Abstract
In semiconductor manufacturing, due to the reason that the manufactured products usually require very long process time, the final output of the quality characteristics cannot be collected immediately. Hence, the traditional online statistical process control (SPC) techniques cannot be implemented directly. Under this situation, how to use the profile data of manufacturing equipment to perform fault detection and classification (FDC), is an important research topic. Recently Lee, et al. (2011) introduced a simple FDC procedure to analyze the profile data, which consists of the target value, level-shift parameter and residuals. The level-shift parameters indicate the amount of drift between batches. A health indicator (health index) is also constructed to address the healthy condition of the manufacturing of equipment. However, this method does not take the autocorrelation of data points into consideration. In addition, the level-shift parameter is not capable of describing the process on-off effect effectively. To overcome those difficulties, this thesis proposed a seasonal time series model together with intervention analysis to analyze the profile data. The proposed method can eliminate the autocorrelation between the data points. Furthermore, this article also constructs three different types of health index to address the wafer-to-wafer, batch-to-batch variations and the residuals performance. It provides engineers a useful way to identify the potential faulty of the manufacturing equipment.