Depending on the history of your LTZ1000 (e.g. solder heating, preliminary heating / cycling), the initial burn-in drift may take months or years.
Frank
Thank You Dr. Frank!!
I'll jump in here with some suggestions & thoughts after a few decades of building these for industrial use. Again: I build custom LTZ circuits for my customers that need to be used - not for PPM-chasing entertainment purposes.
These are just some observations picked up from decades of making various custom test systems:
1. When we assemble an LTZ circuit, we power it up (and add power cycles at least a few times per day) and let it run a couple years. That die inside is going to take a while for the crystal lattice to stress-relieve itself, and there's no way of knowing where your die came from out of the initial wafer. Every die singulated out of the wafer will have a different strain pattern once it becomes a single die, and every LTZ1000x is going to behave differently. It also depends on if the die was diamond-scribed or cut with a wafer saw, lasered, etc. Manufacturing processes change slightly over time, even though the basic die layout hasn't changed in decades. The random crystal edge damage seen along every die will have an effect on final performance, good and bad.
Usually one of these things will happen:
A) The LTZ circuit will be initially jumpy, but after a few year's operation it will settle down and become quite stable. Sometimes these 'grow up' to be the best stable references you've ever seen.
B) The LTZ will be initially fairly stable, but stability doesn't improve as much over time.
C) The LTZ will just stay wobbly and jumpy. If they are still doing this after 2 or 3 years, they probably aren't going to improve.
D) You might find a Golden LTZ now and then. These are born pretty stable form Day One, and then just get better. These are not common, and these are the jewels that would get picked for a Super 3458a (002 option) along with other hand-selected components.
E) There is always the LTZ that starts out rock solid, then as it settles in you see more drift. These aren't common either.
The point is: You have to be patient, and plan on a long time scale before you start taking any serious measures of your circuit.
What we do is assemble the LTZ circuits, just do an initial check to see if they are running, then burn them in for a few years. We use a 'hotter" heater resistor combo with extra 1k's in series with the final top resistor (like 15k~16K over 1k) and then maybe jumper out a 1k or two to get to final heater resistor ratio - and then let that settle in for another 6 months to a year. The final heater resistor value will depend on it's final running application, but minimum is 13k over 1k, and that is only for coolest environments. You might be able to get away with less, depending on application. We have plenty of LTZ circuits running in the field with 14k ~ 15k/1k ratio running 24/7 for decades, and they don't drift maybe 1 or 2ppm per year and get more stable - so I definitely do not agree with Pickering's suggestions about lower heater resistor ratios in every case. Just speaking from experience here after looking at a lot of LTZ's. SOME LTZ's might benefit from a lower heater ratio, but it really depends on where you'll be running these in their work place - and each LTZ is different.
For instance right now in 2016 we're processing LTZ circuits that will be delivered in 2019 to a wafer fab production line test system. This is how HP / Agilent / Keysight Vref boards are done also. It's not like they are soldered together and delivered the next day.
2. Stick to the LT datasheet circuit especially if this is your first time building an LTZ circuit. Suggest that you DO NOT use any other op-amp besides LT1013 - that amp is tailor made for LTZ circuit, and will get you best noise performance overall. The small Vos of the LT1013 isn't really going to have much impact since it's used as a current driver. If you try another auto-zero amps ('2057 et. al.) you have to give serious consideration to noise generated especially in the rest of you circuit - let alone the current noise on the inputs. I have never seen any other amp work as good at LT1013 in the long haul.
If you call LT applications engineering, they will tell you exactly the same thing, and they've seen a lot fo LTZ circuits also. This is good advise.
KEEP ALL SURFACE MOUNT ceramic caps AWAY from any LTZ circuit. They are microphonic and pick up all sorts of mechanical noise. Use high quality poly film caps or better, and Thru Hole works better over long term, especially if your PCB is going into an un-stable temperature area.
3. Adding slots to your PCB may hurt you - I've made boards with and without 'thermal' slots, and never noticed any huge difference either way. We had one design with slots that all it did was pick up vibration from a vacuum pump running about 20 feet away on the production line - and it showed up on the Vref output as noise. Make the board compact and thermally uniform. Keep drafts away, and don't run a serious LTZ out in the open. Pay attention to about every Linear Tech and HP Schematic that shows star grounding techniques.
4. If your ambient temperature is stable, then something like cheaper PTF56, 5ppm TC resistors work just as well in the end for your LTZ as $80+ Vishay Magical Voodoo resistors. Spending a lot of money on the Vishay Magicals will generally reduce back pain since your wallet is a lot lighter, but that's about it. You will not get 40 X more performance out of an LTZ made with expensive resistors, guaranteed - I've never seen that happen, not even close, and I've compared several different assembly methods and parts BOM's over time.
5. PWW 3ppm resistors work fine especially if your ambient temperature is changeable. I have never seen a Vishay Magical Voodoo work any better in an LTZ circuit than PWW. Remember, the whole point of LTZ is that the resistor Drift RVal is attenuated by over 100 to 1 - and this is conservative. You won't see much difference in the Vref if it was made with 1ppm or 3ppm resistors if your ambient temp is changing even several degrees. The heater resistor ratio is most important to keep an eye on to make sure stays stable AS A RATIO. The other resistor values are not too critical as long as they are stable...so don't waste money on tight tolerance resistors if you don't need them.
6. Don't be afraid of good quality PWW trimmer pots to make adjustments in a VBoost circuit. In fact we've got several 3456a that go in for calibration, and they are still in cal after decades with the pots never been touched since the 80's. You can do the same with LTZ boost circuits IF your final circuit is thermally uniform and doesn't see a lot of vibration.
7. I would NOT recommend ever changing out the heater resistor in a perfectly running 3458a. I have seen people do that, and all that happens is now they have a driftier 3458a, because the problem wasn't in the heater resistor. Know what you're modifying! The ambient temp inside a 3458a is not your lab temperature, and these instruments are designed to run on racks alongside other warm equipment! I wouldn't try putting a 12.5k heater resistor into a 3458a unless you had a very well ventilated area.
8. STAY AWAY FROM eBAY LTZ's. That's only going to end in tears. If you take a used LTZ from an unknown source - and who knows how it's been mis-treated while it was demounted - and you solder it in to a new circuit: Your still looking at a multi-year burn-in cycle to see how it's really going to perform over time. Every time you get a soldering iron anywhere close to an LTZ chip - expect to see at least a year or two to return to stability. Yes, it really takes that long - You can try other thermal cycle techniques to convince yourself you can speed up the process - and then you realize it took 18 to 24 months before you see real stability again.
With that type of time commitment involved, I suggest that it's never worth your time to save a few bucks on unknown, used garbage from eBay. It will never save you a dime.
We have always purchased direct from LT, and I can't think of a single time over hundreds of circuits that we've gotten a bad, out of spec LTZ. Maybe once, but that was an obvious dead transistor we saw right on power up - but LT happily replaced it.
Some LTZ's will be better performers of course - but generally we've never been steered wrong buying direct from the source. Seen enough problems from eBay garbage to last a lifetime, so we will never purchase there.
Final Suggestion: If you are running profitable business and need a good, solid, low drift 10V Voltage Reference, stay with a Fluke 731 / 732 series and keep it cal'd. You won't be able to roll your own and save any money if your time is worth anything.
Don't get me wrong - Of course playing and chasing PPM for entertainment purposes is always fun too! I've just been around this long enough that chasing the final PPM or two drift is basically the point of vanishing returns. In terms of profit margin - that last few PPM is not going to ever pay back except in the most critical applications.