Reservoir computing for image processing based on ion-gated flexible organic transistors with nonlinear synaptic dynamics

Abstract

Reservoir Computing (RC) is an efficient framework for processing sequential data. It captures the dynamic features of complex data through high-dimensional mapping, significantly enhancing machine learning capability. However, due to the lack of suitable materials and device fabrication processes, developing highly energy-efficient, flexible, and wearable RC systems remains a challenging task. In this paper, flexible ion-gated transistors were fabricated by spin-coating process at room temperature, utilizing polyimide (PI) as the flexible substrate, polyvinyl alcohol (PVA) doped with lithium perchlorate (LiClO₄) as the gate dielectric, and poly[(bithiophene)-alternate-(2,5-di(2-octyldodecyl)-3,6-di(thienyl)-pyrrolyl pyrrolidone)] (DPPT-TT) as the organic semiconductor. Based on the electric double-layer (EDL) coupling in ion-gated transistors and the rich ionic dynamics, the device can simulate nonlinear synaptic functions such as excitatory postsynaptic current (EPSC), multi-pulse facilitation, and learning-forgetting-relearning behaviors. Based on the transistors’ nonlinear synaptic functions, the constructed RC system can enhance image features while reducing the image size by half, effectively extracting and amplifying hidden features in the original images. In the handwritten digit recognition task, the RC system improved the recognition rate from 79.8 % to 90.6 %, compared to an Artificial Neural Network (ANN) of the same scale. The effective combination of flexible organic ion-gated transistors and the RC framework will undoubtedly contribute to the further development of the next generation of wearable intelligent systems.