AD780 long term drift

Q 

I have questions about the AD780 long term drift spec need your help.
In the datasheet, the long term stability is +/-20ppm/1000hr (typical), and
there is note for it, says "The long-term stability specification is no
cumulative. The drift in subsequent 1,000 hour periods is significantly lower
than in the first 1,000 hour period."
My questions are:
1) what does the "subsequent 1,000 hour periods " mean in the datasheet. Does
it means, once the part worked cumulately more than 1000hr,  the drift will be
lower(better) than before? Or if the part have worked more than 1000Hr, then we
power down it, then power up it again, does the stability specification could
go back?  Hope I explain my question clearly :)
2) how do we test  the long term stability? And is the +/-20ppm/1000hr spec
just a statistical?

 

A 

Long term drift refers to the warm up time and operation of the part. So once a
part is powered up( in a static or dynamic way) and left alone, its output will
shift until it reaches its stability. The majority of the drift happens at the
early hours. it is usually expressed in ppm/1000 hours. Long term drift  is a
"random walk" phenomenon and increases with the square root of the elapsed
time. There are 8766 hours/year. Therefore 1 year figure will be sqrt(8.766) ≈
3 times the 1000 hour figure. ten year value will be roughly 9 times the 1000
hour value. In practice, things are a little better even than this, as devices
tend to stabilize with age. We usually power up couple of parts in the lab in a
controlled environment and let it sit there for 1000 hrs and collect data at
different time intervals. AN-713 for more details.