Soft errors have become a critical reliability concern for advanced CMOS designs because of the continuous technology scaling. Therefore, it is necessary to develop an approach that correctly estimates soft error rates (SERs) and considers process variation during design sign-off. Because of computational inaccuracy in previous approaches, a fast-yet-accurate framework is proposed in this paper for statistical soft-error-rate (SSER) analysis. This approach consists of two components: (1) intensified learning with data reconstruction, and (2) automatic bounding-charge selection. Experimental results show that the proposed framework increases the SER computation speed by 10 7 X, with only 0.8% accuracy loss when compared to the Monte-Carlo SPICE simulation.Index Terms-soft error, process variation, SVM, Monte Carlo