A newly developed virtual tool is designed to enhance the extraction of meaningful information from characterization technique data and effectively guides the screening of target battery materials based on functional requirements.

Efficient characterization of battery materials is fundamental to understanding the underlying electrochemical mechanisms and ensuring the safe operation of batteries. In this work, an innovative data-driven multimodal generative method is proposed to accelerate the characterization and screening of battery materials. This approach leverages a variant of the latent diffusion model, which combines a variational autoencoder (VAE) and a denoising U-Net. The VAE maps microscale information from characterization techniques, such as atomic force microscopy (AFM), into a common latent space, and the denoising U-net, conditioned on battery properties, guides the screening of battery materials. Together, the data-driven properties of material space, enriched with battery functional properties and formulated in a common latent space, achieve the accurate translation of information from AFM to meaningful material descriptors and accelerate the screening of battery materials to meet the functional needs of the battery system under consideration.