Adaptive Test Planning for Semiconductor Reliability via Bayesian Monte Carlo Tree Search
A new study, detailed in a paper on arXiv (2608.09622), introduces an innovative adaptive framework for planning tests that aim to ensure the reliability of advanced semiconductor devices. This framework addresses the challenge of making sequential stress decisions that balance different characterization objectives against various failure mechanisms. Unlike traditional static test plans that rely on general population models, this method is dynamic and considers individual differences and real-time degradation data. It treats reliability qualification as a sequential decision-making problem, utilizing Monte Carlo tree search (MCTS-SA) and extended Kalman filter (EKF) for estimating belief states. By factoring in variability from bias temperature instability, electromigration, and time-dependent dielectric breakdown, it optimizes stress selection to improve qualification success rates. Developed by researchers, this approach could enhance efficiency and reliability in chip testing, leading to faster market readiness.
Key facts
- Paper arXiv:2608.09622 proposes adaptive test planning for semiconductor reliability.
- Uses Monte Carlo tree search with seed-action simulators (MCTS-SA).
- Employs extended Kalman filter (EKF) for belief-state estimation.
- Models variability in BTI, EM, and TDDB failure mechanisms.
- Formulates reliability qualification as a partially observable sequential decision problem.
- Aims to maximize probability of successful qualification under constraints.
- Addresses limitations of static test plans based on population-level models.
- Published on arXiv as a new announcement.
Entities
Institutions
- arXiv