Revealing crystallographically distinct growth sectors in Na-flux GaN via machine learning-assisted nanobeam x-ray diffraction analysis

Gallium nitride (GaN) is widely used in optoelectronic and power devices, but structural defects limit its performance. To improve crystal quality, the multi-point seed technique and the flux film coated technique have been developed, enabling the growth of low-dislocation GaN wafers via the Na-flux method. However, the coalescence boundary (CB) region, formed by the fusion of multiple growth fronts, exhibits complex structural variations that remain insufficiently understood. In this study, we investigate the CB region of FFC-GaN using synchrotron-based nanobeam x-ray diffraction (nanoXRD). To analyze the high-dimensional diffraction data without prior labeling, we introduce an unsupervised machine learning framework that combines Uniform Manifold Approximation and Projection (UMAP) with agglomerative clustering. This approach identifies six crystallographically distinct clusters that align well with growth sectors inferred from cathodoluminescence imaging. By further applying the Kolmogorov–Smirnov test, we enhance the pixel-level interpretation of structural evolution during different growth stages of CB. Our results demonstrate, for the first time, that the combination of UMAP, clustering, and statistical testing enables direct, label-free crystallographic analysis of crystal growth sectors. This framework is believed to be generalizable to other high-dimensional experimental datasets, opening new avenues for uncovering hidden structural features in complex material systems.