DC load using a CPU cooler

[timb]

@Kevin.D These MOSFETs are specifically designed for linear operation in the SOA, so the graphs in the datasheet are pretty accurate for the conditions we’ll be using them in.

[timb]

@microbug

So here’s one idea I had:







The idea here is your FETs would mount to this copper block (50x75x5 [WxHxT] in this example). The hatched area would be where the A30’s heat pipes press against. It would be wise to use thermal epoxy or thermal compound here. The area on the board the copper slab will mount should be a solid copper pour and have holes drilled so the FETs and slab can be bolted down. Obviously you’d still need to use the heatsink mount itself to keep that stabilized on the board.

Copper has such a low thermal resistance, it would be very close to mounting them straight on the heatsink itself. The advantage here is ease of mounting.

I can also design one to work with only two FETs (one on each side).

[mrflibble]

I'll just accidentally drop these links here, and let the coolness factor do the rest.

http://www.mouser.com/new/panasonic/panasonicthermalgraphite/
http://www.digikey.com/product-highlights/us/en/panasonic-pgs/3953

[timb]

Yeah I saw those the other day... I was toying with the idea of placing one on top of the copper block. The FETs and heatsink would rest on it.

Cool stuff. Pun intended.

Sent from my Smartphone

[microbug]

@Kevin.D These MOSFETs are specifically designed for linear operation in the SOA, so the graphs in the datasheet are pretty accurate for the conditions we’ll be using them in.

Phew!

Tim, your idea looks great! I wonder if thermal graphite is conductive, because if not, it could insulate the drains of the FETs. I probably will use two of the previously mentioned MOSFETs, they look like they should cope with the heat dissipation; the heatsink needs space for two TO-247 packages. Spikee's experience is encouraging.

Thanks!

[Kevin.D]

@Kevin.D These MOSFETs are specifically designed for linear operation in the SOA, so the graphs in the datasheet are pretty accurate for the conditions we’ll be using them in.

No it isn't,  the conditons you will be using them aren't the same conditions that those soa graphs are plotted .

IF you look at the data sheet you see those conditons that the sao are taken in is at a fixed case temp .(you will see  T.C =25 ,or t.c =75 or something) .
For a normal use you can't hold the case at some fixed temp unless your using a nice cryogenic cooling fluid like they do when they do these soa plots .
You have to transfer heat from a transistor to a heatsink via a tab .which has an extra
thermal impedance Rth c-s (case to sink thermal resistance ,also any extra thermal resisitance of sil pad's if you use them) .
IF you'd had looked  at the soa plots on the graphs and  noted the various power's being dissipated at the  junction temp of 150  then you would have noticed this
value (150 - case temp) = Rth j-c * power dissipated ( i.e these soa at dc only plot rth j-c * power .they dont include any other thermal resistance).

Being able to use your Rth's to calculate junction temp @ a certain power dissipation (because junction temp is the one that really matters )is really simple to do . And if you can't do that first basic bit then you aren't yet ready to design a multiple mosfet electronic load.

Regards

[timb]

Actually, IXYS has graphs for Tj @ 25, 75 and 150c...

Anyway, I'm not sure what your point is? Obviously you need to know the Tj at a certain power dissipation. Calculating it is (as you pointed out) simple.

My point is that IXYS also provides graphs for a range of junction temps for these MOSFETs, as they're designed to run in the SOA.


Sent from my Smartphone

[mrflibble]

I wonder if thermal graphite is conductive, because if not, it could insulate the drains of the FETs.

By itself PGS is a good electrical conductor, but Panasonic also manufactures sheets with an insulating film.

[microbug]

What should I do for reverse input protection? The max sink current is 20A - I'm planning to have a high-current diode in a TO-247 package on the heatsink pad (heads up @timb!).


Also, what about the binding posts? I could use 4mm banana jacks which are often rated for 20A, but what about those hex-screw ones on the BK precision loads - would they offer a significant advantage?


I'm currently redesigning the power stage.


EDIT: the DAC I'm using is the DAC8411 from TI. I read that most DACs have buffered output stages, and so can't get lower than 10 or so mV to ground. Is the right approach to add, say 20mV negative offset in hardware and to trim in software (the negative supply voltage for the op-amps is -2V)?

[timb]

By the by, made this up quickly in Pixelmator:



The dimensions are loose, based on an image from Arctic’s site that showed the width and height at both angles.

@microbug Using a positive offset is easier sometimes. TI even makes dual output voltage references now that have Vref and Vref/2 outputs; specifically designed for single supply bi-polar applications. For your application you’d use it like this: Say you want a 0 to 2.048 output range. Get a TI REF2040 and use the 4.096 (Vref) output as the reference for your DAC; then use the 2.048 (Vref/2) output to bias your OpAmp.

Now your control range will be 2.048V to 4.096V which is the same as 0V to 2.048V! The other benefit is it will reduce overall noise between the DAC and OpAmp since they’re being powered from the same source.

Anyway, it’s just one approach, but one that I like. =)

[microbug]

@timb, I can't find a part by the name of 'REF2040'! I have already got a DAC8411 and a 16 bit ADC (I forget the part number - will add it later), both of which have built in references. I can replace them if I have to, but I'd prefer not.

Typical zero-code error @ room temperature (according to the datasheet) for the DAC8411 is 200uV - I won't need that much trimming, if any.

[mrflibble]

Maybe he meant the 2041?

http://www.ti.com/product/ref2041

[microbug]

@mrflibble (I like your username!) Yes, that's likely.

EDIT: Of course, it's unlikely that I'll be using the load below 10mA, however; it would be nice to have accuracy right down to 1mA.

FYI, I'm using two INA194s with  .01 ohm resistors for current sensing; the output feeds into the op-amp (one IC and resistor per driver). I can put both current sense voltages through their own resistor (to get the average current sense voltage), and feed it into the ADC - I'll double it in software. A voltage divider will set the full scale current sense range.

I'm going to include a constant-voltage setting in hardware, as I can get the PSoC to act as an analog mux so all that is required is a few resistors (my plan is to divide the input voltage down and compare it to the set voltage using the op-amps). I did look into constant resistance / power modes in hardware briefly, but I don't know whether it would be worth the effort. I won't be limited for op-amps, though; the PSoC has two so I might be able to make use of them.

I put two MOVs which go at 200V or so (haven't chosen the parts yet) across the input for protection. I don't know what to do for protection against voltage above 100V but below 200V though... suggestions?

EDIT 2: For hardware constant-power, an AD633 multiplying load voltage and current would be fine. Constant resistance, which would require comparing either using a digipot (not much resolution) or doing V/I in hardware (difficult - I can't see how it could be done), is harder. I think constant-power would be more useful, anyway.

[timb]

Measure the current with your INAs and the input voltage with a resistor divider using the PSoC's ADC. Now use your MCU to control the current via the DAC. This way you can do CC, CV, CR and CP through software.


BTW, doesn't the PSoC have like a 16-bit DAC and ADC? Why use external parts?


Sent from my Smartphone

[microbug]

Won't the transient response be much worse? I through that was what C2000s (DSPs) were for.

The PSoC 4200 has a 12 bit ADC and a 12 bit IDAC (current DAC), which is not very useful. I chose 16 bit resolution with external parts because they aren't inhibitingly expensive so I might as well; it will let me get closer to my 1mA resolution ideal.

[Joenuh]

I'm planning on building something similar, also with the 1mA resolution. But thinking about that, wouldn't it be easier to use a 1 ohm shunt resistor instead of 0.1 ohm?
A 1mV error on a 0.1 ohm shunt would already be 10mA if I'm not wrong. A 0.1 ohm shunt would dissipate less power but it still seems like a 1 ohm resistor would make more sense, or is that just me?

[rob77]

I'm planning on building something similar, also with the 1mA resolution. But thinking about that, wouldn't it be easier to use a 1 ohm shunt resistor instead of 0.1 ohm?
A 1mV error on a 0.1 ohm shunt would already be 10mA if I'm not wrong. A 0.1 ohm shunt would dissipate less power but it still seems like a 1 ohm resistor would make more sense, or is that just me?

i built one with a 1 Ohm shunt (50W 1R resistor) and even that is not good enough for 1mA with a jellybean opamp (LM324). so i decided to go for a 10mA resolution for my high power load and in addition going to build a new low power load with high resolution.
btw.. frankly speaking - i don't care if the load is 4.0 Amp or 4.003 Amp - for high currents the 1 mA resolution is useless.

[Joenuh]

Well as you mentioned your opamp, a high precision opamp would probably fix that problem. And I agree that for high currents the 1mA resolution is useless. But if it doesn't get to expensive, I would like to build an all-in-one solution that accurately can go from small currents, as 1mA, to higher currents like 5-10 Amps.

Edit: You also mentioned a 50W 1ohm resistor, I hope that is realised by putting multiple resistors in parallel. If not, that also may be part of your accuracy problem. As power resistors are not really accurate/low tolerance. 

[microbug]

The op-amp I am using is the OPA4188 which has a 6uV (25uV max) offset. The INA194 which is measuring current has a max 2mV offset voltage and typical +/- 500uV. I'm thinking about removing it though.

EDIT: Welcome to the forum, Joenuh!

[rob77]

Edit: You also mentioned a 50W 1ohm resistor, I hope that is realised by putting multiple resistors in parallel. If not, that also may be part of your accuracy problem. As power resistors are not really accurate/low tolerance.

i compensate for inaccuracy of the resistor in hardware - both set voltage coming from a DAC (well... PWM with low-pass) and the readout for the ADC are tunable (voltage divider followed by a opamp buffer with tunable gain), so the exact resistance of the shunt is not relevant. and yes, it's a single big-ass 50W heat-sink mounted resistor
the smaller brother of the DC load will have a spec of 0-250mA and will be built with a 4R7 (or 10R) shunt and OP07 op-amps.

[Joenuh]

Edit: You also mentioned a 50W 1ohm resistor, I hope that is realised by putting multiple resistors in parallel. If not, that also may be part of your accuracy problem. As power resistors are not really accurate/low tolerance.

i compensate for inaccuracy of the resistor in hardware - both set voltage coming from a DAC (well... PWM with low-pass) and the readout for the ADC are tunable (voltage divider followed by a opamp buffer with tunable gain), so the exact resistance of the shunt is not relevant. and yes, it's a single big-ass 50W heat-sink mounted resistor
the smaller brother of the DC load will have a spec of 0-250mA and will be built with a 4R7 (or 10R) shunt and OP07 op-amps.

Yes ofcourse that would fix that problem

Maybe it's my own stupidity but why are people using ADCs in these projects? I know, for measurements.. But if you can accurately set the current you don't need to measure it right? Or are you using it to measure the input voltage in some way? To be able to limit in power?

[microbug]

ADCs are useful for software constant resistance / power / voltage modes. For all of these modes, you'll want to easily see the current and voltage flow.

EDIT: Yes, over power limits come into it as well - software controlling a relay, for example.

[timb]

Won't the transient response be much worse? I through that was what C2000s (DSPs) were for.

The PSoC 4200 has a 12 bit ADC and a 12 bit IDAC (current DAC), which is not very useful. I chose 16 bit resolution with external parts because they aren't inhibitingly expensive so I might as well; it will let me get closer to my 1mA resolution ideal.

Nah, your transient response will be just fine. Personally, I’d use the built in ADC and comparator to control an external DAC. Hook your INA up to a single to differential amplifier and use the built-in 12bit differential ADC.

That PSoC is 48MHz with single cycle multiply, so it should be more than fast enough for your needs. You should be able to get mA resolution with a 12-bit ADC and software calibration.

[microbug]

20A/2^12 = 4.88mA

Like rob77, at higher loads I don't care much about milliamp resolution, so are you suggesting a PGA on the DAC (for multiple ranges)? Unless there's a reason not to, I want to use a 16 bit ADC/DAC as I already have the parts, and if I do there's no need for a PGA or anything. I was thinking of doing away with the INA194s altogether and running the sense voltages from the shunts through a couple of resistors, then scaling them for the ADC with a resistor divider.

Schematic / example to come.

EDIT: The other reason that I'd prefer not to do it in software is that the PSoC will be driving a graphic LCD and sensing a rotary encoder / keypad. This will definitely worsen the transient response!

[Joenuh]

20A/2^12 = 4.88mA

Is that the right way of calculating it? Maybe it depends on how your output is. But the way I see it is that which voltage you set on the output of your DAC is the current that will flow, if you use a 1R resistor ofcourse. So wouldn't that be the Vref of the DAC divided by 2^12? So a Vref of 4.096V with a 12 DAC should give you 1mA steps. The problem is now that you can ''only'' go to 4.096A.

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.