| Summary: | Interfacial heat transport is an increasingly determinant factor controlling the dissipation of heat in many modern electronic, optical, and magnetic devices – driven by their continued miniaturization and corresponding increase in interfacial density. Molecular dynamics (MD) simulations are a highly versatile theoretical tool that can be used to predict, understand, and rationalize interfacial heat transport. However, significant improvements are needed to make MD-based methods accurate enough to predict thermal boundary conductance (TBC) values – the primary quantity characterizing interfacial heat transport – that are consistent with experimental values. In this work, I investigate ways of improving MD simulations using three example interface systems: Ge-GaAs, Al-Al2O3, and AlN-GaN.
First, accurate interatomic potentials are generated for each system using either pure machine-learned interatomic potentials (MLIPs) or a hybrid approach in which MLIPs are combined with a Taylor-expansion potential. Key tradeoffs between these approaches are assessed, with a particular focus on speed, stability and vibrational accuracy. Next, I explored how key choices in the setup and analysis of TBC calculations can affect the comparison to experimental values. From a Landauer perspective, the results show that a 4-probe definition of TBC can exceed the maximum transmission limit, consistent with new experimental measurements of the Ge-GaAs TBC. After, discussing the advantages and disadvantages of using nonequilibrium or equilibrium molecular dynamics (NEMD vs. EMD) to predict TBC, a new protocol is presented that attempts to mitigate the effects of noise by analyzing EMD data in a mode-specific fashion. While providing qualitatively better TBC results, the protocol unfortunately exhibits significant parameter sensitivity, at least with the current data.
Using this protocol and the MLIPs developed in this work, I then predict temperature-dependent TBC values for Al-Al2O3 and Ge-GaAs interfaces, and compare them to experimental measurements. For Al-Al2O3, significant effort is made to establish a correspondence between the modelled atomistic structure and the experimental sample, resulting in a plausible O-terminated and Al-terminated structure. Surprisingly, the predicted TBC of these two structures differ significantly, with correspondingly different behavior in their mode-level contributions. Unfortunately, neither set of TBC results exhibit good agreement with experimental measurements, nor do the TBC predictions for Ge-GaAs, which are hypothesized to be impacted by finite-size effects.
Finally, an applied case is explored whereby 15N/14N isotopic disorder was introduced in an AlN-GaN interface in an attempt to enhance TBC, motivated by a prior work. Using a latticedynamics-based descriptor, it is shown that isotopic disorder does enhance “mode overlap” between different sides of the interface, which in concept can enhance interfacial heat flow. However, by performing NEMD calculations with the accurate MLIP, and paying close attention to how the temperature drop is extracted, the results demonstrate that no TBC enhancement occurs and that rather, TBC deteriorates with the addition of isotopic disorder in AlN-GaN. The significant difference between this result and that of the prior work may be partially attributed to the use of a more accurate potential.
|
|---|