IGBT vs MOSFET for dummy load

[coppercone2]

is it actually worth using mesh around the fans? I thought that lowered their efficiency.

also if you made it longer and put the resistors lengthwise i think they would cool better

and there could be air flow through them. I think the change might be fairly drastic.

I always wondered about those resistors, if you can make fittings for them and connect them to a ducted fan.

[OM222O]

I got an answer:

Thank you for contacting ON Semiconductor.

Unfortunately, it is not a good idea to use IGBTs in the linear region. Threshold voltage between individual pieces will vary considerably so if you set a working point for one IGBT and replace it with another, the performance will be different. For this reason it is difficult, if not impossible, to achieve consistency.

If you have any further questions, please do not hesitate to contact us back.

They're basically worried about the precision of the device if it's being ran from a fixed voltage value. In my case each IGBT is controlled from a separate op amp to get an accurate current through it 
They didn't mention anything about them not being able to handle DC current in their linear region and given the SOA provided in the datasheet, I'm farily confident that these are a better replacement than the mosfets

[coppercone2]

you should ask, maybe they thought to stop investigating or explaining the matter after the why not.

[TurboTom]

@OM222O --

you asked about the suitibility of IGBTs for operation in linear mode (as the pass element in an electronic load in particular). Many knowledgable members of the forum told you their thoughts and experiences. The answer was virtually unison that it's not a good idea to use an IGBT in linear mode, and this for a good reason. Of course, it's your own decision to try it and convince yourself. Maybe you find a solution that all the others didn't think of and succeed. But please don't try to argue against the leading opinion without solid evidence. A good point to start at is reading about failure modes of MOSFETs since they are very closely related to IGBTs (the latter have one doping layer more). Here's a book by NXP that you may want to have a look at and that covers the subject quite well and provides informaton to other literature. Some interesting information can be found on page 43ff.

It's also for a good reason that most of the entry level electronic loads utilise the ancient IRFP250 (non-N version if possible) as their pass elements, and usually many of them, each with its individual driving circuitry. There have been many clever people scratching their heads how to solve this problem as elegant as possible. And it's rather smart to learn from them as much as possible and not to release too much magic smoke that you've got to pay for yourself...

Just my two cents.

Good luck and all the best,
Thomas

[boB]

Mosfets will also have a difference of Vgs threshold and it changes with temperature.  Even if you match your FETs or IGBTs there is no guaranty that they will track precisely over temperature range.

I'd use BJTs

You might get away with using insulated gate parts if it is only one and not more than one in parallel.

 This is where switching and PWM with a load resistor can come in handy.  Then the Vgs doesn't matter so much.

[coppercone2]

load matching transistors in a dynamic circuit by active means is just fugly.

[T3sl4co1l]

Taking Vgs(th) out of the equation by attaching a large source resistor, or single op-amp, to each transistor, is not fugly, it's traditional.

They didn't do it back in the BJT days because, well, actually, they did, with emitter resistors, but op-amps no, because ICs were still expensive back then.  Since the 70s passed, no one cares.

The only non-integrated device in existence, that is well enough matched to not need a common-terminal resistor, is the vacuum tube.  And even back then, they still recommended it!  (Example: 6AS7/6080 regulated supplies, designs abound.  The cathode resistor was more to limit short circuit current though, which on beefy tubes like these, could be several amperes!)

Tim

[coppercone2]

I meant digital adjustment.

[jmelson]

This IGBT is fairly cheap and seems fit for the job. It has a SOA chart unlike what previously has been said, and given the numbers I provided, it seems like a fairly decent choice.

https://www.mouser.co.uk/ProductDetail/ON-Semiconductor-Fairchild/FGH50T65SQD-F155?qs=sGAEpiMZZMv4z0HnGdrLjmAnkg%2f6XuvzWCGzpW%2fDvWZ3y7weEXwbsA%3d%3d


Can you please confirm I'm not making a mistake in reading the datasheet? It has way better SOA for similarly priced mosfets.
(Please keep in mind I'm limiting the power of each part to 50 watts, so @ 50 volts 1A would be the current limit and as you decrease the voltage, the current can go up accordingly, so my intended area of operation falls way below the provided SOA)

Well, there it is, plain as day!  It can handle 100 A with just 2.5 V across it, but if it has 100 V across it, the SOA is only good to about 2 A.  And, presumably that is at 25 C, if you let it get hot, the SOA probably shrinks.

Jon

[T3sl4co1l]

Well, yeah.
Well, yeah.
Well, yeah.
That's precisely how SOA and thermal resistance works!  Gold star!   (Not to be condescending. Just that, alas, if only so many commercial designers understood it so simply...)

250-270W (by my eye, but anyway the RthJC figure is what's at work here) is no slouch of a transistor, if it's affordable at least (and the DC curve is actually real).  500W+ transistors are easy to find, but it's harder to find the cheapest watts per dollar (at a given rating), which is really all that matters here.

Tim

[OM222O]

@OM222O --

you asked about the suitibility of IGBTs for operation in linear mode (as the pass element in an electronic load in particular). Many knowledgable members of the forum told you their thoughts and experiences. The answer was virtually unison that it's not a good idea to use an IGBT in linear mode, and this for a good reason. Of course, it's your own decision to try it and convince yourself. Maybe you find a solution that all the others didn't think of and succeed. But please don't try to argue against the leading opinion without solid evidence. A good point to start at is reading about failure modes of MOSFETs since they are very closely related to IGBTs (the latter have one doping layer more). Here's a book by NXP that you may want to have a look at and that covers the subject quite well and provides informaton to other literature. Some interesting information can be found on page 43ff.

It's also for a good reason that most of the entry level electronic loads utilise the ancient IRFP250 (non-N version if possible) as their pass elements, and usually many of them, each with its individual driving circuitry. There have been many clever people scratching their heads how to solve this problem as elegant as possible. And it's rather smart to learn from them as much as possible and not to release too much magic smoke that you've got to pay for yourself...

Just my two cents.

Good luck and all the best,
Thomas

Well I haven't heard any "good reasons" as why not to run an IGBT in it's linear region. People have mentioned linear SOA which as I showed in the picture of the data sheet matches my requirements and even after asking the manufacturer, they are more worried about matching Vgs more than anything else. Please let me know if there's any other "good reasons" not to do so. I'm just trying different things to maybe make something better than what "everybody else does". If I can make this work, it's gonna be a killer deal in terms of electronic loads.


I still haven't managed to find a decent circuit that uses BJTs 

maybe the same arrangement as the mosfet works fine with an addition of base resistor , but most schematics say beta matters a lot so they use darlington or even 3 BJTs connected to each other. They don't provide any reason as to why tho 

any comments on that?


I have also considered PWMing a few load resistors as Tim mentioned but it won't be true constant current and I don't think iy's gonna be as accurately adjustable as I want it to be (24 bit PWM is not available anywhere that I could find and it's probably gonna cost an arm and a leg if it was), so I'm brushing it to the side for the time being.

[Kleinstein]

The problem with DC SOA curves for parts that are clearly intended for switching applications is that it may be plain wrong. In this case chances are very high the SOA curve is only calculated from the transient thermal resistance using a formula that is no longer valid beyond some 10-100 µs as it ignores thermal instability - exactly the effect that limits the useful SOA in linear operation.

It is quite common to find this type of mistake for modern MOSFET data-sheets from some manufacturers - chances are hight they use the same faulty software to calculate the SOA curve for IGBTs too.

It takes very special IGBTs to have a useful FBSOA - so it is much more likely the SOA curve is a mistake than the IGBT actually useful for linear operation. Unless they explicitly mention linear operation one should not trust an DC SOA that just reflects P_tot: not for a modern low voltage MOSFET (< 400 V), not for an BJT and the least for an IGBT.

Internally IGBTs are build a little like a MOSFET providing the base current to an BJT. BJTs are known to have limitations due to 2 nd breakdown, from a thermal instability. It takes rather special BJTs to get a good SOA beyond 100 V. At the same power level IGBTs are using even smaller dies than BJTs and thus tend to be even more susceptible to thermal instability. So it is obvious that the shown SOA can not be true, at least for the higher voltages like >100 V. It up to you to decide if you trust is up to 5 V or 10 V or maybe 15 V.

If something sound to good to be true, consider the possibility that it is not true.

[coppercone2]

is it possible to drill out the package and add a little germanium window or something so that you can measure temperature and make your own SOA curve?

Or measure the die even faster for feedback. Is it on the surface or something burried where the overheating happens? Can you shorten response time?

[OM222O]

I assume thermal instability comes from inadeqote cooling. like how you get more performance and less power leakage on CPUs and GPUs if you cool them properly. I think it was somewhere in the ball park of 4% power decrease for every 10c drop in temps. I assume other semiconductors work similar to that 

and as I mentioned the calculated temperature in my application would be at high 70s or mid 80s in worst case scenario which is far below the provided SOA, even considering the de-rating factor 

please correct me if I'm wrong

[TurboTom]

I assume thermal instability comes from inadeqote cooling. like how you get more performance and less power leakage on CPUs and GPUs if you cool them properly.
...

please correct me if I'm wrong

If you don't mind, please have a look at this publication - it explains a lot and also points out possible solutions.

Edit: Broken Link fixed

[David Hess]

Well, there it is, plain as day!  It can handle 100 A with just 2.5 V across it, but if it has 100 V across it, the SOA is only good to about 2 A.  And, presumably that is at 25 C, if you let it get hot, the SOA probably shrinks.

Not sure if serious ...

For the reasons I gave earlier, I do not believe the DC curve on that SOA graph.  At a case temperature of 175C, the SOA graph is a point at zero volts and zero amps.  I am kind of suspicious that they do not show a graph of this but I suspect it is either because they never intend this device to be used in linear operation where it would matter or because it actually performs worse than the specifications indicate.

I assume thermal instability comes from inadeqote cooling. like how you get more performance and less power leakage on CPUs and GPUs if you cool them properly. I think it was somewhere in the ball park of 4% power decrease for every 10c drop in temps. I assume other semiconductors work similar to that 

Thermal instability comes from operating at high Vds where the temperature coefficient of the Vgs reverses.  Secondary breakdown comes from operating at high Vce where current crowding causes localized heating raising hfe.  So both effects are temperature related in that they cause thermal runaway through positive feedback.

and as I mentioned the calculated temperature in my application would be at high 70s or mid 80s in worst case scenario which is far below the provided SOA, even considering the de-rating factor 

The junction temperature directly affects the SOA and it affects the secondary breakdown and thermal instability regions of the SOA just as much if not more.

I'd use BJTs

His requirements are not particularly demanding:

I'll most likely aim for 50 volts, 20 amps, 200 watts and spread the 4 into 2 groups of 2 (again for increased surface area and better cooling), each cooled by a CPU cooler, so each device needs to handle roughly 50 watts which is not that large.

I suspect a single power MOSFET driving 4 bipolar transistors in parallel (or 8, see below) with 0.1 to 0.22 ohm emitter ballast resistors would be the least expensive because no extra effort would be needed for current sharing due to consistent Vbe and low emitter ballast resistance.  The bipolar transistors could be purchased in a lot of 10 and matched for Vbe at 5 amps for additional reliability.

MOSFETs or IGBTs may be only slightly more expensive but may need to be driven individually to enforce current sharing.  There are more choices now for big MOSFETs and IGBTs versus bipolar transistors contributing toward closer prices.  It seems the only big non TO-3 bipolar transistors you can buy now are intended for audio.  Conceivably an even less expensive design using the largest TO-220 bipolar transistors but more of them (2 per heat sink?) is possible.

I checked a couple of big power MOSFETs and except for the linear rated IXYS IXTH80N075L2 and IXTH64N10L2, the ones I checked would all have problems with current sharing.  You can see it in the graphs where the ones intended for switching operation have transconductance increase with temperature while with the IXYS parts, it decreases.  Figures 1 and 2 for that IGBT show the same thing but even worse.  The switching MOSFETs also showed straight DC SOA which I do not believe but I do believe it on the IXYS parts.  Who else makes linear MOSFETs?

[T3sl4co1l]

You can argue all you want, but it only ever comes down to this:

What is the component capable of?

It doesn't matter if MOST IGBTs have a shity SOA.  I don't trust that this one has a full SOA, so I would test it.

How do you test it?  Build your circuit and see if it works!  Duh!  Simple as that! 

There are many MOSFET types out there.  A lot have 2nd breakdown within the operating area.  You can't generalize "MOSFETs are free from 2nd breakdown", that used to be true back in the bad old days but it isn't now.

You can't generalize "IGBTs have shit SOA", though until this thread I hadn't seen one with a DC SOA, and so it would seem this generalization is more reasonable, but again not an actual rule.

You can't generalize "BJTs exhibit 2nd breakdown [in their operating area]" (note the often omitted but necessary qualification!), because while a lot of them do, amplifier transistors are made to handle it.  Just as modern (high current density) MOSFETs are sometimes made in linear flavors, or coincidentally happen to have a full SOA.

Tim

[BravoV]

Who else makes linear MOSFETs?

Microsemi does it, although not as many as IXYS/Littelfuse has.

[Siwastaja]

I assume thermal instability comes from inadeqote cooling.

No. The thermal instability discussed here has little to do with external cooling - it's about local hotspotting, which happens so quickly and at microscopic level that it's all about the internal heat spreading of the die, through the substrate, to the device case. You can't do much externally.

This is, as has been explained numerous times, due to the different threshold voltage of the different parts of the transistor, and their positive tempco, meaning that a local part that takes more current through it, starts getting even more current, and runs away thermally.

This is actually the same discussed problem of Vgsth matching, but you can't do anything since it's happening inside the transistor - and they have done their best to minimize the source (emitter) resistance. Can't add separate amplifiers or emitter resistors there!

This effect is rolled into the SOA, and if your part has a DC SOA graph like this one, and it's OK for you, then go ahead. (Although it would be still quite difficult to be in-spec if they only give you a DC SOA for Tc=25 degC. You need a refrigerator cooling, basically. Or you are going outside the specs and are on your own anyway.)

Otherwise, the manufacturer is outright lying, or have no idea about the basics of their field of market.

There is only one problem, and it's the fact you can't believe a manufacturer could actually lie about their parts - or if not outright lie, at least not understand their own parts and what the basic specifications means. When you gain more experience, you'll find out the harsh truth. They lie all the time. And the datasheet writers are not always the brightest geniuses who actually make the transistors. Wrong info on the datasheets is so common that it isn't even a pet peeve of mine anymore - we just need to live with it. It's like a rainy day, we don't even feel like complaining about this would help.

This being said, the chances are the DC SOA is perfectly valid. Go for it.

Vgs (Vge) threshold matching is completely a red herring. There is zero need to match this parameter. Any sane design today uses separate drive per transistor, completely eliminating this parameter. Not using a separate driver is an ugly hack, only there to try to "save cost", which made sense when opamps did cost something. Now they come in miniscule packages and cost a few cents per amplifier. Now, not using dedicated drives only increases cost as you probably need to derate the transistors more, add more source(emitter) resistance, handpick parts or do some other hacks which have their own hidden costs (most likely much more than an opamp).

[Le_Bassiste]

i found this quite helpful in understanding the derating of continuous DC SOA at elevated temperatures (PDF download):
https://www.infineon.com/dgdl/Infineon-ApplicationNote_Linear_Mode_Operation_Safe_Operation_Diagram_MOSFETs-AN-v01_00-EN.pdf?fileId=db3a30433e30e4bf013e3646e9381200

[mzzj]

To satisfy my curiosity I played with my lab power supply and 2pcs IRG4PC50W IGBT for SOA:
1. exemplar: 100V and 1 amp =immediate destruction
2. exemplar 100V and 1 amp = survived for test duration (1 seconds ) 
2. exemplar 200V and 0.5 amp = instant death

All of these were performed at about Tcase ~25c (200W "rated" power for these)

Pretty much as expected, unfortunately didn't have any DC SOA rated IGBT's around in my parts bins.

[boB]

To satisfy my curiosity I played with my lab power supply and 2pcs IRG4PC50W IGBT for SOA:
1. exemplar: 100V and 1 amp =immediate destruction
2. exemplar 100V and 1 amp = survived for test duration (1 seconds ) 
2. exemplar 200V and 0.5 amp = instant death

All of these were performed at about Tcase ~25c (200W "rated" power for these)

Pretty much as expected, unfortunately didn't have any DC SOA rated IGBT's around in my parts bins.

You mean the case was 25 C  before you applied power to it ?

No heat sink ?

[jmelson]

Well I haven't heard any "good reasons" as why not to run an IGBT in it's linear region. People have mentioned linear SOA which as I showed in the picture of the data sheet matches my requirements

OK, the reason for all this is that the output stage of the IGBT is a bipolar transistor.  These have a negative thermal coefficient of voltage drop, so the HOTTEST part of the die carries the most current.  When paralleling multiple bipolar or IGBT transistors, you use ballast resistors to balance current.  Within a single die, you can't do that.  The IGBTs are designed to minimize this effect in full saturation, so the SOA gets much bigger when saturated.  The ability to keep all parts of the die carrying the same current fails in the linear region, so one area of the die gets hotter and takes most of the current.  That's why your 100 A transistor can only handle 2 A with higher C-E voltage across is.

And, that graph is for a case temp of 25 C!  That is important.  At higher temperatures, this effect can get a lot worse.  At 100 C case temp, this transistor may only be safe at 100 mA.

Jon

[mzzj]

You mean the case was 25 C  before you applied power to it ?

No heat sink ?

Large 8mm thick copper plate as heat sink.

[coppercone2]

can you get lucky with luck of the draw for high IGBT linear current or just by their nature their gonna give you <3% of their rating in linear region performance?

does anyone have diagrams of these hot spot formations or thermographs?

what can I study to understand the charge densities and everything to understand the structure of the hot spot formation and why it decays with saturation.