Gradient-Based Experimental Design for Defect Detection in MoS2 Including Emission Potentials for Thermal Diffuse Scattering

This is the authors' abstract. We don't add key points for this paper.

Momentum-resolved scanning transmission electron microscopy (STEM) augmented the methodological portfolio for structural and chemical analysis enormously. Here, we compare and benchmark different methods such as bright field, dark field, as well as in-focus and defocused pixelated STEM, against each other as to their dose efficiency for detecting intrinsic defects in MoS2. To this end, ground truth simulations are subjected to hypothesis tests comprising an atomistically parametrized structural model and computationally efficient scattering approaches. In particular, established loss metrics are used to compare model and observation under multiple configurations of Poissonian noise with the dose systematically varied. A statistical analysis of loss gradients, calculated with respect to continuous weight factors for atomic scattering amplitudes, provides success rates for the detection of sulfur vacancies, selenium monomers, and substituents. A scheme is presented to derive signal-dependent critical doses for confident defect detection, which lie in the order of 103e-/Å2, with the defocused pixelated setup showing the best dose efficiency. The emission potential for thermal diffuse scattering at high angles is revisited and introduced into the gradient calculation as a computationally efficient alternative to the frozen phonon approach to include atomic number contrast in inverse multislice.