An Open-Source Hardware and Software Toolkit to Enable Agentic RHEED-Guided Thin-Film Synthesis
This paper presents an open-source hardware and software toolkit that enables autonomous, AI-driven control and quantitative analysis of Reflection High-Energy Electron Diffraction (RHEED) for thin-film synthesis, thereby overcoming traditional operator dependencies and establishing a foundation for fully agentic material growth.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, scientists have relied on a technique called reflection high-energy electron diffraction to watch how atoms arrange themselves as they form new materials. Imagine a beam of electrons striking the surface of a crystal at a very shallow angle, like a stone skipping across a lake. The way these electrons bounce off creates a pattern of light and dark streaks on a screen, acting as a real-time map of the surface's atomic structure. This map tells researchers if the material is growing smoothly layer by layer or if it is becoming rough and disordered. However, reading this map has traditionally been a slow, manual process. A human operator must constantly adjust the equipment to keep the beam aligned and interpret the shifting patterns by eye, a task that varies from person to person and makes it difficult to automate the creation of new materials.
A team of researchers at Oak Ridge National Laboratory has now built a complete, open-source toolkit that allows both human operators and artificial intelligence agents to control this electron beam and analyze the resulting patterns without human intervention. They developed a system that can automatically align the beam, measure the crystal's orientation, and extract precise data about how fast a film is growing and how ordered its atoms are. By connecting this hardware and software to modern AI, they demonstrated that a computer can be given a simple request in plain English, such as "analyze the growth rate," and it will independently set up the experiment, run the analysis, and return the results. This work removes the need for constant human supervision, turning a complex, subjective visual task into a reproducible, automated process that can be shared and improved by the global scientific community.
The core of this new system is a piece of custom hardware that acts as the nervous system for the electron beam. In the past, adjusting the beam's position required manual knobs or proprietary software that was often locked down or difficult to modify. The researchers replaced this with a low-cost, programmable chip that can send precise electrical signals to the magnets steering the electron beam. This allows the system to sweep the beam across the sample in a controlled, automated way, a process known as collecting a "rocking curve," which helps scientists understand the quality of the surface. The team also created a software routine that calibrates how the magnets work together, ensuring that when the beam angle changes, the spot where it hits the sample stays perfectly still. This level of precision is essential for gathering reliable data, and by making the control code open source, they have allowed other labs to replicate this setup without needing expensive, proprietary equipment.
Once the beam is under control, the system must figure out exactly how the crystal sample is oriented. Traditionally, a scientist would rotate the sample and watch the pattern on the screen, stopping when the streaks looked perfectly symmetrical. This method relies on human judgment and can lead to inconsistencies. The new toolkit uses a different approach: it takes a video of the sample rotating and uses a mathematical measure of symmetry to find the exact moment the pattern is most balanced. The software automatically detects the edges of the sample and the bright spots in the pattern, correcting for any tilts or misalignments before calculating the symmetry. This "training-free" method means it works on any material without needing to be taught what that material looks like first. It finds the perfect alignment by simply looking for the most symmetrical image, a task the computer performs with a consistency that humans cannot match over long periods.
The most significant leap forward is the software application called Auto-RHEED, which serves as the brain of the operation. This program can take raw video footage from the electron microscope and turn it into specific, useful numbers. It can calculate the distance between atoms on the surface, measure how large the ordered regions of the crystal are, and determine the speed at which new layers are being added. What makes this tool unique is its ability to talk to artificial intelligence. The researchers connected the software to a standard interface that allows large language models to understand and use its tools. In a demonstration, an AI agent was given a natural language prompt to analyze a film growth experiment. Without any specific instructions on where to look or how to process the data, the agent successfully identified the relevant parts of the image, set up the measurements, and calculated the growth rate, all in under a minute. The agent did not invent its own way of analyzing the data; it used the same rigorous methods a human would, but it did so by simply understanding the request and executing the steps.
The system also proves that it can integrate new scientific methods as they are developed. The researchers designed the software with a flexible architecture that allows different analysis tools to be plugged in like interchangeable modules. They demonstrated this by adding a method that uses advanced AI models to recognize patterns in the diffraction images without needing to be trained on specific examples. This allowed the system to identify different stages of crystal growth, such as when the material shifts from growing in flat layers to forming three-dimensional islands, simply by observing how the visual patterns changed over time. Because the software records every step of the process and every setting used, the results are fully traceable and reproducible, whether they were generated by a human clicking buttons or an AI agent following a text command.
By combining programmable hardware, automated alignment, and an AI-ready software interface, this toolkit transforms RHEED from a manual, operator-dependent skill into a robust, automated sensor for material synthesis. It bridges the gap between the physical act of growing a material and the digital world of data analysis, allowing for experiments that can run themselves and adapt based on what they see. The researchers have made all their code and designs freely available, ensuring that this capability is not limited to a single laboratory but can become a standard tool for the entire field. This work lays the foundation for a future where the creation of new materials is guided by real-time, automated feedback, accelerating the discovery of advanced technologies that rely on precise atomic structures.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.