▎ 摘　　要

Maintaining stability of single-molecular junctions (SMJs) in the presence of current flow is a prerequisite for their potential device applications. However, theoretical understanding of nonequilibrium heat transport in current-carrying SMJs is a challenging problem due to the different kinds of nonlinear interactions involved, including electron-vibration and anharmonic vibrational coupling. Here, we overcome this challenge by accelerating Langevin-type current-induced molecular dynamics using machine-learning potential derived from density functional theory. We show that SMJs with graphene electrodes generate an order of magnitude less heating than those with gold electrodes. This is rooted in the better phonon spectral overlap of graphene with molecular vibrations, rendering harmonic phonon heat transport being dominant. In contrast, in a spectrally mismatched junction with gold electrodes, anharmonic coupling becomes important to transport heat away from the molecule to surrounding electrodes. Our work paves the way for studying current-induced heat transport and energy redistribution in realistic SMJs.