Flex-SNN: Spiking Neural Network on Flexible Substrate

Abstract

Organic thin-film transistors (OTFTs) are suitable for building flexible sensor systems, since they can be fabricated on thin substrates, such as paper or films. In this study, we propose an organic spiking neural network (SNN) model and circuit implementation called flex-SNN for real-time data processing and for reducing communication costs in flexible sensor systems. The flex-SNN model comprises an input layer, excitatory neurons, and a fully connected synapse matrix. The synapses are made of insulator and contact layers for OTFTs, enabling the compact fabrication of the flex-SNN circuit on flexible substrates. In addition, the flex-SNN circuit is scalable since the synapse matrix can be configured using metal wire crossings. The circuit simulations performed using measurement-based device models of OTFTs and synapses demonstrate that the flex-SNN circuit achieves an inference accuracy of 97.0% for classifying 0’s and 1’s in the modified national institute of standards and technology (MNIST) dataset.