Thank you for the excellent scope shots. As you have already pointed out, the bootstrap configuration is a good starting point for us to look at. Is there any sizable resistance in the path from the 12V source to VDDB that could be causing a slow charging of the bootstrap capacitor? Which UVLO grade are you using?
Is it possible to get a scope capture that contains VDDA? This will either require a floating differential probe, two probes in a math subtract mode, or even just a probe on VDDA referenced to GNDB. This will allow us to see if the on-time of channel B is long enough to fully charge the bootstrap capacitor on the high side.
It is recommended to use at least a 10µF capacitor for decoupling.
It sounds like you have probably found the cause. The 11V UVLO is usually chosen to accommodate threshold voltages commonly seen in IGBTs. Generally, the secondary voltage UVLO is chosen to be slightly above the minimum acceptable voltage the designer wishes the gate of the switch to be driven at.
If you are using IGBTs, your idea of increasing the secondary isolated supply is a good choice. If you are using standard MOSFETs, using a lower UVLO grade ADUM3223A or B is probably the more robust option.
Don't forget the diode forward voltage drop when estimating the gate voltage that will be seen on the high-sides .