On evolving hardware: on-line evolution by cellular programming

Extensive research shows that the implementation of evolutionary algorithms are costly in many cases. This paper reports on an alternative construction, namely non-uniform cellular automatons (CA) with which (1) it is possible to increase the capacity for complex computations while preserving the main features of CAs, (2) evolve parallel cellular machines to perform computational tasks via cellular programming, and (3) attain evolware with implementations centering on hardware. The hardware architecture of the cellular programming evolware involves: (1) a group of individual cells is in operation rather than a single one, (2) genetic operators are implemented on-board rather than on offline computer, and (3) the evolutionary phase proceeds uninterrupted via a continuous machine´s operation. The active components of the evolware complex comprise exclusively the latest cellular FPGA circuit. Accordingly, this paper is organized as follows: Section I presents fundamentals and analysis of computations in quasiuniform cellular automats in a quasi-uniform cellular space; Section II introduces the basic approach for coevolving cellular computation and cellular machines, denoted cellular programming. Section III adopts synchronization as a criteria to attain evolware centered on hardware. Cellular programming is implemented in hardware. The algorithm is slightly modified (without loss of performance): the genetic operators, one-point crossover and mutation, are replaced with the operator uniform crossover. This results in the creation of an offspring genome from two parent genomes (bit strings) with a 50% probability for each parent. Finally, Section IV analyzes performance and results