Direct nano-scale microanalysis is important for photovoltaic functional thin films to characterize their homogeneity and purity. This demands combining spatial resolution in the micro/nano-scale and sensitivity in the trace-level range, which is at the moment beyond state-of-the-art. As dictated by counting statistics, the reduction of the spot size degrades the detection limit. The utilization of a tabletop XUV laser at λ = 46.9 nm has shown to dramatically improve the ablation efficiency with respect to that of visible lasers, such that ablation spot of 1 μm limits. Li-doped Cu₂ZnSn(S,Se)₄ (so-called kesterite) thin films were irradiated across 3D ablation arrays for hyperspectral mapping by means of time-of-flight mass spectrometry. The nominal 3D data node lattices were the initialisation perceptron, filled with measured values, and for a detailed supervised learning postprocessing, the node-to-node links were analysed by means of a 2D-kernel covariance algorithm. The latter permitted to obtain robust 3D elemental distribution functions well below the measurement spacing, giving insights into the inhomogeneity and impurities.