Integrating machine learning into the inverse design, fabrication, and characterization of photonic devices brings computational speed-ups throughout the entire device design process. In this presentation, we report on using machine learning for improving the inverse design of nanophotonic meta-structures and speeding up the characterization of single-photon emitters under sparse measurements. We find that compressing the design space of meta-structures using autoencoders can greatly reduce the time required to compute high-efficiency designs using global optimization methods. With the characterization of single-photon emitters, we find that convolutional neural networks can classify the g2 correlation function of an emitter above or below the 0.5 correlation threshold up to 100 times faster than full characterization with 95% accuracy. Using this fast characterization of single-photon emitters, we can speed up super-resolution imaging up to 12 times faster than conventional methods. Our work paves the way for quantum machine learning-assisted global optimization of nanostructures and super-resolution imaging.
|