[19p-C301-1~14] 9.5 New functional materials and new phenomena

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation AwardEnglish Presentation

[19p-C301-4] Performance of MoSe2-SWNT in-material reservoir computing device on time-series prediction tasks

〇(DC)AlifSyafiq KamarolZaman¹, Saman Azhari^1,2, Yuki Usami^1,3, Hirofumi Tanaka^1,3 (1.Kyutech, 2.Waseda University (IPS), 3.Neuromorphic Center)

Keywords:

Time-series prediction,Carbon nanotube,Non-linear dynamics

Reservoir computing (RC) is a framework derived from recurrent neural networks for efficient temporal information processing with reduced complexity and power consumption. This study presents a high-efficiency temporal signal processing device using a memristor-based MoSe2-SWNT core-shell structure, drop-casted onto a SiO2/Si substrate with aluminum electrodes. The device shows resistive switching in its I-V characteristics within a ±5V potential window. The MoSe2-SWNT combination enhances electrochemical activity, surface area, conductivity, and nonlinear dynamics, which are essential for complex tasks. Huang et al. showed that SWNTs improve electron transport and reduce charge transfer resistance. Standalone MoSe2 or SWNT does not show resistive switching. RC tasks tested the device’s temporal capabilities, with over 90% accuracy in waveform generation and NARMA2 prediction. The nanojunctions create complex electron pathways, contributing to nonlinearity. The MoSe2-SWNT device shows potential for resistive switching memory and machine learning tasks by mimicking reservoir computing architecture.