Device‐Level Implementation of Reservoir Computing With Memristors

The advancement of artificial intelligence through deep learning has enabled large‐scale data processing with human‐like intelligence, but its drawbacks, including inefficient learning and excessive power consumption, are demanding a breakthrough based on interdisciplinary research efforts. Reservoir computing (RC) is an alternative computing scheme specifically suitable for temporal data processing, utilizing a dynamic reservoir and a single readout layer. Without any learning process, the reservoir can extract time‐dependent features from the input and transform them into a high‐dimensional form using its unique characteristics. Memristors, which are two‐terminal devices with dynamic switching characteristics, can effectively operate as the core components of RC systems with a crossbar array structure. In this review, an overview of the memristor‐based implementation of RC is provided. The necessary properties of the reservoir and the switching mechanisms of memristors are discussed, followed by recent progress on actual demonstrations of memristor‐based RC based on such principles. The remaining challenges and solutions for device‐level implementation of RC are subsequently presented, providing an opportunity to broaden the readers’ perspectives on the actualization of this novel computing scheme.