On soft error rate analysis of scaled CMOS designs — A statistical perspective

This paper re-examines the soft error effect caused by cosmic radiation in sub 90 nm technologies. Considering the impact of process variation, a number of statistical natures of transient faults are found more sophisticated than their static ones. We apply the state-of-the-art statistical learning algorithm to tackle the complexity of these natures and build compact yet accurate generation and propagation models for transient fault distributions. A statistical analysis framework for soft error rate (SER) is also proposed on the basis of these models. Experimental results show that the proposed framework can obtain improved SER estimation compared to the static approaches.