A Novel Evolutionary Technique for Multi-objective Power, Area and Delay Optimization in High Level Synthesis of Datapaths

The use of multi-objective approaches in High Level Synthesis has been gaining lot of interest in recent years since the major design objectives such as area, delay and power are mutually conflicting, thereby necessitating trade-offs between different objectives. This paper proposes a methodology for area, power and delay optimization using the Non-dominated Sorting Genetic Algorithm II (NSGA II). A metric based technique has been used to determine the likelihood of a schedule to yield low power solutions during binding. Actual power numbers are not determined since this is computationally expensive. The methodology has been evaluated on standard benchmark Data-Flow Graphs (DFGs) and results indicate that it yields improved solutions with better diversity when compared to a weighted sum GA approach. For the IIR benchmark, it was observed that the NSGA II was able to converge to the true Pareto front obtained from exhaustive search.