Cross-Process Defect Attribution using Potential Loss Analysis

2025-07-27

AI Summary

In semiconductor manufacturing, cross-process root-cause analysis of wafer defects is highly challenging due to heterogeneous process paths and complex causal chains. To address this, we propose the Potential Loss Analysis (PLA) framework—the first to jointly model defect density prediction and defect source attribution. PLA employs counterfactual reasoning to estimate the optimal defect density achievable at each upstream process step under partial processing trajectories; it then attributes observed high defect density to the process step inducing the largest “potential loss.” Methodologically, identifying the optimal outcome is formulated as solving a Bellman equation, efficiently realized via dynamic programming coupled with partial-trajectory regression. Experiments on real-world wafer data demonstrate that PLA significantly improves both root-cause localization accuracy and interpretability, while maintaining high-fidelity defect density prediction.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Cross-process root-cause analysis of wafer defects is among the most critical yet challenging tasks in semiconductor manufacturing due to the heterogeneity and combinatorial nature of processes along the processing route. This paper presents a new framework for wafer defect root cause analysis, called Potential Loss Analysis (PLA), as a significant enhancement of the previously proposed partial trajectory regression approach. The PLA framework attributes observed high wafer defect densities to upstream processes by comparing the best possible outcomes generated by partial processing trajectories. We show that the task of identifying the best possible outcome can be reduced to solving a Bellman equation. Remarkably, the proposed framework can simultaneously solve the prediction problem for defect density as well as the attribution problem for defect scores. We demonstrate the effectiveness of the proposed framework using real wafer history data.

Problem

• Identify root causes of wafer defects in semiconductor manufacturing
• Compare processing trajectories to attribute defect densities
• Solve defect prediction and attribution using Bellman equation

Innovation

• Potential Loss Analysis for defect attribution
• Compares best outcomes from partial trajectories
• Reduces task to solving Bellman equation