Ian Ozsvald's MSc Thesis

Short-Circuiting the Design Process:
Evolutionary Algorithms for Circuit Design using Reconfigurable Analogue Hardware

Ian Ozsvald, 1998. KBS MSc Thesis, School of Cognitive and Computing Sciences, University of Sussex, UK.
Download Postscript (2.7MBytes), or download compressed Postscript (655kBytes).

This on-line copy does not include program listings. Contact A. Thompson for further information.

Abstract

The design of analogue electronic circuits is often accomplished by the use of well-known and well-understood circuits that are joined together to form larger circuits. This, although the most common approach, is not necessarily the best.

Thompson [1996], Koza [1996a] and Higuchi et al. [1996] have each used techniques inspired by Darwinian evolution to successfully create electronics circuits. These circuits have used both discrete components and reconfigurable integrated circuits.

Here, I propose a circuit synthesis tool, a program that takes an input/output behavioural specification and generates the required circuit. Zetex's analogue reconfigurable integrated circuit (IC), the TRAC020, is the target for this tool.

A methodology is presented for the evolution of the required circuit definitions. To this end, a simulator for the Zetex device is presented which works in a similar fashion to Zetex's (physical) TRAC development board. The evolutionary methods shown are applied to the simulation of the TRAC and to the hardware, and this dual approach highlights several important problems that otherwise may have been overlooked.

This system is a major step towards automating the circuit design process for TRACs. Enhancements are suggested both to the methodology and the evolutionary mechanism, especially considering a circuit's genetic representation, and the role of neutral mutations.

Back to Evolutionary Electronics at Sussex