On low power high level synthesis using genetic algorithms

This paper describes a new approach for utilizing genetic algorithms to solve high-level synthesis tasks with multiple voltages. The novelty of the method lies in its approach for modeling and encoding the resulting chromosomes. The proposed system receives a control/data flow graph, hardware library, and the constraints for area, time, and power as inputs. It then uses this information to concurrently solve the scheduling, binding, and allocation problems in order to generate a solution optimized for average and peak power. The new formulation eliminates the need to check for a valid schedule since it always generates a valid one. This leads to a simple evaluation function and a quick solution. The evaluation function is a mix of constraints satisfaction and a consideration of average and peak power. Simulation results show that the lower power solutions are obtained in a reasonable CPU time, which is in the range of seconds.