Abstract

Photonic artificial neural networks (ANN) offer a great potential for efficient image recognition, thanks to the possibility of integration, low energy dissipation, ultra-fast response times, and availability of a wealth of nonlinear optical responses [1]. We consider an ANN architecture consisting of photonic nodes implemented with semiconductor-based micropillar lasers with integrated saturable absorber which exhibit fundamental properties of biological neurons, such as excitability, refractory periods and temporal summation [2]. These nodes operate six orders of magnitude faster as compared to biological neurons. This unique property makes them particularly well-suited for applications where speed is a crucial requirement .

© 2023 IEEE