Near-threshold computing (NTC) is a promising technique to reduce the power consumption of very large-scale integration (VLSI) designs. The continuous reductions in the supply voltage present reliability challenges for modern complementary metal–oxide–semiconductor (CMOS) logic due to the occurrence of soft errors from single-event transients (SETs) and multiple-event transients (METs). A fast yet accurate neural network-based model is presented herein to calculate the soft error rate (SER) in circuits in the near-threshold voltage domain. Recurrent neural networks (RRN) are used to model each gate in a given library. A heuristic method for locating multiple faults and propagating them to the circuit outputs based on these neural network models is also presented. On average, the experimental results show that the SER can be estimated up to 20 times faster compared with HSPICE simulations, with less than 0.2% accuracy loss.

近阈值计算（NTC）是一种很有希望的技术，可以减少超大规模集成（VLSI）设计的功耗。由于单事件瞬态（SET）和多事件瞬态（MET）引起的软错误的发生，电源电压的持续降低对现代互补金属氧化物半导体（CMOS）逻辑提出了可靠性挑战。本文提出了一种基于神经网络的快速但准确的模型，以计算阈值电压域附近电路中的软错误率（SER）。递归神经网络（RRN）用于对给定库中的每个门进行建模。还提出了一种基于这些神经网络模型的定位多个故障并将其传播到电路输出的启发式方法。一般，