Abstract
Brain-inspired neuromorphic computing has emerged as a promising solution to overcome the energy and speed limitations of conventional von Neumann architectures. In this context, in-memory computing utilizing memristors has gained attention as a key technology, harnessing their non-volatile characteristics to replicate synaptic behavior akin to the human brain. However, challenges arise from non-linearities, asymmetries, and device variations in memristive devices during synaptic weight updates, leading to inaccurate weight adjustments and diminished recognition accuracy. Moreover, the repetitive weight updates pose endurance challenges for these devices, adversely affecting latency and energy consumption. To address these issues, we propose a Siamese network learning approach to optimize the training of multi-level memristor neural networks. During neural inference, forward propagation takes place within the memristor neural network, enabling error and noise detection in the memristive devices and hardware circuits. Simultaneously, high-precision gradient computation occurs on the software side, initially updating the floating-point weights within the Siamese network with gradients. Subsequently, weight quantization is performed, and the memristor conductance values requiring updates are modified using a sparse update strategy. Additionally, we introduce gradient accumulation and weight quantization error compensation to further enhance network performance. The experimental results of MNIST data recognition, whether based on a MLP or a CNN model, demonstrate the rapid convergence of our network model. Moreover, our method successfully eliminates over 98% of weight updates for memristor conductance weights within a single epoch. This substantial reduction in weight updates leads to a significant decrease in energy consumption and time delay by more than 98% when compared to the basic closed-loop update method. Consequently, this approach effectively addresses the durability requirements of memristive devices.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Data availability
The data used in this research work are available from the authors by reasonably request.
References
Afshari S, Musisi-Nkambwe M, Esqueda IS (2022) Analyzing the impact of memristor variability on crossbar implementation of regression algorithms with smart weight update pulsing techniques. IEEE Trans Circuits Syst I Regul Pap 69(5):2025–2034. https://doi.org/10.1109/tcsi.2022.3144240
Bao B, Hu J, Bao H et al (2023) Memristor-coupled dual-neuron mapping model: initials-induced coexisting firing patterns and synchronization activities. Cognit Neurodyn. https://doi.org/10.1007/s11571-023-10006-8
Dong S, Chen Y, Fan Z et al (2022) A backpropagation with gradient accumulation algorithm capable of tolerating memristor non-idealities for training memristive neural networks. Neurocomputing 494:89–103
Dong X, Xu C, Xie Y et al (2012) Nvsim: a circuit-level performance, energy, and area model for emerging nonvolatile memory. IEEE Trans Comput Aided Des Integr Circuits Syst 31(7):994–1007. https://doi.org/10.1109/TCAD.2012.2185930
Fu J, Liao Z, Gong N et al (2019) Mitigating nonlinear effect of memristive synaptic device for neuromorphic computing. IEEE J Emerg Sel Top Circuits Syst 9(2):377–387. https://doi.org/10.1109/JETCAS.2019.2910749
Fu J, Liao Z, Wang J (2022) Level scaling and pulse regulating to mitigate the impact of the cycle-to-cycle variation in memristor-based edge AI system. IEEE Trans Electron Devices 69(4):1752–1762
Guan J, Liang G (2023) A research of convolutional neural network model deployment in low-to medium-performance microcontrollers. In: Proceedings of the 2023 10th international conference on wireless communication and sensor networks. ACM, pp 44–50. https://doi.org/10.1145/3585967.3585975
Guo M, Sun Y, Zhu Y et al (2023) Pruning and quantization algorithm with applications in memristor-based convolutional neural network. Cognit Neurodyn. https://doi.org/10.1007/s11571-022-09927-7
Horowitz M (2014) 1.1 computing’s energy problem (and what we can do about it). In: 2014 IEEE international solid-state circuits conference digest of technical papers (ISSCC). IEEE, pp 10–14. https://doi.org/10.1109/ISSCC.2014.6757323
Jacob B, Kligys S, Chen B, et al (2018) Quantization and training of neural networks for efficient integer-arithmetic-only inference. In: Proceedings of the IEEE conference on computer vision and pattern recognition. IEEE, pp 2704–2713
Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. arXiv preprint arXiv:1412.6980
Krestinskaya O, Salama KN, James AP (2018) Learning in memristive neural network architectures using analog backpropagation circuits. IEEE Trans Circuits Syst I Regul Pap 66(2):719–732. https://doi.org/10.1109/TCSI.2018.2866510
Kwon D, Lim S, Bae JH et al (2020) On-chip training spiking neural networks using approximated backpropagation with analog synaptic devices. Front Neurosci 14:423. https://doi.org/10.3389/fnins.2020.00423
Li C, Belkin D, Li Y et al (2018) Efficient and self-adaptive in-situ learning in multilayer memristor neural networks. Nat Commun 9(1):1–8. https://doi.org/10.1038/s41467-018-04484-2
Li C, Hu M, Li Y et al (2018) Analogue signal and image processing with large memristor crossbars. Nat Electron 1(1):52–59. https://doi.org/10.1038/s41928-017-0002-z
Li J, Zhou G, Li Y et al (2022) Reduction 93.7% time and power consumption using a memristor-based imprecise gradient update algorithm. Artif Intell Rev 55(1):657–677. https://doi.org/10.1007/s10462-021-10060-w
Li Y, Ang KW (2021) Hardware implementation of neuromorphic computing using large-scale memristor crossbar arrays. Adv Intell Syst 3(1):2000137. https://doi.org/10.1002/aisy.202000137
Linn E, Rosezin R, Kügeler C et al (2010) Complementary resistive switches for passive nanocrossbar memories. Nat Mater 9(5):403–406. https://doi.org/10.1038/nmat2748
Merced-Grafals EJ, Dávila N, Ge N et al (2016) Repeatable, accurate, and high speed multi-level programming of memristor 1T1R arrays for power efficient analog computing applications. Nanotechnology 27(36):365202. https://doi.org/10.1088/0957-4484/27/36/365202
Nandakumar S, Le Gallo M, Piveteau C et al (2020) Mixed-precision deep learning based on computational memory. Front Neurosci 14:406. https://doi.org/10.3389/fnins.2020.00406
Ni R, Yang L, Huang XD et al (2021) Controlled majority-inverter graph logic with highly nonlinear, self-rectifying memristor. IEEE Trans Electron Devices 68(10):4897–4902. https://doi.org/10.1109/TED.2021.3106234
Peng X, Huang S, Jiang H et al (2020) DNN+ neurosim v2. 0: an end-to-end benchmarking framework for compute-in-memory accelerators for on-chip training. IEEE Trans Comput-Aided Des Integr Circuits Syst 40(11):2306–2319
Seide F, Fu H, Droppo J, et al (2014) 1bit stochastic gradient descent and its application to dataparallel distributed training of speech DNNs. In: Interspeech. https://api.semanticscholar.org/CorpusID:2189412
Soudry D, Di Castro D, Gal A et al (2015) Memristor-based multilayer neural networks with online gradient descent training. IEEE Trans Neural Netw Learn Syst 26(10):2408–2421
Strubell E, Ganesh A, McCallum A (2020) Energy and policy considerations for modern deep learning research. In: Proceedings of the AAAI conference on artificial intelligence, vol 34. AAAI, pp 13693–13696. https://doi.org/10.1609/aaai.v34i09.7123
Tan J, Duan S, Wang L et al (2023) Multigas sensing electronic nose using memristor-based inmemory computing. IEEE Sens J. https://doi.org/10.1109/JSEN.2023.3323943
Wang Y, Wu S, Tian L et al (2020) SSM: a high-performance scheme for in situ training of imprecise memristor neural networks. Neurocomputing 407:270–280
Wei X, Gong R, Li Y, et al (2022) Qdrop: randomly dropping quantization for extremely low-bit post-training quantization. arXiv preprint arXiv:2203.05740
Wu Y, Wang Q, Wang Z, et al (2023) Bulk-switching memristor-based compute-in-memory module for deep neural network training. arXiv preprint arXiv:2305.14547
Xia Q, Yang JJ (2019) Memristive crossbar arrays for brain-inspired computing. Nat Mater 18(4):309–323. https://doi.org/10.1038/s41563-019-0291-x
Xiao T, Bennett C, Feinberg B, et al (2022) CrossSim: accuracy simulation of analog in-memory computing
Xu W, Wang J, Yan X (2021) Advances in memristorbased neural networks. Front Nanatechnol 3:645995
Yao P, Wu H, Gao B et al (2020) Fully hardware-implemented memristor convolutional neural network. Nature 577(7792):641–646. https://doi.org/10.1038/s41586-020-1942-4
Zhang Q, Wu H, Yao P et al (2018) Sign backpropagation: an on-chip learning algorithm for analog RRAM neuromorphic computing systems. Neural Netw 108:217–223. https://doi.org/10.1016/j.neunet.2018.08.012
Acknowledgements
This work was supported in part by the National Natural Science Foundation of China under Grant Nos. U20A20227, 62076207, and 62076208.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors confirm that they do not have any known financial interests or personal relationships that could have influenced the work presented in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Tan, J., Zhang, F., Wu, J. et al. Enhancing in-situ updates of quantized memristor neural networks: a Siamese network learning approach. Cogn Neurodyn (2024). https://doi.org/10.1007/s11571-024-10069-1
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11571-024-10069-1