CMOS Current-Mode Function Generator

for Analog Signal Processing

Cosmin Popa
Faculty of Electronics, Telecommunications and Information Technology, Bucharest

Abstract – A new high-accuracy current-mode function circuit error has important values as a consequence of using
generator circuit will be presented, the proposed computational only the second-order approximation of the implemented
structure being designed for implementing a multitude of circuit functions. The range of the input signal is strongly restricted
functions with many applications in VLSI designs. The from the same reasons.
realization of the original function generator is based on the
Most of function generator circuits are dedicated
utilization of a new third-order approximation function that
allows to strongly reduce the approximation error for the exclusively to the implementation in CMOS technology of the
implemented functions. The circuit requires a much smaller exponential function [15]-[28]. An important class of
silicon area comparing to similar previouslys reported circuits. exponential generators [15]-[18] uses the second-order limited
The current-mode operation and the independence of the output Taylor expansion for approximating the exponential function.
current of the function generator on technological parameters This method presents the advantage of simplicity, but also
and on temperature variations additionally increase the overall with a great disadvantage of an increased approximation error
circuit accuracy. The simulations confirm the theoretical and of a limited dynamic output range. Third-order
estimated results, showing an accuracy of the functional squaring approximated exponential function generators [19] present
core equal with 0.97 %. The new proposed function generator
superior performances, but their complexity is strongly
based on the third-order approximation function permits to
obtain a wide output dynamic range of 28dB for a maximal error increased comparing to the previous class of generators, they
of 1dB, when implementing the exponential function. The usually requiring at least two or three current squaring circuits.
proposed function generator is designed for a low-voltage Another possibility of implementing the exponential function
operation, being supplied at VDD=1V. is based on the utilization of particular approximation
functions [20]-[22] that usually allow the second-order
Index terms – function generator, approximation error, approximation of the exponential function. These circuits
circuit accuracy, exponential function, current-mode operation present the same disadvantage of a relatively large
I. INTRODUCTION approximation error caused by the second-order
approximation.

T he new approach of designing signal processing circuits

from an analog perspective represents an important trend
in latest VLSI designs. The requirements of a low-power
The original proposed function generator is dedicated to the
implementation in CMOS technology of a multitude of
continuous mathematical functions. The new method is based
operation, correlated with a high-speed operation that allows a on the utilization of an original approximation function that is
real-time signal processing, impose the analog signal able to third-order approximate the required functions. As a
processing structures especially for deserving portable and result, the error can be strongly decreased comparing to the
medical equipment. There exist a large number of circuit previous reported function generators. The current-mode
functions that can be implemented in CMOS technology: operation of the proposed circuit and the independence of its
signal amplifying [1]-[3], squaring [4]-[6] or square-rooting output current on technological parameters and on temperature
[7], [8] structures, Euclidean distance [9], [10] or contribute to an additional increasing of the function generator
multiplier/divider circuits [11]-[12]. accuracy. The original function generator presents the
There were reported in literature only a small number of important advantage of generating not only the exponential
function generator circuits able to implement in CMOS function, but also a large number of other mathematical
technology a multitude of mathematical functions. In [13], the functions, very useful in VLSI designs. The new particular
implemented function is approximated by adding the weighted approximation function is built having as one of the main
output currents of a number of basic building blocks, realized objectives the minimization of the overall circuit complexity.
using a basic current squarer and a constant current, the Although a high-order of approximation usually requires large
relatively large complexity representing the most important circuits’ complexity, the proposed function generator can be
disadvantage of the circuit. The circuit reported in [14] uses implemented using only a current squaring circuit and an
the second-order Taylor approximation of the required additional current mirror.
functions, the proposed method being based on adding of the
output currents of a weighted current squarer, a weighted
current amplifier (or attenuator) and a constant current. The

978-1-4673-7488-0/15/$31.00
Authorized licensed use limited to: COCHIN©2015 IEEE OF SCIENCE AND TECHNOLOGY. Downloaded on April 14,2025 at 23:38:46 UTC from IEEE Xplore. Restrictions apply.
UNIVERSITY
 II. THEORETICAL ANALYSIS B. CMOS implementation of the function generator circuit
The proposed method used for realizing the implementation Based on the previous proposed expression of the
of a multitude of continuous mathematical functions, f (x ) , is approximation function, given by (2), the corresponding block
diagram of the original function generator is presented in
based on the utilization of a new g ( x ) approximation function
Fig. 1. The I IN current is the input variable current, the I O
that is able to third-order match the required functions.
Because, in the general situation, the complexity of the circuit current is the reference constant current, while the I OUT
that implements an approximation function is evidently current represents the output current of the circuit.
strongly increasing when the order of approximation is The expression of I OUT current is:
increasing, it is necessary to make a trade-off between the
⎛ p 2 ⎞⎟ ⎛ p 3 ⎞⎟ p3
accuracy of the function generator (correlated with the order I OUT = ⎜ n − I IN + ⎜ m − IO + IC , (3)
of approximation) and its overall complexity. From this point ⎜ q ⎟⎠ ⎜ q 2 ⎟⎠ q2
⎝ ⎝
of view, a third-order approximation function permits to
obtain an accuracy that satisfies most of circuits applications,
where the output current of the current-mode Squaring Core
its implementation in CMOS technology requiring a
(SQ) can be expressed as follows (it will be analysed later):
reasonable circuit complexity.
A. The new approximation function I O2
In order to synthesize a mathematical function that allows a IC = . (4)
qI IN
third-order approximation of any continuous mathematical IO −
p
function, some considerations must be taken into account. The
requirements for the approximation function are firstly
imposed by a facile implementation in CMOS technology. (n-p2/q)IIN + (m-p3/q2)IO
From the perspective of the function generator complexity, the IO-qIIN/p
utilization of a current squaring circuit represents the most IA IOUT
IC
convenient possibility (the complexity of a current squaring IO SC IC CM
circuit is usually smaller than the complexity of a current IB p3IC/q2
multiplier circuit). Second, the number of coefficients that
determines the expression of the approximation function must Figure 1. The block diagram of the new function generator circuit
be equal with four. This number is given by the third-order
approximation that imposes the necessity of an identity The “SC” block is required for implementing the first
between the first four terms from the Taylor series expansions fractional term from (2) and its original realization is
of the g ( x ) approximation function and, respectively, of the presented in Fig. 2. The “CM” block is implemented as a
f (x ) approximated function. Third, the system of equations classical current-mirror circuit, having a transfer factor
resulted from this identity must have a determined solution imposed by the particular approximated function.
that will define an unique approximation function.
Considering the following third-order Taylor series VDD

expansion of the required f (x ) function: K K K/4

IC
IA
f (x ) = m + nx + px 2 + qx 3 + ... , (1) IA
I1 I2

the new proposed approximation function that fulfils all the 8K K K

previous requirements has the following expression:

IB
p3 1 ⎛ p 2 ⎞⎟ p3 8K K
g (x ) = + ⎜n − x+m− . (2)
q 2 1 − q x ⎜⎝ q ⎟⎠ q2
p
Figure 2. The Squaring Core (SC) realization
The limitations of the previous approximation is correlated
with the type of the approximated function. For most Two translinear loops containing MOS transistors biased in
approximated functions, the range of x input variable for an saturation region establish the operation of the squaring circuit
accurate approximated is relatively large. proposed in Fig. 2, resulting:

Authorized licensed use limited to: COCHIN UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on April 14,2025 at 23:38:46 UTC from IEEE Xplore. Restrictions apply.
 I I I2 the complexity of the similar third-order approximated
I1 = A + B + B , (5) exponential generators, previously reported in literature.
8 2 2I A In order to obtain the exponential function using the general
function generator having the block diagram presented in Fig.
I I I2 1, relation (2) must be particularize for the values of m ÷ r
I2 = A − B + B , (6) coefficients specific to the exponential function: m = 1 , n = 1 ,
8 2 2I A
p = 1 / 2 , q = 1 / 6 and r = 1 / 24 , resulting the following
So:
expression of the g ( x ) approximation function:
I I B2
IC = I1 + I2 − A = . (7)
4 IA
9 1 x 7
g (x ) = − − . (10)
2 x 2 2
The original squaring core presents many advantages. First, 1−
3
the I C current is independent on technological parameters and
on temperature variations, this fact realizing an important III. SIMULATED RESULTS
increasing of the accuracy for the functional circuit that it The operation of the functional squaring core from Fig. 2 is
deserves. Second, the simplicity of the new proposed squaring evaluated using specific simulations. The simulation of the
core reduces the overall complexity of the function generator
I C (I A , I B ) dependence expressed by (7) for I A = 10 μA and
realized using this composing block. Third, the current-mode
operation of the functional core strongly increases the a range of variation for I B current between −1μA and 1μA
frequency response of the resulted function generator circuit. is presented in Fig. 3. The simulated error for implementing
The output current of the function generator circuit having the squaring function (5) using the original proposed squaring
the block diagram presented in Fig. 1 and the implementation core from Fig. 2 is 0.97% . The simulation was made using
of the squaring core shown in Fig. 2 can be expressed as Cadence, the supply voltage being V DD = 1V . The simulation
follows: is based on 0.35 μm CMOS process.

p3 I O2 ⎛ p 2 ⎞⎟ ⎛ p 3 ⎞⎟
I OUT = + ⎜n − I IN + ⎜ m − I O . (8)
q 2 I − qI IN ⎜⎝ q ⎟⎠ ⎜
⎝ q 2 ⎟⎠
O
p

Considering the input variable of both f (x ) and g ( x )

functions equal with the ratio between the input current and
the reference current, x = I IN / I O and comparing the general
expression (2) of the approximation function with the
expression (8) of the output current of the block diagram
presented in Fig. 1, it results that I OUT current is proportional
with the g ( x ) approximation function and it represents the
Figure 3. Simulation of the squaring circuit dependence (5),
third-order approximation of the general expression of f (x ) implemented using the squaring core
function:
A comparison between the simulated ratio
I OUT = I O g ( x ) ≅ I O f (x ) . (9) I OUT / I O = g (x ) (representing the approximated exponential
function) and f (x ) = exp(x ) is shown in Table 1. As a result of
C. Applications of the function generator circuit using a third-order approximation of the exponential function,
An immediate application of the function generator circuit the output dynamic range of the proposed exponential function
is represented by a high accuracy exponential function generator is approximately 28 dB , for an error smaller
generator with third-order approximation. Comparing to than 1dB .
previous reported second-order approximated function
generators, the original proposed exponential circuit presents a TABLE 1. COMPARISON BETWEEN IOUT/IO AND f(x)=exp(x)
much better accuracy, as a consequence of its third-order x IOUT/IO exp (x) ε
approximation. The new particular approximation function the [dB] [dB] [dB]
proposed circuit is based on strongly decreases the overall -2.0 -17.38 13.98 3.40
complexity of the new exponential function generator, making -1.6 -13.91 -12.59 1.32
it comparable with the complexity of the second-order -1.2 -10.43 -10.05 0.38
approximated exponential generators and much smaller than -0.8 -6.95 -6.89 0.06

978-1-4673-7488-0/15/$31.00
Authorized licensed use limited to: COCHIN©2015 IEEE OF SCIENCE AND TECHNOLOGY. Downloaded on April 14,2025 at 23:38:46 UTC from IEEE Xplore. Restrictions apply.
UNIVERSITY
 -0.4 -3.48 -3.47 0.01 [8] C.A. De La Cruz-Blas, A. Lopez-Martin, A. Carlosena, “1.5-V MOS
0.0 0.00 0.00 0.00 translinear loops with improved dynamic range and their applications to
current-mode signal processing”, Circuits and Systems II: Analog and
0.4 3.48 3.48 0.00 Digital Signal Processing, IEEE Transactions on, 2003, Pp.918 – 927
0.8 6.95 6.99 0.04 [9] C. Popa, “A New FGMOST Euclidean Distance Computational Circuits
1.2 10.43 10.63 0.20 Based on Algebraic Mean of the Input Potentials”, Lecture Notes in
1.6 13.93 14.56 0.65 Computer Science, Springer, 2009, pp. 459-466
2.0 17.38 19.08 1.70 [10] A. Hyo-Jin, C. Chang-Seok, L. Hanho, “High-speed low-complexity
Folded Degree-Computationless Modified Euclidean algorithm
architecture for RS decoders”, Integrated Circuits, ISIC '09. Proceedings
III. CONCLUSIONS of the 2009 12th International Symposium on, 2009, pp. 582 – 585
The requirements for the implementation in CMOS [11] C. Sawigun, W.A. Serdijn, “Ultra-low-power, class-AB, CMOS four-
quadrant current multiplier”, Electronics Letters, pp. 483 - 484
technology of a multitude of circuit functions have been [12] W. Machowski, S. Kuta, J. Jasielski, W. Kolodziejski, “Quarter-square
solved in this paper by designing a high-accuracy current- analog four-quadrant multiplier based on CMOS invertes and using low
mode function generator circuit. Comparing to the classical voltage high speed control circuits”, Mixed Design of Integrated Circuits
approaches of realizing any circuit function using an and Systems (MIXDES), 2010 Proceedings of the 17th International
Conference, pp. 333 – 336
individual circuit, the method of designing a single [13] M.T. Abuelma'atti, N.M. Al-Yahia, “An Improved Universal CMOS
computational structure for generating a large number of Current-Mode Analog Function Synthesizer”, in Proc. International
circuit functions presents many advantages. First, the silicon Symposium on Integrated Circuits, 2007, pp. 580-583.
area for one implemented function can be strongly reduced [14] M. Taher Abuelma'atti, “Universal CMOS current-mode analog function
synthesizer”, IEEE Trans. Circuits and Systems I: Fundamental Theory
comparing to classical designs. This reduction of the circuit and Applications, vol. 49, Issue 10, pp. 1468-1474, 2002.
complexity permits to increase the circuit accuracy by [15] C.H. Kao, H. Lin, C.S. Hsieh, "Low-voltage low-power current mode
increasing the order of approximation of the required exponential circuit", Circuits, Devices and Systems, IEE Proceedings,
functions. Second, function generator circuits allow a facile 2005, pp. 633–635
[16] A. El Mourabit, M.H. Sbaa, "CMOS exponential V-V converter with
reconfiguration by changing a small number of coefficients, large DC operating range", Electronics Letters, 2008, 44, (15), pp. 901–
accordingly to the implemented function. 902
The original proposed method for generating a large number [17] S. Vlassis, "CMOS current-mode pseudo-exponential function circuit",
of continuous mathematical functions was based on a new Electronics Letters, 2001, 37, (8), pp. 471-472
[18] V. Kalenteridis, S. Vlassis, S. Siskos, “A CMOS linear-in-dB VGA based
approximation function that is able to match, in a third-order on exponential current generator”, Design & Technology of Integrated
approximation, any required function. The new function Systems in Nanoscale Era (DTIS), 2011 6th International Conference on,
generator circuit was designed for a current-mode operation, 2011, Pp. 1 – 4
its output current being independent on technological [19] C. Popa, “CMOS Pseudo-Exponential Circuit Based on A New Average
Potential Principle”, The 8th International Workshop on Symbolic Methods
parameters and on temperature variations. These particularities and Applications to Circuit Design, SMACD 2004, Poland
additionally increase the overall accuracy and improve the [20] K.M. Abdelfattah, A.M., Soliman, “Variable gain amplifiers based on a
circuit frequency response. The proposed function generator new approximation method to realize the exponential function “,Circuits
circuit was designed for a low-voltage operation, being and Systems I: Fundamental Theory and Applications, IEEE Transactions
on, Vol. 49, Issue: 9, 2002, Pp. 1348 – 1354
supplied at VDD=1V. [21] L. Weihsing, L. Shen-Iuan Liu, W. Shui-Ken, “CMOS current-mode
A widely used application of the proposed structure was divider and its applications”, Circuits and Systems II: Express Briefs,
represented by the design of an improved accuracy IEEE Transactions on, Vol. 52, Issue: 3, 2005, Pp. 145 - 148
exponential function generator circuit. As a result of using the [22] A.N. Saatlo, S. Ozoguz, “A new CMOS exponential circuit with
extended linear output range”, Circuit Theory and Design (ECCTD), 2011
third-order approximation function, a wide output dynamic 20th European Conference on, 2011, Pp. 893 – 896
range of 28dB was obtained, for a maximal error of 1dB. [23] C. Popa, “Improved Accuracy Pseudo-Exponential Function Generator
with Applications in Analog Signal Processing”, IEEE Transactions on
REFERENCES Very Large Scale Integration Systems, Volume: 16, Issue: 3, ISSN: 1063-
8210, 2008, pp. 318-321
[1] C. Popa, “Synthesis of computational structures for analog signal [24] C. Popa, “Low-Voltage CMOS Current-Mode Exponential Circuit
processing,” Springer Publishing House, 449 pp., 2011 with 70dB Output Dynamic Range”, Microelectronics Journal, vol.
[2] C. Popa, “Superior-order curvature-correction techniques for voltage 44, Issue 12, December 2013, pp. 1348-1357
references,” Springer Publishing House, 179 pp., 2009 [25] C. Popa, “Improved Accuracy Pseudo-Exponential with Applications
[3] C. Popa, “Current-Mode Analog Nonlinear Function Synthesizer in Analog Signal Processing”, The International Conference on
Structures”, Springer Publishing House, 2013 "Computer as a Tool", EUROCON 2005, Belgrade, Serbia &
[4] L.E. Dermentzoglou, A. Arapoyanni, Y. Tsiatouhas, “A Built-In-Test Montenegro, pp. 1594-1597
Circuit for RF Differential Low Noise Amplifiers”, Circuits and Systems [26] C. Popa, “Pseudo-Exponential Computational Circuit With Improved
I: Regular Papers, IEEE Transactions on, 2010 , Pp. 1549 - 1558
Accuracy and Frequency Response”, 35th Jubilee International
[5] C.A. De La Blas, A., Lopez, “A novel two quadrant MOS translinear
Convention on Information and Communication Technology, Electronics
Squarer-divider cell”, Electronics, Circuits and Systems, 2008. ICECS
and Microelectronics, MEET 2012, Croatia, pp. 83-86
2008. 15th IEEE International Conference on, 2008, Pp.5 – 8 [27] C. Popa, “CMOS Current-Mode Pseudo-Exponential Circuits with
[6] C.A. De La Cruz-Blas, et.al., “1.5-V square-root domain second-order
Superior-Order Approximation”, IEEE-EURASIP Workshop on Nonlinear
filter with on-chip tuning”, Circuits and Systems I: Regular Papers, IEEE
Signal and Image Processing, Italy, 2003
Transactions on, Vol. 52, Issue: 10, 2005, Pp.1996 – 2006 [28] C. Popa, “High Accuracy Exponential Function Generator for Analog
[7] A. Kircay, M.S. Keserlioglu, U. Cam, “A new current-mode square-root-
Signal Processing”, 2005 NSTI Bio Nano Conference, Anaheim,
domain notch filter”, Circuit Theory and Design, 2009. ECCTD 2009.
California
European Conference on, 2009, Pp.229 – 232

Authorized licensed use limited to: COCHIN UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on April 14,2025 at 23:38:46 UTC from IEEE Xplore. Restrictions apply.