Coincidence events (pulse pileup) are making it difficult to extract the Duane-Hunt limit from the spectrum. Lower the beam current and measure again.
Extracting the Duane-Hunt limit is always a challenge because, by definition, the counts go to zero - so zero signal. You can't expect too much precision in the estimate. However, the best approach is to count for a long time at a low count rate and fit the signal to a continuum model in the range of energies right below the DH limit. This is the approach that DTSA-II takes.
I appreciate how DTSA-II and NeXL(Spectrum) takes care of that. The method as is implemented in those are robust to find out sample charging effects as these will be always strong enough to show the difference. What accuracy and precision does it have? I think we don't know, and that is what I want to try to find out. Accuracy is not so much influential for detection of charging of the sample as it will influence D-H a lot.
However, the accuracy and precision need to be well understood if method is used for checking or calibrating of HV voltage as proposed there by Probeman, and also here :
https://probesoftware.com/smf/index.php?topic=1535.msg11937#msg11937 (the point 4). I actually got influenced by reasoning of this post and without a second thought I was insisting on doing such tests last year when acquiring a new SEM for our lab. I got aware about the problem only recently when being forced to dig deeper into how HV tension is produced and what other vendor power supply provides. WDS method indeed provides better means for defining Duane-Hunt limit... but still, how much precise and accurate it is? You can increase precision by decreasing count rate and increasing time, but what about accuracy? The fundamental question is Can an electron be halted to 0 velocity? will that energy difference will be exactly the energy of electron before impact?
The first fundamental question: Does perfectly defined D-H will be exactly same as beam energy? Or is there some substantial energy always left in slowed down electron and thus D-H should be a bit (how much) smaller than beam energy? Is there some fundamental difference?
Then next question is how accurate our measurement is (for precision we know and agree already for some tricks). EDS, especially SDD type will loose efficiency over 10keV. In general Bremstrahlung going toward higher energy on EDS seems diminishing. Thus Probeman suggests of using heavy (Bismuth metal?) standard for increasing the continuum before and thus the interpolation down to 0 would be easier. WDS naturally has increasing intensity toward higher energies (or lower sin theta) - thus the problem of low counts transiting into 0 being not very well pronounced is not there, which is nicely illustrated by example from chenderson. Also WDS pulse-pile-ups do not show up in the spectral (wavescan) form (they affect measurements at single static spectrometer position), thus also there is advantage compared with using EDS for D-H. But I disagree it is accurate - if line is fitted to section before getting to 0 of such wavescan. Also at low sin theta the spectral resolution is not so much different form EDS. But why the spectral resolution was at all mentioned before?
The missing piece is "deconvolution" - the ability to deconvolve that continuum wedge going to 0 will define the real D-H, and will show that D-H is always smaller than set HV beam energy. On WDS it is easy as the wedge is pretty steep, and thus simple half length at the (supposed at given spectral position) peak bottom (i.e. in my case that is about 0.1keV at 10kV for LIF) needs to be subtracted. For EDS due to more complicated Gaussian shape and shallow-angled wedge such retraction is not giving anything, as like Ritchy said - it is complicated with pile-up continuum.
Now I did a small experiment on DTSA-II, I made a MC at 14.88 kV and 15kV - for Bi metal. visually they show sigmoidal edge - where clearly the curve hits 0 intensity just a bit above 15kV and a bit more. Due to lack of deconvolution of that edge when using built-in D-H function it gives about 15kV and 15.1kV, so the method highly overestimates D-H. Going with lighter elements can make underestimated D-H. In heaver element spectra there is temptation of subtracting 3 sigmas equivalent calculated for FWHM at given initially estimated D-H position. However at lighter materials this will highly underestimates (shifts down) the DH.
Then again, HV supply can be calibrated at factory, but HV cable length and other factors will influence final acceleration, column geometry - those will influence the final electron landing energy. The more I dig into this the more I am convinced that 1-2% uncertainty of acceleration voltage is hard to improve.