DC load using a CPU cooler

[microbug]

Yes- I agree that it seems like an all round good idea.

FYI, I'm going with 3 100W rated MOSFETs on a 200W cooler (over voltage/current limits in hardware and temperature sensor on heatsink + emergency temperature cutoff on heatsink). Single channel 16 bit DAC controlling current, dual channel 16 bit ADC measuring voltage through a resistor divider, diode clamp and current limiting resistor and measuring current through an INA196. I chose 16 bit data converters because they are not too expensive (less than £2 each). Precision low offset op-amps are used.

Sorry for the brief wording but I'm typing this on a smartphone.

[eneuro]

bolt the mosfets to some random large heatsink and dump it into a tub of mineral oil

If we connect this tube to such room aluminium heater radiator (<10$ about 0.5m long ) we can probably run at full power without any fans 

Easy and scalable-if we need more power to disipate simply add another one 

[Spikee]

It is alive !



It has a few bugs, i placed the 3d model facing the wrong way. So to fix the pinout i needed to mount the heatsink and mosfets on the bottom.
Also to rise the mosfets up it is advisable to use some discard able 0805 resistors under it so it rises the mosfet by around 1mm.

With some tweaking it is definitely something that is doable.

[rob77]

It is alive !

It has a few bugs, i placed the 3d model facing the wrong way. So to fix the pinout i needed to mount the heatsink and mosfets on the bottom.
Also to rise the mosfets up it is advisable to use some discard able 0805 resistors under it so it rises the mosfet by around 1mm.

With some tweaking it is definitely something that is doable.

where are your current sensing resistors ? even if you use 0R1 then it's 2.5W at 5Amps - can't see any resistors capable of 2.5W on your board
and sorry , but i doubt it's alive - half of the parts are missing yet

[Spikee]

It is alive !

It has a few bugs, i placed the 3d model facing the wrong way. So to fix the pinout i needed to mount the heatsink and mosfets on the bottom.
Also to rise the mosfets up it is advisable to use some discard able 0805 resistors under it so it rises the mosfet by around 1mm.

With some tweaking it is definitely something that is doable.

where are your current sensing resistors ? even if you use 0R1 then it's 2.5W at 5Amps - can't see any resistors capable of 2.5W on your board
and sorry , but i doubt it's alive - half of the parts are missing yet

There are two big ones on the bottom =)
mouser.com/Search/ProductDetail.aspx?R=WSL36372L000FEA

[rob77]

It is alive !

It has a few bugs, i placed the 3d model facing the wrong way. So to fix the pinout i needed to mount the heatsink and mosfets on the bottom.
Also to rise the mosfets up it is advisable to use some discard able 0805 resistors under it so it rises the mosfet by around 1mm.

With some tweaking it is definitely something that is doable.

where are your current sensing resistors ? even if you use 0R1 then it's 2.5W at 5Amps - can't see any resistors capable of 2.5W on your board
and sorry , but i doubt it's alive - half of the parts are missing yet

There are two big ones on the bottom =)
mouser.com/Search/ProductDetail.aspx?R=WSL36372L000FEA

2milliOhm ?  what op-amp are you using ? 1mV input offset means half an Amp error with a 0R002 current sensing resistor.

[Spikee]

The INA225 monitors each current shunt.
 but keep in mind that this is still a prototype. Parts might change.

[kc9qvl]

That looks very nice.

Mine is very meager compared. How many watts can it dissipate?

[Spikee]

with a better heatsink 300-350w should be possible. With the current smal one only around 150-200w.

[Nerull]

Thing is, a CPU cooler is doing a specific job, which is to MOVE a high heat flux heat from a very small area, and then reject it to the ambient.  A typical DC load, to handle a high power dissipation will need multiple parallel power devices, and so you don't have such a small area / concentrated heat source.  Also, in modern computer design, the issue is packaging the heat transfer device into an ever smaller physical space, again, unlikely to be an issue with a typical bench top DC load.

And ultimately, you have to reject that heat to the ambient atmosphere, which for a high power DC load requires a lot of surface area or a lot of fast moving air, as it is (should be) continuously rated.  This means either a large heatsink or a powerful fan.

I suspect that while CPU coolers have nice low thermal transfer co-efficient, they actually don't have amazing total heat rejection  values (because the CPU is still a relatively low power device (it gets "hot" because it is so compact!)

For my DC load i used a nice big heatsink, placed the power devices nicely spread out on it, and used two  powerful speed controlled server cooling fans to blow a LOT of air past it.  This manages pretty much 300W continuous in an 30degC ambient.

With CPUs out there outputting more than 200W of heat under load and gamers who can't stand to see their CPUs top 40C, I'm sure there are heatsinks out there that can handle that.

[microbug]

Yes, there are PC coolers (e.g., Arctic Cooling 13 Pro) that can do 300W. I imagine there are some that can do more, but CPUs which generate more than 300 watts of heat are generally intended for servers (Intel Xeon). I could switch from the Arctic Cooling 13 to the Arctic Cooling 13 pro - it's only £10 more, but I don't really need the extra power. If I ever do, I can just buy the Arctic Cooling 13 Pro and fit it in place of the old cooler, and change the firmware so it doesn't cut out above 200W.


I just realised that a PSoC would be perfect for this application and would cut down on the number of external components I need. I'll order a PSoC 4100/4200 dev board and take a look (I got a 10% off code for Farnell by email )

[microbug]

@spikee - how large is the thermal contact on the Freezer 7 (CPU cooler)? I'm wondering how many TO-220 packages I can get under there. I'll probably use either the Freezer 13 or Freezer A30 (17=200W, A30=320W), which (I think) have similar thermal pad sizes.

I'm going to use a different MOSFET to the one I first chose (and stupidly ordered), as the SOA dictates a max of 10V at full current!


EDIT: What would be better thermally: D2PAK packages on the bottom side with thermal stitching, and a cooler on the top; or upside-down TO-220 packages directly touching the cooler?

[SeanB]

Upside down TO247 packages are better, lower thermal resistance. Plus you can fit a bigger die inside which has lower resistance to case as well. You probably will only get 9 under the thermal contact, you need a little room to bend the leads and board to attach them to. Use the thickest PCB you can get, otherwise you will need to machine a back brace to get even contact to all 9 dies, otherwise the warping of the board will cause the middle one and the ones not in the corners to not have good thermal contact. A back brace made from thick insulating material with cutouts for the lead stubs will probably be needed in any case, or a thinner one with an aluminum backing plate to give even pressure.

[microbug]

OK, thanks. For a back brace, could I use a block of aluminium (6mm thickness, see this eBay listing for source) and drill holes in for the leads? I have a drill press. I could get a thicker aluminium block (1cm for example) if that would be better for the back brace.


Thanks for your help!

[SeanB]

Yes, any plate that is stiff, big enough to act as a load spreader and which you can insulate from the board with something like a blank board with milled holes for the leads. You just need it to press evenly on the board so it presses the transistors evenly onto the heat plate.

[timb]

Upside down TO247 packages are better, lower thermal resistance. Plus you can fit a bigger die inside which has lower resistance to case as well. You probably will only get 9 under the thermal contact, you need a little room to bend the leads and board to attach them to. Use the thickest PCB you can get, otherwise you will need to machine a back brace to get even contact to all 9 dies, otherwise the warping of the board will cause the middle one and the ones not in the corners to not have good thermal contact. A back brace made from thick insulating material with cutouts for the lead stubs will probably be needed in any case, or a thinner one with an aluminum backing plate to give even pressure.

Wait, 9??? Look at the picture of the heatsink portion. (5th Picture)

That looks like *maybe* enough room for two TO-220 packages...

http://www.amazon.com/gp/aw/d/B006TCSH4Y/




Sent from my Smartphone

[microbug]

Yes... I agree. I'll measure it up when I buy the heatsink.

EDIT: that link is correct, if anyone is wondering.

[mrflibble]

Wait, 9??? Look at the picture of the heatsink portion. (5th Picture)

That looks like *maybe* enough room for two TO-220 packages...

Heh, I was wondering the very same thing.

I don't know of any cheapo cpu coolers that would accomodate 9 TO-220 packages. You could do something custom, but that would automatically discard that "cheapo" word in there.

Image of the cooler being discussed. Inline because I am lazy today:

[microbug]

Judging by the details on this page (4x8mm heat pipes), I'd say that the thermal pad is 32mm by 32mm or thereabouts. That's just enough for two TO-247 packages.

EDIT: I think it would just do 6 TO-220 packages, because in one dimension it is slightly more than 32mm. My current design (which needs some significant changes anyway) has an OPA4188 controlling 3 MOSFETs and acting as an amplifier for the DAC. I could stick in another OPA4188 and use 6 op-amps for 6 MOSFETs. I'm sure I can find a use for the two spare op-amps.

[timb]

What you could do is get a 1cm+ thick square of copper and mount 4 to 6 FETs on it with the heatsink mounted in the center. (So basically you'd have 2 or 3 FETs mounted bent-legged-face-up on the copper just past where those heat pipes curve up.)

The copper pad could be thermal epoxied to the board.

This way you wouldn't require any reinforcements on the bottom either!


Sent from my Smartphone

[microbug]

The way I see it, I'm going to have to have a block of metal around the FETs for reinforcement and good thermal contact. If someone could suggest a source for copper blocks, that would be good; a cursory look on eBay didn't yield anything thicker than 3mm. TBH, the deciding factor (if I can find a source for a block of copper of the right dimensions) is likely to be cost. I don't want this thing, in total, to be more than around £100.

Since the PCB will be comparatively large, I'll probably get it manufactured in China and sell the 8 or so spare ones to any willing buyers on this forum, which will help a bit with cost.

Estimated cost breakdown:
Components: £15
MOSFETs: £15 (max, could be substantially less)
Cooler: £30
PCBs (x10): £20-25
Enclosure & mounting hardware: £15 (Aliexpress)

EDIT:
FYI, here are some rough specs:
Max power dissipation: 300W
Max voltage: 50V+
Max current: 20A (4mm binding posts are generally rated for 20A max)
Constant current / voltage in hardware, constant power / resistance in software (I don't know what options I'll have with the PSoC 4, but it might be possible to get an analog control loop for constant power and resistance modes).

[Marco]

Wouldn't it be easier to just get some two component thermal epoxy and glue them on?

[microbug]

I doubt that just thermal epoxy would give the required pressure and thermal contact.

[timb]

You'd be surprised. I've got some arctic silver thermal epoxy here. Once it's dried, shit ain't coming off. Ever.

Epoxy is very, very strong stuff.


Sent from my Smartphone

[rob77]

constant voltage in hardware is kind of dangerous... imagine what will happen if you set CV to 4V while connected a 3cell lipo. i would rather implement the CV in software with cutoff current/power.