Quantifying Microelectronics Reliability Under Cumulative Radiation Damage: Stress-Strength Inference with Type-I Censored Data for Failure Probability Estimation
Champagne, Chloe Alexandra
0000-0001-5438-540X
:
2024-03-21
Abstract
The reliability of microelectronics in harsh environments is a vital yet difficult metric for engineers to quantify. While radiation-hardened parts may satisfy reliability requirements within space environments, other computational needs for the mission may suffer. As such, having a framework with which to analyze limited data from radiation tests on commercial off-the-shelf (COTS) parts is highly desirable. Such data may be limited due to the cost of performing the test or obtaining the part. Furthermore, as space has become more accessible to low-cost missions, risk aversion postures towards device qualification have given way to risk-acceptance out of necessity. With standard piece part hardness assurance methods, overconservative design margins and overtest requirements result in hefty testing requirements and arbitrary reliability metrics based on static environment models. In the ever-evolving landscape of space environment modeling, radiation hardness assurance has changed to accommodate new, probabilistic radiation models within an Earth-bound orbit by utilizing stress-strength statistics and confidence contours on the failure parameter space. In this thesis, a probabilistic radiation hardness assurance framework is presented which capitalizes on type-I censored data to bound failure probability estimates to desired confidence levels. By using type-I censored likelihood ratios, the failure distribution parameter space is bounded based on both failure and survivor data obtained during total dose part testing. By masking the failure probability space with the confidence contour and an upper parameter space bound, a worst-case failure probability can be obtained to the designated confidence level, even if failure data was not recorded before the conclusion of the part test. Examples are given using various orbits and part test data, and a comparison of standard hardness assurance techniques to the probabilistic method is provided.