IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO.

11, NOVEMBER 2009

3039

A 10-Bit 500-MS/s 55-mW CMOS ADC

Ashutosh Verma, Member, IEEE, and Behzad Razavi, Fellow, IEEE
AbstractA pipelined ADC incorporates a digital foreground calibration technique that corrects errors due to capacitor mismatch, gain error, and op amp nonlinearity. Employing a highspeed, low-power op amp topology and an accurate on-chip resistor ladder and designed in 90-nm CMOS technology, the ADC achieves a DNL of 0.4 LSB and an INL of 1 LSB. The prototype digitizes a 233-MHz input with 53-dB SNDR while consuming 55 mW from a 1.2-V supply. Index TermsCalibration by inverse function, foreground digital calibration, low gain op amp, nonlinearity correction, pipelined analog-to-digital converter, resistor-ladder DAC.

I. INTRODUCTION

ECENT work on analog-to-digital converters (ADCs) has made signicant progress toward sampling rates of hundreds of megahertz and resolutions in the range of 10 to 11 bits [1], [2]. Among the reported designs, those employing a single channel face a limited speed [3], [4], while those incorporating interleaving suffer from a high power dissipation [2] or a low signal-to-(noise+distortion) ratio (SNDR) [5]. The need therefore exists for ADC architectures that combine high resolution, high speed, and low power dissipation in a single channel. Of course, interleaving can further increase the speed of such designs. This paper proposes a pipelined ADC calibration technique that allows the use of high-speed, low-power, and yet inaccurate op amps. Designed in 90-nm CMOS technology, a 10-bit prototype digitizes a 233-MHz input with an SNDR of 53 dB, for a power consumption of 55 mW [6]. The ADC derives its performance from a high-speed op amp and a highly-linear resistor ladder topology. Section II of the paper describes the ADC architecture and Section III the calibration algorithm. Section IV presents the implementation of the prototype ADC and its building blocks. Section V summarizes the experimental results. II. ADC ARCHITECTURE Pipelined ADCs, even with 1.5-bit stages, must deal with four noise, capacitor mismatch, nite op amp critical issues:
Manuscript received January 23, 2009; revised July 20, 2009. Current version published October 23, 2009. This paper was approved by Associate Editor Michael P. Flynn. This work was supported by Realtek Semiconductors, Kawasaki Microelectronics, and Skyworks Inc. Fabrication was provided by TSMC. A. Verma was with the Electrical Engineering Department, University of California, Los Angeles, CA 90095 USA, and is now with Marvell Semiconductor Inc., Santa Clara, CA 95054 USA. B. Razavi is with the Electrical Engineering Department, University of California, Los Angeles, CA 90095 USA (e-mail: razavi@ee.ucla.edu). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/JSSC.2009.2031044

gain, and op amp nonlinearity. An efcient architecture selects noise requirements but not capacitor values according to necessarily capacitor matching requirements, handling the latter by calibration. The two op-amp-related issues begin to manifest themselves as it becomes increasingly more difcult to design high-gain, high-swing op amps using deep-submicron devices. Table I summarizes recent work on the correction of these errors and the limitations of each technique [7][16]. It is observed that only [16], [12], and [15] address the amplier nonlinearity issue, but the rst sacrices the input dynamic range, the second does not correct for capacitor mismatch, and the third operates with an op amp gain of several hundred (for 12 bits). An ADC suffering from various errors can be viewed as a system having a general input-output characteristic given by . As shown in Fig. 1(a), an inverse operator in the digital domain can correct for all errors, producing a faithful . However, it is difcult to realize this replica of the input, inverse function with reasonable complexity. For example, if the must entail ADC characteristic is nonmonotonic, then hysteresis. Now consider one stage of a pipelined ADC. We surmise that if, as depicted in Fig. 1(b), the residue is digitized by an ideal digitizer (i.e., an ideal back end), then it can be corrected more easily because its errors are not as complex as those in a general , ADC. Nonetheless, since the digital output of this stage, may bear gain error with respect to the analog residue value (as explained in Section III), an additional gain correction step is necessary [Fig. 1(c)]. We also recognize that the ideal back end is simply formed if the calibration begins with the last stage of the pipeline and progresses backward. That is, calibration of and higher are ideal. stage can assume that stages The arrangement shown in Fig. 1(c) is the basis of the proposed ADC architecture and calibration technique. The architecture is shown in Fig. 2. It consists of 13 1.5-bit stages and one 1-bit stage. The rst two stages are calibrated for residue gain error, digital-to-analog-converter (DAC) gain error, and op amp nonlinearity; the next four stages for residue gain and DAC gain error; and the next seven stages for residue gain error. The digital calibration back end consists of programmable [ideally equal to ], and two progain coefcients, grammable third-order polynomials of the form , which approximates the inverse function of the input/output characteristic of each multiplying digital-to-analog converter (MDAC). With the aid of a highly-accurate on-chip reference DAC, the system applies a number of analog levels at the input (at start-up) and uses a least-mean-square (LMS) and so as to drive a certain error function engine to adjust to zero. The sub-ADC and the MDAC in the rst stage sample simultaneously, thereby obviating the need for an explicit front-end sample-and-hold amplier (SHA) [17][19]. Such

0018-9200/$26.00 2009 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3040

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

TABLE I SUMMARY OF PRIOR ART

, , and the op amp generates the residue. In the ideal of , case, the residue is expressed as denotes the digital output of the sub-ADC. That is, where the sampled analog input can be estimated as (1) Capacitor mismatch and nite op amp gain give rise to a gain error while op amp nonlinearity introduces a nonlinear component. If , , and the open-loop input/output characteristic of the op amp is represented by , then in Fig. 3 is equal to and hence (2) Incorporating the effect of the nite input capacitance of the op amp , we have
Fig. 1. Calibration concept.

a SHA would consume substantial power while contributing signicant noise and nonlinearity. Nevertheless, the sampling networks at the input of the sub-ADC and the MDAC must sustain minimal mismatches such that the resulting discrepancy can be corrected by the 0.5-bit redundancy. Also, the sub-ADC conversion time now limits the time allocated to the settling of the MDAC. These issues are discussed in Section IV. III. CALIBRATION TECHNIQUE Consider the 1.5-bit pipelined stage shown in Fig. 3, where the sub-ADC consisting of two comparators determines whether or not, and the MDAC consisting

(3) Note that capacitor mismatch, , appears in the rst term, leading to a residue gain error, and in the last term, translating to a DAC gain error. In accordance with our notation in Fig. 1(c), we observe that the rst two terms in (3) constitute and the coefcient of is the same as . In the architecture of Fig. 2, is approximated by a third-order polynomial for the rst two stages. The op amp used in this work (Section IV) achieves a high speed and large output swings but suffers from a low gain ( ). We therefore expect a relatively high closed-loop nonlinearity. On the other hand, the architecture of Fig. 2 assumes that

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

VERMA AND RAZAVI: A 10-BIT 500-MS/S 55-MW CMOS ADC

3041

Fig. 2. Pipelined ADC architecture.

Fig. 3. Switched capacitor implementation of 1.5-bit stage.

the inverse transfer characteristic of each stage can be approximated by no more than third-order terms. To verify the validity of this assumption, we rst rewrite (3) as

Fig. 4. Error in V

with third-order polynomial tting.

(4) Next, we perform a transistor-level transient simulation on and forcing the multiplexer output to 0) Fig. 3 (e.g., with whereby is slowly varied from to (in differential implementation) and is measured. Last, and so as to obtain a good we estimate the values of and t (e.g., with minimum mean-square error between ). If the residual error between the actual and ) remains well below 1 the value predicted by (4) ( LSB, then the third-order approximation is justied. Fig. 4 , and plots this error across the input range (with

), revealing a maximum of about 0.15 LSB and implying that the approximation is reasonable. A. Calibration Procedure The calibration begins with the last stage and proceeds toward the front end. As a result, calibration of stage can assume that the subsequent stages constitute an ideal back end. Fig. 5 illustrates the calibration arrangement. The reference DAC applies dc inputs to stage such that the sub-ADC of this stage produces and the back end generates . Note that accurately represents the residue output of stage . Next, is so as to undo the nonlinearity created by subjected to to arrive at the overall stage , and is combined with

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3042

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

The value of is therefore adjusted to minimize the difference and . The proposed calibration operates in the between foreground, potentially suffering from drifts with temperature. This issue is addressed in Section IV-B. D. Back-End Stages The low open-loop gain of the op amp yields a closed-loop residue gain of only 1.7. Owing to this large departure from the ideal value of 2.00, a cascade of pipelined stages provides an overall gain substantially less than , failing to generate digital codes for the lower and upper ends of the range. For this reason, the architecture of Fig. 2 employs eight back-end stages to achieve ve bits of resolution. (The digital outputs are eventually truncated to ve bits after calibration logic.) As indicated in Fig. 2, the last eight stages are calibrated for only residue gain error. Since capacitor mismatch is negligible here, only the nite gain of the op amp produces gain error. Moreover, since the mismatches among the gains of the op amps used in these stages are negligible at this resolution level, we assume that the stages exhibit equal gain errors and hence can be calibrated by a single variable. To calibrate the back end, its rst stage (stage number 7 in Fig. 2) is congured as a multiply-by-two circuit while sensing provided by the calibration DAC. an input equal to are then adjusted so as to minimize the differCoefcients ence between the digital output of this back end and . A ow chart of the steps described above is shown in Fig. 7. IV. ADC DESIGN The architecture of Fig. 2 has been realized in 90-nm CMOS technology. This section presents the implementation details. A. Input Sampling Network As mentioned in Section II, the rst ash stage and MDAC sample the analog input simultaneously, facing potential timing mismatches and hence inconsistencies between the digital and residue outputs. Half a bit of redundancy alleviates this issue considerably, but reasonable path matching must be ensured to leave margin for other errors (e.g., comparator offset). Fig. 8 shows the front-end input sampling network in the acquisition mode. To keep the two paths nominally identical, the input to the op amp or the comparator is disconnected in this and turn off before others, thus permode. Switches , forming bottom-plate sampling. Another critical issue in the front end is that the ash ADC conversion time must be minimized to allow sufcient settling time for the MDAC. The tight timing budget requires careful partitioning of the clock period, , among three operations: sampling, ash conversion, and residue generation (in the rst stage). Since bootstrapped switches can provide a relatively short acquisition time while achieving high linearity, this design allocates 25% of to sampling, and the other 75% to ash conversion and residue generation (in the rst stage). These waveforms are derived from an input clock frequency of 1 GHz, which is divided by 2 so as to generate quadrature phases. Two of the phases are then ANDed, producing the necessary 25% duty cycle.

Fig. 5. Digital calibration.

output, . In the ideal case, must be equal to the digital input of the reference DAC, . Thus, serves as the error function that must be minimized by the LMS algorithm. The calibration procedure consists of two steps: estimation and without interaction with , and estimation of of with and set properly.1 These steps are described below. B. Estimation of and

In the rst step of calibration, dc inputs equal to , , and 0 are produced by the reference DAC in Fig. 5 and applied to stage . Stage is congured as a mul. Thus, in Fig. 5 tiply-by-two circuit such that (5) The LMS algorithm adjusts and so as to drive the mean to zero. Specically, and are square error of updated according to the following equations: (6) (7) Fig. 6(a)(c) depict the convergence of , , and in a system-level simulation of the rst stage.2 C. Estimation of The coefcients correct for the effect of capacitor misconversion [the last term in (4)]. In this case, match on conthe MDAC operates in the regular mode (sampling, version, multiplication by 2). The reference applies a voltage such that the digital equivalent of (4) plus equal to can be written as (8)
that the gain error of the op amp is captured within  .  is estimated rst, and then  . It is later determined that the simultaneous calibration of  and  is a better technique and is explained here.
1Note 2In [6], coefcient

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

VERMA AND RAZAVI: A 10-BIT 500-MS/S 55-MW CMOS ADC

3043

Fig. 6. Simulated convergence of (a)  , (b)  , and (c)

D 0D

The absence of the op amp from the sampling network in Fig. 8 means that its offset is not removed. While benign in general ADCs, such a front end offset does create discrepancy in the calibration mode as the correcting third-order polynomial assumes that there is no offset present in the system. To cancel this offset, a zero dc input is applied and the resulting digital is stored in the memory. This digital output output is then subtracted from the overall output when the calibration coefcients are computed. B. Op Amp The speed and power consumption of the ADC are determined primarily by those of the op amp. This work views the op amp as an amplier having large output swings and maximum speed with little attention to its open-loop gain. The large noise requirements, directly leading output swings relax to a lower power consumption. The need for a high-swing op amp naturally points to a two-stage topology, and the desire for maximum speed, to the smallest number of poles, namely, two. From these observations emerges the op amp shown in Fig. 9. To maintain a true two-pole (uncompensated) behavior, the circuit avoids cascode devices. Moreover, to achieve fast common-mode (CM) settling, each stage employs a simple resistive feedback network. Note that the bias current of the output stage is dened as a multiple of through the current mirror action of the pMOS devices. Also, the output CM level is raised to so . that it reaches approximately In order to maximize the uncompensated pole frequencies, the circuit of Fig. 9 incorporates minimum-length transistors in the signal path, thus exhibiting an open-loop gain of only

Fig. 7. Calibration ow chart.

The following stages in the pipeline operate with a 50% duty cycle. The switching of the second-stage sampling capacitors to the output of the rst MDAC generates a glitch, which subsides over the remaining half cycle. This glitch is, however, relatively small because of the smaller capacitors used in the second stage.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3044

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

Fig. 8. Input sampling networks and timing diagram.

Fig. 9. Two-stage op amp schematic.

Fig. 10. Simulated nonlinearity of the op amp.

plots the simulated open-loop nonlinearity of the op amp for a peak-to-peak differential output swing of 1.2 V. In order to study the behavior of the op amp in the MDAC environment, we construct the equivalent circuit shown in Fig. 11 (without the compensation network). Here, and denote the voltage gain and output resistance of stage , respecmodels the total capacitance at the output of the tively, and rst stage. Capacitors , , and represent those required for MDAC operation. , can be obtained by setting The loop transmission, to zero, breaking the loop at and computing the transfer function around the loop. It follows that (9)

25. As explained in Section II, the low gain yields substantial closed-loop nonlinearity, necessitating calibration. Fig. 10
Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

VERMA AND RAZAVI: A 10-BIT 500-MS/S 55-MW CMOS ADC

3045

Fig. 11. Small-signal model of MDAC before compensation.

Fig. 12. (a) Series poly resistors with current-carrying contacts. (b) Continuous polysilicon ladder (top view).

Fig. 13. Reference DAC.

where

and

(10) (11)

The magnitude of

falls to unity at a frequency given by

(12) In this design, , The phase shift at GHz , , and . Thus, is thus given by GHz .

a negative or near-zero uncompensated phase margin. Compensated for a phase margin of 60 , the op amp of Fig. 9 provides a unity-gain bandwidth of 10 GHz. The device dimensions and bias currents shown in Fig. 9 correspond to those in the rst MDAC. The MDAC is scaled down by a factor of two in the second stage and four in the third stage. Stages 4 to 13 remain unscaled due to the small size of the capacitors (25 fF) and negligible power consumption. The foreground calibration technique proposed herein assumes negligible drift of the MDAC characteristics with temperature. Since the op amp operates in a closed-loop conguration, variation of its characteristics is suppressed by the loop gain. Nonetheless, for a constant bias current, the op amp small-signal gain varies markedly with temperature, degrading the performance.3 On the other hand, it was found from simulations that if the overdrive voltage of the input transistors is kept constant, so is the small signal gain. Using the bias circuit in [21] for this purpose, simulations suggest an SNDR degradation of about 2 dB if the circuit is calibrated at 27 C and the temperature rises to 75 . C. Reference DAC The calibration algorithm proposed here relies on a highly-accurate on-chip DAC. The reference fractions used in the calibration process (Section III) must be generated with a precision higher than approximately 11 bits. The choice of the DAC is governed by the following considerations [20]: 1) capacitor DACs require large units to achieve high accuracy, thus
3The

(13) (14) The key observation here is that the phase margin is about 42 before compensation, thereby requiring only a moderate reducso as to reach an adequate amount, e.g., 60 . This tion of stands in contrast to the behavior of typical two-stage op amps, especially if loaded with a signicant capacitance, which exhibit

nonlinear terms vary negligibly.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3046

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

Fig. 14. INL distributions for ladders of different dimensions, (a) L (d) L = 64 m, W = 20 m.

= 32 m, W = 10 m, (b) L = 32 m, W = 20 m, (c) L = 64 m, W = 10 m, and

slowing down the MDAC if they are tied to the virtual ground; 2) current-steering DACs suffer from both gain error and a limited output swing; 3) resistor ladders can provide a zero-offset rail-to-rail output but with a long settling time. Since this work employs foreground calibration, the third choice proves most suitable. In order to design a resistor ladder having a reproducibly accurate set of outputs, a number of error sources must be considered. First, the very thin layer of silicide deposited on polysilicon can experience large thickness variations from one end of the ladder to the other. Thus, non-silicided poly is preferable. Second, the relatively large contact resistance and the poor definition of its value (depending on how much and how deeply the metal lls each contact window) [Fig. 12(a)] suggest that current-carrying contacts can potentially introduce large errors in voltage division. It is therefore desirable to avoid such contacts along the ladder. Third, even contacts that carry no current and simply sense a voltage must create minimal disturbance in the ow of the current. This concern arises because the silicided area under each contact disturbs the current ow, and misalignment in its position leads to random disturbance [Fig. 12(b)]. Based on these observations, we propose the ladder structure shown in Fig. 13. The voltage-sensing contacts are placed on the edge, introducing negligible disturbance in the current ow. Also, since the poly segment between the end and the tap

is not identical to that between taps and (due to the large number of contacts and their underlying silicided area at ), the , are taken from full-scale references used for the ADC, the rst and last taps rather than from the ends. The ladder structure of Fig. 13 has been fabricated separately with different dimensions and its integral nonlinearity (INL) prole has been measured for differential outputs on 40 samples. Fig. 14 plots the distributions of the maximum measured INL for four sets of dimensions: m and 20 m, and m and 64 m. It is observed that the peak INL falls to a value of 0.027% for m and m. These dimensions are chosen for the reference ladder used in this work. The high linearity of the resistor ladder makes it also attractive for use as an interstage multi-bit DAC in the main signal path (in pipelined or subranging architectures). However, the high resistance of the ladder gives rise to long settling times.4 V. EXPERIMENTAL RESULTS The prototype ADC has been fabricated in 90-nm digital CMOS technology. Shown in Fig. 15 is the die, whose active area measures 700 m 700 m. The resistor-ladder DAC is
4The DAC has 3 bits of resolution and a total resistance of 1.3 k
. It occupies an area of 64 m 20 m, consumes 0.3 mW, and can run at a speed of 100 MHz.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

VERMA AND RAZAVI: A 10-BIT 500-MS/S 55-MW CMOS ADC

3047

Fig. 15. Die photograph.

Fig. 16. Block diagram of synthesized calibration logic.

included on the chip. The digital outputs are downsampled by a factor of 16 to simplify testing. The die has been mounted directly onto a printed-circuit board. Operating with a 1.2-V supply, the ADC draws 55 mW, of which 40 mW is consumed by the op amps and 15 mW by the comparators and the clock buffer. The calibration is run off-chip, but a detailed gate-level synthesis of the calibration logic is performed to estimate the associated power dissipation. Fig. 16 shows a block diagram of the synthesized system. The complexity is about 20,000 gates. , The nonlinearity correction requires the function which, if implemented directly, incurs a considerable area and power penalty. However, the logic can be greatly simplied by

) is only a few LSBs observing that the correction term ( wide. The adder output (shown as input ) is also truncated to is represented by only 4 bits. The 6 bits and the coefcient output of this block is truncated to 6 bits. With a clock frequency of 500 MHz in 90-nm CMOS technology, the total power disand sipation of the logic is about 8 mW with . 5.6 mW with Fig. 17 shows the measured DNL and INL at a sampling rate of 500 MHz before and after gain error, DAC error, and nonlinearity calibration. The uncalibrated prototype suffers from a large number of missing codes and an INL of 40 LSB. After full calibration, the DNL and INL fall below 0.4 LSB and 1 LSB, respectively.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3048

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

Fig. 17. Measured differential and integral nonlinearity at a sampling rate of 500 MHz.

Fig. 18. Output spectrum for 10.7 MHz input frequency at a sampling rate of 500 MHz before calibration.

Fig. 19. Output spectrum for 10.7 MHz input frequency at a sampling rate of 500 MHz after calibration.

Figs. 18 and 19 show the measured output spectrum for an input frequency of 10.7 MHz before and after calibration, respectively. The sampling rate is 500 MHz. (Due to the downsampling factor of 16, the spectrum is shown up to half of 500 MHz/16 31.25 MHz.) Before calibration, the SNDR is only 28.28 dB. After full calibration, the SNDR improves to 55.64 dB. Since the background calibration is performed at a sampling rate of 100 MHz, operation at 500 MHz suffers slightly from uncorrected gain error and nonlinearity. Also, the fth-order harmonic remains uncorrected and relatively signicant. Fig. 20 shows the output spectrum for an input frequency of 233 MHz.

The SNDR is equal to 52.8 dB, yielding a gure-of-merit (FOM) of 0.3 pJ/conversion. The degradation is partially attributed to the ringing on the external reference lines that feed the sampling capacitors of all of the MDAC stages. Even though a bypass MOS capacitor of 500 pF is tied on-chip between and , the total capacitance that MDACs switch to these reference reaches 500 fF, creating considerable ringing for a bond wire inductance of greater than 2 nH. It is expected that an on-chip reference buffer improves the dynamic performance. Fig. 21 plots the measured SNDR as a function of the analog input frequency at a sampling rate of 500 MHz. Table II compares the performance of this ADC with that of prior art.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

VERMA AND RAZAVI: A 10-BIT 500-MS/S 55-MW CMOS ADC

3049

TABLE II COMPARISON OF THE PERFORMANCE OF THIS ADC WITH THAT OF PRIOR ART

method and an accurate resistor ladder topology that can remove residue gain error, DAC error, and op amp nonlinearity. Owing to a high-speed low-power op amp design, the ADC achieves 53 dB SNDR reported for a power consumption of 55 mW. REFERENCES
[1] S. K. Gupta, M. A. Inereld, and J. Wang, A 1-GS/s 11-bit ADC with 55-dB SNDR, 250-mW power realized by a high bandwidth scalable time-interleaved architecture, IEEE J. Solid-State Circuits, vol. 41, no. 12, pp. 26502657, Dec. 2006. [2] C.-C. Hsu, F.-C. Huang, C.-Y. Shih, C.-C. Huang, Y.-H. Lin, C.-C. Leem, and B. Razavi, An 11b 800 MS/s time-interleaved ADC with digital background calibration, in IEEE ISSCC Dig. Tech. Papers, 2007, pp. 464465. [3] S.-C. Lee, Y.-D. Jeon, K.-D. Kim, J.-K. Kwon, J. Kim, J.-W. Moon, and W. Lee, A 10b 205 MS/s 1 mm 90 nm CMOS pipeline ADC for at-panel display applications, in IEEE ISSCC Dig. Tech. Papers, Feb. 2007, pp. 458459. [4] K. Gulati, M. S. Peng, A. Pulinchery, C. E. Munoz, M. Lugin, A. R. Bugeja, J. Li, and A. P. Chandrakasan, A highly integrated CMOS analog baseband transceiver with 180 MSPS 13-bit pipelined CMOS ADC and dual 12-bit DACs, IEEE J. Solid-State Circuits, vol. 41, no. 8, pp. 18561866, Aug. 2006. [5] S. M. Louwsma, E. J. M. van Tuijl, M. Vertregt, and B. Nauta, A 1.35 GS/s, 10b, 175 mW time-interleaved AD converter in 0.13 m CMOS, in VLSI Circuits Symp. Dig., Jun. 2007, pp. 6263. [6] A. Verma and B. Razavi, A 10b 500 MHz 55 mW CMOS ADC, in IEEE ISSCC Dig. Tech. Papers, Feb. 2009, pp. 8485. [7] B. S. Song, M. F. Tompsett, and K. R. Lakshmikumar, A 12-bit 1-Msample/s capacitor error-averaging pipelined A/D converter, IEEE J. Solid-State Circuits, vol. 23, pp. 13241333, Dec. 1988. [8] Y.-M. Lin, B. Kim, and P. R. Gray, A 13-b 2.5-MHz self-calibrated pipelined A/D converter in 3 m CMOS, IEEE J. Solid-State Circuits, vol. 26, no. 4, pp. 628636, Apr. 1991. [9] A. N. Karanicolas, H.-S. Lee, and K. L. Barcrania, A 15-b 1-Msample/s digitally self-calibrated pipeline ADC, IEEE J. Solid-State Circuits, vol. 28, no. 12, pp. 12071215, Dec. 1993. [10] U. Moon and B. S. Song, Background digital calibration techniques for pipelined ADCs, IEEE Trans. Circuits Syst. II: Analog Digit. Signal Process., vol. 44, no. 2, pp. 102109, Feb. 1997. [11] J. Ming and S. H. Lewis, An 8-bit 80-Msample/s pipelined analog-todigital converter with background calibration, IEEE J. Solid-State Circuits, vol. 36, no. 10, pp. 14891497, Oct. 2001. [12] B. Murmann and B. E. Boser, A 12-bit 75-MS/s pipelined ADC using open-loop residue amplication, IEEE J. Solid-State Circuits, vol. 38, no. 12, pp. 20402050, Dec. 2003. [13] X. Wang, P. J. Hurst, and S. H. Lewis, A 12-bit 20-Msample/s pipelined analog-to-digital converter with nested digital background calibration, IEEE J. Solid-State Circuits, vol. 39, no. 11, pp. 17991808, Nov. 2004. [14] E. Siragusa and I. Galton, A digitally enhanced 1.8-V 15-bit 40-MSample/s CMOS pipelined ADC, IEEE J. Solid-State Circuits, vol. 39, no. 12, pp. 21262138, Dec. 2004.

Fig. 20. Output spectrum for 233 MHz input frequency at a sampling rate of 500 MHz.

Fig. 21. Measured SNDR as a function of input frequency at a sampling rate of 500 MHz.

VI. CONCLUSION As each generation of CMOS technology continues to further limit the performance of op amps, calibration techniques that deal with not only gain error but nonlinearity become essential. This paper has introduced a pipelined ADC calibration

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.

3050

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 44, NO. 11, NOVEMBER 2009

[15] C. R. Grace, P. J. Hurst, and S. H. Lewis, A 12-bit 80-Msample/s pipelined ADC with bootstrapped digital calibration, IEEE J. SolidState Circuits, vol. 40, no. 5, pp. 10381046, May 2005. [16] A. Panigada and I. Galton, A 130 mW 100 MS/s pipelined ADC with 69 dB SNDR enabled by digital harmonic distortion correction, in IEEE ISSCC Dig. Tech. Papers, Feb. 2009, pp. 162163. [17] T. Cho and P. R. Gray, A 10 b, 20 Msample/s, 35 mW pipeline A/D converter, IEEE J. Solid-State Circuits, vol. 30, no. 3, pp. 166172, Mar. 1995. [18] D.-Y. Chang, Design techniques for a pipelined ADC without using a front-end sample-and-hold amplier, IEEE Trans. Circuits Syst. I: Reg. Papers, vol. 51, no. 11, pp. 21232132, Nov. 2004. [19] I. Mehr and L. Singer, A 55-mW, 10-Bit, 40-Msample/s Nyquist-rate CMOS ADC, IEEE J. Solid-State Circuits, vol. 35, no. 3, pp. 318325, Mar. 2000. [20] B. Razavi, Principles of Data Conversion System Design. New York: IEEE Press, 1995. [21] M. Yoshioka, M. Kudo, T. Mori, and S. Tsukamoto, A 0.8 V 10b 80 MS/s 6.5 mW pipelined ADC with regulated overdrive voltage biasing, in IEEE ISSCC Dig. Tech. Papers, Feb. 2007, pp. 452453.

Ashutosh Verma (M08) received the B.Tech. degree in electrical engineering from the Indian Institute of Technology, Delhi, India, in 1999, and the M.S.E.E. and Ph.D.E.E. degrees from the University of California, Los Angeles, in 2004 and 2009, respectively. From 2000 to 2002, he was with Biomorphic VLSI Inc., Thousand Oaks, CA, where he designed digital integrated circuits for CMOS imagers. In March 2009, he joined Marvell Semiconductor Inc., Santa Clara, CA, where he is a Senior Analog Design Engineer in the Wireless Baseband group. His research interests are in analog-to-digital conversion and mixed-signal design.

Behzad Razavi (F03) received the B.S.E.E. degree from Sharif University of Technology, Tehran, Iran, in 1985, and the M.S.E.E. and Ph.D.E.E. degrees from Stanford University, Stanford, CA, in 1988 and 1992, respectively. He was with ATT Bell Laboratories and Hewlett-Packard Laboratories until 1996. Since 1996, he has been Associate Professor and subsequently Professor of electrical engineering at the University of California, Los Angeles. He was an Adjunct Professor at Princeton University from 1992 to 1994, and at Stanford University in 1995. His current research includes wireless transceivers, frequency synthesizers, phase-locking and clock recovery for high-speed data communications, and data converters. He is the author of Principles of Data Conversion System Design (IEEE Press, 1995), RF Microelectronics (Prentice-Hall, 1998) (translated into Chinese, Japanese, and Korean), Design of Analog CMOS Integrated Circuits (McGraw-Hill, 2001) (translated into Chinese and Japanese), Design of Integrated Circuits for Optical Communications (McGraw-Hill, 2003), and Fundamentals of Microelectronics (Wiley, 2006) (translated to Korean), and the editor of Monolithic Phase-Locked Loops and Clock Recovery Circuits (IEEE Press, 1996), and Phase-Locking in High-Performance Systems (IEEE Press, 2003). Prof. Razavi served on the Technical Program Committees of the IEEE International Solid-State Circuits Conference (ISSCC) from 1993 to 2002 and the VLSI Circuits Symposium from 1998 to 2002. He has also served as Guest Editor and Associate Editor of the IEEE JOURNAL OF SOLID-STATE CIRCUITS, IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, and International Journal of High Speed Electronics. He received the Beatrice Winner Award for Editorial Excellence at the 1994 ISSCC, the Best Paper Award at the 1994 European Solid-State Circuits Conference, the Best Panel Award at the 1995 and 1997 ISSCC, the TRW Innovative Teaching Award in 1997, and the best paper award at the IEEE Custom Integrated Circuits Conference in 1998. He was the co-recipient of both the Jack Kilby Outstanding Student Paper Award and the Beatrice Winner Award for Editorial Excellence at the 2001 ISSCC. He received the Lockheed Martin Excellence in Teaching Award in 2006 and the UCLA Faculty Senate Teaching Award in 2007. He was also recognized as one of the top 10 authors in the 50-year history of ISSCC. He is an IEEE Distinguished Lecturer.

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on November 5, 2009 at 13:59 from IEEE Xplore. Restrictions apply.