8th WSEAS International Conference on POWER SYSTEMS (PS 2008), Santander, Cantabria, Spain, September 23-25, 2008 Improving Dynamic Performance of a Segmented Current-Steering Digital-to-Analog Converter LUCIAN JURCA1, CONSTANTIN VOLOSENCU2, MIRCEA TOMOROGA1, IOAN FILIP2 1 Faculty of Electronics and Telecommunications Engineering 2 Faculty of Automation and Computer Science and Engineering “Politehnica” University of Timisoara B-dul Vasile Parvan nr.2, Timisoara ROMANIA Abstract: - In this paper we present an optimized structure of a 10-b current-steering digital-to-analog converter (DAC) with very good dynamic performance, especially regarding spurious-free dynamic range (SFDR). To reduce the influence of matching errors, an appropriate dynamic element matching (DEM) technique was implemented and OrCAD simulations were used to confirm more realistic improvements than in related works. Key-Words: Spurious-free dynamic range, dynamic element matching, current-steering DAC. is to use the thermometer code, as in [3], [4], [6], [7], [8] and in our case, but there are other redundant codes as well [3]. In fact all these works adopted segmented architectures where the dynamic randomization or, more exactly, the pseudo randomization was made inside the most significant segment only. This segment was built with unary current sources (unit elements) while the least significant part was designed with binary-weighted current sources. Adopting partial randomization and not full randomization happened because the area of the circuitry grows exponentially with the number of scrambled bits. Typically, the full-randomization techniques are used in lower-bit DACs and for example in the feedback DACs in sigma-delta ADCs. Only few works discuss or implement the full randomization for Nyquist-rate data conversion [3], [4], but there the number of input bits is still limited. In this paper we present a structure of a 10-b currentsteering digital-to-analog converter (DAC) with partial pseudo randomization which uses a low-complexity circuitry to achieve good dynamic performance. We adopted the structure of the converter presented in [8] in which the whole converter was included in a subbandgap mechanism. The segmentation with five most significant bits (MSBs) and five least significant bits (LSBs) was made by means of two identical binaryweighted current source blocks connected at the output node through two resistors of appropriate values to ensure the correct weighting in the output voltage contribution. In the present work the most significant segment of the converter described in [9] was built with unary current sources and it will be controlled by a modified 1 Introduction High-speed, high resolution digital-to-analog converters (DACs) are essential components in a wide range of modern applications such as wired and wireless communications [1], precision analog microcontrollers, direct digital synthesis, high-resolution displays or video signal processing. In particular, the current-steering DACs have proved to be suitable candidates for these applications which demand high spectral purity and small output errors. Another important feature of this type of converter is that it can be built by using the unit element approach where special layout techniques and switching schemes can be used to reduce the influence of matching errors [2], [3], [4], [5]. In this way the effect of the deterministic mismatch-generating process caused by the long correlation distance (the matching is dependent of the distance between transistors over the wafer and originates from wafer fabrication and oxidation process) is averaged and static parameters of DAC as integral nonlinearity (INL) and gain error are improved. Furthermore, in order to obtain a better dynamic performance, especially regarding the spurious-free dynamic range (SFDR) of the converter, the dynamic randomization of the unit current sources or dynamic element matching (DEM) are used. Thus, the effect of the stochastic mismatch-generating process caused by the short correlation distance (arising from local irregularities and characterized by the standard deviation σ of the current generated by one unit element) is dynamically averaged. In this way we can spread the undesired harmonics as white noise over the output spectrum. The technique can be applied if and only if we use redundant codes in the DAC. The classical approach ISSN: 1790-5117 201 ISBN: 978-960-474-006-2 8th WSEAS International Conference on POWER SYSTEMS (PS 2008), Santander, Cantabria, Spain, September 23-25, 2008 digital input according to DEM techniques, case in which each unary current source is controlled by a switch. The segmentation of the DAC was also changed, according to the analysis results obtained in [10] where the area, matching errors and power consumption considerations were simultaneously taken into account. In this way the input code of the proposed DAC contains six MSBs and four LSBs. Lowering the supply voltage of this converter (1.1V) is also possible and easy to implement by adopting the modified structure presented in [11]. The outline of the paper is as follows. In section 2 we present SFDR considerations where we explain way this parameter is representative for the dynamic behavior of a current-steering DAC and we discuss the manner in which it can be evaluated. In section 3 we will present the architecture of the converter and the simulation results regarding SFDR. In section 4 we will conclude the paper. different delays by properly choosing the distribution of switches and current cells in the floor plan of the layout. Regarding the third factor, the element (i.e. the unit current source) mismatch, it affects both the static and dynamic performances of the converter and it introduces significant noise and distortion to the output, limiting the achievable SFDR of the DAC, at least at low frequencies. Because of the stochastic-nature component of the mismatch, one single-tone measure (one single input sinusoid to measure SFDR) may not be sufficient to characterize the dynamic performance. Applying several single-tone sinusoidal inputs at different frequencies represents a better solution. Sometimes dual-tone or even multi-tone measurements are more informative, since more repetitive combinations of non ideal output voltages are possible and the output signal is more realistically distributed through the frequency domain. This is for why the SFDR definition must be revised as we already specified. The effect of matching errors can be diminished by using DEM techniques but the problem which still remains is that these techniques, for the thermometer case, basically destroy the low-glitch properties. However, the current-steering segmented DACs proposed in [9] and [11] present very good immunity to the switching activity. This feature is extremely important because it will allow us to simplify the thermometer encoder while keeping the achieved dynamic performance with a classical encoder. One single aspect is still to be discussed here. Very elaborate theoretical models of DEM DACs [3], [4], [7], and static element matching (SEM) DACs [4], [5] were presented in the field literature and rigorous behaviorallevel simulations such as Matlab simulations confirmed the functionality of the mathematical models. Using Matlab instead of a PSpice simulator a more flexible interface is allowed and you can generate more easily the test digital signals. For example, using OrCAD you can not apply the dB(..) function after the FFT tool was used or you can have convergence problems for complex circuits and long run to time in transient analysis as in our case and the simulations are very slow. But despite of these disadvantages, a PSpice simulator using a model library supplied by an IC manufacturer is incontestably more accurate and gives us information which is closer to the behavior of a real converter. Moreover, in one single cited work [7], a comparison concerning SFDR between simulated (with Matlab) and measured results (for the implemented converter) was presented and the differences for several input frequencies were more than 6dB. All the cited works which presented implemented converters [6], [7] and [8] applied DEM techniques inside of an input segment of at most four bits while for example the theoretical work [7] 2 SFDR Considerations The most recent definition for the spurious-free dynamic range (SFDR) that we found in the field literature is this one: “For a pure sine wave input of specified amplitude and frequency, SFDR is the ratio of the amplitude of the DAC output averaged spectral component at the input frequency to the amplitude of the largest unwanted spectral component observed over a specified frequency band” [12]. Usually, the frequency band is limited to Nyquist frequency. As we can see, this definition requires the specification of amplitude and frequency of the input sine wave. This would mean that we can have different values of SFDR for different inputs and an intrinsic parameter of DAC would be signal dependent. Our experience says that a more correct definition would include “measured SFDR” instead of “SFDR” only. The real SFDR would be close to the maximum measured SFDR for a big number of tests with different inputs. We will explain this further on in this section. Three major factors influence the dynamic behavior of current-steering DACs: finite output impedance, physical position of the current cells (especially for high resolution DACs) and the mismatch between the unit current sources. The first two generate delay differences in the current sources switching and consequently glitches in the output voltage and spurs in the output spectrum. The finite output impedance and its influence depending on the output voltage was analyzed in detail in [3] and [13] and the problem was almost entirely solved by adopting the differential output [13], as we have already done in [9] and [11]. The influence of the second factor was also minutely analyzed in [5] and the authors elaborated a technique to homogenize the ISSN: 1790-5117 202 ISBN: 978-960-474-006-2 8th WSEAS International Conference on POWER SYSTEMS (PS 2008), Santander, Cantabria, Spain, September 23-25, 2008 random control bit PRB (AA4) can produce either a 16-b shift or it doesn’t shift the data, while the bit A4 of the input code can shift or not the data with one bit. Thus the total amount of shift is randomly selected between 0 and 31. The synchronization latch was used to equalize the delays between the four least significant bits and the pseudo-randomized bits. The unary current source block contains 63 identical cells and each cell has a current source and a switch. The low-gain opamp forces almost the same potential at the two outputs of the DAC and leads to a non-critical voltage step during the parasitic capacitance switching when the input code changes. Thus, the output glitches were almost suppressed and we do not need to put much effort on the design of the thermometer encoder anylonger. The bit line A4 will directly feed the output B0, the bit line A5 will directly feed the outputs B1 and B2, the bit line A6 will directly feed the outputs B3, B4, B5, B6 and so on. To simulate the DAC SFDR we used the OrCAD program and a library including PSpice models of the 0.35µm CMOS technology. Special parameters were included in the models of the pMOS transistors used as unary current sources which allowed the Monte Carlo expanded this segment up to six bits, but with an important area penalty. So, despite of an enlarged dynamic-matching possibility, we can conclude that for practical reasons the designers haven’t invested yet effort and time to extend the most significant segment of the DAC. The more complex the converter is, the PSpice analysis can fail. However this type of accurate simulation with low-level of idealization (i. e. OrCAD simulation) can have success for a 6-b DEM segment, but with simpler randomization circuitry, as we will see in the next section. 3 DAC Architecture. Simulation Results The block diagram of the proposed DEM DAC is presented in Fig.1. The output of the thermometer encoder is a 63-bit word corresponding to the 6-bit segment of the input code, but because the simulator doesn’t support a bus with more than 32 bits, we used two buses of 32 bits and 31 bits for the lower and the upper sections respectively. The barrel shifter has five layers and it is controlled in such a way that the pseudo Fig.1 DEM DAC architecture ISSN: 1790-5117 203 ISBN: 978-960-474-006-2 8th WSEAS International Conference on POWER SYSTEMS (PS 2008), Santander, Cantabria, Spain, September 23-25, 2008 analysis using these active components with stochastic matching. To obtain the test digital signal we used a fullscale sinusoid (Voff=256mV, Vamp=256mV and FREQ=8393Hz) and a 10-b ideal ADC. The sampling frequency was 100kHz. To avoid major convergence problems and to reduce the simulation duration we settled to 0 the four input bits of the least significant segment and we simulated the DAC without DEM and with DEM. As we can see in Fig.2 we gain 15dB in SFDR and thus our solution is confirmed. [4] D.H. Lee, Y.H. Lin and T.H. Kuo, Nyquist-Rate Current-Steering Digital-to-Analog Converters with Random Multiple Data-Weighted Averaging Technique and QN Rotated Walk Switching Scheme, IEEE Transactions on Circuits and Systems, Vol.53, No.11, November 2006, pp.1264-1268. [5] T. Chen, G. Gielen, The Analysis and Improvement of a Current-Steering DACs Dynamic SFDR - I: The Cell Dependent Delay Differences, IEEE Transactions on Circuits and Systems – I: Regular Papers, Vol.53, No.1, January 2006, pp.3-15. [6] K. O’Sullivan, C. Gorman, M. Hennessy, and V. Callaghan, A 12-bit 320-MSample/s CurrentSteering CMOS D/A Converter in 0.44 mm2, IEEE Journal of Solid-State Circuits, vol. 39, No.7 July 2004, pp.1064-1072. [7] N.U. Andersson, K.OAndersson, M. Vesterbacka, and J. J. Wikner, Models and implementation of a dynamic element matching DAC, Analog Integrated Circuits and Signal Processing, Vol. 34 , Issue 1, January 2003, pp. 7-16. [8] W. Chen, J. Bauwelinck, P. Ossieur, X.Z. Qiu and J. Vandewege, A Current-Steering DAC Architecture with Novel Switching Scheme for GPON BurstMode Laser Drivers, IEICE Trans. on Electronics, Vol. E90-C, No.4, April 2007, pp. 877-884. [9] M. Tomoroga, L. Jurca, M. Ciugudean and C. Toma, Low Glitch Current-Steering DAC with Split Input Code, Proceedings of the 6th WSEAS Int. Conf. on Electronics, Hardware, Wireless and Optical Communications, Corfu Island, Greece, February 16-19, 2007, pp. 40-45. [10] M. Tomoroga, L. Jurca, Study of Matching Errors in Unit Element Approach of Current-Steering Segmented DAC Design, Proceedings of the 6th WSEAS International Conference on System Science and Simulation in Engineering (ICOSSSE), Venice, Italy, November 21-2, 2007, pp.115-120. [11] M. Tomoroga, L. Jurca, M. Ciugudean, C. Toma, Low Voltage Low Glitch Current-Steering DAC Overlapping the Voltage Reference Circuit, WSEAS Transactions on Circuits and Systems, Issue 3, Vol.6, March, 2007, pp. 273-280. [12] E. Balestrieri, S. Rapuano, Defining DAC Performance in Frequency Domain, Measurement, No.40, 2007, pp.463-472. [13] T. Chen, G. Gielen, The Analysis and Improvement of a Current-Steering DACs Dynamic SFDR - II: The Output Dependent Delay Differences, IEEE Transactions on Circuits and Systems – I: Regular Papers, Vol.54, No.2, February 2007, pp.268-278. (a) (b) Fig.2 Simulated SFDR without DEM (a) and with DEM (b) 4 Conclusion In this paper we presented a simple DEM technique to reduce the influence of matching errors in the dynamic performance of a 10-b current-steering segmented DAC. The structure was optimized and the gain in the SFDR range was 15dB. References: [1] C. Volosencu, Identification in Sensor Networks, 9th WSEAS International Conference on Automation and Information (ICAI'08), Bucharest, Romania, 2008. [2] R. Jacob Baker, CMOS: Circuit Design, Layout, and Simulation, 2nd Edition, Wiley-IEEE Press, 2007. [3] J.J. Wikner, Studies on CMOS Digital-to-Analog Converters, Dissertation No.667, Linköping University, Sweden, 2001. ISSN: 1790-5117 204 ISBN: 978-960-474-006-2