English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Graph Neural Networks (GNNs) are becoming a promising technique in variousdomains due to their excellent capabilities in modeling non-Euclidean data.Although a spectrum of accelerators has been proposed to accelerate theinference of GNNs, our analysis demonstrates that the latency and energyconsumption induced by DRAM access still significantly impedes the improvementof performance and energy efficiency. To address this issue, we propose aMemory-Efficient GNN Accelerator (MEGA) through algorithm and hardwareco-design in this work. Specifically, at the algorithm level, through anin-depth analysis of the node property, we observe that the data-independentquantization in previous works is not optimal in terms of accuracy and memoryefficiency. This motivates us to propose the Degree-Aware mixed-precisionquantization method, in which a proper bitwidth is learned and allocated to anode according to its in-degree to compress GNNs as much as possible whilemaintaining accuracy. At the hardware level, we employ a heterogeneousarchitecture design in which the aggregation and combination phases areimplemented separately with different dataflows. In order to boost theperformance and energy efficiency, we also present an Adaptive-Package formatto alleviate the storage overhead caused by the fine-grained bitwidth anddiverse sparsity, and a Condense-Edge scheduling method to enhance the datalocality and further alleviate the access irregularity induced by the extremelysparse adjacency matrix in the graph. We implement our MEGA accelerator in a28nm technology node. Extensive experiments demonstrate that MEGA can achievean average speedup of 38.3x, 7.1x, 4.0x, 3.6x and 47.6x, 7.2x, 5.4x, 4.5xenergy savings over four state-of-the-art GNN accelerators, HyGCN, GCNAX, GROW,and SGCN, respectively, while retaining task accuracy.

MEGA: A Memory-Efficient GNN Accelerator Exploiting Degree-Aware Mixed-Precision Quantization