Two-Stage OTA Sizing Optimization Using Bio-Inspired Algorithms

Abstract

The analog part of a mixed-signal integrated circuit represents a great amount of the circuit sizing effort. It is necessary to size each device separately and, in cases with several variables, the design space becomes quite large. The analog integrated circuit sizing can be modeled as an optimization problem and solved by optimization heuristics. In this work, we compare three bio-inspired heuristics to size a two-stage CMOS Miller operational transconductance amplifier: Particle Swarm Optimization (PSO), Cuckoo Search (CS) and Firefly Algorithm (FA). The goal is to evaluate the applicability of these heuristics for the analog sizing problem and to determine the best configuration of the algorithms parameters for optimizing performance of the generated circuit, mainly power consumption and silicon area. Results show that PSO and CS are more suitable to find optimized solutions, while FA presents less efficient exploration of the design space. Although PSO is faster and generates good solutions, the best overall solution was achieved with CS algorithm.

A Comprehensive Review on Automation-based Sizing Techniques for Analog IC Design

“…The implementation of meta-heuristics depends on the nature of the problem. In general, design methods for analog integrated circuits use heuristics based on natural processes such as Genetic Algorithms [42], Simulated Annealing, Artificial Neural Networks [43,35], Particle Swarm [44,45] and other bio-inspired approaches [29]. These methods have a high probability of finding a solution close to the global optimal after several iterations [1].…”

A Comprehensive Review of Bio-Inspired Optimization Algorithms Including Applications in Microelectronics and Nanophotonics

“…This constraint demands a reliable method to attain the transistor sizing by balancing all the desired parameters of the design. The use of metaheuristic optimization algorithms to determine the transistor sizes can be a solution to this problem (e.g., [ 239 , 240 ]). The minimization of the parameters can be accomplished by selecting one or more objective functions and considering the rest of the parameters as a constraint.…”