[20p-C102-1~16] 9.5 New functional materials and new phenomena

▲ [20p-C102-3] In-materio demonstration of reservoir computing using a random network of single-walled carbon nanotube/Ag-Ag₂S nanoparticle device

〇(P)Deep Banerjee^1,2, Yuki Usami^1,2, Hirofumi Tanaka^1,2 (1.KYUTECH, 2.Neuromorphic AI Research Center)

Keywords:in-materio, reservoir computing, single walled carbon nanotube, Ag-Ag2S

In-materio reservoir computing is gaining popularity for next generation AI hardware as their neural network architecture is highly brain inspired with fast trainability of only output weights. Many such in-materio RC hardware have been proposed, but one with nanomaterials come close to brain like computing framework. In this regard we fabricate an in-materio RC device with single-walled carbon nanotube (SWNT)/Ag-Ag₂S nanoparticle random network and utilizing the non-linear spatio-temporal information processing emergent from the redox behavior; we proceeded to carry out future time-series task of non-linear autoregressive moving average (NARMA). By a simple regression training of high dimensional outputs obtained from SWNT/Ag-Ag₂S we successfully predicted NARMA-2 with a high accuracy of 85%, while a higher time step prediction of NARMA-10 was also achieved with a relatively lower accuracy of 65%. These results indicate that the random non-linear networks of SWNT/Ag-Ag₂S hold the potential for future power efficient RC based AI applications.