ThermAl uses a hybrid U-Net with positional encoding and Boltzmann regularization to predict circuit thermal maps from activity profiles at 0.71°C RMSE while running up to 200 times faster than FEM.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
2D-ThermAl: Physics-Informed Framework for Thermal Analysis of Circuits using Generative AI
ThermAl uses a hybrid U-Net with positional encoding and Boltzmann regularization to predict circuit thermal maps from activity profiles at 0.71°C RMSE while running up to 200 times faster than FEM.