MOPSO Applied to Architecture Tuning with Unified Second-Level Cache for Energy and Performance Optimization

Design Space Exploration (DSE) have been a suitable strategy to configure a parameterized SoC platform in terms of systems requirements such as energy and performance. In this work, a multi-objective approach (MOPSO) based on Particle Swarm Optimization was applied for DSE problems for supporting architecture tuning in memory hierarchy with unified second level cache. The proposed approach considers two objectives to be optimized: energy consumption and application performance; and allows to reduce the design space by exploring only 2,64% of the exploration space. Results of MOPSO with regard to cost function found solutions approaching Pareto Optimum in terms of energy consumption and performance in the majority of cases, about 66% of the studied cases. Experiments based on simulations were carried out on 18 applications from the Mibench and PowerStone suite benchmarks.