High-Throughput Manufacturing of Flexible 4K-Memristor Crossbar Circuit

Memristor crossbars offer a promising path towards high-density, low-power neuromorphic computing due to their ability to store large neural network models directly on-chip, minimizing off-chip communication bottlenecks. However, the production of high-density circuits predominantly depends on lithography technology, which necessitates lengthy processing times, ranging from days to several weeks, and a clean-room manufacturing environment. Herein, we demonstrate the high-throughput manufacturing of nonvolatile flexible memristors of the 4K-crossbar circuit based on ultrasonic-assisted spray printing under an open environment. This enables low cost and time efficiency and allows manufacturing without a clean-room environment. The as-manufactured 4K-memristor array demonstrates outstanding functionality with a switching error of less than 5% in a gray-scale 4K pixel image programming under mechanical stretching. A vector-by-matrix multiplication with an average 1 % relative conductance accuracy has also been successfully demonstrated experimentally, confirming the validity of analog-grade electrical features. The MNIST image classification with a large-scale multilayer perceptron classifier was also demonstrated in this paper.