The goal of this research is to derive circuits that can recover from component failure. Our approach is to replace a single monolithic computing element with a system of simple, redundant, interconnected processing nodes such as a neural net. Each node will be a hardware device called a Mote that can sense data, do simple processing and wirelessly transmit and receive data from its neighbors. The neural net is trained using an evolutionary algorithm called Particle Swarm Optimization (PSO). This paper discusses the PSO algorithm, simulated results using the algorithm, and its application to the Motebased neural net. We also describe and show results for a new algorithm called Dispersive PSO, which is useful when a neural net needs to be retrained to a different function or when a neural net needs to be retrained due to a node failure.