Aug. 11, 2026 — What if scientists could get a taste of discovery as soon as their experiment finishes? Thanks to a new machine learning tool called DONUT, researchers at the U.S. Department of Energy’s (DOE) Argonne National Laboratory are transforming how experiments are run at the Advanced Photon Source (APS), a DOE Office of Science user facility. By delivering results in real time, DONUT allows scientists to make faster decisions, adapt their experiments on the fly and unlock deeper insights into the structure of advanced materials. This breakthrough will accelerate research and also lower barriers for new users — no sprinkles required.
DONUT makes it easier to cut through layers of complex data and reach the sweet spot of scientific discovery, turning X-ray measurements into real-time insights for researchers at the APS. (Image by ChatGPT.)
DONUT, short for Diffraction with Optics for Nanobeam by Unsupervised Training, is a physics-aware neural network. This means the tool is built with an understanding of the physical laws that govern how focused X-ray beams interact with materials.
Developed and tested using data from the Hard X-ray Nanoprobe beamline shared by the APS and the Center for Nanoscale Materials (CNM), DONUT helps scientists quickly interpret complex X-ray images produced by scanning X-ray nanodiffraction microscopy (SXDM), revealing the internal structure of materials at the nanoscale. The CNM is also a DOE Office of Science user facility at Argonne.
Until now, analyzing this data has been a slow and painstaking process, often taking weeks or months. With DONUT, researchers can get results in real time, sometimes hundreds of times faster than traditional methods.
“DONUT lets us see what’s happening inside materials as the experiment unfolds,” said Aileen Luo, assistant computational scientist at Argonne and Cornell University. “Instead of waiting for weeks to find out if an experiment worked, we can now get answers on the spot. That means more productive experiments and more opportunities for discovery.”
A Sweet Solution to a Tough Problem
SXDM is a powerful technique that uses a focused X-ray beam to scan across a sample, collecting information about its crystal structure. This helps scientists understand how materials behave in technologies such as batteries, catalysts that speed up chemical reactions and advanced electronic or magnetic devices. However, SXDM generates complex data with multiple layers and dimensions, making it challenging for scientists to analyze and interpret.
Traditionally, scientists have relied on manual comparisons between measured and simulated X-ray patterns, a process that is both time-consuming and prone to errors. DONUT changes the recipe by combining artificial intelligence (AI) with a built-in physics model. This allows the system to learn directly from experimental data, without needing labeled training examples, where each X-ray pattern must first be matched with the correct answer by experts or simulations — a major advantage for busy beamline users.
“DONUT is flexible and customizable,” said Mathew Cherukara, a computational scientist and group leader at Argonne. “You can train it on the data you collect at the start of the experiment and even adjust what you want it to predict during the experiment. It’s like having a fresh DONUT recipe for every new scientific question.”
The speed and accuracy of DONUT make it possible for scientists at the APS to try new kinds of experiments, including autonomous “self-driving” research, where the next step is chosen automatically based on the latest results. This is especially helpful for studies that test materials in real-world conditions, where things can change quickly and researchers need to respond right away.
“Being able to analyze data as it’s collected means researchers can make decisions on the fly,” said Luo. “It’s a gamechanger for experiments that need quick feedback. No more waiting for the dough to rise.”
DONUT’s approach also lowers the barrier for new users at the APS and CNM, including graduate students and visiting scientists, by eliminating the need for expert-labeled datasets. Traditionally, preparing labeled data requires significant time and specialized knowledge because experts must carefully analyze or simulate each dataset to assign the correct labels. This process can slow down research and limit participation to those with advanced training.
This capability is especially valuable as the upgraded APS delivers brighter X-ray beams and collects data at much higher speeds. With DONUT, researchers can keep up with the rapid pace of data generation, making real-time decisions and exploring new types of dynamic experiments. This combination of advanced machine learning and the upgraded APS promises to accelerate discoveries across materials science and beyond.
The Next Bite: DONUT’s Future
The team is now working to expand DONUT’s impact from real-time analysis to experimental automation. Efforts are underway on new flavors of DONUT for autonomous microscopy, where AI tools could help guide portions of experiments without constant human input. They are also exploring how DONUT’s physics-aware approach could help solve challenges in other advanced imaging techniques, which generate similarly complex datasets that require sophisticated analysis.
Looking ahead, DONUT’s physics-aware training framework is expected to support major initiatives like the DOE’s Genesis Mission. DONUT forms the foundation for the light and neutrondouble scientific productivity and accelerate innovation through AI. This will help scientists across disciplines tackle new scientific questions and make the most of next-generation research facilities
No actual donuts were harmed in the making of this research. But the results are sure to fuel scientists’ hunger for discovery.
The results of this research were published in npj Computational Materials.
Other contributors to this work include Tao Zhou, Ming Du and Martin Holt (Argonne) and Andrej Singer (Cornell University).
This study was funded by the DOE Office of Science, Advanced Scientific Computing Research and Basic Energy Sciences. This work was also supported by the DOE Office of Science, Office of Workforce Development for Teachers and Scientists.
Sectors:
Government, Science
Tags:
Advanced Photon
The Big Data Inside Amazon’s New Fire Phone
The new Fire phone that Amazon launched this week looks like your ordinary black smartphone,…
Three Reasons to be Scared of the Internet of Things
We know the Internet of Things forecasts: 50 billion connected devices by 2020. Apparently, there’s…
Wanted: Intelligent Middleware That Simplifies Big Data Analytics
We’ve seen tremendous technological innovation in the data analytics space over the past 10 years….
GPUs Tackle Massive Data of the Hive Mind
LIVE from GTC12 — The flock of birds that weaves seamlessly through the sky, propelled…
Why Hadoop on IBM Power
In the quest to achieve data-driven insight, Hadoop running on Intel X86-based processors has emerged…
Datanami’s Leverage Big Data Summit Wraps Up
Dialog and networking were on the Datanami agenda this week as we kicked off our…
NVIDIA’s $500 Billion AI Bet: Jensen Huang Brings Wall Street Into the Race
Jensen Huang believes NVIDIA’s chips are becoming much more than expensive pieces of hardware. As…
Your Data is Not Ready: Solving the First Mile Gap for Enterprise AI
There’s trouble brewing in the enterprise AI world and IT leaders are faced with a…
Samsung Attacks the AI Memory Wall with 3D Packaging
The AI memory wall has emerged as the biggest bottleneck in executing AI inference workloads….
Microchip and Micron Bet AI’s Next Infrastructure Battle Will Be Storage
Earlier this week, Microchip and Micron unveiled a new PCIe Gen6 AI storage architecture that…
What Google DeepMind’s Departures Say About the AI Talent War
Google DeepMind has spent the past few years at the center of the AI boom….
Why AI Still Hasn’t Had Its Einstein Moment
AI is having a remarkable run in science. It’s helping researchers predict protein structures, improve…
