I'm sure everyone knows this, but I would simply add that the absolute thickness of a carbon coat is less important than its reproducibility. That is, even 10-15 nm of carbon is often sufficient to provide sample conduction, but the critical question is: are both the standard and unknown coatings the same mass thickness?
This is particularly critical in two instances. First, when measuring very low energy emission lines, e.g., N ka or O ka, where they are significantly absorbed by carbon, and second, when measuring emission lines with a low over voltage, e.g., Fe Ka at say 15 keV or less, and a small difference in carbon coat thickness can significantly reduce the "landing energy", and hence the generated intensity (see Fe K emission over voltage curve).
The main reason we usually apply a 20 nm thickness is, I suspect, that historically we use the color change on polished brass to determine the carbon coat thickness:
https://probesoftware.com/smf/index.php?topic=921.msg7063#msg7063The color change at around 20 nm (red to blue) is quite obvious (violet) and hence usually provides the best reproducibility while looking at the color from the reflected light of the arc. Of course today we have quartz thickness monitors, but I'm chagrined to say that the thickness monitor on our venerable Edwards 306A carbon evaporator has long since failed, so we continue to determine thickness on polished brass.
http://www.geology.wisc.edu/~johnf/g777/AmMin/Kerrick_C-coat.pdfSpeaking of the color violet, does everyone know why the color spectrum named by Newton (ROYGBIV) has 7 colors? Because really, what exactly is the difference between violet and indigo? Here's a discussion of one possibility why we have seven colors:
https://www.dailybreak.com/break/cabinet-of-curiosities-why-indigo-is-in-the-rainbowThis is also worth a read:
https://en.wikipedia.org/wiki/Indigo