Burn In Change in Characteristics

[rickman]

I have an audio I/O board that I test by looping back the analog I/Os
externally, stimulating the digital input and examining the digital
output. The functional test shows a -0.9 dB drop in gain at 20 Hz and a
-2.5 dB drop in gain at 20 kHz. After a 24 Hr burn in the 20 kHz gain
is the same within measurement error, but the 20 Hz gain is always down
to -1.0 dB.

I don't know what could be causing this. Do capacitors age have an
initial rapid aging of some sort? The low end response is controlled by
a couple of X5R caps on each channel. I can't think of anything else
that would affect the frequency response.

I suppose there could be a small effect due to a drift in the DC bias
point which is set by zener diodes. This would modify the capacitance
of the coupling caps a bit. Do zener diodes shift characteristics in
the initial 24 Hrs of operation? I guess I could measure the DC set
point on some before and after boards.

 

[rickman]

Le Tue, 14 May 2013 04:06:58 -0400, rickman a écrit:




High K ceramics (X5R/X7R,Y5V,Z5U,...) are notorious for exhibiting high
aging rates (counted in % per time decade: about 2% for X7R and worse to
much worse for other dieclectrics).

Aging is reset when you throw the cap above its Curie temeprature which
basically is what you do when soldering the cap.

So you are saying this is purely a function of time from exposure to the
Curie temperature? What I am seeing is happening at a normal operating
voltage not far above room temp. 24 hrs under power and *every* one of
the boards shifts the 20 Hz attenuation from -0.95 dB (give or take a
couple of 0.01 dB) to -1.05 dB (again about 0.02 magnitude variation).
I didn't even notice this for the first few hundred boards tested for
frequency response. But the current order is a lot larger than usual
and I'm staring at these numbers all day long. So I finally realized
how consistent the connection is to before and after burn in. It is
virtually a perfect correlation.

I don't think this is a matter of just time as the boards have been
sitting for varying amounts of time since they were soldered. But the
burn in time is pretty consistent since that is the bottle neck in
testing and after 23.5 hours they are outta there to make way for the
next batch.

 

[[email protected]]

Why not use COG/NPO for the "sensitive" areas?  They're not all that
expensive.

I say that something that means -1dB at 20Hz for audio isn't terrible
critical, maybe more it down to 10 to be sure

-Lasse

 

[George Herold]

I have an audio I/O board that I test by looping back the analog I/Os
externally, stimulating the digital input and examining the digital
output.  The functional test shows a -0.9 dB drop in gain at 20 Hz and a
-2.5 dB drop in gain at 20 kHz.  After a 24 Hr burn in the 20 kHz gain
is the same within measurement error, but the 20 Hz gain is always down
to -1.0 dB.

I don't know what could be causing this.  Do capacitors age have an
initial rapid aging of some sort?  The low end response is controlled by
a couple of X5R caps on each channel.  I can't think of anything else
that would affect the frequency response.

I suppose there could be a small effect due to a drift in the DC bias
point which is set by zener diodes.  This would modify the capacitance
of the coupling caps a bit.  Do zener diodes shift characteristics in
the initial 24 Hrs of operation?  I guess I could measure the DC set
point on some before and after boards.

How big a change in C is that? Could it just be a temperature thing.
Are the boards warmer after 24 hrs of burn-in? Or is the sign in the
wrong direction for that?

George H.

 

[[email protected]]

So you are saying this is purely a function of time from exposure to the
Curie temperature?  What I am seeing is happening at a normal operating
voltage not far above room temp.  24 hrs under power and *every* one of
the boards shifts the 20 Hz attenuation from -0.95 dB (give or take a
couple of 0.01 dB) to -1.05 dB (again about 0.02 magnitude variation).
I didn't even notice this for the first few hundred boards tested for
frequency response.  But the current order is a lot larger than usual
and I'm staring at these numbers all day long.  So I finally realized
how consistent the connection is to before and after burn in.  It is
virtually a perfect correlation.

I don't think this is a matter of just time as the boards have been
sitting for varying amounts of time since they were soldered.  But the
burn in time is pretty consistent since that is the bottle neck in
testing and after 23.5 hours they are outta there to make way for the
next batch.

I read it as the caps aging in use, and reset if you bring them up to
curie temperature

try resoldering the caps on a board that has been burned in and
see if it goes back to start

-Lasse

 

[Robert Baer]

So you are saying this is purely a function of time from exposure to the
Curie temperature? What I am seeing is happening at a normal operating
voltage not far above room temp. 24 hrs under power and *every* one of
the boards shifts the 20 Hz attenuation from -0.95 dB (give or take a
couple of 0.01 dB) to -1.05 dB (again about 0.02 magnitude variation). I
didn't even notice this for the first few hundred boards tested for
frequency response. But the current order is a lot larger than usual and
I'm staring at these numbers all day long. So I finally realized how
consistent the connection is to before and after burn in. It is
virtually a perfect correlation.

I don't think this is a matter of just time as the boards have been
sitting for varying amounts of time since they were soldered. But the
burn in time is pretty consistent since that is the bottle neck in
testing and after 23.5 hours they are outta there to make way for the
next batch.

I think that you will see different patterns of burn-in "drift" with
different manufacturers, and most certainly with different ceramics
(X7R, Z5U, X5R, Y5V, U5V, X7T, and the "gold standard" C0G).
Since your correlation is so good, perhaps a sample as small as 5-10
would be sufficient.

 

[John Devereux]

I have an audio I/O board that I test by looping back the analog I/Os
externally, stimulating the digital input and examining the digital
output. The functional test shows a -0.9 dB drop in gain at 20 Hz and
a -2.5 dB drop in gain at 20 kHz. After a 24 Hr burn in the 20 kHz
gain is the same within measurement error, but the 20 Hz gain is
always down to -1.0 dB.

I don't know what could be causing this. Do capacitors age have an
initial rapid aging of some sort? The low end response is controlled
by a couple of X5R caps on each channel. I can't think of anything
else that would affect the frequency response.

I suppose there could be a small effect due to a drift in the DC bias
point which is set by zener diodes. This would modify the capacitance
of the coupling caps a bit. Do zener diodes shift characteristics in
the initial 24 Hrs of operation? I guess I could measure the DC set
point on some before and after boards.

Is it just a thermal effect, is the board still warm from the burn-in?

 

[rickman]

I say that something that means -1dB at 20Hz for audio isn't terrible
critical, maybe more it down to 10 to be sure

This is not a problem at all. I just noticed that is was very
consistent. I thought this was just variation in the measurement until
I realize the correlation. No, the goal is just to be better than -3 dB
at 20 Hz and 20 kHz so the units are all fine. I'm just curious about
the effect.

BTW, the frequency limiting parts mostly are pushing the
size/capacitance tradeoff and there is no more room on the board for an
extra via much less another huge cap. I think the ones on the output
are 10 uF. Inboard there are 2.2 uF IIRC.

 

[Robert Baer]

This is not a problem at all. I just noticed that is was very
consistent. I thought this was just variation in the measurement until I
realize the correlation. No, the goal is just to be better than -3 dB at
20 Hz and 20 kHz so the units are all fine. I'm just curious about the
effect.

BTW, the frequency limiting parts mostly are pushing the
size/capacitance tradeoff and there is no more room on the board for an
extra via much less another huge cap. I think the ones on the output are
10 uF. Inboard there are 2.2 uF IIRC.

One can double or triple the capacitance,using the SAME capacitors
and the SAME board space very easily; stack them.
With a small change in board area, one can put the SMT caps on the
pads sideways and maybe get three in almost the same area.
Then stack another 3 above...
Any of the above tricks require NO extra via.

BTW, there is nothing wrong in putting a via in a pad - takes ZERO
extra board space.

 

[George Herold]

Le Wed, 15 May 2013 03:11:33 -0400, rickman a écrit:







For the part I was telling, yes. You reset the cap by soldering it, then
let it rest at room temperature and cap value slowly decrease.
Say 2% per decade for X7R.
That's why manufacturers (at least the serious ones) have some appnotes
saying you have to wait before checking caps (in circuit testers for ex.)
after they've been soldered in case you need to do so.


That is 0.1dB, or 1%. This might come from either temperature elevation or
another possibility is that, if your caps are DC biased, that under bias
there's also a slight and _slow_ disaccomodation (takes minutes to tens
of minutes to settle).

Did you changed the caps ref (smaller package)? The bias disaccomodation
depends on the electric field value, so the smaller the cap the bigger it
changes under the same bias conditions.


--
Thanks,
Fred.- Hide quoted text -

- Show quoted text -

Yeah, more data! does the voltage change? What about T?
Say is there enough room an X7R cap?

George H.

 

[rickman]

Le Wed, 15 May 2013 03:11:33 -0400, rickman a écrit:


For the part I was telling, yes. You reset the cap by soldering it, then
let it rest at room temperature and cap value slowly decrease.
Say 2% per decade for X7R.

What does that mean, 2% per decade? Decade of what?

That's why manufacturers (at least the serious ones) have some appnotes
saying you have to wait before checking caps (in circuit testers for ex.)
after they've been soldered in case you need to do so.


That is 0.1dB, or 1%. This might come from either temperature elevation or
another possibility is that, if your caps are DC biased, that under bias
there's also a slight and _slow_ disaccomodation (takes minutes to tens
of minutes to settle).

There is no reason to expect a temperature difference. The boards are
pulled from a chassis where they aren't really warm anyway, and sit as
they are tested one at a time. This takes about an hour and there is
virtually no difference from the first to the last. I expect this is
definitely related to a part being under voltage. Obviously the caps
are the most likely, but I don't know much about such a subtle effect,
so I'm asking here.

Did you changed the caps ref (smaller package)? The bias disaccomodation
depends on the electric field value, so the smaller the cap the bigger it
changes under the same bias conditions.

Changed relative to what? The only factor I see is the time under
voltage. Most of the caps on this board are as small as I can get them.

 

[rickman]

Yeah, more data! does the voltage change? What about T?
Say is there enough room an X7R cap?

What about T? There is only minimal T fluctuation. The boards are
always at room temp, even in the burn in chassis, they are fan cooled
and not very warm to begin with, so the ICs just don't get warm enough
to even feel.

What voltage are you talking about? I don't see any reason for a
voltage to change other than possibly the zeners that set the bias point
of the amps. I think I mentioned them in my first post. I don't have
time right now to mess with it. I'm trying to get product out the door.
I was just curious if anyone knew of such an effect in caps or the
zeners.

 

[rickman]

How big a change in C is that? Could it just be a temperature thing.
Are the boards warmer after 24 hrs of burn-in? Or is the sign in the
wrong direction for that?

I have no reason to think there is any difference in temperature in the
two measurements. It takes me an hour to test a batch of board after
being under voltage. So any temp difference would show as the effect
wearing off with time out of the burn in chassis. This is not elevated
temperature burn in and the chassis is fan cooled being designed for
much higher power boards.

I'm starting to wonder if the zeners are drifting while powered, 6.2
volt units.

 

[rickman]

Is it just a thermal effect, is the board still warm from the burn-in?

The "board" is 51 daughter boards of fairly low power consumption. The
burn in chassis is a production card rack with fans. So I doubt you
could measure 1 degree difference from off to powered on. Then it takes
half an hour to swap out the daughter cards and the testing takes an
hour as each one is tested separately, one at a time. So there would be
lots of time to cool down and a drift in the effect would be seen.