You are on page 1of 4

Analog Integrated Circuits and Signal Processing, 40, 71–74, 2004


c 2004 Kluwer Academic Publishers. Manufactured in The Netherlands.

Very Low Voltage MOS Translinear Loops Based on Flipped Voltage Followers

ANTONIO J. LÓPEZ-MARTÍN,1,∗ ALFONSO CARLOSENA1 AND JAIME RAMIREZ-ANGULO2


1
Department of Electrical and Electronic Engineering, Public University of Navarra, Campus Arrosadı́a. E-31006, Pamplona (Spain)
2
Klipsch School of Electrical and Computer Engineering, New Mexico State University, Las Cruces, NM 88003, USA
E-mail: antonio.lopez@unavarra.es; carlosen@unavarra.es; jramirez@nmsu.edu

Received November 13, 2002; Revised December 23, 2002; Accepted December 23, 2002

Abstract. A novel technique for operating MOS Translinear loops at very low supply voltages is described, based
on the use of Flipped Voltage Followers for biasing the loops. The resulting topologies, suited to standard CMOS
processes, can be successfully applied to a varied repertory of low-voltage analog circuits, such as squarers, multi-
pliers, filters, oscillators, and RMS-DC converters. Measurement results for a geometric-mean and a squarer/divider
circuit demonstrate on silicon the usefulness of this technique.

Key Words: voltage-translinear circuits, low-voltage analog circuits, analog CMOS circuits

1. Introduction as Voltage-TL loops [2]. A novel biasing scheme


based on the Flipped-Voltage Follower (FVF) cell [5]
The ongoing reduction of supply voltage driven by leads to a significant performance improvement and
submicron digital CMOS circuits and battery-operated to reduced supply voltage requirements. The resulting
equipment forces CMOS analog circuits aimed to topology can replace conventional ones employed for
mixed-mode designs to conform to this trend. Analog implementing static nonlinear computational circuits
signal processing at supply voltages close to the MOS (e.g., geometric-mean, squarer/divider, multiplier, vec-
threshold voltage is best performed in current-mode, tor normalization) or dynamic linear and nonlinear cir-
so that voltage swings are nonlinearly compressed and cuits (e.g., companding filters, oscillators, RMS-DC
therefore become less constrained by supply voltage converters) using the techniques described in [3, 6].
limitations. Both static and dynamic circuits based on
the TransLinear (TL) principle [1] show this current-
mode behavior. Unfortunately, most of their proposed 2. Circuit Description
CMOS implementations do not optimally exploit the
potential of this technique for very low supply voltage Figure 1 shows a folded second-order MOS TL loop,
(e.g., [2, 3]). where it is assumed that all transistors are operating in
Recently a TL technique in CMOS technology has strong inversion and saturation. The sum of voltages
been proposed [4], suited to very low supply volt- around the loop is zero, so that
ages. Loop transistors are operated in weak inversion,
which leads to low power consumption but also to VGS1 + VGS2 = VGS3 + VGS4 (1)
poor matching characteristics and bandwidths in the
order of tens of kHz. In this Letter we propose an al- Using the MOS square law, the MOS gate-to-source
ternative approach that does not have the aforemen- voltage V GS is related to its drain current I D by
tioned limitations and still allows low-voltage opera- 
tion. Loop transistors are operated in strong inversion 2 
VGS = I D + VTH (2)
and saturation, so that the MOS square law is exploited, kn (W/L)
leading to the so-called MOS TL loops, also known
where all the parameters have their usual meanings.
∗ Corresponding author. Using (2) in (1) and assuming that all transistors are
72 López-Martı́n, Carlosena and Ramirez-Angulo

From (1), the operation of TL loops only deter-


mines the relative values of nodal voltages, but they
are not set in an absolute sense. However, in prac-
tice a reference DC voltage must be set in some node
of the loop (hence determining that of all the nodes)
for proper operation. This voltage is usually set by
the additional circuitry that is anyway necessary for
injecting the loop currents with the functional depen-
Fig. 1. Second-order folded Voltage-Translinear Loop. dence required for a certain application (e.g., Eq. (4)
in our case). In particular, the well-known structure of
Fig. 2(a) [2, 3] is frequently employed for this purpose.
equal, loop currents become related by However, the diode-connected MOSFET of the cur-
    rent mirror precludes very low voltage operation due
I1 + I2 = I3 + I4 (3) to the stacking of two gate-source voltages. Moreover,
the impedance at its drain terminal, where the reference
Hence, several nonlinear current-mode functions can DC level is established, is not very low. This fact makes
be implemented by properly injecting such currents. the loop nodal voltages dependent on the current sink
For instance, if we force by this diode-connected transistor, thus notably lim-
I1 + I2 + 2I5 iting the range of current levels achievable for a cer-
I3 = I4 = (4) tain supply voltage. In addition, it influences the loop
4
bandwidth.
being I5 a certain current, after squaring both sides in The novel topology shown in Fig. 2(b) imple-
(3) and rearranging it can be shown that the TL loop ments (5) solving all these issues. Now a FVF, formed
forces this current I5 to be by transistors M F1 − M F2 and current source I B , sets
 the proper DC voltage at the loop nodes. It is a very
I5 = I1 I2 (5) simple feedback circuit that creates an extremely low
impedance node (the source of M F1 ) with an ac-
Therefore, a geometric-mean circuit is obtained if I1 curately controllable DC voltage (determined by I B
and I2 are the input currents and the output current is a and the dimensions of transistor M F1 ). Hence volt-
copy of I5 . Alternatively, a squarer/divider is obtained age at this node is kept essentially constant regard-
if the output is a copy of I2 and the inputs are I5 and I1 . less of the input and output current levels. This fact

Fig. 2. Biasing of Voltage-Translinear loops: (a) conventional approach and (b) Proposed approach.
Very Low Voltage MOS Translinear Loops Based on Flipped Voltage Followers 73

Fig. 3. Measured and ideal geometric-mean outputs for different Fig. 4. Measured and ideal outputs of the squarer/divider circuit.
input currents.

2.5%. A similar accuracy is observed for the geometric-


enhances the sinking capability of transistor M F2 mean cell. The total silicon area employed for ei-
and improves loop bandwidth. If the aspect ratio ther the geometric-mean circuit or the squarer/divider
of M F1 is properly chosen, I B can be made small, circuit (pads excluded) was 0.09 mm2 . The power
so that the increase in current consumption is very consumption in both cases was 210 µW, using 10-
modest. µA DC input currents. Bandwidth was only simu-
lated due the influence of breadboard parasitics. It
was relatively poor (approx. 5 MHz in both cases)
3. Measurement Results due to the large gate length of the loop transistors.
Nevertheless, the bandwidth of the geometric-mean
In order to validate the proposed technique, a circuit and the squarer/divider circuit using the con-
geometric-mean circuit and a squarer/divider circuit ventional configuration of Fig. 2(a) and identical tran-
were fabricated in a 2.4-µm DPDM n-well CMOS pro- sistor dimensions and bias currents was even lower
cess. Both topologies correspond to Fig. 2(b), being (3.9 MHz), so that the FVF cell led to a 28% increase in
the only difference between them that the output cur- bandwidth.
rent was a copy of I5 for the geometric-mean cell and A better technology allows further reduction of the
a copy of I2 for the squarer/divider cell, as mentioned supply voltage. Simulation results using BISM3v3
previously. Simple current mirrors were employed for models for a 0.8-µm DPDM n-well CMOS technology
injecting the currents shown in Fig. 2(b). The aspect show similar performance using a single 1.2-V supply
ratio of the loop transistors was 80 µm/4.8 µm, and voltage.
bias current was I B = 2µA. The supply voltage em-
ployed was 1.5 V. Figure 3 shows the output current
of the geometric-mean cell for input current I1 varying 4. Conclusions
from 0 to 10 µA and input current I2 stepped from
2 to 6 µA in 2 µA steps. The theoretical response A novel technique for MOS Translinear loops is
is shown in the dotted curves, whereas the measured proposed, that allows operation at very low sup-
one corresponds to the solid curves. Note how the ply voltages. It can be employed in standard CMOS
circuit can operate as a variable-gain square-rooting processes, so that it is an interesting choice in
cell. mixed-signal designs. Experimental results for a
Figure 4 shows the output of the squarer/divider cell geometric-mean cell and a squarer/divider fabri-
for input currents I5 = 10µA and I1 varying from 0 cated using this technique validate the proposed
to 10 µA. The relative error of the output is less than approach.
74 López-Martı́n, Carlosena and Ramirez-Angulo

Acknowledgment 3. A.J. Lopez-Martin and A. Carlosena, “Current-mode multi-


plier/divider circuits based on the MOS translinear principle.”
This work has been supported by the Spanish Dirección Analog Integrated Circuits and Signal Processing, vol. 28, no. 3,
pp. 265–278, 2001.
General de Investigación under grant TIC2003-07307- 4. E. Seevinck, E. Vittoz, M. Du Plessis, T.H. Joubert, and W.
C02-02. Beetge, “CMOS translinear circuits for minimum supply volt-
age.” IEEE Tran. Circ. Syst. II, vol. 47, no. 12, pp. 1560–1564,
2000.
References 5. J. Ramirez-Angulo, R.G. Carvajal, A. Torralba, A. Galan, A.P.
Vega-Leal, and J. Tombs, “The flipped voltage follower: A useful
1. B. Gilbert, “Translinear circuits: An historical overview.” Analog cell for low-voltage low-power circuit design,” in Proc. ISCAS
Integrated Circuits and Signal Processing, vol. 9, no. 2, pp. 95– 2002, Phoenix, AZ, pp. III615–III618.
118, 1996. 6. A.J. Lopez-Martin and A. Carlosena, “Systematic design of com-
2. E. Seevinck and R.J. Wiegerink, “Generalized translinear circuit panding systems by component substitution.” Analog Integrated
principle.”IEEE J. Solid-State Circuits, vol. 26, no. 8, pp. 1098– Circuits and Signal Processing, vol. 28, no. 1, pp. 91–106,
1102, 1991. 2001.

You might also like