Investigating Ionic Diffusivity in Amorphous LiPON using Machine-Learned Interatomic Potentials.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Electrochemistry

Background:

• Lithium phosphorus oxynitride (LiPON) is crucial for thin-film solid-state batteries due to its amorphous solid electrolyte properties.
• Modeling Li+ transport in amorphous LiPON and across Li||LiPON interfaces is computationally challenging due to material complexity and scale requirements.

Purpose of the Study:

• To develop and validate a machine-learned interatomic potential (MLIP) for accurate simulation of LiPON.
• To investigate Li+ transport in bulk LiPON and across Li||LiPON interfaces using the developed MLIP.

Main Methods:

• Trained a neural equivariant interatomic potential (NequIP) framework using a large dataset of 13,454 density functional theory (DFT) structures.
• Validated the MLIP's accuracy with low energy and force errors, comparable to DFT.
• Utilized the trained potential for molecular dynamics simulations of bulk LiPON and Li||LiPON interfaces.

Main Results:

• Generated amorphous LiPON structures consistent with ab initio molecular dynamics, showing nitrogen incorporation.
• Simulated Li+ diffusivity in bulk LiPON, showing good agreement with existing literature.
• Observed Li+ transport across Li(110)||LiPON and Li(111)||LiPON interfaces to be one order of magnitude slower than in bulk phases, but with minor anisotropy and no significant impedance buildup.

Conclusions:

• Machine-learned potentials, specifically NequIP, are highly effective for high-fidelity, large-scale modeling of complex amorphous materials like LiPON.
• The developed MLIP enables efficient investigation of Li+ transport mechanisms in LiPON-based systems.
• Results suggest minimal interfacial impedance, supporting the use of LiPON in advanced thin-film devices.