Dynamic IGZO-based memristors for cost-effective physical reservoir computing

Abstract

Recent advancements in resistive random-access memory devices have led to their applications in neuromorphic computing by leveraging their synapse-emulating and energy-efficient properties. Some studies have highlighted the multifunctional behaviors of resistive random-access memory (RRAM) devices, such as nociceptor emulation, and edge and reservoir computing. Reservoir computing has gained attention for its ability to tackle efficiently complex tasks, utilizing volatile RRAM devices to create a physical reservoir state with minimal impact between input signals. In this research, a cost-effective reservoir computing implementation is achieved by fabricating a uniform memristor based on the indium gallium zinc oxide (IGZO)-thin film. The ITO/IGZO/TaN device exhibits consistent resistive switching behaviors in terms of cycle-to-cycle and cell-to-cell uniformity. The volatile nature of the ITO/IGZO/TaN device is further explored by using relaxation-assisted set, sequential set processes, paired-pulse facilitation, and learning behavior. The study concludes by demonstrating the capabilities of dynamic memristors in computing systems, showcasing typical 4-bit, cost-effective 7-bit, and high-density multilevel 4-bit reservoir computing states. © 2024 The Physical Society of the Republic of China (Taiwan)