A Novel Waveform Classification Method Using Memristor-Based Reservoir Computing

This paper presents a memristor-based reservoir computing (RC) framework for waveform classification tasks, including synthetic signal and electrocardiogram (ECG) datasets. Leveraging the nonlinear dynamic characteristics of memristors and the processing capabilities of the input layer, the model effectively extracts time-series features from input signals. Experimental results demonstrate that as the number of memristors in the reservoir increases, the classification performance significantly improves and stabilizes after reaching a saturation point. Furthermore, the framework exhibits advantages in both low power consumption and high performance, achieving classification accuracies of 100% and 89% for synthetic waveform and ECG classification tasks, respectively. The findings suggest that this approach strikes a favorable balance between energy efficiency and classification performance, showing great potential for widespread applications in low-power scenarios such as wearable devices and edge computing.