Reservoir Computing for Pattern Recognition and Time‐Series Prediction Based on Nanofluidic Negative Differential Resistance

Physical reservoir computing (RC) employs the dynamics of a system to obtain a nonlinear representation of the input data that is useful for pattern recognition and complex time‐series forecasting. Here, we propose a new RC method for image classification where the reservoir dynamics makes use of the negative differential resistance (NDR) phenomena in a nanofluidic biomimetic pore caused by salt precipitation. The experimental reservoir consists of charged conical nanopores bathed in salt solutions, and the sequential NDR phenomena are characterized by abrupt drops in the ionic current when the applied voltage exceeds a series of threshold voltages. By exploiting these nonlinear characteristics, we demonstrate theoretically an RC realization that is able to classify a broad range of binary images with high accuracy by using a nanofluidic reservoir. Also, we consider temporal data forecasting using a biologically inspired model of voltage‐gated ion channels with NDR characteristics. To this end, we use two generic inward‐ and outward‐rectifying channels characteristic of biological membranes and test the RC forecasting with the Mackey–Glass chaotic time series.