Mapping trace elements

[wrigke]

If one wishes to measure quantitative spots in a sample, the calculations are straight-forward to determine the count time and other parameters to achieve a desired detection limit for a particular trace element.  However, I have customers that would like the distribution of their trace elements mapped.  Given that count times for mapping are on the order of milliseconds, not seconds, I'm not really sure whether this is a reasonable request.   I have seen SEM people approach this problem by mapping the same region many times through the use of "frames".  Is this the most effective approach?  Are others possible, such as using multiple spectrometers to map the same element and combining counts?  With any such approach, is the calculation of detection limit essentially the same process as for quantitative spots?

Karen

[Karsten Goemann]

Hi Karen,

I'd say the detection limit would be calculated the same way, which as you say will make it much more difficult to get sufficient detection limits for trace element mapping.

You'd also have to perform an accurate background correction, so you either have to collect at least one separate off-peak map (and use a slope factor) which basically doubles your acquisition time, or, if the major element composition of the phases you map is very uniform throughout, you might be able to determine an accurate background intensity using point acquisitions.
@John - I've never checked - is there actually a direct way in CalcImage to use externally determined "fixed" background intensities?

To achieve the detection limits required you'd have to use long counting times, high beam currents (hundreds of nA) and/or accelerating voltages. The drawback of the latter is that correction factors get very large. You might also run into problems with sample modification by the beam and/or charging. Some people use uA currents for example to map traces in steel and for that you also need to protect your BSE detector.

Combining spectrometers is certainly possible.

Multiple frames as on an SEM could reduce beam damage and charging, but if you do stage scanning to avoid WDS defocussing you'd lose additional time when the stage drives back to the beginning of the next line. Depending on your microprobe there might also be some shift between frames or even lines (stage and/or beam related), so it might only work well enough at lower magnifications. Maybe John can also chime in, but as far as I can see the only way to collect multiple frames of the same map in Probe Image is to add them as multiple acquisition samples and combine them afterwards?

Cheers,

Karsten

[wrigke]

Hi Karsten,

Thanks for your response.  While I have PFE, I don't yet have probe image.  I have Cameca's Peak Sight, so I'm speaking in more general terms about how one would approach a problem, rather than manipulation of specific software.  What I really want to do is not waste my time trying to "teach a pig to sing" as they say in the states. 

 If one can practically execute an acquisition with 500ms integration times with a 20kV beam and 250nA of current, how can one determine the detection limit for an element of interest?  For a spot, the counts and count rates for peak and background are readily accessible in the software, but that data seems quite buried in a map.  Moreover, what is a "spot" when it comes to mapping?  I assume the integration times are on a per pixel basis, but I'm not sure.  Moreover, there may be many pixels in the actual electron interaction volume so the counts arise from a much larger volume than is implied by the number of pixels.  Am I overthinking this problem? 

Karen

[John Donovan]

@John - I've never checked - is there actually a direct way in CalcImage to use externally determined "fixed" background intensities?

The best solution to get an accurate background correction for one's x-ray maps would be to apply the MAN calibration curve acquired in PFE and utilized in CalcImage.  This saves time and certainly is more precise than off-peak maps. Accuracy becomes the limiting factor in high Z materials, but both Karsten and Dave Wark have demonstrated MAN corrections for high Z materials which they presented at the Nashville M&M Castaing session.

Note that CalcImage is provided with Probe for EPMA, so if you can convert PeakSight x-ray maps into the PrbImg files you would be all set. I know that Philippe Pinard has a Python script for creating PrbImg files from JEOL x-ray maps. 

Multiple frames as on an SEM could reduce beam damage and charging, but if you do stage scanning to avoid WDS defocussing you'd lose additional time when the stage drives back to the beginning of the next line. Depending on your microprobe there might also be some shift between frames or even lines (stage and/or beam related), so it might only work well enough at lower magnifications. Maybe John can also chime in, but as far as I can see the only way to collect multiple frames of the same map in Probe Image is to add them as multiple acquisition samples and combine them afterwards?

Yes, that would be correct.  I would just add that if you are using high beam currents, your sample will probably not tolerate more than one pass!

[Karsten Goemann]

Hi Karen,

As far as I can see it shouldn't matter for the calculation of the detection limits if the interaction volume is larger or smaller than one pixel, because the measurement will still be of the interaction volume, no matter how large the actual "display size" (=pixel) is. If you do stage scanning with continuous movement in x direction, you'll be averaging even more in x direction.

The detection limit would be calculated per pixel, but you could bin pixels together during post-processing to improve precision and detection limits if you realise you have mapped at higher resolution (smaller pixel size) than required.

The dwell time in PeakSight is the acquisition time per pixel in milliseconds. To get the map counts in PeakSight you can export the maps as ASCII. Alternatively in the Results window in mapping mode there's a little mouse icon (tool tip is "Pixel Info" when you move the mouse over it). When you click on that another window opens showing you the counts of each pixel (or binned pixels) as you move the mouse over the map. For an approximate calculation of the detection limit you could use 3 x the square root of the background counts.

I guess you could also just acquire a couple of point measurements at the conditions you intend to use (e.g. as in your example 20kV, 250nA, 500ms counting on peak and background) and let the software calculate the detection limit. You should then have the same detection limit for each pixel in the map (as long as the major element composition is the same).

@ John - we've had some good results with MAN backgrounds in materials with MAN >20, for some elements  <0.02 wt% difference between off-peak and MAN background, but I'm not sure how well it would work for true trace mapping. It would certainly require a very careful setup of the curves with lots of well characterised standards in the MAN range of interest.

Cheers,

Karsten

[Les Moore]

Quant trace element analysis in maps.  A very difficult ask.
1. If you do a background map and do a pixel by pixel subtraction, you are going to seriously degrade your S/N and thus lower your sensitivity.
2. If you use an external estimate you won't know if there is a background variation in the trace areas.   
3. The counts are so low in the maps that you can be using a calibration curve 3-4 orders of magnitude away from the std.

So,
1. Find out if there is a background variation and if not, get a global estimate.  Subtracting one number will not change the S/N. 
2. If possible, get a near composition standard with similar MAC to what you are analysing.

This should just leave the usual ZAF issues and other microanalytical pitfalls.   

[Probeman]

Quant trace element analysis in maps.  A very difficult ask.
1. If you do a background map and do a pixel by pixel subtraction, you are going to seriously degrade your S/N and thus lower your sensitivity.

Hi Les,
You are absolutely correct. But you might be interested to learn that there is a very cool way around this for samples where one can obtain a suitable "blank" standard (relatively pure elements, oxides, simple silicates, etc.):

http://probesoftware.com/smf/index.php?topic=307.msg1551#msg1551

2. If you use an external estimate you won't know if there is a background variation in the trace areas.   

That is why the MAN mapping method linked above works so well down to 100 to 200 PPM accuracy in relatively low to moderate Z materials even *without* a blank standard*. But for trace mapping accuracy in complex matrices (e.g., monazite) below 100 PPM, the off-peak mapping method is required. Fortunately this is very easy to do now:

http://probesoftware.com/smf/index.php?topic=107.msg388#msg388