Reproduce “Domain Adaptation for Image Classification of Defects in Semiconductor Manufacturing” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Adrian Poniatowski, Natalie Gentner, Manuel Barusco, Davide Dalle Pezze, Samuele Salti, Gian Antonio Susto

Researchers applied domain adaptation (DA) techniques, including a newly adapted DBACS approach for computer vision, to classify defects in semiconductor manufacturing using real-world Scanning Electron Microscope (SEM) images, demonstrating improved accuracy by aligning data distributions across different product technologies.

Key insights

  • The effectiveness of Domain Adaptation techniques in UDA scenarios for semiconductor defect classification is highly dependent on the specific source-target domain pair.
  • The adapted DBACS approach successfully translates SEM images between domains, consistently outperforming baseline models without adaptation and demonstrating resilience to significant domain differences.
  • While a small amount of labeled target data (SSDA) generally improves performance, the additional complexity of advanced DA techniques does not always provide substantial gains over simply integrating these labels into a standard supervised model for this specific image classification task.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.