Research On NoC Mapping And Buffer Allocation Algorithm Under Self-similar Traffic

As the semiconductor integrated circuits develop rapidly,the integration degree of chip components is getting higher and higher.The bus structure has already been unable to meet the needs of chip designs.In order to solve this problem,Network on Chip(NoC)is proposed as a new communication architecture.The performance of NoC is highly dependent on the actual traffic,so a series of researches based on traffic have great significance.Based on the characteristics of self-similar traffic in NoC,the problems of NoC mapping and buffer allocation are studied emphatically.The results of NoC mapping largely affect the performance of energy and delay,but the current researches on NoC mapping hardly considered the effect of self-similar on-chip traffic characterization on the energy and delay.For this reason,the energy model is established by minimizing the hop counts between communicating cores under the NoC self-similar traffic.Besides,the delay factor concept is introduced to establish the optimization delay model,and the delay performance is optimized from three aspects:the system load distribution,traffic burst strength and the system non-congestion delay.According to the characteristics of the mapping problem,this paper optimizes the energy and delay factor by using an improved Non-dominated Sorting Genetic Algorithm(NSGA?).Simulation results of several sets of task graph show that the improved algorithm can reduce the energy and delay factor obviously,and achieves better convergence compared with the traditional NSGA? algorithm and MOGA algorithm.As for the problem of NoC buffer allocation,on the basis of analyzing the router structure,a queuing delay calculation model is established under self-similar traffic.Based on the genetic simulated annealing algorithm,an optimized algorithm of buffer allocation is designed.The results show that this algorithm can greatly reduce the average queuing delay of NoC input channel.