Simulation optimization with multiple-demes genetic algorithms in master-slave parallel mode

Simulation optimization for complex space is realized by applying genetic algorithms (GAs) to searching the optimum parameters satisfying the desired characteristics. It is necessary to use parallel calculation to obtain satisfactory results in a reasonable amount of time. The advantages and drawbacks of existing parallel calculation mode are analyzed. Then the multiple-demes GAs in master-slave mode (MDMS) are proposed to make progress in not only the quality of solutions but also the speedup for multiple-processors, compared to conventional master-slave mode. They show less limitations on the number of processors than that or coarse-grained and neighborhood mode. The test example in semiconductor device synthesis shows the efficiency.