Structural optimization strategies based on amorphous matrix for reliable synaptic devices

Abstract

With the advent of the fourth industrial revolution, artificial intelligence based on big data is rapidly improved by leaps and bounds. Despite these advances, the hardware for implementing AI has a conventional von-Neumann architecture in which memory and processing units are physically separated, resulting in many inefficiencies in the movement of the data. Neuromorphic computing can overcome the limitations of von-Neumann architecture by storing and computing data simultaneously, such as human synapses, to benefit efficient big data processing. Therefore, resistance-switching devices that can mimic human synapses hardware-wise are actively investigated. However, a conventional resistance-switching device called memristor has suffered from stochastic conductive filament formation, which provokes resistance-switching, hindering its commercialization. In this study, a highly reliable memristor was fabricated through structural improvement by inserting porous structures and a buffer layer using amorphous material compatible with CMOS technologies. In addition, the role of the pore and buffer layer was experimentally demonstrated to identify the resistive-switching mechanism of the device.