Thermal Design for LFPAK FETs

[Steve K]

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package look quite attractive (low source inductance in particular appeals to me) if I knew how to heatsink them to a significant portion of their rated dissipation (typically ~50W). NXP's LFPAK thermal design document discusses under-1-watt applications only. Has anyone any clever ideas on how to get the heat out ?

 

[Phil Allison]

"Steve K"

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package look
quite attractive (low source inductance in particular appeals to me) if I
knew how to heatsink them to a significant portion of their rated
dissipation (typically ~50W). NXP's LFPAK thermal design document discusses
under-1-watt applications only. Has anyone any clever ideas on how to get
the heat out ?


** The LFPAK is about the same size as the chip used in many TO3P/ TO264
power fets and bjts.

So, you could clamp or solder the LFPAK to a copper header and then bolt or
clamp that onto a large aluminium heatsink.

IOW create your own large flat pak.

But why bother ?



..... Phil

 

[Tim Williams]

Some manufacturers offer copper core PCBs (the core of which can be
soldered to). Don't ask about the price.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://seventransistorlabs.com

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package look
quite attractive (low source inductance in particular appeals to me) if I
knew how to heatsink them to a significant portion of their rated
dissipation (typically ~50W). NXP's LFPAK thermal design document
discusses under-1-watt applications only. Has anyone any clever ideas on
how to get the heat out ?

 

[miso]

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package
look quite attractive (low source inductance in particular appeals to
me) if I knew how to heatsink them to a significant portion of their
rated dissipation (typically ~50W). NXP's LFPAK thermal design
document discusses under-1-watt applications only. Has anyone any
clever ideas on how to get the heat out ?>

Not using bonding wires certainly gets the inductance down. But any
discussion of how to mount the part requires knowing the ambient
temperature. Since you are switching the FET, what is the duty cycle?
Does the "on" period exceed a millisecond? [Generally using PWM in
temperature analysis doesn't work if the on time is too long. For
instance, 1 day on and 10 days off isn't 10% duty cycle in the eyes of a
semiconductor.]

A lot of cheesy consumer grade products spec the maximum ambient at 40
degrees C because they figure if it is used in the presence of a person,
said person will get out of Dodge when the ambient exceeds 40. This is
really common for devices with displays.

BTW, there are plenty of places in the world where the ambient exceeds
40 degrees C!

 

[Martin Riddle]

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package
look quite attractive (low source inductance in particular appeals to
me) if I knew how to heatsink them to a significant portion of their
rated dissipation (typically ~50W). NXP's LFPAK thermal design
document discusses under-1-watt applications only. Has anyone any
clever ideas on how to get the heat out ?

Heatsink?
<http://www.fischerelektronik.de/en/latest-news/press-releases/releases/smt-heatsinks-for-lfpak/>

Cheers

 

[[email protected]]

Heatsink?
<http://www.fischerelektronik.de/en/latest-news/press-releases/release...>

Cheers

if you look at similar solderable heatsinks for D2PAK they are around
25K/W
so they are not much help with 50W

-Lasse

 

[Steve K]

Thanks for all the comments...a couple suggested soldering to a copper tab and bolting that to a heat sink. That's pretty much what I came up with (though I haven't tried it yet), other than a dubious idea of pumping oil over the PCB.

Why bother ? Potentially tens of MHz operation (~50% duty cycle) at a few percent of the component cost of RF power FETs (if the heatsinking kluge can be done cheaply). I was looking mostly at the lowest Qg versions of the NXP PSMN series. I had looked at the Fischer heatsinks and they only seem good for a few watts.

Steve

 

[[email protected]]

Thanks for all the comments...a couple suggested soldering to a copper tab and bolting that to a heat sink.  That's pretty much what I came up with (though I haven't tried it yet), other than a dubious idea of pumping oilover the PCB.

Why bother ?  Potentially tens of MHz operation (~50% duty cycle) at a few percent of the component cost of RF power FETs (if the heatsinking kluge can be done cheaply).  I was looking mostly at the lowest Qg versions of the NXP PSMN series.  I had looked at the Fischer heatsinks and they only seem good for a few watts.

Steve

isn't that something like where directfets shines? they have the
"right" side up so you can put a heatsink on top
Maybe mount them on back side sandwiched between pcb and heatsink?

another trick I've seen is to using the kelvin connection on a sense
fet as reference for the gate drive


-Lasse

 

[miso]

Par for the course when doing datasheets for devices that expect pads of
copper to be heat sinks is to just solder the part to a big piece of PCB
and characterize the theta JA. You dremel wide breaks in the copper to
change the size of the heat sink.

To get the junction temperature on a chip is easy since you can
characterize a parasitic diode. [Force a small constant current into the
pin. Make sure the part is not generating any heat itself, that is,
don't operate it. Then sweep temperature and measure the diode voltage.]
For a discrete FET, this could be harder or maybe impossible since there
is no free diode to play with when you make the chip generate heat.

There may be a way to put two parts on the PCB with one generating heat
and the other to measure junction temperature, but I'd have to mediate
on if that is kosher.

 

[Spehro Pefhany]

Thanks for all the comments...a couple suggested soldering to a copper tab and bolting that to a heat sink. That's pretty much what I came up with (though I haven't tried it yet), other than a dubious idea of pumping oil over the PCB.

Why bother ? Potentially tens of MHz operation (~50% duty cycle) at a few percent of the component cost of RF power FETs (if the heatsinking kluge can be done cheaply). I was looking mostly at the lowest Qg versions of the NXP PSMN series. I had looked at the Fischer heatsinks and they only seem good for a few watts.

Steve

Aluminum-core PCB?

 

[Nico Coesel]

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package look qu=
ite attractive (low source inductance in particular appeals to me) if I kne=
w how to heatsink them to a significant portion of their rated dissipation =
(typically ~50W). NXP's LFPAK thermal design document discusses under-1-wa=
tt applications only. Has anyone any clever ideas on how to get the heat o=
ut ?

I have used these kind of heatsinks on a product:
http://uk.farnell.com/fischer-elektronik/fk-244-08-d-pak/heat-sink-for-d-pak-31-5-c-w/dp/4314001

 

[[email protected]]

Aluminum-core PCB?

$$$$$$!!!!!!!

 

[[email protected]]

Par for the course when doing datasheets for devices that expect pads of
copper to be heat sinks is to just solder the part to a big piece of PCB
and characterize the theta JA. You dremel wide breaks in the copper to
change the size of the heat sink.

To get the junction temperature on a chip is easy since you can
characterize a parasitic diode. [Force a small constant current into the
pin. Make sure the part is not generating any heat itself, that is,
don't operate it. Then sweep temperature and measure the diode voltage.]
For a discrete FET, this could be harder or maybe impossible since there
is no free diode to play with when you make the chip generate heat.

There may be a way to put two parts on the PCB with one generating heat
and the other to measure junction temperature, but I'd have to mediate
on if that is kosher.

I'd think that if you just want to characterize it you could run it in
reverse
using the body diode as both thermometer and to generate heat

toggle between a high current to generate heat and a small current to
measure temperature

-Lasse

Version 4
SHEET 1 1172 680
WIRE 320 -64 208 -64
WIRE 704 -64 432 -64
WIRE 896 -64 704 -64
WIRE 128 -48 -96 -48
WIRE 704 -48 704 -64
WIRE 432 -32 432 -64
WIRE 896 -32 896 -64
WIRE 704 48 704 32
WIRE 208 80 208 -64
WIRE -96 112 -96 -48
WIRE 128 160 128 -48
WIRE 160 160 128 160
WIRE 208 192 208 176
WIRE 208 192 -96 192
WIRE 208 288 208 272
WIRE 432 288 432 48
WIRE 432 288 208 288
WIRE 448 288 432 288
WIRE 896 288 896 48
WIRE 1008 288 896 288
WIRE 208 320 208 288
WIRE 208 320 112 320
WIRE 208 352 208 320
WIRE 112 368 112 320
WIRE 160 368 112 368
WIRE 896 368 896 288
WIRE 896 368 848 368
WIRE 896 384 896 368
WIRE 848 400 848 368
WIRE 208 480 208 448
FLAG 704 48 0
FLAG 208 480 0
FLAG 896 480 0
FLAG 448 288 Vdiode
IOPIN 448 288 Out
FLAG 1008 288 Vdiode1
IOPIN 1008 288 Out
FLAG 320 16 0
SYMBOL voltage -96 96 R0
WINDOW 3 -370 51 Left 2
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR Value PULSE(0 10 0 1u 1u 1.5m 2m 5)
SYMATTR InstName V1
SYMBOL voltage 704 -64 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V3
SYMATTR Value 5
SYMBOL nmos 160 80 R0

 

[miso]

I'd think that if you just want to characterize it you could run it in

reverse
using the body diode as both thermometer and to generate heat

toggle between a high current to generate heat and a small current to
measure temperature

-Lasse

I'm a little unsure about the toggling. I've done this with very slow
bench meters to measure the diode, but you have added time into the
equation.

I'm not ruling this out, but you would probably want to test the scheme
on a device that does have a spare diode. That way you would have a
benchmark on the scheme's accuracy.

 

[legg]

Some of the NXP (and Renesas) MOSFETs in the LFPAK (SOT669) package look quite attractive (low source inductance in particular appeals to me) if I knew how to heatsink them to a significant portion of their rated dissipation (typically ~50W). NXP's LFPAK thermal design document discusses under-1-watt applications only. Has anyone any clever ideas on how to get the heat out ?

If you mount them on a largish copper area, you can couple this to the
package wall or other radiators either directly or through thermally
conductive pads/insulators. The large copper area sor of defeats the
intention of thhe smaller package though, doesn't it?

There's not much use heatsinking them inside a container, unless
there's somewhere for the heat to go, but everything will reduce
thermal impedance.

The Aavid 5731/5733/5734/7109 types, at 4-16 degC/W are similar to the
Fischerelektronik parts.

A 1 in square thermal pad can easily beat the lower end of the above
numbers, even at a considerable thickness.

RL