Recurrent neural networks models for analyzing single and multiple transient faults in combinational circuits

Transient faults analysis is an important step in circuits designing flow. By a fast and accurate scrutiny, it is possible to achieve a cost-effective and soft error tolerant system. In this paper, an efficient and accurate approach is presented to estimate the shapes of transient faults when they are propagating through the gate-level circuits. To provide a reliable prediction of how the shape of a transient fault occurring in a circuit will be, a new method based on recurrent neural networks (RNNs) is proposed. This method can make a confident estimation of the effects that the single/multiple transient faults leave while propagating through a combinational circuit. Results for a sample of 32-bit carry propagation adder shows 22x speed-up with a mean of 0.82 penalty in accuracy loss, compared to the HSPICE simulator.