應用機器視覺於TAB表面線路瑕疵之研究

The Research about TAB Surface Defecting Inspection by Using Automatic Visual Approaches

潘旭純(Shiu-Chuen Pan)

指導教授 ： 江 行 全

元智大學/工程學院/工業工程與管理學系/碩士(2004年)

摘要

IC構裝的技術不斷演進，在晶片與基板的接合方式中，TAB製程為許多連續生產製程所採用。某廠商對於TAB表面線路瑕疵檢測是採用半自動化的檢測，目前，廠商所面臨到的問題有三：一是TAB線路表面有異物存在，增加了AOI機台檢測的誤判率；二是AOI檢測機台的內建檢測演算法之適用性低；三是人力的使用。本研究將應用機器視覺技術於廠商在進行TAB表面線路瑕疵檢測時遇到的問題做改善，最終的目的，主要是要降低黏附於TAB線路表面的異物對檢測結果所造成之誤判率；其次，則是期望把廠商現有的瑕疵檢測過程改為完全自動化檢驗。
就針對TAB線路表面上黏附異物的問題，本研究提出合併式分割法來進行改善，此方法較傳統之區域成長法使用時所需的參數定義更為簡易，在電腦配備為Pentium Ⅳ 2GHz PC、Win2000作業系統下，使用Matlab 6.5程式語言進行分割作業，平均執行時間減少了14.1097秒，只需3.1883秒即完成分割作業。而由實驗結果得知，合併式分割法的使用，大多數有異物之影像，都能做到完全去除異物的目的，讓有異物之TAB影像的檢測正確率提升83.67％，有助於廠商檢測結果之誤判率降低。
另外，在對於廠商現有AOI檢測機台的內建瑕疵檢測演算法的部分，本研究採用了求取線路邊界之共變異矩陣的做法。以這個方法做瑕疵偵測，其好處是可以不需要參考樣本做比對，且由共變異矩陣的變化找出瑕疵發生處後，還能再進一步的對瑕疵做分類，讓完整的檢測流程都由機器自動化來完成，解決廠商在人力使用上的問題。而這個完整的TAB表面線路之檢測系統，目前能檢測TAB線路上的瑕疵正確率為85.43％。所以，本研究所提出的瑕疵檢測流程與方法，是可以提供給廠商作為解決目前所面臨之三個問題的做法。

關鍵字

TAB表面線路 ； 影像分割 ； 區域成長法 ； 合併式分割法 ； 瑕疵偵測 ； 瑕疵分類 ； 共變異矩陣

並列摘要

TAB processing is a continuous produce processing between IC packaging chips and substrate in the interconnection mode. One of the TAB processing companies has semi-automatic inspection in detecting TAB surface defect. Recently, the company has three problems during TAB processing. First, there are just a few remaining photos resist, and the TAB surface circuits may have some dusts on them. So they would increase the error rate for AOI machine detection. Second, it is not flexible to build a detection algorithm for the AOI machine. Third, the production flow costs a lot of labors. In this research, I’ll use automatic visual approaches to improve those problems above. The main objective is to reduce the error rate and lead the automatic defecting detection achieve more complete.
This research has built a new merged segmentation method to solve the problems about the TAB surface circuit. The new segmentation method has merged the “Region Growing” and “Edge Detection” techniques. It is more useful to define parameters by using the new algorithm than by the traditional region growing method. The presented method spends only 3.1883 seconds in using merged segmentation method. The speed of the program running is also faster than the speed of the traditional region growing method. The result of the image test is that using merged segmentation method can make image more clearly. Besides, the conclusion shows that the error rate have been reduced, and the correct rate is raising to 83.67％.
In addition, this research presents a covariance matrix method to detect surface defect. It can increase flexibility for TAB surface defect system, and one of advantages is that pattern matching is not necessary. After finding defecting location by covariance matrix method, we can continue defecting classification. It can achieve a complete automatic defect detection process, and decrease the necessary about labors. In this paper, we can construct a trend of complete automatic defecting detection process, and it will increase the correct rate of detecting defect up to 85.43％.

並列關鍵字

TAB surface circuits ； Image Segmentation ； Region Growing Method ； Merged Segmentation Method ； Defect detection ； Defect classification ； Covariance Matrix

參考文獻

【3】 葉繼豪，「應用共變異數矩陣與小波轉換於BGA基板線路瑕疵檢測」，博士論文，元智大學工業工程與管理研究所，中壢，2001。

【6】 Bae, S. C., I. S. Kweon, and C. D. Yoo, “COP: A New Corner Detector,” Pattern Recognition Letters, Vol. 23, 2002, 1349-1360.

【7】 Chabat, F., G. Z. Yang, and D. M. Hansell, “A Corner Orientation Detector,” Image and Vision Computing, Vol. 17, 1999, 761-769.

【8】 Cao, L., Z. Shi, and E. K. W. Cheng, “Fast Automatic Multilevel Thresholding Method,” Electronics Letters, Vol. 38, 2002, 868—870.

【9】 Jing, F., M. Li, H. J. Zhang, and B. Zhang, “Unsupervised Image Segmentation Using Local Homogeneity Analysis,” Circuits and Systems, ISCAS ''03. Proceedings of the 2003 International Symposium on, Vol. 2, 2003, II-456-II-459.