Deep neural network-enabled bifunctional terahertz metasurface design for absorption and polarization conversion.

• Deep Neural Networks assist in parameter optimization. • The suggested structure has the flexibility to function as either an absorber or a polarization converter. • The polarization conversion bandwidth is 3.7 THz, and the absorption bandwidth is 5.21 THz. • The relative bandwidths in the absorption and polarization conversion states reached 78.9% and 58.3%, respectively. • The suggested structure has a low profile and a wider variety of micro device application scenarios. With the assistance of deep neural networks, a bifunctional metasurface (MS) was designed and optimized, i.e., a broadband absorber and a broadband polarization converter. The MS acts as a wide absorber when the vanadium dioxide (VO 2) is in the metallic state and has an absorption bandwidth of 5.21 THz with an absorption rate ≥ 90%. In contrast, the MS acts as a linear–linear polarization converter when the top VO 2 is in the insulating state and has a bandwidth of 3.7 THz with a conversion efficiency ≥ 90%. The bandwidth in both states is maximum compared to other bifunctional counterparts, while this bifunctional MS has good parametric and angular tolerance characteristics and low material cost. The proposed structure and design method of the bifunctional MS can provide a useful reference for the research of new multifunctional terahertz devices. [ABSTRACT FROM AUTHOR]