BQ24232: Charger stops working after a few uses up to a few months.

Hello Emanuel,
Can you please post you schematic.
1) Can you clarify this statement "With only battery connected, IN is outputing voltage ; about 3.5V and has a sourcing current of about 2.7mA". Do you mean that the input voltage is 3.5V and only supplying 2.7mA?
2) The charger does not disconnect BAT from OUT since the FET is fully on and unregulated. The battery voltage shouldn't be at zero. Do you have a protector on your battery?

I am attaching part of the schematic (specifically the USB Vin Data in portion)

Regarding point No.1)

I disconnect the charger/usb cable. I have a good battery with a voltage charge >3.9V. The output voltage is equal to the VBAT. At pin 13 normally the voltage should zero, but on a couple of boards I can measure a voltage of 3.4-3.6V. The current I mentioned was measured connecting a DMM amp meter between the pin 13 and GND in order to see if it was short from somewhere else on the board, but it is in fact leaking through the BQ24232. If I replace the component it works as expected. 

I am already implementing a TVS at the D+,D- rail, but I will add one between +5V and GND just for extra general safety. Is there anything else that I am missing, or something that could protect it additionally? Could it be that the parts fail during the soldering? Should we specify something additional?

I am attaching the board layout just in case. It is a 4layer board with adequate ground internal layer.

Emmanuel,
When you measured 3.4-3.6V on IN when you expected zero it is possible that the input capacitors have not been fully discharged before your measurement was made which is why you see 3.4V to 3.6V on IN. When you mentioned that you measured the current by connecting the amp meter between IN and GND, is the 2.7mA current continuous? This will short input to Ground.
When you measured 0V on the battery, was the battery still connected to the board. What was the measured voltage when the battery was disconnected from the board/

I used the Ammeter connected between IN and GND and the 2.7mA is continuous. The capacitors are kept charged due to this. This is failure case A.

Failure case B (of which we have more failures 3-5 out of 100) is that we get 0V.  This is  measured with the drained battery still connected. When I disconnect it I can see the battery detection routine (Voltage goes high and low every sec) but any battery even brand new (3.8V charged) are connected the charger does not begin charging. If the battery is disconnected I can see current being drawn from the charger about 15mA (minus an LED drawing 5mA). When I connect a battery even those that are at zero the power draw from the USB goes to 0.

 

Hello Emmanuel,
I tried this on my bench with VBAT present and VIN shorted, and i could only see the very small Iq been pulled from the battery. Is it possible that these units are damages for failure 1? The input FETs are typically off when VIN is not present.
For failure B, this is possibly due to the protector on the battery. Did you try to use a power supply that can sink current instead or a battery simulator?