Synergy of fivefold boost SOT efficiency and field-free magnetization switching with broken inversion symmetry: Toward neuromorphic computing

TL;DR

This paper demonstrates a significant enhancement in spin orbit torque efficiency and field-free magnetization switching by integrating Ruthenium Oxide, enabling practical, energy-efficient neuromorphic computing with high neural network accuracy.

Contribution

The study introduces a novel RuO2 layer to boost SOT efficiency and achieve field-free switching, advancing neuromorphic computing hardware.

Findings

5.2 times increase in SOT efficiency with RuO2 layer

Threefold reduction in switching current density

Achieved high image recognition accuracy (~95%) on MNIST datasets

Abstract

Non-volatile Neuromorphic Computing (NC) elements utilizing Spin Orbit Torque (SOT) provide a viable solution to alleviate the memory wall bottleneck in contemporary computing systems. However, the two challenges, low SOT efficiency and the need for in plane symmetry breaking field for perpendicular magnetization switching, greatly limit its practical implementation. In this work, the enhanced SOT efficiency of Platinum (Pt) SOT layer and field free perpendicular magnetization switching are achieved by integrating thin Ruthenium Oxide (RuO2) layer in our material stack. The optimal RuO2 thickness (0.5 nm) enhances 5.2 times Damping Like (DL) SOT efficiency compared with pure SOT layer (Pt), as determined by hysteresis loop shift measurements, with a relatively low resistivity (90 micro-Ohm-cm). Moreover, we achieve 3 times reduction of critical magnetization switching current density compared to reference sample. Our experimental findings also demonstrate Rashba-induced substantial field-free magnetization switching in the presence of an emergent built-in interfacial field. Notably, reliable multi resistance synaptic states are achieved by tailoring the synergistic effects of enhanced SOT and interfacial magnetism. The functionality of synaptic states has been further evaluated by implementing an artificial neural network and achieved image recognition accuracies of approximately 95% and 87% on the MNIST and Fashion-MNIST datasets, respectively. This systematic study paves the way to energy-efficient, field-free SOT synapses for practical NC applications.