Critical Misinterpretation in Oxygen Vacancy Analysis by X-ray Photoelectron Spectroscopy

Abstract

Recently, oxide semiconductors have drawn significant attention not only in display and flexible device technologies—where research and mass production are already well established—but also in the memory semiconductor field, owing to their low leakage current, high carrier mobility even in the amorphous phase, and back-end-of-line (BEOL) compatibility. In this context, accurately analyzing oxygen vacancies, which play a dominant role in determining the electrical performance and reliability of oxide semiconductors, is of critical importance. Traditionally, oxygen vacancies have been analyzed by deconvoluting the X-ray photoelectron spectroscopy (XPS) O1s spectra, where the higher binding energy (B.E.) shoulder is commonly assigned to oxygen-deficient states. This interpretation has been rationalized by the notion that the removal of neighboring oxygen atoms reduces the screening effect, thereby increasing the effective nuclear charge felt by oxygen and shifting the O1s peak to higher B.E. However, such reasoning fails to explain why the O1s peak position does not continuously shift with increasing oxygen vacancy concentration—despite the further reduction in screening that should accompany it. In this study, we experimentally demonstrate the limitation of this conventional approach by showing that samples exhibiting large differences in carrier concentration, sub-gap states, and donor-like states— attributable to significant variations in oxygen vacancy density—exhibit identical XPS O1s spectra. Furthermore, we reveal that the photoemission spectra of oxygen in metal oxides are strongly influenced by (i) uncontrolled carbon-related contamination on the oxide surface, (ii) variations in cation composition, and (iii) the presence of adjacent layers interfacing with the oxide channel. These factors collectively lead to misinterpretation when oxygen vacancy states are inferred solely from XPS peak deconvolution. Additionally, we show that ion gun cleaning, often employed to mitigate surface contamination in XPS measurements, significantly alters the intrinsic characteristics of metal oxide films, making it unsuitable for probing their native states. Depth-profile XPS analyses using ion gun etching, commonly applied to multilayered structures, can also distort the initial characteristics of the films and thus produce misleading conclusions. Overall, our results demonstrate that conventional XPS O1s analysis cannot directly identify or quantify oxygen vacancies in oxide semiconductors. A critical reassessment of the long-standing interpretation of XPS O1s features is therefore essential to accurately understand defect physics and carrier formation mechanisms in oxide semiconductors.