臺灣博碩士論文加值系統

在快速進步的社會中，科技產業對半導體產品品質的要求越來越高，然而在產線現場還是有許多的檢測是以人工目檢的方式進行。晶圓代工及封測業者為讓晶片在不影響空間占位的前提下，導入高階覆晶封裝製程讓銅柱(Pillar)及晶圓錫球在更小的晶圓開孔中接合，以克服薄晶圓可靠度、應力和低介電材料損壞等問題。在共金黏晶的製程中，使用X-ray檢出未銲或銲點不良則是一件重要的工作。為解決此問題，本研究藉由影像前處理增強影像特徵，接續使用深度學習來進行X-ray樣本的檢測以取代傳統的人工判讀。有關影像前處理，2.5D與2D影像以均值濾波器搭配本實驗室研發的自適應能量分析來增強銲點特徵，少數2D影像則使用色彩映射來凸顯不規則的銲點特徵，最後藉由程式整合呼叫深度學習演算法進行自動判圖，並回傳檢測結果。實驗結果顯示在120張樣本中檢測率高達90%以上，檢測一張影像平均所花費的時間少於5秒，實驗證明本研究可有效提升半導體製程的良率。

In a rapidly advancing society, the technology industry has higher and higher requirements for the quality of semiconductor products. However, there are still many inspections on the production line by humans. Foundry and testing companies have introduced high-end flip-chip packaging processes to allow pillar and wafer solder balls to be joined in smaller wafer openings without affecting the space occupied by the chip. In the process of co-gold bonding, it is an important task to use X-ray to detect empty soldering or defective solder joints. To solve this problem, this research uses image pre-processing to enhance image features and then adopts deep learning to detect X-ray samples to replace traditional manual interpretation. Regarding image pre-processing, 2.5D and 2D images use an average filter combined with the adaptive energy analysis developed by our laboratory to enhance solder joint features while a few 2D images use color mapping to highlight irregular solder joint features. Finally, a program is used to integrate the deep learning algorithm to automatically detect the image and return the result. The experimental results show that the detection rate is as high as 90% in 120 samples, and the average time to detect an image is less than 5 seconds. The experiment proves that this research can effectively improve the yield of the semiconductor process.

中文摘要
Abstract
致謝
目錄
圖目錄
表目錄
第1章 緒論
1.1 簡介
1.2 研究背景與動機
1.3 相關文獻回顧
1.4 覆晶產品X-ray影像自動判圖機制與系統架構
1.5 章節概要
第2章 影像前處理應用於銲點特徵增強
2.1 簡介
2.2 影像處理對背景問題之分析
2.3 自適應能量分析用於影像特徵增強
2.4 色彩映射應用於特定不規則特徵增強
第3章 深度學習檢測與自動判圖整合設計
3.1 簡介
3.2 YOLOv3訓練
3.3 檢測程式
第4章 實驗結果與討論
4.1 簡介
4.2 實驗討論
4.3 系統規格
4.4 實驗結果與討論
4.4.1 YOLOv3訓練與人工分類方法驗證
4.4.2 特徵影像與原始影像經過YOLO檢測的比較
4.4.3 程式檢測結果
4.4.4 檢測時間與檢測效率
第5章 結論與未來工作
參考文獻

[1] E. Reinhard, M. Adhikhmin, B. Gooch, and P. Shirley, "Color transfer between images," in IEEE Computer Graphics and Applications, vol. 21, no. 5, pp. 34-41, July-Aug. 2001.
[2] 吳家華，應用自適應能量分析於光學檢測，國立高雄科技大學，光電與通訊工程研究所，碩士論文，2018。
[3] 陳鵬仁，影像前處理應用於深度學習之研究，國立高雄科技大學，光電工程研究所，碩士論文，2020。
[4] P. Shanmugavadivu and K. Balasubramanian, “Thresholded and optimized histogram equalization for contrast enhancement of images,” Computers and Electrical Engineering, vol. 40, no. 3, pp. 757-768, 2014.
[5] C. Huang, J. Hong, and E. Huang, "Developing a machine vision inspection system for electronics failure analysis," IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 9, no. 9, pp. 1912-1925, Sep. 2019.
[6] A. Kamalakannan and G. Rajamanickam, "Spatial smoothing based segmentation method for internal defect detection in X-ray images of casting components," 2017 Trends in Industrial Measurement and Automation (TIMA), pp. 1-6, 2017.
[7] H. Shi, J. Shao, D. Du, B. Chang, and H. Cao, "Noise reduction of the real-time X-ray image based on modified adaptive local noise reduction filter," 2011 4th International Congress on Image and Signal Processing, pp. 1945-1949, 2011.
[8] S. Nuanprasert, S. Baba, and T. Suzuki, "An efficient method of occluded solder ball segmentation for automated BGA void defect inspection using X-ray images," IECON 2015 - 41st Annual Conference of the IEEE Industrial Electronics Society, pp. 003308-003313, 2015.
[9] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: unified, real-time object detection," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 779-788, 2016.