AI-driven inverse design of materials: Past, present and future

TL;DR

The paper surveys AI-driven inverse design of functional materials, tracing its evolution from experiment- and theory-driven paradigms to AI-enabled data-driven discovery. It surveys progress across superconductors, magnetic and thermoelectric materials, carbon-based nanomaterials, 2D materials, photovoltaics, catalysts, HEAs, and porous materials, highlighting methods such as graph neural networks, diffusion models, generative architectures, and large language models (LLMs) like FlowLLM. It discusses datasets, evaluation benchmarks, and successful closed-loop experiments that couple AI predictions with experimental validation. Finally, it identifies outstanding challenges—stability and relaxation of generative designs, conditional generation, amorphous materials, and dataset/benchmark gaps—and sketches future directions including autonomous design with LLMs and multimodal data integration.

Abstract

The discovery of advanced materials is the cornerstone of human technological development and progress. The structures of materials and their corresponding properties are essentially the result of a complex interplay of multiple degrees of freedom such as lattice, charge, spin, symmetry, and topology. This poses significant challenges for the inverse design methods of materials. Humans have long explored new materials through a large number of experiments and proposed corresponding theoretical systems to predict new material properties and structures. With the improvement of computational power, researchers have gradually developed various electronic structure calculation methods, such as the density functional theory and high-throughput computational methods. Recently, the rapid development of artificial intelligence technology in the field of computer science has enabled the effective characterization of the implicit association between material properties and structures, thus opening up an efficient paradigm for the inverse design of functional materials. A significant progress has been made in inverse design of materials based on generative and discriminative models, attracting widespread attention from researchers. Considering this rapid technological progress, in this survey, we look back on the latest advancements in AI-driven inverse design of materials by introducing the background, key findings, and mainstream technological development routes. In addition, we summarize the remaining issues for future directions. This survey provides the latest overview of AI-driven inverse design of materials, which can serve as a useful resource for researchers.

Figures (6)

• Figure 1: Trends in publications and citations in the field of "Machine Learning Materials Discovery" from 2019 to 2024. The left panel (a) illustrates the growth in the number of publications (represented by blue bars) alongside the total citations (depicted by the red line with markers), reflecting a significant increase in both metrics over the past few years. The right panel (b) presents the variation in the total known materials over time on a logarithmic scale, highlighting the acceleration of material discovery processes facilitated by GNoME merchant2023scaling of Google and OMat24 OMat24 of Meta. This figure underscores the rapid development within the field of machine learning materials discovery
• Figure 2: Materials Science Research Paradigms This figure illustrates the evolution of research paradigms in materials science, emphasizing key milestones across various approaches. It highlights the increasing role of artificial intelligence in driving future materials discovery, with AI becoming the dominant force in shaping the field. The experiment-driven paradigm is exemplified by Marie Curie’s Nobel Prize-winning discovery of radium and polonium. The theory-driven paradigm is represented by Dirac's equation, which predicted the existence of the positron and the quantum spin Hall effect. The computation-driven paradigm is demonstrated through DFT calculations applied to materials such as MgB$_2$, hydride superconductors, and graphene. Finally, the AI-driven paradigm showcases recent breakthroughs in artificial intelligence, including AlphaFold2, GPT-4, and AI-accelerated materials discovery, signaling the frontier of research in the field.
• Figure 3: AI-driven discovery of materials. This figure illustrates the role of artificial intelligence in accelerating the discovery of various types of materials. The examples presented highlight how AI-driven approaches are employed to optimize the identification, design, and property prediction of materials across diverse categories, including superconducting materials, magnetic materials, thermoelectric materials, carbon-based nanomaterials, 2D materials, photovoltaic materials, catalyst materials, high-entropy alloys, and porous materials. By leveraging large datasets and advanced computational techniques, AI methods facilitate more efficient screening and prediction, thereby significantly advancing the pace of material discovery. These examples represent only a subset of the broad potential of AI in transforming materials research, with further discussion provided in the main text.
• Figure 4: The rapid advancement of AI technologies accelerates materials discovery in various ways. Blue indicates traditional machine learning; green represents the development of invariant GNNs and their application in discriminative AI for predicting material properties; deep blue denotes the progress of equivariant GNNs, generative AI techniques, and their integration for structural predictions. Orange signifies the role of LLMs in expediting materials discovery.
• Figure 5: The general process of materials inverse design, which is divided into two main parts: the AI-driven theoretical calculation part and the experimental validation part. The theoretical calculation part is further subdivided into material design and generation, high-throughput screening, and computational modeling.