Battery material predicted by AI shows promise in the lab

A chemical compound originally identified by artificial intelligence has now shown in lab experiments that it may be key to unlocking safer, more energy-dense batteries.  

The work, published in the American Chemical Society’s ACS Energy Letters, concludes a seven year journey by the paper’s co-author Austin Sendek, whose AI algorithms identified candidate materials that may make for more stable, longer lasting batteries. This recent paper, co-led by former Stanford postdoctoral researchers Yinxing Ma, Jiayu Wan, and Xin Xu, tests a specific compound known as “LBS” (because of its chemical descriptor Li₈B₁₀S₁₉) that can hold high levels of electricity without breaking down.

LBS “is the most stable, sulfur-based lithium-ion electrolyte that we've ever seen experimentally, as far as I know,” said Sendek, PhD ’18 and adjunct professor of material science. 

Safety vs stability

The work began because most batteries today are made with lithium-containing liquids. Despite good performance and energy density, the liquid electrolyte at the heart of these lithium-ion batteries can combust and become environmentally toxic. 

Solid electrolyte batteries are safer and store more energy, but until now known solid electrolytes had tended to be unstable. This new solid electrolyte LBS could overcome the stability problem. In studies it appears to store more electricity without breaking down or losing conductivity, even after many battery cycles – overcoming the limitations of existing lithium-sulfur electrolytes which have great conductivity but rapidly deteriorate.

From AI to experimental validation

As a graduate student in 2020, Sendek developed AI algorithms that ultimately predicted LBS might exhibit electrochemical properties that exceed existing materials.

“This is a really exciting validation point,” said Sendek. “It proves that machine learning based screening works and can actually yield real discoveries.”

Under the tutelage of the late Evan Reed, co-senior author of the recent study with Yi Cui and William Chueh, Sendek spent seven years screening lithium-containing materials to find promising candidates for solid electrolytes. From more than 12,000 known lithium-containing materials, Sendek’s algorithms identified 21 electrolyte candidates, including lithium thioborate electrolytes. Through simulations in 2018 and 2020, he and his colleagues further refined their investigation to four new lithium thioborate compounds which showed great potential as solid state battery electrolytes.

LBS is one of those candidates and its real electrochemical properties exceeded simulations, according to the new study. 

The recent validation of LBS reveals how machine learning-based screening efforts can accelerate the unearthing of useful compounds. Without AI, discovering successful new materials would have taken many more years. 

“Experimentally, making a new material and verifying its ionic conductivity takes years of work,” said Cui, senior author of the study, as well director of Stanford Precourt Institute for Energy and the Sustainability Accelerator at the Stanford Doerr School of Sustainability. “So, imagine if I had 1,000 compound candidates. Then the task would become impossible.”

Next steps

Despite being one of the best candidates for a solid electrolyte today, “it still needs to be improved,” said Cui.

The work validating LBS could broaden the types of candidate materials for solid state electrolyte. “We hope this promotes further investigations into sulfide system solid electrolytes for safer solid-state batteries,” said Chueh, professor of material sciences and co-director of Stanford’s StorageX Initiative.

Cui, Sendek and collaborators plan to continue refining the LBS compound. The team is working on a model of LBS with added ingredients that may improve the conductivity of the electrolyte in the hope of  increasing the material’s stability window and ionic conductivity, the scientists said.

Reed, a pioneer of computational material science, passed away in March 2022. He was the senior author on the previous studies led by Sendek.

Other members of the Stanford team not mentioned are: Sarah E. Holmes, Brandi Ransom, Zhelong Jiang, Pu Zhang, Xin Xiao, Wenbo Zhang, Rong Xu,, Yusheng Ye and Emma Kaeli.

This work was supported by the Office of Vehicle Technologies of the U.S. Department of Energy, the China Scholarship Council, and SLAC National Accelerator Laboratory