WF-SPR: A weighted single fanout approach for signal probability-based reliability estimation

With continuous improvement of circuit integration, the circuit fault rate has increased. Consequently, accurately and efficiently analyzing and predicting circuit reliability has become a critical objective in digital integrated circuit design. In recent years, as approximate computing circuits (AACs) are getting closer to practical applications, evaluating their reliability has become particularly important. Whether in conventional circuits or AACs, the signal correlation problem caused by fanout nodes can affect the accuracy of circuit reliability evaluation. Moreover, precise reliability evaluation algorithms often come with high computational costs. In this paper, we propose a reliability evaluation method for conventional circuits considering the weight of single fanout nodes based on the signal probability reliability (SPR) method. Meanwhile, this paper improves the method and proposes a new reliability analysis method for AACs. Experimental results on circuits in the ISCAS85 and EvoApprox8b library show that the proposed method has good accuracy and scalability. On both conventional circuits and AACs, the two proposed methods achieve an average error rate of 1% with a time overhead of 0.1% compared to the Monte Carlo (MC) method. On conventional circuits, the average error rate is reduced by 76% compared to the SPR method, and the time overhead is reduced by 97% compared to the SPR multi-pass (SPR-MP) method.