IGBT Linear operation

[Phil1977]

Hi,

following situation: I have an ATORCH DL24 electronic load and today its MOSFET (CS50N20) blew up when it reached 200W for a few seconds.

I had no appropriate replacement available, but a box of original Infineon IRGP4068. I thought why not give it a try, the gate threshold voltage is similar and in an electronic load it really doesn't matter if they have some higher forward voltage.

The regulation seemed to work, I could start the electronic load with low load settings like 5V 1A and checked with the scope that nothing is oscillating.

But already at 100W and with around 40°C heatsink temperature the IGBT was killed again. Acc. to absolute max ratings it is allowed to dissipate 330W at 25°C and 170W at 100°C - so plenty of margin there.

The power source was a current limited power supply, and I didn't notice any peak in its current. It seems the IGBT died just dissipating around 100W in linear operation.

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?

What MOSFET can be recommended for linear use up to 100V / 20A ?

[Andy Chee]

IGBTs can be used linearly but not very well, and somewhat derated.  The junction isn't designed for linear operation.

For linear MOSFETs, try this:

https://www.littelfuse.com/products/power-semiconductors/discrete-mosfets/n-channel-linear/standard_linearmode.aspx

[schmitt trigger]

Please thoroughly read AND understand the following white paper, there are many more:

https://www.nexperia.com/applications/interactive-app-notes/IAN50006_Power_MOSFETs_in_linear_mode

[wraper]

100W from a single mosfet is wishful thinking. Even more so for IGBT in linear mode.

to absolute max ratings it is allowed to dissipate 330W at 25°C and 170W at 100°C

Not in linear mode and good luck trying to keep such case temperature at this much power dissipation.

[Andy Chee]

Acc. to absolute max ratings it is allowed to dissipate 330W at 25°C and 170W at 100°C - so plenty of margin there.

By the way, just in case you were not aware, the absolute max ratings is NOT the same as SOA.

For SOA, you need to refer to the graph in the datasheet.

[T3sl4co1l]

Please thoroughly read AND understand the following white paper, there are many more:

https://www.nexperia.com/applications/interactive-app-notes/IAN50006_Power_MOSFETs_in_linear_mode

Unfortunately, misinformation on this subject is rampant --

Below the ZTC point, if a small region is at a higher temperature than the rest of the die, it will draw more current and dissipate more power becoming even hotter.

This is blatantly false -- but in a subtle way that is easily missed.

The NTC is a necessary BUT NOT SUFFICIENT factor for instability.

The realization is simple.  If two thermally-independent transistors are wired in parallel, operated in the NTC region, there will be some temperature difference between them -- not necessarily fatal (total power could be less than individual ratings), but also not wholly disparate, i.e. they will still share some current, if just not all of it.

But that temperature difference depends on thermal resistance.  Namely, the differential thermal resistance between devices (or, between different points on a given die).

It's perfectly possible, acceptable, and normal, to operate in the NTC region, at power levels that are low in relation to the thermal resistance within the part.

And since a die (and the copper backing) are quite conductive, it's easily possible to have a full SOA free of 2nd breakdown, in the NTC region.  It's also possible to have very little SOA, where the NTC is just that strong, or thermal resistance is poorer.

Which is also why small devices simply don't have 2nd breakdown; they overheat completely, not enough temperature drop across the thing to activate the instability.

The most correct (but still ultimately absolutely inconclusive -- but necessarily so) appnote I know of is
https://www.infineon.com/dgdl/Infineon-ApplicationNote_Linear_Mode_Operation_Safe_Operation_Diagram_MOSFETs-AN-v01_00-EN.pdf?fileId=db3a30433e30e4bf013e3646e9381200
which concludes in the lamest possible way at the top of page 11: "This means that the temperature coefficient must be known. The latter cannot be easily calculated."  In other words, because manufacturers don't publish mechanical construction of their devices (for obvious reasons?), we can only know about stability by the SOA.

---

Regarding IGBTs, generally you would assume they are worse, as current density is much higher, and the minority carrier mechanism should tend to have even worse tempco.  Indeed, many don't have a SOA plot at all, and of those that do, many only give very short time scales (that might be relevant for overload conditions but not DC).

That said, I've seen more and more lately, it seems like, that show a wide SOA -- https://www.onsemi.com/pdf/datasheet/ngtb40n120fl3w-d.pdf for example.

I might not trust that SOA very much, and would want to run a spot test on a sample device before putting something into production that depends on the behavior.

Tim

[DavidAlfa]

MOSFET works with voltage, IGBT with current, so they can't replace each other.
Not that they can't work in linear mode, but the circuit must be designed to handle them.
Your question is like asking "Is diesel a bad fuel? My gasoline car runs badly with it". - That's what happens after a very long week + bad sleeping 

Not in linear mode and good luck trying to keep such case temperature at this much power dissipation.

What? 330w continuous is 330w  continuous - period.
330w loses in switching is basically linear operation anyways (Rising/falling edges).
Though that power is a bit crazy for a single transistor in "small" package.

200W is just too much anyways, thats why it blowed in first place!
Chinese crap... to handle 300W properly you would need at least 3 or 4 mosfet working in parallel.

[wraper]

Not in linear mode and good luck trying to keep such case temperature at this much power dissipation.

What? 330w continuous is 330w  continuous - period.
330w loses in switching is basically linear operation anyways (Rising/falling edges).
Though that power is a bit crazy for a single transistor in "small" package.

Have you ever heard about SOA? The issue with linear region is hotspots on the die. Not to say you'd need liquid nitrogen cooling to reach conditions for 330W.

[Phil1977]

Thanks for all the answers - it seems this topic is not really easy if there are several long and exhausting application notes..

Just a few comments:

200W is just too much anyways, thats why it blowed in first place!
Chinese crap... to handle 300W properly you would need at least 3 or 4 mosfet working in parallel.

You can say bad things about China as much as you want, but this time it was my fault. This unit was limited to 150W, I deliberately increased the limit due to a simple misunderstanding that is related to the thermal capabilities:

100W from a single mosfet is wishful thinking. Even more so for IGBT in linear mode.

to absolute max ratings it is allowed to dissipate 330W at 25°C and 170W at 100°C

Not in linear mode and good luck trying to keep such case temperature at this much power dissipation.

Of course, to cool 330W from a TO247 component is difficult. But I didn't want to operate it continuously with 200W. And that´s where usually the dissipation value from the absolute max ratings gets relevant. If there is no pulsed operation specified, but if the pulse power stays below the max dissipation then for many semiconductors the world is okay. But obviously not for FETs / IGBTs in linear operation.

Regarding IGBTs, generally you would assume they are worse, as current density is much higher, and the minority carrier mechanism should tend to have even worse tempco.  Indeed, many don't have a SOA plot at all, and of those that do, many only give very short time scales (that might be relevant for overload conditions but not DC).

That said, I've seen more and more lately, it seems like, that show a wide SOA -- https://www.onsemi.com/pdf/datasheet/ngtb40n120fl3w-d.pdf for example.

I might not trust that SOA very much, and would want to run a spot test on a sample device before putting something into production that depends on the behavior.

That´s what´s really confusing me. In my naive sight on the electronic world I usually think that if you follow all the limits in the datasheet then the component should expect a long life. But there obviously are SOA limitations that are not content of the basic datasheet! In case of the IRGP4068D I can find only a SOA diagram for the forward diode and this is just a rectangle 

MOSFET works with voltage, IGBT with current, so they can't replace each other.
Not that they can't work in linear mode, but the circuit must be designed to handle them.
Your question is like asking "Is diesel a bad fuel? My gasoline car runs badly with it".

I thought the same before replacing the part. But in linear mode and controlled by a heavily dampened opamp circuit I think they may be exchangeable. Roughly simplified they are both kind of a voltage controlled current source. Of course the regulation behaviour will differ but that´s why the first thing I checked was if there are any oscillations - and there were none.

In conclusion: I´ll get one of these expensive FETs intended for linear operation from little fuse and I´ll keep the power limit at 150W. But thanks again for all the insights, it´s really nice here that you get so much expert knowledge 

[Slh]

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.

[wraper]

Thanks for all the answers - it seems this topic is not really easy if there are several long and exhausting application notes..

Just a few comments:

200W is just too much anyways, thats why it blowed in first place!
Chinese crap... to handle 300W properly you would need at least 3 or 4 mosfet working in parallel.

You can say bad things about China as much as you want, but this time it was my fault. This unit was limited to 150W, I deliberately increased the limit due to a simple misunderstanding that is related to the thermal capabilities:

Which is still a deception. A half decent electronic load would have 4x the number of MOSFETs for this power rating. 150W is an extreme abuse which won't last long, even if done for a few seconds at a time. Especially with a piss-poor cooler it has. It's would need high-end heat-pipe cooler with liquid metal thermal interface to keep case temperature reasonable.

[Phil1977]

No one forces you to buy this thing - if you have something better then no one wants to talk bad about it.

But please lay down that hubris that everything needs to be thick and heavy and oversized. 150W is a common TDP for CPU coolers, and the T_j of a FET may be significantly higher than of a CPU.

And the effect of "liquid-metal" thermal paste is more a hocus pocus esoteric of the gamer scene. A well applied thin conventional paste is well capable of transferring the heat.

Let´s please come back on topic about FET and IGBT linear operation. I´m absolutely sure that a well defined SOA can tell you how to reliably operate a FET.

[wraper]

^You forgot several things: 1. cooler used in this electronic load is something that would come with 65W TDP CPU at best. CPU silicon die and heat spreader size are much larger than this MOSFET die/package, so there isn't anywhere as much power density in CPU. 3. In linear mode some spots on the die heat up way more than the die on average.

[Andy Chee]

I´m absolutely sure that a well defined SOA can tell you how to reliably operate a FET.

Here's the SOA for your CS50N20



Note the Tj specification.  There is no way your current cooling arrangements are capable of maintaining that specification.

[Phil1977]

The IXTH80N20L specifies the SOA for 75°C Heatsink temperature:



This still looks like 300W of allowed dissipation at 30V/10A. Of course the PC cooler that was delivered with the cheap electronic load can not perform that well, but I can try and monitor the temperature where the FET is mounted.

If this component still gets fried with 150°C I´ll give up and get an other electronic load.

[Andy Chee]

remember to subtract your room temperature from that 75 degree allowance.

For example, if your room temperature is 25 deg C, then your heatsink is only permitted to rise 50 additional degrees before MOSFET failure. 

There's no way to dissipate 300W with only 50 degrees rise from your tiny cooling setup!  That explains the commentary re: exotic cooling techniques.

[moffy]

Monitoring the static temp of the heatsink in such a situation doesn't take into account the impulse response of the heatsink or how quickly a thermal impulse is spread from a point. This is the most likely cause of a failure, excessive junction temperature, pity it doesn't have an integral diode for temperature monitoring.

[Andy Chee]

A lesson in thermal calculations might be in order here (courtesy of TI):

https://www.ti.com/lit/an/slva462/slva462.pdf

[wraper]

remember to subtract your room temperature from that 75 degree allowance.

For example, if your room temperature is 25 deg C, then your heatsink is only permitted to rise 50 additional degrees before MOSFET failure. 

There's no way to dissipate 300W with only 50 degrees rise from your tiny cooling setup!  That explains the commentary re: exotic cooling techniques.

Not heatsink but case temperature that will be significantly higher than heatsink temperature. Not to say heatsink surface under the package by itself will be much hotter than the rest of the heatsink.

[Phil1977]

That´s exactly why I trust the CPU coolers to be quite capable there. They are designed to keep a CPU under 80°C in a case that´s already 45°C or more. As far as I remember typical CPU fans have around 0.2K/W or better from the heat spreader to ambient. That means that 150W dissipation (I never told I want to use the 300W that are allowed in the SOA. It´s just good to have some margin) heat up the heat spreader of the CPU cooler to 30K over ambient.

And the electronic load is not running in a closed housing but freely on the bench with not much more than 25°C. So 55°C at the heat spreader seem feasible.

Please lets stop this discussion about heat management here. As soon as I get the new FET I will insert a thermocouple into its housing. The TO-247 package easily allows to drill a small sinkhole next to the mounting hole. A thermocouple with some thermal paste inside the package should give quite a reliable temperature measurement. I´m quite sure I can keep the temperature of the tab itself below 75°C - if the standard CPU cooler is not enough I still have a quite serious water cooler available.

[schmitt trigger]

Please thoroughly read AND understand the following white paper, there are many more:

https://www.nexperia.com/applications/interactive-app-notes/IAN50006_Power_MOSFETs_in_linear_mode

Below the ZTC point, if a small region is at a higher temperature than the rest of the die, it will draw more current and dissipate more power becoming even hotter.

The NTC is a necessary BUT NOT SUFFICIENT factor for instability.


The most correct (but still ultimately absolutely inconclusive -- but necessarily so) appnote I know of is
https://www.infineon.com/dgdl/Infineon-ApplicationNote_Linear_Mode_Operation_Safe_Operation_Diagram_MOSFETs-AN-v01_00-EN.pdf?fileId=db3a30433e30e4bf013e3646e9381200
which concludes in the lamest possible way at the top of page 11: "This means that the temperature coefficient must be known. The latter cannot be easily calculated."  In other words, because manufacturers don't publish mechanical construction of their devices (for obvious reasons?), we can only know about stability by the SOA.

Learned something additional today. That is the reason I mentioned in my original post that there are several other app notes related to this topic. Each carries some kernels of useful information.

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

[Phil1977]

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.

The FET that originally was in the electronic load was probably overloaded by my 200W-excursion. And it was already running for many hours with 150W before, but that of course doesn't mean it was safe (luckily it didn't blow up when connected to unprotected Makita batteries before. This would probably have created some fireworks  )

[wraper]

That´s exactly why I trust the CPU coolers to be quite capable there. They are designed to keep a CPU under 80°C in a case that´s already 45°C or more. As far as I remember typical CPU fans have around 0.2K/W or better from the heat spreader to ambient. That means that 150W dissipation (I never told I want to use the 300W that are allowed in the SOA. It´s just good to have some margin) heat up the heat spreader of the CPU cooler to 30K over ambient.

Type of the heatsink used in this electronic load is the worst that are barely able to cool low end CPUs under load. There is nothing to trust, it's the simplest shit with no heat pipes.

And the electronic load is not running in a closed housing but freely on the bench with not much more than 25°C. So 55°C at the heat spreader seem feasible.

Only in your dreams. Even high end water cooler will struggle to do it with such heat dissipation over area this small.

[Jeroen3]

If you look at commercially made dc loads they limit the dissipation per transistor to sub 50 W.
Eg this Array one has 6 mosfets for 300W.

Cooling solution quickly becomes incredibly tricky if you go higher. Try to limit your load to 50W, it will save you mosfets.

The limits in the datasheet are measured with an infinity heatsink, taking it to the package limits (Rthj-case).
If the heatsink temperature goes up, the delta between junction and case shrinks, limiting your overall power budget.

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) -> Spirito effect!

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.

[T3sl4co1l]

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.

Again, like IGBTs, their current density is higher than ever; that would seem to be a problem.  My guess is, because the die can be thinner and smaller for the same voltage rating, the thermal resistance across it is smaller.  There may be other tweaks that are possible, that just happen to give better SOA as well (maybe quirks of the design make it more amenable to full SOA, so why not adjust it a little further to get that too? etc.).

Classic HEXFETs are generally fine -- IRFP460 for instance, or friends, if ratings are adequate for your operating range.  They were originally published with full DC SOA curves. Most/all manufacturers have dropped that for some reason, so, no design assurance sadly, but so far I haven't heard of one that didn't pass the test (and was legitimately sourced).  Later generation HEXFETs ("advanced" if at all; they mostly dropped the number after 5th gen) tend to have high enough density not to offer full SOA anymore -- but those are mostly in lower voltages I think. Compare IRFP4768 for instance.

Incidentally -- or, at least AFAIK -- SJ MOSFETs have a lateral structure on top, just like HEXFETs and their striped cousins; this is different from the vertical channel of trench MOS (used for lower voltage parts), but I don't think it has any important consequences like for tempco, as can be seen from the overall curves.

Tim