A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks
Abstract
Inspired by the parallelism and efficiency of the brain, several candidates for artificial synapse devices have been developed for neuromorphic computing, yet a nonlinear and asymmetric synaptic response curve precludes their use for backpropagation, the foundation of modern supervised learning. Spintronic devices—which benefit from high endurance, low power consumption, low latency, and CMOS compatibility—are a promising technology for memory, and domain-wall magnetic tunnel junction (DW-MTJ) devices have been shown to implement synaptic functions such as long-term potentiation and spike-timing dependent plasticity. In this work, we propose a notched DW-MTJ synapse as a candidate for supervised learning. Using micromagnetic simulations at room temperature, we show that notched synapses ensure the non-volatility of the synaptic weight and allow for highly linear, symmetric, and reproducible weight updates using either spin transfer torque (STT) or spin–orbit torque (SOT) mechanisms of DW propagation. We use lookup tables constructed from micromagnetics simulations to model the training of neural networks built with DW-MTJ synapses on both the MNIST and Fashion-MNIST image classification tasks. Accounting for thermal noise and realistic process variations, the DW-MTJ devices achieve classification accuracy close to ideal floating-point updates using both STT and SOT devices at room temperature and at 400 K. Our work establishesmore »
- Authors:
-
- Univ. of Texas, Austin, TX (United States)
- Sandia National Laboratories, Albuquerque, New Mexico 87123, USA
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Publication Date:
- Research Org.:
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Org.:
- USDOE National Nuclear Security Administration (NNSA)
- OSTI Identifier:
- 1784862
- Alternate Identifier(s):
- OSTI ID: 1784352
- Report Number(s):
- SAND-2021-6135J
Journal ID: ISSN 0003-6951; 696352; TRN: US2210344
- Grant/Contract Number:
- AC04-94AL85000
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Applied Physics Letters
- Additional Journal Information:
- Journal Volume: 118; Journal Issue: 20; Journal ID: ISSN 0003-6951
- Publisher:
- American Institute of Physics (AIP)
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 42 ENGINEERING; Spintronics; Magnetic devices; Artificial neural networks; Spintronic devices; Magnetic memories; Machine learning; Spin-transfer-torque; Memory device; Stochastic processes; Magnetic tunnel junctions
Citation Formats
Liu, Samuel, Xiao, T. Patrick, Cui, Can, Incorvia, Jean C., Bennett, Christopher H., and Marinella, Matthew J. A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks. United States: N. p., 2021.
Web. doi:10.1063/5.0046032.
Liu, Samuel, Xiao, T. Patrick, Cui, Can, Incorvia, Jean C., Bennett, Christopher H., & Marinella, Matthew J. A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks. United States. https://doi.org/10.1063/5.0046032
Liu, Samuel, Xiao, T. Patrick, Cui, Can, Incorvia, Jean C., Bennett, Christopher H., and Marinella, Matthew J. Mon .
"A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks". United States. https://doi.org/10.1063/5.0046032. https://www.osti.gov/servlets/purl/1784862.
@article{osti_1784862,
title = {A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks},
author = {Liu, Samuel and Xiao, T. Patrick and Cui, Can and Incorvia, Jean C. and Bennett, Christopher H. and Marinella, Matthew J.},
abstractNote = {Inspired by the parallelism and efficiency of the brain, several candidates for artificial synapse devices have been developed for neuromorphic computing, yet a nonlinear and asymmetric synaptic response curve precludes their use for backpropagation, the foundation of modern supervised learning. Spintronic devices—which benefit from high endurance, low power consumption, low latency, and CMOS compatibility—are a promising technology for memory, and domain-wall magnetic tunnel junction (DW-MTJ) devices have been shown to implement synaptic functions such as long-term potentiation and spike-timing dependent plasticity. In this work, we propose a notched DW-MTJ synapse as a candidate for supervised learning. Using micromagnetic simulations at room temperature, we show that notched synapses ensure the non-volatility of the synaptic weight and allow for highly linear, symmetric, and reproducible weight updates using either spin transfer torque (STT) or spin–orbit torque (SOT) mechanisms of DW propagation. We use lookup tables constructed from micromagnetics simulations to model the training of neural networks built with DW-MTJ synapses on both the MNIST and Fashion-MNIST image classification tasks. Accounting for thermal noise and realistic process variations, the DW-MTJ devices achieve classification accuracy close to ideal floating-point updates using both STT and SOT devices at room temperature and at 400 K. Our work establishes the basis for a magnetic artificial synapse that can eventually lead to hardware neural networks with fully spintronic matrix operations implementing machine learning.},
doi = {10.1063/5.0046032},
journal = {Applied Physics Letters},
number = 20,
volume = 118,
place = {United States},
year = {Mon May 17 00:00:00 EDT 2021},
month = {Mon May 17 00:00:00 EDT 2021}
}
Works referenced in this record:
Recent progress in resistive random access memories: Materials, switching mechanisms, and performance
journal, September 2014
- Pan, F.; Gao, S.; Chen, C.
- Materials Science and Engineering: R: Reports, Vol. 83
Scaling magnetic tunnel junction down to single-digit nanometers—Challenges and prospects
journal, April 2020
- Jinnai, Butsurin; Watanabe, Kyota; Fukami, Shunsuke
- Applied Physics Letters, Vol. 116, Issue 16
Low Energy Magnetic Domain Wall Logic in Short, Narrow, Ferromagnetic Wires
journal, January 2012
- Currivan, Jean Anne; Mascaro, M. D.
- IEEE Magnetics Letters, Vol. 3
Integration of spintronic interface for nanomagnetic arrays
journal, December 2011
- Lyle, Andrew; Harms, Jonathan; Klein, Todd
- AIP Advances, Vol. 1, Issue 4
Hitting the memory wall: implications of the obvious
journal, March 1995
- Wulf, Wm. A.; McKee, Sally A.
- ACM SIGARCH Computer Architecture News, Vol. 23, Issue 1
Phase change memory technology
journal, March 2010
- Burr, Geoffrey W.; Breitwisch, Matthew J.; Franceschini, Michele
- Journal of Vacuum Science & Technology B, Nanotechnology and Microelectronics: Materials, Processing, Measurement, and Phenomena, Vol. 28, Issue 2
Synapse cell optimization and back-propagation algorithm implementation in a domain wall synapse based crossbar neural network for scalable on-chip learning
journal, June 2020
- Kaushik, Divya; Sharda, Janak; Bhowmik, Debanjan
- Nanotechnology, Vol. 31, Issue 36
Magnetic Domain Wall Based Synaptic and Activation Function Generator for Neuromorphic Accelerators
journal, December 2019
- Siddiqui, Saima A.; Dutta, Sumit; Tang, Astera
- Nano Letters, Vol. 20, Issue 2
Logic circuit prototypes for three-terminal magnetic tunnel junctions with mobile domain walls
journal, January 2016
- Currivan-Incorvia, J. A.; Siddiqui, S.; Dutta, S.
- Nature Communications, Vol. 7, Issue 1
The design and verification of MuMax3
journal, October 2014
- Vansteenkiste, Arne; Leliaert, Jonathan; Dvornik, Mykola
- AIP Advances, Vol. 4, Issue 10
Analog architectures for neural network acceleration based on non-volatile memory
journal, September 2020
- Xiao, T. Patrick; Bennett, Christopher H.; Feinberg, Ben
- Applied Physics Reviews, Vol. 7, Issue 3
A brain-plausible neuromorphic on-the-fly learning system implemented with magnetic domain wall analog memristors
journal, April 2019
- Yue, Kun; Liu, Yizhou; Lake, Roger K.
- Science Advances, Vol. 5, Issue 4
Neuromorphic Functions in PEDOT:PSS Organic Electrochemical Transistors
journal, October 2015
- Gkoupidenis, Paschalis; Schaefer, Nathan; Garlan, Benjamin
- Advanced Materials, Vol. 27, Issue 44
Resistive Random Access Memory (ReRAM) Based on Metal Oxides
journal, December 2010
- Akinaga, Hiroyuki; Shima, Hisashi
- Proceedings of the IEEE, Vol. 98, Issue 12
In situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory
journal, April 2021
- Li, Yiyang; Xiao, T. Patrick; Bennett, Christopher H.
- Frontiers in Neuroscience, Vol. 15
Impact of Non-Ideal Characteristics of Resistive Synaptic Devices on Implementing Convolutional Neural Networks
journal, September 2019
- Sun, Xiaoyu; Yu, Shimeng
- IEEE Journal on Emerging and Selected Topics in Circuits and Systems, Vol. 9, Issue 3
Matching domain-wall configuration and spin-orbit torques for efficient domain-wall motion
journal, January 2013
- Khvalkovskiy, A. V.; Cros, V.; Apalkov, D.
- Physical Review B, Vol. 87, Issue 2
A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing
journal, February 2017
- van de Burgt, Yoeri; Lubberman, Ewout; Fuller, Elliot J.
- Nature Materials, Vol. 16, Issue 4
Three-terminal magnetic tunnel junction synapse circuits showing spike-timing-dependent plasticity
journal, September 2019
- Akinola, Otitoaleke; Hu, Xuan; Bennett, Christopher H.
- Journal of Physics D: Applied Physics, Vol. 52, Issue 49
Magnetic domain wall neuron with lateral inhibition
journal, October 2018
- Hassan, Naimul; Hu, Xuan; Jiang-Wei, Lucian
- Journal of Applied Physics, Vol. 124, Issue 15
Spike time dependent plasticity (STDP) enabled learning in spiking neural networks using domain wall based synapses and neurons
journal, December 2019
- Sahu, Upasana; Pandey, Aadit; Goyal, Kushaagra
- AIP Advances, Vol. 9, Issue 12
Phase Change Memory
journal, December 2010
- Wong, H. -S. Philip; Raoux, Simone; Kim, SangBum
- Proceedings of the IEEE, Vol. 98, Issue 12
Multiscale Co-Design Analysis of Energy, Latency, Area, and Accuracy of a ReRAM Analog Neural Training Accelerator
journal, March 2018
- Marinella, Matthew J.; Agarwal, Sapan; Hsia, Alexander
- IEEE Journal on Emerging and Selected Topics in Circuits and Systems, Vol. 8, Issue 1
Scalability of Magnetic Tunnel Junctions Patterned by a Novel Plasma Ribbon Beam Etching Process on 300 mm Wafers
journal, December 2015
- Xue, Lin; Kontos, Alex; Lazik, Christopher
- IEEE Transactions on Magnetics, Vol. 51, Issue 12
Accurate deep neural network inference using computational phase-change memory
journal, May 2020
- Joshi, Vinay; Le Gallo, Manuel; Haefeli, Simon
- Nature Communications, Vol. 11, Issue 1
Proposal for an All-Spin Artificial Neural Network: Emulating Neural and Synaptic Functionalities Through Domain Wall Motion in Ferromagnets
journal, December 2016
- Sengupta, Abhronil; Shim, Yong; Roy, Kaushik
- IEEE Transactions on Biomedical Circuits and Systems, Vol. 10, Issue 6
Resistive memory device requirements for a neural algorithm accelerator
conference, July 2016
- Agarwal, Sapan; Plimpton, Steven J.; Hughart, David R.
- 2016 International Joint Conference on Neural Networks (IJCNN)
Current-induced domain wall motion
journal, April 2008
- Beach, G. S. D.; Tsoi, M.; Erskine, J. L.
- Journal of Magnetism and Magnetic Materials, Vol. 320, Issue 7