[volt-nuts] LM399 Long term drift specification

[Tony]

I've just noticed that TI and Linear's specs for 'Long Term Stability' 
(typical) are different. TI state 20ppm/1000Hr while Linear state 
8ppm/SQRT(kHr). That's  a big difference - is this likely to be a real 
difference or just specmanship?

I note that Linear (in Note 4) also state that "Devices with maximum 
guaranteed long-term stability of 20ppm/SQRT(kH) are available." 
Presumably they would be a special order as there doesn't appear to be a 
unique part no. Would they be likely to be much more expensive?

Then on page 4 Linear show a graph of long term performance of 44 units 
(rather cheekily starting the graph at 2 months or approx 1500 hours!). 
To reproduce something approaching the mean curve using the formulae 
(drift ppm/SQRT(kHr)) x SQRT(month * 24 * 30.5/1000), requires me to use 
2.2ppm/SQRT(kHr). That is way less than the typical 8.5ppm value.

To get a curve that resembles the 3-sigma curve requires a value of 
5.7ppm/SQRT(kHr) which is still better than that 8ppm typical figure.

I'm not sure how to interpret this; what value would you use if you were 
designing a reference that isn't going to be re-calibrated after the 
initial calibration and you don't intend to burn in for several months?

Assuming the equipment is expected to have a 15 year life, operating in 
a range of 0 to 40C, what maximum total drift would you be comfortable 
specifying? I'd prefer it to be less than 100ppm, but that would require 
a drift of < 9ppm/SQRT(kHr), but that assumes that the SQRT(KHr) drift 
characteristic is valid for periods much longer than 12 months.

Are there any other references, at similiar or lower cost, that could be 
reasonably guaranteed to have a total drift of < 100ppm after 15years?

Is it reasonable to assume that there are some types of voltage 
reference will always drift, albeit noisily, in one direction allowing 
the original calibration to be offset to some extent to reduce the 
maximum error over its lifetime?

Having looked at several application notes and lots of datasheets, in 
those that include graphs of drift over 1kHrs or so of several 'typical' 
examples, I have not been able to see any meaningful correlation between 
the specified typical 1k drift figures and the graphs. Eg. in Linear's 
Design Note 229 (Don't Be Fooled By Voltage Reference Long-Term Drift 
and Hysteresis" the graphs of drift for the LT1461S8 and the LT1790SOT23 
show very different drift after 1600 hours - in the range 50 to 130 for 
the former and approx -5 to +45 for the latter, yet the LT1461 is spec'd 
at 60ppm/SQRT(1kHr) and the LT17910 at 50ppm.

I realise that I would probably need to contact the manufacturers for 
real answers but its been my experience that they aren't often 
interested if you're not buying large volumes, and I know that a lot of 
people here have a lot of experience in this area.

Thanks,
   Tony H