A new autonomous laboratory navigated through billions of potential material synthesis recipes to identify brighter, lead-free light-emitting nanomaterials in just 12 hours. The work could accelerate development of safer light-emitting nanoplatelets for use in applications ranging from photodetectors to the production of fuel from solar energy.
Nanoplatelets are sheet-like crystals only billionths of a meter thick; in this case, they belong to a family of lead-free "double perovskites," materials whose atomic recipe can be tuned to control how they absorb and emit light.
"One of the big challenges in developing safer optical nanomaterials is the sheer size of the material universe," says Milad Abolhasani, Alcoa Professor and University Faculty Scholar in the department of chemical and biomolecular engineering at North Carolina State University. Abolhasani is the corresponding author of the research.
"These materials are chemically complex, and the synthesis process is challenging," Abolhasani says. "There are a vast number of possible combinations of ingredients, ratios, temperatures, and reaction environments that need to be explored to synthesize light-emitting nanoplatelets with the desired optical properties. Traditional trial-and-error approaches are slow and can miss important interactions between reaction parameters."
Traditional, human-led discovery and synthesis can take years to discover a handful of promising materials. The AI-guided lab, dubbed PoLARIS (perovskite laboratory for autonomous reaction inference and synthesis), not only synthesizes safer optical nanoplatelets much more quickly, it also analyzes their optical properties and then adjusts variables for the next round of experiments.
The researchers select the precursor materials and set the objective - in this case delivering "safer," meaning lead- or heavy metal-free, double perovskite nanoplatelets with the brightest photoluminescence.
PoLARIS then runs a series of experiments from different "recipes" that change variables such as precursor amounts, temperature, and reaction time. Each recipe produces a tiny flowing droplet that serves as a miniature reaction vessel, which is then analyzed automatically. The analysis is fed back into the AI, which adjusts the nanoplatelets synthesis recipe for the next round of experiments.
Within a single 12-hour campaign, PoLARIS ran 120 experiments, improved the brightness, and identified the best-in-class safer optical nanoplatelets.
"What is exciting about PoLARIS is that it does more than speed up trial and error," Abolhasani says. "It learns from every experiment and builds a map of how chemistry, composition and temperature control material performance. That means we can discover promising materials faster, use less material and understand why the best recipes work.
"Many AI-guided systems can help find an answer, but the scientific value increases dramatically when the system can also help explain the answer. PoLARIS not only found a better recipe; it helped us understand why that recipe worked."
The researchers add that PoLARIS is scalable - not only can it discover best-in-class double perovskite nanoplatelets, it can also switch modes to continuously manufacture the optimized material of interest.
"The beauty of PoLARIS is that it is both a GPS for materials discovery and a miniature materials factory," Abolhasani says. "It can search the chemical landscape, find a promising route, explain why that route works and then continue producing the material. That is the type of human-AI-robot collaboration we need to accelerate discovery of next-generation materials.
"The broader goal is to make autonomous discovery more generalizable," Abolhasani continues. "Many of the materials we need for future energy, electronics and sustainability technologies are too complex to optimize by intuition alone. Self-driving laboratories like PoLARIS give us a way to explore those spaces faster, more efficiently and with deeper scientific understanding."
The work appears in Nature Communications and was supported by the National Science Foundation under grants 2315996 and 2315997. NC State Ph.D. student Junbin Li is first author. Other NC State contributors include Ph.D. students Fernando Delgado-Licona, Jinge Xu, Nikolai Mukhin and Sina Sadeghi, and undergraduate student Hayden Perry. Ou Chen, professor of chemistry at Brown University and Zhenyang Liu, graduate student at Brown University, also contributed to the work.
-peake-
