Responsible defect detection with probabilistic contrastive counterfactuals

In semiconductor manufacturing, defect detection is critical to ensuring high yield and quality. Traditional methods often rely on deterministic thresholds or black-box machine learning models, which lack transparency and fail to provide actionable insights into root causes of defects. This work introduces responsible defect detection with probabilistic contrastive counterfactuals, a novel framework that integrates probabilistic causal modeling with contrastive counterfactual explanations to enhance interpretability and accountability in defect detection systems. By analyzing the process data, the framework identifies not only when and where defects occur, but also why they occur by contrasting actual scenarios with minimally altered hypothetical scenarios where defects are absent. This approach supports data-driven decision-making, fostering trust in AI-driven quality control while promoting actionable insights to mitigate defects and optimize manufacturing processes.