Contrary to bulk materials, nanomaterials show very versatile aspects giving access to greater degrees of freedom on the desired physical effects. In this context, nanoparticles (NPs) have established themselves as the basis for new classes of materials whose properties make a scientific and technological break with their 3D counterparts.
Of all the techniques for analysing these extremely small objects, transmission electron microscopy (TEM) is one of the most appropriate because it allows structural and chemical studies of NPs with unrivalled precision. However, TEM generates images with degraded signal-to-noise ratio, contrast and spatio-temporal resolution, which hinder reliable quantification and interpretation of the data. Furthermore, the extraction of structural information from these images relies on manual acquisition and local structural identification, which does not allow statistical analysis of the data and necessarily introduces human bias in the post-processing phase. Recently, the advent of artificial intelligence (AI) algorithms such as “machine learning” or “deep learning” has shown exceptional performance in visual classification tasks. These new techniques have proved revolutionary in a number of fields and can be expected to have a similar impact on electron microscopy as these techniques seem perfectly suited to deal with the greyscale patterns present in TEM images.
The overall objective of the internship will be to develop a unique framework based on deep learning algorithms for in situ gas and liquid microscopy that allows for automated, high-throughput, real-time acquisition and analysis of TEM image sequences. Our approach will be to build a dataset in two steps. The first step will be to generate structural configurations of NPs of different sizes and shapes by atomic scale simulations. For this purpose, we will perform simulations that are essentially based on a tight-binding formalism integrated in structural relaxation codes such as Molecular Dynamics or Monte Carlo. The second step will consist in simulating TEM images from the configurations resulting from the atomistic calculations, including the instrumental noise and the imperfections of the microscope optics. This numerical database will then be integrated into AI codes to analyse experimental TEM data of NPs in vacuum, in gas or in liquid.
It can be noted that this work will be done in collaboration with IFPEN for technical developments and the MPQ laboratory (University of Paris) for the experimental part where the characterizations of the NPs will be carried out by transmission electron microscopy in-situ.
job : intership (4 to 6 months)
Location : LEM, Châtillon
Expertise: Good knowledge of quantum mechanics, solid state physics and statistical physics, and a strong interest in numerical simulation.
Academic degree : Bac +5
Contacts : Riccardo Gatti, Hakim Amara (Email Us)