TY - JOUR
T1 - Perspective
T2 - Spintronic synapse for artificial neural network
AU - Fukami, Shunsuke
AU - Ohno, Hideo
N1 - Funding Information:
The authors are grateful to W. A. Borders, A. Kurenkov, C. Igarashi, T. Hirata, H. Iwanuma, K. Goto, C. Zhang, S. DuttaGupta, and H. Sato for discussion and technical supports for the studies of artificial-synapse part and H. Akima, S. Sato, and Y. Horio for the part of artificial neural network. This work was supported in part by the R&D Project for ICT Key Technology to Realize Future Society of MEXT, ImPACT Program of CSTI, JST-OPERA, JSPS KAKENHI Grant No. 17H06093, JSPS Core-to-Core Program, and Cooperative Research Projects of RIEC (H28/B01, H28/A16, H29/B15).
Publisher Copyright:
© 2018 Author(s).
PY - 2018/10/21
Y1 - 2018/10/21
N2 - While digital integrated circuits with von Neumann architectures, having exponentially evolved for half a century, are an indispensable building block of today's information society, recently growing demand on executing more complex tasks like the human brain has allowed a revisit to the architecture of information processing. Brain-inspired hardware using artificial neural networks is expected to offer a complementary approach to deal with complex problems. Since the neuron and synapse are key components of brains, most of the mathematical models of artificial neural networks require artificial neurons and synapses. Consequently, much effort has been devoted to creating artificial neurons and synapses using various solid-state systems with ferroelectric materials, phase-change materials, oxide-based memristive materials, and so on. Here, we review an example of studies on an artificial synapse based on spintronics and its application to artificial neural networks. The spintronic synapse, having analog and nonvolatile memory functionality, consists of an antiferromagnet/ferromagnet heterostructure and is operated by spin-orbit torque. After giving an overview of this field, we describe the operation principle and results of analog magnetization switching of the spintronic synapse. We then review a proof-of-concept demonstration of the artificial neural network with 36 spintronic synapses, where an associative memory operation based on the Hopfield model is performed and the learning ability of the spintronic synapses is confirmed, showing promise for low-power neuromorphic computation.
AB - While digital integrated circuits with von Neumann architectures, having exponentially evolved for half a century, are an indispensable building block of today's information society, recently growing demand on executing more complex tasks like the human brain has allowed a revisit to the architecture of information processing. Brain-inspired hardware using artificial neural networks is expected to offer a complementary approach to deal with complex problems. Since the neuron and synapse are key components of brains, most of the mathematical models of artificial neural networks require artificial neurons and synapses. Consequently, much effort has been devoted to creating artificial neurons and synapses using various solid-state systems with ferroelectric materials, phase-change materials, oxide-based memristive materials, and so on. Here, we review an example of studies on an artificial synapse based on spintronics and its application to artificial neural networks. The spintronic synapse, having analog and nonvolatile memory functionality, consists of an antiferromagnet/ferromagnet heterostructure and is operated by spin-orbit torque. After giving an overview of this field, we describe the operation principle and results of analog magnetization switching of the spintronic synapse. We then review a proof-of-concept demonstration of the artificial neural network with 36 spintronic synapses, where an associative memory operation based on the Hopfield model is performed and the learning ability of the spintronic synapses is confirmed, showing promise for low-power neuromorphic computation.
UR - http://www.scopus.com/inward/record.url?scp=85054879249&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85054879249&partnerID=8YFLogxK
U2 - 10.1063/1.5042317
DO - 10.1063/1.5042317
M3 - Article
AN - SCOPUS:85054879249
SN - 0021-8979
VL - 124
JO - Journal of Applied Physics
JF - Journal of Applied Physics
IS - 15
M1 - 151904
ER -