Optoelectronic Nanofluidic Neural Networks for Ionic Computing

Emerging neuromorphic computing technologies aim to replicate the brain's cognitive processes and offer promising capabilities for advanced computation and complex tasks. Building artificial neural networks (ANNs) with robust connections and dynamic plasticity is essential for their implementation. Nevertheless, most existing ANNs rely on solid-state semiconductors, where electrons or holes serve as information carriers. Here, we present a novel ion-based nanofluidic memristor operating in aqueous solutions, which can be integrated to create densely connected ionic neural networks (INNs). The nanofluidic memristor exhibits tunable ionic memory and diverse synaptic plasticity that can be modulated by electrical, optical, or combination signals. We demonstrate that the ionic memory originates from non-trivial hysteretic ion-charged surface interactions in confined nanochannels, supported by experiments and numerical calculations. Utilizing the tunable states of the nanofluidic memristor, we implement reservoir computing to classify static patterns and dynamical motions, achieving ultrahigh accuracies of 91% and 97% on the standard MNIST datasets and homemade moving-particle library, respectively. Multiple nanofluidic memristors are integrated to construct a practical INN capable of real-time logic computation and in-sensor computing, verifying the robustness of nanofluidic computation systems. This ion-based nanofluidic platform offers a promising pathway towards fully connected INNs and advances neuromorphic computing beyond solid-state electronics.