LFpak/D2pak thermal limits question?

[3roomlab]

hi folks, i have been digging around on element14 with my free time. hunting around for some good parts MOSFETS mostly. and i noticed that there are alot of very nicely spec mosfets that are D2pak/LFpak packages. on a number of application notes, i approximated that on a good 4 layer board with vias, the most you could get out of any D2pak/LFpak is about 3 watts of dissipation --> is my estimations correct? (based on many application notes that state a roughly best case scenario of 40Kelvin/watt).

i saw that there are some specialized heatsinks which can be soldered around the packages to reduce this to about 32kelvin/watt. which will give about 1 extra watt of headroom.

i have an idea (see pic). what if we solder a brass base + fins (1mm thick? 2mm thick? definately beats the 2oz copper clad yes?) for extra heatsinking? i assume it will increase the thermal footprint (is that the correct word?) the package sits on. the thermal spread will also go beyond just the 1 inch square that many app notes talk about as a theoretical maximum heatpad that it could sit on, maybe with this, the heat spread pad can now be 2inch x 2inch and possible thermal load permissible can be increased to 10/20 watts? or it does not work because it is a package limitation ultimately (even though some spec sheets say that the device can dissipate over 200watts)

has anyone tried this? or is this even something feasible?

edit -- for spec that specify over 200watts of usable power envelope for that component, how does 1 even use up to 100watts of the envelope if there is no way to go even beyond 10watts on a SMD PCB? for example if we are switching a 20A load, if on steady DC state with a mosfet RDS on of say 0.1ohm, this package will dissipate about 40watts (am i right?), which means the mosfet will most likely pop off the PCB ... ouch

[3roomlab]

or even multiple soldered fins?

[lgbeno]

You can draw the equivalent thermal circuit and fill in numbers to see how much better it will be.  My guess is that you are on the right track by reducing the thermal resistances from the LFpack to the heat sink however bending those leads as such might be a problem.  I think most important thermal things to consider are increasing surface area of heat sink (more fins) the other significant factor is airflow, a fan blowing on the sink is going to have much more effect than fancy materials or assembly technique

[SeanB]

You can do the 200W if you are using a ceramic PCB with a internal heat spreading layer, and keeping the case surface at 25C. In real life any copper further away than about 2 inches does not add much extra cooling, you need to go to a 4 layer or more board with a lot of thermal vias under the pad, along with a large rear surface area, all in the thickest copper you can get the PCB made in, to increase much beyond 3W. Just a function of the thin sheet of copper transferring heat out to a large surface. If you mill a hole in the PCB and have a heat slug of copper that is soldered to the back of the device and attached to a larger heatsink then you can get higher power, but for a plain PCB 3W is about it.

[mariush]

You mean heatsinks like these ?

http://www.digikey.com/product-search/en?pv357=264&pv357=290&pv357=13&pv357=272&pv357=261&pv357=271&FV=fff40012%2Cfff80068&mnonly=0&newproducts=0&ColumnSort=0&page=1&quantity=0&ptm=0&fid=0&pageSize=25

[digsys]

I have been doing a LOT of 15-25A power switching / regulation the last few weeks and can pass on a few findings -
#1 !! Use the LOWEST Rdson you can find. If you're working between 40-60V (or lower), you can find NFETs with 0.55 - 2.5mR !!
That's a massive saving in heat !! I also use 4oz Copper Top and Bottom to get the heat away from the FET, then
solder copper fins to increase exposure to air. IF you're NOT expecting sudden and rapid temperature rises, then that's
plenty good enough. This is where a FLIR E4/8 is DAMN handy :-) .. much clearer than doing stoopid calculations :-)
At 25A, I run at app 45-50C on 30C ambient, with NO other heatsinking.

[SeanB]

Thermistor attached as close as possible to heat tab on back of unit. Easiest to do if you have a 3mm via there in the middle and have a 2mm thermistor in the hole with a little heat sink compound to get good thermal contact.

[tszaboo]

..or it does not work because it is a package limitation ultimately (even though some spec sheets say that the device can dissipate over 200watts)..

It is not a package limitation. D2PAK is a TO220 equivalent, there are TO220 packages with 0,5K/W thermal junction to case resistance. The limitation is the base material, a very thin layer of copper in this case. The only real workaround is IMS pcb (insulated metal substrate), ceramic PCB, copper bond PCB. None of these are cheap and hobbyist friendly, unfortunately. For IMS, the whole shebang must be heated up during reflow, and I wouldn't even try soldering it any other way. If you want good dissipation for cheap, TH is the way to go. If you need lots of current for DC-DC, multi phase converters have superior efficiency, and better heat-spread.

[T3sl4co1l]

Not necessarily sucky.  You can use via-in-pad on both sides (have fun getting the solder in there, right?) and solder the heatsink/spreader on the back side.  Which can be just a hunk of copper (maybe 30 thou, 1" square, say).  That'll probably get into the 20W range pretty easily.

Hmm, should try that actually.  Will take up a lot of footprint area and leave little space on the backside -- pretty much requiring a 4+ layer board, but it sure will make for a low profile build.  Sandwich a few boards like this between sil-pads and water cooled heatsinks and you could build up a whole lot of power in a small volume.

I don't recommend bending leads, just because it's a silly way to do it.  For home made stuff... silly is just fine.   A production nightmare though.

Tim

[digsys]

    "TH is the way to go" ... what is TH?   

TH = Through hole = leaded = "legs" :-)  ie TO220  Gives you HEAPS more cheaper options to H/Sink the sucker.

[3roomlab]

ah thru hole components ... ok thanks

nah was exploring those non TH stuff, like LFpak, D2Pak etc. and on a single side board ... low tech CEM-1 material