Component Reliability Modeling Through the Use of Bayesian Networks and Applied Physics-based Models

The objective of the work presented in this paper is to develop a practical methodology to support order of magnitude probabilistic prediction of Commercial Off-The Shelf (COTS) components with little or no screening, qualification, or operational data. The physics-based approach to degradation/failure modeling of dominant failure mechanisms include: Time-Dependent-Breakdown-Dielectric (TDDB), Hot-Carrier Injection (HCI), Negative Bias Temperature Instability (NBTI) and Electromigration (EM). Moreover, it is crucial to consider other external factors that may eventually reduce the reliability of the device. These factors vary based on the fabrication, process, design and the product itself. This paper proposes an approach to identify and integrate all of the relevant qualitative and quantitative information using Bayesian Networks (BNs). The ultimate goal is to develop and validate an infrastructure of methods and tools for order of magnitude reliability assessment on the range of available information including results of physical models, statistical evidence, and expert opinion. The method is demonstrated by assessing the reliability of SRAM chips.