Acta Optica Sinica, Volume. 43, Issue 3, 0312008(2023)
Lithography Hotspot Detection Method Based on Pre-trained VGG11 Model
Figures & Tables(10)
Fig. 1. Schematic of hotspot detection method based on convolutional neural network
Fig. 5. Model performance comparison of different pre-trained VGG models. (a) Accuracy; (b) recall; (c) precision; (d) F1 score
Fig. 6. Model performance comparison of different methods. (a) Accuracy; (b) recall; (c) precision; (d) F1 score
Table 1. Data composition of ICCAD 2012 benchmark suite
View tableTable 1. Data composition of ICCAD 2012 benchmark suite
Benchmark Training data Test data Hotspots Non-hotspots Hotspots Non-hotspots Benchmark 1（B1） 99 340 226 319 Benchmark 2（B2） 174 5258 498 4146 Benchmark 3（B3） 909 4643 1808 3541 Benchmark 4（B4） 95 4452 171 3386 Benchmark 5（B5） 26 2716 41 2111
Table 2. Model performance comparison of different pre-trained VGG models
View tableTable 2. Model performance comparison of different pre-trained VGG models
Benchmark Model Accuracy Recall Precision F1 score Benchmark 1
（B1）
VGG11 0.989 0.987 0.987 0.987 VGG13 0.961 0.938 0.968 0.953 VGG16 0.961 0.942 0.964 0.953 VGG19 0.958 0.938 0.959 0.949 Benchmark 2
（B2）
VGG11 0.984 0.988 0.879 0.930 VGG13 0.979 0.988 0.843 0.910 VGG16 0.972 0.978 0.807 0.884 VGG19 0.969 0.956 0.794 0.867 Benchmark 3
（B3）
VGG11 0.986 0.992 0.967 0.980 VGG13 0.984 0.988 0.966 0.977 VGG16 0.980 0.989 0.955 0.972 VGG19 0.964 0.986 0.916 0.950 Benchmark 4
（B4）
VGG11 0.992 0.942 0.918 0.930 VGG13 0.992 0.932 0.918 0.925 VGG16 0.992 0.942 0.905 0.923 VGG19 0.990 0.927 0.889 0.908 Benchmark 5
（B5）
VGG11 0.993 1.000 0.724 0.840 VGG13 0.993 1.000 0.724 0.840 VGG16 0.992 1.000 0.712 0.832 VGG19 0.990 1.000 0.677 0.808 Average VGG11 0.989 0.982 0.895 0.933 VGG13 0.982 0.969 0.884 0.921 VGG16 0.979 0.970 0.869 0.913 VGG19 0.974 0.961 0.847 0.896
Table 3. Comparison of training time of different pre-trained VGG models
View tableTable 3. Comparison of training time of different pre-trained VGG models
Model VGG11 VGG13 VGG16 VGG19 Training time /s 279 317 359 497
Table 4. Model performance comparison of different methods
View tableTable 4. Model performance comparison of different methods
Benchmark Method Accuracy Recall Precision F1 score Benchmark 1
（B1）
Ref.［ 23 ］- 0.995 0.324 0.489 Case 1 0.989 0.987 0.987 0.987 Case 2 0.987 0.982 0.987 0.984 Case 3 0.987 0.982 0.987 0.984 Benchmark 2
（B2）
Ref.［ 23 ］- 0.986 0.702 0.820 Case 1 0.984 0.988 0.879 0.930 Case 2 0.982 0.990 0.864 0.923 Case 3 0.982 0.988 0.864 0.921 Benchmark 3
（B3）
Ref.［ 23 ］- 0.982 0.443 0.640 Case 1 0.986 0.992 0.967 0.980 Case 2 0.985 0.992 0.965 0.978 Case 3 0.986 0.992 0.958 0.975 Benchmark 4
（B4）
Ref.［ 23 ］- 0.972 0.355 0.520 Case 1 0.992 0.942 0.918 0.930 Case 2 0.993 0.937 0.932 0.935 Case 3 0.995 0.969 0.954 0.961 Benchmark 5
（B5）
Ref.［ 23 ］- 0.980 0.549 0.635 Case 1 0.993 1.000 0.724 0.840 Case 2 0.993 1.000 0.750 0.857 Case 3 0.992 1.000 0.712 0.832 Average Ref.［ 24 ］- 0.980 0.300 0.458 Ref.［ 23 ］- 0.983 0.475 0.635 Case 1 0.989 0.982 0.895 0.933 Case 2 0.988 0.980 0.900 0.935 Case 3 0.988 0.986 0.895 0.935
Tools
Get Citation
Copy Citation Text
Lufeng Liao, Sikun Li, Xiangzhao Wang. Lithography Hotspot Detection Method Based on Pre-trained VGG11 Model[J]. Acta Optica Sinica, 2023, 43(3): 0312008
Paper Information
Category: Instrumentation, Measurement and Metrology
Received: Jul. 6, 2022
Accepted: Aug. 31, 2022
Published Online: Feb. 13, 2023
The Author Email: Li Sikun (lisikun@siom.ac.cn), Wang Xiangzhao (wxz26267@siom.ac.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20