Consider the case where A is an input and B an output and, by an error, another voltage is applied to B (5V instead of the output 0V with A=0V). B will now be considered as an input because of the transition detected, and A will become an output of 5V, which will create another short circuit. Is this a bug in the chip? A vicious cycle?
There will be no burning shorts circuits. The drive strenght of the device is quite low. The device is also mostly for open drain IO where two pins actively grounding the line isn't a problem. The internal weak pullup won't kill anything.
I have used the TXB part for SDIO. The TXB is made for push-pull applications. Again, this device isn't capable of destroying things. I had issues where the A and B supply rail were turning on/off out of sequence and this could occasionally lock up the TXB but not to a dead short.
If we ignore the edge accelerators, the TXS is a passive switch. If opposite voltages are applied to the two sides, then the switch is on, and there is a short through the device, which might exceed the absolute maximum ratings.
With the edge accelerators, if the A side is driven low, and then the B side is driven high, the rising edge on B will result in the TXS driving A high for a short time. If the TXS is stronger than the other device (or if there is enough capacitance/inductance so that the other device is not fast enough), then the A voltage can rise completely, and when the edge accelerator times out, the other device will drive A low again, which can be detected as a falling edge, which will trigger the edge accelerator in the other direction. So it is possible to make the TXS oscillate.
After testing with a breadboard, here are the results :
- With Va=3.3V Vb=5V
- If A=0.00V and B is forced to 5.00V => A=0.01V and B=4.97V, and the current is 11mA
- If B=0.00V and A is forced to 3.30V => B=0.01V and A=3.29V, and the current is 9.8mA
- Same results with the 2 tests forced to 0.00V
So 4.96/0.011=451ohms and 3.28/0.0098=335ohms, it's confirmation of the datasheet : "N-channel pass-gate transistor (on the order of 300 Ω to 500 Ω)"
My conclusion : There are no oscillation bugs in the cases I was wondering.