Skip to main content

Wafermap Defect Pattern Recognition

The manufacturing yield of semiconductor under silicon technology depends heavily on wafer processing. The wafer map may not only reveal a specific defect pattern for engineering trouble shooting but also provide an opportunity for process control, test optimization, quality improvement.  However, reviewing all the wafer maps could be a tedious task, if done manually. This project is to derive an automated and efficient way to classify the wafer map into signature patterns. In general, the team developed a machine learning model with an accuracy of ~95.07% for predicting 5 different defect patterns (Edge, Random, Hotspot, Scratch, Stripe). Users will be able to use a graphical user interface to convert .csv wafermap data to .png images or classify existing .png wafermap images.

Team Members: 

Jason Fang

Sunguk Hong

Nabil Khan

Zhaofeng Liang

Darshan Poudel

Texas Instruments