Internship assignment: Training CNNS for PQFN package delamination classification and segmentation
Problem statement
Power Quad Flat No-leads (or non-leaded) – PQFN packages are widely-used surface mount non-hermetically sealed package types in electronic devices. They are known for compact size and efficient thermal dissipation. While the reliability challenges of PQFN packages are well documented, concerns arise on effective methods to identify packaging defects.
Confocal Scanning Acoustic Microscopy (CSAM) is a powerful, non-destructive imaging technique for identifying package defects. Despite the benefits of acoustic imaging in electronic packaging, the manual interpretation of failure modes from acoustic images is subjective – particularly in cases where there is an interplay of multiple failure mechanisms.
Research question
How can CNN-based ML models be used for defect detection, classification and segmentation?
Objectives
- Make use of currently available data on failure modes and use image augmentation for data labelling of each failure mode
- Evaluate different ML based models (ResNet, U-Net, etc.) performance
Internship and MSc thesis details
- Location: Nijmegen, the Netherlands
- Collaborators: CITC and Nexperia
- Duration: 6 – 12 months
- Requirements: current MSc students in Electrical Engineering
- Supervisors: Henry Antony Martin (PhD candidate) and Edsger Smits (program manager CITC)
- Internship stipend: to be discussed
Contact us
If you would like to learn more about CITC, this internship assignment or what we can offer you, please contact us.