If someone using EDS is looking to this thread and think "oh those WDS-dudes, we on EDS have none of these dead time problems" - I would like to curb Your enthusiasm, especially if Your machine is running Bruker/Rontec EDS SDD detectors.
I had just discovered that dead time (live time) estimations can have very huge uncertainties in case the beam is blanked before running the analysis or mapping. Long counting (>30s) is not so affected, but short counting (<1s, few hundred of ms of pixel dwell time) i.e. for mappings will be a lot affected. Live time of EDS is counted with pulser method (0eV pulse at set frequency, if it miss - it means the counting electronics is busy). Frequency of pulser on Bruker signal processing unit is set (absolutely automatically, with no programmable access) depending from input counts which are constantly monitored and value just before launching the EDS acquisition is used for automatic setup of pulser frequency. In case it is 0cps (beam is blanked) it decides it is very low count rate and set that to (on SX100 tailored Bruker Nano XFlash) 1000Hz. In case input count rates would be near 1Mcps it would set 50000Hz (and other frequencies in between scaling to the input count rate). It does not depend from selected throughput. For mapping, where pixel dwell time is few hundred ms it will produce severely shortened live time and net counts from such short dwell time maps will be not comparable with long EDS measurements of standard material.
The fix is to add 2s waiting time before the start of acquisition (beam-on), so that SPU would get adequate estimation of input count rate and would select adequate zero peak frequency. There is caveat. if mapping starts in the epoxy (left-top corner) and beam would wait those 2 s there, SPU would get quite underrated count rate and would select underestimated 0eV peak frequency.