Prediction and optimization of the thermal transport in hybrid carbon-boron nitride honeycombs using machine learning

The recently discovered carbon honeycombs (CHCs) and boron nitride honeycombs (BNHCs) are found to have the similar molecular structures but different thermal properties. Thus, through appropriately patching together CHCs and BNHCs, the hybrid carbon-boron nitride honeycombs (C-BNHCs) with tunable thermal conductivity can be achieved. In this paper, the machine learning (ML) method together with molecular dynamics simulations is employed to study the thermal transport property of C-BNHCs, and also utilized to design the structures of C-BNHCs for the specific thermal conductivity. Our forward learning study reveals a big difference in the thermal conductivities of C-BNHCs with the same BNHC doping level but different doping arrangements. Meanwhile, a nonmonotonic relation is observed between the thermal conductivity of C-BNHCs and their doping concentration, which, according to our analyses of the phonon density of states and spectral thermal conductivity, is attributed to the complicated phonon scattering behaviors in C-BNHCs. In addition, our ML-based method exhibits the high accuracy and efficiency in the inverse design of C-BNHCs with any specific thermal conductivity. Moreover, as for a target thermal conductivity, the present ML-based inverse design method can output numerous potential optimal structures at once for C-BNHCs, which will greatly shorten the design period of C-BNHCs. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This study investigates the tunable thermal conductivity of hybrid carbon-boron nitride honeycombs through machine learning, revealing the influence of doping arrangements and concentrations on thermal transport. The research highlights the potential complexity in the design of C-BNHCs, but demonstrates the efficiency of machine learning methods in speeding up the design process.