Swarm intelligence driven Design Space Exploration: An integrated framework for power-performance trade-off in architectural synthesis

摘要

The process of Design Space Exploration (DSE) during architectural synthesis is very intricate and tedious. It involves trade-off between conflicting design objectives of power-performance as well as between orthogonal issues of exploration time and quality of result. This paper proposes a novel integrated framework on swarm intelligence (particle swarm optimization) based DSE for power-performance trade-off during architectural synthesis of control and data intensive applications. The proposed integrated approach comprises of a comprehensive mapping process and reliable solution evaluation strategy. Therefore, the novel contributions of the paper are as follows: i) Introduction of a novel particle swarm optimization driven DSE methodology for power-performance trade-off ii) Novel solution evaluation methodology incorporating power and execution time parameters iii) Novel stochastic based diversity introduction technique (particle mutation algorithm) iv) Novel perturbation technique to control unwarranted exploration drift of particles v) improved results in QoR (> 21%) and reduction in exploration time (> 81%) when compared to recent DSE approaches for the tested benchmarks.