Predicting Lattice Orientation of Lithium Niobate using Machine Learning aided Raman Spectroscopy

Due to its nonlinear, electro-optical properties, Lithium Niobate (LiNbO3) is a widely used materials system for integrated optics. We previously developed a technique in which single crystal LiNbO3 is produced within a glass using femtosecond laser heating. In this process, the orientation of the crystal is of utmost importance to achieve desired functionalities. For this reason, understanding how to measure and alter crystal orientation are crucial. The current method to determine crystal orientations, Electron Backscatter Diffraction (EBSD), requires sample preparation that is time-consuming and partially destroys the crystals under study. Therefore, this method is not practical to investigate large numbers of samples while maintaining their functionality. Raman Spectroscopy is noninvasive; in principle, it can determine crystal orientation although the assignment includes more noise. We employed a K Nearest Neighbors machine learning algorithm to better sort through this noise. This model used orientation data from EBSD to teach the computer orientation behaviors in Raman spectra. Subsequently, we successfully determined crystal orientations, using Raman spectra of crystals embedded within a glass sample, as well as shifts in those orientations.