Rethinking Wafer Inspection as a Self-Supervised Representation Learning Problem
Computer Science Bulletin · 20.02.2026 · 9 (1) · 266-276
DOI: 10.71465/csb221 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer Science Bulletin, online erschienen am 20.02.2026. Verfasst von Yongqiang Li und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).
Abstract (Originalsprache)
Wafer surface inspection in advanced semiconductor fabrication faces a fundamental data bottleneck: supervised deep learning models capable of classifying defect patterns demand large-scale annotated datasets that are prohibitively expensive to construct in industrial practice. This paper proposes rethinking wafer inspection as a self-supervised representation learning (SSRL) problem, where discriminative feature representations are learned from unlabeled wafer maps without relying on human-annotated defect categories. A dual-stream pretraining framework is introduced, combining a non-parametric instance discrimination branch with a masked reconstruction branch to capture both global spatial distributions and local structural irregularities of defect patterns. Experiments conducted on the WM-811K benchmark demonstrate that the proposed method achieves competitive defect classification accuracy against fully supervised baselines while requiring significantly fewer labeled samples during fine-tuning. The results establish that SSRL constitutes a practically viable and conceptually coherent paradigm for scalable AOI in high-volume semiconductor manufacturing.
Zitieren
APA
Li, Y., Hou, Z., & Clarke, B. (2026). Rethinking Wafer Inspection as a Self-Supervised Representation Learning Problem. Computer Science Bulletin, 9(1), 266-276. https://doi.org/10.71465/csb221
Vancouver
Li Y, Hou Z, Clarke B. Rethinking Wafer Inspection as a Self-Supervised Representation Learning Problem. Computer Science Bulletin. 2026;9(1):266-276. doi:10.71465/csb221
BibTeX
@article{Li2026,
author = {Li, Yongqiang and Hou, Zhenkai and Clarke, Brandon},
title = {Rethinking Wafer Inspection as a Self-Supervised Representation Learning Problem},
journal = {Computer Science Bulletin},
year = {2026},
volume = {9},
number = {1},
pages = {266--276},
doi = {10.71465/csb221}
}
EUVIMED