A generative diffusion model enables multi-objective on-demand inverse design of piezoelectric metamaterials

Abstract

Piezoelectric metamaterials have attracted increasing interest in areas of mechanoelectric conversion, such as robotics and medical treatment, due to their powerful performance programmability. However, how to design the metamaterial structure to achieve on-demand regulation among mutually exclusive metrics such as electrical, mechanical, and acoustic properties remains a major challenge. Here, we present a multi-objective design strategy based on latent diffusion models to achieve inverse design of piezoelectric metamaterials under different scenario requirements. This method effectively decouples the interdependencies of four different target parameters, enabling the generation of piezoelectric metamaterials that overcome the limitations of existing datasets and significantly enhance the overall piezoelectric response. By simply inputting the desired electrical, mechanical, and acoustic performance criteria, our method is able to output the ideal metamaterial structures whose properties deviate from the input targets by only 1.06% (mean absolute percentage error, MAPE). This study introduces a versatile framework for the multi-objective, on-demand inverse design of metamaterials, which not only shortens the material development cycle but also opens up new perspectives for the on-demand design of diverse functional materials.