Just curious, but why did you choose the LM399? Back in its day it was just about the last word in voltage references, but that was a long time ago. Today references such as the ADR4550B and the LT6655B achieve the same 2 ppm/°C (max) temperature stability as the LM399 but without any internal heater; also, their noise voltage is about 100X lower than the LM399.
Good question, and I will try to give a equally good answer.
First let me preface with the following: The LM399 is my brassboard/predecessor design for my final design which will use the ovenized LTZ1000, which is considered the reference to end all references
, So I wanted to use a part with similar technology to learn lessons on a lower cost design as my eventual design will use high dollar resistors and the venerable but very expensive LTZ1000.
So with that said, I chose the LM399 based on tempco (.5ppm/degC) not 2ppm as you stated above, secondly, Long term stability(LTS) LM399 is (8ppm/√kH), compare ADR4550 (25ppm/√kH),. Note: Initial accuracy of LM399 actually sucks compared to other references, but LTS spec directed my decision to go with LM399.
As far as output noise, I plan on filtering the noise using a output buffer with filter, I have a reference design showing how to do this, and can post it if anyone wants it. What the heck, posted below
Some other pitfalls of LM399, Warm up time for LM399 is 3 seconds as opposed to 160 uS of ADR4550.
Other factors affecting choice was component package. From what I read, the output voltage of SMT packages are more sensitive to mechanical stress, and thermal gradients created by neighbor components. Lastly, I already purchased the LM399 parts, and created time consuming CAD symbols, (silly reason I know).
Well, that was my thinking at the time I went down this road.
P.S. I would like to model thermal behavior of my design with LTSpice, but I think I will need help with some items. I will follow up this post with some questions/issues that I encounter as I can't recall at the moment.