Voltage Controlled Energy Efficient Domain Wall Synapses with Stochastic Distribution of Quantized Weights in the Presence of Thermal Noise and Edge Roughness

Abstract

We propose energy efficient strain control of domain wall (DW) in a perpendicularly magnetized nanoscale racetrack on a piezoelectric substrate that can implement multi state synapse to be utilized in neuromorphic computing platforms. In conjunction with SOT from to a current flowing in the heavy metal layer, strain is generated by applying a voltage across the piezoelectric. Such a strain is mechanically transferred to the racetrack and modulates the Perpendicular Magnetic Anisotropy (PMA). When different voltages are applied (i.e. different strains are generated), it can translate the DW to different distances for the same current which implements different synaptic weights. We have shown using micromagnetic simulations that 5-state and 3-state synapse can be implemented in a racetrack that is modeled with natural edge roughness and room temperature thermal noise. Such strain-controlled synapse has an energy consumption of few fJs and could thus be very attractive to implement energy-efficient quantized neural networks, which has been shown recently to achieve near equivalent classification accuracy to the full-precision neural networks.

References in corpus (4)

• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• Matching domain wall configuration and spin-orbit torques for very efficient domain-wall motion
• Domain wall pinning and potential landscapes created by constrictions and protrusions in ferromagnetic nanowires
• Domain-wall motion induced by spin transfer torque delivered by helicity-dependent femtosecond laser