4D STEM Strain Mapping
Mechanical and electrical properties of materials are directly related to the variations in their atomic spacing (strain). In a transmission electron microscope (TEM), one can measure the crystal lattice deformation by comparing the interplanar distances (d) in the interest area to an unstrained reference region.
Imaging methods (e.g., high-resolution TEM (HRTEM) or scanning TEM (HRSTEM)) measure the interplanar distances (d) and diffraction methods (e.g., nanobeam electron diffraction (NBED)) measure the reciprocal distances (g = 1/d).
Both of these methods in the TEM provide two-dimensional (2D) measurements of the crystal lattice deformation, which you can represent using a 2 x 2 matrix. You can then calculate the 2D strain maps, as shown below:
4D STEM data collection considerations for strain mapping
For the acquisition of 2D strain mapping data, the specimen and microscope setup are critical. This method depends on a 2D array of non-overlapping diffraction disks, with little contrast within the disks, as well as sharp edges to the disks.
Convergence angle: The spatial resolution is determined by the convergence angle (condenser aperture and lens configurations). A larger convergence angle produces a smaller STEM probe (better spatial resolution) and vice versa. On the other hand, the diameter of the diffraction disks directly relates to the convergence angle. A larger convergence angle gives larger disks, and depending on the d spacing, there is a higher chance of disk overlap (should be avoided).
Camera length: Select a camera length that allows you to image the first 2 or 3 orders of diffraction disks on the camera. Keep in mind that the larger camera length spreads the disks over a larger number of pixels (suitable for disk fitting algorithm), but results in less signal per camera pixel (bad for disk fitting algorithm). You should also consider this when selecting the convergence angle.
Specimen thickness and orientation: Choose a sample thickness thick enough so that there is a reasonable amount of scattering (without stress relaxation), but not so thick as to generate inelastic scattering (too much sub-structure in the diffraction disk and poor disk fitting).
Other considerations: Energy-filtered 4D STEM can reduce the inelastic scattering effects and greatly enhance strain mapping results.
The sample data we use here was provided by the researchers at the Ernst Ruska Center for Microscopy and Spectroscopy with Electrons, Germany.
The sample is a Si/SiGe multilayer with four layers of SiGe with different Ge concentrations (21%, 31%, 38%, 45%). It was epitaxially grown, without any dislocations. Since the lattice parameter of Ge (0.5658 nm) is larger than that of Si (0.5431 nm), we expect positive strain in the y (growth) direction and zero strain in the x (in-plane) direction. Strain in the y-direction is expected to increase with Ge concentration (up to 3% at the highest Ge concentration, based on simulations).
4D STEM diffraction data was collected with a STEM probe of ~3 nm and a semi-convergence angle of 0.6 mrad. Camera length was adjusted to make sure diffraction spots up to the 2nd order were captured with a sufficient signal-to-noise ratio in the diffraction image.
Strain calculation in GMS
Strain calculations inGMS are done on a 4D STEM diffraction data (NBED) using the Strain Mapping technique and palette (screenshot below) and follows the flowchart shown above.
- Define the reference pattern
The first step is to define a reference diffraction pattern from a region of the sample that has no strain. In the above example, we use the Si substrate. GMS calculates an average diffraction pattern from all the pixel positions in the selected unstrained region (image N), then displays three red circles (center, u, and v unit vector selection) and one green circle (spots-spot search area selection) on this image.
Note: If you do not specify an unstrained area in the sample, the reference diffraction pattern will be generated using the full ROI. This means that the strain calculations will use an average (and not necessarily unstrained) reference, and as a result, not the absolute values.
In the reference pattern, specify a few parameters while keeping in mind the following considerations:
- Unit vector selection: The three red circles on the reference diffraction pattern are to specify the center, and the unit vectors (u and v), and are used to create a template to detect the rest of the diffraction spots in the search area.
- The center circle is positioned on the direct beam (center of the diffraction pattern usually). The size of this circle sets the area used to generate the diffraction disk template, so you should choose its size to be large enough to include a margin around the visible disk, but not so large as to get close to any other disk.
- The u and v circles are positioned along the crystallographic directions for which the strain is to be calculated (e.g., in-plane and growth directions, respectively). The size of these circles defines the search radius used when analyzing all of the diffraction patterns to find the spot centers.
Note: As shown below, selection of the higher-order diffraction spots results in longer u and v unit vectors. This means the lower order diffraction spots are excluded in the strain calculations, and the noise will be higher in the resulting strain maps.
Spot search area selection: GMS only searches for diffraction spots in the area selected by the green spot circle. The smaller circle means a smaller search area per diffraction image, which speeds up the strain calculations. On the other hand, the diameter of this circle defines how many orders of the diffraction spots are used for these calculations. As shown below, although the noise is not much affected by this selection, including higher-order spots in the calculations results in strains that are closer to what is expected for this specimen from simulations (strain should increase with Ge concentration).
- Create a disk template
The center spot marked on the reference diffraction pattern is used to generate a template representative of the shape of all the disks in the dataset. This template (image E in the screenshot above) is used along the directions of the u and v unit vectors to find the position of all the spots on diffraction patterns at each STEM probe position to then calculate the strain.
- Calculate the strain maps
Now that all the parameters are defined, strain maps can be calculated. GMS uses the u and v unit vectors and the disk template from the last two steps to calculate the best fit lattice for each diffraction pattern (probe position). The strain is then calculated using the lattice deformation in comparison to the reference lattice (see the Background section). A couple of other points to consider:
- In the current implementation, GMS calculates the projected strains along the horizontal and vertical axis of the diffraction pattern and not the crystallographic directions specified by u and v vectors. The user is encouraged to measure the diffraction rotation angle during the microscope session and input this value so it can be applied to the diffraction patterns prior to strain calculations.
- During strain calculations, if the scattered intensity in diffraction spots for a specific probe position is small, then GMS does not perform any analysis on that diffraction pattern, and the strain components are all set to zero at that location.
As mentioned above, accurate determination of the position of the diffraction spots is critical for these measurements. Currently, GMS uses cross-correlation for this purpose. For a better comparison of different correlation methods, please read Ultramicroscopy 176 (2017) 170–176.
J. Ciston and C. Ophus (National Center for Electron Microscopy, CA, USA) provided consultation for strain mapping technique development and implementation in GMS. We would also like to thank T. Denneulin and V. Migunov (Ernst Ruska Center for Microscopy and Spectroscopy with Electrons, Germany) for providing this dataset, figures, and discussions on parameter optimization for strain calculations.