Sie sind auf Seite 1von 4

A General-purpose Kernel based on Genetic Algorithms

for Optimization of Complex Analog Circuits


B. Vaz1,2, R. Costa1,2, N. Paulino1,2, J. Goes1,2, R. Tavares1,2 and A. Steiger-Garção1,2
1 2
UNINOVA – CRI Faculdade de Ciências e Tecnologia
Campus da Faculdade de Ciências e Tecnologia Campus da Faculdade de Ciências e Tecnologia
2825 – 114 Monte da Caparica – PORTUGAL 2825 – 114 Caparica – PORTUGAL
E-mail: jg@uninova.pt E-mail: pc-dee@uninova.pt

Abstract - This paper presents a general-purpose kernel genetic algorithms. The optimization of any new block is
based genetic algorithms for optimizing complex analog basically carried-out by simply providing two additional
circuits and systems. The developed tool is very flexible files to the kernel. A text file containing the information of
allowing optimizations at different levels of abstraction. The the genes (variables) of the chromosome and, on the other
optimization of any new block is basically carried-out by hand, a dynamic link library (DLL) type file, which is ob-
simply providing two additional files to the kernel. A text tained by compiling a C++ code-source file comprising the
file containing the information of the genes (variables) of the overall fitness function. Two examples are given to illustrate
chromosome and, on the other hand, a dynamic link library the efficiency of the optimizer at both, circuit and system
(DLL) type file, which is obtained by compiling a C++ levels. In section 3 the results of the optimization of a low-
code-source file comprising the overall fitness function. Op- voltage pipelined ADC are shown, taking into account major
timization examples of two low-voltage circuits clearly as- issues, such as, thermal noise, power dissipation and linear-
sess the attractiveness of this tool. ity. In section 4, the results of the optimization of a low-
voltage two-stage Miller compensated amplifier are pre-
1. INTRODUCTION sented and compared with simulated results. Finally, conclu-
sions are drawn in section 5.
According to the Semiconductor Industry Association’s
roadmap [1] a fast scaling-down of the transistor’s minimum 2. DESCRIPTION OF THE GENERAL-PURPOSE
channel lengths is expected. Furthermore, this technology KERNEL STRUCTURE
scaling will be accomplished by a reduction of the supply
voltages of the circuits (1.8 Volt by 2000 and 0.3 Volt by 2.1 – ARCHITECTURE OF THE TOOL
2014). However, this reduction of the supply voltage se-
verely affects the analog circuit design basically due to two
main reasons. Firstly, in smaller technologies, the supply
voltage scales roughly linear with the minimum feature size
but the threshold voltages of the transistors do not. As a con-
sequence, the number of topologies capable of operating at
low voltages is highly reduced. On the other hand, since the
voltage swings have to be proportionally decreased, the sig-
nal-to-noise ratios of the circuits are directly affected. In
order to maintain the current efficiency (power dissipation)
is mandatory to have efficient optimization tools to assure
that the overall performance of the circuit or system is not
degraded. Moreover, from the IC market point of view is
highly desirable to have short redesign cycles either for dif-
ferent technologies or specifications.
Figure 1 – Tool’s Architecture.
Several approaches for automatic design and optimization
(automatic sizing) of analog cells were already proposed The architecture of the proposed kernel is presented in Fig-
using statistical optimization techniques such as simulated ure 1. There are two text files that should be provided to the
annealing, either in equation-based systems [2] or in simula- application: one contains the structure of the chromosome
tion-based systems [3]. The optimization tool proposed in and another the desired parameters that are envisaged. The
this paper follows, as suggested in [4], an equation-based
approach for optimization of analog building blocks using chromosome, x , consists of one or more genes. Basically,
the genes are the variables, xi, that the algorithm will found
that best fit the desired goals. The first file comprises the
information regarding the upper (Lmax) and lower (Lmin) lim-
its of the genes and also the resolution of each gene (Rg).
The second file contains the desired value and the relative
strength of all of them. Additionally it can be provided, de-
pending on the problem, another text file containing specific
parameters – Technology file -. One of the most important
blocks of this architecture is the DLL that contains the fit- a) b)

()
ness function, f x . This file has all the mathematical func-
Figure 2 – (a) Roulette System for 5 individuals;
(b) Rank System for 5 individuals
tions that model the behavior circuit. The chromosome that
best fits the goal (the entire set of the desired parameters) is B – Cross-operator
saved for later simulation and confirmation of the results.
Within each chromosome, genes are represented in binary
2.2 – GENETIC ALGORITHM code where the number of bits is dependent of the desired
precision according to:
The heart of the application is the genetic algorithm. A full
description of the genetic algorithm’s theory is out of the Rg −1
scope of this paper. Basically, this algorithm can be summa- x i = L min + (L max − L min ) ∑b j ⋅2 j (2 Rg
)
−1 (1)
rized in the next steps. j =0

1. Initialize the population P with a set of d individuals


With this type of coding, the cross and the mutation opera-
2. While the stop condition isn’t reached:
• For each individual (chromosome) x , calculate f x () tors can be easily implemented. To cross two selected indi-
viduals (parents), first a random cross point is chosen and
• Select s individuals form P to be in the new population the new individuals are made of genetic material of his par-
Ps. Each individual is selected according with his prob- ents. An example of the cross-operator is shown in Figure 3.
ability of selection
• Select (d-s) individuals to be crossed. Each individual is
selected according to his probability of selection. The
new individuals are placed into the new population Ps
• Select m individuals of Ps and apply the mutation op-
erator
• P = Ps

A - Selection
Figure 3 – The Cross-operator.
One possible approach to select the individuals is to pre-
define a percentage of individuals that are directly copied C – The Mutation Operator
into the next population. The remaining individuals are
crossed and finally the mutation operator is applied to new The mutation operator work as follows. Firstly, one bit of
population. A different approach is that only the best chro- the gene to be mutated is randomly chosen. This bit changes
mosome is copied to next population and it remains un- its digital value as it is represented in Figure 4.
changed. There are two ways of selecting individuals: rou-
lette and rank systems.

In the Roulette system, each individual has a fitness associ-


ated, fi, which consists on a normalized value between 0 and
1. This value represents directly the probability that one in-
dividual has to be chosen. An example is presented in Figure
2 (a). Figure 4 – The Mutation Operator.
Another possibility of selection is the Rank system as illus-
trated in Figure 2 (b). Firstly, all the individuals are ordered There is a possibility of varying the mutation bit according
according to their fitness and, secondly, the probability of to the number of the generation. In the beginning all the bits
one individual to be selected is proportional to its position. might be chosen to be mutated but, after a given number of
Again, Figure 3, illustrates the rank system for an example generations only the least significant bits can be mutated.
of a population with five individuals.
3. SYSTEM-LEVEL OPTIMIZATION OF A LOW- lengths of the transistors, Li, and their saturation voltages,
VOLTAGE LOW-POWER PIPELINE ADC Vdsati. M1 represents the PMOS transistors of the differen-
tial-pair, and M6 is the transistor of the second gain stage.
Previous studies concerning the problem of optimizing the M2, M4 are, respectively, the PMOS and the NMOS current
resolution-per-stage in pipeline ADCs while minimizing the sources. Ibias is the value of the biasing current of the dif-
power dissipation have been already reported [5, 6], but ferential-pair and gene MF represents the mirror factor of
none of these works optimize the distribution of the noise this current to the output branch (second stage). Cc and Rc
contributions of the several pipelined stages to the total input are, respectively, the compensation capacitor and resistor.
referred noise. This is an issue harder to manage by the need
  A _ achieved  
− 0 
 Output _ swing _ achieved  
− 
of keeping the unit capacitances above a minimum for feasi-   A0 _ desired    Output _ swing _ desired 

bility and matching reasons. The methodology and the tool f ( x ) ≈ 1 − e  1 − e 
addressed in this paper is quite general, since it is capable of   
   (3)
exploring different noise distributions while optimizes si-
  2 nd _ pole _ achieved  
− 
 total _ current _ desired  
− 
multaneously several other design parameters, namely,   2 nd _ pole _ desired    total _ current _ achieved 

power dissipation, resolution-per-stage and maximum ex- 1 − e 1 − e 
pectable DNL errors. The fitness function was defined as   
  
  noise _ desired  
−  
 DNL _ desired  
− 
  noise _ achieved    DNL _ achieved 

f ( x ) ≈ 1 − e 1 − e 
  
   (2)
  PowerDiss _ desir .  
− 
  PowerDiss _ achiev.  
 2 
1 − e  − (Re s − Re s _ desir.) (Re s − Re s _ desir.) 
  e + e 
 

For the sake of simplicity, complex equations for calculating


all achieved parameters are not shown. The structure of the
chromosome is depicted in Figure 5 as well as the results of
the best chromosome. The desired specifications were, re-
spectively for the input-referred squared-noise, for the
power dissipation and for the DNL, 79.5 nV2, 13.3 mW and
0.5 LSB. The output of the optimizer is shown in Figure 6
for an example of the optimization of a 10-bit (Res_desired)
20 MS/s 1.5 Volt pipelined ADC designed for a 0.35µm
CMOS technology. The achieved squared-noise, power dis-
sipation and maximum DNL error were, respectively, 77
nV2, 10.4 mW and 0.52 LSB@10bit.

# of stages
(NS)
C
(S&H)
Res.
stage 1
Cunit Res.
(stage 1) stage 2
Cunit
(stage 2)
... Res
(stage NS)

5 0.64pF 3b 0.016pF 3b 0.05pF 3b 0.05pF 3b 0.05pF 2b

Figure 5 – Chromosome used for the optimization of a low-


voltage pipeline ADC and best chromosome results.

4. CIRCUIT-LEVEL OPTIMIZATION OF A LOW-


VOLTAGE TWO-STAGE MILLER-OPAMP

The second example presented in this paper comprises the


design and the optimization of a 1.5 Volt CMOS two-stage
Miller-compensated opamp. The fitness function was
roughly defined according to expression (3). The structure of
the chromosome is depicted in figure 7 as well as the results Figure 6 – Evolution of the squared-noise, power dissipation
of the best chromosome. The genes are basically the channel and maximum DNL error during 120 generations.
L1 Vdsat1 L2 Vdast2 L4 Vdsat4 L6 Vdsat6 Ibias MF Cc Rc simulated values using an electrical simulator (HSPICE) and
1.2 µ m 175mV 1.2 µ m 250mV 1.2 µ m 166mV 1.2 µ m 100mV 65 µ A 1.83 1.79pF 1.16k Ω
BSIM3.v3 models. As it can be observed both achieved and
simulated values are relatively close. The major differences
Figure 7: Format of the chromosome used in the genetic between the desired specifications and the
algorithm for the 2-stage opamp optimization and achieved/simulated occur for the DC gain due to problems
best chromosome results on modeling the drain-source impedance of the transistors at
low voltages. The 2nd. pole frequency achieved was also
larger than the desired since desired total current was larger
than the required value to reach the desired specs.

Table 1. Optimized and post-simulated results for the opamp.


Specs. Specs. Simulated
desired achieved Results
A0 (dB) 60 94 93
2nd. pole freq.(MHz) 60 95 120
Output Swing (V) 1 1.04 1.02
Total current (µA) 200 185 186

5. CONCLUSIONS
This paper presented a general-purpose kernel based ge-
netic algorithms for optimizing complex analog circuits and
systems. The developed tool is very flexible allowing op-
timizations at different levels of abstraction. Optimization
examples of two low-voltage circuits were presented to
clearly assess the attractiveness of this tool.

ACKNOWLEDGMENTS
The research work that led to this implementation was partially
supported by the Portuguese Foundation for Science and Technol-
ogy under ADOPT (PCTI/1999/ESE/33311) Project.

REFERENCES
[1] International Technology Roadmap for Semiconductors, Semi-
conductor Industry Association, 1999.
[2] G. E. Gielen, H. Walsharts and W. Sansen, “Analog circuits
design optimization based on symbolic simulation and simu-
lated annealing”, IEEE Journal of Solid-State Circuits, vol. 25,
pp. 707-713, June 1990.
[3] F. Medeiro, R. Rodríguez-Macías, F. V. Fernández, R. Domín-
guez-Castro, J. L. Huertas and A. Rodríguez-Vázquez, “Global
Design of Analog Cells Using Statistical Optimization Tech-
niques”, Analog Integrated Circuits and Signal Processing, 6,
179-195, Kluwer Academic Publishers, 1994.
[4] Nuno Paulino, João Goes and Adolfo Steiger-Garção, “Design
Methodology for Optimization of Analog Building-Blocks us-
ing Genetic Algorithms”, IEEE International Symposium on
Circuits and Systems (ISCAS’2001), Australia, May 2001.
Figure 8. Evolution of the DC gain (A0), output swing, 2nd. [5] Stephen H. Lewis, “Optimizing the Stage Resolution in Pipe-
pole frequency and total current during 120 generations. lined, Multistage, Analog-to-Digital Converters for Video-Rate
Applications”, IEEE Transactions on Circuits and Systems – II,
Figure 8 show the evolution results during 120 generations, vol. 39, no. 8, pp. 516-523, August 1992.
of the DC gain (A0), output swing, 2nd. pole frequency and [6] João Goes, João C. Vital and José E. Franca, "Systematic De-
sign for Optimization of High-Speed Pipelined A/D Converters
total current. Table 1 compares the obtained values of these
using Self-Calibration", IEEE Transactions on Circuits and
parameters after the optimization concluded with post- Systems, vol. II, pp. 1513-1526, December 1998.

Das könnte Ihnen auch gefallen