Three-terminal ferroelectric synapse device with concurrent learning function for artificial neural networks

Yu, Ning; Kaneko, Yukihiro; Ueda, Mitsuru; Matsukawa, Takashi; Fujii, Eiji

doi:10.1063/1.4729915

Cited by 189 publications

(169 citation statements)

References 22 publications

Supporting

Mentioning

168

Contrasting

Order By: Relevance

“…17,18,25,28 Moreover, synaptic weight modulation is mainly by engineering the pulse width and height by overlapping pulses, rather than frequency and relative timing of the spikes that is utilised in biological systems. However, three terminal synaptic devices, similar to biological neural systems, are able to realise both signal transmission and learning functions, where signal transmission is carried via the channel medium and synaptic weights are modulated independently via the gate terminals.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Nanoionics-Based Three-Terminal Synaptic Device Using Zinc Oxide

Pillai

Souza

2017

ACS Appl. Mater. Interfaces

139

View full text Add to dashboard Cite

Section: Introductionmentioning

confidence: 99%

“…36 TCO devices employing a ferroelectric gate insulator have also been reported to show EPSC behaviour. 28,37 The (fig 1i). It is revealed here that the anomalous counter-clockwise hysteresis arises from the positive gate sweep and application of a negative gate bias restores the charge separation to its initial state.…”

Section: Introductionmentioning

confidence: 99%

Nanoionics-Based Three-Terminal Synaptic Device Using Zinc Oxide

Pillai

Souza

2017

ACS Appl. Mater. Interfaces

139

View full text Add to dashboard Cite

“…10 The main advantage of using three terminal devices is the suitability for simultaneous learning and signal transmission. 18 In conclusion, we present Y-shaped quantum dot floating gate transistors suitable for associative learning. In a network consisting of two devices, association is implemented by connecting the output of one to an input terminal of a second device.…”

mentioning

confidence: 99%

“…The devices separate signal transmission and learning. Hence, three- 17,18,19 or even fourterminal 20 synaptic devices may be beneficial because of their ability to decouple the postsynaptic pulse from the output. Indeed, associative learning with simultaneous learning and signal transmission has been demonstrated with three-terminal synapses based on nano-particle organic memory field effect transistors by applying postsynaptic pulses with amplitudes as large as 30 V to the third terminal.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Associative learning with Y-shaped floating gate transistors operated in memristive modes

Maier

Hartmann

Emmerling

et al. 2017

Applied Physics Letters

View full text Add to dashboard Cite

2Associative learning in neural networks is crucial and required to relate new experience with existing memories and was first observed by I. Pavlov in his famous dog experiment. 1 Stimulating different senses by the sight of food and the sound of a bell, I. Pavlov trained his dog to correlate salivation with the sound of the bell (association), although at the beginning of the experiment salivation was only triggered by the sight of food. In neural networks, association corresponds to an increase of synaptic strength, which is controlled by action potentials emitted by pre-and postsynaptic neurons. 2,3 Associative learning can be emulated with memristive Hopfield neural networks, 4 memristive circuits 5,6,7,8 or magnetic tunnel junctions. 9 Other demonstrations of associative learning implement two artificial synapses with memristor emulators 10 or a resistor and a memristor. 11 Artificial synapses based on memristors enable synaptic modifications by applying pre-and postsynaptic voltage pulses to the two terminals of the device, 12,13 which allows to combine information storage and processing on the same physical platform. 14,15,16 Owing to the two-terminal character, however, the application of postsynaptic voltage pulses blocks the device output and prevents information to be transferred during the learning process. The devices separate signal transmission and learning. Hence,18,19 or even fourterminal 20 synaptic devices may be beneficial because of their ability to decouple the postsynaptic pulse from the output. Indeed, associative learning with simultaneous learning and signal transmission has been demonstrated with three-terminal synapses based on nano-particle organic memory field effect transistors by applying postsynaptic pulses with amplitudes as large as 30 V to the third terminal. 21 In memristor-synapses, extra terminals may be implemented with additional gates that allow tuning the set voltage with an electric field effect. 22 We report associative learning with Y-shaped quantum dot floating gate transistors operated in memristive modes. The geometry of the devices with two input terminals and one output terminal enables realizing associative learning with postsynaptic voltage amplitudes as low as 0.5 V. The 3 conductance of the Y-shaped quantum dot floating gate transistor depends on the amount of localized charges on quantum dots (QDs) that are precisely positioned in the two input terminals. Emulating external stimuli "food" and "bell" with two input voltages allows implementing an association between "bell" and "salivation". The number of required pulses to develop (association) or to forget (extinction) the correlation between "bell" and "salivation" is controlled by the amplitudes of the input voltages. Connecting the voltage applied to the drain contact (i.e. right branch V r ) with the side gates (see Fig. 1(a)) leads to a memristive operation. 25,26,27 The Y-shaped floating gate transistor is characterized by applying the voltage V r = V l to both branches with resistances of R l = R...

show abstract

Artificial Synapses Based on Multiterminal Memtransistors for Neuromorphic Application

Wang

Liao

Wong

et al. 2019

Adv Funct Materials

216

206

View full text Add to dashboard Cite

Neuromorphic computing, which emulates the biological neural systems could overcome the high‐power consumption issue of conventional von‐Neumann computing. State‐of‐the‐art artificial synapses made of two‐terminal memristors, however, show variability in filament formation and limited capacity due to their inherent single presynaptic input design. Here, a memtransistor‐based artiﬁcial synapse is realized by integrating a memristor and selector transistor into a multiterminal device using monolayer polycrys‐talline‐MoS2 grown by a scalable chemical vapor deposition (CVD) process. Notably, the memtransistor offers both drain‐ and gate‐tunable nonvolatile memory functions, which efficiently emulates the long‐term potentiation/depression, spike‐amplitude, and spike‐timing‐dependent plasticity of biological synapses. Moreover, the gate tunability function that is not achievable in two‐terminal memristors, enables significant bipolar resistive states switching up to four orders‐of‐magnitude and high cycling endurance. First‐principles calculations reveal a new resistive switching mechanism driven by the diffusion of double sulfur vacancy perpendicular to the MoS2 grain boundary, leading to a conducting switching path without the need for a filament forming process. The seamless integration of multiterminal memtransistors may offer another degree‐of‐freedom to tune the synaptic plasticity by a third gate terminal for enabling complex neuromorphic learning.

show abstract

Three-terminal ferroelectric synapse device with concurrent learning function for artificial neural networks

Cited by 189 publications

References 22 publications

Nanoionics-Based Three-Terminal Synaptic Device Using Zinc Oxide

Nanoionics-Based Three-Terminal Synaptic Device Using Zinc Oxide

Associative learning with Y-shaped floating gate transistors operated in memristive modes

Artificial Synapses Based on Multiterminal Memtransistors for Neuromorphic Application

Contact Info

Product

Resources

About