Based On Several Key Technologies In The Design Phase Of The Network

Along with the rapid development of deep submicron integration technology,the number of IP cores integrated on a single chip keeps increasing.Therefore,traditional bus structure adopted in SoC cannot meet the following communication requirement.Traditional SoC faced problems on communication bandwidth,communication power consumption and global synchronization.In recent years,to solve the limitations caused by bus structure and then use computer network technology in chip design,NoC was proposed by researchers.Bus structure was replaced by routing and block-switch technology to appease the system performance requirement of hundreds of processing units integrated on a single chip.And by exchanging network protocol and computing protocol,the complete system with computing and communication function was realized.Thus,the shortcoming of shared bus structure and the increasing communication scale inside the system were solved from the architecture.Floorplan is the first step of designing chips;it is the key affecting the performance of the whole chip on delay and power consumption.So the research on better floorplan algorithms is very necessary.This thesis focuses on floorplan algorithms for NoCs,and the main contributions can be summarized as follows.First,this thesis improves simulated annealing algorithm on its low convergence rate and low optimization efficiency:we improve the searching strategy and the possible bad transfers,and lead the improved simulated annealing algorithm into particle swarm optimization algorithm.As the simulation results show,compared with traditional simulated annealing algorithm,the novel hybrid algorithm we proposed can save at most 73.7%time,prolong the reduction ratio at most 51.323%and averagely increase the throughput on 3.67%and at most 13.03%.Second,we propose using multi-objective algorithm into particle swarm optimization algorithm.We coordinate contradictions among sub objectives by the concept of pareto best solution.And the experiments verify that multi-objective particle swarm optimization algorithm is more reliable that single-objective particle swarm optimization algorithm.