Ultrahigh speed artificial intelligence and signal processing using Kerr soliton crystal microcombs (Conference Presentation)

We report optical neural networks (ONNs) and other ultra-high speed signal processing based on integrated Kerr micro-comb sources that is programmable, highly scalable and capable of reaching ultra-high speeds. We demonstrate a convolutional accelerator operating at 11 Tera-OPS/s (TOPS) on 250,000 pixel images for 10 kernels simultaneously, and use the same hardware to sequentially form a deep optical convolutional neural network (CNN) that achieves recognition of the full 10 digit set. Finally, we present new architectures to reach the PetaOP/s regime with > 24,000 neurons using a combination of temporal, wavelength, spatial and polarization multiplexing with standard off-the-shelf telecommunications technology.