A 36 mJ/Inf Convolution Accelerator With Reduced Memory Access and Regrouped Sparse Kernels for Environment Sound Classification on Edge Devices

doi:10.1109/TCSII.2025.3585516

	A 36 mJ/Inf Convolution Accelerator With Reduced Memory Access and Regrouped Sparse Kernels for Environment Sound Classification on Edge Devices
	Feng, Lichen 1; Wang, Tao 1; Cai, Rundong 1; Min, Feng 2; Zhu, Zhangming 1
	2025-09-01
发表期刊	IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
ISSN	1549-7747
卷号	72 期号:9 页码:1258-1262
摘要	Efficient environment sound classification (ESC) on edge devices is valuable for applications requiring continuous, long-term monitoring. Existing ESC processors have demonstrated great reductions in latency and resource occupation. However, model sparsity and computation flow still require further optimization. In this brief, we propose an end-to-end ultra-lightweight Depthwise Separable Convolution (DSC) neural network, E2E-ULDSC-Pruned, which is made publicly available as an open-source release. To implement this model, a customized accelerator featuring pipelined DSC computation and regrouped sparse kernels is developed, achieving 36mJ/Inference in ZCU102 FPGA (254ms latency and 143mW power consumption), which is superior to recent works.
关键词	Accuracy Convolution Kernel Computational modeling Shape Feature extraction Timing Frequency modulation Power demand Pipelines Convolution accelerator depthwise separable pipeline environment sound classification sparse kernel
DOI	10.1109/TCSII.2025.3585516
收录类别	SCI
语种	英语
资助项目	Scientific and Technological Innovation 2030 of China[2021ZD0114401] ; National Natural Science Foundation of China[62474128] ; National Natural Science Foundation of China[U22A2013] ; National Natural Science Foundation of China[U24A20291]
WOS研究方向	Engineering
WOS类目	Engineering, Electrical & Electronic
WOS记录号	WOS:001566925200017
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/41727
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Min, Feng; Zhu, Zhangming
作者单位	1.Xidian Univ, Sch Integrated Circuits, Key Lab Analog Integrated Circuits, Xian 710071, Peoples R China 2.Chinese Acad Sci, Inst Comp Technol, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Feng, Lichen,Wang, Tao,Cai, Rundong,et al. A 36 mJ/Inf Convolution Accelerator With Reduced Memory Access and Regrouped Sparse Kernels for Environment Sound Classification on Edge Devices[J]. IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS,2025,72(9):1258-1262.
APA	Feng, Lichen,Wang, Tao,Cai, Rundong,Min, Feng,&Zhu, Zhangming.(2025).A 36 mJ/Inf Convolution Accelerator With Reduced Memory Access and Regrouped Sparse Kernels for Environment Sound Classification on Edge Devices.IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS,72(9),1258-1262.
MLA	Feng, Lichen,et al."A 36 mJ/Inf Convolution Accelerator With Reduced Memory Access and Regrouped Sparse Kernels for Environment Sound Classification on Edge Devices".IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS 72.9(2025):1258-1262.