Empirical Software Engineering
| Hi-index | 0.00 |
We propose a new optimization method, named genetic simulated annealing (GSA), which combines the local stochastic hill climbing features from simulated annealing (SA) and the global crossover operations from genetic algorithm (GA). We demonstrated the advantages of GSA by solving one of the most difficult problems in layout --- the non-slicing floorplan design problem. Given the same amount of computing resources, our experimental results showed that GSA consistently obtained better results than SA, in terms of both the chip area and the total wire length. We also applied GSA to timing driven floorplan design and experimental results indicated that it achieved the specified wire length bounds for the critical nets with small penalty on the chip area and the total wire length.