Efficient SRAM Failure Rate Estimation Via Markov Chain Important Sampling

The occurrence probability of the rare event is extremely small, but rare-event simulationis very important. It has been widely applied in many fields, such as space engineering,insurance, medicine. As the development of submicron technology, SRAM failure rateestimation is of great importance. As to this special rare event, the study has alwaysfocused on the normal distribution recently. In this paper, we study the simulation ingeneral distributions.In this paper, firstly, we introduce some theories of rare-event simulation (na vesimulation：Monte Carlo、Importance Sampling、Gibbs Sampling). Secondly, we introduceMarkov chain Monte Carlo Sampling, and then apply these theories in the SRAM failurerate simulation. Finally, according to the rules of accuracy and efficiency of the resultsreferred in this paper, through some related simulations in SAS, we inspect the results ofSRAM failure rate simulation in different distributions(multivariate normal distribution、multivariate t distribution). All results indicate that to get a reliable estimation, thesimulation frequency of MCIS Sampling is much lower than that of MC Sampling.Therefore, our simulation is feasible.