high voltage e-load

[Christopher]

For some test gear at work, i am designing a constant current e-load.

I have designed low voltage ones successfully using the age-old mosfet and opamp.

Now I have a challenge... A variable-voltage (800-1380Vdc), maximum output power 305W load.


I have to present my ideas next week.

My current idea is to use three of these:

http://uk.mouser.com/ProductDetail/IXYS/IXTK8N150L/?qs=cvHLLyFtoE1YaOTgCSHAQw==

with the driver circuits in parallel, MOVs across the output, big-ass fuses in series, do some thermal analysis on the heatsink with a changing fan speed dependent on temp.

Things that may become a problem I have found:

• What if Fets go short
• Opamp with capacitive drive and high output voltage may be hard to find...
• 100W per device may be pushing it a bit...

Any ideas on potential problems I may have before I put my ideas forward to three chief engineers next week?

Any thoughts, no matter how small or abusive will help me massively. Many thanks x

[ejeffrey]

That link seems to be broken, but I assume you have a high voltage MOSFET.

In a multi-transistor load, give each FET its own sense resistor and opamp.  That way you force equal current (and therefore power) sharing.

You can boost the drive voltage and current of an opamp with an external power stage, but I doubt you need it.  Use a chunky BJT based opamp and you should have no problem.  The NE5532 for one can supply plenty of current and operates off of a 30 volt supply.  It isn't rail-to-rail so you need a split supply to go down to zero current, but you could run it off of +25/-5 supply if you needed that last few volts to drive the MOSFET.

If a FET blows short circuit, you find out if you sized your fuse appropriately 

[krish2487]

I suggest you try using one of these.

http://www.semikron.com/skcompub/en/productsearch.htm?include=suchergebnisse_technologien.html&group=tech001&filter=%7B%2212%22%3A%5B%22i7%22%5D%7D&gf=Voltage+%28V%29%3A1700%3B&gg=IGBT+Mosfet

Pick one with the single switch.

at 800 - 1400 V, you are looking at 0.25 - 0.5 A of current through the IGBT.

Just make sure, there is adequate heatsinking for the module. Most of these are designed to work as a switch, and not in linear mode.

You will need to go overboard in thermal design and considerations.

[mikeselectricstuff]

If you only need to operate over a limited voltage range, look at dumping some of the power in one or more resistors

[Marco]

• What if Fets go short

If you are really scared of this, how about redundancy? Just have an extra short circuited rated IGBT on the bottom, normally full on but slammed off in case of a short circuit, will be faster than a fuse (but obviously not a complete replacement).

PS. damn, never really took a look at the SEB failure rates for high voltage devices ... this is not as unlikely a failure mode as I expected.

[mzzj]

I suggest you try using one of these.

http://www.semikron.com/skcompub/en/productsearch.htm?include=suchergebnisse_technologien.html&group=tech001&filter=%7B%2212%22%3A%5B%22i7%22%5D%7D&gf=Voltage+%28V%29%3A1700%3B&gg=IGBT+Mosfet

Pick one with the single switch.

at 800 - 1400 V, you are looking at 0.25 - 0.5 A of current through the IGBT.

Just make sure, there is adequate heatsinking for the module. Most of these are designed to work as a switch, and not in linear mode.

You will need to go overboard in thermal design and considerations.

I would not bother with those big IGBT modules. The current sharing among parallel dies inside those is horrible in linear use.
600V 400A IGBT module can get fried with 200W power loss in linear use. (one of the igbt dies gets in thermal runaway and hogs all the current.)

I would probably go with something like  the IXYS fet already mentioned. Maybe something bit smaller and 5pcs instead of 3. 
And remember that you can not connect mosfets in parallel for linear use. So better have active current control invidually for each of them.

[T3sl4co1l]

You can connect MOSFETs in parallel just fine, if you use sufficiently large source resistors for each one.   Same for BJTs.

IGBTs are not recommended for DC use (check the SOA in the datasheet... if they even give one, eh?).

For that matter, the IXTN8N150L aren't either; they only go up to 1kV.  No idea if they'll go higher, or what happens if you try.  But anyway, 1500V isn't much headroom over a 1300V supply.  I suggest dividing it in half, by making a cascode with more of them (they can be 1000V devices instead).  Also, use resistors wherever possible to save volts.  Every watt you don't have to dissipate in a transistor means cheaper transistors and less stress on them.

Tim

[mzzj]

You can connect MOSFETs in parallel just fine, if you use sufficiently large source resistors for each one.   Same for BJTs.

IGBTs are not recommended for DC use (check the SOA in the datasheet... if they even give one, eh?).

For that matter, the IXTN8N150L aren't either; they only go up to 1kV.  No idea if they'll go higher, or what happens if you try.  But anyway, 1500V isn't much headroom over a 1300V supply.  I suggest dividing it in half, by making a cascode with more of them (they can be 1000V devices instead).  Also, use resistors wherever possible to save volts.  Every watt you don't have to dissipate in a transistor means cheaper transistors and less stress on them.

Tim

True, big source resistors do fine in this case because there is plenty of voltage to waste and not much current.

[NiHaoMike]

What about a series string of incandescent bulbs PWMed at a relatively low frequency?

[woodchips]

Ok, a silly suggestion, but isn't this something that a multiple kW transmitter valve, tube, vacuum electron device, could handle with both hands tied behind its back? No problems with voltage, power sharing, drive levels? Companies like E2V (the old English Electric Valve Company) in the UK still make these parts, and I am sure there are many others.

Would also somewhat startle the engineers you are presenting it to!

[David Hess]

Ok, a silly suggestion, but isn't this something that a multiple kW transmitter valve, tube, vacuum electron device, could handle with both hands tied behind its back? No problems with voltage, power sharing, drive levels? Companies like E2V (the old English Electric Valve Company) in the UK still make these parts, and I am sure there are many others.

Would also somewhat startle the engineers you are presenting it to!

For a rugged design where a fault would be catastrophic, I would certainly consider this.  An Eimac 75TH will support 2000 volts, 75 watts, and 225 milliamps.  Several in parallel would be needed but only one would be needed with a more modern tube.

See page 7 of Linear Technology application note 2:

http://www.linear.com/docs/4099

You can connect MOSFETs in parallel just fine, if you use sufficiently large source resistors for each one.   Same for BJTs.

MOSFETs suffer from a problem similar to bipolar secondary breakdown at high voltages where their temperature coefficient reverses causes thermal runaway compromising the current sharing between cells on the same die.  This is especially a problem with vertical designs which are pretty much all that are available.  Check the specifications carefully and do not forget to derate based on operating temperature.

I suggest dividing it in half, by making a cascode with more of them (they can be 1000V devices instead).  Also, use resistors wherever possible to save volts.  Every watt you don't have to dissipate in a transistor means cheaper transistors and less stress on them.

I would use a cascode anyway with a relatively low voltage MOSFET controlling the current.  That will lower the input capacitance and mitigate the effects of reverse transfer capacitance which will be serious at high voltages.  A resistor in series with the output will help as well but the breakdown voltage limits of the resistor itself should be considered.

• What if Fets go short

If you are really scared of this, how about redundancy? Just have an extra short circuited rated IGBT on the bottom, normally full on but slammed off in case of a short circuit, will be faster than a fuse (but obviously not a complete replacement).

PS. damn, never really took a look at the SEB failure rates for high voltage devices ... this is not as unlikely a failure mode as I expected.

Instead of placing it on the bottom, I would investigate using it or a thyristor as a crowbar on the power supply output.

[Christopher]

Many thanks for your replies. I am going to quote all of the interesting points.

That link seems to be broken, but I assume you have a high voltage MOSFET.

In a multi-transistor load, give each FET its own sense resistor and opamp.  That way you force equal current (and therefore power) sharing.

You can boost the drive voltage and current of an opamp with an external power stage, but I doubt you need it.  Use a chunky BJT based opamp and you should have no problem.  The NE5532 for one can supply plenty of current and operates off of a 30 volt supply.  It isn't rail-to-rail so you need a split supply to go down to zero current, but you could run it off of +25/-5 supply if you needed that last few volts to drive the MOSFET.

If a FET blows short circuit, you find out if you sized your fuse appropriately 

IXTK8N150L is the fat I have chosen. It's a high voltage Vds, tested in the linear region.  Of course I will have a seperate gate-drive circuit for each mosfet. I came into thermal runaway problems with the first e-load I designed with 10 TO220 mosfets in parallel (Short answer: It had to be tested yesterday and that was all I had in my drawer).

Would it be better if I use a capaciatve load driving opamp or just stick to something normal with some resistance in series with the gate? Obviously I plan on simulating this with bode plots etc.

If you only need to operate over a limited voltage range, look at dumping some of the power in one or more resistors

I was thinking of dumping some resistance in the drain side of the mosfet, just to drop some volts. Normally we create loads with big panel resistors and FET switches, but obviously this requires calibration, is generally nasty and will not work with a changing voltage.

If you are really scared of this, how about redundancy? Just have an extra short circuited rated IGBT on the bottom, normally full on but slammed off in case of a short circuit, will be faster than a fuse (but obviously not a complete replacement).

I've been thinking a high-voltage FET or IGBT always on with a crowbar type latching circuit on the gate. As well as a fuse. I don't really want these £20 silicon devices blowing up, I don't want to have to change them every 5 minutes :-)...

IXYS does make a few 1.5kV linear fets they aren’t cheap though, cheaper probably to just get 3 to 5 of these. I’ve gotten 80W out of the TO-3P package with a relatively modest heat sink and about a 40CFM fan blowing on it.

For the voltage you can get away with large source resistors and put a 20W or so resistor on the drain of each leg.

That sounds like a good plan to me.. I was thinking the linear-mode fets would be the best option as they are designed to work in that region rather than fully slammed on. With this kind of voltage I don't want anything to go wrong if I can help it!

What about a series string of incandescent bulbs PWMed at a relatively low frequency?

Not really possible, same with valves or transformers. I can see my boss laughing in my face if I propose dumping the power as light .

[T3sl4co1l]

To be fair, a sufficiently large array of incandescent bulbs need not emit visible light at all..

That said, it's worth noting any thoriated filament tube will also light up brightly, so, oh well.  2kV isn't terrifically high for a non-lightbulb type, but I think it's on the margins at best.  And even the darkest ones still glow red in places...

Tim

[NiHaoMike]

Paint the bulbs black with grill paint or similar. Now you have a bunch of "heat bulbs".

[ejeffrey]

Would it be better if I use a capaciatve load driving opamp or just stick to something normal with some resistance in series with the gate? Obviously I plan on simulating this with bode plots etc.

I don't think you need it unless fast tuning is important.

If you only need to operate over a limited voltage range, look at dumping some of the power in one or more resistors

I was thinking of dumping some resistance in the drain side of the mosfet, just to drop some volts. Normally we create loads with big panel resistors and FET switches, but obviously this requires calibration, is generally nasty and will not work with a changing voltage.

Just putting the equivalent of a 2000 ohm power resistor in series will cut the worst case power to the transistors by about 100 watts.  If you want to do more, you need to be more clever (different sized resistors loads and a MCU sequencing them)

What about a series string of incandescent bulbs PWMed at a relatively low frequency?

Not really possible, same with valves or transformers. I can see my boss laughing in my face if I propose dumping the power as light .

Incandescent bulbs are by far the cheapest per watt power resistors you can get and they have a nice history in test equipment going back to the HP200, but the non-linear resistance and breakability are kind of annoying in this application.

You could make a buck converter (aka filtered PWM) with ordinary power resistors which would basically drop the FET dissipation to zero.  You could use a single switching FET and wouldn't have to worry about SOA or anything.  If you can tolerate the worse transient response, this is an avenue that might be worth considering.

[richard.cs]

Ok, a silly suggestion, but isn't this something that a multiple kW transmitter valve, tube, vacuum electron device, could handle with both hands tied behind its back?

It would, but it's quite a bit of anode dissipation and wouldn't be a cheap valve. You're at about the top end of where it's reasonably possible with semiconductors, but probably still in the region where they're most appropriate. It's not a silly suggestion and should be given some serious consideration if there's a possibility of voltage transients on the power supply under test. One faulty power supply taking out your test load could ruin your day.

What about a series string of incandescent bulbs PWMed at a relatively low frequency?

Not really possible, same with valves or transformers. I can see my boss laughing in my face if I propose dumping the power as light .

Then your boss isn't an engineer, often light bulbs are a good option for test loads. Cheap, reliable if under-run, and because they dissipate radiatively no messing around with giant heatsinks and fans. Again I don't think they're particularly suitable for this application but in general they shouldn't be so quickly dismissed.

[krish2487]

Then your boss isn't an engineer, often light bulbs are a good option for test loads. Cheap, reliable if under-run, and because they dissipate radiatively no messing around with giant heatsinks and fans. Again I don't think they're particularly suitable for this application but in general they shouldn't be so quickly dismissed.

+1 to that.

Where I work, We use an array of light bulbs to test UPS systems upto 15 KVA using light bulbs. Beyond that heaters and beyond 45 KVA, water load.

Light bulbs are often used in low to mid power power supply testing. Not just AC but even high voltage DC (200 - 400 V DC).

[NiHaoMike]

Maybe a magnetron can be used if the magnets were removed so it behaves just like a regular tube diode?

[richard.cs]

Maybe a magnetron can be used if the magnets were removed so it behaves just like a regular tube diode?

That makes you a nice vacuum diode with an anode dissipation of about 400W, but how would you control it? There's no grid.

[nctnico]

I prefer to use regular transistors for a dummy load. Constant current comes for free from the collector. You only need a temperature compensation scheme. A problem is the safe operation area. Most transistors and MOSFETs are intended for switching to they can't handle lots of current at high voltages. Transistors intended for audio have a much wider safe operation area than switching transistors. In the past I managed to blow up 1000V/100A transistors by running 50V 2A through them 

The suggestion for using resistors is also a good one. They can handle much higher temperatures. I'm currently working on a project which needs 4 1kW dummy loads. I choose to use resistors on a heatsink (with fan). One of the problems with this setup is that resistors have some self inductance so when switching them with high current you'll get some high voltage spiking which needs to be clamped.

[T3sl4co1l]

Maybe a magnetron can be used if the magnets were removed so it behaves just like a regular tube diode?

That makes you a nice vacuum diode with an anode dissipation of about 400W, but how would you control it? There's no grid.

Variac on the filament transformer

[N2IXK]

That said, it's worth noting any thoriated filament tube will also light up brightly, so, oh well.

Not if you use a "modern" metal/ceramic power tube.  This guy will loaf along forever at only 300W dissipation (It will take up to 1200W continuously).

http://www.cpii.com/docs/datasheets/76/3CX1200A7.pdf

Essentially a ceramic version of a 3-1000Z. A rugged, zero bias industrial triode that would last "forever" and take the abuse it might see as a test load.

[SeanB]

Maybe a magnetron can be used if the magnets were removed so it behaves just like a regular tube diode?

That makes you a nice vacuum diode with an anode dissipation of about 400W, but how would you control it? There's no grid.

Variac on the filament transformer

You limit the temperature of the filament so that you operate the tube in saturation, with the current limited by the ability of the cathode to emit electrons. You can get a small range of control doing that, and basically turn it into a close to constant current generator.

[richard.cs]

Variac on the filament transformer

You limit the temperature of the filament so that you operate the tube in saturation, with the current limited by the ability of the cathode to emit electrons. You can get a small range of control doing that, and basically turn it into a close to constant current generator.

Are magnetrons bright-emmiter tungsten cathodes then? If you do that on an oxide cathode you'll destroy it.

[N2IXK]

AFAIK, oven type magnetrons use thoriated tungsten filaments.