Fig. 1. Structural diagram of metagrating. (a) Structure and function; (b) structural parameters

Fig. 2. Flow chart of grating optimization based on deep learning neural network

Fig. 3. Schematic of neural network training process

Fig. 4. [in Chinese]

Fig. 4. Parameter optimization process. (a) Relationship between A and L_loss when e_p=1000 and b_a=2500; (b) relationship between e_p and L_loss when A=300 and b_a=2500

Fig. 5. Predicted diffraction spectra and actual diffraction spectra of metagratings under different conditions. (a) W₁=2.23 mm, W₂=1.61 mm, H₁=5.21 mm, H₂=0.64 mm, H_sub=1.82 mm;(b) W₁=1.94 mm, W₂=1.70 mm, H₁=2.09 mm, H₂=1.20 mm, H_sub=3.07 mm;(c) W₁=1.51 mm, W₂=0.80 mm, H₁=

Fig. 6. Diffraction spectrum of optimal metagrating screened based on neural network

Fig. 7. Diffraction characteristics of gratings. (a) Relationship between frequency and diffraction angle; (b) relationship between diffraction angle and diffraction efficiency

Fig. 8. Near-field distributions of metagrating at different frequencies. (a) 139.7 GHz; (b) 169.3 GHz