AI-based image segmentation for the characterization of bond pads after

Tuesday, November 18, 2025: 12:50 PM
1 (Pasadena Convention Center)
Mr. Dirk Utess , GlobalFoundries, 01109 Dresden, Saxony, Germany
Dr. Martin Weisheit , GlobalFoundries, 01109 Dresden, Saxony, Germany
Dr. Heiko Stegmann , Carl Zeiss Microscopy GmbH, Oberkochen, Germany

Summary:

The Paper discusses the use of AI-based image segmentation for characterizing bond pads after a Shelf-Life Accelerated Test (SLAT) at Globalfoundries. SLAT simulates long-term storage of wafers under elevated temperature and humidity conditions to ensure product quality. Manual inspection of bond pads is time-consuming and requires expert knowledge, limiting the number of analyzed images and affecting the quality of statistics. To improve efficiency, AI-driven segmentation is explored to replace human inspection and automate the classification of aluminum pads. Various strategies for annotating training images are evaluated to achieve the best segmentation results. The training algorithm assesses segmentation quality using loss, accuracy, and Intersection over Union (IoU) metrics. Initial training shows that only two strategies effectively detect discolorations, and further training is needed to enhance accuracy.