Semiconductor optical cavities for photonic signal processing

Neuromorphic photonics offer much faster and energy efficient data processing compared to more matured neuromorphic electronics. I’ll describe a new photonic platform to implement photonic neuromorphic operations. Our optical microcavities in strong light-matter coupling regime serve as building blocks of optical neurons. Light confined in an optical resonator interacting with excitons in optically active material allow to create coupled quasi-particles, exciton-polaritons. Polaritons due to their bosonic nature undergo a transition to non-equilibrium condensate which result in strong laser-like emission from the cavity. This effect occurs at the ps timescale and at single pJ in pulse energy, making this system ideal to implement fast and energy efficient photonic data processing. In our realization of photonic neuron we use the time-coded series of input pulses to code the input information and time-delayed effects to assure the non-linear signal processing. We provide arguments that two types of polariton neural network can be experimentally realized: the binarized time-coded neural network and spiking neural network.