Paper Info
Title
PIMGCN: A ReRAM-Based PIM Design for Graph Convolutional Network Acceleration
Abstract
Graph Convolutional Network (GCN) is a promising but computing- and memory-intensive learning model. Processing-in-memory (PIM) architecture based on the ReRAM crossbar is a natural fit for GCN inference. It can reduce the data movements and compute the vector-matrix multiplication (VMM) in analog. However, it requires an unbearable crossbar cost to leverage the massive parallelism exhibited in GCNs. This paper explores the design space for GCN acceleration on ReRAM crossbars and presents the first PIM-based GCN accelerator named PIMGCN. PIMGCN employs dense data mapping and a search-execute architecture to take full advantage of the intra-vertex parallelisms with acceptable crossbars cost. We further propose two scheduling strategies for PIMGCN to maximize the inter-vertex parallelisms and optimize the pipeline. The optimal scheduling is reduced to a maximum independent set problem, which is solved by a novel node-grouping algorithm. Compared to the state-of-the-art software framework running on Intel Xeon CPU and NVIDIA RTX8000 GPU, PIMGCN achieves on average 11044 x and 74.3 x speedup, 6.13E+06 x and 5.09E+03x energy reduction, respectively. Compared with ASIC accelerator HyGCN [1], PIMGCN achieves 219 x speedup and 95.3 x energy reduction.
Year
DOI
Venue
2021
10.1109/DAC18074.2021.9586231
2021 58TH ACM/IEEE DESIGN AUTOMATION CONFERENCE (DAC)
Keywords
DocType
ISSN
GCN, PIM, ReRAM, Accelerator
Conference
0738-100X
Citations 
PageRank 
References 
0
0.34
0
Authors
8
Name
Order
Citations
PageRank
Yang, Tao136.50
Dongyue Li211.37
Yibo Han300.34
Yilong Zhao423.08
Fangxin Liu501.69
Xiaoyao Liang658545.81
Zhezhi He713625.37
Li Jiang828631.86