A few customers had problems with the start-up of some Colibri modules.
Due to slightly higher in-rush current and more sensitive reset circuit on the Colibri PXA270 V2.2b a voltage drop below 2.97V could lead to boot problems.
The customers added additional bulk capacitance on their carrier board to the 3.3V and VCC_BATT voltage domain. On the Toradex EvalBoard we have about 450uF of bulk capacitance on the 3.3V rail.
The Marvell PXA processors have a known errata. The processor doesn't boot when performing fast power cycling and the voltage on the VCC_BATT domain falls below 2.4V but doesn't reach 0V before powering the whole processor again. This problem was seen on a very few customer carrier boards. The problem could be related to some back-feeding effects on e.g. GPIOs which are still powered when the main 3.3V is off. We never saw this issue on our carrier boards (EvalBoard, Iris, etc.).
Avoid back-feeding on pins when the Colibri module is powered off.