Crystallinity‐controlled volatility tuning of ZrO2 memristor for physical reservoir computing
Dae Kyu Lee,Gichang Noh,Seungmin Oh,Yooyeon Jo,Eunpyo Park,Min Jee Kim,Dong Yeon Woo,Heerak Wi,YeonJoo Jeong,Hyun Jae Jang,Sangbum Kim,Suyoun Lee,Kibum Kang,Joon Young Kwak
DOI: https://doi.org/10.1002/inf2.12635
2024-10-13
InfoMat
Abstract:Crystallinity‐controlled ZrO2‐based memristor, in contrast to its amorphous counterpart, enables superior memory performance with localized conductive filaments along low‐energy pathways. This configuration promotes a steep switching slope, high resistance state, and forming‐free characteristics. The enhanced volatile switching properties can be extended to short‐term memory functions dependent on pulse conditions for reservoir computing applications. Memristors have been emerging as promising candidates for computing systems in post‐Moore applications, particularly electrochemical metallization‐based memristors, which are poised to play a crucial role in neuromorphic computing and machine learning. These devices are favored for their high integration density, low power consumption, rapid switching speed, and significant on/off ratio. Despite advancements in various materials, achieving adequate electrical performance—characterized by threshold switching (TS) behavior, spontaneous reset, and low off‐state resistance—remains challenging due to the limitations in conductance filament control within the nanoscale resistive switching layer. In this study, we introduce an efficient method to control the ZrO2 crystallinity for tunable volatility memristor by establishing the filament paths through a simple thermal treatment process in a single oxide layer. The effect of ZrO2 crystallinity to create localized filament paths for enhancing Ag migration and improving TS behavior is also investigated. In contrast to its amorphous counterpart, crystallized ZrO2 volatile memristor, treated by rapid thermal annealing, demonstrates a steep switching slope (0.21 mV dec–1), a high resistance state (25 GΩ), and forming‐free characteristics. The superior volatile performance is attributed to localized conductive filaments along low‐energy pathways, such as dislocations and grain boundaries. By coupling with enhanced volatile switching behavior, we believe that the volatility is finely tuned to function as short‐term memory for reservoir computing, making it particularly well‐suited for tasks such as audio and image recognition.
materials science, multidisciplinary