Reconfigurable neuromorphic computing by a microdroplet

Abstract

The emerging fluidic memristor, capable of emulating ion transport and signaling in brains, has shown promising features in neuromorphic computing but is still in its nascent stage of development. We introduce a droplet memristor in which applied voltage drives a non-conductive liquid crystal droplet to penetrate into a microwell, blocking the ionic conduction path and increasing the resistance. Our system exhibits switchable excitatory and inhibitory features, modulated by altering the polarity of the ionic surfactants at the liquid-liquid interface. We find that memory effects are proportional to the voltage amplitude and inversely proportional to the scanning frequency, consistent with predictions by Newton’s dynamic theory. We emulate adaptive learning akin to biological synapses and demonstrate that low-temperature-induced phase changes in droplets reduce the handwriting recognition accuracy in droplet artificial neuron networks, promising in-sensing computing capabilities. The droplet memristor can benefit from the diverse liquid properties to extend the functionalities and applications in future neuromorphic computing.