半導體是現在工業社會中重要的產業之一。舉凡隨手可用的手機、電腦、電視到車用電子，幾乎所有人日常周遭都會使用半導體晶圓製造的產品。摩爾定律之下晶圓越做越小，以目前台灣的半導體生產產業中，以達到精密的3奈米製程技術以下為原則。台灣的半導體生產產業中，以達到精密的3奈米以下。而當半導體中晶圓越做越小的此時，晶圓缺陷更是影響電性與良率的一項重要指標。

本研究利用了深度學習(Deep Learning)當中的二維卷積神經網路(2 Dimension Convolutional Neural Networks,2D CNN)模型進行半導體晶圓缺陷的分類，並進行不同的增生方式將晶圓缺陷分類的技術與深度學習的2維卷積神經網路結合。從水平平移、垂直平移、翻轉、明暗到高斯分布比較了各種圖像增生( Image augmentation)的技術對於半導體晶圓缺陷分類的模型準確度進行探討，目的是為了找到對模型準確度影響最佳的圖像增生方式並以此為基礎提高利用二微卷積神經模型之下半導體晶圓缺陷的分類準確性，預期將可有效的減低晶圓缺陷分類的錯誤以幫助工程師能更快速的判讀晶圓缺陷的來源並找出問題，以減低良率下降的風險。
此外本研究也比較了訓練集與測試集比例對於測試準確度造成的影響。實驗結果顯示，訓練集與測試集的比例儘管會影響到模型預測的準確度但並不顯著，最主要影響模型預測的準確度在於資料的平衡性與圖像增生的方式。在經由本研究的探討之後可以發現各種圖像增生方式當中以結合刪除少量樣本的水平平移增生可以達到70%左右的準確度，而高斯增生的表現能最佳，能使神經網路的模型預測準確度高達99%。

Semiconductors play a vital role in today's industrial society. From smartphones to cars, we rely on semiconductor-based products. With the constant shrinking of wafers under Moore's Law, the quality of semiconductor fabrication becomes increasingly important. In this study, we used Deep Learning's 2D Convolutional Neural Networks to classify semiconductor wafer defects. We explored different image augmentation techniques, including translation, flipping, brightness adjustment, and Gaussian distribution. By integrating these techniques, we aimed to improve the accuracy of defect classification and help engineers quickly identify the source of issues, reducing the risk of yield loss. We also investigated the impact of training set and test set ratios on accuracy. While the ratio affected the model's performance, the key factors were data balance and image augmentation. Horizontal translation combined with sample removal achieved around 70% accuracy, while Gaussian augmentation outperformed others, reaching 99% accuracy. Semiconductors are crucial in modern industries. We used 2D CNNs to classify wafer defects, combining various image augmentation techniques. This enhanced defect classification accuracy and aided engineers in identifying issues more efficiently. Gaussian augmentation yielded the highest accuracy at 99%.

[1] McCluskey, Matthew D., and Anderson Janotti. "Defects in semiconductors." Journal of Applied Physics 127.19 (2020): 190401.
[2] Kyeong, Kiryong, and Heeyoung Kim. "Classification of mixed-type defect patterns in wafer bin maps using convolutional neural networks." IEEE Transactions on Semiconductor Manufacturing 31.3 (2018): 395-402.S
[3] AMD,”介紹半導體” [Online].Available: https://www.amd.com/zh-hant/technologies/introduction-to-semiconductors
[4] Kasap,S.O. ”Electronic materials and Devices” McGraw-Hill New York. (2006)
[5] Chien, Jong-Chih, Ming-Tao Wu, and Jiann-Der Lee. "Inspection and classification of semiconductor wafer surface defects using CNN deep learning networks." Applied Sciences 10.15 (2020): 5340.
[6] Chauhan, Krishan Kumar, et al. "Semiconductor wafer defect classification using convolution neural network: a binary case." IOP Conference Series: Materials Science and Engineering. Vol. 1225. No. 1. IOP Publishing, 2022.
[7] Zhu, Jinlong, et al. "Optical wafer defect inspection at the 10 nm technology node and beyond." International Journal of Extreme Manufacturing 4.3 (2022): 032001.
[8]SEMI Taiwan, “半導體是什麼?晶片產業一次看懂” (2022)
[9] 張瑞棋”半導體的誕生-什麼是半導體”,(2021)
[10] 經濟部工業局智慧電子產業計畫推動辦公室, ”半導體產業現況” " [Online].Available:https://www.sipo.org.tw/industry-overview/industry-state-quo/semiconductor-industry-state-quo.html
[11] 維基百科, "高斯雜訊" [Online]. Available: https://zh.wikipedia.org/zh-tw/高斯噪聲
[12] 維基百科, "半導體" [Online]. Available: https://zh.wikipedia.org/zh-tw/%E5%8D%8A%E5%AF%BC%E4%BD%93
[13] Victor Dey ,Mystery Vault “When to Use One-Hot Encoding in Deep Learning?”(2021)
[14] Medium, ”零基礎自學深度學習：(一)神經網路基本架構” [Online]. Available: https://evan-hsiao.medium.com/從coursera學習深度學習-bd6bad6f5e14
[15] 懶尼陳的技術手札, “AutoEncoder 自動編碼器來建構生成模型” [Online]. Available: https://lennychen.com/ autoencoder-自動編碼器來建構生成模型/
[16] Statology,”Label Encoding vs. One Hot Encoding: What’s the Difference?“ [Online]. Available: https://www.statology.org/label-encoding-vs-one-hot-encoding/
[17]工業技術與資訊月刊,369期 “2023台灣半導體產值上看5兆” [Online]. Available: https://news.cnyes.com/news/id/5056099
[18] Kaempf, Ulrich. "The binomial test: A simple tool to identify process problems." IEEE Transactions on semiconductor manufacturing 8.2 (1995): 160-166.
[19] Mikołajczyk, Agnieszka, and Michał Grochowski. "Data augmentation for improving deep learning in image classification problem." 2018 international interdisciplinary PhD workshop (IIPhDW). IEEE, 2018.
[20] 科技新報,”材料分析如何協助新進製程設備改善缺陷” [Online]. Available: https://technews.tw/2022/06/07/ist-materials-analyze/
[21] Goodfellow, Ian, et al. "Generative adversarial networks." Communications of the ACM 63.11 (2020): 139-144.
[22] Medium,Cinnamon AI Taiwan, “CNN入門-圖像增強” [Online]. Available: https://cinnamonaitaiwan.medium.com/cnn入門-圖像增強-fa654d36dafc
[23] Kaggle,ASHADULLAH SHAWON, ”Wafer Defect Classification by Deep Learning”[Online]. Available: https://www.kaggle.com/code/shawon10/wafer-defect-classification-by-deep-learning
[24] Wu, Ming-Ju, Jyh-Shing R. Jang, and Jui-Long Chen. "Wafer map failure pattern recognition and similarity ranking for large-scale data sets." IEEE Transactions on Semiconductor Manufacturing 28.1 (2014): 1-12.
[25] Fan, Mengying, Qin Wang, and Ben van der Waal. "Wafer defect patterns recognition based on OPTICS and multi-label classification." 2016 IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC). IEEE, 2016.
[26] LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998): 2278-2324.
[27]Jason Brownlee ”A Gentle Introduction to k-fold Cross-Validation”(2020) [Online]. Available: https://machinelearningmastery.com/k-fold-cross-validation/
[28] Fukushima, Kunihiko. "Neocognitron: A hierarchical neural network capable of visual pattern recognition." Neural networks 1.2 (1988): 119-130.
[29] GIGABYTE, “AI人工智慧即將取代人類?五分鐘帶你讀懂深度學習產業現況與應用案例“(2019) [Online]. Available: https://www.gigabyte.com/tw/Article/can-ai-replace-humans-5-minutes-to-understand-dnn-industry-and-applications-part-i
[30] LeCun, Yann. "Generalization and network design strategies." Connectionism in perspective 19.143-155 (1989): 18.
[31]黃啟賢. ”生成對抗網路原理” (2022)
[32] Brownlee, Jason. Statistical methods for machine learning: Discover how to transform data into knowledge with Python. Machine Learning Mastery, 2018.
[33]Marco Chiappetta, “Chips Designed By AI Are The Future Of Semiconductor Evolution Beyond Moore's Law”(2021)
[34] Imoto, K., Nakai, T., Ike, T., Haruki, K., & Sato, Y. (2019). A CNN-based transfer learning method for defect classification in semiconductor manufacturing. IEEE Transactions on Semiconductor Manufacturing, 32(4), 455-459.
[35] Nakazawa, T., & Kulkarni, D. V. (2018). Wafer map defect pattern classification and image retrieval using convolutional neural network. IEEE Transactions on Semiconductor Manufacturing, 31(2), 309-314.
[36] O’Leary, J., Sawlani, K., & Mesbah, A. (2020). Deep learning for classification of the chemical composition of particle defects on semiconductor wafers. IEEE Transactions on Semiconductor Manufacturing, 33(1), 72-85.
[37] Lee, K. B., Cheon, S., & Kim, C. O. (2017). A convolutional neural network for fault classification and diagnosis in semiconductor manufacturing processes. IEEE Transactions on Semiconductor Manufacturing, 30(2), 135-142.