Lawrence Berkeley National Laboratory recently announced Completion of “A-Lab”. The “A” stands for artificial intelligence, automation and acceleration. The $2 million lab is complete with three robotic arms, eight furnaces, and lab equipment all controlled by AI software, and runs 24 hours a day.
If that sounds like a real-life replica of Marvel character Tony Stark’s laboratory, it’s not far off. A lab that accelerates materials science discovery at an unprecedented rate and eases the burden on researchers.
A-lab researchers are currently working on improved battery and energy storage device materials in hopes of meeting the urgent need for sustainable energy use. The lab has the potential to foster innovation in many other industries as well.
“It’s too late to develop materials that are so important to society,” he says. Gert CedarSenior researcher at A-Lab.
Materials science is the field that identifies, develops, and tests materials and their applications, from aerospace to clean energy to medicine.
Materials scientists typically use computers to predict new materials not found in nature and make them stable enough for use. Computers can generate theoretical inorganic compounds, but identifying which new compounds to create, figuring out how to synthesize them, and evaluating their performance is a time-consuming process to do manually. .
[Related: This tiny AI-powered robot is learning to explore the ocean on its own]
In addition, computational tools have made material design virtually a breeze. This means that there is a surplus of new materials that still need to be tested, creating a bottleneck effect.
“If you’re lucky, you can have a two-week trial and be successful, or you’ll be in the lab for six months and can’t go anywhere,” Cedar says. “So it can be very time-consuming to develop a chemical synthetic route to actually make the compound you want in large quantities.”
A-Lab material projecta database of hundreds of thousands of predictive materials operated by Founder Christine ParsonThey provide free access to thousands of computationally predicted new materials, along with information about the structures of compounds and some of their chemistries that researchers can use.
“Computer prediction is not enough to actually design new materials,” says Persson. “I have to prove that this is real.”
An experienced researcher can only scrutinize a handful of samples per day. In theory, A-Lab could make hundreds of samples faster and more accurately. With the help of A-Lab, researchers can allocate more time to big-picture projects instead of doing menial tasks.
Yan Zeng, staff scientist leading the A-lab likens the lab process to cooking a new dish. In this case, the lab is given a new dish (a target compound in this case) to find a recipe for. When researchers identify new compounds with the required quality, they send them to the lab. The AI system creates new recipes with different combinations of over 200 materials or precursor powders such as metal oxides including iron, copper, manganese and nickel.
A robotic arm mixes a slurry of powder with a solvent and bakes a new sample in a furnace to stimulate chemical reactions that may or may not produce the desired compound. After trial and error, the AI system can learn and fine-tune recipes until it creates a successful compound.
[Related: A simple guide to the expansive world of artificial intelligence]
AI software controls the movements of three robotic arms operating in a lab instrument to weigh and mix different combinations of starting materials. The lab itself is also autonomous. That means they can make new decisions about what to do after a failure, and they can independently process new synthetic recipes faster than humans.
“We never thought we would be so successful in synthesizing new compounds,” says Ceder. “And it was kind of a maiden voyage.”
The speed bump by human scientists isn’t just coming from AI-controlled robots, as the software is able to extract knowledge from about 100,000 synthetic recipes from 5 million research papers.
Like a human scientist, A-lab records the details of every experiment and logs any failures.
Researchers do not publish data from failed experiments for a variety of reasons, including limited time and funding, lack of public interest, and the perception that failure is less beneficial than success. However, failed experiments occupy a valuable place in research. They rule out false hypotheses and failed approaches. With easy access to data from hundreds of failed samples created daily, you can better understand what works and what doesn’t.
