Beruflich Dokumente
Kultur Dokumente
2005
I.INTRODUCTION
OW-VOLTAGE analog designs in mixed-signal systems have
been recognized as the essential technique to keep up
with the constant drop in digital supply voltage due to techno
logy scaling [1]. The CMOS channel length scaling brings
higher speed in digital designs. To interface with low supply
voltage and high speed digital circuitries in mixed-signal
chips, high-speed analog-to-digital converter (ADC) operating
at low supply voltage is increasingly needed. In this paper,
low-voltage and high speed methodology have been used in
the design of 6-bit fully-differential flash ADC in 0.25 m
CMOS process for a PRML DVD read-channel signal pro
cessing chip. In order to mitigate mismatch effects, resistive
averaging will be discussed and implemented. The flash ADC
architectures are typically the simplest and the fastest struc
tures that can be used to implement ADCs [3]. However, the
main downfall of flash-type ADCs is of larger die-size and
power consumption. The converter is designed to achieve low
est power consumption possible while operating at a satisfact
ory performance level. The first part of the paper describes the
architecture and building blocks. The optimization techniques
and digital error correction methods have been explained in
the following part. The final part of the paper shows simula
tion results.
A.Design specifications
The specifications required for our design project are listed
below:
Resolution : 6 bits
Sample rate (Fs) : 1 GS/s
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
...
+ - +
.VgvBPdnoiraeus-+fatmplifer
...
ROM
g5
b5
g4
b4
Quasi- b3
QuasiGray-toGray
g2 Binary b2
Encoder
Decoder
g3
+
+
-
g1
b1
g0
b0
...
+ - +
NAND
...
+ - +
Comparator
...
Preamp
...
Ref. Ladder
vim
vinp
clk
C
B
vcg
V
ln
d
ika
o
u
m
s-+
tparator
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
No offset
With offset
R2/R1 = 0.25
R2/R1 = 1
R2/R1 = 2
5.7
4.32
4.34
4.40
4.51
III.RESISTIVE AVERAGING
As planned earlier in the checkpoint report, we design and
implement resistor averaging schemes. Averaging resistors
connecting adjacent amplifiers will allow for effective cancel
lations of low-frequency signals, particularly offset noise. It
can be also regarded as enlarging transistor sizes by providing
several identical amplifiers in parallel, and as a result reducing
random offsets. However, the ones residing at each boundary
have adjacent ones only at one side. It causes averaging to be
less effective on boundary amplifiers and their thresholds to
drift away from the value without averaging.
We work on two different configurations. The one called
triple cross-connection [6] is investigated first in order to see
if it minimizes threshold drift and gain compensation. As
pointed out in class presentation, negative transconductance
(gm) contributed by cross connected dummy amplifiers drops
the gain of amplifiers at the boundary, causing undesirable
systematic gain errors as shown in Figure 5. The top plot
shows the first few transfer curves of preamplifiers without
averaging, with their threshold uniformly apart by 1 LSB. The
middle and bottom sections contain the same with 6 and 8
dummy amplifiers cross-connected, respectively. It is noticed
that the first preamp with averaging has a significantly smaller
gain, 1.5 ~ 2.0 deviating from its original value, 3.1. It turns
out that the cross-connected amplifiers also deteriorate dy
namic performance. Specifically, its presence reduces ENOB
to 4.3 from 5.7 without offset included.
Afterwards, we look into a traditional averaging scheme in
that the additional reference ladder taps (beyond FS voltage)
feed dummy amplifiers. 8 dummies are provided at each
boundary and the terminal two dummies are cross-connected
to maintain symmetry. While burning extra powers, this meth
od preserves the gain of boundary amplifiers and has a uni
form capacitive coupling to every resistor tap. Based on the
available information regarding the effect of the number of
dummies and averaging ratios upon its performance, we
sweep through averaging ratios to find out an optimum value.
Figure 5 Transfer curves of the first six preamplifiers. The top section
contains transfer curves (TF) without averaging. The middle and bottom
ones show TFs for 6 and 8 dummies added at the boundaries, respect
ively.
(1)
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
...
...
...
...
g5
+
D
D
g4
g3
+
D
g1
g0
...
...
...
...
clk
Quasig2
Gray
Encoder
B.Metastability
A comparator of an analog-to-digital converter can often be
metastable, which means by the end of the conversion period
the comparator has not reached a decision. The metastability
of a comparator can cause zero, one or two ROM lines to be
selected, creating severe error in the digital output code [1].
Other than increasing positive regeneration pole of the com
parator, two other methods can be used to reduce metastability
error. One of the ways to reduce the error is to use pipelined
latches. Pipelined latches located after the comparator outputs
increases the regeneration gain of the comparator [1]. As
shown in Figure 6 and Figure 7, buffers made of two wellsized inverters and a D-flip-flop have been placed between
each stages of the design to increase the regeneration gain of
the comparator and to ensure the full scale voltage at the out
put of each stage.
4
vdd
g5
b5
g4
b4
g3
QuasiGray
Encoder
b3
g2
b2
g1
b1
g0
b0
clk
g5
b5
g4
b4
g3
Gray-toBinary
Encoder
b3
g2
b2
g1
b1
g0
b0
clk
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
30
vdd
...
...
...
...
...
...
20
10
Thermometer Code
Dout [LSB]
29
LSB
30
LSB
-10
-20
31
LSB
-30
10
20
30
40
50
60
70
Time [ns]
30
LSB
Figure 10 The digital output reconstructed for 500 MHz input frequency.
64 points are sampled for the subsequent FFT analysis.
g0
...
g1
...
g2
...
g3
...
...
g4
...
g5
-20
V.SIMULATION RESULTS
-40
Amplitude dB
-60
-80
-100
-120
0
Metrics at 512
MHz
Without offset
With offset
Averaging
SQNR [dB]
36.36
28.38
30.03
SNDR [dB]
36.15
27.75
28.93
THD [dB]
49.45
36.39
35.46
SFDR [dB]
40.84
38.35
38.29
ENOB [bits]
5.71
4.32
4.51
0.1
0.2
0.3
0.4
0.5
Frequency [GHz]
TABLE III.
Unit
Target Value
Simulation
Value
Resolution
# bits
Sampling Frequency
GS/s
ENOB
# bits
5.5
4.3
Input Capacitance
pF
0.893
Input Common-mode
Voltage
0 Vcm 1.8
0.85V
Ohm
250
250
Total Reference
Ladder Resistance
Supply Voltage
Power Consumption
1.8
mW
Minimum
1.8
131 (total)
112 (static)
VI.CONCLUSION
A 6-bit 1GS/s flash analog-to-digital converter (ADC) for
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
PRML DVD read-channel applications was designed. The
HSPICE simulations show that ENOB of 5.73 bits (without
offset) and 4.3 bits (with offset) are obtained for a 500MHz
frequency input signal. Resistor averaging schemes with an
optimum averaging factor were applied to mitigate the offset
effects, improving ENOB to 4.5 bits.
-20
Amplitude dB
-40
-60
-80
-100
APPENDIX
FFT analyses are carried on several input frequencies from
100 MHz up to 600 MHz. The ENOB variations are plotted in
Figure 12, and the corresponding FFT graphs are shown as
follows.
-120
0
0.1
0.2
0.3
0.4
0.5
Frequency [GHz]
-20
5
Amplitude dB
ENOB [bit]
-40
4
-60
-80
3
-100
2
0.1
0.2
0.3
0.4
0.5
-120
0
0.6
FIN [GHz]
0.2
0.3
0.4
0.5
Frequency [GHz]
0.1
-20
-20
-40
Amplitude dB
Amplitude dB
-40
-60
-60
-80
-80
-100
-100
-120
0
-120
0
0.1
0.2
0.3
0.4
Frequency [GHz]
0.5
0.1
0.2
0.3
0.4
0.5
Frequency [GHz]
ECE598YC CMOS A-D INTERFACE CIRCUITS, FINAL PROJECT REPORT, DEC. 2005
-20
Amplitude dB
-40
-60
-80
-100
-120
0
0.1
0.2
0.3
0.4
0.5
Frequency [GHz]
-20
Amplitude dB
-40
-60
-80
-100
-120
0
0.1
0.2
0.3
0.4
0.5
Frequency [GHz]
ACKNOWLEDGMENT
Gi Hyun Ko and In Jae Chung express deep appreciations
regarding comprehensive and insightful lectures on a
semester-long lectures offered by Professor Y. Chiu and his
guidance throughout this project.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]