The Limits of EPMA Accuracy

[Probeman]

What exactly limits the accuracy of modern EPMA?

After some discussion with several colleagues and much consideration I have come to a few ideas on what is going on. I think there are several problems with improving accuracy in EPMA, partly consisting of unreliable naturally sourced standard materials, but also outmoded practices which no longer reflect improvements in modern hardware and software. I'd like to discuss these ideas in this topic starting with the use of what some of us refer to as "matrix" matched standards.

We usually speak of using "matrix" matched standards in order to minimize the magnitude of our matrix corrections when extrapolating from our standard to our unknown.  But today's modern matrix corrections are quite accurate as shown in many recent evaluations, even with relatively low energy emission lines, at relatively high beam energies which produce matrix corrections on the order of 50% to 100% or more. Here are just a few recent examples:

https://smf.probesoftware.com/index.php?topic=1823.0

Ritchie, N. W., D. E. Newbury, and S. Leigh. "Breaking the 1% accuracy barrier in EPMA." Microscopy and Microanalysis 18.S2 (2012): 1006-1007.

So why do so many EPMA labs still think that so called naturally sourced "matrix" matched standards are necessary? 

It's true that 30 years ago, matrix corrections were less accurate and at that time there may have been some justification for choosing a standard with a matrix somewhat similar to our unknown samples. Yes, we would not want to utilize a Si metal standard for analyzing Si in silicates, as there is a considerable peak shape/shift for Si Ka between these materials.  And yes, there are a few "black holes" in the periodic table that may require a roughly similar matrix, e.g., Si Ka in Hf due to disagreement in mass absorption coefficients.  But geological silicates and oxides are pretty well handled by modern matrix corrections.

So why not just use synthetic MgO as a primary standard for Mg Ka for ones silicate analyses?  If we have a pure synthetic MgO, the composition should be well known, right?  And it is readily available in kilogram quantities!

In fact, we suspect that our matrix corrections are not the main accuracy issue today.  Instead, part of the problem is that in choosing these "matrix" matched standards, we have traditionally opted for natural materials which we now know are heterogeneous, inclusion filled and also of limited supply:

https://smf.probesoftware.com/index.php?topic=1415.0

As Gene Jarosewich once said of the Smithsonian geological standards: one should always take the average of 10 to 15 separate grains in order to get a standard measurement that accurately reflects the wet chemistry average.  But let's be honest, does *anyone* actually do this in their probe labs?

To avoid these standard accuracy issues, we should instead be moving to high purity, synthetic minerals that are carefully characterized for purity, homogeneity, stoichiometry and are readily available in abundance and *globally* distributed. As is currently being done by Will Nachlas and the FIGMAS group with MAS support:

https://smf.probesoftware.com/index.php?topic=1415.msg10368;topicseen#msg10368

So why do some labs still insist on using these so called "matrix" matched standards even though we know there are better, high purity synthetic materials available today?  See below for some possible explanations. But as Penny Wieser has demonstrated in her "Barometers Behaving Badly" paper, the problem extends beyond the use of heterogeneous and inclusion filled natural "standards":


Wieser, Penny E., et al. "Barometers behaving badly I: assessing the influence of analytical and experimental uncertainty on clinopyroxene thermobarometry calculations at crustal conditions." Journal of Petrology 64.2 (2023): egac126.

There is something else that is not right about our EPMA WDS measurements!

We now suspect that in addition to the accuracy concerns of natural standards, the underlying problems with EPMA accuracy today primarily have to do with differences in the *count rate*, as measured on the instrument, between the standard and unknown. Therefore we propose that instead of calling these "matrix" matched standards, we call them "count rate" matched standards.  Why do we think *count rate* matched is the more appropriate term?

1. Dead Time

Most WDS instruments are not well calibrated for dead time corrections, especially at count rates now commonly seen with modern instruments using large area crystals. This matters when using high purity synthetic standard materials because if you go from an unknown with a count rate that is different than the primary standard, the quantitative accuracy will depend on the accuracy of the dead time correction, even at moderate beam currents. Because when measuring major elements on modern instruments with large area Bragg crystals, one can easily obtain dead time corrections of 10% to 30% or more, even at moderate beam currents.

Without an accurate (logarithmic) dead time calibration as described here:

https://smf.probesoftware.com/index.php?topic=1466.msg11102#msg11102

Donovan, John J., et al. "a new method for dead time calibration and a new expression for correction of WDS Intensities for microanalysis." Microscopy and Microanalysis 29.3 (2023): 1096-1110.

one will be forced to utilize these "count rate" matched natural standards with their documented heterogeneity, Even Si Ka on a normal TAP Bragg crystal can yield significant count rates at moderate beam currents, due to its low sin theta and hence larger subtended angle. 

And what about quantitative mapping at high beam currents? Having a linear PHA response and an accurate dead time correction becomes essential for quantitative mapping accuracy. In fact, with properly adjusted PHAs and an accurate dead time correction, we can now perform quantitative mapping at high beam currents using WDS:

https://smf.probesoftware.com/index.php?topic=1466.msg11629#msg11629

Donovan, John J., et al. "Quantitative WDS compositional mapping using the electron microprobe." American Mineralogist: Journal of Earth and Planetary Materials 106.11 (2021): 1717-1735.

https://epmalab.uoregon.edu/pdfs/Donovan_2021_Amer_Min_2021-7739.pdf

2. PHA Tuning:

We also believe that PHA tuning is being performed improperly by some EPMA labs.  First, some EPMA labs are tuning their PHAs on their unknown sample rather than the primary standard. Why does this matter? Because PHAs are sensitive to count rate.

Therefore tuning ones PHA by centering the PHA peak in the PHA distribution (e.g., 4v on a JEOL instrument) with the baseline level set below the PHA peak and setting the PHA window above the PHA peak (and using differential PHA mode) will not yield quantitative results when the count rates between the standard and unknown are significantly different, as will often be the case with modern EPMA instruments using large area crystals with correspondingly higher count rates.

Specifically, the problem with this PHA tuning method is, if we go to another material, for instance ones primary standard, containing a different concentration of the element and therefore likely a higher count rate, the peak will shift to the left (possibly intersecting the baseline level) due to more pulse height depression.

We believe this is why some EPMA labs have found it necessary to use a standard that is "matrix" matched to their unknown. Because when they look for a standard that is "matrix" matched and therefore usually of a similar composition, they are really "count rate" matching to their unknown. In other words, due to improper adjustment of their PHA settings, they obtain inaccurate results when extrapolating, from well characterized and homogeneous pure synthetic MgO or Al2O3 or Fe2O3 materials, to their natural unknowns usually with lower count rates.  Essentially they are "count rate" matching to avoid pulse height depression effects, rather than actually "matrix" matching to avoid large matrix corrections.

In addition, because these natural materials tend to be heterogeneous and of uncertain accuracy, as described above, that introduces further inaccuracy.

How can we fix this PHA tuning problem?  Well the first clue to this was when SEM Geologist suggested that we should not be running in "Differential" mode generally, and instead run the PHAs in "Integral" mode, that is, without a PHA window level filter. This is particularly important for lower energy emission lines on TAP and LDE Bragg crystals:

https://smf.probesoftware.com/index.php?topic=1466.msg11549;topicseen#msg11549

Yes, differential mode can help with some higher Bragg order interferences, but it doesn't help at all with *same* Bragg order interferences, and only partially with higher Bragg order interferences. In fact there are only a few rare spectral interference situations I can think of where differential mode might help, such as Na Ka 2nd Bragg order interfering when measuring trace oxygen, because it's difficult to find a standard for the interference correction that contains sodium but no oxygen.

Otherwise it makes much more sense to tune your PHAs to obtain a linear response in count rate over a large range of count rate, and then correct for any spectral interferences using the quantitative interference correction in software:

Donovan, John J., Donald A. Snyder, and Mark L. Rivers. "An improved interference correction for trace element analysis." Proceedings of the Annual Meeting-Electron Microscopy Society of America. San Francisco Press, 1992.

That means adjusting your PHAs on a material with the highest count rate you will be observing, which is usually the primary standard for that element, at the highest beam current you will be utilizing. Then adjust the PHA peak position using the gain (Cameca) or bias (JEOL), until the PHA peak is *completely* above the baselines level.  This is critically important as we will see, for obtaining a linear response over a range of count rates.

And utilize PHA Integral mode! That way, when you move to a material with a lower count rate, the PHA peak will shift to the right, and in Integral mode all the photons will still be counted, even if the PHA peak is graphically "cut off" to the right:

https://smf.probesoftware.com/index.php?topic=1466.msg11450;topicseen#msg11450

At "normal" beam currents this graphical "cutting off" of the PHA peak on the right will usually only occur with very low energy emission lines such as O Ka or N Ka, etc. Though when attempting to acquire "constant" k-ratios from low beam currents to 200 nA or more for the dead time calibration (see above), one can see this "cutoff" effect even for Si Ka as shown in the link above.

So even though it appears that the PHA peak is "cut off" in the PHA plots, all photons to the right will still be counted in Integral mode. So it is critical for all emission lines to have their PHA peak adjusted so the PHA peak is *completely* above the baseline level at the highest count rate that is expected to occur, and to always run ones analyses in integral PHA mode!

Summary:

First we need to have our WDS spectrometers carefully aligned mechanically. See here for the Bragg Order k-ratio tests to check spectrometer alignment:

https://smf.probesoftware.com/index.php?topic=1739.0

Your instrument engineer may need to perform these alignments, but check using the Bragg Order k-ratio tests linked above:

Then once your dead time constants are carefully calibrated using the Constant K-ratio method, and your PHAs are properly tuned and using *integral* PHA mode as described above, you can expect excellent accuracy even when extrapolating from synthetic MgO and Al2O3 to MgAl2O4 at 20 keV and 30 nA (8 um):

TYPE:     ANAL    ANAL    ANAL
BGDS:      EXP     LIN     LIN
TIME:    60.00   60.00   60.00
BEAM:    30.06   30.06   30.06

ELEM:        O      Mg      Al   SUM  
   126  44.826  17.098  37.646  99.570
   127  44.779  17.113  37.549  99.441
   128  44.894  17.080  38.119 100.093
   129  44.740  17.109  37.652  99.501
   130  44.798  17.143  37.533  99.473

AVER:   44.807  17.108  37.700  99.616
SDEV:     .057    .023    .241    .271
SERR:     .026    .010    .108
%RSD:      .13     .13     .64

PUBL:   44.985  17.084  37.931 100.000
%VAR:     -.40     .14    -.61
DIFF:    -.178    .025   -.231
STDS:     3012    3012    3013

STKF:    .2100   .4274   .4047
STCT:   256.16 1936.93  925.50

UNKF:    .2176   .1172   .2239
UNCT:   265.44  531.14  511.96
UNBG:     1.55    2.02     .79

ZCOR:   2.0592  1.4597  1.6842
KRAW:   1.0362   .2742   .5532
PKBG:   172.61  264.22  648.77

And here's a more recent example extrapolating from synthetic MgO and SiO2 to synthetic Mg2SiO4 also at 20 keV and 30 nA (10 um):

St  273 Set   2 Mg2SiO4 (magnesium olivine) synthetic, Results in Elemental Weight Percents
 
ELEM:        O      Si      Mg
TYPE:     ANAL    ANAL    ANAL
BGDS:      LIN     LIN     LIN
TIME:    60.00   60.00   60.00
BEAM:    29.90   29.90   29.90

ELEM:        O      Si      Mg   SUM  
   136  45.716  19.786  34.595 100.097
   137  45.822  19.800  34.557 100.180
   138  45.798  19.752  34.578 100.128
   139  45.776  19.769  34.551 100.095
   140  45.945  19.766  34.499 100.210

AVER:   45.811  19.775  34.556 100.142
SDEV:     .084    .019    .036    .051
SERR:     .038    .008    .016
%RSD:      .18     .10     .10

PUBL:   45.486  19.960  34.554 100.000
%VAR:      .72    -.93     .01
DIFF:     .325   -.185    .002
STDS:       12      14      12

STKF:    .2084   .3904   .4269
STCT:   254.13 2242.56  815.27

UNKF:    .2158   .1217   .2295
UNCT:   263.14  699.14  438.19
UNBG:     2.54    3.23     .64

ZCOR:   2.1225  1.6248  1.5060
KRAW:   1.0355   .3118   .5375
PKBG:   104.57  217.60  688.52

Note the %VAR and ZCOR values. These are significant matrix corrections, yet still better than 1% relative accuracy at 20 keV!  The problem is not our matrix corrections, the problem is our natural "count rate" matched standard materials along with our dead time calibrations and PHA tuning.

If you want to focus the beam more, use the TDI correction. It works!  With modern FEG instruments with highly focussed beams, the TDI correction integrated with the quantitative matrix correction is more important than ever:

https://smf.probesoftware.com/index.php?topic=11.msg5910#msg5910

These are simple and easy tests to perform. What do you think of all this?  Do you care about EPMA accuracy? 

Run some tests as seen in the FIGMAS challenge topic linked below and let us know what quantitative accuracy you obtain with a properly calibrated instrument using pure oxide standards:

https://smf.probesoftware.com/index.php?topic=1823.0

[Probeman]

Andrew Ducharme found this review article which I had looked at when it came out, and the section on PHA tuning (which I did not note at the time!) illustrates the issues with counting electronics linearity which we discuss in the above post (yeah, it's long but we think worth reading!).

Here's the first screen shot from the pdf:



and the second:



Llovet, Xavier, et al. "Reprint of: Electron probe microanalysis: A review of recent developments and applications in materials science and engineering." Progress in Materials Science 120 (2021): 100818.

I would comment that yes, if we do not tune our PHA settings using the highest count rates that we expect to measure, then by going to a higher count rate on our pure element or pure oxide standard, we will experience pulse height depression and therefore lose counts as the PHA peak shifts to the left as the PHA peak is filtered by the baseline level.

However, if I read correctly, the authors are not correct that the use of differential mode with a PHA window level will make this situation worse because it is only at lower count rates that the PHA peak will shift to the right!

The use of differential mode will only introduce non-linearity in the counting electronics when tuning the PHA on a on a high count rate standard, and then moving to a low count rate unknown, where the PHA peak will shift to the right.

As stated in the previous post, instead, the best procedure is to NOT use differential mode (that is use Integral mode), and to tune your PHA peak at the highest count rate you expect to measure, and make sure that the PHA peak is fully and completely above the baseline level at that high count count rate.

The suggestion to utilize lower beam currents on the high count rate standard and higher beam currents on the lower unknown is not wrong, but then relies on the linearity of ones picoammeter, which can be problematic:

https://smf.probesoftware.com/index.php?topic=1466.msg11324#msg11324

The bottom line is that one can obtain a fully linear response from ones counting electronics by tuning the PHA at the highest intensity one expects to measure, and leaving the PHA in Integral mode.  As one goes to lower count rates, say on ones unknown, yes, the PHA peak will shift to the right, but in Integral mode all the photons will be counted even though they appear to be "cut off" graphically!

Here are some PHA plots from our quantitative runs where we obtained ~0.5% accuracy starting with Mg Ka (note Mg Ka in MgO = ~60 kcps and in MgAl2O4 = ~16 kcps. So just a bit of a count rate difference!



And Al Ka:



And O Ka:



And here is the quant at analysis at 25 keV:

St 3100 Set   3 MgAl2O4 FIGMAS, Results in Elemental Weight Percents
 
ELEM:        O      Mg      Al
TYPE:     ANAL    ANAL    ANAL
BGDS:      EXP     LIN     LIN
TIME:    60.00   60.00   60.00
BEAM:    30.19   30.19   30.19

ELEM:        O      Mg      Al   SUM  
   141  45.071  17.156  37.426  99.653
   142  45.065  17.161  37.576  99.802
   143  45.192  17.182  37.607  99.981
   144  45.034  17.207  37.353  99.594
   145  45.213  17.198  37.403  99.815

AVER:   45.115  17.181  37.473  99.769
SDEV:     .082    .022    .112    .152
SERR:     .036    .010    .050
%RSD:      .18     .13     .30

PUBL:   44.985  17.084  37.931 100.000
%VAR:      .29     .57   -1.21
DIFF:     .130    .097   -.458
STDS:     3012    3012    3013

STKF:    .1793   .3807   .3710
STCT:   216.43 2019.18 1028.68

UNKF:    .1851   .1027   .1882
UNCT:   223.40  544.84  521.76
UNBG:     1.44    1.98     .77

ZCOR:   2.4376  1.6725  1.9916
KRAW:   1.0322   .2698   .5072
PKBG:   156.40  275.78  678.72

Oxygen and magnesium accuracy is within ~0.5% relative!  Aluminum is a bit worse, but again we suspect an alignment problem on that spectrometer because when running Al Ka on spectrometer 1 we get ~0.5% accuracy as demonstrated here:

https://smf.probesoftware.com/index.php?topic=1823.msg13917#msg13917

[Probeman]

The bottom line is that one can obtain a fully linear response from ones counting electronics by tuning the PHA at the highest intensity one expects to measure, and leaving the PHA in Integral mode.  As one goes to lower count rates, say on ones unknown, yes, the PHA peak will shift to the right, but in Integral mode all the photons will be counted even though they appear to be "cut off" graphically!

This last weekend Andrew Ducharme found in the Cameca Reference guide that in Integral PHA mode, the pulse processing electronics will include all photons up to 10v even though the software only displays photons up to 5 v.

Does anyone know what voltage the JEOL PHA electronics will count up to when in Integral mode?

[sem-geologist]

This last weekend Andrew Ducharme found in the Cameca Reference guide that in Integral PHA mode, the pulse processing electronics will include all photons up to 10v even though the software only displays photons up to 5 v.

Just to add to this: there is nothing larger than 10V at measurement as any pulse which would be larger than 10V is clipped-down to 10V before being digitized. Counters on Cameca EPMA's are also capable to sense some cosmic radiation (very energetic events, recognizable as very steep pulses clipped at max already at pulse shapping stage) - they make that random background noise (2-10cps) in WDS ratemeters when beam is off.

[Probeman]

Just to add to this: there is nothing larger than 10V at measurement as any pulse which would be larger than 10V is clipped-down to 10V before being digitized.

So you are saying, that photons (or other particles) that produce pulses with a voltage greater than 10v would still be counted?  I wonder why Cameca bothered to mention the 10v limit in their reference manual?

Would that also be true for JEOL counting electronics?  I'm guessing yes, because John Armstrong once mentioned seeing cosmic ray counts when the beam is off on his JEOL instrument.

[sem-geologist]

To understand what "clipping" is and to not write lengthy explanations (which I tend to overdo) I just forward to wikipedia:

https://en.wikipedia.org/wiki/Clipping_(signal_processing)

So in older Cameca systems (which I believe those reference manuals could be referencing to) were clipping signals between 0.56V and 10V as ADC was 10V capable. If you have new gen WDS electronics those actually clip signal to 0.56V and 5V range as more modern ADC is 5V.

Practically any analog signals when being fed to modern ADC's needs to be protectively clipped to not over-exceed ADC input voltage limits. I am quite convinced Jeol probe needs to do signal clipping too, even without taking a glimpse to their actual design. They probably clip lower bound differently, where Cameca use diode in series in signal path, and Jeol probably just drains negative part of signal to GND.

Thus yes - be it raw photon have 10keV or 100keV or 1MeV or 1GeV - if it had produced any Townsend avalanche in GFCP, and counting electronics is not in enforced dead-time time gap, it will be registered at clipped upper value of ADC input range in PHA. To clarify even further: all above listed high energy photons will be registered as same 10V height pulse on 10V ADC, or as 5V pulse on 5V ADC. (If bias and gain is set to project Photon energies from 0keV to 10keV into 0V to 10V height pulses).

[John Donovan]

Thus yes - be it raw photon have 10keV or 100keV or 1MeV or 1GeV - if it had produced any Townsend avalanche in GFCP, and counting electronics is not in enforced dead-time time gap, it will be registered at clipped upper value of ADC input range in PHA. To clarify even further: all above listed high energy photons will be registered as same 10V height pulse on 10V ADC, or as 5V pulse on 5V ADC. (If bias and gain is set to project Photon energies from 0keV to 10keV into 0V to 10V height pulses).

OK, thanks.

[aducharme]

Being limited to 5 V on modern electronics seems quite bad for WDS? Differential mode windows of 5-5.5 V (stretching b/w ~0.5-6 V) aren't uncommon, even if I don't think you should use them. Clipping at 5 V could then be quite negative.

On older electronics, how many counts should we expect at 10 V? Would it be worth using "differential mode" with a normal baseline and a high window edge at 9.9 or 9.95 V? I doubt it, since integral mode measurements have not been noticeably worsened by them already. We could measure the effect though by measuring counts in integral and this proposed gigantic differential mode.

[sem-geologist]

Being limited to 5 V on modern electronics seems quite bad for WDS? Differential mode windows of 5-5.5 V (stretching b/w ~0.5-6 V) aren't uncommon, even if I don't think you should use them. Clipping at 5 V could then be quite negative.

The big elephant in the room - component design during last decades had changed, and it changed IMHO to the right direction actually. You had to have plenty of room for Analog signals for them to be "immune" to digital noise around. While chips from 70ies, 80ies and early 90ies were slow and their digital transitions were slow - their emission of noise into analog signal also was moderate. 0-10V dynamic range or even -10 to +10V range was very common then for analog lines, In those times it was a brute force attempt to counteract a completely shitty design of PCBs with star-ground nonsense or other woodoo engineering practices which had then not bite back severely, but would had coupled some perceivable noise to analog lines if they would have lower dynamic range (i.e. 0-5V). Fortunately modern PCB design went in the right direction and started ditching these outdated terrible designs, starting to use full ground plains multi-layer PCB's, high density designs, surface mounted parts, Finally whole PCB can be designed with everything having controlled impedance and controlled noise level. Digital chips got much faster where state transition if designed poorly can emit and couple into everything around, including analog traces. Fortunately majority of electronic producers were forced by EMC/EMI regulations to adapt and follow good design practices and ditch bad ones those one which had not adapted went bancrupt and dropped from the market). That made these very wide range ADC's irrelevant. It is really hard to find any new ADC with such range, and there are only some few slow ADC's for some legacy industrial system interaction. Even 0-5V ADC market is starting to shrink at the moment, as proper designed PCB can have lesser dynamic range of analog signal and still be able to get noise free. It is not how big dynamic range is, but how many bits we can quantitize to so that least significant bit would still be above the noise. That is modern analog signals can be quantitized with 24bit ADC's, where ancient ADC from 70'ies would do 12 - 14bit, sometimes even 16, but those last two most significant bits would be noisy. Common and Fast ADC's had 8 bit resolution. And that is what old Cameca hardware used: 0-10V ADC with 8 bits. interestingly it showed in PHA only up to 5.5V Why it was hiding the rest?

New gen multilayer PCB use 5V ADC, also 8-bit. Practically it is the same amount of information as old design. Nothing more and nothing less.

On older electronics, how many counts should we expect at 10 V? Would it be worth using "differential mode" with a normal baseline and a high window edge at 9.9 or 9.95 V? I doubt it, since integral mode measurements have not been noticeably worsened by them already. We could measure the effect though by measuring counts in integral and this proposed gigantic differential mode.

at 10V? I would expect same amount of counts as on modern 5V., which in both cases are not reported to PHA graph by firmware as would make steep single measurement channel-wide anomalously looking peak. If you are asking about the measurements I did those, and there is difference with wide diff and integral when peak of PHA distribution highly escape the right side. I encourage to do them on your own, you will see on your own the difference. The integral mode actually does not care about height of the pulse, as integral count is triggered by raising edge of pulse. Althought Cameca hardware do both measurement simultaniously despite selected mode and even return both with library when asking for counts.

But in my opnion new generation of WDS card have lesser bottleneck and If I could I would change old hardware to new on our old SX100. To put more fire into oil (or to put the bag into the cat ) I should clarify why. You see this 10V ADC is shared between 3 spectrometer! (1,2,3 spec use one ADC, and 4 and 5th spectrometer use second ADC). On newer hardware every spectrometer have its own ADC. Althought thats not whole picture... ADC's needs to wait for its turn to report the measurement results, as all 5 of them use same digital bus. But so far I could not find that it would influence one spectrometer, if other spectrometer have small or high count rates. On old hardware... well they have some clear additional dead time if other spectrometer is having very high count rates as they need to share ADC's. New generation WDS boards is one of upgrade worth every penny. Other cards of new generation are not so critical, or I would even say could be worse.

P.S. clipping is affecting pulse height, not count rate.

[aducharme]

there is difference with wide diff and integral when peak of PHA distribution highly escape the right side

Yes, you're describing a flipped version of pulse height depression ruining a measurement. I was curious about this

be it raw photon have 10keV or 100keV or 1MeV or 1GeV - if it had produced any Townsend avalanche in GFCP, and counting electronics is not in enforced dead-time time gap, it will be registered at clipped upper value of ADC input range in PHA

and how many counts you would expect to exist at 5 V/10 V (depending on age of electronics) in the case where the PHA is properly set up. In other words, do cosmic rays and other high energy artifacts (if any exist) alone create a measurable amount of counts?

You see this 10V ADC is shared between 3 spectrometer

lol. Obviously the design has worked well enough, but it is a kludge that I'm surprised to see. If new generation WDS boards are worth every penny, are current dead-time corrections, built to work on any microprobe, inaccurate on these ADC-sharing instruments?

[sem-geologist]

Ok, You made me start doubt myself for a moment.

there is difference with wide diff and integral when peak of PHA distribution highly escape the right side

Yes, you're describing a flipped version of pulse height depression ruining a measurement. I was curious about this

To begin with there clearly is few peculiar things going: PHA graph does not report last few channels. And I also am not sure if we can set diff mode window to include the last channel. Graphically in peaksight if we move the the window by hand, be it old peaksight or new peaksight sofware, it pushes window back. The only way to select the higher bound as far as possible to the right is using text field and entering large number, it will then enter maximum possible value (i.e. on SX100 with older WDS card and peaksight 5.1; if base line is 560mV it can set 4999mV as window, which would make the right position of it be at 5.559mV. So, does it include last channel or not? I believe on both of systems (old and new) it does not include the last channel and Cameca libs and software does not allow  to select the window wide enought to include last channel. But my memory could be wrong. If you have time please repeat that experiment.

be it raw photon have 10keV or 100keV or 1MeV or 1GeV - if it had produced any Townsend avalanche in GFCP, and counting electronics is not in enforced dead-time time gap, it will be registered at clipped upper value of ADC input range in PHA

and how many counts you would expect to exist at 5 V/10 V (depending on age of electronics) in the case where the PHA is properly set up. In other words, do cosmic rays and other high energy artifacts (if any exist) alone create a measurable amount of counts?

cosmic rays do in casual conditions about 2 to 10 cps. Maybe there are higher fluctuations depending from sun cycle - I don't know, but if we do normally thousands of counts per second - that is at most 0.02 to 0.1%, and as it influence both, peak and background measurement it is practically removed. Same would apply for other external noise which influence both peak and background measurements the same. However there could be like 50% or even more in the last channel, when we set high gain and shift the distribution to overflow the right edge, and if we loose those counts by using diff - we loose a lot.

One of the key problems when tailoring the PHA for specific diff mode at moderate count rate is relying on setting the PHA only for peak position at moderate count rate. The background position which naturally would have much less counts and would have the PHA distribution more to the right and would escape partly the set narrow window. This is how it is easy to fall into the fallacy that PHA narrow window can significantly increase the peak/bkgd ratio of the measurement (I can't remember the exact papers on this, but there are at least few of them spreading such brilliant wisdom). No - it absolutely does not, as PHA distribution at background position would be shifted to the right and would miss partly the narrow PHA window and thus produce much lower background measurement compared to integer mode measurement, making peak/pkg look in the number artificially better, but missing the point that result is biased.

There is one exception where narrow window could work - if you can manage to make PHA distribution stand in the place (not shift) for counting rate range which includes target peak and background. That is doable only on Cameca hardware to some extent as those have both fine grained gain and bias, and I believe it is unachievable on Jeol hardware as they have only bias fine grained, but the gain is very course grained. There is small complications also as new dead-time calibration proposed by probeman is tailored for integral mode. The problem with narrow window (in case if achieving PHA with no shift) is the pulse pile up, which will make a deficit of counts inside the window with growing count rate. It needs then different dead time model. So efficiently such the common dead time mode is applicable for diff only at low – moderately low count rates, before "photon coincidences" rise in significant numbers. I use integral mode in 99.9% of cases, and use narrow diff window only for Pb Ma in geochronology applications with modified bias and gain for no PHA shift (remember you need to do also high concentration standards right with narrow PHA window). And I guess I would use just integer mode also there if we would use ProbeforSoftware, as PHA can shift with changing weather... I  use PHA filtering there due to OEM Peaksight weakness in circular interference corrections. The Pros of diff window in that single case overcomes its numerous Cons.

You see this 10V ADC is shared between 3 spectrometer

lol. Obviously the design has worked well enough, but it is a kludge that I'm surprised to see. If new generation WDS boards are worth every penny, are current dead-time corrections, built to work on any microprobe, inaccurate on these ADC-sharing instruments?

It actually works really great if knowing its weaknesses. On our old SX100 (with old WDS boards) we try to not allow any of our measurement to go over 15kcps, where clearly unlinearities starts to show up. Using peaksight is a limiting factor due to no ability to change dead time constants, maybe if we would use Probe Software we could do the calibration of deadtimes better and the upper reliable boundary clearly would grow up. We utilize multi conditions a lot to overcome the limitations. Both our probes had produced k-ratios for original FIGMAS tests within 0.5% boundary - that is not bad at all (our mount is #1-14). Also those two ADC's contains separate buses to the FPGA's, thus works in parallel, and Cameca tends to distribute out high intensity capable spectrometer connections across them. In our case the effect is very faint, as our first 3 spectrometers have small XTALS, and only 4th have large crystals capable to produce high count rates. To make 3 small XTALS to produce some super intensive counts to get real clue I had to manually set C1 C2 to make some insanely strong beam (if my memory not fails me around 3000nA). Thus I am not tearing my shirt off to get a new card as this one works correctly if staying within its limits. But if there would be an opportunity I would grab a new gen card for that without any hesitations.

[aducharme]

If you have time please repeat that experiment.

No promises but I will try to do so this weekend.

One of the key problems when tailoring the PHA for specific diff mode

John and I completely agree with you. John has taken to referring to this approach to PHA calibration as "count-matching" (as opposed to matrix matching).

And I guess I would use just integer mode also there if we would use ProbeforSoftware, as PHA can shift with changing weather

You know, I think integral mode is more robust to barometric pressure changes than differential mode. One of the impacts of lowering pressure is shifting the pulse to higher voltages. Llovet et al.'s 2021 review paper includes this plot showing the effect:


With a reasonable baseline, integral mode would not lose counts from shifts due barometric pressure, but differential mode would. It looks like there would still be an effect from the change in pulse shape, but I have no idea how that affects what the instrument reports when actually performing WDS.

[sem-geologist]

I never had seen such a huge shift in PHA depending from weather on Cameca hardware - maybe it have something to do with bubblers which partially would prevent back stream of oxygen and water vapor?

Below I reattach my achieved fixed PHA up to 50kcps for high pressure spectrometers.

First, The initial shift review when relying on Automatic PHA setup:



After reducing bias (gas amplification) that reduces average load on feedback capacitor of preamplifier (makes it more empty than charged), which (skipping all important and lengthy details) delays the type I PHA shift. Type II shifting as we see still kicks in when going above 50 kcps due to increase of photon coincidence (again, this is gross oversimplification):



So practically staying bellow 15 kcps, diff mode on high pressure spectrometers can be used after reducing bias and increasing the gain. The range limitations are good enough for good calibration of well behaved Pb standard, and trace analysis at high current.
But as I said, that is the only case where I use diff mode and only with reduced bias and increased gain.

[Probeman]

I never had seen such a huge shift in PHA depending from weather on Cameca hardware - maybe it have something to do with bubblers which partially would prevent back stream of oxygen and water vapor?

It's difficult to compare, but we had this plot from Brian Joy back in 2022:

https://smf.probesoftware.com/index.php?topic=1109.msg10889#msg10889

I would be interested in seeing PHA scans from 10-30 nA up to several hundred nA on some large TAP/PET crystals.

For example as shown here:

https://smf.probesoftware.com/index.php?topic=1475.msg11330#msg11330

And separately, with the PHA peak fully above the baseline and in integral mode, do you see the same count rate as the gain is increased and the PHA peak is shifted to the right, even when it is cut off graphically? As shown here:

https://smf.probesoftware.com/index.php?topic=1475.msg11343#msg11343

[sem-geologist]

And separately, with the PHA peak fully above the baseline and in integral mode, do you see the same count rate as the gain is increased and the PHA peak is shifted to the right, even when it is cut off graphically? As shown here:

https://smf.probesoftware.com/index.php?topic=1475.msg11343#msg11343

As far I remember I looked into that – it does not influence that aspect. Indeed, the contrary (increasing bias, and reducing gain) could increase count loss at high count rates. As it would saturate C.S.P. and it could start introducing additional dead time there (which in normal conditions is dead-time free).

Albeit there is a limit how far bias can be lowered to still get same count rate, for high pressure counters the plato (where count rate is staying the same) is down to 1600V (it also very depends from X-ray energy), but I never set it this low, I set it rather ~1700V from commonly default auto set ~1830V (by peaksight). Why count rate starts to drop after going below ~1600V? (choose one/or both of reasons) a) as effective field around the anode in GFPC starts to be smaller than focused diffracted beam b) field is too weak to consistently produce Townsend avalanche from every X-ray produced photoelectron, and electrons are either reconnected to positive ions, or drifts without amplification to the anode where they are collected, but because no amplification they are invisible hidden in the noise of the anode.
If you ever looked to this and similar scheme (from wikipedia):



The X axis on that scheme is bias voltage of the detector and as you see the proportional region is one of most broad regions with proportional growth of the curve. What is that curve? "rate of charge collected" writen there on that y axis sounds kind intimidating, does it not? it could simply be replaced by "gas amplification", or simply "amplification". That common used chart is not perfect, as it would suppose there is continous discharge at geiger region (no - its not it is chocked; but what to wish from wikipedia), and streamers or SQS (self quenching streamers) are missing between proportional counters and geiger region.

A small digression:
Could our detectors work in geiger region? maybe, in case of leak from gas chamber if pressure inside the GFPC would drop below 1bar...
Could our counters get pushed into streamer mode? I would not want to find it out as that could damage counter windows. How to get closer to streamer mode: very large bias + very high photon count of high energy X-rays + reduced gas pressure (i.e. localised micro leak in the window).

The one of crucial piece to understand is that GFPC is two devices in one: detector of X-rays and adjustable amplifier.

Simplified equation of final pulse height would be something like this:
A_(PHA) = E * A_bias*A_{C.S.P*shapper}*A_gain

where E is energy of X-ray photon, A_bias is amplification by townsend avalanche, A_{C.S.P*shapper} is fixed amplification by Charge sensitive preamplifier and shapper, and A_gain is analog amplification before pulse detection (integral mode counting) and before PHA in the signal pipeline. Only A_bias and A_gain can be adjusted. So i.e. 2 x 2 or 1 x 4 or 4 x 1 – all of these example multiplications of two different amplifications will result in the same final amplification of 4.

So if we have about 1830V bias, the E*A_bias*A_{C.S.P*shapper} result in this kind of amplitude for middle range of X-rays (i.e. Ti Ka, but I am not sure if it was not Ca Ka or Cr Ka) be something like this:

As we can read graphically the observed (by oscilloscope) amplitude of these two stages of amplification it would be about 3.1V.

Now pay attention: for Ti Ka pulse using LIF, or PET, from both of them it would produce exact same amplitude at that point in pipeline (the above oscilloscope picture) if using that same exact bias. Cameca peaksight use gain to move the pulse to different position in PHA on x axis when using "auto PHA", it would use <1 gain (number <1024 in GUI) if LIF, and >1 gain if using PET (>1024). It tries to place PHA peak to some "theoretical" position so all possible pulses for given selected XTAL in wavelenght would fit within 0-5V scale of PHA – which as we already know has actually no practical meaning. Its kind of irony that those default procedures are shitty for high count measurement: in case of PET we ironically get better set PHA exposing argon escape peak - but being exposed to much higher count rate in default Peaksight we would get massive unaccounted count rate deficit from massive "multi-photon coincidences" (compared to LIF, which would not achieve such high count rate at same beam conditions); and for LIF we would get clipped peak at low baseline, which with increasing count rate would clip it more and more. We would have in both default OEM settings poor performance on both XTALS at high current rate.

Then, this A_gain amplification is  happening on WDS board in electronics box (VME card) and this amplification is in between 0 and ~4; where 1.0 (no change in amplitude) would correspond to 1024 units (max is 4096, correct? I guess 1024bits = 1, i also could be that 1000bits = 1.0 gain). Auto PHA for higher energies x-rays often set that gain below 1000, and that means [E * A_bias] part of amplification is very large, and so amplitude needs reduction before going into PHA, so it would fit inside 0-5V scale or be placed at some bizarre theoretical position.

Why it would be designed like that? I think most probable answer would be that noise in the [E * A_bias] part of signal is not amplified, where A_gain also would amplify the the noise, as it is simple direct amplification. Such assumption (of higher bias benefit) is fine if staying in low count rate (maybe <=5kcps), but it ignores dependency of amplitude multiplier A_{C.S.P*shapper} from [count rate]*[E * A_bias], which bites back with severe PHA type I shift which [C.S.P*shapper] combo introduce.

But if you look again to that oscilloscope snapshot above, you can see that noise floor is really low, and there is rather no problem if Gain amplification would increase it twice or even more – the peak/noise ratio would still be perfectly acceptable enough to discard the noise and pass the pulses in the pipeline. So i.e. reducing that pulses to 1.5V from 3.1V (i.e. bias down to 1700V) and then using gain to increase it to 4V (gain of 2.7 which would be ~2700 units in GUI) would produce same amplitude in that case as default Auto bias/gain of peaksight. However lower bias/higher gain would work better for higher count rates, as CSP*shapper would be less affected by lower [E * A_bias].

Another very important benefits of reduced bias.
GFPC ages - the anode gets contaminated with hydrocarbon gunk which impairs its performance.
The aging is directly proportional to number of ions of methane which breakdown into smaller components and then combines into larger hydrocarbs on the anode, which proportionally scales with gas amplification amplitude. If we use gas amplification amplitude conservatively, by reducing gas amplification we could twice or more prolong functional age of GFCP before needing a replacement (or cleaning). Bias reduction should in particularly be not neglected in case of high current used with measurement of large quantities on large crystals resulting in very high count rates.