All-solid-state batteries (ASSB) are widely viewed as a safer and potentially more energy-dense alternative to traditional lithium-ion batteries. Their performance depends strongly on how quickly ions can travel through solid electrolytes. Identifying materials that enable this rapid ion movement has traditionally required time-consuming synthesis and experimental characterization. Researchers also rely on computer simulations, but existing computational approaches often struggle to accurately model the complex and disordered behavior of ions at high temperatures.
Another major difficulty is detecting and predicting when ions move through crystals in a liquid-like manner. Standard computational techniques that attempt to calculate the properties of such dynamically disordered systems demand extremely high computing power, making large-scale studies impractical.
Machine Learning Predicts Raman Signals of Liquid-Like Ion Motion
To address these challenges, researchers developed a machine learning (ML) accelerated workflow that combines ML force fields with tensorial ML models to simulate Raman spectra. Their findings show that strong low-frequency Raman intensity can act as a clear spectroscopic indicator of liquid-like ionic conduction.
When ions move through a crystal lattice in a fluid-like way, their motion temporarily disturbs the lattice symmetry. This disturbance relaxes the usual Raman selection rules and produces distinctive low-frequency Raman scattering. These spectral signals can be directly connected to high ionic mobility.
The new approach allows scientists to simulate the vibrational spectra of complex and disordered materials at realistic temperatures with near-ab initio accuracy while significantly reducing computational cost. When applied to sodium-ion conducting materials such as Na3SbS4, the method revealed pronounced low-frequency Raman features. These signals arise from symmetry breaking caused by rapid ion transport and provide a reliable indicator of fast ionic conduction. The results also help explain earlier experimental observations and open the door to high-throughput screening for new superionic materials.
Raman Features Reveal Superionic Conductors
The researchers further tested the method using sodium-ion conducting systems. The workflow successfully identified Raman signatures linked to liquid-like ion motion. Materials that displayed strong low-frequency Raman features also showed high ionic diffusivity and dynamic relaxation of the host lattice.
By contrast, materials where ion transport occurs mainly through hopping between fixed positions did not produce these Raman signatures. This distinction highlights how Raman signals can reveal the underlying transport mechanism inside a material.
Accelerating Discovery of Advanced Battery Materials
By extending the breakdown of Raman selection rules beyond traditional superionic systems, the study provides a broader framework for interpreting diffusive Raman scattering across many classes of materials. The ML-accelerated Raman pipeline connects atomistic simulations with experimental measurements, allowing scientists to evaluate candidate materials more efficiently.
This strategy introduces a powerful new route for data-driven discovery in energy storage research. By helping researchers quickly identify fast-ion conductors, the method could accelerate the development of high-performance solid-state battery technologies.
The findings were recently published in the online edition of AI for Science, an international journal focused on interdisciplinary artificial intelligence research.