Circuit Reliability Prediction Based on Dual Autoencoders Combining Hash Coding and Partitioning Strategy

For fast prediction of gate-level circuit reliability,a method based on dual-autoencoder combining Hash coding and partitioning strategy is proposed.Firstly,the main characteristics related to gate-level circuit reliability are analyzed and extracted,and a feature data set for circuit reliability prediction is constructed.Next,facing the difficulties of circuit input vector with variable length and different dimensions,a characteristic standardization method for circuit input vector based on Hash coding and partitioning strategy is proposed.Then stacked autoencoders are used with different activation functions to learn different characteristics of data and establish a dual-autoencoder model.The experimental results on circuits of different scales such as 74 series,ISCAS85 and EPFL show that the proposed characteristic standardization method can improve 1.79%-8.29%prediction performance under the condition of different network layers,and the dual-autoencoder model can improve the prediction performance by 21.10%-47.08%compared with the benchmark model.