Hi,
what MOSFET (or several of them) would be good for an electronic load rated for limiting values 650V/50A/1kW (whichever applies first)? Purely linear operation -- I remember you were saying that ordinary switching FETs aren't good for that purpose. Would SiC FETs work? I like both their high voltage and low R_DS_ON for high-current mode.
I think that the best backage would be TO-247.
Best regards, Piotr
The actual SiC chips are tiny, so can't get rid of much heat. I'm using C2M0280120D in the giant TO247 can, but it can only dissipate 60 watts.
A good silicon fet in a similar package is rated for a kilowatt.
Most switchmode fets are bad as linear amps, especially at high voltages. Check their SOAR graphs. You'd probably be better off using a lot of lower-power rated fets scattered around a heat sink, rather than a few monsters. For linear amps, we just tested a bunch to destruction.
There is some trick where stepped power supplies are used, so no fet is pushed into the upper-right of its SOAR curve. Full-bridge helps there too.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
It?s all about the Spirito effect and keeping very conservative SO A
Lots of stuff out there, but many data sheets does not mention linear mode
Maybe select a NPN instead?
Cheers
Klaus
The issue of concern is R-theta-JC, the thermal resistance from the junction to your heatsink (you have to get rid of the heat once it's in your heatsink!), and our 3-page n-channel power MOSFET table in AoE III, Table 3.4b, page 189, is a good start. Scan the Pdiss and R-theta-JC columns, while eyeing the cost column. The IXFH50N60P3 looks good, 1040 watts, $7.40 at D-K.
Check the "year of intro" column, notice that older devices are better. The newer ones use tighter V-groove line widths, and therefore have smaller dies with low capacitance, better for fast switching, but their small die area (saves manuf $$) is dramatically worse for heat flow.
Often it's best to divide up the current, see section 3.6.3. But ballast resistors use up too much power, smaller sense resistors with active balancing is better, Figure 3.117b, page 214, and it's easy (even better at higher voltages, use series wiring). You might find a pile of FDP7N60NZ at $1 each, makes sense, especially in its smaller TO-220 package.
You may find the best result with a physically-larger package, but if you examine Rth for the different package choices for a given die, you'll see that TO-220 can do as well as TO-247. It's the vertical path of silicon-to-metal frame that counts.
The huge SOT-227 package, with large dies and insulated metal heat surface, can do well, but it's costly. Check eBay.
I have continued to update my huge MOSFET table, now up to 1160 devices. If you can't find a part to your satisfaction, let us know and maybe I can find one in it. Hah, we're days away from turning in the manuscript to our new x-Chapter book and I'm debating now whether to include this massive table as an update to Table 3.4b.
--
Thanks,
- Win
A way to calculate Spirito SOAs:
[snip]
It's really not about Rth, but about the gain of the individual trench cell. If one cell has different gain, you get thermal hotspot and it runs off from there.
We had a circuit where a big FET derated to an order of magnitude below the SOA would fail (we got confirmed data from the manufactor)
You need a device that is MADE to handle linear loads
Cheers
Klaus
For a 1kW system, there should be many parts. After applying the various deratings and considering insulator and heat-sink realities, probably at least 8 to 12 parts would be wise. In this way, 400 to 1kW parts can be used at below the 100W level, good for SOA. Furthermore, I suggested that for high voltages, multiple parts are best deployed in series stacks, e.g. four in series, even better for SOA. Hi-Z MOSFET gates make this easy. With a sufficiently conservative design, well-chosen ordinary power MOSFETs should be fine, working far below the SOR ratings. With this approach, it then becomes all about the summed Rth. I think it would be painful to use fewer specialized parts, closer to a manufacturer's tested and guaranteed SOA curve.
--
Thanks,
- Win
I like how they give a bit of theoretical justification, then provide plots that both confirm and deny the theory.
Why do they not simply say "follow the SOA diagram for the specified time scales"? Are they actually wrong? Manufacturers seem to put in the steeper segments when they have to, though...
Tim
-- Seven Transistor Labs, LLC Electrical Engineering Consultation and Design Website: https://www.seventransistorlabs.com/
Yes, that was cute. But it was for special Hi Rel devices, trust us, they were saying. I think the answer is to stay far, far away from the SOA limits at the higher voltages. Klaus' example of trouble 10x below the SOA limit must be for high voltages, surely that's not a problem below 30 to 100 volts? I certainly haven't seen any such trouble.
--
Thanks,
- Win
SuperJunction FETs all seem to have DC SOA. That seems remarkable, but it also seems to be true.
They also act kind of like IGBTs, in that short-circuit current saturates to a fixed value independent of Vgs (as long as it's high enough to begin with). Something to watch out for, for pulsed applications.
Last time I went looking, FQA9N90C had the best $/W in TO-3P (close enough to TO-247), and happens to handle the voltage you're looking for. I haven't done a qualifying test yet, to verify if the SOA is what is claimed.
Consider using an array of resistors to handle the bulk of the load (a power DAC as it were), and a few transistors to handle the leftovers. This greatly saves on expensive and bulky heatsinking, and further improves SOA use. For an operating range quite that wide, you'll need a few ranks of resistors, or just run everything through transistors anyway.
The absolutely tiny power limit doesn't make any sense to me, and will make precise control more difficult.
Tim
-- Seven Transistor Labs, LLC Electrical Engineering Consultation and Design Website: https://www.seventransistorlabs.com/ "Piotr Wyderski" wrote in message news:q5grs8$h1k$1@node2.news.atman.pl... > Hi, > > what MOSFET (or several of them) would be good for an electronic load > rated for limiting values 650V/50A/1kW (whichever applies first)? Purely > linear operation -- I remember you were saying that ordinary switching > FETs aren't good for that purpose. Would SiC FETs work? I like both their > high voltage and low R_DS_ON for high-current mode. > > I think that the best backage would be TO-247. > > Best regards, Piotr
I don't think I've seen any with SOA problems at quite that low voltage, although in that range, I'm mostly looking at SMTs anyway, where the power limit is quite low to begin with (10s W, except D(2)PAKs which are irrelevant at high power anyway).
Tim
-- Seven Transistor Labs, LLC Electrical Engineering Consultation and Design Website: https://www.seventransistorlabs.com/
I wonder if resistors could be used to take some of the extreme power dissipation peaks off the mosfets.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
With 1 kW worst case dissipation, at 650 V the maximum current is 1.5 A and the required load resistance 430 ohm, at 50 A the voltage drop is 20 V and resistance 0.4 ohm is required.
Use fixed resistors in series, with resistances 0.4 : 0.8 : 1.6 :
3.2 ohm and so on. Each resistor should have a worst case power handling of 1000 W. Use switching transistors across each resistor and control which resistors are shorted out. Optoisolators may be needed to drive the gates with a binary code.3 kilowatt DAC?
-- John Larkin Highland Technology, Inc lunatic fringe electronics
a lot of lower-power rated fets scattered around a heat sink, rather than a few monsters."
Doesn't that cause current sharing problems ?
or you could double all the resistances and connect them in parallel instead of series, this would simplify the gate drive, and possibly allow smaller resistors in some slots.
-- When I tried casting out nines I made a hash of it.
Sure, you have to force sharing to optimize the silicon. I use one opamp per fet, which is good since you need gate drivers anyhow.
When one is dealing with kilowatts and big expensive heat sinks, the cost of good controls is trivial.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
The technique developed and use by Agilent / Keysight, and discussed in AoE III, Figure 3.117, page 214, is more useful than an opamp. They take the MOSFET system control voltage, and make individual gate-voltage changes, so all paralleled MOSFETs are running at the same current. All it takes is a small sense resistor and BJT for each MOSFET, as many as you have, and a modest current sink. The latter can easily work at high voltages, 300V no problem. Hard to do with opamps. The transistor differential error amplifiers have high gain, so modest voltages, e.g., 100mV can be used for the sense resistors, saving power.
At high voltages, series MOSFET wiring makes more sense, see Figure 9.111, page 697. They'll all run at the same current. Both methods are very inexpensive.
--
Thanks,
- Win
In parallel, you need to vary the conductance from 2 mS to 2.5 S.
Put the resistors into a can filled with transformer oil. For short tests, a small can can handle several hundred watts for a few minutes (e.g. 50 ohm oil filled RF dummy loads). For continuous dissipation, you need a larger can (drum) with sufficient outside surface area to transfer the heat into air.
You guys are talking PWM into a bank of load resistors? The (older) HP electronic loads we have here are linear, and pretty impressive. But fast PWM switching, with appropriate inductors, should work reasonably-well for many applications. In that case all the linear MOSFET discussion here is irrelevant. Instead you want good MOSFET switching performance, and very fast switches will greatly simplify the inductor issues, and help to make the result more like a linear system. Si MOSFETs are good, but modern fast super-junction MOSFETs may be fine. There's a huge array of choices out there.
--
Thanks,
- Win
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.