ADS-CNN: Adaptive Dataflow Scheduling for lightweight CNN accelerator on FPGAs

doi:10.1016/j.future.2024.04.038

	ADS-CNN: Adaptive Dataflow Scheduling for lightweight CNN accelerator on FPGAs
	Wan, Yi 1; Xie, Xianzhong 1; Chen, Junfan 2; Xie, Kunpeng 3; Yi, Dezhi 4; Lu, Ye 4,5,6,7; Gai, Keke 2,8
	2024-09-01
发表期刊	FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
ISSN	0167-739X
卷号	158 页码:138-149
摘要	Lightweight convolutional neural networks (CNNs) enable lower inference latency and data traffic, facilitating deployment on resource -constrained edge devices such as field -programmable gate arrays (FPGAs). However, CNNs inference requires access to off -chip synchronous dynamic random-access memory (SDRAM), which significantly degrades inference speed and system power efficiency. In this paper, we propose an adaptive dataflow scheduling method for lightweight CNN accelerator on FPGAs named ADS -CNN. The key idea of ADS -CNN is to efficiently utilize on -chip resources and reduce the amount of SDRAM access. To achieve the reuse of logical resources, we design a time division multiplexing calculation engine to be integrated in ADS -CNN. We implement a configurable module for the convolution controller to adapt to the data reuse of different convolution layers, thus reducing the off -chip access. Furthermore, we exploit on -chip memory blocks as buffers based on the configuration of different layers in lightweight CNNs. On the resource -constrained Intel CycloneV SoC 5CSEBA6 FPGA platform, we evaluated six common lightweight CNN models to demonstrate the performance advantages of ADS -CNN. The evaluation results indicate that, compared with accelerators that use traditional tiling strategy dataflow, our ADS -CNN can achieve up to 1.29 x speedup with the overall dataflow scale compression of 23.7%.
关键词	Lightweight convolutional neural networks FPGA Accelerator Adaptive dataflow Unified computing engine Tiling strategy
DOI	10.1016/j.future.2024.04.038
收录类别	SCI
语种	英语
资助项目	Special Key Project of Technological Innovation and Application Development of Chongqing, China[CSTB2022TIAD-KPX0057] ; National Natural Science Foundation, China[62372253] ; National Natural Science Foundation, China[62002175] ; Natural Science Foundation of Tianjin Fund, China[23JCYBJC00010] ; CCF-Baidu Open Fund, China[CCF-Baidu202310] ; Open Project Fund of State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences[CARCHB202016]
WOS研究方向	Computer Science
WOS类目	Computer Science, Theory & Methods
WOS记录号	WOS:001235188200001
出版者	ELSEVIER
引用统计	被引频次：1[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/40053
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Lu, Ye; Gai, Keke
作者单位	1.Chongqing Univ Posts & Telecommun, Coll Comp Sci & Technol, Chongqing 400065, Peoples R China 2.Chongqing Haiyunjiexun Technol Co Ltd, Chongqing, Peoples R China 3.Nankai Univ, Coll Comp Sci, Tianjin 300350, Peoples R China 4.Nankai Univ, Coll Cyber Sci, Tianjin 300350, Peoples R China 5.Tianjin Key Lab Network & Data Secur Technol, Tianjin, Peoples R China 6.Chinese Acad Sci, ICT, State Key Lab Processors, Beijing, Peoples R China 7.Minist Educ, Key Lab Data & Intelligent Syst Secur, Tianjin, Peoples R China 8.BIT, Sch Cyberspace Sci & Technol, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Wan, Yi,Xie, Xianzhong,Chen, Junfan,et al. ADS-CNN: Adaptive Dataflow Scheduling for lightweight CNN accelerator on FPGAs[J]. FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE,2024,158:138-149.
APA	Wan, Yi.,Xie, Xianzhong.,Chen, Junfan.,Xie, Kunpeng.,Yi, Dezhi.,...&Gai, Keke.(2024).ADS-CNN: Adaptive Dataflow Scheduling for lightweight CNN accelerator on FPGAs.FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE,158,138-149.
MLA	Wan, Yi,et al."ADS-CNN: Adaptive Dataflow Scheduling for lightweight CNN accelerator on FPGAs".FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE 158(2024):138-149.