Swallow: A Versatile Accelerator for Sparse Neural Networks

doi:10.1109/TCAD.2020.2978836

	Swallow: A Versatile Accelerator for Sparse Neural Networks
	Liu, Bosheng 1,2; Chen, Xiaoming 1,2; Han, Yinhe 1,2; Xu, Haobo 1,2
刊名	IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
	2020-12-01
卷号	39 期号:12 页码:4881-4893
关键词	Accelerator convolutional (Conv) layers fully connected (FC) layers sparse neural networks (SNNs)
ISSN号	0278-0070
DOI	10.1109/TCAD.2020.2978836
英文摘要	Sparse neural networks (SNNs) are emerging as a promising technique for resource-limited intelligent embedded systems because of the compact model size and the uncompromised accuracy. Recently, most of the dedicated neural network accelerators are beginning to exploit the sparsity of neural network models for performance boost and energy saving. However, existing sparsity-aware accelerators fail to support both sparse weights and activations in neural networks or support them at the same time for both convolutional (Conv) layers and fully connected (FC) layers, which dominate the computational time of neural networks. In this article, we propose a novel sparsity-aware accelerator architecture, called Swallow, to sufficiently improve the inference performance by eliminating ineffectual weights and activations of neural networks. Swallow comprises: 1) a 2-D systolic architecture that fully utilizes the sparsity of both weights and activations in both Conv and FC layers and 2) a sparsity-aware dataflow which is optimized to reuse both weights and activations and to achieve high processing element (PE) utilization by sparse matrix multiplication tiling. Comprehensive evaluations based on a place-and-route process show that Swallow, with 614 GOP/s peak performance and 1.26-W power, outperforms a state-of-the-art sparsity-aware accelerator Cambricon-X by 1.32x in term of energy efficiency.
资助项目	National Key Research and Development Program of China[2018YFA0701500] ; Key Research Program of Frontier Sciences, CAS[ZDBS-LY-JSC012] ; National Natural Science Foundation of China[61804155] ; National Natural Science Foundation of China[61834006] ; Youth Innovation Promotion Association CAS ; Young Elite Scientists Sponsorship Program by CAST[2018QNRC001] ; Beijing Academy of Artificial Intelligence (BAAI)
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:000592111400045
内容类型	期刊论文
源URL	[http://119.78.100.204/handle/2XEOYT63/16138]
专题	中国科学院计算技术研究所
通讯作者	Chen, Xiaoming
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Ctr Intelligent Comp Syst, State Key Lab Comp Architecture, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Liu, Bosheng,Chen, Xiaoming,Han, Yinhe,et al. Swallow: A Versatile Accelerator for Sparse Neural Networks[J]. IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS,2020,39(12):4881-4893.
APA	Liu, Bosheng,Chen, Xiaoming,Han, Yinhe,&Xu, Haobo.(2020).Swallow: A Versatile Accelerator for Sparse Neural Networks.IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS,39(12),4881-4893.
MLA	Liu, Bosheng,et al."Swallow: A Versatile Accelerator for Sparse Neural Networks".IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS 39.12(2020):4881-4893.