Maker Pro
Maker Pro

Decoupling caps and cold failures

O

Oppie

Jan 1, 1970
0
We've all known about the importance of good grounding and bypass caps but
this was a strange one.

I have a design based on the Analog Devices ADuC841, which has worked very
well for some time. In a recent production lot, we started seeing failures
at room temperature. Digging further, we had many failures at temperatures
of 5C. These were soft failures where you'd get erratic operation and then
have to reset the board after it had warmed up for it to work normally.
Device was rated to -40C and an external clock was applied so weren't even
close to design limit.
Was all set to write it off as bad chips but a later spin of the board
assembled by another vendor had a similar problem. Turned out to be entirely
due to the decoupling caps being too far from the chip. In one case, there
was even a via which further increased the impedance. Put an extra cap right
at the main processor pins and problem gone.
Just didn't expect that the problem would manifest itself when cold.
 
O

Oppie

Jan 1, 1970
0
Phil Hobbs said:
High-K ceramics have horrible tempcos. It was probably changing the
waveform of the ringing at the VDD pins, so that the marginal situation
became a failure.

It's pretty poor engineering, letting something basic like that get
through. Didn't anybody think to check the supply noise at the uC? I
hope nobody's life depends on the units you already have in the field.

Understood <Hangs his head in shame>
It was a design I inherited. I do remember reviewing the original design and
remarking that the decoupling on one channel was too far from the chip and
was told as much, to shut up... Difficult to respond to that when the chief
designer is the company president. I wanted to do a four layer board but got
shot down so it was done in two layers.
No- it's not life support related and low energy.
 
M

miso

Jan 1, 1970
0
We've all known about the importance of good grounding and bypass caps
but this was a strange one.

I have a design based on the Analog Devices ADuC841, which has worked
very well for some time. In a recent production lot, we started seeing
failures at room temperature. Digging further, we had many failures at
temperatures of 5C. These were soft failures where you'd get erratic
operation and then have to reset the board after it had warmed up for it
to work normally. Device was rated to -40C and an external clock was
applied so weren't even close to design limit.
Was all set to write it off as bad chips but a later spin of the board
assembled by another vendor had a similar problem. Turned out to be
entirely due to the decoupling caps being too far from the chip. In one
case, there was even a via which further increased the impedance. Put an
extra cap right at the main processor pins and problem gone.
Just didn't expect that the problem would manifest itself when cold.

Besides the cap tempco, the chip itself is probably noisier at low temp
due to the mobility tempco. Basically faster rise and fall times. Five
deg C isn't all that cold though.

If you make a chip that touchy, you get in a special category known as
the jerk whose chip requires bypass caps on the contactor rather than
DUT board. [If you need bypass on the contactor, then you have a
dedicated contactor per chip.] It happens, but you really try not to
need exceptional bypass since invariably some less than optimal boards
are going to be produced.

You did well to find the problem yourself.
 
M

miso

Jan 1, 1970
0
It does not matter if I am a test engineer. What is your opinion
about the best way to ensure that a critical design is fielded with
the least risk?

BIST perhaps?
 
M

MarkK

Jan 1, 1970
0
it's probably not the change of the capacitor characteristics with temp, but
probably a change in the chip characteristics with temp.

logic chips can go faster etc when cold and so may need better bypassing...

Mark
 
N

Nico Coesel

Jan 1, 1970
0
Oppie said:
Understood <Hangs his head in shame>
It was a design I inherited.

Inherited designs are the worst. I recently got bitten by an Attiny
somebody thought would work at 1.8V which it doesn't. Atmel is not
afraid of putting lies in their datasheets :-(
 
N

Nico Coesel

Jan 1, 1970
0
John Larkin said:
On Sat, 11 Feb 2012 13:38:58 -0800 (PST), brent


What is your opinion


Really good design, reviewed and checked by really good engineers.
It's hard to test quality into a product, especially firmware.

Perhaps, but kicking the crap out of a circuit helps to find flaws you
missed. Besides, putting a simple yet effective test rig together is
fun :)
 
O

Oppie

Jan 1, 1970
0
John Larkin said:
Did the board have power and ground planes, or was that stuff routed
as traces? Boards with solid ground and power planes seldom have
bypassing problems, and the distance between parts and caps doesn't
usually matter.

I had wanted to do a 4 layer with full ground plane and split plane for 3.3
and 5V but got shot down. Mandate was for 2 layer Done in Eagle with 3.3V
and 5V routed. Polygons top and bottom covering the whole board defined as
ground. When a Ratsnest is done with orphans off, it will attempt to flood
any free areas with copper plane. If it can't fill an open area because
traces are blocking it, drop in a few vias to bring in the ground from the
other side...
I still print out the board one layer per sheet when done in B size, put it
on the light table and look for where my power busses are actually winding
up. Helps to show any deficient connections. On the screen, try to show the
ground net and so much of the board lights up, it's actually difficult to
pinpoint problems. I have Alpha Blending ON to better see superimposed
layers but even so, it's difficult.

For troubleshooting, I used freeze spray just to get a ballpark temperature.
Was also good for local cooling. I have some plastic shields I made in the
shape of the chips in question so I could cool just one component or area at
a time. It was definitely the ADuC841 that was cold sensitive when the
decoupling impedance was deficient. Some lot codes were more sensitive than
others but at least they were bona fide and not counterfeit devices as I had
considered at one point.
 
J

John Devereux

Jan 1, 1970
0
John Larkin said:
Putting an ADuC841 on a 2-sided board is, well, brave. Flooding unused
areas on a 2-sided board doesn't help much. Tell whoever "shot it
down" to make it work.

I routinely use it's bigger brothers, the ADUC70xx series on two-layer
boards with no issues. I like to use one layer for groundplane (as
continuous as possible) and the top layer for routing. Re-allocate the
I/Os where possible to get an easier layout. And decouple
noise-generator and noise-susceptible parts locally with R/C or L/C
filters (depending on current and permissible voltage drop).

Thanks,
 

Similar threads

T
Replies
16
Views
1K
whit3rd
W
M
Replies
62
Views
6K
rickman
R
P
Replies
36
Views
3K
Frank Miles
F
Top