1-V Continuously Tunable CMOS Bulk-Driven

Transconductor for Gm-C Filters

Juan M. Carrillo, J. Francisco Duque-Carrillo Guido Torelli
Dept. of Electronics Dept. of Electronics
University of Extremadura University of Pavia
06071 Badajoz, Spain 27100 Pavia, Italy
jmcarcal@unex.es guido.torelli@unipv.it

Abstract—A bulk-driven transconductor including a continuous II. 1-V CMOS BULK-DRIVEN TRANSCONDUCTOR
tuning scheme and operated from a 1-V supply is presented in
this paper. The voltage-to-current conversion is carried out by The proposed fully differential (FD) transconductor is
means of the source degeneration of a bulk-driven differential illustrated in Fig. 1, where Vbn, Vbp, Vcn, and Vcp are constant
pair, while transconductance tuning is obtained by modifying the bias voltages, Vcm is the CM control voltage, and Vtn is a
gain of a current mirror. The proposed circuit has been included tuning voltage, as shown below. The shadowed area in the
in a continuous-time transconductance-C (Gm-C) biquadratic cell middle of the scheme corresponds to the transconductor core,
designed for the audio bandwidth, which features a where the V-I conversion is carried out. The operation
programmability range of more than one decade and a dynamic principle is based on generating a copy of the input differential
range of 63.7 dB while consuming less than 45 μW. voltage (Vi,dm) across a resistor R, with the help of two
additional voltage followers M1L to M3L and M1R to M3R
I. INTRODUCTION [4]. The differential current IR flowing through the resistor is
The interest in low-voltage low-power analog integrated equal to (VA – VB)/R. Therefore, if VA and VB track Vi+ and Vi−,
circuits includes the design of transconductors, which can be respectively, the current generated is proportional to the input
used in applications such as analog multipliers or continuous- differential voltage. This current may be mirrored into the
time transconductance-C (Gm-C) filters. Important features of a output branches of the transconductor, thus obtaining a
transconductor are a very linear voltage-to-current (V-I) transconductance value equal to 2/R, provided that the voltage
characteristic, which must be maintained even for large input gain of the voltage followers is equal to unity.
differential signals in order to obtain an acceptable signal-to- The voltage buffers M1L to M3L and M1R to M3R
noise ratio (SNR), as well as Gm tunability, which allows operate as improved source followers [5], in such a way that a
compensating for fabrication process spreads and/or providing copy of the corresponding input voltage (Vi+, Vi −) is present at
the corresponding application with programmability. the source terminal of the input transistors (M1L, M1R). The
With a supply voltage in the order of VGS + VDSat, it is not operation of these blocks is enhanced thanks to the negative
possible to set the common-mode (CM) input voltage of a feedback involving transistors M1L-M2L and M1R-M2R,
conventional gate-driven transconductor to midsupply (VAGND). respectively, which may be easily made stable [4]. Indeed, the
This fact may severely limit the signal swing, as the use of feedback reduces the value of the output resistance of the
output cascode transistors is often mandatory in order to buffer, thus decreasing the loading effect of resistor R. The use
ensure an accurate V-I conversion. The use of bulk-driven of this configuration leads to two additional advantages. On
MOS transistors has been proposed as an effective approach to the one hand, transistors M2L and M2R may be used as the
operate in such stringent supply conditions [1]-[3]. The input input branch of a current mirror in order to deliver a replica of
voltage range of these devices is higher than in the case of a the differential current IR to the transconductor outputs. On the
gate-driven approach, even though also the input referred noise other hand, a constant current (biasing current IB in Fig. 1)
is larger due to the lower value of the bulk transconductance, flows through the input devices with the consequent linearity
gmb, with respect to the gate transconductance, gm [2]. improvement with respect to a conventional source follower.
Therefore, the real advantage of using bulk-driven devices in As observed in Fig. 1, bulk-driven MOS transistors are used
the design of linear transcondutors is the possibility of setting as input devices in order to set the input CM voltage of the
the DC input/output level at midsupply, thus maximizing the transconductor to midsupply and obtain a wide input differential
signal swing even under low-voltage conditions. This paper voltage swing. Nevertheless, it must be taken into account that
presents a 1-V continuously tunable transconductor based on the value of gmb is from 2 to 5 times smaller than gm.
the source degeneration of a bulk-driven input pair.
This work was co-supported by the Spanish Secretaría de Estado de
Universidades e Investigación del Ministerio de Educación y Ciencia (MEC)
under Mobility Grant PR2007-0509 and MEC under Grant TEC2006-13154.

978-1-4244-1684-4/08/$25.00 ©2008 IEEE 896

 VDD

IB+IR Cz IB–IR
M7L Vbp M2L M2R Vbp M7R
M9L M9R
IBP R IBP
Vcp A B Vcp
M9LC M5L M4L + − M4R M5R M9RC
Vo
+
D C Vi IR Vi E F Vo
−

Vcn Vcn
Io/2 Io/2
M1L M1R
M8LC M8RC
V cm V tn Vbn Vbn Vtn Vcm
M8L ITN IBP - ITN IB IB IBP - ITN ITN M8R
M6L M3L M3R M6R

Continuous Tuning Transconductor Core Continuous Tuning

Figure 1. 1-V bulk-driven transconductor with continuous Gm tuning.

VDD one for each branch of the FD transconductor. The tuning

Vb,cmfb Vb,cmfb
Vb,cmfb IBCM IBCM Vb,cmfb circuit consists of a differential pair, M4L-M5L (M4R-M5R),
IBM IBM connected in unity-gain voltage follower configuration, which
+ VAGND −
Vo Vo allows voltage VD (VF) to track VC (VE). When the value of the
MC1 MC2 MC3 MC4
current ITN, provided by transistor M6L (M6R), is one half the
MC7C MC8C value of IBP, sourced by M7L (M7R), the differential pair is
Vcm fully balanced and VD (VF) is ideally equal to VC (VE). Hence,
MC7 MC8
the current mirror formed by M2L and M9L (M2R and M9R)
has a current gain equal to unity. Nevertheless, if the current
Figure 2. CMFB circuit. ITN is made different from IBP/2, the gain of the above current
mirrors may be continuously tuned. Indeed, as ITN is smaller,
VDD the source-to-gate voltage of M5L (M5R) becomes smaller,
Vb,cmfb M7r Vbp M2r VD (VF) goes closer to VDD with respect to VC (VE) and, hence,
To IBP attenuation in the mirrored output current takes place.
CMFB Similarly, amplification is achieved if ITN is larger than IBP/2.
M5r M4r V
AGND
The CM component of the output voltage is stabilized with
M1rthe help of the CM feedback (CMFB) circuit illustrated in Fig.
Vtn Vbn 2, which is based on a differential difference amplifier [6].
ITN IB
M6r M3r Again, a bulk-driven approach is followed for input devices
MC1 to MC4 in order to manage signals around midsupply
Figure 3. Replica circuit used to bias the CMFB circuit as a function of the
and not to limit the transconductor output swing. The CMFB
operation of the tuning sections. circuit operates by current steering (signal Vcm) and, hence, the
biasing current IBCM is chosen as a matched fraction of the
biasing current IB used in the core of the transconductor. The
Therefore, the voltage gain of the follower circuits in the input branch of the low-voltage current mirror MC8-MC8C,
shadowed area of Fig.1 is not equal to unity, but is given by which provides the signal Vcm, is biased by current IBM.
the ratio gmb,i/(gmb,i+gm,i), where gmb,i and gm,i are the bulk and Transistors MC7 and MC7C are included as dummy devices.
the gate transconductance, respectively, of transistors M1L The use of the continuous tuning scheme in the transconductor
and M1R. This ratio has a value of around 1/6 in our case as, of Fig. 1 makes its output quiescent current dependent on the
in the used technology, gmb is around 5 times smaller than gm. tuning level. Therefore, the biasing currents in the CMFB
The effective transconductance of the transconductor in Fig. 1 circuit, IBCM and IBM, must also be a function of the tuning
may thus be expressed as: circuit operation. A replica of half the transconductor input
branch and the tuning section, illustrated in Fig. 3, is used in
g mb ,i order to properly modify the biasing currents of the CMFB
I 2I R 2
Gm ≡ o = = ⋅ (1) circuit. The transistors and currents in the CMFB and in the
Vi, dm Vi ,dm g mb ,i + g m,i R static replica circuit were down scaled with respect to the
corresponding ones in Fig. 1 in order to save power and area.
The use of a passive resistor to implement R leads to better The scaling factor must be chosen in such a way that the gain
linearity performance as compared to the case when an MOS and gain-bandwidth product of the CM signal path are higher
transistor operating in the linear region is employed. than their counterparts in the differential-mode (DM) path.
Nevertheless, its value cannot be electronically modified and,
hence, an additional tuning circuit must be used to provide the Capacitor Cz in Fig. 1 is used for excess phase compensation
transconductor with programmability. The proposed tuning by generating a left half-plane zero, while cascode transistors
scheme is illustrated in the circuit sections enclosed in a have been included to increase the transconductor DC gain.
dashed box in Fig. 1. There are two identical tuning circuits, Although this solution leads to some limitations in the output

897
 gm4 Io (μA) I TN ( μA)
− + 4.0 1.8 μ A
+ − 1.6 μA

1.4 μ A

2.0 1.2 μA
1.0 μ A
gm1 C1 gm2 C2 0.8 μA
+ + − + − + 0.6 μ A
− 0.4 μA
Vi − + + Vo,bp − +
Vo,lp 0.0
0.2 μ A
− − 0.1 μA

C1 C2
gm3 -2.0
− +
+ −

-4.0
Figure 4. Gm-C biquad with fo and Q programmability. -1.0 -0.5 0.0 0.5 1.0
Vi,dm (V)
(a)
voltage swing, an increase of the DC gain is required in order gm,eff (μA/V)
4.5 ITN (μA)
to obtain a high accuracy in the V-I conversion. Finally, it is
4.0 1.8 μA
also important to mention that the noise behavior of the
transconductor is dominated by the contributions of the 3.5
1.6 μA
current source transistors M3L-M3R and M6L-M6R. Flicker 3.0
noise may be reduced by increasing the length of these 2.5 1.4 μA

transistors. Thermal noise is minimized by decreasing their 2.0 1.2 μA

aspect ratio, taking into account that the minimum size of each 1.0 μA
1.5
device is determined by the maximum allowed value of VDSat. 0.8 μA
1.0 0.6 μA
0.4 μA
III. SECOND-ORDER FULLY Gm-C FILTER 0.5
0.2 μA
0.1 μA
A common approach to implement high-order filters is 0.0
-1.0 -0.5 0.0 0.5 1.0
based on the cascade connection of first- and second-order Vi,dm (V)
sections. For this reason, the performance of the proposed (b)
transconductor has been evaluated by means of the Gm-C Figure 5. Simulated (a) output current and (b) transconductance of the
biquadratic cell shown in Fig. 4, which provides lowpass (Vo,lp) transconductor vs. Vi,dm.
and bandpass (Vo.bp) responses and allows programmability of
the cutoff frequency, fo, and the quality factor, Q. The transfer
functions of the lowpass and bandpass filter, respectively, are: 1-V single supply in standard 0.35-μm CMOS technology,
with nominal values of the threshold voltages equal to 0.5 V
v o,lp (s ) g m1 g m 2 C1C 2 and –0.6 V for NMOS and PMOS transistors, respectively.
H lp ( s) ≡ = (2a) The design phase was carried out taking fabrication process
v i ( s) s 2 + g m3 C1 ⋅ s + g m 2 g m 4 C1C 2 spreads as well as variations in supply voltage and temperature
v o ,bp ( s ) g m1 C1 ⋅ s into account. The transconductance attenuation due to the use
H bp (s ) ≡ = (2b) of bulk-driven transistors, expressed in (1), makes the solution
v i ( s) s + g m3
2
C1 ⋅ s + g m 2 g m 4 C1C 2
suitable for operation in low and medium frequency ranges.
where gmi (i = 1 to 4) represents the transconductance of the i-th Therefore, the circuit was optimized to operate in the audio
transconductor in Fig. 4 and C1 and C2 are the integrating band. The value of the biasing current IB in the transconductor
capacitors. Therefore, the expressions of the gain, fo, and Q are core was 800 nA, while current IBP and the middle value of ITN
in the tuning sections were set equal to 2 μA and 1 μA,
H lp (0) = g m1 g m 4 (3a) respectively (smaller values of IBP and ITN were not used in
order to minimize the noise contribution of the tuning circuit).
H bp (ω o ) = g m1 g m3 (3b)
The scaling factor for the CMFB and the replica circuit was
f o = 1 2π ⋅ g m 2 g m 4 C1C 2 (3c) selected to be equal to 4, which allowed a similar gain and
gain-bandwidth product for the CM loop and the DM path.
Q = g m 2 g m 4 g m2 3 ⋅ C1 C 2 (3d) The value of the polysilicon resistor R was set to 200 kΩ and
the capacitor Cz was equal to 2 pF, while the value of the input
The filter can be designed by taking into account equations CM voltage was 0.5 V (midsupply).
(3) and the value of the effective transconductance, given by
(1). C1 was set equal to 2C2 so that a maximally flat response, The effective value of the transconductance in the case of a
i.e., Q = 1/¥2, is obtained when all transconductors are balanced operation of the tuning circuit, i.e., ITN = 1 μA, was
identical and biased with the same currents. equal to 1.86 μA/V, while a range of Gm values from 0.20
μA/V to 3.87 μA/V was observed when the biasing current ITN
IV. SIMULATED RESULTS was moved from 100 nA to 1.8 μA. For higher values of ITN,
The transconductor in Fig. 1, including the CMFB (Fig. 2) the current source M6L (M6R) starts to work in the linear
and the replica circuit (Fig. 3), was designed to operate with a region and the tuning circuit is not effective. Nevertheless, the

898
 20
TABLE I. SIMULATED PERFORMANCE OF THE 1-V Gm-C BIQUAD
(TECHNOLOGY: 0.35-μM CMOS, IB,1-4 = 800 nA, IBP,1-4 = 2 μA, ITN,1-4 = 1 μA)
0
Parameter Value
fo 17.5 kHz
magnitude (dB)
-20
Q 0.890
fo programmability 1.8 – 36.5 kHz
ITN,1-4 = 0.1 μA
-40 Q programmability 0.045 – 21.000
ITN,1-4 = 0.6 μA
ITN,1-4 = 1.0 μA Input referred noise (1 Hz – 18 kHz) 345 μVrms
-60 ITN,1-4 = 1.8 μA Vin for –40dB-THD @ fi = 1 kHz 1.50 Vpp
Vin for –40dB-IM3 @ fi1 = 14.5 kHz, fi2 = 15.5 kHz 0.88 Vpp
Dynamic Range (–40dB-THD) 63.7 dB
-80
10
2
10
3
10
4
10
5
10
6 Dynamic Range (–40dB-IM3) 59.1 dB
frequency (Hz) DC power consumption 44.3 μW
(a)
20

0
The simulated performance of the second-order filter is
provided in Table I, where all parameters refer to the lowpass
magnitude (dB)

-20
response except for the Q factor, evaluated for the bandpass
configuration. The input referred noise of the lowpass cell in
-40 the 1 Hz to 18 kHz band was 345 μVrms. If the –40dB-THD
ITN,1-4 = 0.1 μA
ITN,1-4 = 0.6 A
value is taken as a distortion limit to determine the maximum
-60 ITN,1-4 = 1.0 μA input signal, the dynamic range (DR) of the Gm-C lowpass filter
ITN,1-4 = 1.8 μA
is 63.7 dB for the case of a 1-kHz input sinewave. The DR value
-80 ranges between 67.3 dB and 54.7 dB when fo is programmed
2 3 4 5 6
10 10 10
frequency (Hz)
10 10 from 1.8 kHz to 36.5 kHz. Besides, when the –40dB-IM3 for
(b) two input tones of 14.5 kHz and 15.5 kHz is used to set the
Figure 6. Gm-C filter fo programability for (a) the lowpass and maximum allowed input signal, the DR is equal to 59.1 dB.
(b) the bandpass response.
V. CONCLUSION
lower bound of the tuning range is not given by the value The use of bulk-driven MOS transistors is a good
indicated above, which has been selected for practical reasons: alternative to design a linear transconductor operating with a
simulations showed that even for ITN values as low as 1 nA, low-voltage supply, as it allows setting the input CM voltage
the circuit is still working with an adequate performance. This to midsupply. A 1-V transconductor based on source
feature is due to the fact that transconductance tuning does not degeneration of a bulk-driven differential pair has been
modify the biasing current of the V-I converter core. The DC introduced. The solution presents continuous tuning capability,
V-I characteristics of the transconductor over the whole input which allows higher accuracy than discrete programmability.
differential voltage, Vi,dm, range are illustrated in Fig. 5(a), The suitability of the approach has been validated by the
while the effective transconductance is shown in Fig. 5(b). design of a 1-V continuous-time Gm-C biquadratic cell.
Both plots are depicted as a function of the tuning current ITN,
which is swept in a range from 100 nA to 1.8 μA. A good ACKNOWLEDGMENT
linearity is obtained, and the tunability range is large enough The authors wish to thank the support of the Integrated
not only to compensate for typical process spreads and Microsystems Laboratory of the University of Pavia (Italy).
possible variations due to the application of the circuit, but
also to provide wide programmability in filtering applications. REFERENCES
A non-linear gm,eff vs. ITN relationship exists, but this fact does [1] A. Guzinski, M. Bialko, and J. Matheau, “Body driven differential
not introduce additional distortion in the signal processing. amplifier for application in continuous-time active-C filter,” in Proc.
European Conf. Circ. Theory and Design (ECCTD), pp. 315–319, 1987.
The Gm-C biquad in Fig. 4 was designed in the same [2] B.J. Blalock, P.E. Allen, and G.A. Rincon-Mora, “Designing 1-V op
technology as above to operate also with a 1 V supply. The amps using standard digital CMOS technology,” IEEE Trans. Circuits
values of capacitors C1 and C2 were set equal to 10 pF and 20 Syst. II, vol. 45, pp. 769-780, July 1998.
pF, respectively, which, along with the transconductance of [3] J.M. Carrillo, G. Torelli, R. Pérez-Aloe, and J.F. Duque-Carrillo, “1-V
rail-to-rail CMOS opamp with improved bulk-driven input stage,”
the V-I converters, leads to a nominal cutoff frequency of 20.9 IEEE J. Solid-State Circuits, vol. 42, pp. 508-517, March 2007.
kHz. Figure 6 shows the fo programmability of the Gm-C [4] A.J. López-Martín, J. Ramirez-Angulo, C. Durbha, and R.G. Carvajal,
biquad for the lowpass [Fig. 6(a)] and the bandpass [Fig. 6(b)] “A CMOS transconductor with multidecade tuning using balanced
response. The representation of the Q programmability has current scaling in moderate inversion,” IEEE J. Solid-State Circuits,
been omitted for room reasons. From the figure, the simulated vol. 40, pp. 1078-1083, May 2005.
[5] J. Ramirez-Angulo, R.G. Carvajal, A. Torralba, J. Galán, A.P. Vega-
lowpass cutoff frequency when ITN,1-4 = 1 μA is equal to 17.5 Leal, and J. Tombs, “The flipped voltage follower: a useful cell for
kHz. The difference with respect to the expected value is due low-voltage low-power circuit design,” in Proc. IEEE Int. Symp.
to the parasitic capacitors at the transconductors output nodes. Circuits Syst. (ISCAS), vol. 3, pp. 615-618, Arizona, May 2002.
However, the presence of these parasitics can be counteracted [6] J.F. Duque-Carrillo, “Control of the common-mode component in
CMOS continuous-time fully differential signal processing,” Analog
thanks to the large tuning capability of the transconductors. Integr. Circ. Sig. Proc., vol. 4, pp. 131-140, 1993.

899