chapter 4 circuit-leveldesign -...
TRANSCRIPT
Chapter 4
Circuit-level Design
4.1 Introduction
This chapter focuses on the circuit level design of the basic bnilding blocks in the
siglIl;Hldta (EL,) modulator. It explores the design considerations in selecting
the key blocks like operational transconductance amplifiers (UTA), bias networks.
common-mode feedback circuit. switches, capacitors, comparators and two-phase
clock generation circuitry. The modulator has been designed and simulated in a
TSMC O.18um Cl\IOS process at 1.8V supply voltage using SPICE.
4.2 Sampled-Data vs. Continuous-Time
Sigma-delta modulators can be implemented either as a sampled-data system
or in t.hc cour inuous-t imc domain. Thc main diffcrcnce is that s.unplcd-rluta
sigma-delta systems employ switched-capacitor integrators while continuous-time
systems usc active RC integrators in the modulators as in Figure 4.1. There are
a number of advantages and disadvantages associated with each option which arc
discussed here.
The need for on-chip resistors (51), (1) (87) with very high linearity makes
continuous-time integrators less attractive when compared to switched capacitor
integrators which use capacitor ratios which arc well-controlled. It also helps to
minimize the thermal noise associated with these resistors which occupies a large
66
4.2 Sampled-Data vs. Continuous-Time
area and makes almost impractical to realize on-chip. The frequency response of
switched-capacitor integrators can be more accurately predicted because the time
constant is a function of capacitor ratios (C.,/C i ) and of the sampling frequency.
The time constant of continuous-time integrators, on the other hand is a product
of tho resistor and t.hc capacitor, and suffers severely froin pr()('(~ss variat.ious.
C
Cs
3---,j~F2Vin I I I I
~f r(a)
t
)-L--<:VOUI
Vin R
+
(b)
Voul
Figure 4.1: (a) Switched-capacitor integrator (b) Continuous-time integrator
Auot.hor C\dvallla~e of switched capacitur L;D systems is that they are less
sensitive to clock jitter since most of the charge transfer takes place during the
first half clock period rather than being spread uniformly over t.hc entire clock
period as in continuous-time integrators. As long as the op-amp settles t.o the
required accuracy. it doe; not matt.er whether the op-amp slews or linearly settles.
Continuous- time integrators must be linear at. all times.
Continuous-time integrators also have their advantages. Since the amplifier
set.tling requirenieuts are generally more relaxed than in switched-capacitor cir
cuits. a very high ovcrsampling ratio is uchievablc. The oversampling ratio in
switched-capacitor int.egrators is limit.ed by t.he achievable bandwidth of t.he op
amps. This uiakc» «outiuuous-t.imc ED modulators very appealing for high-speed
applications. 13ecallse of t.hc above incnt.iouc-d uupl.uncutat.iou dillicultios. the
continuous-time approach were not. adopt.ed in this work.
67
4.3 Switched-Capacitor Integrator
Vout
+
Ci
OTAVcmi
F
+'"F2
---1Ci
Fl
~
F2d2/L l F2dl
Vdac- Vdact
~-----'---1
1Yin
1
4.3 Switched-Capacitor Integrator
The integrator is implemented in a fully dilkrellti,d ronJigmatioll as show» ill
Figure 4.2 and employs a two-phase non-overlapping clock as shown in Figure 4.3.
The input is sampled during phase 1 (F 1 and F 1d ) . During phase 2, the charge
is transferred from the sampling capacitor (C.,) to the integrating capacitor (Ci ) .
At the same time, depending on the output value. the appropriate DAC reference
level is applied by closing either switches labelled F 2dl or F2d2 , thus performing
the subtraction operation and the results are being accumulated in the integrat.ion
capacitors.
The integrator employs the bot.tour-plate sampling technique to numnuzo
signal-dependent charge-injection and clock-feed through. This is achieved through
delayed clocks: F 1d , F2d 1 and F2d2 . \Vhen switches labeled F 1 me first. turned off,
the charge injection from those switches remains, to a first order, independent
of the input sigu.il, B(;CHllSC OlW of plates is now Hoatillg, turning off swinhos
labelled F ld shortly after does not introduce charge-injection errors. Further the
switches F1d , F 2d1 and F2d2 are implement.ed as C:YIOS transmission gates in or-
68
4.3 Switched-Capacitor Integrator
Table 4.1: Switches and capacitors in the integrator
Switch/ Size
capacitor
FI N: 10/0.18 urn
P: 10/0.18 urn
F2 N: 10/0.18 urn
P: lO/0.18 mil
FId N: lO/0.18 mn
P: 40/0.18 urn
F2d1 . N: 10/0.18 mil
P: 40/0.18 mil
C, 2 pF
o, 4 pF
der to ensure a small variation ill 011- rosist.auco across the full input siguul range.
This also serves to reduce signal-dependent charge injection from the switches to
C, and C, t.o a negligible level. Table 4.1 shows the switches and capacitors sizes
used in the integrators.
/ /
-~/Figure 4.3: Two-phase non-overlap dock
4.3.1 Operational Trans-conductance Amplifier (OTA)
The basic building block in a switched-capacitor int.egrator is OTA. Figure 4.4
shows the schcmat.ic: of a fully differcntial folded-uLScode OTA used in the in
t.egrators. OIl" of tho uiajor driving forces hdIind t.li« usc of fully difforcntial
opamps is to help reject noise from the substrate as well as from pass-transistor
69
4.3 Switched-Capacitor Integrator
vdd
IFigur« /IA: Fullv-di.lorcnti.rl fold,'d-mscod,' OTJ\
switrho-. turning oil ill switchcd-cava,citol' applicatious. Sinro llJ(\ DC 1 gaIll re
quirement is not so demanding, this is a good selection because of its high opera
tion speed/power consumption ratio. The output common-moele voltage was also
stabilized using a dynamic switched-capacitor common-mode feedback (C:\IFB)
circuit (64) whose linearity is good enough for this application and does not re
quire extra power consumption. Thc schematic also includes the biasing stage
formed by transistors :\h to !vI4.
Trunsistors M, and :\h forms 1,11<' ]l-1.vl)(1 input difh'J'('lIti,d lHur taking into
account the noise constraints, ;VI:; acts as a current source, !vI ll and ;\112 act as
active load for the input stage. anel their gate voltages are controlled by the
common-mode feedback circuit: ]\19 and !VI l O are cascodo transistors which have
opposite type as the input transistors. 11'1 4 • '.1,,, J\h. !vis are current mirrors acting
as active load. The main advantages of using folded-cascade are high ell' gain even
with a single-stage OTA, ease of biasing with a supply voltage as low as 1.8V
when compared with normal tclcscope-cascoded OTAs. simple structure where
frequency compensation is realized by the load capacitance. Also the common
mode control is easier and more accurate. since there is only one stage. Table 4.2
shows the transistor sizes of folded-cascade OTA.
70
4.3 Switched-Capacitor Integrator
r -
Transistor Size (urn)
Ml, M2 720/0.48
M3 432/0.48
1'vJ4, M5, 1'vJ7, M8 216/0.48
1'vJ9. 1'vJ 10 144/0.48
1'vJ1l, M12 72/0.48
Table 4.2' Transistor sizes of folded cascade OTA
4.3.2 Common-mode Feedback
Common-mode feedback is required in Iully-dilfcrontial .unplifior-. to dofiu« t.ho
voltages at the high-impedance output nodes. The amplifier employs dynamic or
switched-capacitor common-mode feedback as shown in Figure 4.5. Capacltors
labelled C, generate the average of t.he output voltages. which is used to create
control voltages for the opamp current sources. The de volt.ag,e across C, is de
terrnined by capacitors C,. which are switched between bias voltages and desired
counuou-un«k- voltage. Thr- bin,s voltag('s an' d(;sigllCd t.o \)(' ('qual to t 1H: differ
ence between the desired common-mode volt.age and the desired control volt.age
used for the opamp current. sources.
The capacitors being switched, C,. might. be between one-quarter and one
tenth the sizes of the non-switched capacitors. C,. Using larger capacitance val
ues overloads the opamp more than is necessary during the phase ()2' and their
size is not. critical to circuit performance. Reducing the capacit.ors t.oo much
causes ('()IIlUlOll-11H)(lc ollsot voltages dur to charge iujor.t.ion of t.ho swii.chos. Nor
mally, all of the switches would be realized by minimum-size n-rhanncl transistors
(O.22um/O.18um) only. except. for t.he switches connect.ed t.o t.he outputs, which
might. be realized by transmission gat.es t.o accommodat.e a wider signal swing. In
applications where the opamp is being used to realize switclied-capacitor circuits.
switched-capacitor CJvIFB circuits arc generally preferred over their continuous
time counterparts since t.hey allow a larger output signal swing. Table 4.3 shows
the capacitor sizes used in the swil.cheri-capacitor common-mode feedback
71
4.3 Switched-Capacitor Integrator
Vout+ Vout-
cmfb
Figure 4.5: A switched-capacitor common-mode feedback
Table 4.3: Capacit.or sizes in t.he switched-capacitor C:\JFB
Capacitor Size
C8 20f
C, 100f
4.3.3 Bias
Figure 4.6 shows the biasing scheme for the OTA which incorporates a wide
swing cascode current. mirror without. restrict.ing t.he signal swings much. The
n-chunncl wide-swing cascade current mirror consists of transistors QI-Q4: along
with t.he diode-connected biasing transistor Qs. The pair Q3. Q4 act.s similarly t.o
a diode-connect.ed transistor at. tho input side of the mirror. The out.put current
comes from Q,. The gate voltages of cascode transistors Ql and Q" an: derived
by the diode-connected transistor Qs. The current for this biasing transistor is
act.uully derived from the bias loop via Ql0 and Qu.
Similarly, t.he p-channel wide-swing cascade current mirror is realized by Q6
Qa. Transistors Qs and Q9 operate as a diode-connected transistor at the input
side of t.he mirror. The current-mirror output. current is t.he drain current. of
Q6' The cascade transistors Q6 and Qo have gate voltages derived from diode
connected Q14, which has a bias current derived from the bias loop via Q12 and
72
4.3 Switched-Capacitor Integrator
08
.-__--oBP1
BP2
017
-=--++-,--__-+-__-+1--=013
05 -yI" :" BN'
L- ~ OBN1
01
Figure 4.6: Wide owing constant transconductance bias circuit.
Table 4 4' Transistor sizes in the bias circuitryC • . " A
Transistor Size Transistor Size Trasistor Size
(um ) (urn) (urn)
Q1, Q2 1D/D.48, 3D/0.36 Q8, Q9 30/0.36, 30/D.48 Q15 10/0.36
Q3, Q4 1O/D36. 10/0.48 Q6, Q7 30/0.48, 30/0.36 Q16 10/0.36
Q5 1.5/0.48 Q14 5/0.48 Q17 10/0.36
Q12, Q13 1O/D.36. 10/0.48 Q10, Q11 3D/D.48, 30/0.36 Q18 20/2
Ql:1. Table 4.4 shows the transistor sizes used in the bias circuitry.
The bias loop does have the problem that at. start-up it is possible for the
current to be zero in all transistors, and the circuit will remain in this stable
state forever. To ensure this condition does not happen, it is necessary to include
start-up circuitry that aneds only the bias-loop in t.ho case that all currnnt.s in
the loop are zero. Here the start-up circuitry consists of transistors Q15, Q16, Q17
and Q18. In the event that all currents in the bias loop are zero, Q17 will be: ofI'.
Since Q18 operates as a high-impedance load that is always on, the gates of Q15
and Q16 will be pulled high. These transistors then will inject currents into the
73
,t.a Switched-Capaclt or Int .egrut or
bias loop which will sta rt up the circuit. Once th e loop sta rts III'. Q17 will come
on. sinking all of the curren t from Q llol . pull ing the gates of Q uo; an d Q 16 low. and
the reby tu rning tln -m off so t hey 110 longer affect the bias loop.
4 .3 .4 Slmulat ion Results of OTA
Table -1 .5 gives the performance summary of OT:\.. Figu re -1 .7 shows the freq uency
respo nse of OTA. Simulat ion res ul ts show that OTA has more than 60 dB gain.
unity gain-bandwidt h (VG D) of -I-l2 :\[Hz and a phase margin of 62 degree. Figure
-l.9 show the transient re-pon-e which shows that the maximum ou tpu t swin g
of a TA is 2\' {different ial]. Figure -l .1O shows the Slew Ra te (S R ) of OT.-\.
whose value is 300\'/ u:,; at loacl capacitance CL=2pF. OTA draws a maximum
current of 2.1 mA dissipating 3.'S m\\' in O.lIS um C:\IOS technology. .-\ second
configura tion i..also considered which is shown in Figu re -I .S [Cascodo OTA circuit
with high output resist ance }. T his OTA design is more stable with increasedload
capacit " TH·e.
.. ~,-------...-... ........_........, .- L
.," ~-------------------+-
· OM ••••_ •••••_ _ _ _ _ _ •••••_ ••
· ,...------------•"•
! • I· ,•.. .- .. -'--." '-. _..- ..- _.-.-.. .._-"'--" -..
_ I .. ).... .......Figure -1 .7: Frequr-ncv res pon-a- of 01'..\.
4 .3 Switched- Capaci to r Integ rato r
""''''•Jf-~ '
~,
Ie.""'''.......
Figure ..1.8: Cascode OT.-\ circuit with high output resistance,
OTA Differential Output Swing
~ - 500ml
- 1.0 f,0.0
>
1.1:1
500m~
0.0
. 1
5.0 u. 1
10ulime ( s )
I
15uI
20u
Figure -1 .9: Tran sient response of OT:\.
rc
4 .3 Switched-Capacito r Integrator
I
/
2.00
1.70
~
> 1.40~
1.10 •
800m492n
--L.498n
time ( s
I
504n)
J
510n
4.3.5
figure ·t 10: Slew rat e of QTA,
P rocess Variation Analysis
Severe uncertaiutv is involved ill nano-scale C).IOS fabri cation which causes vari
at ion in de vice param eters (7). (36), (58), (55). T his leads to variation in dif
Icrent proces s parameters such as chanlid length. ga te-oxide t hickuos s. t hrcshold
vol ta ge. metal win' thickness etc. T I1PSC variation arc ca tr-gorizcd into intr-r-dir
variati on and intra-die [sys tcmatic or random). Xlorcover . tilt' iucreasi ug st a
t ist ical variat ion ill t he pro cess parameter!'; has emerged as a prohleiu ill deep
sub-micron circuit design and cal l cause significant difference ill perform ance of
the fabri ca ted chip compa red to designed , Hence. variability llllal,vsis and mod
eling of t he factors r'outrihuting to the total performance. considering the effects
of random vari ation i ll the process parameters . is important for des igning analog
C),lOS circu its as well. Since c pa mp is the key build ing block in t his design. ill
thi s section. opamp variability is discussed . First. gai n variat ion corresponding
to different loading effect is considered. Figure -l.Ll shows parametric a nalys is
of the ga in a nd ph ase var iat ion corresponding to loml capaci tauo- lfF to lOfF.
As the load capacitance i.... increased the gain slightlv decreased. clue to var ious
second orde r effects.
76
4.5 Circuit Noise
values of sampling and integration capacitors are 2pF. IpF, 0.5pF, 0.2pF and
4pF, 2pF, IpF, 0.4pF respectively.
In addition to sizing the switches for linearity and charge-injection reasons, the
sampling network also needs to b" d"sign"d tor sufficient bandwidth. In('n,asing
the switch sizes decreases the resistance but increases the parasitic capacitance
of the network. An optimum value of the switch size which yields a minimum
R-C time constant can be determined.
4.5 Circuit Noise
There exist mainly two sources of circuit noise (50), (37) which degrades the
porformuncc of wdl-dcsi~l}('d hi~h-r"solut.ion L:6 modulator. They arc thermal
noise (kT/C) and flicker (1/t') noise in the transistors comprising the modulator.
Thermal noise is the result of the random motion of electrons in a conductor. It. is
proportional to absolute temperature. it has a Gaussian amplitude distribution.
and its power spectral density is white.
The source of flicker noise is believed to be due to imperfections in the crystal
lattice at the interface between the oxide and the silicon lavers. The power spec
tral density of flicker noise is approximately inversely proportional to frequency
and is independent of the device biasing conditions. Its amplitude distribution
tends to vary from device to device and may not be Gaussian. The flicker noise
of a ~IOS transistor is inversely proportional to its gate area. So the l/f noise
in an amplifier can be reduced to a great extent simply by increasing the size of
transistors that contribute significantly to the input-referred flicker noise. There
fore the total power spectral density of the equivalent input-referred noise can
be found by adding the two independent noise sources and is given by equation
4.5 where k is Boltzmanns constant, T is the absolute temperature, '/ is a bias
and technology dependent factor, gm is the transconductance of the device, Kf
is an experimentally determined constant that is bias independent but highly
technology dependent, and long-channel transistor behaviour is assumed.
-;? 1 f{~ = 4kT'Y~ + _ f/';} 9m W LCox !
80
(4.5)
4.6 Sampling and Integrating Capacitors
Vdd
1
ICs
Vout
ModelofMOStransistor used as a
switchr---------.I Rs II III
: I
L~-~----lICs
Vout
Fig (a) A switched-capacitor samplingnetwork
(b) Circuit model for sampling noise
Figure 4.14: A switched-capacitor sampling network
4.6 Sampling and Integrating Capacitors
The sizes of the sampling and integrating capacitors are governed by the noise
requirements. The ideal input.-referred thermal noise of the integrator is given by
equation 4.6
kTPN i = 4-
C(4.6)
-:,Figure 4.14a represents the switched-capacitor sampling network in the inte
grator of Figure 4.2 when F1 is high. The power of the sampled noise can be
estimated using the circuit in Figure 4.14b, where the i\lOS switch is modelled
as a resistor in series with a noise voltage source. When the sampling switch in
Figure 4.2 is opened. the noise voltage is stored on C,. If the sampling period
is much longer than the time constant formed by R, and C.,. the high frequency
components of the noise are aliased into the frequency bane! from 0 to f,. As a
result, the full power of the sampling noise appears in this bane! with an approxi
mately white spectral density. kTIC, is thus the total power of the input-referred
81
4.7 Incomplete Linear Settling and Slew Rate Limitation
sampling noise introduced during F1 in the integrator. Since the sampling noise
is distributed uniformly across the Nyquist band, if the signal is oversampled by a
factor M the thermal noise power appearing in the base band is given by equation
4.7
- kT'1)2 = __
AI ]vIC,(4.7)
When the charge that was sampled during F I is transferred from C, to C,
during F 2 , additional thermal noise is introduced by the switches. The thermal
noise introduced by the switches in the circuit during F 2 is band limited by the
response of the operational amplifier as well as tho time constant formed by the
on-resistance of tho switches and the sampling capacitor
4.7 Incomplete Linear Settling and Slew Rate
Limitation
The finite bandwidth of operational amplifiers translates into incomplete linear
sett.ling, in the time domain when the amplifiers arc; used in switched-capacitor
integrators. This causes an integrat.or gain error as given in equation 4.8.
(4.8)
where
(1 - E) is the gain error.
n T = -In(2-N ) , (Nnmber· of Time Constants)
C.,C = feedback factor.z
(4.9)
(4.10)
(4.11 )
From equation 4.9, it. is clear t.hat higher the required resolution (N) of the
ADC, the smaller t.he tolerable settling error will be. This means a higher bias
current. is needed in amplifiers.
82
(4.12)
4.8 Integrator Leakage and Amplifier Gain Requirements
If we substitute the unity-gain frequency of the OTA we designed, we can
compute the number of time constants using the equation 4.12.
, - 1 >- 2fi.;.;uT - ~ ':/" snT
442MH, ;:, 2 x 64iVI H z x nT' nr :( 3.45
which gives a resolution of around 5 bits using the equation 4.10.
This shows how important it is to improve the unity gain bandwidth and de
gain of the opamp to have small sett.ling error as the clock frequency increases,
and to get. the required accuracy. There exists a relationship between the slew
rate'S' and settling time constant 'T' such that slewing never occurs and the
settling process is entirely linear and is given by equation 4.12
S = (1 + C, + Cp ) x Vc;s - VT
C, T
where Vc;s is tbe gate-to-oo\11'ce voltage, VT io the thresho.d voltage of the
input transistors. awl Vas:VT is referred to as the transisr.or overdrive voltage.
C, is the sampling capacitance, C, is the integrating capacitance and Cp is the
parasitic capacitance in equation 4.12.
If we compute T by substituting the sn (300 VIus) of 0\11' OTA. and assuming
an overdrive voltage of 0.5 and parasitic capacitance of 1 pF. we get
T = (1 + 2P,~,1/') X ~:~ = 2.92ns, which is the sottliug time constant. Hence
it is clear that it io necessary to employ relatively large overdrive voltages to
simultaneously achieve both a high slew rate and a large settling time constant.
4.8 Integrator Leakage and Amplifier Gain Re
quirements
Integrator leak, (100) which is a consequence of finite amplifier de gain, limits the
extent to which low-frequency qn.uitizatiou noise is shil]Jcd in a L:6 modulator,
thereby increasing the base band noise. The transfer function of a leaky int.egrator
is given by equation 4.13,
(4.13)
83
4.9 Quantizers
where (' is referred to as the leakage factor.
In a switched-capacitor integrator of Figure 4.2 , the integrator leakage factor
is given by equation 4.14,
1 C,f = AD + 1 x c; (04)
where An is the amplifiers open loop gain. An amplifier gain of 63 db (1.5 K)
gives t.ho intcgrur.or leakage factor as O.02'1r· in order to ensure sufficient linoaritv
and parasitic insensitivity in the integrator response.
4.9 Quantizers
The quantizers (2:3) in the cascaded modulator comprises of a single comparat.or
(68) and a l-bit D/A convert.er in the first stage and a multi-bit A/D and D/A
in tho second and third stages (:30). The comparator is irupleuicut.cd as a cascade
of a regenerative latch followed by an SR latch. while the I-bit DAC is a simple
switch )1<'1 work c01l1ll,cl..cd to off~chip n,i'<:rcnc(' vo]tagc's. Enors int rodurr«l bv 1Ill'
DAC in the first stage of t.he modulator arc added to t.he input signal and directly
degrade the performance of t.he modulator. Therefore, this DAC is designed to
settle to t.he resolution of the modulator. However. errors associated wit.h the
other components of the quantizcrs are great.ly attenuated in the base band by
noise shaping. Tlms t.he principle design objective for these circuits is low power
consumption.
4.9.1 Comparators
The regenerative latch (cross-coupled inverters) in Figure 4.15 is made up of l\!:J,M4 , M.s. and :VI". When phi) is low, t.he p-channcls, l\Is and l\h of the latch
are isolated from the n-channels, l\!:J and ;\14 . In addition, the outputs of the
decision circuit are pulled high, that is, t.he outputs of the comparator arc low,
Q = Q = O. When phi) signal transitions high, t.he regenerative action of t.he
lat.ch causes an imbalance in the decision circuit, forcing the outputs int.o a state
determined by V+ and V-. Table 4.6 shows t.he transistor sizes in t.he regenerative
latch. Figure 4.17 shows the time domain response of the comparator. A second
84
4.9 Quantizers
Transistors
Transistor sizes in the regenerative latch
I Size (11m ) ~
:';11 , lYh, lYb, lvl.1 09/0.45
:\In, lvllO 2.25/0.45
lvl5 , lvl6 , :'vh, :\I~ 2.25/0.45
:\J11 , lvl12 , M1.3' lvl14 1.2/0.18
Table 4.6:
~
configuration iii abo considered which is shown in Figure 4.16 (Clamped push-pull
output comparator). The layont design of clamped push-pull output comparator
is simpler compared to Figure 4.15.
vdd
s
phlt o--tt-"c.:...-------I M10
R
s--oR~
Figure 4.15: Regenerative comparator.
The need for pre-amplification stage depends on the comparator sensitivity
and speed. The preamp stage amplifies the input signal to improve the com
parator sensit.ivit.y and isolates the input. of t.he comparat.or from swit.ching noise
coming from t.he positive feedback st.age.
Sigma-delta modulators show little sensitivity to t.he errors induced during the
85
4.10 Clocks
All the comparators are identical that are used in the first and second stage of
the quantizers. Without a pre-amplification stage, a comparator may provide a
resolution not much better than 50mV, which is enough to guarantee the INL
requirements, taking into account the value of the least significant bit (LSB).
However, if the LSB decreases, as would happen if the gain of the ADC had to
be increased, the hysteresis inherent in the latched comparators might become
a problem. In that case, more complex topologies with pre-amplification stages
must be used, with a considerable increase in the power consumption. That is
the reason why it is interesting to keep the ADC gain equal to unity. After the
comparator array, sixteen A?'lD gates gerwrat.e a l-of-16 code which is translated
t.o binary by an encoding logic using an encoder. To implement t.he DAC. the
l-of-If code is used to select. through analog switches the voltages generat.ed
in a resistive ladder. also used t.o generate the reference voltages for the ADC.
The resistances in the ladder an' H/2 = 100 ohm: this value is low enough to
ensure that the settling (,ITO!' of t.he voltages in the ADC input capacitors arc not
excessive. All switches are complerueut.arv with a size of lOum/0.18um for both
Nl'vIOS and P:\IOS transistors. The value of the t.wo sampling capacitance was
chosen to be 500 fF.
4.9.3 Digital-to-Analog Converters
The I-bit D/ A converters are implemented as switch networks that. are cont.rolled
by the comparator outputs. The outputs of the SR latches used in the compara
tors an' hllfr"r"d hv inverters that .Iriv.: t.wo swit ('Ii.,s ('<Jllll.,(·t(,d t o tho positive
and llq~ativ(~ n>f(~rcIJ('(~ voltages which arc g{~lH'rat('(l on chip.
4.10 Clocks
In switched-capacitor circuit two non-overlap clock phases are needed. In order to
reduce the influence of charge injection (also called clock ked through), a delayed
version is needed for each clock phase. A complement version is also needed for
each clock phase when a transmission gate is used, Normally 4-8 clock phases are
needed in a L;,I'; modulator. These clock phases are usually generated on-chip.
89
4.11 Simulation Results
In a normal signal generator the delayed version clock phase has a same delay at
both rising edge and falling edge. However in most modulators only the falling
edge needs to be delayed in order to reduce the signal-dependent charge injection.
The rising edge does not have to be delayed. The advantage of non-delayed rising
edge is that t.he settling time can be increased. which reduces tho requirement
on the OTA driving capability. The rising edges of the delayed clocks should be
lined up with the rising edges of the non-delayed versions to increase the amount
of available settling time for t.he OTA, which is given by tho equation 4.16.
T,T.sdtleavuilablc = :2 - fnol - tf' - tf (416)
where T, is sampling period, t n ol ='lonoverlap Time, t, =Rise Time and
t r =Fall Time. A circuit to realize two-phase non-overlapping clocks is shown
in Figure 4.20. The two basic non-overlap phases F j and F 2 arc generated by
latches. The delayed versions of t.!H,se two phases are genemted by inverters
working as delay-line. The complement. versions of t.hese phases are also gmwrat.,ed
by inverters, but these inverters are large awl cause very short. extra delay. The
special two-input "inverter" guarantees that the delay happens only at falling
edges.
4.11 Simulation Results
The cOllfigur;lb!<- I:6 modulator has been designed in TSI\IC 0.18Uln C;\IOS
technology, operating from 1.8V supply voltage using SPICE. The cirruit-Iovcl
iiuplouu-utatiou of tho 2nd orde-r I:6 modulator with feed forward signal path
used in t.he first stage is shown in Figure 4.21.
Circuit simulation of a 2nd order modulator with a single-hit. quantizer (GS:vI)
of feed-forward t.opology was carried out. after verifying the results of each block
separately. The l-bit D/A converters are implemented as switch networks that
arc cont.rolled by the comparator outputs. The input. signal had amplitude of 0.5
V (p-p) and a frequency of 100 KHz. The sampling frequency was oct to be 64
:vIHz with an oversampling ratio of HiO. The modulator design employs feedback
reference voltages equal to t.he supply voltages since lowering tho feedback levels
90
4.11 Simul ation Resul ts
Modulated output
1.90
>
1.40
900m
400m
- 100m0.00 4.00u
time8.00u
( s )12.0
Figure -1 .23: Modulator response wit h the modulator input an d qu auti zer ou tput
PSD d !he SIgma-Delta Modulator in !he GSM mode0,.-- - _ _ '- - __-...,
-20
- 40
- 80
~o - 100~
- 120
- 140
-'60
-'60
,,' ,,'Frequeocy [Hz]
,,'
Figure -1 .2-1: OutPU Tspectrum ill GS ),I mode
93
4 .11 Simulat ion R esults
- '00
""~e,
-'50
- 200
,,'-250'-- ...,-- - - ....,.--- .....,-- - -,,----J
Figure .1 .25: Output spectrum ill \\-CD~ I A mode
-~
Figure 4.26: Output spect rum in \\-LA:'\" mode
In order to implement the second and third stage- for t ill' \\"CD~I:\ /\Y LA.:'\
mode. a 2-2 cascadedf fourt h-order] and a 2-2-2 cascaded [sixt h-order] modulator
4 .11 Si m ulatio n R esults
with 2·bit and -l-bit quantizers has been simulated respect ively. T he ..t·bit ADC
employs a flash arc hitecture with sampling network in front of t he 16 comparators
for the comparison of the fully differential input and the reference voltages. The
output of the comparators ace given to an A:'\D gate which generates the ther
mometer code which is furt her converted iuto a binary code using an encoder.
T he -l-bit feedback D.\C circuits are implemented using capacitors. T ho sam
pling capacitances are combined wit h sixteen small unit capacitances to realize
t he -l-bit DAC.
FigUTPs -1.24. 4.25 and 4.26 shows t he modulat cr out put spect rum obtui nod
from circuit -level simulatiou for GS~I I \\'CD~IA/ \\ ' LA:'\ modes fOTa O.5V. 0. 1/ l /S
~ lHz input signal at a sampling frequency of 6..t /6-l/ 200 ~ lHz respectivclv. T he
simulation res ults indicates that the proposed a rchitect ure diss ipates llm\\' /
20m\\" I 35m\\' at 1.8V supply and achieves a peak S:\ DR of 82/G8 /54 d ll ill
GS~I / \\'CD~IA/ II"LA;": modes respectively.
(·l :.)
~
" c a " ",-."y.
11 ' 'o~
c" fj n ,, 1 ~" " e ""[.J,, .v. c
"=" "
.,
Figure 4.27: Opamp layout
95
4 .11 Simulat io n Results
, .-.--;•
• • •• .,· • • •· · • ••; • • ·•• , • •• • ••• • ;• • • • •• ;• • , •· •• • • • •..•• ., • • •• •• • • ;• ••·
•· •• ·• " •• •' .• • • • •~_ ,, _ ..,..: __ L : : • •• • • • •
•
.,
•
•
..
...
Figure 4 .2~ ; Two hit flash converter layout
'. '~!1( I'--.~ I I I 1/'
[. - I .n ! I ~l1-',
~ I, I ..JI '-I
I III
"
f,
I,
I "; ,I II,
:,.
\ i Ii/v
I I U I \, ,I'· ,
I I
I0,
1c- u I t--I I II
, ,, , ;.. ," .. .. .. .. 00 ' 00 ' '" ., -Figure -I .2U: Two hit flash converter layout sinmla tiou out put
4.12 Layout Design
Table 4 7· Porfonnauco comparison with other nmlti-stunrlarrl I:6 ADCsParameters Dual-mode ..igma~delta ADC Triple I\lode
(11 ) (29) (53) (20) (95) (103) (33) Proposed
Orop! 3 2 2 2/(2-1) 2-1-1 2 1 1 1 2-2 2-2-2
No. of hit" 1 2.5 6 1/2.3 1/1/1 Ij1.5j:J 2.5/2.5 1 2 -4
F,.; (1\.IHz) 104/184.32 26/46 23/46 39/38.4 138/245/320 51.2/100!lOO 23/4G G4jG4/20UBW (l\lHz) D,2/3.84 0.2/2 0.2/1.92 O.1/U12 0.271/3.84/20 0.2/5/20 0.2/1.5/1.92 0.2/2/20
DR (dB) 86/54 79/50 HI/70 H2/70 103/82/06 94/88/,56 70/,51/.50 H2/t;Hjr:i4
C:\IOS Pm"""" c.as llm) n.1:,lllm 0.1811m O.\J 11m 0.35 um 0,18 '"ll D,IS lllll 0.18 11111
Pow"r (m\V) 11.5/13..'" 2.4/2,Y 30(50 2.4/4.3 5S/82/12il ."0.8/-5/1 J ll/ZOj:J.S
4.12 Layout Design
A number of layout strategies have to be adopted while doing analog and mixed
signal layout. A digital layout is made by interconnecting simple blocks with
less regard to individual transistors. In contrast. the layout of analog circuits
mainly involves optimizing transistor layouts with much less concern for inter
connections. Important design criteria influenced by the layout are accmacy and
noise immunity,
'When analog and digital circuitries are integrated onto the same chip. addi
tional problems arise. The noise generated by the digital circuitry may couple into
the analog circuitry and corrupt the overall analog circuit performance. There
fore. the control of the noisy interaction is vital in mixed circuits. This can be
done by careful circuit design, for example, by achieving good PSRR.
Fignres 4.27 and 4.28 shows the layout of OTA and flash ADC obtained using
the tool called \IICROWIND in 0.18um technology respectively. Sinmlation re
sults indicate that the OTA layout occupies an area of 6.3um x 5.1um consuming
1m\V of power under 1.8 V supply as illustrated in Fignre 4.27. The simulation
output of 2-bit flash converter is shown in Figure 4.29. Further, the 2-bit flash
ADC dissipates 2.5 m\V occupying an area of 13.8um x 15.6um.
4.13 Performance Comparison
Table 4.7 shows the already publixlu-d mult.i-st.uuiard I:6 ADC in terms of
both dual-mode ,uHI triple-mode architcrturos. TI,,: t riplo-urock: ~6 which has
been published in (95) uses 0.35mn technology and consumes considerably large
97
4.14 Chapter Summary
amount of power. Though (33) uses O.18um technology, it achieves less dynamic
range compared to our performance.
4.14 Chapter Summary
This elmpter desnihes tho circuit kvcl d('si";ll olt.lu- basic building blocks in a E6
modulator based on the results obtained from the Matlab" simulations. The de
sign considerations involves selecting the key blocks like operational transconduc
tance amplifiers (OTA). bias networks. common-mode feedback circuit, switches,
capacitors, comparators and two-phase elock geueration circuitry. The modulator
has been designed and simulated in a TSMC O.18um CMOS process at 1.8V sup
ply volta,.;e. Circ-uit situulat ion of a seco]HI-ol"(kr E6 modulator for GSM with
feedforward topology was carried out after vcrifyirn; t.he results of each block
separately. Later it was extended to simulate a fourth-order and sixth-order
modulator for WCDMA uud WLA:'oI respectively.
98