Integrated particle swarm optimization (i-PSO): An adaptive design space exploration framework for power-performance tradeoff in architectural synthesis

Abstract

This paper presents a novel adaptive design space exploration (DSE) framework called `integrated particle swarm optimization (i-PSO)' for power-execution time tradeoff during architectural synthesis of data (and control) intensive applications. The proposed i-PSO besides introducing a novel DSE methodology, integrates a number of novel algorithms that guides in convergence to a high quality solution without compromising the exploration speed. The major sub-phases of proposed i-PSO that facilitates in faster convergence to an optimal solution are: a) algorithms to control unwarranted exploration drift - i) adaptive end terminal perturbation algorithm that preserves the ability of the exploration process to operate in the valid design space interval ii) clamping algorithm to manage excessive velocity outburst during searching b) algorithm to restrict boundary constraints violation c) rotation based mutation algorithm for particle diversification d) pre-tuning of i-PSO baseline parameters to achieve superior results. Additionally, the paper also reports a novel sensitivity analysis based on the variation of different parameters such as inertia weight and termination condition and its impact on proposed i-PSO based DSE. Finally, the proposed approach when verified on benchmarks yielded an average improvement in quality of results (QoR) (>21%) and reduction in exploration time (> 80%) compared to recent approaches.