Darlingtons used as Audio output devices....?

[IanMacdonald]

Anyway, who need so much power for a single speaker ?

DJs and mixing desk operators. Because they're all stone deaf.

Seriously, there is a problem developing with massively over-loud sound at most venues. The last time I went to the cinema, the sound was so loud as to be painful with earplugs in. A jazz concert I went to a short while before was so loud that the sound was distorting in my ears. Both must have been well over 100dB, probably approaching 110dB.

At that sort of level, it doesn't take all that long to cause permanent hearing damage. Unless you're a sadomasochist there is no pleasure whatsoever in it either, when the sound is so loud as to be distorted anyway.

I've always opposed sound level legislation in that once the nanny state gets in on the act, the limit is set so low that no live music would comply, not even acoustic. Commonsense needs to be used here. Unfortunately, bureaucracy and commonsense never mix. If is getting to the stage though, that a whole generation of hard-of-hearing people are going to be produced if something isn't done to curb this stupidity.

Part of the problem is that DJs and mixing guys typically wear highly-insulating headphones so they don't suffer the pain of their own misdeeds. Maybe they need to insist that open headsets be used. At least then the problem would be self-limiting. In most cases, anyway.

[David Hess]

As for MOSFETS, they seem like a good idea in theory but have two major issues, one that the SOA is usually much moire limited -as in can handle a jawdropping 600V or 50A, but only 2.5A and 30V when applied at the same time.

I think you've got that backwards. MOSFETs generally have a wider safe operating area, than BJTs and are capable of handling far higher currents, at the full voltage specification. Check out a few data sheets for high voltage MOSFETs and BJTs and you'll see what I mean.

For obsolete lateral MOSFETs this is almost the case but not for vertical MOSFETs.  At low currents, the temperature coefficient of the gate threshold voltage reverses so current sharing is not enforced even on a single die.  At low drain voltages this is irrelevant because power dissipation and heating are low but at high drain voltages, it results in thermal instability and current hogging.  Most MOSFET datasheets do not bother to show this in their safe operating curves because they are intended for saturated operation where this does not apply.

https://www.fairchildsemi.com/application-notes/AN/AN-4161.pdf
http://www.irf.com/technical-info/appnotes/an-1155.pdf
http://www.ixys.com/Documents/Articles/Article_Linear_Power_MOSFETs.pdf

I didn't say that the safe operating area is not a limitation with MOSFETs, just that it's generally better than BJTs, especially at higher voltages. Check out the data sheet for a typical HV MOSFET such as the IRF740B. Can you find a BJT with a safe operating area of 400V at over 300mA?
https://www.mouser.com/ds/2/308/FairchildSemiconductor_1614842276095-1191888.pdf

The IRF740B is an excellent example of what I described which I highlighted above.  Its safe operating area curve does not show the thermal instability zone or its affect on safe operating area at high voltage and low current.  Figure 2 in the datasheet shows the origin of the problem which encompasses the entire linear operating range of the IRF740B; below about 15 amps (on a 10 amp part!), the temperature coefficient of the gate threshold voltage reverses yielding positive feedback as the temperature increases.  MOSFETs which are good for linear operation have that point at a much lower current compared to their maximum current.

This is in contrast to the on resistance versus temperature curve shown in figure 8 which applies during saturated operation where the current decreases as temperature rises preventing thermally runaway.  This is what makes MOSFETs operate well in parallel switching applications without source ballast resistors.

The Toshiba TK8A50D datasheet shows what the safe operating curve should look like.  Check out the safe operating area curve of the IXYS IXTP8N50P for a very conservative example which I would trust.

[David Hess]

Some mosfet technologies are a bit limited regarding DC operation but, Infineon still makes fully DC  operable high voltage mosfets.

For example IPW90R1K0C3. I have used those in my high voltage power supply.

How do you know that Infineon part is rated to operate over the entire DC operating area?  The datasheet does not say anything about linear operation.

Some manufacturers (like BK Precision) think it is okay to use non DC rated mosfets in linear mode. Just look up one of the last Shahriars video, where he repaired the blown HV PSU. There was a STW15NK90Z used.

Oh, it can definitely be done.  But the MOSFETs need to be derated significantly beyond what the safe operating area curves imply if they do not take thermal instability into account which is normally the case.

[Kleinstein]

The chances that the Infineon MOSFET is Ok for DC operation is good. The modern very high voltage MOSFETs (e.g. > 600 V) use a special technique, that makes them less sensitive to 2 nd breakdown. Also Infineon in many cases includes the effect of thermal instability in there SOA curves - so there is a good  chance that the SOA curve is trustworthy.  Some other companies tend to calculate there SOA from measured/calculated thermal impedance and thus ignore possible thermal instability in there SOA.

One might still need individual testing of the chips for high power linear operation, as there is always a small chance to have something like small voids in the die attachment (or local contamination) that could cause premature failure for a few rare samples, even with other wise good types. This also applies to BJTs - here audio transistors with individual FBSOA testing are available.

[janoc]

Part of the problem is that DJs and mixing guys typically wear highly-insulating headphones so they don't suffer the pain of their own misdeeds. Maybe they need to insist that open headsets be used. At least then the problem would be self-limiting. In most cases, anyway.

That's not quite that simple - the reason for the insulating headphones is that the DJ has to be able to hear whatever are they setting up over the noise in the venue. So open headset would be unusable.

But otherwise I fully agree with you - both on many DJs being stone deaf ("louder is better!"), many having the intelligence of a stoned monkey (clipping indicators being permanently on on everything from the console to the amps and the perp was not concerned - until the amp finally blew in the middle of the show with a ton of smoke causing an evacuation ...), etc.

The cinema techs are a bit different category - in most cases with the new fully digitalized cinemas these are just regular employees without any special training. They were only shown how to turn the thing on and off and which button to push to start a movie. The idea of actually walking down do the room and checking the sound there instead of only relying on whatever monitors they have in the projection room above is likely something that has never occurred to them (and if it did, they wouldn't know how to adjust the sound anyway). They may have one or two better trained guys that handle loading of new films and basic maintenance, but those are not there all the time.

[T3sl4co1l]

I didn't say that the safe operating area is not a limitation with MOSFETs, just that it's generally better than BJTs, especially at higher voltages.

I had read your post as: "MOSFETs are better than BJTs", which isn't to say they are perfect, just that they are less bad.

I suppose it's still good to make note that a beginner might read it and say "MOSFETs are flawless", which would be a mistake on their part, but one which can be avoided if the matter is emphasized.

For anyone keeping track: all transistors suffer from 2nd breakdown, it's just a matter of "how much?".  IGBTs are categorically the worst, not being suitable for DC operation at all (at least at any useful voltage).  BJTs are better, but it depends on type.  MOSFETs are better still, but it again depends.

The root cause is too much power in too little die area, and an unfavorable tempco that causes hot spots and runaway destruction.  All devices suffer this, but designs can be made in such a way that runaway falls beyond the power dissipation limit.

You rarely see 2nd breakdown limitations on very small devices -- SOT-89s and TO-220FPs, say -- because there's not enough power dissipation in the first place to get into the runaway region.

Lateral MOSFETs avoided this by being such sucky designs in the first place -- such low power density that there wasn't enough temp difference across the huge die to get close.

BJTs have long been made in two styles: those for linear operation, and those for switching.  (There are other differences, making the switching types even better suited for switching; this is a gross simplification.)  A typical linear example might be a part rated 250V, 20A, 150W, with a small derating (due to 2nd breakdown) above 150V. 
In contrast, a switching part might be 800V, 20A, and 100W but derated sharply above 50V.  At 200V, the linear device might handle a full 100W still, while the switching part must be limited to a couple of watts at the same voltage!

Paradoxically, the newest generation of high voltage MOSFETs (SuperJunction type -- if the datasheet headline doesn't say SuperJunction, CoolMOS, QFET, MDmesh M2 or other trademarks, look at the capacitance curve, which is distinctive: Coss tanks above 20-50V, then rebounds slightly at high voltages), despite having higher power density than ever (more amps in less die area, while still handling high voltages -- and high voltage * high current = super high power).  I guess the difference is in tempco.  That or they're all lying...

Tim

[Kleinstein]

There are very different MOSFET types. Some of the modern switching types for low voltage (e.g. 30 V) have a really poor FBSOA. Infineon sometimes even has that curve for the small types (E.g. SOT223), and some are really poor showing a reduction due to 2nd breakdown starting at below 12 V.

When looking for linear power use of a MOSFET, definitely look for higher voltage (e.g. > 200 V) types.

[David Hess]

The chances that the Infineon MOSFET is Ok for DC operation is good. The modern very high voltage MOSFETs (e.g. > 600 V) use a special technique, that makes them less sensitive to 2 nd breakdown. Also Infineon in many cases includes the effect of thermal instability in there SOA curves - so there is a good  chance that the SOA curve is trustworthy.  Some other companies tend to calculate there SOA from measured/calculated thermal impedance and thus ignore possible thermal instability in there SOA.

But the Infineon datasheets and applications notes do not say anything about linear operation and safe operating area of Coolmos parts and if they were suitable, it sure seems like marketing would make it known considering they costs from 1/4th to 1/10th as much as similar linear rated MOSFETs.  Higher voltage parts have a much larger die size for lower junction to case thermal resistance for the same Rds(on) and greater resistivity which helps.  High voltage parts effectively include derating and source ballasting but that applies to any high voltage MOSFET and not just Infineon's Coolmos parts.

I think it much more likely that like most other manufacturers, they leave that part off of the safe operating area curve because it is irrelevant for the intended switching applications and difficult to quantify.  It also makes their parts look worse so why advertise it?

One might still need individual testing of the chips for high power linear operation, as there is always a small chance to have something like small voids in the die attachment (or local contamination) that could cause premature failure for a few rare samples, even with other wise good types. This also applies to BJTs - here audio transistors with individual FBSOA testing are available.

FBSOA testing is pretty destructive.  I think I read an application note which mentioned screening using x-ray imaging to look for die attachment and drain metalization problems in linear rated parts.  It seems like a high resolution thermal camera would work to look for hot spots before encapsulation.

[floobydust]

Integrated Darlingtons are rarely used in audio, instead the discrete Darlington or Sizlaki or triple emitter follower, so 2-3 discretes.

I think it's because there is no access to the middle node, at the base of the last transistor. This usually connects to a resistor on the opposite side, even in STK's.

[T3sl4co1l]

But the Infineon datasheets and applications notes do not say anything about linear operation and safe operating area of Coolmos parts

They say something: if there's a DC SOA curve.  Which most (all I've seen?) CoolMOS parts do.

For a single data point, I've tested a Fairchild (now On Semi) QMOS to destruction, which failed IIRC 10% beyond RthJC spec.  This was done at high voltage (300V or more, I forget).  QMOS (the ones I've seen) also have DC SOA curves.

Higher voltage parts have a much larger die size for lower junction to case thermal resistance for the same Rds(on) and greater resistivity which helps.

Compared to... lower voltage parts in a given generation?  Compared to prior generations?  Or what?  Not sure what you're getting at here.

High voltage parts effectively include derating and source ballasting but that applies to any high voltage MOSFET and not just Infineon's Coolmos parts.

Hmm, could you explain this more?  What do you mean by derating and ballasting?

I think it much more likely that like most other manufacturers, they leave that part off of the safe operating area curve because it is irrelevant for the intended switching applications and difficult to quantify.  It also makes their parts look worse to why advertise it?

Another part I tested to destruction at voltage, a Siliconix IRF740, does not provide DC SOA -- even though prior models of the part (IR's original) did document DC SOA.  It failed at something like 40 or 60% over ratings.  It seems, they're not kidding when they say RthJC maximum -- the typical value is evidently much lower.  Either that, or Tj(max) is over 200C, which seems less likely.

The QFET had a nice small die, like 1.3 mm square.  The IRF740 was huge, like 2.4 x 4.0 mm.  Clearly they hadn't die shrunk that part, even though they have plenty of room to do so, given how much extra RthJC they're making.

FBSOA testing is pretty destructive.  I think I read an application note which mentioned screening using x-ray imaging to look for die attachment and drain metalization problems in linear rated parts.  It seems like a high resolution thermal camera would work to look for hot spots before encapsulation.

AFAIK, there's a method to perform SOA testing without destruction -- just barely toeing the line, with a control circuit to protect it.  I forget the exact mechanism though.

The manufacturer does have the advantage of working with unencapsulated parts so they can measure die temperatures through alternative means.  External sensing isn't going to be fast enough to protect the device, though.

I think the SOA test uses electrical properties -- watching d/dt (runaway) tendancies, or increasing pulse durations followed instantly by measuring die temp via body diode tempco (which is a useful method, as the hottest spot has the least voltage drop, so you're mostly measuring the peak temp of the die, rather than the average over the die area).

Anyway, worth investigating if you'd like to know more.  I'm not testing these things on a commercial basis, so I only know what I've seen.

Tim

[David Hess]

Integrated Darlingtons are rarely used in audio, instead the discrete Darlington or Sizlaki or triple emitter follower, so 2-3 discretes.

I think it's because there is no access to the middle node, at the base of the last transistor. This usually connects to a resistor on the opposite side, even in STK's.

That is one good reason.  Performance is better with the emitter shunt resistors, R5 and R6 in your example, connected between emitters and not to the output as they would be with integrated Darlingtons.  They still work though for lower performance at a lower cost.

The predriver and driver emitter resistors are each returned to the emitter of the complementary device. This allows for improved turnoff of the subsequent device by reverse bias. It also increases the input impedance of the output stage because the emitter resistors have essentially no signal across them. - Bob Cordell

I have deliberately called this an Emitter-Follower (EF) rather than a Darlington con?guration, as this latter implies an integrated device that includes driver, output, and assorted emitter resistors in one ill-conceived package.  (Ill-conceived for this application because the output devices heat the drivers, making thermal stability worse.) - Douglas Self

The shared driver emitter resistor, with no output-rail connection, allows the drivers to reverse-bias the base-emitter junction of the output device being turned off. - Douglas Self

[David Hess]

But the Infineon datasheets and applications notes do not say anything about linear operation and safe operating area of Coolmos parts

They say something: if there's a DC SOA curve.  Which most (all I've seen?) CoolMOS parts do.

Most power MOSFET specifications include a DC SOA curve but leave out the effect of thermal instability.  The cake curves are a lie.  The thermal instability also applies under pulsed operation just like secondary breakdown in a bipolar transistor but they usually do not show that either.  I linked a pair of examples where this is shown.

Higher voltage parts have a much larger die size for lower junction to case thermal resistance for the same Rds(on) and greater resistivity which helps.

Compared to... lower voltage parts in a given generation?  Compared to prior generations?  Or what?  Not sure what you're getting at here.

For the same Rds(on), the die size is roughly proportional to the voltage rating squared which is *not* the case with bipolar transistors and IGBTs.  This is what makes bipolar transistors and IGBTs more economical at higher voltages despite their slower switching speeds.  High voltage MOSFET dies take a lot of area compared to low voltage ones and other transistor types.

For linear applications, this means high power MOSFETs are likely to be very voltage derated.  I usually find it more useful to search for power MOSFETs based on power and then current.  Otherwise, it is easy to end up with say an 80 watt n-channel part and a 120 watt p-channel part with greater dynamic differences.  Unfortunately, some distributors like Mouser do not list transistors by power rating.

High voltage parts effectively include derating and source ballasting but that applies to any high voltage MOSFET and not just Infineon's Coolmos parts.

Hmm, could you explain this more?  What do you mean by derating and ballasting?

Many of the characteristics which make a power MOSFET with a smaller thermal instablity region are shared with high voltage MOSFETs.  A larger die decreases the junction to case thermal resistance helping to prevent hot spots.  The higher voltage construction increases resistivity which compensates for some of the Vgs thermal coefficient.

I think the SOA test uses electrical properties -- watching d/dt (runaway) tendancies, or increasing pulse durations followed instantly by measuring die temp via body diode tempco (which is a useful method, as the hottest spot has the least voltage drop, so you're mostly measuring the peak temp of the die, rather than the average over the die area).

Motorola has a lot of old application notes discussing it but maybe techniques have gotten better.  I do not remember any articles discussing power MOSFET safe operating area in that kind of detail but more recent stuff tends to be behind a paywall or not published publicly.

There are some interesting comments in the Fairchild application note I linked.  High current induced failures occur under the bond wires where the current is greatest while thermal instability failures occur away from the bond wires which act as a heat sink helping to prevent the formation of hot spots.  Imperfections in the die attachment and metalization provide points for likely thermal instability failures.

[T3sl4co1l]

Most power MOSFET specifications include a DC SOA curve but leave out the effect of thermal instability.  The cake curves are a lie.  The thermal instability also applies under pulsed operation just like secondary breakdown in a bipolar transistor but they usually do not show that either.  I linked a pair of examples where this is shown.

Hmm, so you're saying manufacturers publish DC SOA, and are lying?

Then you must be accusing me of lying.  I tested two devices to failure outside of their DC SOA, just as it should be.  I don't much appreciate that.

For the same Rds(on), the die size is roughly proportional to the voltage rating squared which is *not* the case with bipolar transistors and IGBTs.

Ah, previous generations were -- SuperJunction is proportional.  Which is why it so thoroughly destroys the specs of earlier parts -- compare, say, a 600-1200V PolarHV part to a MDmesh M2 part of the same rating.  (Or, IXYS licenses their own line of SJ now, IIRC, if you prefer to stay within the same brand.  Speaking of IXYS, a lot of their PolarHV or HiPerFET or whatever parts, didn't even bother with an SOA at all -- you have no way to know if those beasts are even safe for a single switching event!  Scary!)

And which makes it all the more remarkable that SJ has a full SOA, and that's no lie.

Many of the characteristics which make a power MOSFET with a smaller thermal instablity region are shared with high voltage MOSFETs.  A larger die decreases the junction to case thermal resistance helping to prevent hot spots.  The higher voltage construction increases resistivity which compensates for some of the Vgs thermal coefficient.

Ah, okay.

Why do SuperJunction FETs have full SOA?

There are some interesting comments in the Fairchild application note I linked.  High current induced failures occur under the bond wires where the current is greatest while thermal instability failures occur away from the bond wires which act as a heat sink helping to prevent the formation of hot spots.  Imperfections in the die attachment and metalization provide points for likely thermal instability failures.

Sounds reasonable.

I didn't see anything about this "instability region" you're talking about, though.  Is this an imaginary region on the SOA where thermal drift (increasingly, until failure) occurs?

All SOAs shy away from this region; I'm not sure how you got the idea that manufacturers lie about it.

Have you tested any parts to destruction that showed otherwise?  I've shared my data.

We should be so lucky that everything in our domain can be proven by measurements!

Tim

[David Hess]

Most power MOSFET specifications include a DC SOA curve but leave out the effect of thermal instability.  The cake curves are a lie.  The thermal instability also applies under pulsed operation just like secondary breakdown in a bipolar transistor but they usually do not show that either.  I linked a pair of examples where this is shown.

Hmm, so you're saying manufacturers publish DC SOA, and are lying?

It is routine for manufacturers to leave it out.  This is even somewhat justified for parts intended for switching applications and not linear applications but it is misleading.

Note that like bipolar transistors, low voltage MOSFETs do have a square SOA because at low voltages it is never possible to operate out of saturation with enough power density to cause thermal runaway.  And like bipolar transistors, it is the high voltage parts which have this problem.  The Vgs temperature coefficient is greater at lower currents where it requires a higher voltage to produce enough power for instability to be a problem.

I was rather alarmed a couple of years ago however when I found power MOSFETs where the thermal instability region of the SOA curve started at voltages lower than secondary breakdown of a mediocre bipolar power transistor.  At least it was documented.

Then you must be accusing me of lying.  I tested two devices to failure outside of their DC SOA, just as it should be.  I don't much appreciate that.

Safe operating area is difficult to test and maintain in a production environment.  Motorola published some hair raising application notes about operating transistors outside of their published safe operating area curves after sample testing which could be done by the end user.

For the same Rds(on), the die size is roughly proportional to the voltage rating squared which is *not* the case with bipolar transistors and IGBTs.

Ah, previous generations were -- SuperJunction is proportional.  Which is why it so thoroughly destroys the specs of earlier parts -- compare, say, a 600-1200V PolarHV part to a MDmesh M2 part of the same rating.  (Or, IXYS licenses their own line of SJ now, IIRC, if you prefer to stay within the same brand.  Speaking of IXYS, a lot of their PolarHV or HiPerFET or whatever parts, didn't even bother with an SOA at all -- you have no way to know if those beasts are even safe for a single switching event!  Scary!)

And which makes it all the more remarkable that SJ has a full SOA, and that's no lie.

Are the SuperJunction parts proportional?  I know they are improved but that would be a rather large improvement.  By itself however, that would make their safe operating area worse because of an increase in junction to case thermal resistance.  See below about the thermal instability curve.

I never had any preference for IXYS until I discovered that they published more detailed SOA curves and now that is something I look for if it matters.  The old IRF SOA curves showed *nothing* but Siliconix admitted that there was a potential issue in their even older application documentation but I do not remember now if they identified thermal runaway outside of paralleling separate power MOSFETs.  Everybody of course stressed operation in saturation where the temperature coefficient of Rds(on) enforces current sharing and thermal runaway is not a problem even between separate devices.

Many of the characteristics which make a power MOSFET with a smaller thermal instablity region are shared with high voltage MOSFETs.  A larger die decreases the junction to case thermal resistance helping to prevent hot spots.  The higher voltage construction increases resistivity which compensates for some of the Vgs thermal coefficient.

Ah, okay.

Why do SuperJunction FETs have full SOA?

If you mean full SOA within the power dissipation bounds, I am not convinced that they do.  Not even all of the IXYS parts intended for linear applications have a square SOA but maybe they do not use SuperJunction construction.  Instead IXYS makes a point in the datasheet that the shown SOA is guaranteed which is an odd thing to say unless other datasheets included an SOA which was not guaranteed and the later is my experience.  IXYS charges a premium for that guarantee.

There are some interesting comments in the Fairchild application note I linked.  High current induced failures occur under the bond wires where the current is greatest while thermal instability failures occur away from the bond wires which act as a heat sink helping to prevent the formation of hot spots.  Imperfections in the die attachment and metalization provide points for likely thermal instability failures.

Sounds reasonable.

I didn't see anything about this "instability region" you're talking about, though.  Is this an imaginary region on the SOA where thermal drift (increasingly, until failure) occurs?

This is the area where the combination of Vgs temperature coefficient produced by operating out of saturation and thermal resistance allows for thermal runaway.  At lower power or lower temperature, thermal runaway will not occur despite a high Vgs temperature coefficient.  Lowering the transconductance via source ballasting, lowering the junction to case thermal resistance, and further suppressing the parasitic bipolar transistor all reduce the area of the instability region.

All SOAs shy away from this region; I'm not sure how you got the idea that manufacturers lie about it.

Have you tested any parts to destruction that showed otherwise?  I've shared my data.

We should be so lucky that everything in our domain can be proven by measurements!

I got the idea because there are tons of old and new high voltage power MOSFET datasheets which show a completely square safe operating area meeting the power dissipation limit when that cannot be the case.  And I have a drawer full of the remains of blown up plastic packaged MOSFETs from when I investigated why they were failing.  (1) Expediency meant that we never tracked down the exact reason as it simply did not matter; increased power derating solved the problem nicely.  Knowing what I know now, I suspect thermal runaway during current limiting was triggering the parasitic bipolar transistor which is a sure path to part failure.  I think the older parts originally specified had a better incompletely documented SOA despite the same voltage, current, and power ratings but newer generations of effectively the same part had a worse incompletely documented SOA.  I recall this being explained by greater cell density of newer processes but have forgotten the details.

(1) Not all of the blown up transistor pieces were recovered.  Some pieces stuck in the walls, ceiling, me, etc.