Modified Bayesian Approach to Modeling Commercial Parts with TID Response Variability in Analog Systems

Abstract

Modeling systems intended for radiation environments is an important and challenging reliability problem. Due to their cost, commercial off the shelf (COTS) parts are more prevalent in systems today. However, due to their variability in manufacturing, each part may differ from others in both pre-rad parameters as well as the manner in which it degrades. This proves to make modeling systems made up of these components difficult, as each part has a distribution of potential parameter degradation for any given total ionizing dose (TID).

Much work has been done on modeling systems using a form of Bayesian statistics, and this thesis provides a modified approach to this modeling to predict the system performance as a function of TID. First, a radiation model for an IRF510 power MOSFET is generated. This model is then sampled for each of the identical transistors in an example circuit, which are then used in a SPICE model to emulate an irradiated system. Finally, this model is verified using an experimentation method that involves simulating a large population of test circuits by interchanging irradiated parts.