Intelligent process diagnosis based on end-of-line electrical test data

Abstract

The goal of this research is to develop a fuzzy logic-based system for a first-cut end-of-line diagnosis function. Based on measured abnormal electrical test data, the system provides the engineers a list of prioritized causes (process steps) for further investigation. The intelligent diagnosis system consists of three major modules: fuzzy modeling, knowledge base and inference engine. Experienced engineers diagnosis knowledge is captured in the knowledge base using fuzzy logic knowledge representation models. Each major processing step's fault possibility is calculated in the inference engine. The intelligent diagnosis system has been validated against 23 real fab cases. Results show that version 2.0 of the system identifies the real causes as the top three causes in 20 cases. Our analysis indicates that the inference engine is robust but the knowledge base is insufficient. Improvement strategy has been to periodically update the knowledge base by field engineers based on lessons learned from the case study.