<p>Placement and Routing (PnR), a computationally intensive NP-hard challenge in Very Large-Scale Integration (VLSI) design, is often bottlenecked by the lengthy simulation times of conventional EDA tools. Machine learning-driven congestion prediction provides an effective approach to speed up PnR. Current methods, relying on CNN and GNN architectures, struggle to balance accuracy and efficiency. To address this, we propose ST–FPN, a novel framework integrating Swin Transformer and Feature Pyramid Network (FPN) to enhance congestion prediction. We further enhance congestion-related features, such as Rectangular Uniform wire Density. Built on a streamlined architecture with just 2.74M parameters, <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7641_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="31" /> </InlineMediaObject> <EquationSource Format="TEX">$70 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>70</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> smaller than Swin Transformer UNet, our model improves score by <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7641_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="43" /> </InlineMediaObject> <EquationSource Format="TEX">$72.2 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>72.2</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7641_Article_IEq3.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="43" /> </InlineMediaObject> <EquationSource Format="TEX">$19.3 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>19.3</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> over GAN and RouteNet, respectively. Additionally, the model is extended to CircuitNet-N14 (14nm dataset), achieving a Structural Similarity Index Measure of 0.89, showcasing robust generalization across diverse process nodes.</p>

ST–FPN: a Swin Transformer-based lightweight model for accurate VLSI congestion prediction

期刊论文
发表日期：

2025年7月21日

查看全文

Qiang Cai,
Shizeng Zhang,
Ping Ding,
Xinyu Wu,
Pingyang Huang,
Tianyi Zhang,
Haitao You,
Zhiyuan Cheng,
Qiang Cui

摘要

Placement and Routing (PnR), a computationally intensive NP-hard challenge in Very Large-Scale Integration (VLSI) design, is often bottlenecked by the lengthy simulation times of conventional EDA tools. Machine learning-driven congestion prediction provides an effective approach to speed up PnR. Current methods, relying on CNN and GNN architectures, struggle to balance accuracy and efficiency. To address this, we propose ST–FPN, a novel framework integrating Swin Transformer and Feature Pyramid Network (FPN) to enhance congestion prediction. We further enhance congestion-related features, such as Rectangular Uniform wire Density. Built on a streamlined architecture with just 2.74M parameters, $70 \%$ $70 %$ smaller than Swin Transformer UNet, our model improves score by $72.2 \%$ $72.2 %$ and $19.3 \%$ $19.3 %$ over GAN and RouteNet, respectively. Additionally, the model is extended to CircuitNet-N14 (14nm dataset), achieving a Structural Similarity Index Measure of 0.89, showcasing robust generalization across diverse process nodes.