Professional Documents
Culture Documents
springerDAC10b
springerDAC10b
net/publication/225440755
CITATIONS READS
20 7,100
2 authors, including:
Peiman Aliparast
Aerospace Research Institute of Iran
25 PUBLICATIONS 116 CITATIONS
SEE PROFILE
All content following this page was uploaded by Peiman Aliparast on 30 April 2016.
Received: 8 August 2008 / Revised: 12 January 2009 / Accepted: 21 January 2009 / Published online: 20 March 2009
Springer Science+Business Media, LLC 2009
Abstract In this paper a 10-bit 1.2-GSample/s Nyquist integration of digital and analog systems on one chip.
current-steering CMOS digital-to-analog converter (DAC) Among available DAC structures like R-string, RC-hybrid
is presented. Segmentation (90%) has been used to get the and other structures, current-steering DAC’s are preferred
best DNL and reduce glitch energy. This segmentation ratio because of their lower area, power consumption and com-
guarantees the monotonicity. Higher performance is patibility with digital CMOS process [1]. Current-steering
achieved using a novel 3-D thermometer decoding method DACs are based on a array of matched current sources which
which reduces the area, power consumption, and the number are unity decoded or binary weighted [2]. Figure 1 shows a
of control signals of the digital section. Simulation results typical block diagram of an n-bit segmented current-steering
show that the spurious-free-dynamic-range (SFDR) in DAC which uses the advantages of both architectures. Input
Nyquist rate is better than 65 dB for sampling frequency up word is segmented between b less significant bits that switch
to 1.2-GSample/s. The analog voltage supply is 3.3 V while a binary weighted array and m = n - b most significant bits
the digital part of the chip operates at only 2.4 V. Total that control switching of a unary current source array. The m
power consumption in Nyquist rate measurement is input bits are thermometer decoded to switch individually
149 mW. The chip has been processed in a standard 0.35 lm each of the unary sources [3, 4, 5]. A dummy decoder is
CMOS technology. Active area of chip is 1.97 mm2. placed in the binary weighted input path to equalize the
delay. A latch is placed just before the switch transistors of
Keywords Current steering DAC each current source to minimize any timing error [6].
3-D thermometer decoding High accuracy The performance of the DAC is specified through static
Integrated circuits CMOS parameters: Integral Non-Linearity (INL), Differential Non-
Linearity (DNL) and parametric yield; and dynamic
parameters: glitch energy, settling time and SFDR [7]. Static
1 Introduction performance is mainly dominated by systematic and random
errors. Systematic errors caused by process, temperature and
Modern telecommunication systems, e.g. GSM and HDTV
electrical slow variation gradients are almost cancelled by
require high-speed, high-accuracy DAC’s to enable proper layout techniques [8]. Random errors are determined
P. Aliparast (&) solely by mismatch due to fast variation gradients. The
Islamic AZAD University of Sofian, Sofian, Iran design of current-steering DAC starts with an architectural
e-mail: p.aliparast@sofianiau.ac.ir; selection to find the optimum segmentation ratio (m over n)
paymun.aliparast@gmail.com
that minimizes the overall digital and analog area [9, 6, 10].
P. Aliparast The INL is independent of the segmentation ratio and
University of Tabriz, Tabriz, Iran depends only on the mismatch if the output impedance is
e-mail: p-aliparast@tabrizu.ac.ir made large enough [2]. The DNL specification depends on
the segmentation ratio but it is always satisfied provided that
N. Nasirzadeh
SERAJ Institute of Higher Education, Tabriz, Iran the INL is below 0.5 LSB for reasonable segmentation
e-mail: nnasirzadeh@seraj.ac.ir ratios. The glitch energy is determined by the number of
123
196 Analog Integr Circ Sig Process (2009) 60:195–204
– Negative
• Digital decoding with more area and power
consumption
• Increased number of control signals
Binary:
– Positive
• Low digital power consumption
• Small number of control signals
binary bits b, being the optimum architecture in this sense a • Monotonicity not guaranteed
totally unary DAC. However, this is unfeasible in practice • Larger DNL errors
due to the large area and delay that the thermometer decoder • Large glitch energy
would exhibit. The minimization of the glitch energy is then
done in circuit level design and layout of the switch and latch
Figure 3 also summarizes these points graphically.
array and current source cell [11].
Usually, to leverage the clear advantages of the thermo-
meter-coded architecture and to obtain a small area
simultaneously, a compromise is found by using segmen-
2 Binary weighted architecture versus unary decoded
tation [6]. The DAC is divided into two sub-DACs, one for
architecture
the MSBs and one for the LSBs. Thermometer coding is
used in the MSB where the accuracy is needed most.
Current-steering architectures differ from current dividing
Because of the reduced number of bits in this section, the
architectures in that they replicate a reference current
size is considerably smaller than a true thermometer coded
source rather than divide it. As shown in Fig. 2, the ref-
design. The LSB section can either be done using the bin-
erence source is simply replicated in each branch of the
ary-weighted or the thermometer-coded approach. We will
DAC, and each branch current is switched on or off based
refer to a fully binary-weighted design as 0% segmented,
on the input code. For the binary version, the reference
whereas a fully thermometer-coded design is referred to as
current is multiplied by a power of two, creating larger
100% segmented. For conventional designs, optimum seg-
currents to represent higher-magnitude digital signals. In
mentation is 75% [6]. We have used 90% segmentation to
the unit-element version, each current branch produces an
achieve the best performance in high-speed design. Using a
equal amount of current, and thus 2 N current source ele-
novel 3D decoding method we have improved digital area
ments are needed. Advantages and disadvantages of these
and power consumption which allows our 90% segmenta-
structures are summarized below:
tion to be as optimum as other designs.
Thermometer:
3 Novel thermometer decoding architecture
– Positive
• Low glitch energy Figure 4 shows a block diagram of a conventional row and
• Monotonicity column decoded 10-bit current-steering DAC. In this block
• Small DNL errors diagram, the lowest significant bit is applied to a dummy
123
Analog Integr Circ Sig Process (2009) 60:195–204 197
decoder [12]. This decoder creates a delay proportional to 15-output Binary-to-Thermometer Decoder. Output of these
the Binary-to-Thermometer decoder and causes the signal decoders controls 511 current cells in the main matrix. But if
to arrive at the switches synchronously. The five LSB bits we think about Binary-to-Thermometer Decoder structure
are column decoded and the four MSB bits are row deco- we understand that with b-bit increase of the input of the
ded. Column decoder is a 5-input 31-output Binary- decoder, area, complexity, number of control signal and
to-Thermometer Decoder and row decoder is a 4-input power consumption of the decoder increase with 2b. In fact
123
198 Analog Integr Circ Sig Process (2009) 60:195–204
power and area are doubled with only one bit increase in the Að4 to 15 BTDÞ ¼ 2 Að3 to 7 BTDÞ ð8Þ
input of the decoder and we can write:
þAð5 to 31 BTDÞ ¼ 4 Að3 to 7 BTDÞ ð9Þ
Pð4 to 15BTDÞ ¼ 2 Pð3 to 7BTDÞ ð1Þ
then:
Að4 to 15 BTDÞ ¼ 2 Að3 to 7 BTDÞ ð2Þ
Að4 to 15 BTDÞ þ Að5 to 31 BTDÞ ¼ 6 Að3 to 7 BTDÞ
thus: ð10Þ
Pð5 to 31 BTDÞ ¼ 4 Pð3 to 7 BTDÞ ð3Þ
In this structure three LSB bits are column decoded,
Að5 to 31 BTDÞ ¼ 4 Að3 to 7 BTDÞ ð4Þ three middle bits are row decoded and three MSB bits are
where BTD is Binary-to-Thermometer Decoder, P is the height decoded. On the other hand, we have only used 21
power consumption of the decoder and A is active area that control signals instead of 46 control signals thus the
the decoder uses. Now consider Fig. 5 that shows a 3-D number of control signals has been decreased by 55 percent
decoding architecture. In this block diagram three BTD so we can achieve the best speed and performance.
have been used. Three bits for height, three bits for row and
three bits for column and every cell is selected with three
parameters (R, C and H). In fact we have only used three (3 4 The current cell, latch and driver
to 7 BTD) instead of two (5 to 31 BTD) and (4 to 15 BTD)
thus power consumption and area of the circuit have been Static and dynamic performance of current-steering DACs
improved two times because: is mostly determined by the accuracy of the current sour-
ces, finite output impedance, and switching time. Figure 6
Pð4 to 15 BTDÞ ¼ 2 Pð3 to 7 BTDÞ ð5Þ
shows a current source transistor MCS, an additional cas-
þPð5 to 31 BTDÞ ¼ 4 Pð3 to 7 BTDÞ ð6Þ code transistor MCAS that increases the output impedance
and two complementary switch transistors MSW.
then:
Figure 6(a) shows cascode current source and switch
Pð4 to 15 BTDÞ þ Pð5 to 31 BTDÞ ¼ 6 Pð3 to 7 BTDÞ structure for 1LSB while Fig. 6(b) shows the same struc-
ð7Þ ture for two LSB. In proposed 10-bit DAC one bit is binary
And for the area we have: weighted so it uses the current source of Fig. 6(a) and
123
Analog Integr Circ Sig Process (2009) 60:195–204 199
123
200 Analog Integr Circ Sig Process (2009) 60:195–204
and use only mismatch Eq. 14 to reach a minimum sizing of crossing point are fed from the left and SRD makes a
current cell. With this method, the speed of switch is high non-symmetrical crossing point which reduces the spike at
also INL \0.5 LSB is satisfied. node VX considerably. In SRD circuit, MSRD1 is always on
The small-signal output impedance for the current and when MSRD2 is off, VgSW approaches 2.4 V (power
source topology of Fig. 6(a) is given by: supply value of digital part). When MSRD2 is on with proper
Rout gmSW gmCAS rdsSW rdsCAS rdsCS ð18Þ sizing of MSRD2 ; VgSW can be set to desired value because
VgSW in this case will be equal to VSG of MSRD2 transistors.
The optimum MSW and MCAS gate bias voltages concerning In this circuit for complete switching of MSW transistors we
the output impedance are found by differentiating Rout with need 350 mV differential voltage, so VSG of MSRD2 is set to
respect to VgSW and VgCAS . For the SW and CAS gate bias 2.05 V. On the other hand for non-symmetric crossing its
voltages that maximize output impedance are found as: enough to choose bigger size for MSRD1 than MSRD2. Size of
1 MSRD1 and MSRD2 has been given in Table 1, also SRD
VgCAS ¼ VT þ ðVomin þ 2MVCS þ 2MVCAS MVSW Þ ð19Þ
3 output wave forms and its effect in reducing spike in node
1 VX is shown in Fig. 8(d). The capacitive coupling to the
VgSW ¼ VT þ ð2Vomin þ MVCS þ MVCAS þ MVSW Þ ð20Þ analog output is minimized by limiting the amplitude of the
3
control signals just high enough to switch the tail current
Figure 7 shows the biasing scheme for the cascoded completely to the desired output branch of the differential
current sources. The PMOS sections of the biasing circuits pair. In addition the switch transistors are kept relatively
are labelled as Global biasing while the NMOS sections are small in order to avoid large parasitic capacitances.
labelled as Local biasing. In the actual implementation, the
global biasing is realized using a common-centroid layout
to reduce effects of gradients. The local biasing is separated 5 A few techniques to achieve very high speed
into four quadrants. There is no direct connection between
any two quadrants. This will improve both DNL as well as Clock distribution for 1.2 GHz is very difficult and getting
INL performance [6]. A driver circuit with a reduced swing data in this speed is very hard thus we have used two
placed between the latch and the switch reduces the clock channels for digital section. Every channel works at
feedthrough to the output node as well [14, 15]. Figure 8(a) 600 MHz and then results of two channels are combined at
shows a current source, switch, latch and driver cell. A new the input of the switch to get 1.2 GHz. Figure 9 shows the
swing-reduced-device (SRD) circuit is designed (shown in structure used for digital section of the DAC. Channel 1
Fig. 8(b)). The latch circuit complementary output levels samples input data with clock and channel two samples
and non-symmetrical crossing point are designed to input data with clock-not. A buffer just before switch
minimize glitches [11]. The waveforms of the different combines the output of two digital channels. It sends the
nodes are shown in Fig. 8(c) without SRD circuit and output of digital channel1 with clock and the output of
Fig. 8(d) with SRD circuit. Signals with symmetrical digital channel two with clock-not to the input of switch. In
123
Analog Integr Circ Sig Process (2009) 60:195–204 201
Fig. 8 Non-symmetrical crossing point reduces current source drain spike and clock feedthrough scheme, (a) Current source, switch, latch and
driver cell, (b) SRD circuit, (c) drain spike and driver voltages without SRD circuit, and (d) drain spike and driver voltages with SRD circuit
Table 1 Current source and SRD transistors dimensions and currents fact in one period of clock we take two samples of the input
code and in the output it seems that the circuit works at
Transistor Size ID
1.2 GHz. On other hand, we use master-slave operation in
MCS W = 2 lm, L = 10 lm 10 lA all digital circuits and use pipelining scheme, so in overall
MCAS W = 2 lm, L = 0.35 lm 10 lA the digital circuit only senses one gate-delay. For example
MSW W = 0.5 lm, L = 0.35 lm – the structure of one of 3-input 7-output Binary-to-Ther-
MSRD1 W = 1.5 lm, L = 2 lm – mometer Decoder has been shown in Fig. 10. Layout of all
MSRD2 W = 1 lm, L = 2 lm – digital section has been done manually to guarantee the
best speed, low power and minimum area.
123
202 Analog Integr Circ Sig Process (2009) 60:195–204
123
Analog Integr Circ Sig Process (2009) 60:195–204 203
Resolution 10 bit
Update rate Up to 1.2 GS/s
Max. output swim 1Vpp diff.
DNL Better than 0.19LSB
INL Better than 0.37LSB
SFDR (589 MHz@1.2 GS/s) 65 dB
Analog Power consumpion (at 1.2 GS/s) 66 mW (20 mA from 3.3 V)
Digital Power consumpion (at 1.2 GS/s) 83 mW (34.5 mA from 2.4 V)
Total Power consumpion (at 1.2 GS/s) 149 mW
Analog/Digital voltage supply 3.3 V/2.4 V
Active area 1850 lm 9 1070 lm
123
204 Analog Integr Circ Sig Process (2009) 60:195–204
presented. A novel 3-D thermometer decoding scheme has 10. Gonzalez, J. L., & Alarcon, E. (2001). Clock-jitter induced dis-
been used in digital section which reduces the area power tortion in high-speed CMOS switched-current segmented digital-
to-analog converters. In Proceedings of International Symposium
consumption and number of control signals of this part, on Circuits and Systems (ISCAS’01), May 2001, pp. 1512–1515.
considerably. Simulations have been performed to analyze 11. Bastos, J., et al. (1996). A high yield 12-bit 250-MS/s CMOS
and solve some of important dynamic linearity limitations. D/A converter. In Proceedings of IEEE Custom Integrated
Using two digital channels that work in parallel, one Circuits Conference (CICC), May 1996, pp. 431–434.
12. Bosch, A. V., et al. (2001). A 10-bit 1-GS/s Nyquist current-
operating with clock and the other operating with clock-not steering CMOS D/A converter. IEEE Journal of Solid-State
enables the sampling rate to be 1.2 GS/s while each Circuits, 36(3), 315–324.
channel operates only at 600 MHz. This clocking strategy 13. Pelgrom, M. J. M., et al. (1989). Matching properties of MOS
makes clock distribution much easier. Analog switches and transistors. IEEE Journal of Solid-State Circuits, 24(5), 1433–
1440.
SRD circuits have been optimized not only to get minimum 14. Khono, H., et al. (1995). A 350-MS/s 3.3-v 8-bit CMOS D/A
area and maximum speed but also to improve dynamic converter using a delayed driving scheme. In Proceedings of
behavior of the DAC. Segmentation (90%) decreases DNL CICC. IEEE Custom Integrated Circuits Conference, pp. 211–214.
error and glith energy considerably and guarantees needed 15. Luh, L., et al. (2000). A high-speed fully differential current
switch. IEEE Transactions on Circuits and Systems-II: Analog
improvement of SFDR. Separate power supplies have been and Digital Signal Processing, 47(4), 358–363.
used for digital and analog parts. Digital section operates at
lower supply voltage than analog part. This increases speed
and reduces power consumption of the digital part and at Peiman Aliparast was born in
the same time decreases power supply noise and improve Tabriz, Iran. He received B.Sc.
degree from Tabriz Islamic
the performance of the analog part. The technology used is AZAD University, Tabriz, Iran,
a 0.35 lm, single-poly four-metal, 3.3 V, standard TSMC in 2005 and M.Sc. degree from
Mixed Mode CMOS process. The active area of the DAC, Urmia University, Urmia, Iran,
as shown in Fig. 11, is 1850 lm 9 1070 lm. in 2007, both in electronics
engineering. He is currently
electronics PhD student in Tab-
Acknowledgments The authors would like to thank the referees for riz University. His research
their helpful suggestions and comments. interests are analog and digital
integrated circuit design for
fuzzy and neural network
References applications, analog integrated
filter design and high-speed high-resolution digital to analog con-
1. Johns, D., & Martin, K. (1997). Analog integrated circuit design verter. He is currently with Islamic AZAD University of Soofian,
(pp. 445–484). USA: Wiley & Sons. ISBN:0-471-14448-7. Soofian, Iran, and Nanotechnology Research Laboratory in Tabriz
2. Razavi, B. (1995). Principles of data conversion systems University, Tabriz, Iran. He is also staff of Microelectronics Research
(pp. 45–94). New Jersey: IEEE Press. Laboratory in Urmia University, Urmia, Iran.
3. Sumanen, L., et al. (1998). A 10-bit high-speed low-power
CMOS D/A converter in 0.2 mm2. In Proceedings of ICECS.
IEEE International Conference on Circuit and Systems, Vol. 1,
pp. 15–18. Nasser Nasirzadeh received
4. Nakamura, Y., et al. (1991). A 10-b 70-MS/s CMOS D/A con- B.S. degree in 2001 at Tabriz
verter. IEEE Journal of Solid-State Circuits, 26(4), 637–642. University and M.S. Degree in
5. Albiol, M., et al. (2004). Mismatch and dynamic modelling of 2005 at Urmia University both
current sources in current-steering CMOS D/A converters: An in Electronics Engineering.
extended design procedure. IEEE Transaction on Circuits and While pursuing his M.S. degree
Systems -I: Regular Papers, 51(1), 159–169. he worked for Urmia Micro-
6. Lin, C. H., & Bult, K. (1998). A 10-bit, 500-MS/s CMOS DAC in electronics Research Center and
0.6 mm2. IEEE Journal of Solid-State Circuits, 33(2),1948–1958. Urmia MEMS Lab. both at Ur-
7. Hendriks, P. (1997). Specifying communication DACs. IEEE mia University. Currently, he is
Spectrum, 34(7), 58–69. a teaching staff of the Electrical
8. Cong, Y., & Geiger, R. L. (2000). Switching sequence optimi- and Computer Engineering
zation for gradient error compensation in thermometer-decoded Department at the Seraj Institute
DAC arrays. IEEE Transaction on Circuits and Systems-II, 47(7), of Higher Education, Tabriz,
585–595. Iran. His research interests are high-speed CMOS integrated circuits
9. Vandenbussche, J., et al. (2001). Systematic design of high- and MEMS.
accuracy current-steering D/A converter macrocell for integrated
VLSI systems. IEEE Transaction on Circuits and Systems-II,
48(3), 300–309.
123