.png){kind=icon}
.png){kind=icon}
Return
A steady-state temperature solving method for multi-chip modules components based on an improved U-Net
.png){kind=icon}
PRE
AI
DOI：10.1016/j.mejo.2025.106940.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Convolutional neural networks (CNNs) can achieve fast temperature prediction for multi-chip modules (MCMs). However, when the number of chips is large and the layout is complex, the prediction accuracy drops significantly. This study is based on the U-Net architecture, using chip power maps as input and incorporating techniques such as attention mechanisms and residual connections. These modifications improve the model's ability to capture complex relationships between chip power, layout, and temperature fields, enabling fast and accurate predictions. Experimental results show that for MCMs with 4, 8, 12, and 16 chips, the average mean absolute percentage error (MAPE) of steady-state temperature field predictions is below 0.12 %, with the maximum error not exceeding 1.08 K(K). Additionally, for MCMs with 5 or 14 chips, which were not included in the training set, the MAPE does not exceed 0.1 %, and the maximum error is below 1.01 K. The model can accurately predict the MCM temperature field within 1 ms, significantly improving the speed of MCM design.
Keywords:
Multi-chip module
Temperature field prediction
U-Net
Deep learning
Journal
IF:
2.3
Papers:
296
Citations:
4.7K