Ovonic Threshold Switching for Ultralow Energy Physical Reservoir Computing

In this article, we demonstrate physical reservoir computing (RC) in ovonic threshold switching (OTS) devices. We show that SiGeAsSe OTS is suitable as a physical reservoir because of the nonlinear change in the number of delocalized defects. With the combination of phase space reconstruction (PSR), our algorithm can project data into high-dimensional spaces, thereby enhancing the distinguishability of the data. Such ability is suitable for high-accuracy authentication and classification. Our algorithm can be implemented using both crossbar arrays or individual devices, and achieves a significantly low (0.08%) equal error rate (EER) on gait authentication in simulation. Furthermore, we validated our concept by successfully implementing the algorithm on nine hardware OTS devices and achieved an EER of 4.2% on gait authentication. The low leakage current level of OTS, the fast learning of RC, and interval-based readout responses all contribute to a significantly reduced energy consumption of our proposed method.