Fig. 1. Network architecture flowchart for MMA inverse design based on BDNN. (a) Schematic diagram and side view of the structural unit of MMA. (b) Schematic diagram of an inverse design based on BDNN. (c) Schematic diagram of controlling the graphene chemical potential with an applied bias voltage. (d) Calculated graphene chemical potential as a function of applied bias voltage. (e) Schematic diagram of the FNN and PNN.

Fig. 2. Prediction results and errors of the single-band metasurface absorber inverse design based on BDNN. (a) Loss function and R² for the training and test sets based on BDNN. (b), (c) Results of the BDNN prediction of the frequency and absorptance of the test set. (d) Absorption spectra (1–10 THz) obtained by simulation modeling based on the parameters predicted by BDNN. (e), (f) Frequency and absorptance of the absorption peak obtained by simulation modeling based on the parameters predicted by BDNN.

Fig. 3. Errors of absorption spectra in the inverse design based on KNN and RF algorithms as well as accuracy and stability tests of different algorithmic models. (a), (b) Absorption spectra (1–10 THz) obtained by simulation modeling based on the parameters predicted by KNN and RF. (c), (d) Absorption peak frequencies of the absorption spectra obtained by simulation modeling based on the parameters predicted by KNN and RF. (e), (f) Absorptance of the peaks in the absorption spectra obtained by simulation modeling based on the parameters predicted by KNN and RF.

Fig. 4. Validation of optimal values for each structural parameter of the MMA. (a)–(g) d₁, p, h₁, μ_c, d₂, d₃, and d₄.

Fig. 5. Absorption spectra, normalized equivalent surface impedance, and electric field distribution for the absorption peak. (a) Absorption spectra of MMA in single-band and broadband absorption modes. (b), (c) Normalized equivalent surface impedance in single-band and broadband absorption modes. (d)–(i) Electric field distribution in the x–y and x–z planes in single-band and broadband absorption modes (5.0 THz).