Expert-guided background processing framework for depth-resolved X-ray magnetic circular dichroism analysis

Abstract Depth-resolved characterization of buried magnetic interfaces is vital for advancing spintronic devices, yet consistent analysis remains challenging. Fluorescence-yield depth-resolved X-ray magnetic circular dichroism (XMCD) enables quantitative profiling of spin and orbital moments, but conventional background (BG) subtraction is empirical and operator-dependent. We present a systematic framework using unsupervised t-distributed stochastic neighbor embedding similarity mapping to automate and optimize BG subtraction. Validation tests confirm robustness against spectral noise while revealing strong sensitivity to BG subtraction, highlighting the critical role of optimized BG treatment. Applied to a Graphene/L1 0 -FePd thin film, the method reproduces interfacial spin suppression and orbital enhancement with expert-level consistency. This framework thus provides a standardized and reproducible approach to depth-resolved XMCD, broadly applicable to interfacial magnetism in spintronic materials.