2026/07/23 by Yan Meng Chong, Atreya Majumdar, Manuel Zahn +3
#cond-mat.mtrl-sci
Physical reservoir computing represents an energy efficient approach for processing temporal signals by exploiting the intrinsic nonlinear dynamics and fading memory of a physical system. Recently, ferroelectric semiconductors moved into focus as reservoir materials motivated by their versatile electronic responses to external stimuli. Here, we explore the fundamental possibility to recognize time-varying light pulses via photo-induced currents, using the small-band-gap p-type semiconductor ErMnO3 as a model system. Under white light illumination, ErMnO3 exhibits non-linearly evolving photo-induced currents and controllable relaxation dynamics that naturally realize the high-dimensional projection and fading memory capabilities required for reservoir computing. The reservoir capability of ErMnO3 is reflected by the improved recognition accuracy of "Past" input pulses, which increases from ~33% to ~93% after applying reservoir transformation to the input signal. The results present ferroelectric hexagonal manganites as a promising platform for photo-induced current-based reservoir computing and highlight the potential of light-driven oxide semiconductors for temporal information processing.