Learn how an innovative new AI model is helping researchers unlock the structure of powdered crystalline materials, accelerating progress within batteries, magnets and renewable energy industries.
Transforming Crystal Structure Analysis
X-ray crystallography has been used for over a century to determine the structure of crystalline materials. But it only works where a chunk of crystal is still there. The job becomes especially hard when the materials are in powdered forms (the only versions available to scientists), because fragments can be randomly oriented, leaving it up to them to painstakingly piece together the whole structure one small bit at a time.
Now, researchers at MIT have designed an innovative generative AI model that makes it much easier to identify the structures of these powder crystals. This system, which they named Crystalyze, is trained on a massive database of materials and can output several potential structures for a given X-ray diffraction pattern, helping researchers to more easily identify the true structure. Long-term, the researchers believe this methodology may be useful to the field of materials science more broadly, where understanding how atomistic structures lead to properties and function is critical.
Revealing the inner secrets of 2D crystalline materials
For instance, the crystalline materials (like metals and virtually all other inorganic solida) are made up of many identical repeating units lattices. Think of these units of measuring or ‘boxes’ that have a very certain shape to them and each different type has atoms placed exactly within these boxes. When X-rays are beamed onto these lattices, they are scattered off the atoms in particular directions and intensities on making the details on atomic positions and bonds between them visible.
When the materials are only available as powdered crystal, solving their structures becomes more difficult. But those grains are not the original 3D crystal, and so EBSD software cannot detect the full structure of the lattice. That is where the new AI model comes in.
Crystalyze is trained on the Materials Project, 150k+ materials containing database. The model breaks the prediction of structures down into several subtasks: predicting the size and shape of the lattice “box,” what atoms will go in that box, and how those atoms are arranged within the box. The model can create many candidate structures and subsequently “fit” them to an input X-ray diffraction pattern to correctly identify the structure of even uncharacterized materials.
Conclusion
The Crystalyze AI model is an essential advancement in the field of materials science. This has the potential to accelerate the discovery and characterization of new materials with unique properties, by allowing researchers to more readily determine the structures of powdered crystalline materials. From battery technologies to permanent magnets, every day new materials are discovered and if we understand their underlying structures its will greatly help us in innovating new ideas and advancing with profound applications. With further development of the model and expansion to more materials, new discoveries in materials science result.
