模糊關聯記憶神經網路應用於半導體晶圓之FMECA決策支援系統設計

Using FAM Neural Networks as a decision support system for FMECA in Semiconductor Manufacturing Process

陳增儒(Tseng-Ju Chen)

指導教授 ： 鍾雲恭

元智大學/工程學院/工業工程與管理學系/碩士(1998年)

摘要

因應世界貿易組織的加入，產品多元化與彼此競爭是無法避免的；國內半導體產業的自動化技術仍在不斷地開發中，特別是在提昇產品品質的管理輔助系統方面，其研發的空間仍大，其中半導體之失效模式、效應與危害度分析(Failure Mode, Effects and Criticality Analysis,FMECA)的實施，也應採用自動化技術；因此，本文運用神經網路技術建構一套FMECA決策支援系統，用以辨識晶圓(wafer)電路缺陷訊號的失效模式、推論失效原因、診斷失效效應、區別失效模式之嚴重度等級和分析危害度指標。用來分類缺陷的神經網路稱為模糊關聯記憶(Fuzzy Association Memory, FAM)神經網路，利用它同時具有神經網路及模糊專家系統的特性，將之應用於FMECA系統設計內，相信對往後晶圓設計與製程環境的改進，應可預期是有效的。FAM神經網路之完成分類的工作，乃藉由過去的失效資料中，學習其中的模糊分類法則(fuzzy rules for classification)，並從現今的測物資料決定其失效的因素，其研究價值顯而易見。

關鍵字

半導體晶圓 ； 失效模式效應與危害度分析 ； 模糊關聯記憶神經網路 ； 人工智慧 ； 可靠度工程

並列摘要

It is inevitable that the various multiple products are manafactured and they highly compete one another in market that is opened to the world. To cope with this open market, an automatic or a computerized management assistance system used to increase product quality is necessary for the decrease of the product cost. This thesis, from the viewpoint of the opened semiconductor product market, presents an automatic FMECA technique using FAM neural networks in order to win the market competition. A decision support system (DSS) consisted of FAMs, which have both neural learning and fuzzy inference natures, is proposed to recognize semiconductor circuit probed by the inspection instrument used to generate normal and defect signals, and then to identify die circuit failure modes, to infer the corresponding failure causes, and to diagonize the failure effects. Another back-propagation network (BPN) discriments the criticality ranks of the failure mode identified. The proposed DSS not only learn the failure record history, but classify the failure modes based on the failure history. This functions the proposed DSS applied to FMECA.

並列關鍵字

Semiconductor Wafer ； Failure Mode ； Effects and Criticality Analysis ； Fuzzy Association Memory ； Artificial Intelligence ； Reliability Engineering

參考文獻

1.Bahrami, A. and C. H. Dagli, “From fuzzy input requirements to crisp design,” Int. J. Adv. Manuf. Technol., vol. 8, pp.52-60, 1993.

2.Chang, H. C., H. C. Chen and J. H. Fang, “Lithology determination from well logs with fuzzy associative memory neural network,” IEEE Transactions on Geoscience and Remote Sensing, vol. 35, no 3, pp.773-780, 1997.

5.Gu, Z., Y. F. Zhang and A. Y. C. Nee, “Generic from feature recognition and operation selection using connectionist modelling,” Journal of Intelligent Manufacturing. London., vol. 6, pp.263-273, 1995.

6.Hess, C.,”Strategy to optimize the development, use, and dimension of test structures to control defect appearance in backend process steps,” IEEE/SEMI Advanced Semiconductor Manufacturing Conference, pp.282-289,1994.

Klir, G. J. and Folger, T. A., Fuzzy Sets, Uncertainty, and Information, Prentice Hall Inc., 1992.