Study on pore features in sintered die-attach microstructures based on machine learning.

Area of Science:

• Materials Science
• Nanotechnology
• Computational Materials Science

Background:

• Low-temperature sintering of nano-sized metal particles is crucial for die-attach bonding.
• The porous microstructure significantly impacts bonding performance.
• Understanding microstructural feature relationships is key to optimizing this process.

Purpose of the Study:

• To investigate the internal relationships among various microstructural features in nano-particle sintered materials.
• To develop effective descriptors for characterizing material morphology using machine learning.
• To assess the impact of microstructural features on die-attach bonding performance.

Main Methods:

• Extraction of pore-related features from Scanning Electron Microscopy (SEM) images.
• Application of correlation analysis and Principal Component Analysis (PCA) for feature reduction and descriptor generation.
• Training and evaluation of four machine learning models (KNN, SVM, Random Forest, ANN) on the transformed dataset.

Main Results:

• Identified mathematical relationships among pore features, categorizing them into pore distribution and shape groups.
• Achieved over 90% accuracy in classifying SEM images using machine learning models.
• Demonstrated the effectiveness of proposed physical feature extraction compared to a variational autoencoder-ANN framework.

Conclusions:

• The developed descriptors accurately and pertinently assess microstructure-property relationships in nano-particle sintered materials.
• Machine learning-based analysis of microstructural features offers a powerful approach for optimizing die-attach bonding.
• This study provides a foundation for predictive modeling of material performance based on microstructural characteristics.