Domain Adaptation for Image Classification of Defects in Semiconductor Manufacturing

2025-06-18

AI Summary

In semiconductor manufacturing, defect image classification faces significant cross-domain generalization challenges—where source domains consist of synthetic data or labeled production-line images, and target domains comprise unlabeled or sparsely labeled scanning electron microscope (SEM) images from new fabrication lines. To address this, we propose DBACS, a domain adaptation method built upon the CycleGAN framework. DBACS introduces a novel adversarial loss coupled with a feature-level consistency loss, jointly optimizing unsupervised or semi-supervised domain alignment and discriminative structure preservation. The approach substantially improves model robustness and accuracy for defect classification across diverse SEM devices and process nodes, while reducing reliance on manual annotations and enhancing deployability in industrial settings. Extensive experiments demonstrate that DBACS outperforms state-of-the-art unsupervised domain adaptation methods under multiple cross-domain configurations, offering an efficient and practical solution for vision-based quality inspection in semiconductor manufacturing.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Application Domains: Security

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

In the semiconductor sector, due to high demand but also strong and increasing competition, time to market and quality are key factors in securing significant market share in various application areas. Thanks to the success of deep learning methods in recent years in the computer vision domain, Industry 4.0 and 5.0 applications, such as defect classification, have achieved remarkable success. In particular, Domain Adaptation (DA) has proven highly effective since it focuses on using the knowledge learned on a (source) domain to adapt and perform effectively on a different but related (target) domain. By improving robustness and scalability, DA minimizes the need for extensive manual re-labeling or re-training of models. This not only reduces computational and resource costs but also allows human experts to focus on high-value tasks. Therefore, we tested the efficacy of DA techniques in semi-supervised and unsupervised settings within the context of the semiconductor field. Moreover, we propose the DBACS approach, a CycleGAN-inspired model enhanced with additional loss terms to improve performance. All the approaches are studied and validated on real-world Electron Microscope images considering the unsupervised and semi-supervised settings, proving the usefulness of our method in advancing DA techniques for the semiconductor field.

Problem

• Adapting defect classification models across semiconductor domains
• Reducing manual labeling in semiconductor defect detection
• Enhancing DA techniques for electron microscope images

Innovation

• Domain Adaptation for defect classification
• CycleGAN-inspired DBACS approach
• Unsupervised and semi-supervised settings