Memristive Organic Bismuth Iodides for Neuromorphic Computing with High Performance

Abstract

We report here synthesis of memristive organic bismuth iodide and its application to neuromorphic computing. AxBiyIx+3y (A = organic monovalent cation) thin films are prepared by spin coating, where structural dimensionality and electrical properties are found to depend on composition. The solid solution synthesized by mixing 2 dimensional (2D) composition with 0 dimensional (0D) one exhibits bipolar switching behavior, where electrical conductivity is significantly lowered as compared to the individual 2D and 0D phase. In addition, activation energy for resistive switching is also substantially lowered by mixing the dimensionality. Memristor is coupled with transistor (1t-1r system), which results in the resistance of 60 MΩ for writing and the operating voltage of 0.6 V alongside linear and symmetric weight update and selectivity. This indicates that the organic bismuth iodide is highly promising for large scale neural network. Furthermore, a crossbar array prepared based on the synthesized organic bismuth iodide shows linear and symmetric weight update. Simulation using MNIST (Modified National Institute of Standards and Technology) data set based on single-layer perceptron model confirms that average of accuracy rate of the synthesized organic bismuth iodide in the crossbar array is close to ideal rate accuracy.