Hello,

in EDS quantification is done under the assumption that the sample is (1) thick, (2) flat, and (3) homogeneous.

This is not the case for layered systems, therefore the numbers are garbage.

An attempt to overcome this is using measurements with different energies. A free tool is BadgerFilm

https://smf.probesoftware.com/index.php?topic=1373.0

​

​Reviews of EDS quantification (The vendors are usually reluctant in giving details of the calculations):

https://www.cambridge.org/core/journals/microscopy-and-microanalysis/article/abs/electron-probe-microanalysis-a-review-of-the-past-present-and-future/F20CC447BC24851C41356A47464446CA

​

https://www.academia.edu/72671015/Is_Scanning_Electron_Microscopy_Energy_Dispersive_X_ray_Spectrometry_SEM_EDS_Quantitative

In EDS, low x-ray energies favor lighter elements. There are fewer electrons carrying enough energy to stimulate high energy x-rays, and more of the lower energy x-ray sources are favored. If your EDS system is set to higher energies, more of the higher energy x-rays are stimulated.

When I wanted both low and higher energies In my spectra, I would use both lower KV’s on my EDS system, and co-analyze with micro-XRF. My XRF system uses a 50KV x-ray probe. This dual technique works very well, but your software needs to handle a known standard that is close to your unknown for quant analyses, or you can create a standard out of the material you are analyzing by doing several controlled quants of your material and averaging them, then entering this average into your system as a standard. I hope this helps.

A second observation : Based upon your previous analysis, as shared (above). Your EDS analysis is indicating a need for your EDS software to be re-aligned, or recalibrated. This should allow your readouts to be centered on the appropriate voltages, rather than being slightly off centered. An Unaligned EDS system can render erroneous results, as some elements are near-overlapping in their x-ray energies. Your EDS system instructions should explain how to do the realignment process, which should likely only take a few minutes to do.

Some more considerations and questions:

Is there anything below the Si/SiO2? You indicate 300 nm of thickness. Will the beam penetrate through the entire sample. If so, you are breaking the assumption that the sample is infinitely thick with regard to the beam.

You show a significant Au peak at 5 kV. It is playing a big role in the interactions. The beam has to pass through to get to the sample then the x-rays have to traverse the gold to get detected. That is not a homogenous sample.

I would also question whether the gold is really 5 nm thick. Why do you think it is? It seems thicker. You should not trust your coater documentation unless you have verified it. Our coater has a thickness monitor, but users were not all using the same geometry. Samples set up closer to the target than prescribed will get a thicker coating than expected.

I use Casino to calculate expected intensities of each element based on expected thicknesses. I also turn it around and determine the actual thickness based on the measured intensities. I vary the layer thickness until I get a good match.

Your count rate seems quite low for the 5 kV spectrum. The strobe peak near zero is higher than anything else in your spectrum. If you have a W gun SEM, you may not be able to do much about it, but try to increase the current in the beam. That would be done with the condenser lens and may be referred to as spot size. (That is not the same as the current driving the filament.)

And you have a carbon signal in the spectrum. On our Oxford Aztec, we are able to deconvolute and ignore that peak during quantitation. We presume it is adventitious C on the surface.

You should probably have done something like that with the gold coating. But it is better to tell the system that you have a gold layer of known thickness on the surface. Again, that is present in Aztec; I don't know if that feature was in Inca.

All said, I would try simulating the spectrum using Casino (v2.54) and see what compositions and thickness give you the best match. You will need the raw, net integral for each of the peaks. That should definitely be available. I think the reason your Cu-F stoichiometry is off is because the software is trying to figure out the composition of your not-so-homogeneous Si-O-Cu-F-Au mass. You are probably lucky that the 10 kV results are close to your expections.

Regarding the statement about "alignment", the right term would be energy calibration. The marker lines in your spectra seem to be a little off-center. But I am not sure what those lines are about. They are not there for all elements and one is below zero. That makes no sense. Still a calibration check with Cu at 15 or 20 kV is probably a good idea.

Too, using your actual sample material as your control, by averaging 6 or more analyses and entering that as your soft ware as a control will do a lot to eliminate analysis faults.

Var St. Jeor , I don't recall that you addressed the issue of trying to perform quantitative analyses on a layered sample. This is not so much a question of the statistics and trying to make up your own standards. Instead, it is about the correction model chosen and the appropriateness of its assumptions to the geometry of this material.

If the software is using a normal, bulk model, it is assuming that all x-rays are coming from a common composition as they report and that all x-rays are encountering the same matrix, but they are not. The Au x-rays are not encountering any absorption on their way to the detector. The gold is on the surface. Cu and F x-rays will be absorbed by Au but not by Si or O. In short, the bulk analysis tool was never meant for this situation.

Some EDS manufacturers claim to have packages suitable for analyzing layered material, but what are the assumptions and claims? If you know the compositions of the layers, you may be able to determine the thicknesses of the layers. Or maybe you know the thicknesses with certainty but don't know all of the compositions. If thickness and composition are both in question, I think you have too many variables and not enough equations.

FWIW, I will use a Monte Carlo program like CASINO (v2.48) and simulate the measured intensities for various geometries and compositions and compare them to the measured intensities from the EDS spectrum. I play around with the inputs until I can get values that match at a couple different voltages. Then, I might trust the results.

It just occurred to me, I do not know if this was done in SEM or TEM (reflected or transmitted) mode. In reflected mode the ratio must have the ability to penetrate both layers, going in and back out again assuming the top layer is of a consistent thickness. I would think enough of the second layer ends up being scattered by the top layer to question the accuracy of lower Kv. Remember, there is scattering going on in both directions.

If this is transmitted mode (detector below the sample) then the top layer is scattering some of the probe In lower KV.

Var St. Jeor , you should safely be able to assume this was done in SEM mode. Ewelina described a combined 355 nm of thickness of Au, Cu-F, and SiO2 over an Si substrate. You would not attempt TEM on such a thick sample at 5 or 10 kV. I have heard of doing TEM in an SEM at 30 kV. More likely TEM would have been done at 200 kV or more.

Rather than just talk about how things work, I used Casino to estimate interactions and x-ray production for 5 and 10 kV at the stated thicknesses. The folks at the University of Sherbrooke have written the physical rules into their software. I have found the results to be remarkably good.

The results are attached below.

I did not find the much larger Cu:F ratio at 5 kV. I found rather consistent ratios.

They showed the Si:Au intensity ratio is roughly 1 at 5 kV. I simulated a ratio of 4-1/2. They seem to have much more than the expected 5 nm of Au.

I did not factor a C layer into the simulation, but they must have had some C contamination on the surface.

I have to ask how certain are they of these numbers - either thicknesses or compositions? To me, there are simply too many questions to try to massage the EDS results and try to conclude much of anything from them.

Warren Straszheim,

modern SEM offer (as an option) doing STEM at the usual beam energies (up to 30 kV) with a dedicated STEM detector.

For FIB/SEM instruments some vendors offer the possibility to do FIB lamella preparation and investigate the lamella - without braking the vacuum - in STEM mode (energies up to 30 kV) and (if you have a EDX detector at the right position) doing EDX in the STEM position as well.

It's amazing to get this results within hours, of course a dedicated TEM will have more options, like HRTEM, and diffraction mode.

I am aware that, “for layered materials,“ Kv does have an affect on quant analyses. This Is because of penetration issues, and x-ray exit issues. One way to overcome this layered affect is to do a quant analyses on a sectional view, one that allows equal surface area analyses. Then, one knows what the target should be. This would be a good guide in the choosing of proper KV for EDS “through” the sample’s surface. One good way to answer one’s own questions.

Var St. Jeor , I generally disagree.

Stefan Baunack and I are both pointing out that trying to use bulk analysis software to determine the composition of the layers in a structure is very risky and the results may be nothing but garbage. You are violating several assumptions implicit in the development of the software, namely that the atoms are intermixed in a way that all elements will have affect on all other elements. The average composition determines how deep the beam can penetrate into the sample. The x-rays that are generated are absorbed on the way out of the sample by material of the same average composition. And there may be secondary fluorescence of the atoms by x-rays from nearby.

Turning the layers onto their side only exchanges one type of inhomogeneity for another.

Monte Carlo programs like Casino deal with the effects on a electron by electron basis. How far have we penetrated? What direction are we going now? What are the atoms that I am encountering? You cannot package that up neatly into an analysis program without supplying some a priori information like some layer compositions and/or thicknesses.

Now, if you can make a cross section and determine the true layer thicknesses, that is a big step toward completing the picture. I will say that trying to measure a 5-nm and 50-nm layer will be tough. An ion beam helps to prepare a good surface. Mechanical polishing will not be enough. Then, you are still pushing the resolution limits.

And remember that any compositions you measure from that section will be wrong. Your sample is now not homogeneous in the Y direction. I eyeball the diameter of the interaction volume as 150 nm at 5 kV and probably 300 nm at 10 kV. You will not get a clean analysis. In such cases, I set aside any hopes of quant. I will often use a line-scan across the interface to see if the elements are at least going up and down in the right places.

Warren Straszheim

Such as you mention is very good reason for multiple scans, which are averaged together. Depending upon the characteristics of the material, I have used 6 to 12 separate scans, that fit with my material. As long as the average is within reasonable reach of a predetermined known quant, all is good. That pre-known quant must be pre-establish before a routine test can be established.

Too, with the subject material involved, it is not necessary for your “through” analysis to be correct, as long as you know that the results are reproducible, and “represent” a consistent that has been proven with more specific analyses.

I know this may sound odd, but the method works and is reproducible. A little correct knowledge can take you places as yet unknown. This is called using a “consistent fault” “in your favor.”

I will accept those comments to a point.

Results need to be consistent first. If they are not reproducible, there is no point in talking about accuracy.

From a quality control perspective, I would agree that the results do not have to be accurate to be useful. As long as they are consistent with a known good sample, the 'garbage' or 'wrong' EDS results can be helpful in validating the material.

I'm not sure that I understand your point about "6-12 separate scans". More data should always mean more precision in the results. For a given amount of total time, you probably want to split it between multiple points rather than doing a single, longer analysis at one point. I used such an approach and analysis of variance to determine the amount of variation from particle to particle and within particles for my 1981 Masters Thesis.

The quant software should give you an estimate of reproducibility based on the x-ray count. That would help in planning the acquisition. I don't see it in the original report above, but we had it in our ISIS EDS software, which was the predecessor to the INCA version. (Footnote: I notice that the total spectrum integral was much lower at 5 kV than at 10 kV. Therefore, the 5 kV results will be much more variable.)

But back to the initial question, why weren't the Cu:F ratios the same between the two voltages? I would like to presume it was the exact same area, but that is a presumption. If the area is different and the layers vary in thickness by location, then all bets are off. Measuring top-down should reveal that there are significant differences even if it cannot reveal the details.

Since the basic question was about the reported Cu:F ratio (rather than a quality test), I have been directing my comments to the requirements for good EDS results. Over my 47 years of EDS at the university, I have seen a lot of things. They have taught me to be careful about loosening the requirements. You might get away with it, but probably not.

Now, maybe I should go run a simulation on the effect of the Au layer thickness on a stoichiometric SiO2 sample. Then I should check it with actual measurements. I know that our current Aztec software allows for entry of the coating and its thickness, and that the thickness affects the reported atomic ratios.