Domain wall-based artificial neurons for neuromorphic computing

• Main Author: Mah, William Wai Lum
• Other Authors: S.N. Piramanayagam
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Physics; Neuromorphic computing; Spintronics; Artificial neurons
• Online Access: https://hdl.handle.net/10356/173673
• Institution: Nanyang Technological University

Artificial Intelligence (AI) has been gaining popularity recently. However, the execution of AI in conventional devices with the von Neumann architecture incurred high energy consumption. Consequently, brain-inspired neuromorphic computing (NC) is expected to be a more energy efficient alternative as well as a suitable platform for AI. NC consists of two main components – neurons and synapses. Currently, more studies are focused on artificial synapses. The challenge facing the realization of artificial neurons is the leakage function, which is the automatic domain wall (DW) motion in the absence of applied field or electrical current. The focus of my work is on DW-based artificial neurons, where four designs are presented. The first is a ferromagnetic wire with an anisotropy field gradient, and the second is a trapezoid shaped device. Both designs demonstrated the leakage function by virtue of DW energy reduction. The third device utilizes a synthetic antiferromagnetic (SAF) sample, where the antiferromagnetic coupling gives rise to the leakage function in both experiments and simulations. Likewise, the final design involves two ferromagnetic layers, but it is the stray field from one of the layers that triggers the leakage function. The simulation and experimental results suggest that these four designs have potential applications in the field of NC.