Dec 2, 2024
4:30pm - 4:45pm
Sheraton, Second Floor, Independence West
Jaehyeong Lee1,Hyeongjin Moon1,Jinha Choi1,Jeonghoon Son2,Seungkun Kim2,Jongchan Ryu1,Hongju Kim1,Seyoung Kim2,Sangbum Kim1,Yun Seog Lee1
Seoul National University1,Pohang University of Science and Technology2
Jaehyeong Lee1,Hyeongjin Moon1,Jinha Choi1,Jeonghoon Son2,Seungkun Kim2,Jongchan Ryu1,Hongju Kim1,Seyoung Kim2,Sangbum Kim1,Yun Seog Lee1
Seoul National University1,Pohang University of Science and Technology2
Electrochemical random-access memory (ECRAM) has emerged as a promising synaptic transistor for energy-efficient analog artificial neural networks, with the potential to significantly reduce the cost required to train large-scale models. Among various types of ECRAM, oxygen-ion-conducting metal-oxide-based ECRAM offers advantages over proton-conducting counterparts, including higher thermal and chemical stability, as well as improved retention. Moreover, the compatibility of oxygen ion-conducting materials with existing CMOS technology facilitates easier integration into current manufacturing processes for integrated circuit devices. However, conventional oxygen ion-conducting electrolytes such as HfO<sub>2</sub> and ZrO<sub>2</sub> often exhibit low ionic conductivity at room temperature, resulting in slow speed in weight-updating operations.<br/><br/>In this study, we employ a Sm-doped CeO<sub>2</sub> (SDC) thin film as the electrolyte layer. SDC has demonstrated promising oxygen-ion conductivity as well as high surface exchange coefficient in solid oxide fuel cell applications. We deposit SDC thin films with varying Sm concentrations using a co-sputtering technique, which allows control of the doping level by adjusting the RF power during the deposition process. Electrochemical impedance spectroscopy is carried out to evaluate their electrochemical characteristics to determine the optimal doping level for device operations at room temperature. We fabricate metal-oxide ECRAM devices utilizing the optimized SDC electrolyte film, achieving weight-updating operating speeds in the tens of nanoseconds. Furthermore, we model the ECRAM device to analyze the significantly improved speed and further enhance the performance of metal-oxide ECRAM for high-speed neuromorphic computing applications.