Research On Mapping Technology For The Three-dimensional NoC

With the development of integrated circuit technology,more components with a specific function will be connected to the network for the goal of Internet of Everything(Io E).No C is a new communication mode that can realize the integration of ultra-large scale systems on a single chip.However,there's increasing in the number of IP cores.The power consumption,performance,packing density and area of 2D No C have reached a bottleneck because of its limitations.And 3D No C has become a very important research direction for MPSo C with its advantages of its shorter interconnection,higher performance and greater packing density.The latency and power consumption of 3D No C has been the subject of study.Currently,the majority of research to explore 3D No C advantages is in terms of topology,routing algorithm and switching mechanisms.But they are all on one single level of research.In this paper,power consumption and delay are researched via two aspects in two steps,one is in the application layer with mapping technology and another is in link layer.In this paper,we exploit the application layer to consider No C power consumption optimization,which has a high degree of abstraction.And the number of IP cores in 3D No C will be more,the search space will be larger,the calculation of an amount of data will be more complex.In this paper,mapping algorithm based on adaptive genetic algorithm for mesh No C is presented to minimize power consumption for the difficulties brought about by the rapid search in the huge search space,where the adaptive crossover and mutation operator is used to improve the convergence of the algorithm,and finally the power consumption is minimized on the basis of the simple genetic algorithm,In this paper,the proposed algorithm is verified with three examples of MPEG-4,VOPD and I27,respectively.The application mapping to the 2D and 3D mesh No C.Compared with the mapping algorithm based on the simple genetic algorithm,the optimal solution of AGAM has different degree of optimization,which is as high as 24.26%.For 3D mesh No C,with the growth of the number of cores,the optimized power consumption after mapping is more and more significant.And the convergence rate of the algorithm is as high as 82.84%,the average optimization is 60.61%,and the convergence is greatly enhanced.For the No C delay problem,because increasing the network link bandwidth can reduce the congestion caused by excessive data in the process of network traversal,thereby improving the performance and the communication efficiency of the entire network.Based on this,we proposed a delay optimization method using configurable links to construct unbalanced networks which was based on mapping algorithm mapping result AGAM generated,built a low-power No C with configurable links.The network performance is optimized by increasing the bandwidth of the maximum traffic nodes of this network and constructing the unbalanced network.The method still was verified by MPEG-4,VOPD I27 three application examples,which built the corresponding 2D and 3D mesh No C by the tool of OPNET.The simulation results show that the delay rate of single link delay of unbalanced link network is higher than that of balanced network,which is up to 8.68%,which is 3.47 times of the single delay optimization rate of balanced network.This method reduces the network delay,while effectively reducing the waste of bandwidth resources.