Device degradation and resilient computing

In this paper, we study the impact of recent scaling trends on device degradation effects as well as their impact on microarchitecture. A new binning strategy is proposed that takes into account the individual degradation of the processing nodes. The vitality of each processing node is derived from its duty-cycle. By means of random sampling and statistical analysis, the duty-cycle is captured in a resource and time efficient way so that it is feasible to observe more than 100 processing nodes with a single sampling unit. Applications of this kind of degradation handling can primarily be found where conventional reliability verification and life-test acceleration techniques such as burn-in cannot be used anymore due to their complexity and inherent constraints.