Variational Label-Correlation Enhancement for Congestion Prediction

Biao Liu, Congyu Qiao, Ning Xu, Xin Geng, Ziran Zhu, Jun Yang

Abstract

The physical design process of large-scale designs is a time-consuming task, often requiring hours to days to complete, with routing being the most critical and complex step. As the complexity of Integrated Circuits (ICs) increases, there is an increased demand for accurate routing quality prediction. Accurate congestion prediction aids in identifying design flaws early on, thereby accelerating circuit design and conserving resources. Despite the advancements in current congestion prediction methodologies, an essential aspect that has been largely overlooked is the spatial label-correlation between different grids. The spatial label-correlation is a fundamental characteristic of circuit design, where the congestion status of a grid is not isolated but inherently influenced by the conditions of its neighboring grids. In order to fully exploit this inherent spatial label-correlation, we propose a novel approach, VALCE (VAriational Label-Correlation Enhancement), which associates the estimated congestion value of each grid with a local label-correlation weight influenced by its surrounding grids. VALCE leverages variational inference techniques to estimate this weight, thereby enhancing the regression model's performance by incorporating spatial dependencies. Experiment results validate the superior effectiveness of VALCE on the public available ISPD2011 and DAC2012 benchmarks using the superblue circuit line.

Introduction

In Electronic Design Automation (EDA), predicting routing congestion early—specifically during the logic synthesis and placement stages—is a "holy grail" for saving time and resources. If a design is too congested, the routing process might fail or lead to excessive power consumption and delay. Current prediction methods generally fall into two camps: topological (using Graph Neural Networks to look at logical connections) and geometrical (treating the chip layout like an image).

However, a critical gap exists: label-correlation. In a real chip, congestion is not isolated; if one grid cell is highly congested, its neighbors are very likely to be congested too because circuit elements and wires are spatially continuous. Most existing models treat grid cells as independent targets, missing this inherent spatial "smoothness." VALCE addresses this by treating congestion as a continuous spatial field rather than a collection of discrete points.

The VALCE Framework

Figure 1: Overview of the VALCE framework, illustrating the integration of topological and geometrical features through an encoder-decoder architecture to predict congestion maps.

The core innovation of VALCE is the integration of Variational Inference (VI) to estimate the degree of influence neighboring grids have on one another. The framework consists of several key components:

  1. Latent Label-Correlation: The relationship between a grid and its neighbors is modeled as a latent tensor, assuming a Gamma distribution.
  2. Variational Inference (VI): Since calculating exact relationships across all grids is computationally prohibitive, VI approximates the posterior distribution of these correlation weights.
  3. Regularized Risk Estimator: Instead of a standard Mean Squared Error loss, VALCE uses a regularized risk approach that penalizes the model for failing to respect learned spatial correlations.
  4. Hybrid Feature Extraction: The model extracts topological features from a "Circuit Graph" and geometrical features from "RUDY" maps (wire density representations).
  5. Stage Flexibility: The framework can operate even when geometrical data is missing (e.g., during logic synthesis) by relying solely on topological data.

Experimental Evaluation

VALCE was evaluated on industry-standard benchmark suites, ISPD2011 and DAC2012, using the Superblue circuit line. Ground truth congestion maps were generated using NCTU-GR 2.0, a popular global router.

Performance Metrics

VALCE consistently outperformed established baselines like GCN, GraphSAGE, and CircuitGNN across multiple metrics. As shown in the comparative analysis, even the version of VALCE without geometric features significantly outperforms traditional GNN architectures in grid-level Spearman and Kendall correlations.

Figure 6: Performance comparison of VALCE (without geometric features) against various baseline models, demonstrating superior correlation metrics at both cell and grid levels.

Key results include:

  • Grid-Level Accuracy: On the DAC2012 benchmark, VALCE achieved a Spearman correlation of 0.644, nearly doubling the performance of CircuitGNN (0.321).
  • Cell-Level Precision: In the ISPD2011 benchmark, it reached a Pearson correlation of 0.974, effectively identifying specific components in "hot zones."
  • Computational Efficiency: Despite its mathematical complexity, VALCE maintained inference times (~10-11 seconds per epoch on an NVIDIA RTX-3090) comparable to simpler models, making it viable for real-world EDA workflows.

Conclusion

VALCE represents a significant advancement in congestion prediction by explicitly modeling spatial dependencies through variational inference. By treating the correlation between adjacent areas as a learnable latent variable, the model achieves state-of-the-art accuracy. Future work may involve exploring adaptive neighborhood sizes to capture non-square congestion patterns and integrating VALCE directly into placement algorithms to resolve predicted congestion in real-time.

Summary updated

Abstract

The physical design process of large-scale designs is a time-consuming task, often requiring hours to days to complete, with routing being the most critical and complex step. As the the complexity of Integrated Circuits (ICs) increases, there is an increased demand for accurate routing quality prediction. Accurate congestion prediction aids in identifying design flaws early on, thereby accelerating circuit design and conserving resources. Despite the advancements in current congestion prediction methodologies, an essential aspect that has been largely overlooked is the spatial label-correlation between different grids in congestion prediction. The spatial label-correlation is a fundamental characteristic of circuit design, where the congestion status of a grid is not isolated but inherently influenced by the conditions of its neighboring grids. In order to fully exploit the inherent spatial label-correlation between neighboring grids, we propose a novel approach, {\ours}, i.e., VAriational Label-Correlation Enhancement for Congestion Prediction, which considers the local label-correlation in the congestion map, associating the estimated congestion value of each grid with a local label-correlation weight influenced by its surrounding grids. {\ours} leverages variational inference techniques to estimate this weight, thereby enhancing the regression model's performance by incorporating spatial dependencies. Experiment results validate the superior effectiveness of {\ours} on the public available \texttt{ISPD2011} and \texttt{DAC2012} benchmarks using the superblue circuit line.
BibTeX
@inproceedings{liu2024variational,
  author = {Biao Liu and Congyu Qiao and Ning Xu and Xin Geng and Ziran Zhu and Jun Yang},
  title = {{Variational Label-Correlation Enhancement for Congestion Prediction}},
  booktitle = {ASPDAC},
  year = {2024},
  doi = {10.1109/asp-dac58780.2024.10473930},
}
Figure