On-demand customization of the enhancement of the photonic spin Hall effect

The photonic spin Hall effect (PSHE) has been widely researched, exhibiting great potential in applications of precision measurement, nanophotonic devices, and quantum information processing. However, the enhancement of PSHE in previous work occurs only in the case of limited incident angle, wavelength, and polarization, hence saying nothing of on-demand customization. In this work, we propose an Al-Si hybrid dielectric metasurface and a bidirectional long short-term memory (Bi-LSTM) neural network model to achieve the on-demand customization of the enhancement of PSHE. First, an Al-Si hybrid dielectric metasurface is proposed for the flexible manipulation of generalized Brewster effect (GBE) with an ultra-wide range of incident angle, wavelength, and polarization (nearly arbitrary), which breaks through the limitation on the enhancement of PSHE. Then, in the forward prediction by employing a Bi-LSTM neural network model, the reflectivity, phase difference, and transverse displacement of PSHE can be accurately predicted by inputting the structural parameters of the metasurface. With the same Bi-LSTM neural network and extra screening algorithm (SE), the real-time on-demand customization of the enhancement of PSHE can be achieved besides without the non-uniqueness problem. Our results show a flexible way to achieve the enhancement of PSHE at nearly arbitrary conditions and provide a robust tool for customizing and optimizing PSHE performance for specific application requirements.