材料探索のためのグラフネットワーク (GNoME)

AI Tool GNoME Discovers Millions of New Crystals with Potential for Future Technologies

In a groundbreaking research published in Nature, scientists Amil Merchant and Ekin Dogus Cubuk introduce the Graph Networks for Materials Exploration (GNoME), an AI tool that has discovered 2.2 million new crystals with potential applications in various technologies. These crystals, equivalent to nearly 800 years’ worth of knowledge, have the potential to revolutionize industries such as computer chips, batteries, solar panels, and electric vehicles.

Crystals are essential components of modern technologies, but finding stable crystals for new applications can be a time-consuming and expensive process. GNoME addresses this challenge by using deep learning techniques to predict the stability of new materials. With GNoME, the number of technologically viable materials known to humanity has multiplied, with 380,000 of the predictions being the most stable and promising candidates for experimental synthesis.

The potential impact of GNoME’s discoveries is significant. Among the stable materials predicted by GNoME are compounds similar to graphene, which have the potential to revolutionize electronics with the development of superconductors. Previously, only about 1,000 such materials had been identified, but GNoME has discovered 52,000 new layered compounds with similar properties. Additionally, GNoME has found 528 potential lithium-ion conductors, 25 times more than previous studies, which could greatly improve the performance of rechargeable batteries.

The research team has made GNoME’s predictions available to the research community and is contributing 380,000 predicted stable materials to the Materials Project, an online database. By sharing these resources, the team hopes to drive forward research into inorganic crystals and unlock the potential of machine learning tools in guiding experimentation.

GNoME uses two pipelines to discover stable materials: the structural pipeline, which creates candidates with structures similar to known crystals, and the compositional pipeline, which follows a more randomized approach based on chemical formulas. The predictions from both pipelines are evaluated using established Density Functional Theory calculations, which are used to assess the stability of crystals.

During the training process, GNoME’s performance was dramatically boosted using a technique called active learning. GNoME generated predictions for the structures of novel, stable crystals, which were then tested using computational techniques known as Density Functional Theory. The resulting high-quality training data was fed back into the model training, improving its accuracy and efficiency.

The impact of GNoME’s predictions goes beyond theoretical discoveries. External researchers have independently created 736 of the new materials identified by GNoME, demonstrating the accuracy of the model’s predictions. In collaboration with Google DeepMind, a team of researchers at the Lawrence Berkeley National Laboratory has also shown how GNoME’s predictions can be leveraged for autonomous material synthesis.

The potential applications of GNoME’s discoveries are vast. By providing scientists with a catalog of promising candidate materials, GNoME aims to drive down the cost of discovering new materials and accelerate the development of greener technologies. With the help of AI tools like GNoME, the future of materials discovery and synthesis looks promising.

Disclaimer: This blog post is a summary and interpretation of the original research article. The views and opinions expressed in this blog post are those of the author and do not necessarily reflect the official policy or position of the original authors or organizations.

注意

  • この記事はAI(gpt-3.5-turbo)によって自動生成されたものです。
  • この記事はHackerNewsに掲載された下記の記事を元に作成されています。
    Graph Networks for Materials Exploration (GNoME)
  • 自動生成された記事の内容に問題があると思われる場合にはコメント欄にてご連絡ください。

コメントする