Analysis of S defect structures in WS₂ thin films using neural network potential

Abstract

In recent years, two-dimensional transition metal dichalcogenides (TMDCs) have been actively studied because of their potential applications in thermoelectric materials and MISFETs utilizing their excellent mobility and Seebeck coefficient [1, 2]. Among them, WS₂ is known to have relatively high mobility and stability [3] and has attracted much attention. However, the structural disorder of WS₂ thin films, which is considered to have a significant impact on device performance, has not fully been understood yet from experiments, and thus computational calculation to deepen such understanding is desirable. In this study, we aim to clarify the structural disorder in WS₂ thin films due to S defects using the high-dimensional neural network potential (HDNNP) [4], which can achieve high prediction accuracy and feasible computational cost simultaneously.

Molecular dynamics (MD) calculations with the HDNNP and NVT ensemble at the temperatures of 500, 1000 and 1500K were performed for WS₂ slab models. The ratio of S/W was set to 1.7 or 1.9 based on experimental observation. The MD calculations show that the introduced S vacancies were spontaneously transformed into structures consisting of non-six-membered ring structures (in most cases, 5-membered and 9-membered rings), as shown in Figure 1. These structures moved around frequently during the MD calculations, and were also confirmed to be more stable than the S point defect (i.e. atomic vacancy) structures. In the presentation, correlation between neighboring defect structures will also be discussed.

This study was supported by KAKENHI Grant (Nos. 21H05552 and 23H04100), MEXT, Japan.

Reference

[1] G, Kogo, et al., Sci. Rep. 10, 1067 (2020)

[2] T. Hamada, et al., Jpn. J. Appl. Phys. 61 (2022) SC1007

[3] W. Zhang, et al., Nano Res. 7, 1731 (2014)

[4] J. Behler, and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007)