Structured light, with its programmability in multiple degrees of freedom including amplitude, phase, space, frequency and polarization, represents a frontier field in modern optics. The problem is that most structured-light systems still rely on free-space components such as lenses, spatial light modulators, and carefully aligned optics. They work well in the lab, but they are difficult to miniaturize, stabilize, and mass-produce.
In their new work, researchers from Zhejiang University, Westlake University, Peking University, and collaborators present a simple idea with a powerful payoff: treat a complex vectorial light field like a finite set of coefficients in a Hilbert space, and then let a computer search for a compact chip pattern that performs the desired transformation. In practice, the team builds a transmission-matrix target (what the device should do), and uses adjoint-based inverse design to automatically discover the best permittivity layout inside a tiny design window.
They demonstrate the approach on a valley photonic crystal platform, where topological edge states provide robust transport and extra degrees of freedom (valley pseudospin). The goal is to efficiently convert a standard single-mode waveguide into a chosen pseudospin edge state, enabling controlled, on-chip vectorial structured light.
Key performance highlights include:

• Two ultra-compact topological couplers (about 5 micrometers by 5 micrometers) designed for different domain-wall configurations (Type-I and Type-II).
• Near-zero insertion loss in ideal simulations: 0.04 dB (Type-I) and 0.09 dB (Type-II) at 1550 nm, with broad 3-dB bandwidths of 132 nm and 65 nm.
• Fabrication-aware designs that preserve performance in real devices, with measured losses below 0.6 dB (1550 nm, 3-dB bandwidth >60 nm) for Type-I and below 0.8 dB (1550 nm, 3-dB bandwidth 87 nm) for Type-II.

Efficient and broadband coupling is a long-standing bottleneck for connecting conventional waveguides to topological photonic crystals, because their modes do not naturally match. By solving this mismatch automatically through inverse design, the new framework turns topological structured-light states into practical building blocks that can be integrated with standard photonic circuits.
Looking forward, the method is general. Once a target vector field is defined, the same design workflow can be used to create chip-scale components for polarization multiplexing, advanced optical signal processing, and potentially on-chip interfaces between free-space structured beams and topological photonic circuits. In short, the work offers a scalable route for bringing the richness of structured light from optical tables to manufacturable photonic chips. The work entitled “On-chip vectorial structured light field manipulation by inverse design” was published in Frontiers of Optoelectronics (published on Feb. 11, 2026).
DOI：10.2738/foe.2026.0011