A fuzzy logic based dynamic reconfiguration scheme for optimal energy and throughput in symmetric chip multiprocessors

Embedded systems architectures have traditionally often been investigated and designed in order to achieve a greater throughput combined with minimum energy consumption. With the advent of reconfigurable architectures it is now possible to support algorithms to find optimal solutions for an improved energy and throughput balance. As a result of ongoing research several online and offline techniques and algorithm have been proposed for hardware adaptation. This paper presents a novel coarse-grained reconfigurable symmetric chip multiprocessor (SCMP) architecture managed by a fuzzy logic engine that balances performance and energy consumption. The architecture incorporates reconfigurable level 1 (L1) caches, power gated cores and adaptive on-chip network routers to allow minimizing leakage energy effects for inactive components. A coarse grained architecture was selected as to be a focus for this study as it typically allows for fast reconfiguration as compared to the fine-grained architectures, thus making it more feasible to be used for runtime adaption schemes. The presented architecture is analyzed using a set of OpenMP based parallel benchmarks and the results show significant improvements in performance while maintaining minimum energy consumption.