Gallium nitride (GaN) is a material used in many electronic and optoelectronic devices, such as LEDs and power transistors. During the production of GaN, tiny imperfections in its crystal structure can greatly affect its performance. Detecting these defects is important but not always easy, especially when the data collected is noisy or hard to interpret with traditional analysis tools. In our research, we explored a new way to study GaN crystals using a technique called nanobeam X-ray diffraction, which provides detailed information about the crystal structure at very small scales. Instead of relying on traditional fitting methods—which often fail when the signal has a complex shape—we applied machine learning to discover patterns directly from the raw data. This approach allowed us to reveal data structural features resulted from material's structural features that had not been clearly identified before. Specifically, we found signs of structural changes likely related to defects such as stacking faults and the formation or elimination of dislocations—types of imperfections that can occur during the growth process. These insights help explain how the material changes over time and why certain regions might perform better or worse in a device. Overall, our method shows that machine learning can be a powerful tool for analyzing complex measurement data, offering new ways to monitor and improve crystal quality during manufacturing—especially when conventional methods fall short.
This research is the first to show how an unsupervised machine learning method can be used to directly connect the structure of experimental data (like patterns in X-ray signals) with the structure of crystals inside materials. This is a major step forward because it allows researchers to uncover hidden structural changes—such as defects or irregularities in how crystals grow—without needing prior knowledge or labels. What makes this approach even more powerful is that it's not limited to X-ray diffraction (XRD). The method can potentially be applied to other types of measurements, such as optical, spectroscopic, or imaging data. This means it could help scientists in many different fields analyze complex materials more reliably—especially when the data is noisy or traditional analysis techniques don’t work well. By focusing on patterns that naturally emerge from the data, this method opens the door to more objective, scalable, and accurate ways of understanding materials—making it a valuable tool for both research and industry.
Read more on KudosThe following have contributed to this summary: ZHENDONG WU and Tetsuya Tohei