Reservoir Computing Using Interfacial Memristors with Native SiO_x Nanostructures Modified by Room-Temperature Plasma Oxidation

We present an interface-type memristor utilizing native silicon oxide nanostructures achieved through controlled plasma oxidation engineering at room temperature. This process effectively reduces the concentration of oxygen vacancies on the surface of the native silicon oxide, facilitating oxygen-vacancy-mediated interfacial resistive switching with excellent COMS compatibility and low switching variability (1.03% for C2C and 3.96% for D2D variability). Moreover, this dynamic memristor exhibits favorable nonlinearity and short-term memory (STM) effects, which are crucial properties for the development of memristor-based reservoir computing (RC) systems. Compared with the conventional neural network, the RC system requires fewer trainable weights to achieve an accuracy of 96.7% for the Iris classification task. By the addition of virtual nodes to the reservoir with masks, the accuracy can be further increased to 100%. A simple digit recognition task is also employed, and noisy images that are not included in the original training set can be recognized correctly. These results highlight the great potential of our device for constructing reservoir computing hardware.