@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Applied Science, Faculty of"@en, "Electrical and Computer Engineering, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Afkhami Ardakani, Hasan"@en ; dcterms:issued "2017-08-28T22:36:06Z"@en, "2017"@en ; vivo:relatedDegree "Master of Applied Science - MASc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """Implantable sensors have been used to improve monitoring and diagnosis of health-related parameters while allowing patients to lead a relatively normal life. Using data from such sensors, one can detect abnormal conditions at early stages and facilitate the prevention of potentially serious consequences. Recent technological advances in integrated circuits, wireless communications, and physiological sensing allow miniature, lightweight, ultra-low-power, intelligent monitoring devices. In this thesis, we focus on an electro-thermally active stent technology for management of in-stent restenosis (i.e., re-narrowing of artery at the stented site). Various studies reporting hyperthermia treatments of restenosis through stent heating have shown promising results, i.e., moderate local heating prevents restenosis by limiting cell proliferation. To remotely warm up the stent, we intend to harvest power from a dedicated source outside of the patient’s body and convert it to heat. However, if there is no control over temperature, the stent temperature may increase unboundedly, which would have adverse effects. The main objective of this thesis is to design a low-power, accurate temperature sensing system with a small footprint. Further, the required power to operate the temperature sensor should be harvested. In this work, two different temperature telemonitoring systems have been designed and laid out in a 65-nm CMOS technology. Both systems have been fabricated and successfully validated. The first telemonitoring system converts the sensed temperature directly to a frequency in an unlicensed band and transmits it to an external reader. The system operates from a supply voltage of 0.7 V and a power consumption of 100 µW. The measured sensitivity of the system is 1.1 MHz/°C within the frequency band of 902 to 928 MHz. This system is capable of detecting temperature change to as low as 1 °C. The sensor interface circuit of our second telemonitoring system converts the temperature to duty-cycle and sends sensory data out using an on-off-keying modulation system. The pulse width of the transmitted signal is proportional to e temperature. Measurement results of a proof-of-concept prototype show that the system operates from a supply voltage of as low as 0.6 V while consuming 115 µW."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/62851?expand=metadata"@en ; skos:note "Low-power Temperature Sensing System for Biomedical Applications by Hasan Afkhami Ardakani B. Sc., Isfahan University of Technology, 2014 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (ELECTRICAL AND COMPUTER ENGINEERING) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) August 2017 © Hasan Afkhami Ardakani, 2017 ii Abstract Implantable sensors have been used to improve monitoring and diagnosis of health-related parameters while allowing patients to lead a relatively normal life. Using data from such sensors, one can detect abnormal conditions at early stages and facilitate the prevention of potentially serious consequences. Recent technological advances in integrated circuits, wireless communications, and physiological sensing allow miniature, lightweight, ultra-low-power, intelligent monitoring devices. In this thesis, we focus on an electro-thermally active stent technology for management of in-stent restenosis (i.e., re-narrowing of artery at the stented site). Various studies reporting hyperthermia treatments of restenosis through stent heating have shown promising results, i.e., moderate local heating prevents restenosis by limiting cell proliferation. To remotely warm up the stent, we intend to harvest power from a dedicated source outside of the patient’s body and convert it to heat. However, if there is no control over temperature, the stent temperature may increase unboundedly, which would have adverse effects. The main objective of this thesis is to design a low-power, accurate temperature sensing system with a small footprint. Further, the required power to operate the temperature sensor should be harvested. In this work, two different temperature telemonitoring systems have been designed and laid out in a 65-nm CMOS technology. Both systems have been fabricated and successfully validated. iii The first telemonitoring system converts the sensed temperature directly to a frequency in an unlicensed band and transmits it to an external reader. The system operates from a supply voltage of 0.7 𝑉 and a power consumption of 100 µ𝑊. The measured sensitivity of the system is 1.1 𝑀𝐻𝑧/°𝐶 within the frequency band of 902 to 928 MHz. This system is capable of detecting temperature change to as low as 1 °𝐶. The sensor interface circuit of our second telemonitoring system converts the temperature to duty-cycle and sends sensory data out using an on-off-keying modulation system. The pulse width of the transmitted signal is proportional to e temperature. Measurement results of a proof-of-concept prototype show that the system operates from a supply voltage of as low as 0.6 𝑉 while consuming 115 µ𝑊. iv Lay Summary The primary objective of this research is to design a low-power temperature sensing system for biomedical implants. In particular, the focus of our work is on a smart stent, which is a tube-like device implanted in the blocked or narrowed artery to keep the lumen open. The stent is wirelessly heated and uses a temperature sensor for controlling its temperature. The required power for the temperature sensor operation is harvested from outside of the patient’s body. Two approaches for transferring the sensor information (temperature) from the implanted device to outside of the patient’s body are investigated. We have designed and implemented the proposed integrated temperature sensing system using complementary metal-oxide-semiconductor (CMOS) technology and have experimentally validated the performance of the system. v Preface This thesis is submitted for the degree of Master of Applied Science at the University of British Columbia. The research described herein was conducted under the supervision of Professor Shahriar Mirabbasi, in the Department of Electrical and Computer Engineering, the University of British Columbia, between September 2015 and August 2017. Professor Sudip Shekhar provided technical and editing assistance for Chapter 4. In addition, Amir Masnadi Shirazi provided technical assistance in the design and measurements of the low power voltage-controlled oscillator (VCO) that is presented in Chapter 4. This work, to the best of my knowledge, is original, except for where references are made to previous works. Part of this work has been presented in the following publication: H. Afkhami, A. Masnadi Shirazi, S. Shekhar, S. Mirabbasi, “A Low Power Temperature Sensing System for Implantable Biomedical Applications, ” in 2017 IEEE International New Circuits and Systems Conference (NEWCAS), 2017, pp. 1–4 (Chapter 4). vi Table of Contents Abstract .......................................................................................................................................... ii Lay Summary ............................................................................................................................... iv Preface .............................................................................................................................................v Table of Contents ......................................................................................................................... vi List of Tables ................................................................................................................................ ix List of Figures .................................................................................................................................x List of Abbreviation ................................................................................................................... xiii Acknowledgements .................................................................................................................... xiv Dedication .....................................................................................................................................xv Chapter 1: Introduction to Implantable Biomedical devices .....................................................1 1.1 System Overview ............................................................................................................ 2 1.1.1 General Requirements ................................................................................................. 2 1.2 Wireless Communication Technologies for Implanted Devices .................................... 4 1.2.1 Modulation Methods ................................................................................................... 5 1.2.1.1 AM and ASK Modulation−Demodulation ......................................................... 6 1.2.1.2 FM and FSK Modulation − Demodulation ......................................................... 7 1.2.1.3 PSK Modulation and Demodulation ................................................................... 8 1.2.1.4 Pulse Modulation Encoding ................................................................................ 9 1.3 Conclusion .................................................................................................................... 11 Chapter 2: Temperature effects on Silicon Devices ..................................................................13 2.1 Inductors ....................................................................................................................... 13 vii 2.1.1 Parasitic Resistance ................................................................................................... 15 2.1.2 Parasitic Capacitances ............................................................................................... 17 2.2 Capacitors ..................................................................................................................... 22 2.2.1 Varactors ................................................................................................................... 22 2.3 Inductor Models with Temperature Effect .................................................................... 23 2.4 Temperature Effects on Silicon .................................................................................... 26 2.4.1 Threshold Voltage ..................................................................................................... 27 2.4.2 Mobility..................................................................................................................... 28 2.4.3 Leakage Currents ...................................................................................................... 29 2.4.4 Electrical Conductivity ............................................................................................. 30 2.5 MOSFET Temperature Dependences ........................................................................... 31 2.5.1 On-resistance of MOSFET ....................................................................................... 32 2.5.2 Transconductance (gm) of a MOSFET ...................................................................... 33 2.5.3 Parasitic Capacitances ............................................................................................... 33 2.6 Zero Temperature Coefficient ....................................................................................... 34 2.7 Conclusion .................................................................................................................... 38 Chapter 3: Low-Power VCO for Biomedical Application .......................................................39 3.1 RLC Circuit ................................................................................................................... 39 3.2 Temperature Effects on LC-VCO ................................................................................. 42 3.3 Low Power VCO/Buffer for Biomedical Application .................................................. 48 3.4 Conclusion .................................................................................................................... 54 Chapter 4: A Low-Power Temperature Sensing System for Implantable Biomedical Applications ..................................................................................................................................55 viii 4.1 Introduction ................................................................................................................... 55 4.2 Temperature Sensor Architecture ................................................................................. 57 4.3 Low-Power FM Transmitter ......................................................................................... 61 4.4 Measurement Results .................................................................................................... 64 4.5 Conclusion .................................................................................................................... 64 Chapter 5: Conclusion .................................................................................................................67 5.1 Future Works ................................................................................................................ 68 Bibliography .................................................................................................................................70 Appendices ....................................................................................................................................78 Appendix A BJT based Temperature Sensor ............................................................................ 78 A.1 CMOS-compatible temperature sensors ................................................................... 79 A.2 BJT based temperature sensors ................................................................................. 82 A.3 Duty cycle modulation and sigma delta ADC .......................................................... 84 Appendix B Two-Stage Folded Cascode OTA ......................................................................... 91 ix List of Tables Table 2-1 Conductivity and temperature coefficient of various materials at 20 °C [61]. ............. 31 Table 3-1 Performance comparison of OOK transmitter. ............................................................. 52 Table 4-1 Temperature sensor performance summary and comparison. ...................................... 61 Table 4-2 Performance summary and comparison. ...................................................................... 65 Table 5-1 Performance summary of the proposed systems. ......................................................... 68 x List of Figures Figure 1-1 A summary of the potential power sources and the total power from various body-centered actions [7]. ........................................................................................................................ 3 Figure 1-2 Classification of the communication links based on the physical connection between TX and RX [3]. ............................................................................................................................... 5 Figure 1-3 AM modulation. ............................................................................................................ 7 Figure 1-4 Principle of ASK modulation. ....................................................................................... 7 Figure 1-5. FM modulation. ............................................................................................................ 8 Figure 1-6 Principle of FSK modulation. ....................................................................................... 8 Figure 1-7 PSK techniques often applied in biotelemetry. ............................................................. 9 Figure 1-8 Constellation diagrams of FSK, ASK, and PSK. .......................................................... 9 Figure 1-9 Pulse modulation encoding techniques. ...................................................................... 10 Figure 2-1 Planar spiral inductors. ................................................................................................ 14 Figure 2-2 Lumped model including magnetic coupling between the spiral and the substrate. ... 15 Figure 2-3 Current distribution in a conductor. ............................................................................ 17 Figure 2-4 Compact frequency-independent inductor model. ...................................................... 18 Figure 2-5 Patterned ground shield (PGS). ................................................................................... 19 Figure 2-6 Lumped one-port inductor model (left) and its equivalent (right). ............................. 20 Figure 2-7 Inductor model at different frequencies and corresponding Q behavior. .................... 21 Figure 2-8 CV characteristic of a MOS varactor, its Q variation and Lumped model. ................ 23 Figure 2-9 Normalized substrate resistance vs. temperature. ....................................................... 25 Figure 2-10 Quality factor vs. frequency. ..................................................................................... 25 xi Figure 2-11 Change in the threshold voltages of N-channel and P-channel MOSFETS vs. temperature. .................................................................................................................................. 28 Figure 2-12 Simulation results of IDS –– VGS characteristic at VDS = 0.6 V and at various temperatures (in TSMC 65nm). .................................................................................................... 35 Figure 2-13 Simulation results of gm –– VGS characteristics at VDS=0.6 V and at various temperatures (in TSMC 65nm). .................................................................................................... 37 Figure 3-1 Ideal LC circuit (left), Capacitor energy in an ideal LC circuit (center), Pole locations of an LC circuit in the s-plane (right). .......................................................................................... 40 Figure 3-2 Lossy LC circuit (left), Capacitor energy in lossy LC circuit (center), Pole locations of an RLC circuit in the s-plane (right). ........................................................................................ 40 Figure 3-3 Linear LC oscillator. ................................................................................................... 41 Figure 3-4 System pole locations on the pole-zero plot and impulse response of the linear LC oscillator. ....................................................................................................................................... 42 Figure 3-5 Negative feedback system. .......................................................................................... 42 Figure 3-6 A simplified model of LC-tank. .................................................................................. 43 Figure 3-7 Simulation results of a VCO frequency vs. Temperature (a) large inductor (after compensation) (b) small inductor. ................................................................................................ 47 Figure 3-8 Proposed LC-oscillator/buffer schematic. ................................................................... 49 Figure 3-9 Die photo of the proposed VCO/buffer. ...................................................................... 52 Figure 3-10 Simulation results of the proposed PWM-OOK TX. ................................................ 53 Figure 3-11 Measurement results of the proposed PWM-OOK TX. ............................................ 53 Figure 4-1 A temperature sensor and transmitter for smart-stent implants. ................................. 57 Figure 4-2 Proposed low-power CMOS-based temperature sensor. ............................................ 58 xii Figure 4-3 Output current versus temperature for the proposed sensor........................................ 59 Figure 4-4 Proposed FM transmitter. ............................................................................................ 62 Figure 4-5 Chip micrograph. ......................................................................................................... 64 Figure 4-6 Measured TCO frequency versus temperature. ........................................................... 65 Figure 4-7 Measured TX output at 914.4 MHz (top) and 926.5 MHz (bottom). .......................... 66 Figure A-1 Structure of a basic electro thermal filter. .................................................................. 80 Figure A-2 CMOS temperature sensor based on temperature-dependent delays of CMOS inverters......................................................................................................................................... 80 Figure A-3 Cross-section of (a) Lateral PNP BJT; (b) Vertical PNP BJT; and (c) Vertical NPN BJT. ............................................................................................................................................... 83 Figure A-4 Basic principle of a BJT-based temperature sensor (a) Block diagram of a bandgap temperature sensor (b) Biasing a BJT pair in a current ratio of p, the single-ended voltages are CTAT while the differential voltage is PTAT. ............................................................................. 84 Figure A-5 Principle of duty-cycle modulation. ........................................................................... 85 Figure A-6 Principle of sigma-delta ADC. ................................................................................... 85 Figure 5-7 Kelvin-to-Celsius converter implementation. ............................................................. 86 Figure A-8 Detailed circuit diagram of the temperature sensor. .................................................. 88 Figure A-9 Two-stage folded cascode opamp. ............................................................................. 88 Figure A-10 Die photo of the temperature sensor ........................................................................ 89 Figure A-11 Simulation results (Duty Cycle vs. Temperature) .................................................... 90 Figure B-1 Small signal model for two stage folded cascode OTA. ............................................ 91 xiii List of Abbreviations ADC Analog-to-digital converter BJT Bipolar junction transistor CMOS Complementary metal-oxide-semiconductor CTAT Complementary to absolute temperature ISM band Industrial, scientific, and medical radio band MOSFET Metal–oxide–semiconductor field-effect transistor OOK On-off keying OTA Preoperational transconductance amplifier PTAT Proportional to absolute temperature PSG Patterned-ground-shield PWM Pulse width modulation RF Radio frequency ST Schmitt trigger 𝑉𝑇ℎ Threshold voltage 𝑔𝑚 Transconductance VCO Voltage-controlled-oscillator ZTC Zero temperature coefficient xiv Acknowledgements First and foremost, I wish to express my special gratitude to my supervisor, Professor Shahriar Mirabbasi, for providing me with direction and technical support. I appreciate his countless advice on both research as well as on my career. I am grateful for his inspiration, encouragement and continuous support, throughout my master studies. I would also like to thank Dr. Sudip Shekhar for his scientific advice and insightful suggestions. I would like to thank the members of the SoC group with whom I had the opportunity to work. They provided a friendly and cooperative atmosphere in our research group and offered useful feedback and insightful comments on my work. I would like to thank Dr. Roberto Rosales for his technical assistance, and Roozbeh Mehrabadi for computer-aided design (CAD) tools support. I would also like to thank Canadian Microelectronics Corporation (CMC Microsystems) for providing CAD tools support and facilitating chip fabrication. I would also like to acknowledge the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Canadian Institutes of Health Research (CIHR) for funding this project. My sincere thanks goes out to all of my friends and family who supported me in my journey and incentivized me to strive to achieve my goals. Special thanks are owed to my parents, whose love and guidance are with me in whatever I pursue. I appreciate their endless support and encouragement. Dedication xv Dedication To my parents 1 Chapter 1: Introduction to Implantable Biomedical devices Recent technological advances in integrated circuits and wireless communication have changed the concept of healthcare, and have revolutionized the realization of biomedical devices for health monitoring, diagnosis, and wireless telemetry sensors. Significant progress has been made in the development and improvement of implantable devices (IDs), despite numerous challenges such as power consumption and power delivery [1]. These devices aim to provide patient safety and comfort, and to minimize the cost and risk associated with repeated and invasive surgical procedures [2]. Implantable devices may be powered by batteries or wireless telemetry. Rechargeable and battery-less implantable devices are preferred, as batteries can contribute to the overall size and weight of the device. In addition, non-rechargeable batteries must be surgically replaced. These implantable devices usually communicate with a connection outside of the body through inductive coupling links. Given the available power to the implantable device, choosing the proper modulation and data transmission methods can assist in the further reduction of power consumption and can facilitate secure and fast data transmission. We note that implantable devices should also be biocompatible to prevent any toxic reactions or infections. In addition, longevity and reliability of implantable devices are essential given the cost and time associated with surgical implantation procedures and the patient's recovery [3]. In general, the design of implantable sensors and the corresponding wireless telemetry system is driven by achieving simplicity, a small footprint, low weight, low power operation and efficient transceiver architecture. 2 In this chapter, a discussion on the challenges of designing implantable devices and a brief overview of the possible solutions to these challenges are presented. We discuss analog and digital modulation techniques that can potentially be used for implantable devices. 1.1 System Overview Implantable systems and wireless telemetry devices generally comprise of two fundamental components; an external part located outside of the host body and an internal (implanted) part. The internal component detects, collects and transfers the information to the external receiver via a wireless link (typically an inductive coupling link). The external component is usually used to supply power for the internal component, and/or to analyze and transmit the data to the internal component [2], [4]. In this research project, we focus on the internal (implantable) block. 1.1.1 General Requirements When designing a biomedical implantable device, several requirements should be considered. They are listed as follows: • Low Power Consumption: Power consumption is the main requirement for IDs, as extensive dissipation can drain batteries quickly and may damage soft tissues. IDs can be powered using batteries or wireless power transfer. However, replacement of batteries may require several costly and invasive surgeries. On the other hand, frequent recharging is inconvenient and time-consuming [2]. Wireless power can provide continuous power as an alternative, although the low power restriction should also be applied to ensure that IEEE human tissue exposure standards are met [5]. More recently, much research has been 3 focused on an emerging approach, so-called “energy harvesting”, that exploits ambient energy, natural body motion. or physiological environment (phenomena of the inner body) to generate energy [6]. Figure 1-7 [7], shows potential power sources and the total power from various body-centered actions. The choice of frequency and suitable power supply generally depends on the loss associated with specificities of the ambient condition, the power transfer efficiency, the distance between the internal and external coils, the device geometry, and the package loss [3]. • Minimal Size and Weight: Biodevices should be as small as possible, to be less invasive and result in less discomfort or pain for the host. Excessive size and weight can not only put pressure on tissues, but can also exacerbate tissue damage. With modern CMOS technology, while the overall size and weight of electric circuit components have significantly decreased, battery and package size still pose a barrier [4]. `Body heat (2.4 W)Exhalation (< 1 W)Breathing band (< 0.83 W)Blood pressure (< 0.93 W)Arm motion (< 60 W)Finger motion (6.9 mW)Footfall (<67 W ) Figure 1-1 A summary of the potential power sources and the total power from various body-centered actions [7]. 4 • Biocompatibility: In general, integrity and reliability of IDs can be provided by proper packaging within all unexpected and hostile environments inside the human body. Proper packaging can also protect host tissues from potentially harmful elements of the device, and can offer mechanical support for the implantable device [3], [4]. • Low Voltage Signal and Low Frequencies: Natural signals inside the human body are in the mV or µV range. Hence, low noise systems should be designed to detect small biological signals with minimal power consumption and size. The frequency span of biological signals is between the range of a few hertz to a few kilohertz. In addition, the medical implant communication system (MICS) and the industrial, scientific, and medical radio (ISM) band frequencies have been specifically designated for in vivo and in vitro medical devices [2]. Low voltage and frequency signals demand special care during sensing, amplifying, modulating and transferring. • High Reliability: A failure in biomedical devices can result in pain, damage or even death for the patient. Device maintenance is also complicated and costly, and risks the health of the patient [2]. Therefore, long-term implantable devices with high reliability are essential. 1.2 Wireless Communication Technologies for Implanted Devices Wireless communication between the implanted device (internal) and the external component can be divided into three classes: wave propagation, electrical conduction, and near-field coupling, as shown in Figure 1-2. In this project, only radio-frequency wireless communication is used for data transmission. Further discussion on wireless communication is beyond the scope of this research project and can be found in [3]. 5 A single frequency does not contain any information. As a result, modulation techniques are employed to impress the data signal onto the carrier signal for the transmission. In this section, a brief discussion is presented on different modulation techniques that can potentially be used for data transmission in biomedical devices. 1.2.1 Modulation Methods Analog modulation directly modulates the amplitude (AM), frequency (FM) or phase (PM) of the carrier. Digital modulation uses discrete signals to modulate a carrier wave and can be divided into three groups: amplitude shift keying (ASK), phase shift keying (PSK) and frequency shift keying (FSK). Digital modulation has significantly higher noise immunity compared to the analog counterpart. Figure 1-2 Classification of the communication links based on the physical connection between TX and RX [3]. Wireless communication Wave propagation Conduction Near-field coupling Electromagnetic Acuostic Inductive Capacitive Optical RF ou i 6 Modulation techniques can provide a high data rate, security, low power consumption, good performance, and noise immunity. The proper modulation methods can be selected based on available power, bandwidth, system efficiency, and considering the channel characteristics. 1.2.1.1 AM and ASK Modulation−Demodulation In amplitude modulation the amplitude of the carrier is modulated, as depicted in Figure 1-3. Although an AM based system is the simplest to implement, as the demodulator is only using an envelope detector, it is rarely used in biomedical devices due to weak noise immunity. The digital form of AM, so-called ASK, is significantly less sensitive as it has only two possible carrier amplitudes. Nowadays, the simplest digital modulation used in biomedical devices is ASK or on-off keying (OOK). Figure 1-4 shows the principle of ASK modulation. ASK is the most commonly used modulation technique for wireless telemetry devices because of its simplicity and low power consumption. Several ASK demodulators have been proposed and developed; however, they suffer from high power consumption and/or large area overhead. In general, ASK demodulators used in biomedical applications consist of an envelope detector, digital shaper, and load driver. ASK modulation is also widely used for inductive power transfer, as the tuned coupled coils can operate in the most efficient way if they work continuously. Further, ASK modulation has strong noise performance as its input is pulse modulated (only zero or one) [3]. 7 1.2.1.2 FM and FSK Modulation − Demodulation In frequency modulation, phase or frequency of the carrier is modulated with the source signal. In analog modulation, it is difficult to distinguish frequency modulation from phase modulation. In this modulation, a voltage controlled oscillator (VCO) generates a carrier and its frequency depends on the control voltage (source signal). Since the information contains frequency, FM is not as sensitive to amplitude noise. A phase locked-loop (PLL) can also be used to generate modulated frequencies; however, because of power consumption, PLL is not usually recommended for implanted devices. For instance, the center frequency can vary from a few kilohertz to a few gigahertz. Alternatively, much research has been focused on MICS band designated for biomedical devices. More details of a practical circuit for frequency modulation are discussed in chapter 4. TimeAmplitude AmplitudeTimeBaseband signalAM Figure 1-3 AM modulation. 1 0 1 1 0 1 1 1 Data (Baseband) ASK SignalTimeAmplitudeTimeAmplitude0 0 Figure 1-4 Principle of ASK modulation. 8 FSK is one of the earliest digital modulation techniques used for biomedical applications. The principle of FSK modulation is shown in Figure 1-6. In this method, a digital source modulates a VCO input by changing the varactor between two values. 1.2.1.3 PSK Modulation and Demodulation In the last few decades, much research has been focused on PSK modulation for low power applications. In this modulation, the carrier phase is modulated by 180 degrees (depicted in Figure 1-7) which can be implemented by using an active/passive mixer or balun transformer. The detected signal is compared with a reference signal generated by the carrier recovery circuit that is synchronous to the received signal. In biomedical devices, the absolute received phase is not known and therefore differential PSK (DPSK) is commonly used [3]. As noted previously, when comparing FSK and ASK the former is less sensitive to amplitude noise. PSK has also been proven superior to FSK concerning noise immunity. To clarify, phase is Baseband signalTimeAmplitudeFMTimeAmplitude Figure 1-5. FM modulation. 1 0 1 1 0 1 1 1 Data (Baseband) ASK SignalTimeAmplitudeTimeAmplitude0 0 Figure 1-6 Principle of FSK modulation. 9 the time integral of frequency, and it can be interpreted that PSK averages out noise over the bandwidth of interest. Figure 1-8 shows the constellation diagram of ASK, PSK and FSK, from which the concept of noise immunity can be comprehended. 1.2.1.4 Pulse Modulation Encoding Thus far, pure analog and pure digital modulation have been discussed, which exhibit several drawbacks such as: complex demodulators, large appetite for power, and sensitivity to noise, to name a few. Pulse modulation is an alternative approach, which combines both pure modulations to achieve better signal to noise ratio at the cost of larger bandwidth occupation [2]. 1 0 1 1 0 1 1 1 Data (Baseband) ASK SignalTimeAmplitudeTimeAmplitude0 0 Figure 1-7 PSK techniques often applied in biotelemetry. QIQI'0' '1'QI'0' '1''0' '1' Figure 1-8 Constellation diagrams of FSK, ASK, and PSK. 10 In this modulation, the pulse modulated signal remains analog, but the transmission takes place at discrete times. Analog pulse modulation can be classified into four groups: pulse-amplitude (PAM), pulse-width (PWM) or pulse duration (PDM), pulse-position (PPM) or pulse interval modulation (PIM), and pulse-frequency modulation (PFM). For digital pulse modulation, called pulse-code-modulation (PCM), the analog signal is first quantized and then converted to a pulse train [3]. When a signal is sampled and held for a constant time, PAM can be achieved. Nonetheless, holding the signal value limits PAM’s required bandwidth, causing signal distortion and increasing the reception circuitry’s complexity. PWM is usually achieved by comparing the original signal with a sawtooth waveform, where the duty cycle of the PWM signal is proportional to the sampled value. A 15-channel neural recording interface using PWM time division multiplexed FM, and a 15-channel PDM-FM modulated telemeter for biomedical monitoring, were described in [8] and [9], respectively. For PPM and PIM, a PWM signal should be generated first, followed by the transmission of falling edges (PPM) or both edges (PPM). A blood pressure sensor, bladder pressure telemetry system [10], and eye pressure sensor [11] were described using PPM and PIM modulation, respectively. Baseband signalTimeAmplitude PAMTimeAmplitude PWM/PDMTimeAmplitude PPM/PIMTimeAmplitude PFM Figure 1-9 Pulse modulation encoding techniques. 11 1.3 Conclusion In general, most biomedical devices have similar agreed-upon requirements, such as exhibiting low power consumption, a small footprint, biocompatibility, and high reliability. To minimize invasive effects of bio-devices, the devices themselves need to be as small as possible. Researchers have been working on various solutions to meet these requirements, such as power harvesting and inductive coupling with the aim of eliminating batteries from implanted devices. This direction has been chosen due to the relevant problems associated with batteries in implanted devices, such as limited lifetime, large size, and chemical side effects. Addressing this, a package and encapsulation layer can protect the implanted device under the body’s harsh environment. Furthermore, upholding biocompatibility ensures that host organs, such as tissues, vessels, etc., will not react to the aforementioned potentially harmful elements of the device. In addition to the safety and comfort of the patient, the economy of an implantable device is also important, especially with the increased use of these devices. In this chapter, different modulation techniques suitable for an implantable device’s data communication have been discussed. The proper modulation is selected based on the available power, the distance between transmitter and receiver, and the nature and type of the implanted device. We see that digital modulations are less sensitive to noise compared to analog modulations. FSK modulation can also offer a high data rate; however, it suffers from complicated transceivers/receivers and size issues. The ASK modulation is utilized predominantly for short-range communication because of noise sensitivity issues, while PSK can be used for long distance transmission owing to its superior noise performance. However, PSK may not be suitable for high data rate applications due to bandwidth limitations and demodulator power consumption. 12 Carrier frequency is also very important as it can alter power consumption and size. In general, the human body can safely be exposed to RF electromagnetic fields between 3 kHz – 30 GHz. The Medical Implant Communication System (MICS) band and the second ISM (Industrial, Scientific, and Medical) band are specified between the frequencies of 402–405 MHz and 902–928 MHz, respectively, and are commonly used for biomedical devices. 13 Chapter 2: Temperature effects on Silicon Devices A major success of today’s integrated circuits has been the ability to incorporate numerous on-chip elements, such as resistors, capacitors, and more importantly, inductors, with active devices. In this chapter, on-chip inductors and capacitors used in analog design are discussed. A brief discussion of the physical models of passive components is also presented, in addition to a more detailed study of the temperature-related aspect of integrated circuits. We conclude this chapter by presenting the principles and design tradeoffs of circuits less sensitive to temperature based on the zero-temperature coefficient point. 2.1 Inductors Circular spiral integrated inductors show relatively good performance among chip inductors, but due to fabrication limitations, hexagonal and octagonal structures are typically used. The topmost metal layer, which incorporates the thickest available metal, is usually employed to minimize losses and to achieve a better quality factor (Q). In this section, a brief discussion on inductor models is provided. We begin with the self-inductance of multi-turn spiral inductors [12], 𝐿 ≈ 𝜇2𝜋𝑙53[𝑙4𝑁 +𝑊 +(𝑁 − 1)(𝑊 + 𝑆)]13 𝑊0.083 (𝑊 + 𝑆)0.25 (2.1) where 𝑁 is the umber of turns and 𝑆 is the spacing between two adjacent legs (all units are metric). 14 The quality factor is defined as energy stored in a capacitor or inductor to the average power loss for a sinusoidal excitation, 𝑄 = 𝜔𝑒𝑛𝑒𝑟𝑔𝑦 𝑠𝑡𝑜𝑟𝑒𝑑𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑝𝑜𝑤𝑒𝑟 𝑙𝑜𝑠𝑠= 2𝜋𝑒𝑛𝑒𝑟𝑔𝑦 𝑠𝑡𝑜𝑟𝑒𝑑𝑒𝑛𝑒𝑟𝑔𝑦 𝑙𝑜𝑠𝑡 𝑝𝑒𝑟 𝑐𝑦𝑐𝑙𝑒 (2.2) Since only resistive components dissipate power, various resistances within or around the inductor should be studied. These loss mechanism studies lead us to develop a model for integrated inductors. An equivalent circuit model for inductors can aid designers in developing a simple RLC circuit that can be used in the simulation. Here, lumped Π models for spiral inductors are commonly used, which represent the physical mechanisms taking place in the inductor. In addition, the approximation from this method is valid over a wide range of frequencies. Figure 2-2 reviews the single Π structure. This model consists of low-frequency series inductance (𝐿), the series ohmic resistance (𝑅𝑆), the feedforward capacitance (𝐶𝐹) which models the capacitance between metal lines, the oxide to substrate capacitance (𝐶𝑂𝑋), the substrate capacitance (𝐶𝑆𝑖), and substrate DoutSDinW Figure 2-1 Planar spiral inductors. 15 resistance (𝑅𝑆𝑖). (𝑅′𝑆𝑖) represents substrate resistance due to magnetic coupling to the substrate. In this thesis, we ignore the effects of magnetic coupling. We will now discuss analytical expressions for an inductor’s parasitic elements. 2.1.1 Parasitic Resistance We write the series resistance of an inductor at low frequency as, 𝑟 = 𝑅□𝑙𝑊 (2.3) where 𝑅□ is the metal sheet resistance, 𝑙 is the length of the wire, and 𝑊 is the line width. A wider metal line is usually used to lower the metal’s resistance at the cost of higher capacitance to the substrate. If only the quality factor is a concern, in addition to using the maximum allowable metal width, two or more metal layer can be placed in parallel to reduce the series resistance. In reality, other loss mechanisms also manifest themselves and yield a lower quality factor. At high frequencies, the skin’s effect results in a lower Q. The high frequency current flows at the surface of the conductor and thus results in a greater resistance. The resistance of a conductor due to the effect of the skin is given by, LRSCOX COXCSiCSi RSiRSiCFMRSi' Figure 2-2 Lumped model including magnetic coupling between the spiral and the substrate. 16 𝑅𝑠𝑘𝑖𝑛 =1𝛿𝜎 (2.4) where 𝜎 denotes conductivity and 𝛿 is the skin depth. Skin depth is given by, 𝛿 =1√𝜋𝑓µ𝜎 (≈ 2.5 𝜇𝑚 𝑎𝑡 1 𝐺𝐻𝑧) (2.5) where 𝑓 is the frequency and µ is the permeability. Therefore, as a first order approximation, the series resistance expression can be modified to include the skin effect as, 𝑅𝑆 = 𝑟𝑡𝛿 (1 − 𝑒−𝑡𝛿) (2.6) where 𝑡 is the metal thickness. Eddy current produced by the magnetic field of the adjacent turns also alters the current concentration in metal, which is the so-called “proximity effect”. Considering the skin and proximity effect, one can show that the resistance of a multi-turn spiral inductor is highly frequency-dependent. In practice, the proximity effect can be ignored, as it is not significant compared to the skin effect. In [13] an analytical equation is derived from fundamental electromagnetic principles, 𝑅𝑒𝑓𝑓 = 𝑅𝑜 [1 +110(𝑓𝑓𝑐𝑟𝑖𝑡)2] (2.7) where 𝑅𝑜 is the DC resistance and 𝑓𝑐𝑟𝑖𝑡 can be calculated from the geometrical size of the inductor. In this project, because of the limited operational frequency range (902~928 𝑀𝐻𝑧), and a relatively large 𝑓𝑐𝑟𝑖𝑡 (~1.7 𝐺𝐻𝑧), we can assume that the series resistance is constant and frequency-independent over the abovementioned frequency range. 17 2.1.2 Parasitic Capacitances Apart from the ohmic loss described above, parasitic capacitance also limits an inductor’s performance by limiting the maximum frequency in which the inductor can be used (called the “self-resonance frequency”). Parallel plate capacitances and fringe capacitances are connected to the lossy substrate which can degrade the quality factor. Capacitive and magnetic coupling to the substrate can also create displacement and eddy current in the substrate, respectively, as well as degrading the quality factor. Both the eddy and displacement currents can be reduced using a grounded-shield plate, although the effective inductance will fall in this case and yield a low Q. These effects manifest themselves at a multi Gigahertz regime; however, because of the frequency operation of our devices, we can ignore these effects and simplify the lumped model shown in Figure 2-4. ````DC ConditionSkin EffectProximity EffectFrequencyNumber of parallel linesCurrent Density Figure 2-3 Current distribution in a conductor. 18 𝐶𝐹 represents fringe capacitance and the overlap capacitance between the spiral and the underpass required to connect the inner end of the spiral inductor to external circuitry. This capacitor can be approximated by, 𝐶𝐹 ≈𝐴𝑜𝑣𝑡𝑜𝑥𝑎𝑑 𝜀𝑜𝑥 (2.8) where 𝐴𝑜𝑣 is the overlap area, 𝜀𝑜𝑥 is the permittivity of the oxide layer (𝜀𝑜𝑥 = 3.45 ∗ 10−13𝐹/𝑐𝑚) between the spiral and the underpass, and 𝑡𝑜𝑥𝑎𝑑 is the oxide thickness between the two metal layers. Fringe capacitance between two adjacent legs can be neglected due to its usual small size. 𝐶𝑜𝑥 is the capacitance between the spiral and the lossy substrate, which accounts for most of the inductor’s parasitic capacitance. It is given by, 𝐶𝑜𝑥 =12𝜀𝑜𝑥𝑑𝑜𝑥𝑙𝑊 (2.9) where 𝑑𝑜𝑥 is the distance between the spiral and the substrate. The substrate capacitance and resistance can be expressed as, 𝐶𝑆𝑖 =12 𝐶𝑠𝑢𝑏𝑙𝑊 (2.10) LRSCOX COXCSiCSi RSiRSiCF Figure 2-4 Compact frequency-independent inductor model. 19 𝑅𝑆𝑖 =2𝐺𝑠𝑢𝑏𝑙𝑊 (2.11) where 𝐶𝑠𝑢𝑏 and 𝐺𝑠𝑢𝑏 are the substrate capacitance and resistance per unit area, respectively. Both fit the parameters and constants for a given substrate. We can improve inductor performance using a patterned ground shield (PSG). In this approach, a poly or metal layer is inserted beneath the spiral inductor and is connected to the ground. The ground shield reduces the distance between the substrate and spiral metal, and thereby reduces the effective resistance. The ground shield can then be broken to cut the eddy current loop. In other words, the ground shield should be patterned so that flux can pass through while grounding the electric field [14]. Such a PSG is shown in Figure 2-5. A PSG slightly affects inductance and increases the peak Q; however, it reduces the self-resonant frequency due to increasing parasitic capacitance since the ground shield is closer to the spiral. In general, in order to minimize PSG resistance, a metal with lowest resistance and furthest distance away from the substrate should be used. Parallel metal strips can be used to further reduce the resistance. Figure 2-5 Patterned ground shield (PGS). 20 To avoid unnecessary complex calculations, we assume a single-ended configuration where one of the terminals of the inductor is grounded, as shown in Figure 2-6 on the left side, to determine the quality factor of the inductor. On the right side of Figure 2-6, a simplified model is depicted where, 𝑅𝑆𝑖, 𝐶𝑆𝑖 and 𝐶𝑜𝑥 are replaced with an equivalent shunt resistance (𝑅𝑃) and capacitance (𝐶𝑃). 𝑅𝑃 and 𝐶𝑃 are expressed by, 𝑅𝑃 =1𝜔 𝐶𝑜𝑥2 𝑅𝑆𝑖+𝑅𝑆𝑖 (𝐶𝑜𝑥 + 𝐶𝑆𝑖)2𝐶𝑜𝑥2 ≈1𝜔 𝐶𝑜𝑥2 𝑅𝑆𝑖+ 𝑅𝑆𝑖 ≈ 𝑅𝑆𝑖 (2.12) 𝐶𝑃 = 𝐶𝑜𝑥1 + 𝜔2𝑅𝑆𝑖2 𝐶𝑆𝑖 (𝐶𝑜𝑥 + 𝐶𝑆𝑖)1 + 𝜔2𝑅𝑆𝑖2 (𝐶𝑜𝑥 + 𝐶𝑆𝑖)2 ≈𝐶𝑜𝑥𝐶𝑆𝑖𝐶𝑜𝑥 + 𝐶𝑆𝑖= 𝐶𝑜𝑥|| 𝐶𝑆𝑖 (2.13) 𝐿𝑠 does not decrease significantly with increasing frequency because it is predominantly determined by the magnetic flux external to the conductor [15]. Consequently, it is valid to model 𝐿𝑠 as a constant. Additionally, because of our small bandwidth of interest we can assume that the inductance is constant. An ideal inductor is expected to be a pure energy storage element. In reality, however, parasitic resistances result in power dissipation and parasitic capacitances reduce the inductance. Therefore, the definition of a quality factor includes a description of how an inductor works as a storage element [14]. For an inductor, the quality factor is defined as [16], COXCSiRSiLSRSCF CPRPLSRSCF Figure 2-6 Lumped one-port inductor model (left) and its equivalent (right). 21 𝑄 = 2𝜋. (𝑃𝑒𝑎𝑘 𝑀𝑎𝑔𝑛𝑒𝑡𝑖𝑐 𝐸𝑛𝑒𝑟𝑔𝑦 − 𝑃𝑒𝑎𝑘 𝐸𝑙𝑒𝑐𝑡𝑟𝑖𝑐 𝐸𝑛𝑒𝑟𝑔𝑦 𝐸𝑛𝑒𝑟𝑔𝑦 𝐿𝑜𝑠𝑠 𝑖𝑛 𝑂𝑛𝑒 𝑂𝑠𝑐𝑖𝑙𝑙𝑎𝑡𝑖𝑜𝑛 𝐶𝑦𝑐𝑙𝑒) (2.14) 𝑄𝑜𝑛−𝑐ℎ𝑖𝑝=𝜔𝐿𝑅𝑠.𝑅𝑃𝑅𝑃 + 𝑅𝑆[1 + (𝜔𝐿𝑅𝑆)2]. [1 − (𝐶𝐹 + 𝐶𝑃)(𝑅𝑆2𝐿+ 𝜔2𝐿)] (2.15) The first term here represents the series loss in the spiral. The second term accounts for the silicon substrate loss and the last term is the self-resonant factor representing the reduction in Q due to the increase in peak electric energy with increasing frequency [16]. The self-resonant frequency ωOsc is 𝜔𝑂𝑠𝑐 = √1𝐿(𝐶𝐹 + 𝐶𝑃)[1 −𝑅𝑆2𝐿(𝐶𝑃 + 𝐶𝐹)] (2.16) Given the foregoing equation, we can sketch an approximation of Q as a function of frequency, as shown below in Figure 2-7. At low frequencies, the series resistance (𝑟𝑆) defines Q. The quality factor increases linearly up to a point where the skin effect becomes significant. At high frequencies, 𝑅𝑆𝑢𝑏 shunts the inductor and limits the Q [16]. SRLskinS RRLLRsubQω LRS L RS RSkinL RS RSkinRSub Figure 2-7 Inductor model at different frequencies and corresponding Q behavior. Substrate Loss Factor Self-Resonance Factor 22 2.2 Capacitors The gate capacitance of MOSFET can be used to realize the nonlinear capacitors with highest density. MOS capacitors (MOSCAP) may be utilized where linearity and power consumption is not a concern. We note that channel resistance limits the Q of the capacitors, and gate leakage can drain the battery. Fringe capacitors may be used if the Q or linearity of MOSCAP is not adequate. 2.2.1 Varactors Varactors are commonly used in LC-VCOs (voltage controlled oscillators) to tune the resonate frequency. Two main characteristics of varactors are capacitance range and quality factor. The capacitance range is the ratio of the maximum to the minimum capacitance of the varactor. The quality factor is limited by parasitic resistances. The junction capacitance of a reversed biased 𝑝𝑛 junction is given by, 𝐶𝑗 =𝐶𝑗𝑂(𝑎 +𝑉𝐼𝑉𝐵𝐼)𝑚 (2.17) where 𝑉𝐼 is the input voltage, 𝑉𝐵𝐼 is the build-in potential, 𝐶𝑗𝑂 is the capacitance at zero bias, and 𝑚 is an empirical factor. From equation (2.17), because of the low supply voltage, the capacitance range is small. An accumulated 𝑝𝑛 junction is then preferred, with which a higher tuning range can be achieved. The CV characteristics of accumulation-mode varactors is plotted in Figure 2-8. The quality factor of the varactors is also determined by channel resistance between the source and drain. Figure 2-8 (right) also shows the lumped model for a varactor. A larger tuning or capacitance range can be achieved if the larger length is used, at the cost of a lower quality factor. The trade-23 off between the capacitance range and Q of varactors ultimately leads to another trade-off between the tuning range and phase noise of LC VCOs [12]. 2.3 Inductor Models with Temperature Effect In this section, the temperature dependency of on-chip, planar, spiral inductors is analyzed and characterized. The temperature dependence of the quality factor can be explained in the context of the temperature coefficient of the parasitic resistance. The series and shunt resistances exhibit a strong dependence on temperature and frequency. Throughout this section, we will examine how temperature affects on-chip inductors. We will then discuss the problem of temperature variation in inductors and a temperature model for inductors will be presented. As discussed previously, the inductance of a planar spiral is frequency and geometry-dependent; however, because of our frequency and the frequency range of interest, we can assume that the inductance is frequency-independent. The geometry of inductors varies with the number of turns, line spacing, line width, line thickness, and the outside radius of the inductor. Therefore, inductance behavior is well understood and is not expected to vary significantly with temperature VGSCGSCmaxCminVGSQ00RvarCvarSG Figure 2-8 CV characteristic of a MOS varactor, its Q variation and Lumped model. 24 [17]. Our simulation shows that inductance will change approximately 0.013 𝑛𝐻/°𝐶 up to the self-resonant frequency, 𝐿 = 𝐿𝑂(𝜌𝐿(𝑡𝑒𝑚𝑝𝑟𝑒𝑎𝑡𝑢𝑟𝑒 − 25) + 1) (2.18) where 𝐿𝑂 is the inductance at 25 °𝐶. Parasitic resistance, as we have mentioned, defines the quality factor of the inductor. Metal series resistance is linearly temperature-dependent. Series resistance modeling the effect of skin, however, has a lower temperature coefficient, yet it is highly frequency-dependent. Our simulations show that substrate resistance has a positive temperature coefficient for temperatures below roughly 100 °𝐶, while for temperatures greater than 100 °𝐶 the substrate resistance shows a negative temperature coefficient. Hence, Q is expected to vary with temperature as parasitic resistances change. Further, Q decreases with increasing temperature at low frequencies because of the positive temperature coefficient of series resistances. At high frequencies, the primary power loss of the inductor is dominated by the substrate resistance (where the overlap capacitances shunt out the series resistances), and the substrate resistance increases with the temperature increase. In order to modify the inductor model, we obtained a linear equation for the parasitic resistances [18]. It should be noted that, although the substrate and skin effect resistances are significantly frequency-dependent, we assume that they are constant over the bandwidth of interest. The substrate resistor shows higher order nonlinearity, as shown in Figure 2-9. This is modeled linearly owing to the limited temperature range. Metal series resistance is given by, 𝑟(𝑇) ≈ 𝑟𝑜(𝛼. ∆𝑇 + 1) (2.19) The substrate resistance can be expressed as, 25 𝑅𝑆𝑖 ≈ 𝑅𝑆𝑖𝑜(𝛽1. ∆𝑇2 + 𝛽2. ∆𝑇 + 1) → 𝑅𝑆𝑖𝑜 ≈ 𝑅𝑠𝑢𝑏(𝛽3. ∆𝑇 + 1) (2.20) The resistance modelling the metal track’s skin effect is, 𝑅𝑆𝑘𝑖𝑛𝑛(𝑇) ≈ 𝑅𝑆𝑘𝑖𝑛𝑛𝑜 ∗ (𝜂1. ∆𝑇 + 1) (2.21) Figure 2-9 Normalized substrate resistance vs. temperature. SRLskinS RRLLRsubQω LRSL RS RSkin L RS RSkinRSubLRSub-0.0034 °C -1-0.0046 °C -1-0.001 °C -1temp < 100 °C+0.003 °C -1(temp < 100 °C)Temperature Coefficient Figure 2-10 Quality factor vs. frequency. 26 where ∆𝑇 = 𝑇 − 𝑇𝑛𝑜𝑚𝑖𝑛𝑎𝑙(25 °𝐶), 𝑟𝑜 , 𝑅𝑆𝑖𝑜and 𝑅𝑆𝑘𝑖𝑛𝑛𝑜 are resistance at 25 °𝐶, and 𝛼, 𝛽1, 𝛽2, 𝛽3 and 𝜂1 are the temperature coefficients (𝛼 = 0.0034 °𝐶−1, 𝛽1 = −0.00003 °𝐶−1, 𝛽2 =0.004 °𝐶−2, 𝛽3 = 0.003 °𝐶−1 and 𝜂1 = 0.0012 °𝐶−1 ). Simulation results show that PSG can aid in reducing inductor coupling with the substrate and thereby decreases the effect of the parasitic substrate. We note that PSG reduces the electric coupling, while flux still passes through the PSG to the substrate. In other words, at very high frequencies and high temperature, the substrate’s parasitic resistance affects the quality factor. This can be explained by the fact that the coupling of the inductor to the substrate (eddy current) is neglected, as shown in Figure 2-2. The difference in temperature coefficients between metal layers (inductor layer and the patterned ground shield layer) is overridden by the substrate loss at high temperature and high frequency, as the Si substrate is more sensitive to temperature [19]. 2.4 Temperature Effects on Silicon A change in temperature can generally affect the MOSFET threshold voltage, leakage current, mobility, carrier diffusion, interconnect resistance, velocity saturation, energy band gap, current density, carrier density, and electromigration. In other words, temperature variation can impact the power, speed, and reliability of a system [20]. We will examine temperature effects on the critical parameters of MOSFETs, such as threshold voltage, leakage current, mobility, and interconnect resistance. We will show that MOSFETs can demonstrate a positive, negative, or zero temperature coefficient. The effects of temperature on the dynamic responses of MOSFETs are also provided. 27 2.4.1 Threshold Voltage A precise evaluation of temperature dependence of the threshold voltage is important, not only because of a MOSFET’s voltage-current characteristics (ie., a small change in threshold voltage causes a large change in the output current), but also because a system should be able to operate over a wide range of temperatures. Therefore, an accurate model for temperature changes and the effects of such changes on threshold voltages is required for circuit design. The threshold voltage of a MOSFET is found to be linearly increased with decreasing temperature. Accordingly, we can model the threshold voltage by, 𝑉𝑡ℎ(𝑇) = 𝑉𝑡ℎ𝑜 − 𝛼𝑉𝑡ℎ(𝑇 − 𝑇𝑜) (2.22) where 𝑇 is temperature, 𝑉𝑡ℎ𝑜 is the threshold voltage at nominal temperature (𝑇𝑜), and 𝛼𝑉𝑡ℎ(≈2.9 𝑚𝑉.𝐾−1) is the empirical parameter titled as temperature coefficient of threshold voltage. It is worthwhile to note that the threshold voltage of P-channel and N-channel MOSFETs change in opposite directions with increasing temperature, as illustrated in Figure 2-11. In addition to temperature, the threshold voltage also depends on the potential distribution of the channel. It is known that the threshold voltage for submicron transistors linearly decreases with an increase in drain voltage [21]. In this project, however, we assume that the average threshold voltage is independent of the applied voltage and only changes by temperature. It is shown that 𝛿𝑉𝑇ℎ(𝑇)/𝛿𝑇 (the threshold voltage’s sensitivity to variations in temperature increase) decreases when downscaling from 3.5 𝑚𝑉/℃ for 6 𝜇𝑚 processes to 2 𝑚𝑉/℃ for 2 𝜇𝑚 processes [22]. The threshold temperature coefficient for normal transistors in 65 𝑛𝑚 process is about 0.7 𝑚𝑉/℃ ). 28 2.4.2 Mobility The mobility of a MOSFET has a highly complex temperature dependence. At low temperatures the mobility increases as temperature increases, while at high temperatures the mobility decreases (300 – 600𝐾). There is also a region where mobility is relatively constant with increasing temperature. Our range of operation is > 300𝐾 and thus the mobility will decrease as temperature increases. µ(𝑇) = µ𝑜 (𝑇𝑇𝑜)−𝑛 (2.23) where 1.5 < 𝑛 < 2.5, and µ𝑜 is the mobility at nominal temperature 𝑇𝑜. Figure 2-11 Change in the threshold voltages of N-channel and P-channel MOSFETS vs. temperature. 29 2.4.3 Leakage Currents The total off current of MOSFETs can be divided into two groups: • Source-drain current: includes subthreshold current and punch through current. • Bulk current: includes the impact ionization effect, gate induced drain leakage current, and conventional pn-junction leakage [23]. In this section, we will only focus on subthreshold current, because it dominates in modern off-state leakage currents and is significantly-increased with the scaling of technology. Other sources of leakage currents are beyond the scope of this project, and further information can be found in [23]. The temperature dependence of gate leakage current has been shown as minor compared to that of subthreshold leakage current. When the gate-source voltage of a MOSFET is lower than 𝑉𝑡ℎ, a subthreshold current occurs. In a similar way to bipolar transistors, the carriers here distribute from areas of high concentration to areas of lower concentration, which is called the “diffusion current”. MOSFET subthreshold current can be expressed as, 𝐼𝑠𝑢𝑏 = µ𝐶𝑜𝑥𝑊𝐿(𝜂 − 1)𝑉𝑇2𝑒(𝑉𝐺𝑆−𝑉𝑡ℎ𝜂𝑉𝑇)(1 − 𝑒−𝑉𝐷𝑆𝑉𝑇ℎ ) ≈ µ𝐶𝑜𝑥𝑊𝐿(𝜂 − 1)𝑉𝑇2𝑒(𝑉𝐺𝑆−𝑉𝑡ℎ𝜂𝑉𝑇) (2.24) where µ is mobility, 𝐶𝑜𝑥 is the gate oxide capacitance, 𝑉𝑇 is the thermal voltage (=𝐾𝑇𝑞), and 𝜂 is a parameter representing capacitive coupling between the gate and silicon surfaces. 𝜂 is a fitting constant and its typical values range from 1 to 2. From equation (2.24), it is evident that the threshold voltage is considerably reduced with technology scaling and as a result the subthreshold current exponentially increases. Furthermore, mobility, thermal voltage, and threshold voltage are all temperature dependent parameters and can influence the temperature response of the 30 subthreshold current [23]. As a rule of thumb, leakage current doubles for every 10 degree rise in temperature [20]. We will later discuss a practical circuit to utilize the temperature behavior of the subthreshold current for temperature sensors. 2.4.4 Electrical Conductivity The conductivity of a semiconductor is written as, 𝜎 = 𝑞(µ𝑛𝑛 + µ𝑝𝑝) (2.25) where 𝑞 is the charge of the electron, 𝑛 and 𝑝 are charge densities of electrons and holes, and 𝜇𝑛 and 𝜇𝑝 stand for the mobility of the electrons and holes, respectively. Both the carrier density and mobility are temperature-dependent. This semiconductor conductivity is complicated and thus a brief discussion on temperature dependence of the Si conductor is provided below. Detailed discussions of the underlying physical phenomena can be found in [24]. In general, there are undoped or intrinsic semiconductors, lightly-doped semiconductors, and heavily-doped semiconductors. For intrinsic semiconductors, conductivity increases or resistivity decreases with increasing temperature. For lightly-doped conductors, up to about 1021, conductivity reduces or resistivity increases with increasing temperature. For heavily doped semiconductors, ≫ 1021, conductivity also increases or resistivity also decreases with increasing temperature. We know that the conductivity of a metal decreases with increasing temperature. This is because all charge carriers are free electrons and thus density will not alter significantly with temperature. Since resistivity is reversely proportional to conductivity, it can be expressed as, 31 𝜌 =1𝜎→ 𝜌(𝑇) = 𝜌𝑜(𝛼𝑅(𝑇 − 𝑇𝑜) + 1) (2.26) where 𝑇 is the temperature, 𝑅𝑜 is the resistance at nominal temperature 𝑇𝑜, and 𝛼𝑅 is the temperature coefficient of resistance. The effective temperature coefficient varies with the temperature and purity level of the metal. In consequence, 𝛼𝑅 is empirically fitted using the measurement data. 2.5 MOSFET Temperature Dependences As mentioned, a MOSFET can show a positive, negative or zero temperature coefficient, depending on the bias voltage. This is mainly because the carrier concentration increases while the carrier mobility decreases with increasing temperature. In this section, we will examine the effect of temperature on a MOSFETs transconductance (𝑔𝑚), on-resistance, and critical parasitic capacitances. We will then discuss the zero-temperature coefficient behavior of a MOSFET. Material 𝜌(Ω.m)𝑎𝑡 20 ℃ 𝜎 (𝑆𝑚) 𝑎𝑡 20 ℃ Temperature coefficient (𝐾−1) Silver 1.59×10−8 6.30×107 0.0038 Gold 2.44×10−8 4.10×107 0.0034 Copper 1.68×10−8 5.96×107 0.003862 Aluminum 2.82×10−8 3.50×107 0.0039 Table 2-1 Conductivity and temperature coefficient of various materials at 20 °C [61].  𝐴𝑙 and 𝐶𝑢 have relatively similar values of 𝛼𝑅 (𝑎𝑡 25 ℃) –– 0.004308 and 0.00401, respectively. 32 2.5.1 On-resistance of MOSFET The on-resistance of a MOSFET rises when temperature increases. The on-resistance is usually considered by the dominant resistance (channel resistance), although the actual resistance is a combination of many resistors in series, such as metallization resistances, wire resistances, and substrate resistances. We can write the on-resistance operating in deep triode (𝑉𝐷𝑆 ≪ 𝑉𝐺𝑆 − 𝑉𝑡ℎ) as, 𝑅𝑜𝑛 =1µ𝑛𝐶𝑜𝑥 (𝑊𝐿 )(𝑉𝐺𝑆 − 𝑉𝑡ℎ) (2.27) In this equation, the threshold voltage and mobility are temperature-dependent, and the 𝑅𝑜𝑛(𝑇) can be written as, 𝑅𝑜𝑛(𝑇) =1𝜇𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿 ) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜 + 𝛼𝑉𝑡ℎ(𝑇 − 𝑇𝑜)) → 𝑅𝑜𝑛(𝑇) =1𝜇𝑜𝐶𝑜𝑥(𝑊𝐿)(𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜)×(𝑇𝑇𝑜)+𝑛1+𝛼𝑉𝑡ℎ(𝑇−𝑇𝑜)𝑉𝐺𝑆 − 𝑉𝑡ℎ(𝑇)= 𝑅𝑜𝑛𝑜×(𝑇𝑇𝑜)+𝑛1+𝛼𝑉𝑡ℎ(𝑇−𝑇𝑜)𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜 ≈ 𝑅𝑜𝑛𝑜 (𝑇𝑇𝑜)+𝑛 (2.28) The increase in on-resistance can be used to control the leakage current. That is, the current increases as temperature rises. However, the increased on-resistance will automatically lower the current being carried [24]. In chapter 0, we will show that the temperature dependence of on-resistance is the main barrier for implementation of a low power, high-performance phase modulation system. 33 2.5.2 Transconductance (gm) of a MOSFET Transconductance (𝑔𝑚) represents the MOSFET’s sensitivity to a small change in gate-source voltages. In other words, 𝑔𝑚 is a figure of merit that shows how well a MOSFET can convert a voltage (𝑉𝐺𝑆) to a current (𝐼𝐷𝑆). 𝑔𝑚 = (𝜕𝐼𝐷𝜕𝑉𝐺𝑆)𝑉𝐷𝑆 𝑐𝑜𝑛𝑠𝑡.= µ𝑛𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜)= 𝜇𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜 + 𝛼𝑉𝑡ℎ(𝑇 − 𝑇𝑜)) → 𝑔𝑚 = 𝜇𝑜𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜) [1 +𝛼𝑉𝑡ℎ(𝑇−𝑇𝑜)𝑉𝐺𝑆 − 𝑉𝑡ℎ(𝑇)] (𝑇𝑇𝑜)−𝑛= 𝑔𝑚𝑜 [1 +𝛼𝑉𝑡ℎ(𝑇−𝑇𝑜)𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜] (𝑇𝑇𝑜)−𝑛 (2.29) The mobility will decrease 𝑔𝑚, while the threshold voltage will increase 𝑔𝑚 with increasing temperature. We will discuss in section 2.6 that threshold voltage effects are counterbalanced by threshold voltage in a manner similar to the well-known zero-temperature-coefficient (ZTC) bias point for MOSFET currents. 2.5.3 Parasitic Capacitances In general, MOSFET parasitic capacitances can be classified into two groups: overlap and junction capacitances. Shoucair [25] showed that overlap capacitances have a very weak temperature dependence (~25 𝑝𝑝𝑚/℃), while junction capacitances have a weak temperature dependence (~100~150 𝑝𝑝𝑚/℃). Shoucair [25] also formulated the temperature dependence of junction capacitances as, 1𝐶.𝜕𝐶𝜕𝑇≈ −1𝑉𝑏𝑖 + 𝑉𝑟𝑘2𝑞[ln {𝑁𝐴𝑁𝐷1.5×1033𝑇3} − 3] (2.30) 34 At 𝑇 = 300 𝐾, using 𝑉𝑏𝑖 = 0.7, 𝑉𝑟 = 0, 𝑘𝑞= 8.62×10−5 𝑒𝑉 𝐾−1, 𝑁𝐴 = 1×1016𝑐𝑚−3, 𝑁𝐷 =2×1015𝑐𝑚−3, 𝑛𝑖 = 1.45×1010𝑐𝑚−3, and 𝐸𝑔 = 1.12 𝑒𝑉, we have, 1𝐶𝐷𝑆.𝜕𝐶𝐷𝑆(𝑇)𝜕𝑇= 1.295×10−4 ℃−1 (2.31) In this project, we assume that MOSFET capacitances are temperature-independent for the first-order approximation. 2.6 Zero Temperature Coefficient The Zero Temperature Coefficient (ZTC) is a general condition where a particular device parameter or circuit performance becomes temperature-independent. It can be analytically expressed as 1 𝑥𝜕x(𝑇)𝜕𝑇= 0 where 𝑥 is the particular device parameter or circuit performance. For example, MOSFET drain-source current exhibits zero or an amount with the least sensitivity to temperature at a particular gate-source voltage. Figure 2-12 illustrates the 𝐼𝐷𝑆 of an NMOS as a function of 𝑉𝐺𝑆 in 65 nm CMOS technology. The 𝑍𝑇𝐶𝐼𝐷𝑆 operation point can be seen at around 𝑉𝐺𝑆 ≈ 610 𝑚𝑉. In addition to the MOSFET drain-source current, this behavior can be found in some circuit performances, such as oscillation frequency in an inverter ring oscillator. The oscillation frequency is temperature-independent at a particular supply voltage. For 130 nm CMOS technology, the 𝑍𝑇𝐶𝑓𝑜𝑠𝑐 of a three-stage ring oscillator operating at 2.4 GHz takes place when the supply voltage is about 0.74 𝑉 [26]. The ZTC point can also be found in other devices such as the Zenner Diode [27]. In this text, we only focus on a MOSFET’s ZTC point. Shoucair [25] has laid out the guidelines for designing a 35 temperature-independent two-stage topology of a CMOS op-amp. In [28] and [29], the 𝑍𝑇𝐶𝐼𝐷𝑆 bias point was reported for the first time, in both linear and saturation regimes. Analytical and experimental results were presented to obtain the accurate 𝑍𝑇𝐶𝐼𝐷𝑆 bias point in CMOS technology. Osman [30] also obtained a more accurate 𝑍𝑇𝐶𝐼𝐷𝑆 point considering the temperature dependency of mobility degradation within a vertical field. µ𝑒𝑓𝑓(𝑇) =µ𝑜1 + 𝜃(𝑇). (𝑉𝐺𝑆 − 𝑉𝑡ℎ(𝑇)) (2.32) where µ𝑜 denotes low-field mobility and 𝜃(𝑇) is a fitting parameter representing the applied transverse electric field. Although remarkable efforts have been undertaken to improve the accuracy of the 𝑍𝑇𝐶𝐼𝐷𝑆 by considering temperature dependence of all model parameters (such as 𝑉𝑡ℎ, µ, 𝜃 and contact resistances), all of the presented equations are not user friendly for analog Figure 2-12 Simulation results of IDS –– VGS characteristic at VDS = 0.6 V and at various temperatures (in TSMC 65nm). 36 design purposes [24]. 𝑍𝑇𝐶𝐼𝐷𝑆 can be interpreted as the bias point that compensates for the threshold voltage shift when temperature mobility is reduced. This intuitive interpretation was made by Filanovsky [31], who extracted a simple equation for the 𝑍𝑇𝐶𝐼𝐷𝑆 . Ignoring the higher order non-ideality, such as velocity saturation, we perform a simplified analysis from [31] to obtain the MOSFET drain-source current in the deep triode region as, 𝐼𝐷𝑆 ≈ 𝜇𝑛𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ)𝑉𝐷𝑆 𝑓𝑜𝑟 𝑉𝐷𝑆 ≪ (𝑉𝐺𝑆 − 𝑉𝑡ℎ) (2.33) We substitute the temperature-dependent expressions for mobility and threshold voltage in the current equation [24], 𝐼𝐷𝑆 = µ𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜 + 𝛼𝑉𝑡ℎ(𝑇 − 𝑇𝑜))𝑉𝐷𝑆 (2.34) → 𝐼𝐷𝑆 = µ𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜)𝑉𝐷𝑆 + µ𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿) 𝛼𝑉𝑡ℎ𝑇𝑉𝐷𝑆 −µ𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿)𝛼𝑉𝑡ℎ𝑇𝑜𝑉𝐷𝑆 (2.35) Thus we have, 𝜕𝐼𝐷𝑆𝜕𝑇= µ𝑜 (1𝑇𝑜)−𝑛(−𝑛𝑇−𝑛−1)𝐶𝑜𝑥 (𝑊𝐿) (𝑉𝐺𝑆 − 𝑉𝑡ℎ𝑜)𝑉𝐷𝑆 +µ𝑜 (1𝑇𝑜)−𝑛(−𝑛𝑇−𝑛−1) 𝐶𝑜𝑥 (𝑊𝐿)𝛼𝑉𝑡ℎ𝑇𝑉𝐷𝑆 + µ𝑜 (𝑇𝑇𝑜)−𝑛𝐶𝑜𝑥 (𝑊𝐿) 𝛼𝑉𝑡ℎ𝑉𝐷𝑆 −µ𝑜 (1𝑇𝑜)−𝑛(−𝑛𝑇−𝑛−1) 𝐶𝑜𝑥 (𝑊𝐿)𝛼𝑉𝑡ℎ𝑇𝑜𝑉𝐷𝑆 = µ𝑜𝐶𝑜𝑥 (𝑊𝐿) (𝑇𝑇𝑜)−𝑛[−𝑛𝑇(𝑉𝐺𝑆 −𝑉𝑡ℎ𝑜) + 𝛼𝑉𝑡ℎ (1 − 𝑛 +𝑛𝑇𝑂𝑇)] 𝑉𝐷𝑆 (2.36) Based on the ZTC bias point definition (𝜕𝐼𝐷𝑆(𝑇)𝜕𝑇= 0), at the bias in which the drain-source current exhibits zero variation with temperature, we can obtain a 𝑉𝐺𝑆 value that corresponds to the ZTC as, 37 𝑉𝐺𝑆(𝑍𝑇𝐶) = 𝑉𝑡ℎ(𝑇) +𝑇𝛼𝑉𝑡ℎ𝑛 (2.37) 𝐼𝐷𝑆 =µ𝑜𝑇𝑜2𝐶𝑜𝑥2(𝑊𝐿)𝛼𝑉𝑡ℎ2 (2.38) As a result, for 65nm technology we can see 𝑉𝐺𝑆(𝑍𝑇𝐶) ≈ 0.55 𝑉 by taking 𝛼𝑉𝑡ℎ =0.7𝑚𝑉 𝐾−1 and 𝑉𝑡ℎ = 0.4 𝑉. It can be shown that there exists two separate 𝑍𝑇𝐶𝐼𝐷𝑆 for a MOSFET; one located within the saturation, and one within the linear region. We also define the 𝑍𝑇𝐶𝑔𝑚 point, when the transconductance (𝑔𝑚 –– 𝑉𝐺𝑆) characteristics of the MOSFET remain constant when temperature varies, as 𝜕gm(𝑇)𝜕𝑇= 0. Figure 2-13 depicts the 𝑔𝑚 of an NMOS as a function of 𝑉𝐺𝑆 in 65 nm CMOS technology. The 𝑍𝑇𝐶𝑔𝑚 operation point can be seen around 𝑉𝐺𝑆 ≈ 420 𝑚𝑉 for an NMOS transistor. Figure 2-13 Simulation results of gm –– VGS characteristics at VDS=0.6 V and at various temperatures (in TSMC 65nm). 38 The 𝑍𝑇𝐶𝑔𝑚 and 𝑍𝑇𝐶𝐼𝐷𝑆 can be utilized in analog circuit design for high temperature applications. For example, 𝑍𝑇𝐶𝑔𝑚can be used to achieve stable circuit parameters while 𝑍𝑇𝐶𝐼𝐷𝑆can be used to maintain the DC bias current. We note that 𝑍𝑇𝐶𝑔𝑚 and 𝑍𝑇𝐶𝐼𝐷𝑆 are highly process-dependent and we can not obtain both conditions together [32]. 2.7 Conclusion In this chapter we have provided a lump model for passive devices in standard CMOS technology. The inductor can be modelled as an RLC circuit whose resistances are temperature-dependent while the inductance and parasitic capacitances are temperature-independent. At frequencies below 2 𝐺𝐻𝑧, the parasitic resistance increases and quality factor decreases, while at high frequencies and high temperature the quality factor increases with increasing temperature. Silicon is inherently temperature dependent. The threshold voltage, mobility, and substrate leakage current are the most important temperature-dependent parameters. A MOSFET current can exhibit positive, negative or zero-temperature-coefficients. Therefore, temperature effects can be minimized by properly biasing a transistor around the ZTC point. Similar to Shoucair [25], we can follow guidelines for designing temperature-independent circuits. We see that ZTC characteristics can be employed to design a temperature-independent circuitry. 39 Chapter 3: Low-Power VCO for Biomedical Application In this chapter, we briefly discuss basic oscillator concepts, particularly focusing on LC-VCOs (voltage-controlled oscillators). We will also present a brief discussion of the effects of temperature on the operation of LC-VCOs, and we propose a low power VCO/buffer that can be used for implantable biomedical applications. We will show that the proposed circuit can be used for OOK-pulse width modulation systems, and with circuit modification it is capable of being used in phase modulation systems as well. 3.1 RLC Circuit An ideal LC (lossless) circuit is shown in Figure 3-1. Assuming an impulse current is applied to the circuit, based on the law of conservation of energy the total energy at any point of time is constant and equal to the initial energy stored in the capacitor. That is, in an LC circuit the energy is only exchanged between capacitor and inductor. From the circuit’s point of view, the capacitor voltage can be obtained as, 𝜕2𝑣𝑐𝜕𝑡2+1𝐿𝐶𝑣𝑐 = 0𝐿𝑎𝑝𝑙𝑎𝑐𝑒 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚→ 𝑠2 +1𝐿𝐶= 0 (3.1) 𝑠1,2 = ±𝑗√𝐿𝐶= ±𝑗𝜔𝑜 (3.2) 𝑠1,2 = ±𝑗√𝐿𝐶= ±𝑗𝜔𝑜 (3.3) 40 As mentioned previously, in practice both the capacitor and inductor are lossy. For simplicity, we model the total loss as a parallel resistor as shown in Figure 3-2. In this case, the energy is exchanged between capacitor and inductor, albeit at a lower rate 𝜔𝑑 , eventually decaying to zero. 𝜕2𝑣𝑐𝜕𝑡2+1𝑅𝐶𝜕𝑣𝑐𝜕𝑡+1𝐿𝐶𝑣𝑐 = 0𝐿𝑎𝑝𝑙𝑎𝑐𝑒 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚→ 𝑠2 +1𝑅𝐶𝑠 +1𝐿𝐶= 0 (3.4) 𝑠1,2 = 𝛼 ± 𝑗𝜔𝑑 = −𝜔𝑜2𝑄± 𝑗𝜔𝑜√1 −14𝑄2 (𝑤ℎ𝑒𝑟𝑒 𝑄 =𝑅𝐿𝜔𝑜= 𝑅𝐶𝜔𝑜 (3.5) → 𝑣𝑐(𝑡) = 𝑉𝑂 𝜔𝑜𝜔𝑑e−αt cos(𝜔𝑑𝑡 + cos−1𝜔𝑑𝜔𝑜) (3.6) LC+-VC(t)Energy ωot ½ CVo2π/2 πσ jω Figure 3-1 Ideal LC circuit (left), Capacitor energy in an ideal LC circuit (center), Pole locations of an LC circuit in the s-plane (right). LC+-VC(t)Energy ωot ½ CVo2π/2 πσ jω -α jωd R-jωd e-2αt Figure 3-2 Lossy LC circuit (left), Capacitor energy in lossy LC circuit (center), Pole locations of an RLC circuit in the s-plane (right). 41 A negative resistance generated by an active circuit can compensate for the power dissipated in the resistor (due to the lossy LC) in order to sustain the oscillation. Figure 3-3 shows the feedback model of a linear LC oscillator. The active circuit is typically formed by a transconductance circuit. A start-up is usually induced to trigger the circuit, and thus the aperiodic start-up and noises can be modeled as the inputs to the feedback system, whose transfer function is [15], 𝐻(𝑠) =𝑠𝐶𝑠2 + 𝑠 (1 + 𝐺𝑚𝑅𝑅𝐶 ) +1𝐿𝐶 (3.7) From feedback theory, • If the poles of 𝐻(𝑠) are in the left half-plane, the system loop is stable ( 𝑅 > −1𝐺𝑚). • If the poles of 𝐻(𝑠) are in the right half-plane, the system loop is unstable (𝑅 < −1𝐺𝑚). • If the poles of 𝐻(𝑠) exist on the imaginary axis, the system loop is marginally stable (𝑅 = −1𝐺𝑚). Assuming an impulse current start-up is applied to the linear LC oscillator, the response of this system can be stable, unstable, or marginally stable, as shown in Figure 3-4. The marginally-stable loop produces a constant oscillation, while the stable loop will eventually decay and the unstable loop will unboundedly grow. LCRLossy TankActive Circuit (-Gm)ActiveCircuit (-Gm)Vout(t)Start-up and noiseVout(t) Figure 3-3 Linear LC oscillator. 42 We can also explore the oscillator system in terms of feedback. A negative-feedback amplifier, shown in Figure 3-5, can oscillate if the 𝐻(𝑠 = 𝑗𝜔1) = −1. That is, the circuit can sustain when the “Barkhausen’s criteria” are satisfied, |𝐻(𝑠 = 𝑗𝜔1)| = 1 (3.8) ∡𝐻(𝑠 = 𝑗𝜔1) = 𝜋 (3.9) 3.2 Temperature Effects on LC-VCO The oscillation frequency of an LC-VCO is commonly assumed as the resonant frequency of the LC-tank. In practice this is not the case. The oscillation frequency not only changes with energy losses in the resonator inductor and capacitor, but it is also affected by temperature. σ jω R > -1/Gm R < -1/GmR = -1/Gm Figure 3-4 System pole locations on the pole-zero plot and impulse response of the linear LC oscillator. H(s)X Y Figure 3-5 Negative feedback system. 43 As already explained, if the Barkhausen criteria are satisfied then the oscillation is sustained. This occurs when the loop has a phase shift of 180° and a magnitude of one. In practice, a loop–gain of greater than one is needed to build up to the steady state, and the amplitude growth will eventually cease due to the circuit’s nonlinearity. Considering the inductor model and linear LC-oscillator discussed previously, we simplify the tank as shown in Figure 3-6, where 𝑅𝐿 and 𝑅𝐶 represent the series resistor due to the losses of the inductor and capacitor, respectively. The effects of temperature on the active circuit will be discussed in further detail later in this chapter. Currently, we assume that the active circuit is temperature-independent. Therefore, 𝑍𝐿 = 𝑗𝜔𝐿 + 𝑅𝐿 (3.10) 𝑍𝐶 =1𝑗𝜔𝐶+ 𝑅𝐶 (3.11) 𝑍𝑜 = 𝑗𝜔𝑀 −1𝐺𝑚 (𝑃𝑢𝑟𝑒 𝑟𝑒𝑠𝑖𝑡𝑖𝑣𝑒 𝑀 = 0) (3.12) The operation frequency can be obtained by satisfying the Barkhausen’s criteria, 𝑌𝐶 + 𝑌𝐿 + 𝑌𝑍𝑂 = 𝑌𝑡𝑜𝑡 (3.13) 𝑖𝑚𝑎𝑔 (𝑌𝑡𝑜𝑡) = 0 (3.14) LCRCActive Circuit Vout(t) ZoRL Figure 3-6 A simplified model of LC-tank. 44 →−𝜔𝐿𝑅𝐿2 + (𝜔𝐿)2 +1𝜔𝐶𝑅𝐶2 +1(𝜔𝐶)2 = 0 (3.15) We define the quality factor of the capacitor and inductor as, 𝑄𝐿 =𝜔𝐿𝑅𝐿 (3.16) 𝑄𝐶 =1𝑅𝐶𝜔𝐶 (3.17) Therefore, we can simplify the equation as, 𝜔2𝐿𝐶 =(1 +1𝑄𝐶2)(1 +1𝑄𝐿2) (3.18) 𝜔𝑜 =1√𝐿𝐶→ 𝜔𝑂𝑆𝐶2 = 𝜔𝑜2((1+1𝑄𝐶2)(1+1𝑄𝐿2)) (3.19) This can be further simplified using first order Taylor series expansion, 𝜔𝑂𝑆𝐶2 = 𝜔𝑜2(1 +1𝑄𝐶2)(1 −1𝑄𝐿2) ≈ 𝜔𝑜2(1 +1𝑄𝐶2 −1𝑄𝐿2) (3.20) Evidently, both capacitor and inductor losses will affect the oscillation frequency presented in equation (3.20). In general, the former causes an increase in oscillation frequency while the latter decreases oscillation frequency. The quality factor of available capacitors in CMOS 65nm is usually much greater than that of the inductors at our frequency of interest. Therefore, the equation can be rewritten as, 𝑄𝐶 ≫ 1 → 𝜔𝑂𝑆𝐶2 ≈ 𝜔𝑜2(1 −1𝑄𝐿2) (3.21) 45 We conclude that the oscillation frequency deviates from the resonant frequency due to losses in the tank. Although we ignore the loss of the capacitor owing to the high Q value, the inductor’s parasitic resistors can affect the oscillation frequency, as they are highly frequency and temperature-dependent. As we know, the frequency dependence of the parasitic resistance can be ignored at frequencies below 1 𝐺𝐻𝑧. From the presented inductor temperature model in the previous chapter, we establish that as temperature increases, resistance in the spiral metal causes the quality factor to decrease, while resistance of the substrate improves the quality factor at high frequencies. In other words, at low frequencies the dominant resistance is a series metal resistor which linearly increases (or reduces the quality factor) with increasing temperature. Consequently, the frequency of operation decreases in respect to temperature. At high frequencies, however, the substrate resistance is dominant. The substrate resistance also increases with temperature (up to a point). In this case, the quality factor also rises (or the frequency of operation increases) with increasing temperature. It should be noted that using a large inductor results in large substrate capacitance and resistances. Therefore, eddy current and displacement current are significant and cannot be ignored at low frequency. As shown in Figure 3-7 (a), the frequency of oscillation is linearly increasing, and our simulation shows that at the ISM band, the frequency varies approximately +/−5% over the range of [−20 140] °C. We need to also investigate the temperature dependence of the active circuit. For simplicity, we assume that the active circuit is memory-less, meaning that it contains no reactive or energy-storage components. In order to sustain the oscillation frequency, the Barkhausen’s criteria should be satisfied. From our previous discussion, the active circuit must be able to replenish the energy lost in the tank (𝐺𝑚𝑅 > −1). If this energy conservation requirement is not satisfied, we will 46 observe a growing or decaying oscillation. We know that both the transconductance and the energy loss (𝑅) are temperature dependent; 𝐺𝑚 reduces (1Gm increases) while 𝑅 increases with increasing temperature. Therefore, we can surmise that the temperature effect of an active circuit on oscillation amplitude and frequency is counterbalanced by tank resistance, which is negligible. From another point of view, the oscillation amplitude is equal to 𝑅×𝐼, where 𝑅 is the (parallel) tank resistance and 𝐼 is the fundamental current provided by the active circuit. Assuming that the current is temperature-independent using the ZTC bias point concept, we expect that the amplitude will increase as temperature goes up. However, the transconductance decreases and because of the compressive nature of the active circuit, the oscillation amplitude cannot arbitrarily increase, resulting in the active circuit attempting to compensate for this increase. We note that in reality the “large-signal” transconductance must be used instead of the small-signal conductance. In [33] it is shown that it is practically impossible to achieve a marginally-stable state with small-signal transconductance in a linear oscillator. Yet the large-signal transcendence does indeed balance the tank loss in a nonlinear oscillator. Due to stringent power constraints, large devices should be avoided for biomedical VCOs. As discussed previously, parasitic capacitances of MOSFET are weakly temperature-dependent, thus we ignore the temperature dependence of parasitic capacitances. 47 (a) (b) Figure 3-7 Simulation results of a VCO frequency vs. Temperature (a) large inductor (after compensation) (b) small inductor. 48 3.3 Low Power VCO/Buffer for Biomedical Application We have acknowledged that the main concern of implantable biomedical devices is the power budget. Such stringent power requirements result in simplified approaches for handling wireless communication. That is, only the bare necessities are left for the transmitter and receiver. It is practically impossible to use a power amplifier in an implantable biomedical system to drive the load. On the other hand, if the VCO is directly connected to an antenna, the hostile environment of the body could affect oscillation frequency. Thus, a buffer could potentially help to enhance drivability and could provide a reverse isolation for the LC-VCO. In this case we must note that a buffer dissipates power and increases the total power consumption. Therefore, it makes sense to employ the current re-use technique to meet the power budget. We propose a LC-oscillator/buffer with enhanced load drivability. Total power consumption can be reduced by sharing the current between circuits. In our circuit, the current is shared between an NMOS-only LC-VCO and a common source buffer. Using an ac coupling network, containing 𝑅2 and 𝐶2, the buffer can be dc-isolated from the VCO while the signal can pass through. A decoupling capacitor (𝐶𝐷) is also used to avoid any coupling, especially noise, between the VCO and buffer. It should be noted that the supply noise coupled to the oscillator is lower than the conventional NMOS-only oscillator. This is due to the voltage division between the buffer and oscillator. The buffer is capable of driving of a 50 𝛺 load and can withstand a large capacitive load. Additionally, it can provide an adequate output swing while isolating the LC tank and load. Figure 3-8 illustrates the schematic of this circuit. 49 Although an off-chip bias voltage 𝑉𝐵 is used to explore and improve the effect of temperature on the LC-oscillator, one can instead use a conventional NMOS-only oscillator to decrease the area at a cost of higher temperature decency. Simulation results show that an accurate bias voltage 𝑉𝐵 can significantly enhance the temperature and supply voltage effects on the oscillator. Simulations show that the frequency of a conventional NMOS-only oscillator varies up to 650 𝑀𝐻𝑧/𝑉 with an increasing supply voltage, while our circuit varies by only 80 𝑀𝐻𝑧/𝑉. VB (Temp.)M3M4L1 L1C2 C2R2 R2L2 L2C3 C3M1M2C1R1C1R1CDC4 C4cVDSW1CLKSW1CLKCL RLCLRL Figure 3-8 Proposed LC-oscillator/buffer schematic. 50 In Figure 3-8, transistors 𝑀3 and 𝑀4 are loaded with an inductor to maximize the output swing and to provide the matching network. Moreover, the body-source of 𝑀3 and 𝑀4 are tied to ensure that there is no nonlinearity caused by the body’s effect. From our previous discussion in chapter 2, we see that 𝑉𝐷 is designed in such a way to be approximately 0.4 𝑉 in order to minimize temperature’s effect on the transconductance of NMOS transistors. The temperature coefficient of the oscillator frequency is about −0.3 𝑀𝐻𝑧/℃. If a well-regulated supply voltage is not available, one can improve efficiency using a tail current combined with Hegazi’s technique for noise reduction at the cost of output swing, higher supply voltage, and higher power consumption. However, our priority here is to minimize power consumption and maximize the output swing. In consequence, the bias current source is not being used. Patterned ground shield (PSG) planar spiral inductors are also employed. As discussed previously, PSG can improve the effect of temperature on the inductor by reducing capacitive coupling to the substrate. The quality factor and inductor values remain relatively constant at the ISM band frequency. That said, differential inductors can be used in order to enhance the quality factor and area efficiency. The VCO-buffer is fabricated in CMOS 65 nm technology (Figure 3-9). The die size is 690× 670 𝜇𝑚2 and the power dissipation is approximately 115 µ𝑊 at 0.6 𝑉 (minimum supply voltage). The buffer is connected to the load with wire bond (dashed-line). The output capacitance (𝐶𝐿) includes the parasitics from the PCB trace and bond wire. The output signal power is approximately −31 𝑑𝐵𝑚 at 0.6 𝑉 supply voltage. The simulation shows that the phase noise is 51 −110 𝑑𝐵𝑐/𝐻𝑧 at 1 𝑀𝐻𝑧 offset. The key performance metrics and a comparison with other low power oscillators for biomedical applications are summarized in Table 3-1. The proposed VCO-buffer uses an NMOS-based source-follower buffer stacked on top of an NMOS-only oscillator. It can directly connect to a 50 Ω load. This is used for a pulse width modulation-on/off keying (PWM-OOK) system for the implantable biomedical transmitter. In order to bring the leakage power down, power gating can be applied to all blocks (temperature sensor and transmitter) using thick oxide NMOS/PMOS switches. Additionally, low 𝑉𝑇ℎ transmission-gate switches at the output of the buffer are used to implement the OOK system. Given the low output load (50 𝛺), wide transistors should be employed to prevent degrading the signal transmission; although, smaller switches can be used at the cost of higher ripple and performance degradation. Charging and discharging of these wide transistors may also be problematic, and careful consideration is required − i.e. to improve switch performance, a voltage booster for the clock can be used. As discussed previously, phase modulation has improved noise performance compared to amplitude or frequency modulation. However, phase modulation requires an additional mixer or balun. Active mixers increase power consumption, while passive mixers suffer from low gain and degrade efficiency. In addition, a major problem associated with a passive mixer is the temperature-dependence of MOS switches, which can considerably reduce the signal delivered to the load. 52 Performance This work [62] [34] [63] Supply (V) ≥ 0.6 V 0.56 0.7 1.8 Power 115 µ𝑊 150 350 840 Frequency (MHz) 965 (@0.6V) 𝑀𝐼𝐶𝑆 𝑀𝐼𝐶𝑆 𝑀𝑒𝑑𝑅𝑎𝑑𝑖𝑜 Output signal power (𝑑𝐵𝑚) −31 −16 −16 −17.19 Modulation PWM/OOK OOK MSK/OOK 𝑂𝑂𝐾 Area (𝑚𝑚2) 0.46 − 0.5 0.55 Technology 65 nm CMOS 65 𝑛𝑚 𝐶𝑀𝑂𝑆 90 𝑛𝑚 𝐶𝑀𝑂𝑆 180 𝑛𝑚 𝐶𝑀𝑂𝑆 Table 3-1 Performance comparison of OOK transmitter. 690 µm670 µmPCBWire-bond Figure 3-9 Die photo of the proposed VCO/buffer. 53 Figure 3-10 Simulation results of the proposed PWM-OOK TX. 8.5 mV1.2 VDataSignal (single ended)D.E. Figure 3-11 Measurement results of the proposed PWM-OOK TX. 54 3.4 Conclusion The principle of an oscillator from the perspective of circuit point and feedback has been discussed in this section. We have explained that oscillation can sustain if the Barkhausen’s criteria is satisfied, or similarly, when the active circuit counterbalances tank loss. We have shown that, in practice, the frequency of oscillation shifts away from the resonant frequency of the LC-tank due to inductor and capacitor losses. We also have seen that the parasitic resistors are temperature-dependent, as well as the frequency of oscillation; the oscillation frequency will decrease with increasing temperature. That said, at high frequencies and high temperatures, the oscillation frequency increases as temperature increases. This occurs due to the displacement and eddy currents in the substrate. We have proposed a practical circuit that minimizes the effect of temperature while achieving high performance for low-power application. 55 Chapter 4: A Low-Power Temperature Sensing System for Implantable Biomedical Applications Chapter 5 This chapter presents a low-power system that senses ambient temperature and wirelessly transmits the sensed information to a nearby receiver. Although intended for an implantable smart coronary stent, it can be used for other sensory applications that require similar temperature sensing. A thermal sensor is presented that utilizes MOSFET’s biased in sub-threshold along with a combination of p+-poly and n+-poly resistors to generate a PTAT (proportional to absolute temperature) source that can operate at low supply voltages. Leveraging the central limit theorem, the output currents of several PTAT sources are combined to further reduce error. The resulting PTAT source is then used to implement a temperature-controlled oscillator for frequency modulation and transmission. A prototype in 65-nm CMOS can sense and transmit temperature values in the range of 30 to 50 °𝐶, with an average resolution of 1.1 𝑀𝐻𝑧/°𝐶. The sensor-transmitter system consumes 100 𝜇𝑊 of DC power and delivers −34 𝑑𝐵𝑚 of power to a 50 − Ω load. 4.1 Introduction Recent technological advances in integrated circuits (ICs) and wireless communications have revolutionized the realization of implantable sensors for health monitoring and diagnosis, and biomedical wireless telemetry [34]–[36]. We know that in practice, an implantable sensor should be small and should operate robustly inside the human body with reasonable operation longevity. Thus, the design of such a sensor and its wireless telemetry system is also driven by the 56 aforementioned traits of simplicity, having a small footprint, being lightweight, operating at low-power, and having efficient transmitter (TX) architecture. The work presented here is intended for an implantable smart coronary stent. To clarify, a stent is a mesh-like tube which is inserted into a blocked or a narrowed artery site to keep the lumen open. Conventional stents can be considered as passive elements. Recently, a class of intelligent active stents have been proposed that can be remotely heated with wireless control of their temperature. Such stents enable wireless endo-hyperthermia, which facilitates inhibition of in-stent restenosis (re-narrowing of the stented location) [37], [38]. In other words, they use moderate local heating to prevent restenosis by limiting cell proliferation [37]. To remotely warm up the stent, power is harvested from a dedicated radio frequency (RF) source that is outside of the patient’s body and is converted to heat. However, if there is no control over such heating, a thermal runaway may incur, leading to adverse effects [38]. Therefore, a robust implantable temperature sensor (TS) is required for monitoring and controlling temperature. Although modern CMOS IC technology has made it feasible to realize small-size TSs, they are typically power hungry, require relatively high supply voltages (𝑉𝐷𝐷), and are rather inaccurate, preventing their widespread use. In this work, we present a low-power TS along with a low-power 915 𝑀𝐻𝑧 frequency-modulated (FM) TX (Figure 4-1). The proposed system is designed to use the stent as an antenna for data communication. The stent is also used for energy harvesting (the energy harvesting unit is outside the scope of this paper). Temperature data is FM-modulated using a temperature-varying capacitor which sets the frequency of a temperature-controlled oscillator (TCO). The proposed system is designed and fabricated in a 65-nm CMOS process. Measurement results show that the prototype offers an average resolution of 1.1 𝑀𝐻𝑧/°𝐶 and senses temperature 57 changes in the specified range of 30 to 50 °𝐶. The TX dissipates 100 𝜇𝑊 of DC power while delivering −34 𝑑𝐵𝑚 of power to a 50 − Ω antenna. In this chapter, the proposed TS and FM TX circuit are described. Measurement results are presented and concluding remarks are also provided. 4.2 Temperature Sensor Architecture Due to the high level of integration, low cost, and digital signal processing capabilities of CMOS technology, CMOS-compatible sensors are attractive for biomedical implants. Physical and electrical properties of silicon, which are temperature-dependent, can be exploited to implement a TS. For instance, MOSFET’s threshold voltage (𝑉𝑇ℎ) and mobility (𝜇) have a negative temperature CvarPower Link Data LinkStentTemperature SensorBodyT = 37oC T = 38oCT = 36oC Figure 4-1 A temperature sensor and transmitter for smart-stent implants. 58 coefficient (TC). Also, parasitic PNP and NPN bipolar junction transistors (BJTs) in CMOS technology can be used for TS. However, BJT-based TSs operate at a relatively higher 𝑉𝐷𝐷, posing serious power limitations for a single-supply implantable system that is often powered through energy harvesting. Low-power CMOS-based TSs on the other hand suffer from increased nonlinearity and errors. Figure 4-2 shows the proposed low-power CMOS TS operating at 𝑉𝐷𝐷 = 0.6 𝑉. The sensor uses thick-oxide MOS devices (to minimize leakage) that are biased in the subthreshold region. In this region, the drain current of a transistor can be written as; 𝐼𝐷 ≈ µ0COXWLVT2(ζ − 1)eVGS−Vthζ𝑉𝑇 (5.1) Figure 4-2 Proposed low-power CMOS-based temperature sensor. VDDRN-PolyRP-PolyMN2MN1MP1 MP2RP-PolyTemp.RN-PolyTemp.RN-Poly+RP-PolyTemp.TS 1TS 2TS NTemp. SignalIOUTIINIOUT α TTS Temperature sensor59 where 𝑉𝑇 is the thermal voltage (=KTq ~ 25𝑚𝑉/°𝐶) and ζ is a parameter that represents the capacitive coupling between the gate and silicon surface. The output current of the TS can be written as: 𝐼𝑜𝑢𝑡 = VGSN1 − VGSN2RP,Poly + RN,Poly =VGSN1 − VGSN2Req (5.2) Using (5.1) and (5.2), and assuming that all transistors use the same lengths, 𝐼𝑜𝑢𝑡 can be simplified as: 𝐼𝑜𝑢𝑡 = ζVTReq × ln (WP1WN2WP2WN1) (5.3) In (5.3), there are still two main parameters, namely, VT and Req, that are temperature dependent. The TC of a poly-resistance can be made positive or negative depending on its doping Figure 4-3 Output current versus temperature for the proposed sensor. 60 [39]. Thus, as shown in Figure 4-3, a series combination of p+-poly and n+-poly-resistances can, to the first order, equalize the dependency of Req to temperature. Figure 4-3 shows the simulated 𝐼𝑜𝑢𝑡 versus temperature for the proposed TS. As can be seen from the figure, within the temperature range of 30 to 50 °𝐶 (which exceeds that of the human body), the curve is relatively linear with a resolution of approximately 5 𝑛𝐴/°𝐶. However, for robust operation, a higher order of temperature dependency, along with process variations and mismatches, must be taken into account. It should be noted that TS linearity specifications for implantable devices are not as stringent as other thermal sensors [40]–[42] due to the relatively narrow range of operating temperatures within the human body. Thus, as compared to one highly-linear and accurate sensor consuming large power and area, we implement multiple TSs that are ultra-compact, operate at low 𝑉𝐷𝐷, and consume ultra low power. According to the central limit theorem, the sum of many independent random variables tend to be distributed according to one of a small set of attractor distributions [43]. In this design, ten TSs are distributed on the chip to ensure that the adverse effects of the process variation will also be averaged out and the desired accuracy will be obtained. As each TS consumes 0.2 𝜇𝑊 (worst case at about 50 °𝐶) and occupies 0.0004 𝑚𝑚2, operating all ten TSs is preferred over choosing the best single TS, as this avoids complex calibration schemes. For the sake of completeness, Figure 4-4 compares the performance of the proposed TS with state-of-the-art designs, keeping in mind that the different temperature sensing requirements of this work. 61 4.3 Low-Power FM Transmitter Figure 4-4 shows the proposed transmitter where an FM transmission is selected over amplitude-modulated (AM) schemes due to its superior noise immunity. The aggregated TS output current of is first amplified, integrated, and converted to a voltage. Chopping is used to improve the offset and noise performance. Next, the resulting voltage signal is then used in the second stage to change the value of a varactor. The combination of the TS, the integrator, and the varactor is equivalent to a temperature-variable-capacitor (TVC), whose capacitance (𝐶𝑉𝑎𝑟) can be approximately shown as: 𝐶𝑉𝑎𝑟 =𝐶𝑉0(1 +𝜅𝑇𝑇0)𝑚 (5.4) where 𝐶𝑉0is the capacitance at room temperature, 𝑇0 = 273 °𝐾, m is an empirical fitting parameter for the varactor, and 𝜅 is a parameter dependent on the gain of the preamplifier. Utilizing the TVC Parameter This work [41] [40] [42] Technology 65 nm CMOS 0.16 μm CMOS 65nm CMOS 0.5 μm SOI BiCMOS VDD (V) 0.6 1.5 1.2 5 Error (oC) ±0.8 ±0.06 ≈ ± 2.8 ±0.6 Temperature (oC) 30 ~ 50 -55 ~ 125 -40 ~ 120 -70 ~ 225 Sensor Type Subthreshold BJT Ring Oscillator Thermal Diffusivity Area (mm2) 10*0.0004 0.16 0.0013 1 Power (μW) 10*0.12 6.9 400 3.5 Table 4-1 Temperature sensor performance summary and comparison. 62 in the resonant tank of an LC oscillator results in a TCO, whose operating frequency is expressed as: 𝑓𝑜𝑠𝑐 =12𝜋 √𝐿𝑇(𝐶𝑇 + 𝐶𝑉𝑎𝑟) (5.5) As the minimum attenuation of body tissue on RF signals occurs in the frequency range of 0.8 ~ 1 𝐺𝐻𝑧 [38], the resonant tank of the TCO is tuned at a center frequency of 915 𝑀𝐻𝑧 (Figure 4-6), which is also within an unlicensed Industrial, Scientific, and Medical (ISM) band. It can be shown that for the given oscillator, the bias current, 𝐼𝑏𝑖𝑎𝑠, is inversely proportional to the product CvarLTankStentM1M2CdRfVDDgmpgmnVoutABA BCTS NTemp. SensorsPush-Pull PAΦCh ΦCh +− RSTVDDM1 M2M3 M4ΦCh ΦCh ΦCh = LPFVin-Vin+VDDVB1CC RCVoutΦCh ChopperTS 1TS 2CMOS VCOChopper-Stabilized Op-AmpData LinkT = 37oCT = 36oC T = 38oC Figure 4-4 Proposed FM transmitter. 63 of the tank quality factor (𝑄𝑇) and inductance (𝐿𝑇) [44]. As a result, in this design we have maximized 𝑄𝑇𝐿𝑇 at 915 𝑀𝐻𝑧 to minimize the power consumption. Through 3𝐷 electromagnetic simulations, an optimized 𝐿𝑇 of 180 𝑛𝐻 at 915 𝑀𝐻𝑧 is designed with a 𝑄𝑇 of about 5. Using this overall power consumption, the TCO is minimized to ~120 𝜇𝐴. To ensure that 𝑓𝑜𝑠𝑐 varies predominantly due to temperature change sensed by the TS, an on-chip low-dropout (LDO) regulator with 30 𝑑𝐵 of supply rejection and up to 100 𝑚𝑉 of dropout is used to generate a 𝑉𝐷𝐷 of 0.6 𝑉 (from a harvested supply of 0.7 𝑉) for the LC oscillator, and small cross-coupled devices in the oscillator core are used to minimize their parasitic capacitance. However, the relatively large inductor represents a resistive loss in the inductor 𝑅𝐿, which can also introduce a positive TC to the oscillation frequency. The substrate capacitance is also temperature-dependent, but its variations are negligible for microstrip structures operating at the frequency of about 1 𝐺𝐻𝑧. In many industrial temperature sensors, a robust system must have the ability to function within a broad temperature range (typically, – 40 °𝐶 to 125 °𝐶) and the effect of temperature on the series resistance and substrate capacitance should be considered. Biomedical sensors do need to operate within a much lower temperature range (e.g., 30 to 50 °𝐶), and thus, adverse effects of temperature variations are less pronounced. We have co-designed the TSs and LC oscillator over the operating temperature range of this application (30 to 50 °𝐶) to meet the desired resolution and error requirements. The output of the TCO is connected to two current-re-used push-pull [45] amplifiers to drive the stent. For our proof-of-concept prototype, the active stent impedance is chosen to be 50 𝛺 at 915 𝑀𝐻𝑧. The push-pull amplifier delivers 3.8 𝜇𝑊 of RF power to the stent, large enough to be detected with a sensitive external reader circuit. 64 4.4 Measurement Results As a proof-of-concept, the proposed low-power sensor along with the FM-modulator is designed and laid out in a 65-nm CMOS process. Figure 5-5 shows a die micrograph with a total area of 700 m × 350 m. Figure 4-5 demonstrates transmitted output frequency versus temperature. As can be seen in Figure 5-6, for the temperature range of 30 to 50 °C, the TX tunes the output frequency by 1.1 𝑀𝐻𝑧/°𝐶. shows the output frequency of the TX at 914.4 and 926.5 𝑀𝐻𝑧 with the peak output power of – 24.48 and – 24.62 𝑑𝐵𝑚 to a 50 − 𝛺 load, respectively. The overall power consumption of the TX is 100 𝜇𝑊. Measurement results of the TX along with a comparison with prior-art TX designs are tabulated in Table 4-2. 4.5 Conclusion An ultra-low-power ultra-compact low-voltage TS is implemented using a combination of MOSFETs and p+/n+ poly resistors. The effects of mismatch, process variations, and other non-idealities are reduced by using multiple distributed sensors. Driving a varactor in an LC oscillator, the TSs realize a TCO; temperature variations, therefore, result in an FM TX. Further, a 65-nm CMOS proof-of-concept prototype has been designed for a smart stent application. Measurement results show that our proposed technique can tune the output frequency with a resolution of 1.1MHz °C⁄ , when temperature changes from 30 to 50 °C. 700 µm350 µm Figure 4-5 Chip micrograph. 65 Figure 4-6 Measured TCO frequency versus temperature. Parameter This work [1] [2] [3] [15] Technology 65 nm 90 nm 180 nm 180 65nm Supply Voltage 0.7 (0.6 after LDO) 0.7 0.7 2.1 to 3.5 0.56 Frequency Band (MHz) 915 ISM MICS MICS 433 ISM MICS Modulation FM MSK FSK BFSK OOK Data Rate 10 kb/s 120 kb/s 250 kb/s 800 kb/s 250 kb/s Output Power (dBm) –24 NA –16 –17 ~ –4 -16 PDC (μW) 100 350 400 > 10,000 150 Table 4-2 Performance summary and comparison. 66 Figure 4-7 Measured TX output at 914.4 MHz (top) and 926.5 MHz (bottom). 67 Chapter 5: Conclusion The focus of this research is on the design of low-power temperature sensing systems that can be wirelessly powered, and among other applications, can be used in biomedical implants. Although intended for an implantable smart coronary stent, the techniques presented in our research can be used in other sensory applications requiring similar temperature sensing. In [38] the optimum frequency of operation and the maximum deliverable power for restenosis telemonitoring have been discussed. Therefore, with a power budget estimation for the telemonitoring system, we have designed and implemented two separate temperature-sensing systems in a 65-nm CMOS process. The first system, which employs an analog frequency modulation technique, benefits from a simple design and its low consumption of power. This system monitors the smart stent temperature with a supply voltage of 0.7 𝑉 and power consumption of 100 µ𝑊. This system transmits the temperature data at the ISM band frequency, while offering 1.1 𝑀𝐻𝑧/°𝐶. Higher resolution may be obtained at the expense of power consumption and receiver complexity. The focus of the second architecture is on further improving adaptability and drivability of the proposed system to different loads. Our simulation results show that the second architecture (PWM-OOK) can achieve higher resolution (0.2 ℃) while also reducing receiver complexity. With an ultra-low-power design in mind, the configuration of this second system is based on pulse width modulation-on/off keying (PWM-OOK) and low head-room analog and digital blocks, which allow operation of this system with rectified supply voltages as low as 0.6 V. The total power consumption of such a system with low supply voltage is about 115 µ𝑊. The receiver system can be implemented by an envelope detector, low noise amplifier, and a counter. In 68 addition, despite the former (FM) system, PWM-OOK is more robust to the process spread. A calibration may be required to align and synchronize the transmitter and receiver. Trimming the receiver located outside of the body is much easier than adjusting the transmitter implemented inside the patient’s body. Further, the PWM-OOK system is less sensitive to the supply voltage and load variation. Preliminary measurements are performed for this system, and as expected, the system can indeed provide more efficient transmission in comparison to conventional implantable systems. Table 5.1 summarizes the performance of the designs proposed in this work. 5.1 Future Works As the main objective of this work is to monitor temperature, conducting in-vivo tests are envisioned as one of the main future goals. The proposed smart stent should be implanted inside live animal test subjects, and the effects of biocompatible coating on the performance of relevant Performance PWM-OOK FM Supply ≥ 0.6 𝑉 0.7 (0.6 𝑎𝑓𝑡𝑒𝑟 𝐿𝐷𝑂) Power (𝜇𝑊) 115 100 Frequency (MHz) 965 915 𝐼𝑆𝑀 Output signal power (𝑑𝐵𝑚) −31 – 34 Area (𝜇𝑚^2) 690× 670 700× 350 Technology 65 𝑛𝑚 𝐶𝑀𝑂𝑆 65 𝑛𝑚 𝐶𝑀𝑂𝑆 Table 5-1 Performance summary of the proposed systems. 69 systems should be explored. A reliable connection between systems and the stent is also a subject of future work. Design modifications are also envisioned to improve the performance of circuits. As discussed in Section 0, a robust bias voltage can improve the supply and temperature sensitivity of the system. Another possible improvement to such a design is to replace the single-ended inductor with a differential one. Differential inductors lead to smaller chip area and larger power saving. Moreover, a low dropout voltage regulator can further reduce the supply sensitivity of the PWM-OOK system. As discussed in Chapter 1, PSK modulation can further be implemented as an alternative to the proposed OOK system. Owing to its superior noise performance, long distances between receiver and transmitter can be achieved. Finally, if a higher resolution for temperate sensors is required, BJT-based TS, presented in Appendix A can be a possible solution. In addition, all MOSFET-based temperature sensors presented in [46] can potentially be used. 70 Bibliography [1] W. H. Ko, “Early History and Challenges of Implantable Electronics.,” ACM J. Emerg. Technol. Comput. Syst., vol. 8, no. 2, p. 8, Jun. 2012. [2] M. A. Hannan, S. M. Abbas, S. A. Samad, and A. Hussain, “Modulation Techniques for Biomedical Implanted Devices and Their Challenges,” Sensors, vol. 12, no. 12, pp. 297–319, Dec. 2011. [3] H.-J. Y. · C. van Hoof, Bio-Medical CMOS ICs, vol. 53, no. 9. 2013. [4] K. Bazaka and M. Jacob, “Implantable Devices: Issues and Challenges,” Electronics, vol. 2, no. 1, pp. 1–34, Dec. 2012. [5] “IEEE Std C95.3 TM -2002 (R2008) IEEE Recommended Practice for Measurements and Computations of Radio Frequency Electromagnetic Fields With Respect to Human Exposure to Such Fields, 100 kHz–300 GHz IEEE International Committee on Electromagnetic Safety (Standards Coordinating Committee 28 on Non-Ionizing Radiation) IEEE-SA Standards Board,” 2002. [6] Y.-H. Joung, “Development of implantable medical devices: from an engineering perspective.,” Int. Neurourol. J., vol. 17, no. 3, pp. 98–106, Sep. 2013. [7] “Energy Harvesting from Human Power a roadmap to new research challenges The Energy Harvesting Network Executive Summary,” 2011. [Online]. Available: http://eh-network.org/files/human_power.pdf. [Accessed: 01-Jan-2011]. [8] M. Yin and M. Ghovanloo, “Using Pulse Width Modulation for Wireless Transmission of Neural Signals in a Multichannel Neural Recording System,” in 2007 IEEE International Symposium on Circuits and Systems, 2007, pp. 3127–3130. 71 [9] D. C. Jeutter, “Overview of Biomedical Telemetry Techniques,” Eng. Med. Biol. Mag., vol. 2, no. 1, pp. 17–24, Mar. 1983. [10] W. Sansen, B. Puers, and R. Govaerts, “Design and realisation of an ultra-miniature hybrid for a pressure telemetry capsule,” in Proceedings of International Conference on Microelectronics, 1984. [11] K. Van Schuylenbergh, E. Peeters, B. Puers, and W. Sansen, “An implantable telemetric tonometer for direct intraocular pressure measurements,” in 1st European Conference on Biomedical Engineering, 1991. [12] B. Razavi, “RF MICROELECTRONICS Second Edition.” [13] Yu Cao et al., “Frequency-independent equivalent-circuit model for on-chip spiral inductors,” IEEE J. Solid-State Circuits, vol. 38, no. 3, pp. 419–426, Mar. 2003. [14] J. Chen, “On-chip spiral inductor/transformer design and modeling for RF applications,” Thesis Univ. Cent. florida, p. 124, 2006. [15] Hooman Darabi, “Radio Frequency Integrated Circuits and Systems - Hooman Darabi - Google Books.” . [16] C. P. Yue and S. S. Wong, “On-chip spiral inductors with patterned ground shields for Si-based RF ICs,” IEEE J. Solid-State Circuits, vol. 33, no. 5, pp. 743–752, May 1998. [17] R. Groves, D. L. Harame, and D. Jadus, “Temperature dependence of Q and inductance in spiral inductors fabricated in a silicon-germanium/BiCMOS technology,” IEEE J. Solid-State Circuits, vol. 32, no. 9, pp. 1455–1459, 1997. [18] D. He, Jenhao Cheng, and L. Chen, “Global scaling inductor models with temperature effect,” in 2008 9th International Conference on Solid-State and Integrated-Circuit Technology, 2008, pp. 520–522. 72 [19] Y. Z. Xu and J. T. Watt, “A Compact Model for Temperature and Frequency Dependence of Spiral Inductor,” Tech. Proc. 2007 NSTI Nanotechnol. Conf. Trade Show, vol. 3, pp. 674–677, 2007. [20] D. Wolpert and P. Ampadu, Managing temperature effects in nanoscale adaptive systems. Springer, 2012. [21] G. Groeseneken, J.-P. Colinge, H. E. Maes, J. C. Alderman, and S. Holt, “Temperature dependence of threshold voltage in thin-film SOI MOSFETs,” IEEE Electron Device Lett., vol. 11, no. 8, pp. 329–331, Aug. 1990. [22] F. S. Shoucair, “Scaling, subthreshold, and leakage current matching characteristics in high-temperature (25 degrees C-250 degrees C) VLSI CMOS devices,” IEEE Trans. Components, Hybrids, Manuf. Technol., vol. 12, no. 4, pp. 780–788, 1989. [23] O. Semenov, A. Vassighi, and M. Sachdev, “Leakage Current in Sub-Quarter Micron MOSFET: A Perspective on Stressed Delta IDDQ Testing,” J. Electron. Test., vol. 19, no. 3, pp. 341–352, 2003. [24] V. K. Khanna, Morgan & Claypool Publishers, and Institute of Physics (Great Britain), Extreme-temperature and Harsh-environment Electronics : Physics, Technology and Applications. . [25] F. Shoucair, “Design Consideration in High Temperature Analog CMOS Integrated Circuits,” IEEE Trans. Components, Hybrids, Manuf. Technol., vol. 9, no. 3, pp. 242–251, Sep. 1986. [26] P. Toledo, “MOSFET Zero-Temperature-Coefficient (ZTC) Effect Modeling and Analysis for Low Thermal Sensitivity Analog Applications,” no. September, 2015. [27] D. L. Butler and R. J. Baker, “Low-voltage bandgap reference design utilizing Schottky 73 diodes,” in 48th Midwest Symposium on Circuits and Systems, 2005., 2005, p. 1794–1797 Vol. 2. [28] F. S. Shoucair, “Analytical and experimental methods for zero-temperature-coefficient biasing of MOS transistors,” Electron. Lett., vol. 25, no. 17, p. 1196, 1989. [29] Z. D. Prijić, S. S. Dimitrijev, and N. D. Stojadinović, “The determination of zero temperature coefficient point in CMOS transistors,” Microelectron. Reliab., vol. 32, no. 6, pp. 769–773, Jun. 1992. [30] A. A. Osman, M. A. Osman, N. S. Dogan, and M. A. Imam, “Zero-temperature-coefficient biasing point of partially depleted SOI MOSFET’s,” IEEE Trans. Electron Devices, vol. 42, no. 9, pp. 1709–1711, 1995. [31] I. M. Filanovsky and A. Allam, “Mutual compensation of mobility and threshold voltage temperature effects with applications in CMOS circuits,” IEEE Trans. Circuits Syst. I Fundam. Theory Appl., vol. 48, no. 7, pp. 876–884, Jul. 2001. [32] M. El Kaamouchi, G. Dambrine, M. Si Moussa, M. Emam, D. Vanhoenacker-Janvier, and J.-P. Raskin, “Body-Biasing Control on Zero-Temperature-Coefficient in Partially Depleted SOI MOSFET,” in 2008 IEEE Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems, 2008, pp. 114–117. [33] Hooman Darabi, Radio Frequency Integrated Circuits and Systems. Cambridge University Press, 2015. [34] J. L. Bohorquez, A. P. Chandrakasan, and J. L. Dawson, “A 350 uW CMOS MSK Transmitter and 400 uW OOK,” IEEE J. Solid-State Circuits, vol. 44, no. 4, pp. 1248–1259, Apr. 2009. [35] J. Bae, N. Cho, and H.-J. Yoo, “A 490uW fully MICS compatible FSK transceiver for 74 implantable devices,” 2009 Symp. VLSI Circuits, pp. 36–37, 2009. [36] P. D. Bradley, “An ultra low power, high performance Medical Implant Communication System (MICS) transceiver for implantable devices,” 2006 IEEE Biomed. Circuits Syst. Conf., pp. 158–161, Nov. 2006. [37] C. Brasselet et al., “Effect of local heating on restenosis and in-stent neointimal hyperplasia in the atherosclerotic rabbit model: a dose-ranging study,” Eur. Heart J., vol. 29, no. 3, pp. 402–412, Jan. 2008. [38] Kamyar Keikhosravy, “Biomedical Telemonitoring Systems with an Emphasis on In-Stent Restenosis Monitoring,” 2014. [39] Hung-Ming Chuang, K.-B. Thei, Sheng-Fu Tsai, and Wen-Chau Liu, “Temperature-dependent characteristics of polysilicon and diffused resistors,” IEEE Trans. Electron Devices, vol. 50, no. 5, pp. 1413–1415, May 2003. [40] K. Kim, H. Lee, S. Jung, and C. Kim, “A 366kS/s 400uW 0.0013mm2 frequency-to-digital converter based CMOS temperature sensor utilizing multiphase clock,” in 2009 IEEE Custom Integrated Circuits Conference, 2009, pp. 203–206. [41] B. Yousefzadeh, S. H. Shalmany, and K. A. A. Makinwa, “A BJT-Based Temperature-to-Digital Converter With ±60 mK ( 3σ ) Inaccuracy From −55 °C to +125 °C in 0.16-μm CMOS,” IEEE J. Solid-State Circuits, vol. 52, no. 4, pp. 1044–1052, Apr. 2017. [42] C. P. L. van Vroonhoven and K. A. A. Makinwa, “An SOI thermal-diffusivity-based temperature sensor with ±0.6°C (3σ) untrimmed inaccuracy from −70°C to 225°C,” Sensors Actuators A Phys., vol. 188, pp. 66–74, Dec. 2012. [43] F. Hiai and D. Petz, “The Semicircle Law, Free Random Variables, and Entropy,” Am. Math. Soc., vol. 77, pp. 1–376, 2000. 75 [44] T. Taris, H. Rashtian, A. H. Masnadi Shirazi, and S. Mirabbasi, “A low-power 2.4-GHz combined LNA-VCO structure in 0.13-μm CMOS,” in 2013 IEEE 11th International New Circuits and Systems Conference (NEWCAS), 2013, pp. 1–4. [45] C. Ma, C. Hu, J. Cheng, L. Xia, and P. Y. Chiang, “A near-threshold, 0.16 nJ/b OOK-transmitter with 0.18 nJ/b noise-cancelling super-regenerative receiver for the medical implant communications service.,” IEEE Trans. Biomed. Circuits Syst., vol. 7, no. 6, pp. 841–50, Dec. 2013. [46] K. Souri, Y. Chae, F. Thus, and K. Makinwa, “A 0.85V 600nW all-CMOS temperature sensor with an inaccuracy of ±0.4°C (3σ) from −40 to 125°C,” in 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC), 2014, pp. 222–223. [47] K. A. A. Makinwa, “Smart temperature sensors in standard CMOS,” Procedia Eng., vol. 5, pp. 930–939, 2010. [48] Y. W. Li and H. Lakdawala, “Smart integrated temperature sensor - mixed-signal circuits and systems in 32-nm and beyond,” in 2011 IEEE Custom Integrated Circuits Conference (CICC), 2011, pp. 1–8. [49] C. van Vroonhoven and K. Makinwa, “Thermal diffusivity sensors for wide-range temperature sensing,” in 2008 IEEE Sensors, 2008, pp. 764–767. [50] S. M. Kashmiri, S. Xia, and K. A. A. Makinwa, “A Temperature-to-Digital Converter Based on an Optimized Electrothermal Filter,” IEEE J. Solid-State Circuits, vol. 44, no. 7, pp. 2026–2035, Jul. 2009. [51] Poki Chen, Chun-Chi Chen, Chin-Chung Tsai, and Wen-Fu Lu, “A time-to-digital-converter-based CMOS smart temperature sensor,” IEEE J. Solid-State Circuits, vol. 40, no. 8, pp. 1642–1648, Aug. 2005. 76 [52] P. Chen, C.-C. Chen, Y.-H. Peng, K.-M. Wang, and Y.-S. Wang, “A Time-Domain SAR Smart Temperature Sensor With Curvature Compensation and a 3σ Inaccuracy of −0.4°C ∼ +0.6°C Over a 0°C to 90°C Range,” IEEE J. Solid-State Circuits, vol. 45, no. 3, pp. 600–609, Mar. 2010. [53] Kyoungho Woo, S. Meninger, T. Xanthopoulos, E. Crain, Dongwan Ha, and Donhee Ham, “Dual-DLL-based CMOS all-digital temperature sensor for microprocessor thermal monitoring,” in 2009 IEEE International Solid-State Circuits Conference - Digest of Technical Papers, 2009, p. 68–69,69a. [54] K. Ueno, T. Asai, and Y. Amemiya, “Temperature-to-frequency converter consisting of subthreshold mosfet circuits for smart temperature-sensor LSIs,” in TRANSDUCERS 2009 - 2009 International Solid-State Sensors, Actuators and Microsystems Conference, 2009, pp. 2433–2436. [55] M. A. P. Pertijs, A. Niederkorn, X. Ma, B. McKillop, A. Bakker, and J. H. Huijsing, “A CMOS smart temperature sensor with a 3σ inaccuracy of ±0.5 °C from -50 °C to 120 °C,” IEEE J. Solid-State Circuits, vol. 40, no. 2, pp. 454–461, Feb. 2005. [56] K. Souri, Y. Chae, and K. A. A. Makinwa, “A CMOS Temperature Sensor With a Voltage-Calibrated Inaccuracy of ±0.15 °C (3σ ) From -55 °C to 125 °C,” IEEE J. Solid-State Circuits, vol. 48, no. 1, pp. 292–301, Jan. 2013. [57] G. C. M. Meijer, R. Van Gelder, V. Nooder, J. Van Drecht, and H. Kerkvliet, “A three-terminal intergrated temperature transducer with microcomputer interfacing,” Sensors and Actuators, vol. 18, no. 2, pp. 195–206, 1989. [58] A. Heidary, G. Wang, K. Makinwa, and G. Meijer, “A BJT-based CMOS temperature sensor with a 3.6pJ·K2-resolution FoM,” in 2014 IEEE International Solid-State Circuits 77 Conference Digest of Technical Papers (ISSCC), 2014, pp. 224–225. [59] I. M. Filanovsky and H. Baltes, “CMOS Schmitt trigger design,” IEEE Trans. Circuits Syst. I Fundam. Theory Appl., vol. 41, no. 1, pp. 46–49, 1994. [60] M. A. P. Pertijs and J. H. Huijsing, Precision temperature sensors in CMOS technology. Springer, 2006. [61] G. W. Stewart, Principles of Physics, vol. 41, no. 1061. Saunders College Pub, 1998. [62] J. Cheng et al., “A near-threshold, multi-node, wireless body area sensor network powered by RF energy harvesting,” in Proceedings of the Custom Integrated Circuits Conference, 2012, pp. 1–4. [63] L. C. Liu, M. H. Ho, and C. Y. Wu, “A MedRadio-band low-energy-per-bit CMOS OOK transceiver for implantable medical devices,” in 2011 IEEE Biomedical Circuits and Systems Conference, BioCAS 2011, 2011, pp. 153–156. 78 Appendices Appendix A BJT based Temperature Sensor Temperature is one of the most commonly measured environmental quantities. A temperature sensor can be found in various places, ranging from a heating system to the monitoring of perishable foods, to in-vivo applications. Integrated circuits (ICs) have facilitated the realization of low-cost, small-size temperature sensors and interface electronics on a single chip, so-called “smart sensors”. The need for robust and accurate temperature sensors for thermal management in laptops and personal computers has given a tremendous boost to the development of smart temperature sensors during the last two decades. In multi-core devices, temperature needs to be detected and adjusted to avoid overheating and irreversible damages. In addition, it can significantly boost system performance if thermal runaway is avoided. As discussed, most of silicon’s physical properties are temperature-dependent. Thus, standard CMOS technology can potentially be used to detect temperature with no additional fabrication required, such as MEMS temperature sensors. However, the design of accurate and low-cost CMOS smart temperature sensors is challenging, as output signals are small and sensitive in order to process spread and packaging [47]. The output signal is usually digitized using an analog to digital converter (ADC), which can be done using ratiometric measurements. That is, the temperature-dependent signal is compared with a reference signal to produce a digital signal. To facilitate this, sigma-delta (𝛴𝛥) and duty cycle converters are widely used, since low speed and high resolution is required. Necessary accuracy can be achieved by using dynamic element matching (DEM), correlating double sampling and chopping techniques. 79 It should be noted that the most accurate temperature sensor (TS) is not always the optimal choice. By considering information such as the required accuracy, operational range, linearity, and calibration effort, a proper choice can be made. For example, a highly accurate sensor consumes more power and area compared with a less accurate TS. If a narrow dynamic range needs to be measured, a less-linear TS can be used to save power and area. As explained in the next chapter, one can fit multiple less-accurate sensors, or an accurate counterpart, into the same power and area budget. If the thermal gradient across the chip is large, less-accurate distributed sensors may result in better overall accuracy compared to the accurate TS. This may, however, cause significant error due to the temperature gradient difference between the sense point and temperature sensor [48]. In this appendix chapter, we will briefly discuss the CMOS-compatible temperature sensors and we will present more details on BJT-based temperature sensors. We describe the design of a low power, energy-efficient, low-cost BJT-based temperature sensor and the simulation and measurement results are presented. A low power temperature sensing system for implantable biomedical application is also described. A.1 CMOS-compatible temperature sensors A sensor should be compatible with existing CMOS technology. Temperature sensors (TSs) should be generally small, dissipate low power, and be located close to hotspots to avoid the need for large amounts of routing resources and accurate measurements. Supply and substrate noise should also be minimized. In this appendix section, we will discuss the CMOS compatible sensor. 80 Thermal diffusivity (i.e. the rate of heat diffusion) of silicon can be used to sense temperature. This can be done by measuring an Electro Thermal Filter (ETF), which consists of a heater and a temperature sensor shown in Figure 1-1. In this structure, when driving at a constant frequency 𝑓𝑟𝑒𝑓, the phase shift of the ETF can be expressed as, 𝜙𝐸𝑇𝐹 ∝ 𝑟√𝑟𝑟𝑒𝑓𝐷 (A.1) where 𝐷 is the thermal diffusivity of silicon [42]. Thermal delay is determined by the ETF’s geometry and the thermal diffusivity of the silicon substrate. This type of temperature sensor is not sensitive to leakage current as it is inherently a time-based TS and thus can operate up to 160 ℃ [49]. Although ETF sensors have been shown to achieve inaccuracies less than ±0.2 ℃ (3𝜎) [50] and ±0.7 ℃ (3𝜎) [51] in 0.18 𝜇𝑚 and 0.7 𝜇𝑚 CMOS technology, respectively, with a low cost batch calibration they suffer from significant power dissipation and low speeds. SiO2Temp. SensorsHeaterrSiliconPHeatVETF Figure 5-1 Structure of a basic electro thermal filter. CounterOutTriggerClockTdelay=f (μ, VT, VDD) Figure 5-2 CMOS temperature sensor based on temperature-dependent delays of CMOS inverters. 81 The propagation delay of a chain of inverters, or alternatively the frequency of a ring oscillator, is temperature-dependent and can be used as a temperature sensor. As discussed previously, depending on the supply voltage a positive or negative temperature coefficient for the frequency of the oscillator can be obtained. The average propagation delay of a CMOS inverter, which drives a load capacitance 𝐶𝐿, may be expressed as [51], [𝜇𝑃 = 𝜇𝑁 = 𝜇] & [(𝑊𝐿)𝑃= (𝑊𝐿)𝑁= (𝑊𝐿)] → 𝑇𝑃 =(𝑊𝐿)𝐶𝐿𝜇𝐶𝑜𝑥(𝑉𝐷𝐷−𝑉𝑇ℎ) 𝑙𝑛(3𝑉𝐷𝐷−4𝑉𝑇ℎ𝑉𝐷𝐷) (A.2) From equation (A.2), clearly the delay is sensitive to the supply voltage (about 10 𝑉/℃) and process spread. Therefore, it requires two [52] or one [53] point calibration to achieve inaccuracies of about ±0.5 ℃ and ±2.5 ℃, respectively. The subthreshold current is also temperature-dependent and can potentially be used for temperature sensing. The gate-source voltage 𝑉𝐺𝑆 of a MOSFET operating in subthreshold is, 𝑉𝐺𝑆 − 𝑉𝑇ℎ = (𝜂𝐾𝑇/𝑞) 𝑙𝑛(𝐼𝐷𝐼𝑜) (A.3) where 𝜂 is the subthreshold slope factor and 𝐼𝑜 is a process-dependent parameter. After one point calibration, inaccuracies of about ±2 ℃ have been achieved from 10 ℃ to 80 ℃ [54]. Resistors are also temperature-sensitive. Poly-resistance can be proportional-to-absolute-temperature (PTAT) or complementary-to-absolute-temperature (CTAT) depending on doping, while metal resistance is solely PTAT. A resistor-based TS requires an accurate voltage that is usually formed by a bandgap circuit. Therefore, using a BJT TS is a more direct approach than using a resistor-based sensor relying on BJTs for biasing [48]. 82 The base-emitter voltage of a BJT is temperature-dependent. BJT-based temperature sensors have been used commonly over the past few decades. Although they suffer from requiring a high supply voltage (𝑉𝐷𝐷 > 1.2 𝑉), an accurate, low power, and low cost temperature sensor can be designed. We will discuss this in more details in the next section. A.2 BJT based temperature sensors BJT-based smart temperature sensors are one of the most often-used integrated sensors. Both NPN and PNP BJTs can be used for temperature sensing. However, NPN based sensors are more accurate at a lower supply voltage and can potentially be used at supply voltages below one volt [48]. The same characteristic in the band-gap circuit can be used to generate a temperature dependent signal. BJT temperature sensitivity is stable and linear, yet second order nonlinearity is small <0.5 ℃ without trim. Consequently, these sensors are the predominant choice in temperature sensing. Figure A-3 shows both parasitic NPN and PNP BJTs available in the CMOS process. Similar to MOSFET, the lateral BJT current is contaminated by channel doping due to surface traps and lattice dislocations. Thus, vertical BJTs are preferred to lateral BJTs. Depending on availability, both NPN and PNP can be used. Although an NPN BJT requires a triple well process, it can operate with the supply voltage below one volt. The basic principle of operation is shown in Figure A-4. The saturation current (𝐼𝑆) has a positive temperature coefficient, while 𝑉𝐵𝐸 has a negative temperature dependence, which is almost linear with a slope of approximately−2𝑚𝑉/℃. By one point calibration, the process spread of 𝐼𝑠 and 83 𝑉𝐵𝐸 can be compensated. The difference in base-emitter voltages of two BJTs 𝛥𝑉𝐵𝐸 is process-independent. As shown in Figure A-5, with the help of a gain factor 𝛼 (about 16 for 𝑝 = 5) we can generate a temperature-dependent 𝑉𝑃𝑇𝐴𝑇 and a temperature-independent reference voltage 𝑉𝑅𝐸𝐹. These voltages are applied to an ADC, whose output code 𝜇 is a digital representation of temperature. BJT based temperature sensors can achieve inaccuracies of ±0.1 ℃ (3𝜎) in ceramic packages after one-point calibration [55]. With no calibration, the accuracy is in the order of a few degrees depending on the process used [56]. OxideEmitterBasen-welln+p+ p+Gaten+ p+Collector (a) OxideEmitterBasen-welln+p+ p+Gate Collectorn+ p+ (b) OxideEmitterP-welln+p+Basen+CollectorDeep n-well (c) Figure A-3 Cross-section of (a) Lateral PNP BJT; (b) Vertical PNP BJT; and (c) Vertical NPN BJT. 84 A.3 Duty cycle modulation and sigma delta ADC The principle of the temperature to duty cycle converter is shown in Figure A-5. The charge at the end of each period is always the same. Therefore, the duty cycle can be calculated as follows, ΔVBE +-+-VBE2 +-VBE1 I pIα ADC µ (a) Temperature (K)VREF = VBE1 + VPTATVBE2VBE1 VPTAT = α ΔVBEΔVBE = VBE1 - VBE2Voltage (V)1.2Operating range (b) Figure A-4 Basic principle of a BJT-based temperature sensor (a) Block diagram of a bandgap temperature sensor (b) Biasing a BJT pair in a current ratio of p, the single-ended voltages are CTAT while the differential voltage is PTAT. * 𝑽𝑩𝑬𝟏 =𝒌𝑻𝒒𝒍𝒏 (𝑰𝑰𝑺), 𝑽𝑩𝑬𝟐 =𝒌𝑻𝒒𝒍𝒏 (𝒑𝑰𝑰𝑺) , 𝜟𝑽𝑩𝑬 = (𝒌𝑻𝒒) 𝒍𝒏 (𝒑) 85 𝐼1𝑇1 = 𝐼2(𝑇 − 𝑇1) (A.4) µ = 𝐷𝑜𝑢𝑡 =𝑇1𝑇=𝐼1𝐼1 + 𝐼2 (A.5) where (𝐼1 + 𝐼2) is temperature independent and 𝐼1 is PTAT. This type of converter is also called a charge balancer, as the total charge is constant at the end of each period. The great advantage of this circuit is that the absolute value of resistors and capacitor are not important. A sigma-delta analog to digital converter (ADC) can also be used to convert the temperature to the duty cycle. The difference between 𝛴∆ ADC and the duty cycle converter is that the output of the Schmitt-trigger in the former is sampled by the clocked DFF and is synchronized with the system clock. With the help of oversampling, higher resolution may be obtained at the expense of power consumption. Figure A-6 demonstrates the principle of sigma-delta 𝛴∆ ADC. CI1I2GNDVDD t (s) VV2V1VoutVoutVCT1 T Figure A-5 Principle of duty-cycle modulation. CGNDVDDVoutD Q MCClkI1I2 Figure 5-6 Principle of sigma-delta ADC. 86 It should be noted that the output signal is related to the temperature in Kelvin and it covers a wide range of temperatures. However, we are interested in the temperature range of [−55 ℃ (218 𝐾),+130 ℃ (403 𝐾)] and a large part of the dynamic range is not being used here. This problem has already been solved by Meijer [57] by shifting the signal to the desired region to maximize the dynamic range. This can be done by subtracting 𝐼𝑃𝑇𝐴𝑇 and 𝐼𝐶𝑇𝐴𝑇 instead of directly using 𝐼𝑃𝑇𝐴𝑇. A Kelvin to Celsius converter implementation of our temperature sensor is shown in Figure A-7. The capacitor 𝐶 is charged by a current 4𝐼𝑃𝑇𝐴𝑇 − 𝐼𝐶𝑇𝐴𝑇 and is discharged by a current 2𝐼𝐶𝑇𝐴𝑇 − 2𝐼𝑃𝑇𝐴𝑇. Cascode current mirrors are used to ensure accuracy over a wide supply range. Charge balancing can be applied to obtain the duty cycle as follows, 𝐼1 = 4𝐼𝑃𝑇𝐴𝑇 − 𝐼𝐶𝑇𝐴𝑇 (A.6) 𝐼2 = 2𝐼𝐶𝑇𝐴𝑇 − 2𝐼𝑃𝑇𝐴𝑇 (A.7) 𝐷 =𝐼1𝐼1+𝐼2=4𝐼𝑃𝑇𝐴𝑇−𝐼𝐶𝑇𝐴𝑇𝐼𝐶𝑇𝐴𝑇+2𝐼𝑃𝑇𝐴𝑇=4𝛥𝑉𝐵𝐸𝑅𝑃𝑇𝐴𝑇−𝑉𝐵𝐸𝑅𝐵𝐸𝑉𝐵𝐸𝑅𝐵𝐸+2𝛥𝑉𝐵𝐸𝑅𝑃𝑇𝐴𝑇→ 𝐷 =4𝑅𝐵𝐸𝑅𝑃𝑇𝐴𝑇𝛥𝑉𝐵𝐸−𝑅𝑃𝑇𝐴𝑇𝑅𝐵𝐸𝑉𝐵𝐸𝑅𝑃𝑇𝐴𝑇𝑅𝐵𝐸𝑉𝐵𝐸+2𝑅𝐵𝐸𝑅𝑃𝑇𝐴𝑇𝛥𝑉𝐵𝐸 (A.8) It can be established that the output bitstream is not sensitive to the resistor values as a ratio of resistors are being used. The ratio is set by using large devices (200 𝐾𝛺) and a careful layout. The output duty signal is a linear function of temperature and cycle, and is independent of Schmitt trigger’s (ST) threshold voltage to the first order [58]. C2IPTATICTATGNDVDDICTAT2IPTAT Figure 5-7 Kelvin-to-Celsius converter implementation. 87 A binary temperature reading can be obtained from the output bitstream and by counting the number of ones in a sequence of 𝑁 bits. The duty cycle then can be expressed as 𝐷 = 𝑁𝑇 + 𝐾 or similarly 𝐷 = 𝐴𝑇 + 𝐵 where 𝐴 = 0.0035, 𝐵 = 0.36, where 𝑇 is the temperature in degrees Celsius. Meijer et al. shows that the nonlinearity in a smart temperature sensor can be reduced by making the reference current (𝐼𝐶𝑇𝐴𝑇 + 2𝐼𝑃𝑇𝐴𝑇) slightly temperature-dependent. In addition to that, a proper choice of 𝐴 and 𝐵 ensures that the trimmed sensor has minimum curvature. Therefore, the curvature of 𝑉𝐵𝐸 is reduced and a systematic non-linearity less than 0.1 ℃ can be achieved. The PTAT current is derived from the voltage difference between the base-emitter of two BJTs, while the CTAT current is derived from the base-emitter voltage of a bipolar transistor. Figure A-8 shows the temperature sensor circuit. Substrate NPN transistors are used, as they have a better linearity behavior compared to a PNP counterpart. A PTAT current (0.1 µ𝐴 at room temperature) is generated by 𝑄1, 𝑄2, 𝑅𝑃𝑇𝐴𝑇 , and 𝑂𝑃1. The voltage difference between the base-emitters is in the order of hundreds of microvolts and can be affected by the low-frequency noise of 𝑂𝑃1. Thus, chopping is used to mitigate offset and flicker noise effects. A gain of 90 𝑑𝐵 over PVT with a current consumption of less than 2 µ𝐴 is achieved by a two-stage folded-cascode operational transconductance amplifier (OTA), as shown in Figure A-9. Similarly, opamp 𝑂𝑃2 and 𝑅𝐵𝐸 is converted from the base-emitter voltage of 𝑄3 into a CTAT current. The gain of 𝑂𝑃2 should be greater than 70 𝑑𝐵 to keep errors below 0.1 ℃. A two stage OTA can provide the gain with a power consumption of less than 1 µ𝐴. 88 Errors in the current mirrors are mitigated by dynamic element matching (DEM). The DEM and chopping state machines can be self-clocked (by Schmitt-trigger), although we used an off-chip clock in our study. The dynamic errors can be alleviated using layout techniques, such as common centroid, interdigitating, and dummy strips. The Schmitt-trigger (ST) is designed based on two RPTATQ1 Q2OP1Q3CC1CC2IBIASRBEM1 M3 M5 M7M2 M4 M6 M8 M10 M12M9 M11M14 M18M13 M17M19 M21 M23 M25M20 M22 M24 M26SW2SW1ICTAT ICTATSW3M28M27IPTAT µ M33M34Start-up (SU)RB1RB2(SU)IPTATsuQCalCalibrationM29M30CϕchopϕchopIPTAT ICTATOP1(SU)ϕchopϕchopM16M15 M31M32 Figure 5-8 Detailed circuit diagram of the temperature sensor. MT1M1 M2M3 M4MT2M7M8M9 M10M11 M12Vi-Vi+VB2VB4MB2MB1RC CCM13M14IVB3M5 M6 Figure 5-9 Two-stage folded cascode opamp. 89 inverters in series with positive feedback. More detail on ST can be found in [58], [59]. The nonlinearity of the integrator capacitor is not an issue as it is followed by a comparator [60]. The total power consumption of the circuit is around 10 µ𝐴 at continuous operation (at a nominal supply voltage of 1.2 𝑉). Powering down the system can significantly reduce power consumption and self-heating. For example, a system that is powered down 90% of the time has its power consumption reduced by a factor of ten. Therefore, we aimed for the 10 sample/s (100ms). An accurate result can be achieved by averaging out 8 periods (1.6 𝑚𝑠). Additionally, using the simplified average can alleviate the DEM and chopping residuals. The digitization can be completed using a microcontroller or FPGA. After one point calibration, an error of ± 0.4 ℃ can be achieved. Figure 5-10 shows the simulation results of the output bitstream when temperature varies from −20 ℃ to 120 ℃. The circuit complexity and calibration methodology are affected by temperature sensitivity and linearity. A sensitive converter may be required for low-temperature sensitivity at the expense of power consumption and noise performance. A repeatable nonlinear characteristic can be directly used, but temperature dependence is usually process-dependent and not repeatable. Hence, we need to linearize the response by calibrating at multiple points [48]. 390 µm350 µm Figure 5-10 Die photo of the temperature sensor 90 A BJT-based temperature sensor implemented in standard CMOS is presented in this section. A continuous-time duty cycle modulator is used whose output can be easily interfaced to a microcontroller instead of the discrete-time ∆Σ modulators. By using dynamic element matching, chopping, and a single room trimming, the sensor achieves a spread of less than ± 0.4 ℃ (3𝜎) from −20 ℃ to 120 ℃. The sensor occupies 350×390 µ𝑚2 and was implemented in a 65𝑛𝑚 CMOS process. The sensor outputs a rail-to-rail square-wave, and varies from about 1 𝑘𝐻𝑧 to 5 𝑘𝐻𝑧 at a continuous time over temperature and supply voltage. With a linear fit, the output duty cycle can be expressed as 𝐷 = 𝐴𝑇 + 𝐵 where 𝐴 = 0.0035, 𝐵 =0.36 and 𝑇 is the temperature. We note that a one-time programmable memory requires storing the trim data. Figure 5-11 Simulation results (Duty Cycle vs. Temperature) 91 Appendix B Two-Stage Folded Cascode OTA The suitable OTA configuration is selected based on requirements and boundary configurations. We will provide more details here on the design of two stage folded cascode OTA. The small signal model is shown in the figure below, Av0 = gm1(R9||R7)gm13RL (B. 1) ωp1 ≈1(R9||R7)gm13RLCC (B. 2) if ωp2 ≫ ωu → ωu ≈ Av0ωp1 =gm1CC (B. 3) ωo = √gm7gm13CLCB (B. 4) 1Q= 2ξ = √gm7CBgm13CL (1 +CLCC) (B. 5) z1,2 ≈ ±√gm13gm7CBCC (B. 6) ξωo =gm7CL(1 +CLCC) → TScriterion(2%) =4ξωo & Slew Rate =ItailCC (B. 7) gm1vi/2 gmvi/2 gm7vi/2 rd5rd2 1/gm11i10rd6rd1 CArd8R9 RL CLgm8vgs8 gm13 vgs13 i10Cc RcR2 = rd1||rd6R7CA Figure 5-12 Small signal model for two stage folded cascode OTA. "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "2017-11"@en ; edm:isShownAt "10.14288/1.0354984"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Electrical and Computer Engineering"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "Attribution-NonCommercial-NoDerivatives 4.0 International"@* ; ns0:rightsURI "http://creativecommons.org/licenses/by-nc-nd/4.0/"@* ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Low-power temperature sensing system for biomedical applications"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/62851"@en .