在半導體製造上需要對製程機台進行即時資料之趨勢監控外，需要更多能精確判斷製程與機台狀態的新製程偵測技術與方法，特別是在先進的12吋晶圓廠上。在半導體製程技術持續的微縮下，製程的飄移或缺陷通常不是單一製程機台參數所造成的，更可能是由多個參數所造成的綜合性影響。因此使用傳統的單一變數分析去找尋問題根由是困難的；再者單一變數的即時規格管制通常無法偵測出製程機台的異常行為。本文旨在探索如何利用製程機台群組化之參數資料進行多變量分析，然後利用解析結果建構出監控模型，以偵測半導體晶圓在量產過程之製程缺陷問題，並整合此監控模型與機台即時缺陷偵測分類系統(Fault detection and classification system, FDC)，以提供晶圓廠即時製程缺陷監控偵測。
本論文的實例研究驗證是以Ti/TiN金屬膜剝落缺陷為例，以探討此利用多變量分析手法所建構之即時監控模型的效益。Ti/TiN金屬膜剝落缺陷是由於膜層與晶片基層的應力不協調所造成的。經由缺陷檢測發現，主要發生在SiN鍍膜製程之後，其晶片檢測圖之分佈區域在於晶片邊界約有50~400的單元(die)損害且造成5~40﹪的良率損失。為了有效解決此一問題，本文利用FDC系統建構出效率化之監控模型，以找出正常與不正常晶圓批(lots)之間的差異。
本文中透過Ti/TiN金屬膜剝落缺陷的實例研究，闡明如何將缺陷分類，如何挑選模型構建之解釋變數，如何應用此模型預測診斷半導體製程缺陷，並進一步延伸FDC系統之應用開發。

The rapid innovation of new process technologies in the semiconductor industry, especially 12 inches Fab, along with a tendency of monitoring real time process equipment data has led to the need to more accurately determine process and equipment health. To go along with shrinking of semiconductor process window, a process shift (or issue) is not usually caused by a single parameter of process equipment. It is may be caused by the synthetic effect of more than one parameter of process equipment. So, it is difficult to find root cause of a process issue by traditional univariate analysis method. And, again, univariate real time SPC generally can’t detect abnormal behavior of process equipment. This article discusses about how to cluster parameters of process equipment together for a specific process issue by statistical multivariate analysis methods, then creates a monitoring model to detect the issue of semiconductor manufacture process according to these clusters, and finally integrates the model with the FDC system in a Fab, in order to detect process issues in real time. A case study of Ti/TiN peeling is used to demonstrate these techniques and the resulting benefits; this case study builds on previously reported work on univariate analysis of this process.
Ti/TiN film peeling occurs due to stress caused by a mismatch between the film and the substrate and it has been detected near the wafer edge after SiN deposition. There are about 50~400 dies per wafer suffer from this issue and the resulting yield loss is about 5~40%. These abnormal lots are often processed after self-clean (dry or wet clean) of machine. In order to solve this issue, an FDC system with efficient models is used to find the difference between normal lots and abnormal ones. The approach is to identify the major contributors and use data from identified good and bad wafers to develop models that predict the wafer quality (with respect to peeling) from measured FDC parameters.
This thesis illustrates, through the example of the Ti/TiN analysis, the methodology by which the faults are categorized, parameters are chosen, models are developed and utilized to predict faults. This methodology could be extended in a generic fashion to other FDC applications in the fab as well.

中文部份
1.陳順宇，「多變量分析」，第2-1~2-64頁，台北，華泰書局，二版，2000年7月。
2.陳耀茂，「多變量解析方法與應用」，第89~153頁，台北，五南圖書出版公司，民國90年7月。
英文部份
1.Cherry, G., Good, R., Qin, S. J., 〝Semiconductor process monitoring and fault detection using recursive multi-way PCA," TWMCC (2004).
2.Joyce-Wohrmann, R., 〝A pragmatic approach to managing APC /FDC in high volume logic production, 〞 Advanced Semiconductor Manufacturing Conference and Workshop, 2003 IEEEI/SEMI, pp: 303(2003).
3.Kalin, M., 〝Benefits of APC at Infineon application examples from production fab Villach, 〞 Advanced Semiconductor Manufacturing, 2004. ASMC '04. IEEE Conference and Workshop, pp.437 – 442(2004).
4.Latin, J., Carroll, D., Green, P. E., 〝Analyzing Multivariate Data, 〞 2003, Duxbruy.
5.Lee, J., Baek, H., Song, Y., Lee, P. S., Gasser, D., Price, D.W., 〝A study on rapid development and ramp-up of high density DRAM using an advanced e-beam inspection system, 〞Advanced Semiconductor Manufacturing, 2004. ASMC '04. IEEE Conference and Workshop, pp.227 – 231(2004).
6.Lo, J., Turk, H., 〝 Defect inspection on rare-earth garnet surfaces, 〞 Magnetics, IEEE Transactions on, Volume: 18, Issue: 6, pp.1304 – 1306(1982).
7.Montgomery, D. C., Peck, E. A., Introduction to Linear Regression Analysis, John Wiley & Sons, NEW York, pp290~297 (1992).
8.Rencher, A. C., 〝Multivariate Statisical Inference and Applications, 〞1998, John Wily & Sons.
9.Scanlan, B., 〝Defect inspection sampling plans-which one is right for me? , 〞Advanced Semiconductor Manufacturing Conference and Workshop, 1998IEEE/SEMI, pp. 103 – 108(1998).
10.Williams, R., Gudmundsson, D., Monahan, K., Shanthikumar, J.G., 〝Optimized sample planning for wafer defect inspection, 〞Semiconductor Manufacturing Conference Proceedings, 1999 IEEE International Symposium on, pp.43 – 46(1999).
11.Wu, H.C., Chang, C.T., Chen, B.H., Lee, C.Y., Chang, C.J., Ko, J.C., Zhou, M.S., Liang, M.S.,〝Fault detection and classification of plasma CVD tool,〞Semiconductor Manufacturing, 2003 IEEE International Symposium on ,pp.123 – 125(2003).
12.Yue, H.H., Qin, S.J., Markle, R.J., Nauert, C., Gatto, M., 〝Fault detection of plasma etchers using optical emission spectra, 〞Semiconductor Manufacturing, IEEE Transactions on 13(3),pp. 374-385(2000).
13.Zimpel, J., Voightländer, K., Steinbach, A., 〝Using multi way PCA (MPAC) for advanced monitoring and diagnosis for Plasma processing based on optical emission spectroscopy, 〞 SEMATECH AEC/APC Symposium XII, (2000).