Using Neural Networks to Deal with Three Phonon Scattering in Phonon Monte Carlo Simulation

The Boltzmann transport equation can well characterize the sub-micron heat transfer, and the motion and interaction of phonons are simulated by the Monte Carlo method. The approximate theory of relaxation time is used to simulate the phonon scattering process, and the energy and momentum conservation of the scattering event is considered through the method of Neural Networks. In this work, the phonon scattering process is calculated first, and all the scattering processes which can satisfy the conservation of energy and momentum are searched and used as training samples. A neural network is trained to quickly search for phonons that conform to the conservation of energy and momentum in a Monte Carlo simulation. The thermal conductivity of bulk silicon obtained by simulation is consistent with the experimental results.