This paper proposes a group-based very fast simulated annealing (GB-VFSA) algorithm to achieve higher-quality placement in a shorter CPU time. We first introduce a temperature-dependent perturbation model based on Cauchy distribution to generate new solutions. Then, we put high-connected blocks into one group and use a group as a unit for placement. In order to avoid premature convergence of the algorithm, multiple potential solutions are used to search the solution space at the same time. The idea "pheromone" which comes from ant colony optimization is used to realize the communication between multiple potential solutions. Experimental results using MCNC beachmarks show that GB-VFSA achieved 23% reduction in CPU time and 3.6% improvement in maximal time delay over traditional simulating annealing algorithm.
|ジャーナル||International Journal of Machine Learning and Computing|
|出版ステータス||Published - 2018 12 1|
ASJC Scopus subject areas
- Computer Science Applications
- Information Systems and Management
- Artificial Intelligence