SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 17: Machine Learning in Dynamics and Statistical Physics I
DY 17.7: Vortrag
Dienstag, 28. März 2023, 11:45–12:00, ZEU 160
A 3-layer injection-locked multimode semiconductor laser neural network — •Elizabeth Robertson1,3, Romain Lance2, Anas Skalli2, Xavier Porte2, Janik Wolters1,3, and Daniel Brunner2 — 1Deutsches Zentrum für Luft-und Raumfahrt, 12489 Berlin, Germany — 2Institut FEMTO-ST, Université Bourgogne Franche-Comté, CNRS UMR6174, Besançon, France — 3Technische Universität Berlin, Institut für Optik und Atomare Physik, 10623 Berlin, Germany
Optical hardware implementations of artificial neural networks (ANNs) have become a hot topic of research due to the inherent parallelism, potentially high speed and energy efficiency of optics [1]. Semiconductor laser networks are of specific interest as they are highly non-linear systems, which can be modulated at high throughput [2]. Previous work using spatial modes as nodes of an ANN, illustrated the use of multimode large area VSCELs for neural network computing in a fully parallel substrate, without pre- or post-processing [3,4]. We further expand this concept to a three-layer network consisting of mutually coupled multimode VSCELs, injection locked to a DFB laser. Here, information is fed into the network by modulating the injection laser, and boolean output weights are implemented using a digital micromirror device. We present an outline of the system, investigate its locking behavior and non-linear response. [1] Huang C. et al., Advances in Physics: X 7, 1 (2022) [2] Skalli A. et al., Opt. Mater. Express 12, 2395-2414 (2022) [3] Porte X. et al., J. Phys. Photonics 3 024017 (2021) [4] Skalli A. et al., Opt. Mater. Express 12, 2793-2804 (2022)