Inverse design of reconfigurable varifocal moiré polariton metalenses in a reflective Gires–Tournois interferometer

The development of active and reconfigurable planar optical systems remains a central objective in modern photonics. Here, we propose and theoretically validate a computational inverse-design paradigm that integrates adjoint-based optimization with the linear Stark effect of moiré interlayer excitons embedded within a reflective Gires–Tournois Interferometer (GTI) architecture. By treating a user-defined target light field as the input, our framework automatically compiles it into a spatially resolved voltage map that programs the complex reflection coefficient of the moiré polariton microcavity on demand. The non-symmetric reflective GTI configuration overcomes the Ohmic losses inherent in transmissive designs, enabling full 2π phase modulation while maintaining a high spatially averaged reflectivity (>70%) and operating strictly within safe dielectric limits (E_z ≤ 0.8 V/nm). Leveraging this methodology, we demonstrate the fully automated, ab initio design of a high-numerical-aperture, aberration-managed varifocal reflective metalens with continuous focal-length tuning (up to Δf ≈ 344 μm) and a peak focusing efficiency of 77%. Systematic material screening identifies the twisted WSe₂ homobilayer and the MoTe₂/WSe₂ heterostructure as optimal candidates. This work establishes a direct bridge between computational inverse design and programmable quantum materials, paving the way toward next-generation, chip-scale, field-programmable photonic systems for optical computing, adaptive imaging, and LiDAR applications.