Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
Machine learning for materials discovery and optimization
Discovering new materials with customizable and optimized properties, driven either by specific application needs or by fundamental scientific interest, is a primary goal of materials science. Conventionally, the search for new materials is a lengthy and expensive manual process, frequently based on trial and error, requiring the synthesis and characterization of multiple compositions before a desired material can be found. In recent years this process has been greatly improved by high-throughput approaches. Advances in artificial intelligence, such as machine learning for materials science, data-driven materials prediction, automated or autonomous combinatorial synthesis, and data-guided high-throughput characterization, can now significantly accelerate materials discovery.
This collection brings together the latest computational and experimental advances in machine learning and big data-driven approaches for high-throughput prediction, synthesis, characterization and optimization of new materials.
Graph neural networks are machine learning models that directly access the structural representation of molecules and materials. This Review discusses state-of-the-art architectures and applications of graph neural networks in materials science and chemistry, indicating a possible road-map for their further development.
Machine learning is an increasingly important tool for materials science. Here, the authors suggest that its contextual use, including careful assessment of resources and bias, judicious model selection, and an understanding of its limitations, will help researchers to expedite scientific discovery.
Quantum materials host many exotic properties, which might be utilized for new electronic devices. Here, artificial intelligence for the discovery of quantum materials is discussed, covering both materials and property prediction, and high-throughput synthesis.
In data-driven approaches for materials discovery, it is essential to account for phase stability when predicting synthesizability. Here, by combining density functional theory calculations and machine learning, the authors predict the synthesizability of unreported half-Heusler compositions.
Fractionally doped perovskites oxides are interesting for their energy production and storage applications. Here, the authors develop an easy-to-use machine learning model for materials discovery from limited data, identifying and synthesizing 7 new compositions relevant for solar thermochemical hydrogen production.
Predicting the synthesizability of unknown crystals is important for accelerating materials discovery. Here, the synthesizability of crystals with any given composition and structure can be predicted by a deep learning model that maps crystals onto color-coded 3D images processed by convolutional neural networks.
Experiments and simulations can reveal energetic barriers during atomic-scale growth but are time-consuming. Here, machine learning is applied to single images from kinetic Monte Carlo simulations of sub-monolayer film growth, allowing diffusion barriers and binding energies to be accurately determined.
Controlling the microstructure of thin films is vital for tuning their properties. Here, machine learning is applied to obtain synthesis-composition-microstructure relationships in the form of structure zone diagrams for thin films, enabling microstructure prediction.
High-throughput prediction and synthesis are vital for obtaining new materials that deviate from existing compositions. Here, machine learning is combined with high-throughput synthesis to identify superionic conductors based on Ca-(Nb,Ta)-Bi-O.
Biodegradable polyhydroxyalkanoates are promising replacements for non-degradable plastics. Here, neural network property predictors are applied to a search space of approximately 1.4 million candidates, identifying 14 polyhydroxyalkanoates that could replace widely used petroleum-based plastics.
Collagen is known to play a key role in the fracture resistance of bone. Here, in situ synchrotron tomography during the mechanical testing of bone is combined with deep learning to mitigate radiation damage, revealing that a compromised collagen network lowers the efficacy of crack deflection.
Characterizing fission products in uranium dioxide nuclear fuel is important for predicting its long-term properties. Here, machine learning is used to mine microscopy images of precipitates and nanoscale gas bubbles in high-burn-up fuels, providing detailed structural insight of nanoscale fission products.
High-throughput materials discovery can reduce the time taken to identify high-performing materials. Here, compositionally-graded films are fabricated in a binary halide perovskite system, of interest for solar cells, and their stability investigated during artificial aging.
Graph neural networks are an accurate machine learning-based approach for property prediction. Here, a geometric-information-enhanced crystal graph neural network is demonstrated, which accurately predicts the formation energy and band gap of crystalline materials.
In glass formation, the dynamics of extended structures beyond atomic short-range order is yet to be understood. Here, persistent homology, combined with machine learning, reveals superstructures made of 3-to-9 prism-type atomic clusters which undergo drastic changes according to the glass cooling rate.
The discovery of new alloys with desirable mechanical properties is traditionally a time consuming process. Here, machine learning is applied to the discovery of aluminum alloys, revealing a compositionally-lean alloy with an ultimate tensile strength of 952 MPa and 6.3% elongation.
Traditionally, machine learning for materials science is based on database-specific models and is limited in the number of predictable parameters. Here, a versatile graph-based neural network can integrate multiple data sources, allowing the prediction of more than 40 parameters simultaneously.
Brick-and-mortar composite structures found in nature are known for their superior mechanical performance. Here, computational approaches are used to understand the key design features that control mechanical behavior, providing guidance for the design of improved composites.
Designing and understanding quantum materials requires continuous feedback between experimental observations and theoretical modelling. Here, a machine learning scheme integrates experiments with theory and modelling on experimental timescales for extracting material parameters and properties of Dy2Ti2O7 spin-ice under pressure.
Artificial intelligence may significantly accelerate the discovery of new materials but is not easily applicable to non-periodic structures. Here, a deep learning framework is proposed to predict properties of tangible carbon nanotubes by generating virtual structures at different scales and compositions.
Finding materials with large magnetization is highly desirable for technological applications. Here, a machine learning autonomous search and automated combinatorial synthesis reveal that multi-element alloys with Ir and Pt impurities have a magnetization exceeding the Slater-Pauling limit of Fe75Co25.