Fig. 1. Schematic diagram of angle resolved scattering system

Fig. 2. Scattering geometry schematic diagram

Fig. 3. Electromagnetic simulation model of angular resolution scattering system

Fig. 4. Cascade machine learning algorithm framework based on decision trees

Fig. 5. Angle-resolved scattering system experimental platform

Fig. 6. Comparison between real data and simulated data

Fig. 7. Geometric schematic diagrams of three types of defects. (a) class-0; (b) class-1; (c) class-2

Fig. 8. Angular resolution scattering signals of three types of defects

Fig. 9. Angular resolution scattering signals of class-0. (a) d=0.05 µm; (b) d=0.2 µm; (c) d=0.35 µm; (d) d=0.5 µm; (e) d=0.65 µm;(f) d=0.8 µm

Fig. 10. ROC curves of defect type prediction results

Fig. 11. Predicted results of width and depth regressors. (a) class-0 width regressor; (b) class-1 depth regressor; (c) class-1 width regressor; (d) class-1 depth regressor; (e) class-2 width regressor; (f) class-2 depth regressor

Fig. 12. Error curve of defect width inversion for cylindrical defects

Fig. 13. Angular resolution scattering signals at various defect depths