[24p-31A-1~14] FS.1 Focused Session "AI Electronics"

4:15 PM - 4:30 PM JST(7:15 AM - 7:30 AM UTC)English Presentation

[24p-31A-12] Application of GNN and CNN to CiM-based Accelerators

〇Hanxi Xue¹, Naoko Misawa¹, Chihiro Matsui¹, Ken Takeuchi¹ (1.Univ. Tokyo)

Keywords:

Computation-in-Memory(CiM),Graph Neural Networks(GNN),hardware accelerator

This presentation is about the use of the Computation-in-Memory(CiM) accelerator on the GNN. It analyzes problems and solutions for the CiM-based accelerator's implementation on the GNN considering the inherent physical characteristics, and the circuit construction. Regarding the structure of GNN, what are the characteristics and differences of GNN compared with CNN, and what challenges does it pose to the design of its hardware accelerators. In response to these problems, compared with the common CiM accelerator for CNN, what needs to be considered when designing specifically for GNN will be analyzed in this presentation.