Abstract
For semiconductor industry, there are several factors influencing the increasing cost, such as complicated Wafer process, expensive raw materials and strict production environment. To improve yield rate and decrease production cost, wafer fabrication factory puts emphasis on the method of its process control and analysis. In recent years, many semiconductor foundries have invested a large sum of capital in Advanced Process Control (APC). In the field of APC, the induction and development of Fault Detection & Classification (FDC) is definitely one of the important parts. FDC can rapidly detect abnormal situation of operation machine, so as to improve the yield rate. Based on fuzzy system and decision tree, the research tries to construct an FDC system on heating curve monitoring of soft bake in photolithography area. To key in characteristics of certain heating curve to the system, it can obtain one coefficient of determination to identify abnormal status. As the coefficient of determination reveals to be abnormal, rule base can identify what kind of abnormal situation it is. With data collection and extracting of characteristics of abnormal curve, this research adopts Classification and Regression Trees (CART) to construct its rule base. 410 process data collected from one wafer fabrication factory to test and verify the presented FDC system. Experiment results showed that 42 abnormal curves could be successfully identified by the system. Only 13 curves are incorrectly identified out of the other 368 normal curves. And these 13 curves are the augury of process become aberrance possibly. Therefore, the presented system helps semiconductor fabrication companies to detect abnormal heating curves in an efficient way. Process engineers are able to detect abnormal curves immediately with the aid of this system and therefore quickly identify corresponding situations correctly via the rule base.