性能约束下功耗感知的电压频率岛NoC映射  

Power aware mapping for NoC with voltage-frequency islands under performance constraints

作　　者：任向隆[1] 高德远[1] 樊晓桠[1] 安建峰[1] 

机构地区：[1]西北工业大学计算机学院,陕西西安710072

出　　处：《华中科技大学学报（自然科学版）》2012年第12期28-33,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60736012;60773223;61003037;61173047);国家高技术研究发展计划资助项目(2009AA01Z110);西北工业大学基础研究基金资助项目(JC201212)

摘　　要：针对支持电压频率岛(VFIs)的片上网络(NoC)功耗优化问题,定义了性能约束的功耗感知NoC映射问题,并提出一种基于遗传、蚂蚁算法融合的优化方法.通过在映射过程中同时考虑计算功耗、VFIs开销功耗及通信功耗,提高了算法的优化能力,降低了系统的总体功耗;通过将遗传算法与蚂蚁算法融合,利用遗传算法的快速搜索能力、蚂蚁算法精确优化能力,使优化算法兼顾了收敛速度和优化效果.实验结果表明:本算法在满足NoC性能要求的前提下,可显著降低VFIs NoC的功耗;具有收敛速度快,优化精度好的特点,适用于求解大规模NoC映射问题.To solve the power optimization of network-on-cship（NoC） with voltage-frequency islands（VFIs）,power-aware NoC mapping with performance constraints was formulated.An optimization method based on fusion of genetic algorithm and ant algorithm was proposed.With considering the power consumption of processors and level shifters between different VFIs besides the communication during the mapping,the algorithm makes much progress in optimization and reduces the overall power consumption of system.Through fusing genetic algorithm and ant algorithm together,and taking advantage of the fast searching ability of genetic algorithm and precise optimization capabilities of ant algorithm,the optimized algorithm takes convergence speed into account as well as optimization precision.Experimental results show that the proposed algorithm can significantly reduce the power consumption of NoC with VFIs and meet its performance requirements simultaneously.Experimental results also indicate that the algorithm has good optimization precision,fast convergence,and is suitable for solving the large-scale NoC mapping problems.

关 键 词：片上网络 电压频率岛 低功耗 映射 遗传算法 蚂蚁算法 融合 性能 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]