ai generates revolutionary new battery design

will lithium-ion batteries soon be relics of the past?

by rick richardson
technology this week

researchers have now utilized artificial intelligence’s capacity to expedite the discovery and testing of novel materials to create a battery that is less reliant on the pricey and growingly scarce mineral lithium.

more: chatgpt is getting humanlike memory | lenovo readies new thinkbook laptop with a transparent display | microsoft moves further into nuclear development with new director | tack one launches new location tracker for children and seniors | windows is now an app for pcs, apple products | bill gates says america has a math problem | what is an heic file? | congress: tax prep companies shared private data with google, meta for years | what an a.i.-powered workforce means for accountants
exclusively for pro members. log in here or 2022世界杯足球排名 today.

lithium-ion batteries power numerous everyday items, as well as electric cars. additionally, batteries are needed to store renewable energy from solar panels and wind turbines. therefore, they would be an essential component of a green electric grid.

however, mining lithium is costly and harmful to the environment. finding a substitute for this crucial metal may be expensive and time-consuming, as millions of candidates will need to be developed and tested over several years by researchers. nathan baker and his colleagues at microsoft completed the assignment in months using ai. compared to some rival systems, they used up to 70 percent less lithium in the design and construction of their battery.

the researchers searched for novel materials for the electrolyte, the battery component that allows electric charges to flow through, focusing on solid-state batteries. initially, 23.6 million potential materials were created by modifying the composition of known electrolytes and substituting certain lithium atoms with alternative elements. researchers subsequently excluded materials that an ai system predicted would be unstable and those in which the chemical processes necessary for batteries to function would be weak. the researchers also considered the behavior of each material throughout the battery’s active operation. their list had only a few hundred applicants after only a few days, some of which had never been previously investigated.

“but we’re not material scientists,” says baker. “so, i called up some experts who’ve worked on large battery projects with the department of energy … and said, ‘what do you think? are we crazy?’”

among the scientists answering the phone was vijay murugesan of the pacific northwest national laboratory in washington state. he and his associates proposed more ai screening standards. after additional elimination rounds, murugesan’s team finally selected one of the ai’s recommendations to synthesize in the lab. it was notable because half of the sodium atoms that murugesan would have anticipated to be lithium atoms were replaced instead. according to him, this is a unique electrolyte composition, and combining the two components raises some interesting concerns regarding the fundamental physics of how the material functions inside a battery.

even though this material has a poorer conductivity than comparable prototypes that require more lithium, his team was nevertheless able to build a functional battery with it. both baker and murugesan say that there is still plenty to be done to optimize the new battery. but it took almost nine months from the moment murugesan initially spoke with the microsoft team until the battery was strong enough to light a light bulb.

although the techniques used here are cutting-edge in terms of machine learning tools, what sets this apart is that it was developed and tested, according to massachusetts institute of technology professor rafael gómez-bombarelli, who was not engaged in the project. “making predictions is very simple; persuading someone to fund actual experiments is a difficult task.” according to him, the group employed ai to supplement and expedite computations that physicists had performed for decades. however, challenges to this strategy may arise in the future. according to him, materials other than battery components can call for a more intricate method of element combination because the data required to train the ai for this kind of task is frequently minimal.

posted on april 9, 2024