Research On Key Technologies For On-chip Network Mapping

Along with the further development of integrated circuit technology,providing much more possibilities of increasing the number of processing elements?PE?integrated on confined chip area,the huge improvements of the integration level of single chips enhance the performances of the chips to a certain extent.However,based on the system-on-chip fragmented bus and shared bus architecture,the problems of insufficient communication bandwidth,excessive power consumption and communication delays from contention for inter-unit communication links on the chip seriously limit the advance which from high level of integration in the performance of the chip.Network-on-Chip?NoC?which is the extension of conception of the System-on-Chip has become the currently mainstream architecture,replacing the system-on-chip architecture thus linking a large number of processing elements on a single chip reasonably and effficicnlly has become an important research direction of the current Network-on-Chip.Facing the mapping of Network-on-Chip which is the problem of non-deterministic polynomial,in this paper,we first improve the temperature descent Function of Simulated Annealing?SA?algorithm and the generation Strategy of neighborhood solution.The simulation results show that the improved SA algorithm has a significant decline in power consumption,While the convergence of the algorithm has been improved.What is more,in order to decrease the complexity of simulation operation and facilitate the research about multi-objective optimization.The equivalent approximation model for mapping is used in this paper,the situation result shows that comparing the results of the mapping of the equivalent approximation model with the results of simulator AccessNoxim₀.2,the relative crror is within 2.3%.has great reference value.Finally,applying the improved NSGA-? algorithm into the mapping solution setting the load balancing and power consutimption as evaluation indicators to assess the performance of improved algorithm in distribution and convergence.Experiments show that the non-dominated solution obtained by the improved NSGA-? algorithm is much closer to the optimal solution,and the distribution has been significantly improved than the traditional NSGA-? algorithm.