Prospects of voltage-controlled spin Hall nano-oscillators for neuromorphic computing

Mutually synchronized nano-constriction-based spin Hall nano-oscillators (SHNOs) are promising energy-efficient devices for high-quality microwave signal generation and an attractive alternative to conventional spin-torque nano-oscillators [1]. They have recently shown great potential to closely emulate the brain-inspired processes and thus open the way for ultra-fast oscillator-based neuromorphic computing [1,2]. However, interfacing of such SHNO networks and tunability of individual oscillators to perform complex tasks in large oscillator networks remain challenging tasks.

In my talk, I will describe a state-of-the-art voltage-controlled nano-constriction based W(5nm)/ CoFeB(1.7nm)/MgO(2nm) SHNOs, which combine a nano-scale footprint, CMOS compatibility, and energy-efficient fine individual oscillator control in mutually synchronized chains [3,4]. I will further discuss how voltage-induced moderate changes in perpendicular magnetic anisotropy across the CoFeB/MgO interface strongly tune the auto-oscillation frequency and spin-wave localization in the constriction region. As a result, we observe a substantial voltage modulation in threshold current (22%) and a substantial 50 MHz frequency tunability (12 MHz/V) in these SHNOs [3]. Finally, I will demonstrate how such strong voltage-tunability in both frequency and threshold current can lead to the non-volatile tuning of the synchronization state in a chain of four mutually synchronized SHNOs [4]. The demonstrated energy-efficient tuning approach is promising to train large SHNO networks for cognitive tasks and scale oscillator-based neuromorphic computing schemes to more extensive network sizes.