I'm in the process of redesigning a few products around here which currently use a 50ppm/C reference (MCP1525). Long-term aging is not specified (other than a chart showing a +-20ppm aging occurs in the first couple hundred hours after power on). Of course, any redesign is an excuse to look at the parts to determine if we can do better. Plus we've gotten some feedback from customers who would prefer we improve our accuracy over temperature.
So, what I am looking for:
This is a 3.3V design with no additional voltages available. The reference needs to output a voltage somewhere below that. 2.5V seems to work well. A stable Vref/2 output is also a big bonus cause otherwise i'm stuck doing a voltage divider there which means more resistor ppm errors (unless I spend more $ on a matched tempco resistor divider). Of course, if the combination of a voltage ref+appropriate divider is less expensive, I'm definitely open to that as well.
The challenge in this application is that the circuit will be exposed to +-40C swings from the cal point. (-20C to +60C operational range). We are measuring battery array voltages which are +-60V and want to be accurate to +-0.1V (I apologize if I just used the wrong term for accuracy vs precision). If I did my math correctly, I figure I have at most ~1000ppm to work with, and probably a fair bit less than that. A 50ppm reference * 40C is 2000ppm. So obviously I need something better if I want to meet this desired spec - probably something in single digit PPM/C range once one accounts for other component temperature drifts.
This is in a solar-powered application so power consumption is important. This along with the extreme temperature range and cost seems to eliminate most/all ovenized solutions.
Since this is going to be calibrated, I'm not fussy about the initial voltage accuracy - as long as it's in a usable range. A 2.5V+-10% initial accuracy reference with low aging/temperature ppm is no problem at all.
I've found TI's REF2025 which is a 8ppm absolute + 6ppm Vref/2 tracking voltage reference which looks pretty good. It's ~1.75USD@1K which is tolerable if I absolutely had to, but I'd rather find something < $1 at that quantity.
I'm hoping some of the volt-nuts on here can possibly point me toward some parts or other design strategies I've missed.