Classification of defect spatial signatures using independent component analysis and estimation of process/tool malfunctions using χ² test and exact test

Abstract

SUMMARYWe developed a system that detects spatial signatures from the defect inspection data of individual substrates and thus performs fault detection in device manufacturing. Leveraging of independent component analysis facilitates the unsupervised simultaneous classification of any defect distribution generated as a result of one or more tool malfunctions. All substrates are classified according to our proposed coefficient of similarity for each defect distribution. A root cause process is identified through a test of independence between the manufacturing tools and their rates of the number of classified substrates on the basis of the classification results and their fabrication history data. The tests of independence use χ 2 tests in combination with exact tests to decrease the incidence of false-positive errors. The root cause tool is identified in terms of the highest rate between the tools in the identified process. Our system functions automatically and requires no experience or technical skill. We present a case in which for approximately 2 days, our system detected a tool malfunction earlier than the conventional monitoring of substrates, and with greater total defect counts per substrate than a control limit. We present another case in which our system detected a greater number of substrates than conventional monitoring.

A Systematic Literature Review of Unsupervised Fault Detection Approach for Complex Engineering System

“…The result showed a strong modeling capability for FD, but the usage was limited with fewer data points and low dimension. Yamada et al, [30] considered a huge amount of data in their study. The study proposed a combination of information from exact tests and statistical chi-test to determine the faulty condition.…”