Heatsink for DC electronic load

[void_error]

I was looking around for a heatsink for a modular DC load I'm currently designing and ran across this.

How much power do you think it can handle before my 4 TO-220 MOSFETs attached to it blow? I'm guessing more than the 65W it's supposedly rated for since the MOSFETs can be allowed to run hotter than a CPU.
The nice thing about this cooler is its price - US $4 on sale so I ordered 5 of them.

[tszaboo]

Depends on the junction to case thermal resistance of the MOSFETs. You have somewhere in the ballpark of 0.5K/W to 2K/W for a TO220 power MOSFET plus insulation material. The heatsink itself will heat up. Mounting can be less than ideal.
On the other hand, if all conditions are ideal, I think 65W should be possible.

[void_error]

As far as I know MOSFETs can safely work at higher temperatures than a desktop CPU so I'm guessing around 80W is probably possible. Also, the cooler on the CPU is in a case that's above ambient while this will draw air at ambient temperature so it should be able to dissipate more.

[Jeroen3]

The heaviest amd that it is supposed to fit has an 125W TDP.
With an Socket 775 server cooler (with 3 heatpipes) I was able to fume 65 Watts of a TO247 with a few mm oxide insulator.
(Thermal cutoff NTC for 90C glued with paste in the to247 hole)

[void_error]

That's one transistor, I have 4 so theoretically 20W/transistor should be ok. Overall thermal resistance will be lower than for 1 device.

[nctnico]

I'd design it so the heatsink is connected to the input voltage. That way you can bolt the input terminals to the heatsink and avoid electrical insulation between the heatsink and the transistors. Electrical insulation increases the thermal resistance very quick.

[Kevin.D]

These coolers look about same size  as ones I used (I think old athlon cpu coolers)to build an eload.
I measured the thermal Resistance (S-A) of these it was .3 C/W with fan running. And  2.4 C/W fan off.
Originally I designed to use 2 of these sinks/fans in order to mount two  TO3-p  case style Nfets on each sink which would have given me my aimed for ~220 Watts max of dissipation with fans and 40W without fans   (But I only mounted one nfet on each sink in my prototype for testing  (I haven't yet had the
necessity/inclination to mount the other 2 Nfets) and with these two I can do ~ 150W continuous with fans running but this doesnt leave much safety margin .I would say 60W each is comfortable for the fets I chose.

Just do some common sense 'thermal impedance' matching similar sorta thing as you do with power transfer , (i.e theres no large advantage in increasing heatsink rating if the largest thermal bottleneck is the junction to sink resistance of cases and conversely no large benefits by adding more parrallel fets to reduce j-sink Resistance if sink-ambient rating of sink is largest bottlneck.

Choose Mosfets carefully, although if only going for very low input voltages then this is no biggy. Otherwise dont choose Hi Gm, low rdson  fets designed for switching, as there max dissipation rating at highger Voltages in linear mode will be very poor (often the data sheet's won't show this correctly or dont show it all). Good mosfets for linear are older generations with lower gm (especially)  and higher rds on.
Regards . 

[void_error]

I'd design it so the heatsink is connected to the input voltage. That way you can bolt the input terminals to the heatsink and avoid electrical insulation between the heatsink and the transistors. Electrical insulation increases the thermal resistance very quick.

I was thinking about that.

Choose Mosfets carefully, although if only going for very low input voltages then this is no biggy. Otherwise dont choose Hi Gm, low rdson  fets designed for switching, as there max dissipation rating at highger Voltages in linear mode will be very poor (often the data sheet's won't show this correctly or dont show it all). Good mosfets for linear are older generations with lower gm (especially)  and higher rds on.
Regards . 

Thanks Kevin. I'll take that into acount. R_DS[ON] is not very important here. I suppose I can get 20-25W out of a TO-220 package, the whole thing will have thermal shutdown anyway. I'll digging for suitable transistors right away.

[void_error]

Looks like the IRF620 might be a good candidate...
What do you think?
Or the IRF510?

[Kevin.D]

Old favorites irf510 and Irf530 .
Yes if  your satisfied with a to220 case dissipation rating then these would be good choices.
Whilst higher rds(on) is good here note that it  limit's the max current at LOW input Voltages (i.e if Rsense = .1 and min rds (on) = .4 so your min Output R is now .5 so  if DUT is .5V  max I is now  1 A ) .So more in parallel or irf530 (rds(on) .16 if higher current's at low Vins needed.
In the transfer characteristics of those look at the nice slow linear rise in Id Vs Vgs over the full input Vgs range  (Meaning a large variation in temp (which causes a lowering of Vgs(th)) now only causes a relatively mild increase in Id and  so  less chance of thermal runaway internally at high Vds  ( a quicky calculation of the extra power dissipation @ max Vin due to that xtra current makes these to be  almost immune to thermal runaway on a good heatsink . You  could probably  even get away running these in parallel without a opamp each fet if you use one xtra as safety margin  )) .

regards

[void_error]

They both aren't really great for higher power dissapation, look at the thermals.
Max power dissapation at 25C ambient 50W derating at 0.4W/C, Rjc is crap 3.5 so for a realistic operating worst case condtion maybe 10W with a good heatsink and fan.

Oh crap, I overlooked the thermal derating factor.

Try and find one with high 25C power dissapation capability and smal jc thermal resistance these are usually TO247 or T03P type packages a lot also are rated to 175C max junction temp versus 150C in the puny T0220 packages.

Hmm, they're quite expensive...
Haven't made any bill of materials calculation yet but it might be cheaper to use less TO-247/TO3P rather than more TO-220 devices. Each MOSFET will have its own two driving op amps (no, not in parallel).

IRFP150M looks like a good choice at $1

Also, g_m is g_fs in the datasheets?

[void_error]

You should be able to get 70-80W WATTS out of each of those assuming adequate heatsinking. That is what I've gotten out of these

https://www.fairchildsemi.com/datasheets/FQ/FQA24N50.pdf

One blew up on me testing a 200VDC PFC, but then I only paid about a buck for them so if I have to replace one every now and then who cares.

Two of them should be good for 100W total then but I'll double check. They're the cheapest I could find for a TO-247 case.

[Mr.B]

I used IXTH30N50L2 to build my load.
Two TO-247 per CPU cooler.
Although not cheap...

Datasheet below.
The SOA graph says it all.

[void_error]

Yep double check, those do have a higher jc resistance ,almost 1 still much better than the T0220s though, but still I think with two you should be able to get 100W assuming a decent thermal solution fan + heatsink. 

Looks like two won't be good enough for 100W assuming 0.25C/W heatsink-ambient and 0.5C/W case-heatsink at an ambient temperature of 40C. I simply halved the thermal resistance of 1 transistor to get the pair's combined thermal resistance.

I used IXTH30N50L2 to build my load.
Two TO-247 per CPU cooler.
Although not cheap...

Staying away from IXYS mainly because of the price. Building a dirt cheap modular DC load using $4 CPU coolers.

Datasheet below.
The SOA graph says it all.

How much power are they dissipating? They seem overkill.

[Mr.B]

How much power are they dissipating? They seem overkill.

100W per FET.
My "Power Module" has four FETs and two coolers.
The power boards are designed so that they can be daisy chained using FPC cable for the control and feedback signals.
The controller board is separate.
Cooling is provided by Corsair H60 CPU water block cooler units with high velocity fans. One on each side of the Power Module.
At 30 degrees C ambient I estimate that Tj is possibly about 115 degrees.
I say 'estimate' because Corsair will not give me any details on the thermal performance of their cooler, so all I can do is measure it manually as best I can.
I am fairly sure I could get more than 100W per device, but I am not convinced the coolers would perform satisfactory.

[void_error]

Looks like two won't be good enough for 100W assuming 0.25C/W heatsink-ambient and 0.5C/W case-heatsink at an ambient temperature of 40C. I simply halved the thermal resistance of 1 transistor to get the pair's combined thermal resistance.

Two IRFP460s should provide an adequate safety margin.

[Mr.B]

Two IRFP460s should provide an adequate safety margin.

That FET does not look to me to be designed for linear use.
Also, Vishay have not plotted a DC curve on their SOA graph, so there is no way to determine from the datasheet how it might perform as a linear device.

Edit: I am no expert by a long shot, but it has a "linear derating factor" of 2.2 W/deg... That doesn't look good.

[fivefish]

I built this... will do 240W dissipation.


Tested it for at least 30min at 30V @ 6A, 60V @ 3A (180W) and 120V @ 2A (240W).
Around 120V @ 2.5+A, the mosfet gets shorted permanently... oops!
   
Using (2) Mosfets, one per CPU cooler.
I tapped the copper core, and screwed the TO247 directly to the heatsink core.
Using multi-turn pot for fine-grained control.
Also a BNC jack, for connection to a scope so I can monitor DC waveform if needed.
12VDC adapter.

[Jeroen3]

I used an IXTQ22N60P and 3 mm of aluminum oxide insulator plate.

The problem is that you cannot make a load step from ambient temperature to theoretical maximum load. The mosfet has too much thermal inertia (is that the right term?) and it will break. You have to ramp it up slow enough for it to not create a lot of thermal delta and expansion stress.

_{Discussing soa curves is really awkward in dutch... since soa == std.}

[void_error]

Two IRFP460s should provide an adequate safety margin.

Edit: I am no expert by a long shot, but it has a "linear derating factor" of 2.2 W/deg... That doesn't look good.

Here are some calculations I've done:

2xIRFP460 - R_th[j-c]=0.225C/W (0.45/2) P_D=280W, 2.2C/W derating above 25C

2xKAP 218 - R_th[c-h]=0.27C/W - it's 0.15C/W for 1 sq in, scaled it down to the transistor's thermal pad and divided by two because there's two transistors (result is actually 0.535C/W for a kapton pad that's the exact size of the transistor's thermal pad)

DEEPCOOL CK-AM209 - R_th[h-a] = 0.25C/W estimated

R_th[j-a] = 0.225 + 0.270 + 0.250 = 0.745C/W

For 100W that results in a 74.5C temperature rise above ambient.

For a 40C ambient (worst case) T_j = 114.5C. I don't run above 30C ambient so this won't either.

T_c will be T_j-(0.225*100)=92C

Apply the linear derating factor of 2.2W/C and you get 280-((92-25)*2.2)= 146W you can dissipate at a T_c=92C.

If I've done this right then these MOSFETs are overkill but I've probably screwed up somewhere. Please correct me if I'm wrong.

EDIT:
IRFP140 - 99.6W @ T_j=133.5C
IRFP250M - 120.2W @ T_j=127.0C
IRFP450 - 89.5W @ T_j=124.5C
IRFP150M - 99.6W @ T_j=139.5C - cheapest TO-247
IRFP140N - 79.0W @ T_j=147.0C

[void_error]

I have another idea which will go well with they way I plan to mount the transistors.

They'll be bolted onto the heatsink using M3 screws and with the leads bent forward. The whole thing will sit on top of the PCB.

I want to add an I2C temperature sensor to each of them. They'll be mounted on daughter boards in direct contact with the MOSFETs using thermal paste.
The MCP9805 and MCP9843 are the cheapest I could find. They have 3 external address bits so than means with two of them per DC load module I can use up to 4 DC load modules daisy-chained for a total power of 400W.
They also have a pin called "Event" which can be used for hardware shutdown of the DC load module in case the case temperature of the MOSFETs exceeds a preset value.

What do you think?

[fivefish]

They'll be bolted onto the heatsink using M3 screws and with the leads bent forward.

That's what I did, tapped and threaded the heatsink, and bolted my TO247 to it.  There are pads on the PCB that align with the TO247 legs if they're bent forward.  But on my photo above, I just used flying wires to attached to the PCB. 

If you didn't use an insulator (mica, silpad, etc), the heatsink's aluminum body will be at same potential as the V+ of device under test.  It's up to you if that's a big deal or not.

Here's my heatsink's fin temperature after running 30 minutes at 180Watts.  Later on, I realize I should have connected the temp probe at the TO247 itself. The junction temperature will be hotter than this 40C.



Good luck on your project!

[void_error]

Sorry for bumping this up again...

I've been double-checking the MOSFETs I was planning to use and and oops... they're probably gonna blow up. SOA is not DC rated. Found an IXYS device that would do and hold up at DC dissipating the required power but it's bloody expensive so I started looking at darlington BJTs. Much cheaper. Found the BDV65B to be suitable as far as SOA, V_CE and I_C are concerned. Haven't done the thermal calculations yet.

Is it a good idea to use BJTs instead of MOSFETs?

[dom0]

BJTs are a very good option for a current sink, under the following conditions:

- required minimal compliance voltage is >2 V
- required currents are not very high (which would lead to a lot of control current and power)
- note second breakdown at higher Uce. It gives similar limits to DC operation to BJTs as the Spirito effect gives to FETs.

[TopLoser]

Sorry for bumping this up again...

I've been double-checking the MOSFETs I was planning to use and and oops... they're probably gonna blow up. SOA is not DC rated. Found an IXYS device that would do and hold up at DC dissipating the required power but it's bloody expensive so I started looking at darlington BJTs. Much cheaper. Found the BDV65B to be suitable as far as SOA, V_CE and I_C are concerned. Haven't done the thermal calculations yet.

Is it a good idea to use BJTs instead of MOSFETs?

See my teardown of a 1200W 240A that uses MOSFETS, posted it today in the test equipment board.