IC Packaging Material Identification via a Hybrid Deep Learning Framework with CNN-Transformer Bidirectional...

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Miniaturization of electronic components and integrated circuits (ICs) is advancing rapidly due to micro- and nanomanufacturing technologies.
• Accurate identification of IC packaging materials is crucial for ensuring the reliability and performance of miniaturized electronic devices.

Purpose of the Study:

• To develop an automated method for identifying IC packaging materials using deep learning.
• To introduce a novel hybrid deep learning framework that effectively captures both local and global features from IC packaging images.

Main Methods:

• A hybrid deep learning framework, the CNN-transformer interaction (CTI) model, was designed.
• The CTI model utilizes cascaded blocks, each containing Convolutional Neural Networks (CNNs) for local features and Transformers for global and local-window features.
• A bidirectional interaction mechanism facilitates feature transfer between CNN and Transformer branches in both channel and spatial dimensions.

Main Results:

• The CTI model demonstrated high performance in recognizing three types of IC packaging materials.
• The framework achieved an F1-score of 96.16% and an accuracy of 97.92%.
• Performance surpassed existing deep learning methods for IC packaging material identification.

Conclusions:

• The proposed CNN-transformer interaction (CTI) model is effective for automated IC packaging material identification.
• The hybrid approach offers superior performance compared to conventional deep learning techniques.
• This method contributes to ensuring the reliability of miniaturized integrated circuits.