Evolutionary design of analog circuits with a uniform-design based multi-objective adaptive genetic algorithm

Abstract

A novel multi-objective genetic algorithm based approach for evolvable hardware (EHW) is proposed in this paper. It features intensive applications of the Uniform Design Technique (UDT) to fitness function composing and crossover operator construction for the sake of multiple uniform search directions, better quality of offspring and higher computation efficiency. With an efficient and universal representation scheme based on standard component values and PSpice Net-list, it supports generation and optimization of circuit structures and component values. Moreover, its search efficiency is enhanced by an adaptation technique of genetic parameters, which pays attention to both population diversity, loci' effects and development of evolution. Both theoretic analyses and experimental results demonstrated its capability in finding out a set of Pareto-optimal solutions or effective design results via a shorter execution, and showed its potential in fully automated design of circuits concerning multi-objective requirements and preferences.