Maha Khademi, Quantum Device Physics

Main supervisors:

Saroj Prasad Dash, Professor, Quantum Device Physics
Johan Åkerman, Professor, Institutionen för fysik (GU) 

Supervisor: Samuel Lara Avila, Senior Researcher, Quantum Device Physics

Examiner: 
Dag Winkler, Full Professor, Quantum Device Physics

Abstract: 

Over the past decade, spintronic oscillators have gained significant attention as promising candidates for unconventional computing technologies [1]. This is largely due to their high non-linearity, compatibility with CMOS technology, and their ability to synchronize both with external signals and with one another. Among these, nano-constriction spin Hall nano oscillators (NC SHNOs) — composed of a heavy metal (HM) and a ferromagnet (FM) — stand out for their ease of nano-fabrication, broad frequency tunability, and strong mutual synchronization.
In SHNOs, the HM generates a spin current via the spin Hall effect, producing a spin-orbit torque (SOT) on the FM. This torque induces magnetization auto-oscillations, which are subsequently converted into time-varying resistance through anisotropic magneto-resistance (AMR), enabling electrical readout of the output signal.
Despite their potential, SHNOs have faced a key challenge: low output power, which limits their applicability in areas such as neuromorphic computing. One plausible explanation lies in the thermal management of the nano-constriction area, where the current density is highest. To address this, using a high-thermal-conductivity substrate like silicon carbide (SiC) allows for improved heat dissipation. This enables the application of higher current to the device, which, in turn, permits the use of thicker FM layers. As a result, output power has been increased almost 100 times compared to previously reported values for a single SHNO [2].
These high-output-power SHNOs open new pathways for applications such as physical reservoir computing (PRC). Their high non-linearity with respect to electrical current and magnetic fields makes them particularly effective for transformation tasks [3]. However, prediction tasks require memory integration to achieve accurate results.
To incorporate memory into SHNO-based reservoirs, one approach is to combine them with memristor gates. Memristors can store information efficiently through their resistance states: low resistance state (LRS) or high resistance state (HRS). This enables precise tuning of the SHNO’s frequency via the gate. However, placing the gate directly on top of the SHNO can cause irreversible damage to the FM layer due to the formation of memristive filaments. To mitigate this issue, the gates are fabricated away from the NC area. This approach achieves frequency tunability that exceeds previous reports by more than threefold [4].
In conclusion, addressing SHNO challenges such as low output power and enabling memory functionality unlocks new possibilities in efficient and accurate neuromorphic computing, and brings these devices closer to practical applications.

[1] Kumar A., et al. "Mutual Synchronization in Spin-Torque and Spin Hall Nano-oscillators." Nanomagnets as Dynamical Systems: Physics and Applications (2024): 143-182.
[2] Khademi M., et al., in preparation
[3] Sud A., Kumar A., Khademi M., in preparation
[4] Khademi M., et al. "Large Non-Volatile Frequency Tuning of Spin Hall Nano-Oscillators using Circular Memristive Nano-Gates." IEEE Electron Device Letters (2023).