Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 3: Poster
AKPIK 3.4: Poster
Thursday, March 21, 2024, 11:00–14:30, Poster B
Prospects of hybrid atomic-photonic neural networks for neuromorphic computing — •Mingwei Yang1,2, Elizabeth Robertson1,2, Kilian Junicke2, Lina Jaurigue3, Kathy Lüdge3, and Janik Wolters1,2 — 1Deutsches Zentrum für Luft- und Raumfahrt, Institute of Optical Sensor Systems, Berlin, Germany. — 2Technische Universität Berlin, Berlin, Germany. — 3Technische Universität Ilmenau, Institute of Physics, Ilmenau, Germany
Optical neural networks have been identified as promising for neuromorphic computer hardware, attributed to their inherent parallelism, fast processing speeds, and low energy consumption. We discuss how photonic networks can be combined with atomic vapors providing optical non-linearities and memory functionality. In particular, we discuss an implementation of a convolutional neural network with a saturable absorber for optical nonlinearity [1], a reservoir computing architecture with atomic memory [2], and the prospects of such systems as the Ising model solver.
[1] Yang, Mingwei, et al. "Optical convolutional neural network with atomic nonlinearity." Optics Express 31.10 (2023): 16451-16459.
[2] Jaurigue, Lina, et al. "Reservoir computing with delayed input for fast and easy optimisation." Entropy 23.12 (2021): 1560.
Keywords: optical neural networks; optical non-linearity; atomic memory; Ising model; neuromorphic computing