Dynamic STEM-EELS: MoS2

These datasets consist of STEM-EELS point measurements collected from autonomously identified regions of interest in V-doped MoS2, using Nion UltraSTEM100 instruments. This drastically minimizes deposited electron dose and allows to observe and analyze dynamic transformations under the beam, which are typically not able to be measured in standard grid-based approaches due to much higher dose rates.

The data are separated into low-loss and core-loss classifications. 

The low-loss data have used monochromation in order to access the MoS2 band gap energy and observe any changes that defects might have on the local electronic structure as measured by STEM-EELS. In order to increase locality of the low loss EELS measurements, the off-axis EELS configuration was used (by electrically displacing the beam relative to the EELS aperture and camera), but this reduces the scattered signal by several orders of magnitude given that it is the dark field that strikes the EELS detector.

The core-loss data were collected using a smaller dispersion to access wider energy ranges and without monochromation. On-axis EELS configuration was used, where the core loss of structures evolving under the beam, as well as individual V substitutional defects, were collected autonomously. 

The files here exist in .ndata1 format, which effectively combines numpy arrays (.npy files) and metadata (.json files) into one file. 

Each "experiment" is broken into 3 file types: ADF, Label, and Measurements:  The 'ADF' is the image data, the 'Label' is the as-predicted image and classified coordinates, and the 'Measurement' is the EELS measurements. Further, each of these is stacked N times per experiment - in other words, a single experiment consists of N ADFs, N Labels, and N measurements, where N is the number of iterations that consist of one experiment. Typically N is on the order of 5 or 10.

The logic behind this is that the experiment proceeds by: collecting image #1, labeling image #1, measuring selected regions within #1 ---> collect image #2, label image #2, measure selected regions within #2, and so on, where image #2 contains information on if changes have occurred since image #1 (whether this is drift or structural changes).  This is done multiple times because the beam will induce changes between frames and we can explore multiple defect states between frames.