The strategies of filament control for improving the resistive switching performance

With the rapid application of artificial intelligence in daily life and work, the traditional von Neumann architecture device faces the limitation of scalability and high energy consumption. These limitations can be overcome by in-memory computing based on analog resistance switch devices, but the resistive switching behavior depends on the formation and rupture of filaments with spatial and temporal variation. According to the filamentary switching mechanisms, conductive filaments play an irreplaceable role in the resistive switching process, and the stochastic filaments are the main cause of nonuniform performances and variation. Therefore, an efficient way to solve these problems is by controlling the filaments. In recent years, researchers have made many efforts to control the filaments, resulting in numerous feasible methods being invented. Herein, departing from the filamentary mechanisms, the strategies of filament control are discussed from the aspects of electrode optimization, switching layer optimization and channel design. Meanwhile, the challenges of promotion in device performance and application in neuromorphic computing and outlook for future research directions are also discussed. This journal is