Statistical Device Modeling with Arbitrary Model-Parameter Distribution via Markov Chain Monte Carlo

We propose a novel statistical device modeling methodology that can represent model-parameters of arbitrary distribution and correlation. The proposed modeling is based on Markov chain Monte Carlo method in which random samples are drawn from the target probability distribution. The proposed method is also independent of the device models, allowing us to apply the method for any device models. Through the validation, the proposed method successfully reproduced the two peaks of the model parameter distribution that generated the current distribution. In addition, the experiments on the measured current variations following a non-Gaussian distribution demonstrate that the proposed method reduced the modeling error significantly as compared to the conventional method that can only use normal distribution.