Disclaimer: I don't know what I'm doing and I'm doing this for fun. Please keep in mind this is my first metrology project.
I've been working towards building an adjustable DC reference (my goal is "6.5 digit performance": lowest range 1-0v in 1uV resolution) and am in need of a voltage reference board for that purpose. My first thoughts are either LTZ1000, LM399, or MAX6226. LTZ1000 is off the table because as of yet I have not contracted volt-nutieutus. I rejected the LM399 because of the high noise compared to the MAX6226. The issue with the MAX6226 is the high temperature coefficient. I thought I'd rectify this by ovenizing it myself. What could go wrong? Side note, thank god for free samples. They make projects with fun expensive chips much more affordable.
Here's the schematic I came up with, pretty simple. Note that the heater setpoint is derived from the reference voltage with a trimmer. This is bad for two reasons: If the reference had a positive temperature coefficient (it doesn't luckily) a thermal runaway could occur. Second, the trimmer is poor in terms of stability.
I thermally bonded the reference, heater, and temperature sensor together by stacking them on top of eachother. This was very hard, and in the future I will not do this unless it works very well.
I 3D printed a cover to thermally insulate the reference and to squish the reference sandwich together. A piece of foam is inserted into the cover to aid in squishing. No thermal compound was applied because I didn't have any on hand.
With lots of wrangling involving total re-assembly I got it to work. (edit: note that the flux was cleaned off later)
I built a long term test setup to monitor the voltage. Thanks to my university's lab manager for allowing me to steal this multimeter and corner of the lab for this.
I have a raspberry Pi logging voltage data coming from the multimeter and temperature data coming from the oven and ambient temperatures. All of the data is uploaded to the cloud so I can monitor it remotely. I plan to leave it going for at least 1000 hours to stabilize.
This was a great learning experience in soldering, experiment design, and datalogging. Any suggestions in how to improve my absolutely terrible test setup would be great.