The goal of PICOMETRICS was to characterise the 3D atomic structure of nanomaterials in their native state using TEM. However, in order to avoid damage caused by the illuminating electron beam, the number of electrons used to acquire the data needs to be tuned down significantly hampering a quantitative analysis. To overcome this problem, the team combined novel data-driven methods and experimental strategies to locate and identify all atoms in 3D.
A method has been developed to optimise the electron dose. This is a crucial step since on the one hand we need sufficient electrons to get an image of sufficient quality but on the other hand damage caused to the material will increase with a larger electron dose. This problem can be compared to the use of a medical scanner, where one wants to keep the dose as low as possible to prevent damage. This is all much more critical for children than for adults and also here some nanomaterials are more sensitive to damage caused by the electron beam than others.
For beam-sensitive materials, the optimal electron dose is sufficiently lower than the traditionally used electron dose. As a consequence, the signal and contrast that can be observed in the resulting images is lowered as well. To avoid human bias introduced when visually interpreting such images, we developed a statistical model to resolve fuzzy images down to single atoms - even light ones that only weakly scatter electrons and are therefore hard to spot.
Not only do we want to detect light elements from experimental data, we also want to determine the chemical composition in 3D at the atomic scale from low-dose transmission electron microscopy data. Therefore, a combination of statistical techniques and detailed image simulations was proposed. Furthermore, the PICOMETRICS team has developed a novel statistical model to reveal atomic rearrangements in nanoparticles over time. The atoms in a nanoparticle are hiding behind each other in atomic columns. We only see a top view projection of this stack of atoms in the electron microscopy image. Our goal is to find precisely how many atoms hide in each atomic column of the nanoparticle, thus unravelling the atomic structure. Prior research efforts have enabled us to count the number of atoms with single-atom precision. From one snapshot, it is however impossible to gain insight in the dynamics of the nanoparticle. The new model is therefore specifically designed to analyse a series of images. During the recording of this movie, the atoms could move around in the structure. The model estimates how likely it is for an atomic column to lose or gain one or more atoms from one frame of the movie to the next. Using these probabilities, the model reveals the most likely number of atoms in each column at each time as illustrated in the animated image.
In combination with methods developed to extract the 3D atomic structure, this novel strategy of analysing a time series of images opened up new opportunities since we can follow how the structure of nanoparticles changes under realistic conditions such as a gaseous reaction environment or varying temperatures. Such experiments are not only far more complex in practice, they also lead to new challenges since extra distortions appear when imaging in a gas flow. The PICOMETRICS team therefore used artificial intelligence to correct these distortions showing impressive results. Another breakthrough of PICOMETRICS was the development of two highly dose-efficient real-time imaging methods. Both methods have shown to be very promising for investigating beam-sensitive materials, including metal halide perovskites and metal-organic frameworks.
