The Anti-aging Technology For Multicore Systems

As CMOS process aggressively scaling into deep submicron, the reliability of integrated circuits is facing severe challenge from aging effect, and its lifetime is negatively affected by wear-out. In some fields such as the military, aerospace or where the high reliability of circuit is required, Aging of the circuit becomes one of the issues to be addressed,which is an urgent need to improve the reliability. The prediction and prevention of aging effect are both studied in this paper. The Electromigration effects is taken as an example of aging effect, an analytical model is proposed based on the existing RAMP model and an task scheduling based on mixed integer linear programming is presented on heterogeneous multi-core systems for mitigating aging.Existing system-level reliability models are often considered incompletely, or applied only in some particular cases. By the association of temperature, the fixed parameter of RAMP model is substituted. The new model is related to the task scheduling after amending. The proposed model combines many factors that affected reliability such as frequency, voltage, power and stress time. The model divides the aging factors of system-level into two types, one is the intrinsic factors, primarily related to the manufacturing process, etc. and the other is the extrinsic factors, mainly related to the way it is used. The new model is fully considered and related to some key parameter of task scheduling, so it is easy to use it for mitigating the effects of aging by task allocation.The workload balancing algorithms for homogeneous multicore system are inapplicable to heterogeneous multicore systems, and will increase the gap of reliability of the cores. Therefore, considering the impact of these factors, a MILP-based task scheduling method is proposed, the optimal goal is balancing the MTTF of processing cores, instead of workload balance. The corresponding task scheduling is given by solving the MILP and the MTTF of heterogeneous systems is improved. Experiments show that compared to the temperature balance, load balancing algorithms and other balancing algorithms, our method is more applicable to alleviate the problem of aging for heterogeneous multicore systems.