Demonstration of Synaptic Behavior in a Heavy-Metal-Ferromagnetic-Metal-Oxide-Heterostructure-Based Spintronic Device for On-Chip Learning in Crossbar-Array-Based Neural Networks

Abstract

Nanomagnetic and spintronic devices, which make use of physical phenomena in materials and interfaces like perpendicular magnetic anisotropy (PMA) and spin–orbit torque (SOT) to exhibit multiple electrically readable and controllable states, have been widely considered as synaptic elements in analog crossbar arrays for on-chip learning of analog neural networks (ANN). Here, in such a heavy-metal-ferromagnetic-metal-oxide-heterostructure-based (Pt/Co/SiO2) spintronic device, multiple mixed states are first demonstrated experimentally. These mixed states correspond to different magnetization configurations between two saturated states: all magnetic moments vertically up and all moments vertically down (the ferromagnetic layer exhibits PMA). These mixed states are then modulated through in-plane-current pulses, which result in SOT at the heavy-metal–ferromagnet interface, and thus long-term potentiation (LTP) and long-term depression (LTD) are demonstrated in the device (synaptic behavior). The experimentally obtained LTP-and-LTD behavior is explained qualitatively through micromagnetic simulations, which model the interface phenomena phenomenologically. The synaptic bit resolution is then determined experimentally to be 5 (≈ 30 distinguishable states) by measuring the stability of the mixed states. The nonlinearity and asymmetry in the obtained LTP and LTD are quantified experimentally. Next, a crossbar-array-based ANN is simulated using spintronic-synapse-device models based on the experimentally obtained LTP and LTD, and reasonably high classification accuracy is predicted for MNIST and Fashion-MNIST data sets, despite the synaptic nonidealities like nonlinearity, asymmetry, and cycle-to-cycle variations. The impact of nonlinearity and asymmetry on classification accuracy is found to be much higher than that due to limited bit resolution (we do not go below 10 bits per synapse cell though) and cycle-to-cycle variations.

Impact of edge defects on the synaptic characteristic of a ferromagnetic domain-wall device and on on-chip learning

“…Under the assumption that the thickness of the heavy metal layer is higher than the spin diffusion length inside the metal, spin current density is defined as j s = j c × θ SH , where j c is in-plane charge current density and θ SH is the spin Hall angle of the heavy metals. For our simulation of both the devices without defects and with defects, we consider θ SH = 0.1, corresponding to platinum (Pt), which has been used as the heavy metal in the recent experimental demonstrations of such domain-wall synapse by Yadav et al [26]. Yadav et al have, in fact, reported this value of spin Hall angle in Pt in the same report [26] by carrying out hysteresis-loop-shift measurement on their device (following the method described by Hu and Pai [53] and Pai et al [54]).…”

“…For our simulation of both the devices without defects and with defects, we consider θ SH = 0.1, corresponding to platinum (Pt), which has been used as the heavy metal in the recent experimental demonstrations of such domain-wall synapse by Yadav et al [26]. Yadav et al have, in fact, reported this value of spin Hall angle in Pt in the same report [26] by carrying out hysteresis-loop-shift measurement on their device (following the method described by Hu and Pai [53] and Pai et al [54]). Spin diffusion length in Pt has been found to be 3-4 nm [55,56].…”

“…For the device without defects, throughout the ferromagnetic layer (1000 nm × 100 nm), we use the following parameters in our micromagnetic simulation: saturation magnetization (M s ) = 9.5 × 10 5 A m −1 , PMA constant (K) = 1 × 10 6 J m −3 , exchange-correlation constant (A) = 1.5 × 10 −11 J m −1 , and damping factor α = 0.3. These parameters have also been chosen based on the report by Yadav et al, where most of these parameters have been measured experimentally using a Pt/Co/SiO 2 device [26,57]. For our device with defects here, we include triangular regions on either side of the 1000 nm × 100 nm ferromagnetic layer (as shown in figure 1(c)), where PMA constant (K) = 1.05 × 10 6 J m −3 and the rest of the parameters (M s , A, α) have the same values as before.…”

“…We briefly review the recent experimental reports on domain-wall devices next, with our focus being on the linearity and symmetry of the reported LTP and LTD characteristics (requirement (iii) above). Yadav et al [26] have recently quantitatively characterized the non-linearity and asymmetry of the experimentally obtained synaptic characteristic of a Co/Pt/SiO 2 -based domain-wall device (upon the application of identical pulses for LTP, and identical pulse streams for LTD) with the help of the method followed in the literature of the Neuro-Sim simulator [9,27,28], popularly used for this purpose across different NVM synapse technologies. They have reported non-linearity coefficient for LTP α P = 3.875, non-linearity coefficient for LTD α D = −3.84, and thus asymmetry coefficient (|α P − α D |) = 7.715, for their device.…”

Interfacial Origin of Unconventional Spin‐Orbit Torque in Py/γ−$\gamma -$IrMn3

“…[ 1–6 ] Large anti‐damping SOTs are of utmost interest and play a pivotal role in ultrafast magnetization switching, [ 3,7 ] spintronic oscillators, [ 4,6,8–10 ] as well as the emerging area of spintronic‐based neuromorphic computing. [ 11–15 ] In the FM/HM system, the SOT is generated either via the spin Hall effect [ 16–19 ] from the HM layer and/or the Rashba–Edelstein effect [ 20,21 ] from the interface between the FM and HM layers. In both cases, when a charge current flows in the $$x\text{-}$$direction, it generates a spin current in the $$z\text{-}$$direction with the spin polarization along the $$y\text{-}$$direction, which can apply an in‐plane anti‐damping torque on the adjacent FM layer.…”