Title: Thermal boundary conductance of sharp metal-diamond interfaces predicted by machine learning molecular dynamics

Abstract: Thermal transport across sharp metal-diamond interfaces plays a critical role in the thermal management of future diamond-based ultrawide bandgap semiconductor devices. However, experimental thermal boundary conductance (TBC) values are mostly nonexistent, and current theoretical models are inaccurate in predicting the TBCs since accurate interatomic potentials of metal-diamond heterostructures are unavailable. In this letter, we show the prediction of TBCs of several practically promising sharp metal-diamond interfaces using nonequilibrium molecular dynamics (NEMD) simulations by developing accurate machine learning interatomic potentials (MLIPs). The predicted TBCs of Al, Mo, Zr, and Au-diamond interfaces are approximately 316, 88, 52, and 55 MW/m2K, respectively, after quantum corrections. The corresponding thermal boundary resistances (TBRs) are equivalent to 0.75-{\mu}m thick of Al, 1.38-{\mu}m Mo, 0.30-{\mu}m Zr and 5.28-{\mu}m Au, respectively. These low TBC values need to be considered in future diamond-based semiconductor designs. We also find that, the conventional simple models such as the acoustic mismatch model (AMM) and diffuse mismatch model (DMM), even including the full band phonon dispersion from first principles, largely mispredict the TBC because they do not include inelastic transmission as well as interfacial structural and bonding information. The quantum correction of TBC matches well with the quantum correction of phonon specific heat of metals, instead of diamond. Additionally, we reveal that the Debye temperature ratio is a better indicator of TBC than the elastic modulus ratio.