[18a-F104-1~12] 12.2 Characterization and Materials Physics

△ [18a-F104-12] Autoencoder system fabricated with PEDOT:PSS wire

〇Masaru Okada¹, Yasumasa Sugito¹, Wataru Hikita¹, Tetsuya Asai², Yuji Kuwahara¹, Megumi Akai-Kasaya^1,3 (1.Osaka Univ., 2.Hokkaido Univ., 3.PREST)

Keywords:neuromorphic, conductive polymer, neural network

Construction of ANN(Artificial Neural Network) architecture consisting of PEDOT:PSS (poly(3, 4ethylenedioxythiophene):polystyrene sulfonate) wire connection by using machine learning system were performed. The polymer wire growth through polymerization in monomer solution and changes its conductance, whereby it functions as nonvolatile resistive change memory. In this study, an autoencorder ANN were constructed with 54 nodes of polymers by self-learning of 3 kinds of 9 bit binary letters, X, H, T. We compared the resulted structure of the polymer ANN with the computed one and found that the polymer ANN consists of smaller number of weighty connection than that of the computed ANN. The difference of polymer growth condition for each electrodes and influence of neighbors in solution trough might conduct the imbalance polymer growth.