MIT SIAM Seminar
Thursday, October 25, 2018 at 4:00pm to 5:00pm
Semiconductor parameter extraction (and more!) with Bayesian inference
Rachel Kurchin, graduate student, DMSE
Thursday, 10/25, 4-5 PM in room 4-231, refreshments provided
Abstract: Bayesian parameter estimation is a widely-used approach for model optimization in a variety of fields including astrophysics, high-energy physics, and bioinformatics. However, it has not been adopted extensively for electronic device characterization. We have developed a generalized open-source Python code, Bayesim, that accepts sets of observed data as a function of experimental conditions and modeled data as a function of those same conditions as well as a set of parameters to be fit, and outputs a probability distribution over these parameters, accounting for both experimental and model uncertainty. Because models of electronic devices are frequently computationally expensive, we adopt a deterministic and adaptive scheme for sampling the parameter space and computing model uncertainty. I will discuss applications of the code in fundamental characterization of photovoltaic materials as well as current and planned future features, and leave plenty time for discussion of how Bayesim might be useful for your application!