FPN and BAM Module-Enhanced Swin Transformer for Lithography Hotspot Detection

Read full text Summarise with VIRAT
Authors: <p>Lu Chunyan</p>
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

To address the persistent challenges of low F1-score and high false alarm rate (FAR) in lithography hotspot detection, this study proposes a new detection framework based on a Dual Attention Swin transformer (DA-Swin). The proposed architecture leverages the multiscale feature fusion capability of the Feature Pyramid Network (FPN) to enhance the model's sensitivity to scale-invariant features. Meanwhile, the Bottleneck Attention Module (BAM) establishes a channel-spatial cooperative attention mechanism, achieving a 22.2% reduction in FAR. Compared to CrossEntropyLoss, the adoption of BCEWithLogitsLoss significantly reduces output layer parameters by approximately 50%. Benchmark tests on ICCAD demonstrate that our method achieves an accuracy of 98.67% and an F1-score of 0.857, surpassing prior works (TCAD'18, ICCAD'20, and JM3'24). Ablation studies further confirm the synergistic optimization of each module, delivering a highprecision (>98%) and costeffective (14.85% faster inference speed) solution for advancednode IC layout verification.

Fields of Study & Subjects

Industrial Vision Systems and Defect Detection Advancements in Photolithography Techniques VLSI and FPGA Design Techniques

Publication details

Journal / source
Academic Journal of Computing & Information Science
Volume
9
Issue
2
Publication date
1 January 2026
Language
EN