Reviewed by Alex SmithNov 8 2022
By firing two X-Ray pulses that are only a few milliseconds apart, researchers can capture atomic-resolution images of a system at two different points in time using X-Rays as a superfast, atomic-resolution camera.
A speckle pattern typical of the sort seen at LCLS’s detectors. Image Credit: Joshua Turner
Comparing these snapshots demonstrates how a material changes throughout a nanosecond, providing information that scientists could use to develop ultrafast communications systems, computers, and other technologies in the future.
Joshua Turner, a lead scientist at Stanford University and the Department of Energy’s SLAC National Accelerator Center, and ten other researchers used artificial intelligence to speed up the laborious and time-consuming process of resolving the information in these X-Ray snapshots.
This X-Ray probing methodology is accelerated by their machine learning-assisted method, published on October 17th, 2022, in Structural Dynamics. It also expands it to previously inaccessible materials.
The most exciting thing to me is that we can now access a different range of measurements, which we could not before.
Joshua Turner, Lead Scientist, SLAC National Accelerator Laboratory, Stanford University
Handling the Blob
The X-Rays scatter off a material while examining materials using this two-pulse method and are typically detected one photon at a time. A detector measures these scattered photons to create a speckle pattern, a blotchy image that depicts the precise arrangement of the sample at a particular moment in time.
Researchers compare the speckle patterns from each pair of pulses to determine sample fluctuations.
Turner added, “However, every photon creates an explosion of electrical charge on the detector. If there are too many photons, these charge clouds merge together to create an unrecognizable blob.”
To produce a clear comprehension of the speckle pattern, the researchers must gather a vast amount of scattering data due to the cloud of noise.
You need a lot of data to work out what is happening in the system.
Sathya Chitturi, Study Lead and PhD Student, Stanford University
Turner and co-author Mike Dunne, who oversees SLAC’s Linac Coherent Light Source (LCLS) X-Ray laser, provide him advice.
To comprehend the speckle patterns using conventional techniques, all data must be gathered and then analyzed using models that predict how the photons group together at the detector.
On the other hand, the machine learning approach directly extracts fluctuation information from the raw detector image of scattered photons. When combined with upgraded hardware, this new method is 100 times faster than the previous one, making it possible to analyze data more quickly.
Co-author Nicolas Burdet, an associate staff scientist at SLAC who created a simulator that generated data to train the machine learning model, partly contributed to the new approach's success.
Through this training, the algorithm could understand how the charge clouds merge and how many photons per blob and per pulse pair actually hit the detector. Even in extremely blobby conditions, the model held up well.
Seeing Beyond the Clouds
Various materials, such as high-temperature superconductors or quantum spin liquids, have been challenging to investigate because X-Rays scatter off them too weakly for detection. Yet, they can now be studied using the model. According to Chitturi, the new technique could also be used with colloids, alloys, and glasses, which are non-quantum materials.
According to Turner, the research should assist the LCLS-II upgrade, enabling researchers to gather up to a million images, or a few gigabytes of data, every second, as opposed to roughly a hundred snapshots a second for LCLS.
“At SLAC we are excited about this upgrade but have also been kind of worried if we can handle this amount of data,” stated Turner.
In a separate study, the team discovered that their new method ought to be quick enough to handle all that data.
Turner commented, “This new algorithm will really help.”
Artificial intelligence’s promise of increased speed suggests that it will also change the experimental procedure. Researchers will be able to assess data and make modifications while it is being collected, which could reduce the amount of time and money spent on the experiment.
Additionally, it will enable researchers to identify surprises and reroute their studies in real-time to look into unforeseen phenomena.
Chitturi stated, “This method can let you explore more of the materials science you are interested in and maximize scientific impact by letting you make decisions at different points along your experiment about changes in experimental variables such as temperature, magnetic field, and material composition.”
The study is a part of a broader initiative by SLAC, Northeastern University, and Howard University to employ machine learning to advance the fields of materials and chemistry.
The DOE Office of Science and the DOE Early Career Research Program provided funding for the study. LCLS is a user facility for the DOE Office of Science.
Journal References:
Chitturi, S. R., et al. (2022) A machine learning photon detection algorithm for coherent x-ray ultrafast fluctuation analysis. Structural Dynamics. doi:10.1063/4.0000161
Chen, H., et al. (2022) Testing the data framework for an AI algorithm in preparation for high data rate X-ray facilities. arXiv. doi:10.48550/arXiv.2210.10137