Close

Presentation

Functionality Matters in Netlist Representation Learning
TimeTuesday, July 12th11:15am - 11:37am PDT
Location3004, Level 3
Event Type
Research Manuscript
Keywords
RTL/Logic Level and High-level Synthesis
Topics
AI
EDA
DescriptionGate-level netlist representation extracting plays an essential role in many EDA procedures. We argue that existing methods fail to capture underlying semantic information and suffer from generalization issues. To address the problem, we propose a scalable solution that promises to generate generic and comprehensive high-level netlist representations. The proposed flow is based on a novel contrastive scheme that effectively extracts generic functional knowledge about netlists, where a customized Graph Neural Network (GNN) architecture is used as the encoder. Comprehensive experimental results on multiple complex designs demonstrate that our proposed solution significantly outperforms several state-of-the-art netlist representation extracting flows.