My team is using the ADIS16365 IMU in one of our systems. During final assembly of the system, we command the IMU to run the "Gyroscope Precision Automatic Bias Null Calibration" described in the datasheet. Once the calibration is done, we power cycle the system, then verify that the calibration worked by calculating the gyro bias to determine the residual bias.
After doing this on ~1100 IMUs, we can see a pattern developing in the residual gyro bias after calibration. See attached figure.
You can see the residuals are centered at 0.05 deg/sec, which just so happens to be the LSB of the gyro offset register in the IMU. This makes me think that when the IMU calculates the bias, it is truncating the result to the LOWEST (not nearest) LSB. This results in a average residual gyro bias of 0.05 deg/sec (1 LSB).
I would like to see if someone could confirm my hypothesis by inspecting the software in the IMU that performs this function. If ADI can confirm this is what is happening, I can make adjustments to account for this (change pass/fail limits in factory, run calibration outside IMU, apply generic offset, etc.).