A 10 bit 12.8 MS/s SAR Analog-to-Digital Converter in a 250 nrn SiGe BiCMOS Technology Johannes DigeJ, Markus Grozing, Manfred Berroth Institute of Electrical and Optical Communications Engineering University of Stuttgart, Germany Abstract-This paper presents a 10 bit 12.8 MS/s succes­ sive approximation register (SAR) analog-to-digital converter (ADC) implemented in a 250 nm SiGe BiCMOS technology. An energy-efficient switching algorithm with top-plate sampling is applied which reduces the total input capacitance by 50%. High­ Dout impedance inputs with emitter followers and internal reference voltage generation make it suitable for applications that require precise on-chip voltage monitoring. For a low-frequency input signal, measured SNDR and SFDR of the presented SAR ADC are 48.7 dB and 57.8 dB. The effective resolution bandwidth (ERBW) is 19 MHz. The ADC draws 17.4 rnA from a 2.6 V supply including reference voltage Fig. 1. Block diagram of the ADC (analog signals drawn single-ended for simplicity) generation, clock drivers and emitter follower buffers for input and reference voltages. The die area is 2.1 X 0.7 ADC core occupying 1 X 0.5 mm2• mm2 with the A formula for relating static nonlinearity (INL) measure­ ments with dynamic SNDRIENOB measurements is derived. From output codes recorded with constant input voltages, the distortion power caused by nonlinearity and the noise of the reference voltage source and the comparator are determined. After adapting them to sinusoidal inputs, the expected impact on SNDR and ENOB is derived. I. INTRODUCTION Medium-resolution medium-speed ADCs are suitable for a variety of applications. In wireless communication systems an ADC digitizes the received signal before it is passed on to a digital signal processor. Especially narrowband sub1 GHz wireless communication systems can be implemented with the presented kind of converter. Other applications are sensor systems where analog sensor measurements are to be converted into digital domain to be sent via a bus system. Furthermore such ADCs can be integrated into more sophisti­ cated circuits in an assistive manner, e.g. for calibration. In a high-speed, low-resolution time-interleaved ADC a medium­ resolution, medium-speed sub-ADC can be applied to calibrate each single channel. Another application is to monitor analog voltages inside an integrated circuit. High input impedance and internal reference voltage gener­ ation qualify the ADC to be used as an independent assistive block inside complex systems. Its fully differential design enables the integration together with CMOS logic without being prone to supply or substrate noise. The extra amount of supply power can be tolerated because assistive circuits are usually not required permanently and can be switched off when not needed. 978-1-4799-4994-6/14/$31.00 ©2014 IEEE Section II introduces the applied SC-switching algorithm and explains circuit implementation details. Measurement re­ sults with constant and sinusoidal input signals are given in Section III, Section IV relates these results by investigating the impact of noise and nonlinearity on the dynamic effective resolution. Section V concludes the paper. II. CIRCUIT DESIGN AND LAYOUT The proposed ADC works with successive approximation based on a charge redistribution principle similar to the split capacitor algorithm [1]. The block diagram in Fig. 1 shows the components required for the successive approximation. The bit values that are sequentially derived by the comparator are stored in the SAR. They are fed back to a digital-to-analog converter (DAC) and the generated analog corresponding volt­ age is subtracted from the input signal. Track-and-Hold circuit, subtractor and DAC are included in a switched-capacitor (SC) DAC composed of binary weighted capacitors and MOSFET switches. The signal input is matched to 50 S1 (not shown) and drives an emitter follower. Three reference voltages VCM, 1I;.ef+ and 1I;.ef- are generated on-chip and fed to the SC DAC. The entire circuit is designed in a differential manner to reduce distortion due to substrate or supply noise. The differential input pins are connected to emitter followers to drive the capacitive input impedance of the SC DAC. The reference voltages are generated by an operational ampli­ fier (op-amp) with common-mode control [2] and resistive feedback. Two resistor voltage dividers generate the desired common-mode voltage and another voltage controlling the reference voltage difference. The output impedances of the common-mode and reference voltage generators are decreased using emitter followers as output buffers. AA ISF � ISF � R c C DO,a D1,b Fig. 2. DO,b Fig. 4. Three-stage comparator Successive approximation register v;.ef+ to other capa­ r t- citors and "-!l<mp" (a) . .m (b) Fig. 3. Capacitor topology of an N-bit SC DAC for the MSB with (a) split capacitors forming a capacitive voltage divider [l] and (b) the implemented structure with the common-mode voltage VCM as additional reference voltage A. Successive approximation register The SAR in Fig. 2 controls the temporal sequence of the successive approximation process and switches the SC DAC according to the determined bit values. The upper line of delay-flip-flops (DFFs) is connected as a shift register which is reset after each conversion cycle by an internally generated signal. Each conversion cycle has a length of twelve clock periods. The data register, i.e. the lower line of DFFs, stores the bit values received from the comparator via the input C. The ADC operates synchronously with the clock signal elk having a frequency of twelve times the sampling rate. The outputs Di,a and Di,b control the SC DAC. All components are designed pseudo-differentially, i.e. for every logical signal its inverse is generated, too. B. Switched-capacitor digital-to-analog converter The SC DAC is implemented with a differential array of binary weighted capacitors. The capacitor weight is set by a parallel connection of switched capacitor unit cells containing a unit capacitor of Cu 5.3 fF and a three-input switch connecting the bottom plate to one of three voltages. A metal­ insulator-metal (MIM) capacitor with a thin insulator layer is used as unit capacitor, the three-input switch is composed of two cascaded transfer gates. The SC DAC uses top-plate sampling [3] which reduces the input capacitance by 50% and operates similar to the algorithm with split capacitors [1] where each capacitor can be connected to a positive or a negative reference voltage = or v;.ef- as shown in Fig. 3(a). However the binary weighted capacitors are not split but can be connected to the common-mode voltage VCM �(v;.ef+ + v;.ef-) instead, Fig. 3(b). The voltage transition caused by the switching operation after the MSB decision can be explained by Fig. 3. It shows the unit cell array controlled by the MSB, N - 2 binary downscaled unit cell arrays and a reference capacitor which are connected to the central node in parallel are not shown. With split capacitors, either the upper switch toggles to v;.ef- or the lower one to v;.ef+, depending on the MSB. This causes the voltage across the serial capacitors to change by ±(v;.ef+ - v;.ef-). Considering the voltage division by two and the weight of the MSB capacitor array of 0.5 with respect to the total capacitance, the central node voltage changes by ±:j(v;.ef+ - v;.ef-). In the proposed topology, the bottom plate of the capacitor is switched between VCM and one of the reference voltages, i.e. by ± �(v;.ef+ - v;.ef ) Taking the capacitor weight of 0.5 into account, the central voltage changes by ±:j(v;.ef+ - v;.ef ) as well, generating the :j and � threshold levels for the second decision. After reset, the bottom plates of all 512 unit capacitors are connected to the common-mode voltage VCM. According to the comparator decisions, the connections are changed to v;.ef+ to increase the DAC output voltage or to v;.ef- to decrease it. The SAR outputs Di,a select either VCM or one of the reference voltages, the outputs Di,b decide which of them is connected to the bottom plate. In the layout, the switched capacitor unit cells are placed in periodic structures surrounded by unconnected dummy cells for both of the differential arrays. This arrangement provides the same environment for each cell which reduces mismatch and distributes parasitics uniformly. = _ . _ C. Comparator The differential comparator shown in Fig. 4 determines one bit per step. It has three-stages, the first of which is a pair of PMOS source followers preventing a static current flow from the capacitive DAC and shifting up the common-mode level. The second stage is a current-mode-logic (CML) amplifier with NPN HBT input transistors Ql and Q2. Its voltage gain of 23.7 dB improves the comparator's sensitivity and isolates the inputs from the regenerative nodes of the cross-coupled p-channel MOSFETs P3 and P4 in the third stage. These transistors drive the outputs of the third state regeneratively to o \.0 � Fig. 5. Chip photograph and layout M r-----�----�--� . . . . . . . . . . . . • . . . . . . . .j • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • � ',. .... ,ol 1 ....... ..... ................."'"\/'''":..''�.< ...... ......... r'�""�J\/r\II'�\ • • • • • • • • • • • • • • • • • • • • • • • • • . 0 � \ :... ;/{\).�v�y:.. .. ... ". ... -""'- o L-____�____-L____�______�____-L____� 25 o 50 100 75 fin (MHz) 12.8 MS/s Fig. 7. SFDR, SNDR and ENOB at frequency fin 150 over the input signal s r-�--��--�-�--, CD o Z LJ.J <0 . . . . . . . . . . . . : . . . . . . .. . . . . . . . . ... , . . . . . . .. . . . . : . Fig. 6. INL at a sampling rate of fs = 12.8 MS/s rail-to-rail levels after the inverted clock signal elk has become low [4]. A NAND gate is connected to the latch outputs and acknowledges the completed decision. Its output triggers a clocked latch holding the comparator decision during the reset phase. D. Layout N . . . . . . . . ... ; . . . . . . . . . . . . . : . . . . . . , . . . . . . . . . . . ... . . .. . . . . . " . . . . . . . . .. . . . . . .. . . . . ...... ......... .... , oL--L__L-�__��__��__-L__L--L__L-�� o 2 4 6 8 10 12 14 16 18 20 26 fs (MSjs) Fig. 8. ENOB over the sampling rate fs for fin"'" 2 MHz 48.7 dB. Near Nyquist frequency, the SFDR is 55.4 dB and the SNDR is 49.5 dB. ENOB reaches 7.9 at DC and remains over 7 up to an input frequency of fin 80MHz. The effective resolution bandwidth (ERBW) is 19 MHz which exceeds the third Nyquist band. Fig. 8 shows that the ADC can operate at higher sam­ pling rates than 12.8MS/s but its performance decreases. For lower sampling rates and an input signal frequency of fin ;::::; 2MHz, the ENOB reaches a maximum of 8.2 from 4 MS/s to 10MS/s. For sampling rates exceeding 13 MS/s, the performance is limited by incomplete sampling due to the bandwidth of the sampling switch resistance and input capacitance. = A die photograph is shown in the upper half and the chip layout in the lower half of Fig. 5. The die size is 2. 1 x 0.7 mm2. The ADC core occupies approximately 0.5 mm2 in the left half of the chip. Two differential arrays of unit capacitors and switches in the upper and lower half form the SC DAC. SAR and comparator are placed in the gap between these arrays, the output register follows to the right. The reference voltage generator and the emitter followers are placed in the left part of the ADC core. III. MEASUREMENT RESULTS This section presents the measured DC and AC character­ istics of the ADC. Measurements are performed with a DC voltage source and a sinusoidal signal generator connected to the ADC inputs and a clock source providing the twelvefold sampling rate. The digital outputs are recorded by a logic analyzer. At a conversion rate of 12.8MS/s, the ADC draws 17.4 mA from a 2.6 V supply. Fig. 6 shows the integral nonlinearity (INL) at a sampling rate of fs 12.8MS/s [5]. It is worse for negative differential input voltages and approximately remains below ±1 LSB in the positive range. For smaller sampling rates, e.g. 1MS/s, the tracking period is 12.8 times longer and the INL is smaller than 0.6 LSB. This indicates that the sampling switches dominate nonlinearity compared to capacitor matching at 12.8MS/s. Assuming equal probability for all codes, the rms value of the the INL is (}INL,eq 0.54 LSB. The spurious-free dynamic range (SFDR), signal-to-noise­ and-distortion ratio (SNDR) and the effective number of bits (ENOB) at fs 12.8MS/s are given in Fig. 7. For low­ frequency inputs, the SFDR and the SNDR are 57.8 dB and = = = IV. REL ATIONSHIP OF DC AND AC MEASUREMENTS Nonlinearity error and noise deteriorate the dynamic perfor­ mance of the ADC. In this section, the INL measurement is analyzed and its expected value and variance, hence nonlinear­ ity and noise, are related to the maximum achievable SNDR and ENOB. The INL derivation is based on records of 200 samples for a set of equidistant, constant input voltages Vi,D [5]. Subtracting the mean values of these records from an ideal straight line representing an ideal transfer characteristic of an infinite-resolution ADC yields the INL given in Fig. 6. The standard deviations (}Code(Vi,D) given in Fig. 9 show the amount of random variations of the output codes for a constant input voltage Vi,D. For input voltages near zero, the mean of (}Code(Vi,D) has a minimum and it increases towards the edges of the voltage range. These random variations are mainly caused by two sources: comparator noise and reference voltage noise. Comparator noise has the same impact on all comparator decisions, hence its power (}�mp effects the conversion range (4). Thus nonlinearity and random noise degrade the signal­ to-noise-and-distortion ratio by 10 19(O"� +O"fNL +O"�) jO"� dB. With these results, the expected effective resolution from DC measurements is ENOBex = 10 Fig. 9. Standard deviation O"Code(Vi,D) of codes in INL measurement uniformly. Noisy reference voltages directly disturb the outer edges of the input voltage range but have no influence on the middle level. Thus the contribution of the differential reference voltage noise to the standard deviation of measured codes depends linearly on the differential input voltage magnitude. The noise power O"�odeCVi,D) that appears for a constant input voltage difference Vi,D is given by ) Vi,D 2 O"2Code(Vi,D) = O"ref Vi + O"2cmp 1,D,max ( (1) where the maximum allowed input voltage magnitude Vi,D,max is approximately 0.74 V. In Fig. 9, the mean value of O"Code(O) gives the comparator's noise contribution O"cmp ;::::: 0.6 LSB. From O"code(±Vi,D,max) the noise contribution from the ref­ erence voltage generator O"ref ;::::: 0.9 LSB can be determined. For measuring the SFDR, SNDR or effective resolution of an ADC, a sine signal is applied to the input. In this case, the output codes of the ADC are not equally distributed but follow the probability density function (pdt) of a sine signal. The pdf of an unscaled sine wave is Ix(x)= 1 � -x 'IT (2) for x E (-1) 1) and Ix(x) = 0 elsewhere. This pdf can be scaled to fit the input voltage range by replacing x by v; V;,D I,D,lnax The distortion power contained in the digitized sine signal due to nonlinearity can be estimated by the INL values of Fig. 6 weighted by a scaled version of Equ. (2). Summing up the weighted nonlinearity distortion power yields • (3) The amount of random excess noise can be estimated in a similar way by weighting O"Code with Equ. (2) and determining its mean square. For the measurement shown in Fig. 9 this results in (4) The SNDR of an ADC with the signal power P, quantiza­ L tion noise power O"� = �r and uncorrelated excess noise and distortion power O"� = O"fNL + O"� is given by P P SNDR = 10 19 2 = 10 Ig2 O"Q + O"E2 O"Q - 10 19 0"2QO"+Q2 O"� (5) From the INL measurement, the distortion power and excess noise O"� = O"fNL + O"� have been derived in Equ. (3) and 2 2 2 O'Q+O'INL +O'N Ig 0'2 - --------'Q-- = 10 6. 02 8.07 (6) which fits well to the ENOB measurements. The slightly worse measured ENOB is mainly caused by additional nonlinearity of the sampling switch for changing input voltages. The ENOB could be improved by optimizing the noise bandwidth of the reference voltage source and finding a better compromise between comparator speed and noise. V. C ONCLUSION An ADC suitable for the application as an assistive circuit is presented. Having high-impedance inputs and generating its own reference voltages, it can be used to monitor an internal voltage or for calibration of an analog or mixed-signal circuit. The relatively large power consumption of 45.24 mW is dominated by emitter follower buffers and the op-amp that generates the reference voltages. It can be tolerated because the ADC used as an assistive circuit can be turned off during normal operation of the surrounding circuit. The clock frequency determines the flexible conversion rate which only has minor impact on the performance up to 12.8MSjs. Thanks to the fully differential design, the ADC can be integrated together with logical circuits without being prone to substrate or supply noise. At 12.8MSjs and low frequency input signals, the effective resolution is 7.9 bit and remains over 7 bit up to 80MHz. For lower sampling rates, the effective resolution near DC input reaches 8. 4 bit. This agrees well with INL measurements at lower sampling rates showing less nonlinearity distortion error. A method to relate the results of DC-input measurements to dynamic measurements of SNDR and ENOB is presented. This method reliably predicts the ADC's dynamic performance from nonlinearity measurements. VI. A CKNOWLEDGEMENT The authors thank J.-c. Scheytt and IHP for providing chip area in IHP's 250 nm SiGe BiCMOS technology and supporting the circuit design. R EFERENCES [1] B. Ginsburg and A. Chandrakasan, "An energy-efficient charge recycling approach for a SAR converter with capacitive DAC," ISCAS 2005, vol. 1, May 2005, pp. 184-187. [2] J. Digel, M. Masini, M. Grozing, M. Berroth, G. Fischer, O. Sonom, H. Gustat, and J.-c. Scheytt, "An Integrating Digitizer for an IR-UWB Receiver," ANALOG'l1, Nov. 2011. [3] c.-C. Liu, S.-J. Chang, G.-Y Huang, and Y-Z. Lin, "A 0.92 mW lO-bit 50-MS/s SAR ADC in O.13J.im CMOS process," Symposium on VLSI Circuits 2009, Jun. 2009, pp. 236-237. [4] J. Digel, M. Grozing, M. Berroth, H. Gustat, and J.-C. Scheytt, "High­ speed comparators for SAR ADCs in 130 nm BiCMOS," PRIME 2010, Jul. 2010. [5] "IEEE Standard for Terminology and Test Methods for Analog-to-Digital Converters," IEEE SId 1241-2010 (Revision of IEEE SId 1241-2000), pp. 1-139, Jan 2011.