Developing Computational Algorithms to Predict the Quality of High Voltage GaN Diodes Using Data Science and Machine Learning-FY21 Naval Innovative Science and Engineering (NISE) Final Report
Abstract
A major roadblock in developing GaN power electronic devices is the reliable manufacturing of substrates and epitaxial layers, thus there is a need to develop quick, non-destructive techniques for predicting the quality of devices fabricated on wafers. This research project focused on collecting large data sets on vertical PiN diodes to developed computational algorithms for doing this. Specifically, Raman spectroscopy was used to detect high crystal stress points associated with higher leakage currents, and machine learning was applied to optical profilometry images to predict the quality of vertical PiN diodes. It was found that high crystal stress points detected by Raman doubled that chance of a high leakage failure, and machine learning was 91 percent effective at predicting the forward bias conductions using optical profilometry. This results from the research will help future GaN projects screen wafers and epi, saving the costly fabrication process of devices.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2021
- Accession Number
- AD1154852
Entities
People
- Alan G Jacobs
- James C. Gallagher
- Michael A. Mastro
- Travis J. Anderson
Organizations
- United States Naval Research Laboratory