Large-scale stretchable neuromorphic circuits for on-body edge processing of sensory data

The rapid development of intrinsically stretchable electronics for use on the human bodies and robots has significantly enhanced the ability to collect multi-modal data at high spatiotemporal resolutions, over extended periods, and across diverse body locations. This progress has generated a growing demand for enhanced computing capabilities to process sensory data, making near-sensor edge computing an attractive solution. Stretchable organic electrochemical transistors (OECTs) have been demonstrated to be as a viable platform for integrating neuromorphic edge computing functions into these human-interfaced systems. However, the lack of a scalable fabrication method for stretchable OECT arrays and circuits has limited the achievable computing complexity. Here, we address this limitation through synergistic innovations in material designs and device fabrication processes, enabling large-scale, intrinsically stretchable OECT arrays with a high density of up to 10,000 transistors per cm2. These OECT devices exhibit good synaptic performance in terms of linear, precise, and repeatable programming of conductance states, as well as a good retention time. With high performance uniformity at these integration levels, we have unprecedentedly utilized a stretchable circuit to achieve the hardware implementation of artificial neural network (ANN) for processing health data, including physiological data for heart-attack risk assessment and kernel convolution for locating propagation wavefronts in cardiac ventricular fibrillation. Additionally, we explored the potential of implementing reinforcement learning algorithms for robotic applications.