Design Optimisations for a Cellular Automata Model with Programmable Interconnect Structure

Abstract:
FPGA systems are often used as accelerators for software computations and can deliver speed-ups of several orders of magnitude. We use such an FPGA-based system to support a genetic algorithm (GA) running in Java on a host computer to optimise cellular automata currently applied to the density classification task. Essentially, the computational intensive fitness evaluation is performed by the hardware while the GA is running as software. In this paper we concentrate on the optimisation of a programmable interconnect structure for small-world cellular automata and a result aggregation to reduce data transfers. The synthesis results are given and the trade-offs are discussed.