Daniel said:
How cool do they have to be keep to last 10 years?
The capacitor thing, uses a modified Arrhenius relationship. In
chemistry, Arrhenius predicts that chemistry reaction rate, doubles
for every 7C rise in temperature. For capacitors, they do a curve
fit, and the exponent isn't quite the same as the one in chemistry.
Capacitor reliability may double, for anywhere from 10C temperature
rise, to 15C temperature rise.
http://www.rubycon.co.jp/en/catalog/e_pdfs/aluminum/e_MBZ_MCZ.pdf
The data sheet has the basic endurance stated in it. In small print...
2000hrs at 105°C,
Say the temperature inside the computer case is 35C. From 105 down
to 35, gives us 70C to work with. (I'm ignoring core heating in
the cap, for simplicity.)
Just picking a number, if the lifetime doubles for every 10C, that
is 7 doublings or 2**7 times longer life, or 2000 * 2**7 = 256000 hours
or 29 years of operation.
If you pick a different number, like 15C for the modified Arrhenius factor,
you might be looking at 2000 * 2**5 = 64000 hours or 7.3 years of
continuous usage.
Each manufacturer will provide a web page or data somewhere, as to
the modified Arrhenius factor. (They do curve fitting to measured
data, to get it. You cannot assume every manufacturer gets the same
answer. Some of them will cheat and use the measurements of others.)
They mention 10C in this note, plus 15 year service life independent
of the Arrhenius calculation. The 15 year service life is based on the
rubber bung in the bottom of the cap, drying out and failing to maintain
the seal on the capacitor. Rubber is attacked by things like ozone, and
local conditions might cause the rubber to fail before that estimate.
(Consider how long a rubber band lasts in your environment, to get some
idea of what is happening to rubber.) If your Arrhenius calculation comes
out to 29 years, then the 15 year service life overrides that calculation.
And the service life isn't a "guaranteed fail", it would represent
some percentage of the caps in service, needing to be replaced.
http://www.rubycon.co.jp/en/products/alumi/pdf/Life.pdf
When electronic equipment is designed, "Service Life" is a factor
which is stated in the high level architecture documents. Where
I worked, at one time, that number might have been 20 years, while
later the number dropped to 10 years. A change in the Service Life
for a line of products, gives the engineer more room for cost cutting.
To meet 20 years, might take extensive component testing, usage of
more expensive components, or a complete change in implementation
(say a digital VRM with Volterra, rather than the usual Vcore
implementation). I have no idea what the "Service Life" target
for motherboards is. Presumably, it is longer than the 3 year warranty
But not by much.
If you can find similar documents for solid polymer caps, you could
go through a similar analysis, to see what they're buying you.
Paul
