Human decision-substitutable chemometric identification of pixel shrinkage in OLEDs using secondary ion mass spectrometry

Abstract

Secondary ion mass spectrometry (SIMS)-based depth profiling of organic light-emitting diode (OLED) samples composed of multiple layers is a useful analytical tool for identification of pixel shrinkage, which is a frequent defect occurring in OLED production. As an alternative to engineer-based defect identification, which is time-consuming and occasionally suffers from person-to-person discrepancies, a chemometric analysis scheme utilizing the whole SIMS dataset of the OLED sample has been explored in this study. Initially, all the depth-profiled spectra of each sample were assigned (classified) into the corresponding layers (groups) using a Gaussian mixture model (GMM), and the subsequent average spectra of each layer were used for layer-by-layer comparisons. Then, a one-class support vector machine (OC-SVM) was adopted for defect identification, and principal component (PC) scores of the spectra in each layer were used as the inputs for OC-SVM. The occurrence of defects in the 3rd layer was clear in the PC score domain, and the accuracy obtained by predicting the external test samples was 97.4%. Moving forward, the identification of potential peaks significantly contributing to the defect could be valuable clues to track the causes of defects (trouble shooting) during production. Kernel density estimation (KDE) was adopted for this purpose, and the top nine peaks with the lowest p-values (most significant contributions) were finally provided.