9
\$\begingroup\$

I have a chip in which I have selected this footprint: Package_SO: SOIC-8_3.9X4.9mm_P1.27mm. This is what it looks like:

enter image description here

Because this chip gets pretty hot, I thought I should keep the footprint but with vias for thermal relief. Something like this:

enter image description here

However, the chip is all plastic. There is no metal on the bottom to be soldered. Does it make sense to make a change like this? Will it help reduce the temperature?

Or because it cannot be soldered on the bottom, it will have no effect at all?

EDIT: This is a picture of the surrounding layout: enter image description here

EDIT2: This is the layout for the whole portion that gets heated up: enter image description here

EDIT 3: Following are images of the schematic topology, as well as a proposed layout for this boost converter.

enter image description here

enter image description here

EDIT 4: This is my attempt in copying user's user1850479 layout. enter image description here

\$\endgroup\$
14
  • 2
    \$\begingroup\$ That looks like a switching converter. Is it actually getting hot enough that this is a concern? And if so, how much of that heat is from the adjacent diode and inductor? Might make more sense to look at adding more thermal vias/planes to the area generally as heat will be coming from several closely packed components. \$\endgroup\$ Sep 18 at 1:50
  • 1
    \$\begingroup\$ Show more of your layout. I have my doubts that this is actually the part getting hot; I suspect nearby parts are heating it instead, as user1850479 suggests. \$\endgroup\$
    – Hearth
    Sep 18 at 2:41
  • \$\begingroup\$ It is true that the adjacent coil gets really hot as well. When I get back at home later, I will post pictures from the layout. \$\endgroup\$ Sep 18 at 7:31
  • \$\begingroup\$ Actually, I just updated the question with a pic of the layout. \$\endgroup\$ Sep 18 at 8:27
  • 1
    \$\begingroup\$ That layout is not good for a switching converter. Can you post the whole thing, including the diode to capacitor hot loop? I suspect fixing the board will help a lot. \$\endgroup\$ Sep 19 at 0:22

5 Answers 5

16
\$\begingroup\$

The effect will be minimal without anything more thermal conductive than air between the chip and that thermal pad. What you can do is to increase the width of the traces, that will work way better.

\$\endgroup\$
11
\$\begingroup\$

As mentioned you’d have to have grease or another thermal interface material to make this work. There should be some glues or epoxies that are compatible with a production process. You could also glue a small Heatsink to the top.

Also maximize heat flow through the Vcc and GND pins. Tie them well thermally to the planes/pours and you’ll be able to decrease temps a little.

But they do add those power pads for a reason. You’ll want to change packages if you have significant dissipation.

\$\endgroup\$
3
  • 1
    \$\begingroup\$ Thank you! Could you please explain what you mean with this Also maximize heat flow through the Vcc and GND pins. Tie them well thermally to the planes/pours and you’ll be able to decrease temps a little? \$\endgroup\$ Sep 18 at 7:31
  • 2
    \$\begingroup\$ i.stack.imgur.com/YMWPe.png Pin 1 and 8 are regular trace width, the others have more width, and GND pin (4) also Vcc (6) have largest width. GND is also connected through vias to a ground plane that will further spread the heat. So use wide traces from the pins a bit, those will work as a heatsink. Also heed user1850479s advice about hot components around the IC. \$\endgroup\$
    – MiNiMe
    Sep 18 at 7:59
  • 1
    \$\begingroup\$ Thank you. I updated the question with a pic of the surrounding layout. \$\endgroup\$ Sep 18 at 8:28
8
\$\begingroup\$

If you cannot change to a different package, you'll get best heat conduction by having a large copper area for each pin:

Example layout

If the package had exposed metal pad, that would have a direct connection to the chip. Then it would make sense to just connect that to a big ground plane and have normal traces for other pins.

But the side pins connect to the chip through thin bond wires. Heat will have to transfer through the plastic enclosure to the pins, and then to the PCB. When each individual pin is a poor heat conductor, you want to use them all in parallel to make the best out of a poor situation.

Thermal conductive glue between the chip and PCB would be a good idea, but usually comes at an extra manufacturing cost. If you add the glue, adding ground plane vias under the chip is a good idea.

\$\endgroup\$
6
  • \$\begingroup\$ Thank you! For the last paragraph, this stands correct even if the bottom of the chip is just plastic? \$\endgroup\$ Sep 18 at 8:31
  • 1
    \$\begingroup\$ @user1584421 Yes, glue for plastic. If it was metal, you wouldn't need glue, it would get soldered in standard process. \$\endgroup\$
    – jpa
    Sep 18 at 9:06
  • \$\begingroup\$ Even with a metal bottom/case, we wouldn't solder that base to the thermal relief vias. Makes rework more complicated. A thermal pad is our preferred approach. \$\endgroup\$
    – SteveSh
    Sep 18 at 11:48
  • \$\begingroup\$ @SteveSh Yeah, every time you improve heat conduction, rework gets harder. Depends on exact situation what is more important. \$\endgroup\$
    – jpa
    Sep 18 at 13:39
  • \$\begingroup\$ It depends on the chip; there are some leaded devices that have one or more leads connected to the die bond pad, they don't all use wirebonds for all leads. This is common for SOIC-8 and SOT-23-6 (maybe also SOT-23-3) MOSFETs, as I understand it. \$\endgroup\$
    – Hearth
    Sep 18 at 14:32
5
\$\begingroup\$

I think the problem is actually the enormous high current switching loop you have routed through your board. It is so large it does not fit in your picture, but here is the part of it I can see:

enter image description here

Depending on how far it goes to the right of the board before meeting back up and closing the loop it is actually a little surprising the converter works at all.

Here is a better layout for that converter:

enter image description here

By putting the diode and MLCC capacitor right up against the switch, current only flows a few millimeters and all on one layer, so it encounters only a small amount of resistance and inductance which leads to lower losses and better performance. Furthermore that distance is mostly connected with wide planes that have lower resistance/inductance. Note that depending on the switching frequency you may also want to put that big electrolytic capacitor in too, but you can parallel the MLCC to keep the loop area small and put the electrolytic slightly further out.

When I built that converter like this the chip did not get particularly hot.

Edit: If you wanted to try and salvage the existing board, you could try adding an MLCC and a (short!) jumper wire here:

enter image description here

That would reduce the loop to the green area. You could test that and see if the temperature is reduced.

\$\endgroup\$
12
  • 1
    \$\begingroup\$ The 2nd layout is indeed how it should be done. The 1st one had no chance at working well. \$\endgroup\$ Sep 19 at 15:25
  • \$\begingroup\$ First of all, I cannot thank you enough for taking the time to not only answer my question, but experimenting with this by yourself. My gratitude is immense! Secondly, this is the third board that I design, so terms like "high current switching loop" are not only unknown for me, but I don't even know where you learn these things.... It makes sense that I messed this up, as my skill is not even close to being a good PCB designer. One last thing. I am not sure if this is the correct site to suggest this, but I am willing to pay you to make a shared screen review of my board/help me with the \$\endgroup\$ Sep 19 at 21:10
  • \$\begingroup\$ layout. If you are not interested, please excuse my offer. Otherwise, I can make a deposit (paypal?) and we can sit for 30-60 mins to examine the board, if you are up to it. Whatever you decide, I am extremely grateful for your help! Take care! \$\endgroup\$ Sep 19 at 21:11
  • 1
    \$\begingroup\$ @user1584421 I believe we both used the same IC, so you could use my layout exactly, or at least very similar. If you want to do your own layout, draw the circle I did between the switch pins, the output capacitor and the diode and see if they form a tight circle. If they don't, try to improve it until they do. It's only three components, so not too hard to put them into a triangle. \$\endgroup\$ Sep 20 at 14:28
  • 1
    \$\begingroup\$ @user1584421 I paralleled two input capacitors and two output capacitors to hit the capacitance I wanted, so all four of those are capacitors. You can ignore that and use your electrolytics instead. \$\endgroup\$ Sep 20 at 17:42
1
\$\begingroup\$

Does it make sense to make a change like this? Will it help reduce the temperature?

Or because it cannot be soldered on the bottom, it will have no effect at all?

It won't if the package doesn't have a pad directly on the bottom, there will still be air and plastic between the die and the PCB and neither of those will be great a transferring heat.

If you want to get rid of the air, use thermal heatsink compound or a thermal glue on the bottom of the chip. But the PCB isn't the best place to dump heat, because it simply heats up the PCB. The best place is to dump it in a heatsink and the heat ends up in the air (If you dump heat in a pcb it just heats up the PCB and then the PCB heats up the air, but you end up with a hot PCB). Even a small piece of metal can make a significant difference in transfering heat to the air. Even better would be a small heatsink.

The best thing to do for a DC DC switcher would be to reduce the current or make sure the design is correct.

\$\endgroup\$
1
  • \$\begingroup\$ Many applications don't have the availability of free air for cooling, and have to conduct heat through the PCB to a suitable heat sink. Military and satellite applications fall in this category. \$\endgroup\$
    – SteveSh
    Sep 19 at 0:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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