Presentation
CANCELED - Towards Large-Scale Routing: A Novel Learning Based Divide & Merge Approach
DescriptionRectilinear Steiner Minimum Tree (RSMT) is an NP-complete optimization problem that has many important applications for routing in EDA. But most of existing approaches often suffer from the issues like high computational complexity and unstable performance for large-scale input. In this paper, we study this problem from the intersection of the traditional optimization and the emerging machine learning perspectives. We propose a novel learning-based divide and merge framework to solve large-scale routing instance; we develop a graph neural network technique for merging the local regions. Our method can achieve a significant reduction on the total running time.
Event Type
Work-in-Progress Poster
TimeTuesday, July 11th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security