multiscale Fourier-BTE coupling

DeepThermo builds GPU-native, differentiable thermal physics software for microelectronics, spun out of the University of Notre Dame. Its flagship is a GPU-native multiscale non-Fourier thermal model spanning transistor to chip, differentiable end-to-end for gradient-based design optimization. It combines a differentiable phonon Boltzmann Transport Equation solver, neural-field surrogates, and multiscale Fourier-BTE coupling.

Read more

automatic differentiation end-to-endGPU-native FVM solver for phonon Boltzmann Transport Equationinverse designoperator-learning neural surrogatesthermal simulation of transistors and chips