Machine learning assisted nanobeam X-ray diffraction based analysis on hydride vapor-phase epitaxy GaN.

Abstract

Nanobeam X-ray diffraction (nanoXRD) is a powerful tool for collecting in situ crystal structure information with high spatial resolution and data acquisition rate. However, analyzing the enormous amount of data produced by these high-throughput experiments for defect recognition or discovering hidden structural features becomes challenging. Machine learning (ML) methods have become attractive recently due to their outstanding performance in analyzing large data sets. This research utilizes an ML algorithm, uniform manifold approximation and projection (UMAP), to enhance the nanoXRD-based crystal structure analysis of a cross-sectional hydride vapor-phase epitaxy GaN wafer. Compared with the results obtained by conventional fitting, UMAP gives a more precise categorization of crystal structure based on the raw three-dimensional ω-2θ-φ diffraction patterns. The property that UMAP embeds the high-dimensional data while retaining the data structure is valuable in guiding the analysis of nanoXRD profiles. This research also demonstrates the capability of UMAP in analyzing other spectroscopic or diffraction data sets to guide crystal structure investigations.