Professional Documents
Culture Documents
kchuang3@ece.gatech.edu
2
david.yeh@gatech.edu
I. INTRODUCTION
The rapid development of modern communication systems
such as the 60GHz wireless devices and other wireless
products [1] with ultra-high data transmission rates drive the
demand of data converters with ever increasing sampling rates.
In order to simultaneously meet the ultra-low power
requirement for portable electronic devices [2] and extremely
high sampling rate, it is absolutely necessary to simplify the
architecture of such ADCs as much as possible by excluding
all non-essential blocks that increase power consumption and
limit the sampling speed.
ADCs range from the oversampling delta-sigma, to
medium-speed successive approximation and to the highest
sampling speed, flash. Among all of the aforementioned types
of converters, flash converters are the most suitable based on
the requirement for 60GHz wireless applications. Moreover,
flash ADCs guarantee the smallest number of clock cycles per
conversion; and thus minimum latency. This is especially
critical for mixed-signal single carrier phase recovery systems.
The performance of ADCs can be improved by means of
either analog or digital calibrations. Calibration techniques
improve the overall performance at the expense of increased
circuitry complexity, but often degrade the inter-stage
bandwidth, linearity and gain due to the presence of switches
or series-input capacitors [3]. One offline calibration scheme
proposed in [4] requires more than twice the area occupied by
the ADC core. In a highly integrated wireless radio for mobile
applications, the power consumption and area available for
Digital Signal
Processor
data converters are extremely limited. Fig. 1 shows the mixedsignal demodulator integrated for 60GHz applications
requiring low-power, high-speed compact ADCs for digital
signal processing [5]. Fig. 2 shows the potential application
for portable UWB receiver.
The architecture of the ADC is discussed in section II. High
speed optimizations are explained in section III. The
experimental results are demonstrated in section IV followed
by a conclusion in section V.
II. ARCHITECTURE
Typically in the design of ADC, track-and-hold (T/H)
circuit is the first component that provides instantaneous
sampling of the input signal and holds its output at the
sampled level for the rest of the clock period. Charge
redistribution from the nonlinear parasitic capacitance and the
sampling capacitor indicate distortion while this output slowly
settles to its final state. In order to avoid this distortion, T/H
circuit must provide high bandwidth, which requires higher
power consumption. However, mobile applications targeted in
this work require extremely low power consumption; hence in
the proposed ADC, T/H circuit is excluded. Consequently, the
design of high speed comparator becomes critical and requires
optimization. By placing two converters in parallel and
218
Digital back-end
6th
5th
0th
Za[2:0]
Zb[2:0]
2:1 MUX
Z[2:0]
Vin
CLK
phase (clk is high), the gain and the total input noise without
considering flicker noise are derived in (1) and (2)
respectively, where coefficient is a constant depending on
the technologies.
clk
clkb
W5
Ron ,9
W3
Av v
4 2 g m 7 Ron ,9
g m1
interleaving them (in Fig. 3), the clock frequency for each
converter can be reduced while maintaining the same system
throughput.
III. HIGH SPEED OPTIMIZATIONS
Unlike the conventional architecture, track-and-hold circuit
and difference amplifiers are removed to enable ultra lowpower processing [6]-[8]. The block diagram in Fig. 4 consists
of 2N-1 taps of reference voltages, high speed latched
comparators, buffers and digital back-end.
A. Optimization of Comparator
Two practical issues in high-speed comparator design are:
technology dependent gain-bandwidth product (GBW)
constant of an amplifier and the poor isolation between the
latch output and the input of the difference amplifier through
the drain-gate capacitances of input transistors, inducing
strong kick-back noise to the preceding stages. Unlike the
switched-cascode comparator in [8], Fig. 5 shows the currentmode latched comparator implemented for the low-power
ADC operation that can solve these two problems. First, the
bandwidth requirement for high speed operation during
comparator
amplification
phase
is
defined
as
BW INTER ! (n 1) ln(2) f s / 2S assuming half of the sample
period is used for settling with n-bit resolution [9]. With n = 3,
the minimum bandwidth for settling is: BW > 0.88fNyquist for
time interleaving.
Secondly, the input differential pairs before the latch has
the advantage of reducing the input offset voltage of the latch
and the gain is also enhanced from the current mirroring. The
overload recovery time is determined by M9, the output
parasitic capacitance and the transconductance of M7-M8. The
comparator is optimized based on the trade-offs of gain and
total input-referred thermal noise. During the amplification
Vn2,in
'f
v 16kT (
J
g m1
(1)
g m3J
g
2
m1
2 g m7 Ron,9
g m5J
)
W
W
g m21 ( 5 ) 2 g m21 ( 5 ) 2 Ron,9
W3
W3
(2)
219
1
t CLK
2 (t latch
1
t buf t su )
(3)
25
SFDR (dB)
20
0.50
0.25
0.25
INL (LSB)
DNL (LSB)
0.00
-0.25
-0.50
2
Output Code
15
10
0.00
-0.25
5
0.1
-0.50
0
5GS/s
3GS/s
1
Input Frequency (GHz)
10
Output Code
11). From both plots, the error due to the INL is estimated to
be less than 1-b LSB since the fundamental tone is at least 6ndB (in our case 18dB) above all the individual harmonics and
distortions, where n is the theoretical number of bits [10].
The dynamic performance is characterized in Fig. 12 and
13. This ADC achieves spurious free dynamic range (SFDR)
220
16
SNDR (dB)
3GS/s
5GS/s
12
4
0.1
V. CONCLUSION
A time-interleaved high-speed 3-bit flash ADC in 90nm
CMOS is presented. In order to achieve excellent power
efficiency, T/H and difference amplifiers are removed and
only the very essential blocks are retained. Thorough
optimizations are analyzed and the results demonstrate an
ultra low-power (3mW, 3.9mW) and high-speed (3GS/s,
5GS/s) performance. This represents an opportunity for a
multi-gigabit real-time digital processing for portable UWB
systems and 60GHz radio applications, satisfying the stringent
power budget for portable consumer electronic devices.
10
Resolution
Supply Voltage
Input Range
Sample Rate
Power
SFDR (dB)
SNDR (dB)
ENOB (bit)
ERBW
FOM
DNL / INL
Active Area
Technology
This work
3 bit
1V
200 mVpp
REFERENCES
[1]
3 GS/s
5 GS/s
3 mW
3.9 mW
23.2 @fin=346MHz
19.84 @fin=834MHz
16 @fin=346MHz
14.73 @fin=834MHz
2.37 @fin=346MHz
2.15 @fin=834MHz
1.25 GHz
1.15 GHz
0.35 pJ
0.465 pJ
< 0.4 LSB
0.0108 mm2
90 nm
[2]
[3]
[4]
TABLE II
COMPARISON WITH STATE-OF-THE-ART
Resolution
Sample Rate
(GS/s)
Power (mW)
Area (mm2)
Technology
(m)
FOM (pJ)
[5]
[4]
4
[6]
4
[7]
4
[11]
3
THIS
3
1.25
20
10.6
0.021
0.18
CMOS
0.81
2.5
0.033
0.09
CMOS
0.16
43
0.06
0.18
CMOS
2.14
2360
2.55
0.12
SiGe
109
3
0.0128
0.09
CMOS
0.35
[6]
[7]
[8]
[9]
[10]
[11]
221