Heatsinking a power component

Please, DO reinvent the wheel. Heatsink tabs and surface-mount leads can be combined if the heatsink is fixed securely to the PC board, and (sometimes, at least) the assembly can make it through the soldering phase of manufacture with all connections good.

But, half the solutions put the heatsink in firm contact with the PC board ( the heatsink area that contacts the circuit board is wasted because there's no air circulating there), and the other half use standup heatsinks whose mechanical and electrical connections are the first place a repair/rework technician is going to check. They fail. A lot.

Best, probably, is a heatsink that clamps over the power components after assembly/soldering. I don't see those, except in laptop computers (all custom heatpipe and plate gizmos, NOT a good off-the-shelf standard).

I recall Larkin's been soldering copper flags on things for this purpose.

You know those spherical diodes you often see soldered to heatsinks? Get some of those heatsinks. They're either tin plated steel or copper, obviously not aluminum since they're solderable. Find a convienient mounting combination (through hole / surface mount?) and go with that. Bonus points if you come up with a useful product that Digikey wants to stock. ;-)

Tim

-- Deep Friar: a very philosophical monk. Website:

Please, DO reinvent the wheel. Heatsink tabs and surface-mount leads can be combined if the heatsink is fixed securely to the PC board, and (sometimes, at least) the assembly can make it through the soldering phase of manufacture with all connections good.

But, half the solutions put the heatsink in firm contact with the PC board ( the heatsink area that contacts the circuit board is wasted because there's no air circulating there), and the other half use standup heatsinks whose mechanical and electrical connections are the first place a repair/rework technician is going to check. They fail. A lot.

Best, probably, is a heatsink that clamps over the power components after assembly/soldering. I don't see those, except in laptop computers (all custom heatpipe and plate gizmos, NOT a good off-the-shelf standard).

Well, normally you don't and rather go back to TO220. Just had to do that on a design. All transistors are DPAK except one. Better safe than the PCB turning brown after a few years. Remember the cheap TV sets from the 60's that had the H-flyback driver tube in a socket on phenolic, happily dissipating around 20 watts plus filament power? Some wise-crack inventor thought that was the cats meouw and then service techs had to resolder the traces to this socket every few years. If it was really bad you had to solder wires over the traces. I don't want to see my clients go there.

And don't let anyone tell you that FR-4 can't burn. It can and I have seen the aftermath of it (not my design).

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

AAvid or Wakefield has SMT heatsinks for these devices.

Cheers

That was an act of desperation!

We're designing a gadget now that uses a power opamp in a 7-pin DPAK sort of package. It will dissipate about 2 watts max, so we're using a surface-mount heat sink which straddles the opamp, plus a lot of copper in thermal contact with the chip base, topside, and with thermal vias to more copper on the bottom.

But I'd be reluctant to try to get tens of watts out of any dpak type of part. The thermal resistance of 1 oz copper foil is about 70K/w per square, so you've almost lost the battle before you get to the heat sink.

John

Get some bergquist gappad-1500 and mount D2PAKS on the bottom of the PCBA. Use the PCBA to squeeze the gappad down between the heatsink/chassis. This is about the best thermal contact you can get without going to extreme measures.

You can calculate the termal resistance of the gappad interface from the charts on the datasheet.

Alternately you could mount the devices on the top and put a bunch of thermal vias to the bottom and mount the heatsink directly to the board with some thin insulation and heat sink goop.

I have some charred FR4 that backs you up.

I have some photos of a case that could have resulted in serious havoc. But I am not at liberty to share. It was impressive, in some areas the only matter left was the fibers. Copper traces, resins, all gone up in thick black smoke.

This is why I am super-careful with DPAK and other PCB-heatsinkers. One must not only consider normal operation but also what happens in case of erroneous operation. A TO220 goes phut ... *BANG* and that's pretty much all there is to it. A DPAK could burn down a building.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

am not at liberty to share. It was impressive, in

gone up in thick black smoke.

not only consider normal operation but also what

that's pretty much all there is to it. A DPAK could burn

A few low-value resistors at strategic places, if you can afford the current drop, are always a good idea. They burn out very fast, too.

I've seen ceramic SMD cap (RF blocking cap) burn and turn FR4 around it into a black charred conductive area that had to be scratched out to break the short. Plenty of smoke.

M

.

Get

Thanks to everybody... I mentioned only tip of the iceberg. The hot part is IRFS3206. I have proto PCB designed for D2PAK and MOSFETS were blowing up until I kludged up TO220 package with the heat sink there (I made TO220 packages out of D2PAK before ordering right part) After some search and comparison I figured out that Infineon IPB030N08N3 G and especially BSC028N06LS3 G should run much cooler (I hope that they do meet the specs, there is no other way to find out but try them). The parts come in SMT package only. My plan is to make a PCB with the hole for the Cu or Al slug (attached to the heat sink on the opposite side of the PCB) pressed against the tab. I do not visualize it clearly yet, but, hey, I am an EE with the whiteboard/ eraser combo for mechanical design tool. Some pictures of existing designs might help. Somebody mentioned that laptop people do it all the time. Sounds interesting, but I do not have a laptop that I can pry open. :o(

Name a SMD that can produce 10s of watts of heat?

Aluminum core PCB might be just the ticket.

I guess a TO-268 package MOSFET could do that if you ran it at something like 10% of max.

The two square chips upper-right are Hittite 20 GHz distributed amps, which have some insane power density. They have power pads on their bottoms, soldered to many-via copper pours on the pcb top, to pours on the bottom, where we bolt heat sinks.

ftp://jjlarkin.lmi.net/AmpTop.jpg

If you plop the DPAK onto, say, a square inch of topside copper, you could shotgun that with lots of biggish vias and conduct the heat to a bottomside pour, and attach a heat sink there.

I'd suggest a ring of thermal vias close to the dpak, and another ring near the rim of the pour, and maybe even more in-between.

Figure very roughly 50 K/W for a 25 mil via, less for bigger diameters, assuming 1 oz thru-hole plating and no solder inside. The plating thickness is the big unknown... it could easily be half that.

20 vias, say 32 mil dia, might get you around 2 K/w top to bottom, say 4 to be conservative about plating and geometry issues. Add the heatsink theta. 10 watts starts to get scary.

An additional inner-layer pour would help spread the heat laterally, if you're going multilayer.

TO220 sounds safer.

John

ose.

=A0Get

, t .

to

Of course it is. How about SuperSO8 that has 0.9K/W junction to case

fa4c607f8c)? Infineon has really nice parts, all I like are SMT. :o(

seen it done like that in an electric powersteering controller for a Toyota, made by Omron I think

-Lasse

I once had to repair a PCB where a tranny had gone bad and burned a hole clean through the board. I cleaned up the edges of the hole, and just globbed some epoxy into the hole and bypassed the traces. Even at $30.00/hr (our shop rate in 1979) it's cheaper than a new board.

This was the place where there was a sign on the wall:

Our Shop Rates: Regular.............$30.00 If you watch........$40.00 If you help.........$60.00 If your kid helps..$100.00

Cheers! Rich

Define for how many milliseconds you'd want those watts :-)

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

folderId=db3a30431ddc9372011ebafa04517f8b&fileId=db3a30431ddc9372011ebafa4c607f8c)?

Junction to case doesn't really help you here. Ok, you can spec 2oz or more of copper and such but at the end of the day all this thermal energy needs some place to go.

So far I have always found a FET that suited the application and comes in TO220. Even if not 100% perfect it's better than a DPAK or D2PAK running red-hot where your MTBF goes to pots. Broaden the scope to other mfgs like ST, ON, IRF et cetera.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

folderId=db3a30431ddc9372011ebafa04517f8b&fileId=db3a30431ddc9372011ebafa4c607f8c)?

Silicon is cheap; heatsinks and fires are expensive. It's sometimes better to use two or four cheaper (lower power or Rds_on) dpak fets and spread the heat out.

John