Improving Machine Learning-Based Identification of Defects and Thermal Resistance in Microelectronics and Composites
Project Description
Reliability and performance in microelectronics and composite materials are strongly influenced by hidden defects (voids, delaminations, poor interfaces) and local variations in thermal resistance. Conventional inspection techniques often trade off speed, resolution, and interpretability. This project advances machine learning (ML) pipelines for high-fidelity defect detection and thermal mapping by fusing multi-modal datasets (e.g., thermoreflectance, infrared thermography, scanning thermal microscopy, electrical self-heating) with physics-aware feature engineering. The aim is to improve sensitivity to sub-surface anomalies and produce quantitative maps of interfacial thermal resistance, enabling earlier detection, process control, and predictive maintenance. The project will build on our existing experimental datasets and simulation “digital twin-inspired” models to benchmark new ML models and uncertainty quantification, targeting a co-authored publication.
Supervisor
ZHENG, Qiye
Quota
2
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP4100
Applicant's Roles
• Clean and harmonize datasets; implement baselines (e.g., ResNet/U-Net) and advanced models with uncertainty estimation.
• Integrate physics-informed features to improve generalization.
• Run ablation studies; quantify performance across defect types, depths, and SNR conditions; prepare publication-quality plots.
Applicant's Learning Objectives
• Gain practical experience in thermal imaging analysis, signal processing, and deep learning model development.
• Learn uncertainty quantification, cross-validation, and reproducible ML practices.
• Understand structure–property–performance links in microelectronics/composites and how ML can guide process control.
Complexity of the project
Challenging