DC load using a CPU cooler

[metacollin]

I came to this thread kind of late, but I've been reusing CPU heatpipe/sink assemblies with and without fans for years with FETs and transistors with all sorts of packages.  Doesn't matter if they're surface mount or have a screw hole, I use both.  Use whatever package has the lowest junction to case thermal resistance. 

This is what I've learned:

1. Do not use thermal interface compound (TEC).  It's terrible and using it often ends in ritual suicide. TECs do little beyond fill in imperfections between two flat surfaces with something that isn't air.  But otherwise, it's basically about as bad at conducting heat as a solid, non-air filled material/over priced goop can be. Toothpaste performs nearly as well as thermal grease while smelling significantly better (spearmint is my favorite).  There is less than a 10% performance difference between quality thermal paste and shitty toothpaste.  And TEC dries out or otherwise deteriorates, though much more slowly than toothpaste (years vs. hours).

2. Do not drill holes in the heat spreader, it's a huge dickchafe and you'll probably breach a heat pipe and effectively ruin the entire assembly's ability to shed those phat watts. 

3. Brackets? Clamps? Maybe even bears, oh my? Don't bother, and it will be a huge pain in the dick for inferior thermal performance. 

4. Hot air heats heatsinks as effectively as cool air cools them.  Heatguns aren't just for heatshrink and bacon.     

5. The important bit - Just solder the damn FETs directly to the copper heat spreader.  They're *made* to be used that way.  No holes, no brackets, no bullshit, just bitchin' awesome thermal performance and a bond as strong as the heatspreader is bonded to the rest of the assembly and heatpipes.  And you won't have to commit Seppuku to restore honor to your family. (The last one was a joke, of course.  Probably.)

Use low temperature Bismuth-Tin solder.  Sometimes you can find it in paste form in small quantities, but its quicker and easier to just make your own.  Find that old bag of bismuth ingots and that bar of grade A tin you have stashed somewhere in your garage, melt them in an aluminum container on the stove, mixed 42% Tin, 58% Bismuth, Sn42/Bi58.  This is the same thing bonding the heatpipes to the cooling fins and heatspreader.  It melts around 140°C, give or take.  You can tweak the temperature by moving the ratio tinward. Remember, heatpipes work in reverse, just blow a heatgun through the part a fan would blow air through and the heat pipes will eagerly shit all of those watts into the copper heat spreader where you want to solder your FET, and you'll have things up to temp and wetting nicely in tens of seconds.  Just, uh, make sure the whole thing doesn't fall apart due to gravity or something.  Also don't attempt this with normal solder, you'll vent one of the heatpipes and it WILL shoot a jet of methanol directly at your genitals.

Those wattage and SOA curves in the datasheets are all done with those mofos soldered DIRECTLY to some big ass hunk of copper.  If you solder a FET to copper, it adds negligible thermal resistance and, depending on the total thermal resistance of the junction to heatsink/heatpipe assembly to ambient, you will actually achieve the specs in the datasheet.  Now, often the wattage is assuming you cast level 99 Heatsink of the Infinite and have a heatsink with 0 thermal resistance cooled by unicorns breath, but if you run the numbers for your non-magic non-unicorn breath cooled heatsink, you'll get exactly what all the thermals predict.  The heatsink is the limiting factor if its for a CPU, it probably won't handle much more than 100W shed to ambient with a fan, but a TO-247 will dump 100W into those sweet sweet heat pipes without even breaking a sweat, no problem.  Single package.  With a bigger heat pipe assembly, 200W out of a single package is totally doable, I've done 240W out of a TO-264 soldered to a big ass aluminum heatsink (solder wets aluminum wonderfully, just sand it in an aquarium or similar and fill it with CO2 using dry ice or whatever is handy, this will remove the surface layer of aluminum oxide and prevent a new one from forming instantly.  The aluminum oxide surface layer is what makes aluminum seemingly unsolderable, but actual unoxidized aluminum wets nicely).  Not even copper or heatpipes, just shitty old aluminum. 

Some quick examples attached.  In order: 1. A big ass fet soldered directly to a big ass cooler assembly salvaged from an MSI graphics card. 2. One of those shitty chinese 100W LEDs, soldered directly to the spreader.  3.  For a modest reduction in wattage, I quite like laptop coolers.  This one cost me $2.50 USD on ebay.  4.  My ghetto home made Solder Slab™ of low temp Sn42Bi58 solder, with little nibbles from my iron evident at the edge. 5. Finally, a CPU heatsink actually being used as such, AMD processor soldered directly to it.  Yes, it still works.  It's running FreeBSD as I type this.

[Spikee]

I tried the soldering method on my arctic freezer 7 pro rev 2 but it was a no go.
I have a top of the line soldering iron with around 240W of capacity (JBC) but the heatpipes of the cpu cooler are to good. After a fuckload of time the temperature only achieves around 60-80 deg C.

I have tested the Arctic silver thermal adhesive up to 600W dummy load using 4 mosfets and it works fine. The only hassle scraping it off when you don't need it anymore. Just normal thermal compound and a press fit heatsink is the best solution.

[Jeroen3]

I have a top of the line soldering iron with around 240W of capacity (JBC) but the heatpipes of the cpu cooler are to good. After a fuckload of time the temperature only achieves around 60-80 deg C.

As described above, you must use the heatpipes in reverse by heating up the fins using hot air. All energy will channel and concentrate back up to the heatspreader plate. Making it incredibly hot very fast.

It's quite interesting to test if soldering is a lot better. Intel recently changed from solder to glue between the heatspreader and the chip die. This seems to have decreased the performance.

[microbug]

@Jeroen3 Yes, I heard the same story about solder being better.

I can see that solder would be better once it's there. How would the thermal connection between the cpu cooler and the heat spreader be when soldered together? Also, would this be doable with the Atten 858D hot air gun and Hakko FX-888D soldering station (my soldering equipment)?

@Joenuh, multiple gains on the op-amp is one way of doing it. It would make sense as you are unlikely to need high resolution at high power. The reason, however, that I am not using 1 ohm shunts - apart from the heat, which has to go somewhere - is that in constant power mode with a current source of 20A (the max current) connected to the input, they would limit the minimum power to 200W (I²R = 400*0.5 = 200). On a 300W load, that's almost useless.

EDIT: No reply yet from IXYS...

[metacollin]

Thermal paste is often still adequate, but will up your over all thermal resistance and you'll have less headroom.  It just seems like once you've gone through all the trouble of mounting something with sufficient force, drilling a hole, cleaning the surfaces, blah blah, you could have had a much easier time just using the soldering method.  I am really not sure why its not used more (in prototyping/personal project contexts, I am clueless about how well it would scale to some sort of volume production setting, probably not well though).  I started doing it simply because it was *much* easier and quicker. 

Yes, it's totally impossible to get it to temp using an iron.  I use a $20 1500W heatgun from amazon.  Hot Air is terrific for heating any kind of surface-area heavy thing like a heat sink.  The radiator fins suck up the heat if you blow hot air between them like crazy.  All I'm saying is anyone interested should give it a try, and especially if you get ahold of or make some low temp solder, and you'll never go back.  I used to drill and tap stuff and mount it the classic way, but haven't done that in ages, simply because I'm, ahem, a convenience enthusiast and the soldering method got the most done for the least amount of effort. 

But seriously, I've yet to encounter a heat sink large enough that a cheapie heat gun couldn't bring up to temp with relative ease.  Also, it's a lot easier to hit 140° than 180° or higher for lead-free SAC alloys. 

[microbug]

@metacollin, OK. I'd probably have to buy a better heat gun if I connect the heat spreader to the cooler with solder, since the Atten 858D 'only' does 700W.

[Kevin.D]

That's a problem that I'm having. I'm thinking of having multiple gain options on the opamp so that a switch can switch between different feedback resistors of the opamp. That switch could be something like a ADG884.

Theres a couple of ways you can have a multiple range load without any switching .
I did one a year or so ago it's got 4 decade ranges 12A ,1.2 A ,120mA ,and 12mA
on the lowest range you get 10uA adjustment precision . I designed it to be used as a current sink/source as well as use as a eload hence the four lower decade ranges (and it has true reverse polarity protection which is a must to protect the DUT) .It's also fully protable so I can take it to the garage outside where I use it on large lead acid batt capacity tests.
Search forum for 'eload' and you should find it

[microbug]

I had some trouble finding it, so here's the link:
https://www.eevblog.com/forum/projects/fully-portable-150w12a-e-loadcurrent-reg-with-multiple-ranges/

EDIT: I think multiple ranges are a good idea. The end project will be digitally controlled, so I'll use analog switches (such as AD884) to control the range. I have dropped the INA194; the input offset voltage was too high. The 6uV of the OPA4188 should be much better .

EDIT 2: I have been looking into the DAC some more. The DAC8411 uses the supply as the reference, and as it only uses 160uA (max), I'll power it from a 4.096V reference.

EDIT 3: This is better because I can make the current ranges 409.6mA, 4.096A and 40.96A (that one will be software and fuse-limited to 20A) and have adjustment with a resolution of exactly 6.25uA / 62.5uA / 0.625mA (instead of a long decimal number). Hardware constant-voltage ranges will probably be double the values of the current ranges (819.2mV, 8.192V, 81.92V, maybe 819.2V (software limited to 100V, of course!)).

EDIT 4: I'm going to use 100mOhm shunt resistors. With 21mOhm MOSFETs, that's a 61.5mOhms impedance altogether (MOSFETs and shunts are in parallel). If I assume the actual impedance will be 70mOhms (accounting for connections, traces etc), then the minimum power dissipation (I^2*R) is around 30W at max current. Graph attached - x is current, y is minimum power dissipation.

[microbug]

Bump (I often edit my messages rather than replying, so see the last message I posted).

What should I do to generate a small (250-300uV) but stable negative offset for the current/voltage set line (do I have to use a negative voltage reference or can it be something cheaper)?

I'll make the constant current ranges 400mA, 4A and 40(20)A so I have some headroom to trim the DAC in software.

The minimum constant current will still be 1mA (or maybe 500uA), even on the 400mA range, because of the offset.

[rob77]

i strongly suggest a negative rail for your op-amps in the design and strongly suggest to use op-amps with offset compensation pins (all old-school precision op-amps have those available - pins 1+8 if it's a 8 pin package) and trim down the offset to zero.
otherwise forget the accuracy you're trying to achieve

btw... i'm always nulling the offset before i solder the op-amp in - setup a non-invert amp with gain of at least 5000 (non invert input grounded) and tweak the offset to zero. then solder the op-amp into the final board and either solder the 20turn trimpot with it or replace the trimpot with resistors for the final board.

[microbug]

There will be a negative rail for the op-amps, generated by an inverting switching controller. I don't know if an 'old-school' trimmable op-amp will be better overall than an OPA4188 - it has a very low offset anyway and the drift over temperature is extremely low.

EDIT: Do you have any specific suggestions for a better op-amp?

[rob77]

actually those "zero-drift" opamps from TI look pretty good, the thermal stability is impressive. but i'm afraid the 25uV offset is too much for your 16bit DAC. datasheet states 6uV typical , but you know you have to consider the max value which is 25uV but anyways - you can compensate for that even if there are no offset trim pins - but it's less elegant than a trimpot

btw... if you'll use a 4.096V reference , then why not to use a 12bit dac ? 16bit is a overkill - you can't build the circuit stable enough anyways... your opamp will be zero-drift ,but the resistors not someone will sneeze in the room and the whole thing will drift away - rendering the extra 4 DAC bits useless

[microbug]

@rob77, I guess you're right. Besides, the current/voltage set resolution will still be perfectly good - 100uA on 409.6mA; 1mA on 4.096A and 10mA on 40.96A. If I go with a 12 bit DAC, I can use the built in 12 bit ADC on the PSoC 4, both saving components and improving transient response for software constant-power/constant-resistance.

[microbug]

I found a good DAC: DAC7613. This will let me use the REF2041 as timb suggested!

[microbug]

@timb: how should I negate the offset voltage (nominal +/- 4LSB) of the DAC buffer? You mentioned that it's easier to use a 'positive offset voltage' - what do you mean?

EDIT: Unless there's a better way, I'll use a resistor divider between 2.048V and ground to generate 2.045V (6 LSB down), and trim in software.

EDIT 2: I attached my attempt at a negative offset voltage for the DAC output. Will it work?

EDIT 3: ... or would I be better off having the same voltage divider on the VREFL and VREFH inputs?

EDIT 4: Since the range of the DAC is now effectively 0-2.048V, it makes more sense to make the current ranges in multiples of 10 of 2.048V. Therefore, I will have current ranges of 204.8mA, 2.048A and 20.48A.

[eneuro]

I have a top of the line soldering iron with around 240W of capacity (JBC) but the heatpipes of the cpu cooler are to good. After a fuckload of time the temperature only achieves around 60-80 deg C.

That is why I'd like to try spot weld mosfets TO220 tap  in 3-4 spots to copper heatsink or flat heatpipe.
This should prevent cooling spot area, but of course high energy short pulses needed, so there will be no time to disipate heat.
Just wondering if 30A on spot welder trafo primary 230VAC, so close to 7kW will do the trick, while copper has quite low resistance, but cross section area of those spots will be much lower than transformer a few turns secondary, so maybe? 
Anyone tried this?

[macboy]

I have a top of the line soldering iron with around 240W of capacity (JBC) but the heatpipes of the cpu cooler are to good. After a fuckload of time the temperature only achieves around 60-80 deg C.

That is why I'd like to try spot weld mosfets TO220 tap  in 3-4 spots to copper heatsink or flat heatpipe.
This should prevent cooling spot area, but of course high energy short pulses needed, so there will be no time to disipate heat.
Just wondering if 30A on spot welder trafo primary 230VAC, so close to 7kW will do the trick, while copper has quite low resistance, but cross section area of those spots will be much lower than transformer a few turns secondary, so maybe? 
Anyone tried this?

Spot welding will give you a good (and irreversible) mechanical connection to the heatsink, but what about the thermal connection? You will still have air gaps in the imperfections between the device tab and the heatsink. A metal-air-metal thermal interface is very poor. The soldering technique provides a metal-metal-metal (tab-solder-heatsink) thermal interface, which is as close to ideal as you can practically get.

[eneuro]

Spot welding will give you a good (and irreversible) mechanical connection to the heatsink, but what about the thermal connection? You will still have air gaps in the imperfections between the device tab and the heatsink.

Finally found tips how to spot weld those tabs 
How to Spot Weld Copper

"Spot welding copper requires special electrodes and brazing paste in order to achieve the same sturdy welding job that can be more easily achieved with metals such as steel."

Extracted recipe from this link below:

"Apply brazing paste in between the two points to which you will apply the electrodes of the spot welder."
"Lower the electrodes of your spot welder and pinch the two pieces of copper you will spot weld."
"Activate your spot welder for quick bursts, turning the welder off and removing the electrodes after each burst."

This is what I was thinking about and it looks like that Silver Solder Paste could be used for brazing copper like in this video using classic high power heater

I'd like to solder heatsink to tap using powerfull spot welder, while I do not like to put mosfets in high temperature stress when heater is used during assembly, so it is time to try those tricks above and see what happends if high energy short pulses will be apllied to TO220 mosfets taps

Note: Just checked SSQ-6 Silver Solder specs here http://muggyweld.com/ssq6-silver-solder-paste and it looks like it

"Melts at 1050° F"

which means
1050ºF = 565ºC 
Maybe classic solder pase used in SMD assembling will be better for brazing copper to copper , but it is interesting to test spot weld copper to aluminium heatsink using this SSQ-6.

[Joenuh]

@Microbug: Did you get a reply from ixys? Because I sent an email with a sample request yesterday and I already got a reply.

Also, are those ranges correct? I assume from the calculations that you are using a 4.096V Vref and a 16bit DAC? Wouldn't that give a lowest range of 62.5uA instead of 6.25uA? Or are you using a gain of 0.1, which would make no sense to me?

That resolution would be really hard to get I think, considering input offset voltage and stuff like that. I'm even in doubt about getting my (accurate) 1mA resolution with a shunt of 0.1 Ohm.
16bit DAC would be the way to go I think or would it be overkill and am I thinking to hard about this?

[microbug]

I didn't get a reply  from IXYS, I think the email might not have sent. I'll send another!

I'm now using a 12 bit DAC (you can have an external reference, and it's more accurate) with the REF2041, as suggested by timb. There is a mux which controls the range by connecting the appropriate resistor divider. Moe details to come later!

[metacollin]

@microbug: I use a 1300W heat gun.  But anyway, you've got things under control and in the way you want thermally, sorry to keep talking about much earlier parts of the project.  Can't wait to see more of this project though!

@eneuro and the other guy talking about welding/brazing:

I would discourage anyone from spot welding copper for thermal purposes, and even more so actual IC packages to copper.  CPU coolers have heat pipes soldered to a thin copper plate which is intended to have heat from some watt nugget conducted into it.  It then spreads the heat and distributes it into the heat pipes. Then, very thin copper sheets will be soldered to the heat pipes, and sometimes additionally the copper sheets/fins are skived together to make the outer edges of the fins a rigid structure that is more diffidult to bend.  That is all that is holding together the CPU cooler in your computer, laptop, GPU, etc, and its more than enough.

But its soldered using Sn42Bi58, which is used extensively in thermoelectric applications, mainframes (for its tin whisker retardant properties), and niche industrial applications (sometimes used for large BGA packages, fro example).  It is very strong mechanically compared to electrical solders and has excellent wetting, is eutectic, very low thermal expansion, and fairly importantly in the context of thermal performance, it expands slightly as it solidifies, filling in any gaps and making a very secure thermal and mechanical bond between two surfaces.

You are using heat sink/heat pipe assemblies inappropriately if you require them to be stronger than what the thermoelectric soldering used in every commercial passive or forced air CPU and GPU cooler on the market can provide.  Heat conduction is a function of cross section, spot welding is, uh, not known for the cross section of the welds, and any weld will only penetrate so far.  You could not weld a semiconductor package such that the entire back surface was bonded without destroying the device. And beyond that...why? What benefit does welding or brazing heatsinks to each other offer?  Why use silver filler? Surely not for its thermal conductivity.  If you want heat to move from one place to another you don't wait for it to heat up a bunch of mass in between, watching it spread on timescales measured in _minutes_ through solid material.  If you want to move heat out of and quickly away from something, ore move it from one spot to another spot, you  use a heat pipe. Heat pipes transfer almost all heat energy to the 'cold' end in a couple seconds.  The thermal conductivity of heat pipes used for CPU cooling is in excess of 100 times that of copper. 


Anyway, I'm done rambling about heat.  and using a CPU (or GPU for even phatter watts) for a dummy load is a great idea, and the heat pipes they use will keep the actual MOSFET junctions at much lower temperatures as long as your airflow is removing heat as fast as they make it.  And, surplus server coolers, for xeons and the like, which usually need to shed 125W+, are both abundant and cheap on ebay.  I'm eager to see how things develop with the digital control and DAC. 

[macboy]

Heat conduction is a function of cross section, spot welding is, uh, not known for the cross section of the welds, and any weld will only penetrate so far.

I agree and tried to make this point earlier. Spot welding a device to heatsink is a non-starter idea for many reasons, especially the one above.

[microbug]

@metacollin, I'm glad you're interested!

I'm using an ADG715 for the analog mux rather than the PSoC 4; the PSoC 4's 400 ohm resistance makes it unsuitable (the ADG715 has a resistance of 2.5 ohms).

EDIT: Any ideas for removing the op-amp / DAC offset? I'm using the REF2041 to bias the op-amp at 2.048V, and the DAC outputs from 2.048V to 4.096V.

[Jeroen3]

Socket 775 is relatively small compared to the newer ones. If you get some old Pentium 4 heatsinks for servers with heatpipes you'll also have somewhat compact model.
Like these: 66h080000-046.
The 3 heatpipes can all serve 1 TO-220, with a total of 3, or 2 and an LM35 in the middle, not sure if there is need for copper filling plates. Awaiting dimension spec from oem.

Does anyone know of temperature sensor ic's in TO220 that do not have the tab connected to anything?

[microbug]

@Jeroen3, unless I'm wrongly estimating the size, that heatsink won't get close to 300W.

SeanB said this:

Those copper rods are going to make poor contact, so you probably want the 3mm copper plate, machine it smooth on the faces and use the thermal epoxy to hold it and the devices together. You will not get good contact onto those flattened pipes without thermal epoxy or a good thermpad.

I think he probably knows what he's talking about!

EDIT: For the MCU, I'm going to use a PSoC 4 chip (not the PSoC 4 dev board). I can use the dev board to program the PSoC 4 with a bootloader that will allow firmware updates over USB (there's going to be a USB to UART chip too).