Lightweight Hotspot Detection Model Fusing SE and ECA Mechanisms.

Area of Science:

• Semiconductor manufacturing
• Machine learning applications
• Integrated circuit design

Background:

• Hotspot detection is critical in lithography for ensuring integrated circuit (IC) yield.
• Existing deep learning models can be computationally intensive, posing challenges for real-time applications.

Purpose of the Study:

• To propose a lightweight and efficient machine learning model for lithography hotspot detection.
• To enhance feature extraction capabilities for improved accuracy in identifying hotspots.

Main Methods:

• Integration of Squeeze-and-Excitation (SE) and Efficient Channel Attention (ECA) mechanisms within a convolutional neural network (CNN) architecture.
• Utilizing seven convolutional layers and four pooling layers for feature extraction, followed by three fully connected layers.
• Training and evaluation on a custom layout dataset and the ICCAD 2012 dataset.

Main Results:

• The proposed model demonstrates a lightweight architecture with simplified CNN structure.
• Adaptive channel weight adjustment via SE and ECA mechanisms enhances feature extraction without dimensionality reduction.
• Experimental results show superior overall accuracy, recall, and runtime performance compared to ConvNeXt, Swin transformer, and ResNet 50.

Conclusions:

• The developed lightweight model effectively detects lithography hotspots with enhanced feature representation.
• The integration of SE and ECA attention mechanisms provides a significant performance advantage.
• This approach offers a more efficient and accurate solution for hotspot detection in semiconductor manufacturing.