Are you working with the software package that collected the data, or are you reinventing the wheel, so to speak? There are many issues related to EDS analysis. Are you deconvoluting according to the known peak shapes for the elements with the expected peak widths? If you are allowing peak widths and relative peak intensities to vary within a set, you are not doing it correctly. A student might be able to recreate good correction algorithms on their own - many degrees were probably granted based on the work that has led to the current systems - but I think your research work would be better directed elsewhere.

If you are using something like DTSA, DeskTop Spectrum Analyzer, it should provide you many of the basic tools with constraints. I presume it contains peak profiles for the elements.

I think it would be better to find out what went wrong and address that problem. To do that, we need more information to say something relevant. What is your sample? What EDS system were you running? What conditions were you using?

FWIW, I have run into trouble with our system when the accelerating voltage and working distance is not provided to the EDS software. We ran into this with systems from Oxford Instruments, but the problem is not theirs, per se. If we restart the SEM software, we must also restart the EDS software to restore communication. Otherwise, the EDS software reports AV=0.2 keV (to keep from dividing by zero) and 0x magnification. It cannot identify elements or otherwise process the data with that bad voltage input. Now, you should be able to determine the actual voltage from where the background tails off. I don't know how to correct the value stored with the Oxford spectrum. (Maybe you can in DTSA.)

Finally, I will say that I have been performing a lot of good EDS over the last 40-some years. EDS is discounted as only partial reliable compared to WDS and EPMA. That need not be the case. IF the same care was given to performing EDS as is given to performing EPMA, the results could be comparably good. Analysis should be limited to analysis of flat, level, homogeneous samples. Attention should be given to the statistics. However, many users work without much care, i.e., carelessly, then are surprised that the results are not very good.

Warren Straszheim Thank you for the detailed explanation. The device used is from Thermo Fisher Scientific, which performed rectangular mapping. However, the results section only includes distribution, as shown in the attached document. I separately utilized desktop peak analysis software (Fityk) to determine quantitative peak values. Could you provide a method or equation to calculate atomic concentration from this raw data?

Calculating from the EDX profile manually is challenging and prone to errors. To get the most accurate and efficient results, I recommend using Energy Dispersive X-ray Spectroscopy (EDX) from the Scanning Electron Microscope (SEM) while performing the mapping or scanning. The SEM instrument will automatically quantify the results, making the process simpler and more straightforward.

Seems You made an EDS spectral map (at every pixel a full spectrum is recorded). I assume that the "Extracted spectrum" is the sum of all 512×340 recorded spectra.

If the Thermo Fisher software has the capability of doing quantification of this spectrum it should be worth a try.

That additional information helps a lot. I am still curious which particular system it is. That might help. And I still don't know what your sample was or how it was prepared. That will be important.

You should be able to quantify the extracted spectrum, although I don't think that you should.

Any spectrum should be quantifiable. If this is not, then I wonder what happened. I see that the voltage was 17 kV. That should not be3a problem then. Maybe Thermo-Fisher does not like quantifying such sum spectra. It is rather dangerous.

Your sample does not look like much of a candidate for a map. It looks like it could be a thin layer on a substrate. If so, the substrate could contribute much to the map. I don't see much variation in maps. The area is mostly homogeneous.

(Side note: I really don't like colors for evaluating maps. The dark blues and red make it hard to tell the difference between full brightness (a dark color) and black. I use white for all maps until I am ready to emphasize the difference in distribution.)

(Side note 2: I generally do my maps at 256 pixels across to get better data per pixel. At 512 pixels, each pixel would only have one fourth the data. I collected 12M counts to spread across 256x240 pixels for about 200 counts per pixel. You seem to have 1.5M to spread across 512x340 pixels for 8.6 counts per pixel.)

Looking at your spectrum I see Si, O, Au (unlabeled), C, and maybe Al and Cu. What were you expecting? The Au coating could account for the texture in your image. I use 2-5 nm of iridium for our samples. It has less texture and less thickness, so less contribution to the spectrum.

Is the Si and O from the substrate? If it is, you should have used a lower voltage to reduce its involvement. You could have used as low as 5 kV for the peaks I see in your spectrum. I would also use Si, not oxidized Si for a substrate. I would also try for a thicker deposition.

With all that, I would probably not try hard to quantify this spectrum.

Again, what were you expecting of this sample?

Shayan Bayki , I looked briefly at Fityk. While it can probably handle the mathematics of peak fitting, I don't think it will handle the physics of EDS.

There is much that goes into peak width and background modeling of an EDS spectrum. The ratios between peaks in a set (e.g., Si Ka, Kb) are fixed and should not be allowed to vary. The EDS quant routines calculate the depth of penetration which depends on the voltage and the composition. They know of the expected yield of x-rays under the given conditions. They calculate the absorption of x-rays of one element by other elements. They account for secondary fluorescence and detector characteristics.

So, it is much more than just peak fitting. A 10K counts of C peak probably represents much more than 10K counts of Si peak. You might look into DTSA.

Ramesh Sivasamy, Thank you for your suggestion, but I would like to understand the numerical measurement behind it. Is it possible to calculate the concentration from the intensity plot? I would appreciate it if you could provide it.

Warren Straszheim Sir, the sample is carbon tape with a thin PDMS coating (5 microns). The Si distributon is from the PDMS layer, so a higher concentration is expected. Based on your explanation, I understand that quantifying the spectrum specifically for weight concentration by fitting raw data is not possible. However, could you explain (or provide any literature) on the numerical method that the software uses?

I agree with Warren Straszheim that is doesn't make sense to re-invent the wheel.

A good review about the underlying models for EPMA (i.e. WDS, but for EDS the physics is the same) can be found here:

Article Reprint of: Electron probe microanalysis: A review of recent...

If you want to use the DTSA-II program:

For quantification the detector type is taken into account. If you have the spectral data in the vendors format or .msa format, this information may be included, otherwise the results may be limited.

What was your aim?

Do you want to check the stoichiometry of the PDMS?

With a lot of C, H, O quantification is doubtful.

For 17 keV electrons the penetration depth of the electrons in PDMS may be calculated to be ca. 10 µm, i.e. you'll excite the carbon tape.

Thank you so much, Stefan Baunack, for providing the paper. It has been extremely helpful in clarifying my doubts. I treated the samples with different annealing temperatures to measure the Si weight concentration. I already have the data in .msa format and will use DTSA-II for the analysis.

I suggest you may want to go back and recollect the data under different conditions for a better chance of success. You can push on with the data such as you presented it, but you will probably continue to have questions about the results.

Much will depend on the detector and window characteristics. DTSA may be able to handle that, but I would expect your analyzer software to do it better. I was impressed by the first Tracor-Northern TN-2000 EDS system I used. T-N became Noran which was bought out by Thermo-Fisher. They (and others) published several seminal papers back in the 1970s and 1980s (and beyond) that described their quantitative process. I see Reed, Scott, Goldstein, Newbury, and others in the article's bibliography.

Good analysis of C will be difficult. I would say that is more troublesome at higher voltage. You will find the correction factors depart further from 1.00, so you should be more concerned about how good the correction formula and data (e.g., mass absorption factors) are.

Since you are dealing with C, the roughness of the surface may become an issue. Unfortunately, that rough surface is rather inherent in PDMS. I have always found it to shrink and wrinkle under the SEM beam. You may have put down 5 um, but I think you would find the actual thickness much thinner. I found that I had to keep the voltage around 2 keV to avoid changes, but that is too low to do EDS.

And if the PDMS is changing under the SEM beam, will the beam dosage not matter more than the annealing conditions?? Even the vacuum of the SEM may be doing more to the PDMS than your annealing.

I would consider applying the PDMS to a different substrate besides C tape. That could give you an indication that you are penetrating through the PDMS. I tried a quick CASINO (v2.48) simulation of PDMS on Fe. For a 5-um layer at 17kV, the Fe-L line was 1% of the Si-K line. For a 3-um layer, it was 16%.

At 10kV, Fe-L was 1% of the Si-K line for a thickness of just over 2.5 um.

At 5kV, Fe-L was 1% of the Si-K line for a thickness of about 850 nm.

Dear Warren Straszheim and Stefan Baunack I have attached the result from another device (EDAX Co.). The peculiar part is that while the peak intensities for Fe, V, and Ti are recognizable, the atomic percentage results indicate that 53% belongs to C, despite its very small intensity. Are these results reliable? The sample is mild steel (2 mm thickness) machined by the EDM process.

You should discuss your EDS work step-by-step with an experienced operator.

1) peak identification (is done), but a large peak at ca. 2 keV is not labelled. Why? An element could be P, or was the peak excluded by the software because identified as an artifact (escape peak, sum peak). In this case modifying the dead time or reducing the beam current could be helpful.

2) If you have achieved a reasonable peak identification and background correction (for all peaks!, is there a faint blue line for this fit?) you can calculate the concentrations.

Here you could/should/must exclude elements not present in the sample. The carbon signal could stem from carbon coating (not needed for steel), or carbon tape to fix the sample. What will change if you exclude C from the calculation of concentration?

Al might be present or comes from the stub.

IIRC the EDAX software can give concentrations with a confidence level (error), i.e. is the vanadium signal considered to be real?

Additional information: https://myscope.training/EDS_Welcome

I agree with Stefan Baunack, you should enlist the help of an experienced operator for these questions. I think we are serving that purpose on ResearchGate, but you should have a knowledgeable operator at your site. If not, you will find many results that don't make sense but which do have an explanation. Our service lab is setup with staff dedicated to the support of the instruments. Grad students do not have time to learn all the details.

I raise the question again about what the samples were coated with. I expect the peak around 2.1 keV is from a gold coating. It is very bad form not to label that peak.

The graph reminds me of why we did not choose EDAX for our system. Why are their major divisions at 1.3 keV? At least they are not using something even more inane like 1.34 keV. They really should use values like 2, 1, 0.5, 0.2 or 0.1 keV for major divisions. Minor divisions would also help.

The C level also reminds me why we chose Oxford. Even then, the C in this case is probably from contamination on the surface. It should probably be ignored. In the case of your PDMS, it must be dealt with.

I have attached some results for Fe and Cr standards on our system collected at various voltages. Neither have real C in the standard, but they were coated with 10+ nm of carbon. The C reports as ~5 wt% and 18 at%. But my C peaks look bigger than the C in the EDAX spectrum.

It would be nice to examine the correction factors that translated the intensity to mass fraction. While I cannot see the Phi-Rho-Z or ZAF factors directly, it appears that the composite factor must be over 2 in my case. It must be even higher for the EDAX case to come up with 21 wt%. As those factors increase from 1, they should be viewed with suspicion.

BTW, I would seriously doubt the V value. The V-Ka line falls right on top of the Ti-Kb line. It would be nice to see the fit both with and without V in the mix.