1. Introduction
Pipelined analog-to-digital converters (ADCs) are widely used in those fields that require a high conversion rate and medium to high resolution[1]. With the rapid development of CMOS technology and design methodology in the past decades, the performances of reported pipelined ADCs are improved continuously. The pipelined ADCs in Refs. [2, 3] have already achieved a resolution of up to 16-bit with sampling rate of over 160-MSPS. However, these ADCs are implemented with conventional switched-capacitor (SC) techniques, which rely on high-gain operational amplifiers (op-amps) with large bandwidth to ensure required precision and speed. Therefore, the power consumption of these ADCs exceeds 1 W because of the high-performance op-amps used. Obviously, the power consumption of pipelined ADCs that use high-performance op-amps will increase dramatically with further improvement of speed and resolution performances. Additionally, the ADCs in Refs. [2, 3] are implemented in BiCMOS processes, which limit the embedded applications of these ADC in advanced deep submicron CMOS technology.
In order to alleviate the trade-off between power consumption and speed-resolution performances, several new design methods of pipelined ADC have been proposed and verified to be effective in the past years. An effective method is to use low gain op-amps or even open-loop amplifiers to replace the high-performance op-amps in the conversion stages[4-7], and then adopt various digital calibration algorithms to correct the gain error and nonlinearities brought by the low gain op-amps or open-loop amplifiers. In this way, the power consumption in the op-amps can be reduced extensively. However, the calibration algorithms are complex and consume considerable chip area and power. The pipelined ADCs in Refs. [8,9] are implemented based on comparators or zero-crossing methods to eliminate op-amps to reduce power consumption drastically. However, these two solutions show limited dynamic range due to the voltage fluctuations and noise from the ground rail.
The charge-domain (CD) pipelined ADCs based on bucket-brigade devices (BBDs) provide an alternative opamp-less and low-power pipelined ADC architecture[10]. However, the speed of the CD pipelined ADCs based on conventional BBDs is severely limited by the poor charge transfer efficiency[11]. Recently, the charge transfer efficiency of BBDs is improved greatly by a boosted charge transfer circuit (BCT)[12, 13]. With the novel BCTs, the conversion rate of CD pipelined ADCs is improved remarkably. However, the output charge amount of the BCT in Refs. [12,13] is influenced remarkably by process, voltage and temperature (PVT) variations, which can cause large fluctuations in the common-mode (CM) charge when the BCTs are used in CD pipelined ADCs. The CM charge errors reduce the input signal range of the ADC extensively. Furthermore, as CM charge error accumulates stage by stage, several stages at the backend of the pipeline may fail to function when the CM charge error increases to a given extent. In order to relieve the problem, complicated CM charge control techniques are introduced in Ref.[12] to enhance the performance, increasing the design complexity and power consumption.
In order to reduce the CM charge errors due to PVT variations and eliminate the CM charge control circuit, a new BCT topology that adopts a differential difference amplifier (DDA) with a voltage reference to reject the output CM charge errors caused by PVT variations is proposed in this paper. Based on the proposed BCT, a low power 10-bit 125-MSPS CD pipelined ADC without using any CM charge control techniques is implemented in 0.18
2. Proposed boosted charge transfer circuit
2.1 Analysis of the existing BCT
The BCT circuit transmits the input charge packet to the output terminal under the control of the clock signals and acts as the core circuit of a CD pipeline stage. The BCT circuit and its operating waveforms in Ref. [12] are given in Fig. 1(a). The gate of MS is controlled by the output of the amplifier A1, which is composed of M1, M2, and M3. Before
QT=C1{[VA(t2)−VA(t1)]−[VS(t2)−VS(t1)]}=C1[−Vin(t1)+VP1−V0], |
(1) |
where
In CD pipelined ADCs, two BCTs are employed in a conversion stage to perform differential signal processing[12]. Assuming that the input signals of the differential stage are
{QT,diff=C1[Vinp(t1)−Vinn(t1)]QT,CM=C1[−Vin,CM+VP1/2−V0/2], |
(2) |
where
As shown in Eq. (2), the CM charge is determined mainly by
2.2 The proposed PVT insensitive BCT
The proposed BCT structure and its operating waveforms is given in Fig. 1(b), in which a DDA is used to detect the voltage difference between nodes D and S,
QT=C1{[VA(t2)−VA(t1)]−[VS(t2)−VS(t1)]}=C2[VD(t2)−VD(t1)]. |
(3) |
After
VD(t2)−VS(t2)=Vr2−Vr1. |
(4) |
Applying Eq. (3) into Eq. (4) and substituting each voltage term with its corresponding value, the transferred charge can be recalculated as
QT=[−Vin(t1)+(Vp2−Vp1)−(Vr2−Vr1)][C1C2/(C1+C2)]. |
(5) |
For a differential conversion stage composed of two proposed BCTs, the differential and common-mode charge can be expressed as
QT,diff=[Vinp(t1)−Vinn(t1)][C1C2/(C1+C2)],QT,CM=[−Vin,CM+(Vp1−Vp2)/2−(Vr2−Vr1)/2]×[C1C2/(C1+C2)]. |
(6) |
From Eq. (6),
Circuit implementation of the proposed BCT is shown in Fig. 1(c), in which the input and output switch capacitor circuits are omitted for simplicity. Simulated transient waveforms of the proposed circuit with a clock frequency of 125 MHz are given in Fig. 1(c). The charge transferring process is initiated at
The proposed BCT is designed with 0.18
Qe=(VD−VS)C1C2/(C1+C2). |
(7) |
The simulated
As can be seen from Fig. 2, the charge waveforms of the BCT in Fig. 1(a) vary significantly with large PVT variations, while the waveforms of the proposed BCT keep almost unchanged, which verifies that the proposed BCT can reject charge errors due to PVT variations remarkably. The proposed BCT's power consumption is about twice that of the conventional one. However, this power consumption increase is much smaller than the power consumption of the CM control circuit used in conventional CD pipelined ADCs.
3. Architecture and operation of the proposed 10-bit CD pipelined ADC
3.1 Construction of CD sub-stages
Based on the proposed PVT, various sub-stages for a CD pipelined ADC can be constructed by adding appropriate sub-ADC, sub-DAC and conditional charge conditional charge subtracting (or adding) circuits. The abstraction model for a sub-stage with arbitrary resolution and its operation waveforms are given in Fig.3[14]. To simplify the description of the CD sub-stage, a single-ended configuration is used. The CD sub-stage is composed of a charge storage node Xn, two charge storage capacitors
The operation waveform of the sub-stage is given in Fig. 3(b). At
The charge relationship of the ADC sub-stage can be derived as:
Qout=Qi+Cs⋅ΔVdac+(Cc+Cs)[Vn(t0)−Vn(t4)]+Cc[Vc(t4)−Vc(t0)]=Qi+Cs⋅ΔVdac+Qc, |
(8) |
where
Qcmout=(Qi+Cs⋅ΔVdac+2Qc+2Qicm−Qi+Cs[VFdac−ΔVdac)]/2=Qicm+CsVFdac/2+Qc, |
(9) |
where
3.2 Cascading 2 or more CD sub-stages
In order to illustrate the operation of the CD pipelined ADC more explicitly, the structure of two successive 1.5-bit CD sub-stages and their controlling timing are given in Fig. 4. The 1.5-bit sub-stage circuit works under 4 non-overlapping clock phases, reset1 phase (
Similarly, before the transfer phase, the
3.3 Architecture of the 10-bit CD pipelined ADC
The block diagram of the 10-bit CD pipelined ADC in this paper is shown in Fig. 5, which is composed of a high-speed low-distortion sample and hold (S&H) circuit, a 2.5-bit CD sub-stage, 5 consecutive 1.5-bit CD sub-stages and a final 3-bit flash stage. The input differential analog voltage signal,
In order to reduce power consumption while balancing the thermal noise and matching requirements, the BBD capacitor sizes in the pipeline stages are arranged in the manner of a tapered structure[12], i.e., the capacitor values shrink with corresponding ration stage by stage. The ADC includes an on-chip precision bandgap reference voltage generator and buffer amplifiers. As 4-phase non-overlapping clock signals with over 100-MHz frequency are needed in the ADC, a distributed clock generator is employed. A duty cycle stabilizer (DCS) is also adopted to process the input clock signal and reject the clock skew and jitter.
4. Implementation of other building blocks
4.1 Sample and hold circuit
The circuit diagram of the S&H circuit is shown in Fig. 6, which is a passive SC S&H including a pair of BCT circuits. As the input analog bandwidth is needed to be over 400 MHz, in order to enhance the IF sampling performance, Ss1 and Ss2 are implemented with constant
The proposed S&H circuit works under two-phase non-overlapping clock signals,
In the voltage-to-charge transfer process discussed above, the output charge
Qip=Cp(ΔVNip−ΔVNop), |
(10) |
Qin=Cn(ΔVNin−ΔVNon), |
(11) |
where
Qid=Qip−Qin=Cs(ΔVNip−ΔVNin)=CsVid. |
(12) |
From Eq. (12), the input voltage
4.2 Charge comparators
As shown in Fig. 7(a), the charge comparator used in the sub-ADC of each pipelined sub-stage is composed of a commonly used capacitively coupled comparator and 4 charge sensors, which are connected at the inputs of the charge comparator. Charge sensors are used to isolate the charge package signal from capacitors
The input capacitors used for the capacitively coupled comparator is about 0.1 pF. No offset cancellation scheme is employed because large comparator offsets can be tolerated in pipelined ADCs with redundancy. The latched comparator is shown in Fig. 7(c), which includes three stages: input amplifier (M0-M6), regeneration latches (M7-M10), and output S-R latch (M14-M21). The input amplifier is a simple NMOS differential pair with a PMOS active load, which not only amplifies the input signal but also suppresses the kickback noise from the regeneration latches. The NMOS switches (M3 and M4) are turned off during regeneration time to save power consumption. It also helps to reduce kickback noise from the regeneration latches. The combination of PMOS and NMOS regeneration latches speeds up the regeneration compared to the PMOS-only latches. The regeneration latches are reset to a voltage close to the power supply by M11 and M12 during the sampling/resetting phase. One additional reset switch, M13, across the differential latching nodes reduces the offset due to the mismatch of M11 and M12. The NMOS switch M22 disables the NMOS regeneration latch during the resetting phase to avoid large DC current to ground. The output S-R latch holds the comparison result during the whole clock period for the convenience of following encoding logic.
5. Experimental results
The ADC has been fabricated in a 0.18
The measured output spectrum with a 3.79-MHz input sinusoid signal at 125-MSPS sampling rate is shown in Fig. 8(b). The measured SNR and the spurious free dynamic range (SFDR) are about 57.3 dB and 67.7 dB, respectively. The signal-to-noise-and-distortion ratio (SNDR) is about 55.8 dB, therefore the effective number of bits (ENOB) is about 9.0. The measured nonlinearity of the ADC is shown in Fig. 8(c). The maximum integral nonlinearity (INL) is
![]() |
6. Conclusion
A PVT insensitive BCT circuit is proposed in this paper. A low power 10-bit, 125-MSPS CD pipelined ADC based on the proposed BCT is fabricated in 0.18