Abstract
In the realm of flexible substrate technology, inkjet printing has emerged as a highly sought-after method due to its cost-effectiveness, scalability for mass production, simplicity, and environmental sustainability. While recent advancements have led to the development of fully printed artificial neurons, realizing a fully printed Echo State Network (ESN) remains elusive. This study introduces a novel approach involving the fabrication of a low-cost, minimally processed, nonlinear computation element entirely through inkjet printing. Positioned on a polyethylene terephthalate film substrate, this element serves as the activation function for an Echo State Network. The conventional activation function within the ESN architecture is supplanted by the electrical response curve of the inkjet-printed neuron. This neuron is constructed using silver nanoparticle ink, complemented by a layer of hexagonal boron nitrade. To assess the activation performance for time series prediction within the Echo State Network framework, the Mackey-Glass time series data is employed as a benchmark. Comparative analysis is conducted against established conventional activation functions, utilizing test mean squared error as the performance metric. The outcomes of this evaluation highlight the efficacy of the proposed nonlinear computational element as an activation function within the Echo State Network paradigm.