Traditionally, scientists have discovered materials through a combination of intuition, plucking, and simple luck. The approach led to many dead ends, but it also spawned important innovations, including the components of the lithium-ion battery that Crabtree calls “the best battery we’ve ever had”.

From the late 20th century onwards, computers allowed scientists to simulate the structures and properties of molecules and materials and only synthesize the most promising ones, saving time and money. The high throughput screening also enabled tens or hundreds of connections to be tested in quick succession.

But these methods still have one major limitation, namely the imagination and understanding of scientists. The landscape of possible arrangements of elements in a molecular or crystal lattice is unimaginably large, and scientists have only examined fragments of this landscape. “We barely scratched the surface,” says Venkat Viswanathan, a materials scientist at Carnegie Mellon University in Pennsylvania.

Enter artificial intelligence. AI methods like machine learning can sound out huge data sets for subtle correlations or trends that humans may elude. One popular application is face recognition, where algorithms analyze millions of photos to learn how to recognize subtle facial features that identify a person.

In a material context, a machine learning algorithm can be trained to look for arrangements of atoms that yield a desired material property or function. Maintaining these structure-function correlations could enable “inverse design” – a long-held goal of materials science, in which the optimal material for a selected function can be found.

There is one problem, however: Materials science lacks the massive training datasets available for face recognition and other popular AI applications. “Materials science is inherently very data-poor,” says Kristin Persson, a physicist at Lawrence Berkeley National Laboratory in California who leads the Materials Project, the world’s largest public materials database. The synthesis and complete characterization of a material is laborious and time-consuming, which is why material databases often contain at most a few hundred entries for a certain material property.

But with ever-increasing material databases and new methods that can generate large amounts of data quickly, many researchers believe that materials science is ripe for an AI revolution. Climate change could be an area where this revolution is producing great results.

John Dagdelen / Lawrence Berkeley Laboratory

By applying machine learning to material project data, researchers can localize new materials for specific applications. One example is this newly discovered phosphor material that could potentially be integrated into energy-efficient solid-state lighting.By applying machine learning to material project data, researchers can localize new materials for specific applications. One example is this newly discovered phosphor material that may be converted into energy efficient solids … show more

LEAVE A REPLY

Please enter your comment!
Please enter your name here