Inverse design of quasi-bound states in the continuum metasurface for the polarization independent enhancement of Goos-Hänchen shift

Bound states in the continuum (BIC) have been widely researched and applied in optics due to their unique electromagnetic response. However, there are still difficulties in predicting and customizing BIC spectra. To address this issue, we design an efficient combined neural network for highly accurate prediction of quasi-bound states in the continuum (q-BIC) spectrum, as well as for the inverse design of the polarization independent enhancement of the Goos-Hänchen (GH) shift. Firstly, we propose a C₄ symmetric metasurface for achieving q-BIC spectrum and providing the condition of enhanced GH shift. By employing a combined neural network, the intensity, position, shape, and phase of q-BIC spectrum with ultra-narrow resonance can be accurately predicted and on-demand customized, even under a small dataset. Besides, we develop a screening algorithm for the q-BIC spectrum to quickly realize the polarization independent enhancement of GH shift. As an application, an ultra-high sensitivity refractive index sensor has been proposed, whose sensitivity can reach 2.31×10⁷ µm/RIU for TM polarization and 1.03·10⁶ µm/RIU for TE polarization. Therefore, this work brings new solutions for quick prediction of q-BIC spectrum and the development of flexible polarization photonic devices.