针对忆阻神经网络加速器的混合粒度剪枝方法研究

周博; 陈辉

doi:10.12052/gdutxb.240011

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

针对忆阻神经网络加速器的混合粒度剪枝方法研究

信息与通信工程 | 更新时间：2026-05-27

- 针对忆阻神经网络加速器的混合粒度剪枝方法研究
- Research on Mixed-grained Pruning Method for Memristive Neural Network Accelerator
- 广东工业大学学报 2025年42卷第3期页码：123-129
- 作者机构：
  
  广东工业大学计算机学院,广东,广州,510006
- 作者简介：
  
  周博(1998–) ，男，硕士研究生，主要研究方向为忆阻神经网络加速器，E-mail：bobasyu@163.com
  陈辉(1974–) ，男，副研究员，硕士生导师，主要研究方向为控制工程，E-mail：chenhui02@gdut.edu.cn
- 基金信息：
  
  国家自然科学基金面上项目 (62072118)
- DOI：10.12052/gdutxb.240011
  中图分类号： TP389.1
- 纸质出版：2025
- 稿件说明：
移动端阅览
周博, 陈辉. 针对忆阻神经网络加速器的混合粒度剪枝方法研究[J]. 广东工业大学学报, 2025, 42(3): 123–129.

Zhou Bo, Chen Hui. Research on Mixed-grained Pruning Method for Memristive Neural Network Accelerator[J]. 2025, 42(3): 123-129.
周博, 陈辉. 针对忆阻神经网络加速器的混合粒度剪枝方法研究[J]. 广东工业大学学报, 2025, 42(3): 123–129. DOI： 10.12052/gdutxb.240011

Zhou Bo, Chen Hui. Research on Mixed-grained Pruning Method for Memristive Neural Network Accelerator[J]. 2025, 42(3): 123-129. DOI： 10.12052/gdutxb.240011.

摘要

消除冗余以减少无效计算是加速神经网络和提高计算效率的常用方法。权重剪枝是一种常用的模型压缩方法，其通过去除冗余权重来有效降低计算成本。然而，现有的非结构化剪枝方法没有考虑阻性随机存取存储器 (Resistive Random Access Memory，RRAM) 的忆阻交叉阵列 (Memristive Crossbar Array，MCA) 结构；而结构化剪枝方法虽然契合MCA结构，但是其过粗的剪枝粒度容易造成网络精度的下降。本文提出了一种混合粒度剪枝方法，有效地降低了基于RRAM的忆阻神经网络加速器的硬件开销。该方法将权重子矩阵列根据冗余程度不同进行分类，并执行不同的剪枝策略，充分利用卷积神经网络 (Convolutional Neural Network，CNN) 的冗余性。与现有方法相比，该方法在压缩比和能量效率方面分别提高了2.0倍和1.6倍，并且精度损失更低。

Abstract

Reducing redundant computations is a common method to accelerate neural networks and improve computational efficiency. The weight pruning is an effective model compression technique by removing redundant weights. However

most existing unstructured pruning methods do not consider the Resistive Random Access Memory (RRAM) crossbar structure of the memristors. On the contrary

the structured pruning methods fit well with the Memristive Crossbar Array (MCA) structure but may lead to a decrease in network accuracy due to the coarser pruning granularity. In this paper

we propose a mixed granularity pruning method that can effectively reduce the hardware overhead of the RRAM-based accelerators. The proposed method classifies the weight sub-matrix columns based on different levels of redundancy

and applies different pruning strategies for different columns

which makes full use of the redundancy of Convolutional Neural Networks (CNNs) . Compared to existing methods

the proposed method achieves compression ratio and energy efficiency improvements of approximately 2.0× and 1.6×

respectively

with less accuracy loss.

关键词

Keywords

references

MATSUO Y, LECUN Y, SAHANI M, et al . Deep learning, reinforcement learning, and world models[J ] . Neural Networks, 2022, 152: 267-275.

LI B, WU K J, WANG J, et al . Design of graph convolutional network accelerator based on resistive random access memory[J ] . Journal of Electronics Information Technology, 2023, 4 5(1): 106-115.

LECUN Y, JACKEL L D, BOSER B, et al . Handwritten digit recognition: applications of neural network chips and automatic learning[J ] . IEEE Communications Magazine, 1989, 27(11): 41-46.

SIKHA O K, BHARATH B. Vgg16-random fourier hybrid model for masked face recognition[J]. Soft Computing, 2022, 26(22): 12795-12810.

JOARDAR B K, DOPPA J R, LI H, et al . Realprune: reram crossbar-aware lottery ticket pruning for cnns[J ] . IEEE Transactions on Emerging Topics in Computing, 2022, 11(2): 303-317.

YANG S, HE S, DUAN H, et al . Apq: automated dnn pruning and quantization for reram-based accelerators[J ] . IEEE Transactions on Parallel and Distributed Systems, 2023, 34(9): 2498-2511.

SHIN H, PARK R, LEE S Y, et al . Effective zero compression on reram-based sparse dnn accelerators[C ] //Proceedings of the 59th ACM/IEEE Design Automation Conference. New York: Association for Computing Machinery, 2022: 949-954.

LIU Y, TENG Y L, NIU T, et al . Filter pruning algorithm based on deep reinforcement learning[J ] . Journal of Beijing Univer sity of Posts and Telecommunications, 2023, 46(3): 31-36.

PENG Y, KIM K, WU F, et al . Structured pruning of self-supervised pre-trained models for speech recognition and understanding[C ] //ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . Rhodes Island: IEEE, 2023: 1-5.

PENG J, LIU H, ZHAO Z, et al . Cmq: crossbar-aware neural network mixed-precision quantization via differentiable architecture search[J ] . IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022, 41(11): 4124-4133.

吴昊. 深度学习数据并行训练中低秩分解梯度压缩算法的系统支持[D]. 合肥: 中国科学技术大学, 2023

YANG X, YANG H, DOPPA J R, et al . Essence: exploiting structured stochastic gradient pruning for endurance-aware reram-bassed in-memory training systems[J ] . IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022, 42(7): 2187-2199.

PENG J, LI Z, LIU H, et al . Network pruning towards highly efficient rram accelerator[J ] . IEEE Transactions on Nanotechnology, 2022, 21: 340-351.

CHIOU C Y, LEE K T, HUANG C R, et al . Admm-srnet: alternating direction method of multipliers based sparse representation network for one-class classification[J ] . IEEE Transactions on Image Processing, 2023, 32: 2843-2856.

ZHU Z, SUN H, QIU K, et al . MNSIM 2.0: a behavior-level modeling tool for memristor-based neuromorphic computing systems[J ] . IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems., 2023, 42(11): 4112-4125.

HU M, STRACHAN J P, LI Z, et al . Dot-product engine for neuromorphic computing: programming 1t1m crossbar to accelerate matrix-vector multiplication[C ] //Proceedings of the 53rd annual design automation conference. New York: Association for Computing Machinery. 2016: 1-6.

MURALIMANOHAR N, BALASUBRAMONIAN R, JOIPPI N. Optimizing nuca organizations and wiring alternatives for large caches with CACTI 6.0[C]//40th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO 2007) . Chicago: IEEE, 2007: 3-14.

SCHAK M, GEPPERTH A. Gesture mnist: a new free-hand gesture dataset[C]//International Conference on Artificial Neural Networks. Switzerland, Bristol: Springer Cham, 2022: 657-668.

KRIZHEVSKY A, HINTON G. Learning multiple layers of features from tiny images[J]. Handbook of Systemic Autoimmune Diseases, 2009, 1(4): 1-60.

ZHANG C, ZHANG H, TIAN F, et al . Research on sheep face recognition algorithm based on improved alexnet model[J ] . Neural Computing and Applications, 2023, 35(36): 24971-24979.

CHEN P, HE S, ZHANG X, et al . Accelerating tensor swapping in gpus with self-tuning compression[J ] . IEEE Transactions on Parallel and Distributed Systems, 2022, 33(12): 4484-4498.

CHU C, WANG Y, ZHAO Y, et al . Pim-prune: fine-grain dcnn pruning for crossbar-based process-in-memory architecture[C ] //2020 57th ACM/IEEE Design Automation Conference (DAC) . San Francisco: IEEE, 2020: 1-6.

YU S, LIU Y, ZHANG L, et al . High area/energy efficiency rram cnn accelerator with kernel-reordering weight mapping scheme based on pattern pruning[C ] //2021 IEEE 10th Non-Volatile Memory Systems and Applications Symposium (NVMSA). Beijing: IEEE, 2021: 1-6.

YANG S, CHEN W, ZHANG X, et al . Auto-prune: automated dnn pruning and mapping for reram-based accelerator[C ] //Proceedings of the ACM International Conference on Supercomputing. New York: Association for Computing Machinery, 2021: 304-315.

SHEN Z, WU J, JIANG X, et al . Prap-pim: a weight pattern reusing aware pruning method for reram-based pim dnn accelerators[J ] . High-Confidence Computing, 2023, 3(2): 100123.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于神经网络的HEVC帧内预测组合快速算法

基于分离结果信噪比估计与自适应调频网络的单通道语音分离技术

基于BP神经网络的车辆碳排放测算研究

基于多级残差U-Net的稀疏SPECT图像重建

铁塔航拍图像中鸟巢的YOLOv3识别研究