F Silveira D Flandre P G A Jespers September 1996
F Silveira D Flandre P G A Jespers September 1996
F Silveira D Flandre P G A Jespers September 1996
net/publication/2977303
A gm/ID based methodology for the design of CMOS analog circuits and its
application to the synthesis of a silicon-on-insulator micropower OTA
CITATIONS READS
639 8,916
3 authors, including:
All content following this page was uploaded by Denis Flandre on 13 March 2014.
Abstract-A new design methodology based on a unified treat- micropower CMOS OTA. SO1 technology was chosen because
ment of all the regions of operation of the MOS transistor it offers significant assets for low-power circuit design as
is proposed. It is intended for the design of CMOS analog discussed in [8]-[lo]. This is further exemplified from the
circuits and especially suited for low power circuits where the
moderate inversion region often is used because it provides a perspective of the synthesis methodology as well as from the
good compromise between speed and power consumption. The results gathered from the experimental SO1 OTA.
synthesis procedure is based on the relation between the ratio The paper is organized as follows. In Section 11, we intro-
of the transconductance over dc drain current g m / I D and the duce the proposed design methodology and its main features.
normalized current Io /(W / L ) .The g m /ID indeed is a universal The application to the design of an SO1 micropower CMOS
characteristic of all the transistors belonging to a same process. It
may be derived from experimental measurements and fitted with OTA and validation through experimental results are presented
simple analytical models. The method was applied successfully to in Sections 111 and IV. Section V demonstrates the usefulness
the design of a silicon-on-insulator(SOI) micropower operational of the proposed technique for the quick comparison of the
transconductance amplifier (OTA). performances of various CMOS technologies. The SO1 CMOS
results presented in Section IV are compared to those achieved
1. INTRODUCTION with bulk CMOS and the advantages of SO1 technology
over bulk CMOS for low-power analog circuits are further
HE present trend toward portable equipment as well underlined.
as the increasing circuit density and size of integrated
systems tend to make low power consumption a primary
concern [l], [2]. In CMOS analog circuits the minimum power 11. THE gm/ID METHOD
consumption is achieved when MOS transistors operate in the In the proposed method, we consider the relationship be-
weak inversion region [3]. However, the best compromise tween the ratio of the transconductance gm over dc drain
in terms of consumption and speed is achieved in moderate current ID and the normalized drain current IO I o / ( W / L )
inversion. as a fundamental design tool. The choice of g m / I D is based
On the contrary to complex synthesis tools based on SPICE- on its relevance for the three following reasons.
like models and simulators coupled with optimization routines,
1) It is strongly related to the performances of analog
analytical or quick hand methods for synthesizing analog
circuits.
circuits give an insight on the design problem. However, these 2) It gives an indication of the device operating region.
procedures generally suppose that the MOS transistors are
3) It provides a tool for calculating the transistors dimen-
either in strong inversion or in weak inversion. Mainstream
sions.
methods assume generally strong inversion and use the transis-
tor gate voltage overdrive (GVO) as the key parameter, where In order to illustrate this, let us consider the so-called
GVO = VG - V,, Vi being the gate voltage and V, the “intrinsic gain stage” as a simple example. The intrinsic gain
threshold voltage [4]-[6]. Micropower design techniques, on stage consists of a single transistor in the common source
the other hand, exploit known weak inversion models [33, [7]. configuration loaded by an ideal current source (delivering a dc
This paper presents a new design methodology that allows a current I D )and a capacitance C L .We call gm the small-signal
unified synthesis methodology in all regions of operation of the transconductance and V, the Early voltage which controls the
MOS transistor. It provides an alternative taking full advantage transistor small-signal output conductance, gd = I D/VA.
of the moderate inversion region to obtain a reasonable speed- The dc gain (Ao) and transition unity-gain frequency ( f ~ )
are given by
power compromise. The method exploits the transconductance
over dc drain current ratio ( g m / I o ) relationship versus the
normalized current [ID/(W / L )1.
An example of the proposed methodology is presented in the and
paper considering an experimental silicon-on-insulator (SOI)
fT=Ec,.
1 Sm
(2)
Manuscript received September 19, 1995; revised April 15, 1996.
F. Silveira is with the Instituto de Ingenieria ElCctrica, Universidad de la
Repliblica, Montevideo, Uruguay. The g m / I D ratio is a measure of the efficiency to translate
D. Flandre and P. G. A. Jespers are with the Laboratoire de Mi- current (hence power) into transconductance; i.e., the greater
croklectronique, Universitk Catholique de Louvain, B- 1348 Louvain-la-Neuve,
Belgium. the g m / I D value, the greater the transconductance we obtain
Publisher Item Identifier S 0018-9200(96)06472-4. at a constant current value. Therefore, the g m / I D ratio is
0018-9200/96$05.00 0 1996 IEEE
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 31, NO. 9, SEPTEMBER 1996 1315
35
I I BVDD
30
25
10
the g,/ID ratio is also an indicator of the mode of operation SILICON-ON-INSULATORMICROPOWER CMOS OTA
of the transistor. The proposed methodology was applied to the synthesis
Let us now consider the dependence of 9,110 on tran- of the cascoded OTA [12] shown in Fig. 2 . We assume that
sistor size. The normalized current In is independent of the the total supply current (Itot) is known a priori and equal
transistors size. According to (3) the g m / I D ratio is also to 2 PA. Furthermore, we assume that the load capacitor
size independent. Therefore, the relationship between g m / I D ( C L )is equal to 10 pF and the supply voltage (VDD)equal
and the normalized current is a unique characteristic for all to 3 V. The design procedure is illustrated here aiming at
transistors of the same type (nMOS or PMOS) in a given batch. the best performances in terms of: dc open loop gain (Ao),
Of course, this statement must be revised when dealing with transition frequency ( f ~ ) phase
, margin ( P M ) , and slew
short channel transistors. rate ( S R ) .However, it can straightforwardly be modified to
The “universal” quality of the g m / I D versus In curve can take into account other performance aspects (like noise or
be extensively exploited during the design phase, when the common mode rejection) that may be relevant to the particular
transistor aspect ratios ( W / L ) are unknown. Once a pair of application, as long as they are directly related to the unified
values among g m / I D , gm, and I D has been derived, the g m / I D versus In relationship, i.e., to the current and small-
W / L of the transistor can be determined unambiguously. An signal parameters. For large-signal related performance aspects
example of this procedure is described in the next section. such as signal swing, an I D - V, or g m / I o - V, relationship
1316 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 31, NO. 9, SEPTEMBER 1996
TABLE I
OTA TRANSISTORS
DIMENSIONS
f T
A0 T2 156 57.6
SR
1 Ti- I 1 1
I 0.26
.f Transistor T6I 7 1 1 2 ) 0.52
(w/L) Lengths Early CL 18z.5
T9 9.5 1.58
I T10 I 657.5 1 3 I 243.3 I
T11 I 94 I 3 I 34.6 I
current mirror factor B . The B factor multiplies fT and
SR, as shown in (5), and its maximum value is limited
by its influence on the OTA stability. The designed OTA
has B equal to two.
Choosing the values for g m / I D , 10 is determined for
each transistor from the experimental g m / I D versus
10 curves. Then, with the drain current value found
Fig. 3 . OTA design methodology. in point 1, the W / L of each transistor is found. The
intended values of g m / I ~are chosen accordingly to
is required. This extension of the methodology was applied to their effect on the OTA performance. The dc gain is
predict the output swing. proportional to the product of ( g m / I D ) l and ( g m / I ~ ) 2
The methodology is graphically illustrated in Fig. 3, where as shown in (4); fT is proportional to g,l as shown in
all the data to be provided by the designer are shown in bold (5), therefore, as the current is given, it is proportional
italics. To evaluate An, SR, and fT we use the following to ( h / I D ) l . Hence ( h / I D ) i and ( g m / l D ) 2 should
expressions: be taken as high as possible. The maximum value we
for the gain An: may choose is limited on one hand by the weak inversion
1 maximum value of the technology (about 35 V-' in thin-
. 1 film fully-depleted SO1 MOS transistors and 25 V-l in
+
vA6 . v A 7 v A 9 . VAlO
bulk MOS transistors) and on the other hand by the
stability requirements because as we increase gm / I D ,
where (gnL/lo)l, ( g m / r D ) 2 are the g m / I D ratio
with fixed current, the transistor sizes and parasitic
of the input and cascode transistors, respectively,
capacitances are increased and the phase margin is
V A ~V,A ~V&J,
, and VAlO are the Early voltages of
reduced. The value of the couple [ ( g m / I D ) l , ( g m / 1 0 ) 2 ]
T6, T7, Tg, and Tin. The Early voltages will be
considered proportional to the transistor length with applied in the implemented design was the result of an
exploration of the design space to obtain the combination
a typical value of the constant of proportionality of
that optimizes the trade-off between the dc gain and
7 V/pm. Our experimental results have confirmed this
transition frequency for a given phase margin. The
approximation to be satisfactory for L values in our
chosen values are ( g 7 n / I D ) 1equal to 28 and ( g m / I D ) 2
range of interest from 3-12 ym.
* for SR and the first order approximation of f ~ :
equal to 30, which correspond to an operating point in
the moderate inversion region close to weak inversion.
The current mirror transistors are operated in strong in-
version, with ( g m / 1 ~ ) 3equal to eight, to guarantee good
matching and noise properties.
The transistor lengths are determined by a trade-off between
where B is the current mirror gain shown in Fig. 2, 101 area and stability on one side and dc gain (due to the
is the current through the input pair transistors, and gml dependence of the Early voltage on the transistor length)
is the input transistors transconductance. on the other side. The chosen values are shown in Table 1.
The sequence followed for the synthesis is as follows. Knowing the transistor lengths, all the transistor dimensions
1) The drain current of each transistor is determined from are determined, and the OTA frequency response can be
the specified total current ltot and the choice of the predicted either by a SPICE or a symbolic analysis. When
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 31, NO. 9, SEPTEMBER 1996 1317
TABLE I1
CALCULATED, AND MEASURED
SIMULATED, so1 OTA
RESULTSOF THE MICROPOWER CIRCUIT
* Test version of the OTA with bonding PADS at the sources of the cascode transistors.
TABLE 111
COMPARISON
OF RESULTS
OF g m / I o BASEDSYNTHESIS STRONG
WITH CONVENTIONAL AND WEAKINVERSION
INVERSION SYNTHESIS
a SPICE analysis is applied, a model valid in all regions of The strong inversion synthesis was based on the quadratic
operation (like the one presented in [7]) is selected. If the phase expression for the drain current in saturation as a function
margin ( P M )is not acceptable, the values of B , ( g m / I D ) , or of the gate voltage. Extended to moderate inversion, this
transistor lengths must be modified. approximation is known to overestimate the g,/ID related to
a certain normalized current when compared to the real data.
IV. EXPERIMENTAL
RESULTS So that reciprocally, specified g m / I D and I D are achieved
The resulting design was realized in the 3-pm CMOS by underestimated transistor sizes ( W / L ) and amplifier area
on SO1 process of the Microelectronics Laboratory of the (estimated as the “total” W . L ) as shown in Table 111.
Universitk Catholique de Louvain. The weak inversion synthesis considered the exponential
The dc gain, transition frequency, and phase margin of the approximation for the drain current versus the gate voltage.
circuit prototype were measured. The transconductance of the This approximation predicts a value of gm / I o equal to 35
input and cascode transistors were also determined in order to ( n = 1.1) independent of the current. Therefore, the g m / I D
obtain the g m / I ~values. value does not determines an I D value in this simplified
Table I1 shows the excellent agreement observed between model. The In value must be chosen to guarantee weak
the performance figures calculated by the synthesis program, inversion operation. The criterion applied was to choose Io
calculated by SPICE, and measured in the circuit prototype. ten times smaller than the value corresponding to the limit
The SPICE simulation used the HSPICE level 34 model which between the weak and strong inversion approximations equal
implements the model described in [7], with a set of parameters to 2npC0,U;4, which is a classical criterion to guarantee
optimized to fit the SO1 MOSFET characteristics. weak inversion operation [6]. This procedure leads to an
In order to further demonstrate the interest and potential of overestimation of the transistor sizes required to achieve the
the new methodology, we compare its results with the results specified g n / I D and I D values.
obtained from the conventional design methods limited to the The OTA performances achieved by the SI and WI synthesis
strong and weak inversion regions. These are shown in the are then estimated injecting the transistor sizes they provided
columns “SI synthesis” and “WI synthesis” of Table 111. into the OTA model which uses the real g m / I D versus In
1318 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 31, NO. 9, SEPTEMBER 1996
TABLE IV
SO1 AND BULKLOW-POWEROTA PERFORMANCE COMPARISON
2.9
72.9
2
72.0
2
79.0
2.9
84.0
1
PM (“)
CW.L (pm21 2628 4900 2881 1808
I 35
21
30
19
h
2s
17 g
3
2 20
h
15 & v
13 7 i15
11 9
2 10
9
5
7
, , , , 1
, ,, ,, , , I
, ,
,
,
,
, ,
n
IOOK 1OM
IOOK 1M 1OM
fl(W
Fig. 5. Input pair g m / I D and current consumption versus transition fre-
Fig. 4. Current consumption ratio and dc open loop gain difference between quency for SO1 (solid line) and Bulk (dashed line) CMOS OTA.
the bulk and SO1 OTA versus transition frequency with phase margin of 60
degrees and CI; = 10 pF.
nMOS and PMOS junction capacitances per unit area C,, and
data. These estimations are shown in the columns “SI real” and C,,, and the nMOS and PMOS sidewall capacitances per unit
“WI real” of Table 111. The comparison shows that the strong length C,,,, and C,sw,. In our SO1 technology (resp. Bulk)
inversion approach overestimates the transition frequency by n = 1.1(1.5), C,, = 0.06 fF/pm2 (0.18 fF/pm2), C,, = 0.06
about 50% and the gain by about 7 dB as the reduced transistor fF/pm2 (0.4 fF/pm2), C,,,, = 0.05 @/pm (0.4 fF/pm),
sizes correspond to larger normalized currents and hence Cjswp = 0.05 @/pm (0.5 fF/pm). All other technology data
smaller real g m / I D . The differences are lower if we consider are supposed the same for both technologies: the gate oxide
the weak inversion approach because in the design under thickness (tax = 30 nm), the effective mobility [p, = 500
consideration the input pair and cascode transistors operate e-4
m 2 /(V.s), p, = 190 e-4m2/(V.s)], the drain and source
in moderate inversion close to the weak inversion region, region extensions ( X , = X , = 8pm) which multiplied by
nevertheless, the transition frequency is overestimated by 15% the transistor width give an estimation of the drain and source
and the estimated die area of the resulting design (estimated areas, the source-gate and drain-gate overlap (ovn = ovp =
as the sum of the transistor active areas) is 25% larger than 0.15pm) and the Early voltages (VA, = VA, = 20 V @ 3
the area of the circuit obtained by the g m / I D approach. pm length). The current mirror factor B is equal to two in
both implementations.
v. ADVANTAGES OF THE SILICON-ON-INSULATOR Table IV compares the performance of one bulk and three
TECHNOLOGY FOR LOW POWER ANALOGCIRCUITS SO1 implementations of the OTA. The four designs were op-
In this section we first compare the designed SO1 OTA timized to achieve the same transition frequency, considering
performance with the achievable performance of a similar bulk the same transistor lengths to simplify the comparison. SO11
implementation. Then the current consumption and gain as is the OTA designed in Section I11 with the aim to maximize
a function of the transition frequency are compared for the the gain. This design has therefore the maximum values of
considered OTA architecture in SO1 and bulk technologies. ( g m / I o ) l and ( g m / I ~ ) acompatible with the required phase
This section will demonstrate the usefulness of the pro- margin. The SO12 design sought to trade-off a slight decrease
posed design methodology for the quick comparison of the in gain by a decrease in the OTA die area. We defined, for the
performances of SO1 versus bulk CMOS processes. We only sake of comparison, the objective of having approximately the
need to introduce the technological data differences between same die area as the bulk implementation. The decrease in area
both technologies, i.e., mainly the n factor, the “bottom-plate’’ is obtained with a slight decrease in ( g m / I o ) l and ( g m / I D ) 2 .
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 31, NO. 9, SEPTEMBER 1996 1319
The SO13 design was aimed to have the same consumption is a universal characteristic of all the transistors of the same
and gain as the bulk implementation [i.e., the same values of type (nMOS or PMOS) belonging to the same process. This
(gm/1o)l and ( g m / I D ) a , ] , in order to evaluate the reduction approach allows accurate sizing of analog circuits with a
in die area that can be achieved for the same performances. unified treatment from the strong to the weak inversion region.
Fig. 5 shows the g m / I D of the differential pair transistors Particularly for low-power analog circuits, it is possible to take
as a function of the transition frequency for both technologies, advantage of the moderate inversion region to optimize the
which explain the results of Fig. 4 following our design power-speed compromise.
method. Fig. 5 also shows the current consumption for both The proposed methodology was validated by the design of
technologies. an SO1 micropower OTA. The experimental results showed
The results show that in order to obtain the same transition excellent correspondence with the predicted design data. When
frequency and stability figures as SO11, the bulk implemen- compared with the results achievable in bulk CMOS technolo-
tation requires an increase of 45% in current consumption, gies, it appears that the increased g m / I D values and lower
whereas the gain is about 8 dB lower. The lower g m / I D parasitic capacitances of fully depleted SO1 devices allow a
values achievable in bulk have indeed to be compensated reduction of 72% of consumption with increased gain for the
by an increase of I D to maintain the same gm. The SO11 same bandwidth.
OTA occupies, however, a larger die area than the bulk
implementation due to the fact that the SO1 lower parasitic ACKNOWLEDGMENT
capacitances allow to work closer to the weak inversion
region (i.e., with bigger transistors for a given current) with The authors wish to thank J.-P. Eggermont and B. Gentinne
acceptable phase margin. Nevertheless, the SO12 OTA which for their contributions in device characterization.
has about the same area as the bulk OTA, still provides a much
higher gain with a reduced power dissipation and an increased REFERENCES
phase margin than the bulk OTA. The design SO13 shows [I] E. A. Vittoz, “Low-power low-voltage limitations and prospects in
that if the same performances as in the bulk implementation analog design,” in Proc. Workshop Advances in Analog Circuit Design,
are desired, the estimated die area is reduced by about 40%, Eindhoven, Mar. 1994.
[2] 2. J. Lemnios and K. J. Gabriel, “Low-power electronics,” IEEE Design
without taking into account the area savings in SO1 technology and Test of Computers, vol. 11, no. 4, pp. 8-13, Winter 1994.
due to the absence of wells. [3] E. A. Vittoz and J. Fellrath, “CMOS analog integrated circuits based on
Fig. 4 shows the current consumption ratio and dc open loop weak inversion operation,” IEEE J. Solid-State Circuits, vol. SC-12, no.
3, pp. 224-231, June 1977.
gain difference between the bulk and SO1 implementations of [4] P. R. Gray and R. G. Meyer, “MOS operational amplifier design-A
the OTA as a function of the specified transition frequency tutorial overview,” IEEE J. Solid-state Circuits, vol. SC-17, no. 6, pp.
657-665, Dec. 1982.
for the same phase margin of 60 degrees. As higher transition [5] P. E. Allen and D. R. Holberg, CMOS Analog Circuit Design. New
frequencies were sought, all the transistors were taken of min- York Holt Rinehart and Winston, 1987.
imum length (3 pm) in both SO1 and bulk implementations, [6] K. R. Laker and W. M. C. Sansen, Design ofAnalog Integrated Circuits
and Systems. New York: McGraw-Hill, 1994.
and the g m / I o of the current mirrors was taken equal to five. [7] E. A. Vittoz, “Micropower techniques,” in Design of VLSZ Circuits
The above examples show that the increased g m / I D values for Telecommunications and Signal Processing, J. E. Franca and Y. P.
provided by the SO1 technology, as well as the lower parasitic Tsividis, Eds. Englewood Cliffs, NJ: Prentice-Hall, 1993.
[8] D. Flandre, J.-P. Eggermont, D. De Censter, and P. Jespers, “Compari-
capacitances which allow to use higher values of g m / I D with son of SO1 versus bulk performances of CMOS micropower single-stage
acceptable phase margin lead to a reduction in current con- OTA’s,” Elm. Lett., vol. 30, no. 23, pp. 1933-1934, Nov. 10, 1994.
sumption of up to about 72% with a simultancous increase in [9] J. P. Colinge, J. P. Eggermont, D. Flandre, P. Francis, and P. G. A.
Jespers, “Potential of SO1 for analog and mixed analog-digital low-
gain of up to 22 dB. On the other hand, the same performances power applications,” in IEEE Proc. ISSCC, San Francisco, Feb. 1995,
as in bulk technology can be achieved with a much smaller pp. 194-195.
[IO] D. Flandre, L. F. Ferreira, P. G. A. Jespers, and J.-P. Colinge, “Modeling
circuit. and application of fully-depleted SO1 MOSFET’s for low-voltage low-
power analog CMOS circuits,” accepted for publication in Solid-State
Electronics, Special Issue on Low-Power Electronics.
VI. CONCLUSION [ 111 F. Forti and M. E. Wright, “Measurement of MOS current mismatch in
The presented design methodology exploits the fact that the the weak inversion region,“ IEEE J. Solid-state Circuits, vol. 29, pp.
138-142, Feb. 1994.
relationship between the ratio of the transconductance over dc [12] F. Krummenacher, “High voltage gain CMOS OTA for micropower SC
drain current g m / I D and the normalized current I D / ( W / L ) filters,” Elec. Lett., vol. 17, no. 4. pp. 160-162, Feb. 19, 1981.