There are a tiny handful of switch FETs in the 3456A that have generic JEDEC part numbers given in the "replaceable parts" table. The
overwhelming majority of them have a custom HP part number, which I agree refers to some kind of special selection... but I haven't seen any record of what it is they're selecting for. Gate leakage is probably a good guess.
Thanks for pointing out the BAV199--those look like nice little diodes! It's too bad you'd have to get the lead stretcher out (i.e. solder leads onto a SOT-23) to use them in this application.
As an experiment I tried lifting the gates of Q106, Q108, and Q118. That didn't seem to have any effect, so I reconnected the gates again... and found I had modestly increased the offset somehow. Flux residue? Heating from two solder operations altering the components? No idea, but I'll try thoroughly cleaning the input switching area with isopropyl alcohol later.
I tried readjusting the charge balancing circuit (R147 and R148) but this seemed to have almost no effect.
During that last adjustment I found that with autozero and the analog filter enabled and an integration period of 100 cycles, the displayed offset changed from about -3.0 μV or whatever to about +35 μV. This
only occurred with all three factors; changing any one of them brought the displayed offset back to sanity. I saw a bunch of high-frequency spikes on the input amplifier's output (TP303) that correlated perfectly in time with the ADC's slope changes, which were only present with the analog filter enabled. I
think this might be because the TLE2071 I used in U307 is too fast/performant. I'll try swapping it with an OP07 just to see what happens.
All that aside, I'll think aloud here for a moment. As has been established previously, gate leakage is the primary failure path we suspect. That implies we're looking for FETs that are 1.) off at least part of the time during a measurement, especially low-range DC Volts; 2.) have their gates pulled to -18 V when off; and 3.) are connected directly to one of the two main measurement nodes. By my reckoning, that makes my list of suspects the parts below, listed in no particular order:
Q109 (changed; no improvement)- Q101 (don't remember if checked or changed)
Q114 (changed; greatly reduced offset)- Q116, DC V low-range switch (doubtful; offset in high DC V ranges is similar to that of low ranges)
Q115 (changed; no improvement)- Q105, 4-wire ohms low switch (unchecked)
Q102 (changed; no improvement)Q103 (changed; no improvement)- Q110, AC V switch (unchecked)
- Q112, self-test #7 switch (unchecked)