Deep-Learning-Assisted Design of Dynamically Tunable Near-Infrared Dual-Band Lithium Niobate Metamaterial Absorber for Optical Communication.

To overcome the fixed operating wavelengths and limited tunability of conventional metamaterial absorbers, this study proposes a dynamically tunable near-infrared dual-band lithium niobate (LN) metamaterial absorber platform at optical communication wavelengths. Using an "end-to-end + gradient optimization" deep learning inverse-design framework, we efficiently and accurately derive the structural parameters matching target absorption spectra. The designed LN metamaterial absorber (LN-MMA) exhibits dual-band near-perfect absorption peaks at 1310 and 1550 nm, achieving peak efficiencies of 99.9% and 99.5%, respectively. Owing to the strong electro-optic effect of LN, the resonance wavelengths can be dynamically tuned via an external electric field. Furthermore, by integrating the LN-MMA with coherent technology, a dynamically tunable coherent perfect absorber (LN-CPA) is realized, exhibiting absorption peaks at 1550 and 1704 nm with efficiencies of 98.5% and 98.1%. In this configuration, the absorption response can be actively tuned by the phase difference between two incident coherent beams, enabling all-optical modulation. These results demonstrate a compact absorber with both electrical and optical tunability, showing strong potential for inter-satellite laser communication, optoelectronic signal conversion, optical detection, and all-optical signal processing.