AI develops battery using 70% less lithium

Employing an AI algorithm, scientists eliminated unstable materials and those exhibiting weak chemical reactions critical for battery operation
The image shows a battery. — Unsplash
The image shows a battery. — Unsplash

Using the power of artificial intelligence, researchers have expedited the exploration and testing of novel materials, leading to the creation of a battery that relies less on the expensive mineral lithium.

Lithium-ion batteries serve as the energy source for numerous everyday devices and electric vehicles. They're essential for a sustainable electric grid, storing renewable energy generated from sources like wind turbines and solar panels. However, the expense and environmental impact of lithium mining makes finding an alternative crucial. 

Traditionally, this pursuit is costly and time-consuming, demanding years of research and testing millions of potential candidates. Yet, Nathan Baker and his team at Microsoft achieved this feat in mere months using AI. They engineered a battery design consuming up to 70% less lithium than certain existing models.

The researchers concentrated on developing a battery composed entirely of solid components, focusing on the electrolyte — a vital part that facilitates the movement of electric charges. They began with 23.6 million material variations, altering the structure of established electrolytes and substituting lithium atoms with other elements.

Employing an AI algorithm, they eliminated unstable materials and those exhibiting weak chemical reactions critical for battery operation. Additionally, they evaluated how these materials would perform during battery operation. Within a short span, they narrowed down the list to a few hundred candidates, some previously unexplored.

Despite lacking expertise in material science, Baker sought insights from specialists involved in large-scale battery projects. One such expert, Vijay Murugesan from the Pacific Northwest National Laboratory, collaborated with the team, enhancing the AI's screening criteria. After additional rounds of elimination, Murugesan's team synthesized a suggested material in the lab. This material notably replaced half the anticipated lithium atoms with sodium, posing questions about its unique behaviour within a battery.

The resultant battery, while exhibiting lower conductivity than lithium-rich prototypes, was functional enough to power a light bulb within approximately nine months from the project's inception. 

Rafael Gómez-Bombarelli from the Massachusetts Institute of Technology praised the project for bridging the gap between AI predictions and practical experiments, emphasising the challenge of securing investments for experimental verification. He highlighted the cutting-edge AI tools' role in accelerating and enhancing calculations rooted in decades of physics, acknowledging potential obstacles due to limited data availability and the complexity of materials beyond battery components.