As integrated photonic circuits grow more complex, measuring the properties of each individual component becomes harder. Researchers at Aalto University and the University of Central Florida have shown that a neural network can infer the losses and resonance-frequency shifts of every ring in an array of coupled ring resonators, using only spectral measurements of the whole array.

The supervised machine-learning approach worked with high accuracy across several experimental configurations, offering a scalable and non-invasive alternative to traditional calibration, and could pave the way for automated calibration and control of photonic chips. The QMAT authors are Elizabeth Louis Pereira, Faluke Aikebaier and Jose Lado of the Correlated Quantum Materials group at Aalto, working with Andrea Blanco-Redondo's team at the University of Central Florida.

Figure: Pereira et al., APL Photonics (2026); image from arXiv:2601.09423, CC BY 4.0