Abstract

In the context of developing new architectures for Neural Networks (NNs) in Photonics, optical reservoir computing is a suitable candidate to fully integrate those systems. In telecommunications, it could potentially allow to reduce energy consumption as well as the number of data processing operations and suppress the need to transduce signals between electronics and optics [1]. Semiconductor lasers (SLs) can provide a highly nonlinear environment, thus a hardware reservoir for computation [2]. A common issue in trying to fully integrate photonic reservoirs is the absence of adjustable hardware output weights. This is tackled here using a digital micromirror device, which allows to spatially select spectral components of the signal, providing tuneable Boolean output weights. A schematic of the experiment is shown in Fig1(a).

© 2023 IEEE