GPIC: A Genetic Programming Intrinsic Circuit Design Platform

In the realm of analog evolvable hardware, the potential of a coarse-grained, topological search approach which employs intrinsic testing has been unexplored to date. Our evolvable hardware design platform named GPIC (i.e. Genetic Programming of Intrinsic Circuits) is designed for this approach which makes it suitable for adaptive, fault tolerant system design as well as CAD flow applications. On a workstation GPIC evolves circuit designs which are each intrinsically realized and tested on a Commercial Off-The-Shelf (COTS) field programmable device, the Anadigm AN221E04. This intrinsic approach provides speed over software circuit simulators that employ computationally expensive models. To interface with our COTS chip, we have implemented a resource manager within GPIC and configure it via the vendor’s design GUI. For circuit search and optimization GPIC uses a cyclic graph representation that explores the topological space of analog circuits. Each circuit is represented using high level functional (i.e. coarse grained) analog components, such as adders and amplifiers. For component value optimization, GPIC uses particle swarm optimization. As demonstration of the platform’s capabilities, we describe how GPIC has evolved a simple controller for a linear, third order plant.