High Performance of an In-Material Reservoir Computing Device Achieved by Complex Dynamics in a Nanoparticle Random Network Memristor
Oradee Srikimkaew, Deep Banerjee, Saman Azhari, Yuki Usami, Hirofumi Tanaka
Abstract
High Resolution Image Download MS PowerPoint Slide An in-material reservoir computing (RC) device with an Ag–Ag 2 S core–shell nanoparticle (NP) network is proposed. Network-wide nonlinear sine-wave outputs of higher frequencies and varying phases were produced from the different Ag + ion diffusion rates and filament formation caused by the heterogeneous NP size in the thiol layer. Such emergent dynamics of multiple information regimes enabled the reconstruction of Fourier waves, with a maximum accuracy of 99% achieved only for trained outputs with mixed spatiotemporal complexities. Additionally, the device showed stable retrieval of past information with a two-times-step delay and successfully computed a two-step time-series prediction task with 87% accuracy.