Investigating SEM Interaction Volume by Studying Oxides on Silicon

William D. Hughes

University of Rochester, Institute of Optics

1. Introduction

A Scanning Electron Microscope (SEM) utilizes an electron beam to image a sample and generate X-rays for compositional analysis. These require the beam to interact with the sample of interest. The region where this occurs is the interaction volume. The interaction volume depth and size are dependent on the type of signal that is detected and the accelerating voltage of the beam.

The SEM requires properly grounded conducting surfaces in order to image. Typically, insulating surfaces, like silicon dioxide, cause charging on the surface which creates imaging artifacts. However, if the insulating layer is thin enough, most of the interaction volume will lie in the substrate beneath. This project aimed to find more about the thinness requirements for good imaging. The project took a conducting substrate, silicon, and grew different levels of silicon dioxide on top of it. The cross sections of each sample were imaged for thickness measurements. Once thickness measurements were obtained, the interaction volume was modeled for various accelerating voltages. X-ray compositional analysis was then taken to compare the model to experiment. Finally, images were taken with the oxide layer on top to show if and when poor quality images occur.

Figure 1: This diagram illustrates the interaction volume of an electron beam on a sample.

2. Procedure

Sample Preparation:

Silicon from off of the shelf was used in this project. The first step was to cleave it into several pieces. These pieces were about 5 cm x 2 cm in size. There were about 15 samples. All of the samples except one were placed into the small tube furnace in the cleanroom at URNano. The furnace was operated with normal procedures. Samples were taken out of the furnace after 1 hr. 15min, 2 hr. 15 min, 3 hr. 15 min, 4 hr. 45 min, 6 hr. 30 min, and 8 hr. It is important to note that these were done over two days. After 3 hr. 15 min, all remaining samples were removed as time had run out on the first day. Samples were replaced on another day. The time of the rest of the samples is a sum of the two runs on the furnace. This time included warm up time of the furnace on both days so a repeated procedure must be aware of this. Next, the different samples were cleaved again to get small enough pieces to go into the SEM. The six oxide samples were placed on a hex nut with a cleaved side facing upwards (shown in Figure 2). These samples were secured with carbon tape on both sides, to ensure proper grounding. The hex nut was then placed in the sputtercoater in the Institute of Optics at the University of Rochester. The samples were coated with a fine grain gold target. Next, pieces of the six oxide samples and plain silicon were placed on sample stubs with carbon tape. These samples were not coated.

Figure 2: Samples placed on hex nut for oxide thickness measurements.


The thickness of each sample was found by imaging the hex nut samples in the Zeiss-Auriga SEM at the Institute of Optics. Next, the electron flight simulator was used to model the interaction volume of the various samples. The normal (oxide on top) samples were then placed in the SEM. X-ray data was taken a 1,000x magnification on an apparently uniform section of the sample. These X-ray measurements were taken at 5kV, 10kV, and 20 kV. Then, regular secondary electron images were taken of these samples at various accelerating voltages to see the effects the oxide may cause.

3. Thickness Measurements

The cross sections of the samples were imaged to find the oxide layer that existed on the edge. These needed to be quality images so the samples were properly prepared with a gold coating. Imaging conditions were varied to maximize the clarity. For five of the six samples, the InLens detector was used to image at a close working distance. This allowed better resolution to find the oxides. The largest oxide sample was imaged with the backscattered electron detector. This was used because it clearly showed an atomic weight contrast between the bulk and the thin film. The backscattered electron detector detects atomic weight differences best of the three electron detectors. X-ray spectra were qualitatively taken on each sample to prove that it was an oxide that was being imaged. Additionally, some of these images were colorized using ImageJ.

Figure 3: Six different oxide samples imaged with thickness measurements. The samples are ordered from shortest time in the furnace to longest. They are lettered B to G. Sample B has a thickness of 119 nm, Sample C has a thickness of 232 nm, Sample D has a thickness of 275 nm, Sample E has a thickness of 354 nm, Sample F has a thickness of 688 nm, and Sample G has a thickness of 1047 nm.

The results in the images make physical sense. The rate was slower than anticipated for the first four samples. Sample E was expected to be closer to 500 nm. The likely cause of the difference from expectation was the fact that the furnace had to warm up on the second day, and Sample E was taken out an hour and a half after being put in the furnace. It only grew an additional 80 nm on top of what was already there as seen in Sample D. The rate increased for the last two samples as the rate towards the end was well over 100 nm/hr. This was an unexpected result as the rate should in general decrease as the oxide grows thicker. This could have happened because the samples were not taken out as often on the second day in the furnace. This allowed the samples to stay hot as the set up required removing every sample temporarily in order to remove any sample.

The images above required time to obtain. The magnification of each image was high which increased the difficulty of obtaining quality images. The aperture size was too large when first obtaining the images. The quality of images greatly increased once the aperture size was reduced. This decreased the negative effects of spherical and chromatic aberrations as the exterior of the beam path was cut off. Additionally, it blocked scattered electrons which increased the contrast of the image.

4. Interaction Volume Model

The computer on the SEM has electron flight simulation software that models where electrons go when they hit a particular surface. The samples in this project easily could be modeled because only silicon and silicon dioxide were the materials in the sample. The models are depicted below. The bars to the right indicate the X-ray emission strength of silicon at that particular depth. It allows for a qualitative assessment of where the X-rays should be emitting from.

Figure 4: Sample B modeled at different accelerating voltages.

Figure 5: Sample C modeled at different accelerating voltages.

Figure 6: Sample D modeled at different accelerating voltages.

Figure 7: Sample E modeled at different accelerating voltages.

Figure 8: Sample F modeled at different accelerating voltages. The 5 kV model did not work as the oxide was too thick for that voltage.

Figure 9: Sample G modeled at different accelerating voltages. The 5 kV model did not work as the oxide was too thick for that voltage.

The models predict some general trends. At 20 kV, no matter the oxide thickness, a majority of the interaction volume is in the substrate. However, the percentage naturally decreases as the oxide gets thicker. At 10 kV, the thinnest oxide (Sample B) has an interaction volume mostly within the substrate. However, as the oxide gets thicker, a majority of the interaction volume is in within the oxide layer. At 5 kV, the thinnest oxide layer has an interaction volume that is roughly equally within the oxide layer and substrate. The thickest samples have the interaction volume completely within the oxide layer.

5. X-Ray Data

The X-ray data was taken with each of the oxides facing upwards, toward the electron beam. At zero tilt, this allowed the beam to hit the surface at normal incidence. X-rays are generated when the high energy electrons knock a low energy level electron out of the electron shell of a particular atom, allowing a higher energy electron to relax to that level and emit an X-ray photon of the energy difference. Each type of atom has its unique set of X-rays it can produce, so it is a way to determine compositional characteristics. The samples in this project contain silicon and oxygen. The oxide is one part silicon to 2 parts oxygen.

The X-ray data was taken at 1000x magnification, at 5 kV, 10 kV and 20 kV. It was important to check the image to ensure that it was a uniform, clean surface that was being analyzed.

Table 1: X-ray Data.

This data shows where the interaction volume actually is for characteristic X-rays. The % oxide column indicates what percentage of the X-rays are coming from the oxide layer (assuming perfect uniformity within it). The data qualitatively agrees with the model. At 20 kV, a majority of the signal is consistently in the substrate. The model predicted a similar characteristic. The model was slightly off when compared at 10 kV. The model predicted that at the thickness that was equal to Sample F, a majority of the interaction volume would be in the oxide. The X-ray data indicated that this did not happen until Sample G, where the thickness was over a micron. At 5 kV, the data generally agrees with the model. Though, Sample B data indicated less oxide contribution than the model did. Also, the model predicted that Samples F and G would have an interaction volume completely within the oxide but the data showed a small excess in silicon for this to be the case. potential cause of this is nonuniformity within the oxide layer. There may have naturally been excess silicon because some silicon got trapped within the oxide layer as it grew or there possibly was a diffusion of silicon across the interface.

6. Imaging Effects

As Figure 1 shows, secondary electrons have a smaller interaction volume than characteristic X-rays. This means that the percentage of the interaction volume that is an oxide that was found in the previous section would be more for secondary electrons. The oxide layer is insulating, which if the electrons build up on the surface, artifacts occur because of deflections of the beam due to the charge. If more of the interaction volume is in the insulating layer, poor images would be expected. The images below test this concept.

Figure 10: Low voltage causes charging artifacts in samples C (left), E (middle), and G (right).

Figure 11: High voltage (20 kV) images of samples F (left), and G (right). Charging is nonexistent.

Figure 10 shows the outcome of an interaction volume mostly within an oxide layer. The surface built up charge and the quality of the image degraded. The streaks across the edges and the image are evidence of the charging. The electrons were left in the insulating layer. Generally, a strategy to reduce charging is to reduce the accelerating voltage. This causes fewer electrons to hit the surface and less charge will build up. However, in this particular circumstance, charging can be reduced by increasing the accelerating voltage, as seen in Figure 11. This occurs as more electrons went into the conducting substrate and thus did not sit on the surface. Additionally, the interaction volume near the surface usually narrows at higher voltage because the electrons penetrate deeper into the sample before spreading out.

7. Conclusions

An insulating surface does not necessarily need to be gold coated to be imaged. Despite more electrons hitting the insulating surface at higher voltage, charging can be reduced when the interaction volume penetrates into a conducting part of the sample, like silicon. In the case of the oxides presented in this project, the thickest oxide was about 1 micron. Low accelerating voltage (5kV) imaged very poorly, as the electrons only penetrated the insulating oxide. Higher accelerating voltage (20 kV) imaged better. Electron flight simulation served as a guide to understanding the interaction volume. However, the X-ray data gave a more accurate picture as to how deep the electrons in the electron beam go.

The lesson learned from this project is that a counterintuitive approach could solve charging issues in an SEM. Typically, reducing voltage reduces charging. However, the results here show that a thin insulating surface requires the opposite remedy.


I would like to thank Brian McIntyre for the humor and constant advice throughout the semester. I would like to thank my TA, Rakan, for guiding me through each of the labs. Also, I would like to thank Jim Mitchell for his help in the cleanroom with the furnace.

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

Your name:

Email address:


OPT407 Class Notes