A Bifunctional Memristor Enables Multiple Neuromorphic Computing Applications

As a promising building block of the emerging neuromorphic computing hardware, memristive structures with multi-functionalities are highly desired to implement diversified computing applications in a single device. However, the demonstration of such multi-functional structures remains limited. In this work, an Ag/GeS/Pt-based bifunctional memory structure with both long-term and short-term memristive behaviors is reported, enabling multiple neuromorphic computing applications in a single device. It is found that the unexpected short-term switching in Ag/GeS/Pt can not only be used to simulate learning/relearning and forgetting behavior but can also be implemented for reservoir computing. While for long-term switching memristive behavior, its application is demonstrated as the traditional memory. The work reveals a novel coexistence of the two types of resistive switching, shedding light on various neuromorphic computing applications such as reservoir computing and traditional memory realized in a single memristive device.