隨著5G 和人工智慧的推動, 台灣半導體產業也正蓬勃發展。且近年由於晶片市場需求量的上升, 使得封裝產業更加受到注目。在進行封裝之前, 每一道製程皆需檢驗, 以確保產品的品質沒有任何瑕疵, 才能進行封裝。目前, 大部分的封裝半導體工廠每一製程主要還是依靠人工的方式進行瑕疵檢測。但在競爭激烈的環境下, 使用人工檢測的方式是缺乏效率的且人工成本高的。因此, 如何透過數位影像處理的方式進行檢測瑕疵, 取代原先以人工檢測瑕疵的方式, 進而降低人事成本, 是值得研究的議題。本研究目標是開發影像系統辨識, 辨識晶片的良品與壞品,也就是系統自動辨識取代現階段人工檢查。假設檢驗人員每日出勤8員, 需支付總月薪約新台幣28萬元, 而改成電腦影像判斷只需一個操作人員月薪約新台幣3.5萬元。
本研究提出之影像辨識系統, 採用了深度學習演算架構的卷積神經網路(Convolution Neu-ral Network, CNN), 利用實驗設計得到最佳的CNN 中所需參數, 並找出預測模型。模型預測結果之績效: 敏感度(Sensitivity) 為96 %、特異度(Specificity) 為93% 與準確度(Accuracy)為94%。因此本研究所提出之辨識分析方法具有相當的準確性, 能有效提升生產效率並減少人力成本。

With the promotion of 5G and artificial intelligence, Taiwan’s semiconductor in-dustry is booming. In recent years, due to the increase of demand in the wafer market,the packaging industry has attracted more attention. Before the packaging, each process needs to be inspected to ensure that the quality of the product is free of defects before it can be packaged. At present, most of the packaging semiconductor factories mainly rely on manual methods for defects detection. However, in a highly competitive environment,manual detection is inefficient and labor intensive. Therefore, how to detect by digital image processing, instead of manually detecting, and thus reducing personnel costs, is worthy of study. The goal of this research is to develop image recognition system and identify the good and bad products of the wafer, that is, the automatic identification of the system replaces the manual inspection at this stage. Assume that there are 8 inspec-tors per day, and the total monthly salary is about TWD 280,000, if the company utilizes the computer image recognition, there will be 1 operator with a monthly salary of TWD 35,000 only.
The image recognition system proposed in this study uses the convolutional neural network (CNN) of the deep learning framework. To obtain the optimal parameters in the CNN, we apply design of experiment (DOE) and find the prediction model. The performance of the model prediction results: sensitivity is 94 %, specificity is 93%, and accuracy is 96%. Therefore, the identification analysis method proposed in this study has considerable accuracy, which can effectively improve production efficiency and reduce labor costs.

[1] 范哲綸. (2017). 發展自動車牌辨識之機器學習系統. 中原大學機械工程研究所學位論
文, 1-45.
[2] 陳冠位. (2007). 影像處理及類神經網路於晶圓缺陷分析之應用. 元智大學資訊管理學
系學位論文, 1-75.
[3] Chang, C. Y., Li, C. H., Chang, Y. C., & Jeng, M. (2011). Wafer defect inspection
by neural analysis of region features. Journal of Intelligent Manufacturing, 22 (6),
953-964.
[4] Ko, S. J., & Lee, Y. H. (1991). Center weighted median filters and their applications
to image enhancement. IEEE transactions on circuits and systems,38 (9), 984-993.
[5] Peng, X., Sun, B., Ali, K., & Saenko, K. (2015). Learning deep object detectors
from 3d models. In Proceedings of the IEEE International Conference on Computer
Vision (pp. 1278-1286).
[6] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning
applied to document recognition. Proceedings of the IEEE, 86 (11), 2278-2324.
[7] I.E. Sobel, Camera Models and Machine Perception, Ph. D. dissertation, Stanford
University, Palo Alto, Calif., 1970
[8] Fu, K. S., & Mui, J. K. (1981). A survey on image segmentation. Pattern recognition,
13 (1), 3-16.
[9] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning
applied to document recognition. Proceedings of the IEEE,86 (11), 2278-2324.
[10] Lawrence, S., Giles, C. L., Tsoi, A. C., & Back, A. D. (1997). Face recogni-
tion: A convolutional neural-network approach. IEEE transactions on neural networks,
8 (1), 98-113.
[11] Bahlmann, C., Zhu, Y., Ramesh, V., Pellkofer, M., & Koehler, T. (2005, June). A
system for traffic sign detection, tracking, and recognition using color, shape, and
motion information. In IEEE Proceedings. Intelligent Vehicles Symposium, 2005.
(pp. 255-260).IEEE.
[12] Haralick, R. M., Sternberg, S. R., & Zhuang, X. (1987). Image analysis using
mathematical morphology. IEEE transactions on pattern analysis and machine
intelligence, (4), 532-550.