Exploiting Neural-Network Statistics for Low-Power DNN Inference

IF 2.4 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC IEEE open journal of circuits and systems Pub Date : 2024-04-12 DOI:10.1109/OJCAS.2024.3388210
Lennart Bamberg;Ardalan Najafi;Alberto Garcia-Ortiz
{"title":"Exploiting Neural-Network Statistics for Low-Power DNN Inference","authors":"Lennart Bamberg;Ardalan Najafi;Alberto Garcia-Ortiz","doi":"10.1109/OJCAS.2024.3388210","DOIUrl":null,"url":null,"abstract":"Specialized compute blocks have been developed for efficient nn execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39 %. These power improvements are achieved with no loss of accuracy and negligible hardware cost.","PeriodicalId":93442,"journal":{"name":"IEEE open journal of circuits and systems","volume":null,"pages":null},"PeriodicalIF":2.4000,"publicationDate":"2024-04-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10498075","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE open journal of circuits and systems","FirstCategoryId":"1085","ListUrlMain":"https://ieeexplore.ieee.org/document/10498075/","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0

Abstract

Specialized compute blocks have been developed for efficient nn execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39 %. These power improvements are achieved with no loss of accuracy and negligible hardware cost.
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
利用神经网络统计实现低功耗 DNN 推断
为了实现高效的 nn 执行,我们开发了专门的计算模块。然而,由于大量数据和参数的移动,互连和片上存储器构成了另一个瓶颈,损害了功耗和性能。本研究针对这一瓶颈,为边缘人工智能推理引擎提供了一种低功耗技术,该技术将无开销编码与神经网络数据和参数的统计分析相结合。对于最先进的基准,我们的方法将用于数据存储和移动的逻辑、互连和内存块的功耗降低了 80%,同时为计算块节省了 39% 的额外功耗。在实现这些功耗改进的同时,不会降低精度,硬件成本也可忽略不计。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
自引率
0.00%
发文量
0
审稿时长
19 weeks
期刊最新文献
Double MAC on a Cell: A 22-nm 8T-SRAM-Based Analog In-Memory Accelerator for Binary/Ternary Neural Networks Featuring Split Wordline A Companding Technique to Reduce Peak-to-Average Ratio in Discrete Multitone Wireline Receivers Low-Power On-Chip Energy Harvesting: From Interface Circuits Perspective A 10 GHz Dual-Loop PLL With Active Cycle-Jitter Correction Achieving 12dB Spur and 29% Jitter Reduction A 45Gb/s Analog Multi-Tone Receiver Utilizing a 6-Tap MIMO-FFE in 22nm FDSOI
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1