# Manipulating transient SOT-MRAM switching dynamics for efficiency improvement and probabilistic switching

## TL;DR

This paper explores how manipulating transient dynamics in SOT-MRAM devices can improve switching efficiency and enable probabilistic computing.

## Contribution

The novel contribution is identifying a quasi-stochastic regime and demonstrating how to use transient phenomena to lower switching currents and enable probabilistic operation.

## Key findings

- Minor changes to device geometry and material properties can significantly reduce switching currents in SOT-MRAM.
- At higher temperatures, the quasi-stochastic regime becomes a tunable probabilistic transition band.
- The framework enables the creation of a fast and efficient probabilistic bit for computing.

## Abstract

In this paper, we investigate the effect of transient dynamics in the switching process for spin-orbit torque magnetic random-access memory (SOT-MRAM) devices stabilized by in-plane uniaxial magnetocrystalline anisotropy. We develop theory for the interaction between spin torques and effective fields during a magnetization write trajectory and apply this framework to find regions of failed and successful switching. We focus particularly on a “quasi-stochastic” regime located between regions of deterministic failed and successful switching and caused by the interplay between torque-driven and precession-driven magnetization evolution during the switching process. We demonstrate a series of minor alterations to device geometry, material characteristics, and electrical inputs that use transient phenomena to lower the switching barrier—thereby allowing for SOT-MRAM switching with significantly lower currents and faster write speeds than the traditional architecture. Furthermore, we demonstrate that at elevated temperatures, the unpredictable stochastic regime evolves into a probabilistic “transition band” with clearly defined, montotonic, and tunable regions of probabilistic operation. Through this addition of control mechanisms through electrical inputs, our framework paves the way for the creation of a fast, efficient probabilistic bit (p-bit) for the field of probabilistic computing.