Resistive Random Access Memory (RRAM), with its inherent processing-in-memory capability, has emerged as an efficient hardware platform for neural network acceleration.Pruning techniques effectively compress neural networks by removing redundant weights, thereby reducing the hardware cost of RRAM-based accelerators.However, existing RRAM- oriented structured pruning methods often suffer from excessively coarse granularity, which can lead to accuracy degradation.Moreover, they typically neglect the numerical patterns shared among weights, leaving potential redundancy underexploited and limiting further improvements in compression and hardware efficiency.To address these challenges, we propose a pruning method for RRAM neural networks based on Weight Reconstruction.Specifically, an integer scaling weight reconstruction strategy is designed to extract and share common numerical structures among weights, while discarding components with minimal impact on model accuracy.The essential weight information is then mapped onto the RRAM crossbar for network inference, achieving a compact weight representation.Furthermore, a progressive retraining mechanism is introduced, where the discarded components are leveraged as guidance signals that are gradually attenuated to refine the model, thereby recovering accuracy while maintaining high compression and hardware efficiency.Experiments show that, compared to the state-of-the-art works, the proposed method achieves up to 1.2×, 1.2×, and 1.3× improvements in compression rate, area efficiency, and energy efficiency, respectively, with negligible accuracy loss.
我们观察到,在多个典型的神经网络中,滤波器权重之间存在显著的整数倍或近似比例的数值关系,这种结构性数值特征可以用最大公约数(Greatest Common Divisor, GCD)进行描述。图1a展示了几种主流神经网络中满足最大公约数关系的权重所占比例,其中在Alexnet和Plain20网络中,甚至有超过90%的权重可以通过某一公约数映射到统一的比例结构。进一步地,图1b展示了在各个网络中,不同最大公约数值所对应的权重占比情况。这些比例结果表明,权重中存在的数值规律可以被有效利用,以实现更细粒度的结构压缩。
在基于GCD的权重重构过程中,余数 r 被舍弃以压缩存储开销。然而,尽管余数在数值上相对较小,却仍可能包含对推理精度敏感的信息,若直接丢弃,可能导致模型性能退化。因此,本文提出渐进式余数裁剪重训练策略,在重训练初期保留余数作为精度补偿信号,并在训练过程中按比例因子逐步裁剪,从而使网络平稳适应无余数的稀疏权重表示。
KaushalS, TammineniD K, RanaP, et al.Computer vision and deep learning-based approaches for detection of food nutrients/nutrition: New insights and advances[J].Trends Food Sci Technol, 2024, 146: 104408.
[2]
KheddarH, HemisM, HimeurY.Automatic speech recognition using advanced deep learning approaches: A survey[J].Inf Fusion, 2024, 109: 102422.
[3]
ZhangS Y, ZhangZ Z, LiuY L, et al. Dialect speech recognition method based on few-shot learning[J]. Journal of Jiangsu University(Natural Science Edition),2025,46(6): 692-698.
LauriolaI, LavelliA, AiolliF.An introduction to deep learning in natural language processing: Models, techniques, and tools[J].Neurocomputing, 2022, 470: 443-456.
[6]
BaiJ, XueW, FanY, et al.Partial sum quantization for computing-In-memory-based neural network accelerator[J].IEEE Trans Circuits Syst Ⅱ Express Briefs, 2023, 70(8): 3049-3053.
[7]
ChiP, LiS, XuC, et al.PRIME: A novel processing-in-memory architecture for neural network computation in ReRAM-based main memory[C]//2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA).Seoul, Korea: IEEE, 2016: 27-39.
[8]
ShafieeA, NagA, MuralimanoharN, et al.ISAAC: A convolutional neural network accelerator with in situ analog arithmetic in crossbars[C]//2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA).Seoul, Korea:IEEE, 2016: 14-26.
[9]
LiuF, WangZ, ChenY, et al.SoBS-X: Squeeze-out bit sparsity for ReRAM-crossbar-based neural network accelerator[J].IEEE Trans Comput Aided Des Integr Circuits Syst, 2023, 42(1): 204-217.
[10]
JoardarB K, DoppaJ R, LiH, et al.ReaLPrune: ReRAM crossbar-aware lottery ticket pruning for CNNs[J].IEEE Trans Emerg Top Comput, 2023, 11(2): 303-317.
[11]
QuS, LiB, ZhaoS, et al.A coordinated model pruning and mapping framework for RRAM-based DNN accelerators[J].IEEE Trans Comput Aided Des Integr Circuits Syst, 2023, 42(7): 2364-2376.
[12]
MengJ, YangL, PengX, et al.Structured pruning of RRAM crossbars for efficient in-memory computing acceleration of deep neural networks[J].IEEE Trans Circuits Syst Ⅱ Express Briefs, 2021, 68(5): 1576-1580.
[13]
ShenZ, WuJ, JiangX, et al.PRAP-PIM: A weight pattern reusing aware pruning method for ReRAM-based PIM DNN accelerators[J].High Confid Comput, 2023, 3(2): 100123.
[14]
IelminiD, PedrettiG.Resistive switching random-access memory (RRAM): Applications and requirements for memory and computing[J].Chem Rev, 2025, 125(12): 5584-5625.
[15]
ZhuZ, SunH, LinY, et al.A configurable multi-precision CNN computing framework based on single bit RRAM[C]//2019 56th ACM/IEEE Design Automation Conference (DAC).Las Vegas, NV, USA: IEEE, 2019: 1-6.
YangC, LiuH.Channel pruning based on convolutional neural network sensitivity[J].Neurocomputing, 2022, 507: 97-106.
[18]
JiH, SongL, JiangL, et al.ReCom: An efficient resistive accelerator for compressed deep neural networks[C]//2018 Design, Automation & Test in Europe Conference & Exhibition (DATE).Dresden, Germany:IEEE, 2018: 237-240.
[19]
WangP, JiY, HongC, et al.SNrram: An efficient sparse neural network computation architecture based on resistive random-access memory[C]//2018 55th ACM/ESDA/IEEE Design Automation Conference (DAC).San Francisco, USA: IEEE, 2018: 1-6.
[20]
YangT H, ChengH Y, YangC L, et al.Sparse ReRAM engine: Joint exploration of activation and weight sparsity in compressed neural networks[C]//2019 ACM/IEEE 46th Annual International Symposium on Computer Architecture (ISCA).Phoenix, AZ, USA: IEEE, 2019: 236-249.
[21]
YangS, HeS, DuanH, et al.APQ: Automated DNN pruning and quantization for ReRAM-based accelerators[J].IEEE Trans Parallel Distrib Syst, 2023, 34(9): 2498-2511.
[22]
LiuF, ZhaoW, WangZ, et al.ERA-BS: Boosting the efficiency of ReRAM-based PIM accelerator with fine-grained bit-level sparsity[J].IEEE Trans Comput, 2024, 73(9): 2320-2334.
[23]
ZhuZ, SunH, XieT, et al.MNSIM 2.0: A behavior-level modeling tool for processing-In-memory architectures[J].IEEE Trans Comput Aided Des Integr Circuits Syst, 2023, 42(11): 4112-4125.
YuS, ZhangL, WangJ, et al.High area/energy efficiency RRAM CNN accelerator with pattern-pruning-based weight mapping scheme[C]//2021 IEEE 10th Non-Volatile Memory Systems and Applications Symposium (NVMSA).Beijing, China: IEEE, 2021: 1-6.
[26]
YangS, ChenW, ZhangX, et al.AUTO-PRUNE: Automated DNN pruning and mapping for ReRAM-based accelerator[C]//Proceedings of the ACM International Conference on Supercomputing.Virtual Event, USA: ACM, 2021: 304-315.