Fault Prediction And Fault Tolerant Routing In 3D NoC

With the expansion of the scale of three-dimensional network-on-chip(3D-NOC)and the improvement of its process technology,the performance of 3D-NOC has been doubled.However,the importance of fault prediction mechanism,which was delayed due to insufficient performance and cost problems,has gradually increased.In recent years,the fault prediction of three-dimensional network-on-chip(3D-NOC)and the routing algorithm corresponding to the fault points have become an important direction.A successful fault prediction system can not only prevent the performance loss and cost increase caused by core damage,but also greatly improve the service life of 3D-NOC with perfect fault-tolerant routing algorithm.The existing prediction models and routing algorithms do not meet the goal of improving the overall reliability of the network in the actual network-on-chip environment.In this paper,fault prediction and fault-tolerant routing in three-dimensional network-on-chip are briefly introduced,and the research status of these two problems at home and abroad is analyzed.A prediction model with long validity and high accuracy for three-dimensional network-on-chip with mesh structure is designed,and an algorithm for fault-tolerant routing based on two-dimensional fault zone and tunnel is proposed,which has low power consumption and high accuracy.The advantages of delay and low packet loss rate.The prediction model and fault-tolerant routing are integrated into a complete prediction-fault-tolerant system,which realizes a series of measures from fault prevention to response.Firstly,this paper summarizes the achievements and research status of the existing prediction models at home and abroad,and concludes that the existing research results can not meet the two requirements of prediction accuracy and prediction delay at the same time.Therefore,the reliable prediction model can not be used in the actual network-on-chip operation environment.In order to solve this problem,a prediction model for the actual operating environment of network-on-chip is designed based on the structure of deep learning network,which is relatively simple.This prediction model can provide a more accurate prediction result in a shorter time than other neural networks and complex deep learning models.Then,this paper predicts the results of the model and designs a more reliable routing algorithm based on the low reliability of the current routing algorithm.This algorithm uses the advanced fault zone and tunnel model of NoC,and uses the tunnel priority strategy to ensure the routing performance of NoC in high failure rate environment.Finally,a complete prediction-fault-tolerant routing system will be established to adapt the actual operating environment of the existing three-dimensional network-on-chip,so that when the number of fault nodes increases in the network-on-chip environment,it can accurately predict the fault nodes and maintain a certain routing performance to ensure the reliability of the network-on-chip...