Examples of quantification for a EDSSEMSpectrum.

[CompRhys]

Describe the functionality you would like to see.

I am working with a EDSSEMSpectrum parse from the ELID output of a Phenom benchtop SEM. I would like to quantify the composition but the quantification methods are only supposed implemented for the EDSTEMSpectrum spectrum. I am not sure of any particular reason why we couldn't apply the same method to the SEM spectra.

Describe the context

We are working with the Phenom API but it only outputs data in their ELID format. Ideally we would like to be able to analyze the data ourselves on main compute infrastructure rather than having to do it on the windows machine paired to the microscope. Hyperspy seems like our best bet for doing this but it's not clear that the functionality we need is available for the SEM.

[ericpre]

The quantification implemented in for the EDS_TEM signal assumed a thin foil geometry, which is most of the time not the case foe EDS measurements in the SEM.

There are other methods but these are not implemented in exspy. Is it something that you would be interested in contributing?

[CSSFrancis]

One other thing that might be nice with SEM EDS support is increased support for things like WDS as well. I don't think it would be terribly difficult to add that support but it requires someone with the right knowledge/motivation to add that feature.

[CSSFrancis]

I transferred over some former discussion on this:
#118
#117

General consensus was that for thick SEM EDS it would require standards in order to do any quantification. In general, standardless EDS is very very inaccurate and I think the comment that we don't want to be contributing to bad EDS results is probably a good one.

[CompRhys]

The quantification implemented in for the EDS_TEM signal assumed a thin foil geometry, which is most of the time not the case foe EDS measurements in the SEM.

There are other methods but these are not implemented in exspy. Is it something that you would be interested in contributing?

I read into the linked issues and I think that I am not well placed to help in the development of standardless quantification. Whilst we would ideally avoid having to do all our data analysis in proprietary formats I think that I will be unable to get any development time to learn and then implement an open alternative.

Thanks for your efforts in maintaining this package.

[jeinsle]

@CompRhys My group battles this all the time. At the moment the workaround we use is saving a phase spectra as a MSA file, and then converting to SPX. At that point we use Bruker’s Esprit to quantify.

Recently, I had a student do this on an old data set where I had personally treated the SEM data as TEM and faked it using Cliff-Lorimer. (This technically is the Castig(sp?) method and this was what motivated @dnjohnstone to originally open the older thread). For a comparison of standardless approaches, Esprit and this ‘fake’ were close enough that we know what mineral we are looking at (large composition ranges which are valid). The question, comes down to how absolutely quantitative do you want to be? If you want to be accurate out to more than 2 decimal places than standards are absolutely required. If you need to be within about 0.5% then you are probably good enough. The other part here to think about is that in an SEM running mapping you are likely not to have the count totals needed to really push beyond about 0.5% accuracy anyway. (This even includes advantages of quantifying on a sum spectra for a phase which you could see when you calculate the standard error for each of the peaks you want to quantify)

Feel free to email me directly if you want to discuss these workarounds. I would be really curious to see som the Phenom data as we are looking at trying to acquire one for student projects.

[CompRhys]

keep things on public forums is good in my opinion!

We're interested in tracking potential composition shifts between our inputs and outputs and so ideally we need to have as quantitative results as possible (given the limitations of EDS).

I would say that the Desktop Phenom that we are using is a highly capable and fast machine with good programmatic interface to allow us to drive it with python scripts. The issue I was attempting to find a solution for is that at our startup the majority of system runs on linux whereas the programming interface for the SEM only is available on windows. As we would like to decouple our analysis and data collection I wanted to find alternatives. It seems that for our progress in terms of the start-up needs we simply need to run our analysis using the phenom provided software. My only prior SEM experience was a 25+ year old machine in the first year material science labs during undergrad nearly 10 years ago so I maybe not the best to ask. We make almost negligible use of the other thermofisher software for phenom beyond the python programming interface, these may be more important for your workflows especially for things like large scale mapping or particle/precipitate detection.