Using a scope, we see a spike on the output (which explains all our functional problems), but we don't see a spike on the input we supply. So this spike comes from inside the FPGA, probably from the IBUF.
The spike is correlated with activity on other pins, so we immediately suspected something like ground-bounce issues. We have ensured that the FPGA is well grounded (by connecting all ground pins to a common ground reference).
We have tried moving the input pin to different IO banks. We have tried moving other pins. We can see no difference.
We tried holding the rest of the design in hard reset, so only the few signals in question are active, and the problem persists.
This occurs after DONE goes high? Befor the device is configured, and while it is being configured, the pin should either be tristate, or have the week pullup enabled if thst is selected by the HSWAPEN pin.
On power-up, the IO pin is undefined until the supply voltages are detected to be OK. Before that, the device holds IO pins in the tristate condition, but a low Vcco voltages, a pin ,might appear as a logic 0 due to the protection diode being forward biased.
Have you che checked the signal integrity of your layout? Might this be cross talk noise from an adjacent signal? What happens when you load it with 1K ohms?