Probabilistic modeling approaches for nanoscale devices

The continual downsizing of silicon technology to nanoscale has enabled the realization of ultra high density, low power chips. However, such devices are inherently unreliable, contingent and prone to soft transient errors. As the deterministic approaches fail to model their behavior, and estimate the effect of soft transient errors on nanoscale devices, many probabilistic approaches have been proposed in literatures. In this manuscript, a comparative study of many of these approaches is presented. A computational framework based on Markov Random Field, Probabilistic Transfer Matrices and Probabilistic Decision Diagram is developed using MATLAB for design and analysis of combinational circuits at nanoscale. It is observed that Bayesian Network and Probabilistic Decision Diagrams have least time complexity among these approaches. The Probabilistic Transfer Matrices and Markov Random Fields are difficult to scale as they require lot of memory and long simulation time. However, Probabilistic Transfer Matrices provide more accurate output error probability.