Output transistor safe operating area

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Jump to Latest
B

Bob Cordell

Member
Joined 2006
cordellaudio.com
megajocke said:
I agree, great post Bob! Thanks.

I'm quite interested in the safety margins myself so have done some research on the issue. As for what happens at different temperatures the 150 degree (or 175 for some parts) limit is, from what I can find, a limit that comes from the plastic material itself. Temperature should be kept below this so that the package does not degrade. Keeping average junction temperature below 150 degrees should even be enough out of concern for the plastic.

Just as you say, the destruction temperature is much higher than this value. For example, the "SPiKe" protection in the LM3886 is set to activate at a junction temperature of 250 degrees C and this part uses BJT outputs, so it's not just MOSFET:s that survive high temperatures. (SPiKe AN)

John Larkin on sci.electronics.design did some tests with power MOSFET:s similar to those you have made and he came to the conclusion that failure happens at around 350 degrees C for those. At 300 degrees, they wouldn't turn off however so in a typical amplifier circuit they would be doomed already by then. (Try this link to Google Groups.)

I've come to similar conclusions myself too. After torturing a couple of lateral MOSFET:s at 100 degrees case temperature and somewhat more than 150% of rated room temperature dissipation for a couple of minutes I gave up. They still worked.

As for second breakdown there seems to be two different effects that get grouped into this. One is that as collector voltage increases the current will crowd towards the emitter finger edges, effectively increasing the thermal resistance. (OnSemi AN) This is essentially a thermal limitation; the temperature at these hottest spots must be kept low enough. If it gets high enough the transistor will, just as a MOSFET, be unable to turn off at these spots.

The other one is the often described hotspotting. If the positive thermal feedback becomes too great somewhere hotspots will form and the transistor will be doomed. Vertical MOSFET:s may suffer from this too, not only BJT:s, like those trench parts linked to earlier in this thread. However, this behaviour does not necessarily degrade with temperature. Above a certain Ic at a certain Vce, the transistor may get destroyed, regardless of if it is cold or hot.

In the following datasheets there are some interesting graphs. Page 3 of the LM12 datasheet and page 9 of the LM3886 datasheet contain graphs showing thermal resistance VS Vce and even transient thermal plots at different collector voltages. Here it can be seen how the thermal resistance increases with collector voltage and also the the degradation towards higher voltages gets less severe the hotter the transistor gets.

Many datasheets show that the second breakdown need not be derated as much as the power dissipation rating. This fits well with the 100ms pulse thermal resistance plot for different temperatures shown in the LM3886 datasheet.

A Sanken transistor handbook I saw once said in passing that the second breakdown ratings need not be derated as much because the destruction temperature is in the range 250-350 degrees. They suggested that the second breakdown curves be derated as if they would result in a junction temperature of 250 degrees. This would mean a derating to ~65% at 100 degrees case temperature, something that fits well with many datasheets I've seen which have different derating factors for S/B and thermal limits. The OnSemi appnote is even less restrictive: 75%.

It is sometimes unclear how much margin there is in the SOA plots of BJT:s however. MOSFET:s almost always have calculated SOA plots, both DC and pulse, based on 150 or 175 degree peak temperature so in that case there are large margins as long as they are thermally stable. But for BJT:s, at least in the past, the plots could be based on destructive testing with a little margin added.

In some cases, such as for the 2SA1943 the SOA plot and transient thermal resistance can be compared. Here it seems like the thermal limits in the SOA plot, both DC and pulse, are based on a ~150 degree peak junction temperature which would be nice and conservative.

How the second breakdown limits were arrived at can't be determined this way though. If they too are based on a peak 150 degree temperature at the hottest spot they would be quite conservative. If they are based on destruction or 250 degree peak temperature I wouldn't want to push it on the other hand. It is probably safest to assume this is the case, build a circuit to measure (transient) thermal impedance by junction drop at a couple of different voltages, or test the margins by blowing up a couple of them under controlled conditions. The last option might be the most fun, but maybe a bit expensive. :D

Another way of coming to conclusions of how much is needed is to look at what's done in commercial equipment of good quality and reputation. Some manufacturers have 5 year warranties for example, and in that case it's much cheaper to add a couple of output devices than risk warranty returns. Still, these products often have less output device SOA than what many frequents of this site consider an absolute minimum for reliability. Still these amplifiers hold up in club applications where they are used night in and night out.

Designing for an absolute worst case average junction temperature of 150 degrees (at impedance dips) often seem to give similar results in the needed amount of output devices as what commercial manufacturers with good reputation use. Actually, that ends up with, for 150W transistors on a 60 deg C heatsink, about 150W of output per pair at the lowest rated impedance (dips to half of that are allowed) if rail loss is negligible. That's about 75W per pair at 8 ohms if 4 ohm speaker loads are to be allowed. :) Hmm... I belive I saw that number just some minutes ago somewhere else... :p

If I did the calculations correctly, the absolute worst case average output stage dissipation (if idle is reasonably low) for any input signal can be determined as:
Total output stage dissipation = Vcc ^ 2 / (4 * Re)
where Vcc = rail voltage of a single rail (e.g. 40V for +-40V rails) and Re = the impedance at the lowest impedance dip. This is the same dissipation as for a half rail voltage square wave into a resistive load with the same resistance as the impedance at the dip. This dissipation is a little bit, but not much, higher than the worst case sinewave into a resistive load of Re. It is about 10-20% (I don't remember the exact number) lower for a worst-case sine wave than for the square wave. Music will be even lower than the sine, so there is even more conservativeness than those tens of percent built into this.

Oops, this became a long post. I hope it will be useful :)
Click to expand...

Hi Megajocke,

Thanks for this great post! There's a lot of good stuff in here, so don't apologize for the length. I always enjoy seeing this kind of in-depth down-and-dirty technical stuff on this forum.

Your post made it cross my mind that it would be interesting to look at a couple of the Bryston amps as examples of how many output pairs are used. They have a good reputation for reliability.

Cheers,
Bob
 
tiefbassuebertr

tiefbassuebertr

Member
Joined 2005
tiefbassuebertragung.de
[B said:
tiefbassuebertr[/B]: MJ21193/MJ21194 (MJ 21193 21194) - no data for SOA during dc conditios
There is only the condition for 1 sec, but not for DC in the SOA diagram (please refer fig. 13, page 4) about
http://www.onsemi.com/pub_link/Collateral/MJ21193-D.PDF
In the SOA-diagram of the datasheet MJ15015/15016/2N3055A
there are different conditions (for 30us,100us, 1ms and 100ms) so as DC condition (please refer fig. 12 and fig. 13, page 5
http://www.onsemi.com/pub_link/Collateral/2N3055A-D.PDF
Are there certainly calculation steps to determine self the dc conditions for MJ21193/MJ21194 SOA?
Thank you very much for your advices
wg_ski: The DC curve can always be generated for any device. In this case 250W at T=25C derating to zero at 200C. They usually put the 25C DC curve on the SOA graph for reference (MJ15015, etc.) but they don't have to because they've already supplied that information.
tiefbassuebertr:1914176]Here I have both SOA Diagrams MJ21193 and MJ15025 from
ON Semiconductor MJ15025: Power 16A 250V Discrete PNP
for directly compare, but I understand only this one from MJ15025
AndrewT: Hi T, For Tc=25degC,
I see a current limit that matches the specification.
I see a power limit that matches the specification.
I see a plot of second breakdown that is not described in the specification.
Those SOA graphs seem to give all the information I need without any ambiguity. What is the problem?
tiefbassuebertr: The problem is, that I have only the SOA for 1 sec and I want to understand, how I can generrated other times and dc conditions for directly compare to the other mentioned power devices.
Additional it would be of interest for me, why the SOA for MJ21194 is only for one second and this one for the older MJ types a wide range of different times for SOA (open the weblink in the beginning of this thread)
Additionaly I don't understand the follow:
=======================================
"The DC curve can always be generated for any device. In this case 250W at T=25C derating to zero at 200C. They usually put the 25C DC curve on the SOA graph for reference (MJ15015, etc.) but they don't have to because they've already supplied that information."
=======================================
what means this exactly concerning the comparable of SOA with different conditions for each types?
AndrewT: you cannot infer the shorter single shot durations from the DC or 1second ratings.
You cannot infer the second breakdown curve from the DC ratings.
You can easily find the low voltage and current ratings from the DC specification.
A 250W 17A device can pass <17A at <14.7V when Tc=25degC.
At >=14.7V the device can pass [250/Vce] amperes continuously when Vce is <= DC second breakdown limit.
(from this thread:
http://www.diyaudio.com/forums/soli...93-21194-no-data-soa-during-dc-conditios.html)
AndrewT say's, that isn't a problem; but after a look to the discussion here one must see different - so I think.
AndrewT;2048080]what are you referring to.
I cannot see what the questions you asked in that thread have anything to do with the discussion in this thread.
Please explain why you think these threads are related.
Click to expand...

Actually to the whole thread here. I don't understand, why SOA diagrams from the same manufacturer are not direct comparable. What is the reason, why Onsemi show us only the SOA graph for 1 sec. by MJ21193/21194 ??
What must I do, if I see either the graph only for 100 milli-sec or only for 1 sec or only for the DC conditions? In some cases I see all graphs, unfortunately not in all cases (MJ21193 shows only the graph for 1 sec in their SOA diagram).
 
Last edited:
B

Bob Cordell

Member
Joined 2006
cordellaudio.com
tiefbassuebertr said:
Actually to the whole thread here. I don't understand, why SOA diagrams from the same manufacturer are not direct comparable. What is the reason, why Onsemi show us only the SOA graph for 1 sec. by MJ21193/21194 ??
What must I do, if I see either the graph only for 100 milli-sec or only for 1 sec or only for the DC conditions? In some cases I see all graphs, unfortunately not in all cases (MJ21193 shows only the graph for 1 sec in their SOA diagram).
Click to expand...

This is a tough question to answer. I always prefer datasheets that show SOA for DC and a variety of pulse times. For audio purposes, I would assume that 1 second SOA is about the same as DC SOA. Audio excursions that stress SOA will normally be fairly brief. Bear in mind that the period of a 20 Hz sinusoid is 50 ms, and that even a quarter cycle is only 12.5 ms. In the absence of short-circuit issues, a 10 ms SOA is probably the most relevant to audio when driving a loudspeaker.

Another thing I like to see on datasheets, but which is often only presented on MOSFET datasheets, is transient thermal impedance. This can sometimes give a hint as to the relationship of DC SOA to pulsed SOA.

Cheers,
Bob
 
tiefbassuebertr

tiefbassuebertr

Member
Joined 2005
tiefbassuebertragung.de
Bob Cordell said:
This is a tough question to answer. I always prefer datasheets that show SOA for DC and a variety of pulse times. For audio purposes, I would assume that 1 second SOA is about the same as DC SOA. Audio excursions that stress SOA will normally be fairly brief. Bear in mind that the period of a 20 Hz sinusoid is 50 ms, and that even a quarter cycle is only 12.5 ms. In the absence of short-circuit issues, a 10 ms SOA is probably the most relevant to audio when driving a loudspeaker.

Another thing I like to see on datasheets, but which is often only presented on MOSFET datasheets, is transient thermal impedance. This can sometimes give a hint as to the relationship of DC SOA to pulsed SOA.

Cheers,
Bob
Click to expand...

Thank you for this statement. In opposite to AndrewT's comments this I can clearly understand (perhaps because I don't speak and understand Scottish English).
Today's engineers no longer take everything so well - so the earlier data sheets of semiconductors in general were considerably more extensive than the currently datasheets
go to
MJ4502 Datasheet pdf - Verbleiter Energie Transistor-Universeller Zweck - Central Semiconductor
and you will see the differences (e. g. MJ4502/802 from ST vs. ONsemi/Motorola)
by the way - the third datasheet by this URL is actually an old Motorola application note from the aera where launched the MJ4502/802 production (AN485 - High Power Audio Amp with circuit protection - an other approach for short protect as usual) - actually found only by chance and interesting to read - so I think.
 
G

GK

Disabled Account
Joined 2006
B

Bob Cordell

Member
Joined 2006
cordellaudio.com
tiefbassuebertr

tiefbassuebertr

Member
Joined 2005
tiefbassuebertragung.de
how to assess the diagrams of the company "MOSPEC"?
By their website are to find a wide range of "vintage" BjT devices e. g. Motorola's MJ15015/15016
http://www.mospec.com.tw/pdf/power/2N3055A.pdf
or Sanken's 2SC3519(A)/2SA1386(A)
http://www.mospec.com.tw/pdf/power/2SC3519.pdf
http://www.mospec.com.tw/pdf/power/2SA1386.pdf
the complete overview are there:
:: Welcome :: MOSPEC SEMICONDUCTOR::pOWER TRANSISTORS
Because I have never heard about their quality standart at whole, I have start this thread:
http://www.diyaudio.com/forums/soli...-amplifiers-mospec-quality-standart-like.html
 
A

AndrewT

R.I.P.
Joined 2004
Pingrs said:
Try section2.4 Output Stage
ETI-466 Amplifier Technical Review

Brian.
Click to expand...
ref16 & 18 refer to David Eather's two part paper on developing a hand calculation method of determining de-rated SOAR and reactive load lines and comparing the two. Try to find the original papers.

Bensen's spreadsheet is on this Forum for doing the FET version of Eather's Method.
I have modified versions that handle BJToutputs or FET outputs or BJT drivers.
Email if you want copies.
 
tiefbassuebertr

tiefbassuebertr

Member
Joined 2005
tiefbassuebertragung.de
wahab said:
tiefbassuebertr said:
how to assess the diagrams of the company "MOSPEC"?
By their website are to find a wide range of "vintage" BjT devices e. g. Motorola's MJ15015/15016


their datasheets seems also to be vintages, as they
seems to be paste/copy of the old manufactureres datasheets...
Click to expand...

....and also the "know how" of the old manufacturers?
Click to expand...
 
D

djoffe

Member
Joined 2008
akitika.com
Trench FETS in the Linear Region

One of the posters in the forum gave the link to:
http://powerelectronics.com/mag/Ely January 2004 PET.pdf
"Are TrenchFets too fragile for Linear Applications".

Does anyone out there have experience with these kinds of failures, where the SOA says yes, but the TrenchFET devices say NO to linear applications?

Does anyone have a list of preferred mosfet devices, e.g. those that are rugged in linear applications?
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Top Bottom