Question on mosfet in linear mode, does "hot spotting" a serious problem ?

[BravoV]

Its about using a power mosfet in linear voltage regulator as the pass element, I remember I read it somewhere about the hot spotting problem or better worded "reliability issue" when ordinary mosfet is contantly working in linear mode and dissipating a relatively high power, even the power dissipation is still "within ratings" and for the sake of the argument its also properly seated with an overkill sized heatsink as well. 

I'm aware there are specially designed & made what so called "linear" mosfet, but that is not my question, its just I'm curious and asking if its safe and "good for long term" using those jelly bean power mosfet for this purpose ?

Appreciate if anyone could share some insight or even better your own experience regarding this topic.

[bfritz]

Power MOSFET's are typically constructed as vertical structures, and are actually many parallel MOSFETs in a single package.  If the various MOSFETs have highly different characteristics, this could be a problem, but the manufacturers purposely design the structures so that metal lengths to the individual FETs is fairly well matched, and all the doping of the materials and oxide thicknesses are equivalent, leading to well matched devices.  There is some variation in current between the FETs, but this is accounted for in the specification of the typical operating specifications and absolute maximum ratings.

MOSFETs also have a nice current sharing benefit.  As a MOSFET carries more current, it heats.  As it heats, the Rds-on increases, which results in the other devices in parallel taking a larger share of the current.  So, the system shares current in a way that minimizes total heat generation for the small variations in current in the devices.

MOSFETs are operated in their linear region and show excellent reliability.  Almost every single Low Dropout Linear Regulator uses a P-channel FET as the pass element.  (Most people mistakenly refer to all linear regulators as LDOs, while the term LDO truly stands for Low Drop-Out.)

[ejeffrey]

It sounds like you are talking about second breakdown, which is a problem with BJTs (not MOSFETS) operated in the linear regime.  The problem with BJTs is that their gain increases with temperature, so the hot spots in the junction area conduct more current, leading to more heating and thus more current.  When this effect exceeds the rate of thermal diffusion, you get thermal runaway and the destruction of the device.  You can see how to deal with this issue by looking at the SOA plot in the data sheet.  At high Vce, the maximum power dissipation and junction temperature are reduced.  MOSFETs are inherently stable in this type of circuit and can be used up to their maximum rated junction temperature, although MOSFETs can display a similar thermal runaway in switching circuits.

[SeanB]

I worked on a device that had a hot spotting problem in one component. Design defect, in that the device was poorly heatsinked, but ran at a high current as a saturated switch. Not helped by the designer not including a base pull down resistor, so leakage current would eventually turn it on fully. I tried a few with a 10R pulldown, it helped with the leakage, but the longest lasting cure was to select the driver 2N2219 for low Vcesat rather than for gain. Was hard to teach the others that in a darlington you do need the resistor, and need the driver to saturate with low voltage rather than the power element, as this would still run as a linear part. The unit was a light dimmer, running a 12V 100W halogen lamp with pulse width dimming provided by using unijunction oscillators. Common fault was - lamp was bright, went out, selected spare lamp and it did not work.

[T4P]

I worked on a device that had a hot spotting problem in one component. Design defect, in that the device was poorly heatsinked, but ran at a high current as a saturated switch. Not helped by the designer not including a base pull down resistor, so leakage current would eventually turn it on fully. I tried a few with a 10R pulldown, it helped with the leakage, but the longest lasting cure was to select the driver 2N2219 for low Vcesat rather than for gain. Was hard to teach the others that in a darlington you do need the resistor, and need the driver to saturate with low voltage rather than the power element, as this would still run as a linear part. The unit was a light dimmer, running a 12V 100W halogen lamp with pulse width dimming provided by using unijunction oscillators. Common fault was - lamp was bright, went out, selected spare lamp and it did not work.

Isn't light dimmers supposed to be using triacs ?

[BravoV]

Thanks for the comments and inputs, really appreciate it.

Its been a while since I looked into this but there are a few things I recall. Most fets SOA are not tested they extrapolate. In reality the SOA  graph isn't a straight line particularly when you get out at about 60% VDS it craps out. Some devices are more suitable to linear operation logic level fets aren't one.

So for your regulator if you are building a standard 0-30V one if you select a fet with a VDS in the 150-200V range it should be more reliable and less likely to self destruct.

AcHmed99, from your experience about on 60% VDS, so this should be the "true" SOA ?

Just for an example in this discussion, I randomly grabbed a mosfet in my components junk bin, its a cheap P-mos IRF9540 VDSS=-100V, -23 Amp (datasheet -> www.irf.com/product-info/datasheets/data/irf9540n.pdf), attached two particular charts here :



Actually for constant working at linear zone, not pulsing, according to your experience on 60% mark, so these yellow regions are not even safe ? Mind explain and elaborate more on this, I'm really interested and eagerly to learn & hear from you.

I have used 500V T0-3P Fets in an electronic load blocking 200V 2 devices dissipating 80W each while blocking 200V for extended periods case temp never got much past 100C so the junction was maybe at 110C or so. These were bolted to a brick heatsink with an 80mm fan blowing right on the heatsink in free air. They are still being used periodically. I'm  using it right now actually one did just fail quite spectacularly actually Drain to gate short put 200V on my control board vaporised pretty well everything including a 50V tant that did what they do best went up in flames.

I don't believe this was due to thermal failure a wire may have inadvertently touched were it shouldn't. They were only dissipating about thirty watts each I've had them at much higher.

Sorry to hear that incident toasted your control circuit, my stuff won't even near that working voltage and current level.

Damn, drain to gate short on failure, sounds really terrible and nasty. 

[SeanB]

I worked on a device that had a hot spotting problem in one component. Design defect, in that the device was poorly heatsinked, but ran at a high current as a saturated switch. Not helped by the designer not including a base pull down resistor, so leakage current would eventually turn it on fully. I tried a few with a 10R pulldown, it helped with the leakage, but the longest lasting cure was to select the driver 2N2219 for low Vcesat rather than for gain. Was hard to teach the others that in a darlington you do need the resistor, and need the driver to saturate with low voltage rather than the power element, as this would still run as a linear part. The unit was a light dimmer, running a 12V 100W halogen lamp with pulse width dimming provided by using unijunction oscillators. Common fault was - lamp was bright, went out, selected spare lamp and it did not work.

Isn't light dimmers supposed to be using triacs ?

Triacs do not work well if your supply is 28VDC.

[T4P]

I worked on a device that had a hot spotting problem in one component. Design defect, in that the device was poorly heatsinked, but ran at a high current as a saturated switch. Not helped by the designer not including a base pull down resistor, so leakage current would eventually turn it on fully. I tried a few with a 10R pulldown, it helped with the leakage, but the longest lasting cure was to select the driver 2N2219 for low Vcesat rather than for gain. Was hard to teach the others that in a darlington you do need the resistor, and need the driver to saturate with low voltage rather than the power element, as this would still run as a linear part. The unit was a light dimmer, running a 12V 100W halogen lamp with pulse width dimming provided by using unijunction oscillators. Common fault was - lamp was bright, went out, selected spare lamp and it did not work.

Isn't light dimmers supposed to be using triacs ?

Triacs do not work well if your supply is 28VDC.

Right .  Didn't read it was a LVDC supply .

[SeanB]

Moving map display, made in the 1960's and using a whole lot of servo technology. Luckily servo spares were easy, we had a lot of old obsolete autopilot and air data computers which used the same synchros and resolvers, along with the 10 000 gears as well. Fun was the film cassette, where you had a 35mm film running in a magazine which had to rotate 360 degrees along with feeding the film through accurately. Stretch and such, along with blooming in areas exposed to the lamp for long periods.

[BravoV]

I am starting to see more realistic SOA graphs on a few FETS where they actually show reduced power capability in the right hand of the SOA they are few and far between though and it still isn't specified whether the data is actually from bench test or just less optimistic extrapolation.

Your best bet considering the scarcity of data for linear operation would be to do what I did and see if it can survive monitor the case temperature under worst case conditions you plan on using them or spend 18 to 20 bucks per fet from say IXYS in which they've done it for you. They have also taken manufacturing steps to minimize hotspots which I can't elaborate on without reading through there literature again. Oh and it's a good idea to wear safety Glass's if you plan on doing your own testing.

Actually my application is only for relatively low voltage < 30 volt dc ldo regulator, and not for commercial.

I guess its easier and safer way is to use them in parallel, rather than spending time & effort to search those linear data for these cheap jellyebean mosfet, thanks AcHmed.


Edit :

Just took a glance at the APL502B2 linear mosfet datasheet, looking at it's SOA chart below that has the extra officially bold "DC border line" which common switching mosfet doesn't have, I guess its not that overly impressive compared to ordinary mosfet.

Datasheet -> www2.microsemi.com/datasheets/APL502B2&L.pdf

[BravoV]

Achmed, thanks for those links, good reading.

So I’m using three; maximum Vds under short circuit conditions would be about 32V max current about 6A SO PMAX is 192W That’s the worst it would have to survive and they would only have to survive long enough for protection circuit to kick in. So I will test if they can survive those conditions for say a minute then I know a few seconds is survivable.

Please share the results once you finished the short circuit testing, must be really interesting to learn from there.

[BravoV]

AcHmed99, just a bump to hear from you how is that expensive mosfet doing now ? 

[digsys]

BravoV - Hi mate, ltns ... I bump into you again :-)
That bridge driver IC you found is absolutely awesome, and we're pretty much using it in all units over 5A.
Let me know if you need any more free sample PCBs :-) ... I have them with 0,1or2 35mm caps (suited for 22,000uf 35V 105c or less)
Now I've found a new high side NFET Linear reg IC which is also impressive, but run into the exact same damn problem you're
talking about. I've done dozens of hrs of research, but as you've found, it's a NOT well documented spec.
I have found devices that "work" (from trial and error .. I mean studying specs :-) ), but I can't go into production until I 100% know WHY!
Did you get anywhere with your search?

Added: I originally suspected this failure mode - www.vishay.com/docs/69087/an911.pdf  but adding G-S Zeners, resistors didn't help.

[BravoV]

BravoV - Hi mate, ltns ... I bump into you again :-)
That bridge driver IC you found is absolutely awesome, and we're pretty much using it in all units over 5A.
Let me know if you need any more free sample PCBs :-) ... I have them with 0,1or2 35mm caps (suited for 22,000uf 35V 105c or less)

Damn, its been a while since the last post, thanks for bumping this up. Also I've been quite busy lately.

Now I've found a new high side NFET Linear reg IC which is also impressive, but run into the exact same damn problem you're
talking about. I've done dozens of hrs of research, but as you've found, it's a NOT well documented spec.
I have found devices that "work" (from trial and error .. I mean studying specs :-) ), but I can't go into production until I 100% know WHY!
Did you get anywhere with your search?

What problem did you have ? Blown fet ? How much current did you use ?

Unfortunately there is no progress at all at my side, as enthusiast and also lack of experience and knowledge, all I can do is reading and asking around like this thread does.

Please keep us in the loop on your work there.

Added: I originally suspected this failure mode - www.vishay.com/docs/69087/an911.pdf  but adding G-S Zeners, resistors didn't help.

Thanks, nice document. 

Shit Sorry Bravo I missed that. Oh well 2 years late is better then never. 

I have used it in my current ELOAD I just replaced the Fairchild fets with the IXYS ones. I haven't really used them to their extremes though. I posted pictures and schematics of my bodge ELOAD  before, I can't find them now though. I haven't done it up in a nice enclosure with a user interface, but yep I have been using them, when I need to test one of my supplies.

2 years, LOL, thanks though.

Please, repost the circuit and photo of the circuit even it looks ghetto. Talking about extreme, what is the planned limit at your circuit there ?

[BravoV]

AcHmed99, thanks for the photos, much appreciated. 

Please, if you had a chance to torture it near its limit, please update us.

[David Hess]

MOSFETs also have a nice current sharing benefit.  As a MOSFET carries more current, it heats.  As it heats, the Rds-on increases, which results in the other devices in parallel taking a larger share of the current.  So, the system shares current in a way that minimizes total heat generation for the small variations in current in the devices.

It sounds like you are talking about second breakdown, which is a problem with BJTs (not MOSFETS) operated in the linear regime.  The problem with BJTs is that their gain increases with temperature, so the hot spots in the junction area conduct more current, leading to more heating and thus more current.  When this effect exceeds the rate of thermal diffusion, you get thermal runaway and the destruction of the device.  You can see how to deal with this issue by looking at the SOA plot in the data sheet.  At high Vce, the maximum power dissipation and junction temperature are reduced.  MOSFETs are inherently stable in this type of circuit and can be used up to their maximum rated junction temperature, although MOSFETs can display a similar thermal runaway in switching circuits.

At high Vds, the Rds temperature coefficient is overcome by the Vgs temperature coefficient and current sharing fails so hot spots may develop in a way similar to bipolar junction transistors in secondary breakdown.  Vertical MOSFETs are more susceptible to this than lateral MOSFETs and it is rarely shown on the SOA graph for devices intended for switching applications.

I do not think this will be a problem however with a 30 volt linear regulator.