Domain wall magnetic tunnel junction-based artificial synapses and neurons for all-spin neuromorphic hardware

Abstract

We report a breakthrough in the hardware implementation of energy-efficient all-spin synapse and neuron devices for highly scalable integrated neuromorphic circuits. Our work demonstrates the successful execution of all-spin synapse and activation function generator using domain wall-magnetic tunnel junctions. By harnessing the synergistic effects of spin-orbit torque and interfacial Dzyaloshinskii-Moriya interaction in selectively etched spin-orbit coupling layers, we achieve a programmable multi-state synaptic device with high reliability. Our first-principles calculations confirm that the reduced atomic distance between 5d and 3d atoms enhances Dzyaloshinskii-Moriya interaction, leading to stable domain wall pinning. Our experimental results, supported by visualizing energy landscapes and theoretical simulations, validate the proposed mechanism. Furthermore, we demonstrate a spin-neuron with a sigmoidal activation function, enabling high operation frequency up to 20 MHz and low energy consumption of 508 fJ/operation. A neuron circuit design with a compact sigmoidal cell area and low power consumption is also presented, along with corroborated experimental implementation. Our findings highlight the great potential of domain wall-magnetic tunnel junctions in the development of all-spin neuromorphic computing hardware, offering exciting possibilities for energy-efficient and scalable neural network architectures.

Fig. 1. iDMI dependence on thickness of heavy metal layer.

a OOP magnetic hysteresis loops of W(t)/CFB(0.9)/MgO(2)/W(3)/Ru(2) film stack measured by VSM. The arrows indicate field sweeping direction. b K_u, M_s dependence on t_W derived from VSM data in a. c AHE loops with read DC I = ± 8 mA and an in-plane bias field of H_x =  −1584 Oe. Inset shows the measurement schematic. d H_z^eff as function of I under different H_x. e SOT efficiency χ versus H_x, and f iDMI constant obtained from samples with different t_W, the corresponding error bars associated were defined experimentally. Inset shows experimentally measured D in W/CFB heterojunctions adapted from literature reports. The dash line is for guidance of eyes.

Fig. 2. First principles calculations of iDMI in W/CoFe heterojunction.

a Schematic diagram of three sites mechanism of iDMI (left panel) and crystal structure of W/CoFe bilayer (right panel). b Plane-averaged electron difference density Δρ(r) (per unit cell) showing the charge transfer in CoFe/W structure with different t_w. c Gain (peak1) and loss (peak2) of electron peaks extracted from b with difference W thickness. d Theoretically calculated D dependence on W thickness.

Fig. 3. p-MTJ film stacks characterization, devices fabrication and testing.

a Schematic of PMA film stack, and b corresponding HRTEM bright-field (BF) and HAADF images. c OOP and IP M-H loops measured by VSM. Inset shows magnified loops at low fields. d Magnetoresistance versus H_z of p-MTJ device.

Fig. 4. Spin-synapse devices based-on DW-pMTJs.

a Schematic diagram of measurement configuration of four-state synaptic device, representative MOKE image of State III enclosure with magnified PCs. b TEM images of etched profiles at PC edge transition and bottom regions and the corresponding HRTEM image in PC bottom with inverse FFT patterns. c K-H (purple-green) and R-H (blue-red) loops as a function of H_z swept (forth-back) from pristine device before PCs formation. Blue curve denotes Kerr signal measured by p-MOKE. Colorful arrows represent respective magnetization orientation of PL, RF, FL at different states. d R-H loop as a function of H_z of the synaptic device with PCs. Dashed arrows illustrate sweep direction of H_z. e Magnified R-H loop as marked by green dash-box in d. f State-by-state switching of developed four-state synaptic device. Red, green and blue curves signify state1 (I) to state2 (II), state1 (I) to state3 (III), state1 (I) to state4 (IV) switching, respectively.

Fig. 5. Spin-neuron based on DW-pMTJ sigmoidal activation function generator.

a MOKE image, the measuring configuration and protocol of spintronic sigmoid neuron based on DW-pMTJ. b Different resistance states with corresponding DW position. The green arrows indicate the state switching direction. Inset illustrates the resistance curve vs sweeping H_z. c Resistance as a function of H_z of sigmoidal device with central top electrode covering the whole channel (inset image). The arrows indicate field sweeping direction. d Resistance as a function of the amplitude and numbers of pulsed magnetic field for sigmoidal device in c. Insets depict the correspond array and magnified photos of the DW-pMTJ sigmoidal devices.

Fig. 6. SOT and iDMI synergistic effect on DW dynamics with quantitative visualization.

a Schematic of proposed DW pinning mechanism with DMI profile. b Effective fields (left) and torques (right) in FM where m_DW (green), DL (red), DMI (yellow), DWE (blue) denote center magnetization of domain wall, damping-like term of SOT effect, domain wall energy and iDMI, respectively. c Quantitative determination and illustration of DW motion and energy landscape with and without J_e. Inset depicts the schematic of DW coordinates of q and φ.

Fig. 7. Circuit simulation and experimental verification of spin neuron circuit.

a Schematic of a simplified ANN network comprising 16×16 DW-MTJ synapses and 16×1 16-state DW-MTJ neurons. b. Transient simulation results of the spin neuron circuit. c Hardware implementation of with developed 4 binary SOT-MTJs as synapse and 1 spin-neuron device (right panel). d Waveform of the pre-neuron signal output by MCU and corresponding output of the operational amplifier. e MOKE Kerr images captured after each pre-neuron signal applied.