Acta Optica Sinica, Volume. 43, Issue 23, 2315001(2023)
Lithography Hotspot Detection Based on Improved YOLOv3
Figures & Tables(12)
Fig. 1. Hotspot detection task, the left figure is the mask, the block area on the right is the hotspot area
Fig. 7. Results of using YOLOv3 after training to detect lithography hotspots. (a) Original mask; (b) output image of the model
Table 1. Data of ICCAD 2012 dataset
View tableTable 1. Data of ICCAD 2012 dataset
Name Tech /nm Training dataset Testing dataset HS NHS HS Area /nm2 Benchmark 1 32 99 340 226 12516 Benchmark 2 28 174 5285 498 106954 Benchmark 3 28 905 4642 1796 122565 Benchmark 4 28 95 4453 177 82010 Benchmark 5 28 26 2716 41 49583 Benchmark 6 28 1200 17096 2512 361112
Table 2. Dataset after data augmentation
View tableTable 2. Dataset after data augmentation
Name Tech /nm Training dataset Validation dataset Testing dataset HS NHS HS NHS HS Area /nm2 Benchmark 1 32 743 2579 33 141 226 12516 Benchmark 2 28 1251 4776 141 509 498 106954 Benchmark 3 28 6421 33428 728 3708 1796 122565 Benchmark 4 28 673 4013 79 440 177 82010 Benchmark 5 28 187 2445 21 271 41 49583 Benchmark 6 28 8616 15378 948 1718 2512 361112
Table 3. Experimental result
View tableTable 3. Experimental result
Name Method Recall Rrecision F1 Benchmark 1 YOLOv3 1.00 0.97 0.98 YOLOv3+SENet 1.00 0.98 0.99 Benchmark 2 YOLOv3 1.00 1.00 1.00 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 3 YOLOv3 1.00 0.38 0.55 YOLOv3+SENet 1.00 0.44 0.61 Benchmark 4 YOLOv3 1.00 0.98 0.99 YOLOv3+SENet 1.00 0.97 0.98 Benchmark 5 YOLOv3 1.00 0.93 0.965 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 6 YOLOv3 1.00 0.35 0.52 YOLOv3+SENet 1.00 0.45 0.62
Table 4. Experimental result for different attention mechanisms
View tableTable 4. Experimental result for different attention mechanisms
Name Method Recall Rrecision F1 Benchmark 1 YOLOv3+CBAM 1.00 1.00 1.00 YOLOv3+SENet 1.00 0.98 0.99 Benchmark 2 YOLOv3+CBAM 1.00 0.93 0.97 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 3 YOLOv3+CBAM 1.00 0.44 0.61 YOLOv3+SENet 1.00 0.44 0.61 Benchmark 4 YOLOv3+CBAM 1.00 0.51 0.68 YOLOv3+SENet 1.00 0.97 0.98 Benchmark 5 YOLOv3+CBAM 1.00 1.00 1.00 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 6 YOLOv3+CBAM 1.00 0.24 0.38 YOLOv3+SENet 1.00 0.45 0.62
Table 5. Comparison of experimental results of different methods
View tableTable 5. Comparison of experimental results of different methods
Name Method Recall Precision F1 Benchmark 1 Method in Ref.［ 9 ］0.95 0.36 0.52 Method in Ref.［ 14 ］1.00 0.32 0.49 YOLOv3+SENet 1.00 0.98 0.99 Benchmark 2 Method in Ref.［ 9 ］0.98 0.22 0.35 Method in Ref.［ 14 ］0.99 0.70 0.82 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 3 Method in Ref.［ 9 ］0.98 0.20 0.33 Method in Ref.［ 14 ］0.98 0.44 0.64 YOLOv3+SENet 1.00 0.44 0.61 Benchmark 4 Method in Ref.［ 9 ］0.94 0.16 0.27 Method in Ref.［ 14 ］0.97 0.36 0.52 YOLOv3+SENet 1.00 0.97 0.98 Benchmark 5 Method in Ref.［ 9 ］0.98 0.18 0.30 Method in Ref.［ 14 ］1.00 0.55 0.70 YOLOv3+SENet 1.00 1.00 1.00 Benchmark 6 Method in Ref.［ 9 ］0.96 0.22 0.35 Method in Ref.［ 14 ］0.98 0.48 0.64 YOLOv3+SENet 1.00 0.45 0.62
Tools
Get Citation
Copy Citation Text
Mu Lin, Fanwenqing Zeng, Xiaoxuan Liu, Fencheng Li, Jun Luo, Yijiang Shen. Lithography Hotspot Detection Based on Improved YOLOv3[J]. Acta Optica Sinica, 2023, 43(23): 2315001
Paper Information
Category: Machine Vision
Received: May. 5, 2023
Accepted: Aug. 29, 2023
Published Online: Dec. 8, 2023
The Author Email: Yijiang Shen (yjshen@gdut.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20