Delay & Jitter through a discrete gate

[PwrElectronics]

Anyone ever see this?

So, a 74HC type buffer.  I can get specifics on the mfg and logic family.  Its one of those 1 gate types used for glue logic.

So, normally the propagation delay through this is something <10ns.  Usually more like 2ns.  Got a unit returned with faults and the issue traced down to a specific buffer that has variable and often quite jittery and excessive propagation delay.  I've seen anywhere from 80 to 800ns.  A colleague who first found this was seeing delays approaching 10us on occasion.  I got the board pulled out of the assembly to take a closer look.

I've looked at the rise/fall of the input to the buffer and no issues.  Power is good.  The signals checked at both some convenient vias on the board as well as the device pins directly and no difference.  Tried heating and cooling the device with no effects.  Thermal camera shows its not running hotter than a good device on the same board.

Problem seems to be with the buffer and not something with the solder or traces.

What could be wrong with this to cause this behavior?

[PCB.Wiz]

It sounds like some internal contact that should be low impedance has gone hi, due to damage or corrosion.
Strangest CMOS failure I saw was a MCU pin that flipped polarity on input direction.
Still ‘worked’ but upside down.

[PwrElectronics]

This is a pretty new unit and no evidence of corrosion on the board.  Solder joints look OK under microscope.

I've seen a few strange things too in my career but this is a new one on me.   

[PCB.Wiz]

This is a pretty new unit and no evidence of corrosion on the board.  Solder joints look OK under microscope.

I've seen a few strange things too in my career but this is a new one on me.   

I was meaning inside the cmos part.

[RoGeorge]

The propagation delay is heavily dependent on the supply voltage.  I've made once a ring oscillator, for the fun of it, with 74HC00.  It was able to do 100 000:1 frequency sweep by simply sweeping the Vdd.   

Check you have good decoupling against any power noises, with no voltage spikes or surges measured near the chip pins, and that the voltage regulator doesn't transiently go into current limiting or self oscillating (for small amounts of time, small enough to not be visible on a casual look on a slow oscilloscope).

Another thing is the slew rate, try using a trigger Schmitt input and see if it makes any difference.  Maybe only some edges are too slow, not all of them, and that is why on the oscilloscope the raise/fall time appears to be OK, when in fact for some rare edges it might not be OK.

Beware that sometimes the IC fabrication process changes (upgrades), and the newer chips are much faster.  This can be very bad, because the fastest a gate is, the more sensitive it will be to reflections in the PCB traces, noises and glitches.  So, a circuit that used to work perfectly well with the old/slower chips, might not work properly, or might not work at all with the newer/faster version of the same chip.

Also, 10us instead of 10ns propagation at a constant Vdd is not possible.  If it were to be so, no digital circuit would be able to work.  Something else is happening with that board.  Maybe those with 10us are from the CD4000 family and not from the 74HC, so fake chips, I don't know.