AI Breakthrough Boosts Self-Driving Labs, Speeds Materials Discovery by Ten Times
Image Credit: Girl with Red Hat | Splash
Researchers at North Carolina State University have developed a technique enabling self-driving laboratories to gather at least 10 times more data than earlier methods, potentially shortening materials discovery timelines from years to days.
The method, outlined in a paper published on July 14, 2025, in Nature Chemical Engineering, combines artificial intelligence with continuous flow reactors for real-time experimentation, allowing AI to process data and modify parameters seamlessly.
Background on Self-Driving Laboratories
Self-driving laboratories have roots in late-19th-century automation efforts, starting with devices for tasks like unattended filtrate washing in 1875. These evolved through 20th-century instrumentation into integrated systems. Modern iterations emerged around 2020, exemplified by Ada, a self-driving lab for accelerated discovery of thin-film materials.
A key advancement occurred in 2023 with the A-Lab at Lawrence Berkeley National Laboratory, an autonomous setup that synthesized 41 new inorganic materials in 17 days by leveraging AI for recipe generation from literature, machine learning for X-ray diffraction phase identification, and active learning to address synthesis failures. This system drew on ab initio calculations and past synthesis data, attaining a 71% success rate for 58 targets spanning 33 elements.
Further progress included Argonne National Laboratory's Polybot in February 2025, employing AI to refine electronic polymer films for enhanced conductivity and reduced defects. These developments responded to inefficiencies in traditional materials science, where manual processes limit data generation and require substantial human input.
Development of the New Technique
Led by Milad Abolhasani, an ALCOA Professor of Chemical and Biomolecular Engineering at NC State, the technique transitions from steady-state flow experiments—requiring reaction stabilization prior to measurement—to dynamic flow experiments. Co-lead authors are Ph.D. student Fernando Delgado-Licona, master's student Abdulrahman Alsaiari, and former undergraduate Hannah Dickerson, with funding from the National Science Foundation. It was validated using cadmium selenide colloidal quantum dots as a model for inorganic synthesis.
Steady-state configurations in self-driving labs pause for each reaction to complete, producing one data point per run. The dynamic method continuously alters chemical mixtures in microfluidic channels, correlating transient states to steady-state outcomes via monitoring every half second. This yields up to 20 data points for a 10-second reaction, facilitating immediate AI-driven adjustments.
The approach addressed constraints in batch and steady-state systems, which experience downtime during stabilization and yield limited data, impeding AI model training. By embedding in situ characterization and autonomous operations, it aimed to boost data efficiency for machine learning predictions of optimal experiments.
How the AI-Powered System Operates
The system centers on AI orchestration: machine learning algorithms evaluate data to propose next actions, while robotics manage fluid mixing and flow regulation. Sensors deliver ongoing feedback on optical traits via in-situ characterization, enabling AI to tweak variables like temperature or concentration during flow.
Differing from prior self-driving labs that segregate synthesis and analysis, this merges them to minimize delays. The AI applies probabilistic models to handle data variability and focus on promising routes, incorporating active learning akin to the A-Lab. Following initial setup, it pinpoints optimal materials in the first cycle, unlike conventional labs' iterative approaches.
Impact on Materials Discovery
By supplying denser datasets, the technique expedites AI convergence on suitable candidates. It decreases costs through reduced chemical usage—up to 10 times less than advanced fluidic labs—and curbs waste, supporting sustainable research amid resource pressures.
Broadly, such self-driving labs have yielded results like the A-Lab's confirmation of predicted materials, bridging computational and experimental divides. For sectors with material constraints, this hastens innovation requiring swift prototyping.
Applications in Chemistry and Environmental Materials
The technique suits nanocrystal synthesis for electronics, LEDs, and photovoltaics, demanding exact control of particle dimensions and properties. In chemistry, it aids kinetic studies under diverse conditions, advancing catalyst or polymer development.
For environmental materials, it enables low-waste processes for energy storage or pollutant-absorbing compounds. Polybot's work on conductive polymers suggests applications in flexible electronics for renewables, while quantum dots may improve solar cell performance.
Self-driving labs are projected to advance toward greater autonomy, diminishing human supervision via sophisticated AI, though hybrid models with human oversight will continue for intricate tasks. Developments encompass scaling for varied materials, from organics to biologics, and fusing multi-modal data like spectroscopy.
Policy initiatives could emphasize infrastructure to capitalize on AI advancements, fostering discovery cycles. Obstacles include standardization and reproducibility, yet potential extends to tackling issues like climate change through expedited material breakthroughs.
TheDayAfterAI News
We are a leading AI-focused digital news platform, combining AI-generated reporting with human editorial oversight. By aggregating and synthesizing the latest developments in AI — spanning innovation, technology, ethics, policy and business — we deliver timely, accurate and thought-provoking content.