ABSTRACT Widefield interferometry offers non‐destructive, scalable nanometrology for semiconductor photonics, but prevailing pipelines require multi‐frame scanning (or phase‐shifting) and postprocessing of reconstructed noise‐limited phase, and do not provide single‐shot, geometry‐level uncertainties. We introduce an uncertainty‐aware Bayesian computational imaging framework that estimates semiconductor waveguide geometry (e.g., height and width) directly from a single widefield interferogram, coupling an end‐to‐end intensity forward model with Dynamic Nested Sampling to return full posterior distributions and model evidence. Operating in the intensity domain avoids noise transfer to reconstructed topography and remains reliable under low‐signal and sub‐pixel fringe‐shift conditions. Working in a widefield mode is a vital advantage of our Bayesian method, due to fully developed statistics over many pixels in a large field of view, significantly reducing the estimation uncertainties. We successfully validate performance in simulations showing sub‐nanometer height precision and nanometric width accuracy, and in experiments on a metrologically certified 15 nm calibration step and a rib waveguide (design height 8 nm). The framework is model‐agnostic and, given an appropriate forward model and priors, is in principle extendable to other nanostructures. By unifying single‐shot acquisition with probabilistic inference, we establish Bayesian computational nanometrology as a potential route to widefield, uncertainty‐quantified measurements for semiconductor nanophotonics and process‐level monitoring.
Suski et al. (Sun,) studied this question.