EM-Reliability System Modeling and Performance Optimization for High-Performance Microprocessors

This article presents a new approach for system-level reliability management technique for multi-core microprocessors. In the new approach, the electromigration (EM) induced mean time to failure (MTTF) at the system level is modeled as a resource, which is abstracted from a more physics-based EM model. Based on such resource-based EM models, we propose a novel task migration method to explicitly balance the consumption of EM resources for all the cores. The new method leads to the equal chance of failure of these cores, which will maximize the life time of the multi-core system. Low power mode control is enabled to compensate the excessively consumed life time of all the cores, giving the multi/many core system more flexibility to handle heavy task assignment under a given reliability requirement. Experimental results on a 36-core processor platform shows that the proposed task migration scheme could balance and compensate the life time consumption of all the cores as desired.