AI and materials: How artificial intelligence is turning data into breakthrough smart materials?

Evolution of AI-related publications and patents

Over the last decade, AI has seen exponential growth. Benefiting from strong research dynamic and advancements in technology and data availability, the number of scientific publications and patents in this field has witnessed a dramatic surge in the last 10 years (CAGR 20% and 30% respectively). The United States and China are leading the AI race with significant contributions from academic institutions such as the Chinese Academy of Science, the Massachusetts Institute of Technology (MIT) or the Carnegie Mellon University. Moreover, due to the huge potential and the bright future it has ahead of it, numerous consolidated enterprises and start-ups are entering the field, further driving innovation and highlighting AI’s vast potential across different industries.

Improvement of the material discovery process using artificial intelligence

The material discovery process is a comprehensive approach used to identify and develop new materials with desirable properties for various applications. This process, which can be decomposed in different stages, found some limitations in the past due to the use of traditional methods that limited the speed of the discovery cycle. However, with the advent of AI, the whole cycle is accelerated and enriched at all stages, leading to an unprecedent speeding of the discovery process.

AI for the identification of future trends

Thanks to algorithms, Natural Language Processing (NLP), text-mining or techniques like Latent Dirichlet Allocation (LDA), AI-tools help to analyze market trends, competitor activities, customer reviews, economic indicators… to identify emerging requirements, future trends and new questions based on needs.

AI for literature review and background research

Thank to NLP or machine learning algorithms, AI-powered tools like ChemDataExtractor, tmChem or IBM DeepSearch can work with overwhelmingly vast databases that are fed with historical scientific knowledge (patents, papers, reports, …) and are continuously updated.

These tools can ingest and analyze all the data to extract and identify relevant information, key concepts, patterns, correlations… and even summarize the findings to provide a comprehensive overview of existing knowledge. So far, this stage has been one of the most benefited, and two different approaches could be identified:

• Open source: The AI-powered tools exclusively look for information in the existing public literature (ex. Citrine informatics with HRL Laboratories).
• Open source + internal information: The AI-powered tools are complementary fed with internal data provided by the final client (ex. Chemintelligence)

AI for hypothesis formulation

Based on the existing data previously analyzed and theoretical frameworks, AI-driven analytics can identify plausible relationships between materials, properties and other variables, and propose hypothesis that expand the discovery space.

AI for the experimentation of new materials

• Prediction: AI-trained models such as neural networks and regression algorithms, can predict the properties and behavior of new materials based on several variables like their composition or structure. Then, AI can rapidly screen vast libraries of potential material compositions, predicting their properties and identifying promising candidates for further study.
• Simulation: Running AI-driven material simulations at different scales (atomic, molecular, macro) can challenge the previously stated predictions by modeling its behavior under various conditions. Then, the AI can recover and analyze the simulation data, and thanks to generative models like GAN (Generative adversarial Network) or VAEs (variation Autoencoders) propose new material structures that are likely to have the desired properties. This allows to create a close loop that can be fed until a desired number of solutions are predicted that meet the requirements established.
• Test: Once the predicted solution complies with the set characteristics that are sought, the AI can conduct virtual experiments to test material properties under various simulated conditions. This will allow to optimize the design of real-life physical experiments (DoE) by selecting the most informative and efficient set of evaluations to perform, maximizing the information gained.

AI for the data analysis of the performance of new materials

AI-powered tools allow data preprocessing and cleaning. These tools can perform different types of analysis (statistical, multivariate, real-time…) and even be integrated with simulation and experimentation models, providing immediate feedback and insights that allow to optimize the experimental conditions dynamically.

AI for the creation of customized results reports

AI, thanks to NLP can generate and audience-customized reports that summarize large volumes of experimental data, include interactive and advanced visualizations (3D plots, heatmaps…), and even integrate insights from textual data such as lab notes or research articles.

Finally, since the potential of the AI can be maximized with the knowledge of a human expert on the field by determining some necessary features and fine-tuning some parameters, the researchers are still at the heart of the innovation process. The expert presence allows to enhance the chances of the AI-powered generative models to propose an array of possible materials that will have to be evaluated and synthesized in the lab.

Find out how our team can support you in your AI consulting needs >

3 examples of initiatives using AI to discover new materials

Initiative n°1: A-lab, an innovative lab powered by Berkeley focused on the synthesis of novel materials

The Department of Materials Science and Engineering of BerkeIey developed the A-Lab, an autonomous laboratory that leverages artificial intelligence (AI) to significantly accelerate the synthesis of novel materials, and close the gap between the rates of computational screening and experimental realization of novel material.

The A-Lab, which was designed to expedite the synthesis of novel inorganic materials, synthesized 41 novel compounds from a set of 58 targets within only 17 days. By utilizing data from the Materials Project and Google DeepMind, along with natural-language models trained on historical literature, the autonomous laboratory generated and optimized synthesis recipes using active learning grounded in thermodynamics.

The A-Lab demonstrated the transformative impact of AI in material science, enabling faster discovery, continuous optimization, and high-throughput scalability. The integration of AI and robotics allowed autonomous, efficient, and precise synthesis processes, making the experiments reproducible with minimal human intervention.

Initiative n°2: Altrove, a French startup aiming at creating alternatives to rare earth-containing materials

Altrove is a French start-up that leverages artificial intelligence (AI) to accelerate the discovery and creation of new materials that are crucial for technologies like EV or advanced electronics. They are particularly focused on creating alternatives to rare earth-containing materials, since to this date, the rare earth elements are surrounded by many geopolitical and economic challenges.

The company addresses the complex challenges of material science by not only predicting potential new material, but also developing precise recipes for their synthesis. Its approach involves using artificial intelligence models to predict stable inorganic materials, followed by automated lab processes to synthesize, test and optimize these materials.

Initiative n°3: Citrine informatics, a US enterprise using machine learning algorithms to support scientists in material discovery

Citrine Informatics is a US enterprise focused on the material industry that applies artificial intelligence and data-driven methods to accelerate materials discovery and development. It utilizes advanced machine learning algorithms to analyze vast datasets related to materials properties, compositions, and performance. Thanks to their four different products, they can create an AI-powered ecosystem that enables scientists and engineers to make data-driven decisions more effectively and efficiently.

Citrine’s AI-driven platforms have already been successfully implemented by several entities across the material industry to optimize their R&D efforts.

Case study examples:

1. New materials: HRL Laboratories partnered with Citrine informatics to accelerate the development of a 3D printable Aerospace-Grade alloy. HRL laboratories wanted to find nanoparticles that nucleate a microstructure less prone to hot cracking. Given the specific properties sought, Citrine AI-powered software employed classical nucleation theory, lattice spacing rules, thermodynamic stability, alongside materials informatics, to efficiently explore 11.5 million combinations of powders and nanoparticles, of which 100 promising candidates were identified. Finally, the resulting material, AL 7A77 was commercialized within two years from the beginning of the project, which represent a drastic decrease in time when compared to classic R&D procedures.
2. Improved properties: A global leader in specialty chemicals and plastics turned to Citrine informatics to improve their capacity to dynamically respond to one customer’s requirements. Their challenge was to increase the mechanical properties of a glass fiber reinforced polymer, while maintaining the rest of its property profile. The Citrine Platform was updated with customer test data and recipe information from its portfolio, and its AI models were retrained through a process call Sequential Learning. At the end, among the trillions of potential candidates, 10 experimental candidates were proposed that improved the performance of the previous material on an average 21%.

In conclusion, the integration of artificial intelligence (AI) into materials science is revolutionizing the R&D landscape of novel materials. AI-driven methodologies facilitate the rapid analysis of extensive datasets, predictive modeling of material properties, and optimization of experimental protocols, thereby accelerating the innovation cycle.

However, despite these advancements, the journey towards fully realizing AI’s potential in R&D is still ongoing, and there remain challenges in data integration, algorithm development and translation of AI knowledge into practical applications, among others. If you have a project related to IA and would like to discuss it with our team, don’t hesitate to contact us!

About the author,

Vincent, Director of Alcimed’s Chemicals and Materials Business Unit in France