Deep Learning-Based Early-Stage IR-Drop Estimation via CNN Surrogate Modeling

TL;DR

The paper tackles the need for fast early-stage IR-drop estimation in VLSI by presenting a CNN surrogate that predicts dense IR-drop heatmaps from three spatial layout feature maps. Formulated as pixel-wise regression, the approach uses a U-Net to capture multi-scale spatial patterns, trained on a physics-inspired synthetic dataset to learn realistic IR-drop behavior without costly signoff simulations during inference. Results show millisecond inference with high spatial fidelity (e.g., $MSE \approx 4.9 \times 10^{-4}$ and $PSNR \approx 33.3$ dB), enabling rapid hotspot identification and design exploration while remaining a complementary tool to signoff analyses. The framework is deployed via an interactive Streamlit app and released with dataset-generation scripts, illustrating practical impact for fast pre-signoff screening and iterative optimization.

Abstract

IR-drop is a critical power integrity challenge in modern VLSI designs that can cause timing degradation, reliability issues, and functional failures if not detected early in the design flow. Conventional IR-drop analysis relies on physics-based signoff tools, which provide high accuracy but incur significant computational cost and require near-final layout information, making them unsuitable for rapid early-stage design exploration. In this work, we propose a deep learning-based surrogate modeling approach for early-stage IR-drop estimation using a CNN. The task is formulated as a dense pixel-wise regression problem, where spatial physical layout features are mapped directly to IR-drop heatmaps. A U-Net-based encoder-decoder architecture with skip connections is employed to effectively capture both local and global spatial dependencies within the layout. The model is trained on a physics-inspired synthetic dataset generated by us, which incorporates key physical factors including power grid structure, cell density distribution, and switching activity. Model performance is evaluated using standard regression metrics such as Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). Experimental results demonstrate that the proposed approach can accurately predict IR-drop distributions with millisecond-level inference time, enabling fast pre-signoff screening and iterative design optimization. The proposed framework is intended as a complementary early-stage analysis tool, providing designers with rapid IR-drop insight prior to expensive signoff analysis. The implementation, dataset generation scripts, and the interactive inference application are publicly available at: https://github.com/riteshbhadana/IR-Drop-Predictor. The live application can be accessed at: https://ir-drop-predictor.streamlit.app/.

Figures (3)

• Figure 1: Three spatial feature maps capturing the key physical factors influencing IR-drop distribution: power grid strength (left), cell density (middle), and switching activity (right).
• Figure 2: U-Net-based CNN for IR-drop heatmap prediction. The model performs pixel-wise regression from three physical layout feature maps to a continuous IR-drop distribution using an encoder-decoder architecture with skip connections.
• Figure 3: Predicted IR-drop heatmap with spatial distribution of voltage drops and identified hotspot regions. The visualization shows maximum IR-drop of 1.0904, average IR-drop of 0.1281, with 17 hotspots identified, resulting in a HIGH RISK classification.