DC electronic load 1000V 1A 1000Watt DIY Arduino Project

[oz2cpu]

Arduino Uno with TFT for readout of Voltage and Current and Watt calculation.
Touch panel removed to get the two needed analog input pins.
Input Reverse voltage protection added using diode.
maybe it is an idea to add input fuse too, think about that if you make one.
Note Arduino usb connector is = GND terminal, so remove input high voltage when working with software.
See my web page for more.
http://webx.dk/oz2cpu/dcload1kv/dcload1kv.htm

see youtube video here:
https://youtu.be/rsDLfKjoBAs

[Kleinstein]

The 4 wire sensing for directly from the input therminals is good for voltage, but as shown in the schematics it makes the current read out and regultaion worse. With the rather high voltage and low current  version this is not really a problem, as the current is low and the voltage is high. So it does not hurt much, but it also not needed.  A lower voltage, higher current version would need more care there and may want correct sensing for both.

IGBTs have a notorious poor SOA. So they may fail at rather low power at higher voltage (e.g. 50 mA at 500 V).  The power rating for the used type also is rather low.
For high voltage use I would definitely consider having part of the power from just series resistors. They are much easier with the SOA than FETs or IGBTs.
In a simple version some of the transistors would have an intentional lower threshold and a series resistor. So more current would flow there and the other would only get active when the voltage drop at the resistors approaches the external voltage.

For the regulator stability is gets usually better with a little extra resistance in series to the C1 and C2 capacitors. A capacitor with added loss is more effective to dampen oscaillations as one without.

The temperature regulation is really simple - it depends on the NTC type.

[oz2cpu]

YES you are right about the tricky GND wire, is shared for voltage / current measurement
and not correctly 4 point, in this case the wire is thick and short
the limited adc resolution and low max current, save the day here..

we have been playing with this yesterday, for some practical real lab things, and it is considered super for the job,
we are also super happy about the smooth function of the temperature regulation,

only complain so far is lack of wrong usage protection, like over power, over heat cut off
if this unit is ever used by no brainers, then i should clearly add a cut off system.
but so far i dont own any powersupplies with this much voltage and this much current capeabilities

[T3sl4co1l]

IGBTs?  That don't even publish an FBSOA, let alone a DC curve on it?  Fuckin' yikes...

You haven't tested it at... nearly a fraction of its full ratings (especially voltage), right..?

For high voltage use I would definitely consider having part of the power from just series resistors. They are much easier with the SOA than FETs or IGBTs.

I did one better on my build:
https://www.seventransistorlabs.com/Images/ActiveLoad2.jpg
The 1k 300W vitreous resistors are wired in parallel with the load, as needed; the remainder is then taken up by three TO-3P (FQA9N90C) in linear mode.  The transistors are ballasted by the three smaller (100R 100W) resistors in turn.  This gives a full scale 500V 4A range, with full current capacity down to 100V or so, and reduced capacity at lower voltages.

A little 160x128 display furnishes readout, and some pots control the V/I curve and limits.  It's also serial programmable.

The load terminals are completely isolated from chassis, and in turn, the rest of the circuit, except for the isolator required for the serial port.  The metal chassis is galvanically grounded for safety.

Note the pink brackets: the heatsink is further isolated from the circuit as well as the chassis, to avoid using insulators or heat spreaders on the transistors.  The drains all connect together here.

I need to write up an article on this some day... or maybe make a few and sell them, hah.  Until then, no schematics or code to show off, unfortunately.

Tim

[oz2cpu]

I dont understand your post tim ?
IGBT's run fine in linear mode, in a feedback system like i have made.
and i tested my product to 1kV input, that is 20% under their max specified voltage,
and i tested isolation to chassis to 2kV so i know it is safe,
and the chassis and heat sink is is on purpose earth, for safety,
you are right it is better thermal, to mount the hot items directly on a heatsink, with no electrical isolation at all,
but I liked to try do it this way.

Here is a picture with 1kV input

[Kalvin]

IGBT's run fine in linear mode, in a feedback system like i have made.

Running an IGBT / Transistor / MOSFET in a linear mode may be very stressful for the device. You should check the device data sheet for the FSOA figures.

Check this document on running an IGBT in linear mode and how to determine Forward Safe Operating Area (FSOA):
https://www.richardsonrfpd.com/docs/rfpd/Microsemi%20How%20to%20Make%20Linear%20Mode%20Work.pdf

[Kalvin]

Kleinstein and Tim are probably trying to say that you should use some high power resistors as a load connected to IGBTs collectors, so that the high power resistors will dissipate some/most of the load, and the IGBTs will just regulate the current through the resistors, and dissipate only a fraction of the total power so that you won't violate FSOA.

[JohanH]

IGBTs aren't meant to be used in linear mode (except some exotic devices like audio transistors Toshiba GT20D101/201). But they will work as long as you run them WAY below their SOA.

I use such an IGBT module (Semikron SKM 200 1200V 1380W) in an old DC load that I built 20 years ago, works fine for low power.

[T3sl4co1l]

I dont understand your post tim ?
IGBT's run fine in linear mode, in a feedback system like i have made.
and i tested my product to 1kV input, that is 20% under their max specified voltage,
and i tested isolation to chassis to 2kV so i know it is safe,

It's a technical point, mildly subtle.  The difference between tossing stuff together, and proper engineering.

Question: show where in the datasheet you are recommended to operate in this manner.

Feedback is irrelevant; that's more or less implied by "linear mode".  Anyway, what's done externally, has no effect on internal current sharing within the die itself.

Can you get away with it?  Maybe.  If they don't give an SOA at all, who knows, maybe it's full DC with no 2nd or 3rd breakdown?  Or maybe it blows up not long after the 10µs short-circuit rating.

The point is, you don't know.  It's not a "you idiot, that will guaranteed never work" complaint, it's an information complaint.

(The worst part about this type of argument is, when the person does get away with it, they gain a smug sense of success, not realizing in what sense they have nonetheless failed in the design process, despite apparently succeeding in the build.  Meanwhile, the expert looks like a bit of an asshole to everyone, except to other experts who understand the argument being made.)

At best, you need to do a full SOA evaluation (at least including the area you intend to operate in, and somewhat beyond) to see if they're usable at all.  (22mA means nothing -- you have a 1kV 1A source to properly test this, right?)  And re-test periodically, as the chip design may change over time, or the design moves between different factories, etc.  (The latter only relevant to a production context, of course.)

There are some very good reasons why not to use IGBTs.  General, not specific, mind: an individual design can be tweaked to respond differently.  But it remains true that Vge(th) has a negative tempco, and that IGBTs have the highest power density among Si devices.  The active die area is minuscule.

It's fairly rare to find SOA curves on them, at all; most of them you'll find, are (or were*) very severe in deratings, for exactly these reasons.

Let alone spookier phenomena, like parasitic latchup.  But that's been largely solved for decades, I think IGBTs since the (late?) 80s have always addressed that.  You still see lots of books/appnotes repeating that...

*I have seen newer, SuperJunction type IGBTs, not only with SOA curves shown, but the full DC curve at that.  And it's not like they're doing internal ballasting (extra (distributed) emitter resistance), Vce(sat) is still attractively low.  I haven't read anything explaining this remarkable property.  SJ MOSFETs benefit similarly, it seems; I don't think I've seen one that doesn't offer full DC SOA.  But again, IGBTs operate on a completely different mechanism, and there's no reason to suspect this behavior would carry over to them.

For the record, IRG4PH20K is not this type; a contemporary MOSFET of similar rating would have significant 2nd breakdown limitations.

Tim

[T3sl4co1l]

Kleinstein and Tim are probably trying to say that you should use some high power resistors as a load connected to IGBTs collectors, so that the high power resistors will dissipate some/most of the load, and the IGBTs will just regulate the current through the resistors, and dissipate only a fraction of the total power so that you won't violate FSOA.

Oh yeah, there's another point of interest I didn't mention: if/when the transistors do eventually fail (whether poor design / thermal management, or transients), the resistor limits fault current from the external supply.  Less sparking, explosion, smoke; less damage to drive circuitry.

Tim

[oz2cpu]

>Kleinstein and Tim are probably trying to say that you should use some high power resistors as a load connected to IGBTs collectors

adding resistors in series with Collectors, will affect the current capabilities at low input voltage,
so that is not what i wanted initially, could of course live fine with just using my siglent dc load at all voltages under 150V (its max rating)
and this way, only use this load for all voltages over 150V

the attached snip, and red markings are my main concern !
the current is very well shared, so it is fair to x 8 here.
60W peak pr unit, while cold (60 x 8 = 480W)
but only 24W pr unit, while hot (24 x 8 = 192W)
since the device at 60W peak is not going to stay cold, maybe only a few secs,
So that is important to remember, never to load it over 200W
or compromise working current / voltage area, and add series resistors, to pump up the total power load capabilities..
A such series resistance could be external, just pick its value and power size for the lab experiment,
and then use the electronic load only for fine tuning of the wanted current, this way most voltage and also power
could be burned in the resistors.

So thanks a lot friends, you came with some very good, and very use full points.

[Kleinstein]

The problem with the SOA specs is that individual units can be quite different. The relatively poor SOA ratings come from problmes with current sharing on the chip. So the current may concentrate to a small part of the chip. This usually gets worse, the high the voltage is.  Most chips may be OK, starting from a rather homogenous material, but some may be poor, e.g. parts from the edge.  SOA testing is a bit tricky and the parts are this not checked to the ability to withstand significant power at a high voltage.

The 60 W rating is already rather low for the quite large case, but because of the SOA limitations the 60 W limit likely only applies to low votlages. With 500 V the safe limit is likely way lower. At least there is good chance than some will fail early.

At the very least it would be a good idea to do a stress test under safe conditions, before actually using the load with a supply that is not happy with a dead short. Fuses for 1000 V DC are already a bit tricky. The stress test could be done with 1 IGBT at a time - so less power (but still the high voltage) needed.

Yes, series resistors would limit the used for lower voltage. So one sould also have an alternative path without the resistors (or a much lower value) in parallel. With higher voltage the transistors plus series resistors are used and at lower voltages other transistors without would do most of the job.

With 8 IGBTs in parallel, there is quite some capacitive load to the OP-amp. This does not help with the stability and may be the reason for needing so much capacitance to get it stable.

[JohanH]

60W peak pr unit, while cold (60 x 8 = 480W)
but only 24W pr unit, while hot (24 x 8 = 192W)

The thing is, dissipation figures are almost meaningless in linear operation. They are specified for pulsed operation only. IGBT die can develop local hotspots in linear operation, regardless of total package temperature. That's why they almost never specify DC SOA.

[oz2cpu]

>At the very least it would be a good idea to do a stress test under safe conditions, before actually using the load with a supply that is not happy with a dead short. >Fuses for 1000 V DC are already a bit tricky.

that is why i did draw the fuse into the schematic, but omitted it in my own unit :-)
I actually got a few of the fuses from microwave ovens, they sound perfect for this,
so maybe i should add one.

>The stress test could be done with 1 IGBT at a time - so less power (but still the high voltage) needed.

Good point, and very interesting topic,
i do happen to have a very large bag of the igbt's
Free, from one of my regular sponsors :-)
So i could perform power pr unit tests,

and one of my local friends just wrote me, he got a 0-1000V 0-10A supply available, i can come pick it up :-)
but i need to bring a few extra friends for the transport job i think..

[uer166]

the attached snip, and red markings are my main concern !

Those red markings are irrelevant, they'll fail due to hotspotting and other effects at much less than the rated power dissipation (I've done this and in my case a 600W rated IGBT brick failed at ~60W).

[oz2cpu]

thanks uer166, i look forward to perform some real life stress watt tests on the devices i use.
How many times did you repeat this experiment ?
at what voltage versus its max rating ?
it is of course much easier for me to test at 50% voltage rating, if the results are expected to be just as bad ?

PS: i took it apart yesterday and added external reference on the Arduino, changes sense resistors, added trimmers,
more hardware calibration, and even ended added software calibration, using MAP feature, 3 point,
since the ADC in the arduino is really bad, I am now super happy the readouts are very accurate :-) 

 

[Martinn]

As others have pointed out repeatedly, you still don't seem to have the SOA on the radar. This is a common pitfall when trying to build an electronic load (as I also had to learn).
One hint: Ixys makes some nice linear mode FETs which are explicitly designed for your type of usage: https://www.littelfuse.com/products/power-semiconductors/discrete-mosfets/n-channel-linear.aspx
This https://www.littelfuse.com/products/power-semiconductors/discrete-mosfets/n-channel-linear/standard_linearmode/ixtk8n150l.aspx device for example has a DC SOA of 1000V, 0.5 A at 60 °C.
They even sent me free samples for a HV electronic load project I once planned to do. So you might try that once you are through your bag of IGBTs...

[oz2cpu]

thanks Martinn, where did i say i dont get it ?
the IXTK8N150L is a really nice device, good thinks just dont come cheap :-)
none at RS, none at Farnell, none at Mouser, only 4 at Digikey, and they are far from cheap,
and all my normal sponsors dont have any thing like this too :-)
and by the way, still no luck burning any of my IGBT's seems like i am lucky
or just dont have enough power available

[Martinn]

Well, almost all modern FETs have a miserable DC SOA (or are not specified at all), which seems to be a byproduct of optimizing them for switching applications. So the few remaining analog FETs sell for $$$$... Try requesting samples, it worked for me.

[T3sl4co1l]

Well, almost all modern FETs have a miserable DC SOA (or are not specified at all), which seems to be a byproduct of optimizing them for switching applications. So the few remaining analog FETs sell for $$$$... Try requesting samples, it worked for me.

FYI, this is a couple generations outdated -- this was true at the height of ordinary VDMOS, more or less.  To which OP's IGBTs will be contemporary; take your pick of any late-90s MOSFETs for examples.

I don't know what exact mechanisms allow it, but a lot of SuperJunction (what I would consider "modern" Si) offer DC SOA curves.  This despite ever-higher power density (which was the main reason conventional VDMOS broke the conventional "wisdom" of "MOSFETs free from 2nd breakdown" -- ye olde lateral, and most e.g. HEXFETs (i.e., IR designs of the 70s-80s), just didn't have high enough power density to extend into the region of instability.

Astonishingly, even some SJ IGBTs are publishing SOA curves, including DC.  I haven't tested any as such, and I wouldn't trust them offhand for such purposes.  Anyway, linear operation is all about die area, and you get more die per buck among MOSFETs.

SJ is most readily identifiable either directly by name (or in-house equivalents, e.g. Infineon's CoolMOS), or by the uniquely severe, then slightly rebounding, capacitance curve (Crss in particular drops to nearly nothing by modest Vds (20-100V?) then rises slightly; curves may not extend to high enough voltages to see the rebound).

Like I said earlier, FQA9N90C works just fine.  At least, I think "QFET" is a SJ type?  I see "planar stripe DMOS", which doesn't seem to rule that out, at least.

Tim

[oz2cpu]

I think it is a super interesting topic, and very important to have up in the open
so we all hear about it,
thanks again to you all for repeating emphasize the lack of power capeability of IGBT's

[Martinn]

I don't know what exact mechanisms allow it, but a lot of SuperJunction (what I would consider "modern" Si) offer DC SOA curves.  This despite ever-higher power density (which was the main reason conventional VDMOS broke the conventional "wisdom" of "MOSFETs free from 2nd breakdown" -- ye olde lateral, and most e.g. HEXFETs (i.e., IR designs of the 70s-80s), just didn't have high enough power density to extend into the region of instability.

Tim, are you sure SJ is a benefit for linear operation? Vishay mentions in an app note that SJ has lower chip sizes (for lower Qg), but consequently also lower robustness and thermal capacity.
Infineon Coolmos P7 950V (SJ) are also rather miserable:
https://www.infineon.com/cms/de/product/power/mosfet/n-channel/500v-950v/coolmos-p7/950v-coolmos-p7/?redirId=144097
https://www.infineon.com/dgdl/Infineon-IPD95R450P7-DataSheet-v02_02-EN.pdf?fileId=5546d462636cc8fb01643b1b778a55d4
This (IPD95R450P7) 950V 14A device has a DC SOA of 950V 0.8 mA! But I guess with Infineon one has to be content there's a DC SOA at all.
Maybe this plastic case is not ideal for power dissipation, although 950V*0.8 mA is just 0.76 W.
This one https://www.infineon.com/dgdl/Infineon-IPW90R120C3-DS-v01_00-en.pdf?fileId=db3a3043183a955501185000e1d254f2 is much better, SOA 950V 0.45 A, TO-247

- Martin

[T3sl4co1l]

Smaller chip size just means less power dissipation for the same switching capacity.  Like I said, linear just needs die area <--> power dissipation.  So you'll tend to need lower Rds(on) or higher Vds(max) ratings than you might otherwise expect. But those don't matter, as long desired operating conditions are covered.

Like, in comparison, RF transistors have the opposite problem, they're built for power, and are practically useless at switching -- the Rds(on) is basically just enough to deliver design peak output power.  You can buy bigger ones sure (for deeper class D/etc. operation), but you spend a lot of money to gain that efficiency.

There's not really any inbetween, from switching transistors at low frequency, to RF transistors wholly uneconomical except when you absolutely need it (i.e., at RF).  Well, not in Si anyway; GaN is expanding switching roles to higher frequencies.  But with even higher stakes: the power density is incredible, there's no SOA to speak of, the slightest misstep and it's a puff of smoke.

Anyway, last time I was shopping for watts/dollar, FQA9N90C was one of the top contenders.  Whatever process that is, seems to do fine.  That was a few years ago, there may be better options now; or for other ratings.  TO-220s for example seem better at lower voltages (<200V), and even BJTs are viable there (even some switching types, which retain enough SOA at voltage to be competitive).

Tim

[Wolfgang]

I followed this advice and made a smaller version of it (1000V, 50W cont. 250W pulse for 5s).
This works fine, see here:

https://electronicprojectsforfun.wordpress.com/high-voltage-test-load-is-scpi-controllable/