MultiObjective Optimization of Standard Cell Placement using Memetic Algorithm

Abstract

Beyond the optimization of single parameter (usually the wire-length) in Standard Cell Placement (SCP), focus in the present work is laid on the optimization of speed, power, and the wire length. As discussed in our previous work of hybrid algorithms for single objective optimization of SCP the main advantage of hybridization is the improvement in convergence speed to Pareto front although it leads to increase in computation time per generation. Memetic Algorithm (MA) is a hybrid of Genetic Algorithm (GA) & Local search (LS) wherein we need to strike a right balance of the two for optimum solution. In this paper we work on our previous GA based multi-objective SCP algorithm [2] for simultaneous optimization of power, speed and wire length while maintaining the layout width as constant by choosing initial population to be alleles of high fitness value and apply proper local search to all the members of initial population. Further we compare the results with previously established GA based algorithm by applying the two on ami20, ami33 and ami120 cell library instances. The Memetic algorithm is found to give better results with 10% improvement in wire-length, 7.5% lesser delay and power consumption reduction by nearly 6%.