiates the tuning of the multi-phase current estimator parameters and identifies the actual phase component values. Fig presents the tuning procedure applied on the first phase, while the second phase current is frozen. Initially, both the gain and time constant of the current estimator are incorrect. After injecting the step and performing the multi-phase calibration procedure described in Section 4.6, the parameters are identified. From Fig it can be observed that the current step of the sink fully excites the active phase. After, the calibration is performed; the corrective action is taken and the estimated

92 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 78 current is now properly tuned. The identified output capacitance is 500 µf, which is 37.5% lower than its nominal value due to the capacitor DC bias effect [54]. Figure 4.24: Current estimator tuning and component identification with the multi-phase converter Ch1: Output converter voltage (50 mv/div); Ch2: actual phase 1 inductor current i L1 (t) 10 A /V; Ch3: actual phase 2 inductor current i L2 (t) 10 A /V; Ch4: estimated current i sense1 [n] 10 A/V; D1 - sink enable signal. Time scale is 20 μs/div. After the tuning and power-stage identification are completed, the response of the non-linear multi-phase controller is tested with a large load step (0 A 45 A) as shown in Fig As soon as the sharp voltage drop is detected and ΔI load is estimated, the transient compensator first turns on both phases to block the voltage deviation. It can be noticed that phase inductances are significantly mismatched ( 20%), since i L1 (t) is able to reach i L2 (t) at the end of the turn-on action. The transient compensator now measures the voltage deviation and performs the control actions based on the change of current references producing duty ratio increments ±Δd i. The current-loop compensators recognize the mismatch of phase inductances and current undershoot in the second phase. To compensate they provide a larger initial current step for the second phase as shown in Fig Once the output voltage reaches the new steady state within 5 μs, the current compensator regulate inductor currents until they match the current ratio specified by their equivalent resistances. Finally, the effectiveness of the conduction loss-based current scheme that equalizes the thermal stress of converter phases is verified. Initially, equal duty cycles of control signals c 1 (t) and c 2 (t) are enforced to show that phase equivalent resistances are indeed mismatched. As shown in Fig a), the average phase currents are now unequally distributed based on the

93 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 79 ratio of R eq1 and R eq2. Phase 1, taking significantly larger portion of the output load current than Phase 2, has also much higher operating temperature (ΔT 12 = 17.8 C) as demonstrated in Fig b). Then, the conduction-loss based current sharing scheme is enabled. The load current Figure 4.25: Transient response of the nonlinear average current-mode controller Ch1: Output converter voltage (50 mv/div); Ch2: inductor current i L1 (t) 10 A /V; Ch3: inductor current i L2 (t) 10 A /V; Ch4: estimated phase 2 current i sense2 [n] 10 A/V; D1-D3- sink enable and control signals. Time scale is 5 μs/div. Figure 4.26: The effectiveness of the conduction-loss based sharing scheme a) current sharing with equal duty cycles of control signals c 1 (t) and c 2 (t); b) mismatched phase temperatures with equal duty cycles of c 1 (t) and c 2 (t); c) conduction-loss based current sharing; d) phase temperatures with the conduction-loss based sharing scheme.

94 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 80 is now distributed between phases based on the square root ratio of R eq1 and R eq2 (Fig c)) to equalize the conduction losses of both phases. Consequently, the phase temperatures are now well balanced (ΔT 12 = 1.4 C) as demonstrated in Fig d) Accuracy of Current and Temperature Estimation The accuracy of the current estimation is assessed by changing the load current and comparing it to the estimated value. Before the accuracy measurements are performed, the estimator is calibrated only once at 30% of the maximum load current. Fig a) shows that a fairly good estimate is obtained. For the full load, the relative error of the estimation is less than 5% while for 90% of the full load it is less than 10%. At lighter load, the error increases due to quantization effects. If the sink step size is increased from 1 A to 2 A, the error is reduced on average by 1.3% as illustrated in Fig b). Fig shows results of the temperature estimation. It can be seen that the absolute error of the temperature estimation is limited to ±7 C while the relative error is less than 10%. The precision of this measurement depends on the size of the look-up tables and the accuracy of the data stored in the estimator look-up tables. Figure 4.27: a) Estimated inductor current versus the output load current with a 1 A sink step size b) Relative error of the estimated current for two sink step sizes: 1 A and 2 A.

95 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION Conclusion Figure 4.28: Estimated temperature versus sensor temperature. This chapter introduces a combined self-tunable digital current and temperature estimator and the non-linear sensorless average current-mode controller with for multi-phase dc-dc converters. Based on the measurement of the inductor voltage and its processing in the digital domain the current estimator calculates the average value of the inductor current over one switching cycle. To compensate for the variations in power stage parameters that currently limit the accuracy of equivalent analog methods, the estimator automatically extracts the main converter parameters and self-adjusts its parameters by using a test current sink and a tunable IIR filter. Compared to previous iterative-based tuning methods, the tuning procedure is much faster since it requires only a single action of the sink circuit. From the extracted equivalent resistance, the temperature of the MOSFETs is remotely estimated and continuously monitored. The fast and accurate operation of the current and temperature estimator is demonstrated on a digitally controlled buck converter prototype. The non-linear average mode controller utilizes the identified power stage parameters and estimated phase currents to provide fast transient response and dynamic current sharing. To equalize the thermal stress between phases, a conduction-loss based current sharing scheme is implemented. The operation of the controller is verified with a two-phase point-of-load, 80-W, buck converter showing the response time under 5 μs and a voltage deviation smaller than 100

96 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 82 mv for a 45-A load step. The digital blocks of the controller are suitable for an on-chip implementation since they occupy only 0.43 mm 2 of silicon area when synthesized in a CMOS 0.18-μm process. REFERENCES [1] H. Peng, D. Maksimović, Overload protection in digitally controlled DC-DC converters," in Proc. IEEE Power Electronics Specialist Conf., Jun. 2006, pp [2] J.Xiao, A.V. Peterchev, J. Zhang, S.R. Sanders, A 4-μA quiescent-current dual-mode digitally controlled buck converter IC for cellular phone applications, IEEE Journal of Solid-State Circuits, vol. 39, pp , Dec [3] N. Rahman, A. Parayandeh, K. Wang, A. Prodić, Multimode digital SMPS controller IC for low-power management, in Proc. IEEE International Symposium on Circuits and Systems, 2006, pp [4] S. Bibian, H. Jin, High performance predictive dead-beat digital controller for DC power supplies, IEEE Trans. Power Electron., vol. 17, pp , May [5] Kelvin Ka-Sing Leung, Henry Shu-Hung Chung, Dynamic hysteresis band control of the buck converter with fast transient response, IEEE Trans. Circuits Syst. II, vol. 52, pp , July [6] A. Soto, P. Alou, J.A. Cobos, Nonlinear digital control breaks bandwidth limitations, in Proc. IEEE Applied Power Electronics Conf., 2006, pp [7] G. Feng, E. Meyer, Y.-F. Liu, A new digital control algorithm to achieve optimal dynamic performance in DC-to-DC converters, IEEE Trans. Power Electron., vol. 22, pp , July [8] Feng Su, Wing-Hung Ki, Chi-Ying Tsui, Ultra fast fixed-frequency hysteretic buck converter with maximum charging current control and adaptive delay compensation for dvs applications, Proc. IEEE Journal of Solid-State Circuits, vol. 43, pp , Apr [9] S. Saggini, P. Mattavelli, M. Ghioni, M. Redaelli, Mixed-signal voltage-mode control for DC-DC converters with inherent analog derivative action, IEEE Trans. Power Electron.,vol. 23, pp , May [10] S. Saggini, P. Mattavelli, G. Garcea, M. Ghioni, A mixed-signal synchronous/ asynchronous control for high-frequency DC-DC boost converters, IEEE Trans. Ind. Electron., vol. 55, pp , May [11] L. Corradini, P. Mattavelli, E. Tedeschi, D. Trevisan, High-bandwidth multisampled digitally controlled DCDC converters using ripple compensation, IEEE Trans. Ind. Electron.,vol. 55, pp , Apr

97 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 83 [12] Zhenyu Zhao, A. Prodic, Continuous-time digital controller for high-frequency DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , Mar [13] E. Meyer, Zhiliang Zhang, Y.-F. Liu, An optimal control method for buck converters using a practical capacitor charge balance technique, IEEE Trans. Power Electron., vol. 23, pp , July [14] V. Yousefzadeh, A. Babazadeh, B. Ramachandran, E. Alarcon, L. Pao, and D. Maksimovic, Proximate time-optimal digital control for synchronous buck DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , July [15] A. Costabeber, L. Corradini, P. Mattavelli, S. Saggini, Time optimal, parametersinsensitive digital controller for DC-DC buck converters, in Proc. IEEE Power Electronics Specialist Conf., 2008, pp [16] S. Effler, A. Kelly, M. Halton, T. Kruger, and K. Rinne, Digital control law using a novel load current estimator principle for improved transient response, in Proc. IEEE Power Electronics Specialist Conf., 2008, pp [17] L. Corradini, E. Orietti, P. Mattavelli, S. Saggini, Digital hysteretic voltage-mode control for DC-DC converters based on asynchronous sampling, IEEE Trans. Power Electron., vol. 24, pp , Jan [18] Santa C. Huerta, P. Alou, J. A. Olivier, O. Garcia, J. A. Cobos, A. Abou-Alfotouh, A very fast control based on hysteresis of the cout current with a frequency loop to operate at constant frequency, in Proc. IEEE Applied Power Electronics Conf., 2009, pp [19] B. Tomescu, H.F.Vanlandingham, Improved large-signal performance of paralleled DC- DC converters current sharing, IEEE Trans. Power Electron., vol. 14, pp , May [20] Peng Li, B. Lehman, A design method for paralleling current mode controlled DC-DC converters, IEEE Trans. Power Electron., vol. 19, pp , May [21] J. Abu-Qahouq, Mao Hong, I. Batarseh, Multiphase voltage-mode hysteretic controlled DC-DC converter with novel current sharing, IEEE Trans. Power Electron., vol. 19, pp , Nov [22] Jaber A. Abu Qahouq, Lilly Huang, Doug Huard, Allan Hallberg, Novel current sharing schemes for multiphase converters with digital Controller implementation, in Proc. IEEE Applied Power Electronics Conf., Feb. 2007, pp [23] Y. Zhang, R. Zane, D. Maksimović, Current sharing in digitally controlled masterless multi-phase DC-DC Converters, in Proc. IEEE Power Electronics Specialist Conf., Jun. 2005, pp [24] Mao Hong, Liangbin Yao, Caisheng Wang, I. Batarseh, Analysis of inductor current sharing in nonisolated and isolated multiphase dc-dc Converters, IEEE Trans. Power Electron., vol. 54, pp , Dec 2007.

98 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 84 [25] V. Choudhary, E. Ledezma, R. Ayyanar, R.M. Button, Fault tolerant circuit topology and control method for input-series and output-parallel modular DC-DC converters, IEEE Trans. Power Electron., vol. 23, pp , Jan [26] Zhenyu Zhao, A. Prodić, Limit-cycle oscillations based auto-tuning system for digitally controlled DC-DC power supplies, IEEE Trans. Power Electron., vol. 22, pp , Nov [27] L. Corradini, P. Mattavelli, W. Stefanutti, and S. Saggini, Simplified model referencebased autotuning for digitally controlled smps, IEEE Trans. Power Electron., vol. 23, pp , July [28] J. Morroni, R. Zane, D. Maksimović, Design and implementation of an adaptive tuning system based on desired phase margin for digitally controlled DCDC converters, IEEE Trans. Power Electron., vol. 24, pp , Feb [29] H. Forghani-zadeh, G. Rincón-Mora, Current sensing techniques for DC-DC converters, in Proc. IEEE MWSCAS, Aug. 2002, pp [30] P. Givelin, M. Bafleur, On-chip over-current and open-load detection for a power MOS high-side switch: a CMOS current-source approach, in Proc. European Conf. Power Electronics and Applications, 1993, pp [31] S. Yuvarajan, L.Wang, Power conversion and control using a current-sensing MOSFET, in Proc. Midwest Symp. Circuits and Systems (MWSCAS), 1992, pp [32] J. Chen, J. Su, H. Lin, C. Chang, Y. Lee, T. Chen, H.Wang, K. Chang, P. Lin, Integrated current sensing circuits suitable for step-down DC-DC converters, IEE Electron. Letters, pp , Feb [33] C. Lee, P. Mok, A monolithic current-mode CMOS DC-DC converter with on-chip current-sensing technique, IEEE Journal of Solid-State Circuits, vol. 39, no. 1, pp. 3 14, Jan [34] P. Midya, P. T. Krein, M. F. Greuel, Sensorless current mode control-an observer-based technique for DC-DC converters, IEEE Trans. Power Electron., vol. 16, pp , Jul [35] E. Dallago, M.Passoni, G. Sassone, Lossless current sensing in low-voltage high-current DC/DC modular supplies, IEEE Trans. Ind. Electron., vol. 47, pp , Dec [36] L. Hua, S. Luo, Design considerations of time constant mismatch problem for inductor DCR current sensing method, in Proc. IEEE Applied Power Electronics Conf., 2006, pp

99 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 85 [37] Data Sheet, MAX 6697, 7-Channel Precision Temperature Monitor, Maxim. [38] Data Sheet, TMP 100, I2C Temperature Sensor, Texas Instruments. [39] G. Garcea, S. Saggini, D. Zambotti, and M. Ghioni, Digital auto-tuning system for inductor current sensing in VRM applications, in Proc. IEEE Applied Power Electronics Conf., 2006, pp [40] H. P. Forghani-zadeh, G. A. Rincón-Mora, An accurate, continuous, and lossless selflearning CMOS current-sensing scheme for inductor-based DC-DC converters, IEEE Journal of Solid-State Circuits, vol. 42, pp , Mar [41] Z. Lukić, N. Rahman, A. Prodić, Multi-bit Σ- PWM digital controller IC for DC DC converters operating at switching frequencies beyond 10 MHz, IEEE Trans. Power Electron., vol. 22, pp , Sept [42] A. Parayandeh, A. Prodić, Programmable analog-to-digital converter for low-power DC- DC SMPS, IEEE Trans. Power Electron., vol. 23, pp , Jan [43] Z. Lukić, Z. Zhao, S M Ahsanuzzaman, A. Prodić, Self-tuning digital current estimator for low-power switching converters, in Proc. IEEE Applied Power Electronics Conf., 2008., pp [44] N. Rahman, A. Parayandeh, K. Wang, and A. Prodić, Multimode digital SMPS controller IC for low-power management, in Proc. IEEE International Symposium on Circuits and Systems, 2006, pp [45] O. Trescases, W. T. Ng, H. Nishio, M. Edo, T. Kawashima, A digitally controlled DC-DC converter module with a segmented output stage for optimized efficiency, in Proc. IEEE Power Semiconductor Devices and ICs, 2006, pp [46] A. Prodić, D. Maksimović, Digital PWM controller and current estimator for a low-power switching converter, in Proc. IEEE Computers in Power Electronics Conf., 2000, pp [47] Z. Lukić, A. Stupar, A. Prodić, D. Goder, Current estimation and remote temperature monitoring system for low power digitally controlled DC-DC SMPSs," in Proc. IEEE Power Electronics Specialist Conf., Jun. 2008, pp [48] J. Morroni, A. Dolgov, R. Zane, D. Maksimović, Online health monitoring in digitally controlled power converters, in Proc. IEEE Power Electron. Specialists Conf., 2007, pp

100 CHAPTER 4. RECONFIGURABLE DIGITAL CONTROLLERS WITH MULTI-PARAMETER ESTIMATION 86 [49] J. Chen, A. Prodić, R.W. Erickson, D. Maksimović Predictive digital current programmed control, IEEE Trans. Power Electron., vol. 18, no. 1, pp , Jan [50] PIP212-12M data sheet, NXP Semiconductors, Eindhoven, Netherlands. [51] FDMF8700 data sheet, Fairchild Semiconductors, San Jose, USA. [52] C. Nesgaard, M. A. E. Andersen, Optimized load sharing control by means of thermalreliability management, in Proc. IEEE Power Electronics Specialist Conf., 2004, pp [53] K. Rinne, A. Russel, A power converter, International Patent, "WO/2009/010476", Jan [54] Technical Note, Mechanism of DC bias characteristics, Murata.

101 Chapter 5 Conclusions and Future Work This thesis presents advanced reconfigurable digital controllers of multi-phase dc-dc converters utilized for power management systems in low power applications, up to 100 W. The proposed architectures are suitable on-chip integration and provide practical hardware solutions for a broad range of implementation issues related to: controller power consumption, dynamic converter efficiency optimization, current estimation, temperature monitoring, converter parameter estimation, and combined fast transient response and dynamic current sharing. In the following sections the thesis contributions are summarized, and future work is discussed. 5.1 Thesis Contributions The most significant contributions of this dissertation are: 1) Multi-Use and Fault-Tolerant Digital Controller IC: A flexible four-phase digital PWM controller IC that can be used with interleaved, multi-output, and parallel dc-dc switching converters operating at frequencies up to 10 MHz is proposed. The IC can be programmed to operate with any number of converter phases and it is fault-tolerant. During interleaved mode, if a failure in one of the phases occurs, it automatically switches to operation with reduced number of phases by disabling the faulted phase and adjusting the angles of the remaining ones. The key element of this IC is a new multiphase digital pulse-width modulator (MDPWM) that utilizes a programmable counter, programmable delay line, and digital logic with variable numbers representation. The MDPWM produces four pulse-width modulated control signals whose switching frequency can be adjusted between 100 khz and 10 MHz with an effective resolution of 11 bits of input duty ratio commands. The controller IC is realized in standard 0.18 μm CMOS process and exhibits low power consumption of μa/mhz per phase. Due to its flexibility and low-power consumption, the IC can be utilized in various applications including low-power, battery-powered electronic devices where multiple supply voltages are required. 87

102 CHAPTER 5. CONCLUSIONS AND FUTURE WORK 88 2) System for Dynamic Efficiency Optimization of Multi-phase DC-DC Converters: A novel current-program mode digital controller and a multiphase dc-dc converter with non-uniform current sharing that dynamically optimize converter efficiency over the full range of operation are introduced. The converter phases are operated as binary-weighted constant current sources, i.e. scaled in a binary-logarithmic fashion. To minimize the system size and provide fast dynamic response, the phases switch at different frequencies and their components are selected so that the most efficient operating points correspond to the set currents. The digital controller operates on a modification of the phase shadowing principle. Depending on the output load, the number of active phases is dynamically configured. The new architecture of the controller does not require an analog-to-digital converter for current measurement and is suitable for high-frequency low power converters. An experimental 4-phase, 70-W, 12 V-to-1.8 V buck converter utilizing the digital control architecture with logarithmic current sharing is built. A comparison of the efficiency with an equivalent uniform current shared converter shows that, at medium and light loads, the presented system results in efficiency improvements of up to 6% and 25%, respectively. 3) Reconfigurable Digital Controllers with Multi-Parameter Estimation: A reconfigurable digital controller architectures providing estimation of inductor currents, identification of key converter parameters, and temperature monitoring of power switches in single-phase and multi-phase dc-dc converters are presented. To perform component parameter estimation and current estimator calibration with multi-phase converters, a method based on freezing phase currents is proposed. The non-linear sensorless multiphase average current-mode controller utilizes the identified power stage parameters and estimated inductor currents to provide fast transient response, while maintaining dynamic current sharing and equalized thermal stress between converter phases. The equal thermal stress of converter phases is obtained by regulating phase currents in order that the phase conduction losses are balanced. The new controller eliminates the need for accurate output voltage valley point detection and can operate at modest oversampling rates of the output voltage. Both controller architectures are well-suited for on-chip implementation. They provide a solution for combining a digital controller, current estimator and remote temperature monitor on a single integrated circuit. The estimation of the average inductor

103 CHAPTER 5. CONCLUSIONS AND FUTURE WORK 89 current value over one switching cycle is based on the analog-to-digital conversion of the inductor voltage and consequent adaptive signal filtering. The adaptive digital IIR filter is used to compensate for variations of the inductance and dc resistance emulating converter losses and affecting accuracy of the estimation. The operation of both controllers is verified with a single-phase 5.5 V-to-1.5 V, 15-W buck converter and dual-phase 12 V- to-1.8 V, 80-W, buck converter, both operating at a switching frequency of 500 khz. The results show that the current and temperature estimations are accurate with error less than 10%. The current estimation has the response time of one switching cycle. Dynamic current sharing combined with small transient response time, under 5 µs for a 45-A load current step, are obtained with the dual-phase buck converter having an output capacitor of 500 µf. The digital blocks of the new multi-phase controller occupy only 0.43 mm 2 of silicon area when implemented in 0.18μm CMOS process. 5.2 Future Work Future work could include a development of a digital controller IC that now intelligently combines the best features of all presented controllers in this thesis. For instance, a system for dynamic efficiency optimizations of multi-phase converters could be merged with the sensorless non-linear average current-mode controller. The operation of the current-mode controller could be also modified to dynamically balance converter losses from the information about phase components and phase currents in the presence of high-frequency load steps. Finally, the current sink for current estimator calibration and oversampling output voltage ADC for load transient detection and estimation could be potentially eliminated or simplified (in case of the ADC) by establishing a digital communication link between the switch-mode power supply and its load. From the information passed to the controller about the operating mode, the specific activity of the load device, and its current consumption, the current estimator could be periodically calibrated while the requirement for oversampling of the output voltage would be eliminated.

104 Appendix A Filter Time Constant Calculation To derive the equation that links the filter time constant τ f and inductor time constant τ L with properties of the estimated current in order to perform the filter time constant tuning, the following approximation is made. It is assumed that during the initial transient, the average inductor current behaves as a ramp signal with a slope k. The Laplace transform of i L (t) in that case is equal: I L 1 = 2 (A.1) s ( s) k. The sensed current according from the output of the RC filter according to (4.1) and (A.1) is then equal to: I sense 1+ sτ 1 1+ sτ ω 1 s + ω = L 2 2 (A.2) 1+ sτ s 1+ sτ ω s s + ω L L f L ( s) I ( s) = k = k, f f L f where ω f =1/τ f and ω L =1/τ L. The sensed current I sense (s) now can decomposed as: I sense A B C = 2 (A.3) s s s +ω ( s) + +, f where coefficients A, B, and C are: ω f ωl A = k, (A.4) ω ω f L B = k, (A.5) ω f ωl C = k. (A.6) ω ω f L 90

105 APPENDIX A. FILTER TIME CONSTANT CALCULATION 91 By using the fact that the middle term of I sense (s) in (A.3) is equal to I L (s) from (A.1) and after applying the inverse Laplace transform, the sensed inductor current becomes equal to: i sense ω ω f L ω f t ( t) = k ( 1 e ) + i ( t). ω ω f L L (A.7) The difference ΔI peak between the estimated current and actual inductor current when the output voltage peak point occurs at t= ΔT peak is then equal: (A.8) where the slope k is equal to ΔI/ ΔT peak. From an approximation that e -x 1-x+x 2 /2, where x= ω f ΔT peak, to simplify (A.8), it can be shown the ratio between ω f and ω L is: (A.9) Finally, by rearranging (A.9) the inductor time constant τ L now can be expressed in terms of the filter time constant τ f as: (A.10) 91