High-throughput screening and machine learning classification of van der Waals dielectrics for 2D nanoelectronics

Abstract

Van der Waals (vdW) dielectrics are promising for enhancing the performance of nanoscale field-effect transistors (FETs) based on two-dimensional (2D) semiconductors due to their clean interfaces. Ideal vdW dielectrics for 2D FETs require high dielectric constants and proper band alignment with 2D semiconductors. However, high-quality dielectrics remain scarce. Here, we employed a topology-scale algorithm to screen vdW materials consisting of zero-dimensional (0D), one-dimensional (1D), and 2D motifs from Materials Project database. High-throughput first-principles calculations yielded bandgaps and dielectric properties of 189 0D, 81 1D and 252 2D vdW materials. Among which, 9 highly promising dielectric candidates are suitable for MoS₂-based FETs. Element prevalence analysis indicates that materials containing strongly electronegative anions and heavy cations are more likely to be promising dielectrics. Moreover, we developed a high-accuracy two-step machine learning (ML) classifier for screening dielectrics. Implementing active learning framework, we successfully identified 49 additional promising vdW dielectrics. This work provides a rich candidate list of vdW dielectrics along with a high-accuracy ML screening model, facilitating future development of 2D FETs.