A large-scale stretchable neuromorphic circuit for on-body edge computing

Abstract

Intrinsically stretchable electronics can be used to make wearable devices that collect large amounts of multimodal sensory data. This has led to a demand for enhanced near-sensor computing capabilities that can process such data. One potential solution is neuromorphic edge computing implemented using stretchable organic electrochemical transistors, but the lack of a scalable fabrication method for these transistors has limited the size and complexity of the systems. Here we report the fabrication of large-scale, intrinsically stretchable organic electrochemical transistor arrays with a density up to 10,000 transistors per square centimetre. The transistors exhibit good synaptic performance, including linear, precise conductance programming and good retention. The performance uniformity of the array enables the hardware implementation of a stretchable artificial neural network for processing health data, including heart attack risk assessment and kernel convolution for locating propagation wavefronts. We also explore the potential of implementing reinforcement learning algorithms on the neuromorphic circuits for use in soft robotics. A photolithographic fabrication process can be used to create soft and stretchy organic electrochemical transistor arrays that have a density up to 10,000 devices per square centimetre, and can perform edge computing tasks in wearable devices and soft robots.