A Generic Object-Oriented System for Wafer-Level FDC

摘要

The purpose of fault detection and classification (FDC) is to improve processrnoperations by quickly and accurately determining when process abnormalities havernoccurred and identifying the cause of the process abnormalities. Increased yield andrnimproved product quality result from reducing the number of wafers produced underrnsub-optimal operating conditions. An additional benefit of FDC is in the time andrnmoney saved troubleshooting, thereby decreasing product cycle times and ensuringrnengineering resource allocation to process/equipment improvement as opposed torneradicating faults. This work presents a comprehensive object-oriented FDCrnapplication that combines fault detection using multi-block principal componentrnanalysis (PCA) with automated fault classification using rule-based predicate logic.rnThe prescribed system, using PCA and arbitrary logic rules, has inherent flexibility tornwork equally well on equipment as well as wafer state data for FDC.rnAs the volume of wafers produced and the amount of available online data hasrnincreased, monitoring individual online parameters has become infeasible andrnmultivariate process monitoring has emerged as the method of choice. The philosophyrnbehind multivariate process monitoring is that process data is inherently correlated. Ifrnthis correlation structure is disrupted then it can be assumed that a fault hasrnoccurred. It is this disruption in the correlation structure that yields an overallrnperformance metric as well a contribution vectors which can be remembered inrnvarious patterns to identify classified faults by the predicate logic.