250A electronic load

If we need 20 expensive high-performance MOSFETs, let's try to make a good part choice. Here's my narrowed-down parts spreadsheet. Besides power capabilities, price matters. Why spend $200-400, if $50 to $100 might be enough?

Resistors are good at higher voltages, where you can afford the lost low-voltage range. (Say a 400V load capable of full current above 50 or

Mind that large packages are much more expensive, and we aren't talking much silicon here, so you're basically paying for the package.

TO-220 is quite affordable and shouldn't be discounted offhand. It's good for ~50W in typical application. You would need enough at this scale (40) that a mechanical solution will be desirable, e.g. a bent plate or machined bracket that applied spring force to the transistor body.

MAX-247s are nice also; price, YMMV, and they are only spring-mountable. Clips like MAX07NG are cheap and plentiful; it's just a pain screwing down oodles of them.

I wouldn't worry about TO-264, SOT-227, etc. 264 maybe worth leaving in the search criteria but doubtful anything economical will show up.

As for fringe ideas, you could SMT a heat spreader bar very close to (touching, or overlapping, the tabs, of) an array of D2PAKs, which might be acceptable on assembly cost. Don't think it would be any more compact than conventionally assembled TO-220s though, and if you need compactness you're better off arranging water cooling with some bigger devices (perhaps including 264s or 227s where the added cost is justified by the size reduction).

But I digress.

Performance, what's that? I think you will find old school types -- IRFPxxx in particular -- will show up quite prominently in Pd(max)/$ metrics. Die area too, relevant to pulsed operation.

Last time I was looking for high voltage types, FQA9N90C was the cheapest, most powerful (>200W), DC SOA type on DK. And for low voltage TO-220s with high energy, STP50N06 and IRFZ46N were top. (Should be good for something like 7ms on-time at 30V, 10A. Modern transistors only do maybe 1-2ms, if that.)

The FQA9N90Cs I'm using in a DC load:

Which, I have only partially tested so far, and am procrastinating through a combination of lack of need (ah, but I want an electronic load to finish the power supplies I want to test with it--), and shortcomings of the programmable interface (I need to make an adapter card so I can have the controller on the front panel, while connecting to the base:

Tim

Spreading heat from a small area to big - blown - heat-sink is always trick y.

Heat pipes can soak up a lot of heat, and transfer it - as fast moving wate r-vapour - along relatively long copper pipes to quite substantial fan-cool ed heat-sinks for eventual dissipation to the atmosphere.

You do have to make sure that piping doesn't contain any non-condensible ga ses.

Sloman A.W., Buggs P., Molloy J., and Stewart D. ?A microcontroller

-based driver to stabilise the temperature of an optical stage to 1mK in th e range 4C to 38C, using a Peltier heat pump and a thermistor sensor? ? Measurement Science and Technology, 7 1653-64 (1996)

eventually went for this approach, and the guy who got it working had to se t up a test rig to make sure that assemblies that got delivered in really h ad been pumped out hard before the water was put in and the system sealed o ff.

It wasn't anything complicated - it just put in a little heat and made sure that it was being shifted to the blown end of the assembly without there b eing too much of a temperature difference.

I don't like the phase-change stuff. It doesn't squish down thin. The best cooling is using regular thermal grease between the transistor and a very flat piece of copper as the first-level heat spreader.

Agree about multiple sense resistors and an opamp per mosfet.

You may as well dump as much voltage/power as possible in the sense resistors. Heat sink them too. You can get 25 watt DPAK resistors, or stand-up 20 watt wirewounds. Use multiple of those!

We have a couple of Kikusui load boxes, and they are very handy for testing power supplies. They can work in CC and programmable ohms modes, and pulse mode to display supply dynamics. A good load box should be fast and stable and not very capacitive.

Commercial wirewound resistors may be a bit inductive etc, but linear R wire & strip are relatively well behaved. Sure is a cheap way to diss most of that heat.

NT

Thankfully, most all modern current-sense resistors are low-inductance. Tim Williams gave us a tour of his "active" load project, which seems to be based on PWMing massive banks of huge power resistors, plus a subset of linear MOSFETs. But I've convinced myself it makes more sense to do the dissipation mostly with MOSFETs, rather than with resistors. 50 and 100-watt resistors are more expensive than high-power MOSFETs, and much less flexible if you want all-linear w/o PWM. A MOSFET handles a wide range of voltage and current, leting us keep our accurate sub-us current-pulse speed.

As my design takes shape, I'm easing the current-sense resistor dissipation problem with range-switch MOSFETs (2.5mR FDP8860, etc.), plus a 3157 spdt signal switch. E.g., each of 20 banks can handle 25A (500A total), by switching to a 5mR sense resistor. I'll setup the PCB to handle either 5W SMT resistors or TO-220 heat-sink- mounted resistors. Bourns PWR221T-30 are under $2 each.

Here's a drawing of a single bank.

There's a potential shortcoming of my simplified circuit, in that paralleling 50mR R1 with 5mR R1a includes another 2.5mR from Q2's Rds(on), and there's some uncertainty in that value. But an 1mR error in the 52.5mR total creates under 0.2% error in the resulting 4.56mR R1a || R1 value.

I think even that could have been avoided by placing R1a between GND and the bottom side of R1, and using Q2 just to short R1. The feedback of the measured current would then come from (R1+R1a) for lo range and (R1a) for hi range, without a FET resistance. Or is there a special reason for the chosen circuit?

Arie de Muijnck

No, that sounds like a good idea.

Do we really want to burn power in testing? Perhaps, below 100W, or for transient testing it's the quick solution; but nowadays, so is a buck or boost converter feeding a battery, rather than a resistor bank.

If the DUT's source is DC, couple the load battery back to the source and get less drain on it.

If the DUT's source in from the AC line, then Couple the load battery to a grid-tied inverter, sharing the same wall socket as the DUT, and you just get reduced wall current draw for the test or burn-in environment.

The load battery absorbs power and supplies it to cover transients, turn-on, protection and control state machine delays/deviations that may be unpredictable. Doesn't need a lot of capacity.

The net power loss can still be a large fraction of what may have been burnt. There are static losses and noise issues, even when nothing is hooked up, but the hardware costs are probably comparable, and the battery/inverter can multiply function as a charger/UPS.

Nailing down the commodity parts for a safe grid tie is probably the biggest stumbling block, but the recent growth in DIY PV should make things easier. If you've already got one, then its battery bus could be an invaluable two-wire addition to your test bench, rather than a lot of heatsinks and fans.

RL

Does it do any good to put a heat sink on the topside of a TO-220 as in, clamp it between heat sinks. I know it's plastic and has poorer heat transfer, but I still wonder.

Mikek

We rarely load-test a gadget longer than a few thermal time constants. Transient response testing can be done in minutes. Why go to all that trouble and expense to recycle 15 cents worth of electricity?

Who needs a bench full of charged batteries?

This power supply test took about an hour:

It ran at 500 watts for maybe half an hour.

Last time I did something like this, I brought in the control signal differential. That just added one resistor quad-pack.

You could lay out a smallish PCB with a heat sink and fan all built in, with fastons for the high-current pins, maybe optional bus bar ties. Use as many of those as needed.

I like DPAK power resistors. Caddock MP725 and Riedon PFC are cheap and nanoseconds fast, 25 watts heat sunk. I have TRDs somewhere.

It may not help much, but I'm contemplating designing the PCB for several build possibilities, one of which is small standing-up heat sinks. Dual heat-sink mounting of Q1 and Q2 makes sense, because only one will work hard at a time. There are a number of attractive choices available.

Another architecture:

Several, at least 4 but maybe more, little fet-opamp-senseresistor circuits scattered around a smallish board with a heat sink and a fan. Possibly a heat sink per fet. Enable the identical current sink circuits as needed, on one board or many.

Small fets are cheap. Heat sinks are expensive and develop local hot spots. High currents are hard to handle on a PCB. Just spread out a bunch of cheap parts electrically and thermally.

We have three class-D amps here, 250 watts per, each with its own fan, ribbon cable and fastons.

Last year I bought a birthday present, a Yokogawa WT310E digital power meter. This instrument takes 0.1% power measurements on waveforms of any type. It's especially well suited for accurate efficiency measurements on things like the inefficient Meanwell power supplies. Taking such measurements requires a high-power load. Like John, we have collections of big power resistors well-suited for such a task. But the lashups are a nuisance, and not well suited to quick flexible changes in load, so you take what you can get by re-wiring resistors. Which usually means one or two max-power tests. I'm looking for better capability, like my Kikusui electronic loads.

As for operating duration, it'd be nice to run for an hour at full power, but 30 seconds may be long enough for transient tests, and five minutes to do a suite of efficiency measurements. Thermal mass is our friend. My plan is to design a PCB that can be built up to cover all the possibilities.

Also, we can make a say 2kW heatsink, but whoa!! How to get the heat out of the heatsink. Our lab has big circulating-fluid chillers that can handle that, if we ever need to implement that capability.

Oh-- just to clarify, it's a unary power DAC. Paired with a 10-bit unary ADC, it's the perfect [approximate] solution; no (repetitive) switching necessary. Or, it would be if the ADC weren't obsolete now. ;o) (A little thought should be able to determine which "ADC" I used, and exactly how it's connected. Nice feature, that!) (But I thought of that, and the resistors can be controlled via MCU as well.)

Then the linear sink simply fills in the gaps between bits.

If you're concerned about dynamics, beware that the ESL of those leaded parts will set a pretty serious upper limit on performance. Enough I think that it can dominate over MOSFET capacitance. This can be compensated out to some extent (put an RC on the feedback sense, thus making the opamp overshoot just enough), as long as enough voltage is available to do so.

Incidentally, a switching load would be an interesting approach, and needs only one resistor and a converter; it has the downside of awful dynamics, because the EMI filter can only be designed for one fixed impedance while the equivalent DC resistance is supposed to be variable over a huge range. The further mismatched those impedances are, the worse the filter performs, in terms of both attenuation and cutoff frequency. Would be good for slow stuff like battery testing, maybe motors, but definitely not supply transient testing.

Tim