Work

Physics-informed sparse observations enable spatial-state inference for intelligent photonic alignment

Year: 2026
Type: article
Abstract: Optical alignment remains a bottleneck in high-throughput photonic-chip packaging. Conventional active-alignment methods rely on dense power-feedback scans, where local power measurements mainly guide incremental motion rather than reveal the underlying spatial coupling state. Here, we present a physics-informed sparse-observation framework for fiber-to-chip alignment. By interpreting directional power variations through mode-overlap coupling, axial field evolution, and direction-dependent displacement sensitivity, the framework enables inference of the dominant three-dimensional spatial coupling state from fewer than ten local power values per inference. The framework combines physics-informed reparameterization with an LRC–ResNet–MTL network to extract spatial-state information from sparse power features. Analytical modeling and full-wave simulations validate the coupling model under displacement and small angular perturbations. Experiments on measured fiber-to-chip coupling data over a 2–20 μ m axial range achieve RMSEs of 0.048 μ m, 0.048 μ m, and 0.101 μ m along the x -, y -, and z -directions, respectively. Additional validations across coupler designs, material platforms, wavelengths, and polarization states support the applicability of the proposed framework under different practical coupling conditions. Compared with dense feedback-based probing, the proposed strategy substantially reduces the required number of observations while achieving component-wise RMSEs below 0.11~ μ m. These results recast local optical-power probing from a feedback-only measurement process into a state-inference strategy, providing a data-efficient route toward photonic-chip packaging and precision optical testing.
Language: English
Open Access status: gold
Estimated APC paid: $2,270
OpenAlex ID: W7225885418
FWCI:
Cites: 41
Cited by:
Related to: 10