Chiral qBICs Induced Near-Infrared Exciton–Polaritons Engineering via a Structure–Spectra Framework

Exciton–polaritons (EPs), hybrid quasi-particles usually arising from the strong coupling between quasibound states in the continuum (qBICs) and excitons in transition-metal dichalcogenides (TMDCs), enable photon spin-controlled light–matter interactions with distinctive properties. Traditional empirical reasoning and trial-and-error procedures for a high-quality (Q) chiral qBIC metasurface for EPs are inefficient and time-consuming. Particularly, computational requirements scale exponentially with the number of design parameters for a complex metasurface structure, severely limiting the exploration of optimal configurations. In this work, we construct a structure–spectra framework based on a residual deep neural network to engineer near-infrared EPs within a bulk MoTe₂-based cut-cuboids metasurface featuring a high-Q chiral qBIC dominated by a magnetic dipole mode with a near-perfect circular dichroism (∼0.97). The structure–spectra framework accurately predicts the strong coupling of the self-hybridized EPs and Rabi splitting with only 1.5% error and robust performance across unexplored parameter space. Notably, the transition from weak through intermediate to strong coupling regime is successfully simulated and predicted by modulating the exciton oscillator strength. This flexible tuning of the spin-controlled self-hybridized EPs by oscillator strength and tilt angle demonstrates a strong dependence on the quality factor-to-effective mode volume ratio (Q/V_eff) of the proposed self-hybridized chiral metasurface. The proposed framework provides a new paradigm for engineering chiral qBIC metasurfaces and EPs in TMDCs, accelerating the explorations and applications of strong light–matter interactions in the near-infrared region.