114-1 SEMINAR 楊朝堯(陽明交大材料工程)-1141001

Time:1141001 (Wed.) 14:20~16:20
Speaker:Prof.楊朝堯/ Yang, Chao-Yao
國立陽明交通大學材料科學與工程學系所/Department of Materials Science and Engineering, National Yang Ming Chiao Tung University
Title:A spin-orbit torque pulley embedded in an exchange-spring-coupled antiferromagnet/ferromagnet bilayer with pulse-dependent plasticity for efficient neuromorphic computing
Abstract:
Utilizing spin–orbit-torque (SOT) switching of exchange bias (Hex) in ferromagnet (FM)/antiferromagnet (AFM) bilayers has become a mainstream approach to study AFM order and its associated interaction with SOT. However, Hex chiefly reveals interfacial spins of AFM, rather than depth-dependent order formed under exchange-spring coupling throughout the FM/AFM bilayer. In this work, we explore the hidden AFM order in Pt/Co/IrMn trilayers and its effect on SOT switching. Although Hex is fully reversible under SOT as a fully dynamic interface, reducing the driving stimuli of SOT such as a small in-plane field (Hx) and short writing-current pulses (IWrite) enables to reveal a robust subsurface IrMn order that gives rise to the switching asymmetry. The trilayer device exhibits ratchet-type SOT switching upon performing a multipulse characterization with a minimized pulse width at 20 μs, where analog-like forward and digital-like backward reversal can be observed, with the polarity set and reversibly flipped by the initial magnetization-Hex state. The ratchet-type switching accords with an exchange-spring picture in which progressively stronger depth-dependent bulk IrMn orders functions as a buried pinning backbone in the trilayer while the interfacial Hex remains mobile. Leveraging pronounced pulse-width–dependent plasticity (10–100 μs), the measured SOT switching characteristics with tunable threshold and steepness can be applied for the activation functions in convolution neural network (CNN) framework. It has been shown the dynamic annealing, adopting the SOT switching curves obtained from short-to-long pulses for different activation phases, achieves image-recognition accuracy approaching that processed with software ReLU function (~99%). These results show that ratchet-type SOT switching together with pulse-width-dependent plasticity not only reveals a specific hidden AFM order behind Hex in the Pt/Co/IrMn trilayer but also establish a reconfigurable, energy-efficient SOT building block for neuromorphic computing.
Place:S101, Gongguan Campus, NTNU
  • Post author: