Rapid design of terahertz stereo-metamaterial devices with tandem neural network for efficient polarization conversion

Polarization conversion devices are crucial for manipulating the polarization status of electromagnetic waves in photonics. Artificial intelligence (AI)-based data-driven approaches are revolutionizing nanophotonics by allowing efficient inverse design methods. We propose using the power of AI for rapid and high-efficiency inverse design of THz stereo-metamaterial polarizers, thereby accelerating the analysis of the desired device. For this purpose, we utilized a stereo-metamaterial (SMM) structure for new-generation THz polarizers, converting the polarization of waves from linear to circular and elliptically polarized waves at the THz frequency range in reflection mode. A tandem neural network (TNN) with a weighted loss function is employed to design the SMM-based THz polarizer device inversely, based on a desired ellipticity angle spectrum rather than the structure's reflection and phase spectra, thereby obtaining the corresponding structural parameters for the first time. Training and testing our TNN with the simulated datasets created for the inverse design of the device, design parameters were obtained by providing an ellipticity angle spectrum or vice versa more efficiently and rapidly.