Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
FM: Fachverband Funktionsmaterialien
FM 12: German-French Focus Session: Materials Research in Polar Oxides – From Domain Engineering to Photonic and Electronic Devices I
FM 12.4: Vortrag
Mittwoch, 11. März 2026, 10:30–10:45, BEY/0138
Photocurrent-based recognition in time-series tasks using reservoir computing — •Yan-Meng Chong1, Atreya Majumdar2, Ingvild Hansen1, Karin Everschor-Sitte2, and Dennis Meier1,2,3 — 1Department of Materials Science and Engineering, Norwegian University of Science and Technology (NTNU), Trondheim, Norway — 2Faculty of Physics and Center for Nanointegration Duisburg-Essen (CENIDE), University of Duisburg-Essen, Duisburg, Germany — 3Research Center Future Energy Materials and Systems, Research Alliance Ruhr, Bochum, Germany.
In reservoir computing, input data are mapped into higher dimensional space, translating non-linear problems into linearly solvable ones. In general, any physical system that possesses non-linearity, complexity, short-term memory, and reproducibility can serve as reservoir. Here, we investigate the ferroelectric semiconductor ErMnO3 as a potential candidate material for reservoir computing. We show that the system displays pronounced non-linear changes in photocurrent under varying light intensity. The response can be tuned by changing the metal-semiconductor contacts (Schottky- or Ohmic-like behavior) used for readout, determining the timescale on which the photocurrent vanishes after illumination. This relaxation behavior in the OFF state gives fading memory. We perform training on variations in the output (photocurrent), which allows for recognition in time-series tasks. Interestingly, ferroelectric domain walls can also be used as reservoirs with characteristic photocurrent signals, giving new opportunities to down-scale or enhance the complexity of physical reservoirs.
Keywords: ferroelectric; reservoir computing; ErMnO$_3$; photoconductivity; fading memory