Aug. 11, 2026 — What if scientists may get a style of discovery as quickly as their experiment finishes? Because of a brand new machine studying device referred to as DONUT, researchers on the U.S. Division of Vitality’s (DOE) Argonne National Laboratory are reworking how experiments are run on the Superior Photon Supply (APS), a DOE Workplace of Science consumer facility. By delivering ends in actual time, DONUT permits scientists to make quicker choices, adapt their experiments on the fly and unlock deeper insights into the construction of superior supplies. This breakthrough will speed up analysis and likewise decrease boundaries for brand spanking new customers — no sprinkles required.
DONUT makes it simpler to chop by way of layers of advanced information and attain the candy spot of scientific discovery, turning X-ray measurements into real-time insights for researchers on the APS. (Picture by ChatGPT.)
DONUT, brief for Diffraction with Optics for Nanobeam by Unsupervised Coaching, is a physics-aware neural community. This implies the device is constructed with an understanding of the bodily legal guidelines that govern how centered X-ray beams work together with supplies.
Developed and examined utilizing information from the Laborious X-ray Nanoprobe beamline shared by the APS and the Center for Nanoscale Materials (CNM), DONUT helps scientists shortly interpret advanced X-ray pictures produced by scanning X-ray nanodiffraction microscopy (SXDM), revealing the interior construction of supplies on the nanoscale. The CNM can also be a DOE Workplace of Science consumer facility at Argonne.
Till now, analyzing this information has been a gradual and painstaking course of, usually taking weeks or months. With DONUT, researchers can get ends in actual time, generally tons of of instances quicker than conventional strategies.
“DONUT lets us see what’s occurring inside supplies because the experiment unfolds,” mentioned Aileen Luo, assistant computational scientist at Argonne and Cornell College. “As a substitute of ready for weeks to seek out out if an experiment labored, we will now get solutions on the spot. Which means extra productive experiments and extra alternatives for discovery.”
A Candy Answer to a Robust Downside
SXDM is a strong method that makes use of a centered X-ray beam to scan throughout a pattern, amassing details about its crystal construction. This helps scientists perceive how supplies behave in applied sciences reminiscent of batteries, catalysts that pace up chemical reactions and superior digital or magnetic units. Nonetheless, SXDM generates advanced information with a number of layers and dimensions, making it difficult for scientists to investigate and interpret.
Historically, scientists have relied on guide comparisons between measured and simulated X-ray patterns, a course of that’s each time-consuming and susceptible to errors. DONUT adjustments the recipe by combining synthetic intelligence (AI) with a built-in physics mannequin. This enables the system to be taught immediately from experimental information, while not having labeled coaching examples, the place every X-ray sample should first be matched with the right reply by consultants or simulations — a serious benefit for busy beamline customers.
“DONUT is versatile and customizable,” mentioned Mathew Cherukara, a computational scientist and group chief at Argonne. “You’ll be able to prepare it on the info you accumulate firstly of the experiment and even modify what you need it to foretell through the experiment. It’s like having a contemporary DONUT recipe for each new scientific query.”
Rolling Out Actual-Time Science
The pace and accuracy of DONUT make it doable for scientists on the APS to strive new sorts of experiments, together with autonomous “self-driving” analysis, the place the following step is chosen robotically primarily based on the most recent outcomes. That is particularly useful for research that take a look at supplies in real-world circumstances, the place issues can change shortly and researchers want to reply instantly.
“With the ability to analyze information because it’s collected means researchers could make choices on the fly,” mentioned Luo. “It’s a gamechanger for experiments that want fast suggestions. No extra ready for the dough to rise.”
DONUT’s method additionally lowers the barrier for brand spanking new customers on the APS and CNM, together with graduate college students and visiting scientists, by eliminating the necessity for expert-labeled datasets. Historically, making ready labeled information requires vital time and specialised data as a result of consultants should rigorously analyze or simulate every dataset to assign the right labels. This course of can decelerate analysis and restrict participation to these with superior coaching.
This functionality is very helpful because the upgraded APS delivers brighter X-ray beams and collects information at a lot increased speeds. With DONUT, researchers can sustain with the speedy tempo of knowledge technology, making real-time choices and exploring new varieties of dynamic experiments. This mixture of superior machine studying and the upgraded APS guarantees to speed up discoveries throughout supplies science and past.
The Subsequent Chew: DONUT’s Future
The staff is now working to increase DONUT’s impression from real-time evaluation to experimental automation. Efforts are underway on new flavors of DONUT for autonomous microscopy, the place AI instruments may assist information parts of experiments with out fixed human enter. They’re additionally exploring how DONUT’s physics-aware method may assist remedy challenges in different superior imaging methods, which generate equally advanced datasets that require subtle evaluation.
Trying forward, DONUT’s physics-aware coaching framework is predicted to assist main initiatives just like the DOE’s Genesis Mission. DONUT varieties the muse for the sunshine and neutron supply challenge of Genesis, a daring nationwide initiative that goals to double scientific productiveness and speed up innovation by way of AI. This can assist scientists throughout disciplines deal with new scientific questions and take advantage of next-generation analysis amenities.
No precise donuts have been harmed within the making of this analysis. However the outcomes are certain to gasoline scientists’ starvation for discovery.
The outcomes of this analysis have been revealed in npj Computational Materials.
Different contributors to this work embody Tao Zhou, Ming Du and Martin Holt (Argonne) and Andrej Singer (Cornell College).
This research was funded by the DOE Workplace of Science, Superior Scientific Computing Analysis and Primary Vitality Sciences. This work was additionally supported by the DOE Workplace of Science, Workplace of Workforce Improvement for Academics and Scientists.
Source: Amber Rose, Argonne Nationwide Laboratory
