Inside the microchips powering the device you’re reading this on, the atoms have a hidden order all their own. A team led by Lawrence Berkeley National Laboratory (Berkeley Lab) and George Washington University has confirmed that atoms in semiconductors will arrange themselves in distinctive localized patterns that change the material’s electronic behavior.

The research, published in Science, may provide a foundation for designing specialized semiconductors for quantum-computing and optoelectronic devices for defense technologies.

On the atomic scale, semiconductors are crystals made of different elements arranged in repeating lattice structures. Many semiconductors are made primarily of one element with a few others added to the mix in small quantities. There aren’t enough of these trace additives to cause a repeating pattern throughout the material, but how these atoms are arranged next to their immediate neighbors has long been a mystery.

Do the rare ingredients just settle randomly among the predominant atoms during material synthesis, or do the atoms have preferred arrangements, a phenomenon seen in other materials called short-range order (SRO)? Until now, no microscopy or characterization technique could zoom in close enough, and with enough clarity, to examine tiny regions of the crystal structure and directly interpret the SRO.

“It’s an interesting scientific question because SRO dramatically changes the properties of a material. Our colleagues have predicted SRO theoretically in semiconductors, but this is the first time the individual structure of these SRO domains has been shown experimentally,” said co-lead author Andrew Minor, director of the National Center for Electron Microscopy at Berkeley Lab’s Molecular Foundry and a professor of Materials Science and Engineering at UC Berkeley.

Minor’s lab is part of the Center for Manipulation of Atomic Ordering for Manufacturing Semiconductors (µ-Atoms), a Department of Energy (DOE) Energy Frontier Research Center focused on understanding atomic ordering in semiconductors. “Our results are exciting because the property that’s being changed by this local ordering is the most important property for microelectronics, the band gap, which is what controls the electronic properties,” he said.

The breakthrough moment came when first author Lilian Vogl, who was then a postdoctoral researcher in Minor’s lab, was studying a sample of germanium containing a small amount of tin and silicon using a powerful type of electron microscopy recently pioneered by the group called 4D-STEM. The initial results were too muddled to parse the faint signals from the electrons diffracting off the tin and silicon from the strong signals off the tidily arranged germanium, so she implemented an energy-filtering device on the system to improve contrast.

When the next dataset started appearing on her monitor, she quickly realized there was a new kind of result. The faint signals were clearer, and repeating patterns emerged, indicating that the atoms have preferred order after all.

To validate her findings and learn what these patterns meant, Vogl collected more data with the energy-filtering 4D-STEM and used a pre-trained neural network to sort the diffraction images. The tool identified six recurring motifs representing particular atomic arrangements in the sample material, but the Berkeley Lab team still couldn’t determine the exact atomic structures that were generating the motifs. To interpret their experimental results, they turned to µ-Atoms collaborators at George Washington University led by co-lead author Tianshu Li, a professor of Civil and Environmental Engineering.

Li’s team generated a highly accurate and efficient machine-learning potential capable of modeling millions of atoms in the material’s structure, allowing Vogl to perform simulated 4D-STEM on different possible structural arrangements until she found matches for the motifs in the experimental data.

“It’s remarkable that modeling and experiment can work seamlessly to unravel SRO structural motifs for the first time,” said Li, whose team had previously predicted SRO and its impact and helped motivate the current study.

“Proving SRO experimentally is not an easy task, let alone identifying its structural motifs. Signals from SRO can easily be obscured by defects or inherent movement of atoms at room temperature, and until now there was no clear way to separate them. This work represents the first step toward our broader goal.”

Shunda Chen, a research scientist in Li’s group who developed the model, said, “With these models, which combine machine learning with first-principles calculations, we can replicate experimental procedures with high fidelity and pinpoint the structural motifs that would otherwise remain hidden.”

Follow-up work initiated by other µ-Atoms members at the University of Arkansas and at Sandia National Laboratories is already yielding insights into how these short range-order motifs affect the semiconductor’s electronic properties, and the scientists hope that manipulating the order to enable new types of devices and processing routes will be possible soon.

“We’re going to be able to really push the boundaries beyond current capabilities by designing semiconductors at the atomic scale,” said Vogl, who is now group leader of the Environmental & Analytical Electron Microscopy Group at the Max Planck Institute for Sustainable Materials.

“We are opening the door to a new era of information technology at the atomic scale, unlocking the deterministic placement of SRO motifs for tailoring of band structures that could impact a wide variety of technologies, from topological quantum materials to neuromorphic computing to optical detectors.”

At its annual Oktane customer conference in Las Vegas, Nevada, identity and access management specialist Okta has been expanding its vision to be a first-port-of-call when it comes to securing non-human identities (NHIs) as a swelling wave of artificial intelligence (AI) agents causes their numbers to swell dramatically.

Among the announcements made today are new capabilities within both the Okta and Auth0 platforms that the supplier said will enable users to integrate AI agents seamlessly into their identity security fabrics.

A study released earlier in 2025 by Okta revealed that although 91% of organisations were already deploying agentic AI in search of productivity gains, but also that just 10% of organisations are today putting any form of cyber governance in place to manage agents – so Okta believes the risk is also rising, and fast.

Such risks are no longer theoretical; Okta cited incidents such as the now infamous breach which an AI bot built on the Paradox AI platform and used by fast-food giant McDonalds in its hiring process exposed the personal data of millions of job applicants to hackers who correctly guessed that its password was ‘123456’.

Okta CEO Todd McKInnon compared unleashing AI agents on an organisation’s environment to creating a lot of individual new insider threats.

“AI agents are a powerful new identity type. They can act independently, on their own or on behalf of a user or a team or a company,” said McKinnon. “They can access tools, apps or data, they can plan or complete tasks on their own. The pace here of innovation is absolutely stunning.

“These AI agents and the potential here, are getting very, very powerful and it’s happening very quickly.

“Without identity security AI security collapses. AI security is identity security, you can’t be successful in one without the other,” said McKinnon.