2N3055 failure (power supply)

Getting a little testy are we? Of all the things that have been suggested so far, the only thing that has any merit is the over-voltage protection. Everything else has been hogwash. After all the time and expense, the OP would have very little (in the way of reliability) to show for it. This would be very easy for the OP to prove to himself. If he were to change the one power transistor that went bad, put the power supply under full load and measure the voltages across each of the ballast resistors (as you call it), so he could see what each of the power transistor current draw is. Then go ahead and change the rest of the power transistors and do the same test again. Look at the data sheet for the 2N3055, and see how close he is to the specks. I bet he will have plenty of safety factor. If that is so, then what has he gained? I bet, not much.

Brian Ellis

This circuit might work well with logic level power MOSFETs, something like STP20NF06, which is only $0.60, compared to about $1.30 for 2N3055, if you can use the TO-220 package instead of TO-3. MOSFETs work very well for current sharing, as their resistance increases by a factor of about 2 from

If you use bipolars, it is very important to keep all the parallel transistors at the same temperature. It may even be helpful to arrange the emitter resistors so they will heat up a different area of the heat sink to minimize thermal runaway. Thermistors can also help by reducing base current drive to the hottest devices.

Vbe should be about 0.5 to 0.7 for any similar transistor, so the 0.1 ohm resistor should provide current equalization of +/- 1 amp from the nominal

Good luck,

Paul

Guessing is clearly your greatest asset. I bow in admiration.

No it isn't!! Seems you cant read a data sheet either.

What you gonna set it at? 10degC?

Yours too it would seem. You seem to be obsessed with assholes, are you a little gay?

The UK is just crawling with the vile pukes.

Aussie just hate them to death.

Yanks completely despise them.

Other poms just quietly loathe them.

YOU can use you Killfile to eliminate them anytime.

...... Phil

Actually, no. RdsON isn't relevant here, because they're not saturated, and if they're on the same heatsink most local thermoregulation would cancel out anyhow. Even "identical" fets have to be hand-matched (and expect to throw some away) to get safe levels of linear-mode sharing. Huge source resistors help the sharing at high currents but it's hard to find a value that works over a decent load range.

But it's easy to use paralleled mosfets: just use an opamp per fet as closed-loop gate drivers, with feedback from small source resistors. That will linearize and equalize them to microvolts precision.

Switching (saturated) mosfets can share a load reasonably well. [1]

John

I tried a simulation (LTSpice) using three very different MOSFETs, and current sharing at high levels was reasonably good. At low levels, one or more were essentially turned off, but that's not a problem. I have no way to simulate device variations or temperature effects, however. I used a 20 volt raw source and 15 volts on the gates. The MOSFETs are IRF7811, IRF7468, and IRF9410. Here's the results:

Rload Vload I(R1) I(R2) I(R3)

Paul

I assume you hooked them hard in parallel, with no source resistors.

My experience in building NMR power amps is that it's ugly to parallel "identical" power fets even with the largest feasible source resistors. In my situation, a class AB push-pull amp, quiescent current ran about 10% of peak output current, and with a little bad luck, one of the fets (out of 4 on each side) would wind up furnishing most of Iq and getting a lot hotter than the others. We wound up matching parts, a real nuisance. After that experience on our first amp, we went to closed-loop control of each fet, which has a number of side benefits.

If you don't care which fet does most of the work at low loads, and if you can afford to drop a volt (or preferebly 2) in each source resistor at full load, it can be made to work. But at 1 volt drop, there could well be some intermediate load that's embarassing.

I just wouldn't trust those Spice numbers. The last couple of lines have three different parts whose transfer curves spread less than 20%. I don't think that's realistic.

John

I used the same 0.1 ohm source resistors as in the OP's schematic. The last two lines are at and considerably above the rating of the supply. At 8 amps each, the source resistors drop 0.8 volts, and there is a large difference in transconductance over that range. At 90 amps out, the source resistors are doing most of the work, dropping 3 volts each. Some intermediate values:

The current sharing gets crappy at low currents, but at that point it doesn't matter very much. What I don't know is how the transconductance varies with temperature. I don't do much with linear power anymore, now that PWM is so easy to implement, but linear circuits are still useful. Any real data, experimental or theoretical, that supports or refutes these results, would be appreciated.

Just for fun I made a similar circuit with bipolar transistors: ZTX1048A,

Rload Vload I(R1) I(R2) I(R3)

That's pretty good current sharing for such a wide range of transistor types. I would expect no more than 1 ampere difference between any two devices at the nominal 4 amperes each for five, with 0.1 ohm resistors.

It is quite possible that an amplifier running at audio frequency may require much better device matching. There are probably considerable variations in gain for transistors used for that purpose.

Paul

"Paul E. Schoen = CRIMINAL F****IT "

All 3 mosfets are in SO-8 pack & have power ratings of 2.5 watts max !!

Shows what happens when you put a dumb simulator in the hands of an utter imbecile.

....... Phil

"Paul E. Schoen = CRIMINAL F****IT "

This is even STUPIDER !!

A 115 watt TO3, a 1 watt TO92 and 3 watt SOT223 ??

Shows what happens when you put a dumb simulator in the hands of a f****ng imbecile.

ROTFLMAO !!

....... Phil

Since *someone* pointed out that the previous MOSFETs were small packages, I ran another simulation with honking big devices. STB120NF10 (D2PAK), IRFP2907 (TO-247), and IRF1405 (TO-220).

Rload Vload I(R1) I(R2) I(R3)

Even with such major device differences, the current sharing is not too bad at high levels where it is critical.

Paul

"Paul E. Schoen F****ng MORON from HELL "

Your totally ASININE simulator is **** NOT **** modelling the production differences that are **specified ** to EXIST with real MOSFET devices.

Read the flaming spec sheet you BLOODY IMBECILE !!!!!!

The *gate threshold* voltage is speced to range from * 2 to 4 volts

• - for all 3 device types.

With transconductance values from 70 to 130 amps per volt - that means that parallel connected, unmatched devices operating in linear mode have NO CHANCE of sharing load current.

A 1 volt gate threshold voltage differential equates to circa 100 amp drain current error !!!!

See specs on page 4 and figure 4 on page 6.

Draw two extra curves, spaced one volt each side of the curve shown - that is the region where all real devices fall.

Paul Schoen = a COMPLETE F****NG ASS !!!!

F*ck Off Imbecile !!!!!!

....... Phil

For more technical analysis of MOSFETs in parallel, read this from IR:

From what I gather after a quick scan, worst case current imbalances of about 40% may be expected for devices without additional circuitry such as source resistors.

Paul

If you're still assuming 0.1 ohms in each source, the high-current point results in about 1.5 volts drop in each source resistor, or about 22 watts dissipated per resistor. Sure, that works if you don't mind all that heat. But the low-current sharing is correspondingly terrible.

It's interesting that the TO-247 part (which will have the fattest leads and source wirebonds by far) comes out worst at all currents. I certainly wouldn't do a production design based on Spice simulations like this.

John

I added the calculated Vgs for these devices with the 0.1 ohm source resistors:

STB120NF10 IRFP2907 IRF1405 D2PAK TO-247 TO-220 Rload Vload I(R1)/Vg1 I(R2)/Vg2 I(R3)/Vg3

These voltages are generally below the typical cutoff voltages on the device data sheets, so the results might not be accurate at the low current levels.

These are high current devices, so I used lower values to get more current:

With 0.01 ohm source resistors:

With 1 uohm source resistors (essentially zero):

So, even without negative feedback current sensing, there is adequate sharing at levels where it counts, and the largest device does finally take its proper share of the load. Also, the positive temperature coefficient of device resistance even above saturation will provide even better sharing as devices heat up.

The main problem with parallel devices appears to be for higher frequency and switching parameters, and not steady state DC conditions. There could be problems with a non-resistive dynamic load where currents change quickly. That could also be simulated, but I don't think it is necessary unless there is an actual application where this would be required.

Paul

Sorry, I just don't believe these numbers. Spice doesn't include real-life device-device variations. Transfer curves can be all over the place, even for parts out of the same tube.

I have used multiple mosfets in parallel in switching apps. That works fine.

John

I guess someone will need to do actual experimentation to see what happens in real life. I have not found very much information on device variations, particularly for gain (transconductance) and gate threshold, which are probably critical for this application. It would probably be a valuable exercise to do this. I was very surprised at the current sharing without source resistors. I would think that the "knee" of the Vgs to I curve is a useful point. I would think it would vary no more that about 0.2 volts from the nominal 4.5 volts, especially among similar devices. The source resistors should take care of that much variation without too much power, and for linear applications one might as well dissipate some of the power in resistors rather than the MOSFET. Switching applications are a much different animal, where efficiency is paramount, and transients are critical.

An interesting discussion. There is some good information at:

Paul

"Paul E. Schoen F****ng MORON from HELL "

The author only considers switching operation - not LINEAR operation.

F*ck Off - YOU BLOODY IMBECILE !!

........ Phil

In article , Paul E. Schoen wrote: [snip]

I had a serious rant here a few years ago about the way in which manufacturers of VMOS devices perpetuated the myth that "MOSFETs allow easy paralleling".

In fact nothing could be further from the truth.

For the VMOS, at fixed Vgs, dId/dT is positive. This means that if any one MOSFET in a bank runs slightly warmer it will take more current, getting hotter, so taking more current..... and so on. Thermal runaway.

Look at Fig.3 of the IRF1405. At Vgs= 4.5V, Id is 4.2A at Tj=25C, rising to 28A at 175C. That's roughly +0.15A/C. For the IRF1405 dI/dT does not approach zero (and go negative) until Id is about 180A.

Hitachi realised this problem 20-odd years ago and they produced power MOSFETs where dI/dT went negative at only about 100mA. ISTR they were called lateral or long channel MOSFETs, an entirely different process to the VMOS and it's derivatives. The negative dI/dT did allow easy paralleling and those Hitachi MOSFETs were widely used in successful high power audio amps.