Scanning Electron Microscopy (SEM) imaging is the current reference method for quality control in the microelectronic industry, due to the size of the objects involved and to the yield expected when these tools are used in production lines. In order to improve our knowledge on physics in play during imaging and to develop more performant post-processing software, synthetic images are necessary, representative of real images obtained in clean room conditions. Various state-of-the-art models are available to produce such images (by Monte-Carlo or mathematical approaches) but they are limited by their medium runtime performances or their lack of some typical imaging artifact. Such models are available in our laboratory and can be used during the PhD. Another solution deployed in other fields to rapidly produce images with realistic features is the image generation by deep learning neural network.
The goal of this thesis is to develop one or multiple deep learning models able to produce realistic SEM images from a chip design. Model hyperparameters as well as input dataset will be optimized to obtain images as close as possible to reality, according to metrics to define. This model will then be used to infer the chip design from the SEM image and/or reconstruct the 3D topography of the sample based on a top-down image. This thesis address the following question: to which extent a deep learning model is able to extract advanced metrology information from a SEM image, and what are the associated optimal conditions?
The Computational Patterning laboratory host this thesis. It is specialized in numerical methods to optimize manufacturing processes in clean room. The supervisor team from CEA is specialized in lithography process and numerical modelling, and is associated with Gipsa Lab for their expertise in neural network. This work is a follow-up of proof of concept developed since 2018 in the lab. The thesis contract is a fixed-term contract of 3 years, with a gross salary of 2043.54€ the first two years and 2104.62€ the final year. The PhD student will take part in scientific publications and international conferences. Competences developed during the PhD will be valuable for future positions in diverse technological domains, in particular in the current context of expansion of artificial intelligence.