Scalable analytical model of the reliability of multi-core systems-on-chip by interacting Markovian agents

The reliability of multi-core systems-on-chip has been the object of several studies in recent years since these devices are heavily utilized in modern digital equipment at any level of complexity. This level of integration has caused a reduced time to failure due to the rapid scaling down of the dimension with consequent increase of the cores temperature and current densities. Past studies have utilized discrete event simulation: a technique very difficult to master in this scenario due to the number of components and the rarity of the failure events. The present study proposes an analytical framework based on Markovian Agent Models (MAM), able to capture systems with the big number of cores possible with the today and tomorrow's technology while at the same time considering the effects caused by the position of the cores (center, border, corner) on the temperature level, and the dynamic redistribution of the workload with the progressive failure of the cores. The paper presents the model, adopting realistic parameters and interaction phenomena taken from the literature.