Answers are grounded in this paper and cite their sources.
Amorphous materials offer unique functionalities but remain poorly understood because structural disorder obscures links between local environments and macroscopic properties. Electron energy-loss spectroscopy (EELS) probes local chemistry, yet fine structure is often broadened and overlapping, limiting interpretation. Here we present a manifold-learning framework in which a variational autoencoder (VAE) embeds paired structural descriptors and O K-edge spectra in amorphous In–Ga–Zn oxide (a-IGZO) into a shared latent space. Gradient-guided latent perturbations quantify conditional spectral sensitivities and descriptor interdependencies on a data-constrained manifold, recapitulating established O K-edge fingerprints, including cation-specific modulations consistent with In-, Ga-, and Zn-oxide–like local environments and enhanced pre-edge features associated with peroxide-like O–O configurations. An auxiliary network maps experimental spectra to this space and enables descriptor-level inference, allowing spatial mapping of local descriptors that are difficult to access directly from experiment. Applied to a-IGZO devices, the approach resolves interfacial and growth-condition trends that are independently corroborated by EDS and EELS edge-integral quantification. This work establishes an interpretable platform for systematic, data-driven exploration of a-IGZO.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: