This reminds me an old (12 years or so) design of mine, where I needed variable gain cells, DC to say 10 MHz, 12 bit, 1:50 factor or so. I used paired 2N4391-s; one for compensation feedback, the other as a variable resistor (having the gates shorted was handy as it allowed me to solder the cans together for temperature coupling). Buying about a 100 transistors inevitably bought me most if not all of them from the same wafer and I could find decent pairs easily enough. Now obviously I did not get 12-bit temperature stability, but I had anticipated this sort of trouble and had put an AD792 temperature sensor next to them :-). I had an MCU and it did turn the fine gain to compensate; this gave me a 10x+ advantage over competitors (who were _awarded_ all major deals in spite of that, but that's another story). Now the rest of the story is that I got no 12-bit integral non- linearity, either. Obviously even the 4391, in the reasonable low-voltage range where I used it, was about 0.5 to may be 1% non-linear, which was not acceptable. Fortunately I had enough processing power (on a 16.67 MHz 68340 :-) to acquire the non-linear data into a somewhat longer spectrum (8192 was the norm; I did something like 8500 or so) and continuously (IIRC 10 times a second) converted it down to a linearized version.... Worked wonderfully, some are still in use. Got carried away by just a phrase, I guess, but I see this as a "heroic" design of mine (there are more reasons except the above for me to think so :-) and I felt like telling about it although being only very remotely related to the context...
The zener voltage for a given transistor type does not seem to change from batch to batch; sometimes one can even change vendors. Changes seen seem to be less than 1%. In the case cited, i believe you will not see oscillation or negative resistance effects. Run the current down toward zero to see some spetacular "bazz-fazz".
No, there's no guarantee of this between batches. It's only guaranteed to be above a certain minimum- that's what's on the spec sheet.
But if they decide to cook the next batch a little longer, or change the device geometry to improve some other spec, the zener voltage could really change a lot. As long as it remains above the spec, you have nothing to legitimately complain about if the zener voltage zooms up to 15 volts in the next batch.
As an example, I have a batch of MPSA42's, spec says 300 volts maximum, but a half-dozen I tested work fine up to about 830 volts plus or minus about 10 volts. Does this mean I can depend on this? No.
Better to not go into production if you're depending on this!
I remember doing that as an experiment in college. My recollection was you only need 100uA or so. Maybe up the resistor a bit, say 47k to be safe that it breaks down. [Of course, you need to do the experiment. ;-) ]
I strongly do not recommend darlingtons for that; too variable and a high zener due to two E-B junctions to break down. I have tested a number of transistors, and for a given brand and type, the berakdown voltage (in the *stable* region for the part) is remarkably "constant" between devices and between batches (even when a year or so seperated). All transistor E-B junctions have a range of currents where they exhibit negative resistance as well as oscillate; some are really nasty. It seems that a majority *start* with a large negative resistance characteristic (even in the nanoamp region), and slowly go towards zero then finally positive. Then again, most zeners that are made also have similar problems.