🤖 AI Summary
Poor generalization in wafer defect detection for semiconductor manufacturing—stemming from end-to-end models’ neglect of dynamic inter-stage process dependencies and structural commonalities across products and fabrication processes—is addressed in this work. We propose a manufacturing-stage-decoupled modular temporal neural network. Our architecture decomposes quality prediction into composable, interpretable stage-specific modules that jointly learn from multimodal sensor time-series data, enabling stage-aware feature extraction and cross-process knowledge transfer. Evaluated across diverse wafer types and fabrication processes, our method achieves a 12.7% average improvement in fault detection accuracy over state-of-the-art end-to-end models. Moreover, it supports root-cause attribution for quality deviations, delivering both strong generalization capability and process-level interpretability—key requirements for practical deployment in high-mix, low-volume semiconductor manufacturing environments.
📝 Abstract
The growing availability of sensors within semiconductor manufacturing processes makes it feasible to detect defective wafers with data-driven models. Without directly measuring the quality of semiconductor devices, they capture the modalities between diverse sensor readings and can be used to predict key quality indicators (KQI, extit{e.g.}, roughness, resistance) to detect faulty products, significantly reducing the capital and human cost in maintaining physical metrology steps. Nevertheless, existing models pay little attention to the correlations among different processes for diverse wafer products and commonly struggle with generalizability issues. To enable generic fault detection, in this work, we propose a modular network (MN) trained using time series stage-wise datasets that embodies the structure of the manufacturing process. It decomposes KQI prediction as a combination of stage modules to simulate compositional semiconductor manufacturing, universally enhancing faulty wafer detection among different wafer types and manufacturing processes. Extensive experiments demonstrate the usefulness of our approach, and shed light on how the compositional design provides an interpretable interface for more practical applications.