Predicting Kerr Soliton Combs in Microresonators via Deep Neural Networks

Formation of the Kerr soliton combs is a widely recognized important but complex issue, which relates to cross-influences among intra-cavity nonlinearities, chromatic dispersions, mode interactions, and pumping effects. Here, we propose and demonstrate a deep neural network model to predict Kerr comb spectra in silica microspheres statistically, via training their transmission spectra. Such a scheme enables soliton comb identification under a particular pump scanning, with error <8%, verified by experimental measurements. This study bridging the deep learning and the microcomb photonics, may provide a powerful and convenient tool for photonic device test and investigation.