Network On Chip Mapping Based On Adaptive Routing Algorithm

Since its birth,semiconductor process technology and integrated circuit design and manufacturing technology have maintained a high-speed development and iteration speed different from the traditional industry.Especially since the 1990 s,with the rapid development of single Chip integration technology,System on Chip(SoC),which is characterized by the integration of the whole System on a single Chip to achieve specific tasks,has gradually become a hot spot in the development of integrated circuits.Network on Chip(NoC),a new System communication architecture,comes along with the development and demand of SoC,especially Multiprocessor System on Chip(MPSoC).As an important part of SoC design and research,NoC has become an important research field of SoC.The object of this paper is the on-chip network mapping optimization problem in NoC research,which belongs to the application modeling optimization problem.The general idea of this problem is to determine the corresponding mathematical model according to the optimization objective,and solve the problem approximately by heuristic algorithm according to the mapping of the given application task to the network topology.The work of this paper mainly includes the following three points:Firstly,three important parameters,power consumption,delay and load,which are used to measure the performance of the target system under different routing strategies and switching mechanisms,are selected for mathematical modeling,and their respective characteristics and similarities and differences are discussed.According to the target system studied in this working group,a load model based on Markov transfer matrix and the multi-objective weighted optimization function of the shortest path power model are proposed as the evaluation scheme of the target system.Secondly,according to the evaluation scheme of the target system proposed in this paper,the Wolf pack algorithm(WPA),a new type of swarm intelligence algorithm,is designed as a discrete algorithm,which is used as an approximate optimization algorithm to solve the NoC mapping problem of the target system,and the discrete design principle and execution steps of the algorithm are given.Finally,select a number of commonly used application task as the experimental target simulation experiment,and select based on standard genetic algorithm to improve theadaptive simulated annealing genetic algorithm and random mapping algorithm as the control experiment,and according to the experimental results verify the rationality of the optimization model and different routing policy and exchange mechanisms on the result of optimization.