What about the current example using source resistors? These provide negative feedback, so Vgs is no longer fixed. How do we calculate the minimum value needed to prevent runaway?
Also, source resistors dissipate some heat, which reduces the amount dissipated in the MOSFET. So in linear applications such as the current topic, the source resistor can be useful.
A positive drain-current tc doesn't guarantee runaway. If the "loop gain" is less than 1, it won't run away. The math depends on the Id/t slope, heat sinking, mutual heat sinking, stuff like that. Actually, true thermal runaway in paralleled mosfets is rare, almost impossible.
In that it reduces power supply headroom and efficiency at peak output, all it can do is make heat and hurt. In a lot of situations, there is no value of source resistor that gives decent idle-current sharing but doesn't introduce huge peak-output losses. The old nonlinear resistor trick (ie, parallel the emitter resistor with a diode) often won't work for fets.
Tony was talking about a single MOSFET with fixed Vgs. His explanation made good sense.
For paralled MOSFETS, aren't you referring to the fully-saturated condition where the Rds has a positive temperature coefficient? I don't think that is the case in this thread.
How much headroom is lost? What is the minimum value required? If it's only 0.1V to 1V, it may be a very good tradeoff due to the simplicity. If it's greater, there may be some other tradeoffs to consider.
I'm looking for the equations and design procedures to evaluate these tradeoffs.
Isn't that a good thing? The power dissipated at idle should be quite low. Why not have one device carry most of the load? That should improve linearity and slew rate. Let the other devices come on line as the load current increases. If they turn on smoothly, it should add little distortion.
Again, where are the equations? What makes it work in some cases and fail in others?
This could turn into a very useful thread if we could pin down some numbers:)
The same datasheet shows threshold voltage as a function of temperature, but at a drain current of only 250 uA. I think this is a worst case scenario. So the change in threshold voltage is 3.3 to 2.7 from 25C to
100C. A 0.1 ohm source resistor will drop 0.4 volts at 4 amps (per the OT application), so it should ensure reasonable current sharing. Device temperature difference should not exceed 25C on a good heatsink, so threshold should only change about 0.2 volts, or 2 amps with the existing resistors.
Actually a lower current rated MOSFET would probably work better. The IRF2903Z (75A,30V) shows threshold change from 4.0 to 3.6 volts from 25 to
100C, with drain current of 1 amp, and 3.4 to 2.7 at 1 mA.
The FQB30N06 (30A,60V) shows a current increase from 5A to 9A at fixed gate voltage of about 4.8V from 25 to 175C, and no change at currents of 25A with 6.2V Vgs, and negative change above that. Current varies with gate voltage from 4.5A at 5.0Vgs to 11A at 5.5Vgs.
Given the 22 VDC input voltage (probably about 16 V at full load), and 20 amp 14 volt maximum output, efficiency is not really a factor. The resistors provide current limiting of about 40 amps if any one of the devices should short out, if the load is a battery. Also the resistors can run hotter, and are less expensive than the semiconductor devices. You lose only about 0.4 V of headroom. And current sharing at idle levels does not really matter.
Looking at the specs for the bipolar 2N3055, the Vbe changes sharply as a function of collector current at moderate levels. It is 0.7V at 0.5A, 0.8V at 2A, 1.0V at 5A, and 1.5V at 10A. This shows that the 0.1 ohm resistors are more than adequate to ensure safe current sharing at the rated output of 20A. Much below that it is not so critical. There does appear to be a need for current limiting or fusing to protect against severe overloads and short circuits, however.
Another severe limitation with the 2N3055 is the sharp drop of hfe at high currents. It takes about 1 ampere to drive the output to 10 amps, and 100 mA to drive it to the rated 4 amps per device. Thus you need a solid 2-15 volt 500 mA base drive circuit to get decent output regulation. For the MOSFETs, you can use a simple voltage divider or pot.
Luckily I ran the simulations for only 10 mSec. Otherwise I might have burned up my computer. I did find burned spots on my CRT where the MOSFETs were!
So I did an experiment. I connected four MOSFETs, HUF75645P, 100V, 75A,
310W, 0.014 ohm. Sources to GND. Gates tied together, through 100 ohms to a pot across 10 VDC supply. Each drain to a 100 ohm resistor and a red LED to
+10 VDC.
LEDs were all about equal brightness and were barely lit at 2.73V and fairly bright at 3.00V to 3.20V.
The spec sheet shows threshold from 2.0 to 4.0 V, but this is probably more the spread over temperature, which is 0.6 to 1.1 normalized, or 2.1 to 3.85 from -40 to +160C. My experiment shows threshold at about 2.8 to 3.2 V at
20C.
I even tried heating one device with a soldering iron. At 2.76 Vgs, the Vd changed from 8.60V to 7.90V. This does show a positive temperature coefficient of current to temperature at near threshold, which would be detrimental to current sharing, but it appears that the reverse is true at higher currents. The LED for the hot device was a bit brighter, but not too much.
I'm pretty well satisfied that MOSFETs can be paralleled successfully without extreme measures. Moderate source resistors dropping 0.2 volts or so at maximum current should be enough to provide adequate current sharing without loss of headroom or efficiency for most applications. It would be a good idea to try this same type experiment at higher levels, but I think these (or similar) MOSFETs would work fine for the OP's power supply. I bought 50 of them on eBay for $20, although shipping from Sweden added about $6.
I think it is a question of doing sums to calculate the dP(fet)/dT and making sure that that value is less than the reciprocal of the Theta(junction-case) of the heatsinking allocated to that individual FET.
I've just scribbled out a few sums, based on the assumption that the main Id tempco comes from the negative tempco of Vgs(Threshold), (values given in various publications ranging from 1mV/C to 2.8mV/C).
The sums looked plausible but some example calcs resulted in negative values for the external Rs, so it is back to the dwg board on that. :(
AFAIK the worst case for potential thermal runaway is a MOSFET with high transcoductance (gfs), running with high voltage and low current, with inadequate heatsink.
This is the typical circumstance of (audio) power amps, with paralleled big Hexfets, in class AB, running idling.
See which is an article proposing the use of Hexfets in paralleled complementary format....... but then read to the end, at the Update, to see the remarks after one was built.
Phil responds to that by calling you a "stinking, monstrously autistic, kiddie fiddling pile of dog vomit!!! " I simply think you're badly misinformed, despite our attempts to rectify that. But, you do get serious points for trying a bench experiment.
Sadly, your experiment was nearly meaningless, because the power dissipation was so low it didn't increase the junction temperature sufficiently to get interesting. There's a smidgin of interesting data when you compare the next-to-last two columns. The Vgs = 3.00V column at about (10-1.2)/100 = 8mA shows modestly-matched currents, but the next column with higher 34 to 58mA currents shows a 70% mismatch. However, these tiny currents are silly: these are 75A 310W mosfets, for Pete's sake!
Device #3 drawing 58mA had only 0.17W of power dissipation, and Intersil's datasheet tells us this will raise the FET's junction temperature only 0.17*62 = 10C, which is not enough to learn anything. I suggest you replace the LED+100 with 0.1-ohm 3-W current-sensing resistors, etc., to maintain the MOSFETs at the same Vds. You can run them at currents of say 0.2 to 2A each, for 2 to 20W dissipation, without heatsinks, natch, and measure the relative voltage drops across the 0.1 sense resistors. Then perhaps you'll learn something interesting.
The threshold voltage is usually specified at very low current, such as 250 uA. I expected the greatest mismatch to occur here, at what I would call the "knee" of the curve. The point is that a few tenths of a volt Vgs take the device from just barely conducting to a full ON condition. Certainly I would not recommend running in parallel without adequate source resistors, which would equalize the currents due to negative feedback.
OK. I changed the circuit to source follower. Devices 1, 2, and 3 are connected through 12 ohm 10 watt resistors to GND, device 4 has a 100 ohm to GND. All drains connected to a 10 VDC supply. Gates in parallel through
This seems to be a reasonable test within the lower normal limits of the power supply being discussed. Current in each device varies from 0.18 A to
0.60 A. As a reference, device 4 is running at about 1/8 the current of the others. It does have some thermal connection to the others, as I have them mounted in free air next to each other (and probably touching). I started measuring at 7.50 (labeled cold), and those values were drifting as the devices heated up, so they are not very accurate, but do show the increased current as temperature rises. Also, device 2 is probably the hottest since it is sandwiched between 1 and 3, so with good heat sinking the current sharing may be even better.
An exhaustive test would involve hours of work taking measurements or setting up a data acquisition system to plot curves at many points. Also more devices would be needed to achieve true statistical confidence. But I would be comfortable using these devices for the OP's application, and I'd expect current sharing to be within 20% or better. With the 0.1 ohm resistors, it would be easy to try.
Right now there is a tube of 50 pieces on eBay item 170062103958, for $0.01 plus $8.00 shipping from Sweden, if you'd like to give it a Schottky.
Hey guys, if you insist on insulting eachother, please lose the expletives. They only show that you don't know enough english to make meaningful comments !.
As far as 3055's are concerned...I'd bin them..they may be cheap, but they are also horrible devices..use a TIP, or better still, a HexFET
hey , at least we dont have to pay medical bills here.........broke my arm allit cost me is $00.00 dollars, better than usa , dont know what merry old england is like ?
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.