What is it about?
We propose a scheme to optimize the parameters of a macrospin-type spin-torque oscillator using the gradient descent method with automatic differentiation. First, we numerically created time series data for teaching and tuned the physical parameters of the oscillator to reproduce the dynamics. Next, we implemented a high-accuracy image recognition by connecting a coupled spin-torque oscillator system to the input and output layers and training the entire network through gradient descent.
Featured Image
Photo by Alexandre Debiève on Unsplash
Why is it important?
Optimizing physical parameters serves various purposes in developing devices, including system identification. Spin-torque oscillators have been experimentally and theoretically applied to neuromorphic computing. However, their physical parameters are usually optimized via grid search procedures. We developed an efficient parameter optimization method using gradient-based optimization. Combining our gradient-based approach with experimentation enables the design of an experimental setup and physical system to solve a task with high precision.
Perspectives
Optimizing the parameter values of a physical system is an important issue for many researchers. For example, some researchers may want to find the parameter values of a physical system that maximizes a desired physical effect, while others may want to find the parameter values of a physical system that maximizes computational power. We hope that this article will inspire many researchers who are interested in optimizing the parameter values of physical systems.
Yusuke Imai
The University of Tokyo
Read the Original
This page is a summary of: Gradient-based optimization of spintronic devices, Applied Physics Letters, February 2025, American Institute of Physics,
DOI: 10.1063/5.0238687.
You can read the full text:
Contributors
The following have contributed to this page
