Researchers have conducted a systematic review of advanced artificial intelligence algorithms and hardware acceleration techniques applied to material structure design, highlighting significant progress in material property prediction, material structural optimization, material discovery, information extraction in material text. Published in AI & Materials, this review provides crucial guidance for accelerating data-driven materials research and fostering the interdisciplinary development of next-generation functional materials.
Artificial intelligence and hardware acceleration technologies are profoundly transforming the research paradigm of material structure design. Traditional experimental trial-and-error methods and theoretical computational approaches often face limitations when confronting large-scale, complex material systems, constrained by lengthy development cycles and prohibitive computational costs. Addressing this challenge, Professor Shiyu Du from China University of Petroleum (East China), in collaboration with researchers from China University of Petroleum (East China), University of Electronic Science and Technology, University of Colorado Denver, Tongji University and Milky-Way Sustainable Energy Ltd., has conducted a systematic review of advanced deep learning algorithms and hardware acceleration techniques applied to material structure design, providing crucial guidance for accelerating data-driven materials research.
"Deep learning has far outpaced the processing capacities of conventional experimental and computational approaches," explains Professor Shiyu Du, emphasizing that the data-driven paradigm of "Big Data + AI" has become the fourth paradigm for materials science development.
The research team systematically categorizes applications across four key domains. In material property prediction, deep learning models accurately forecast bandgaps, mechanical behavior, thermal conductivity, and catalytic performance, often achieving accuracy comparable to density functional theory calculations while dramatically reducing computational costs. In structural optimization, generative models and convolutional neural networks enable inverse design of metamaterials and elastic structures to achieve targeted properties. In materials discovery, Transformer-based architectures and generative adversarial networks accelerate the identification of novel molecules and crystal structures by efficiently navigating vast chemical spaces. Additionally, natural language processing techniques, particularly BERT-based domain-specific models such as MatSciBERT, now enable automated knowledge extraction from the rapidly expanding corpus of materials science literature. Professor Du notes, "Establishing sizable, high-quality, openly accessible datasets remains a critical priority for advancing deep learning research in materials, alongside the growing importance of explainable AI techniques."
The research team further examines the hardware acceleration platforms essential for deploying these computationally intensive models. Graphics processing units (GPUs) remain the workhorse for large-scale training, while field programmable gate arrays (FPGAs) offer customizable efficiency for edge deployment. Application-specific integrated circuits (ASICs) achieve peak energy efficiency, and neuromorphic chips inspired by biological neural architectures promise ultra-low-power operation for specialized applications.
This review provides a foundational reference for researchers seeking to leverage artificial intelligence in material structure design, while charting a course toward more interpretable, generalizable, and computationally efficient approaches for developing next-generation functional materials.
Zhang J, Yu J, Qiu N, Liu Y, Song Y, et al. Advanced artificial intelligence algorithms and hardware acceleration techniques applied to material structure design. AI Mater. 2026(1):0003, https://doi.org/10.55092/aimat20260003.
AI & Materials
Literature review
Not applicable
Advanced artificial intelligence algorithms and hardware acceleration techniques applied to material structure design
24-Mar-2026