Comparator Design Ken Martin Dept. of Elec. and Comp. Eng. University of Toronto Toronto, Canada M5S 1A4 martin@eecg.toronto.edu (416) 978-6695 © K.W. Martin, 1997 1 Comparators Using an Op-Amp for a Comparator • Can use an uncompensated op-amp for a comparator. • This comparator has a resolution limited to the input-offset-voltage of the op-amp on the order of 2mV to 5mV for typical MOS processes. • In many applications, this would be inadequate. V in Vout • An alternative architecture which can resolve signals with accuracies much less than the input-offset-voltages of op-amps is shown below. φ1' φ2 Sr C V in φ1 Vout Bottom Plate • During φ 1 the bottom plate of the capacitor C (i.e. the left side of capacitor C) is connected to ground and the top plate is connected to the inverting input of the op-amp. At the same time, the output of the op-amp is also connected to the inverting input of the op-amp by closing switch S r . • During the comparison phase, the reset switch, S r , is turned off, and the top plate of the capacitor is connected to the input voltage. If the input signal is greater than zero, the output of the op-amp will saturate negative; if the input signal is less than zero, the output of the op-amp will saturate positive. These two cases are easily resolved and stored using a simple digital latch. © K.W. Martin, 1997 2 • The limitations of this approach become apparent when one considers non-ideal op-amps, having finite gains, and needing compensation in order to be stable during the reset phase. • For example, consider the case where a 0.5mV signal must be resolved. After comparison, the output of the op-amp should have a 5V difference between the cases when the input signal is -0.25mV and +0.25mV. This means the gain of the op-amp must be at least 10,000. Thus, the open-loop time constant of the comparator will be on the order of 10 4 ⁄ ω ta . This limits the clock frequency to about 500kHz which is normally unacceptably slow. • During the comparison time, the compensation capacitor should be disconnected as shown below. Vin Vin Q1 φ1 Cc Vout • Transistor Q1 is used to achieve lead compensation when it is on during the reset phase, φ 1 . During the comparison phase, Q1 is turned off which disconnects the compensation capacitor C c thereby greatly speeding up the op-amp during this phase. © K.W. Martin, 1997 3 Multi-Stage Versus Single Stage Approach Assume that a gain stage has a unity-gain frequency ω t given by W µ n C ox -----V eff g 1 1 µ n V eff L 1 m ω t-i = --- --------- = --- --------------------------------- = --- ---------------3 2 2 L2 3 C gs --- WLC ox 3 (1) Note that g m ⁄ C gs is the approximate unity-gain frequency of a transistor. This approximation is optimistic for single stages having very large gain. Further assume each stage is approximated by a first order transfer function A i-0 A i(s) = ------------------------------1 + s ⁄ ω -3dB (2) From (2) and using A i(ω t) = 1 , we have ω t-i ω -3dB ≅ --------A i-0 (3) The settling time constant of a single stage is 1 ⁄ ω -3dB . A i-0 L 2 τ i = 2 ---------------µ n V eff (4) Assuming we need 3τ i for settling time in half a period T ⁄ 2 , we have ω t-i 1 µ n V eff 1 1 F max = --- = ------- = ------------- = ------ ---------------12 A L 2 6A i-0 T 6τ i i-0 (5) For a single stage, assume we have A 0 = 10 4 , L = 0.5µm , V eff = 0.25V , µ n = 0.05M 2 ⁄ V ⋅ S . This gives F max = 417kHz . For a cascade of three stages, A i-0 = 3 A 0 = 3 10 4 = 21.5 , but we now need three times the settling time as for a single stage and we get F max = 1 ⁄ 18τ i = 65MHz . Thus, the cascade of three lower-gain comparators is 155 times faster than a single high-gain stage. © K.W. Martin, 1997 4 Clock-Feedthrough Errors • Perhaps the major limitation on the resolution of comparators is due to what is called clock-feedthrough. • This error is due to transistor channel-charge flow when transistors turn off. • When they turn off, there are two mechanisms whereby charge errors occur. The first is due to the channel charge which must flow out from the channel region of the transistor to the drain and the source junctions. The channel charge of a transistor having zero V DS is given by Q CH = W 3 L 3 C V eff = W 3 L 3 C ox ( V GS – V t ) (6) ox • Another, usually somewhat smaller charge, unless V eff is very small, is due to the overlap capacitance between the gate and the junctions. • Shown below is the comparator using n-channel switches and parasitic capacitors to model the overall charge efffects when transistors turn off. φ'1 φ2 C3a C 1a C3b C1b Q1 Vin V' C V'' Q3 Vout φ1 C2a Q2 C2b The comparator of shown. with n-channel switches and the parasitic capacitances • Consider first when Q 3 turns off. If the clock waveform is very fast, the channel charge due to Q 3 will flow equally out through both junctions [Shieh/87]. • The charge that goes to the output node of the op-amp will have very little effect other than causing a temporary glitch. © K.W. Martin, 1997 5 • However, the charge that goes to the inverting-input node of the op-amp will cause the voltage across C to change which introduces an error. Since this charge is negative, for an n-channel transistor, the node voltage V″ will go negative. The voltage change due to the channel charge is given by Q ch V eff3 C ox W 3 L 3 ( V DD – V tn )C ox W 3 L 3 ∆V″ = ---------- = – --------------------------------------- = – --------------------------------------------------------(7) 2C 2C 2C since the effective gate-source voltage of Q3 is given by V eff3 = V GS3 – V tn = V DD – V tn . The above voltage change in V″ is based on the assumption that Q 2 turns off slightly after Q 3 . More will be said about this assumption shortly. • To calculate the change in voltage due to the overlap capacitance, we have – ( V DD – V SS ) C ov ∆V″ = -----------------------------------------------C ov + C (8) • The clock feedthrough of transistors Q 1 and Q 2 may cause temporary glitches but will have much less effect than the clock feedthrough of Q 3 , if it is assumed that Q 2 turns off slightly after Q 3 [Haigh/83]. (This time difference is the reason why the clock voltage of Q 2 is denoted φ 1 , whereas the clock voltage of Q 3 is denoted φ 1 ′ ). • The argument for this is slightly complicated but goes as follows. When Q 2 turns off, its clock feedthrough will cause a negative glitch at V′ , but this will not cause any change in the charge stored in C since the right side of C is connected to an effective open circuit assuming Q 3 has already turned off. Later, when Q 1 turns on, the voltage V′ will settle to V in irrespective of the previous clock feedthrough of Q 2 . Thus, the voltage at the left of C is unaffected by the clockfeedthrough of Q 2 , and the voltage across C is also unaffected by the clockfeedthrough of Q 2 . Therefore, the voltage V″ is unaffected by the clock feedthrough of Q 2 . This would not be the case if Q 2 was turned off at the same time or before turning off Q 3 . The clock-feedthrough of Q 1 has no effect due to a simpler argument, the comparison has already been made when it turns off. In addition, its clock-feedthrough has no effect when it turns on because the right side of C is connected to an open circuit at that time, assuming non-overlapping clocks. © K.W. Martin, 1997 6 Input-Offset-Voltage Cancellation • Another source of error is due to the input-offset voltage of the op-amp. • In switched-capacitor comparators, such as that shown above, it is not a problem as it is stored onto the capacitor during the reset phase as shown below. During φ1: φ1' C φ1 Sr Voff Vout Voff Voff During φ2: φ2 V in + Voff φ1' Sr C V in Vout Voff Voff • • Not only does this technique eliminate input-offset-voltage errors, but it also minimizes errors caused by low-frequency 1/f noise which can be quite large in CMOS IC’s. © K.W. Martin, 1997 7 Minimizing Errors Due to Clock Feed-through • The simplest way to reduce errors due to clock feedthrough is to use larger capacitors. For our previous example, capacitors on the order of 100 pF, or so, would guarantee that the clock-feedthrough errors were less than 0.5 mV. • An alternative approach for minimizing errors due to clock feedthrough is to use fully-differential design techniques for comparators as shown below. φ2 φ1' C Q3a φ1 Vin Vout φ2 φ1 Q3b C φ1' • Ideally, when the comparator is taken out of reset mode, the clock-feedthrough of reset switch Q 3a will match the clock-feedthrough of Q 3b . The only errors now will be due to mismatches in the clock-feedthrough of the two switches which will be at least ten times smaller, typically, than in the single-ended case. Normally, comparators are always designed as fully differential, although we will often show them as single-ended to simplify schematics. • A third alternative, that can be used along with fully-differential design techniques, is to realize a multi-stage comparator where the clock-feedthrough of the first stage is stored on coupling capacitors between the first and second stage, thereby eliminating its effect. Also, the clock feed-through of the second stage can be stored on the coupling capacitors between the second and the third stages, etc. A single-ended version is shown below. • Consider the time when φ1' goes low and the switch of the first stage has charge injection. When φ1' goes low, Q 1 will inject charge into both the inverting input and the output of the first stage. The charge injected at the first stage output will only cause a temporary glitch there. The charge injected at the inverting input © K.W. Martin, 1997 8 φ 1' Vin φ2 C1 φ1 φ1'' C2 OA1 φ1''' C3 OA2 Vout OA3 φ1 φ1' φ1'' φ1''' φ2 will cause this node to go negative. The amount this node voltage will be in the range of tens of millivolts or so. After the inverting input goes negative, the output of the first stage will go positive by an amount equal to the negative transition of the inverting input multiplied by the first stage’s gain. However, at this time φ1'' is still high. Therefore, the second-stage is still being reset. Therefore, C 2 is charged up to whatever the error caused by the clockfeedthrough of the first stage is. This eliminates its effect. In a similar manner, when φ1'' turns off and the second-stage goes from closed-loop reset mode to open-loop comparison mode, the third stage is still in reset mode and the clockfeedthrough of the second stage is stored on coupling capacitor C 3 . Finally, when the third stage turns off, its clock feedthrough is not cancelled. However, the error it causes in resolving an input voltage to all three stages is equal to the voltage transition at the inverting input of the third stage divided by the negative of the gains of the first two stages. (i.e. This is the input voltage needed to cancel the effect of the clock-feedthrough of the third stage coming out of reset mode.) © K.W. Martin, 1997 9 Latched Comparators • Modern comparators will have 1-3 gain stages followed by a track-and-latch stage. The track-and-latch stage operates as an amplifier during φ 1 and then is reconfigured into a positive-feedback latch during φ 2 . Latch-Mode Time-Constant • Consider two inverting amplifiers connected in a positive-feedback loop. Vy Vx • This can be modelled by small-signal model where each inverter has a transconductance given by G m = A V ⁄ R L where A V is the inverter gain. AV -------V y RL Vy CL RL RL CL Vx AV -------V x RL • We have AV dV x V x -------V y = – C L ---------- – ------- RL dt R L (9) AV dV y V y -------V x = – C L ---------- – ------- RL dt R L (10) and Multiplying (9) and (10) by R L and rearranging gives © K.W. Martin, 1997 10 dV x τ ---------- + V x = – A V V y dt (11) dV y τ ---------- + V y = – A V V x dt (12) and where τ = R L C L is the time-constant at the output node of each invertor. Subtracting (12) from (11) and rearranging terms gives τ d∆V --------------- = ∆V A – 1- ---------dt (13) V where ∆V = V x – V y is the voltage difference between the output voltages of the invertors. Equation (13) is a first-order differential equation with no forcing function. Its solution is given by ∆V = ∆V 0 e AV – 1 --------------t τ - (14) where ∆V 0 is the initial voltage difference at the beginning of the latch phase. Thus, the voltage difference increases exponentially in time with a time-constant given by RL CL CL τ τ ltch = ---------------- ≅ --------------- = --------AV – 1 AV Gm (15) where again G m = A V ⁄ R L is the transconductance of each invertor. Note that τ ltch is roughly equal to the inverse of the unity-gain frequency of each invertor. • In the case of MOS devices, normally the output load will be proportional to the gate-source capacitance of a single transistor, or specifically, C L = K 1 WLC ox © K.W. Martin, 1997 11 (16) Here, K 1 is a proportionality constant which might be between 1 and 2. Also, the invertor transconductance will be proportional to the transconductance of a single transistor given by W G m = K 2 g m = K 2 µ n C ox -----V eff L (17) where K 2 might be between 0.5 and 1. Substituting (16) and (17) into (15) gives K1 L 2 L2 τ ltch = ------ ----------------- = K 3 ----------------K 2 µ n V eff µ n V eff (18) where K 3 might be between 2 and 4. Note that (18) implies that τ ltch is dependant primarily on the technology and is not very dependant on the design (assuming a reasonable design is used that maximizes V eff and minimizes C L ). Note also the similarity between (18) and (4), the equation for the time-constant of a cascade of gain stages. For a given technology, (18) is very useful in determining a rough estimate for the maximum clock frequency of a latch-andtrack comparator. • If it is necessary for a voltage difference of ∆V logic to be obtained for succeeding logic circuitry to safely recognize the correct output value, then through the use (14), the time necessary for this to happen is given by C L ∆V logic ∆V logic L2 T ltch = --------- ln ------------------ = K 3 ----------------- ln ------------------ G m ∆V 0 µ n V eff ∆V 0 This analysis is the basis for understanding Comparator Metastability. © K.W. Martin, 1997 12 (19) Examples of Modern High-Speed CMOS Comparators • A modern high-speed CMOS comparator will typically have one or two stages of pre-amplification followed by a track-and-latch stage, as is shown below [Yukawa, 85]. Q5a Q4a Q4b Q5b Vltch Vltch Vout+ V2a Vin+ Vltch Vin- V1a V2b Q3a Q3b Vltch Vout- V1b Q1b Q1a Preamplifier Stage Q2aQ2b Track-and-Latch Stage • The preamplifier (or preamplifiers) are required to obtain higher resolution and to minimize the effects of ‘flash-back’. The output of the preamplifier, although larger than the comparator input, is still much smaller than the voltage levels needed to drive digital circuitry. The track-and-latch stage then amplifies this signal further during the track-phase, and then also during the latch-phase when the positive feedback is enabled which regenerates the analog signal into a fullscale digital signal. The use of the track-and-latch stage minimizes the total number of gain stages required, even when good resolution is needed, and, thus, is faster than the multi-stage approach just described. • Flash-back denotes the charge transfer either into or out of the inputs when the track-and-latch stage goes from track-mode to latch-mode. It is caused by the charge needed to turn the transistors in the positive feedback circuitry on, and by the charge that needs to be removed in order to turn transistors in the tracking circuitry off. Without having a preamplifier or buffer, this flash-back will go into the driving circuitry causing very large glitches, especially in the case when the impedance seen looking into the two inputs is not perfectly matched. © K.W. Martin, 1997 13 • In the above design, the pre-amp is a simple p-channel input differential amplifier. During reset mode, its outputs are shorted together. At the same time, Q 3a and Q 3b are off. This causes nodes V 1a and V 1b to be reset low since Q 1a and Q 1b are on. At the same time, Q 5a and Q 5b are on causing V 2a and V 2b to be reset high. Next, during the sample and latch mode, the two crosscoupled latches are connected together by Q 3a and Q 3b and nodes V 2a and V 2b will dicscharge negatively until the cross-coupled pair of Q 4a and Q 4b is activated. When this happens, the positive feedback of this cross-coupled pair and of Q 2a and Q 2b quickly amplifies the small differential signal to full logic levels. • In high-resolution applications, capacitive coupling and reset switches will also typically be included to eliminate any input-offset-voltage and clock-feedthrough errors, in a manner similar to what was described previously. • Another example of the above architecture is shown below [Song/90]. This Vout Vin Trk Preamp Gain Stage Positive Feedback A two-stage comparator having a preamplifier and a positive-feedback track-andlatch stage. comparator actually has the positive feedback of the second stage always enabled. In track mode, when the two diode-connected transistors of the gain-stage are © K.W. Martin, 1997 14 enabled, the gain around the positive-feedback loop is less than one and the circuit is stable. The combination of the diode-connected transistors of the gain stage and the transistors of the positive-feedback loop act as a moderately large impedance giving gain from the preamplifier stage to the track-and-latch stage. The diode-connected loads of the preamplifier stage give a limited amount of gain in order to maximize speed, while still buffering the flash-back from the input circuitry. • A third example is shown below. . [Norsworthy/89]. This design also uses diode- Latch Vout Digital SignalLevel Restoration Vin Latch Preamp Positive Feedback A two-stage comparator from [Norsworthy/89]. connected loads in order to keep all nodes relatively low impedance (similar to current-mode circuit-design techniques), thereby keeping all node time constants small and giving fast operation. This design also uses pre-charging to eliminate any memory from the previous decision. For example, the positive-feedback stage © K.W. Martin, 1997 15 is pre-charged low, whereas the digital-restoration stage is pre-charged high somewhat similar to what is done in Domino CMOS logic. • A fourth example was designed by one of the authors [Martin/84] and is shown in . The appropriate clock-waveforms for this design are shown in . This design First-Stage Common-Mode Feedback φ1 ′ φ2 Vin C1A φ1 ′ φ1 φ2 C1B First SC Gain Stage Second Gain Stage C2A C2A φ1 ″ Positive Feedback Stage φ1 ″ Trk Trk A two-stage comparator with capacitive coupling to eliminate input-offset voltage and clock-feedthrough errors, along with positive feedback for fast operation. © K.W. Martin, 1997 16 φ1 φ1 ′ φ1 ″ Trk The clock waveforms required by the comparator of . eliminates any input offset voltages form both the first and second stages by using capacitive coupling. It also has common-mode feedback circuitry for the first preamplifier stage which allows for input signals having large common-mode signals. Unlike in fully-differential op-amps, the linearity of the common-mode feedback circuitry in fully-differential comparators is non-critical as whenever large signals are present (and the common-mode feedback circuitry becomes nonlinear) there is no ambiguity in resolving the sign of the input signal. Measured (but unpublished) performance of this circuit resulted in a 0.1mV resolution at a 2MHz clock frequency, despite a very old 5µm technology. The performance measured was limited by the test set-up rather than the circuitry. • A fifth example, realized in a BiCMOS technology, that exhibited very good performance, was described in [Razavi/92]. Indeed, at the time of publication, this comparator is one of the better BiCMOS designs to date. It is based on the realization that it not necessary to reset the first stage and the input-offset errors of the first stage can still be eliminated by resetting the right side of the coupling capacitors between the first stage and the second stage, based on the assumption that the gain of the first stage is not too large [Poujois, 1978] [Vittoz, 1985]. A simplified schematic of this comparator is shown in . During reset phase, the inputs to the preamplifier are connected directly to ground (or a reference voltage), while the outputs of the coupling capacitors are connected to ground as well. This stores any offset voltages of the first stage on the capacitors. When the comparator is taken out of reset phase, then the effect of the clock-feedthrough of S 5 and S 6 on the input resolution is divided by the gain of the first stage. In the realization described in [Razavi/92], the first stage was a BiCMOS preamplifier consisting of MOS source followers followed by a bipolar differential amplifier and emitter-follower output buffers as is shown in . Notice that the circuit operates between ground and a negative voltage supply. Note also that the switches are realized using p-channel transistors. The track-and-latch stage of © K.W. Martin, 1997 17 S1 Vin S3 φ1 S5 φ1 PositiveFeedback Latch φ1 S2 φ1 S4 φ1 Vout S6 φ1 A simplified schematic of the comparator described in [Poujois, 1978] [Razavi/92]. φ1 Vout- Vout+ φ1 Vin+ Vinφ1 φ1 The BiCMOS preamplifier of [Razavi/92]. [Razavi/92] has both a bipolar latch and a CMOS latch as is shown in . During the reset phase, φ 1 is low and φ 2 is high, X 1 and Y 1 are connected to ground through switches M 1 and M 2, C 3 is discharged to the minus supply through M 11, and X 2 and Y 2 are discharged to the minus supply through M 9 and M 10. Also, at this time, M 3 and M 4 are off. Next, φ 1 goes high which leaves X 1 and Y 1 floating and connects the inputs of the preamplifier to its input signal. After a short delay, which is needed for the transient response of the preamplifier, φ 2 goes low, which turns on M 12 thereby activating the positive feedback action of Q 5 and Q 6. This develops a differential voltage between X 1 and Y 1 of about 200mv, since C 3 is about 1/5 the size of C 1 and C 2. The offset of this bipolar latch is quite small, typically on the order of one millivolt or less. Also, note that, when activated, the © K.W. Martin, 1997 18 φ1 X1 M1 M2 φ1 Y1 Vin+ VinC1 C2 Q5 M12 Q6 M3 φ2 C3 M4 M5 M11 M6 X2 Y2 Vout+ Voutφ2 φ2 M9 M7 M8 M10 The BiCMOS track-and-latch stage of [Razavi/92]. only power dissipated is that required to charge C 3; there is no d.c. power dissipation. A short time after the bipolar latch is activated, the inverter connected to C 3 will change state and its output will go high. This turns on M 3 and M 4 which activates the CMOS part of the latch consisting of cross coupled M 5 and M 6 and cross-coupled M 7 and M 8. The reason for including cross-coupled M 5 and M 6 is to prevent nodes X 1 and Y 1 from being discharged very far from ground towards the minus power supply while at the same time amplifying the 200mV difference signal developed by the bipolar latch to almost a full-level CMOS voltage change. Thus, the latch is accurate, relatively fast, and low power. Also described in [Razavi/92] is an interesting CMOS only comparator which will not be described here, but which interested readers might want to investigate. © K.W. Martin, 1997 19 References • W. Colleran and A. Abidi, “A 10-b, 75-MHz Two-Stage Pipelined Bipolar A/D Converter,” IEEE J. of Solid-State Circuits, Vol. 28, No. 12, pp. 1187-1199, Dec. 1993. • J. Fernandes, S. Miller, M. Mallison, and G. Miller, “A 14-bit 10-µs Subranging A/D Converter with S/H,”IEEE J. of Solid-State Circuits, vol. 24, No. 6, pp 1485-1491, Dec. 1989. • D.G. Haigh and B. Singh, “A Switching Scheme for Switched-Capacitor Filters which Reduces Effect of Parasitic Capacitances Associated with with Control Terminals,” Proc. IEEE Int. Symp. on Circuits and Systems, vol. 2, pp. 586-589, Jun. 1983. • R.H. Krambeck, C. Lee, and H.S. Law, “High-Speed Compact Circuits with CMOS,”IEEE J. of Solid-State Circuits, Vol. 17, No. 3, pp. 614-619, Jun. 1982. • K. Martin, “Improved Circuits for the Realization of Switched-Capacitor Filters,” IEEE Trans. Circuits and Systems, Vol. CAS-27, No. 4, pp. 237-244, April 1980. • K. Martin, L. Ozcolak, Y.S. Lee and G.C. Temes, “A Differential Switched-Capacitor Amplifier,” IEEE J. SolidState Circuits, SC-22, No. 1, pp. 104-106, Feb. 1987. • K. Martin, “New Clock-Feedthrough Cancellation Technique for Analog MOS Switched-Capacitor,” Elctron. Lett., no. 18, 1992. • A. Matsuzawa, et. al., “A 10b 30MHz Two-Step Parallel BiCMOS ADC with Internal S/H,” IEEE Intern. SolidState Circuits Conf., pp. 162-163, Feb. 1990. • J. McCreary and P.R. Gray, “All-MOS Charge Redistribution Analog-to-Digital Conversion Techniques-Part 1, IEEE J. of Solid-State Circuits, vol. SC-10, pp. 371-379, Dec. 1975. • G. Miller, M. Timko, H.S. Lee, E. Nestler, M. Mueck, and P. Ferguson, “An 18b 10µs Self-Calibrating ADC,” IEEE Intern. Solid-State Circuits Conf., pp. 168-169, Feb. 1990. • S. Norsworthy, I. Post, and S. Fetterman, “A 14-bit 80-KHz Sigma-Delta A/D Converter: Modeling, Design, and Performance Evaluation, IEEE J. of Solid-State Circuits, vol. 24, No. 2, pp. 256-267, Apr. 1989. • R. Poujois and J. Borel, “A Low Drift Fully Integrated MOSFET Operational Amplifier”, IEEE J. of Solid-State Circuits, Vol. SC-13, pp. 499-503, Aug. 1978. • B. Razavi and B. Wooley, “Design Techniques for High-Speed, High-Resolution Comparators,” IEEE J. of SolidState Circuits, vol. 27, No. 12, pp 1916-1926, Dec. 1992. • D. Robertson, P. Real, and C. Mangelsdorf, “A Wideband 10-bit 20Msps Pipelined ADC using Current-Mode Signals,” IEEE Intern. Solid-State Circuits Conf., pp. 160-161, Feb. 1990. • J. Shieh, M. Patil, and B.L. Sheu, “Measurement and Analysis of Charge Injection in MOS Analog Switches,” IEEE J. of Solid-State Circuits, vol. 22, No. 2. pp 277-281, Apr. 87. • K. Sone, Y. Nishida, and N. Nakadai, “A 10-b 100-Msample/s Pipelined Subranging BiCMOS ADC,” IEEE J. of Solid-State Circuits, Vol. 28, No. 12, pp. 1180-1186, Dec. 1993. • B. Song, H.S. Lee, and M. Tompsett, IEEE J. of Solid-State Circuits, vol. 25, No. 6. pp. 1328-1338, Dec. 1990. • E. J. Swanson, “Method and Circuitry of Decreasing the Recovery Time of an MOS Differential Voltage Comparator, U.S. patent 5247210, Sept. 1993. • K. Tan, et. al., “Error-Correction Techniques for High-Performance Differential A/D COnverters,” IEEE J. of Solid-State Circuits, Vol. 25, No. 6, pp. 1318-1327, Dec. 1990. • T.L. Tewksbury, H.S. Lee, and G. Miller, “The Effects of Oxide Traps on the Large-Signal Transient Response of Analog MOS Circuits,” IEEE J. of Solid-State Circuits, Vol. 24, No. 2, pp. 542-543, April 1989. © K.W. Martin, 1997 20 • E.A. Vittoz, “Dynamic Analog Techniques,” in Design of MOS VLSI Circuits for Telecommunications, ed. Y. Tsividis and P. Antognetti, Prentice Hall, 1985. • P. Voerenkamp and J. Verdaasdonk, “A 10b 50MS/s Pipelined ADC,” IEEE Intern. Solid-State Circuits Conf., pp. 32-33, Feb. 1992. • R. Van de Plassce and P. Baltus, “An 8-bit 100-MHz Full-Nyquist Analog-to-Digital Converter, IEEE J. of SolidState Circuits, Vol. 23, No. 6, pp. 1334-1344. • T. Wakimoto, Y. Akazawa, and S. Konaka, “Si Bipolar 2-GHz 6-bit Flash A/D Conversion LSI,” IEEE J. of SolidState Circuits, Vol. 23, No. 6, pp. 1345-1350, Dec. 1988. • A. Yukawa, “A CMOS 8-Bit High-Speed A/D Converter IC,” IEEE J. of Solid-State Circuits, Sc-20, No. 3, pp. 775-779, June 1985. © K.W. Martin, 1997 21