Artificial Chemist, a technology invented by researchers at North Carolina State University (NC State) and the University at Buffalo, integrates artificial intelligence and an automated system for performing chemical reactions to speed R&D and manufacturing of solution-processed materials. In proof-of-concept experiments, the researchers showed that Artificial Chemist could identify and produce made-to-measure inorganic perovskite quantum dots (QDs) for any color in 15 minutes or less. The platform demonstrated it could run 500 QD synthesis experiments per day, and the researchers believe it could run as many as 1000 per day. The QDs were synthesized and tuned by the new technology. The Artificial Chemist has a “body” for performing experiments and sensing the experimental results and a “brain” for recording and applying data. The brain is an AI program that characterizes the materials being synthesized by the body, and then uses the data to make autonomous decisions about what the next set of experimental conditions should be. The brain bases its decisions on what it determines will most efficiently move it toward the material composition that best exhibits the desired properties and performance metrics. A technology called Artificial Chemist incorporates AI and an automated system for performing chemical reactions to accelerate R&D and manufacturing of commercially desirable materials. Courtesy of Milad Abolhasani. “The Artificial Chemist is similar to a self-driving car, but a self-driving car at least has a finite number of routes to choose from in order to reach its preselected destination,” NC State professor Milad Abolhasani said. “With Artificial Chemist, you give it a set of desired parameters, which are the properties you want the final material to have. Artificial Chemist has to figure out everything else, such as what the chemical precursors will be and what the synthetic route will be, while minimizing the consumption of those chemical precursors.” Use of knowledge transfer enables the Artificial Chemist to become smarter and faster over time at identifying the right material. It stores the data generated from every request it receives and draws on the data to speed the process of identifying the next candidate material. The researchers tested nine policies for directing how the AI could use data to decide what the next experiment would be. They then ran a series of requests, each time asking Artificial Chemist to identify a QD material that was the best fit for three different output parameters. “We found a policy that, even without prior knowledge, could identify the best quantum dot possible within 25 experiments, or about one-and-a-half hours,” Abolhasani said. “But once Artificial Chemist had prior knowledge — meaning that it had already handled one or more target material requests — it could identify the optimal material for new properties in 10 to 15 minutes.” The knowledge transfer strategy also mitigates the issues of batch-to-batch precursor variability. “We found that Artificial Chemist could also rapidly identify the boundaries of materials properties for a given set of starting chemical precursors, so that chemists and materials scientists do not need to waste their time on exploring different synthesis conditions,” Abolhasani said. The Artificial Chemist “not only helps you find the ideal solution-processed material more quickly than any techniques currently in use, but it does so using tiny amounts of chemical precursors,” Abolhasani said. “That significantly reduces waste and makes the materials development process much less expensive.” The technology is currently designed for solution-processed materials including QDs, metal and metal oxide nanoparticles, and metal organic frameworks (MOFs). Abolhasani believes that autonomous materials R&D enabled by Artificial Chemist could “reshape the future of materials development and manufacturing.” He is currently looking for partners to help his team transfer the technique from the lab to the industrial sector. The research was published in Advanced Materials (www.doi.org/10.1002/adma.202001626).