Distinct thin film growth characteristics determined through comparative dimension reduction techniques

Abstract

Reflection high energy electron diffraction (RHEED) information is critical for the growth of thin films; however, only a small percentage of the data from RHEED videos is typically used. The use of full videos in machine learning can require dimension reduction techniques. In this paper, three dimension reduction techniques, principal component analysis (PCA), non-negative matrix factorization (NMF), and kmeans clustering, are compared to investigate their benefits to the analysis of RHEED data. Three different heterostructures with different growth modes, all deposited on Ti-terminated strontium titanate by pulsed laser deposition, were used for the analysis: lanthanum aluminate with layer-by-layer growth, lithium cobalt oxide with island growth, and strontium ruthenate with a transition from layer-by-layer to step-flow growth. A phase shift in intensity fluctuations of different RHEED spots was discovered and discussed in terms of their sensitivity to the film growth characterization. The diffraction spots that were more sensitive to the growth were differentiated from the spots that are affected by the substrate as a function of film thickness. It was concluded that NMF provides the analysis that is easiest to interpret without the loss of detailed physical information due to its non-negativity constraint and lack of forced orthogonality such as in PCA. Analysis of the full RHEED videos enables a more detailed understanding of growth characteristics and control of growth processes as aided by dimension reduction.