Microring Resonator Dispersion Metrology with Neural Networks

Published

August 31, 2026

Author(s)

Ergun Simsek, Shao-Chien Ou, Gregory Moille, Kartik Srinivasan

Abstract

Precise control of geometry and material properties is critical for the reliable fabrication of high-performance photonic integrated circuits. Such control is essential in nonlinear integrated photonics, where a key quantity for devices such as microresonator frequency combs (microcombs) is the resonator dispersion, which not only dictates device behavior but also provides a sensitive fingerprint of both ring dimensions and material refractive index, making it an attractive observable for post-fabrication characterization. In this work, we develop a machine learning framework to solve three complementary problems: (i) predicting resonator width and height from integrated dispersion calculations/measurements, (ii) identifying the correct material dispersion model associated with different precursor gas ratios, and (iii) predicting the coefficients of a sixth-order polynomial to reconstruct the integrated dispersion spectrum directly from ring dimensions. These three neural networks together enable both inverse and forward characterization of microring resonators. Using numerically generated datasets based on Sellmeier-type material models, we show that: (i) in the absence of noise, ring dimensions can be predicted with sub-nanometer accuracy; (ii) in realistic scenarios with relatively precise resonator frequency measurements (±50 MHz or better), <8 nm accuracy in dimension prediction can be achieved using only ≈45 dispersion samples; and (iii) for less precise systems with higher noise levels (±200 MHz), prediction errors increase to the ≈16 nm range. Sellmeier model classification remains highly robust in all cases, achieving accuracies above 99 %. Importantly, we find that dispersion samples taken far from the pump resonance provide the most informative input, reducing the need for full spectrum characterization. Furthermore, the forward prediction network reconstructs the dispersion spectrum directly from ring dimensions with high accuracy, providing a fast alternative to time-consuming numerical simulations and enabling rapid assessment of whether and where anomalous dispersion can be achieved in a given design. By combining both inverse and forward predictions, our results highlight the potential of machine learning applied to integrated dispersion data as a rapid, non-destructive tool for wafer-scale quality control and process monitoring in photonic foundries.

Citation

Nanophotonics

Volume

15

Pub Type

Journals

Download Paper

https://doi.org/10.1002/nap2.70269

Local Download

Citation

Simsek, E. , Ou, S. , Moille, G. and Srinivasan, K. (2026), Microring Resonator Dispersion Metrology with Neural Networks, Nanophotonics, [online], https://doi.org/10.1002/nap2.70269, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=961345 (Accessed September 29, 2026)

Additional citation formats