Energy-Efficient Mapping for 3D NoC Using Logistic Function Based Adaptive Genetic Algorithms  

作　　者：WANG Jiawen LI Li WANG Zhongfeng ZHANG Rong ZHANG Yuang 

机构地区：[1]Institute of VLSI Design, LAPEM, Nanjing University

出　　处：《Chinese Journal of Electronics》2014年第2期254-262,共9页电子学报（英文版）

基　　金：supported in part by the National Nature Science Foundation of China(No.61176024,No.61006018);Research Fund for the Doctoral Program of Higher Education of China(No.20120091110029);A Project Funded by the Priority academic program development(PAPD)of Jiangsu Higher Education Institutions

摘　　要：The problem of mapping application tasks is one of key issues in 3D Network on chip(3D NoC) design. A novel Logistic function based adaptive genetic algorithm(LFAGA) is proposed for energy-aware mapping of homogeneous 3D NoC. We formulate the mapping problem and show the Standard genetic algorithm(SGA). The LFAGA is presented in detail with the goal of obtaining higher convergence speed while preventing the premature convergence. Experimental results indicate that the proposed LFAGA is more efficient than previously proposed Chaos-genetic mapping algorithm(CGMAP). In the experiments, a randomly generated task graph of size 27 is mapped to a 3D NoC of size 3×3×3, the convergence speed of LFAGA is 2.55 times faster than CGMAP in the best condition. When the task size increases to 64 and the 3D NoC size extends to 4×4×4, LFAGA is 2.31 times faster compared to CGMAP. For the No C sizes in the range from 3×3×2 to 4×4×4, solutions obtained by the LFAGA are consistently better than the CGMAP. For example, in the experiment of size 4×4×4, the improvement of final result reaches 30.0% in term of energy consumption. For a real application of size 3×4×2, 18.6% of energy saving can be achieved and the convergence speed is 1.58 times faster than that of the CGMAP.The problem of mapping application tasks is one of key issues in 3D Network on chip （3D NoC） de- sign. A novel Logistic function based adaptive genetic al- gorithm （LFAGA） is proposed for energy-aware mapping of homogeneous 3D NoC. We formulate the mapping prob- lem and show the Standard genetic algorithm （SGA）. The LFAGA is presented in detail with the goal of obtaining higher convergence speed while preventing the premature convergence. Experimental results indicate that the pro- posed LFAGA is more efficient than previously proposed Chaos-genetic mapping algorithm （CGMAP）. In the ex- periments, a randomly generated task graph of size 27 is mapped to a 3D NoC of size 3×3×3, the convergence speed of LFAGA is 2.55 times faster than CGMAP in the best condition. When the task size increases to 64 and the 3D NoC size extends to 4×4×4, LFAGA is 2.31 times faster compared to CGMAP. For the NoC sizes in the range from 3×3×2 to 4×4×4, solutions obtained by the LFAGA are consistently better than the CGMAP. For example, in the experiment of size 4×4×4, the improvement of final result reaches 30.0% in term of energy consumption. For a real application of size 3×4×2, 18.6% of energy saving can be achieved and the convergence speed is 1.58 times faster than that of the CGMAP.

关 键 词：Energy-aware mapping 3D NoC Logisitc function Adaptive genetic algorithm Convergence speed. 

分 类 号：TN791[电子电信—电路与系统]