In some of the EFM32 device (Gecko/GG/LG/WG) datasheets, there is a graph showing the comparator response time against various biasprog settings that could be confusing -
Going by the plot above, the response time would be very small (~0 µs) when the biasprog setting is 4 and HYSTSEL is 0.
These graphs were generated using a test where the sampling frequency was 2.5 MHz, which in turn made the response time unit 0.4 µs. That's why we don't see any values between 0 to 0.4 µs. However, more tests were performed for newer EFM32 devices (like Zero Gecko) with increased resolution. The graphs in these datasheets are more realistic in terms of response time.
The fastest response time found in internal tests for the Zero Gecko device was 34 ns. This is the average result of a very small number of devices at typical conditions. The settings used were -
FULLBIAS = 1
HALFBIAS = 0
BIASPROG = 15
HYSTSEL = 0
Supply voltage = 3.0 V
Temp = room
Common mode voltage = 1.25 V
ACMP input: VDD to 0 V transition
This is subject to process and temperature variations, and a system with 34 ns response time cannot be expected to function reliably.