Rapid classification of quantum sources enabled by machine learning

Rapid and deterministic nanoscale assembly of quantum emitters remains to be a daunting challenge for the realization of practical, on-chip quantum photonic devices. The major bottleneck is the time-consuming second-order photon autocorrelation measurements for the classification of solid-state quantum emitters into "single" and "non-single" photon sources during the quantum device assembly. We have adapted supervised machine learning algorithms to perform such classification in an efficient sub-second process based on sparse autocorrelation data. We demonstrate an ~80% fidelity of emitter classification based on datasets containing on average only one co-detection event per bin. In contrast, the conventional fitting classification method based on Levenberg-Marquardt fitting typically requires two-orders of magnitude longer collection times, and it fails entirely when applied to the same datasets. We anticipate that machine learning-based classification will provide a unique route to enable rapid and scalable assembly of quantum nanophotonic devices and can be directly extended to other quantum optical measurements, promising breakthroughs in quantum information, sensing and super-resolution microscopy.