Highly Efficient Classification of Time‐Series Based on Resistive Switching Cluster‐Assembled Materials

The need for high accuracy time-series classification and pattern recognition tasks is rapidly increasing due to the ubiquitous deployment of the internet of things (IoT) paradigm, such as in the development of brain–computer interfaces. Several statistical methods have been proposed for the analysis of large amounts of data, including artificial neural networks. However, the amount of energy required for the training of these models is extremely high, and the request for data storage in the cloud is even more demanding. Edge computing solutions are currently regarded as viable alternatives to mitigate this energetically unfavorable condition. Herein, a random-assembled resistive switching (RS) device is reported based on a nanostructured nanocomposite Au/ZrOx film used as a preprocessing element of a highly efficient classifier of time-series. As the resistance of the device evolves in a complex way under voltage variable input, a time-series statistical analysis is applied to extract the key features from the nonlinear electrical response of the nanostructured device. The potential of combining these nanocomposite RS devices is demonstrated by their ability to accurately and in real-time classify neuronal traces, corresponding to physiological and evoked local field potentials and spiking activity recorded from the rat barrel cortex.