IGBT Linear operation

[T3sl4co1l]

The conclusion: don’t exceed the data sheet’s SOA. With adequate margin.

That´s what is still confusing me: In the datasheet of the mentioned IGBT is no SOA is given but it was killed at around 30% of it´s max dissipation.

Yeah they're telling you the SOA is "don't".

An IGBT is physically incapable of being used in linear mode. They are highly tuned switching devices so their structure is optimized for turn on and off.
In linear mode you will burn it out because only a portion of the device will conduct and that is a guarantee to overheat and fail-short. There is a name for this effect, but I can't remember! (positive feedback loop in transistors that increases gain with current/temperature)

This is false -- IGBTs are continuous devices like any other transistor.  Please do not spread misinformation; there is already dangerously much in this subject already, as this thread plainly shows.

IGBTs *can* exhibit latchup, in which case it fucking shorts out (controlled switch on, uncontrolled switch off) -- it becomes an SCR.  IGBTs have been largely latchup-free since the 80s or 90s, so much so that datasheets don't even bother to proclaim their freedom of it.

IGBTs typically have small (or N/A!) SOAs, but some are showing full SOA.

IGBTs have been used in audio amplifiers; I can't say this is a very wise or effective choice, there's often no accounting for that among audiophile types, y'know, but more to say it's possible, and has been done.  I believe there was even a line of Toshiba parts intended for linear operation.

Some IGBT also have a maximum switching time specified in their datasheet. It's hard to make it switch slower, which can be real challange in EMC tests.

I don't even know where one comes up with this--?

MOSFETs may have max time as well, it's a characteristic not a rating.  Max is just the worst case parameter spread, not a "you must obey".

Tim

[David Hess]

My questions: Is this to be expected? Are IGBT not suited for linear operation? Can a circuit that´s designed for a MOSFET somehow kill an IGBT without exceeding its SOA?

Modern IGBTs have poor SOA so are not suited to high power linear operation.  Substitution between MOSFETs and IGBTs is possible depending on the matching of the specifications, but for equal voltage and current specifications, the IGBT will be smaller and have a lower power rating.  This suggests looking for a replacement with an equal or higher power rating, even if this means a higher voltage and current rating.

IGBTs have been used in audio amplifiers; I can't say this is a very wise or effective choice, there's often no accounting for that among audiophile types, y'know, but more to say it's possible, and has been done.  I believe there was even a line of Toshiba parts intended for linear operation.

Toshiba made n-channel and p-channel IGBTs for complementary class-AB linear audio power amplifiers and they would have been well characterized for linear operation.  I think it was more of a marketing thing rather than because of better performance.

[Kleinstein]

The datasheet shows a full SOA for the CS50N20 MOSFET.  However chances are that this graph is derived from transisent thermal resistance and this way ignoring possible thermal instability.  It is unfortunate that such false SOA graphs are sometimes shown.

There are special, but quite expensive MOSFETs specified for linear operation. If more standard FETs are wanted, one should look for either rather old types with relatively large chip and large R_on or types for a rather high voltage, like the superjunction types. An old time favorite when MOSFET audio amplifiers came up was the IRFP250, that could still be not too bad, even though the IRFP250N one often gets now is not as good for linear operation.
The IRFP460 mentioned in another answer before would be more powerful.

A problem may be that those fets that are better suited for linear operation tend to have a higher gate capacitance, which can make the replacement a bit tricky. One should also derate the power to something like 50% or maybe < 100 W for a TO247 case. A direct parallel connection is not practical - so spreading the heat would  need a change to the circuit.

[Phil1977]

Thanks for the well qualified last posts!

The IXTH80N20L is explicitly specified for linear operation and IXYS (now littlefuse) gives a so called "Forward Bias Safe Operating Area (FBSOA) - Conveniently they specify it for 25°C heat sink and 75°C heat sink temperature. This part is not crazily expensive (11€ at mouser). The datasheet explicitly states its suited for DC loads and current regulators.

I well know what it means to cool parts with crazy power densities. I worked on industrial projects with high power LEDs, and these have forward currents up to 7A/mm². That means up to 10W/mm² heat dissipation, that´s 1kW/cm² - much more than a standard CPU. The LED dies really needed to stay below 80°C to stay efficient, we did lots of FEM-simulations and did experiments with diamond heat spreaders. And I´m quite sure you don't need a FEM simulation to get rid of 150W of roughly one cm² of the TO247 tab. A decent copper heatspreader with a standard 0.2K/W CPU fan will do it.

PS: The IXTH80N20L is not in stock - I´ll try the IRFP460 as proposed by Kleinstein.

[Phil1977]

Just found this on the Infineon homepage:

IGBT Linear Mode

 
The modern trench IGBT cells (also micro-pattern trench) are not suitable for linear mode operation. Such an operation must be strictly avoided in applications. Modern IGBTs are designed for switching applications only.

 
Exceptional case is a short circuit (desaturated IGBT) which is a special use-case of linear mode. The main difference to a regular linear mode application is that the short circuit time is specified with a limited time. In this short time it is ensured that thermal runaway of individual cells will not occur and thus this short term linear mode is allowed for exceptional cases like short circuit.. desaturated IGBT)

https://community.infineon.com/t5/Knowledge-Base-Articles/Is-linear-mode-allowed-i-e-Desaturated-IGBT/ta-p/491055#.

[Phil1977]

As an interim solution I exchanged the blown FET by a 20N60C3 in a TO220-package.

Together with a CPU cooler with a copper heat spreader it runs at 120W with a T_case of 75°C - measured with a thermocouple at the upper side of the tab and confirmed by a thermal imager.

The 20N60C3 is also not explicitly qualified for linear operation but it seems the FETs for higher voltage with a bit higher RDS_on have less problem with that.


For sure there are better electronic loads available - but so far the DL24 is working fine for me.

[Jeroen3]

...

An IGBT is physically incapable of being used in linear mode. They are highly tuned switching devices so their structure is optimized for turn on and off.
In linear mode you will burn it out because only a portion of the device will conduct and that is a guarantee to overheat and fail-short. There is a name for this effect, but I can't remember! (positive feedback loop in transistors that increases gain with current/temperature)

This is false -- IGBTs are continuous devices like any other transistor.  Please do not spread misinformation; there is already dangerously much in this subject already, as this thread plainly shows.
...

Okay, I will tell Semikron to update their application manual.

Stationary module operation in the active region is not permissible, because VGE(th) falls when the temperature rises, meaning that even small differences between the individual chips may cause thermal instability

And they mention it several other times as well.

[Phil1977]

...

An IGBT is physically incapable of being used in linear mode. They are highly tuned switching devices so their structure is optimized for turn on and off.
In linear mode you will burn it out because only a portion of the device will conduct and that is a guarantee to overheat and fail-short. There is a name for this effect, but I can't remember! (positive feedback loop in transistors that increases gain with current/temperature)

This is false -- IGBTs are continuous devices like any other transistor.  Please do not spread misinformation; there is already dangerously much in this subject already, as this thread plainly shows.
...

Okay, I will tell Semikron to update their application manual.

I think it´s a misunderstanding: You can bring an IGBT into linear operation (what T3SL has mentioned and why I tried to use it as the FET-replacement...), but the manufacturers do not allow it because it quickly leads to islanding effects with following destruction (as mentioned by Infineon and proofed by the destruction at comparatively low dissipation).

But anyhow it´s different compared to e.g. a SCR, which has no linear operation mode from the beginning.

[Jeroen3]

If you can get an SCR into linear mode I'll buy you a beer, haha.

[wraper]

If you can get an SCR into linear mode I'll buy you a beer, haha.

Although I did not try, I suspect it may be possible if current is kept below latching current.

[Zero999]

If you can get an SCR into linear mode I'll buy you a beer, haha.

Although I did not try, I suspect it may be possible if current is kept below latching current.

Too lower trigger curent can cause an SCR to operate in linear mode, when it behaves like a BJT. This can kill it, if the SOA is exceeded.

[T3sl4co1l]

Okay, I will tell Semikron to update their application manual.

Stationary module operation in the active region is not permissible, because VGE(th) falls when the temperature rises, meaning that even small differences between the individual chips may cause thermal instability

And they mention it several other times as well.

Read more closely:

1. Apparently this is a multi-chip module.
2. The dies are probably matched for Vce(sat), and maybe tempco thereof (if it's also subject to variation), but not also Vge(th) and its tempco (which may be independent of the other parameters).
3. They are separate dies, no different than paralleling discrete transistors, which isn't what we're talking about here.
4. It can still turn out well, if the thermal resistance is good enough -- modules generally are quite good at thermal, so there's still a possibility, perhaps if they had used (linear mode-) matched dies; but
5. They don't specify this mode of operation, and don't intend to support it, just nip it in the bud, discourage operation regardless of the considerations (specifically because, the above considerations are not something they want to bother supporting).

Tim

[Jeroen3]

They also explain their dies are a hexagonal pattern of junctions. With multiple chips it is obvious a problem.

Can you provide an example of an IGBT suitable for linear usage, I can't seem to find any.

[T3sl4co1l]

I mean hex doesn't matter, HEXFETs have had wide SOA for a long time, only relatively recently (early 2000s?) they've achieved such density (and other optimizations) to affect SOA.  Over SOA-relevant time scales, there's not nearly enough curvature in the temperature field to matter, it's just the overall power density and tempco that matters; it's treatable as a continuous medium.

The Toshiba mentioned earlier was, https://www.alldatasheet.com/datasheet-pdf/view/30904/TOSHIBA/GT20D201.html
A newer one, https://www.onsemi.com/pdf/datasheet/fga50t65shd-d.pdf (I have some of these, I might test one or two)
Another, https://www.onsemi.com/pdf/datasheet/ngtb40n120fl3w-d.pdf

"Suitable" might be overstating it. I don't really know why they show a DC SOA. Or how seriously they mean it. But there it is.

Tim

[Kleinstein]

The toshiba part seems to be made for linear operation.
The SOA in the Onsemi parts is likely just calculated from the thermal transients.
The same problem happens with some MOSFET SOA curves. Derive the maximum power for a given pulse length from transient thermal resistance and
than get a useless SOA curve that misses the important part  (possible thermal instability / 2nd break-down).

Even the Toshiba part shows the break at high voltage where the SOA is no longer limited by power or pulse energy, but lower for the higher voltages.
With a more normal IGBT a SOA curve without that extra limitation is more like a red flag indicating non trustworthy SOA curve.
So I consider the 2 onsemi part no better than the typical IGBTs, just a worse data-sheet.

[Phil1977]

Is it a stupid idea to replace an IGBT by a small FET + large BJT for linear applications? Like in its equivalent circuit?

[Andy Chee]

Is it a stupid idea to replace an IGBT by a small FET + large BJT for linear applications? Like in its equivalent circuit?

I'd imagine the linearity of the darlington-like FET/BJT combo, would be worse than a single IGBT device alone, even if the IGBT isn't specified to operate linearly.

[T3sl4co1l]

The toshiba part seems to be made for linear operation.
The SOA in the Onsemi parts is likely just calculated from the thermal transients.
The same problem happens with some MOSFET SOA curves. Derive the maximum power for a given pulse length from transient thermal resistance and
than get a useless SOA curve that misses the important part  (possible thermal instability / 2nd break-down).

I've heard this paranoia before, but never any justification for it.

Do you know of any evidence to support the claim that an SOA might be calculated without taking instability into account?

How can you tell when one is calculated, and another is legitimate?

Or put another way -- should one put blind trust in the marketing blurb that e.g. "Linear L2" MOSFETs are, in fact, suitable for linear operation?  How much veracity does that claim carry, versus a Superjunction MOSFET, or IGBT, that doesn't make such a statement but carries an identical(ly shaped) SOA?

I've certainly been burned by marketing claims before (granted, mostly on much more complicated things -- ICs), and a lack of reading the fine print, the data tables, the characteristic curves.  I've not seen a scenario yet where the inverse is true -- but that doesn't mean they don't exist, and I would be very interested to see counterexamples.

With a more normal IGBT a SOA curve without that extra limitation is more like a red flag indicating non trustworthy SOA curve.
So I consider the 2 onsemi part no better than the typical IGBTs, just a worse data-sheet.

So you categorically deny that they might have wide SOA?  Why do you know this with such certainty---you must have proof to this effect?  Is it a fundamental semiconductor issue, are there journal articles discussing this?  (I admit I haven't searched journals for IGBT SOA before.)  Is it proven that hotspotting can occur from the resistance of the die thickness itself, thus independent of mounting plate resistance? Could it occur at multiple points simultaneously, or even individual cells, rather than a mass effect biased towards the center of the die (as is widely documented to happen with BJTs and MOSFETs)?

Or perhaps it's much more mundane than this -- you simply asked an FAE and they said yeah it's full of shit.  Could you quote the e-mail so we can name-and-shame?

Tim

[T3sl4co1l]

Is it a stupid idea to replace an IGBT by a small FET + large BJT for linear applications? Like in its equivalent circuit?

Yes, but for interesting reasons: one of the ways they vary speed grades (and consequently optimize voltage drop and other parameters), is basically by varying the hFE of the bipolar effect.  The gate-switched (majority carrier) current might be as much as half the bipolar-switched (minority carrier) current.  This allows significant lifetime-killing (reduces hFE, storage time) but reduces overall current density (increases cost).

Since low-hFE parts aren't available, you can't really do this; let alone high voltage PNPs, that just plain don't exist (I think?).

Basically, imagine you had an upside-down line-output transistor: even then, hFE might be in the 10-20 range at modest currents, though dropping to just a few at high current in saturation.

There's also the JFET effect, that I keep forgetting what role it plays in IGBT (and MOSFET to some extent) behavior... but hey, I use IGBTs so rarely, that would happen.

You'd expect IGBT SOA to be generally comparable to BJT SOA for such reasons, but you never know if they're doing things that have a strong impact on that, or what.  Field stop seems to have some benefit, positive Vce(sat) tempco seems correlated, and they could always be doing something gross/stupid but effective like emitter degeneration resistance -- unlikely given the pressure to low Vce(sat), but who knows, lots of tweaks are possible.

Tim

[Kleinstein]

I don't claim that the SOA for the onsemi IGBTs is for sure false, but there is a large chance that it is. False SOA curve are not so rare with MOSFETs. So for me the claimed SOA curve has zero credebility.

An only power limited SOA would be extraordinary for a high voltage BJT or IGBT.  The current concentration / thermal instability can happen aready with only the thermal resistance of the silicon die. The countemeasures (emitter or base resistance) should comes with a higher saturation voltage and thus not so good switching performance. In a BJT there may be addition (e.g. hot electrons) unstable effects.
So it is unlikely that a IGBT with an ideal SOA would not specially mark this as a special feature.

The linear marked MOSFETs may internally not be different from superjunction MOSFETs, at least not much. The manufacturers should have a reasonable good idea on which types are well suitetable for linear operation and they may have slight differences in the parameters, like a thicker substrate. One difference should be in stricter testing, e.g. for a flaw-less die attach and ideally an individual testing of at least the pulsed SOA. At least for audio BJTs they do (or did) offer parts that are individually SOA tested. The need for stricter testing can explain the extra costs for linear rated MOSFETs.

[mtwieg]

It might be worth looking at superjunction MOSFETs. I've noticed that a lot of those have good DC SOA. I haven't tested any to the limits but I do have a to-247 superjunction that's done a few seconds at 50W and minutes at 15W.

Superjunction FETs are typically the 650V ones.

I haven't looked into the physics of why superjunctions have such good DC SOA (or why they bother to test them as they're mostly for switching applications) but someone on here might know why.

I had been using a superjunction FET (FCA35N60) in a linear amplifier with pretty good results, but unfortunately they're now obsolete with no form-fit-equivalent available.

I'm not sure superjunction FETs have higher SOA specs in general. They're certainly not marketed that way AFAIK.

Classic HEXFETs are generally fine

Indeed, older HEXFETs have excellent SOA specs compared to modern parts. For example, the IRF630 from Vishay can handle Ids=12A at Vds=50V for 1ms. But the IRF630 from STM can only handle Ids=6A under the same conditions! So even with the exact same part number you have to be careful.

Those are "third generation" HEXFETs. The IRF630NPBF from International Rectifier (now Infineon) is a "fifth generation" HEXFET, and can only handle 4A under those conditions.

I always check the curves carefully in every datasheet if I think SOA is important.

[David Hess]

The oldest lateral and vertical power MOSFETs had a square SOA limited only by power and temperature, and this was a big feature for marketing, along with no secondary breakdown.

Lateral power MOSFETs used to exist which had all kinds of advantages like square SOA making them useful in linear applications and low capacitance for higher frequency operation, but vertical power MOSFETs had smaller dies making them more economical and quickly displaced them.

Is it a stupid idea to replace an IGBT by a small FET + large BJT for linear applications? Like in its equivalent circuit?

I doubt there will be any advantage.  Both parts have to sustain the full voltage, and discrete PNP transistors are not the highest voltage.

In some applications using a low voltage n-channel MOSFET in combination with a high voltage NPN cascode transistor can have some performance advantages over a single high voltage n-channel MOSFET or single high voltage NPN, but MOSFETs are always getting better and IGBTs exist now so this configuration is less common than it used to be.

[Phil1977]

Is it a stupid idea to replace an IGBT by a small FET + large BJT for linear applications? Like in its equivalent circuit?

I doubt there will be any advantage.  Both parts have to sustain the full voltage, and discrete PNP transistors are not the highest voltage.

In some applications using a low voltage n-channel MOSFET in combination with a high voltage NPN cascode transistor can have some performance advantages over a single high voltage n-channel MOSFET or single high voltage NPN, but MOSFETs are always getting better and IGBTs exist now so this configuration is less common than it used to be.

It was just loudly thinking about enhancing the power of the electronic load. But if I would convert it to BJTs I could probably completely omit the FET and control the BJTs in a Darlington circuit.

But as I said, it was only loud thinking. There are a lot of possibilities and ideas for DIY electronic loads or even better DC-DC-converters with controllable input power.

Anyhow, it´s really nice to learn about the basics and obviously the operation of modern power semiconductors in linear mode is not trivial.

[Kleinstein]

As a replacement in position a cascode circuit with 2 MOSFETs could be possibility. A lower voltage one for low capacitance and the high voltage one with the gate at a fixed voltage of some 5 to 15 V. There may still be issues turning off fast with a low votlage.

[David Hess]

It was just loud thinking about enhancing the power of the electronic load. But if I would convert it to BJTs I could probably completely omit the FET and control the BJTs in a Darlington circuit.

I looked at the economics of that problem a couple years ago.  Using bipolar transistors is a competitive solution, with big ring emitter bipolar transistors up to 200 watts available for linear audio and servo applications at reasonable prices.  Power MOSFETs have the advantage of being available in larger sizes, up to about 500 watts, so fewer packages are required, but costs are higher.