Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LMM-IR: Large-Scale Netlist-Aware Multimodal Framework for Static IR-Drop Prediction

Kai Ma, Zhen Wang, Hongquan He, Qi Xu, Tinghuan Chen, Hao Geng

TL;DR

This work tackles the costly, chip-wide static IR drop analysis by proposing LMM-IR, a multimodal, end-to-end framework that fuses circuit-level data with large-scale netlist information. It introduces a 3D point-cloud netlist representation processed by a Large-scale Netlist Transformer (LNT) and pairs it with a circuit encoder, enabling effective cross-modal fusion and accurate IR drop prediction across designs with hundreds of thousands to millions of nodes. A two-stage training regime (reconstruction followed by IR drop prediction) further enhances generalization, yielding state-of-the-art F1 and MAE on ICCAD 2023 benchmarks while maintaining competitive inference times. The approach demonstrates the practical impact of integrating netlist topology and circuit context for robust, scalable PDN analysis, with potential extensions leveraging large language models for text-rich PDN data.

Abstract

Static IR drop analysis is a fundamental and critical task in the field of chip design. Nevertheless, this process can be quite time-consuming, potentially requiring several hours. Moreover, addressing IR drop violations frequently demands iterative analysis, thereby causing the computational burden. Therefore, fast and accurate IR drop prediction is vital for reducing the overall time invested in chip design. In this paper, we firstly propose a novel multimodal approach that efficiently processes SPICE files through large-scale netlist transformer (LNT). Our key innovation is representing and processing netlist topology as 3D point cloud representations, enabling efficient handling of netlist with up to hundreds of thousands to millions nodes. All types of data, including netlist files and image data, are encoded into latent space as features and fed into the model for static voltage drop prediction. This enables the integration of data from multiple modalities for complementary predictions. Experimental results demonstrate that our proposed algorithm can achieve the best F1 score and the lowest MAE among the winning teams of the ICCAD 2023 contest and the state-of-the-art algorithms.

LMM-IR: Large-Scale Netlist-Aware Multimodal Framework for Static IR-Drop Prediction

TL;DR

This work tackles the costly, chip-wide static IR drop analysis by proposing LMM-IR, a multimodal, end-to-end framework that fuses circuit-level data with large-scale netlist information. It introduces a 3D point-cloud netlist representation processed by a Large-scale Netlist Transformer (LNT) and pairs it with a circuit encoder, enabling effective cross-modal fusion and accurate IR drop prediction across designs with hundreds of thousands to millions of nodes. A two-stage training regime (reconstruction followed by IR drop prediction) further enhances generalization, yielding state-of-the-art F1 and MAE on ICCAD 2023 benchmarks while maintaining competitive inference times. The approach demonstrates the practical impact of integrating netlist topology and circuit context for robust, scalable PDN analysis, with potential extensions leveraging large language models for text-rich PDN data.

Abstract

Static IR drop analysis is a fundamental and critical task in the field of chip design. Nevertheless, this process can be quite time-consuming, potentially requiring several hours. Moreover, addressing IR drop violations frequently demands iterative analysis, thereby causing the computational burden. Therefore, fast and accurate IR drop prediction is vital for reducing the overall time invested in chip design. In this paper, we firstly propose a novel multimodal approach that efficiently processes SPICE files through large-scale netlist transformer (LNT). Our key innovation is representing and processing netlist topology as 3D point cloud representations, enabling efficient handling of netlist with up to hundreds of thousands to millions nodes. All types of data, including netlist files and image data, are encoded into latent space as features and fed into the model for static voltage drop prediction. This enables the integration of data from multiple modalities for complementary predictions. Experimental results demonstrate that our proposed algorithm can achieve the best F1 score and the lowest MAE among the winning teams of the ICCAD 2023 contest and the state-of-the-art algorithms.

Paper Structure

This paper contains 16 sections, 5 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Circuits Representation
  4. Multimodal Representation Learning
  5. Attention Module in IR Drop Prediction
  6. Problem Formulation
  7. The Proposed IR Drop Prediction Flow
  8. Embedding PDN with Circuit Encoder
  9. Embedding the netlist into point cloud
  10. Encoding netlist point cloud with LNT
  11. Multimodality Fusion Model for Static IR Drop
  12. Experiments
  13. Experiments Settings and Benchmarks
  14. Comparisons with SOTA works
  15. Ablation studies
  16. ...and 1 more sections

Figures (5)

  • Figure 1: The workflow of three models for IR drop prediction. From top to bottom: commercial tool, traditional U-Net, and the proposed LMM-IR model.
  • Figure 2: Our LMM-Net framework.
  • Figure 3: Illustration of ordinary netlist representation vs. our proposed netlist embedding.
  • Figure 4: Ablation studies on ICCAD-2023 contest dataset for different techniques' application.
  • Figure 5: The IR drop prediction visualizations of SOTA works and ours.

Theorems & Definitions (3)

  • Definition 1: F$1$
  • Definition 2: Mean absolute error (MAE)
  • Definition 3: Turn around time (TAT)