Google DeepMind has developed an artificial intelligence model that has successfully predicted 2.2 million potential new inorganic crystal structures, a monumental stride that could significantly impact the production of cutting-edge microprocessors, electric batteries, solar panels, and other technologies. Among these structures, the AI has highlighted 380,000 candidates predicted to possess adequate stability for future technological applications.
The GNoME Project
Based on a graph neural network and trained on data from 69,000 known crystals, the sophisticated AI model, known as GNoME, aids researchers in identifying novel compounds that could revolutionize various industries. The software predicts the atomic structure and chemical formula of new materials, then assesses these predictions using Density Functional Theory (DFT), a computational technique widely recognized in physics, chemistry, and materials science. The active learning approach employed during development improved the model's performance, refining its predictive capability with DFT-tested feedback.
In a research paper, Google DeepMind says GNoME has assembled a list of structures with prevalent commercial potential, including 52,000 compounds structurally akin to graphene and over 500 lithium ion conductors. It has even spotlighted 15 lithium transition-metal oxides, poised to advance the manufacturing of superconductors and rechargeable batteries. The empirical validity of GNoME's predictions is corroborated by 736 of the forecasted crystals matching chemically-verified structures documented in previous experiments, with 184 of these discoveries being new.
Robotics Meets AI for Material Synthesis
Translating predictions into physical reality remains a challenge, which researchers from the Lawrence Berkeley National Laboratory took head-on. They initiated a project using GNoME's outputs to select 58 inorganic compounds. A robotic system, A-Lab, manufactured these compounds based on instructions generated by another AI trained on academic texts. Operating continuously for 17 days, A-Lab successfully synthesized 41 out of 58 targeted materials, reaching an impressive 71 percent success rate with potential for future improvement.
While challenges persist, such as reaction times and material volatility, the collaboration between AI and robotics has proven effective. The enhancements in AI-driven synthesis methods foretell an era where new material creation will proceed at an unprecedented pace, facilitating progress in environmental and climate solutions. Innovations might include recyclable plastics, efficient energy harvesting technologies, superior batteries, and more durable, cost-effective solar panels, signaling a transformative future for material science and technology.
