Towards model-driven reconstruction in atom probe tomography

Reconstructions in atom probe tomography (APT) are plagued by image distortions arising from changes in the specimen geometry throughout the experiment. The simplistic and inaccurate geometrical assumptions that underpin the conventional reconstruction approach account for much of this distortion. Here we extend our previous work of modelling APT experiments using level set methods to three dimensions (3D). This model is used to generate and subsequently reconstruct synthetic APT datasets from electron tomography (ET) of an $Al\textit{-}Mg\textit{-}Si$ multiphase specimen. Finally, we apply our model to the reconstruction of an experimental field-effect transistor (finFET) dataset. This model-driven reconstruction successfully reduces density distortions compared to conventional methods. By combining prior knowledge about the specimen geometry from sources such as ET, such an approach promises new distortion correcting APT reconstruction applicable to complex specimen geometries.