Semi-automatic tools for nanoscale metrology and annotations for Deep Learning automation on Electron Microscopy images
Résumé
For semiconductor applications, billions of objects are manufactured for a single device such as central processing
unit (CPU), storage drive or graphical processing unit (GPU). In order to obtain functional devices, each individual element of the device has to follow precise specifications at nanoscale. For that, it is essential to obtain
efficiently a pipeline to automate measurements on a set of similar images of semiconductor objects captured
with electron microscopy. Generally, the pipeline consists to annotate an object in a image and then take the
measurements of the object. Manually images annotation is extremely time consuming. In this paper, we propose
a semi-automatic method to annotate an object in a microscopy image based on deep learning with the precision
of the measurement. The approach is a deep learning contour-based method able first to detect the object and
after find the contour thanks to a constraint loss function. This constraint improves the quality of boundary
detail of the vertices of each object by matching the predicted vertices and the most likely ground truth. The
loss is computed during training for each object using a proximal way of our dataset. The experiments showed
that our approaches can achieve a state-of-the-art performance on several microscopy images dataset.