Title Page
ABSTRACT
초록
Contents
1. Introduction 14
2. Related Works 16
2.1. NAND 16
2.1.1. NAND Structure and Principle 16
2.1.2. NAND Process defects 18
2.2. Semiconductor Automatic Analysis Techniques 19
2.2.1. FE-SEM 20
2.2.2. FIB-SEM 22
2.3. Deep Learning 23
2.3.1. Deep Learning Principle 23
2.3.2. Convolutional Neural Network 25
3. Method 29
3.1. Manual Feature Design 29
3.1.1. Sobel filter 29
3.1.2. Canny Edge detector 30
3.2. Template Matching 33
3.3. YOLOv8 34
4. Experiments 40
4.1. Experimental Setup 40
4.2. Experimental Results 41
4.2.1. Manual Feature design 43
4.2.2. Template Matching 45
4.2.3. YOLOv8 46
5. Conclusion 48
Reference 49
Table 1. YOLOv8 pretrained Detect models result 39
Table 2. 4 Pattern Recognition Function Evaluation Results for 2 Devices 41
Figure 1. Nand memory array 17
Figure 2. (a) NAND flash memory array. (b) Channel Etch Failure Case. 18
Figure 3. Schematic diagram of the NAND with concave surface polished by dimple grinder from top view and cross section 20
Figure 4. Full sequence 21
Figure 5. Configuration outlines of this new method and the flow from acquisition of 3D coordinate by CSI to observing the selective FOV by SEM. 22
Figure 6. Comparison of Conventional FIB-SEM and Orthogonally-arranged FIB-SEM 22
Figure 7. Ethos NX5000 FIB-SEM : Auto Micro Sampling Tool 23
Figure 8. Comparing deep learning with machine learning. 24
Figure 9. Hierarchically-structured taxonomy. 26
Figure 10. LeNet introduced by Yan LeCun 27
Figure 11. AlexNet introduced by Krizhevsky 2014 27
Figure 12. (a) 2-layer ResNet block. (b) 2 generalized residual blocks (c) 2-layer ResNet block from 2 generalized residual (d) 2-layer RiR block. 27
Figure 13. (a) lena.jpg grayscale image (b) after sobel 30
Figure 14. (a) original image (b) after Canny 31
Figure 15. Manual Sequence 32
Figure 16. TM Sequence 34
Figure 17. YOLO: Release Dates Timeline 35
Figure 18. Model structure of YOLOv8 detection models(P5) - yolov8n/s/m/l/x 36
Figure 19. Anchor Free Detection 37
Figure 20. Mosaic 38
Figure 21. YOLOv8 pretrained Detect models 39
Figure 22. 3D NAND 5-row image a) Slit patten b) Hole pattern 40
Figure 23. Pattern Recognition Function Evaluation Results for 2 Devices 42
Figure 24. Sobel filter x2 Image Processing Results 43
Figure 25. Canny Edge Detector Image Processing Results 44
Figure 26. Training and validation curves for YOLOv8 46