Spin Wave Computing using pre-recorded magnetization patterns

TL;DR

This paper introduces a spin wave computing device utilizing a bilayer structure with a pre-recorded magnetization pattern to perform complex computations like Fourier Transform and Grover search more efficiently than traditional methods.

Contribution

It presents a novel spin wave computing architecture that uses a bias layer to control spin wave interference, enabling high-performance computational operations.

Findings

Demonstrated Fourier Transform, Vector-Matrix multiplication, and Grover search using the device.

Achieved operational parameters surpassing conventional designs by orders of magnitude.

Showed that adjusting the bias layer magnetization controls spin wave behavior.

Abstract

We propose a novel type of a spin wave computing device, based on a bilayer structure which includes a bias layer, made from a hard magnetic material and a propagation layer, made from a magnetic material with low damping, for example, Yttrium Garnet (YiG) or Permalloy. The bias layer maintains a stable pre-recorded magnetization pattern, generating a bias field with a desired spatial dependence, which in turn sets the equilibrium magnetization inside the propagation layer. When an external source applies an RF field or spinwave to the propagation layer, excited spin waves scatter on the magnetization's inhomogenuities, resulting in a complex interference behavior. One thus has the ability to adjust spin wave propagation properties simply by altering the magnetization in the bias layer. We demonstrate that the phenomenon can be utilized to perform a variety of computational operations, including Fourier Transform, Vector-Matrix multiplication and Grover search algorithm, with the operational parameters exceeding conventional designs by orders of magnitude.