AI-Driven Customized Material Design: Bridging The Gap in The Chemical Space for Targeted Functionalities

Due to the extensive breadth of the chemical space, the endeavor of discovering materials with specific functionalities proves to be a formidable challenge. The infinite combinations of chemical composition and crystalline structures contribute to a vast realm of unexplored materials. In light of these challenges, there is a pressing demand for AI models capable of designing materials tailored to specific needs. Our work primarily addresses this demand by introducing an adaptive material encoding scheme, which facilitates the customized design of materials based on varying functional requirements. By adopting this encoding scheme with AI models, we have successfully trained a system that achieves benchmark accuracy in material properties prediction. To further exemplify the efficacy of our approach, we selected piezoelectric materials as a case study for developing novel material compositions and crystal structures that meet requirements. Leveraging cutting-edge generative models, we accomplished a material generation task with a recovery rate exceeding 90%. In our concluding efforts, we integrated material prediction and generation models to scout for potentially high-performance novel piezoelectric materials. These newly uncovered materials have undergone validation through Density Functional Theory (DFT) computations, underscoring the promising horizon of utilizing AI-guided sustainable approaches in navigating the intricate landscape of material design.