Engineering of multiple bound states in the continuum by latent representation of freeform structures

We demonstrate a neural network capable of designing on-demand multiple symmetry-protected bound states in the continuum (BICs) freeform structures with predefined symmetry. The latent representation allows tuning geometry differentiable, continuous way. show rich band inversion and accidental degeneracy these by interacting directly. Moreover, high design accuracy is demonstrated for arbitrary control BIC frequencies using photonic property readout to interpret representation.