Principles of analog neuromorphic computing: from components to systems and algorithms
- Authors: Demin V.A.1, Emelyanov A.V.1,2, Nikiruy K.E.1, Surazhevsky I.A.1, Sitnikov A.V.1,3, Rylkov V.V.1, Kashkarov P.K.1,2, Kovalchuk M.V.1,2,4
-
Affiliations:
- National Research Centre “Kurchatov Institute”
- The Moscow Institute of Physics and Technology
- Voronezh State Technical University
- Lomonosov Moscow State University
- Issue: Vol 18, No 4 (2023)
- Pages: 794-797
- Section: Conference proceedings
- URL: https://genescells.ru/2313-1829/article/view/623345
- DOI: https://doi.org/10.17816/gc623345
- ID: 623345
Cite item
Abstract
This report presents the current state of affairs in the implementation of artificial intelligence hardware accelerators based on practically successful neural network algorithms of the first and second generations based on formal artificial neural networks (ANNs). The shortcomings of existing solutions are noted and ways to overcome them using analog neuromorphic architectures are outlined.
The latter are created on the principles of the structuring and functioning of a living nervous system, using artificial neurons and models of synaptic contacts - the so–called memristors, electrically rewritable nanoscale elements of non-volatile memory [1-3]. With the use of these elements, it is possible to significantly increase the performance and energy efficiency of algorithm accelerators based on the ANNs [4-6], as well as the formation of promising computing systems based on bioplausible 3rd generation neural network algorithms - Spiking Neural Networks (SNNs) [7-9].
The original method of substantiating the optimal rules for local tuning SNNs with frequency encoding and the possibility of their implementation in the form of the Spike-Timing-Dependent Plasicity (STDP) are discussed [10]. The results of SNN learning stability to a variability of analog memristors, as well as the use of noise as a constructive factor in the fine-tuning and maintenance of SNN memristive weights are demonstrated [7, 11].
Also, approaches to the implementation of local plasticity rules with dopamine-like modulation as a type of SNN reinforcement learning are discussed. The latter is necessary for the formation of imitative "needs" of an agent in the process of its autonomous functioning [12, 13, 14]. The first results on the creation of a prototype of a memristive implantable neuroprosthesis of the motor activity are considered [15, 16].
Finally, possible hardware solutions for both neuronal elements and synaptic connections based on suitable memristive devices are demonstrated. The concept and first results on the creation of an analog neuromorphic computing system based on the above components are presented.
Thus, an attempt is made to systematize the existing and original methods of implementing energy-efficient and compact analog neuromorphic computing systems for real-time and life-learning artificial intelligence.
Full Text
This report outlines the current implementation status of hardware accelerators for artificial intelligence, focusing on successful neural network algorithms of the first and second generations that use formal artificial neural networks (ANNs). Identified shortcomings of current solutions are addressed, with proposed solutions using analog neuromorphic architectures.
The latter are designed based on the structural and functional principles of a living nervous system, using artificial neurons and models of synaptic connections, commonly referred to as memristors. These are electrically rewritable nanoscale components of non-volatile memory [1–3]. By using these components, it is feasible to significantly enhance the effectiveness and energy efficiency of algorithm accelerators which are based on ANNs [4–6]. Additionally, it enables the development of promising computing systems relying on bioplausible third-generation neural network algorithms, namely Spiking Neural Networks (SNNs) [7–9].
The paper discusses the original approach to establishing optimal rules for tuning local SNNs with frequency encoding. It also explores the potential implementation of said rules using Spike-Timing-Dependent Plasticity (STDP) [10]. The study demonstrates the stability of SNN learning when subjected to analogue memristors’ variance, and it highlights noise as an effective tool for fine-tuning and sustaining SNN memristive weights [7, 11].
Approaches to implementing local plasticity rules with dopamine-like modulation are discussed as a type of SNN reinforcement learning. This approach is necessary for forming imitative “needs” of an agent during autonomous functioning [12–14]. In addition, the first results of the creation of a prototype of a memristive implantable neuroprosthesis for motor activity are examined [15, 16].
Finally, potential hardware solutions for both neuronal components and synaptic connections using suitable memristive devices are demonstrated. The concept and initial findings of an analog neuromorphic computing system created with the aforementioned components are presented.
Thus, this paper aims to organize current and novel approaches for implementing energy-efficient and compact analog neuromorphic computing systems that can enable real-time processing and lifelong learning in artificial intelligence.
ADDITIONAL INFORMATION
Authors’ contribution. All authors made a substantial contribution to the conception of the work, acquisition, analysis, interpretation of data for the work, drafting and revising the work, final approval of the version to be published and agree to be accountable for all aspects of the work.
Funding sources. The work was carried out with the financial support of the Ministry of Science and Higher Education, agreement No. 075-15-2023-324.
Competing interests. The authors declare that they have no competing interests.
About the authors
V. A. Demin
National Research Centre “Kurchatov Institute”
Author for correspondence.
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow
A. V. Emelyanov
National Research Centre “Kurchatov Institute”; The Moscow Institute of Physics and Technology
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow; Dolgoprudny, Moscow region
K. E. Nikiruy
National Research Centre “Kurchatov Institute”
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow
I. A. Surazhevsky
National Research Centre “Kurchatov Institute”
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow
A. V. Sitnikov
National Research Centre “Kurchatov Institute”; Voronezh State Technical University
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow; Voronezh
V. V. Rylkov
National Research Centre “Kurchatov Institute”
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow
P. K. Kashkarov
National Research Centre “Kurchatov Institute”; The Moscow Institute of Physics and Technology
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow; Dolgoprudny, Moscow region
M. V. Kovalchuk
National Research Centre “Kurchatov Institute”; The Moscow Institute of Physics and Technology; Lomonosov Moscow State University
Email: demin.vyacheslav@mail.ru
Russian Federation, Moscow; Dolgoprudny, Moscow region; Moscow
References
- Strukov D, Snider G, Stewart D, Williams R. The missing memristor found. Nature. 2008;453(7191):80–83. Corrected and republished from: Nature. 2009;459(7250):1154. doi: 10.1038/nature06932
- Martyshov MN, Emelyanov AV, Demin VA, et al. Multifilamentary character of anticorrelated capacitive and resistive switching in memristive structures based on (Co-Fe-B)x(LiNbO3)100−x nanocomposite. Phys Rev Applied. 2020;14:034016.
- Minnekhanov AA, Emelyanov AV, Lapkin DA, et al. Parylene-based memristive devices with multilevel resistive switching for neuromorphic applications. Sci Rep. 2019;9(1):10800. doi: 10.1038/s41598-019-47263-9
- Prezioso M, Merrikh-Bayat F, Hoskins BD, et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature. 2015;521(7550):61–64. doi: 10.1038/nature14441
- Emelyanov AV, Lapkin DA, Demin VA, Erokhin VV. First steps towards the realization of a double layer perceptron based on organic memristive devices. AIP Advances. 2016;6(11):111301. doi: 10.1063/1.4966257
- Yao P, Wu H, Gao B, et al. Fully hardware-implemented memristor convolutional neural network. Nature. 2020;577(7792):641–646. doi: 10.1038/s41586-020-1942-4
- Emelyanov AV, Nikiruy KE, Serenko AV, et al. Self-adaptive STDP-based learning of a spiking neuron with nanocomposite memristive weights. Nanotechnology. 2020;31(4):045201. doi: 10.1088/1361-6528/ab4a6d
- Wang W, Pedretti G, Milo V, et al. Learning of spatiotemporal patterns in a spiking neural network with resistive switching synapses. Sci Adv. 2018;4(9):eaat4752. doi: 10.1126/sciadv.aat4752
- Gerasimova SA, Mikhaylov AN, Belov AI, et al. Design of memristive interface between electronic neurons. AIP Conf Proc. 2018;1959(1):090005. doi: 10.1063/1.5034744
- Demin VA, Nekhaev DV, Surazhevsky IA, et al. Necessary conditions for STDP-based pattern recognition learning in a memristive spiking neural network. Neural Netw. 2021;134:64–75. doi: 10.1016/j.neunet.2020.11.005
- Surazhevsky IA, Demin VA, Ilyasov AI, et al. Noise-assisted persistence and recovery of memory state in a memristive spiking neuromorphic network. Chaos, Solitons and Fractals. 2021;146:110890. doi: 10.1016/j.chaos.2021.110890
- Nikiruy KE, Emelyanov AV, Demin VA, et al. Dopamine-like STDP modulation in nanocomposite memristor. AIP Advances. 2019;9(6):065116. doi: 10.1063/1.5111083
- Minnekhanov AA, Shvetsov BS, Emelyanov AV. Parylene-based memristive synapses for hardware neural networks capable of dopamine-modulated STDP learning. J Phys D: Appl Phys. 2021;54(48):484002. doi: 10.1088/1361-6463/ac203c
- Vlasov D, Rybka R, Sboev A. Reinforcement learning in a spiking neural network with memristive plasticity. In: Reinforcement learning in a spiking neural network with memristive plasticity. 6th Scientific School Dynamics of Complex Networks and their Applications (DCNA); 2022; Kaliningrad, Russian Federation. P. 300–302. doi: 10.1109/DCNA56428.2022.9923314
- Mikhaylov A, Pimashkin A, Pigareva Y, et al. Neurohybrid memristive CMOS-integrated systems for biosensors and neuroprosthetics. Front Neurosci. 2020;14:358. doi: 10.3389/fnins.2020.00358
- Masaev DN, Suleimanova AA, Prudnikov NV, et al. Memristive circuit-based model of central pattern generator to reproduce spinal neuronal activity in walking pattern. Front Neurosci. 2023;17:1124950. doi: 10.3389/fnins.2023.1124950