Particle Swarm Optimization for Run-Time Task Decomposition and Scheduling in Evolvable MPSoC

Decomposition and scheduling are two major challenges in hardware and software developments of multi¬processor systems. The introduced EvoMP (Evolvable Multi-Processor) is a novel NoC-based homogeneous MPSoC system that performs decomposition and scheduling using evolution¬ary algorithms at run-time. A hardware genetic core was used in first version of this platform to perform these two tasks. This core tries to find an efficient solution for decomposition and scheduling of the target application among available computational resources. This paper presents the new version of this system in which a hardware Particle Swarm Optimization (PSO) core is exploited to perform evolutionary decomposition and scheduling. The principle of operation and architecture of the EvoMP platform and the PSO core is briefly explained in this paper. The simulation and synthesis results of this PSO-based EvoMP is also presented and compared with prior genetic-base approach.