AiTPO: KAN-UNet Heterogeneous Network for Timing Prediction and Optimization at Global Routing
ACM Transactions on Design Automation of Electronic Systems · 31.05.2026 · 31 (3) · 1-28
DOI: 10.1145/3735639 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACM Transactions on Design Automation of Electronic Systems, online erschienen am 11.02.2026. Verfasst von He Liu und 7 weiteren Autor:innen. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 29 Literaturangaben. Gefördert von: Major Key Project of PCL, Natural Science Foundation of Fujian Province.
Abstract (Originalsprache)
Routing is a critical stage in achieving timing closure in integrated circuit design. Due to the time-consuming flow of detailed routing (DR), the lack of accurate routing information, and the impact of congestion during global routing (GR), rapidly obtaining precise timing information at the global routing stage to guide subsequent timing optimization is a significant challenge. These challenges lead to substantial discrepancies between the estimated timing at GR stage and the actual results after post-DR, resulting in inaccurate evaluations of chip performance. To address this issue, we propose an effective timing prediction and optimization framework, AiTPO. The innovative KAN-UNet heterogeneous timing prediction model effectively combines UNet and KAN networks. By fusing spatial features extracted by UNet with numerical data, the model gains the capability to learn complex relationships across multi-modal data, thereby enhancing robustness and accuracy. Additionally, with the accurate timing evaluation, we introduce two timing optimization strategies during global routing to enhance timing performance. The first strategy involves net ordering based on predicted significant delay nets, prioritizing the routing of more timing-critical nets to reduce detours caused by congestion. The second strategy employs timing estimation to select the most optimal topology from multiple candidates generated by the enhanced A* algorithm, where congestion is considered as a cost factor. Which contributes to optimizing Worst Negative Slack (WNS) and Total Negative Slack (TNS). Experimental results on the real circuits under 28nm process node show that the wire delay prediction accuracy with the proposed KAN-UNet model improves by 34.6% and 25.4% in terms of Mean Absolute Error (MAE) and Max Absolute Error (MaxAE), respectively, compared to GR-based estimations and demonstrate the effectiveness of our timing optimization strategies, which lead to a 2.0% and 4.2% improvement in TNS and WNS, respectively.
Förderung
- Major Key Project of PCL
- Natural Science Foundation of Fujian Province
Zitieren
APA
Liu, H., Zeng, Z., Tao, S., Huang, Z., Li, Y., Xie, B., Gao, W., & Li, X. (2026). AiTPO: KAN-UNet Heterogeneous Network for Timing Prediction and Optimization at Global Routing. ACM Transactions on Design Automation of Electronic Systems, 31(3), 1-28. https://doi.org/10.1145/3735639
Vancouver
Liu H, Zeng Z, Tao S, Huang Z, Li Y, Xie B, et al. AiTPO: KAN-UNet Heterogeneous Network for Timing Prediction and Optimization at Global Routing. ACM Transactions on Design Automation of Electronic Systems. 2026;31(3):1-28. doi:10.1145/3735639
BibTeX
@article{Liu2026,
author = {Liu, He and Zeng, Zhisheng and Tao, Simin and Huang, Zhipeng and Li, Yifan and Xie, Biwei and Gao, Wei and Li, Xingquan},
title = {AiTPO: KAN-UNet Heterogeneous Network for Timing Prediction and Optimization at Global Routing},
journal = {ACM Transactions on Design Automation of Electronic Systems},
year = {2026},
volume = {31},
number = {3},
pages = {1--28},
doi = {10.1145/3735639}
}
EUVIMED