Well, you could encase the smd part in foam on it's adapter PCB and run the wires to the PCB. That will give you some isolation from shock and also from temperature changes and air currents. This may turn out to be a blessing in disguise.
No, cut some foam into a compartment and close the chip & PCB inside. That is good enough for ovenized x-tal oscillators. Use flexible leads into it. Then, mount the little encapsulated assy inside. I'm trying to suggest something that is mostly COTS with minimal messing around for you.
When I am matching pairs of transistors, a foam cap does a wonderful job of keeping the two parts isolated from the outside environment - and that isn't even sealed. The most important thing is to have freedom from air currents, which this idea will give you. As a side benefit, you get a sealed space that can be kept at whatever temperature the heater (or a heater) will be set at. So a quiet voltage reference, coupled with a heater if it doesn't have one, gives you a stable temperature with no air currents to disturb your reference. Some of the better references, like the LTZ1000 has.
Other references with a heater are the http://www.linear.com/product/LM129 family and the LM129 family. For your 5V reference, you might have to use a voltage divider or a buffer with gain with the reference.
Some of the temperature compensated references are getting quite good also. If you keep one of those at a set temperature, it would be extremely stable.
So, what are the main reasons why you needed the metal can? I think Linear Technology has made the new products equal or better in performance. Not unless you need the LTZ1000 level of performance, and then you should probably just buy those. Expensive, I know. Just think of us poor Canadians looking at the exchange rate.
Thanks Chris, that's a pretty good explanation. I think I'm getting it now.
The reference ends up being a box within a box, i.e., case. Like we
see when we start taking apart better built gear.
When we open it up, we see another enclosure. So it's okay and I guess
we should say mandatory that it doesn't have air currents/derating that circulate over the precision resistive and measurement components.
Will have to look into heaters, does a mini incandescent bulb fit the bill?
Thinking out loud here. Gotta run.
BTW - Please give an update on the new bench when you get the chance.
Yes, you have it. Yes, a light bulb makes a dandy heater. That's what they produce the most of, heat. Thinking aloud as well, maybe consider using the fuse type lamps. They are available in 6 and 12 volt types. The 12 volt ones are commonly 300 mA, which would make a really nice heater. Because the filament is stretched out straight, you get a larger heat radiating surface. A normal fuse clip, 1 1/4" by 1/4" size, will hold them fine. You want a distributed radiator so you avoid creating your own weather pattern inside that little box. Watch the temperature because some foam products will just fall apart with a little heat.
You could probably get away with sticking the lamp in with the chip and resistors so that everything is stable. You actually need some heat loss, but not too much. A single box should be all you need, lined with closed cell foam or made out of the stuff. If you use a cast Al box, surrounded with foam, you can also shield the circuit. Heat the box with the lamp and it will radiate evenly. Of course you could also use screw down resistors (to the case) and also mount the driver transistors. That way you get heat from everything. Free heat as this is what you actually want. You could also do what Agilent did with one of their ovenized 10 MHz reference oscillators. Just use transistors and use them as the heater. Mosfets might be better for this. Once you have a sensing element (varistor, diode, thermistor) near your circuit, use feedback to control the temperature. This takes some effort, but if you want a really high quality reference that is close to temperature immune, this is one way to do it. Make sure you limit the cold current to a safe level and reduce the heat output before it reached the set point. This will greatly reduce temperature oscillations. You want a proportional temperature controller, not a "bang - bang" type.