Using FETs in parallel

Keep them very close together on the same heatsink and use a proper set of current sharing resistors (same thing as emitter resistor with linear parts) and you'll be fine.

Steve Roberts

FETs usually have a wide range of transconductance and gate ON voltage. Looking at my old favorite, the IRF820, they show a 100% range of typical onvoltage, 2 to 4 volts. They don't even give a maximum transconductance, just a minimum. That should raise your hackles.

I'd use enough resistance in each source lead to give you at least a 4 volt drop across each resistor. That will help ballast the ON voltage differences and Gm.

I'm sorry, Roger, but you are wrong. While it's true that MOSFETs used in their classic switching mode with high gate voltages tend to thermally self-balance, it's NOT true for the same MOSFETs used in the linear mode at lower currents, with the much necessarily-lower gate voltages. Here the tempco in fact works against you, pushing the MOSFET towards hogging more current as it heats up. This fact is apparent when looking at the Transfer Characteristics plots in common power MOSFET datasheets. You'll see for gate voltages below some high value, like 5.5V, the drain current increases with temperature. For example, I'm looking at a 75n06, which conducts 5A with Vgs = 4V and Tj = 25C. The current increases by 140% to 12A for Tj = 100C, a rather dramatic change in the wrong direction.

Any power fet used in linear mode needs a gate resistor!

John

As saturated switches, yes. As linear amps, they share current very badly, especially so at lower currents. In a relatively high drain-voltage situation, one can easily parallel, say, four "identical" power fets, run up the gate voltages, and have one fet exceed its power rating and fry while the other three do pretty much nothing. And as Win says, the tc's can make it worse, not better.

I like to use an opamp per fet to servo their drain currents, with a low-ohms current-sensing resistor in each source. That avoids using

John

In Mospower Applications Handbook by Siliconix ISBN 0-930519-00-0

Section 5.3.3 pg 5-48

An Analysis and Experimental Verifications of Parasitic Oscillations in Paralleled Power MOSFETs

To be published in IEEE Transactions in Electron Devices , Vol. ED-31, No. 7 July 1984 (seems like yesterday)

Paralleled mosfets can behave as differential pairs. I have seen the smoke and shed the tears.

Regards,

Boris Mohar

Got Knock? - see: Viatrack Printed Circuit Designs (among other things)

void _-void-_ in the obvious place

That's correct, but one dare not be satisfied if they are simply "audio fets" - the datasheet must show a correctly-moved TC curve.

It is not that the parts are well matched, but FETs tend to self balance to some extent. Basically, if a device runs hotter than the rest, the gain changes in such a way as to make it heat less. This is the opposite of the behaviour with bipolar transistors. This effect is what is relied on inside power FETs, which are typically many hundreds of actual FET junctions all wired in parallel. Adding circuitry to 'help', is certainly well worth considering, but is ofte unneccesary.

Best Wishes

And even if the tempcos are compensating, gate threshold voltages are all over the place.

John

All the power mosfets I use, which are modern vertical cellular structures.

No? Is something - gate oxide thickness, doping, whatever - better controlled in a lateral part? The Hitachi 2SK105x parts are pretty wimpy for a TO3P package, and Vgs at 100 mA is spec'd at 0.15 to 1.45. And the transfer curve tc is in the hogging direction below 0.6 amps.

John

True, especially if FETs come from different production runs. One could be almost fully on while the others do little. That one may blow and then there may not not be enough margin left for the others. Consequently they could also fail.

Regards, Joerg

Aargh!. You are correct. I did not actually fully read the post, and _assumed_ that anyone paralleling FET's, was probably using them for switching!. I must admit, I'd still guess that the 'reference design', was probably a switching circuit of some sort...

Best Wishes

Fets have a very nice property ,when they heat,they draw less. So you get automatic distribution of current between them. Oldtime transistors do the reverse, and to stop that ,each needs a small emittor resistor,loosing a bit of efficiency.

The old Hitachi lateral types designed for audio actually match very well. They may have stopped making them though. Certainly they dropped the TO-3 can versions ages back.

I don't know how well the 'substitutes' from Semelab-Magnatec and Exicon match.

Phil probably knows the score.

Graham

That's an old 4-page 1996 datasheet with poor graphics, compared to the newer 8-page 1999 version.

In article , John Larkin wrote: [...]

well almost. If the there are modest source and drain resistances near the part the gate resistance becomes optional.

Use bipolar transistors like the TIP35 with emitter resistors. They have several advantages:

(1) TIP35s are so slow they almost never oscillate.

(2) Vbe(on) is a lot better matched than Vgs(th).

(3) They are cheap.

(4) They are hard to break.

And we don't see any link to your "reference design"- you little piss-ant pseudo-intellectual. There are any number of ways to balance currents and just because *you* don't see it doesn't mean it's not there.

........... Phil