In-situ integrated nonlocal Huygens metasurfaces enable robust vortex beams from multimode fibers

The paper demonstrates a robust C-band optical vortex generator by integrating a silicon nonlocal Huygens metasurface directly onto the end facet of a commercial multimode fiber (MMF). The metasurface supports overlapped Mie resonances and nonlocal bound states in the continuum (BICs), yielding high-transmission bands by satisfying the generalized nonlocal Huygens condition. Through momentum-space coupling between MMF modes and Huygens-BICs with topological polarization-vortex nature, the output maintains doughnut-shaped profiles under static and dynamic perturbations, demonstrating exceptional robustness against fiber deformation. This approach provides a compact, alignment-tolerant platform for robust structured light generation in MMFs for endoscopy, optical trapping and communications.

The key challenge addressed is that MMFs, which support hundreds to thousands of spatial modes, produce random and unstable speckle-like outputs due to mode coupling from perturbations such as bending, non-uniform stress, or core-diameter fluctuations. Conventional real-space pixelated phase- or polarization-type diffractive elements cannot be directly applied to MMFs because their performance requires precise pixelated pre-compensation and degrades severely under environmental variations. The proposed nonlocal metasurface operates through collective resonant modes sustained across the entire periodic lattice, with optical signatures intrinsically encoded in momentum space and protected by global symmetries, conferring exceptional robustness against real-space local perturbations.

The metasurface design uses a square lattice of triangular silicon meta-atoms with broken in-plane inversion symmetry, which converts ideal BICs into radiative q-BICs with controlled linewidths while creating momentum-space polarization vortices carrying topological charges. Two structural variants (meta-1 and meta-2 with heights of 259 nm and 518 nm) are presented, showing how meta-atom height controls spectral overlap between q-BICs and broad Mie resonances. Both geometries support two q-BICs near the C-band: q-BIC 1 is a magnetic-dipole mode (in-plane current loop) with topological charge +1, and q-BIC 2 is an electric-dipole mode (out-of-plane oscillation) with topological charge −1. As meta-atom height increases, the narrow transmission dip of q-BIC 1 converts into a transmission peak due to coupling between the narrow magnetic-dipole q-BIC and broad Mie resonances, while the high-transmission peak of q-BIC 2 persists.

The fabrication process uses electron beam lithography (EBL) and inductively coupled plasma (ICP) etching with CMOS-compatible holders and baseplates for high-precision fiber alignment. Amorphous silicon films are deposited onto the fiber end-facet via electron beam evaporation to avoid thermal damage. The measured angle-resolved reflection and transmission spectra of the on-fiber nonlocal metasurface agree well with calculated results accounting for fabrication-induced non-idealities such as inward recess of meta-atom sidewalls from ICP etching.

For vortex generation, the metadevice exploits momentum-space coupling between guided modes of the MMF and the nonlocal metasurface, exciting k-dependent states of polarization near q-BIC 2. Due to spin-orbit locking inherent in the BIC topological configuration, the left circular polarization component exhibits a vortex with l = −1 while the right circular polarization component possesses a vortex with l = +1. Far-field intensity profiles exhibit a clear doughnut shape, and linear polarization imaging reveals the winding of the electric-field vector around the singularity, consistent with a −1 topological charge.

Robustness testing involved deliberately bending the MMF body into configurations at angles of 0°, 45°, 90°, 135°, and 180°. The characteristic doughnut-shaped vortex profiles are essentially preserved across all bending angles, while a bare MMF under the same conditions exhibits randomly varying speckle patterns. Further testing with complex, randomly combined deformations including twisting, bending, and stretching shows the doughnut-shaped intensity profiles remain remarkably stable, while bare MMF output fluctuates randomly. Under continuous dynamic perturbations over a 1.8-second exposure period, the vortex mode retains high modal fidelity in the time-averaged sense with even better doughnut profile.

The work concludes that the demonstrated concept can be extended to other forms of momentum-space light shaping, such as structured beam generation, beam shifting, or complex wavefront engineering. The high-Q dielectric resonances also open avenues for ultrasensitive sensing, low-threshold lasers, and nonlinear frequency conversion on a fiber tip. Future integration of active materials could enable dynamic tunability of optical modes, leading to reconfigurable metafiber devices for adaptive optics and intelligent light manipulation.

Improvements to AI Systems:

1. Physics-Informed Inverse Design for Metasurface Topology: Train a generative AI model (e.g., diffusion or GAN) on simulated and measured spectra of nonlocal metasurfaces (varying meta-atom height, lattice symmetry, and sidewall angle) to predict optimal geometries for arbitrary target transmission/reflection profiles. The AI would learn the generalized nonlocal Huygens condition and q-BIC coupling physics, enabling rapid design of new metasurfaces for other wavelength bands (e.g., visible or mid-IR) without brute-force electromagnetic simulations.

2. Robustness-Aware Mode Decomposition and Prediction: Develop a neural network that takes the physical state of a multimode fiber (bending angle, twist, strain) as input and predicts the output modal field distribution, trained on data from the paper’s robustness tests (0°–180° bends, combined deformations). This AI would learn the invariant topological properties (polarization vortices, doughnut profiles) and separate them from random speckle, enabling real-time prediction of output fields for any arbitrary fiber perturbation—useful for adaptive optics in endoscopy or telecommunications.

3. Momentum-Space Feature Extraction for Structured Light Classification: Build a convolutional neural network (CNN) that operates on angle-resolved reflection/transmission spectra (momentum-space images) to classify the topological charge and linewidth of q-BICs, even with fabrication non-idealities (e.g., sidewall recess). The AI would be trained on the paper’s measured vs. simulated spectra to generalize to other nonlocal metasurface designs, enabling automated quality control and characterization of fabricated devices.

4. Spin-Orbit Locking Emulator for Polarization-Vortex Generation: Create a physics-constrained recurrent or transformer model that learns the mapping from input fiber mode excitation (e.g., LP01, LP11) to output polarization-vortex states (l = ±1) based on the metasurface’s BIC topology. This AI could then simulate and optimize new metasurface designs for generating higher-order vortices (e.g., l = ±2) or vector beams with custom topological charges, accelerating development of structured light sources for optical trapping and quantum communications.

5. Dynamic Perturbation Compensation via Reinforcement Learning: Implement a reinforcement learning agent that controls a deformable mirror or spatial light modulator (SLM) to maintain a stable doughnut-shaped output from a bare MMF, using the paper’s time-averaged fidelity data as a reward signal. The agent would learn to counteract random mode coupling in real time, effectively mimicking the robustness of the nonlocal metasurface but in a reconfigurable electronic platform—enabling adaptive fiber systems for dynamic environments.

What the Improved AI System Can Do:

• Automate the design of nonlocal metasurfaces for any target wavelength or application (sensing, lasing, beam shaping) by generating and validating thousands of candidate geometries in minutes, reducing reliance on manual parameter sweeps.

• Predict and stabilize structured light output from multimode fibers under arbitrary physical perturbations, enabling real-time correction in medical endoscopes (e.g., beating heart motion) or free-space optical links (e.g., atmospheric turbulence).

• Characterize fabricated metasurfaces non-destructively by analyzing their momentum-space spectra, flagging fabrication errors (sidewall angle, height mismatch) and predicting performance degradation before deployment.

• Generate novel vortex beams with tailored topological charges and polarization textures on demand, useful for high-capacity mode-division multiplexing in fiber communications or for creating optical tweezers with complex force landscapes.

• Adaptively maintain high-quality beam profiles in real-world fiber systems without physical metasurfaces, by learning to pre-compensate for perturbations electronically—offering a flexible, reconfigurable alternative to fixed photonic designs.