Modeling and optimization of two-dimensional semiconductor hBCN self-switching diodes

Abstract

In this work, a data-driven modeling and optimization framework is proposed for nanoscale self-switching diodes (SSDs) based on two-dimensional hexagonal boron-carbon-nitrogen (hBCN). Owing to its relatively large bandgap, hBCN is a promising channel material for enhancing rectification performance and suppressing leakage current in SSDs. To reduce the computational cost of conventional quantum-transport-based design approaches, artificial neural networks are employed to develop surrogate models that map key geometrical parameters to critical device characteristics, including forward-bias resistance, reverse-bias conductance, and turn-on voltage. A dataset of 768 distinct device geometries is generated, and three multilayer perceptrons are trained to capture the nonlinear relationships between the geometrical parameters and device characteristics. The developed surrogate models achieve R2>0.99 on the test dataset, demonstrating high predictive accuracy. The trained models are then integrated with a gradient-based optimization procedure to identify an optimal device geometry within the prescribed parameter ranges. The effects of individual geometrical parameters on the current–voltage characteristics and rectification ratio (RR) are also investigated. The results show that the RR is strongly influenced by the channel length, channel width, and leaning angle, whereas the groove thicknesses have only minor effects. The optimized V-shaped hBCN SSD exhibits significantly reduced reverse-bias conductance and lower turn-on voltage compared with graphene, MoS 2, silicene, and silicon nanowire SSDs, resulting in the lowest overall figures of merit among the compared technologies. These results demonstrate the effectiveness of machine-learning-assisted optimization and the potential of hBCN for high-performance SSD application.