Photonic neuromorphic computing using symmetry-protected zero modes in coupled nanolaser arrays

Photonic neuromorphic computing has emerged as a promising approach toward energy-efficient artificial neural networks (ANN). Nanolasers, in particular, have become attractive candidates due to their ultra-low power consumption and intrinsic nonlinear characteristics. In this work, we propose a photonic neuromorphic computing architecture based on symmetry-protected robust zero modes at the center of the optical spectrum in coupled semiconductor nanolaser arrays. We experimentally demonstrate that even a small set of coupled nanolasers inherently provides non-convex classification capabilities, enabling it to solve non-trivial classification tasks. As a benchmark, we show that a 2 × 2 nanolaser array, acting as a hidden nonlinear layer with recurrent coupling is able to solve the XNOR logical gate. Our results further highlight the computation capabilities of such nanolaser array by showing robust classification performance even under challenging conditions, such as the classification of highly compressed handwritten digits with significantly overlapping feature boundaries. These findings suggest that symmetry or topologically protected modes in nanolaser arrays can leverage robust optical connections to tackle complex problems without the need of scaling up the number of neurons. Nanolasers, valued for ultra-low power and nonlinearities, are promising for neuromorphic computing. Authors show coupled nanolaser arrays with symmetry-protected modes act as hidden nonlinear layers in photonic neural networks, enabling XNOR logic gates and highly compressed digit classification.