Parallel Computation of the Phonon Boltzmann Transport Equation for Simulation of Frequency Domain Thermo-reflectance (FDTR) Experiments

.The phonon Boltzmann Transport Equation (BTE) is the most appropriate model for extracting thermal properties of semiconductors from state-of-the-art frequency-domain thermo-reflectance (FDTR) experimental data. However, solving the BTE is computationally expensive because of its high-dimensionality, resulting in prohibitive time-to-solution. Scalable parallelization of the BTE solver that can avail supercomputing resources is desirable. We developed an adaptive parallelization strategy for the implicit solution of the multidimensional phonon BTE in the time domain. This adaptive method minimizes load imbalance and unnecessary computations in previously used band and cell-based partitioning schemes, achieved an average speedup of \(1.9×\) compared to the nonadaptive counterpart, and maintained strong scalability at \(25{,}600\) cores. This new solver was used to simulate an FDTR experiment. The results (phase lag between the probe and pulse laser signals) are validated against experimental measurements and show good agreement. Our developed strategies can be applied to other computationally challenging problems constrained by limited parallelism.KeywordsBoltzmann transport equationFDTRnanoscaleparallel computingPDEsemiconductorMSC codes65M0868W10