In this combined theoretical and experimental project, you will be addressing one of the most pressing problems in modern electron microscopy: the problem of beam damage.
To obtain transmission electron microscopy (TEM) images with atomic resolution, at least 100 high-energy electrons pass through the space occupied by each atom. The energy of these electrons are many orders of magnitude higher than the energy required to break a chemical bond. So it is not surprising that the electron beam influences the sample, both by directly damaging it, and by inducing diffusion and chemical reactions.
Unfortunately, beam damage is poorly understood, both theoretically and phenomoenologically. It is for example not even known when it is the total dose (electrons per area) or the dose rate (electrons per area per time) that is most critical for inducing beam damage. A major problem in studying beam damage it to move from qualitative observations of beam damage, to quantitiative measurements. Here, deep learning comes to the aid.
Deep neural networks allow us to analyze TEM images automatically, identifying where the atoms are, and whether they are moving [1,2]. This motion of the atoms will be influenced by the beam, and by measuring the diffusion as a function of the beam parameters (dose, dose rate and electron energy) we will be able to obtain quantitative measurements of the influence of the beam.
In the project you will be recording image sequences on some of the best transmission electron microscopes in the world, and will be analysing them using modern neural networks. You will be training neural networks suitable for analyzing your own data, and will be writing Python scripts to process the output from the networks and to do the statistical analysis.
Knowledge of Python programming is an advantage, and so is an interest in machine learning / neural networks. Prior knowledge of machine learning is not necessary. You should at least have passed a course in solid state physics (e.g. 10303) OR in electron microscopy (10250).
Figure: A TEM image of a gold nanoparticle on a CeO2 substrate, taken from an image sequence (a TEM video). Analysis of the entire video showing how often the individual surface atoms diffuse. Is this diffusion induced by the electron beam?
 J. Madsen et al.: A Deep Learning Approach to Identify Local Structures in Atomic-Resolution Transmission Electron Microscopy Images, Adv. Theory Simul
, 1800037 (2018). Preprint available at https://arxiv.org/abs/1802.03008
Knowledge of solid state physics OR electron microscopy. Some Python programming.