Analog Fault Diagnosis Using Bayesian Framework and Structural BIST
Abstract
The integration of analog and digital components in monolithic SoCs makes analog circuits increasingly susceptible to manufacturing defects and in-field failures. While prior work has focused on defect detection, accurate defect diagnosis at transistor level remains a critical challenge, particularly during yield learning and manufacturing ramp where fine-grained localization is necessary. This paper presents a Bayesian framework for transistor-level defect diagnosis using delay-based structural Built-In Self-Test (BIST). The proposed method leverages multiple delay measurements, captured between injected perturbations and observed responses, as diagnostic signatures. These signatures are processed using a probabilistic inference engine to estimate the likelihood of each defect, enabling robust diagnosis even in the presence of process variations and overlapping defect signatures. The framework is validated on a 65nm analog circuit with 416 injected transistor-level defects. Results demonstrate high diagnostic efficiency (up to 0.996) while maintaining minimal hardware overhead (0.002%) and sub-second execution time. The proposed approach is scalable and uses generic algorithm making it valuable for wide range of applications.