隨著半導體製程微縮技術的提升，半導體製造廠對整體產品品質的要求亦隨之升高，為同時兼顧高品質及高產能情況下，透過機台感測器蒐集製程數據，並依其進行故障檢測及快速改正是常見的策略之一。本研究提出一種錯誤偵測與分類(FDC)架構，將機台量測數據以卷積神經網路 (CNN)進行分類，可快速且有效地判別晶圓好壞以達成高良率及高產能之目的。
傳統的錯誤偵測與分類架構，是由工程師將製造過程所蒐集到的狀態識別變量(SVID)，依據個人經驗轉換成不同的統計量稱為FDC參數，再藉由觀察FDC參數的變化以監測晶圓狀況。目前在執行錯誤偵測與分析的過程可分為四個階段，分別為製程確認、資料蒐集、統計分析及卡控規則定義，本研究是以優化統計分析階段為目的，而多數的文獻研究也都致力於優化該階段。本研究先將FDC數據以晶圓為單位轉換成圖像集合，再使用CNN方法對其進行樣型識別，以取得故障模式進而分類晶圓好壞。
本研究提出一種錯誤偵測與分類的架構，藉由樣型識別方法提取關鍵故障特徵，以快速進行晶圓的好壞分類。為確認該研究架構效度，本研究已實證於台灣某先進半導體公司，可有效進行晶圓故障識別，其準確率達95%以上。

Semiconductor manufacturing process has a high demand for quality. Faults that are quickly identified and rectified results in yield enhancement. Tool sensors collect profile data of wafers during the manufacturing process. This study aims to propose a fault detection classification (FDC) framework that allows semiconductor industry to use convolutional neural networks (CNN) to efficiently classify the condition of the wafers as it passes each tool.
Detecting and classifying faults provides critical information for yield enhancement. Traditionally, FDC is a process conducted by engineers transforming profile data called status variable identification (SVID), gathered by the sensors in tools, into FDC statistical results. Using domain knowledge, the engineers will be able to detect the patterns that determines the condition of the wafer. Modern FDC process is commonly categorized into four stages; process tool, data collection, statistical analysis, and business rules, with many variations of research striving to optimize the statistical analysis stage. The intention of this study is to transform the FDC sensor data into a collection of wafer-based images and using CNN image recognition methodology to capture fault patterns from the image database to classify the conditions of the wafers.
The proposed FDC framework utilizes image recognition methodology to swiftly identify the conditions of the wafers and extract the features to determine the rules of fault classification. To validate the methodology an empirical study was conducted at a leading semiconductor company in Taiwan. The model aims to predict the condition of the wafer and extract the feature of the images to highlight the rules for fault classification.

1. Introduction 1
1.1 Research Background and Motivation 1
1.2 Research Objective 2
1.3 Thesis Organization 3
2. Literature Review 4
2.1 FDC Statistical Analysis 4
2.1.1 K Nearest Neighbor 4
2.1.2 Principal Component Analysis 5
2.2 Convolutional Neural Network – Image Recognition 8
2.3 Convolutional Neural Network in FDC Analysis 9
3. Research Framework 11
3.1 Problem Definition 13
3.2 Data Preparation 14
3.3 Convolutional Neural Network – VGG-16 CNN 15
3.4 Validation 16
4. Empirical Study 18
4.1 Problem Definition 18
4.2 Data Preparation 19
4.3 VGG-16 Model Construction 22
4.4 Performance Evaluation 23
4.5 Case Summary 28
5. Conclusion 29
5.1 Contribution 29
5.2 Future Work 29
6. References 31

Agrawal, A., et al. (2017). What to Expect From Artificial Intelligence.
Chien, C.-F., et al. (2012). "Semiconductor fault detection and classification for yield enhancement and manufacturing intelligence." Flexible Services and Manufacturing Journal 25(3): 367-388.
Chien, C.-F., et al. (2007). "Data mining for yield enhancement in semiconductor manufacturing and an empirical study." Expert Systems with Applications 33(1): 192-198.
He, Q. P. and J. Wang (2007). "Fault Detection Using the k-Nearest Neighbor Rule for Semiconductor Manufacturing Processes." IEEE Transactions on Semiconductor Manufacturing 20(4): 345-354.
He, Q. P. and J. Wang (2010). "Large-Scale Semiconductor Process Fault Detection Using a Fast Pattern Recognition-Based Method." IEEE Transactions on Semiconductor Manufacturing 23(2): 194-200.
Hijazi, S., et al. (2015). "Using Convolutional Neural Networks for Image Recognition."
Jing, C. and J. Hou (2015). "SVM and PCA based fault classification approaches for complicated industrial process." Neurocomputing 167: 636-642.
Lee, K. B., et al. (2017). "A Convolutional Neural Network for Fault Classification and Diagnosis in Semiconductor Manufacturing Processes." IEEE Transactions on Semiconductor Manufacturing 30(2): 135-142.
Mansouri, M., et al. (2016). "Kernel PCA-based GLRT for nonlinear fault detection of chemical processes." Journal of Loss Prevention in the Process Industries 40: 334-347.
Moyne, J., et al. (2016). "Big Data Capabilities Applied to Semiconductor Manufacturing Advanced Process Control." IEEE Transactions on Semiconductor Manufacturing 29(4): 283-291.
Nakazawa, T. and D. V. Kulkarni (2018). "Wafer Map Defect Pattern Classification and Image Retrieval Using Convolutional Neural Network." IEEE Transactions on Semiconductor Manufacturing 31(2): 309-314.
Sheriff, M. Z., et al. (2017). "Fault detection using multiscale PCA-based moving window GLRT." Journal of Process Control 54: 47-64.
Vijayaraghavan, R. (2003). "."
Wu, M. and L. Chen (2015). Image recognition based on deep learning.
Yann LeCun, C. I., NYU, et al. "THE MNIST DATABASE of handwritten digits." from http://yann.lecun.com/exdb/mnist/.
Zhou, F., et al. (2016). "Differential feature based hierarchical PCA fault detection method for dynamic fault." Neurocomputing 202: 27-35.