1
Low Energy and Low VoltageADC Design Strategy
Akira Matsuzawa
Tokyo Institute of Technology
2Outline
• Overview of ADCs
• OpAmp based ADC design
• Comparator based ADC design : SAR ADCs
• Flash and sub-ranging ADCs
• Summary
3ADC performance
0 10 20 30 401 10 10
1 10 9
1 10 8
1 10 7
1 10 6
1 10 5
1 10 4
1 10 3
0.01
0.1
1
6)
SNR (dB)
BE
R
QPSK 16QAM 64QAM 256QAM
N
S
PPBWC 1log2
Shannon’s theory to determine capacity
N
ADCN
S
PP 225.1
SNR of ADC is
sNfC
srate NfD Higher data-rate can be realizedby higher multi-level modulation.It result in increase of ADC resolution.
2sfBW
Conversion frequency is determined by signal bandwidth.
Therefore
Data rate is proportional to the product of fs and N
fs: Sampling frequencyN: Resolution
4Performance and architectures of ADCs
4 6 8 10 12 14 16
10M
1M
100k
10k
100M
1G
10G
184 6 8 10 12 14 16
10M
1M
100k
10k
100M
1G
10G
18
Con
vers
ion
freq
uenc
y (H
z) Flash
Pipeline
Delta-sigmaSAR
Sub-ranging
Resolution (bit)
ADC has a suitable performance domain.SAR ADC expands the performance area
OpAmp based
Comparator based
5Strategy of energy efficient ADC design
Reducing static powerResistor DAC Capacitor DACOpAmp based Dynamic comparator based
2DDd CVE
Reducing capacitance
Flash Sub-range SAR# of CMP
TR sizeG
T CV 1 Large TR Small TR with compensation
Noise
CVn
1
Use complementally ckt.
Clock Use self clocking
Reducing voltageEffective to digital gates
Use forward or adaptive body biasing
6SNR vs. signal bandwidth
50
60
70
80
90
0.1 1 101 102
135 dB/Hz
143 dB/Hz
150 dB/Hz
BCDEFGHIJK
[8] VCO[9] VCO[10]CT[11]DT[12]CT[13]CT/DT[14]DT[15]CT[16]CT[17]DT
BCDEFGHIJK
[8] VCO[9] VCO[10]CT[11]DT[12]CT[13]CT/DT[14]DT[15]CT[16]CT[17]DT
fb (MHz)
SN
R (d
B)
bfdBSNRdBSNR log10)()( 0
N
s
PPdBSNR log10)(
bNN fnsityspectrumdePP )('
bN
s fPPdBSNR log10'
log10)(
SNR0
bfdBSNRdBSNR log10)()( 0
SNR of ADCs is inversely proportional to signal bandwidth, fb. Higher bandwidth results in lower SNR and effective resolution.
7SNR vs. signal bandwidth
143dB
133dB
123dB
Timmy Sundstrom, PhD thesis, Linkoping 2011.
fs=4fb
Same tendency to higher BW.
8Fundamental Energy of sampling circuit
Signal
Switch Capacitor
Track Hold
NFSqn
qnVVP 2
22
21212
NFS
qnVV2
TCVn k21
21 2
Sampling circuit
Electrical energy=Thermal energy
CTVn
k2
Quantization voltage
Quantization noise power
qnn PV 2
2
22k12FS
N
VTC
C C
NFSd TCVE
22 2k242
Noise balance
Capacitance
Pd of sampling circuit
Ns TE
22k24Fundamental energy of sampling is often used. However this neglects power for comparison.
9Energy consumption of ADC
Timmy Sundstrom, PhD thesis, Linkoping 2011.
Nss TE
22k24
Energy of ADC is reaching 100x of the fundamental sampling energy,and 10x of the fundamental ADC energy consumption.
Consumed energy of ADC is mainly determined by the resolution.
192 102 NsEEs
sADC ENE
EADC
Conventional fundamental sampling energy
Fundamental ADC conversion energyinvolving energy consumption of comparator
192 102 NADC NE
10
OpAmp based ADC design
11Mega-technology trend of ADCs
2C
4C
8C
16C
16CC
Major conversion scheme is now changing from pipeline to SAR.
Pipeline ADC
SAR ADC
CMPDAC
-
-+
+
Op amp
CMPDAC
-
-+
+
Op amp
1st stage 2nd stage
Sample Amplify
Cf
Cs
Cf
Cs
1st stage 2nd stage
OpAmp based design
Comparator based design
Consumes static power
No static power
12Amplifier for pipeline ADC
Vsig
VDAC
Cf
Vout
Cs
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
1stage
OpAmp
Vref
Comp
DataVsig
Vou
t
221 DACsigDAC
f
s
f
ssigout
VVVCC
CCVV
Conventionally Cs=Cf=C0
0,refDAC VV
Unit conversion circuit
I/O transfer characteristics
An OpAmp realizes an accurate voltage amplification.
13Low voltage OpAmp: Headroom and Pd
n
VkTVNf
VIVP
II
thn
effcDDtotDDda
tot
25
25
2_
2
nV
kTVNfVNfCI
NfCVIg
NfCgGBW
thn
effceffc
ceff
m
cm
22
2
424
2_
02
02
2
0
2
VDDVbp
Out
Vbn
In Vs
Veff
Veff
Two stage OpAmp
gm2
gm1
I1
I2
I2I1
n=4
n
VkTVNf
VPPthn
effcDDdampdpipe 2102 2
_
effDDpps VVV 22_ TGSeff VVV
VVeff 15.0
Total Pd of ADC
A two stage cascade OpAmp can realize low voltage operation.However, the output voltage swing become lower at low voltage operation.
2_
0 2thnV
kTnC
1212 44 IIgg mm
14Required performances
222
22
31
231
N
effDDNsig
q
VVVV
12222_ ENOBqthn VV
2_
0 2thnV
kTnC
Quantization voltage
Thermal noise
Unit capacitance
31:
#:2:
factorfeedbackβ
sourcesnoiseofntcoefficiennoiseγ
cNfGBW
106)( NG dB70dB: 10b94dB: 14b
N: Resolutionfc: Conversion freq.
Open loop gain
ENOB: Effective # of bit
GBW
Required gain and bandwidth of OpAmp and capacitance are determined bythe resolution and conversion frequency.
idealfromnDegradatioENOB :
15C0 and Pd at low voltage operation
12bit, 100MHz ADC
Capacitance should be larger at low voltage operation to keep sufficient SNR. This results in rapid increase of power dissipation.
effDDpps VVV 22_ kTVVC
kTVC
SNR effDDpps2
02_0 22 20 2 effDD VV
SNRkTC
eff
Dmc VC
ICgGBWf
00
2
20 2 effDDDDeffc
DDeffcDDDd VVVSNRVTf
VVfCVIP
Low voltage doesn’t make sense for pipeline ADC.
0.4 0.6 0.8 1 1.21 10 12
1 10 11
1 10 10
1 10 9
Vdd
C0
(F)
20 21
effDD VVC
P d(W
)
0.4 0.6 0.8 1 1.20.01
0.1
1
10
Vdd
22 effDDDD
d VVVP
16FoM: Figure of Merit of ADC
222
2122230
effDD
DDeffENOB
N
cd VVVVkTnNfP
Nc
ENOBd
ENOBc
d
fP
fPJFoM
22
2)(
idealfromnDegradatioENOB : 22 212
2230effDD
DDeffENOB
ENOB
VVVVkTnFoM
0.4 0.6 0.8 1 1.21 10 13
1 10 12
1 10 11
FoM
(J/c
onv.
ste
ps)
VDD (V)
FoM stands for consumed energy normalized by the effective steps.
12bit, 100MHz ADC
Low voltage operation for OpAmp based ADC increases FoM.
17
eff
Dmc VC
ICgGBWf
00
2
20 2 effDDDDeffc
DDeffcDDDd VVVSNRVTf
VVfCVIP
FoM vs. VDD
Nc
ENOBd
ENOBc
d
fP
fPJFoM
22
2)(
An OpAmp based ADC consumes much conversion energy
12bit ADC
0.4 0.6 0.8 1 1.21 10 15
1 10 14
VddVDD (V)
FoM
(J/C
onv.
ste
ps)
SAR ADC 6fJ
0.4 0.6 0.8 1 1.21 10 13
1 10 12
1 10 11
FoM
(J/c
onv.
ste
ps)
VDD (V)
Pipelined ADC
150fJ
18
Comparator based ADC design : SAR ADC
19Basic idea for low energy analog design
sDDd
LDD
stogle
IVPCVIf
CML, OpAmp
VDD
Vi+ Vi-
Vo-CL
RL
CL
RLVo+
Is
RO
CL
Vi Vo
VDD
CMOS
2DDL
dtogle VC
Pf
2
21
DDLdd VfCfEP
Conventional analog circuits consume larger energy. Dynamic circuits doesn’t consume larger energy.
2
21
DDLd VCE
CMOS: Consumed energy is independent of delay time.
Lortogle CRTf 11
Lotogle CRf 1
20SAR ADC
Vin
Vref
Comparator Logic
Switches
Capacitors
S11 S12 S13 S14 S15
S2
S0
2C
4C
8C
16C
16C
VrefVin
NA
C
Cn1
nC
10 n
refsigx VnVV VxNA
Sample 221
refCVE
SAR can be designed to consume no static power.
Generating subtracted signal
No resistorsNo static current !Potentially full swing
inCVQ
SAR can realize larger signal swing compared with pipeline ADC.
Not OpAmp based,but comparator based
21Performance overview of SAR ADCs
FoM has lowered rapidly due to the progress of SAR ADC.
1/200 during three years.
FoM trend of 10bit ADC
0.1
1
10
100
1000
2005 2006 2007 2008 2009 2010
Recent FoM rangeFoM
(fJ/
conv
. ste
ps)
stepsconvfJseveral ./10
22Issue of comparator for SAR ADCs
A comparator has noise and this results in conversion error.
V. Giannini, P. Nuzzo, V. Chironi, A. Baschirotto, G. van der Plas, and J. Craninckx, “An 820uW 9b 40MS/s Noise Tolerant Dynamic-SAR ADC in 90nm Digital CMOS,” IEEE ISSCC 2008, Dig. of Tech. Papers, pp.238-239, Feb. 2008.
23Dynamic comparator
CLK
CLK CLK
CLK
INP
SPSN
INNFN FP
2iV
2iVID ID
CL
CL
VDD
A dynamic comparator is widely used to reduce static power.
The difference in input voltages causes a difference in discharging speed.
D. Schinkel, E. Mensink, E. Klumperink, Ed Van Tuijl, B. Nauta,
“A Double-Tail Latch-Type Voltage Sense Amplifier with 18ps Setup-Hold Time,” ISSCC Dig. of Tech. Papers, pp.314-315, Feb., 2007.
24Deriving noise equation
12
2
eff
dd
ds
Lt V
VIkTC
d
2
02
22
22 1
d
d
t
nD
vnD
Lt dtiII
C
22
222 k,k
D
L
L
D
nt
Ln I
TC
CI
vδ
CTv
d
1) Sampling noise of Switch
2) Transistor noise
dt
nL
n dtiCv
0
1
Noise voltage of output by current noisevICtD
L
td
ni
VDD
VDD/2
td
td
VL
time
Vi VL
ni ID CL
S
VDDimVg
Equivalent circuit
TR noiseVoltage and timing
Transistornoise
Samplingnoise
1
k422
222
eff
dd
ddL
eff
d
tdeffin V
VαγVCα
TVtδ
αV
Vδ
A. Matsuzawa," IEEE 8th International Conference on ASIC(ASICON), pp. 218-221, Oct. 2009.
25Match with noise simulation
DDL
effin VC
kTVV
2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0 10 20 30 40
CL [fF]
δV
in(σ
) [
mV
]
Estimation(α=2)Estimation(α=1)Simulation
0
10
20
30
40
50
60
70
80
90
100
-1.0 -0.6 -0.2 0.2 0.6 1.0
ΔVin [mV]
P (out=
hig
h) [%
]
The derived equation has a good match with simulation.
Noise in comparator
1
k422
22
eff
dd
ddL
effin V
VαγVCα
TVVδ
26Required capacitance and consumed Energy
6 7 8 9 10 11 12 13 140.1
1
10
100
1 103
1 104
1 105
CL(
fF),
Ec(f
J)
Resolution (bit)
CL
Ec
VDD=1V, Veff=0.2V
Node capacitances should be increased to realize higher ADC resolution.This results in increase of consumed energy of the dynamic comparator.
2fF & 4fJ @8bit40fF & 80fJ @10bit0.6pF & 1pJ @12bit10pF & 20pJ @14bit
Flash ADC: Ec determines the minimum FoMSAR ADC: Ec cannot be neglected for higher resolution ADC
Ec: conversion energy
27Noise improvement of dynamic comp.
CLK
VINP_a VINP_b VINN_b VINN_a
Gate weighted interpolation
Offset CALVdd
Voutp
VoutnWPa WNaWNbWPb
V n(
) [m
V]
M. Miyahara, Y. Asada, D. Paik, and A. Matsuzawa, "A Low-Noise Self-Calibrating Dynamic Comparator for High-Speed ADCs," A-SSCC, Nov. 2008.
Yusuke Asada, Kei Yoshihara, Tatsuya Urano, Masaya Miyahara, and Akira Matsuzawa, "A 6bit, 7mW, 250fJ, 700MS/s Subranging ADC," A-SSCC, 5-3, pp. 141-144, Taiwan, Taipei, Nov. 2009.
90nm CMOS
Dynamic comparatorNoise of comparator
Noise of comparator can be reduced by complementary ckt. and an optimizationof the node capacitance.
CMOS input
28Pd estimation of SAR ADC
Vin
Vref
Comparator Logic
Switches
Capacitors
222 DDLcdc VCfNp 22 DDscds VCfp
22 DDgcdc VCNfp
Cs: Total sampling capacitanceCL: Load capacitance of comparatorCg: Effective capacitance of gates and switches
Cg: const
2
22
DD
N
s VC 3
22
DD
N
L VC
1) S/H&CDAC 2) Comparator
2) Logic gates and switch drivers
Divide SAR ADC into three different circuits.
Akira Matsuzawa, ISSCC Forum, San Francisco, CA, Feb. 2011.
29Equations to estimate the ADC performance
222 DDLcdc VCfNp
22_ 1
4
DD
eff
eff
DD
Lthn V
VVV
CkTV
22
231
N
DDq
VV
222_ 12 qENOBthn VV
2
2_
14
DD
eff
eff
DD
thnL V
VVV
VkTC
N
c
ENOBdgdcds
fPPP
FoM2
2
2_
4
thns V
kTC
22 DDscds VCfp
22 DDgcdg VCNfp
Quantizationvoltage
Permitted thermalnoise
ThermalNoise of COMP.
Sampling capacitor
LoadCapacitorOf COMP.
Pd of S/H
Pd of COMP.
Pd of Gate
30C, Pd, and FoM vs. VDD
0.4 0.6 0.8 1 1.21 10 13
1 10 12
1 10 11
Vdd
0.4 0.6 0.8 1 1.21 10 5
1 10 4
1 10 3
0.01
Vdd
Cs,
CL
(F)
VDD (V) VDD (V)
P d(W
)
Cs
CL S/H
Comp
Logic
Cg=0.1pF
Cg=0.3pF
2
1
DDs V
C
3
1
DDL V
C
DDdc VP 1 2
DDdg VP
.constPds
Cs and CL increase with reducing VDD,since the quantization voltage decreases with reducing VDD.
Pd of S/H is constant for VDD,however Pd of comparator increases with reducing VDD.Pd of logic gate decreases rapidly with reducing VDD.
31FoM vs. VDD
0.4 0.6 0.8 1 1.21 10 15
1 10 14
VddVDD (V)
FoM
(J)
Cg=0.1pF
Cg=0.3pF
VVKT
bitENOBMHzF
bitN
eff
c
15.03002
5.010012
FoM can be lowered by reducing VDD, if Pd of logic gate is dominant.
Thus the voltage lowering is effective to reduce Pd for low resolution ADC,However, it is still difficult to reduce Pd by reducing VDD for high resolution ADC,even if SAR ADC architecture is used.
Gate dominant
COMP dominant
32Example: An ultra-low power CDCWe have developed an ultra-low power Capacitance to Digital Converter.
1. 10b SAR like architecture2. Self-clocking3. Single to differential
① ②③
3nA @ 30 times/secTuan Minh Vo,Yasuhide Kuramochi, Masaya
Miyahara,Takashi Kurashina, and Akira Matsuzawa“A 10-bit, 290 fJ/conv. Steps, 0.13mm22, Zero-Static
Power, Self-Timed Capacitance to Digital Converter.”SSDM 2009, OCT.
33Self clocking technique
Comparison is ended if the output voltages are not same.
Output voltage of dynamic comparator
Self-clocking scheme
Self-clocking scheme is very useful1) Reducing power consumption (Clock circuits, routing clock, )2) Just an enable command signal is required. No need of clock.
Suitable for micro controller.
34
60GHz Short range mm-wave system
Background
35Frequency plan in 60 GHz range
Ch.1 Ch.2 Ch.3 Ch.4
57 58 646362616059 65 66
2160 MHz
1760 MHz
Frequency [GHz]
BW: 1.8GHz, 4 channels
IEEE 802.15.3c
BPSK: 1.7 GbpsQPSK: 3.5 Gbps16QAM: 7 Gbps
802.15.3c-2009, IEEE Std., Oct. 2009. [Online]. Availablehttp://standards.ieee.org/getieee802/download/802.15.3c-2009.pdf
Wide frequency range for 60 GHz short distance communication
36System block diagram
RF Front-end IC (65nm CMOS) Baseband IC (40nm CMOS)
Digital Base Band60GHz ILO20GHz PLL 3456MHz PLL
PLL PLL
LPF ADC
VGA ADC
Buff
BuffLNA
VGA
DAC
LPF
DAC
DAC
DACDAC
LPF DAC
DACPA
LPF
DAC
36MHz TCXO
DAC
8bDAC
DAC
Reg. bank addr/data
chch
8b 8b
8b
8b+6b 8b+3b
By Sony
Matsuzawa and Okada lab. developing
37Equipment image
Gain: 5.6 dBi
Two chips solution on one PCB with antenna
Low cost system
38
40nm CMOS (BB)
4.2m
m3m
m
Tokyo Tech SONY 38
Developed chips
65nm CMOS (RF)
LNAQ MIXER
I MIXER
LO BUF.
LO BUF.
Q.OSC.
LogicI MIXER
Q MIXER
LO BUF.
LO BUF.
Q.OSC.PA
PLL LO BUF.
DCTR
DAC(Qch)
ADC(Qch)VGA
VGA
DAC(Ich)PLL
Digital BB
LVDS(11ch)
LVDS(11ch)
LDPC
ADC(Ich)RAM
K. Okada, .., A. MatsuzawaISSCC 2012
39
39
Block Diagram of RF Chip
Tx Output
LNA
I Mixer
RF Amp.Rx input
BB LNA
PFD
20GHz PLL19.44GHz, 20.16GHz,20.88GHz, 21.60GHz
Q Mixer
I Mixer BB Amp.
LO Buf.
BB Amp.
RF Amp.
RF Amp.PA
Q MixerRx input BB LNA
RF Amp.
36MHz
LO Buf.
CP LPF
÷4 CML÷5÷(27,28,29,30)
LogicChannel selectionGain controlPower managementTDD control
Controlsignals
I+
I-
Q+
Q-
I+
I-
Q+
Q-
Ref.Clk
60GHz QILO
40
40
Block Diagram of BB Chip
I+
I-
Q+
Q-
LPF
LPF
Analog BB
ClockPLL
2fs
(4/3)fs
Digital BB
RxFilter
AGC
DCTR*
SyncDetector
LDPCDecoder
BB/RFControlPRBS generator
Saw-tooth wave gen.BER calc.AWGN generatorRAM
Test modules
TxFilter
SyncGenerator
LDPCEncoder
LVDS
LVDS
Max. 6.3Gb/sTx PHY data
Max. 6.3Gb/sRx PHY data
Controlsignals
I+
I-
Q+
Q-
Controlsignals
VGALPF
VGALPF
ADC
ADC
DAC
DAC
Ref.Clk
*digital carrier and timing recovery
(35 MHz)
41
41
RF Measurement Setup
I/Q
Control signals
RF board(Tx mode)
I/Q
Control signals
RF board(Rx mode)
Power supply Power supply
AWGAgilent M8190A
OscilloscopeAgilent DSA91304A
Laptop PC Laptop PC
I/Q output (Rx)
I/Q input (Tx)
DC supply
DC supply16.3mm x 14.4mm
6-dBi antennaTx
[3] R. Suga, et al., EuMC 2011
Rx
42
42
RF+BB Measurement Setup
AbsorberRF boardRF board BB boardBB board
BB chip BB chipRF chipwith 6dBi antenna [3]
[3] R. Suga, et al., EuMC 2011
I/Q
Control signals
RF boardI/Q
BB PHYControl(FPGA)
Laptop PC
Power supply
I/Q
Control signals
I/Q
RF board
Power supply
BB PHY Control(FPGA)
Laptop PC
Tx mode Rx mode
43
-40-30-20-10
010
55.08 58.32 61.56-40-30-20-10
010
57.24 60.48 63.72-40-30-20-10
010
59.40 62.64 65.88
-40-30-20-10
010
61.56 64.80 68.04
-40-30-20-10
010
59.40 62.64 65.88
7.0Gb/s 16QAM (max 10Gb/s)Channel ch.1 ch.2 ch.3 ch.4 Max rate
Constellation
Spectrum
Back-off 4.4dB 4.6dB 5.0dB 5.7dB 5.0dB(ch.3)Data rate* 7.0Gb/s 7.0Gb/s 7.0Gb/s 7.0Gb/s 10.0Gb/s(ch.3)
EVM** -23.0dB -23.0dB -23.3dB -22.8dB -23.0dB (ch.3)Distance*** 0.3m 0.5m 0.5m 0.3m >0.01m (ch.3)
*The roll-off factor is 0.25. The bandwidth is 2.16GHz except for Max rate.**EVM through Tx and Rx boards. ***Maximum distance within a BER of 10-3. The 6-dBi antenna in the package is used.2012/03/05
44
2012/03/05
Performance comparison (RF+BB)
[1] K. Okada, et al., ISSCC 2011 [4] H. Asada, et al., A-SSCC 2011 [5] A. Siligaris, et al., ISSCC 2011 [6] S. Emami, et al., ISSCC 2011 [12] C. Marcu, et al., ISSCC 2009
Integration Data rate (16QAM) Ch. PDC (Tx/Rx)
CEA-LETI [5] RF (Hetero) 3.8Gb/s - 1,357mW/ 454mW
SiBeam [6] RF (Hetero) 3.8Gb/s Ch.1-2 1,820mW/ 1,250mW
Tokyo Tech(This work)
RF (Direct) +analog BB+digital BB
RF+BB:6.3Gb/s Ch.1-4
RF:319mW/ 223mW
BB:196mW/ 398mW
World’s fastest data rate with low power
45Comparison of 60GHz RF front-end
02468
101214161820
2007 2008 2009 2010 2011 2012 2013
Data
rate
[Gb/
s]
Year
UCB
NEC OOK
Univ. of Toronto
FSKOOK
SiBeam, CEA-LETI
16QAM
QPSK+16QAMTokyo Tech
Toshiba
IMEC
direct-conversionother arch.
2012/03/05
We attained world’s fast data rate of 16Gbps
46Developing baseband SoC
40nm CMOS technologyADC
Developed chip integrating ADC, DAC, VGA,and PLL, using 40nm CMOS technology.
5b, 3GSps, 12mW/chRX: 300mW, TX: 110mWTokyo tech developed
47Analog Baseband : ADC
47
D-FF 5bit x 8
On-chip foreground calibration
-80-70-60-50-40-30-20-10
0
0 288 576 864 1152Nor
mal
ized
Pow
er [d
B]
Frequency [MHz]
Fin = 100 MHzVGA Gain = 12 dB
*single channel inc. S/P
*
48ADC Comparison
Architecture Cal. fs[GS/s]SNDR[dB]
Power[mW]
FoM [fJ/-c.s.]
Process[nm]
Area[mm2]
[1] Flash - 3.5 31.2 98 946 90 0.149[2] SAR Internal 2.5 34.0 50 489 45 1[3] Folding Internal 2.7 33.6 50 474 90 0.36
[4] Pipeline, Folding External 2.2 31.1 2.6 40 40 0.03
[5] Flash Internal 2.88 27.8 36 600 65 0.25
This work Flash Internal 2.3 26.1 12 316 40 0.06
[1] K. Deguchi, et al., VLSI Circuits 2007 [2] E. Alpman, et al., ISSCC 2009 [3] Y. Nakajima, et al., VLSI Circuits 2007 [4] B. Verbruggen, et al., ISSCC 2010[5] T. Ito, et al., A-SSCC 2010
48
49
Flash and sub-ranging ADCs
50Flash ADC
Comparator Array
1
1
0
0
• Expecting highest speed• Comparator determines the ADC performance
6b: 63 NFS
qVV2
Vq=16mV, Mismatch
51FoM of Flash ADC
ENOBcENOBN
s
Nsc
ENOBs
d Ef
fEf
PFoM
22
22
Ec: Energy/Comparator
FoM of flash ADC is determined by energy consumption of unit comparator and the degradation of effective bit.
Reduction of consumed energy and increase of ENOB are very important.
2DDc CVE
2
q
n
2
q
off2
)()(121log21
VV
VVENOB
Offset mismatch Thermal noise (can be neglected)
52Performance of flash ADC
2
q
off2
)(121log21
VVENOB
N
ENOBd
fP
22FoM
c
FoM is degraded by the offset mismatch voltage of the comparator.Offset mismatch voltage should be reduced at low voltage operation.
Voff(): Offset mismatch
Vq : 1LSB voltage
53Tradeoff: mismatch and energy consumption
Transistor size (um2)
0
5
10
15
20
25
0 0.2 0.4 0.6 0.8 1
0
50
100
150
200
250
300
350
オフセット
消費電力
Offs
et m
ism
atch
(mV)
E c(fJ
)
LWVo
1)(ffset
LWCE cc
Serious tradeoff between mismatch of transistor and gate area. Larger transistor is required to reduce mismatch voltageand results in increase of gate area and consumed energy.
21
offsetc VE
Ec=50fJ3mV
Mismatch compensation
Example
6bit ADC: Voff
54FoM vs. Area
0.01
0.1
1
10
0.01 0.1 1 10
Area (mm2)
FoM
(pJ/
conv
.ste
p)
Occupied area should be reduced to lower the FoM.We must pay much attention to the occupied area.
AreaCEc
5bit and 6bit ADCs
55Mismatch compensation for the dynamic comparator
FP or FN
Vi
gmVi IDVL
CLVDD
CLK
CLK CLK
CLK
INP
SPSN
INNFN FP
2iV
2iVID ID
CL
CL
VDD
VDD
VDD/2 td
VL
time
dt
D
LDDd I
CVt2
Delay time
effd
D
m
D
LDD
i
D
D
d
i
d
Vt
Ig
ICV
dVdI
dIdt
dVdt
2
effD
m
eff
D
eff
Dm
effTgsD
VIg
VI
VdIg
VVVI
D
D
L
L
d
d
II
CC
tt
eff
i
d
d
VV
tt
D
D
L
Leffi I
ICCVV
effi
d
d
VV
tt
Mismatch can be compensated by change of capacitance or current
Equivalent circuit for the fist stage
56Digital calibration methods for mismatchResistor ladder type Capacitor array type
Binary weighted capacitor arrayY. Asada, K. Yoshihara,T. Urano, M. Miyahara and A. Matsuzawa,
“A 6bit, 7mW, 250fJ, 700MS/s Subranging ADC” A-SSCC, pp. 141-144, Nov. 2009.
57Effect of digital mismatch compensation
Measured resultV o
ffset
Voffset
Voffset
The mismatch voltage can be reduced from 14mV to 1.7mV.
M. Miyahara, Y. Asada, D. Paik, and A. Matsuzawa, "A Low-Noise Self-Calibrating Dynamic Comparator for High-Speed ADCs," A-SSCC, Nov. 2008.
58Area comparisonPenalty area for digital compensation will be reduced with technology scaling.
Comprator CAP Array
Register 4b
5m
65 m
25 m 10 m 30 m
5m
85 m
Comprator
25 m 30 m 30 m
Register 4bDecorderMUX
4.5m
Comprator
30 m
120 mStrage Capacitor & Charge Pump
90 m
2.9m
41 m
UpDownCounter5b
22 m
Comprator &Cap Array
19 m
90 nm
40 nm
Resistor ladder
Capacitor array
Capacitor array
59Issue of resistive DAC to generate VREF
T/HIN
7
7
7
7
15
Upper bit
Lowerbit
Resistive DAC consumes static power and has a serious tradeoff between Pd and speed.
pronref
refpronref CRI
VCRR
44max I
τ 1
60Advantage of capacitive DAC to generate VREF
Operating as S/H circuit• No static power consumption ( 360W@1GHz )• Smaller Cu realize faster settling time
(tDAC= 3.4 ronCU < 80ps @ rON = 1k, CU= 15fF)
REFIN
tot
pout VnV
CCV
1
1 CRon2DDd CVE
Capacitor DAC doesn't consume static power and has no trade offbetween Pd and speed.
61Settling time and power
CDAC realizes faster settling time to RDACwith low power consumption.
0
20
40
60
80
100
120
140
160
180
0 50 100 150 200
time [ps]
α=0.5 vout(CDAC)
α=5 vout(CDAC)
α=10 vout(CDAC)
RDAC
0
200
400
600
800
1000
1200
1400
1600
1800
RDAC
CDAC
CDAC
CDAC
Time response Power dissipation
626bit sub-ranging ADC using CDAC
6 bit ADC has been realized in a 90 nm 10M1P CMOS technology with a chip area of 0.13mm2
Y. Asada, K. Yoshihara,T. Urano, M. Miyahara and A. Matsuzawa,
“A 6bit, 7mW, 250fJ, 700MS/s Subranging ADC” A-SSCC, pp. 141-144, Nov. 2009.
63Performance comparison
[1] C-Y. Chen, VLSI Circuits 2008.[2] B-W. Chen, A-SSCC 2008.[3] F. C. Hsieh, A-SSCC 2008.[4] Z. Cao, ISSCC 2008.[6] Y. C. Lien, A-SSCC 2008.
[1] [2] [3] [4] [6] This WorkResolution(bit) 6 6 6 6 6 6fs(GS/s) 0.8 1.2 0.7 1.25 1 0.7SNDR(DC/Nyq.) 35/32 34/33 31/30 34/28 35/33 35/34Pd (mW) 12 75 24 32 30 7Active area(mm 0.13 0.43 0.052 0.09 0.18 0.13VDD(V) 1.2 1.2 1.2 1.2 1.2/1.0 1.2FoM(pJ) 0.44 2.17 1.31 1.22 0.8 0.25CMOS Tech.(nm 65 130 130 130 90 90Architecture Flash Flash Pipeline 2b-SAR Subrange Subrange
Attain lowest FoM at that time
64Voltage lowering: FoM vs. VDD
VDD
Ec
Voff
ENOB
)(bitENOB cfSVIV
VCE
TcDD
2DDcc
exp
Ec:Energy consumption for each comparator and followed logic circuits.
FoM can be reduced drastically by reducing supply voltage VDD.
ENOBcE
2FoM
ENOB is degraded by the reduction of VDD, however little affects the FoM.
Energy reduction by reducing VDD is dominant.
2
2
)(121log
21
q
off
VV
ENOB
65FoM delay (FD) product
TDDDD
d VVVkT
Delay time
The FD product suggests the balance between the number of interleaving and decrease of energy consumption.
VT = 0.4 V DelayFoMFD
Delay is increased and the operating speed is lowered by reducing VDD
We should investigate the optimum VDD by FD product.
Optimum range
66Forward body biasingForward body biasing can decrease the delay time (1/2)and can be used easily at 0.5 V operation.
VDD
IN OUTBB
FBB
BB
VTN = 100mV
VTP = 60mV
Increased leakage current in the proposed ADC is 0.32 mA by forward body biasing.
1/2
67ADC Structure
Block diagram of ADC Chip microphotograph
5bit 0.5V 600MSps Flash ADC is designed and fabricated in 90nm CMOS.
S/H circuits use gate boosted switches.
M. Miyahara , J. Lin, K. Yoshihara, and A. Matsuzawa,“A 0.5 V, 1.2mW, 160fJ, 600 MS/s 5 bit Flash ADC”A-SSCC, pp. 177-180, Nov. 2010.
68Performance Summary
Reference # [7] [8] [9] [10] This workResolution (bit) 5 5 5 5 5fs (GS/s) 0.5 1.75 1.75 0.06 0.6SNDR (dB) 26 30 30 26 27Pd (mW) 5.9 2.2 7.6 1.3 1.2Active area (mm2) 0.87 0.017 0.03 - 0.083Vdd (V) 1.2 1 1 0.6 0.5FoM(fJ) 750 50 150 1060 160CMOS Tech. (nm) 65 90 90 90 90Architecture SAR Fold+Flash Flash Flash Flash
[7] B. P. Ginsburg, J. Solid-State Circuits 2007.[8] B. Verbruggen, ISSCC 2008.[9] B. Verbruggen, VLSI Circuits 2008.[10] J. E. Proesel, CICC 2008.
FoMFmax = 160fJ @ 600MSpsFoMBest = 110 fJ @ 360MSps
A high speed and low FoM 0.5V flash ADC has been realized.
69Summary of energy efficient ADC design
Reducing static powerResistor DAC Capacitor DACOpAmp based Comparator based
2DDd CVE
Reducing capacitance
Flash Sub-range SAR# of CMP
TR sizeG
T CV 1 Large TR Small TR with compensation
Noise
CVn
1
Use complementally ckt.
Clock Use self clocking
Reducing voltageEffective to digital gates and low resolution ADC
Use forward or adaptive body biasing