9

I want to use a linear regulator for some application, it's in a SOT-223 (not SOT-89) package. How do I keep it cool, preferably without a bulky heatsink? The regulator may be dissipating 2-3W of heat. I've heard that you can use copper traces under the regulator on the PCB to keep the regulator cool; does anyone have any references on this?

Thomas O
  • 31,546
  • 57
  • 182
  • 320

7 Answers7

14

You won't be able to dissipate that much heat with only copper traces to wick away the heat. (A SOT-89 is also a very small package, are you sure that specific part in that specific package is rated for 3W?)

I use D-Pak sized packages with lots of copper on the four layers and arrays of vias to try to give the device a lot of copper for a heat sink.

alt text

This works reasonably well for low duty cycle loads but doesn't work well for continuous load applications (there is high thermal resistance to the air). For high dissipation requirements you need fins and air moving over them, and unless you're building circuit boards differently than I know how to, you're going to need a heat sink to get those fins.

akohlsmith
  • 11,162
  • 1
  • 35
  • 62
  • Thanks for your answer, so it looks like linear is out for this. I'd be okay using a DPAK. I think I'm getting SOT-89 confused with another type, SOT-223, which is much bigger. – Thomas O Oct 31 '10 at 00:29
  • 1
    You have lots of vias for heatsinking.... but then you're thermally relieving the heatsink contact pad. Wut? – Connor Wolf Oct 31 '10 at 06:21
  • The vias are to connect the 4 layers of copper together as well as to add more metal. I thermally relieve the contact pad so I can actually solder the damn thing to the board. :-) The idea is to give it some mass for the heat to go but to keep a little bit of thermal resistance for actual manufacture. – akohlsmith Oct 31 '10 at 19:38
8

I do realize this isn't answering your question directly, but something you might want to consider.

Instead of dissipating so much energy, you can put a buck converter before your linear regulator. Get the buck to output at a voltage just above what is required by your linear regulator.

This will not only decrease the amount of heat you have to dissipate, but also improve the efficiency of your design.

As far as heat sinking goes, I tend to put several vias directly to my ground plane. The ground plane seems to be very good at dissipating heat. If you go to a 4+ layer board and have the ground plane internal then heat dissipation wont be nearly as good.

Kellenjb
  • 17,509
  • 5
  • 51
  • 87
  • 1
    I'm trying to avoid a buck, but I think I'll have to go with buck anyway. – Thomas O Oct 31 '10 at 17:52
  • 1
    I have avoid bucks for awhile, but I have come to realize that I just have to use them some times. They do add some complexities and cost, but they are well worth it when you are having to step down a large voltage or output lots of current. – Kellenjb Oct 31 '10 at 19:12
  • 1
    If you're using the buck regulator why do you want to keep the linear (which can be an LDO)? – stevenvh Jul 04 '12 at 17:53
  • 1
    I think I remember Thomas mentioning in other questions/answers/comments that he was concerned about the noise with bucks. I also think I remember mention of multi power rail applications... but this was 2 years ago. – Kellenjb Jul 05 '12 at 01:25
5

Its the laws of physics. You need to dissipate 3W through devices with large thermal resistance there will be a temperature rise. Using copper traces can take the heat away from surface mount devices into the printed circuit board. But that heat still needs to be sunk.

Looking at a SOT223 device, they have a Rj-a of 91 K/W which means at two to three watts a temperature of rise of 273 K can be expected. This will cook your device. The Rj-s (junction to solder point resistance) is 10 K/W so provided your board can dissipate the heat the device will be 30 K above ambient.

If your board is mounted in metal enclosure, you can with a bit of design effort, align the large thermal pads on the circuit board with islands on the metal enclosure.

        /---\                        hot device    
==================================   PCB
_______/     \______/    \______     Metal enclosure

Using large copper pads on each layer with lots of via will aid in transferring the heat. The only other issue is to clamp the circuit board to the metal enclosure and apply sufficient pressure and thermal compound so the board can conduct heat into the enclosure.

Doing this effectively transfers the heat from the component to the board and into the enclosure. So the enclosure effectively becomes the heatsink.

Without a heatsink on the board, you will reduce the Rj-a from 91 K/W to a lower value. What this value is, you will need to determine experimentally. Make a simple circuit board with the device in question on it, and thermal pads on each layer with vias, then ramp up the amount of power you are running through the device from less than one watt gently to two/three watts and using a thermocouople, record the temperature on the board and device. This will enable you to calculate the Rj-a of the device on your circuit board.

smashtastic
  • 3,336
  • 21
  • 31
3

As described in Low Profile Power Dissipation, you might be able to waste some heat in another component (upstream resistor or second regulator), so your regulator doesn't have to dissipate as much. You'll have to do the calculations for the min and max voltages and min and max loads you expect to see.

Community
  • 1
endolith
  • 28,494
  • 23
  • 117
  • 181
  • I've seen large resistors used to "drop" the voltage so the regulator doesn't have to get as hot. Ultimately, it's just moving the thermal dissipation to something else. – ajs410 Nov 01 '10 at 19:49
  • Yep, that's exactly what it's doing. But the percentage dissipated by each component varies depending on the load. The worst case for the regulator will occur when resistor and regulator are dissipating the same amount. – endolith Nov 01 '10 at 20:00
3

Yes you can cool the device using the board. Note that this takes a reasonably amount of surface area to do so. Don't expect that the whole of your board will give the component a cooling effect, for example if it's tab is on the ground plane. The only effective area is within 6cm's to 8cm's I believe.

The via's or small holes you usually see in those planes are thermal via's. On the other side of the board there probably is a copper plane as well. It increases the thermal cooling, but might be hard to make when you prototype your own boards inhouse. The holes can't be that big (in the order of a few tenth's of mm).

I've made a switching regulator the other day that needed some cooling as well. It was in a TO-263 casing, which is a little bit bigger. But anyway, the datasheet of national at page 4 and 5 specified that with 1 sq inch of copper area I had a cooling resistance of 26C/W. That's J-A, which isn't too bad. If you disspiate 3W that would add 75C above ambient, which is good enough. In this particular case I was making the PCB on an amateur etching machine, so I made the area twice as big because the soldering connection to the pab is harder to make.

Hans
  • 7,238
  • 1
  • 25
  • 37
  • Thermal vias increase thermal impedence - they are designed to make it easier to solder because they conduct less heat, right? – Thomas O Oct 31 '10 at 17:19
  • Thermal relief on vias increase thermal impedance. A bed of vias to the ground plane on the other side really decreases thermal conductance. – Kortuk Oct 31 '10 at 23:54
2

This can be a big brutal and I haven't done any thermal guesstimates on your requirements but one option if the physical size of a heatsink is a problem is to pot the board or device area with a compound that has a low thermal resistance. I have seen this done with plain old Araldite to spread the thermal load. If the potting is done inside a metal case then you have the benefit of the metalwork too. Mind you - this makes reworking a tad difficult!

Linker3000
  • 1,289
  • 9
  • 11
1

While researching the same question for a SOT-223 package switching transistor, I stumbled on the ON Semiconductor Soldering and Mounting Techniques Reference Manual (find it here: http://www.onsemi.com/pub_link/Collateral/SOLDERRM-D.PDF). This is a compiled set of articles on thermal and mounting considerations and includes dozens of footprints for common package types (including SOT-223). It also includes articles on how to prepare PCB heatsink mounts, thermal grease, and other techniques that I had not considered before. The document was recently revised, July 2014.

I found it worthwhile to look through.

M. Abel
  • 326
  • 2
  • 7