One-Shot Defect Fingerprint Comparability Using Siamese Networks for Wafer Map Similarity in Semiconductor Manufacturing

In the field of semiconductor manufacturing, the complex nature of the fabrication processes poses a significant challenge in maintaining a steady and high-quality yield. Microchips are fabricated on a silicon circular tray, known as a wafer, which can show specific defect patterns, based on the patterns of failures of groups of chips. Traditional classification algorithms are effective to classify distinct visual wafer patterns. However, in manufacturing, patterns are rarely fully distinct but represent composition of patterns which are significant for composing complex scenarios. Decomposing such patterns into feature vectors, can be utilized to identify and distinguish defect root causes. To address this challenge, the focus of this research is the development of a similarity-based classification, utilizing Siamese Networks and a comparison methodology, aiming to decompose patterns into composing core patterns as a defect fingerprint, their degree of intensity and, finally, enabling to compare current to past patterns for which defect root causes are already known.