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.
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.
Tim
-- Deep Friar: a very philosophical monk. Website:
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).
C> Guess that plain old copper clad will dissipate maybe 1W/in^2. =A0(What w=
as
ut
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.
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.
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?
Large pads would help, I think.
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?
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.
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.
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.
A quick search did not bring anything very special up, but I'll have to look some more.
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.
If you look at the manufacturers' details for surface mount power 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).
I put an LM20 (temperature sensor from NatSemi) on the board. I only fit it on the prototype.
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.
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.
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?
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.
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.
The modeling works, but of course experience is needed to get accurate enough results. The price you can get from Mentor ;)
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.