Machine learning accelerated discovery of superconducting two-dimensional Janus transition metal sulfhydrates

Abstract

The MoSH monolayer, one of the Janus transition metal sulfhydrates synthesized by stripping the top-layer S of ${\mathrm{MoS}}_2$ and replacing it with H atoms [Wan et al., ACS Nano 15, 20319 (2021)], has been predicted to host strong coupling two-gap superconductivity with a calculated critical temperature $T_c$ of about 28.58 K at atmospheric pressure. In this work, by using machine learning aided high-throughput calculations, we narrow down 180 possible configurations of two-dimensional Janus transition metal sulfhydrates ($MX\mathrm{H}$ monolayers, where $M=\text{transition}$ metal group elements and $X=\mathrm{S}$, Se, and Te) to 20 stable metals. Among them, we identify six low-energy monolayers that are potential high-$T_c$ superconductors. Notably, the $1T$-TiSH monolayer stands out with the highest $T_c$ of approximately 48 K, surpassing the superconducting properties of $1H$-MoSH ($T_c=28.58$ K) and the well-known ${\mathrm{MgB}}_2$ superconductor ($T_c=39$ K). By solving the anisotropic Migdal-Eliashberg equations, we find that $1T$-TiSH naturally exhibits a one-gap superconducting nature with strong electron-phonon coupling ($\lambda =2.79$) originating from the interactions of Ti $d_{xz,yz}$ orbitals and in-plane vibrations, which is different from and better than the $1H$-MoSH monolayer ($\lambda =1.60$). The presented results enrich families of Janus transition metal sulfhydrates and accelerate the design of novel two-dimensional superconductors.