Interconnects parasitic extraction using modified Genetic Algorithm

Abstract

Three new Genetic Algorithm (GA) are proposed and used to solve a Curve fitting problem for Parasitic Extraction Macro-modeling application. The first proposed approach, Diagonal GA (DGA); is based on replacing the traditional random population initialization method with a deterministic diagonal-like one. The second proposed approach, Elite Condensation GA (ECGA); is based on fine tuning the GA by explicitly condensing the population around a number of elite individuals. The third proposed approach, ECGA2, is a modified version of ECGA; that chooses elite members among all the population in each generation, then it divides the population into a number of sub-populations where each sub-population is composed of a single elite and a condensed population around it. Then, it performs GA operations on each of those subpopulations separately before merging them all into one population and keep repeating that divide-merging sequence. The performances of these three proposed approaches were measured on an extensive real data sets and used along with the understanding of the physical problem to offer various explanations of the theoretical aspects of the new algorithms.