IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO.

2, FEBRUARY 2014 343

Analysis and Design of a Low-Voltage Low-Power

Double-Tail Comparator
Samaneh Babayan-Mashhadi, Student Member, IEEE, and Reza Lotfi, Member, IEEE

Abstract— The need for ultra low-power, area efficient, and ADCs. Many techniques, such as supply boosting methods [2],
high speed analog-to-digital converters is pushing toward the [3], techniques employing body-driven transistors [4], [5],
use of dynamic regenerative comparators to maximize speed and current-mode design [6] and those using dual-oxide processes,
power efficiency. In this paper, an analysis on the delay of the
dynamic comparators will be presented and analytical expres- which can handle higher supply voltages have been developed
sions are derived. From the analytical expressions, designers can to meet the low-voltage design challenges. Boosting and
obtain an intuition about the main contributors to the comparator bootstrapping are two techniques based on augmenting the
delay and fully explore the tradeoffs in dynamic comparator supply, reference, or clock voltage to address input-range
design. Based on the presented analysis, a new dynamic com- and switching problems. These are effective techniques, but
parator is proposed, where the circuit of a conventional double-
tail comparator is modified for low-power and fast operation they introduce reliability issues especially in UDSM CMOS
even in small supply voltages. Without complicating the design technologies. Body-driven technique adopted by Blalock [4],
and by adding few transistors, the positive feedback during removes the threshold voltage requirement such that body-
the regeneration is strengthened, which results in remarkably driven MOSFET operates as a depletion-type device. Based
reduced delay time. Post-layout simulation results in a 0.18-µm on this approach, in [5], a 1-bit quantizer for sub-1V 
CMOS technology confirm the analysis results. It is shown that in
the proposed dynamic comparator both the power consumption modulators is proposed. Despite the advantages, the body-
and delay time are significantly reduced. The maximum clock driven transistor suffers from smaller transconductance (equal
frequency of the proposed comparator can be increased to 2.5 to gmb of the transistor) compared to its gate-driven counter-
and 1.1 GHz at supply voltages of 1.2 and 0.6 V, while consuming part while special fabrication process, such as deep n-well is
1.4 mW and 153 µW, respectively. The standard deviation of the required to have both nMOS and pMOS transistors operate
input-referred offset is 7.8 mV at 1.2 V supply.
in the body-driven configuration. Apart from technological
Index Terms— Double-tail comparator, dynamic clocked modifications, developing new circuit structures which avoid
comparator, high-speed analog-to-digital converters (ADCs), stacking too many transistors between the supply rails is
low-power analog design.
preferable for low-voltage operation, especially if they do not
I. I NTRODUCTION increase the circuit complexity. In [7]–[9], additional circuitry
is added to the conventional dynamic comparator to enhance
C OMPARATOR is one of the fundamental building blocks
in most analog-to-digital converters (ADCs). Many high-
speed ADCs, such as flash ADCs, require high-speed, low-
the comparator speed in low supply voltages. The proposed
comparator of [7] works down to a supply voltage of 0.5 V
with a maximum clock frequency of 600 MHz and consumes
power comparators with small chip area. High-speed compara-
18 μW. Despite the effectiveness of this approach, the effect
tors in ultra deep submicrometer (UDSM) CMOS technologies
of component mismatch in the additional circuitry on the
suffer from low supply voltages especially when considering
performance of the comparator should be considered. The
the fact that threshold voltages of the devices have not
structure of double-tail dynamic comparator first proposed
been scaled at the same pace as the supply voltages of the
in [10] is based on designing a separate input and cross-
modern CMOS processes [1]. Hence, designing high-speed
coupled stage. This separation enables fast operation over a
comparators is more challenging when the supply voltage is
wide common-mode and supply voltage range [10].
smaller. In other words, in a given technology, to achieve
In this paper, a comprehensive analysis about the delay of
high speed, larger transistors are required to compensate the
dynamic comparators has been presented for various architec-
reduction of supply voltage, which also means that more
tures. Furthermore, based on the double-tail structure proposed
die area and power is needed. Besides, low-voltage opera-
in [10], a new dynamic comparator is presented, which does
tion results in limited common-mode input range, which is
not require boosted voltage or stacking of too many transistors.
important in many high-speed ADC architectures, such as flash
Merely by adding a few minimum-size transistors to the
Manuscript received September 4, 2012; revised December 7, 2012; conventional double-tail dynamic comparator, latch delay time
accepted January 15, 2013. Date of publication February 11, 2013; date of is profoundly reduced. This modification also results in con-
current version January 17, 2014.
The authors are with the Electrical Engineering Group of Engineering siderable power savings when compared to the conventional
Department, Ferdowsi University of Mashhad, Mashhad 91775-111, Iran dynamic comparator and double-tail comparator.
(e-mail: babayan@ieee.org; rlotfi@ieee.org). The rest of this paper is organized as follows. Section II
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. investigates the operation of the conventional clocked
Digital Object Identifier 10.1109/TVLSI.2013.2241799 regenerative comparators and the pros and cons of each
1063-8210 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
344 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 2, FEBRUARY 2014

Fig. 1. Schematic diagram of the conventional dynamic comparator.

Fig. 2. Transient simulations of the conventional dynamic comparator for

structure is discussed. Delay analysis is also presented and input voltage difference of Vin = 5 mV, Vcm = 0.7 V, and VDD = 0.8 V.
the analytical expressions for the delay of the comparators are
derived. The proposed comparator is presented in Section III.
Section IV discusses the design issues. Simulation results are and M4 , M6 ). Thus, Outn pulls to VDD and Out p discharges
addressed in Section V, followed by conclusions in Section VI. to ground. If VINP < VINN , the circuits works vice versa.
As shown in Fig. 2, the delay of this comparator is com-
prised of two time delays, t0 and tlatch . The delay t0 represents
II. C LOCKED R EGENERATIVE C OMPARATORS the capacitive discharge of the load capacitance CL until
Clocked regenerative comparators have found wide appli- the first p-channel transistor (M5 /M6 ) turns on. In case, the
cations in many high-speed ADCs since they can make voltage at node INP is bigger than INN (i.e., VINP > VINN ),
fast decisions due to the strong positive feedback in the the drain current of transistor M2 (I2 ) causes faster discharge
regenerative latch. Recently, many comprehensive analyses of Out p node compared to the Outn node, which is driven by
have been presented, which investigate the performance of M1 with smaller current. Consequently, the discharge delay
these comparators from different aspects, such as noise [11], (t0 ) is given by
   
offset [12], [13], and [14], random decision errors [15], and CL Vthp CL Vthp
kick-back noise [16]. In this section, a comprehensive delay t0 = ∼
=2 . (1)
analysis is presented; the delay time of two common struc- I2 Itail
tures, i.e., conventional dynamic comparator and conventional In (1), since I2 = Itail /2 + Iin = Itail /2 + gm1,2 Vin , for
dynamic double-tail comparator are analyzed, based on which small differential input (Vin ), I2 can be approximated to be
the proposed comparator will be presented. constant and equal to the half of the tail current.
The second term, tlatch , is the latching delay of two cross-
coupled inverters. It is assumed that a voltage swing of
A. Conventional Dynamic Comparator Vout = VDD /2 has to be obtained from an initial output
The schematic diagram of the conventional dynamic com- voltage difference V0 at the falling output (e.g., Out p). Half
parator widely used in A/D converters, with high input of the supply voltage is considered to be the threshold voltage
impedance, rail-to-rail output swing, and no static power of the comparator following inverter or SR latch [17]. Hence,
consumption is shown in Fig. 1 [1], [17]. The operation of the latch delay time is given by, [18]
the comparator is as follows. During the reset phase when    
CL Vout CL VDD /2
CLK = 0 and Mtail is off, reset transistors (M7 –M8 ) pull tlatch = · ln = · ln (2)
gm,eff V0 gm,eff V0
both output nodes Outn and Out p to VDD to define a start
condition and to have a valid logical level during reset. In the where gm,eff is the effective transconductance of the back-to-
comparison phase, when CLK = VDD , transistors M7 and M8 back inverters. In fact, this delay depends, in a logarithmic
are off, and Mtail is on. Output voltages (Out p, Outn), which manner, on the initial output voltage difference at the begin-
had been pre-charged to VDD , start to discharge with different ning of the regeneration (i.e., at t = t0 ). Based on (1), V0
discharging rates depending on the corresponding input volt- can be calculated from (3)
age (INN/INP). Assuming the case where VINP > VINN , Out p  
V0 = Vout p (t = t0 ) − Voutn (t = t0 )
discharges faster than Outn, hence when Out p (discharged  
  I2 t0   I2
by transistor M2 drain current), falls down to VDD –|Vthp| 
= Vthp −  
= Vthp 1 − . (3)
before Outn (discharged by transistor M1 drain current), the CL I1
corresponding pMOS transistor (M5 ) will turn on initiating the The current difference, Iin = |I1 − I2 |, between the
latch regeneration caused by back-to-back inverters (M3 , M5 branches is much smaller than I1 and I2 . Thus, I1 can be

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
BABAYAN-MASHHADI AND LOTFI: ANALYSIS AND DESIGN OF A LOW-VOLTAGE LOW-POWER DOUBLE-TAIL COMPARATOR 345

approximated by Itail /2 and (3) can be rewritten as

  Iin
V0 = Vthp 
I1
  Iin
≈ 2 Vthp 
I
tail
  β1,2 Itail
= 2 Vthp  Vin
I
 tail
  β1,2
= 2 Vthp  Vin . (4)
Itail
In this equation, β1,2 is the input transistors’ current factor
and Itail is a function of input common-mode voltage (Vcm )
and VDD . Now, substituting V0 in latch delay expression and
considering t0 , the expression for the delay of the conventional
dynamic comparator is obtained as
tdelay = t0 +tlatch
    
CL Vthp  CL VDD Itail
=2 + · ln   . (5)
Itail gm,eff 4 Vthp Vin β1,2
Fig. 3. Schematic diagram of the conventional double-tail dynamic com-
parator.
Equation (5) explains the impact of various parameters. The
total delay is directly proportional to the comparator load
capacitance CL and inversely proportional to the input dif-
ference voltage (Vin ). Besides, the delay depends indirectly large tail current would be desirable to enable fast regeneration
to the input common-mode voltage (Vcm ). By reducing Vcm , in the latch [10]. Besides, as far as Mtail operates mostly in
the delay t0 of the first sensing phase increases because lower triode region, the tail current depends on input common-mode
Vcm causes smaller bias current (Itail ). On the other hand, (4) voltage, which is not favorable for regeneration.
shows that a delayed discharge with smaller Itail results in
an increased initial voltage difference (V0 ), reducing tlatch .
B. Conventional Double-Tail Dynamic Comparator
Simulation results show that the effect of reducing the Vcm on
increasing of t0 and reducing of tlatch will finally lead to an A conventional double-tail comparator is shown in
increase in the total delay. In [17], it has been shown that an Fig. 3 [10]. This topology has less stacking and therefore can
input common-mode voltage of 70% of the supply voltage is operate at lower supply voltages compared to the conventional
optimal regarding speed and yield. dynamic comparator. The double tail enables both a large
In principle, this structure has the advantages of high input current in the latching stage and wider Mtail2 , for fast latching
impedance, rail-to-rail output swing, no static power consump- independent of the input common-mode voltage (Vcm ), and
tion, and good robustness against noise and mismatch [1]. Due a small current in the input stage (small Mtail1 ), for low
to the fact that parasitic capacitances of input transistors do offset [10].
not directly affect the switching speed of the output nodes, The operation of this comparator is as follows (see Fig. 4).
it is possible to design large input transistors to minimize the During reset phase (CLK = 0, Mtail1 , and Mtail2 are off),
offset. The disadvantage, on the other hand, is the fact that due transistors M3 -M4 pre-charge fn and fp nodes to VDD , which
to several stacked transistors, a sufficiently high supply voltage in turn causes transistors MR1 and MR2 to discharge the output
is needed for a proper delay time. The reason is that, at the nodes to ground. During decision-making phase (CLK =
beginning of the decision, only transistors M3 and M4 of the VDD , Mtail1 and Mtail2 turn on), M3 -M4 turn off and volt-
latch contribute to the positive feedback until the voltage level ages at nodes fn and fp start to drop with the rate defined
of one output node has dropped below a level small enough to by IMtail1 /Cfn(p) and on top of this, an input-dependent
turn on transistors M5 or M6 to start complete regeneration. At differential voltage Vfn(p) will build up. The intermediate
a low supply voltage, this voltage drop only contributes a small stage formed by MR1 and MR2 passes Vfn(p) to the cross-
gate-source voltage for transistors M3 and M4 , where the gate- coupled inverters and also provides a good shielding between
source voltage of M5 and M6 is also small; thus, the delay time input and output, resulting in reduced value of kickback
of the latch becomes large due to lower transconductances. noise [10].
Another important drawback of this structure is that there is Similar to the conventional dynamic comparator, the delay
only one current path, via tail transistor Mtail , which defines of this comparator comprises two main parts, t0 and tlatch .
the current for both the differential amplifier and the latch The delay t0 represents the capacitive charging of the load
(the cross-coupled inverters). While one would like a small capacitance C Lout (at the latch stage output nodes, Outn and
tail current to keep the differential pair in weak inversion and Out p) until the first n-channel transistor (M9 /M10 ) turns on,
obtain a long integration interval and a better Gm/I ratio, a after which the latch regeneration starts; thus t0 is obtained

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
346 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 2, FEBRUARY 2014

The differential voltage at nodes fn/fp (Vfn/fp) at time t0

can be achieved from
 
Vfn/fp = Vfn (t = t0 ) − Vfp (t = t0 )
IN1 − IN2
= t0 ·
C L ,fn(p)
gm1,2 Vin
= t0 · . (9)
C L ,fn(p)
In this equation, IN1 and IN2 refer to the discharging
currents of input transistors (M1 and M2 ), which are dependent
on the input differential voltage (i.e., IN = gm1,2 Vin ).
Substituting (9) in (8), V0 will be
gmR1,2
V0 = 2VThn Vfn/fp
Itail2
 2
2VThn C Lout
= · · gmR1,2 gm1,2Vin . (10)
Itail2 C L ,fn(p)
Fig. 4. Transient simulations of the conventional double-tail dynamic
This equation shows that V0 depends strongly on the
comparator for input voltage difference of Vin = 5 mV, Vcm = 0.7 V, transconductance of input and intermediate stage transistors,
and VDD = 0.8 V. input voltage difference (Vin ), latch tail current, and the
capacitive ratio of C Lout to C L ,fn(p). Substituting V0 in latch
regeneration time (2), the total delay of this comparator is
from
VThn C Lout VThn C Lout achieved as follows:
t0 = ≈2 (6)  
IB1 Itail2 VThn C Lout C Lout VDD /2
tdelay = t0 + tlatch = 2 + · ln
where IB1 is the drain current of the M9 (assuming VINP > Itail2 gm,eff V0
VINN , see Fig. 3) and is approximately equal to the half of the VThn C Lout C Lout
=2 +
tail current (Itail2 ). I gm,eff
After the first n-channel transistor of the latch turns on (for  tail2 
VDD · Itail2
2 ·C
L ,fn(p)
instance, M9 ), the corresponding output (e.g., Outn) will be · ln 2 ·C
. (11)
discharged to the ground, leading front p-channel transistor 8VThn Lout gmR1,2 gm1,2 Vin
(e.g., M8 ) to turn on, charging another output (Out p) to the From the equations derived for the delay of the double-tail
supply voltage (VDD). The regeneration time (tlatch ) is achieved dynamic comparator, some important notes can be concluded.
according to (2). For the initial output voltage difference at 1) The voltage difference at the first stage outputs (Vfn/fp)
time t0 , V0 we have at time t0 has a profound effect on latch initial differen-
  IB2 t0 tial output voltage (V0 ) and consequently on the latch
V0 = Vout p (t = t0 ) − Voutn (t = t0 ) = VThn −
C delay. Therefore, increasing it would profoundly reduce
  Lout
IB2 the delay of the comparator.
= VThn 1 − (7) 2) In this comparator, both intermediate stage transistors
IB1
will be finally cut-off, (since fn and fp nodes both
where IB1 and IB2 are the currents of the latch left- and right-
discharge to the ground), hence they do not play any
side branches of the second stage, respectively.
role in improving the effective transconductance of the
Considering Ilatch = |IB1 − IB2 | = gm R1,2 Vfn/fp, (7) can
latch. Besides, during reset phase, these nodes have to
be rewritten as
be charged from ground to VDD , which means power
Ilatch Ilatch gmR1,2
V0 = VThn ≈ 2VThn = 2VThn Vfn/fp consumption. The following section describes how the
IB1 Itail2 Itail2 proposed comparator improves the performance of the
(8) double-tail comparator from the above points of view.
where gmR1,2 is the transconductance of the intermediate
stage transistors (MR1 and MR2 ) and Vfn/fp is the voltage III. P ROPOSED D OUBLE -TAIL DYNAMIC C OMPARATOR
difference at the first stage outputs (fn and fp) at time Fig. 5 demonstrates the schematic diagram of the proposed
t0 . Thus, it can be concluded that two main parameters dynamic double-tail comparator. Due to the better performance
which influence the initial output differential voltage (V0 ) of double-tail architecture in low-voltage applications, the
and thereby the latch regeneration time are the transcon- proposed comparator is designed based on the double-tail
ductance of the intermediate stage transistors (gmR1,2 ) and structure. The main idea of the proposed comparator is to
the voltage difference at the first stage outputs (fn and fp) increase Vfn/fp in order to increase the latch regeneration
at time t0 . In fact, intermediate stage transistors amplify speed. For this purpose, two control transistors (Mc1 and
the voltage difference of Vfn/fp causing the latch to be Mc2 ) have been added to the first stage in parallel to M3 /M4
imbalanced. transistors but in a cross-coupled manner [see Fig. 5(a)].

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
BABAYAN-MASHHADI AND LOTFI: ANALYSIS AND DESIGN OF A LOW-VOLTAGE LOW-POWER DOUBLE-TAIL COMPARATOR 347

(a) (b)

Fig. 5. Schematic diagram of the proposed dynamic comparator. (a) Main idea. (b) Final structure.

A. Operation of the Proposed Comparator

The operation of the proposed comparator is as follows (see

Fig. 6). During reset phase (CLK = 0, Mtail1 and Mtail2 are off,
avoiding static power), M3 and M4 pulls both fn and fp nodes
to VDD , hence transistor Mc1 and Mc2 are cut off. Intermediate
stage transistors, MR1 and MR2 , reset both latch outputs to
ground.
During decision-making phase (CLK = VDD , Mtail1 , and
Mtail2 are on), transistors M3 and M4 turn off. Furthermore, at
the beginning of this phase, the control transistors are still off
(since fn and fp are about VDD ). Thus, fn and fp start to drop
with different rates according to the input voltages. Suppose
VINP > VINN , thus fn drops faster than fp, (since M2 provides
more current than M1 ). As long as fn continues falling, the
corresponding pMOS control transistor (Mc1 in this case) starts
to turn on, pulling fp node back to the VDD; so another control
transistor (Mc2 ) remains off, allowing fn to be discharged
completely. In other words, unlike conventional double-tail Fig. 6. Transient simulations of the proposed double-tail dynamic comparator
for input voltage difference of Vin = 5 mV, Vcm = 0.7 V, and VDD = 0.8 V.
dynamic comparator, in which Vfn/fp is just a function of
input transistor transconductance and input voltage difference
(during the reset phase), both switches are closed and fn and
(9), in the proposed structure as soon as the comparator detects
fp start to drop with different discharging rates. As soon as the
that for instance node fn discharges faster, a pMOS transistor
comparator detects that one of the fn/fp nodes is discharging
(Mc1 ) turns on, pulling the other node fp back to the VDD .
faster, control transistors will act in a way to increase their
Therefore by the time passing, the difference between fn and
voltage difference. Suppose that fp is pulling up to the VDD
fp (Vfn/fp) increases in an exponential manner, leading to
and fn should be discharged completely, hence the switch in
the reduction of latch regeneration time (this will be shown in
the charging path of fp will be opened (in order to prevent any
Section III-B). Despite the effectiveness of the proposed idea,
current drawn from VDD ) but the other switch connected to fn
one of the points which should be considered is that in this
will be closed to allow the complete discharge of fn node. In
circuit, when one of the control transistors (e.g., Mc1 ) turns on,
other words, the operation of the control transistors with the
a current from VDD is drawn to the ground via input and tail
switches emulates the operation of the latch. This will be more
transistor (e.g., Mc1 , M1, and Mtail1 ), resulting in static power
discussed in the following section.
consumption. To overcome this issue, two nMOS switches are
used below the input transistors [Msw1 and Msw2 , as shown in
Fig. 5(b)]. B. Delay Analysis
At the beginning of the decision making phase, due to the In order to theoretically demonstrate how the delay is
fact that both fn and fp nodes have been pre-charged to VDD reduced, delay equations are derived for this structure as

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
348 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 2, FEBRUARY 2014

previously done for the conventional dynamic comparator and transconductance of the latch is increased. In other words,
the conventional double-tail dynamic comparator. The analysis positive feedback is strengthened. Hence, tlatch will be
is similar to the conventional double-tail dynamic comparator,  
C Lout Vout
however; the proposed dynamic comparator enhances the tlatch = · ln
gm,eff + gmR1,2 V0
speed of the double-tail comparator by affecting two important  
factors: first, it increases the initial output voltage difference C Lout VDD /2
= · ln . (16)
(V0 ) at the beginning of the regeneration (t = t0 ); and gm,eff + gmR1,2 V0
second, it enhances the effective transconductace (gmeff ) of Finally, by including both effects, the total delay of the
the latch. Each of these factors will be discussed in detail. proposed comparator is achieved from
1) Effect of Enhancing V0 : As discussed before, we define
t0 , as a time after which latch regeneration starts. In other tdelay = t0 + tlatch
 
words, t0 is considered to be the time it takes (while both latch VThn C Lout C Lout VDD /2
=2 + · ln
outputs are rising with different rates) until the first nMOS Itail2 gm,eff + gmr1,2 V0
transistor of the back-to-back inverters turns on, so that it will VThn C Lout C Lout
pull down one of the outputs and regeneration will commence. =2 +
I gm,eff + gmR1,2
According to (2), the latch output voltage difference at time ⎛tail2 ⎞
t0 , (V0 ) has a considerable impact on the latch regeneration VDD /2
× ln ⎝   gmR1,2 gm1,2 Vin   ⎠.
time, such that bigger V0 results in less regeneration time.  
4VThn VThp Itail2 exp
G m,eff1 ·t0
Similar to the equation derived for the V0 of the double-tail Itail1 C L,fn(p)

structure, in this comparator we have (17)

Ilatch By comparing the expressions derived for the delay of the
V0 = VThn
IB1 three mentioned structures, it can be seen that the proposed
Ilatch comparator takes advantage of an inner positive feedback
≈ 2VThn in double-tail operation, which strengthen the whole latch
Itail2
gmR1,2 regeneration. This speed improvement is even more obvious
= 2VThn Vfn/fp. (12) in lower supply voltages. This is due to the fact that for larger
Itail2
values of VTh /VDD, the transconductance of the transistors
In order to find Vfn/fp at t = t0 , we shall notice that the
decreases, thus the existence of an inner positive feedback
combination of the control transistors (Mc1 and Mc2 ) with two
in the architecture of the first stage will lead to the improved
serial switches (Msw1 , Msw2 ) emulates the operation of a back-
performance of the comparator. Simulation results confirm this
to-back inverter pair; thus using small-signal model presented
matter.
in [18], Vfn/fp is calculated by
3) Reducing the Energy Per Comparison: It is not only the
Vfn/fp = Vfn(p)0 exp((Av − 1)t/τ ). (13) delay parameter which is improved in the modified proposed
comparator, but the energy per conversion is reduced as well.
In this equation, Avτ−1 ∼
C L,fn(p)
= and Vfn(p)0 is the initial
G m,eff1 As discussed earlier, in conventional double-tail topology, both
fn/fp node difference voltage at the time when the correspond- fn and fp nodes discharge to the ground during the decision
ing pMOS control transistor is started to be turned on. Hence, making phase and each time during the reset phase they should
it can be shown that Vfn(p)0 is obtained from be pulled up back to the VDD . However, in our proposed
  gm1,2 Vin comparator, only one of the mentioned nodes (fn/fp) has to
Vfn(p)0 = 2 VThp  . (14) be charged during the reset phase. This is due to the fact that
Itail1
during the previous decision making phase, based on the status
Substituting (13) in (12), V0 will be
of control transistors, one of the nodes had not been discharged
gmR1,2 and thus less power is required. This can be seen when being
V0 = 2VThn Vfn/fp
Itail2   compared with conventional topologies [see Figs. 9(b) and
  gmR1,2 gm1,2 Vin G m,eff1 · t0 10(b)].

= 4VThn VThp  exp .
Itail2 Itail1 CL,fn(p)
(15) IV. D ESIGN C ONSIDERATIONS
Comparing (15) with (10), it is evident that V0 has been In designing the proposed comparator, some design issues
increased remarkably (in an exponential manner) in compare must be considered. When determining the size of tail transis-
with the conventional dynamic comparator. tors (Mtail1 and Mtail2 ), it is necessary to ensure that the time
2) Effect of Enhancing Latch Effective Transconductance: it takes that one of the control transistors turns on must be
As mentioned before, in conventional double-tail comparator, smaller than t0 (start of regeneration)
 
both fn and fp nodes will be finally discharged completely. VThp · CL,fn(p) VThn C Lout
In our proposed comparator, however, the fact that one of the ton,Mc1(2) < t0 → <
I IB1
first stage output nodes (fn/fp) will charge up back to the   n1,2
VThp · CL,fn(p) VThn · C Lout
VDD at the beginning of the decision making phase, will turn → < . (18)
I I
on one of the intermediate stage transistors, thus the effective Tail1
2
Tail2
2

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
BABAYAN-MASHHADI AND LOTFI: ANALYSIS AND DESIGN OF A LOW-VOLTAGE LOW-POWER DOUBLE-TAIL COMPARATOR 349

This condition can be easily achieved by properly designing threshold voltage mismatch is obtained as follows:
the first and second stage tail currents. Even if possible in the gmc1,2VThc1,2
fabrication technology, low-threshold pMOS devices can be Veq,dueVThc1,2 =
gm1,2
used as control transistors leading to faster turn on.
μp WC1,2 VODC1,2
In designing the nMOS switches, located below the input = VThc1,2 (20)
transistors, the drain-source voltage of these switches must μn W1,2 VOD1,2
be considered since it might limit the voltage headroom, where VOD refers to the overdrive voltage of the transistors.
restricting the advantage of being used in low-voltage applica- 2)Effect of Current-Factor Mismatch MC1 , MC2 , i.e.,
tions. In order to diminish this effect, low-on-resistance nMOS β C1,2 :
switches are required. In other words, large transistors must In order to calculate the input-referred offset due to the
be used. Since the parasitic capacitances of these switches do current factor mismatch of MC1,2 , β C1,2 is modeled as a
not affect the parasitic capacitances of the fn/fp nodes (delay channel width mismatch W , i.e., β/β = W/W . The
bottlenecks), it is possible to optimally select the size of the differential current that W generates can be obtained as
nMOS switch transistors in a way that both low-voltage and expressed in (21).
low-power operations are maintained. 1 W
The effect of mismatch between controlling transistors on μp Cox
i diff = (Vgsc1,2 − Vthc1,2 )2 . (21)
the total input-referred offset of the comparator is another 2 L
important issue. When determining the size of controlling Note that the controlling transistors are in saturation since
transistors (MC1 − MC2 ), two important issues should be |VGDc1,2| = |Vfn − Vfp | < |Vthp|. So the input-referred offset
considered. First, the effect of threshold voltage mismatch voltage due to the current factor mismatch is calculated from
and current factor mismatch of the controlling transistors i diff
on the comparator input-referred offset voltage. Second, the Veq,dueβC1,2 =
gm1,2
effect of transistor sizing on parasitic capacitances of the
0.5μp WC1,2 (Vgsc1,2 − Vthp )2
fn/fp nodes, i.e., CL,fn(p), and consequently the delay of the =
comparator. While larger transistors are required for better μn W1,2 (Vgs1,2 − Vthn )
0.5μp WC1,2 VOD2
matching; however, the increased parasitic capacitances are
delay bottlenecks. In order to study the effect of threshold = C1,2

μn W1,2 (Vcm − RCLK K nC1,2 VOD 2 − Vthn )

C1,2
and current factor mismatch of control transistors on the
(22)
total input-referred offset voltage, a brief mismatch analysis
is presented here. where Vcm is the input common mode voltage and Rclk is the
equivalent on resistance of the tail transistor.
Assuming both mismatch factors, the total input-referred
A. Mismatch Analysis offset due to the mismatch of the controlling transistors can
be found from
In principle, the effect of threshold voltage mismatch and 
current factor mismatch of controlling transistors is almost σtotal = σV2
ThC1,2
+ σβ
2
C1,2
. (23)
negligible in most cases except for the situation where input
From (20) and (22), it can be concluded that the ratio of
differential voltage (Vin ) is very small where fn and fp have
the controlling transistor sizes to the input transistor size, i.e.,
approximately similar discharging rates. This is true because
(WC1,2 /W1,2 ), is effective in reducing the offset. Due to the
by the time that the controlling transistor (Mc1 or MC2 ) turns
fact that the transconductance of the input transistors (gm1,2)
on, the differential input signal is already amplified to large
is important in amplifying the input differential voltage and
amplitude compared to the mismatches. In other words, offset
due to the dominant role of the size of these transistors on
due to the controlling transistor mismatches is divided by the
total input-referred offset, usually large input transistors are
gain from the input to the output. However, in case of small
designed, which results in diminishing the effect of controlling
Vin , when fn and fp follow each other tightly, the mismatch
transistors mismatch.
of the controlling transistors might influence the result of the
comparison. Hence, the following brief analyzes the effect
of threshold and current factor mismatches of controlling B. Kickback Noise
transistors on the total input-referred offset voltage. Principally in latched comparators, the large voltage vari-
1)Effect of Threshold Voltage Mismatch of MC1 , MC2 , i.e., ations on the regeneration nodes are coupled, through the
VThC1,2: parasitic capacitances of the transistors, to the input of the
The differential current due to the threshold voltage mis- comparator. Since the circuit preceding it does not have zero
match can be obtained from output impedance, the input voltage is disturbed, which may
degrade the accuracy of the converter. This disturbance is
i diff = gmc1,2 VThc1,2 (19) usually called “kickback noise.” In [16], it has been shown
that the fastest and most power efficient comparators gen-
where gmc1,2 is the transconductance of the controlling tran- erate more kickback noise. This is true about our proposed
sistors. So, the input-referred offset voltage due to the Mc1,2 dynamic comparator. Although it improves the double-tail

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
350 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 2, FEBRUARY 2014

(a)

Fig. 7. Peak input voltage error due to kickback noise. (b)

Fig. 9. (a) Post-layout simulated delay and (b) energy per conversion as a
function of supply voltage (Vin = 50 mV, Vcm = VDD − 0.1).

(a)
Fig. 8. Layout schematic diagram of the proposed dynamic comparator.

topology in terms of operation speed and thus energy per

comparison, the kickback noise is increased in comparison to
conventional double-tail structure (Fig. 3). Fig. 7 presents the
peak disturbance as a function of differential input voltage of
the comparator in three studied architectures. While double-
(b)
tail structure takes advantage of input-output isolation and
thus the minimum kickback noise, the conventional dynamic Fig. 10. (a) Post-layout simulated delay and (b) energy per conversion as a
comparator and our proposed structure has nearly similar function of input common-mode voltage (Vin = 50 mV, VDD = 1.2).
kickback noise. However, in our proposed comparator since
control transistors are not supposed to be as strong as the latch TABLE I
transistors in conventional dynamic comparator, it is possible S UMMARY OF THE C OMPARATOR P ERFORMANCE
to determine the size of those transistors in a way that keeps
Item Value
the advantages of the speed enhancement and power reduction,
Technology 180-nm CMOS
while reducing kickback noise. Besides, for some applications
where kickback becomes important, it is possible to apply sim- Supply voltage 1.2 V
ple kickback reduction techniques, such as neutralization [16] Average power dissipation per conversion 329 μW
@ freq. = 500 MHz
to remarkably reduce the kickback noise (See Fig. 7, proposed
Worst case delay (Vcm = 0.6 V, 550 ps
dynamic comparator with neutralization). Vin = 1 mV)
Delay/log(Vin ) 69 ps/dec
V. S IMULATION R ESULTS Offset standard deviation 7.8 mV
In order to compare the proposed comparator with the (1-sigma) (σ os )
conventional and double-tail dynamic comparators, all cir- Energy efficiency 0.66 pJ
cuits have been simulated in a 0.18-μm CMOS technology
with VDD = 1.2 V. The comparators were optimized and
the transistor dimensions were scaled to get an equal offset found in [10]). Fig. 8 shows the layout of the comparator.
standard variation of σ OS = 8 mV at the input common- Particular care was taken in the layout to avoid affecting delay
mode voltage of V cm = 1.1 V (the same conditions that are and power of the comparator. Fig. 9(a) and (b) demonstrates

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
BABAYAN-MASHHADI AND LOTFI: ANALYSIS AND DESIGN OF A LOW-VOLTAGE LOW-POWER DOUBLE-TAIL COMPARATOR 351

TABLE II
P ERFORMANCE C OMPARISON

Comparator structure Conventional Dynamic Double-tail Dynamic Proposed Dynamic Comparator

Comparator Comparator
Technology CMOS 180 nm 180 nm 180 nm
Supply voltage (V) 0.8 V 0.8 V 0.8 V
Maximum sampling frequency 900 MHz 1.8 GHz 2.4 GHz
Delay/log(Vin ) (ps/dec.) 940 358 294
Peak transient noise voltage 215 n 221 n 219 n
at regeneration time(nV)
Kickback noise voltage (at Vin = 10 mV) 51.3 mV 5.3 mV 43 mV With neutralization: 13 mV
Energy per conversion (J) 0.3 p 0.27 p 0.24 p Without Msw1 and Msw2 : 0.265 p
Input-referred offset voltage (mV) 7.89 mV 7.91 mV 7.8 m
Estimated area 16 μ × 16 μ 28 μ × 12 μ 28 μ × 14 μ

the post-layout simulation results of the delay and the energy

per conversion of the mentioned dynamic comparators versus
supply voltage variation. As shown in Fig. 9(a), in comparison
with the other two structures, the delay of the proposed double-
tail dynamic comparator is significantly reduced in low-voltage
supplies. It is obvious that at high supply voltages, all struc-
tures have approximately similar performances, about 200 ps
clk-to-output delay (including clock buffer) with 0.65 pJ/bit-
conversion for 8-mV offset. However, by decreasing the supply
voltage, three structures start to behave differently. It is evident
that the double-tail topology can operate faster and can be used
in lower supply voltages, while consuming nearly the same
power as the conventional dynamic comparator. The case is
even much better for the proposed comparator when compared
to the conventional double-tail topology. For instance, the
Fig. 11. Delay of the proposed comparator versus supply voltage (VDD ).
proposed comparator can operate in 0.6 V supply at the
cost of 106 fJ/conversion with 840 ps delay versus 1.81 ns
for the conventional double-tail comparator and 3.5 ns for
the conventional topology. Our simulations show that if the
circuit is optimized for VDD = 0.6 V, the results would
be even better for the proposed circuit. Fig. 10 shows the
simulated performance as a function of input common-voltage
(Vcm ). Generally in the double-tail topologies, the delay of
the comparator is less influenced by the variation of the input
common-mode voltage in comparison with the conventional
dynamic topology and thus has a wider common-mode range.
The power consumption is nearly equal.
Fig. 11 depicts the dependence of the comparator delay
on power supply level at various differential input voltages.
For Vin = 10 mV, the delay is 460 ps at VDD = 0.9 V.
This delay drops from 460 to 162 ps when VDD changes
from 0.9 to 1.5 V. In addition, at a given VDD , the larger the
differential input voltage, the smaller the comparator delay will
be. Fig. 12 shows the simulated delay of the comparator versus
differential input voltage under different conditions of input Fig. 12. Delay of the proposed comparator versus input voltage difference
common-mode voltage (Vcm ) at VDD = 1.2 V. The delay of the (Vin ).
comparator at Vin = 1 mV and Vcm = 700 m is 413 ps. For
a given value of Vcm , the delay decreases as differential input
voltage increases. Furthermore, the delay is also dependent Fig. 13, the standard deviation of the offset of the proposed
on the variation of common-mode voltage. For example, at comparator is achieved to be σ OS = 7.8 mV using Monte
Vin = 10 mV, the delay increases by 64 ps, from 239 to Carlo simulations for a run of 300 samples. Table I summarizes
303 ps, as Vcm decreases from 900 to 700 mV. As shown in the performance of the proposed dynamic comparator. Finally,

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.
352 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 2, FEBRUARY 2014

[8] B. Goll and H. Zimmermann, “A 65nm CMOS comparator with modified

latch to achieve 7GHz/1.3mW at 1.2V and 700MHz/47μW at 0.6V,” in
Proc. IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, Feb. 2009,
pp. 328–329.
[9] B. Goll and H. Zimmermann, “Low-power 600MHz comparator for
0.5 V supply voltage in 0.12 μm CMOS,” IEEE Electron. Lett., vol. 43,
no. 7, pp. 388–390, Mar. 2007.
[10] D. Shinkel, E. Mensink, E. Klumperink, E. van Tuijl, and B. Nauta,
“A double-tail latch-type voltage sense amplifier with 18ps Setup+Hold
time,” in Proc. IEEE Int. Solid-State Circuits Conf., Dig. Tech. Papers,
Feb. 2007, pp. 314–315.
[11] P. Nuzzo, F. D. Bernardinis, P. Terreni, and G. Van der Plas, “Noise
analysis of regenerative comparators for reconfigurable ADC archi-
tectures,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 55, no. 6,
pp. 1441–1454, Jul. 2008.
[12] A. Nikoozadeh and B. Murmann, “An analysis of latched comaprator
offset due to load capacitor mismatch,” IEEE Trans. Circuits Syst.
II, Exp. Briefs, vol. 53, no. 12, pp. 1398–1402, Dec. 2006.
[13] S. Babayan-Mashhadi and R. Lotfi, “An offset cancellation technique
for comparators using body-voltage trimming,” Int. J. Analog Integr.
Circuits Signal Process., vol. 73, no. 3, pp. 673–682, Dec. 2012.
Fig. 13. Histogram diagram of Monte Carlo simulation of the offset in [14] J. He, S. Zhan, D. Chen, and R. J. Geiger, “Analyses of static and
proposed dynamic comparator. dynamic random offset voltages in dynamic comparators,” IEEE Trans.
Circuits Syst. I, Reg. Papers, vol. 56, no. 5, pp. 911–919, May 2009.
[15] J. Kim, B. S. Leibowits, J. Ren, and C. J. Madden, “Simulation and
analysis of random decision errors in clocked comparators,” IEEE Trans.
Table II compares the performance of the proposed comparator Circuits Syst. I, Reg. Papers, vol. 56, no. 8, pp. 1844–1857, Aug. 2009.
[16] P. M. Figueiredo and J. C. Vital, “Kickback noise reduction technique
with the conventional dynamic and double-tail comparators. for CMOS latched comapartors,” IEEE Trans. Circuits Syst. II, Exp.
In 0.18-μm CMOS, the proposed comparator provides the Briefs, vol. 53, no. 7, pp. 541–545, Jul. 2006.
maximum sampling of 2.4 GHz at 0.8-V supply voltage. [17] B. Wicht, T. Nirschl, and D. Schmitt-Landsiedel, “Yield and speed
optimization of a latch-type voltage sense amplifier,” IEEE J. Solid-State
Circuits, vol. 39, no. 7, pp. 1148–1158, Jul. 2004.
VI. C ONCLUSION [18] D. Johns and K. Martin, Analog Integrated Circuit Design, New York,
USA: Wiley, 1997.
In this paper, we presented a comprehensive delay analy-
sis for clocked dynamic comparators and expressions were
derived. Two common structures of conventional dynamic
comparator and conventional double-tail dynamic comparators
were analyzed. Also, based on theoretical analyses, a new
dynamic comparator with low-voltage low-power capability Samaneh Babayan-Mashhadi was born in Mash-
had, Iran, in 1984. She received the B.S. and M.S.
was proposed in order to improve the performance of the degrees (Hons.) in electrical engineering from the
comparator. Post-layout simulation results in 0.18-μm CMOS Ferdowsi University of Mashhad, Mashhad, Iran, in
technology confirmed that the delay and energy per conversion 2006 and 2008, respectively, where she is currently
pursuing the Ph.D. degree, from the Electrical Engi-
of the proposed comparator is reduced to a great extent in neering Department, with research on design of low-
comparison with the conventional dynamic comparator and power high-resolution, high-speed analog-to-digital
double-tail comparator. converters.
Her current research interests include low-power
low-voltage analog and mixed-signal integrated
R EFERENCES circuits.

[1] B. Goll and H. Zimmermann, “A comparator with reduced delay time in

65-nm CMOS for supply voltages down to 0.65,” IEEE Trans. Circuits
Syst. II, Exp. Briefs, vol. 56, no. 11, pp. 810–814, Nov. 2009.
[2] S. U. Ay, “A sub-1 volt 10-bit supply boosted SAR ADC design in
standard CMOS,” Int. J. Analog Integr. Circuits Signal Process., vol. 66,
no. 2, pp. 213–221, Feb. 2011. Reza Lotfi received the B.Sc. degree from Ferdowsi
[3] A. Mesgarani, M. N. Alam, F. Z. Nelson, and S. U. Ay, “Supply University of Mashhad, Mashhad, Iran, the M.Sc.
boosting technique for designing very low-voltage mixed-signal circuits degree from the Sharif University of Technology,
in standard CMOS,” in Proc. IEEE Int. Midwest Symp. Circuits Syst. Tehran, Iran, and the Ph.D. degree from the Uni-
Dig. Tech. Papers, Aug. 2010, pp. 893–896. versity of Tehran, Tehran, in 1997, 1999, and 2004,
[4] B. J. Blalock, “Body-driving as a Low-Voltage Analog Design Technique respectively, all in electrical engineering.
for CMOS technology,” in Proc. IEEE Southwest Symp. Mixed-Signal Since 2004, he has been with the Ferdowsi Univer-
Design, Feb. 2000, pp. 113–118. sity of Mashhad, where he is currently an Associate
[5] M. Maymandi-Nejad and M. Sachdev, “1-bit quantiser with rail to rail Professor. From 2008 to 2009, he was with the
input range for sub-1V  modulators,” IEEE Electron. Lett., vol. 39, Electronics Research Laboratory, Delft University
no. 12, pp. 894–895, Jan. 2003. of Technology, Delft, the Netherlands, as a Post-
[6] Y. Okaniwa, H. Tamura, M. Kibune, D. Yamazaki, T.-S. Cheung, Doctoral Scientific Researcher involved in research on ultralow-power analog
J. Ogawa, N. Tzartzanis, W. W. Walker, and T. Kuroda, “A 40G- and mixed-signal integrated circuits for biomedical applications. His current
b/s CMOS clocked comparator with bandwidth modulation technique,” research interests include low-voltage low-power analog integrated circuit
IEEE J. Solid-State Circuits, vol. 40, no. 8, pp. 1680–1687, Aug. 2005. design for biomedical applications and high-speed data converters for telecom-
[7] B. Goll and H. Zimmermann, “A 0.12 μm CMOS comparator requiring munication systems.
0.5V at 600MHz and 1.5V at 6 GHz,” in Proc. IEEE Int. Solid-State Dr. Lotfi has been an Associate Editor of the IEEE T RANSACTIONS ON
Circuits Conf., Dig. Tech. Papers, Feb. 2007, pp. 316–317. C IRCUITS AND S YSTEMS —PART I: R EGULAR PAPERS since 2010.

Authorized licensed use limited to: Intel Corporation via the Virtual Library. Downloaded on July 08,2020 at 16:53:06 UTC from IEEE Xplore. Restrictions apply.