Predictive Age-Aware Runtime Remapping for Many-Core Systems

Transistor scaling has enabled the development of many-core architectures, allowing significant advancements in computational capabilities. However, this progress has introduced critical challenges related to system reliability, as uneven core and communication channel usage can accelerate aging and degrade performance. In this paper, we propose a predictive thermaland traffic-aware aging management strategy that operates at runtime to remap tasks when a core is detected to exceed a predetermined temperature threshold or when communication channels experience overload. By balancing the usage of cores and communication channels, the strategy improves system reliability and longevity. The experimental results demonstrate that our remapping algorithm fits runtime and scalability requirements and effectively mitigates circuit aging compared to systems without this strategy, mitigating circuit aging and enhancing the reliability of many-core systems. This confirms the feasibility of maintaining long-term system reliability through proactive runtime task remapping.