Applied Physics Letters · 2015 · 86 citations · 33 references
Natural SciencesComputer EngineeringElectrical EngineeringElectromagnetismMagnetic MaterialsNeural EngineeringMagnetization ProcessesMemristive MaterialsSpiking Neural NetworksSpintronic MaterialsSpintronicsSynaptic TransmissionCmos Digital Circuit DesignCurrent-induced Magnetization SwitchingHandwritten Character RecognitionMemristive Crossbar ArraysNeuromorphic DevicesNeuromorphic HardwareSpin-orbit TorquesSpin-transfer Torque
The authors propose a spin‑orbit torque device that functions as an electronic neuron. The SOT device implements an artificial neuron’s thresholding function through a two‑step switching scheme that moves a nanomagnet’s magnetization from an unstable hard‑axis point to one of two stable states, with synaptic currents encoded as bipolar currents from a resistive crossbar array. Simulations of a two‑layer feed‑forward neural network using the SOT neuron show approximately three‑times lower power consumption than a 45 nm CMOS implementation while achieving about 80 % accuracy on MNIST handwritten digit classification.
A device based on current-induced spin-orbit torque (SOT) that functions as an electronic neuron is proposed in this work. The SOT device implements an artificial neuron's thresholding (transfer) function. In the first step of a two-step switching scheme, a charge current places the magnetization of a nano-magnet along the hard-axis, i.e., an unstable point for the magnet. In the second step, the SOT device (neuron) receives a current (from the synapses) which moves the magnetization from the unstable point to one of the two stable states. The polarity of the synaptic current encodes the excitatory and inhibitory nature of the neuron input and determines the final orientation of the magnetization. A resistive crossbar array, functioning as synapses, generates a bipolar current that is a weighted sum of the inputs. The simulation of a two layer feed-forward artificial neural network based on the SOT electronic neuron shows that it consumes ∼3× lower power than a 45 nm digital CMOS implementation, while reaching ∼80% accuracy in the classification of 100 images of handwritten digits from the MNIST dataset.
33
Gradient-based learning applied to document recognition
Yann LeCun, Léon Bottou, Yoshua Bengio et al. · Proceedings of the IEEE · 1998 · 58.2K citations · Full text
Nanoscale Memristor Device as Synapse in Neuromorphic Systems
Sung Hyun Jo, Ting‐Chang Chang, Idongesit E. Ebong et al. · Nano Letters · 2010 · 4.1K citations
Life Sciences, Electrical Engineering, Materials Science +17
Spin-Torque Switching with the Giant Spin Hall Effect of Tantalum
Luqiao Liu, Chi‐Feng Pai, Yi Li et al. · Science · 2012 · 4.1K citations · Full text
Natural Sciences, Materials Engineering, Materials Science +17
J. E. Hirsch · Physical Review Letters · 1999 · 3.1K citations · Full text