Published on

05-Oct-2016View

212Download

0

Transcript

- 956 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 57, NO. 12, DECEMBER 2010 A Frequency Model of a Continuously Driven Clocked CMOS Comparator Shunsuke Okura, Hajime Shibata, Member, IEEE, Tetsuro Okura, Toru Ido, Member, IEEE, and Kenji Taniguchi, Member, IEEE Abstract—A frequency model of a continuously driven clocked CMOS comparator with the effect of the input signal during regeneration is presented. The model utilizes a small-signal linear model derived from the theoretical analysis of the comparison error caused by the transition from the tracking mode to the regeneration mode. The comparison error voltage is a function of input signal frequency and is represented with the transfer function. The correctness of the model is assured by several transistor-level simulation results. The model provides a valuable insight for the design of high-speed comparators. Index Terms—Clocked comparator, CMOS, frequency re- sponse, metastability, regeneration. I. INTRODUCTION FOR THE design of analog-to-digital converters (ADCs),a comparator is a crucial and often a limiting component toward high-speed operation because of its finite accuracy and comparison speed. Among the main building blocks of comparators, flip-flops and latches have been well analyzed. A formula describing the probability of a metastable state occur- rence has been presented in [1]. A formula for the sensitivity of a latch in a sense amplifier has been described in [2]. The latch has been theoretically analyzed with small-signal models in [3] and [4]. Moreover, there have been many reports on high- speed ADC or comparators that discuss the effect of the static offset voltage, the metastability, the reset time of the latch [5]– [10], the frequency response of the preamp [7], [11], and design optimizations of high-speed comparators [4], [11], [12]. This brief presents a frequency model of continuously driven clocked CMOS comparators taking into account the effect of the input signal during regeneration. A clocked comparator has two modes, namely, the tracking mode and the regener- ation mode. According to the study on the transition from the tracking mode to the regeneration mode, the comparison result depends on the input signal frequency. This dependence is caused by the input signal change just after the beginning of regeneration while the regeneration gain is still small. The Manuscript received April 19, 2010; revised July 24, 2010; accepted October 11, 2010. Date of current version December 15, 2010. This work was supported in part by a grant from the Global COE Program, “Center for Electronic Devices Innovation,” from the Ministry of Education, Culture, Sports, Science and Technology of Japan. This paper was recommended by Associate Editor J. Paramesh. S. Okura, T. Okura, T. Ido, and K. Taniguchi are with the Graduate School of Engineering, Osaka University, Suita 556-0871, Japan. H. Shibata is with Analog Devices, Inc., Toronto, ON M5G 2C8, Canada. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TCSII.2010.2087972 Fig. 1. Schematic of a clocked CMOS comparator. comparison error that depends on the input signal frequency is named ac error in this brief. Based on the small-signal analysis, the ac error voltage is given by a simple function of input signal frequency. The frequency response of the comparator is modeled with the derived ac error. The presented model is useful for designs using high-speed comparators. In previous work, the dynamic nature of the comparator that includes the frequency response of the latch is analyzed [13]. This brief delivers a frequency model based on the detailed analysis. This brief is organized as follows. Section II shows that the comparison result depends on the input signal frequency. The ac error voltage is derived by the theoretical analysis with a small-signal model, and the comparator is modeled to represent the frequency dependence in Section III. Section IV provides SPICE simulation results with a 0.18-μm CMOS process model to confirm the derived model. The conclusion is drawn in Section V. II. EFFECT OF THE INPUT SIGNAL CHANGE Fig. 1 shows a schematic of the clocked CMOS comparator having two parts, a preamp and a regenerative latch. The input signal is continuously applied to the comparator. The preamp drives the latch according to the voltage difference between vin and vref . When CLK is high, the latch is in a reset state and tracks the difference signal with nonzero ON-resistance of the switch. At this tracking mode, the comparator shows a negative exponential response since vout is shorted to ground. Just at the time of the switch turning to open, vout is regeneratively amplified to the rail-to-rail level by a positive feedback loop in the amplifier of latch, and finally, the comparison result of vin and vref is stored in the latch. Since the regeneration time constant is a finite nonzero value, the signal generated by the preamp during regeneration introduces the ac error voltage, which depends on the input signal frequency. Fig. 2 shows the transient simulation results of the clocked comparator, in which the ac error voltage is described. The 1549-7747/$26.00 © 2010 IEEE
- OKURA et al.: FREQUENCY MODEL OF A CONTINUOUSLY DRIVEN CLOCKED CMOS COMPARATOR 957 Fig. 2. Comparator transient response depending on the input signal fre- quency. (a) Input signal vin. (b) Comparator output vout. input signal is a sine wave whose frequency is 200, 160, 120, and 80 MHz, which is continuously applied to the comparator input [see Fig. 2(a)]. The reference signal is a 40-mV dc voltage. The comparator switches from the tracking mode to the regeneration mode at 50 ns. At this instance, all input signals are 0 mV, and all tracked signals are negative. As shown in Fig. 2(b), the outputs of 200- and 160-MHz cases increase, whereas the outputs of 120- and 80-MHz cases decrease. This means that the comparison result depends on both the difference signal and the ac error voltage, which becomes larger as the frequency increases. The ac error voltage is a function of input signal frequency because it is introduced by the input signal changes in a short period just after the clock edge while the regeneration gain is still small. III. SMALL-SIGNAL ANALYSIS A. Theoretical Derivation The ac error voltage is studied with a small-signal circuit on the transition from the tracking mode to the regeneration mode. For simplicity, the comparator switches from the tracking mode to the regeneration mode at 0.0 s. The comparator is first analyzed in the tracking mode to derive the initial value in the regeneration mode. The regeneration process is then analyzed. The ac error voltage is derived from the zero-to-one transition point V0to1, which is the reference voltage when the comparator is in a metastable state for a given input signal. The comparison result is logic high if vref < V0to1, whereas it is logic low if vref > V0to1. The regeneration response is represented with the positive exponential term. When the coefficient of the positive Fig. 3. Equivalent circuit of the comparator. exponential term is 0.0, the comparator results in a metastable state, and V0to1 is derived. The ac error voltage is given by Verr,ac = V0to1 − vin(0) (1) where vin(0) is the sampled input signal. The comparator is simplified as the equivalent circuit shown in Fig. 3. The preamp is composed of an amplifier and a transconductance gmin [14]. The amplifier is not broken down into components since it is conventionally well known. The switching conductance is gsw. The negative conductance −gml provides regenerative feedback of the latch. The variable γ, which is the ratio of gsw to gml, is kept above 1.0 so that the latch can reset the prior data stored. The static dc offset voltage Vos,static is in series to the negative conductance because the comparator has a potentially large offset voltage originating from the latch whose device size is kept small to minimize the regeneration time constant and power consumption. The output capacitance is represented as CL. A sinusoidal input signal applied to the comparator is given by vin(t) = sin(ωint + θin) (2) where ωin and θin are the frequency and the phase, respectively. The input signal amplitude is normalized to 1.0 V. The transfer function of the amplifier in the preamp Hpreamp(s) is given by Hpreamp(s) = Apreamp 1 + sτpreamp (3) where Apreamp and τpreamp are the dc gain and the time constant, respectively. The amplifier output is given by the Laplace transform of (2) and (3) as Vx(s)= Apreamp 1+sτpreamp ( s·sin(θin)+ωin ·cos(θin) s2+ω2in − vref s ) . (4) Based on KCL, the latch output during the tracking mode is given by sCLVout(s)+γgmlVout(s)+(−gml)· ( Vout(s)− Vos,static s ) = gminVx(s)(t
- 958 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 57, NO. 12, DECEMBER 2010 which is presented as vout(0−) =Apreamp gmin (γ − 1)gml · ( K1K2 sin(θin + θ1 + θ2) − vref − gml Apreampgmin Vos,static ) (8) K1 ≡ ( ω2inτ 2 preamp + 1 )−1/2 (9) θ1 ≡ −arctan(ωinτpreamp) (10) K2 ≡ ( ω2inτ 2 track + 1 )−1/2 (11) θ2 ≡ −arctan(ωinτtrack). (12) The tracking error, which has been analyzed in [7], [10], and [11], is not considered in this analysis, so that the comparator is supposed to be at steady state until 0.0 s. In the regeneration mode, KCL gives the output signal sCLVout(s)+(−gml)· ( Vout(s)− Vos,static s ) =gminVx(s)+CLvout(0−) (t≥0) (13) ⇒Vout(s)=− τregvout(0−)+ gmingml Vx(s)− Vos,static s 1−sτreg (14) τreg = CL gml (15) where τreg is the regeneration time constant [12]. Substituting (4) and (8) into (14), the inverse Laplace transform of the equation gives the output presented as vout(t) = ac1 · exp(−jωint) + ac1 · exp(jωint) + ac2 + ac3 · exp ( t τreg ) (t ≥ 0) (16) where the coefficients ac1, ac2, and ac3 are independent of time (see the Appendix). vout(t) either increases or decreases exponentially with time depending on the sign of ac3, whereas the first three terms in the right side of (16) do not contribute to the regeneration result. If ac3 > 0, vout(t) ends up with logic high, whereas vout(t) does with logic low if ac3 < 0. When ac3 = 0, the comparator is in a metastable state. From ac3 = 0, the zero-to-one transition point is represented as V0to1 =K1K2K3 sin(θin + θ1 + θ2 + θ3) + gml Apreampgmin Vos,static (17) K3 ≡ ( 1 + ω2inτ 2 reg )−1/2 (18) θ3 ≡ arctan (ωinτreg) . (19) The first term in the right side of (17) shows the magnitude (K1K2K3) and the phase shift (θ1 + θ2 + θ3) for sin(θin). The variables K1, θ1, K2, θ2, K3, and θ3, which are given by (9)–(12), (18), and (19), respectively, are the functions of the input signal frequency ωin. The second term in the right side of (17) indicates the input-referred voltage of the static dc offset, which does not affect the ac error. The ac error voltage is therefore represented as Verr,ac = K1K2K3 sin(θin + θ1 + θ2 + θ3)− sin(θin) (20) Fig. 4. Frequency model of the comparator. (a) With an additional summing block. (b) Replacement of (a). referring to (1). It is noted that the sampled input signal is given by vin(0) = sin(θin). The ac error voltage is the attenuated input signal with phase shifts. B. Behavioral Model The frequency response of the comparator is modeled with the derived ac error voltage. The comparator is conventionally modeled as a summing block with quantization noise added to the input signal [15]. When the ac error is taken into account, the comparator is modeled with an additional summing block, as shown in Fig. 4(a), where the sum of the input signal and the ac error is noted as a zero-to-one transition point. From (20), the sum of the sampled input signal and the ac error voltage is given by V ′0to1(ωin) = vin(0) + Verr,ac =K(ωin) sin (θin + θ(ωin)) (21) K(ωin) ≡K1 ·K2 ·K3 (22) θ(ωin) ≡ θ1 + θ2 + θ3 (23) noting that the variables Ki and θi (i = 1, 2, 3) are functions of the input signal frequency ωin. Since the sampled signal is sin(θin), (21) replaces the model of the comparator shown in Fig. 4(a) with that in Fig. 4(b). The transfer function H(s) is given by the magnitude K(ωin) and the phase θ(ωin). In order to derive H(s), magnitude Ki and phase θi (i = 1, 2, 3) are transformed as follows: K1 and θ1, which were given by (9) and (10), respectively, are transformed to K1 = ∣∣∣∣ 11 + jωinτpreamp ∣∣∣∣ (24) θ1 =∠ ( 1 1 + jωinτpreamp ) (25) K2 and θ2, which were given by (11) and (12), respectively, are transformed to K2 = ∣∣∣∣ 11 + jωinτtrack ∣∣∣∣ (26) θ2 =∠ ( 1 1 + jωinτtrack ) (27)
- OKURA et al.: FREQUENCY MODEL OF A CONTINUOUSLY DRIVEN CLOCKED CMOS COMPARATOR 959 Fig. 5. Plot of (30) when τpreamp = τreg. and K3 and θ3, which were given by (18) and (19), respectively, are transformed to K3 = ∣∣∣∣ 11− jωinτreg ∣∣∣∣ (28) θ3 =∠ ( 1 1− jωinτreg ) . (29) Therefore, H(s) is derived as H(s) = 1 1 + sτpreamp︸ ︷︷ ︸ Hpreamp(s) · 1 1 + sτtrack︸ ︷︷ ︸ Htrack(s) · 1 1− sτreg︸ ︷︷ ︸ Hreg(s) . (30) The three poles of (30) correspond to the poles of the preamp, the latch in the tracking mode, and the latch in the regeneration mode, respectively. The plot of (30) is shown in Fig. 5, setting τpreamp = τreg as a simple example. It is noted that τtrack is given by τreg/(γ − 1) according to (7) and (15). The bandwidth is saturated when γ > 2, because the bandwidth is limited by Hpreamp(s) and Hreg(s), whose poles are located at a lower frequency than that of Htrack(s). On the other hand, the phase delay at −3 dB decreases as γ increases because the sum of angles of Hpreamp(s) and Hreg(s) is zero and the phase shift depends only on Htrack(s). The phase shift in the comparator is kept small by setting γ high at τpreamp = τreg. This is useful for the design of the comparator, which is sensitive to the phase shift. IV. SIMULATION RESULT Fig. 6 presents an example schematic of a fully differen- tial high-speed comparator. Two differential input pairs (Mn1, Mn2, and Mn3, Mn4) sense the difference between the analog inputs and the reference voltages and generate currents accord- ingly. These currents are mirrored from load transistors (Mp1 and Mp2) to p-channel transistors (Mp3 and Mp4). A pair of p-channel cascode transistors (Mp5 and Mp6) biased at a given voltage Vb prevents kickback noise [16]. The latch utilizes an Fig. 6. Example of a fully differential high-speed CMOS comparator. TABLE I VARIABLES OF THE COMPARATOR DERIVED FROM SPICE SIMULATIONS Fig. 7. Comparison of zero-to-one transition points between the model and simulated results for input signal frequencies. n-channel cross-coupled transistor (Mn6 and Mn7) and a CMOS switch (Mn5 and Mp7). Table I shows the variables of the comparator derived from SPICE simulations with the transistor-level netlist and a 0.18-μm CMOS process model. It is noted that the two poles of the preamp, namely, the mirror pole and the cascode pole, are represented with τpreamp1 and τpreamp2. The simulated zero-to-one transition points versus the input signal frequency ωin are shown in Fig. 7, where the input signal phase is 0 rad. The dots are found with transient simulation runs. The solid line shows the presented frequency model of the comparator, where the variables listed in Table I are used. The solid line is close to the dots, so that the presented model pre- cisely estimates the zero-to-one transition points. The dashed line shows the conventional model only with the frequency response of the preamp. The conventional model significantly deviates from the dots because the effect of the input signal during regeneration is large at high frequencies.
- 960 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 57, NO. 12, DECEMBER 2010 Fig. 8. Comparison of zero-to-one transition points between the model and simulated results for input signal phases at ωin = 3.14 Grad/s. Fig. 8 shows the zero-to-one transition points versus the input signal phase θin at ωin = 3.14 Grad/s. The ac error voltage is not negligible at this high frequency. The presented model of the solid line shows good fitting to the simulation results of dots and provides fine estimation of the zero-to-one transition points at each input signal phase, whereas the mismatch between the conventional model and simulation results is considerably large. These simulation results demonstrate the accuracy of the presented model. V. CONCLUSION A novel frequency model of a continuously driven clocked comparator has been presented. The model includes a transfer function to represent the effect of the input signal during regen- eration, which was derived from the small-signal analysis on the transition from the tracking mode to the regeneration mode. The transfer function is given by the product of three poles, which are functions of basic design parameters. In addition, a comparison between the model estimation and simulated results for an example design has been described. The presented model shows a precise estimation of a comparator response with high-frequency input signals. Therefore, the presented model provides a valuable insight for high-speed comparator design. APPENDIX The time response of the comparator during the regeneration mode is given by (16) with coefficients ac1, ac2, and ac3. These coefficients are respectively given by ac1 = −ApreampK1 gmin gml sin(θin + θ1) + j cos(θin + θ1) 2(1 + jωinτreg) , (31) ac2 =Apreamp gmin gml vref + Vos,static (32) ac3 =Apreamp gmin gml (K1K3 sin(θin + θ1 + θ3)− vref) − Vos,static + vout(0−). (33) The first two terms in the right side of (16) is represented with (31) as ac1 · exp(−jωint) + ac1 · exp(jωint) = −ApreampK1K3 gmin gml sin(ωint + θin + θ1 + θ3). (34) ACKNOWLEDGMENT S. Okura would like to thank T. Wakimoto and C. W. Mangelsdorf of Analog Devices, K. K., for making possible the internship during which this work was performed and for their assistance, support, and encouragement throughout this project. This work was partly carried out at the Frontier Research Base for Global Young Researchers, Osaka University, on the Pro- gram of Promotion of Environmental Improvement to Enhance Young Researchers’ Independence, the Special Coordination Funds for Promoting Science and Technology, Japan Ministry of Education, Culture, Sports, Science and Technology. REFERENCES [1] H. J. M. Veendric, “The behavior of flip-flops used as synchronizers and prediction of their failure rate,” IEEE J. Solid-State Circuits, vol. SSC-15, no. 2, pp. 169–176, Apr. 1980. [2] R. Sarpeshkar, J. L. Wyatt, Jr., N. C. Lu, and P. D. Gerber, “Mismatch sensitivity of a simultaneously latched CMOS sense amplifier,” IEEE J. Solid-State Circuits, vol. 26, no. 10, pp. 1413–1422, Oct. 1991. [3] W. A. M. Van Noije, W. T. Liu, and S. J. Navarro, Jr., “Precise final state determination of mismatched CMOS latches,” IEEE J. Solid-State Circuits, vol. 30, no. 5, pp. 607–611, May 1995. [4] G. M. Yin, F. O. Eynde, and W. Sansen, “A high-speed CMOS comparator with 8-b resolution,” IEEE J. Solid-State Circuits, vol. 27, no. 2, pp. 208– 211, Feb. 1992. [5] B. Zojer, R. Petschacher, and W. A. Luschnig, “A 6-bit/200-MHz full Nyquist A/D converter,” IEEE J. Solid-State Circuits, vol. SSC-20, no. 3, pp. 780–786, Jun. 1985. [6] B. Peetz, B. D. Hamilton, and J. Kang, “An 8-bit 250 Megasample per second analog-to-digital converter: Operation without a sample and hold,” IEEE J. Solid-State Circuits, vol. SSC-21, no. 6, pp. 997–1002, Dec. 1986. [7] R. J. V. D. Plassche and P. Baltus, “An 8-bit 100-MHz full-Nyquist analog- to-digital converter,” IEEE J. Solid-State Circuits, vol. 23, no. 6, pp. 1334– 1344, Dec. 1988. [8] C. W. Mangelsdorf, “A 400-MHz input Flash converter with error correction,” IEEE J. Solid-State Circuits, vol. 25, no. 1, pp. 184–191, Feb. 1990. [9] I. Mehr and D. Dalton, “A 500-MSample/s, 6-bit Nyquist-rate ADC for disk-drive read-channel applications,” IEEE J. Solid-State Circuits, vol. 34, no. 7, pp. 912–920, Jul. 1999. [10] M. Choi and A. A. Abidi, “A 6-b 1.3-Gsample/s A/D converter in 0.35-μm CMOS,” IEEE J. Solid-State Circuits, vol. 36, no. 12, pp. 1847– 1858, Dec. 2001. [11] K. Uyttenhove, J. Vandenbussche, E. Lauwers, G. Gielen, and M. Steyaert, “Design techniques and implementation of an 8-bit 200-MS/s interpolating/averaging CMOS A/D converter,” IEEE J. Solid-State Circuits, vol. 38, no. 3, pp. 483–494, Mar. 2003. [12] B. Razavi and B. A. Wooley, “Design techniques for high-speed, high- resolution comparators,” IEEE J. Solid-State Circuits, vol. 27, no. 12, pp. 1916–1926, Dec. 1992. [13] S. Okura, T. Ohkura, K. Taniguchi, and H. Shibata, “Frequency response analysis of latch utilized in high-speed comparator,” in Proc. IEEE Int. Conf. Electron., Circuits, Syst., 2006, pp. 1077–1080. [14] R. J. Baker, H. W. Li, and D. E. Boyce, CMOS Circuit Design, Layout, and Simulation. New York: Wiley-IEEE Press, 1997. [15] R. J. Baker, CMOS, Mixed-Signal Circuit Design. New York: Wiley- IEEE Press, 2002. [16] K. Sushihara, H. Kimura, Y. Okamoto, K. Nishimura, and A. Matsuzawa, “A 6b 800 MSample/s CMOS A/D converter,” in Proc. Int. Solid State Circuits Conf., Feb. 2000, pp. 428–429. /ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /DocumentCMYK /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ] >> setdistillerparams > setpagedevice