Materials informatics approach for design of Si/Ge layered nanostructures with low thermal conductivity

We report an efficient method based on the materials informatics approach to predict low thermal conductivity structures using a descriptor obtained by phonon mode calculations. For a small number of Si/Ge layered structures, we perform normal mode analysis to calculate the participation ratio for each phonon mode and calculations of thermal conductivity in the stacking direction using the perturbed molecular dynamics (MD) method. The descriptor for the thermal conductivity is defined using the participation ratios averaged in the acoustic phonon frequency ranges with their lower and upper limits independently optimized. By employing the descriptor and genetic algorithm, low thermal conductivity structures are recommended among a huge number of possible structures. The recommended structures are confirmed to have very small thermal conductivities from the results of the perturbed MD calculations. By employing the method, we can design Si/Ge layered structures with low thermal conductivity at very low computational cost.