Vulnerability-aware Dynamic Reconfiguration of Partially Protected Caches

Yuanwen Huang and Prabhat Mishra
University of Florida


Abstract

Cache vulnerability is a serious design concern due to exponential increase in soft errors with technology scaling. Partially Protected Caches (PPC) is a promising solution to mitigate vulnerability to soft errors in resource-constrained embedded systems. However, PPC suffers from both performance and energy overhead. Dynamic Cache Reconfiguration (DCR) is widely used in embedded systems to save energy and improve performance. In this paper, we propose a methodology which takes advantage of the protected cache to reduce vulnerability, while utilizes reconfigurability to explore the trade-off between vulnerability, energy and performance. Experimental results demonstrate that our proposed method can significantly reduce both vulnerability (up to 87%) and energy consumption (up to 41%) without affecting the performance.