Emulation of Synaptic Functions with Poly(Ionic Liquid) Heterojunction for Visual Pattern Recognition

Abstract

The progress in artificial intelligence has driven the development of bioinspired iontronics for neuromorphic computing, offering scalability and energy efficiency. Ionic-liquid-based iontronic devices have emerged as capable of emulating the complex functions of neurons and synapses. However, many mechanisms stop at millisecond pulses to trigger ion transport spikes. To open possibilities toward fast in-memory computing, we report a poly(ionic liquid)s (PILs) heterojunction artificial synapse with a sensitive response to submillisecond biases. It exhibits bidirectional modulation driven by voltage-tunable gradual formation and destruction of an ionic depletion layer at the interface. The device-extracted parameters are implemented in an image recognition task using an artificial neural network, resulting in 90% accuracy. It is also successfully applied to perform convolutional neural network inference and convolutional image processing. The solvent independence of PILs facilitates thermal stability while showcasing low energy consumption of 2.96 fJ per spike under 10 μs of 5 mV voltage pulse by leveraging the highly delocalized charge of large ions. This work highlights the reliability of all-ionic artificial synapses as an energy-efficient pattern-recognition hardware.