Diffusion-Model-Driven Discovery of Ferroelectrics for Photocurrent Applications.

Area of Science:

• Materials Science
• Solid State Physics
• Computational Chemistry

Background:

• Ferroelectric materials are crucial for advanced memory and photovoltaic devices.
• The discovery of novel ferroelectrics is limited, hindering technological progress.

Purpose of the Study:

• To develop a computational framework for discovering new ferroelectric materials.
• To identify novel ferroelectric candidates with high polarization and potential for applications.

Main Methods:

• Utilized a diffusion-model-based crystal generation framework (MatterGen) to create 12,800 candidate structures.
• Employed a multi-fidelity screening pipeline integrating machine learning and density functional theory (DFT) calculations.
• Performed HSE06 calculations to determine bandgaps and assess photovoltaic potential.

Main Results:

• Identified two promising ferroelectric candidates: Ca3P2 and LiCdP.
• Ca3P2 exhibits a polarization of 8.9 µC cm⁻², and LiCdP shows a high polarization of 144.1 µC cm⁻².
• Calculated bandgaps of 1.58 eV for Ca3P2 and 1.13 eV for LiCdP, indicating suitability for photocurrent applications.

Conclusions:

• The study presents Ca3P2 and LiCdP as new, promising ferroelectric materials.
• The findings highlight the efficacy of generative models and computational screening in accelerating materials discovery.
• The identified materials hold significant potential for next-generation memory and photovoltaic technologies.