Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 6: Machine Learning in Dynamics and Statistical Physics I

DY 6.6: Vortrag

Montag, 9. März 2026, 10:45–11:00, HÜL/S186

Physical Reservoir Computing with Ferroelectric Oxides for Time-series Classification Tasks — •Atreya Majumdar¹, Yan Meng Chong², Dennis Meier^{1, 2, 3}, and Karin Everschor-Sitte¹ — ¹Faculty of Physics and Center for Nanointegration Duisburg-Essen (CENIDE), University of Duisburg-Essen, Duisburg, Germany — ²Department of Materials Science and Engineering, Norwegian University of Science and Technology (NTNU), Trondheim, Norway — ³Research Center Future Energy Materials and Systems, Research Alliance Ruhr, Bochum, Germany.

Physical reservoir computing leverages the intrinsic complexity, non-linearity, and fading memory of material systems to process temporal data for solving time-series pattern recognition tasks. Magnetic and ferroelectric materials have recently emerged as promising reservoir computers, offering dynamics well suited for processing time-dependent signals [1]. Here, we demonstrate that the photocurrent dynamics of the ferroelectric semiconductor ErMnO3 can be harnessed as an effective physical reservoir for real-time time-series classification. Moreover, the relaxation time of the photocurrent can be controllably tuned, providing flexibility to capture different temporal features and thereby enhancing performance. Altogether, the results highlight the potential of ferroelectric oxides as scalable, energy-efficient platforms for real-time physical reservoir computing.

[1] K. Everschor-Sitte, et al., Topological magnetic and ferroelectric systems for reservoir computing. Nat. Rev. Phys. 6, 455 (2024).

Keywords: Physical reservoir computing; Machine learning; Ferroelectric materials; Pattern recognition; Time-series classification