在半導體封裝生產製造系統之中，由於製程設備和產品種類日新月異，如何掌控製程參數的變異和維持製程設備的有效一直是很重要的課題。製造過程中，人員需熟悉操作方法，了解維修保養和故障排除；同時，若要提高產品的品質及降低不良率，除了最佳的印字製程參數設定外，就是讓設備保持在最佳的運轉狀況。因此需使用適當的集群分析方法來將龐大的失效資料進行萃取與歸納，以及從原有基礎的失效模式知識，透過適當的統計分類方法，進行失效狀態的關聯與趨勢分析。
本文針對半導體封裝製程中所產生之失效模式知識，探討知識分群特性以及維度變數的決定對於群集分類的影響。對於群集的資料分佈意義和參數對群集的意義與各群集間的關係，做更進一步的分析與描述，並選取重要之群集分析方法將失效模式進行萃取分析，如切割式分群法、階層式分群法、自組織映射網路分群法等將失效模式進行萃取分析，以歸納出有效的群集分析準則，作為失效趨勢的分類取樣之重要依據。協助工程師對設備與產品上進行問題排除，針對不周延之處加以改進。本文在進行實務專家系統建構時奠定相當的基礎，在群集分類的結果可作為失效模式與效能分析分類之基礎，使整體異常事件處置時間能大幅縮短並增進產線效能。
最後本研究於成果方面，對於失效事件經由切割式分群法與關聯法則的分析，將危險程度高的失效事件抽離出來，並從各群集分析方法的特性進行探討，可作為工程人員改善異常與提升生產效能之基礎。

In the manufacture system of semiconductor assembly, because the manufacture equipment and production type are rapid changed all the time, how to control the variety of process parameters and effectiveness of process equipment is important issue for discussing. In the manufacture process, people need to familiar with equipment, such as operated method, maintenance and troubleshooting. At the same time, in order to improve quality and reduce the yield rate, it not only needs to set up the best parameters in the making process but also has to keep the machine in the best running status. Based on this critical issue, it needs to use the appropriate clustering analysis method to extract and compress the huge failure data. Moreover, exerting the statistic analyzes the connection and trend of failure state in the original failure model knowledge.
The research is focus on the failure model knowledge in the semiconductor assembly process. It describes how the factors of the attributes of knowledge clustering and the selecting dimension variables affect the clustering classification. Moreover, it explains the meaning of clustering data distribution and the effect between clustering and parameters. Furthermore, it generalizes the effective principle of clustering analysis by using the method of cluster analysis, which are Partition Clustering, Hierarchical Clustering and Self Organization Map, in the failure model. This principle of clustering analysis is the important basis for classifying the sample in the failure trend. The purpose of this method is to help engineers solve the problems effectively and improve the function for the abnormal status of equipment or production. The research is established in the basis of reality expert system. Thus, the result of clustering categorization could be the foundation of Failure Mode and Effects Analysis classification. Expect that can shorten the solving time when the abnormal affair happens, deeply, and improve the capacity in production line.
Final achievement, failure event analyze by Partition Clustering and associate rule, pick the high danger failure event, proceed to probe for the characteristic of every clustering analysis method, be the basis of improve deviation and promote production efficacy for engineering personnel.

參考文獻
丁一賢、陳牧言，民94，資料探勘，滄海書局，臺中市。
小野寺勝重，民86，FMEA手法概要，日本科學技術連盟。
李培端，2002，半導體製程資料特徵萃取與資料挖礦應用，科技管理學刊，第七卷，第一期，137-160。
林文俊，民93，以資料探勘及線上分析處理技術支援醫療管理決策，國立屏東科技大學高階經營管理研究所，碩士論文。
林志杰，民95，應用資料探勘技術建立中西醫腦中風診斷模型之研究，長庚大學資訊管理研究所，碩士論文。
陳順宇、鄭碧娥，民93，統計學，華泰書局，臺北市。
葉怡成，民90，類神經網路模式應用與實作，儒林圖書，臺北市。
鍾秋英，民92，資料挖掘應用於產品失效模式與效應分析以印刷電路板業為例，私立元智大學工業工程與管理研究所，碩士論文。
鍾文仁、陳佑任，民92，IC封裝製程與CAE應用，全華科技圖書股份有限公司，臺北市。

Arunajadai, S.G.., Uder, S.J., Stone, R.B., and Tumer, I.Y., 2004, Failure Mode Identification through Clustering Analysis, QUAL. Reliab. Engng. Int., 20, 511-526
Feldmann, K., and GÖhringer, J., 2001, Internet-Based Diagnosis of Assembly Systems, Annals of the CIRP, 50, 5-8.
Ganti, V., Ramarkrishnan, R., Gehrke, J., Powell, A., and French, J., 1999, Cluster Large Datasets In Arbitrary Metric Spaces, Proceedings of the 15th International Conference on Data Engineering, Sydney, Australia, 502-511.
Guha,S., Rastogi, R., and Shim, K., 1998, CURE: An Efficient Clustering Algorithm For Large Databases, Proceedings of CM SIGMOD International Conference on Management of Data, New York, 73-84.
Guha,S., Rastogi, R., and Shim, K., 1999, ROCK: A Robust Clustering Algorithm ForCategorical Attributes, Proceedings of the 15th International Conference on Data Engineering.
Han, J. and Kamber, M., 2001, Data Mining：Concepts and Techniques, Morgan Kaufmann Publishers, San Francisco.
Huang, G..Q., Shi, J., and Mak, K.L., 2000, Failure Mode and Effect Analysis （FMEA） Over the WWW, Int J Adv Manuf Technol, 16, 603-608.
Kaufman, L. and Rousseeuw, P.J., 1990, Finding Groups in Data: An Introduction to Cluster Analysis.
Kohonen, T., 1981, Self- Organization and Associative Memory, 2nd edition, Springer-Verlag, New York .
Kusiak, A. and Kurasek, C., 2001, Data mining of printed-circuit board defects , IEEE Transactions on Robotics and Automation and Cybernetics, 17（2）, 191-196.
McQueen, J.B., 1967, Some Methods of Classification and Analysis of Multivariate Observations, Proceedings of the 5th Berkeley Symposium on MathematicalStatistics and Probability, 281-297.