A 15-b - Msample/s Digitally Self-Calibrated Pipeline ADC: Hae-Seung Lee, and
A 15-b - Msample/s Digitally Self-Calibrated Pipeline ADC: Hae-Seung Lee, and
A 15-b - Msample/s Digitally Self-Calibrated Pipeline ADC: Hae-Seung Lee, and
1207
I. INTRODUCTION IPELINE analog-to-digital converters (ADCs) present advantages compared to flash or successive approximation techniques because potentially high resolution and high speed can be achieved at the same time. A 1-b-per-stage design is particularly amenable because each stage is very simple and fast. The primary limitations on the accuracy of a switchedcapacitor pipeline ADC are capacitor mismatch, charge injection, finite operational amplifier gain and comparator offset. Previous 1-b-per-stage ADCs, including algorithmic and pipeline ADCs, removed some of these errors by using extra clock cycles with ratio independent [ 11, reference refreshing [2], error averaging [3] and analog calibration [4] techniques. Although the analog calibration does not require extra clock cycles during normal conversions, a weighted capacitor array is needed for each stage to be calibrated. For pipeline ADCs, where many stages are calibrated, the added complexity and capacitive load is significant. This paper presents a digital self-calibrationtechnique based on a radix 1.93 and one comparator per stage conversion algorithm. A nonradix two conversion algorithm was previously employed in a successive approximation converter [ 5 ] . That technique required a precise extemal calibration source and hence could be factory calibrated only. The technique described here is self-calibrating, simple, and tolerant of comparator errors. The digital calibration presented here may be applied to pipeline or cyclic ADC architectures. A l-bper-stage or multi-b-per-stage design may be employed with either ADC architecture.
Manuscript received May 17, 1993; revised July 23, 1993. This work was supporteded by the Semiconductor Research Corporation under Contract 9 1SP-080, Analog Devices, and General Electric. A. N. Karanicolas and H.-S. Lee are with the Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02 139. K. L. Bacrania is with Hams Semiconductor, Palm Bay. FL 32905. IEEE Log Number 92121 15.
The key advantage of the digital calibration reported in this paper is that the errors at the carry transitions are direcrly measured under the same condition as during the normal conversion. Therefore, this technique offers potentially higher calibration accuracy than other calibration techniques [6]-[8] that measure the error at different conditions than the actual conversion. Another important aspect of this design is that the calibration is performed in the digital domain, so no extra analog circuitry, such as weighted capacitor arrays, is needed and no extra clock cycles are necessary during the conversion [9], [6]. The digital calibration reported here automatically accounts for capacitor mismatch, capacitor nonlinearity contributing to DNL, charge injection, finite op-amp gain, and comparator offsets. The nominal offsets of the op amp and comparator are reduced by standard offset cancellation and subsequently eliminated by digital calibration. Unlike analog calibration, digital calibration does not correct or create analog decision levels. Therefore, the uncalibrated ADC must provide decision levels spaced at no more than 1 LSB at the intended resolution. In 1-b-per-stage pipeline ADCs, missing decision levels result when the input of any of the stages exceeds the full scale due to capacitor mismatches, capacitor nonlinearity, charge injection, finite op-amp gain and comparator offsets. The missing decision levels cannot be removed by digital calibration alone. Missing decision levels can be eliminated, however, by using gain less than 2 and two to three more stages of pipeline, which gives enough redundancy in the analog decision levels. With gain less than 2, missing codes are introduced rather than missing decision levels. The missing codes that result with a gain less than 2 are eliminated by the digital calibration. In contrast to an elaborate calibration reported previously using gain less than 2 [9], the calibration reported here is much simpler and more accurate. No multiplication is needed in the calibration algorithm, and only a small digital memory is needed.
11. PIPELINE ARCHITECTURE
The pipeline architecture used is shown in Fig. 1. A l-b-perstage design is employed. The pipeline begins with a samplehold amplifier (SHA) and is then followed by multiply-by-two (MX2) stages. The prototype ADC consists of an input SHA and 17 MX2 stages that provide decision level redundancy for a 15-b resolution. Each MX2 stage has an analog input and a 1-b digital input as well as an analog output and a 1-b digital output. The gain-of-1 and gain-of-2 sampled-data functions for the SHA and the MX2 stages are, respectively, performed using operational amplifiers in switched-capacitor
1208
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 28, NO. 12, DECEMBER 1993
SHA Stage
MX2 Stage 1
MX2 Stage 2
................................................................................................................................................................
....................................................
V
3
D(2)
WO)
W1)
architecture.
closed-loop configurations. The SHA and MX2 stages each utilize a comparator to determine the respective output bits. Only single-ended systems are drawn for simplicity, though the actual system is fully differential. The radix 2 1-b-per-stage ADC algorithm is next explained. As shown in Fig. 1, the analog input V, of the ADC is first sampled by the SHA. A comparator monitors the SHA output Vout(0), resulting in D ( 0 ) = 1 if Vn is positive or D ( 0 ) = 0 if V;, is negative. The SHA output V,,t(0) is then passed to MX2 stage 1. Thus the input to MX2 stage The 1 is Vn(l) = Vout(0). output VoUt(l) of MX2 stage 1 is 2Vn(l) - Vref if D ( 0 ) = 1 or is 2V,(l) Vref if D ( 0 ) = 0. A comparator monitors VoUt(l) in MX2 stage 1 and results in D ( 1 ) = 1 if Vout(l) is positive; it results in D(1) = 0 if VoUt(l) negative. These data are then passed on to the is remaining MX2 stages in a similar manner. The description of some of the switched-capacitor circuits used to implement this basic radix 2 I-b-per-stage algorithm is found in [3].
Vin
Vin
(-Vref, -Vret)
D=O
D=1
Fig. 2.
vn.
along the Vout axis and kV,,f along the V,, axis. Charge injection offset causes a vertical shift of the residue plot. Near the major carry transition point, the residue exceeds the reference boundary, resulting in missing decision levels. This is because the remaining pipeline section is saturated so that the output code does not change for the corresponding range of analog input. Near the major carry transition point, the residue minimum does not extend to -Vref, resulting in a gap from the minimum to the reference boundary; missing codes result. This is because the full input range of the remaining pipeline section is not accessed. Comparator offset causes a shift of the major carry transition point. This leads to the residue exceeding the reference boundary as well as to a gap in the reference boundary. Again, missing decision levels and missing codes, respectively, result. Finally, capacitor mismatch as indicated causes the residue to exceed the reference boundary near the major carry transition point, resulting in missing decision levels. Capacitor mismatch could also lead to a gap from the residue extrema to the reference bou ndary near the major carry transition point, resulting in missing codes.
1209
Ideal
(+Vref, +Vref)
(-Vref, -Vref)
#+
Comparator Offset Capacitor Mismatch
m vout
Vin
Fig. 3.
Errors in radix 2 1-bit-per-stage ADCs. Missing codes and missing decision levels present.
decision levels occur, which cannot be eliminated by digital calibration alone. The key is to use a nominal gain of less than 2 such that the output of each stage never exceeds kV,,,. For the case of radix < 2, the principal errors affect the residue output in a similar manner as with radix 2. Since the residue is maintained within the reference boundary, no missing decision levels result. The missing codes that result due to the gaps from the residue extrema to the reference boundary are eliminated with digital calibration.
+#
Sl(11) SZ(11)
(+Vref, +Vref)
m vout
s1
vin
D=O
D=l
Fig. 4.
calibration logic system along with calibration constants and S2 determined for stage 11. The two quantities S I and 5 are identified on the residue plot. SIand 5 correspond to 2 2 the quantized representation of Vout, or the quantity X , when I, = 0 with D = 0 and D = 1, respectively. & The digital self-calibration algorithm is now described:
=1
(1) (2)
where D is the bit decision, X is the raw code and Y is the transformed code. This transform ensures that the output code Y with = 0 is the same for D = 0 and D = 1,eliminating missing codes. Note that no multiplication is required in this calibration scheme. Two constants S and S2 are needed per 1 stage. The entire converter uses a 132-b memory to store S - S2 for each stage being calibrated. To determine SI, 1 the analog input is set to zero and the input bit is forced In to 0. The quantity X in this condition is SI. an analogous manner, S2 is determined when the input bit is forced to 1. The calibration constants SI S2 are each obtained by averaging and 2 048 samples without truncation and then truncating the final result.
vn
1210
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 28, NO. 12, DECEMBER 1993
I
1 1
Sl(11) S2(11)
- - ....
Sl(10) S2(10)
5
Fig. 5.
Digital calibration of higher level stages.
to the first MX2 stage. In this way, the entire pipeline ADC is calibrated. Quantization and truncation errors are avoided by averaging the calibration data without truncation. The calibration algorithm described here results in an input referred offset. However, this offset is constant and hence eliminated by digitally subtracting the offset measured with 0 V at the SHA input. z Since the calibration aligns points SI and S using measured values under the same condition as during normal conversion, the Calibration automatically accounts for capacitor mismatch, charge injection and finite op-amp gain. Capacitor nonlinearity causes only integral nonlinearity (INL) error, not differential nonlinearity (DNL) error. It will be shown that comparator errors up to the gain reduction factor (3.5% of full scale) have no effect on the conversion accuracy. The nonunity gain that results from the radix < 2 pipeline can be easily compensated at the ADC digital calibration logic or elsewhere in the system. In the prototype ADC, input voltages corresponding to f0.95Vref give full-scale output codes for 15 b. Note that only digital addition, subtraction, and a small digital storage are needed for the digital calibration algorithm. With a 1 Msample/s conversion rate and 2 048-point averaging, the total calibration time for the converter is approximately 70 ms.
t sl/+l
vOut +Vref
sl
7 s3
-Vref
Fig. 6 . Residue plot of a radix < 2 MX2 stage with comparator offset Lb.; present in the previous stage.
unchanged. This is important because the comparator offset requirement can be relaxed. In the present design, comparator offset up to 1.75% of Kef is corrected by calibration; this is equivalent to a 3.5% residue output referred error. Notice that the comparator offset in question can be static or dynamic. The digital self-calibration will remove this error as long as the comparator offset is low enough so that the residue does not exceed the reference boundary. It must be noted that when capacitor mismatch, charge injection and comparator offset are simultaneously present, the combined output referred error must remain less than 3.5% of Kef for the calibration to be feasible. A finite op-amp gain can be tolerated with the digital selfcalibration. This effect represents a gain error and is not distinct from capacitor mismatch if the gain is finite, but constant. Temperature can cause the gain to drift and this will require recalibration. To avoid this need of recalibration, the pipeline ADC uses a high d-c gain op-amp design.
KARANICOLAS
er
1211
are used to avoid a start-up difficulty in the output common mode level when the power supply is tumed on. If the output common mode level is near the A ~ D power-supply rail, then D A . O p Amp and Comparator Pre-Amp the input differential pair bias current source M I Scould be cut BiCMOS is used to provide n-p-n bipolar devices for high- off and the op-amp output common mode level could remain speed, high-gain analog capability while providing CMOS at the AvDD power-supply rail. However, M25 and M26 tum devices for a switched-capacitorenvironment. The operational on in this condition to drive the output common mode level amplifier is shown in Fig. 7. The op amp is a two-stage, fully away from A ~ D D , desired. Once the output common mode as differential design with a 100 MHz unity-gain bandwidth and a level is stabilized, devices M25 and A426 remain cut off for 125-dB dc gain [lo]. The main signal path is indicated in bold. normal op-amp operation. A PMOS differential input stage, composed of devices M I and The comparator preamplifier is shown in Fig. 8. A twoM2, is used for lower l/f noise and lower threshold voltage stage, open-loop design is used. The main signal path is relaxation, compared to NMOS [ l l ] . Focus will be placed emphasized in bold. A PMOS differential input pair, composed on the right half of the remainder of the op-amp signal path. of devices M1 and M2, is used for reasons analogous to those An n-p-n bipolar second stage Q2 is used to achieve a high- for the op amp. Focus will be placed on the right half of the frequency nondominant pole. The second stage is cascoded remainder of the pre-amp signal path. Emitter follower Q 2 is with device Q 3 and then actively loaded with a cascoded used as a buffer between the first-stage output and the secondcurrent source, composed of devices Mg and Mlo, to obtain stage input. A bipolar differential pair, composed of devices the high-output resistance used to help achieve large op-amp Q3 and Q 4 , is principally used to achieve the gain in the dc gain. This large dc gain is desirable to avoid the need for pre-amp. Emitter follower Q 6 is used to prevent loading of recalibration due to gain drift arising, for instance, from a the second-stage output. Devices Q 7 - 1 0 are used to establish temperature change. Device Q1 is used as an emitter follower sufficient common mode input level to bases of Q 1 and Q 2 so to prevent the second stage from excessively loading the first- that VCB Q 1 3 remains positive. The comparator uses two for stage output and thus reducing the dc gain. Compensation preamplifiers in cascade, connected using coupling capacitors, capacitor C is used for pole-split compensation of the op which then drive a simple CMOS cross-coupled latch. Open , amp [ 121. A dynamic common mode feedback scheme is used loop offset cancellation [ 141 is used to reduce the input referred to sense the output common mode level of the op amp [lo], offset of the preamplifiers well within the calibration range. [13]. The voltage VCM represents the output common mode The gain of each preamplifier is approximately 60. Since level. A differential pair composed of devices M3 and M4 is passive loads are used in the preamplifier design with low used to steer common mode current to the first stage of the gain, a common mode feedback scheme is not required. op amp. This completes the common-mode negative feedback loop that tends to drive VCMequal to VC~IREF, stabilizing thus the output common-mode level. The voltage VCMREFis set B . Pipeline ADC Timing between the power-supply rails, or ground, for a positive The basic pipeline ADC timing is shown in Fig. 9. Twoand negative power-supply scheme. Devices M25 and M26 phase, nonoverlapping clocks 41 and 4 2 are used. The analog
V. CIRCUIT DESIGN
1212
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 28, NO. 12, DECEMBER 1993
sample
amplify compare
sample
amplify compare
sample
amplify compare
sample
amplify compare
sample
amplify compare
e e e
amplify compare sample amplify compare sample
input to the pipeline ADC is presented to the SHA while 41 is high. At the end of phase $1, the analog input is sampled. When 4 2 is high, the SHA switches to the amplify mode and its output is presented to the input of MX2 stage 1. At the end of phase 4 2 , the analog input presented to MX2 stage 1 is sampled. Also at the end of phase 42, the comparator monitoring the output in the SHA is strobed and the input bit D(0) for MX2 stage 1 is determined. This process continues until the ADC input data in question reaches the end of the pipeline. New ADC input data is sampled at the end of each phase 41.Thus, the throughput of the ADC is the period of the clock, or 1 ps. The latency of the ADC is 18 clock periods, or 18 ps. Shift registers can be used to time align the pipeline output data.
the actual implementation is fully differential, a single-ended version is shown for simplicity. The capacitors C and C2 are 1 nominally equal. A small capacitor C3, added in the circuit to reduce the nominal gain to 1.93, C2 can be expressed as C = (1 2 a)Cl, where a indicates the mismatch between C and C2, thus / a ( << 1. Nominally, C3 = pC1 where 1 /3 = 0.035. The top plates of the double-poly capacitors C1, Cz, and C3 are connected to the op-amp input. The nominal 1 value of C and C, is 1 pF. During the sample phase, the op amp is in the unity-gain connection and the analog input V;, is presented to the bottom plates of the input capacitors C1 and C2. The bottom plate of C3 is grounded. During the amplify phase, the op amp is in the inverting connection with the bottom plates of capacitors C2 and C3 connected to the opamp output, and the bottom plate of Cl is connected to +V,,f if D = 1 or -Vref D = 0. Additional clock phases derived if from the two-phase nonoverlapping clocks are used in order to minimize the charge injection in the SHA and MX2 stages [3].Performing a charge balance between sample and amplify phases, for the case D = 1 the output can be shown to be
Vout
(2
+ a)%
Kef
l+a+/3
NN
0.965(25/;, - Vr,f)
(3)
VI.
EXPERIMENTAL
RESULTS
The prototype ADC uses extemal logic circuits and software to perform the addition, subtraction, and data storage for the digital self-calibration algorithm. In order to operate the ADC, calibration data is first obtained. The converter is then run at the maximum possible speed. Calibration mode data and run mode data are obtained at 1 Msample/s. The chip was designed for a maximum sampling rate of 8 Msampleh. However, the present experimental set-up is limited to 1 Msample/s. It is surmised that the 121-pin PGA package utilized is the primary source of circuit settling degradation. C. Switched-Capacitor Multiply by Two The total harmonic distortion (THD) is computed from the Fig. 10 shows an MX2 amplifier for an even-numbered ratio of the non-sine-wave input spectral power up to the stage in the (a) sample phase and (b) amplify phase. Although Nyquist rate, or 500 kHz, to the sine-wave power. Fig. 11
1213
0.25 I
I
0
0.5 1 1.5 2 2.5 3 3.5 CALIBRATED OUTPUT CODE
XI 0 4
a2 x
I
, Vout
1.5,
' 7
Vin
+Vref -Vref
w-amp
(b) Fig. 10. Single-ended version of an MX2 amplifier for an even-numbered stage in the (a) sample phase and (b) amplify phase.
-1.5 0
2
0.5 1 1.5
2
2.5
3.5
XI 0 4
1-
1 8 0 ,0 0.5 1 1.5
.-I---.-
2.5
3.5
4.5
5
xi05
FREQUENCY Hz
Fig. 1 1 .
shows the FFT of the measured output data with a 98.756kHz sine-wave input at -2-dB full-scale (FS) input. The THD in this case is -90 dB. The input frequency of 98.756 kHz is close to the limit of the sine-wave generator, which is rated at approximately -85 dB THD+N up to 100 kHz. The input referred noise of the ADC was measured as 1.25 LSB rms at
15 b. Due to reduced power-supply operation and a capacitor voltage coefficient of 62 ppm/V, the differential reference voltage was limited to 4 V. Fig. 12 shows the plot of the measured differential nonlinearity of the ADC. The peak DNL is within h0.25 LSB. Fig. 13 shows the plot of the measured integral nonlinearity of the ADC. The peak INL is within f1.25 LSB. The DNL and INL were obtained using a sine-wave code density test [I51 with an FS sine-wave input at 9.8756 kHz. The sine-wave code density test was run for approximately 8 h collecting a 32 million sample histogram. Thus the INL shown in Fig. 13 reflects not only the capacitor voltage coefficient but also cumulative drift of the calibrated transfer characteristic over the 8-h period. The power supply of k5 V is reduced to f4 V on some parts due to changes in the process ground rules. The power dissipation reflects two design parameters. First, the ADC operates on a f 4 - V power supply. Second, to achieve high resolution, large capacitors are used, resulting in over 7 pF of loading for each operational amplifier. To settle to 15-16 b of accuracy at the target sampling rate of 8 MHz, relatively
1214
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 28, NO. 12, DECEMBER 1993
high power is necessary for the operational amplifier. Although capacitance and amplifier power can be scaled down for later stages of the pipeline, identical designs are employed for all 17 MX2 stages and the SHA for simplicity. No attempts were made to optimize the die area. Fig. 14 shows the die photograph of the ADC. The summary of performance parameters is listed in Table I. VII. CONCLUSION This paper has demonstrated a digital self-calibration technique based on radix < 2 applied to a 1-b-per-stage pipeline ADC. This technique accounts for capacitor mismatch, comparator offset, finite op-amp gain, and for DNL error contributed by circuit nonlinearities. A 15-b, 1-Msample/s pipeline ADC prototype was demonstrated utilizing an 1 1-V, 4-GHz, 2.4-pm BiCMOS process. No component matching or offsets better than 7-b accuracy are needed to attain 15-b linearity with this technique. The digital self-calibration algorithm
TABLE I CHIPPERFORMANCE
Resolution
Conversion Rate
THD
DNL INL Input Referred Noise Input Range Power Supply Power Dissipation IC Dimensions Component Count Process
15 b I Msample/s -90 dB< M.25 LSB < k1.25 LSB 1.25 LSB rms 4 v k4 v 1.8 W 9.3 mm x 8.3 mm
5 955 1 1 V, 4 GHz, 2.4 p m BiCMOS
requires only addition, subtraction, and small data storage. This technique may be applied to pipeline or cyclic ADC architectures. A 1-b-per-stage or multi-b-per-stage design may be employed with either ADC architecture.
1215
ACKNOWLEDGMENT The authors thank G. Fisher, A. Ito, and M. Snowden at Harris Semiconductor, M. Mueck at Analog Devices, and J. Bulzacchelli, S. Nadeem, J. Lutsky, and Prof. C. Sodini at MIT for technical suggestions and discussions. Fabrication of the experimental prototype was provided by Harris Semiconductor.
REFERENCES
[ I ] P. W. Li, M. J. Chin, P. R. Gray, and R. Castello, A ratio independent algorithmic analog-to-digital conversion technique. IEEE J . Solid-Stare Circuits, vol. SC-19, no. 6. pp. 1138-1143. Dec. 1984. [2] C. C. Shih er al., Reference refreshing cyclic analog-to-digital and digital-to-analog converters, IEEE J . Solid-Stare Circuits, vol. SC-21, pp. 544-554, Aug. 1986. 131 B . 3 . Song, M. F. Tompset, and K. R. Lakshmikumar, A 12-b 1Msample/s capacitor error-averaging pipelined A/D converter, IEEE J . Solid-State Circuirs , vol. 23, no. 6. pp. 13241333, Dec. 1988. [4] H. Ohara et al., A CMOS programmable self-calibrating 13-b eightchannel data acquisition peripheral, IEEE J . Solid-Stare Cif-cuits,vol. SC-22, pp. 93C938, Dec. 1987. [SI Z. G. Boyacigiller, B. Weir, and P. D. Bradshaw, An error-correcting 14b/20ps CMOS A/D converter, in Proc. IEEE Int. Solid-Srate Circurt.7 Cont.., 1981. [6] H.-Seung Lee, D. A. Hodges, and P. R. Gray. A self-calibrating 15-b CMOS A/D converter, lEEE .I. Solid-State Circurts. vol. SC-19. no. 6, pp. 813-819, Dec. 1984. [7] S.-Hoon Lee and Bang-Sup Song. A code-error calibrated two-step A D converter, in Proc. IEEE Int. Solid-Starc, Circuits Coif.. 1992. 181 Y.-M. Lin, B. Kim, and P. R. Gray. A 13-b 2. 5 MHz self-calibrated pipelined A D converter in 3-pm CMOS. lEEE .I. Solid-Stute Cfrc.uit.c. vol. 26, pp. 6 2 8 4 3 6 , Apr. 1991. [9] R. H. McCharles, Charge circuits for analog LSI. Ph.D. di\sertation. Electronics Research Laboratory, College of Engineering. Univ. of Califomia, Berkeley, Mar. 1981. [IO] A. N. Karanicolas, K. K. 0, J. Y. Wang. H.-Seung Lee. and R. L. Reif. A high-frequency fully differential BiCMOS operational amplifier. /E J . Solid-Stare Circuits. vol. 26, pp. 203-208, Mar. 19Y 1. [ 1 I ] T. L. Tewksbury. H.-Seung Lee, and G. A. Miller. The elfects of oxidc traps on the large-$ignal transient response of analog MOS circuits. IEEE J . Solid-State Circ.uir.c. vol. 24, no. 2. pp. 542-544, Apr. 1989. 1121 J. E. Solomon, The monolithic op amp: A tutorial study. IEEE .I. Solid-Stare Circiiits, vol. SC-9, no. 6, pp. 314-331. Dec. 1974. [I31 R. Castello and P. R. Gray, A high performance micropower switchedcapacitor filter, IEEE .I. Solid-State Cir.cuir.7, vol. SC-20. no. 6. pp. 1122-1 132, Dec. 1985. [ 141 A. B. Grebene, Bipolar and MOS Analoy Inte,yrared Circ.urt De.si,qn. New York: Wiley, 1984. [IS] J. Doemberg, H.-Seung Lee, and D. A. Hodges. Full-speed testing of A/D conveters, IEEE J . Solid-Stare Cirwirs. vol. SC-19. no. 6. pp. 82C827, Dec. 1984.
Hae-Seung Lee (M85-SM92) was bom in Seoul, South Korea, in 1955. He received the B.S. and M.S. degrees in Electronic Engineering from Seoul National University, Seoul, South Korea, in 1978 and 1980, respectively. He received the Ph.D. degree in electrical engineering from the University of California, Berkeley, in 1984, where he developed self-calibration techniques for A/D converters. In 1980 he was Member of Technical Staff in the Department of Mechanical Engineering at the Korean Institute of Science and Technology, Seoul, South Korea, where he was involved in the development of altemative energy sources. Since 1984, he has been with the Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, where he is now Professor. Since 1985 he has acted as Consultant to Analog Devices, Inc., Wilmington, MA, and MIT Lincoln Laboratories, Lexington, MA. His research interests are in the areas of analog integrated circuits, early vision circuits, fabrication technologies, and solid-state sensors. Dr. Lee is a recipient of the 1988 Presidential Young Investigators Award. He has served a number of technical program committees for various IEEE conferences, including the lntenational Electron Devices Meeting, the Intemational Solid-State Circuits Conference. and the Custom Integrated Circuits Conference. Currently, he is a Technical Program Committee member for the IEEE Symposium on VLSI Circuits. Since 1992, he has been an associate editor of the IEEE Journal of Solid-State Circuits.
Kantilal L. Bacrania (M82) recelved the B Sc (Hans ) degree in physics/electronics from the Univerwsity of Manchester, Manchester, UK. and the M Sc. Degree in digital techniques from HerriottWatt University, Edinburgh, Scotland He was with Burroughs Corporation and CPI, both in the UK. where his work was related to magnetic recording. read-wnte amplifiers, and data separators for disk and tape drivers He IS currently a Senior Principal Engineer at the Semiconductor Divi\ion of Harris Corporation, where he IS Group Leader for high-speed data acquisition circuit design. He has been with the company since 1980 and has worked on various high-speed A/D converters and related products, Presently he is involved in design of high-speed. lowpower converters in BiCMOS technology. Mr. Bacrania has authored or coauthored articles in JSSC and ISSCC journals and trade magazines and holds seven patents related to circuit and cystem design.
--
Andrew N. Karanicolas (S86M.92) was bom in Philadelphia, PA, on July 28, 1965. He received the S.B. and S.M. degrees in electrical engineering from the Massachusetts Institute of Technology. Cambridge, in 1987 and 1990, respectively. He is currently a candidate for the Ph.D. degree in electrical engineering at the Mas5achusetts Institute of Technology. From 1987 to 1988 he was a Teaching Assistant in the Department of Electrical Engineering and Computer Science at the Massachusett\ Institute of Technology. Since June 1988, he has been a Research Assistant at the MIT Microsystems Technology Laboratory. He was a Research lntem at Ham\ Semiconductor, Melboume, FL. from July 1991 tu December 1991. Hi\ research interests include data conversion systems. solid-state sensor interface systems, high-performance BiCMOS. CMOS and bipolar analog integrated circuits, and analog and digital signal processing.