Unveiling structure-property correlations in ferroelectric $Hf_{0.5}Zr_{0.5}O_2$ films using variational autoencoders

TL;DR

This paper introduces a combined phase-field and variational autoencoder framework to analyze and optimize the structure-property relationships in ferroelectric Hf0.5Zr0.5O2 films, facilitating targeted material design for nanoelectronic applications.

Contribution

It presents a novel integration of PF modeling with VAEs to encode, analyze, and inverse-design ferroelectric properties based on microstructural parameters.

Findings

VAEs effectively encode hysteresis loops into low-dimensional space

The framework enables inverse design of ferroelectric properties

Systematic exploration of material parameters improves device performance

Abstract

While $Hf_{0.5}Zr_{0.5}O_2$ (HZO) thin films hold significant promise for modern nanoelectronic devices, a comprehensive understanding of the interplay between their polycrystalline structure and electrical properties remains elusive. Here, we present a novel framework combining phase-field (PF) modeling with Variational Autoencoders (VAEs) to uncover structure-property correlations in polycrystalline HZO. Leveraging PF simulations, we constructed a high-fidelity dataset of $P-V$ loops by systematically varying critical material parameters, including grain size, polar grain fraction, and crystalline orientation. The VAEs effectively encoded hysteresis loops into a low-dimensional latent space, capturing electrical properties while disentangling complex material parameters interdependencies. We further demonstrate a VAE-based inverse design approach to optimize $P-V$ loop features, enabling the tailored design of device-specific key performance indicators (KPIs), including coercive field, remanent polarization, and loop area. The proposed approach offers a pathway to systematically explore and optimize the material design space for ferroelectric nanoelectronics.