UFDC Home  myUFDC Home  Help 



Full Text  
A LOW VOLTAGE SUBTHRESHOLD ALL DIGITAL PHASE LOCKED LOOP FOR ULTRA LOW POWER BIOMEDICAL MICROSYSTEMS By TANUJ AGGARWAL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010 2010 Tanuj Aggarwal To my Mom, Dad, Bro and Megha ACKNOWLEDGEMENT With great gratitude I would like to thank my academic advisor Prof. Rizwan Bashirullah for giving me the opportunity to be a part of the ICR research group. His continuous support and guidance during the past two years have helped me gain insight in the field of IC design and adopt a professional approach towards my work. I would also like to thank Dr. Robert Fox and Dr. Gijs Bosman for being members of my master's thesis committee. During my stay at UF I thoroughly enjoyed classes like Bipolar and MOS IC design design by Dr. Fox, Advanced VLSI design by Dr. Bashirullah. It was a pleasure working as a teaching assistant for the Bipolar course for Dr. Fox, thank you for teaching a complicated topic like feedback in a very simple way. Special thanks to ChunMing Tang for all his support throughout the implementation of this chip. Working with him was an extremely good learning experience. I would like to thank ChungChing Peng and Quizhong Wu for teaching me everything about the synthesis and design tools. Interacting with Chris and Walker has always been fun and I will always remember all the late night dinners that we had at I Hop. Thanks Jikai Chen, Hang Yu, Xiao Zhiming and Xue Lin for your inputs with various design related issues and all the basketball sessions. I would specially like to thank Prof. Dipankar Nagchoudhuri for introducing me the world of VLSI circuits and showing his faith in me. I am deeply indebted to him. Finally I thank my roommates Manu Rastogi, Vaibhav Garg and Ravi Shekhar for always being there. Megha you have been so supportive and patient at every step, I can't thank you enough. I would like to thank my parents and my brother for all their love and support. TABLE OF CONTENTS Page A C KN O W LED G EM ENT ... ...... .................................. ......................... ............... 4 LIST OF TABLES ......... ................ .................... ............... 7 L IS T O F F IG U R E S ....................................................... 8 A B S T R A C T .............. ..... ............ ................. .................................................. 1 0 CHAPTER 1 INTRODUCTION .................. ...... ......... ......... 12 1.1 Overview ............... ......... ................... 12 1.2 Motivation ........................ .................. 12 1.3 Thesis Organization ............... ................................. 14 2 LITERATURE REVIEW ................... ............................... 15 2.1 Introduction to PLL's...... .................................. 15 2.1.1 Charge Pump Based Phase Locked Loop..................................... 16 2.1.1.1 PLL Response to a Phase Step.................................. 20 2.1.1.2 PLL Response to a Frequency Step ........... .. .. ............... 20 2.1.2 Digital Phase Locked Loops ................................... 21 2.1.3 BangBang Digital Phase Locked Loop ....................... ......... 22 2.2 Design Considerations for Low Power PLLs ........ ............................... 25 2.2.1 Low Power Oscillator.......................... ......... 25 2.2.2 Subthreshold Operation of Digital Circuits.................... 26 2.2.3 Jitter and Phase Noise in Ring Oscillators...................... 27 3 SYSTEM ARCHITECTURE .............. ..................................35 3.1 Basic Architecture of the System ......... ........................ 35 3.2 Digital Phase Locked Loop............................... ...................... 36 3.2.1 BangBang Phase Frequency Detector .................. ....... ......... 37 3.2.2 Digital Loop Filter.... ............................................ ....... 38 3.2.3 Sigma Delta Modulator ......... ............................ 40 3.2.4 D igitally C controlled O scillator............................................. 40 3.3 Sizing of Digital Circuits in Subthreshold Region ......... .... ................... 43 4 SIMULATION AND MEASUREMENT RESULTS .............. .... ...............46 4 .1 S im ulation E nvironm ent ..................................................................... .......... 46 4.1.1 Behavioral M odel................................................... .................... 46 4.1.2 Mixed Signal Simulations .............. ........................... 50 4.1.3 Synthesis of Digital Blocks ......................................... 52 4.2 M easurem ent R results ............................................. .................... .... .... 53 4.2.1 Test Measurement Setup ............. ............................... .............. 53 4.2.2 T est C ases ........................................... ............................... .. 55 5 CO NCLUSIO NS .................................... .................. ..... .......... 62 APPENDIX A PHASE NOISE SIMULATION IN CADENCE................................. ............... 63 LIST O F REFERENCES ................................................................... ....... 65 B IO G RA PH ICA L SKETC H .......................................................................... 69 6 LIST OF TABLES Table page 21 Performance summary of various low power oscillators................ ............... 25 22 Comparison of phase noise at 1 MHz offset frequency at different oscillation frequencies .............. ...... ............ ...... .................... ....... 34 41 Comparison of theoretical and measurement results for ring oscillator.............. 58 LIST OF FIGURES Figure Page 21 Block diagram of a PLL .......... ...... ......... ....................... .............. 15 22 Charge pump PLL ................................. ......... 16 23 Magnitude response of the closed loop transfer function for different damping factors................................... ........ ........... 18 24 Magnitude response of the error transfer function for different damping factors.......................................... ............... 19 25 D igital phase lock loop................................................................... ......... 22 26 Bangbang PLL.............................. ............... 23 27 Single ended ring oscillator with identical N stages............... ............... 28 28 An inverter stage in the ring oscillator..................................... ...................... 29 29 Noise represented by a parallel current source in a transistor........................ 30 210 Ring oscillator to verify the phase noise model........................... ............ ... 32 211 Comparison of simulated and expected phase noise for a ring oscillator running at 34.8M Hz ........... ........... .......................... ....... ........ 33 31 Block diagram of a typical biomedical microsystem with sensing and data communication capability ............ ......... .......... ...... .......... 35 32 Im plem ented digital phase lock loop.............................................. .. ................... 36 33 BangBang PFD ............. ......... .................................. ............... 38 34 Digital loop filter............................. ....... ................ 39 35 S igm a delta m odulator ............................ ......... ................ .............. 40 36 Digitally controlled oscillator .................. .................... ............... 42 37 Frequency versus control w ord.................................................... .... .. ............... 42 38 ION to IOFF ratio in subthreshold region........................... ....... ......... ......... 43 39 Leakage induced problems in a NAND gate operating in subthreshold region 45 41 Control word versus time for two different values of Kp.................................. 48 42 Control word versus time with the sigma delta modulator disabled .................... 49 43 DPLL response to a frequency step ............................................ 49 44 DPLL response to a phase step .......................... ..... ... ............... 50 45 Mixed signal simulation for lock acquisition ....... ..... ... ..................... 51 46 Mixed signal simulation for frequency step........ ....... ...... ............... 52 47 Chip layout .................... ...... .......... ........................... ........ 53 48 Die photo of passive transceiver with DPLL ............ .. .... .................. 54 49 Packaged chip mounted on a PCB for testing ...... ........ .................... 54 410 Measured waveforms for reference frequency of 40kHz and multiplier ratio set to 16 ...... ................... ............................... 55 411 Spectrum of PLL output with oscillation frequency of 640kHz.......................... 56 412 Jitter histrograms for 40kHz reference clock A) Divided clock B) PLL output clock ............. ............................................ .......... ....... 56 413 Spectrum of PLL output with oscillation frequency of 321kHz.......................... 57 414 Jitter histrograms for 20kHz reference clock A) Divided clock B) PLL output clock ............. ............................................ .......... ....... 57 415 Peak to peak jitter for different supply voltages and oscillation frequencies....... 59 416 PLL lock range and power consumption at different supply voltages A) typical typical TT design corner B) slowslow SS design corner................................. 61 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science A LOW VOLTAGE SUBTHRESHOLD ALL DIGITAL PHASE LOCKED LOOP FOR ULTRA LOW POWER BIOMEDICAL MICROSYSTEMS By Tanuj Aggarwal August 2010 Chair: Rizwan Bashirullah Major: Electrical and Computer Engineering Biomedical microsystems have very stringent space and low power constraints and need to accomplish data sensing and communication in a cost effective way. A clock signal is required by onchip analog and digital components for sampling or processing the gathered data as well as for synchronizing the system. Although the power constraints are rigid, the clock frequency requirements are relaxed as the majority of biomedical signals reside at low frequencies. Thus low power, low area, low frequency, low voltage operation and simple implementation are the key features required by a clock generator in order to be incorporated in implantable biomedical systems or sensor networks. Offchip components such as crystal oscillators can provide a very stable clock, but they are not suitable for these systems mainly because of their large size and high power consumption. Monolithic oscillators such as relaxation oscillators which consume low power and occupy significantly low area are preferred for clock generation. An efficient frequency calibration scheme is also needed to reduce drift in the oscillation frequency due to process variations [1]. Motivated by these factors, we present the design of a very low power subthreshold digital phase locked loop (DPLL) employing a ring oscillator, for clock generation in biomedical microsystems. The DPLL can be used as a frequency multiplier with programmable gain factors to generate a signal at higher frequencies which is an exact multiple of the reference clock. Thus the local clock of the system can be synchronized to a wireless low frequency signal and enable data communication. The main advantage of a digital implementation is that it remains functional even as the operating voltage is scaled down for decreased power consumption. A passive transceiver system employing the proposed DPLL was implemented in a 130nm CMOS process. It has a tunable output frequency range of 384 kHz 1.54 MHz. All blocks of the DPLL operate from a 260 mV supply in the subthreshold region and consume an average power of 200 nW while producing an output frequency of 1.28 MHz. A Register Transfer Level (RTL) behavioral model of the DPLL was developed and its functionality was verified using mixed signal simulation tools in the Cadence suite. All blocks besides the digitally controlled oscillator (DCO) were implemented using synthesis and automated place and route tools. CHAPTER 1 INTRODUCTION 1.1 Overview The focus of this work is on the usage of a low voltage digital PLL as the clock generator block in low power biomedical microsystems. PLLs have traditionally been made from analog building blocks, but with recent advancements in IC design technology the focus has shifted to their digital implementation as a result of various advantages such as reduced power and area. Low power and area are the primary requirements of a clock generator in biomedical systems as they have limited power resources and should incorporate functionalities such as sensing, data processing and communication in a simple and efficient way. Operating circuits in the subthreshold region reduces the power consumption significantly. Due to the relaxed frequency constraints in the targeted applications, sub threshold logic can be used for implementing the digital circuits. Motivated by these factors we investigated the use a subthreshold DPLL for synchronized clock generation in biomedical systems. Behavioral modeling and mixed signal simulation methodologies have been employed to quickly validate the functionality of the design. Automated place and route tools have also been used to reduce the turnaround time. Also, the behavioral model of the DPLL can be easily ported to other technologies without much modification. 1.2 Motivation Wirelessly powered embedded systems with data transfer capability have been widely used in various applications. For example, a design of an implantable device which mimics the functionality of a photoreceptor is presented in [2], while a remotely controlled implantable drug delivery system is described in [3]. In both these applications, the DC supply voltage is generated by rectification of the RF signal received at the antenna interface. In order to maximize the communication range, these systems require high RFDC conversion efficiency and low power consumption building blocks. A local clock is required by the baseband signal processing block for decoding the received data as well as for sending the generated data back. It can also be used by an analog interface which is connected to a temperature or humidity sensor in a wireless monitoring application. Reconfigurable property of the local clock can also prove to be very useful. With this additional feature, the system can adapt itself to communicate with other systems working at a different clock rate. Also, onchip data can be modulated and sent back at a frequency away from the carrier to mitigate the effect of interference. The clock frequency generated in these systems can drift with process and temperature variation, leading to synchronization failures. In this situation, the remotely sent commands cannot be interpreted properly to perform the desired operation by the system and the interrogator is unable to recover the backscattered data sent from the monitoring device. Thus, a careful design procedure is required to keep the power consumption and overall cost of implementation down. Various oscillator topologies with different tuning techniques have been proposed in the past to address these issues. However, PLLs have not been incorporated mainly because they are considered to consume too much power and area [4]. An extremely low power Digital Phase Locked Loop (DPLL) with a negligible area overhead is presented in this thesis to lock the oscillator to a desired frequency. 1.3 Thesis Organization This thesis presents a design of a very low power DPLL suitable for biomedical microsystems and addresses some of the implementation issues. A brief overview of analog and digital PLLs is presented in Chapter 2 along with a literature survey of various DPLL designs and oscillator topologies. A detailed analysis for deriving the timing jitter expressions for single ended ring oscillators working in the subthreshold regime is also presented in this chapter. Implementation details of the system are described in Chapter 3 for explaining the design choices for each block. A functional verification methodology along with various simulation and measurement results is described in Chapter 4. Finally, various conclusions are discussed in Chapter 5. CHAPTER 2 LITERATURE REVIEW 2.1 Introduction to PLL's A Phase Locked Loop (PLL) is a control system with negative feedback that synchronizes an output signal with respect to both frequency and phase of an input signal. In locked state, the phase difference between the PLL's output and reference signal is either zero or remains constant [5]. A general block diagram of a PLL is shown in Figure 21, which consists of three main blocks, namely a phase detector (PD), a loop filter and a tunable voltage controlled oscillator. Initially, the PLL is in an unlocked state and the phase detector compares the two signals Uin(t) and Udiv(t). Over time as the phase error (Uerr(t)) builds up, the system adjusts the oscillator frequency such that this phase error is mitigated(ideally zero). The closed loop phase transfer function for the system is given by equation 21. Os,,(s) Kd F (s)/ N H( (s) s + KK, F(s)N Where Ko, Kd /s and F(s) are Laplace transforms of transfer functions of the phase detector block, voltage controlled oscillator and the loop filter respectively. (t) ............................................... Uin(t) I, [. Figure 21. Block diagram of a PLL In order to understand certain system concepts it is important to consider a particular PLL implementation. 2.1.1 Charge Pump Based Phase Locked Loop The charge pump based PLL shown in Figure 22 is a classic example of an analog implementation. It consists of the combination of a phase frequency detector (PFD) and a charge pump which either charges or discharges the loop filter capacitor through switches (S1 and S2) based on the UP and DOWN signals. A phase detector block is an integral part of PLL systems as shown in Figure 21, however a frequency detection loop can be used in conjunction with the phase detector loop to increase the acquisition range [6]. Sequential logic phase and frequency detectors with a charge pump circuit are a cost effective solution to increase PLL system performance [7]. finI C 12   Divide by N L  Counter %  I Figure 22. Charge pump PLL The average value or duty cycles of the UP and DOWN signals can easily identify whether the input reference frequency (fin) is less than or greater than the output frequency (fout). Initially when fdiv is far away from fin, the PFD acts as a frequency detector and continuously pumps current in one direction into the loop filter to charge the capacitor which increases the output frequency fout. When fin and fdiv are close, the PFD acts as a normal phase detector and a steady control voltage is supplied to the oscillator to maintain the locked state. This discrete time system can be approximated as a continuous time system as described in [6] whose open loop and closed loop transfer functions are given by equations 22 and 23. H (s)op R + 1 Kvco (22) e" 2z Cs sN Since the open loop transfer function has two poles at the origin one due to the oscillator and the other one due to the loop filter, it is a Type2 PLL. IKv (RCs +1) H(S)closed 2nCN K(23) s2+ v CRs + VC 2z N 2nC N The closed loop transfer function can also be expressed in terms of damping factor ( ) and natural frequency ( c,) as in equation 24 and easily analyzed as second order systems. H(s) = 2 2ons 2 (24) S2 + 2Gos + o,2 By comparing equation 23 with 24, the natural frequency and damping factor can be expressed as in equations 25 and 26, where R and C are the passive components of the loop filter and 11 and 12 are the charge pump currents (1=11=12). O IKvc (25) 2zCN R ICK =R C, (26) 2 27rN A magnitude response of the closed loop transfer function H(jm)l for different damping factors is shown in Figure 23. The frequency axis is normalized by the natural frequency co, which allows this plot to be valid for second order Type 2 PLL's in general [5]. 10 9 15  =0.2 __ 5 =0.5 0 20 =0.707 e=1 25 =2  =4 30 I 10i 100 101 Normalized Frequency (clm / Figure 23. Magnitude response of the closed loop transfer function for different damping factors Low values of z (less than 1) result in poles with complex values, and the peaking in magnitude is mainly due to poles being located close to the imaginary axis of the s plane. However at higher values of z, the slight peaking in IH(j)) arises mainly due to the location of the zero. By increasing loop gain (in turn damping factor) the nearest pole and zero come closer, effectively reducing peaking of IH(j]) [8]. We can see in Equation 23 that the system has a zero at (1/RC) which is necessary to make the system stable. Without the resistor in place, the system will simply have two poles at the origin which will contribute to a total phase shift of180 at the unity gain frequency, making the system oscillatory. In general for a typen PLL, n1 zeroes are required to make it stable [8]. By choosing values of R, I and Kvco, a sufficiently high damping factor can be realized and location of the zero can also be adjusted. As stated earlier, a PLL system tends to minimize the phase error between the input (fin(t)) and the divided signal (fdiv(t)). The error transfer function is given by equation 27 and the magnitude plot is shown in Figure 24. For input frequencies less than the natural frequency (co,,), the phase error is quite small and is further suppressed by keeping the damping factor high. H(s) = 1 H(s)= 2 S 2 (27) s2 + 2c,,s + c,2 0 1 =0.707 .o =2  __40_ 50 10 100 101 Normalized Frequency (/o ) Figure 24. Magnitude response of the error transfer function for different damping factors The performance of charge pump PLLs is limited by various circuit level non idealities such as current mismatches between 11 and 12, as well as clock feed through. All these factors introduce ripple into the control voltage which disturbs the locked state of the PLL. Normally the loop bandwidth is kept around one tenth of the input frequency. Thus for low frequency signals, the size of the loop filter capacitor and resistor has to be made very large to meet this bandwidth criteria. 2.1.1.1 PLL Response to a Phase Step When the PLL is in a locked state and there is a change in the phase of the input reference clock, the PLL is always able to recover from such a perturbation. This can be understood by the following analysis. If a phase change of AO occurs in the incoming signal as shown in equation 28 (Laplace transform shown in equation 29), The error transfer function of the PLL is given by equation 210. By substituting s=0 it can be seen that the error always evaluates to zero. O,(t)= u(t)A (28) 0i(s) = (29) S lim (t)=Iim A (210) t sm Soso +KoKd F(s) N s 2.1.1.2 PLL Response to a Frequency Step If the input frequency changes by a factor of A,, then the PLL experiences a phase change of Ac *t at the input. The Laplace transform for this phase change is given by equation 211 and the error transfer function is given by equation ,(s) = (211) SS2 lim (t)=Iim N2 (212) sm e s +)KoKd F(s)IN s2 As can be seen for s=0, the steady state error does not reduce to zero. This means that there will always be a steady state error due to the frequency step at the input which can be reduced by keeping the loop gain of the PLL high. 2.1.2 Digital Phase Locked Loops Several digital PLLs for different applications have been reported in [911], [12] because of their numerous advantages, some of which are lower chip area, lower power consumption, faster behavioral simulations and most importantly robustness against technology parameter variations. A block diagram for a typical linear DPLL is shown in Figure 25. It consists of a phase detector followed by a time to digital converter (TDC) to digitize the phase error followed by a digital loop filter and a digitally controlled oscillator. The resolution of the TDC is critical as it determines the amount of phase error that can be measured and ultimately filtered out. There are several ways to implement a TDC. A simple version can be made by using inverters as delay elements in the signal path as in [13]. Two counters clocked by a high frequency signal have been used in [9] for digitizing the phase error. The sdomain loop transfer functions for this class of DPLL can be obtained by applying the linear analysis techniques already described in Section 2.1.1 however, a discrete time zdomain model can more accurately predict the system behavior [14] since it is a sampled system. A detailed design procedure is described in [15] to obtain the value of various loop parameters by using an analogy between a charge pump PLL and a linear digital PLL. Also, the s domain transfer functions can be converted to the continuous time zdomain model by using bilinear transforms. The main advantage of a digital PLL is that it requires a considerably small loop filter compared to that of the charge pump based PLL. The digital components can also be operated at lower supply voltages to reduce the power consumption. Digital Phase Converter  I t I I PD TDC Loop Filter DCO , t ^ fdiv Divide by N _ Counter  Figure 25. Digital phase lock loop 2.1.3 BangBang Digital Phase Locked Loop A PFD with a TDC is not only hard to design but also occupies substantial chip area and consumes excessive power, therefore it is not suitable for low power applications. An alternative to this approach is to use a binary phase and frequency detector (BPFD) where the phase/frequency difference is represented by only a single bit. Several bangbang PLL (BBPLL) designs are described in [10], [16] and [17]. Some limitations of this topology include increased frequency acquisition time and limited jitter performance. A block diagram for a BBPLL with a proportional (Kp) and integral (KI) path loop filter is shown in Figure 26. The single bit PFD output indicates whether fin is leading or lagging the fdiv signal. This information is used by the loop filter which operates at the divided down clock frequency to generate a control word to adjust the DCO frequency. In the locked state, a BBPLL does not maintain a fixed phase difference, rather the control word changes between two relatively close values on each reference clock cycle. Because the BPFD makes the system highly nonlinear, system dynamics of a second order digital BBPLL cannot be analyzed using either s or zdomain models. Some of the key results of the timedomain analysis of [18] are discussed here. One of the necessary conditions for locking is given by equation 213.  fin I Loop Filter K E out fdiv BPFD DC O DCO ,^ K __ Divide by i N Counter I I \__ Figure 26. Bangbang PLL 1< T, NT (213) 0 NKpK T Where Tr is the reference clock period, TDCO is the DCO free running period, KT is the period gain constant of the DCO and NKpKT is the quantization step of the divided down clock. If this condition is not met, the time difference (At) between Tr and the divided down clock period (TDIv) will never converge and the PLL will not lock. This implies that if the free running oscillator frequency is initially far away from the reference, the proportional path constant has to be sufficiently high. The ratio of the proportional (Kp) and integral (KI) path constants determine the stability of the BBPLL. The other necessary condition for locking is given by equation 214, where D is the delay in the loop. K 2 K < (214) Kp 2D +1 The expression for the peak to peak jitter (Jpp) is given by equation 215 and it can be seen here that by minimizing K, D and N, the jitter can be reduced in the system. Increasing KI helps in decreasing the locking time but this comes at the expense of increased jitter. Although having a higher value of Kp helps in ensuring stability, it cannot arbitrarily be kept very high as it also increases the quantization step of the proportional path which results in more jitter. J,, = NKK, 2(1+ D) + (1+ D) + (1+ D)3 + O j (215) {Kp K, Kp These findings are in close agreement with those of [16] which mainly highlight the effect of the loop filter parameters on the stability of a low power, compact and low jitter DPLL. In the locked state, a linear discrete time model has been used to account for the results. The loop filter transfer function for the integral and proportional path digital filter (F(z)) is given by equation 216 and the closed loop transfer function for the entire DPLL (H(z)) is given by equation 217. From the root locus analysis described in this paper, it was shown that increasing Kp drives the system towards stability while increasing KI mainly affects the closed loop bandwidth. F(z)= Kp +K (216) 1z Kdco (Kp + K,) K + K H(z)2 2 K=d(K +K 1 Kd K (217) 2 K dco(KP +K) K  dcoK N N 2.2 Design Considerations for Low Power PLLs After gaining an insight in the working of a PLL we now discuss some of the design issues that need to be considered for low power design. 2.2.1 Low Power Oscillator Several onchip clock generation schemes for low power applications have been reported, some of which are summarized in Table 21. Table 21. Performance summary of various low power oscillators Reference Oscillator Supply Power Operating topology Voltage Consumption Frequency [19] Injection locked 0.5 V 6.7 pW 52 kHz625 kHz divider / [20] Relaxation 1 V 1.5 pW 52 kHz625 kHz [21] Relaxation 0.8 V 320 nW 1.52 MHz [22] Current starved 1.5 V 40 pW 2.2 MHz with digital caliberation [23] Current starved 0.8 V1 V 191 nW306 nW 1.28 MHz ring [4] Current starved 0.4 pW 500 kHz ring [20] Current starved 0.7 V1.2 V 200 nW 2.45 MHz ring Ref. [19] describes a dualpath clock generator composed of injection locked dividers and a RC resonator. Here, the reference clock is derived directly from the RF carrier which guarantees high accuracy, however this scheme consumes too much power (7 pW approximately). By using a RC relaxation oscillator described in [21], the power consumption can be reduced. However, a huge area is required by the onchip resistors and capacitors. Since, the output frequency is mainly determined by the value of these passive elements, it is unreliable. A voltage controlled oscillator consisting of a current starved ring oscillator with digitally calibrated bias current has been used in [22]. Although the ring oscillator alone consumes about 9.5 pW, the digital calibration scheme requires about 31 pW of power. Current starved ring oscillator based topologies described in [4] and [23] can be a good choice as they strike a balance between low power, area and frequency deviation [20] and also because they do not rely on passive components such as resistors, capacitors and inductors [24], [27] and [29]. LC oscillators are also a popular choice in PLLs because of their superior phase noise properties. However, in order to obtain a low oscillation frequency in the range of a few Megahertz, the size of the onchip inductor and capacitor has to be kept substantially high. Also, the tuning range of LC oscillators is only in the range of 1020% [25]. Thus, an LC oscillator cannot cater to the requirements of biomedical sensor network systems. Ring oscillators are an attractive alternative mainly because of their simple architecture, low area, wide tuning range and ease of integration. Here, an odd number of inverters are connected in feedback to generate a periodic signal whose frequency is determined by the delay of each inverting stage. By increasing the delay of each cell, low oscillation frequencies can be easily obtained. However, this comes at the expense of a poor phase noise resulting in timing jitter, as single ended ring oscillators are more susceptible to variations in supply voltage. Due to the low speed requirements in the biomedical systems, this timing jitter can be tolerated since the emphasis is on low power consumption. Although ring oscillators with differential delay cells are more immune to various noise sources, they are not suitable in the applications of interest mainly because of their high power consumption and area requirements. 2.2.2 Subthreshold Operation of Digital Circuits Subthreshold operation refers to operating circuits at a supply voltage (VDD) lower than the threshold voltage (VT) of a transistor. It involves charging and discharging the load capacitor with the subthreshold leakage current and is able to achieve minimum energy consumption with limited speed performance [2627]. It has been incorporated in low power applications such as a FFT processor and hearing aids [2829], and can be applied in biomedical microsystems which have very limited power available. For the 130nm technology, the threshold voltages for nMOS and pMOS are 0.38mV and 0.33mV respectively. The expression for subthreshold leakage current is given by equation 218 [29]. GS VT VDS IDS oDSOe nVth (1_e ) (218) Where IDSO is the drain current when VGS is equal to VT (equation 219) [30] W DSO Cox (n 1)V2 (219) L Vth is the thermal voltage and n is the subthreshold slope factor given by equations 220 and 221 respectively. kT Vth = (220) q n =1+ Cd (221) Cox For VDs >4Vth, equation 218 can be reduced to equation 222. At higher values of VDs the exponential term becomes negligible. VGS VT ID = IDOe nVth (222) 2.2.3 Jitter and Phase Noise in Ring Oscillators Besides power savings, there are some additional benefits of using single ended ring oscillators (Figure 27) in terms of spectral characteristics. Ref. [31] derives the expressions for phase noise and timing jitter for both single ended and differential ring oscillators by using impulse sensitivity functions (ISF is a timevarying constant that can determine the phase shift due to a noise source) and states that single ended ring oscillators have lower phase noise than their differential counterparts for a given power and frequency. It also states that the timing jitter in single ended oscillators can be minimized by equalizing the rising and the falling times. Total N Stages > 1 2 N Figure 27. Single ended ring oscillator with identical N stages Ref. [32] derives the expression for phase noise in ring oscillators in terms of power dissipation, temperature, frequency of oscillation and offset frequency by analyzing the time domain jitter. Some of the key results are discussed here. The variance of timing jitter for switching based relaxation and ring oscillators is directly proportional to the variance in the control voltage at the input. This variance in the control voltage at any time t is given by equation 223 [32]. (2 kTR ( 2tRC 2 2t/RC (223) CRn Where k is the Boltzmann constant, T is the temperature, R and C are the net resistance and capacitance at the input node, Rn is the equivalent thermal noise resistance and o2 is the variance of the control voltage at t=0. The variance in switching time jitter can be calculated by substituting equation 223 in equation 224 [32]. 2 2 dV 2 AT2 = AVc(t) 2 (224) In the model used for calculating the expressions for timing jitter of a ring oscillator similar to the one shown in Figure 27, each inverter (Figure 28) can either be in ON or OFF states. In the ON state, a constant current I either charges or discharges the load capacitor while in the OFF state no current is drawn from the supply. 1 2 Figure 28. An inverter stage in the ring oscillator In the ON state the output resistance is l/gdswhile in the OFF state the output resistance will be 1/gdo. Where gdO is given by equation 225. gd0 = gds VDS (225) If the oscillator is working in the subthreshold region then the output conductance gds can be calculated as in equation 226 by substituting the value of IDS from equation 218. GSVT VDS dls I e nVth .e Vh g, = DSO (226) dVos Vct The output conductance (gdo) at VDS =0 is then given by equation 227. VGS VT IDSe nVth gd0 (227) th The thermal noise for a transistor can be represented by a parallel current source connected between the drain and source of a transistor as shown in Figure 29. A/2 =4kTGAf= 4kTAf Figure 29. Noise represented by a parallel current source in a transistor For a transistor operating in the subthreshold region, the equivalent noise resistance (Rn) is given by equation 228 [33]. 2 Rn (228) gdO After representing the exponential function in terms of its Taylor series and 2kT substituting o02 = assuming t << RC equation 223 can be represented as equation C 229 [32]. C RC RC229) Now substituting Rn = 2/gd R = 1/gd and t/C = VDD/21 in equation 229, it can be represented as equation 230. For the condition VDS > 4Vth, gds can be neglected. /V,(t)2 = kT + gdoVDD gdsVDD (230) C 21 / Using equation 230, the variance switching time jitter (equation 224) is given by equation 231. 2 kTC gd.oVDD (231) AT1 1+ DD (231) In a ring oscillator with N stages, there are 2N independent switching events in each period. Thus, the net timing jitter for a ring oscillator operating in the subthreshold region is given by equation 232. A 2 2kTTO2 (1gdVDD (232) 0 NCV2D 21 ) Here the nominal period oscillation (To) is given by equation 233 [32]. T NCDD (233) Now substituting the values of gdo and I corresponding to the subthreshold region of operation from equations 227 and 222 respectively, the expression of timing jitter for the ring oscillator is given by equation 234. AT2 2kTT02 V(234) AT0 = 1 + 2Vt, (234) For an oscillation frequency of fo, the power consumption for a ring oscillator with N stages is approximately given by equation 235 Thus, the relationship between timing jitter and power consumption at a given oscillation frequency (fo) is given by equation 236. We can clearly see that the timing jitter is inversely proportional to the power consumption and should decrease in value at the expense of more power. P = NCVD (235) 2 2kT Vth ATo = 1+ DD (236) Pfo 2Vth Once the variance of timing jitter is determined, the phase noise at a given offset frequency can also be calculated by using equation 237 [32]. f2 f03 AT PN(Af) = T (237) (03 A To2) +(Af)2 Thus at much higher frequency offsets, the phase noise can be approximated by equation 238. 2kT V fo )2 PN(Af) = 1 + V f (238) P 2 Vth Af For validating this model a ring oscillator with three inverters (Figure 210), operating in the subthreshold region was designed and simulated in Cadence SPECTRE (APPENDIX A). The W/L for the pMOS and nMOS was set to (2pm/120nm) and (1.58 pm/120nm) respectively. At 250mV the oscillation frequency was 34.8MHz and the power consumption of the inverters in the ON state turned out to be 80nW approximately. VQQ VDD OUT GND Figure 210. Ring oscillator to verify the phase noise model The simulated and the predicted phase noise from equation 237, for this oscillator are compared in Figure 211. From this plot, we can see that the phase noise at an offset frequency of 1 MHz is 93.74dBC/Hz. The expected value of the phase noise at this offset frequency from equation 238 is 90.2dBc/Hz (at room temperature). 40   I SSimulated 50 ....... Predicted  70 171~ ~ ~~ 80 '__ "> 90o   ..... So 90 i .100 _  "'1 1 110 1 120 104 105 106 107 Frequency Offset (Hz) Figure 211. Comparison of simulated and expected phase noise for a ring oscillator running at 34.8MHz Further, the oscillation frequency was varied by adding more inverter stages in the ring oscillator and keeping the supply voltage the same i.e at 250mV. The expression for minimum achievable phase noise for ring oscillators is given by equation 239 [32]. pN(Af)7= kT PN(Af) 733kT (239) The phase noise is computed for different oscillation frequencies by using both formulas (equation 239 and equation 238) and are results are mentioned in Table 41. We can clearly see that the value given by equation 238 is more close to the simulation result. Thus, we can use this model for a better estimate of the phase noise performance of a ring oscillator working in the subthreshold region for a given power constraint. Table 22. Comparison of phase noise at 1 MHz offset frequency at different oscillation frequencies Number of Oscillation Power PN at 1MHz PN at 1MHz PN at 1MHz stages in Frequency Consumption Offset Offset Offset the (MHz) (nW) (dBc/Hz) (dBc/Hz) (dBc/Hz) Oscillator (eq. 239) (eq. 238) (Simulation) (N) 3 34.8 58 91.9 97.62 93.74 5 20.3 52.5 96.25 101.87 98.3 7 14.43 52.9 99.23 104.86 101.8 9 11.22 53.25 101.45 107 104.5 CHAPTER 3 SYSTEM ARCHITECTURE 3.1 Basic Architecture of the System A simplified block diagram for a typical passive biomedical microsystem is shown in Figure 31. It consists of an antenna port, a power management unit comprised of a voltage rectifier, voltage regulator and bias circuitry, an ASK modulator/demodulator, a clock generator and a digital baseband signal processor. Power Management Unit   S I VDDH I .34 Analog Interface ensor ASK Clock CLK ........... Demodulator Generator ........ ; Signal Processor and ASK DN Control Logic Modulator 1 Do" " IModulator DOUT .......................... Figure 31. Block diagram of a typical biomedical microsystem with sensing and data communication capability The power management unit is one of the most important components of the entire system as it provides the required voltages needed for operation by the various blocks and its efficiency is critical to system performance. The device is powered up by continuously sending a RF wave, which is rectified into a DC voltage. The power efficiency of the system can be improved by separating the supplies for analog and digital blocks (VDDH and VDDL). Digital blocks can be operated at a lower supply voltage which results in power savings. This allows the analog blocks to be unaffected by the voltage spikes which appear due to signal transitions in the digital blocks. Although, the dynamic power dissipation decreases with the supply voltage scaling, the leakage current increases as it is exponentially dependant on the supply voltage [34]. The reference clock and commands for configuring the system are sent as ASK modulated signals along with the carrier. This information is extracted by the ASK demodulator and provided to relevant blocks. The backscattering scheme described in [22] is generally employed for sending the data back to the interrogator where the antenna impedance is varied between either a perfect match or complete mismatch i.e a short. 3.2 Digital Phase Locked Loop Figure 32 shows the functional block diagram of the subthreshold DPLL operated from a 260mV supply. LOOP FILTER ..............................LK COARSE Figure 32. Implemented digital phase lock loop ^RX 10 7 .32 A binary phase and frequency detector (BPFD) compares the divided down clock Z1 FINE . P3F Figure 32. Implemented digital phase lock loop A binary phase and frequency detector (BPFD) compares the divided down clock from the digitally controlled oscillator (DCO) with the extracted reference clock (CLKRx) from the ASK demodulator to generate a single bit early/late signal. The digital loop filter with programmable proportional (Kp) and integral (KI) path gains is updated on every cycle of the divided down (1/(M*N)) clock producing a 10bit output based on the early/late signal. Out of these 10 bits, the 3 least significant bits are used by a sigma delta (XA) sampled at 1/M of the output clock (CLKTx) frequency to produce a bit stream. The sigma delta output, along with the remaining 7 bits from the loop filter output, forms an 8bit control word for the DCO. The five most significant bits (MSBs) of the control word are used for tuning the coarse delay stage through a 5to32 bit decoder (COARSE [31:0]) and the three least significant bits (LSBs) control the fine tuning cell through a 3to8 thermometric decoder. In order to keep the hardware complexity of the loop filter low the programmable constants are chosen to be a power of 2. Multiplication/division operations are performed by left/right shifting of the bits based on the value of K1. 3.2.1 BangBang Phase Frequency Detector A digital phase detector can be implemented in many ways, such as a XOR gate or a JK flip flop where the duty cycle of the output indicates the phase difference between the two signals being compared. These phase detectors can track the phase error only when the error is confined within a very small range [5]. This limitation can be overcome by using a phase frequency detector (PFD). A low pass filter is required to be used with the previously mentioned phase detectors to determine the DC content of the output. A bangbang phase and frequency detector (BBPFD) is a special case where the phase difference between its input signals is quantized by a single bit resolution. In this DPPL design, a BBPFD similar to [35] has been used and is shown in Figure 33. In this implementation, a conventional PFD is followed by a sampling flip flop. The UP signal goes high whenever FREF arrives and the DOWN signal goes high at the rising edge of the FDIv signal. By sampling the UP signal with the DOWN signal, the sign of the error is determined. Thus a '1' output specifies that the divided down clock is lagging the reference signal and DCO frequency needs to be increased. The output (E/L) remains either high or low for the entire period of the divided down clock. The output signal does not have a 2 7 periodicity as it provides just the direction of error corresponding to all phase differences at the input and because of this, the transition dynamics are very smooth [10]. TD UP 3.2.2 Digital Loop Filter A digital loop shown in Figure 34, with programmable integral and proportional path constants, has been used in this design. It operates at the divided down clock frequency and while in the locked state, it computes the control word every reference clock cycle. Based on the PFD output (E/L) the value stored in a 13bit accumulator is CLR >CLK FDIV DOWN incremented or decremented on each rising edge of the clock, followed by a 3.2.2 Digital Loop Filter A digital loop shown in Figure 34, with programmable integral and proportional path constants, has been used in this design. It operates at the divided down clock frequency and while in the locked state, it computes the control word every reference clock cycle. Based on the PFD output (E/L) the value stored in a 13bit accumulator is incremented or decremented on each rising edge of the clock, followed by a multiplication by the integral path constant (K). In order to keep the implementation simple, KI is chosen to be a power of 2 and multiplication is performed by simply shifting the bits of the accumulator which generates a 10bit result. The proportional path constant (Kp) is directly added to this result and a 10bit control word (CW) is obtained after checking for overflows. The values of Kp and KI affects the closed loop characteristics of the entire DPLL and a root locus based analysis approach is described in [16] and [10]. Kp affects the damping factor and KI affects the bandwidth. A larger value of KI also helps in fast frequency acquisition in the initial phase when the PLL is unlocked but results in a longer phase capture time [36]. As discussed earlier, a higher damping factor is required to make the system more stable. However, a higher value of Kp also increases the jitter in the system. This is due to the control word changing by a factor of (2*Kp +1) whenever the PFD output toggles. A higher value of Kp/Ki is required to meet the stability criteria described by equation 214. In the present design, Kp and KI are both programmable and their ratio can assume a maximum value of (7/0.125=28) corresponding to different system requirements. 3 10CW 133 K,1 13 Z1 2 Figure 34. Digital loop filter 3.2.3 Sigma Delta Modulator A first order sigma delta modulator shown in Figure 35 has been used to obtain an increased frequency resolution for the DPLL. It operates at a higher frequency than the loop filter and oversamples the loop filter output by a factor of M. The 3 LSBs from the 10bit control word output of the loop filter (CW) are added to the previous value of the accumulator and the generated carry out bit is supplied as the output (SD_OUT). The sigma delta also has a high frequency noise shaping transfer function that shifts the phase noise toward higher frequencies; the noise is ultimately filtered out because of the low pass phase response of the PLL [10]. CW[2:0] SD_OUT + Q CLK Figure 35. Sigma delta modulator 3.2.4 Digitally Controlled Oscillator As discussed in section 2.2.3, a ring oscillator topology consists of an odd number of inverters are connected in a feedback loop and the oscillation frequency (equation 31 ) is determined by the delay of each stage (r ). In order to have a low oscillator, frequency either a large number of stages has to be used or the delay of each stage has to be increased substantially. A ring oscillator based DCO topology with tunable coarse and fine delay stages, shown in Figure 36, has been implemented to work at very low supply voltages. By operating the inverters in the subthreshold region, the delay of each stage increases dramatically. Further, connecting the outputs of all the coarse delay stages together increases the output capacitance of the coarse delay stage. 1 fos (31) 2Nr The coarse tuning stage is implemented as a 32to1 delay select path architecture with tristate buffers acting as selection switches. The shortest coarse delay path consists of 3 inverters while the longest path consists of 65 inverters. The five most significant bits of the control word are used for tuning the coarse delay stage through a 5to32 bit decoder (COARSE[31:0]). A variable number of inverters have been used for coarse delay selection in [37] and [11]. However, the frequency resolution of the coarse delay stage alone is not sufficient and a fine delay stage is required to obtain an increased resolution. The fine tuning stage has tristate buffers connected in parallel with inverters that are activated by an 8bit thermometric code (FINE [7:0]), based on the remaining 3bits of the control word. The shunted tristates control the current drive strength at each node and determine the delay in the loop. The fine delay stage covers one coarsetuning step to obtain a monotonically rising frequency response, failure to ensure this could lead to an unstable PLL [9]. The 32 coarse delay steps and 8 fine delay steps together allow the DCO to have the capability of generating 256 different frequencies. Each course delay stage adds a delay of approximately 28ns, while each tristate buffer in the fine delay stage changes the delay by about 3ns. COARSE DELAY Figure 36. Digitally controlled oscillator The simulated DCO response is shown in Figure 37 for the TT corner with a supply voltage of 260 mV. As can be seen, the DCO frequency changes linearly for lower values of the control word but the response becomes nonlinear for higher values of the control word. The DCO frequency is also very sensitive to the supply voltage variations while operating in subthreshold region. This problem was overcome by using a voltage regulator to produce a constant supply voltage. x 106 3.5 I I I I 3 v'2.5 S2 0 1.5 Q) 0 50 100 150 Control Word Figure 37. Frequency versus control word FINE DELAY 3.3 Sizing of Digital Circuits in Subthreshold Region All the blocks of the DPLL described in previous sections are operated in the sub threshold region to significantly reduce the power consumption. Also by reducing the supply voltage, leakage current due to the Drain Induce Barrier Lowering (DIBL) effect can be significantly reduced since it is directly proportional to the supply voltage. The current equations for the subthreshold region of operation are discussed in Section 2.2.3. The focus of discussion of this section is the ION to IOFF ratio which can be used as a measure of robustness for digital circuits as in [29]. ION (ON current) is the current that the MOSFET can source when it is fully turned on. IOFF (OFF current) is the current that leaks through the MOSFET even when it is turned off. At high operating voltages this ratio is significantly large and ensures robustness of the circuit. But in the subthreshold region of operation this ratio can get quite low, leading to slow charging or discharging of the output node and can cause circuit failures. Figure 38 shows that the ION to IOFF ratio for an nMOS transistor (130nm) can approximately have a maximum value of 4000 in the subthreshold region. 4000 3500 3000 S2500 2000 1500 1000 500 0 0.05 0.1 0.15 0.2 0.25 Supply Voltage (V D Figure 38. ION to IOFF ratio in subthreshold region Thus, care has to be taken while sizing the gates to ensure functionality across all process corners. The effect of ION to IOFF ratio on the sizing of the digital circuits operating in the subthreshold region can be understood from the following discussion. In the superthreshold regime, CMOS digital circuits are usually sized such that the drive strengths of PMOS and NMOS transistors are roughly equal. A typical sizing ratio of 2:1 is often chosen, where the PMOS is sized twice the size of the NMOS. This sizing is chosen so that the PMOS and NMOS can source the same amount of current to charge or discharge the load. This is usually done to have roughly equal rise and fall delays resulting in a trip voltage of VDD/2 and obtain a high noise margin. This sizing ratio doesn't hold well in subthreshold regime. This is because the current doesn't follow the same quadratic profile with respect to the drain, source, gate and bulk voltages in the subthreshold region. For proper functioning of the digital circuit, the ON complimentary NMOS and PMOS networks should be able to provide enough drive current to pull down or pull up a node when the other network is OFF within the time period of the desired frequency of operation. With process scaling, the leakage current of the transistors increases. Added to the subthreshold current is the gate leakage current whose impact becomes more significant with every successive technology generation. This OFF state leakage current can be significantly large enough to prevent a node from being pulled up or down in the subthreshold regime, where the ION and IOFF ratios are drastically low. For example in Figure 39, consider the inputs A=1 and B=1 which are applied to the NAND gate. When operated at very low voltages (in subthreshold), the two PMOS's in parallel might supply enough leakage current to prevent the ONstate NMOS's from pulling down the output node. The fact that the NMOS's are in series does not help this situation either because it further reduces the available current. With the added unpredictability of the PMOS's being faster than the NMOS's, the output node might always be stuck at logic '1'. Increasing the operating voltage would mitigate the problem, as the effect of the leakage current (IOFF) on the overall charging/discharging behavior of the output node would be significantly reduced. This would happen because operation at higher voltages increases ION, allowing it to be much larger than IOFF. In the light of the discussion above, a minimum operational voltage for different gates was found with an output swing limited to 10%90% of the supply voltage across different process corners. It was found that for 130 nm technology, standard cells were functional for 160 mV supply with a P/N ratio of 2:2. Therefore the digital blocks are guaranteed to function by operating at 250 mV. VDD I I OUT H I  IOFF B ION GND Figure 39. Leakage induced problems in a NAND gate operating in subthreshold region CHAPTER 4 SIMULATION AND MEASUREMENT RESULTS 4.1 Simulation Environment For mixed signal systems, the designer has to be extremely careful of the interactions between the digital and analog blocks. As stated in Section 2, PLLs can have pure analog, digital or mixed signal implementations. It is extremely difficult to simulate a PLL system because of the "mixed analogdigital" nature of internal signals [38] and feedback. It needs to also be simulated for a large number of reference clock cycles to capture the transient response since a PLL shows a highly nonlinear behavior prior to locking. In order to verify the correct functionality (whether it ever locks to a frequency) and evaluate other performance parameters such as timing jitter, device level simulations using circuit simulators such as SPECTRE or SPICE result in extremely impractical and long simulation time. Behavioral modeling techniques described in [3940], and mixed signal simulation techniques described in [35] and [38] offer a promising solution to the problem of effectively verifying the functional behavior of the PLL as well as expedite the entire simulation process. 4.1.1 Behavioral Model Behavioral modeling involves describing the function of a block with the use of concise, approximated mathematical equations or simple pseudo logic like code. There are no transistor level schematics, rather tools such as MATLAB, C, Verilog or System Verilog are used to describe the functionality of a block. It empowers the designer to quickly integrate the blocks and perform system level simulations to analyze the feasibility and try out different configurations. Functionally equivalent Verilog codes for various sub blocks such as the PFD, loop filter, DCO and programmable divider corresponding to the bangbang digital PLL described in Section 3.2.1 were written. The entire DPLL system was simulated in ModelSim for verifying the locking characteristics and analyzing the effect of the loop parameters on the system stability and acquisition time. A linear model of the DCO, producing a square wave and having 256 control steps was developed. The relation between the DCO output frequency (FOUT) and the control word (CW) from the digital loop filter is given by equations 41 and 42,where FH and FL are the highest and lowest frequencies of the DCO respectively. Fout = FL +KDCo *CW (41) KDCO _ (FH FL )(42) K,= H (42) 256 The effect of increasing the proportional path constant Kp can be seen in Figure 41. In this particular simulation setup, a reference frequency of 16 kHz was chosen with a multiplier factor of 64 to obtain a 1.024 MHz output clock. As stated earlier in section 2.1.3, increasing Kp effects the damping factor and results in a reduced frequency acquisition time, however this comes at the cost of increased jitter. Since each time the output of the PFD changes, the control word changes by a factor of(2 x Kp K,), thus higher the value of Kp, the higher the control word ripple will be. Next, to analyze the benefit of using a sigma delta modulator, it was shut off without changing any of the previous set of input parameters. As seen in Figure 42, the ripple in control word increases. This is because the sigma delta modulator pushes the noise towards higher frequencies which is ultimately filtered out due to the low pass response of the PLL. 1 / t 50 1 0 U 100 50 0 200 400 600 800 1000 1200 1400 1600 Reference Clock Periods 300 Kp=7,K1=1 250 N=64 200 ao U 100  50 0 200 400 600 800 1000 1200 1400 160 Reference Clock Periods Figure 41. Control word versus time for two different values of Kp To analyze the PLL response to a frequency step, a 16 kHz reference clock was applied at the input followed by a 12.5 kHz clock. As can be seen in Figure 43, the PLL first locks to the 16 kHz clock and the control word settles down to produce an output frequency of 1.024 MHz. The PLL locks to the new frequency after the reference clock changes. ,nn. 100 50 Figure 42. 0 200 400 600 800 1000 1200 1400 1600 Reference Clock Periods Control word versus time with the sigma delta modulator disabled 300 250 ....  Locked 150o C/ V =. 2 '"^ so 4 r Frequency Ste 0 0 100  50  o 1II I IIIIlllB 0 500 1000 1500 Reference Clock Periods Figure 43. DPLL response to a frequency step For phase step simulation, the reference frequency is later a 75 degree phase shift is applied at the input. As can initially set to 16 kHz and be seen in Figure 44, the Kp=4,K=1 N=64,SD_EN=0 nnnnnnnnnninni 200 PLL is able to quickly recover from the phase step and the control word reaches its previous value again. 300 I I Locked States 250  ,,  S200 , SPhase Step 150 z UR o CL U 100 50 0 500 1000 1500 2000 2500 Reference Clock Periods Figure 44. DPLL response to a phase step 4.1.2 Mixed Signal Simulations After behavioral modeling is complete and system specs have been decided, critical blocks in the system can be replaced by their transistor level schematics for simulations. Cadence design suite comes with a large number of simulators which facilitate integration and simulation of subblocks implemented at different levels of abstraction. For this particular system, the Verilog model for the DCO was replaced by its transistor level counterpart operating from a 260 mV supply. In the mixed simulation mode, the blocks are connected to each other and then the hierarchical editor is invoked to notify the tool which view (abstraction level) of a particular block is to be used in the simulation. Cadence uses Spectre Verilog to simulate such a schematic which contains some blocks with only Verilog models and some with complete transistor level implementation. Next, design partitioning takes place and necessary interface elements are placed between the transistor level and code level implementations. These interface elements behave as ADCs or DACs and it is possible to specify properties such as propagation delay, rise/fall times and input/out levels To demonstrate a locked state in the first simulation setup, a reference frequency of 20 kHz is applied at the input and the multiplier gain is set as 64. Figure 45 shows that the PLL locks with the DCO frequency of 1.28 MHz. Next, a frequency step simulation is performed by first setting the reference clock to 40 kHz and later changing it to 33 kHz with the multiplier gain set as 32. As can be seen in Figure 46, the PLL first locks with an output frequency of 1.28 MHz and later with 1.06MHz. Although mixed signal simulations are more accurate, they take significantly more time to run than behavioral level simulations. However, the required run time and resources are far less than the complete transistor level simulations. Transient Response :O N LF< 9 > ......... ....... ...... ......... :ON_LF<8> I .,,,,,, ,,, ,, ...... .... 1.40M ON LF<5> I. , O N _L > ... .. .. .1 ~ oNLF<5>LV1J^j7J J 1J 1IjUI` J I ON_LF<4>r. n.nn.U.nl...i.ir 1.20M Locked to 1.29 MHz /SD J ON_LF<2> D N S1.00M ON_LF<0> , x: /CLOCK_PD : /net061 o: /FREF  90Qm /RST : /EL 900K 700m 800K 500m 300m 700K 100m 600K 10m 600K___ .___ .______ .600K 0.0 5.0m 10m 15m 20m 0.0 10m 20m time ( s ) time ( s ) Figure 45. Mixed signal simulation for lock acquisition I r Frequency Step 1JBH /VL Locked States  S'  Figure 46. Mixed signal simulation for frequency step 4.1.3 Synthesis of Digital Blocks After verifying the functionality of the system by using the above mentioned simulation techniques, synthesis of digital blocks is carried out using various tools. Synopsys DC compiler was used to produce an optimized netlist for the behavioral Verilog model based on the provided 130nm standard cell library information. After putting the timing constraints in place, timing slack was checked for any set up and hold time violations. The optimized netlist can also be tested using the same set of test fixtures which were incorporated with the previous behavioral simulations. Next, Cadence Encounter was used to perform automated place and route on the netlist to generate the layout for this design. The layout of the chip is shown in Figure 47, where the top half consists of the synthesized components and a custom DCO is integrated at the bottom. E O 162 pm Figure 47. Chip layout 4.2 Measurement Results The passive transceiver system, consisting of a RFDC multiplier, onchip power management circuitry, ASK demodulator (envelope detector) along with the digital PLL, was implemented using a130 nm CMOS process. A die photo of the entire chip, with dimensions of 1.67 mm by 1.27 mm, is shown in Figure 48. The DPLL occupies an area of 0.17 mm2 4.2.1 Test Measurement Setup The packaged chip was mounted on a PCB which consists of various input and output terminals connected to an oscilloscope through SMA connectors (Figure 49). The necessary operating voltages for the DPLL, I/O pads and level converters were supplied through DC voltage generators. An ammeter with a 1nA resolution was connected in series with the PLLs supply voltage to measure the average power consumption. An 8channel ADC bus on the oscilloscope was used to observe the control word from the loop filter. 1670pm Figure 48. Die photo of passive transceiver with DPLL SWITCH PROGR AND K, Figure 49. Packaged chip mounted on a PCB for testing 4.2.2 Test Cases First, a reference frequency of 40kHz was applied at the input with the division ratio (M x N) set to 16. The value of Kpwas set to 3'b001 (1) and K to 2'b01 (0.5). The expected DCO frequency is 640kHz in the locked state, this can be seen from the timing waveforms shown in Figure 410, where the rising edge of the reference clock (CLK_REF) is properly phase aligned with the rising edge of the divided clock (CLK_DIV) from the DCO. The observed control word (CW) toggles between (0XA2 and OXA3) and the DCO frequency (CLK_OUT) is 640MHz approximately. Signal E_L is the output of the binary phase and frequency detector. All these signals are internally 250mV and have been level converted to 1.2 V before sending them to the digital output pad. The measured spectrum of the DCO signal is shown in Figure 411. The jitter measurements results for this set up are shown in Figure 412, the RMS and peakto peak jitter for the PLL output are 84.44ns and 518.18ns respectively while consuming 100nW of power at a supply voltage of 250mV. CLK_REF CLK_DIV 1V/div CLK_OUT EL ........ .. ..... ............. ....................... ........... I ........................ ........... ......... CW <7:0> A. 201is/div Figure 410. Measured waveforms for reference frequency of 40kHz and multiplier ratio set to 16 15 dBIdiv Ref 60 dBm Mkrl 640 kHz 1.0272 dBm LOg 4/\  75 Center 640 kHz Span 1.358 MHz Figure 411. Spectrum of PLL output with oscillation frequency of 640kHz Figure 412. Jitter histrograms for 40kHz reference clock A) Divided clock B) PLL output clock Next, the input frequency was changed to 20kHz without changing the rest of the input parameters such as the supply voltage, multiplier ratio, Kp and KI values. The expected PLL output frequency is 320kHz. The measured spectrum of the PLL output signal shown in Figure 413 confirms that the PLL locks with the DCO frequency fixed at 321kHz. 20 dBIdiv Ref 60 dBm Mkrl 321.72994 kHz 2.3299 dBm Log 20 2nd H rmoni at 0 Q__ __ A 0 _1,0 1 1 20 . 80 inn Center 320 kHz Span 1.766 MHz Figure 413. Spectrum of PLL output with oscillation frequency of 321 kHz S. .. .... .. ................................................... S. ........................ .......................... Figure 414. Jitter histrograms for 20kHz reference clock A) Divided clock B) PLL output clock " The jitter measurements results for this set up are shown in Figure 414. The RMS and peakto peak jitter for the PLL output frequency of 320 kHz approximately are 122ns and 927.23ns respectively. It can be seen here that jitter increases with decrease in the oscillation frequency. Next, the control word of the DCO was fixed at 126 by keeping the RESET signal low and the supply voltage was varied. The oscillator frequency corresponding to this control word increases with each successive supply voltage step. In Section 2.2.3 the expressions for the variance of timing jitter for a ring oscillator operating in the sub threshold region were derived. The standard deviation of the timing jitter can be calculated by substituting T=3000K and k=1.38 x 1023 JK1 and taking the square root of equation 236 The measurement results for the timing jitter are compared against the expected result in Table 41. Table 41 Comparison of theoretical and measurement results for ring oscillator Supply Oscillation Power 0TO TO Voltage (mV) Frequency (fo) Consumption (Theoretical)) (Measured) 150 46.28kHz 16.5 nW 6.48ns 725.36 ns 200 149kHz 32 nW 2.89ns 210 ns 250 453.58kHz 75 nW 1.18ns 42.57 ns 300 1.3MHz 225 nW 0.43ns 27 ns 350 3.7MHz 784 nW 15ps 12 ns 400 6.47MHz 1.7 pW 8.9ps 3.44 ns 450 11.52MHz 3.73 pW 4.3ps 2 ns 500 19.57MHz 7.88 pW 2.38ps 1.46ns We can see that the measured standard deviation of the timing jitter is much higher compared to the expected value. This is because this model only considers the noise due to the output resistance at each node and the effect of other noise sources such as the power supply voltage noise and substrate noise has not been accounted for. The level converter and the digital output pad buffer in the signal path also contribute to the noise component. As predicted by the model the timing jitter reduces with an increase in the supply voltage and oscillation frequency. The measurement results for peaktopeak jitter at different supply voltages are shown in Figure 416. 104 fo=46.28KHz, P=16.5nW fo: Oscillation Frequency P: Power Consumption 3 fo =149KHz,P=32nW 10  So = 454KHz,P=75nW / fo = 1.3MHz,P=225nW ,fo = 3.7MHz,P=784nW 10 101 CL fo = 6.47MHz,P=4.25uW / fo = 11.52MHz,P=3.73uW fo = 19.57MHz,P=7.88uW 100 10I I I I I 150 200 250 300 350 400 450 500 Supply Voltage (mV) Figure 415. Peak to peak jitter for different supply voltages and oscillation frequencies The change in the oscillation frequency at each voltage step can be understood as follow. The oscillation frequency for a ring oscillator is roughly determined by equation 43, where VDD is the supply voltage and I is the ON current (equation 222) as explained in Section 2.2.3. The ratio of the frequencies corresponding to two different supply voltages (VDD1 and VDD2) should roughly be given by the relation in equation 44. fo (43) NC VDD f2 2 VDD1 (44) f /1 VDD2 Substituting the values of the ON current corresponding to subthreshold region of operation equation 44 reduces to equation 45. (VGs2 VGsl) f2 e nVth DD1 (45) f, VDD2 Since VGS =VDD here, relationship between the two frequencies is given by equation 46, where n~1.5 and Vth~26mV. (VDD2 VDD1) f2 e nVth DD1 (46) f1 VDD2 Thus, when the supply voltage changes from 200mV to 250mV the oscillation frequency should change by a factor of 2.9. As can be seen from Figure 415 that indeed the oscillation frequency changes by a factor 3. To determine the locking range of the PLL, the reference frequency was swept for different supply voltages. The division ratio was fixed at 32 and the value of Kp and KI were fixed at 3'b001 and 2'b00 respectively. For the typicaltypical (TT) and slowslow (SS) design corners, the measurement results are shown in Figure 416 (A) and (B) respectively. We can see that at 250mV supply voltage, the tuning range of the PLL is from 310kHz to 1.5MHz and the power consumption ranges from 62.5nW to 185nW. The highest supply voltage at which the PLL is functional is 500mV, after which the oscillator stops working. This is because at this voltage, the pMOS becomes too strong and the output node gets stuck at logical '1' and is not able to discharge through the n MOS. req 10 9MHz384MHz Power 4 64uW14 6uW ,Freq 7 68MHz33 8MKH Power 3 72uW1067uW Freg 428MHz18 14MHz, Power 1 24uW4 73uW Freq 2 12MHz6 9MHz Power 490nW1 48uW Freq 864KHz4MHz Power 171 nW636nW Freq 310KHz1 5MHz Power 62 5nW185nW ,Freq 102 42KHz512KHz Power 28nW56nW Freq 35 2KHz179 2KHz,, Power 18nW24nW 3 10 Frequency (KHz) 10 102 103 104 105 Frequency (KHz) Figure 416. PLL lock range and power consumption at different supply voltages A) typicaltypical TT design corner B) slowslow SS design corner F S40U 400 0 2 350 300 Q. 250 20) nn 400 P350 : 300 . 250 CO _req 95MHz37MHz Power 4 07uW1522uW Frq5 24MHz21 3Mz Power 1 76uW7 04uW Freq2 8MHz11 9MHz Power 772nW3 12uW .req 1 25MHz5 4MHz Power 272nW1 luW Freq 473KHz2 3MKI Power 90nW330nW Freq 158KHz700KHz Power 31 2nW86 4nW Freq 50KHz240KHz Power 15 2nW26 6nW . i i i .. LUU 200 [. *J^U 500 A054   CHAPTER 5 CONCLUSIONS Miniature embedded systems used in various biomedical applications have very stringent power, size and cost constraints. They are powered by a wireless RF signal which can also contain additional information such as the reference clock. Due to the limited power available and DC voltage fluctuations, designing a clock generator is extremely challenging. In this thesis, a lowpower onchip clock generator was proposed that can provide a 385kHz to 1.54MHz clock at 260mV for digital baseband processing blocks and a backscattering modulator. The clock generator consumes ~200nW and is based on a subthreshold DPLL which synchronizes the onchip clock to an externally controlled lowfrequency ASK signal that modulates the incident RF carrier. A single ended ring oscillator has been utilized in this DPLL design mainly because of the low area and power overhead. There are several benefits of using the proposed clock generator. It enables the frequency tuning over a large range of frequencies. By using the DPLL as a frequency multiplier simultaneous communication with more than one peer can take place as each one can be programmed to work at different frequencies. While the area overhead is negligible, a relatively stable clock can be obtained in a cost effective way. The Verilog model of the DPLL can be easily ported to other technologies. Most importantly, a high system efficiency is achieved by operating the digital components of the DPLL in the subthreshold region. Because of low voltage headroom, the static power dissipation of digital circuits is decreased considerably as it is exponentially dependant on the VDS of the transistors. The expressions derived in Section 2.2.3 can be used for designing a subthreshold ring oscillator to meet the phase noise specifications of a design. APPENDIX A PHASE NOISE SIMULATION IN CADENCE The phase noise simulations were performed by performing the following steps which are described in the Cadence help documents. Both PSS and PNOISE analyses should be chosen to simulate phase noise in Cadence. The steps for PSS (periodic steady state) analysis are described as follows. Perform the transient simulation and determine the frequency of oscillation and the time after which the frequency stabilizes. In the analog design environment window chose the analysis type as "pss". Set "Beat Frequency" as the oscillation frequency obtained from transient simulations Set "Output harmonics" as "Number of harmonics" and type in a number between 35. Choose "Accuracy Defaults" as either conservative or moderate. Put a sufficiently large value in the "Additional Time for Stabilization" textbox. The "Save Initial Transient Results" tab can be left blank. Click on the check box to enable the "Oscillator" tab and select the output node from the schematic. The reference node should be set as ground. The "Sweep" checkbox can be left blank Click on the "Enabled" checkbox. Once this is done, the next step is to complete the set up for the "pnoise" simulation for which the following steps need to be performed. In the analog design environment window chose the analysis type as "pnoise". Select "Sweep Type" as "relative" and "Relative Harmonic as 1. Specify the frequency offset range for which the phase noise has to be determined. Select "Sweep Type" as automatic. Select the number of "Sidebands" between 3 to 5. Select "Output" as voltage and select the output node from the schematic. Select ground as the negative output node. Select "Input Source" as none and select sources in the "Noise Type" tab. Activate the "Enabled" checkbox. After completing all the above mentioned steps, click the "netlist and run" button from the analog design environment window. After the simulation is complete go to the results tab and select "Main Form". Choose "pnoise" in the "Analysis" section and select "Phase Noise" as the function. Click the "Plot" button. The phase noise plot should pop up. LIST OF REFERENCES [1] K. Choe, O. D. Bernal, D. Nuttman and Minkyu Je, "A precision relaxation oscillator with a selfclocked offsetcancellation scheme for implantable biomedical SoCs," in SolidState Circuits Conference Digest of Technical Papers, 2009, pp. 402403,403a. [2] W. Liu, K. Vichienchom, M. Clements, S. C. DeMarco, C. Hughes, E. McGucken, M. S. Humayun, E. De Juan, J. D. Weiland and R. Greenberg, "A neurostimulus chip with telemetry unit for retinal prosthetic device," IEEE Journal of SolidState Circuits, vol. 35, pp. 14871497, 2000. [3] T. B. Tang, S. Smith, B. W. Flynn, J. T. M. Stevenson, A. M. Gundlach, H. M. Reekie, A. F. Murray, D. Renshaw, B. Dhillon, A. Ohtori, Y. Inoue, J. G. Terry and A. J. Walton, "Implementation of wireless power transfer and communications for an implantable ocular drug delivery system," Nanobiotechnology, IET, vol. 2, pp. 7279, 2008. [4] MengLin Hsia, YuSheng Tsai and O. T.Chen, "An UHF passive RFID transponder using A lowpower clock generator without passive components," in IEEE International Midwest Symposium on Circuits and Systems, 2006, pp. 11 15. [5] R. E. Best, PhaseLocked Loops: Design, Simulation and Applications. New York: McGrawHill, 2007. [6] B. Razavi, Design of Analog CMOS Integrated Circuits. McGraw Hill Higher Education, 2003. [7] F. Gardner, "ChargePump PhaseLock Loops," IEEE Transactions on Communications, vol. 28, pp. 18491858, 1980. [8] F. M. Gardner, Phaselock Techniques. Hoboken, NJ: Wileylnterscience, 2005. [9] T. Olsson and P. Nilsson, "A digitally controlled PLL for SoC applications," IEEE Journal of SolidState Circuits, vol. 39, pp. 751760, 2004. [10] J. A. Tierno, A. V. Rylyakov and D. J. Friedman, "A Wide Power Supply Range, Wide Tuning Range, All Static CMOS All Digital PLL in 65 nm SOI," IEEE Journal of SolidState Circuits, vol. 43, pp. 4251, 2008. [11] Duo Sheng, ChingChe Chung and ChenYi Lee, "A fastlockin ADPLL with highresolution and lowpower DCO for SoC applications," in IEEE Asia Pacific Conference on Circuits and Systems, 2006, pp. 105108. [12] R. B. Staszewski, J. L. Wallberg, S. Rezeq, ChihMing Hung, O. E. Eliezer, S. K. Vemulapalli, C. Fernando, K. Maggio, R. Staszewski, N. Barton, MengChang Lee, P. Cruise, M. Entezari, K. Muhammad and D. Leipold, "Alldigital PLL and transmitter for mobile phones," IEEE Journal of SolidState Circuits, vol. 40, pp. 24692482, 2005. [13] R. B. Staszewski, S. Vemulapalli, P. Vallur, J. Wallberg and P. T. Balsara, "Time todigital converter for RF frequency synthesis in 90 nm CMOS," in Radio Frequency Integrated Circuits (RFIC) Symposium, 2005, pp. 473476. [14] J. P. Hein and J. W. Scott, "zdomain model for discretetime PLL's," IEEE Transactions on Circuits and Systems, vol. 35, pp. 13931400, 1988. [15] V. Kratyuk, P. K. Hanumolu, UnKu Moon and K. Mayaram, "A Design Procedure for AllDigital PhaseLocked Loops Based on a ChargePump PhaseLocked Loop Analogy," IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 54, pp. 247251, 2007. [16] A. M. Fahim, "A compact, lowpower lowjitter digital PLL," in Proceedings of the 29th European SolidState Circuits Conference, 2003, pp. 101104. [17] Yehui Sun and Hui Wang, "Analysis of digital bangbang clock and data recovery for multigigabit/s serial transceivers," in Custom Integrated Circuits Conference, 2009, pp. 343346. [18] N. Da Dalt, "A designoriented study of the nonlinear dynamics of digital bang bang PLLs," IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 52, pp. 2131, 2005. [19] L. Lincoln, K. Leung and H. C. Luong, "A 7pW clock generator in 0.18pm CMOS for passive UHF RFID EPC G2 tags," in 33rd European Solid State Circuits Conference, 2007, pp. 412415. [20] Chi Fat Chan, Weiwei Shi, KongPang Pun, Lai Kan Leung, Ka Nang Leung and ChiuSing Choy, "A lowpower signal processing frontend and decoder for UHF passive RFID transponders," in IEEE International Symposium on Circuits and Systems, 2009, pp. 15811584. [21] R. Barnett and Jin Liu, "A 0.8V 1.52MHz MSVC relaxation oscillator with inverted mirror feedback reference for UHF RFID," in Custom Integrated Circuits Conference, 2006, pp. 769772. [22] JongWook Lee and Bomson Lee, "A LongRange UHFBand Passive RFID Tag IC Based on High Design Approach," IEEE Transactions on Industrial Electronics, vol. 56, pp. 23082316, 2009. [23] F. Cilek, K. Seemann, G. Holweg and R. Weigel, "Impact of the local oscillator on baseband processing in RFID transponder," in International Symposium on Signals, Systems and Electronics, 2007, pp. 231234. [24] ChiFat Chan, KongPang Pun, KaNang Leung, Jianping Guo, Lai Kan Leung and ChiuSing Choy, "A LowPower ContinuouslyCalibrated Clock Recovery Circuit for UHF RFID EPC Class1 Generation2 Transponders," IEEE Journal of SolidState Circuits, vol. 45, pp. 587599, 2010. [25] X. Zhao, R. Chebli and M. Sawan, "A wide tuning range voltagecontrolled ring oscillator dedicated to ultrasound transmitter," in Proceedings. the 16th International Conference on Microelectronics, 2004, pp. 313316. [26] MyeongEun Hwang and K. Roy, "ABRM: Adaptive Ratio Modulation for ProcessTolerant Ultradynamic Voltage Scaling," IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 18, pp. 281290, 2010. [27] A. Wang, A. P. Chandrakasan and S. V. Kosonocky, "Optimal supply and threshold scaling for subthreshold CMOS circuits," in IEEE Computer Society Annual Symposium on VLSI Proceedings, 2002, pp. 59. [28] C. H. . Kim, H. Soeleman and K. Roy, "Ultralowpower DLMS adaptive filter for hearing aid applications," IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 11, pp. 10581067, 2003. [29] A. Wang and A. Chandrakasan, "A 180mV subthreshold FFT processor using a minimum energy design methodology," IEEE Journal of SolidState Circuits, vol. 40, pp. 310319, 2005. [30] B. H. Calhoun, A. Wang and A. Chandrakasan, "Modeling and sizing for minimum energy operation in subthreshold circuits," IEEE Journal of SolidState Circuits, vol. 40, pp. 17781786, 2005. [31] A. Hajimiri, S. Limotyrakis and T. H. Lee, "Jitter and phase noise in ring oscillators," IEEE Journal of SolidState Circuits, vol. 34, pp. 790804, 1999. [32] R. Navid, T. H. Lee and R. W. Dutton, "Minimum achievable phase noise of RC oscillators," IEEE Journal of SolidState Circuits, vol. 40, pp. 630637, 2005. [33] S. Liu, J. Kramer, G. Indiveri, T. Delbruck and R. Douglas, Analog VLSI: Circuits and Principles The MIT Press, 2002. [34] A. Ashry and K. Sharaf, "Ultra low power UHF RFID tag in 0.13 pm CMOS," in International Conference on Microelectronics, 2007, pp. 283286. [35] D. Bhatia, Digital Aided Synchronization and Mixed Signal Modeling of High Frequency DcDc Converters [Electronic Resource].Gainesville, FL: University of Florida, 2009. [36] Jri Lee, K. S. Kundert and B. Razavi, "Analysis and modeling of bangbang clock and data recovery circuits," IEEE Journal of SolidState Circuits, vol. 39, pp. 15711580, 2004. [37] ChingChe Chung and ChenYi Lee, "An alldigital phaselocked loop for high speed clock generation," IEEE Journal of SolidState Circuits, vol. 38, pp. 347 351, 2003. [38] B. A. A. Antao, F. M. EITurky and R. H. Leonowich, "Mixedmode simulation of phaselocked loops," in Proceedings of the IEEE Custom Integrated Circuits Conference, 1993, pp. 8.4.18.4.4. [39] F. Herzel and M. Piz, "Systemlevel simulation of a noisy phaselocked loop," in European Gallium Arsenide and Other Semiconductor Application Symposium, 2005, pp. 193196. [40] M. Hinz, I. Konenkamp and E. H. Horneber, "Behavioral modeling and simulation of phaselocked loops for RF front ends," in Proceedings of the 43rd IEEE Midwest Symposium on Circuits and Systems, 2000, pp. 194197, vol.1. BIOGRAPHICAL SKETCH Tanuj Aggarwal was born in 1984 in Rishikesh, India. He received his bachelor's degree in Information and Communication technology from Dhirubhai Ambani Institute of Information and Communication Technology (DAIICT), Gandhinagar, India in 2007 and master's degree in Electrical and Computer Engineering from University of Florida, Gainesville, Florida in 2010 respectively. He worked as a Research Assistant in the Integrated Circuit Research Lab (ICR) from June 2008 to May 2010. The focus of his research was the implementation of low power digital circuits for wireless communication applications. His research interests include design and implementation of low power digital circuits, development of design flows to aid fast simulation and verification of mixed signal systems and computer architecture. PAGE 1 1 A LOW VOLTAGE SUBTHRESHOLD ALL DIGITAL PHASE LO CKED LOOP FOR ULTRA LOW POWER BIOM EDICAL MICROSYSTEMS By TANUJ AGGARWAL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010 PAGE 2 2 2010 Tanuj Aggarwal PAGE 3 3 To my Mom, Dad, Bro and Megha PAGE 4 4 ACKNOWLEDGEMENT With great gratitude I would like to thank my academic advisor Prof. Rizwan Bashirulla h for giving me the opportunity to be a part of the ICR research group. His continuous support and guidance during the past two years have helped me gain insight in the f ield of IC design and adopt a professional approach towards my work. I would also like to thank Dr. Robert Fox and Dr. Gijs Bosman for being members of my m asters thesis committee. During my stay at UF I thoroughl y enjoyed classes like Bipolar and MOS IC design design by Dr. Fox, Advanced VLSI design by Dr. Bashirullah. I t was a pleasure working as a teaching assistant for the B ipolar course for Dr. Fox, thank you for teaching a complicated topic like feedback in a very simple way. Special thanks to ChunMing Tang for all his support throughout the implementation of this chip. Working with him was an extremely good learning experience. I would like to thank Chung Ching Peng and Quizhong Wu for teaching me everything about the synthesis and design tools. Interacting with Ch ris and Walker has always been fun and I will always remember all the late night dinners that we had at I Hop. Thanks Jikai Chen, Hang Yu, Xi a o Zhiming and Xue Lin for your inputs with various desig n re lated issues and all the basket ball sessions. I would specially like to thank Prof. Dipankar Nagchoudhuri for introducing me the world of VLSI circuits and showing his faith in me. I am deeply indebted to him. Finally I thank my roommates Manu Rastogi Vaibhav Garg and Ravi Shekhar for always being there. Megha you have bee n so supportive an d patient at every step, I cant thank you enough. I would like to thank my parents and my brother for all their love and support. PAGE 5 5 TABLE OF CONTENTS Page ACKNOWLEDGEMENT .................................................................................................. 4 LIST OF TABLES ............................................................................................................ 7 LIST OF FIGURES .......................................................................................................... 8 ABSTRACT ................................................................................................................... 10 CHAPTER 1 INTRODUCTION .................................................................................................... 12 1.1 Overview ........................................................................................................ 12 1.2 Motivation ....................................................................................................... 12 1.3 Thesis Organization ....................................................................................... 14 2 LITERA TURE REVIEW .......................................................................................... 15 2.1 Introduction to PLLs ....................................................................................... 15 2.1.1 Charge Pump Based Phase Locked Loop ........................................... 16 2.1.1.1 PLL Response to a Phase Step .............................................. 20 2.1.1.2 PLL Response to a Frequency Step ....................................... 20 2.1.2 Digital Phase Locked Loops ................................................................ 21 2.1.3 Bang Bang Digital Phase Locked Loop ............................................... 22 2.2 Design Considerations for Low Power PLLs .................................................. 25 2.2.1 Low Power Oscillator ........................................................................... 25 2.2.2 Sub threshold Operation of Digital Circuits .......................................... 26 2.2.3 Jitter and Phase Noise in Ring Oscillators ........................................... 27 3 SYSTEM ARCHITECTURE .................................................................................... 35 3.1 Basic Architecture of the System ................................................................... 35 3.2 Digital Phase Locked Loop ............................................................................. 36 3.2.1 Bang Bang Phase Frequency Detector ............................................... 37 3.2.2 Digital Loop Filter ................................................................................. 38 3.2.3 Sigma D elta Modulator ........................................................................ 40 3.2.4 Digitally Controlled Oscillator ............................................................... 40 3.3 Sizing of Digital Circuits in Sub threshold Region .......................................... 43 4 SIMULATION AND MEASUREMENT RESULTS ................................................... 46 4.1 Simulation Environment ................................................................................. 46 4.1.1 Behavioral Model ................................................................................. 46 4.1.2 Mixe d Signal Simulations .................................................................... 50 4.1.3 Synthesis of Digital Blocks .................................................................. 52 PAGE 6 6 4.2 Measurement Results .................................................................................... 53 4.2.1 Te st Measurement Setup .................................................................... 53 4.2.2 Test Cases .......................................................................................... 55 5 CONCLUSIONS ..................................................................................................... 62 APPENDIX A PHASE NOISE SIMULATION IN CADENCE .......................................................... 63 LIST OF REFERENCES ............................................................................................... 65 BIOGRAPHICAL SKETCH ............................................................................................ 69 PAGE 7 7 LIST OF TABLES Table page 2 1 Performance summary of various low power oscillators ..................................... 25 2 2 Comparison of phase noise at 1MHz offs et frequency at different oscillation frequencies ......................................................................................................... 34 4 1 Comparison of theoretical and measurement results for ring oscillator .............. 58 PAGE 8 8 LIST OF FIGURES Figure Page 2 1 Block diagram of a PLL ...................................................................................... 15 2 2 Charge pump PLL .............................................................................................. 16 2 3 Magnitude response of the closed loop transfer function for different damping factors ................................................................................................................. 18 2 4 Magnitude response of the error transf er function for different damping factors ................................................................................................................. 19 2 5 Digital phase lock loop ........................................................................................ 22 2 6 Bang bang PLL ................................................................................................... 23 2 7 Single ended ring oscillator with identical N stages ............................................ 28 2 8 An inverter stage in the ring oscillator ................................................................. 29 2 9 Noise represented by a parallel current source in a transistor ............................ 30 2 10 Ring oscillator to verify the phase noise model ................................................... 32 2 11 Comparison of simulated and expected phase noise for a ring oscillator running a t 34 8MHz ............................................................................................ 33 3 1 Block diagram of a typical biomedical microsystem with sensing and data communication capability ................................................................................... 35 3 2 Implemented digital phase lock loop ................................................................... 36 3 3 Bang Bang PFD ................................................................................................. 38 3 4 Digital lo op filter .................................................................................................. 39 3 5 Sigma delta modulator ........................................................................................ 40 3 6 Digitally controlled oscillator ............................................................................... 42 3 7 Frequency versus control word ........................................................................... 42 3 8 ION to IOFF ratio in subthreshold region ............................................................... 43 3 9 Leakage induced problems in a NAND gate operating in subthreshold region 45 4 1 Control word versus time for two different values of KP ...................................... 48 PAGE 9 9 4 2 Control word versus time with the sigma delta modulator disabled .................... 49 4 3 DPLL response to a frequency step ................................................................... 49 4 4 DPLL response to a phase step ......................................................................... 50 4 5 Mixed signal simulation for lock acquisition ........................................................ 51 4 6 Mixed signal simulation for frequency step ......................................................... 52 4 7 Chip layout .......................................................................................................... 53 4 8 Die photo of passive transceiver with DPLL ....................................................... 54 4 9 Packaged chip mounte d on a PCB for testing .................................................... 54 4 10 Measured waveforms for reference frequency of 40kHz and multiplier ratio set to 1 6 .............................................................................................................. 55 4 11 Spectrum of PLL output with oscillation frequency of 640kHz ............................ 56 4 12 Jitter histrograms for 40kHz reference clock A) Divided clock B) PLL output clock ................................................................................................................... 56 4 13 Spectrum of PLL output with oscillation frequency of 321kHz ............................ 57 4 14 Jitter histrograms for 20kHz reference clock A) Divided clock B) PLL output clock ................................................................................................................... 57 4 15 Peak to peak jitter for different supply voltages and oscillation frequencies ....... 5 9 4 16 PLL lock range and power consumption at different supply voltages A) typical typical TT design corner B) slow slow SS design corner .................................... 61 PAGE 10 10 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science A LOW VOLTAGE SUBTHR ESHOLD ALL DIGITAL PHASE LO CKED LOOP FOR ULTRA LOW POWER BI OMEDICAL MICROSYSTEM S By Tanuj Aggarwal August 2010 Chair: Rizwan Bashirullah Major: Electrical and Computer Engineering Biomedical m icro systems have very stringent space and low power constraints and need to accomplish data sensing and communication in a cost effective way. A clock signal is required by onchip analog and digital components for sampling or processing the gathered data as well as for synchronizing the system. Although the power constraints are rigid, t he clock fr equency requirements are relaxed as the majority of biomedical signals reside at low frequencies. Thus low power, low area, low frequency, low voltage operation and simple implementation are the key features required by a clock generator in order to be inc orporated in implantable biomedical systems or sensor networks. O ff chip components such as crystal oscillators can provide a very stable clock, but they are not suitable for these systems mainly because of their large size and high power consumption. M onolithic oscillators such as relaxation oscillators which consume low power and occupy significantly low area are preferred for clock generation. An efficient frequency calibration scheme is also needed to reduce drift in the oscillation frequency due to p ro cess variations [1]. Motivated by these factors, we present the PAGE 11 11 design of a very low power sub threshold digital phase locked loop (DPLL) employing a ring oscillator, for clock generation in bi omedical microsystems. The DPLL can be used as a frequency multiplier with programmable gain factors to gener ate a signal at higher frequencies which is an exact multiple of the reference clock. Thus the local clock of the system can be synchronized to a wireless low frequency signal and enable data communication. The main advantage of a digi tal implementation is that it remains functional even as the operating voltage is scaled down for decreased power consumption. A passive transceiver system employing the proposed DPLL was implemented in a 130nm CMOS process I t h as a tunable output frequency range of 384 k Hz 1.54 MHz. All blocks of the DPLL operate from a 260 mV supply in the subthreshold region and consume an average power of 200 nW while producing an output frequency of 1.28 MHz. A Regist er Transfer Level (RTL) behavioral model of the DPLL was developed and its functionalit y was verified using mixed signal simulation tools in the Cadence suite. All blocks besides the digitally controlled oscillator (DCO) were implemented using synthesis and automated place and route tools. PAGE 12 12 CHAPTER 1 INTRODUCTION 1 1 Overview The focus of this work is on the usage of a low voltage digital PLL as the clock generator block in low power biomedical microsystems. PLLs have traditionally been made from analog building blocks, but with recent advancements in IC design technology the focus has shifted to their digital implementation as a result o f various advantages such as reduced power and area. Low power and area are the primary requirements of a clock generator in biomedical systems as they have limited power resources and should incorporate functionalities such as sensing, data processing and communication in a simple and eff i cient way. Operating circuits in the subthreshold region reduces the power c onsumption significantly. Due to the relaxed frequency constraints in the targeted applications sub threshold logic can be used for implementing the digital circuits Motivated by these factors we investigated the use a subthreshold DPLL for synchronized clock generation in biomedical systems. Behavioral modeling and mixed signal simulation methodologies have been employed to quickly validate the functionality of the design. A utomated place and route tools have also been used to reduce the turnaround tim e Also, the behavioral model of the DPLL can be easily ported to other technologies without much modification. 1 2 M otivation Wirelessly powered embedded systems with data transfer capability have been widely used in various applications For example, a design of an implantable device which mimics the fu nctionality of a photoreceptor is presented in [2] while a remotely PAGE 13 13 controlled implantable drug delivery system is described in [3] In both these applications, the DC supply voltage is generated by rectification o f the RF signal received at the antenna interface. I n order to maximize the communication range these systems require high RF DC conversion efficiency and low power consumption building blocks. A local clock is required by the baseband signal processing block for decoding the received data as well as for sending the generated data back. It can also be used by an analog interface which is connected to a temperature or humidity sensor in a wireless monitoring application. Reconfigurable property of the local clock can also prove to be very useful. With this additional feature the system can adapt itself to communicate with other systems working at a different clock rate. Also, onchip data can be modulated and sent back at a frequency away from the carrier to mitigate the effect of interference. The clock frequency generated in these systems can drift with process and temperature variation, leading to synchronization failures In this situation, the remotely sent commands cannot be interpreted properly to perform the des ired operation by the system and t he interrogator is unable to recover the backscattered data sent from the monitoring device. Thus a careful design procedure is required to keep the power consumption and overall cost of implementation down Various oscillator topologies with different tuning techniques have been proposed in the past to address these issues H owever PLLs have not been incorporated mainly because they are considered to consume too much power and area [4] An extremely low power Digital Phase Locked Loop (DPLL) with a negligible area overhead is presented in this thesis to lock the oscillator to a desired frequency. PAGE 14 14 1 3 Thesis Organization This thesis presents a design of a very low power DPLL suitable for biomedical micro systems and address es some of the implementation issues. A brief overview of analog and digital PLL s is presented in Chapter 2 along with a literature survey of various DPLL designs and oscillator topologies. A detailed a nalysis for deriving the timing jitter expressions for single ended ring oscillators working in the subthreshold regime is also presented in this chapter. Implementation details of the system are described in Chapter 3 for explainin g t he design choices fo r each block A functional verification methodology along with various simulation and measurement results is described in Chapter 4. Finally, various conclusions are discussed in Chapter 5. PAGE 15 15 CHAPTER 2 LITERATURE REVIEW 2 1 Introduction to PLLs A Phase Locked Loop (PLL) is a control system with negative feedback that synchronizes an output signal with respect to both frequency and phase of an input signal. In locked state, the phase difference between the PLLs output and reference signal is either zero or remains cons tant [5] A general block diagram of a PLL is shown in Figure 2 1 which consists of three main blocks, namely a phase detector (PD), a loop filter and a tunable voltage controlled oscillator. Initially the PLL is in an unlocked state and t he phase detector compares the two signals Uin(t) and Udiv(t). Over time as the phase error (Uerr(t)) builds up, the system adjusts the oscillator frequency such that this phase error is mitigated(ideally zero). The closed loop phase transfer function for the system is gi ven by equation 2 1 () ()od div in odKKFsN s Hs ssKKFsN ( 2 1 ) Where Ko, Kd /s and F(s) are Laplace transforms of transfer functions of the phase detector block, voltage controlled oscillator and the loop filter respectively. Figure 2 1 Block diagram of a PLL PD Loop Filter Oscillator Divide by N Counter i Uin(t) Udiv(t) Uout(t) Uerr(t) PAGE 16 16 In order to understand certain system concepts it is important to consider a particular PLL implementation. 2 1 1 Charge Pump Based Phase Locked Loop The charge pump based PLL shown in Figure 2 2 is a cl assic example of an analog implementation. It consists of the combination of a phase frequency detector (PFD) and a charge pump which either charges or discharges the loop filter capacitor through switches (S1 and S2) based on the UP and DOWN signals. A ph ase detector block is an integral part of PLL systems as shown in Figure 2 1 however a frequency detection loop can be used in conjunction with the phase detector lo op to increase the acquisition range [6] Sequential logic phase and frequency detectors with a charge pump circuit are a cost effective solution to increase PLL system performance [7]. Figure 2 2 C harge p ump PLL The average value or duty cycles of the UP and DOWN signals can easily identify whether the input reference frequency (fin) is less than or greater than the output Divide by N Counter i D Q CLK D Q CLK UP DOWN R C VCO I1I2 finfoutfdiv PFDS1 S2 PAGE 17 17 frequency (fout) Initially when fdiv is far away from fin, the PFD acts as a frequency detector and continuously pumps current in one direction into the loop filter to charge the capacitor which increases the output frequency fout. When fin and fdiv are close, the PFD acts as a normal phase detector and a steady control voltage is supplied to the oscillator to maintain the locked state. This discrete time system can be approximated as a continuous time system as described in [6] whose open loop and closed loop transfer functions are given by equations 2 2 and 2 3 1 () 2VCO openK I HsR CssN ( 2 2 ) Since the open loop transfer function has two poles at the origin one due to the oscillator and the other one due to the loop filter, it is a Type2 PLL. 21 2 () 22vco closed vco vcoIK RCs CN Hs KK II sRs NCN ( 2 3 ) The closed loop transfer function can also be expressed in terms of damping factor ( ) and natural frequency ( n ) as in equation 2 4 and easily analyzed as second order systems. 2 222 () 2nn nns Hs ss ( 2 4 ) By comparing equation 2 3 with 2 4 the natural frequency and damping factor can be expressed as in equations 2 5 and 2 6 where R and C are the passive components of the loop filter and I1 and I2 are the charge pump currents (I=I1=I2). 2vco nIK CN ( 2 5 ) PAGE 18 18 22vcoICK R N ( 2 6 ) A magnitude response of the closed loop transfer function () Hj for different damping factors is shown in Figure 2 3 The frequency axis is normalized by the natural frequency n which allows this plot to be valid for second order Type 2 PLLs in general [5]. Figure 2 3 Magnitude response of the closed loop transfer function for different damping factors Low values of (less than 1) result in poles with complex values, and the peaking in magnitude is mainly due to poles being located close to the imaginary axis of the s plane. However at higher values of the slight peaking in () Hj arises mainly due to the location of the zero. By increasing loop gain (in turn damping factor) the nearest pole and zero come closer affectively reducing peaking of () Hj [8]. We can see in Equation 2 3 that the system has a zero at ( 1 RC ) which is necessary to make the system stable. Without the resistor in place, the system will simply have two poles at the 101 100 101 30 25 20 15 10 5 0 5 10 Normalized Frequency (/n)Magnitude (dB) =0.2 =0.5 =0.707 =1 =2 =4 PAGE 19 19 origin which will contribute t o a total phase shift of 180 at the unity gain frequency, making the system oscillatory. In general for a typen PLL, n1 zeroes are required to make it stable [8] By choosing values of R, I and KVCO, a sufficiently high damping factor can be realized and location of the zero can also be adjusted. As stated earlier, a PLL system tends to minimize the phase error between the input (fin(t)) and the divided signal ( fdiv(t)). The error transfer function is given by equation 2 7 and the magnitude plot is shown in Figure 2 4 For input frequencies less than the natural frequency ( n ), the phase error is quite small and is further suppressed by keeping the damping factor high. 2 22()1() 2err nns HsHs ss ( 2 7 ) Figure 2 4 Magnitude response of the error transfer function for different damping factors The performance of charge pump PLLs is limited by various circuit level nonidealities such as current mismatches between I1 and I2, as well as clock feed through. 101 100 101 50 40 30 20 10 0 10 Normalized Frequency (/n)Magnitude (dB) =0.2 =0.5 =0.707 =1 =2 =4 PAGE 20 20 All these factors introduce ripple into the control voltage which disturbs the lock ed state of the PLL. Normal ly the loop bandwidth is kept around one tenth of the input frequency. Thus for low frequency signals the size of the loop filter capacitor and resistor has to be made very large to meet this bandwidth criteria. 2 1 1 1 PLL Response to a Phase Step When the PLL is in a locked state and there is a change in the phase of the input reference clock, the PLL is always able to recover from such a perturbation. This can be understood by the following analysis. If a phase change of occurs in the incoming signal as shown in equation 2 8 (Laplace transform shown in equation 2 9 ) The error transfer function of the PLL is given by equation 2 10 By substituting s=0 it can be se en that the error always evaluates to zero. ()()itut ( 2 8 ) ()is s ( 2 9 ) es od ts t sKKFsNs2 0lim()lim ( 2 10) 2 1 1 2 PLL Response to a Frequency Step If the input frequency changes by a factor of then the PLL experiences a phase change of *t at the input. The Laplace transform for this phase change is given by equation 2 11 and the error transfer function is given by equation 2()is s ( 2 11) te s ods sKKFsNst2 2 0lim() lim ( 2 12) PAGE 21 21 A s can be seen for s=0, the steady state error does not reduce to zero. This means that there will always be a steady state error due to the frequency step at the input which can be reduced by keeping the loop gain of the PLL high. 2 1 2 Digital Phase Locked Loops Several digital PLLs for different applications have been reported in [9 11] [12] because of their numerous advantages, some of which are lower chi p area, lower power consumption, faster behavioral simulations and most importantly robustness against technology parameter variations A block diagram for a typical linear DPLL is shown in Figure 2 5 It consists of a phase detector followed by a time to digital converter (TDC) to digitize the phase error followed by a digital loop filter and a digitally controlled oscillator. The resolution of the TDC is critical as it determines the amount of phase error that can be measured and ultimately filtered out. There are several ways to implement a TDC. A simple version can be made by using inverters as delay elements in the signal path as in [13 ] T wo counter s clocked by a high frequency signal have been used in [9] for digitizing the phase error. T he s domain loop transfer functions for this class of DPLL can be obtained by applying the linear analysis techni ques already described in Section 2. 1.1 however a discrete time z domain model can more accurately predict the system behavior [14] since it is a sampled system A detailed design procedure is described in [15] to obtain the value of various loop parameters by using an analogy between a charge pump PLL an d a linear digital PLL. Also the s domain transfer functions can b e converted to the continuous time z domain model by u sing bilinear transforms. The main advantage of a digital PLL is that it requires a considerably small loop filter compared to that of the charge pump based PLL. T he PAGE 22 22 digital components can also be operated at lower supply voltages to reduce the power consu mption. Figure 2 5 Digital phase lock loop 2 1 3 Bang Bang Digital Phase Locked Loop A PFD with a TDC is not only hard to design but also occupies substantial chip area and consumes excessive power, therefore it is not suitable for low power applications. An alternative to this approach is to use a binary phase and frequency detector (BPFD) where the phase/frequency difference is represented by only a single bit. Several bang bang PLL (BB PLL) designs are described in [10] [16] and [17] Some limitations of this topology include increased frequency acquisition time and limited jitter performanc e. A block diagram for a BB PLL with a proportional (KP) and integral (KI) path loop filter is shown in Figure 2 6 The single bi t PFD output indicates whether fin is leading or lagging the fdiv signal T his information is used by the loop filter which operates at the divided down clock frequency to generate a control word to adjust the DCO frequency. In the locked state, a BB PLL does not maintain a fixed phase difference, rather the control word changes between two relatively close values on each reference clock cycle. PD Loop Filter DCO Divide by N Counter i finfdivfout TDC Digital Phase Converter PAGE 23 23 Because the BPFD makes the system highly nonlinear, system dynamics of a second order digital BB PLL cannot be analyzed using either s or z domain models. Some of the key results of the timedomain analysis of [18] are discussed here. One of the necessary conditions for locking is given by equation 2 13. Figure 2 6 Bangb ang PLL 011rdco PTTNT x NKK ( 2 13) W here Tr is the reference clock period, TDCO is the DCO free running period, KT is the period gain constant of the DCO and NKPKT is the quantization step of the divided down clock If this condition is n ot met, the time difference ( t ) between Tr and the divided down clock period (TDIV ) will never converge and the PLL will not lock. This implies that if the free running oscillator frequency is initially far away from the reference, the proportional path constant has to be sufficiently high. The ratio of the proportional (KP) and integral (KI) path constants determine the stability of the BB PLL. T he other necessary condition for locking is given by equation 2 14, where D is the delay in the loop. DCO Divide by N Counter i finfout Z1 + + KPKI Loop Filter BPFD fdiv PAGE 24 24 2 21I PK KD ( 2 14) The expression for the peak to peak jitter (JPP) is given by equation 2 15 and it can be seen here that by minimizing KI, D and N the jitter can be reduced in the system. Increasing KI helps in decreasing the locking time but this comes at the expense of increased jitter. A lthough having a higher value of KP helps in e nsuring s tability it cannot arbitrarily be kep t very high as it also increases the quantization step of the proportional path which results in more jitter. 23 32(1)(1)(1)III ppPT PPPKKK JNKKDD DO KKK ( 2 15) These findings are in close agreement with those of [16] which mainly highlight the effect of the loop filter parameters on the stability of a low power, compact and low jitter DPLL. I n the locked state, a linear discrete time model has been used to account for the results. The loop filter transfer function for the integral and proportio nal path digital filter (F(z)) is given by equation 2 16 and the closed loop transfer function for the entire DPLL (H(z)) is given by equation 2 17 From the root locus analysis described in this paper it was shown that increasing KP drives the system towards stability while increasing KI ma i nl y affects the closed loop bandwidth. 1() 1I PK FzK z ( 2 16) P dcoPI PI dcoPI dcoIK KKKz KK Hz KKKKK zz NN2() () () 21 ( 2 17) PAGE 25 25 2 2 Design Considerations for Low Power PLLs After gaining an insight in the working of a PLL we now discuss some of the design issues that need to be considered for low power design. 2 2 1 Low Power Oscillator S everal onchip clock generation schemes for low power applications have been re ported, some of which are summarized in Table 2 1 Table 2 1 Performance summary of various low power oscillators Reference Oscillator topology Supply Voltage Power Consumption Operatin g Frequency [19] Injection locked divider / 0.5 V 6.7 W 52 k Hz 625 k Hz [20] Relaxation 1 V 1.5 W 52 k Hz 625 k Hz [21] Relaxation 0.8 V 320 nW 1.52 MHz [22] Current starved with digital caliberation 1.5 V 40 W 2.2 MHz [23] Current starved ring 0.8 V 1 V 191 nW 306 nW 1.28 MHz [4] Current starved ring 0.4 W 500 k Hz [20] Current starved ring 0.7 V 1.2 V 200 nW 2.45 MHz Ref. [19] describes a dual path clock generator composed of injection locked dividers and a RC resonator. Here, the reference clock is derived directly from the RF carrier which guarantees high accuracy, however this scheme consumes too much power (7 W approximately ) By using a RC relaxation oscillator described in [21] the power consumption can be reduced. However, a huge area is required by the o n chip resistors and capacitors. Since, the output frequency is mainly determined by the value of these passive elements it is unreliable. A voltage controlled oscillator consisting of a current starved ring oscillator with digitally calibrated bias current has been used in [22] Although the ring oscillator alone consumes about 9.5 W, the digital calibration PAGE 26 26 scheme requires about 31 W of power. Current starved ring oscillator based topologies described in [4] and [23] can be a good choice as they strike a balance between low power, area and frequency deviation [20] and also because they do not rely on passive components such as resistors, capacitors and inductors [24], [27] and [29]. LC oscillators are also a popular choice in PLLs because of their superior phase noise properties. However, in order to obtain a low oscillation f requency in the range of a few M ega h ertz, the size of the onchi p inductor and capacitor has to be kept substantially high. Also, the tuning range of LC oscillators is only in the range of 1020% [25] Thus, an LC oscillator cannot cater to the requirements of biomedical sensor networ k systems. Ring oscillators are an attractive alternative mainly because of their simple architecture, low area, wide tuning range and ease of integration. Here, an odd number of inverters are connected in feedback to generate a periodic signal whose frequency is determined by the delay of each inverting stage. By increasing the delay of each cell, low oscillation frequenc ies can be easily obtained. However this comes at the expense of a poor phase noise resulting in timing jitter as single ended ring oscillators are more susceptible to variations in supply voltage. Due to the low speed requirements in the biomedical systems this timing jitter can be tolerated s ince the emphasis is on low power consumption. Although ring oscillators with differential delay cells are more immune to various noise sources, they are not suitable in the applications of interest mainly because of their high power consumption and area r equirements. 2 2 2 Subthreshold Operation of Digital Circuits Sub threshold operation refers to operating circuits at a supply voltage (VDD) lower than the threshold voltage (VT) of a transistor. It involves charging and discharging the PAGE 27 27 load capacitor with the sub threshold leakage current and is able to achieve minimum energy consumption with limited speed performance [26 27] It has been incorporated in low power applications such as a FFT processor and hearing aids [28 29] and can be applied in biomedical microsystems which have very limited power available. For the 130nm technology the threshold voltages for nMOS and pMOS are 0.38mV and 0.33mV respectively. The expression for sub threshold leakage current is given by equation 2 18 [2 9] DS GST thV VV V nVth DSDSIIee0(1) ( 2 18) Where IDS0 is the drain current when VGS is equal to VT ( equation 2 19 ) [30] DS ox thW ICnV L2 00(1) ( 2 19) Vth is the thermal voltage and n is the subthreshold slope factor given by equations 2 20 and 2 21 respectively. thkT V q ( 2 20) d oxC n C 1 ( 2 21) For VDS > 4Vth, equation 2 18 can be reduced to equation 2 22 A t higher values of VDS the exponential term becomes negligible. GSTVV nVth DSDSIIe0 ( 2 22) 2 2 3 Jitter and P hase N oise in R ing O scillators Besides power saving s, there are some additional benefits of using single ended ring oscillators ( Figure 2 7 ) in terms of spectral characteristics Ref. [31] derives the PAGE 28 28 expressions for phase noise and timing jitter for both single ended and differential ring o scillators by using impulse sensitivity functions (ISF is a time varying constant that can determine the phase shift due to a noise source) and states that single ended ring oscillators have lower phase noise than their differential counterparts for a given power and frequency. It also states that the timing jitter in single ended oscillators can be minimized by equalizing the rising and the falling times. Figure 2 7 Single ended ring oscillator with ident ical N stages Ref. [32] derives the expression for phase noise in ring oscillators in terms of power dissipation, temperature, frequency of oscillation and offset frequency by analyzing the time domain jitter. Some of th e key results are discussed her e The variance of timing jitter for switching based relaxation and ring oscillators is directly proportional to the variance in the control voltage at the input. This variance in the control voltage at any time t is given by equation 2 23 [32] tRC tRC c nkTR Vtee CR2 2/22/ 0()(1) ( 2 23) Where k is the Boltzmann constant, T is the temperature, R and C are the net resistance and capacitance at the input node, Rn is the equivalent thermal noise resis tance and 2 0 is the variance of the control voltage at t=0. The variance in switching tim e jitter can be calculated by substituting equation 2 23 in equation 2 24 [32] Total N Stages 1 2 N PAGE 29 29 c ocdV TVt dt2 22() ( 2 24) In the model used for calculating the expressions for timing jitter of a ring oscillator similar to the one shown in Figure 2 7 each inverter ( Figure 2 8 ) can either be in ON or OFF states. In the ON state a constant current I either char ges or discharges the load capacitor while in the OFF state no current is drawn from the supply. Figure 2 8 An inverter stage in the ring oscillator In the ON state the output resistance is 1/gds while in the OFF state the output resistance will be 1/gd0. Where gd0 is given by equation 2 25 DSdds Vgg0 0 ( 2 25) If the oscillator is working in the subthreshold region then the output conductance gds can be calculated as in equation 2 26 by substituting t he value of IDS from equation 2 18. DS GST thV VV V nVth DSDS ds DS thdIIee g dVv0. ( 2 26) The output conductance (gd0) at VDS =0 is then given by equation 2 27. 1 2 PAGE 30 30 GSTVV nVth DS d thIe g V0 0 ( 2 27) T he thermal noise for a transistor can be represented by a parallel current source connected between the drain and source of a transistor as shown in Figure 2 9 Figure 2 9 Noise represented by a parallel current source in a transistor For a transistor operating in the subthreshold region, the equivalent noise resistance (Rn) is given by equation 2 28 [33] n dR g02 ( 2 28) After representing the exponential function in terms of its Taylor series and substituting kT C2 0 assuming tRC equation 2 23 can be represented as equation 2 29 [32] c nkTtt Vt CRCRC222 ()1 ( 2 29) Now substituting ndRg02 dsRg 1 and DDtCVI 2 in equation 2 29, it can be represented as equation 2 30 For the condition DSthVV 4 gds can be neglected. dDDdsDD cgVgV kT Vt CII2 0()1 2 ( 2 30) Using equation 2 30 the variance switching time ji tter (equation 2 24 ) is given by equation 2 31 nkTf IkTGf R24 4 PAGE 31 31 doDD sgV kTC T II2 21 2 ( 2 31) In a ring oscillator with N stages there are 2N independent switc hing events in each period Thus, the net timing jitter for a ring oscillator operating in the subthreshold region is given by equation 2 32. doDD DDkTTgV T NCVI2 2 0 0 22 1 2 ( 2 32) Here the nominal period oscillation (T0) is given by equation 2 33 [32] DDNCV T I0 ( 2 33) Now substituting the values of gd0 and I corresponding to the subthreshold region of operation from equations 2 27 and 2 22 respectively, the expression of timing jitter for the ring oscillator is given by equation 2 34. DD DD thkTT V T NCVV2 2 0 0 22 12 ( 2 34) For an oscillation frequency of f0, the power consumption for a ring oscillator with N stages is approximately given by equation 2 35 Thus the relationship between timing jitter and power consumption at a given oscillation frequency (f0) is given by equation 2 36 We can clearly see that the timing jitter is inversely proportional to the power consumption and should decrease in value at the expense of more power. DDPNCV2 ( 2 35) DD thV kT T PfV2 0 02 1 2 ( 2 36) PAGE 32 32 Once the variance of timing jitter is determined, the phase noise at a given offset frequency can also be calculated by using equation 2 37 [32] fT PNf fTf2 3 00 2 2 2 3 00() ( 2 37) Thus at much higher frequency offsets the phase noise can be approximated by equation 2 38 DD thf V kT PNf PVf2 02 ()1 2 ( 2 38) For validating this model a ring oscillator with three inverters ( Figur e 2 10) operating in the subthreshold region was designed and simulated in Cadence SPECTRE (APPENDIX A) The W/L for the p MOS and nMOS was set to (2m/120nm) and (1.58 m/120nm) respec tively. At 250mV the oscillation frequency was 34.8MHz and the power consumption of the inverters in the ON state turned out to be 8 0 nW approximately. Figure 2 10. Ring o scillator to verify the phase noise model VDDGND OUT PAGE 33 33 The simulated and the predicted phase noise from equation 2 37 for this oscillator are compared in Fig ure 2 11 From this plot we can see that the phase noise at an offset frequency of 1 MHz is 93.74dBC/Hz. The expected value of the phase noise at this offset frequency from equation 2 38 is 90.2 dBc/Hz (at room temperature). Figure 2 11. Comparison of s imulated and expected phase noise for a ring oscillator running at 34.8MHz Further, the oscillation frequency was varied by adding more inverter stages in the ring oscillator and keeping the supply voltage the same i.e at 250mV. The expression for minimum achievable phase noise for ring oscillat ors is given by equation 2 39 [32]. f kT PNf Pf2 07.33 () ( 2 39) The phase noise is computed for different oscillation frequencies by using both formulas (equation 2 39 and equation 2 38) and are results are mentioned in Table 4 1 We can clearly see that the value given by equation 2 38 is more close to the simulation result T hus, we c an use this model for a better estimate o f the phase noise 104 105 106 107 120 110 100 90 80 70 60 50 40 Frequency Offset (Hz)Phase Noise (dBc/Hz) Simulated Predicted PAGE 34 34 performance of a ring oscillator working in the subthreshold region for a given power constraint. Table 2 2 Comparison of phase noise at 1MHz offset frequency at different oscillation frequencies Number of stages in the Oscillator (N) Oscillation Frequency (MHz) Power Consumption (nW) PN at 1MHz Offset (dBc/Hz) ( eq 2 39) PN at 1MHz Offset (dBc/Hz) (eq. 2 38) PN at 1MHz Offset (dBc/Hz) (Simulation) 3 34.8 58 91.9 97.62 93.74 5 20.3 52.5 96.25 101.87 98.3 7 14.43 52.9 99.23 104.86 101.8 9 11.22 53.25 101.45 107 104.5 PAGE 35 35 CHAPTER 3 SYSTEM ARCHITECTURE 3 1 Basic Architecture of the S ystem A simplified block diagram for a typical passive biomedical microsystem is shown in Figure 3 1 It consists of an antenna port, a power management unit comprised of a voltage rectifier, voltage regulator and bias circuitry, an ASK modulator /demodulator, a clock generator and a di gital baseband signal processor. Figure 3 1 Block diagram of a typical biomedical microsystem with sensing and data communication capability The power management unit is one of the most important components of the entire system as it provides the required voltages needed for operation by the various blocks and its efficiency is critical to s ystem performance. The device is powered up by continuously sending a RF wave, which is rectified into a DC voltage. The power efficiency of the system can be improved by separating the supplies for analog and digital blocks (VDDH and VDDL) Digital blocks can be operated at a lower supply voltage which results in power saving s. This allows the analog blocks to be unaffected by the voltage spikes which appear due to signal transitions in the digital blocks. Although, the Analog Interface VDDLVDDH Sensor CLK DOUTDIN Power Management Unit ASK Demodulator ASK Modulator Clock Generator Signal Processor and Control Logic PAGE 36 36 dynamic power dissipation decreases with the supply voltage scaling, t he leakage current increases as it is exponentially dependant on the supply voltage [34] The reference clock and commands for configuring the system are sent as ASK modulated signals al ong with the carrier. This information is extracted by the ASK demodulator and provided to relevant blocks. The b ackscattering scheme described in [22] is generally employed for sending the data back to the interrog ator where the antenna impedance is varied between either a perfect match or complete mismatch i.e a short. 3 2 Digital Phase Locked Loop Figure 3 2 shows the functional block diagram of the sub threshold D PLL operated from a 260mV supply Figure 3 2 Implemented digital phase lock loop A binary phase and frequency detector ( B PFD) compares the divided down clock from the digitally controlled oscillator (DCO) with the extracted reference clock (CLKRX) from the ASK demodulator to generate a single bit early/late signal. The digital loop filter Z1 + + KP 3 KI 2 PFD CLKRX 10 7 3 1 DECODER COARSE 32 8 FINE DCO 1/M 1/N LOOP FILTER CLKTX PAGE 37 37 with programmable proportional (KP) and integral (KI) path gains is updated on every cycle of the divided down (1/(M*N)) clock producing a 10bit output based on the early/late signal Out of these 10 bits, the 3 least significant bits are used by a sigmadelta ( ) sampled at 1/M of the output clock (CLKTX) frequency to produce a bit stream The sigma delta output, along with the remaining 7 bits from the loop filter output, forms an 8 bit control word for the DCO. The five most significant bits (MSBs) of the control word are used for tuning the coarse delay stage through a 5to 32 bit decoder (COARSE [31:0]) and the three least significant bits (LSBs) control the fine tuning cell through a 3to 8 thermometric decoder. In order to keep the hardware complexity of the loop filter low the programmable constants are chosen to be a power of 2. Multiplication/division operations are performed by left/right shifting of the bits based on the value of KI. 3 2 1 B angBang P hase Frequency D etector A digital phase detector can be impl emented in many ways, such as a XOR gate or a J K flip flop where the duty cycle of the output indicates the phase difference between the two signals being compared. These phase detectors can track the phase error only when the error is confined within a very small range [5] T his limitation can be overcome by using a phase frequency detector (PFD). A low pass filter is required to be used with the previously mentioned phase detectors to determine the DC content of the output. A bang bang p hase and frequency detector (BB PFD) is a special case where the phase difference between its input signals is quantized by a single bit resolution. In this DPPL design a BBPFD similar to [35 ] has been used and is shown in Figure 3 3 In this implementation, a conventional PFD is followed by a sampling flip flop. The UP signal PAGE 38 38 goes high whenever FREF ar rives and the DOWN signal goes high at the rising edge of the FDIV signal. By sampling the UP signal with the DOWN signal, the sign of the error is determined. Thus a 1 output specifies that the divided down clock is lagging the reference signal and DCO frequency needs to be increased. The output (E/L) remains either high or low for the entire period of the divided down clock. The output signal does not have a 2 periodicity as it provides just the direction of error corresponding to all phase differences at the input and because of this the transition dynamics are very smooth [10] Figure 3 3 BangBang PFD 3 2 2 Digital L oop F ilter A digital loop shown in Figure 3 4 with programmable integral and proportional path constants has been used in this design. It operates at the divided down clock frequency and while in the locked state, it computes the contr ol word every reference clock cycle Based on the PFD output (E/L) the value stored in a 13bit accumulator is incremented or decremented on each rising edge of the clock, followed by a multiplication by the integral path constant (KI). In order to keep th e implementation simple KI is chosen to be a power of 2 and multiplication is performed by simply shifting FREF D Q CLR D Q UP DOWN R D Q CLK E/LFDIV CLR PAGE 39 39 the bits of the accumulator which generates a 10 bit result. The proportional path constant (KP) is directly added to this result and a 10bit contr ol word (CW) is obtained after checking for overflows The values of KP and KI affec ts the closed loop characteristics of the entire DPLL and a root locus based analysis approach is described in [16] and [10] KP affects the damping factor and KI affects the bandwidth. A larger value of KI also helps in fast frequency acquisition in the intial phase when the PLL is unlocked but results in a longer phase capture time [36] As discussed earlier, a higher damping factor is required to make the system more stable. However, a higher value of KP also increases the jitter in the system This is due to the control word changing by a factor of (2*KP +1) whenever the PFD output toggles A higher value of KP/ KI is required to meet the stability criteria described by equation 2 14. In the present design, KP and KI are both programmable and their ratio can assume a maximum value of (7/0.125=28) corresponding to different system requirements. Figure 3 4 Digital loop filter Z1 + + KPKIE/L 3 2 3 10 10 13 13CW PAGE 40 40 3 2 3 Sigma D elta M odulator A first order sigma delta modulator shown in Figure 3 5 has been used to obtain an increased frequency resolution for the DPLL. It operates at a higher frequency than the loop filter and oversamples the loop filter output by a factor of M. The 3 LSBs from the 10bit control wor d output of the loop filter (CW ) are added to the previous value of the accumulator and t he generated carry out bit is supplied as the output (SD_OUT). The sigma delta also has a high frequency noi se shaping transfer function that shifts the phase noise toward higher frequencies; the noise is ultimately filtered out because of the low pass phase response of the PLL [10] Figure 3 5 Sigma delta modulator 3 2 4 Digitally C ontrolled O scillator As discussed in section 2. 2.3 a ring oscillator topology consists of an odd number of inverters are connected in a feedback loop and t he oscillation frequency (equation 3 1 ) is determined by the delay of each stage ( ). In order to have a low oscillator frequency either a large number of stages has to be used or the delay of each stage has to be increased substantially. A ring oscillator based DCO topology with tunable D Q CLK 3 3 3 CW[2:0]+SD_OUT PAGE 41 41 coarse and fine delay stages shown in Figure 3 6 has been implemented to work at very low supply voltages. By operating the inverters in the subthreshold region, the delay of each stage increases dramatically Further, connecting the outputs of all t he coarse delay stages together increases the output capacitance of the coarse delay stage. oscf N 1 2 ( 3 1 ) The coarse tuning stage is implemented as a 32to 1 delay select path architecture with tri state buf fers acting as selection switches The shortest coarse delay path consists of 3 inverters while the longest path consists of 65 inverter s. The five most significant bits of the control word are used for tuning the coarse delay stage through a 5 to 32 bit decoder (COARSE[31:0]). A variable number of inverters have been used for coarse delay selection in [37] and [11] However, the frequency resolution of the coarse delay stage alone is not sufficient and a fine delay st age is required to obtain an increased resolution. The fine tuning st age has tri state buffers connected in parallel with inverters that are activated by an 8bit thermo metric code (FINE [7:0]), based on the remaining 3bits of the control word. The shunted tri states control the current drive strength at each node and determine the delay in the loop. The fine delay stage covers one coarsetuning step to obtain a monotonically rising frequency response, failure to ensure this could lead to an unstable PLL [9] The 32 coarse delay steps and 8 fine delay steps together allow the DCO to have the capability of generating 256 different frequencies. Each course delay stage adds a delay of approximately 28ns, while each tri state buffer in the fine delay stage changes the delay by about 3ns. PAGE 42 42 Figure 3 6 Digitally controlled oscillator The simulated DCO response is shown in Figure 3 7 for the TT corner with a supply voltage of 260 mV. As can be seen the DCO frequency changes linearly for lower values of the control word but the response becomes non linear for higher values of the control word. The DCO frequency is also very sensitive to the supply voltage va riations while operating in subthreshold region. This problem was overcome by using a voltage regulator to produce a constant supply voltage. Figure 3 7 Frequency versus control word COARSE DELAY FINE[ 1 ] FINE[ 0 ] FINE[ 2 ] FINE[ 3 ] FINE[ 5 ] FINE[ 4 ] FINE[ 6 ] FINE[ 7 ] FINE DELAY OUT C31 C0 COARSE[3 1 ] COARSE[0] 0 50 100 150 200 250 0.5 1 1.5 2 2.5 3 3.5 x 106 Control WordDCO Frequency (Hz) PAGE 43 43 3 3 Sizing of Digital Circuits in Subthreshold R egion All the blocks of the DPLL described in previous sections are operated in the subthreshold region to significantly reduce the power consumption. Also by reducing the supply voltage, leakage current due to the Drain Induce Barrier Lowering (DIBL) e ffect can be significantly reduced since it is directly proportional to the supply voltage. The current equations for the subthreshold region of operation are discussed in Section 2.2.3. T he focus of discussion of this section is the ION to IOFF ratio which can be used as a measure of robustness for digital circuits as in [29] ION (ON current) is the current that the MOSFET can source when it is fully turned on. IOFF (OFF current) is the current that leaks through the MOSFET even when it is turned off. At high operating voltages this ratio is significantly large and ensures robustness of the circuit. But in the subthreshold region of operation this ratio can get quite low, leading to slow charging or discharging of the output node and can cause circuit failures Figure 3 8 sho ws that the ION to IOFF ratio for an nMOS transistor (130nm) can approxi mately have a maximum value of 4 000 in the subthreshold region. Figure 3 8 ION to IOFF ratio in subthreshold region 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 500 1000 1500 2000 2500 3000 3500 4000 4500 Supply Voltage (VD D)IO N/IO F F PAGE 44 44 Thus, care has to be taken while sizing the gates to ensure functionality across all process corners The effect of ION to IOFF ratio on the sizing of the digital circuits operating i n the sub threshold region can be understood from the following discussion. In the super threshold regime, CMOS digital circuits are usually sized such that the drive strength s of PMOS and NMOS transistors are roughly equal. A typical sizing ratio of 2:1 is often chosen where the PMOS is sized twice the size of the NMOS. This sizing is chosen so that the PMOS and NMOS can source the same amount of current to charge or discharge the load. This is usually done to have r oughly equal rise and fall delays resulting in a trip voltage of VDD/2 and obtain a high noise margin. This sizing ratio doesnt hold well in subthreshold regime. This is because the current doesnt follow the same quadratic profile with respect to the d r ain s ource, g ate and b ulk v oltages in the subthreshold region. For proper functioning of the digital circuit the ON complimentary NMOS and PMOS networks should be able to provide enough drive current to pull down or pull up a node when the other network is OFF within the time period of the desired frequency of operation. With process scaling the leakage current of the transistors increases. Added to the sub threshold current is the gate leakage current whose impact becomes more significant with every successive technology generation. This OFF state leakage current can be significantly large enough to prevent a no de from being pulled up or down in the subthreshold regime, where the ION and IOFF ratios are drastically low. For example in Figure 3 9 consider the inputs A=1 and B=1 which are applied to the NAND gate. When operated at very low voltages (in subthreshold), the two PMOSs in parallel might supply enough leakage current to prevent the ON state NMOSs from PAGE 45 45 pull ing down the output node. The fact that the NMOSs are in series does not help this situation either because it further reduces the available current. With the added unpredictability of the PMOSs being fast er than the NMOSs the output node might always be st uck at logic I ncreasing the operating voltage would mitigate the problem, as the effect of the leakage current (IOFF) on the overall charging/ discharging behavior of the output node would be significantly reduced. This would happen because operation at higher voltages increases ION, allowing it to be much larger than IOFF. In the light of the discussion above, a minimum operational voltage for different gates was found with an output swing limited to 10% 90% of the supply voltage across different process corners. It was found that for 130 nm technology standard cells were functional for 160 mV supply with a P/N ratio of 2:2. Therefore t he digital blocks are guaranteed to function b y operat ing at 250 mV. Figure 3 9 Leakage induced problems in a NAND gate operating in subthreshold region VDDA OUT B A B I OFF I ON GND PAGE 46 46 CHAPTER 4 SIMULATION AND MEASU REMENT RESULTS 4 1 Simulation Environment For mixed signal systems, the designer has to be extremely careful of the interactions between the digital and analog blocks. As stated in Section 2 PLLs can have pure analog, digital or mixed signal implementations. It is extremely difficult to simulate a PLL system because of the mixed analog d igital nature of internal signals [38] and feedback. I t needs to also be simulated for a large number of reference clock cycles to capture the transient response since a PLL shows a highly nonlinear behavior prior to locking. In order to verif y the correct functionality ( whether it ever locks to a frequency ) and evaluate other performance parameters such as timing jitter, device level simulations using circuit simulators such as SPECTRE or SPICE result in extremely impractical and long simulation time. Behavioral modeling techniques described in [39 40] and mixed signal simulation techniques described in [35] and [38] offer a promising solution to the problem of effectively verifying the functional behavior of the PLL as well as expedite the entire simulation process 4 1 1 Behavioral Model Behavioral modeling involves describing the functi on of a block with the use of concise, approximated mathematical equations or simple pseudo logic like code. There are no transistor level schematics, rather tools such as MATLAB, C, Verilog or System V erilog are used to describe the function ality of a blo ck. It empowers the designer to quickly integrate the blocks and perform system level simulations to analyze the feasibility and try out different configurations. PAGE 47 47 Functionally equivalent V erilog codes for various sub blocks such as the PFD, l oop f ilter, DCO and programmable di vider corresponding to the bang bang dig ital PLL described in Section 3.2.1 were written. The entire DPLL system was simulated in Model Sim for verifying the locking characteristics and analyzing the effect of the loop parameters on the system stability and acquisition time. A linear model of the DCO producing a sq uare wave and having 256 control steps was developed. The relation between the DCO output frequency (FOUT) and the control word (CW) from the digital loop filter is given by equations 4 1 and 4 2 ,where FH and FL are the highest and lowest frequencies of the DCO respectively. *outLDCOFFKCW ( 4 1 ) () 256HL DCOFF K ( 4 2 ) The effect of increasing the proportional path constant KP can be seen in Figure 4 1 In this particular simulation setup, a reference frequency of 16 k Hz was chosen with a multiplier factor of 64 to obtain a 1.024 MHz output clock As stated earlier in section 2.1. 3 increasing KP e ffects the damping factor and results in a reduced frequency acquisition time however this comes at the cost of increased jitter. Since each time the output of the PFD changes the control word changes by a factor of PIKK (2) thus higher the value of KP, the higher the control word ripple will be Next, to analyz e the benefit of using a sigma delta modulator it was shut off without changing any of the previous set of input parameters. As seen in Figure 4 2 the ripple in control word increases This is because the sigma del ta modulator pushes the noise t owards higher PAGE 48 48 frequencies which is ultimately filtered out due to the low pass response of the PLL. Figure 4 1 Control w ord v er sus time for two different values of KP To analyze the P LL response to a frequency step, a 16 k Hz reference clock was applied at the input followed by a 12.5 kHz clock. As can be seen in Figure 4 3 the PLL first locks to the 16 kHz clock and the control word settles down to produce an output 0 200 400 600 800 1000 1200 1400 1600 0 50 100 150 200 250 300 Reference Clock PeriodsControl Word 0 200 400 600 800 1000 1200 1400 1600 0 1 PFD Output KP=4, KI=1 N=64 0 200 400 600 800 1000 1200 1400 1600 0 50 100 150 200 250 300 Reference Clock PeriodsControl Word 0 200 400 600 800 1000 1200 1400 1600 0 1 PFD Output KP=7,KI=1 N=64 PAGE 49 49 frequency of 1.024 MHz The PLL locks to the new frequency after the reference cl ock changes. Figure 4 2 Control word v ersu s time with the sigma delta modulator disabled Figure 4 3 DPLL r esponse to a frequency step For phase step simulation, the reference frequency is initially set to 16 kHz and later a 75 degree phas e shift is applied at the input. A s can be seen in Figure 4 4 the 0 200 400 600 800 1000 1200 1400 1600 0 50 100 150 200 250 300 Reference Clock PeriodsControl Word 0 200 400 600 800 1000 1200 1400 1600 0 1 PFD Output KP=4,KI=1 N=64,SD_EN=0 0 500 1000 1500 2000 2500 0 50 100 150 200 250 300 Control Word 0 500 1000 1500 2000 2500 0 1 Reference Clock PeriodsPFD Output Locked States Frequency Step PAGE 50 50 PLL is abl e to quickly recover from the phase step and the control word reaches its previous value again. Figure 4 4 DPLL response to a phase step 4 1 2 Mixed Signal Simulations After behavioral modeling is complete and system specs have been decided, critical blocks in the system can be replaced by their transistor level schematics for simulations Cadence design suite comes with a large number of simulators which facilitate integration and simulation of subblocks imple mented at different levels of abstraction. For this particular system the V erilog model for the DCO was replaced by its transistor level counterpart operating from a 260 mV supply. In the mixed simulation mode the blocks are connected to each other and t hen the hierarchical editor is invoked to notify the tool which view (abstraction level) of a particular block is to be used in the simulation. Cadence uses Spectre Verilog to simulate such a schematic w hich contains some blocks with only V erilog model s and some with complete transistor level implementation. Next design partitioning takes place and necessary interface elements 0 500 1000 1500 2000 2500 0 50 100 150 200 250 300 Reference Clock PeriodsControl Word 0 500 1000 1500 2000 2500 0 1 PFD Output Locked States Phase Step PAGE 51 51 are placed between the transistor level and code level implementations. These interface elements behave as ADCs or DACs and it is p ossible to specify properties such as propagation delay, rise/fall times and input/out levels To demonstrate a locked state i n the first simulation setup, a reference frequency of 20 k Hz is applied at the input and the multiplier gain is set as 64. Figure 4 5 shows that the PLL locks with the DCO frequency of 1.28 MHz. Next a frequency step simulation is performed by first setting the reference clock to 40 k Hz and later chang ing it to 33 kHz wit h the multiplier gain set as 32. A s can be seen in Figure 4 6 the PLL first locks with an output frequency of 1.28 MHz and later w ith 1.06MHz. Although mixed signal simulations are more accurate, they take significantly more time to run than behavioral level simulations However, the required run time and resources are far less than the complete transistor level simulations Figure 4 5 Mixed signal simulation for lock acquisition Locked to 1.28 MHz PAGE 52 52 Figure 4 6 Mixed signal simulation for frequency step 4 1 3 Synthesis of D igital B locks A fter verifying the functional ity of the system by using the above mentioned simulation techniques synthesis of digit al blocks is carried out using various tools. Synopsys DC compiler was used to produce an optimized netlist for the behavioral Verilog model based on the prov ided 130nm standard cell library information. After putting the timing constraints in place, timing slack was checked for any set u p and hold time violations. The optimized netlist can also be tested using the same set of test fixtures which were incorporated with the previous behavioral simulations Next Cadence Encounter was used to perform automated place and route on the netlist to generate the layout for this design The layout of the chip is shown in Figure 4 7 where the top half consi sts of the synthesized components and a custom DCO is integrated at the bottom. Locked States Frequency Step PAGE 53 53 Figure 4 7 Chip layout 4 2 Measurement Results The passive transceiver system, con sisting of a RF DC multiplier, on chip power management circuitry, ASK demodulator ( envelope detector ) along with the digital PLL, was implemented using a 130 nm CMOS process. A die photo of the entire chip, with dimensions of 1.67 mm by 1.27 mm is shown in Figure 4 8 The DPLL occupies an area of 0.17 mm2 4 2 1 Test Measurement Setup The packaged chip was mounted on a PCB which consists of various input and o utput terminals connected to an oscilloscope through SMA connectors ( Figure 4 9 ) The necessary operating voltages for the DPLL, I/O pads and level c onv erters were supplied through DC voltage generators. An ammeter with a 1nA resolution was connected in series with the PLLs supply voltage to measure the average power 162 m110 m PAGE 54 54 consumption. An 8channel ADC bus on the oscilloscope was used to observe the control word from the loop filter. Figure 4 8 Die photo of passive transceiver with DPLL Figure 4 9 Pack ag ed chip mounted on a PCB for testing Programmable Transceiver DPLL 1670 m1270 mMOS CAPs SWITCHES FOR PROGRAMMING KPAND KI REF. CLOCK DIV. CLOCK PLL OUTPUT CONTROL WORD <7:0> (ADC BUS) I/O VDD VDD FOR PLL PAGE 55 55 4 2 2 Test Cases First a reference frequency of 40k Hz was applied at the input with the division ratio (M x N) set to 16. The value of KP was set to 3b001 (1) and KI to 2b01 (0.5). Th e expected DCO frequency is 640kHz in the locked state, this can be seen from the timing wav eforms shown in Figure 4 10, where the rising edge of the reference clock (CLK_REF) is properly phase aligned with the rising edge of the divided clock (CLK_DIV) from the DCO. The observed control word (CW) toggles between (0XA2 and 0XA3) and the DCO frequency (CLK_OUT) is 640MHz approximately. Signal E_L is the output of the binary phase and frequency detector. All these signals are internally 250mV and have been level converted to 1.2 V before sending them to the digital output pad. The measured spectrum of the DCO signal is shown in Figure 4 11. The jit ter measurements results for this set up are shown in Figure 4 12, the RMS and peak to peak j itter for the PLL output are 84.44ns and 518.18ns respectively while consuming 100nW of power at a supply voltage of 250mV. Figure 4 10. Measured waveforms for reference frequency of 40kHz and multiplier ratio set to 16 CLK_REF CLK_DIV CLK_OUT E_L CW <7:0> 1V/div 20s/div PAGE 56 56 Figure 4 11. Spectrum of PLL output with oscillation frequency of 640kHz Figure 4 12. Jitter histrograms for 40kHz reference clock A) Divided clock B) PLL output clock Next the input frequency was changed to 20kHz without changing the rest of the input parameters such as the supply voltage, multiplier ratio, KP and KI values. The expected PLL output frequency is 320kHz The measured spectrum of the PLL output CLK_REF CLK_DIV CLK_OUT CLK_REF A B PAGE 57 57 signal shown in Figure 4 13 confirms that the PLL locks with the DCO frequenc y fixed at 321kHz Figure 4 13. Spectrum of PLL output with oscillation frequency of 321kHz Figure 4 14. Jitter histrograms for 20kHz reference clock A) Divided clock B) PLL output clock 2ndHarmonic at 642 KHz CLK_REF CLK_DIV CLK_OUT CLK_REF A B PAGE 58 58 The jitter measurements results for this set up are shown in Figure 4 14. The RMS and peak to peak jitter for the PLL output frequency of 320 kHz approximat ely are 122ns and 927.23ns respectively. It can be seen here that jitter increases with decrease in the oscillation frequency. Next the control word of the DCO was fixed at 126 by keeping the RESET signal low and the supply voltage was varied. The oscillator frequency corresponding to this control word increases with each successive supply voltage step. In Section 2.2.3 the expressions for the variance of timing jitter for a ring oscillator operating in the subthreshold region were derived. The standard deviation of the timing jitter can be calculated by substituting T=3000K and k=1.38 x 1023 JK1 and taking the square root of equat ion 2 36 The measurement results for the timing jitter are compared against the expected result in Table 4 1 Table 4 1 Comparison of theoretical and measurement results for ring oscillator Supply Voltage (mV) Oscillation Frequency (f0) Power Consumption T 0 (Theoretical)) T 0 (Measured) 150 46.28 kHz 16.5 nW 6.48ns 725.36 n s 200 149 kHz 32 nW 2.89ns 210 ns 250 453.58 kHz 75 nW 1.18ns 42.57 ns 300 1.3MHz 225 nW 0.43ns 27 ns 350 3.7MHz 784 nW 15ps 12 ns 400 6.47MHz 1.7 W 8.9ps 3.44 ns 450 11.52MHz 3.73 W 4.3ps 2 ns 500 19.57MHz 7.88 W 2.38ps 1.46ns W e can see that the measured standard deviation of the timing jitter is much higher compared to the expected value. This is because this model only considers the noise due to the output resistance at each node and the effect of other noise sources such as the power supply voltage noise and substrate noise has not been account ed for. The level converter and the digital output pad buffer in the signal path also PAGE 59 59 contribute to the noise component. A s predicted by the model the timing jitter reduces with an increase in the supply voltage and oscillation frequency. The measurement r esults for peak to peak jitter at different supply voltages are shown in Figure 4 16. Figure 4 15. Peak to peak jitt er for different supply voltages and oscillation frequencies The change in the oscillation frequency at each voltage step can be understood as follow. The oscillation frequency for a ring oscillator is roughly determined by equation 4 3 where VDD is the supply voltage and I is the ON current ( equation 2 22) as explained in Section 2.2.3 The ratio of the frequencies corresponding to two different supply voltages (VDD1 and VDD2) should roughly be given by the relation in equation 4 4 DDI f NCV0 ( 4 3 ) DD DDfIV fIV221 112 ( 4 4 ) Substituting the values of the ON current corresponding to subthreshold region of operation equation 4 4 reduces to equation 4 5 150 200 250 300 350 400 450 500 100 101 102 103 104 Supply Voltage (mV)PeakPeak Jitter ( ns ) fo = 46.28KHz,P=16.5nW fo = 149KHz,P=32nW fo = 454KHz,P=75nW fo = 3.7MHz,P=784nW fo = 1.3MHz,P=225nW fo = 19.57MHz,P=7.88uW fo = 11.52MHz,P=3.73uW fo = 6.47MHz,P=4.25uW fo : Oscillation Frequency P : Power Consumption PAGE 60 60 GSGS thVV nV DD DDfV e fV21() 21 12 ( 4 5 ) Since VGS =VDD here relationship between the two frequencies is given by equation 4 6 where n~1.5 and Vth~26mV. DDDD thVV nV DD DDfV e fV21() 21 12 ( 4 6 ) Thus, when the supply voltage changes from 200m V to 250mV the oscillation frequency should cha nge by a factor of 2.9. As can be seen from Figure 4 15 that indeed the oscillation frequency changes by a factor 3. To determine the locking range of the PLL the ref erence frequency was swept for different supply voltages. The division ratio was fixed at 32 and the value of KP and KI were fixed at 3b001 and 2b00 respectively. For the typical typical (TT) and slow slow (SS) design corners the measurement results are shown in Figure 4 16 (A) and (B) respectively. We can see that at 250mV supply voltage, the tuning range of the PLL is from 310kHz to 1.5MHz and the power consumption ranges from 62.5nW to 185nW. The highest supply voltage at which the PLL is functional is 500mV after which the oscillator stops working. This is because at this voltage, the p MOS becomes too strong and the output node gets stuck at logical 1 and is not able to discharge through the nMOS PAGE 61 61 Figure 4 16. PLL lock range and power consumption at different supply voltages A) typical typical TT design corner B) slow slow SS design corner 102 103 104 100 150 200 250 300 350 400 450 500 550 Frequency (KHz)Supply Voltage ( mV ) Freq:35.2KHz179.2KHz Freq:102.42KHz512KHz Freq:310KHz1.5MHz Freq:864KHz4MHz Freq:2.12MHz6.9MHz Freq:4.28MHz18.14MHz Freq:7.68MHz33.8MKHz Freq:10.9MHz38.4MHz Power:18nW24nW Power:28nW56nW Power:62.5nW185nW Power:171nW636nW Power:490nW1.48uW Power:1.24uW4.73uW Power:3.72uW10.67uW Power:4.64uW14.6uW 101 102 103 104 105 100 150 200 250 300 350 400 450 500 550 Frequency (KHz)Supply Voltage ( mV) Freq:50KHz240KHz Power:15.2nW26.6nW Power:31.2nW86.4nW Freq:158KHz700KHz Freq:473KHz2.3MKHz Power:90nW330nW Power:272nW1.1uW Freq:1.25MHz5.4MHz Freq:2.8MHz11.9MHz Power:772nW3.12uW Power:1.76uW7.04uW Freq:5.24MHz21.3MHz Freq:9.5MHz37MHz Power:4.07uW15.22uW A B PAGE 62 62 CHAPTER 5 CONCLUSIONS Miniature embedded systems used in various biomedical applications have very stringent power, size and cost constraints. They are powered by a wireless RF signal which can also contain additional information such as the reference clock. Due to the limited power available and DC voltage fluc tuations designing a clock generator is extremely challenging. In this thesis a low power onchip clock generator was p roposed that can provide a 385 kHz to 1.54MHz clock at 260mV for digital baseband processing blocks and a backscattering modulator. The clock generator consumes ~200nW and is based on a sub threshold DPLL which synchronizes the onchip clock to an externally controlled low frequency ASK signal that modulates the incident RF carrier A single ended ring oscillator has been utilized in this DPLL design mainly because of the low area and power overhead. There are several benefits of using the proposed clock generator. It enables the frequency tuning over a large range of frequencies. By us ing the DPLL as a frequency multiplier simultaneous com mun ication with more than one peer can take place as each one can be programme d to work at different frequencies While the area overhead is negligible, a relatively stable clock can be obtained in a cost effective way. The Verilog model of the DPLL can be easily ported to other technologies. Most importantly a high system efficiency is achieved by operating the digital components of the DP LL in the sub threshold region. Because of low voltage headroom the static power dissipation of digital circuits is decreased considerably as it is exponentially dependant on the VDS of the transistors. The expressions derived in Section 2.2.3 can be used for designing a sub threshold ring oscillator to meet the phase noise specifications of a design. PAGE 63 63 APPENDIX A PHASE NOISE SIMULATI ON IN CADENCE The phase noise simulations were performed by performing the followi ng steps which are described in the Cadence help documents. B oth PSS and PNOISE analyses should be chosen to simulate phase noise in Cadence. The steps for PSS (periodic steady state) analysis are described as follows. Perform the transient simulation and determine the frequency of oscillation and the time after which the frequency stabilizes. In the analog design environment window chose the analysis type as pss. Set Beat Frequency as the oscillation frequency obtained from transient simulations Set Output harmonics as Number of harmonics and type in a number between 3 5. Choose Accuracy Defaults as either conservative or moderate. Put a sufficiently large value in the Additional Time for Stabilization textbox. The Save Initial Transient Res ults tab can be left blank. Click on the check box to enable the Oscillator tab and select the output node from the schematic. The reference node should be set as ground. The Sweep checkbox can be left blank Click on the Enabled checkbox. Once this is done, the next step is to complete the set up for the pnoise simulation for which the following steps need to be performed. In the analog design environment window chose the analysis type as pnoise. Select Sweep Type as relative and Relative H armonic as 1. Specify the frequency offset range for which the phase noise has to be determined. Select Sweep Type as automatic. PAGE 64 64 Select the number of Sidebands between 3 to 5. Select Output as voltage and select the output node from the schematic. Se lect ground as the negative output node. Select Input Source as none and select sources in the Noise Type tab. Activate the Enabled checkbox. After completing all the above mentioned steps, click the netlist and run button from the analog design environment window. After the simulation is complete go to the results tab and select Main Form. Choose pnoise in the Analysis section and select Phase Noise as the function. Click the Plot button. The phase noise plot should pop up. PAGE 65 65 LIST OF REF ERENCES [1] K. Choe, O. D. Bernal, D. Nuttman and Minkyu Je, "A precision relaxation oscillator with a self clocked offset cancellation scheme for implantable biomedical SoCs," in Solid State Circuits Conference D igest of Technical Papers, 2009, pp. 402403,403a. [2] W. Liu, K. Vichienchom, M. Clements, S. C. DeMarco, C. Hughes, E. McGucken, M. S. Humayun, E. De Juan, J. D. Weiland and R. Greenberg, "A neurostimulus chip with telemetry unit for retinal prosthetic device," IEEE Journal of SolidState Circuits, vol. 35, pp. 14871497, 2000. [3] T. B. Tang, S. Smith, B. W. Flynn, J. T. M. Stevenson, A. M. Gundlach, H. M. Reekie, A. F. Murray, D. Renshaw, B. D hillon, A. Ohtori, Y. Inoue, J. G. Terry and A. J. Walton, "Implementation of wireless power transfer and communications for an implantable ocular drug delivery system," Nanobiotechnology, IET, vol. 2, pp. 72 79, 2008. [4] Meng Lin Hsia, Yu Sheng Tsai and O. T.Chen, "An UHF passive RFID transponder using A low power clock generator without passive components," in IEEE International Midwest Symposium on Circuits and Systems, 2006, pp. 1115. [5] R. E. Best, PhaseLocked Loops: Design, Simulation and Applic ations. New York: McGraw Hill, 2007. [6] B. Razavi, Design of Analog CMOS Integrated Circuits. McGraw Hill Higher Education, 2003. [7] F. Gardner, "ChargePump Phase Lock Loops," IEEE Transactions on Communications, vol. 28, pp. 18491858, 1980. [8] F. M. Gardner, Phaselock Techniques. Hoboken, NJ: Wiley Interscience, 2005. [9] T. Olsson and P. Nilsson, "A digitally controlled PLL for SoC applications," IEEE Journal of Solid State Circuits, vol. 39, pp. 751760, 2004. [10] J. A. Tierno, A. V. Rylyakov and D. J. Friedman, "A Wide Power Supply Range, Wide Tuning Range, All Static CMOS All Digital PLL in 65 nm SOI," IEEE Journal of Solid State Circuits, vol. 43, pp. 4251, 2008. [11] Duo Sheng, Ching Che Chung and ChenYi Lee, "A fast lock in ADPLL with high resolution and low power DCO for SoC applications," in IEEE Asia Pacific Conference on Circuits and Systems, 2006, pp. 105 108. PAGE 66 66 [12] R. B. Staszewski, J. L. Wallberg, S. Rezeq, ChihMing Hung, O. E. Eliezer, S. K. Vemulapalli, C. Fernando, K. Maggio, R. Staszewski, N. Barton, Meng Chang Lee, P. Cruise, M. Entezari, K. Muhammad and D. Leipold, "All digital PLL and transmitter for mobile phones," IEEE Journal of SolidState Circuits, vol. 40, pp. 24692482, 2005. [13] R. B. Staszewski, S. Vemulapalli, P. Vallur, J. Wallberg and P. T. Balsara, "Timeto digital converter for RF frequency synthesis in 90 nm CMOS," in Radio Frequency Integrated Circuits (RFIC) Symposium, 2005, pp. 473476. [14] J. P. Hein and J. W. Scott, "z domain model for discretetime PLL's," IEEE Transactions on Circuits and Systems, vol. 35, pp. 13931400, 1988. [15] V. Kratyuk, P. K. Hanumolu, UnKu Moon and K. Mayaram, "A Design Procedure for All Digital Phase Locked Loops Based on a ChargePump Phase LockedLoop Analogy," IEEE Tr ansactions on Circuits and Systems II: Express Briefs, vol. 54, pp. 247251, 2007. [16] A. M. Fahim, "A compact, low power low jitter digital PLL," in Proceedings of the 29th European Solid State Circuits Conference, 2003 pp. 101 104. [17] Yehui Sun and Hui Wang, "Analysis of digital bang bang clock and data recovery for multi gigabit/s serial transceivers," in Custom Integrated Circuits Conference, 2009, pp. 343346. [18] N. Da Dalt, "A design oriented study of the nonlinear dynamics of digital bang bang PLLs," IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 52, pp. 2131, 2005. [19] L. Lincoln, K. Leung and H. C. Luong, "A 7CMOS for passive UHF RFID EPC G2 tags," in 33rd European Solid State Circuits Co nference, 2007, pp. 412 415. [20] Chi Fat Chan, Weiwei Shi, Kong Pang Pun, Lai Kan Leung, Ka Nang Leung and Chiu Sing Choy, "A lowpower signal processing front end and decoder for UHF passive RFID transponders," in IEEE International Symposium on Circuits and Systems, 2009, pp. 15811584. [21] R. Barnett and Jin Liu, "A 0.8V 1.52MHz MSVC relaxation oscillator with inverted mirror feedback reference for UHF RFID," in Custom Integrated Circuits Conference, 2006, pp. 769772. [22] Jong Wook Lee and Bomson Lee, "A Long Range UHF Band Passive RFID Tag IC Based on High Design Approach," IEEE Transactions on Industrial Electronics, vol. 56, pp. 2308 2316, 2009. PAGE 67 67 [23] F. Cilek, K. Seemann, G. Holweg and R. Weigel, "Impact of the local oscillator on baseband process ing in RFID transponder," in International Symposium on Signals, Systems and Electronics, 2007, pp. 231234. [24] ChiFat Chan, Kong Pang Pun, KaNang Leung, Jianping Guo, Lai Kan Leung and ChiuSing Choy, "A Low Power Continuously Calibrated Clock Recover y Circuit for UHF RFID EPC Class 1 Generation2 Transponders," IEEE Journal of Solid State Circuits, vol. 45, pp. 587599, 2010. [25] X. Zhao, R. Chebli and M. Sawan, "A wide tuning range voltagecontrolled ring oscillator dedicated to ultrasound transmit ter," in Proceedings. the 16th International Conference on Microelectronics, 2004, pp. 313316. [26] Myeong Eun Hwang and K. Roy, "ABRM: Adaptive Ratio Modulation for ProcessTolerant Ultradynamic Voltage Scaling," IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 18, pp. 281290, 2010. [27] A. Wang, A. P. Chandrakasan and S. V. Kosonocky, "Optimal supply and threshold scaling for subthreshold CMOS circuits," in IEEE Computer Society Annual Symposium on VLSI Proceedings, 2002 pp. 5 9. [28] C. H. Kim, H. Soeleman and K. Roy, "Ultra lowpower DLMS adaptive filter for hearing aid applications," IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 11, pp. 10581067, 2003. [29] A. Wang and A. Chandrakasan, "A 180mV subthreshold FFT processor using a minimum energy design methodology," IEEE Journal of SolidState Circuits, vol. 40, pp. 310319, 2005. [30] B. H. Calhoun, A. Wang and A. Chandrakasan, "Modeling and sizing for minimu m energy operation in subthreshold circuits," IEEE Journal of SolidState Circuits, vol. 40, pp. 17781786, 2005. [31] A. Hajimiri, S. Limotyrakis and T. H. Lee, "Jitter and phase noise in ring oscillators," IEEE Journal of SolidState Circuits, vol. 34, pp. 790804, 1999. [32] R. Navid, T. H. Lee and R. W. Dutton, "Minimum achievable phase noise of RC oscillators," IEEE Journal of SolidState Circuits, vol. 40, pp. 630637, 2005. [33] S. Liu, J. Kramer, G. Indiveri, T. Delbruck and R. Douglas, Analog VLSI: Circuits and Principles The MIT Press, 2002. [34] International Conference on Microelectronics, 2007, pp. 283286. [35] D. Bhatia, Digital Aided Synchronization and Mixed Signal Modeling of High Frequency Dc Dc Converters [Electronic Resource]. Gainesville, FL: University of Florida, 2009. PAGE 68 68 [36] Jri Lee, K. S. Kundert and B. Razavi, "Analysis and modeling of bang bang clock and data recovery circuits," IEEE Journal of SolidState Circuits, vol. 39, pp. 15711580, 2004. [37] Ching Che Chung and ChenYi Lee, "An all digital phaselocked loop for highspeed clock generation," IEEE Journal of SolidState Circuits, vol. 38, pp. 347351, 2003. [38] B. A. A. Antao, F. M. El Turky and R. H. Leonowich, "Mixedmode simulation of phaselocked loops," in Proceedings of the IEEE Custom Integrated Circuits Conference, 1993, pp. 8.4.18.4.4. [39] F. Herzel and M. Piz, "System level simulation of a noisy phaselocked loop," in European Gallium A rsenide and Other Semiconductor Application Symposium, 2005, pp. 193196. [40] M. Hinz, I. Konenkamp and E. H. Horneber, "Behavioral modeling and simulation of phaselocked loops for RF front ends," in Proceedings of the 43rd IEEE Midwest Symposium on Ci rcuits and Systems, 2000, pp. 194197, vol.1. PAGE 69 69 BIOGRAPHICAL SKETCH Tanuj Aggarwal was born in 1984 in Rishikesh, India. He received his b achelor s degree in Information and C ommunication technology from Dhirubhai Ambani Institute of Information and Communication Technology (DA IICT), Ga ndhinagar, India in 2007 and masters degree in Electrical and Computer Engineering from University of Florida, Gainesville, Florida in 2010 respectively. He worked as a Research Assistant in the Integrated Circuit Research Lab (ICR) from June 2008 to May 2010. The focus of his research was the implementation of low power digital circuits for wireless communication applications. His research interests include design and implementation of low power digital circuits, development of design flows to aid fast simulation and verification of mixed signal systems and computer architecture. 