J
JohnH
Arno Wagner said:Exactly.
and based on what I see coming through for recovery.
With the emphasys on "see".
This is also supported by semiconductor quality today. Except for some
known cases (the last one was the problem with new flame retardant
that killed a series of Fujitsu drives several years back),
semiconductors reliability is quite predictable.
However they have a limited lifetime
'Limited' as in decades.
and that derates with the operating temperature. Rule of thumb is half
the lifetime for each 10C more.
(Applies to logic and memory.
As if memory is no logic.
Power semiconductors are less sensitive.)
And guess what produces heat the most?
The base point seems to have shifted up in the last decades, at the
moment it possibly is around 50 years at 50C. There is also
catastrophic failure temperature at 150...200C chip temperature,
depending on technology.
So for 7 years lifetime you can turn the heat up to 80C and
leave it on 24/7. Switch it off at nights and you can go to 90C.
Today, if a chip works for some months or years and then fails, it is
allmost allways due to temperature.
Or just old age in combination with a production flaw.
Example: Run a chip at 90C (chip, not package surface temperature),
and get something like 1.9 years of lifetime.
You gotta love Swiss math:
50y 50C
25y 60C
12.5y 70C
6.75y 80C
3.4y 90C
1.7y 100C
(Everything works differently in Swissland apparently, even math)
In addition chips have "hot spots", were the local temperature is
significantly higher than the overall chip temperature. Especially
pocessor lik structures suffer from this, which includes controller
chips on disks and chipsets.
In short, if a semiconductor fails prematurely, but after working
for some time,
it was in most cases insufficiently cooled
Because of a design or production flaw in the chip.