A 14-bit MOS DAC with Current Sources free from Power-Line Voltage Drop and with Output Circuits free from Code-dependent Variable Time Constant Toru Sai and Yasuhiro Sugimoto* Department of Electrical, Electronic, and Communication Engineering, Chuo University 1-13-27, Kasuga, Bunkyo-ku, 112-8551, Tokyo, Japan E-mail: sai@comp.elect.chuo-u.ac.jp, *sugimoto@elect.chuo-u.ac.jp designs, all the output of the current sources was tied to one output terminal where a load resistor is connected. However, the number of currents that tie to the output terminal depends on the digital input code of the DAC. Therefore, the number of parasitic capacitances, each of which is associated with a specific current source and current switch connected to the output terminal of the DAC, changes depending on the digital input code. This is one of the major reasons that highfrequency reconstructed waveforms of DACs have large harmonic distortions [5]. In this paper, we have designed a DAC that realizes 14-bit resolution without calibrating the unit current sources in a matrix, and have also designed output circuits for the DAC that do not have a time-constant change even when the digital input code changes. Abstract-A 14-bit MOS DAC is described that has current sources that are free from the non-linear current mismatch caused by ground-line voltage drop and output circuits that do not suffer from time-constant change at the output terminal when the input digital code changes. Base current sources are locally classified into two groups with two different values, and the unit current source is constructed by adding one base current source from one group and another base current source from the other group. Eight buffer amplifiers are distributed between the output terminal and current switches to stabilize the voltage at the node where several outputs of unit current sources are tied together and to eliminate the influence of stray capacitances associated with unit current sources and current switches. A 14-bit, 3.3-V DAC was fabricated using 0.35-um CMOS devices. The results show that the DNL and INL are from +0.7 to -0.75 LSB and +1.5 to -1 dB, respectively, and that the SFDR for 2.5-MHz and 10-MHz reconstructed signal waveforms were 77 dB and 67 dB, respectively, with a 50-MHz clock. The current consumption was 25 mA. II. THE 14-BIT DAC ARCHITECTURE The developed DAC adopts segmentation and R-2R ladder architecture to realize a 14-bit resolution. A block diagram is shown in Figure 1. I. INTRODUCTION In order to realize a highly accurate MOS D-to-A converter (DAC) such as one with 14-bit resolution, precise current matching among MOS current sources is needed [1]. Current sources are commonly segmented and placed in a matrix form [2] in such a high-resolution DAC, and the foreground calibration method [4], [5] has been used for MSB current sources to reduce mismatches; however, foreground calibration is inconvenient to manufacturers, and it should be avoided. Segmentation means that current sources occupy a large area on an LSI chip. The ground line, as a power line for NMOS current source transistors, should go through each current source to obtain the current which flows out of the current source; as a result, a voltage drop along the ground line is observed. This voltage drop causes a non-linear mismatch in the current of the unit current source. This problem needs to be solved if calibration is not desired. Moreover, in a highly accurate DAC, the frequency characteristics in the Spurious-Free-Dynamic-Range (SFDR) of the output signal must also be investigated. In previous 978-1-4244-3896-9/09/$25.00 ©2009 IEEE Fig. 1. Block diagrams of the designed 14-bit DAC. 49 Given the adopted segmentation, the current sources and current switches for the upper 6 bits are placed in the matrix form as shown in Figure 1. Sixty-three unit current sources are used to form 63 steps for the upper 6 bits from Bit1(=MSB) to Bit6. The lower 8 bits are constructed using 8 current sources, current switches and R-2R ladder resistors. The output circuit is prepared between the “DA-out” or “DAout/” terminal and the current switches. The outputs of two rows of unit current sources-that is, 16 unit current sourcesare connected into one and lead to one output circuit. Due to the differential configuration, in reality, two output circuits are needed for every 16 unit current sources. The output circuit stabilizes its input voltage so that the waveforms at the “DA-out” or “DA-out/” terminal need not see a stray capacitance change. Fig. 2. Non-linear current mismatch along the ground line. III. ELIMINATING CURRENT MISMATCH DUE TO GROUND LINE VOLTAGE DROP As we cannot have an equal length of ground line from the ground pads for each unit current source, the voltage at the ground line for each unit current source may differ due to the line resistance RG. Suppose that 8 unit current sources in the current source matrix are placed in the configuration shown in Figure 2, with the ground line connecting them and the ground terminal at one end, and that they are numbered 1, 9, 17, 25, 33, 41, 49, and 57. In this design, each current source consists of one NMOS transistor, and its gate terminal is connected to the bias circuit, which is common to all the current sources in the current source matrix. Then, the ground line voltage on the left side of the current source becomes higher than that on the right side of the current source. As a result, the current values of the unit current sources become non-linearly distributed along the positions as shown in Figure 2. In order to eliminate the current error caused by the ground line voltage drop, a new structure for the current source array is proposed. Figure 3 shows a new layout, and the connection structure for the current sources is shown in Figure 2. In Figure 3, a unit current source consists of two base current sources. For example, unit current source 57 consists of the parallel connection of base current sources 57a and 57b. Each unit current source was divided into two base current sources, which were labeled “a” and “b.” In Figure 3, column “a” has 8 base current sources, and they are further split into two subgroups: 1a, 9a, 17a, 25a and 33a, 41a, 49a, 57a. The current values of 1a, 9a, 17a, and 25a are equal because the ground lines are laid out in equal length using the first and second metal layers as shown in Figure 3. Point B, the origin point, is on the third metal layer and is common to 4 base current sources: 1a, 9a, 17a, and 25a. The same discussion holds for each subgroup of the base current sources. The current values for base current sources 33a, 41a, 49a, and 57a are equal with respect to the origin point A; the current values for base current sources 1b, 9b, 17b, and 25b are equal with respect to the origin point C; and the current values for base current sources 33b, 41b, 49b, and Fig. 3. New structure to realize unit current sources. 57b are equal with respect to the origin point D. The thirdlayer metals connect those origin points. As the third metal is connected to the ground pads of the chip, we assumed that the voltages at one end of the two third-layer metals in Figure 3 are the same. Then, the voltages at origin points A and C become equal, and the voltages at origins B and D become equal. Although the voltages are different between A and B, or C and D, all the unit current sources come to have equal current values when one unit current source is constructed by the parallel connection of one base current source from subgroup A and another from subgroup D, or one from subgroup B and another from subgroup C, etc., as shown in Figure 3. IV. ELIMINATING THE TIME-CONSTANT CHANGE AT THE OUTPUT TERMINAL A DAC needs to output analog signals without distorting the waveforms. In reality, however, the high-frequency signal is greatly distorted. In order to take this characteristic into account, the SFDR was observed. Harmonics are major components of spurs that deteriorate the SFDR. One of the major causes of distortion is considered to be stray capacitances associated with unit current sources and current switches [3]. However, the analyses in reference 3 were based on a small-signal equivalent circuit. We took the dynamic operation into account, that is, the time-constant change at the output terminal depending on the digital input code. This is illustrated in Figure 4. Figure 4 shows the conventional DAC circuit with stray capacitors associated with current sources and current switches to realize the upper bits. Depending on the digital input code, switches from SW1 to SWn turn on and off and the amount of current proportional to 50 the value that the input digital code represents flows into the load resistor RL, resulting in the analog output voltage. However, the stray capacitors (Cs) become connected and disconnected as the switch turns on and off, and the time constant at the output terminal (Signal Out in Figure 4) changes depending on how many switches turn on. This phenomenon was SPICE-simulated and is shown as a curve labeled “conventional” in Figure 5. The curve indicates that the time constant for the rising interval changes as the output voltage changes. As the number of current sources is tied to the output terminal when the output voltage becomes low, that is, when the MSB of the input digital code is high, the time constant becomes large. This result is coincident with the fact that a number of switches turn on and a number of stray capacitors are tied to the output terminal. In order to avoid this situation, the buffered cascode circuit, which consists of an amplifier A, a MOS transistor M1, and the bias voltage V3 in Figure 6, is placed between the output terminal and current switches. The impedance at the minus input terminal of amplifier A-that is, the point where all the currents are combined-becomes low. This means there is no signal swing at the point even when a large analog signal is output at the output terminal. The time-constant change at the output terminal is eliminated. This characteristic is shown as the curve labeled “with gain-boosted cascode circuit” in Figure 5. The impedance at the minus input terminal of amplifier A-that is, the point where all the currents are combined-becomes low. The time-constant change for the rise time of the output waveform has been eliminated. SFDR frequency characteristics were further simulated by using 0.35-um CMOS device parameters, as shown in Figure 7. The SFDR at high frequencies is expected to be improved greatly when the time-constant change is suppressed by using the circuit in Figure 6. V. DESIGN PARAMETERS OF THE CIRCUIT The design parameters of the circuit in Figure 6 are described. The current of the unit current source in the matrix is chosen 185 uA and about 12 mA flows in the 50 Ω load RL, resulting in 600 mVp-p output at the full-scale signal level. The circuit is differentially configured. The stuck-up elements from the ground to 3.3 V supply (VCC) are a current source transistor, a current switch, a buffered cascode circuit and a load resistor. The size, gm and the output conductance of the MSB switch are 16u/0.5u, 1.4 mS and 52 uS, respectively. Its source capacitance is 0.6 pF, and this value is relatively large because a cascode transistor for the current source is not used for the reason of testing the low-voltage operation of this current source part. The voltage gain of amplifier A is 26 dB; its -3 dB frequency bandwidth is 220 MHz and consumes 0.8 mA. In place of a cascode transistor for each current source, only the 8 buffered cascode amplifiers are needed. Fig. 4. Conventional DAC circuit with stray capacitors for realization of upper bits. Fig. 6. Modified DAC output circuit in which the time-constant variation is eliminated. Fig. 7. Simulated SFDR characteristics of the circuits in Fig. 4 and Fig. 6. Fig. 5. The influence of stray capacitors. 51 The advantage is that the influence of the source capacitance of the current switch is reduced. To guarantee the operation of the buffered cascade circuit when all switches turns off, 0.6 mA of constant current source is connected to the source of M1 in Figure 6. The supply voltage of the logic circuit is 2 V, and the control signal for current switches is 2 V and 0 V. Transistors in the current switch remain in a state of saturation. VI. CONCLUSION A new current source matrix which realizes high accuracy without calibration was designed. An output circuit without time-constant change due to the digital input code-dependent stray capacitance change was also designed. The use of a 0.35 μm CMOS chip verified the effectiveness of the designs. REFERENCES V. EXPERIMENTAL RESULTS [1] G.A.M.Van der Plas, J.Vandenbussche, W.Sansen, M.S.J.Steyaert and 2 G.G.E.Gielen , “A 14-bit intrinsic accuracy Q random walk CMOS In order to determine the effects of the new current source matrix and the distortion-free output circuit, a 14-bit DAC was designed and fabricated using 0.35-um CMOS devices. Although we could observe its operation up to 200 MS/s in a SPICE simulation, the actual chip is subject to a great deal of digital noise generated by the large-size devices used, and relatively good performance has been obtained only in the case up to 50 MS/s although it operates at a speed up to 100 MS/s. Figure 8 shows the measurement results of DNL and INL at the 14-bit level. The DNL ranged within ±0.7 LSB and the INL from +1.5 to -1.0 LSB when the clock was 50 MHz. A relatively good INL performance was obtained. This may be the result of having good matching in the unit current sources. Figure 9 shows the frequency spectrum of the 2.5 MHz full-scale signal output with a clock speed of 50 MHz. The second harmonic was the largest; however, it was about the same as the noise level. Figure 10 shows the frequency characteristics of the measured SFDR, the second (white circle) and the third (black triangle) harmonics for the fullscale output signal with a clock speed of 50 MHz. The SPICE simulation result in SFDR with a 200 MHz clock is superimposed on the graph. The slope is similar to up to onefifth of the clock speed. However, a small amount of degradation is observed when the reconstructed signal is more than 10 MHz. This might be the reason that the DAC adopts R-2R ladders for the lower 8-bit realization. The measured second and third harmonics levels are almost the same although the differential outputs are taken. These points should be investigated further. Table 1 summarizes the IC performance. DAC”, IEEE JSSC, vol.SC-34, no.12, pp.1708-1718, December 1999. [2] T.Miki, , “An 80-MHz 8-bit CMOS D/A converter”, IEEE JSSC, vol.SC21, no.6, pp.983-988, December 1986. [3] A.V.Bosch, M.A.F.Borremans, M.S.J.Steyaert, and W.Sansen, “A 10-bit 1GSample/s nyquist current steering CMOS D/A converter”, IEEE JSSC, vol.SC-36, no.3, pp.315-324, March 2001. [4] W.Schofield, D.Mercer, L.St.Onge, “A 16b 400MS/s DAC with <-80dBc IMD to 300MHz and <-160dBm/Hz Noise Power Spectral Density”, ISSCC 2003, paper No. 7.1, pp.126-127, February 2003. [5] D.A.Mercer, “Low-Power Approaches to High-Speed Current-Steering Digital-to-Analog Converters in 0.18-um CMOS”, IEEE JSSC, vol.SC42, no.8, pp.1688-1698, August 2007. (a) DNL (b) INL Fig. 8. Measured DNL and INL of a 14-bit DAC Table 1. Overall IC performance. Fig. 9. Spectrum of the reconstructed 2.5-MHz analog output signal with a 50-MHz clock. Fig. 10. SFDR frequency characteristics of the reconstructed analog output signal with a 50-MHz clock. 52