Post-Routing Path Statistical Delay Estimation Based on SGAT-GRU Prediction Framework

Due to the absence of actual interconnect parasitic parameters in pre-routing timing analysis tools, timing analysis results exhibit significant errors. This necessitates extensive Statistical Static Timing Analysis (SSTA) and repeated routing iterations, substantially prolonging the design cycle. Therefore, a fast and accurate prediction framework is urgently needed to forecast post-routing path statistical delays before the routing stage. Although existing models utilize Graph Neural Networks (GNN) to accelerate timing analysis, they suffer from limitations such as restricted receptive fields and feature smoothing. To address these challenges, this paper proposes SGAT-GRU – a rapid and accurate framework for post-routing path statistical delay prediction. The framework employs a Self-supervised Graph Attention Network (SGAT) and Gated Recurrent Unit (GRU) to embed local and global information respectively. The resulting cell embeddings enhance prediction accuracy of path statistical delays. These embeddings are fed into two Multi-Layer Perceptrons (MLPs) to obtain path mean delay and path standard deviation delay, which are then combined to derive the statistical path delay. Ablation experiments demonstrate the superiority of the SGAT-GRU framework. Comparative results with existing models confirm that our approach achieves high-precision and efficient prediction: SGAT-GRU achieves an MAE range of 0.02–3.77 ps and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> range of 0.93–0.99, with speed improvements of 277.48× to 1536.68× compared to conventional commercial tools.