Exciton-polariton multilayered neural networks

Abstract

Exciton-polaritons are hybrid light-matter quasiparticles. Being such hybrid, they inherit the fast dynamics of light and strong nonlinearities of matter. Nonlinearity is essential for neural network models to solve high complexity tasks. Thus, excitonpolaritons are believed to be a promising platform for realising high throughput neural network hardware. In this thesis, we present a theoretical scheme for multilayered neural network realization using exciton-polaritons. We then demonstrate the system’s ability to emulate multilayer perceptron and solve the nonlinear XOR problem.