Aluminum PCB for Thermally Challenging Design?

[meshtron]

I'm working on a product where I have a chip that could dissipate up to 9W.  And I'm trying to get it IP68 ingress rated...

My current prototypes aren't well sealed up and the (BGA pin-style) 28x28x10mm heat sink is adhesive taped to the top of the DRV8144 chip (which is what's generating heat).  Transfer from the chip to the heat sink is okay (though the small chip + large heat sink isn't ideal, but nothing I can do there) but the heat sink is undersized by at least half.  It runs fine at ~5W dissipation but somewhere around 7W (in 30C ambient) it can't keep up and overtemps.

I'm considering switching to make the board as a 2-layer aluminum board with this kind of stack up:


That would let me put all my components on one side, then put the heat sink on the opposite side attached directly to the aluminum board.  That would turn the board into a heat spreader and (hopefully) reduce overall thermal resistance between the heat pad on the chip and the heat sink.  It actually also solves some other ingress-related issues for me too.  I'd probably also solder-mask the bottom side (in the pic above) and have a large "pad" where the heat sink is going to attach (with adhesive + mechanical retainer).

I just can't convince myself that passing heat from the bottom (solder) side of the IC all the way THROUGH the board to the heat sink is actually less resistance (or about the same) as what I have now.  Any thoughts from thsoe of you dealing with heat and heat sinks more than I have (I'm new to all of this)?

EDIT: Here's a video of the current 2-layer, 2oz copper (both sides), lots of stitching board starting up with a 10A DC load so ~4.5W dissipation. The bright part on the left is the linear regulator that should've been (and now is) a buck converter.  The DRV is just under the front edge of the heat sink.  Also, attached a picture of the current board for reference.

[Benta]

Use a ceramic substrate instead of a PCB (or rather: as a PCB). Done all the time.
Example: IGBT power modules.
Excellent isolation and heat transfer to your heatsink.

[meshtron]

Hmm - I don't know anything about ceramic substrates - I will do some digging.  Thank you!

[Benta]

Most common is Alumina (Al₂O₃).
Used mostly in power, HV or RF applications.
Earlier very common as substrate in hybrid modules.

[meshtron]

So far I've used PCBWay and JLCPCB to fab my boards, looks like neither of them offer ceramic.

The thermal conductivity sounds appealing, but the brittleness might be an issue - this is a relatively high-vibration environment (think automotive).  Looks like copper substrate IS available and has an even higher thermal conductivity.  So, maybe I'll try that.  2-layer (with similar stackup as above) copper isn't that much more expensive than 2-layer aluminum.

[PCB.Wiz]

I'm working on a product where I have a chip that could dissipate up to 9W.  And I'm trying to get it IP68 ingress rated...

9W sounds tough from such a small SMD package.
You may need both top and bottom heat spreaders ?
Or, buy a better part with lower power loss ?

[David Hess]

9W sounds tough from such a small SMD package.
You may need both top and bottom heat spreaders ?
Or, buy a better part with lower power loss ?

The datasheet says 16.7 C/W top thermal resistance and 6.4 C/W bottom thermal resistance, making 9 watts feasible through the bottom with an aluminum board, but not the top no matter what heat sink is used.

[meshtron]

9W sounds tough from such a small SMD package.
You may need both top and bottom heat spreaders ?
Or, buy a better part with lower power loss ?

I think the part is decent - at least not worse than others I found doing similar work for me.  I'm just asking a lot of it.  I started this phase of design tweaking wondering if perhaps having two DRV chips in parallel would be beneficial.  But, since Rdson is linear-ish with both current and temperature, I'd still have 9W to deal with so thought it didn't matter.  BUT - I didn't take into account it would also give me double the heat source footprint on the heat sink, that might be enough reason to do it alone.

The datasheet says 16.7 C/W top thermal resistance and 6.4 C/W bottom thermal resistance, making 9 watts feasible through the bottom with an aluminum board, but not the top no matter what heat sink is used.

Great point!  I think it's worth trying the aluminum "back side" heat sink placement, unfortunately a bit of an expensive experiment.  I did realize, though, that I can't use the stackup I originally proposed (because any PTH wouldn't be electrically isolated) so I'll have to switch to this one.  Perhaps a little more thermal resistance but should still spread well.  And, if I need to, I could go to copper core for a bit more money which not only has way less thermal resistance but also they'll fab with basically exposed (and leveled) substrate pads so I could essentially attach the heat sink directly to the substrate.  I'd rather have the mechanical properties of aluminum if I can get away with it, but it's another option.

[peter-h]

Be aware that thermal cycling is the biggest killer of electronics. So you need to keep the temp variations minimal.

[meshtron]

Be aware that thermal cycling is the biggest killer of electronics. So you need to keep the temp variations minimal.

Does thermal cycling bother the chips, or the board, or?  I know what you're saying is correct, I just don't know the various mechanisms that cause problems.  I think for this application I'm just going to have thermal cycling, it's inevitable.  But everything I can do to keep the high temperatures lower (extracting more heat) reduces that delta so should help.  Also when looking at the ceramic substrate mentioned by Benta, I saw multiple references to very low coefficient of thermal expansion - it's easy for me to imagine how higher CoE would cause some long-term issues too.

I think I will lay out the next board so I can put two DRV chips in parallel if I want, but it will work fine with one.  That should let me do some good experiments to see if we're making progress.

[peter-h]

For stuff running hot, I would use off-PCB devices, and mount them on high grade (aluminium oxide) washers, onto an aluminium case. I've just built some one-offs in this way



Be careful to not introduce stresses and vibration modes however, between the PCB and the device package. Use wires like above. There are other ways to solve that e.g. PCB mounted power device and a metal plate under it which extends off-PCB and bolts to the case, so that assembly forms a part of the way the PCB is fixed in the box.

For IP68 you want a stiff box with an o-ring seal all around, special connectors, etc. Good fun but not exactly cheap. Done it many times...

You don't want to be dissipating 9W into a PCB.

For a really high grade job use metal devices e.g. TO3, not TO220 like I have above (but I am dissipating about 1W).

[langwadt]

afaict most copper/alu boards are single layer only

why not thinner FR4 board with more copper layers ?

[meshtron]

afaict most copper/alu boards are single layer only

why not thinner FR4 board with more copper layers ?

PCBWay will do 2 (and I think 4) copper layers on an aluminum or copper board.  But really I'm trying to minimize the thermal resistance between the top of the board (where the chip will be generating heat) and the bottom of the board where I'd attach the heat sink.  Spreading the heat through the board is part of that, but getting it to dissipate into (and away from) the heat sink is the real goal.

A thinner 4-layer FR4 might indeed do that, but I'm not savvy enough with all the math to really know.  Having a low thermal-resistance substrate makes intuitive sense to me as a mechanism to improve heat transfer.  But maybe with enough vias, a thinner FR4 board would do the same.

[mtwieg]

Might be helpful to do some simple back-of-the envelope math.

FR4 thermal conductivity is around 0.3W/(m-k). The area of the full chip is 3x6mm, and the FR4 thickness is 0.15mm. So that sliver of FR4 is going to have a thermal conductivity of 0.3*(0.006*0.003/0.00015) = 0.036 W/k, or 27.8 C/W. That's already too much to support 9W (even before adding the additional 6.4 C/W from the package itself). If you go for a stackup with two layers on the same side, that doubles the dielectric thickness as well.

Unless the stackup supports vias connecting the copper layers directly to the aluminum substrate, I don't think the aluminum substrate will help much.

[David Hess]

Might be helpful to do some simple back-of-the envelope math.

FR4 thermal conductivity is around 0.3W/(m-k). The area of the full chip is 3x6mm, and the FR4 thickness is 0.15mm. So that sliver of FR4 is going to have a thermal conductivity of 0.3*(0.006*0.003/0.00015) = 0.036 W/k, or 27.8 C/W. That's already too much to support 9W (even before adding the additional 6.4 C/W from the package itself). If you go for a stackup with two layers on the same side, that doubles the dielectric thickness as well.

Unless the stackup supports vias connecting the copper layers directly to the aluminum substrate, I don't think the aluminum substrate will help much.

I suspect Texas Instruments is expecting the addition of copper vias at the pads to take advantage of the package's bottom thermal resistance, as would be done with a package that has a thermal pad on the bottom.

It would not hurt to add a heat sink to the top as well.  Every little bit helps.

[meshtron]

Might be helpful to do some simple back-of-the envelope math.

FR4 thermal conductivity is around 0.3W/(m-k). The area of the full chip is 3x6mm, and the FR4 thickness is 0.15mm. So that sliver of FR4 is going to have a thermal conductivity of 0.3*(0.006*0.003/0.00015) = 0.036 W/k, or 27.8 C/W. That's already too much to support 9W (even before adding the additional 6.4 C/W from the package itself). If you go for a stackup with two layers on the same side, that doubles the dielectric thickness as well.

Unless the stackup supports vias connecting the copper layers directly to the aluminum substrate, I don't think the aluminum substrate will help much.

I'm still a little unclear as to whether or not that layer is actually FR4.  On the source page of the image I posted above, someone asked a question at the bottom about the type of dielectric and got this answer:


Going up (in thermal conductivity) to any of those values makes the back-of-napkin math more palatable.  There's not a good way to run vias through to the substrate on aluminum, but on the copper ones (which must have some different capability) they show a stackup like this:


In my case, this wouldn't help much since the package I'm using doesn't have a thermal pad, but it seems like there might be a creative and more direct way to move heat through the board itself.

EDIT: Also, there is indeed a HTSSOP version of the beefier DRV8145 that DOES have a thermal pad, so I might be able to take advantage of this straight shot from thermal pad to copper substrate.