Fig. 1. Schematic diagram of a deep learning based digital lithography autofocus system
Fig. 2. Image of the centroid of each out-of-focus range on the CCD
Fig. 3. Coarse check focus network structure
Fig. 4. Bottleneck modules
Fig. 5. Confusion matrix for network models with different layer structures on the test set
Fig. 6. Training results of ResNet28+FF
Fig. 7. Normalized evaluation curves for different definition evaluation functions in a set of out-of-focus images
Fig. 8. Search algorithm flow chart
Fig. 9. Experimental system diagram
Fig. 10. Coarse focus check process display
Fig. 11. Precision check focus process display
Network model | Accuracy/% |
---|
ResNet18 | 70.0 | ResNet34 | 77.5 | ResNet50 | 77.5 |
|
Table 1. Accuracy of network models with different layer structures on the test set
Network model | Accuracy/% |
---|
ResNet18+FF | 77.5 | ResNet34+FF | 83.1 | ResNet50+FF | 84.5 |
|
Table 2. Accuracy of different layer structured network models with added feature fusion modules on the test set
Image definition evaluation function | Time taken to evaluate a single image /s |
---|
Laplacian | 0.005 | Energy | 0.998 | Brenner | 0.247 | SMD | 0.991 | variance | 0.771 |
|
Table 3. Time required to process images with different definition evaluation functions
Off-focus volume/μm | Coarse check focus results | Steps used for precision focus checks | Total time/ms | Focus check results |
---|
-40 | Grade 4 negative defocus 98% | 6 | 263.8 | True | -29 | Grade 3 negative defocus 53.8% | 5 | 238.4 | True | -18 | Grade 2 negative defocus 88% | 4 | 212.1 | True | -10 | Grade 1 negative defocus 58.4% | 3 | 191.4 | True | -1 | Coarse focal plane 80% | 2 | 168.9 | True | 9 | Grade 1 positive defocus 54.8% | 3 | 188.8 | True | 17 | Grade 2 positive defocus 78.6% | 4 | 212.9 | True | 28 | Grade 3 positive defocus 73.5% | 5 | 241.1 | True | 39 | Grade 4 positive defocus 96.8% | 6 | 263.5 | True |
|
Table 4. Focus detection performance of the same pattern in different out-of-focus situations using the proposed method
Off-focus volume /μm | Steps used for focus checks | Total time/ms | Focus check results |
---|
-40 | 41 | 1 023.1 | True | -29 | 30 | 729.3 | True | -18 | 19 | 478.1 | True | -10 | 11 | 302.4 | True | -1 | 2 | 75.6 | True | 9 | 10 | 298.1 | True | 17 | 18 | 520.5 | True | 28 | 29 | 804.1 | True | 39 | 40 | 1 055.2 | True |
|
Table 5. Focus detection performance of the same pattern in different out-of-focus situations using conventional methods
Focus check graphics | Coarse check focus results | Time for coarse focus checks /ms | Steps used for precision focus checks | Total time for precision focus checks /ms |
---|
| Grade 3 positive defocus78.6% | 93.7 | 5 | 148.6 | | Grade 3 positive defocus73.5% | 93.7 | 5 | 145.5 | | Grade 3 positive defocus61.8% | 93.7 | 5 | 147.2 |
|
Table 6. Comparison of focus detection performance of different patterns at 28 μm out of focus
Focus check graphics | Coarse check focus results | Time for coarse focus checks/ms | Steps used for precision focus checks | Total time for precision focus checks/ms |
---|
| Grade 2 negative defocus68.4% | 93.7 | 4 | 123.6 | | Grade 2 negative defocus88% | 93.7 | 4 | 123.3 | | Grade 2 negative defocus76% | 93.7 | 4 | 125.6 |
|
Table 7. Comparison of focus detection performance of different patterns at -18 μm out of focus
Focal plane detection error/μm | Number of times each error occurs in the coarse check focus | Number of times each error occurs in the precision check focus |
---|
0 | 1 | 2 | ±1 | 3 | 15 | ±2 | 4 | 3 | ±3 | 3 | 0 | ±4 | 2 | 0 | ±5 | 3 | 0 | ±6 | 1 | 0 | ±7 | 3 | 0 |
|
Table 8. Results of the coarse and precise focus detection errors in 20 tests