A Different Approach To TEM Sample Prep

Trevor O'Loughlin

University of Rochester, The Institute of Optics

1. Introduction

Ordinarily, crystalline structures, such as semiconductor detectors, require one to go through the lengthy process of focused ion beam ablation to create a TEM sample. This sample can be used for many different purposes, such as confirming the correct layer thickness while looking for any possible problems with the growth. This project was meant to take a step back and investigate whether the FIB process is necessary, or whether a simpler process may work just as well, with less user effort.

A view of the design

Figure 1: The structure is grown by a process known as Molecular Beam Epitaxy (MBE). The structure shown here is the sample investigated.

The design looked at here is a III-V infrared mirror in the mid wave infrared (2-5 microns) region, consisting of alternating layers of AlAsSb and GaSb grown on a GaSb substrate.

For these mirror stacks to work as intended, both the growth thickness and the composition must be controlled. While the conditions before and during growth are carefully calibrated, post growth analysis is oftentimes difficult, as the many layers can cause uncertainty in what exactly went wrong. A secondary goal of this project is to see what information may be gained from TEM analysis.

Several techniques were used in the process of studying these samples. Sample preparation was especially important and an ultramicrotome was used to try to prepare small enough pieces for the TEM. The samples were imaged in the TEM in brightfield, as well as a diffraction mode. The Scanning TEM mode was used as well to gain additional information and the EDS tool was used in conjuction to create atomic composition maps of the surface. Finally, a few of the images were colorized using ImageJ.


2. Sample Preparation

TEM samples require special consideration when preparing. It is important that at least sections of the sample are thin enough that the electron beam may pass through, as that is the primary imaging tool. To try to create these samples easily, an ultramicrotome was used. This tool consists of a sample holder and a glass knife, formed by scoring the surface and allowing the atomic bonds of the glass material to break, resulting in an extremely sharp surface. Usually, this is used to cut off small strips that trail together over water. However, such a knife is also capable of chipping off small pieces of a harder material. Due to the ability to set an angle, preference can be given to a side of the sample too. So, in theory, the ultramicrotome could be used to chip off what are small chunks of mostly the top layers of the structures (called epitaxial layers). One problem with this technique is that the hard sample chips the blade, which means the blade stops cutting and simply pushes the sample around.

Figure 2: Left: View of the microtome in action, chipping off pieces of the sample. Right: Zoom in on the knife, showing a chip in the blade.

A second idea was formed during the project. Since these samples are only a few microns thick, it should be possible to scratch off the surface and collect the particles. One or more of these particles should allow for viewing of the structure. In order to accomplish this, a diamond scribe was scratched over the surface and the particles were collected in methanol then deposited onto a TEM grid.

Figure 3: Blurry picture of the scratched sample.

3. Brightfield TEM

Transmission electron microscopy (TEM) is a technique used for imaging thin (~100nm thick) samples using an electron beam. Brightfield imaging functions similarly to traditional microscopy, except electrons pass through the sample and are collected on the other side to form an image rather than photons. Because of the high voltage used in the column (and therefore short wavelength), extremely small features can be made visible. A FEI Technai TEM was used to take these images.

From the images, it is noticeable that both processes produce samples that can achieve atomic resolution. This is important for accessing device growth, as the interface between different layers can have a large impact on device performance, and thus looking at that interface, down to the atom, can be of interest. It also allows one to see problems created during growth, such as dislocations in the lattice.

As mentioned earlier, most mictotoming is done on softer samples, where the samples form a train. Here though, it looks like the samples stack on each other, making imaging difficult. Therefore, these samples can only be imaged on their edges. X-ray analysis also shows that there is no aluminum present, meaning that this is bulk material. This was found throughout the sample and no epitaxial layers were found with this technique. This implies that even though the microtome was set to prefer the epitaxial region, in imaging they do not come across.

It is also interest to note that there is a thin amorphous layer around the edge of the sample. A clear suspect is that it is a thin oxide layer, as oxidation is a problem with devices, but no oxygen was seen in the X-ray measurements. Another suspect could be water contamination as these were floated on water during the microtoming process, but again, no traces were found in X-ray measurements. It is not clear what this amorphous layer is.

Figure 4: High Resolution pictures of the microtomed sample. The circled part shows a possible dislocation point in the lattice, as the regular order is shifted here. It should be noted X-ray analysis showed this region as all GaSb, so this is part of the substrate. The outside amorphous layer is also visible.

The scratching technique had a bit more success. Atomic resolution is possible here too, and because this technique creates smaller samples, there were some inner structures visible. Epitaxial layers were also very visible too, making interface investigation possible. The strange amorphous layer is visible here too.

Figure 5: High resolution picture of the scratched sample, taken on the epitaxial region. No differences between the two regions are visible, indicating good growth.

Finally, it is worth noting that the beam exposure seems to distort the image. Different parts of the image are brought in and out of focus over time. This could be especially troublesome for long exposures or long working times, as we do not want the structures to change.

image 1 image 2
image 3 image 4

Figure 6: Time progresses left to right, top to bottom. Notice how the lines in image switch from a top-left to bottom-right slant to a top-right to bottom-left slant between the 2nd and 3rd image. It is still unclear what exactly is going on here.

4. Diffraction patterns

Diffraction patterns are especially important in crystalline structures, as they can show the angle of viewing along the lattice.

Figure 7: The first two figures are the computed Fourier transforms of the right picture of Figure 4 and Figure 5, respectively. Due to the hexagonal nature of the image, it is probably being viewed along the (111) plane. The bottom picture is a true diffraction image, with the first order blocked. The square nature implies viewing along the (100) plane.

At the high magnification the bright field images were taken at (>300k magnification), the Fourier transform of the image should basically be the diffraction pattern.

5. STEM/EDS

The initial plan was to use EDS (energy dispersive spectroscopy) to deduce the composition of the grown layers. Theoretically, this is still possible, however, the gold grid is a problem. Arsenic EDS peaks are right around gold peaks EDS peaks, meaning that there is too much interference to get a good idea of how much arsenic there actually is in the sample.

Figure 8: As can be seen here, gallium, antimony and aluminum can all be clearly visible. However, the arsenic peaks are overshadowed by gold and aluminum.

The use STEM was a much greater success. The different layers are clearly visible and the use of EDS allows maps of the surface to be made. The layers may now be measured for thickness.

Figure 9: STEM picture of epitaxial layers of the growth.

Figure 10: STEM picture and element map. The colors represent aluminum (red), gallium (orange), arsenic (green), and antimony (blue) placement. It can be seen that the gallium and aluminum are organized into bands. The arsenic displays a lot of noise. Interestingly, it appears the antimony shows up more in the AlAsSb layers than in the GaSb layers.

Figure 11: STEM picture and element map (drift corrected). The colors represent aluminum (red), gallium (orange), and antimony (green) placement. There appears to be a missing layer on the right, possibly from sample prep.

It is here that another problem becomes apparent. Figure 1 shows the desired structure and thickness of these materials. Optical measurements put the materials to be around 10% thicker than anticipated, so the GaSb layers should be around 269 nm and the AlAsSb layers should be around 312 nm. Measurements on Figure 9 and 10 put the thicknesses around 337 nm for Al and 437 nm for Ga however, while measurements on figure 11 put the thicknesses around 325 nm for Al and 219 nm for Ga. These numbers do not line up with each other or the expected thicknesses. It could be that viewing angle is playing a part, as well as some fracturing from the preparation process. Basically, some of the layers fracture off, leaving empty space behind. Then, the viewing angle allows the layers to hang over the empty space, making one layer look bigger and one layer look smaller. This seems like the probably cause for the discrepancy, but further investigation would be needed to confirm this.

6. Colorization

ImageJ was used to colorize a few of the images.

Figure 12: Colorized pictures of a brightfield TEM lattice, low magnification STEM image of material, and a higher magnification STEM image of the epitaxial layers.

7. Conclusion

Though not as thorough as FIB may be, these alternate techniques for TEM processing nonetheless provide information to the user at a much lower time commitment. They can be designed to show epitaxial structures or preference the substrate. They are not without their faults, as if stoichiometric data is desired, care must be taken with the grid choice and the accuracy of thickness measurements is currently under investigation as well. Nevertheless, this experiment has shown that alternate, easier techniques to FIB do exist for crystalline samples.

Acknowledgments

I would like to thank Brian McIntyre for his help and my TA, Rakan Ashour, for helping me with the SEM labs.

Please enter any comments, criticisms, questions, etc. below.

Your name:

Email address:

References

B. Satpati, J.B. Rodriguez, A. Trampert, E. TourniƩ, A. JoulliƩ, P. Christol, J. Cryst. Growth. 301-302 (2007) 889-892.