A massively parallel reconfigurable co-processor for computationally demanding Particle Swarm Optimization

The Particle Swarm Optimization or PSO is a heuristic based on a population of individuals, in which the candidates for a solution of the problem at hand evolve through a simulation process of a social adaptation simplified model. Combining robustness, efficiency and simplicity, PSO has gained great popularity as many successful applications are reported. The algorithm has proven to have several advantages over other algorithms that based on swarm intelligence principles. The use of PSO solving problems that involve computationally demanding functions often results in low performance. In order to accelerate the process, one can proceed with the parallelization of the algorithm and/or mapping it directly onto hardware. This paper presents a novel, massively parallel PSO co-processor which was implemented by using reconfigurable hardware. The implementation results show that the proposed architecture is very promising as it achieved superior performance in terms of execution time when compared to the direct software execution of the algorithm.