2
\$\begingroup\$

I'm currently designing a PCB with a couple of fast electronic elements (600MHz ARM, DDR3 ram etc.) and I'm quite puzzled which layer should those elements be placed upon.

The product will be closed in not-so-airtight plastic enclosure, but airflow will probably be rather constrained.

One might imagine that air coming in contact with warm IC package should be directed upwards, which would help the heat to be dissipated by convection. If such IC would be placed upside down, on the bottom layer, the heated air would have "nowhere to go", and the element would become considerably hotter.

The question is, does this effect really exist? Even so, is it noticeable enough to considerably affect working temperature of such IC packages?

\$\endgroup\$
  • \$\begingroup\$ If your planes are chopped up, then heatflow and heat-spreading thru those copper sheets will be very poor, causing high temperatures. \$\endgroup\$ – analogsystemsrf Apr 16 at 17:09
  • \$\begingroup\$ Do you have the option of mounting the PCB vertically to aid convection? \$\endgroup\$ – Andrew Morton Apr 17 at 15:52
  • \$\begingroup\$ Can you include the datasheet of the part in question? \$\endgroup\$ – Voltage Spike Apr 17 at 17:27
0
\$\begingroup\$

Airflow is critical in most electronic cooling, but if your product can be destroyed by simply turning it over, it is badly designed.

The right way of doing it is by calculating the power dissipation, both component by component, and in total, accounting for all use-cases and safety margins, compare it with their dissipation capabilities, and add heat sinking where necessary (whether via radiators with or without forced airflow, copper pours, or more "aggressive" methods like Peltier elements or watercooling.

Component datasheet often give a lot of information in this regard, both for calculating the dissipated power and ways to sink it away from the component.

\$\endgroup\$
0
\$\begingroup\$

If you must chop up the various planes, then have overlapping regions (insulated, of course) for easy heatflow.

In this drawing, the various colors of rectangles are used to indicate two different layers of copper foil.

schematic

simulate this circuit – Schematic created using CircuitLab

The thermal resistance of epoxy-fiberglass (FireRating-4) is 200X higher (worse) than copper, but for THIN layers of FR-4 and with large overlap regions of copper, the overlaps become quite good paths for heat flow.

\$\endgroup\$
0
\$\begingroup\$

The IC package itself has no concept of "up" or "down" (although the pins will all be on one side!), so the question boils down (no pun intended) to how you remove the heat. Convection is a way to acheive air flow passively, and this movement removes the heat. Convection generates flow upwards (hot air rises), so you would arrange the components such that this air flow is impeded as little as possible.

It absolutely affects the temperature rise, but you would have to model that to get definitive answers. Datasheet temperature rises will indicate under what conditions their model is valid.

In general, your layout is always a compromise of optimum routing, optimum placement for connector access, optimum thermal management etc. Have a look at some reference designs to get an idea of how they place it and try not to deviate too far from that!

\$\endgroup\$
  • \$\begingroup\$ Convection definitely has a notion of up and down. Warm air rises. Open air above a heat source (here, the IC package) is much more effective at dissipating heat than open air below it. \$\endgroup\$ – Pete Becker Apr 16 at 12:38
  • \$\begingroup\$ @PeteBecker Get your point, slightly awkward wording. I was trying to say the IC package itself doesn't care if it's facing up or down, it's the airflow that matters. \$\endgroup\$ – awjlogan Apr 16 at 12:46
  • \$\begingroup\$ @StainlessSteelRat Hah, in my (increasingly limited) experience it just smells bad..! \$\endgroup\$ – awjlogan Apr 17 at 15:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.