A Physics-Guided Generative Diffusion Model for Super-Resolution AFM of Trap Dynamic Behavior

Atomic force microscopy (AFM) offers nanoscale insights into material properties, yet its intrinsic speed-resolution trade-off severely restricts the observation of dynamic processes. Although deep learning-based super-resolution holds promise, existing models often fail on experimental data due to a "paradigm mismatch" between synthetic training degradations and the physical subsampling of AFM. Here, we present StableAFM, a physics-guided generative framework designed to break this barrier. By integrating a physically consistent degradation model with a novel latent-space diffusion posterior sampling strategy, our approach restores high-fidelity features from rapid, sparse scans while suppressing generative artifacts. We demonstrate the method's transformative potential by accelerating Kelvin probe force microscopy (KPFM) acquisition fourfold to map charge dynamics in multilayer ceramic capacitors. The super-resolved ISPD analysis further resolves an annular-like depletion pattern within the deep-level electron-trap aggregates. This ability to resolve complex, dynamic sub-micron structures establishes StableAFM as a robust, high-throughput pathway for advanced functional material characterization.