Scalable multi-cores with improved per-core performance using off-the-critical path reconfigurable hardware

Scaling the number of cores in a multi-core processor constraintsthe resources available in each core, resulting in reduced percoreperformance. Alternatively, the number of cores have to be reducedin order to improve per-core performance. In this paper, we propose atechnique to improve the per-core performance in a many-core processorwithout reducing the number of cores. In particular, we integrate aReconfigurable Hardware Unit (RHU) in each core. The RHU executesthe frequently encountered instructions to increase the core's overall executionbandwidth, thus improving its performance. We also propose anovel integrated hardware/software methodology for efficient RHU reconfiguration.The RHU has low area overhead, and hence has minimalimpact on the scalability of the multi-core. Our experiments show thatthe proposed architecture improves the per-core performance by an averageof about 12% across a wide range of applications, while incurringa per-core area overhead of only about 5%.