Learning-Effective Mixed-Dimensional Halide Perovskite QD Synaptic Array for Self-Rectifying and Luminous Artificial Neural Networks

Abstract

A mixed-dimensional heterostructure comprising nanomaterials with varying dimensions provides a promising structure for an artificial synapse for reconfigurable neuromorphic functions. In this study, an 8 x 8 memristor crossbar array based on a mixed-dimensional heterostructure comprising Cs(1-x)FA(x)PbBr(3) (0.00 <= x <= 0.15) quantum dots (QDs) and different dimensional interfacial nanomaterial layers between the Al and ITO electrodes is designed and fabricated. This array device exhibits a high yield and reliable self-rectifying analog switching characteristics with low synaptic-coupling (SC, up to 5.19 x 10(-5)) and light emission, facilitating stimuli response visualization and preventing undesired pathways in the network array. Furthermore, because the formamidinium (FA) concentration alters the QD size, thereby engineering interfacial band alignment in the heterostructure, the essential synaptic properties such as dynamic range, SC, and nonlinearity can be improved. Especially, as x increases from 0 to 0.11, the recognition accuracy for the MNIST patterns increases significantly, from 68.97% to 89.08%, even for single-layer ANNs. The energy consumption required for a specific accuracy level is reduced by a factor of 25.15. The utilization of mixed-dimensional perovskite QD-based heterostructures in neural networks may provide desirable neuromorphic electronic functions with enhanced learning capability and energy efficiency, while preventing unwanted neural signals.