graph-hotspot-detection

This repository presents my complete undergraduate graduation project (2025) from Guangdong University of Technology (Automation), titled
“Research on Lithography Hotspot Detection Algorithm Based on Deep Learning”, focusing on lithography hotspot detection using deep learning and graph-based methods.

The project includes full documentation, implementation code, and trained model weights, aiming to provide a comprehensive reference for research and engineering practice in computational lithography and EDA.

本仓库展示了我在广东工业大学（自动化）完成的本科毕业设计项目（2025年），项目名称为 “基于深度学习的光刻热点检测算法研究”。该项目主要研究基于深度学习和图论的光刻热点检测方法。

项目包含完整的文档、实现代码和训练好的模型权重，旨在为计算光刻和EDA领域的研究和工程实践提供全面的参考。

该项目获得2025届自动化学院自动化专业本科毕业设计创新奖（排名：5/10，专家评审组评分：91/100）。

🔍 Overview

Lithography hotspot detection is a critical problem in IC manufacturing. Traditional approaches rely heavily on pattern matching or rule-based methods, which often struggle with generalization.

This project explores a Graph Neural Network (GNN)-based framework to model layout patterns and detect hotspots more effectively.

Key features:

• Graph-based layout representation

• Deep learning-driven hotspot classification

• Comparison with prior work (Geng et al., ICCAD 2020)

• Full undergraduate research workflow documentation

📁 Repository Structure

.
├── docs/               # Complete academic documents
├── src/
│   ├── comparation/   # Reproduction of prior work (Geng2020)
│   ├── models/        # Trained model weights (.pt)
│   ├── scripts/       # Core implementation (train/test/backbone)
│   └── utils/         # Utility functions
├── checkpoints/       # Additional model checkpoints
├── assets/            # Figures and diagrams for visualization
├── scripts/           # Entry scripts

📄 Documentation

The /docs directory provides a complete set of academic materials from an undergraduate graduation project in a Chinese university. It covers the full lifecycle of the project, from proposal to final evaluation and award application.

All personal information has been removed for public release.

📘 Thesis

• undergraduate_thesis.pdf （Chinese：毕业论文原文）
• undergraduate_thesis.docx

🎤 Defense

• defense_slides.pdf（Chinese：答辩演示文稿）
• defense_slides.pptx
• defense_record.pdf（Chinese：答辩记录表）
• defense_record.docx

📝 Reports

• proposal_report.pdf（Chinese：开题报告书）
• proposal_report.docx
• midterm_report.pdf（Chinese：中期检查表）
• midterm_report.docx

🏆 Awards

• innovation_award_application.pdf（Chinese：毕业论文创新奖申请表）
• innovation_award_application.docx
• innovation_award_brief.pdf（Chinese：毕业论文创新奖“小作文”）
• innovation_award_brief.docx

📑 Evaluation

• evaluation_form.pdf（Chinese：毕业论文评阅表）
• evaluation_form.docx

📋 Supervision Records

• supervision_record.pdf（Chinese：导师指导记录）
• supervision_record.docx

📎 Appendix

• appendix_I.pdf（Chinese：“附表 I”）
• appendix_I.docx

Note: "appendix_I" is used to illustrate the positive significance of the graduation project's solution in social responsibility, engineering management, and economic decision-making.

💡 These materials may be useful for:

• Understanding the structure of undergraduate thesis projects in Chinese universities
• Reference for academic writing and documentation standards
• Educational and research purposes in engineering disciplines

🧠 Method

The proposed method models IC layout patterns using graph structures and applies deep learning techniques for classification.

Highlights:

• Structured representation of layout geometry
• GNN-based feature extraction
• End-to-end hotspot detection pipeline
• Comparative study with baseline methods

📊 Dataset

The dataset is not included in this repository due to size constraints.

You can obtain it from: 👉 https://github.com/Intelectron6/Lithography-Hotspot-Detection

🚀 Usage

1. Training

Source file located at: src/scripts/Train.py

2. Testing

Source file located at: src/scripts/Test.py

📦 Model Weights

Pretrained model weights are provided in:

src/models/

📚 Reference

If you are interested in the baseline method:

• Hao Geng, Haoyu Yang, Lu Zhang, Jin Miao, Fan Yang, Xuan Zeng, and Bei Yu. 2020. Hotspot detection via attention-based deep layout metric learning. In Proceedings of the 39th International Conference on Computer-Aided Design (ICCAD '20). Association for Computing Machinery, New York, NY, USA, Article 16, 1–8. https://doi.org/10.1145/3400302.3415661

🏆 Acknowledgement

This work received the Outstanding Innovation Award for undergraduate graduation design at Guangdong University of Technology.

The research results have been published in the following journal:

• Liang, L., Cai, S., & Zhang, H. (2025). Hotspot detection algorithm based on graph-family modeling and graph neural networks. Journal of Guangdong University of Technology, 42(6), 34–43. https://doi.org/10.12052/gdutxb.250111

⚠️ Disclaimer

This repository is intended for academic and educational purposes only.

📬 Contact

If you find this project helpful, feel free to open an issue or contact me.