Multi-objective Intrinsic Hardware Evolution

Evolutionary design of digital circuits reveals the poten-tial to provide autonomous systems with the properties of self-adaptation, self-optimization and functional recovery. Especially for functions that are rated by a data-dependent and time-varying functional quality, rather than an absolute measure of correctness, evolutionary techniques are attrac-tive. In this paper, we present a novel approach to evolvable embedded systems that is able to adapt to both slow and rad-ical changes in the environment and the system state, respec-tively. First, a multiobjective evolutionary search algorithm with a selection scheme based on Pareto dominance is used to compute a set of reasonable trade-offs. Then, the deci-sion is made which solution to use for the present situation. During operation, the system adapts to slowly changing en-vironmental conditions by the evolutionary search process. To handle radical changes, precomputed dominant solutions are stored in the system. When a radical change occurs, the system switches to a ”good-enough ” solution, and the online evolutionary process is restarted. In our work, we treat evolutionary circuit design as a multiobjective optimization problem that tries to evolve func-tionally good or correct FPGA circuits with minimal area and maximal speed. Additionally, the optimization problem can be specified by constraints as, for example, usable area of a chip or a minimal functional quality of a circuit. 1