TAC-UNet: A Partially Dense UNet for Thermal Analysis of Chiplet-Based System

In this paper, a novel deep learning framework called TAC-UNet is proposed for the Thermal Analysis of Chiplet-based systems. To enhance solution accuracy, the down-sampling and up-sampling modules in the TAC-UNet are designed to be partially dense. We compare the performance of TAC-UNet with other networks. The results demonstrate that the TAC-UNet can generate the temperature distribution in an extremely short inference time, with an average absolute error of 0.38 K.