Machine-Learning-Optimized Aperiodic Superlattice Minimizes Coherent Phonon Heat Conduction

Abstract

Lattice heat conduction can be modulated via nanostructure interfaces. Although advances have been made by viewing phonons as particles, the controllability should be enhanced by fully utilizing their wave nature. By considering phonons as coherent waves, herein we design an optimized aperiodic superlattice that minimizes the coherent phonon heat conduction by alternatingly coupling coherent phonon transport calculations and machine learning. The thermal conductivity of the fabricated aperiodic superlattice agrees well with the calculations over a temperature range of 77-300 K, indicating that complex aperiodic wave interference of coherent phonons can be controlled. The thermal conductivity of the aperiodic superlattice is significantly smaller than the conventional periodic superlattice due to enhanced phonon localization. The optimized aperiodic structure is formed by connecting weakly correlated local structures that introduce interference over broad phonon frequencies. Controlling coherent phonons by aperiodic interferences opens a new route for phonon engineering.