| With the semiconductor technology scaling into the Nano Era,the performance growth of single-core processors is severely limited by the power dissipation,the interconnection wire delay and the design complexity,etc.Therefore,it is becoming a main processor designing trend to integrate massive relatively simple processor cores on a chip,rather than the more complex single-core.However,limited by the process level,many-core processors are also facing the problem of low yield and reliability reduction,which can cause difficulty using many-core processors in safety-critical areas.Meanwhile,using fault-tolerant technology to improve the yield and the reliability is a feasible way to reduce the difficulty.At present,a kernel-level redundancy mechanism,called N+M,is considered as a very effective kernel-level hardware fault tolerance technology.Based on this idea and considering the defect clustering effect,topology reconfiguration algorithms for the optimization objective of reliability,or that of both reliability and time performance,are studied deeply on the model of 2D-Mesh homogeneous many-core processor.And results are as follows:(1)Due to the smaller size of transistors,the original nonsignificant defects on the wafer become more and more significant and show the clustering characteristics,which lead to the clustering failures of cores.Therefore,it is necessary to demonstrate the clustering characteristics of failure cores objectively when establishing the failure cores distribution model of the many-core processor,and that is the first prerequisite to evaluate the efficiency of the relevant topology reconfiguration algorithms more objectively.In respect of shortcomings of the existing modeling methods,a new one based on the defects clustering effect is proposed in this paper.The experiments show that the proposed method can effectively reflect the clustering characteristics of failure cores,and those characteristics also affect the repair rate of some core-level redundancy topology reconfiguration algorithms significantly.(2)In order to ensure that the systems' service is highly dependable,reliability is the basic design and acceptance standard of these safety-critical systems in fields of aerospace,military equipment,automatic driving,etc.The existing core-level redundancy topology reconfiguration algorithms revolves around the time performance.In order to further improve the reliability of a many-core processor after reconfiguration,and expand its application in safety-critical areas,a reliability-centred core-level redundancy topology reconfiguration algorithm,which is established on the basis of a 2D-Mesh many-core processors reliability model and using the dagger sampling Monte Carlo method as the reliability evaluation approach,is proposed.Experiments show that the reliability of many-core processors configured by this method is significantly improved.(3)Based on the study of reliability single-objective optimization problem,a dual-objective optimization problem model for the reliability and time performance of many-core processors core-level redundancy topology reconfiguration is established in order to meet the requirements of both reliability and performance in some areas.Then,by combining the advantages of constraint method with that of stratified sequencing method,we proposed a solution of the dual-objective optimization problem.Finally,a reliability and time performance dual-objective optimization algorithm for many-core processors topology reconfiguration is designed and implemented by Simulated Annealing idea.Experiments show that the reconfiguration algorithm can greatly improve the time performance of the system under the premise of effectively guaranteeing the reliability target.That is because random mapping phenomenon of candidate cores is bettered.The research results in this paper provide some references for engineering practice,and also build foundation for designing more efficient algorithms of many-core processors topology reconfiguration. |
PDF Full Text Request