Tilted magnetic anisotropy-tailored spin torque nano-oscillators for neuromorphic computing

Abstract

Spin torque nano-oscillators (STNOs) hold significant promise for communication and bio-inspired computing applications. However, their modulation capability is constrained by a dilemma between frequency window and linewidth reduction, particularly in hypercritical conditions like the presence of an external magnetic field. This poses a notable challenge in the practical application of STNOs. Here, we report a unique type of all-electrical compact STNOs that employ the tilted magnetic anisotropy (TMA), which can efficiently promote the linewidth Δf reduction and precisely modulate oscillation frequency ranging from 495 to 556 MHz. The developed STNOs consist of a ferromagnetic reference layer with tunable TMA, wherein the spin transfer torque along the tilted spin polarization direction elaborates a self-oscillation of magnetic moments in the free layer without application of magnetic field. The free layer equips in a magnetic droplet oscillation mode, and the oscillation frequency can be modulated either synergistically or independently by varying the current intensity and/or the TMA angle. Nevertheless, the TMA angle primarily governs the deformation of the magnetic droplet and the corresponding oscillation frequency and linewidth. Moreover, a unique 4 × 4 STNO array with optimized input current and TMA configuration is proposed to execute the reservoir computing hardware training based on nonlinear dynamic oscillation phase-coupling characteristics, promising a diverse synchronization map with high kernel quality and low generation rank for highly reliable pattern classification implementation. The developed STNOs possess a simple structure, nonlinearity, high frequency tunability, and compatibility with CMOS processes, enabling them a fundamental component for large-scale integration of advanced hardware in neuromorphic computing.

Granular Magnetization Switching in Pt/Co/Ti Structure with HfOx Insertion for In-Memory Computing Applications

“…Another approach is to control the DW motion in the free layer of MTJ, causing the parallel and antiparallel composition change between the free layer and fixed layer to tune the output resistance. ,− Recently, the MTJ devices utilizing DW positions for in-memory computing, artificial synapse, and spiking neuron functionalities have been successfully demonstrated. − However, this method requires larger device sizes and complex device structures to produce multiple DW locations. The presence of multiple DW locations is associated with an increased number of output resistance states for synapses and the realization of neuron functionalities such as leaky integrate-and-fire and self-reset. − Aside from device fabrication and miniaturization, this approach also poses challenges in the readout window due to the limited tunnel magnetoresistance (TMR) ratio …”