How much can a small SMD component really dissipate power
LM
to spread somewhere if I beliewe manufacturers, but in a full board
there is very little space to spread. Expesially if other side is full
too.
If I think about a simple case, there is a very small space around a
resistor to heat air above it. A small space because everywhere around
and under the part there are similar parts. What then is thermal
resistance of say square millimeter of FR4, or square inch if you want.
Does this make sense? A small component in the middle of similar parts,
all getting warm. The real life must be a mess.
Are there programs to simulate temperature of a full two sided board and
clever enough to be future proof.
IR cameras are one thing I can think to confirm how hot a part is.
Any other comments?
Regards
LM
Tim Williams
the empirical figure, something like 150 K*in^2/W in free air?)
Adding bumps to the board will increase its power dissipation slightly,
and cutting up the copper foil reduces heat spreading ability (leading to
hotspots).
So if you have 100 little SMTs in a square inch, dissipating a total of
1W, you'll be pretty much at ratings. That's an average of 10mW per, but
the statistical distribution is probably a "power" law (of a different
kind of power), where 10% of the transistors dissipate 90% of the power,
or whatever. Keep this in mind when sizing thermal pads.
Tim
--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms
"LM" <L...@somehost.somedomain> wrote in message
news:4ca8fb3c$0$32120$9b53...@news.fv.fi...
LM
1 watt per square inch. It is not much, I think there could be more
than 100 parts in that area. 150 kelvins hotter than ambient, seems
quite hot, easily over 200 centigrade (Celsius).
Continues below.
On 4 loka, 01:02, "Tim Williams" <tmoran...@charter.net> wrote:
> Guess that plain old copper clad will dissipate maybe 1W/in^2. (What was
> the empirical figure, something like 150 K*in^2/W in free air?)
>
> Adding bumps to the board will increase its power dissipation slightly,
> and cutting up the copper foil reduces heat spreading ability (leading to
> hotspots).
>
> So if you have 100 little SMTs in a square inch, dissipating a total of
> 1W, you'll be pretty much at ratings. That's an average of 10mW per, but
> the statistical distribution is probably a "power" law (of a different
> kind of power), where 10% of the transistors dissipate 90% of the power,
> or whatever. Keep this in mind when sizing thermal pads.
I had to start with a simple case to get any idea of this . It will
different in real life when there are different power dissipations and
different ratings.
Leif
John Larkin
Most surface-mount resistors are the same thickness, about 20 mil
alumina, and most have about the same aspect ratio, 2:1. So if you can
heat sink the end caps to PCB copper, an 0603 resistor can dissipate
about the same power as a 2010 for the same maximum hot-spot
temperature. Most surface mount part cooling is dominated by the
copper leads/pours, not the part itself. For example, a polyfuse trip
current changes radically as a function of lead sizes.
Lateral thermal conduction of 1 oz copper is about 70 K/w per square.
That can dominate some situations.
Thermal resistance of a copper patch to ambient is complex, vaguely
values like 50-100 K/w per square inch. As you note, adjacent parts
and other-layer copper planes can affect part temperature a lot. It's
indeed a mess.
John
Robert Baer
LM
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Mon, 04 Oct 2010 00:52:37 +0300, LM <L...@somehost.somedomain> wrote:
> >I ask because there are more and more full boards here. Heat is supposed
> >to spread somewhere if I beliewe manufacturers, but in a full board
> >there is very little space to spread. Expesially if other side is full
> >too.
>
> >If I think about a simple case, there is a very small space around a
> >resistor to heat air above it. A small space because everywhere around
> >and under the part there are similar parts. What then is thermal
> >resistance of say square millimeter of FR4, or square inch if you want.
> >Does this make sense? A small component in the middle of similar parts,
> >all getting warm. The real life must be a mess.
>
> >Are there programs to simulate temperature of a full two sided board and
> >clever enough to be future proof.
>
> >IR cameras are one thing I can think to confirm how hot a part is.
>
> >Any other comments?
>
> >Regards
>
> >LM
>
> Most surface-mount resistors are the same thickness, about 20 mil
> alumina, and most have about the same aspect ratio, 2:1. So if you can
> heat sink the end caps to PCB copper, an 0603 resistor can dissipate
> about the same power as a 2010 for the same maximum hot-spot
> temperature. Most surface mount part cooling is dominated by the
I think you are saying that as long as parts have equal area to spread
the heat, it does not matter how big the part body is?
>Lateral thermal conduction of 1 oz copper is about 70 K/w per square.
>That can dominate some situations.
Large pads would help, I think.
>..don't fogetto put it all inside a closed plastic box..
Hah, we going to put the cards in vacuum. A (little) more seriously, I
read that if there is empty space between the box and pcb, air will
move and spread the heat. But how do you check how hot it is inside a
box. Hire some miniature chinese with thermometers?
John Larkin
wrote:
Not exactly. The copper foil has a finite lateral spreading thermal
conductivity (the 70 K/w per square value) so small footprint parts
will get hotter, per watt, than bigger ones. The resistor example is
sort of a special case.
It would be interesting to do a few experiments here. There is a
National datasheet somewhere that has a bunch of examples of
various-shaped copper pours and their thetas. Maybe someone here
remembers which it is.
>>Lateral thermal conduction of 1 oz copper is about 70 K/w per square.
>>That can dominate some situations.
>Large pads would help, I think.
Yes. If lateral heat spreading becomes a limit, use copper pours on
multiple layers and stitch them together with flooded-over (no spoke)
vias, close to the part.
You can barely make out the vias in the copper pours here...
ftp://jjlarkin.lmi.net/Chimera.JPG
Vias under parts are great, if they don't hog too much solder paste.
>
>>..don't fogetto put it all inside a closed plastic box..
>Hah, we going to put the cards in vacuum. A (little) more seriously, I
>read that if there is empty space between the box and pcb, air will
>move and spread the heat. But how do you check how hot it is inside a
>box. Hire some miniature chinese with thermometers?
A thermal imager is a fabulous tool, and they are getting a little
closer to affordable lately.
ftp://jjlarkin.lmi.net/IR_0026.jpg
Otherwise, use some fine-wire or foil stick-on thermocouples, or build
thermal sensors (surface mount RTDs or thermistors, LM45s, like that)
into your design. You can not stuff them in production.
John
LM
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Mon, 4 Oct 2010 02:30:19 -0700 (PDT), LM <sala.n...@mail.com>
environment. But usefull still.
Very interesting. I have learned new things and perhaps got some
things confirmed. It looks like there is not a good thermal simulator
software.
One problem remains however, whether to top post or not.
Thank you all (so far)
Leif M
LM
> National datasheet somewhere that has a bunch of examples of
> various-shaped copper pours and their thetas. Maybe someone here
> remembers which it is.
to look some more.
>Not exactly. The copper foil has a finite lateral spreading thermal
>conductivity (the 70 K/w per square value) so small footprint parts
>will get hotter, per watt, than bigger ones. The resistor example is
>sort of a special case.
And if there is no copper but plain laminate, hotter still.
I saw a 'nice' resistor network modelling heat in one National
document, I hope I can find some/any other way to check temperatures.
Nemo
resistors, in the small print they say e.g. "for 1 watt dissipation, the
pads must be connected to 1 square cm of copper".
I believe you can buy surface mount heat sinks these days (ie, ones
which go on the board, to increase its effective surface area, not on
components).
> how do you check how hot it is inside a
> box. Hire some miniature chinese with thermometers?
I put an LM20 (temperature sensor from NatSemi) on the board. I only fit
it on the prototype.
> Hah, we going to put the cards in vacuum.
This is an interesting point, even though you are joking. Does anyone
know what factor to increase power rating by for resistors in vacuum?
I've had to do this a couple of times, and used a factor of 3 or 4, and
they survived - but I think they were only dissipating for a few seconds
every minute.
> It looks like there is not a good thermal simulator software.
I once worked at a company where we wanted to know what the average
temperature inside a box would be (for MTBF calculations). Someone
pulled a diagram out of a folder - a photocopy of a photocopy of an old
book - and said "well... we're dissipating an average of 100 watts in
that box. Its surface area is 1.3 square metres. So it'll be... um... 20
degrees hotter than ambient in there." This took all of 15 minutes to
gather the data and look at the graph. This was Too Easy, so the
managers bought a thermal CAD package and detailed an engineer to learn
how to use it, model the cards in the 3D space, and give them an answer
they could believe in. 1 week later he reports, "22 degrees". This is
suspiciously close to the non-credible first result, so they order a
full scale mock up (we don't have the cards yet, we had to make dummy
ones with resistor loads). This gives an answer of 20 degrees.
If anyone has a bit of web server room where we can upload that graph, I
have a JPG image of it, contact me here. It's always been spot on for me.
Jan Panteltje
<Ne...@nocannedmeatproducts.nosirree> wrote in
<jYqqo.18640$kL4....@newsfe29.ams2>:>
>If anyone has a bit of web server room where we can upload that graph, I
>have a JPG image of it, contact me here. It's always been spot on for me.
Email the graph to gr...@panteltje.com and I will put it on the ftp server.
Jan Panteltje
<Ne...@nocannedmeatproducts.nosirree> wrote in
<jYqqo.18640$kL4....@newsfe29.ams2>:>
>>If anyone has a bit of web server room where we can upload that graph, I
>>have a JPG image of it, contact me here. It's always been spot on for me.
>Email the graph to gr...@panteltje.com and I will put it on the ftp server.
I received the graph, it is available here:
ftp://panteltje.com/pub/Temperature_Rise.jpg
Interesting, even data for different altitudes!
Kim Enkovaara
> Very interesting. I have learned new things and perhaps got some
> things confirmed. It looks like there is not a good thermal simulator
> software.
There are, but they are not cheap, for example
http://www.mentor.com/products/mechanical/products/flotherm
--Kim
John Larkin
Not cheap, and they tend to have a huge learning curve. And they are
only as accurate as their input, which can be bad. Unless you do
thermal design full-time for IBM or something, it's usually faster and
cheaper, and certainly more accurate, to hack a physical model of your
situation (cardboard, duct tape, fans, heat sinks, copperclad) and
test it.
What does Flotherm cost? Is the air flow modeling really any good?
John
LM
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
A quick note
I found an Application Note which gave thermal resistance for a 0402
something like 800degrees/watt at worst, then I calculated the area
the resistor would take(150degrees/watt/in^2), with very little leads
and I got like 29000 degrees per watt. That is a bit of difference
between measured and calculated values.
Thank you for the Flowtherm link and info too.
Kim Enkovaara
> On Fri, 08 Oct 2010 09:41:10 +0300, Kim Enkovaara
>> http://www.mentor.com/products/mechanical/products/flotherm
>>
> only as accurate as their input, which can be bad. Unless you do
> thermal design full-time for IBM or something, it's usually faster and
> cheaper, and certainly more accurate, to hack a physical model of your
> situation (cardboard, duct tape, fans, heat sinks, copperclad) and
> test it.
The tools are not that difficult but need experience in creating the
models etc. and also in understanding the results. There are also
consultants doing these things, if there is no inhouse experience
(or licenses).
The problem with hw models is that what-if analysis is hard to do. For
very power dense designs tens of rounds of different placement scenarios
are needed before things look better. It's hard to simulate in head
how different heatsinks and other mechanical structures affect to the
airflow of other components, how much the pcb conducts heat, would
heatpipes help etc. Also testing what happens if different fans fail is
interesting simulation in telecom designs that need high availability.
In very dense designs also hw models are needed, but they can be built
from the few good results from the simulations.
> What does Flotherm cost? Is the air flow modeling really any good?
The modeling works, but of course experience is needed to get accurate
enough results. The price you can get from Mentor ;)
--Kim