logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

MSDF-Based MAC for Energy-Efficient Neural Networks

delete2025-01-01
delete0
PRE
AI
S
Sahar Moradi Cherati
M
Mohsen Barzegar
L
Leonel Sousa
DOI：10.1109/TVLSI.2025.3539445delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
本文提出了一种基于最高有效位优先（MSDF）方法的高能效串行乘累加（MAC）单元，专门针对在资源受限和低功耗环境下运行的神经网络。所提出的MAC单元已集成到两种神经网络架构中：多层感知器（MLP）和去噪自编码器。对于MLP，我们采用基于Modified National Institute of Standard and Technology（MNIST）数据集的手写数字分类预训练模型。采用45 nm CMOS技术的硬件综合结果表明，基于MSDF的MLP在硬件指标（如减小电路面积和每神经元能耗）方面取得了良好的权衡，并达到了97.4%的良好分类准确率。在自编码器的情况下，所提出的MAC单元被用于同一数据集的预训练去噪自编码器。该自编码器采用基于MSDF的混合精度跨层设计，以减少计算资源及负载，同时保持高质量图像重建。对图像正确性、信号保真度及视觉质量指标进行了全面评估，以评估不同层中不同精度水平的影响。结果表明，所提出的方法在实现有效去噪性能的同时，满足了均衡的硬件要求。
Keyword:
Computer arithmetic
denoising
energy efficiency
most significant digit first (MSDF) arithmetic
multiply-accumulate (MAC)
neural networks

期刊

I
IEEE Transactions on Very Large Scale Integration (VLSI) Systems
IF:
3.1
论文数:
505
被引数:
7.3K

机构

I
inesc-id, lisbon, portugal
学者数:
2
论文数: 1
被引数: 0
Instituto Superior Técnico 封面图
Instituto Superior Técnico
学者数:
238
论文数: 117
被引数: 2.5K
引用论文

引用论文

Asymmetric Autoencoders: An NN alternative for resource-constrained devices in IoT networks
errAd Hoc Networks
IF0
err2024-04-01
err0
PREAI
errGilbert,Mateus S.; de Campos,Marcello L.R.; Campista,Miguel Elias M.
err分享
err收藏
Stripes: Bit-serial deep neural network computing
err2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO)
IF0
err2016-10-01
err0
PREAI
errPatrick Judd; Jorge Albericio; Tayler Hetherington; Tor M. Aamodt; Andreas Moshovos
err分享
err收藏
Model-driven Cluster Resource Management for AI Workloads in Edge Clouds
errACM Transactions on Autonomous and Adaptive Systems
IF0
err2023-03-27
err0
errOAAI
errQianlin Liang; Walid A. Hanafy; Ahmed Ali-Eldin; Prashant Shenoy
err分享
err收藏
A design methodology for networks of online modules and its application to the Levinson-Durbin algorithm网络在线模块的设计方法及其在Levinson-Durbin算法中的应用
errIEEE Transactions on Very Large Scale Integration (VLSI) Systems
IF0
err2004-01-01
err0
PREAI
errR. Galli; A.F. Tenca
err分享
err收藏
An Efficient Design of 16 Bit MAC Unit using Vedic Mathematics
err2019 International Conference on Communication and Signal Processing (ICCSP)
IF0
err2019-04-01
err0
PREAI
errAkella Srinivasa Krishna Vamsi; S R Ramesh
err分享
err收藏
Ultra low-voltage low-power CMOS 4-2 and 5-2 compressors for fast arithmetic circuits
errIEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
IF5.2
err2004-10-01
err258
errOAAI
errChang, CH; Gu, JM; Zhang, MY
err分享
err收藏
A Power-Aware Digital Multilayer Perceptron Accelerator with On-Chip Training Based on Approximate Computing
errIEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTING
IF5.4
err2017-04-01
err35
errOAAI
errKim, Duckhwan; Kung, Jaeha; Mukhopadhyay, Saibal
err分享
err收藏
Custom Hardware Architectures for Deep Learning on Portable Devices: A Review
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2022-11-01
err36
PREAI
errZaman, Kh Shahriya; Reaz, Mamun Bin Ibne; Ali, Sawal Hamid Md; Bakar, Ahmad Ashrif A.; Chowdhury, Muhammad Enamul Hoque
err分享
err收藏
Most Significant Digit First Multiply-and-Accumulate Unit for Neural Networks
err2024 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)
IF0
err2024-07-01
err0
PREAI
errCherati,Sahar Moradi; Barzegar,Mohsen; Sousa,Leonel
err分享
err收藏
Image Quality Assessment through FSIM, SSIM, MSE and PSNR—A Comparative Study
errJournal of Computer and Communications
IF0
err2019-01-01
err0
errOAAI
errUmme Sara; Morium Akter; Mohammad Shorif Uddin
err分享
err收藏
学者 查看更多内容