Some time ago, I measured the startup time of the TLV3492AIDCNT to be about 1 ms. We now have a number of parts that take 3-25 seconds to generate a valid output after power up. I've verified good power and inputs. Some devices start with a low output and after some time begin drifting up slowly (analog output) toward our supply rail (3.3 V) over 3-4 seconds, then snap to ground, another few seconds pass, then the output is valid. Once started it works fine. Powering off and back on too quickly will result in a normal start. Cold spray and a heat gun have no effect. One of the inputs is a steady 2.2 V. The other input is a Hall effect sensor that is mostly around 2 but increases to as much as 3.3V when a magnet passes by.
Parts that exhibit this problem constantly one day can test fine the next day. This design that has been in production for several years and this is our first experience with this problem.
More detail:
The negative terminal is driven by a voltage divider 4.99k (top) and 10k (bottom) along with a 0.1uF cap. The top of the voltage divider is tied to the same 3.3V rail powering the TLV3492.
The positive terminal is driven by a 5V Hall effect sensor after it's output has been scaled down with a 4.99k / 10k voltage divider with a 100 pF cap. This prevents more than 3.3V from reaching the TLV3492's inputs.
The output goes to an Altera CPLD i/o pin configured as input only. There's nothing else on the net.
We substituted parts from another reel which appeared to cure the problem, only to see a failure again in subsequent testing. Not sure if the date codes were the same.
Thanks for your time and any advice you can offer.