基于两级遗传算法的特定应用片上网络拓扑优化  

Application-specific Network-on-Chip Topology Optimization Based on Two-level Genetic Algorithm

作　　者：赖国明[1] 林小拉[2] 

机构地区：[1]韩山师范学院计算机应用与技术系,潮州521041 [2]中山大学信息科学与技术学院,广州510006

出　　处：《计算机科学》2013年第2期44-48,70,共6页Computer Science

基　　金：国家自然科学基金(60773199;61073055;60903215);广东省自然科学基金(10152104100004);韩山师范学院博士启动项目资助

摘　　要：大规模片上系统面临着性能、同步、能耗等方面的通信问题,片上网络为复杂的片上系统通信问题提供了一种有效的解决方案。由于网络拓扑优化是一种NP难问题,针对片上系统主要面向特定应用的特点,提出一种基于两级遗传算法来求解特定应用片上网络的拓扑优化问题近似解。实验结果表明,相对于现有的三级遗传,两级遗传算法在最小能耗上平均有1.1%的改进,而且仿真运行时间大大减少,平均减少97%。Large scale system-on-chip is facing several communication problems, such as performance, synchronization and power dissipation. Network-on chip provides a most promise solution to the communication challenges of complica- ted system-on-chip. Because network topology optimization is known to be an NP-hard problem,a^ng at the charactericatic that most system on-chips are application-specific, this article proposed a two-level generic algorithm to solve the ap proximate solutions to the topology optimization problem. Experiments show that the proposed method can, on avera- ge, improve about 1.1 % the minimal power consumption and 97% simulation time which is greatly reduced by compw ring to the existing three-level generic algorithm.

关 键 词：特定应用片上网络 遗传算法 拓扑优化 

分 类 号：TP393.03[自动化与计算机技术—计算机应用技术]