初期解をファジイ推論するGA/SAを用いたVLSIフロアプラン手法の精度向上

Abstract

Rapid increase of the scale of integration requires higher knowledge and well trained skills of experienced design engineers. However it is usually difficult for novice engineers to perform optimized design of initial and macroscopic placement in floorplanning. This paper proposes to apply the soft computing technology mostly fuzzy inference and genetic algorithms to automate the floorplanning design which decides a macroscopic placement of the top layer of LSI physical implementation. ISPD98 benchmark data is used for evaluation. The relation among several parameters of fuzzy inference and genetic algorithms and placement cost is discussed. The relation between I/O pins and the cost is also discussed. Simulated annealing is employed after genetic algorithms to avoid local optimization.