Role of Domain Walls on Imprint and Fatigue in HfO2-Based Ferroelectrics

TL;DR

This paper develops a machine learning model to understand domain wall dynamics in HfO2 ferroelectrics, revealing how electric field orientation influences fatigue and imprint phenomena, thereby guiding improved device performance.

Contribution

It introduces a novel machine learning approach for modeling HfO2 ferroelectrics and uncovers the atomic mechanisms behind fatigue and imprint effects, proposing strategies to mitigate them.

Findings

Fatigue and imprint are linked to E-path and T-path switching pathways.

Inclined electric fields can suppress fatigue and improve device performance.

The model accurately predicts interatomic potentials and Born effective charges.

Abstract

HfO2-based ferroelectric materials are promising for the next generation of memory devices, attracting significant attention. However, their potential applications are significantly limited by fatigue and imprint phenomena, which affect device lifetime and memory capabilities. Here, to accurately describe the dynamics and field effects of HfO2, we adopt our newly developed DREAM-Allegro network scheme and develop a comprehensive machine-learning model for HfO2. Such model can not only predict the interatomic potential, but also predict Born effective charges. Applying such model, we explore the role of domain dynamics in HfO2 and find that the fatigue and imprint phenomena are closely related to the so-called E-path and T-path switching pathways. Based on the different atomic motions in the two paths, we propose that an inclined electric field can sufficiently suppress fatigue and enhancing the performance of HfO2-based ferroelectric devices.