Fabrication And Study Of Artificial Synapses And Probabilistic Neuron Devices Based On Spin-orbit Torque

When compared to von Neumann architecture where processing and memory units are physically separated,artificial neuron networks(ANNs)that can perform brain-like computing in memory,are expected to construct a lower power and faster architecture.According to biological neural system,ANNs consist of two building blocks:artificial synapses and neurons.Owing to synaptic plasticity,it is necessary that artificial synaptic devices must have a continuously tunable resistance.On the other hand,artificial neurons are binary devices with a threshold.In particular,stochastic neurons,or probabilistic bits(p-bits),can be used to implement stochastic neural networks to perform tasks that cannot be directly solved by traditional circuits such as integer factorization and invertible logic.As an emerging non-volatile threshold device,spin–orbit-torque(SOT)device has advantages of ultrafast speed,low power consumption,and high stability and endurance.The results show that SOT can induce multilevel switching in the ferromagnetic film with multi domain structure,while the single domain structure is switched randomly without the assistance of external magnetic field.Accordingly,utilizing SOT devices,it is possible that the artificial synaptic device and the p-bits can be implemented in the same material system.However,the experimental demonstration of the SOT based artificial synapse is rarely reported,and the p-bits without weakening the anisotropy is almost blank.Therefore,this paper proposes SOT based artificial synapses and p-bits via studying the SOT induced magnetization switching of multi-domain devices and single-domain devices,respectively.Firstly,based on the perpendicular magnetized Ta/Co Fe B/Mg O with multi domain,magnetic field induced multilevel switching by SOT was studied from three aspects:principle design,micromagnetic simulation and experimental demonstration.When a write current higher than 3.5 MA/cm~2 is applied to the device,a domain wall along the current direction will be formed in the center of the device.The amplitude of SOT effective field can tune the position of domain wall.When the current is fixed,SOT depends on the in-plane magnetic fields.The different domain wall positions along with the different values of the anomalous Hall effect(AHE)resistance of the device under various magnetic fields within the range of±20 Oe were observed by using magneto-optical Kerr microscope and AHE measurement system,respectively.Moreover,the initialization-free multilevel storage is realized.Secondly,based on the perpendicular magnetized Ta(W)/Co Fe B/Mg O with multi domain,current induced synaptic behavior by SOT was studied.When the in-plane magnetic field is fixed to 50 Oe,domain wall is driven back and forth in a continuous manner in the Co Fe B layer by SOT that the current exerts on the Co Fe B magnetization.Hence,the magnetization and consequently the AHE resistance are modulated in an analog manner,being controlled by the pulsed current characteristics including amplitude,duration,and repetition number.Utilizing magneto-optical Kerr microscope,the current dependent domain wall moving velocity was studied.Furthermore,using positive and negative voltage pulse pair,the STDP and anti-STDP learning rules were realized in the same device by changing the direction of assisting magnetic field.Finally,based on the perpendicular magnetized film Ta/Co Fe B/Mg O with single domain,the dependence of magnetic field and current on the switching probability was studied,and the SOT based probabilistic neuron device and reconfigurable physical unclonable function(RPUF)have been proposed.After a write current higher than 5.8MA/cm~2 is applied to a single domain device,the magnetization direction of Co Fe B will be switched randomly.The switching probability can be tuned by SOT that is determined by the magnetic field and current.Accordingly,the field induced and current induced probabilistic neuron devices have been demenstrated.Then,an RPUF was constructed using a device array composed of nanomagnets working at 50%switching probability.The inter-Hamming distance,the reconfigure Hamming distance and the correlation coefficient between reconfigure cycles of the proposed RPUF were evaluated.