GPU-accelerated solution to the transient phonon Boltzmann Transport Equation (BTE) with application to finFETs

We present a GPU-accelerated Discrete Ordinate Method (DOM) based explicit transient phonon Boltzmann Transport Equation (BTE) to model non-diffusive heat transport in nanoscale semiconductor devices. Phonon dispersion relations, group velocities, and scattering rates are obtained from first-principles density functional theory calculations, enabling physically consistent transport simulations. The explicit time-marching approach avoids large linear-system solves and substantially reduces memory requirements while maintaining sub-picosecond temporal resolution. The solver is accelerated on a GPU using OpenACC and applied to transient heat transport in an 8-nm-wide finFET. The simulations capture ultrafast hotspot formation and reveal region-dependent thermal time constants of approximately 50 ps during heating and 50–120 ps during cooling. Sensitivity studies further show that the transient thermal response is strongly influenced by the choice of thermalizing versus reflecting boundary conditions. We discuss the interpretation of these time constants and their dependence on the assumptions. The GPU implementation exhibits near-linear scaling with problem size and significantly lower memory requirements than implicit Krylov-subspace-based approaches, enabling three-dimensional transient simulations involving up to approximately 3.1 billion unknowns on a single 48 GB GPU. These results demonstrate the importance of resolving time-dependent, non-equilibrium phonon transport in advanced transistor geometries and establish explicit GPU-accelerated phonon-BTE solvers as a practical framework for large-scale, high-fidelity transient thermal analysis of next-generation semiconductor devices.