Artificial Intelligence Aided Rapid Design of THz Stereo-Metamaterials Based Polarization Conversion Devices for Highly Sensitive and Efficient Strain Sensors

Polarization conversion devices, pivotal in the field of photonics, are crucial for manipulating the polarization state of electromagnetic waves. These devices have significantly contributed to advancements in technology, specifically in the realms of communication, imaging, and remote sensing. They have facilitated applications ranging from space exploration to the detection of hidden objects, the identification of explosives, and even cancer detection using the Terahertz (THz) spectral range, which spans from 0.1 to 10 THz. Owing to its wide bandwidth and remarkable ability to penetrate dielectric materials, the THz wave is garnering more interest for noninvasive screenings, high-resolution imaging, and improved precision in data collection. Metamaterials (MM), which are artificially synthesized materials composed of repeated subwavelength-sized meta-atoms, offer a promising avenue for the advancement of these devices. Metamaterial-based THz polarization conversion devices have an advantage over conventional devices because of their thin and compact structure. They are not only easier to integrate but also offer flexibility, marking a new era in the evolution of polarization conversion technology. Stretchable electronics have recently obtained rising interest because the mechanical deformation of elastomeric substrates has been used to induce spectral shifts in the resonance of nanophotonic structures such as nanoparticle dimer extinction and gratings. With the advancement of polymer technology and lithographic techniques, it is possible to fabricate metamaterials on thin, flexible substrates, including Kapton polyimide or polyimide (PI), polyethylene naphthalate (PEN), and polydimethylsiloxane (PDMS). Research indicates that crumpling an unbending metamaterial on a soft backing layer surface can influence the response. The ability to realize such structures on elastic substrates presents further opportunities for tuning via mechanical straining.<br/><br/>In light of that information, we utilized a stereo-metamaterial (SMM) structure for new-generation THz polarizers converting the polarization from linearly to circularly and elliptically polarized wave at the THz frequency range in reflection mode. In this work, we present the processes and results of the inverse design of SMM using the artificial neural network (ANN), trained by various parameters, including polarization status and ellipticity angle, to achieve highly efficient device performance. Training and testing our ANN with the created datasets by simulation for the inverse design of the device, design parameters were obtained by giving an artificial EM response or ellipticity angle spectrum or vice versa more efficiently and rapidly. This study was designed to use a tandem neural network (TNN). It is created by two deep neural networks (DNNs); an inverse neural network (INN) predicts design parameters from the target spectrum, and a forward neural network (FNN) predicts the target spectrum from design parameters, as demonstrated in this work. During the training process of TNN, FNN is trained and tested first. Then, after FNN’s weight freezes, it is cascaded with INN for training and testing TNN. The training error can be derived by comparing the target and prediction spectra, solving the non-unique input problem.<br/><br/>The SMM devices based on the ANN-powered design were successfully fabricated on the stretchable substrate. We demonstrated efficient sensing of different polarization statuses using THz polarimetry spectroscopy under the different strains. The sensitivity reaches to 0.1% strain sensing. More details on the results will be presented.