3D reconstruction of nanometric objects from stereoscopic electron microscope images
Published : 8 October 2019
Keywords : Images treatment, GPU programmation, optimisation, inverse problem, stereovision, neural networks.
Robust, non-destructive and fast 3D metrology is a world-wide major challenge of microelectronics industry for defects inspection, optical lithography fidelity and process control. Fast reconstruction methods from stereoscopic electron microscope (SEM) images based on geometrical considerations allow to reconstruct 3D topography of micronic objects. However, those technics cannot be applied on nanometric objects because of local physical phenomena which disturb the placement of the points of interest . Alternative methods based on the resolution of inverse problem have been already prototyped. Significant improvements of the computation time are expected after their implementation on the GPUs of our group. Model calibration methods must also be developed, potentially based on neural networks.
3D metrology based on SEM images arouses the interest of several LETI industrial partners, and this thesis is intended to be a key element for present and future collaborations in this field.
The objective of this thesis is to develop a 3D metrology from SEM images the most precise and robust as possible. For this, the PhD student will initially use the group’s theoretical and simulation resources to improve and develop new SEM imaging models. The scope of these models is broad, from the simulation of micrometric objects to nanometric structures. CEA-LETI has a new generation of SEM that allows to image patterns at different points of view. These multi-stereo images allow an increase of data on the image structure which facilitates its 3D reconstruction, compared to the case of a single SEM image taken in top view.
The PhD student will train the SEM models with a collection of stereoscopic SEM images of patterns, which 3D topographies will be known from reference 3D metrology. The student will investigate, in a second time, different mathematical strategies for the 3D reconstruction, allowing fast and precise convergence.
Eventually, 3D reconstruction will be applied on different customer products of interest.