Presentation Information
[9a-B21-5]Investigation of a CNT-COOH/Polyoxometalate Polymer Composite for Physical Reservoir Computing
〇(DC)Adrian Dy Go1, Seiya Watanabe1, Hiroyuki S. Kato1, Megumi Akai-Kasaya1 (1.UOsaka)
Keywords:
memory capacity,carbon nanotubes,physical reservoir computing
Physical reservoir computing (PRC) exploits the nonlinear and history-dependent dynamics of physical materials to perform temporal computation without training internal weights. This approach offers a potential route to energy-efficient hardware-based computation by delegating processing to the intrinsic physics of the material. Composite materials based on conducting polymers are of interest as candidate reservoirs owing to their tunable electrical properties and low-cost fabrication. In this work we investigate a CNT-COOH/PEDOT-MeOH:SDBS:POM composite as a candidate reservoir material. The CNT-COOH component introduces charge transport pathways and nonlinearity while POM clusters contribute redox-active character to the transient electrical response. A 64-pin array was used to interface with the material enabling multipoint electrical access. Randomly selected pins served as physical nodes with 20 virtual nodes per node for state readout. Random voltage pulse stimuli were applied to elicit a dynamic current response. Preliminary benchmarking using memory capacity (MC) and NARMA2 was performed to evaluate reservoir characteristics. The composite yielded an MC1 of 2.96 compared to 1.73 for a control without CNT-COOH. The MC(k) approaches 1 at delays k = 0 and k = 1 before declining at k = 2 suggesting the reservoir retains information for at least one preceding timestep. NARMA2 prediction yielded an NMSE of 0.2419 indicating some degree of nonlinear processing capability. These results suggest the composite exhibits behavior relevant to reservoir computing and warrant further investigation.