Formal Quantification of the Register Vulnerabilities to Soft Error in RTL Control Paths

Abstract

Radiation-induced soft error is a significant reliability issue in nanoscale technology nodes. As the sequential registers are major contributors to the system soft error rate, accurate analysis of their vulnerabilities at early design phases is essential for cost-efficient error mitigation. In this paper, a novel approach is proposed to quantify the soft error vulnerabilities of the registers in control paths at Register-Transfer Level (RTL). By modeling the control path as a state transition system, formal probabilistic model checking is adopted to compute the register vulnerabilities by taking the workload dependency into consideration. Efficient RTL abstraction and model simplification techniques are proposed to achieve an exponential reduction of the state space, enabling our methodology to analyze large control modules in a typical embedded processor. The experimental results show the effectiveness of the proposed techniques, which can successfully quantify the non-uniform register vulnerabilities in RTL designs.