PSU damages HDD

[Arno Wagner]

Hi,

just to document a problem I debugged today:

I have a Linux server that suddenly had unreadable areas
on its primary disk (a Maxtor DiamondMax 9 plus 200GB). Upon
closer inspection these turned out to be a total of 114
defect clusters (4 sectors each) in 4 longer bursts plus
some individual clusters. The rest of the disk works
perfectly fine. SMART status is not failed, but the
self-test fails with a read error.

After replacing the disk and copying the data, the
computer crashed and had problems getting a DHCP address
(it is statically assigned DHCP). Replacing the network
card did not change anything. The power levels were fine,
but a check with an oscilloscope recvealed about 400mVpp
of noise and additional 500mV negative spikes on the 12V
line with no load. Replacing the PSU solved the problem.

I strongly suspect that the original HDD is not to blame
and that the PSU caused the damage. This is the first
PSU I have that did exhibit normal average voltages,
but a high enough level of noise and power-spikes to
cause real trouble. I suspect I can salvage the disk
with a complete overwrite, but have not had the time
for that yet.

Arno

 

[TE Cheah]

| noise and power-spikes to cause real trouble

I had 2 Quantum hdd 's IC`s damaged by voltage spikes when self
-assembled fans' wires came loose & sparks jumped. [ii] I feed +12
v to a 5cm fan to suck out heat from my HP Ll925 ( LCD ) monitor.
So I added 16000 & 11500 µF to my +5 & +12v rails, to damp these
2 voltages : then no more IC gets damaged by voltage spike,
whatever comes loose.

 

[Arno Wagner]

| noise and power-spikes to cause real trouble

I had 2 Quantum hdd 's IC`s damaged by voltage spikes when self
-assembled fans' wires came loose & sparks jumped.

Ok, _tht_ is not surprising ;-)

[ii] I feed +12
v to a 5cm fan to suck out heat from my HP Ll925 ( LCD ) monitor.
So I added 16000 & 11500 µF to my +5 & +12v rails, to damp these
2 voltages : then no more IC gets damaged by voltage spike,
whatever comes loose.

Yes. But you might blow up the rectifier in a PSU with that.
Unlikely, but possible.

Arno

 

[Impmon]

Yes. But you might blow up the rectifier in a PSU with that.
Unlikely, but possible.

Not likely unless cap failed and even then failed cap tended to
develop a short so the PSU would either shut down or blow a fuse.

 

[Arno Wagner]

Not likely unless cap failed and even then failed cap tended to
develop a short so the PSU would either shut down or blow a fuse.

Actually that is not the mechanism. The added caps could increase
the load current on start-up into the output filter of the PSU too
much.

Arno

 

[Jim P Sharma]

Actually that is not the mechanism. The added caps could increase
the load current on start-up into the output filter of the PSU too much.

Nope, the surge rating of all diodes is MUCH more than the static current
allowed.

 

[Arno Wagner]

Nope, the surge rating of all diodes is MUCH more than the static current
allowed.

Who talks about static current?

Arno

 

[Jim P Sharma]

Who talks about static current?

The designers of the power supply. Thats what determins which
diode is used on the output of the power supply for each rail and
the surge current rating you get is much higher than the static
rating with any semiconductor diode. So the size of the filter
caps isnt limited by what the diode can supply, its determined
by other factors, whats needed to filter adequately.

 

[Arno Wagner]

The designers of the power supply. Thats what determins which
diode is used on the output of the power supply for each rail and
the surge current rating you get is much higher than the static
rating with any semiconductor diode. So the size of the filter
caps isnt limited by what the diode can supply, its determined
by other factors, whats needed to filter adequately.

It is not quite that simple. True, rectifier diodes have higher
impulse current ratings, than static ratings. But the impulse
rating drops off fast when the pulse gets longer. If you (say)
triple the output filter capacity, then you get a three times
as long initial high load current when the PSU starts.

Arno

 

[Jim P Sharma]

It is not quite that simple.

Fraid so.

True, rectifier diodes have higher impulse current ratings,
than static ratings. But the impulse rating drops off fast
when the pulse gets longer. If you (say) triple the output
filter capacity, then you get a three times as long initial
high load current when the PSU starts.

Its much more complicated than that. The initial current to
charge the filter caps is determined by what the switching
regulator delivers to those diodes. That cannot be higher
than the mains end of the power supply can deliver, so
the filter caps dont in fact have any effect at all on the
current thru the final diodes on each rail, a larger filter
cap just means that it takes longer at the SAME
current thru the final diodes to end up fully charged.

And that just affect the time till the power
supply puts up the PWR GOOD line.

 

[Arno Wagner]

Fraid so.

Its much more complicated than that. The initial current to
charge the filter caps is determined by what the switching
regulator delivers to those diodes. That cannot be higher
than the mains end of the power supply can deliver, so
the filter caps dont in fact have any effect at all on the
current thru the final diodes on each rail, a larger filter
cap just means that it takes longer at the SAME
current thru the final diodes to end up fully charged.

Yes, that is what I just said. Have a look into an actual rectifier
diode datasheet to understand why that is a potential problem.

As I also said, it is unlikely to be a problem in practice, since
the disodes are usually far overdimensioned anyways.

Arno

 

[Jim P Sharma]

Yes, that is what I just said.

No it isnt. You said something completely
different and you are just plain wrong.

Have a look into an actual rectifier diode datasheet
to understand why that is a potential problem.

Doesnt help. THE CURRENT THRU THE OUTPUT DIODE
DOESNT INCREASE WITH BIGGER OUTPUT FILTER CAP
BECAUSE THAT CURRENT IS LIMITED BY WHAT THE
REGULATOR CAN PROVIDE.

As I also said, it is unlikely to be a problem in practice,
since the disodes are usually far overdimensioned anyways.

They arent on the static current rating.

 

[Dave (from the UK)]

So I added 16000 & 11500 µF to my +5 & +12v rails, to damp these
2 voltages : then no more IC gets damaged by voltage spike,
whatever comes loose.

That reminds me of the time when I did my MSc which was a lot of computer
modeling. I wanted the PC to run 24/7, but drops in mains voltage (probably
power supply pushed to its limits too), meant the machine would often reboot.

I solved it by adding external capacitance. In my case not across the 5 and 12
V, but on the 350 V DC immediately after the mains rectifier. The capacitors
were in an external box. I forget what I had, but there was sufficient to run
for several seconds with no main input. The CPU was either a 386 or perhaps a
486, but nothing newer.

But as for voltage "spikes" large electrolytic capacitors are not much use. Such
capacitors are far from ideal at high frequencies and have very low
self-resonate frequencies. Whereas school physics lessons tells you the
reactance decreases with increasing frequency, in practice the reactance can
rise so they are basically open-circuit at a few kKz.

I don't actually believe the 400 mv pk-pk noise, which is roughly 80 mV RMS
would cause damage.


--
Dave K MCSE.

MCSE = Minefield Consultant and Solitaire Expert.

Please note my email address changes periodically to avoid spam.
It is always of the form: month-year@domain. Hitting reply will work
for a couple of months only. Later set it manually.

 

[Arno Wagner]

[...]

But as for voltage "spikes" large electrolytic capacitors are not
much use. Such capacitors are far from ideal at high frequencies and
have very low self-resonate frequencies. Whereas school physics
lessons tells you the reactance decreases with increasing frequency,
in practice the reactance can rise so they are basically
open-circuit at a few kKz.
Agreed.

I don't actually believe the 400 mv pk-pk noise, which is roughly 80
mV RMS would cause damage.

I think it is boderline for crashes. But I also had these negative
spikes. And the numbers I posted were on a completely unloaded system,
with CPU and disk working this could get far worse.

Arno

 

[Arno Wagner]

No it isnt. You said something completely
different and you are just plain wrong.

Doesnt help. THE CURRENT THRU THE OUTPUT DIODE
DOESNT INCREASE WITH BIGGER OUTPUT FILTER CAP
BECAUSE THAT CURRENT IS LIMITED BY WHAT THE
REGULATOR CAN PROVIDE.

Yes, but it lasts longer. Allowable diode peak currents
are dependent on the time they are applied.

Arno

 

[Jim P Sharma]

Yes, but it lasts longer.

Irrelevant when that current is well within the STATIC rating of that
diode.

Allowable diode peak currents are dependent on the time they are applied.

Yes, but we arent talking about peak currents in this case, the initial
charge current of the filter caps is determined by what the regulator
can supply and that wont exceed the rating of that output diode.

 

[Jim P Sharma]

Dave (from the UK) <[email protected]>
wrote

I think it is boderline for crashes. But I also had these negative
spikes. And the numbers I posted were on a completely unloaded
system, with CPU and disk working this could get far worse.

Dunno, they may not have even been there with a loaded system.

 

[TE Cheah]

| > in practice the reactance can rise so they are basically
| > open-circuit at a few kKz.
| Agreed.
Crap. How can capacitive reactance rise with frequency ?

| > 400 mv pk-pk noise, which is roughly 80 mV RMS
400mv pp is 200mv peak, RMS is 200 ÷ sq rt of 2 = 141.42 mv.

 

[Dave (from the UK)]

| > in practice the reactance can rise so they are basically
| > open-circuit at a few kKz.
| Agreed.
Crap. How can capacitive reactance rise with frequency ?

It is not ****, but a fact.

Capacitive reactance does increase with frequency - the reactive of the
'capacitors' do. They are not perfect capacitors - in fact, they are very far
from perfect.

If you look on circuit diagrams, you often find 100 uF in parallel with 0.1 uF
used for decoupling.

Do you think that is because the designer wants 100.1 uF, then uses a capacitor
with a 20% tolerance? No, he/she does it because the 100 uF cap is useless at
high frequencies, so the 0.1 uF is more effective at decoupling at higher
frequencies than one where the value marked on the case is 1000 x bigger.

| > 400 mv pk-pk noise, which is roughly 80 mV RMS
400mv pp is 200mv peak, RMS is 200 ÷ sq rt of 2 = 141.42 mv.

For a sine wave yes, but not noise. If you assume the noise is uniformily
distributed, the peak is rarely more than 2.5 to 3.0 x the standard deviation -
see a normal distribution. 63% of the time it is within 1 SD. It is within 3
SD's about 99% of the time.

So for peak to peak, it will rarely exceeed 5 to 6 x the standard deviation.
That is why I used the word 'roughly' - there is no way to exactly map
peak-to-peak to rms in the case of random noise.


--
Dave K MCSE.

MCSE = Minefield Consultant and Solitaire Expert.

Please note my email address changes periodically to avoid spam.
It is always of the form: month-year@domain. Hitting reply will work
for a couple of months only. Later set it manually.

 

[Arno Wagner]

| > in practice the reactance can rise so they are basically
| > open-circuit at a few kKz.
| Agreed.
Crap. How can capacitive reactance rise with frequency ?

They have pretty high inductivity. A few kHz is maybe
too low for low-esr, but from personal observations
ar 250kHz 10uF ceramic filter as good or better as
1000uF low-ESR electrolyte.

| > 400 mv pk-pk noise, which is roughly 80 mV RMS
400mv pp is 200mv peak, RMS is 200 ÷ sq rt of 2 = 141.42 mv.

For noise RMS is irrelevant, really. PP is what you want.

Arno