Optimization of interfacial thermal transport in Si/Ge heterostructure driven by machine learning

Heat dissipation through interfaces becomes challenging in nanodevices which impedes the dissipation of waste heat. Accordingly, effective approaches are needed to optimize interfacial thermal transport. In this work, by combining the molecular dynamics simulations and machine learning technique, we systematically study the optimization of interfacial thermal transport in Si/Ge heterostructures through interfacial nanostructuring. Three structural parameters are proposed to describe the nanostructures at interfaces and applied to the machine learning driven predictions. The results demonstrate that the interfacial thermal transport significantly depends on the interfacial nanostructures and diverse guidances are discovered for the optimization. When fixing the density of nanostructures, the interfacial thermal resistance has a minimum at specific heights of nanostructures with small angles, while the minimum is gradually disappeared for the nanostructures with larger angles. When fixing the height of nanostructures, there is also a minimum versus density but gradually disappeared with increasing angles. The nonmonotonic dependences on density and height open spaces for the optimization of interfacial thermal transport. Our spectral decomposition analysis provides physical insights into machine learning predictions and optimizations. Finally, we also summarize the machine learning predictions from the perspective of contact area, in which the distinct dependencies on nanostructuring angle and height manifest the feasibility for the further optimization of interfacial thermal transport. Our machine learning driven study provides comprehensive knowledge and guidances for the optimization of interfacial heat dissipation in nanodevices through nanostructuring.