UC Irvine

Encapsulating a permeable layer of oxide support on metal catalysts, as a type of strong metal-support interaction (SMSI), is often adapted to design heterogeneous catalysts with enhanced reactivity and catalytic activity. The success of such a design highly relies on the thickness of the encapsulation layer, often only one or two atomic layers, and its chemical composition and structure, which is highly delicate given its thickness and complex chemical environment. Precisely detecting such a trace layer and determining its chemistry however, is challenging. Scanning transmission electron microscopy (STEM), the most commonly used technique for such analysis, also suffers from a low signal strength due to the thickness and the electron beam sensitivity of the surface encapsulation layer. This can lead to the potential misinterpretation or overlooking of the encapsulation signal. Here, using Pd-TiO2 as a prototype system, we develop and demonstrate an unsupervised machine learning method that allows us to reveal the presence and chemical information of the SMSI encapsulation layer that is otherwise hidden in STEM-electron energy loss spectroscopy (EELS) datasets. This method not only provides a robust tool for the analysis of trace SMSI in catalysts, but is generally applicable to any materials and spectroscopy datasets of any material systems where revealing a trace signal is critical.