Jiyeon Kim1,Jae Hak Lee1,Minho Jin1,Haeyeon Lee1,Ji Soo Kim1,Eunho Lee2,Youn Sang Kim1
Seoul National University1,Kumoh National Institute of Technology2
Jiyeon Kim1,Jae Hak Lee1,Minho Jin1,Haeyeon Lee1,Ji Soo Kim1,Eunho Lee2,Youn Sang Kim1
Seoul National University1,Kumoh National Institute of Technology2
Neuromorphic computing is emerging compared to conventional computing systems based on Von Neumann structure which have limitation of processing delay when massive data is delivered between CPU and memory. To overcome this Von Neumann bottleneck, human-brain imitating synaptic devices have been attracted in neuromorphic system. Especially, memristors have attracted great attention due to the low switching voltage, enabling low-power consuming devices. Also, non-volatile memory characteristics of memristor can emulate potentiation process in synapse. In order to determine whether the memristor exhibits neuromorphic properties, synaptic behaviors such as synaptic plasticity and linearity need to be accompanied. To implement synaptic memristor, molybdenum disulfide (MoS<sub>2</sub>), which is a representative example of transition-metal dichalcogenides (TMDs), has been used for channel layer of synaptic memristor because its eletrical properties such as high mobility and large bandgap can be easily controlled by number of layers, which is available due to its two-dimensional structure. Among its distinctive characteristics, many researchers are focusing on precisely controlling sulfur vacancies of MoS<sub>2</sub>, which is directly related to synaptic memristor operation according to conductance change. However, traditional fabrication methods to make sulfur vacancies showed some limitations such as plasma damage, complicated process by chemical vapor deposition (CVD) and contamination of CVD reactor, suggesting necessity of concise and non-destructive method. Herein, we introduced facile solvent-dipping method immersing the samples in the solvent to control sulfur vacancies, considering polarity index and Hansen solubility parameter (HSP). We selected three solvents with similar HSP but different polarity: tetrahydrofuran, chlorobenzene, and toluene. Toluene has been reported as a common solvent to make sulfur vacancies in MoS<sub>2</sub>. Moreover, in this study, polarity index was newly proposed to confirm if the polarity of solvent affects sulfur vacancy generation in MoS<sub>2</sub> when selected solvents have similar HSP values. We confirmed tetrahydrofuran-treated synaptic memristor generated the highest sulfur vacancies because bipolar solvent can remove both polar and nonpolar sulfur. This unique solvent-dipping process resulted in great performance of tetrahydrofuran-treated synaptic memristor; non-volatile memory characteristics such as high programming/erasing current ratio (>10<sup>3</sup>) and long retention (>10<sup>4</sup> s) and synaptic behaviors of paired-pulse facilitation (PPF) about 170% and 8-times increased linearity. Neuromorphic computing was conducted by Modified National Institute of Standards and Technology (MNIST) to show synaptic behavior clearly, demonstrating increased recognition accuracy after solvent treatment. Visual simulation was illustrated to verify long term potentiation/depression (LTP/LTD) of synaptic devices. Flexible synaptic memristor was also fabricated and maintained its performance without noticeable degradation while bending test was operated, suggesting possible application in next-generation flexible devices. We suggested concise, non-destructive, and reliable solvent-dipping method to control sulfur vacancies in TMDs, and this is a novel study to our knowledge. Our research provides the framework for TMD-based neuromorphic devices, requiring further research for real application in brain-imitating neuromorphic system.