Crystallinity-Controlled Hexagonal Boron Nitride-Based Memristors for Fully Integrated Reservoir Computing Processors.

Memristors based on 2D materials (2DMs) have attracted considerable research interest due to their excellent switching performance. Former synthesis methods for 2DMs aimed to synthesize 2DMs with a large grain size. However, these methods cause a stochastic distribution of defects in high-density memristor arrays, resulting in device nonuniformity. Moreover, high synthesis temperatures and mechanical transfer make it difficult to implement large-area memristor arrays and additional integration. Therefore, synthesis methods of nanocrystalline 2DMs for memristors are essential. In this study, crystallinity-controlled hexagonal boron nitride is directly synthesized on metal electrodes, and a fully integrated memristor-based reservoir computing processor is implemented. Memristors using nanocrystalline hexagonal boron nitride (NC h-BN) exhibit volatile switching and reliable reservoir dynamics. Memristors using amorphous boron nitride (a-BN) exhibit nonvolatile switching and linear potentiation/depression curves (α = -0.475, α = 0.656). By integrating NC h-BN and a-BN memristors in three dimensions, an efficient reservoir computing processor with integrated reservoir and readout layers is realized. Overall, the neural network of the integrated processor shows high accuracy for inferring temporal data. Hence, the crystallinity-controlled hexagonal boron nitride synthesis method paves the way for the realization of fully integrated reservoir computing processors.