AI-Guided Design and Fabrication of Diamond Microdisk Resonators for Spin–Photon Interfaces

We present a machine-learning–guided design and fabrication framework for vertically loaded diamond microdisk resonators (VLDMoRt) that enables high-efficiency spin–photon interfaces. These devices combine high-Q performance, efficient free-space coupling, and compatibility with scalable photonic integration; however, discrepancies between predicted and measured optical responses have limited their reliability. Our approach implements a closed-loop feedback pipeline linking electromagnetic simulation, diamond nanofabrication, and experimental validation. A UNet-based segmentation model trained on scanning electron microscopy images reconstructs precise 3D geometries, enabling accurate optical simulations that incorporate realistic surface roughness and etching profiles. By iteratively refining both structural parameters and fabrication conditions, we achieve ~90% consistency between simulated and measured quality factors and resonance characteristics. This workflow establishes a systematic route toward reproducible, high-performance photonic cavities for quantum information applications.