So a fet heater circuit. Vin = 63 V (dang moving specs, it started at 50V) Then a series of three ~20W resistors. TO-220 (At the moment these are 15 ohms, 45 total though writing I think my solution is raise this to 22 ohms ea.) Then a FET and feedback/ control loop. The max. current at 1A.
This worked OK on the lab bench, but my boss blew up the fet running it in vacuum. The fet was running close to the edge on the bench. (obviously)
Anyway I started looking at TO-247 fets on DK. This got me again thinking about fets as heaters. And it seems to me I want to pick a fet with an on resistance that is relatively large 0.1 to 1 ohm. That way I need a larger change in G-S voltage to change the current. On the other hand I built this with a low Ron fet and it works just fine. What's wrong with my thinking?
I think there are probably two issues here: worst-case heat dissipation by the FET, and possible hot-spotting.
As I recall, the worst-case dissipation for the FET will be when it's dropping half of the total voltage across itself. You'll have 31.5 volts across the FET, and 31.5 volts across the resistors, with 700 mA of current flowing. Power dissipated in the resistors will equal power dissipated in the FET - 22 watts each.
That's a lot of heat for a TO-220 package to get rid of. The Great Font of Dubious Knowledge says that one of these packages can get rid of 50 watts or more on an "infinite heat sink", but those are a bit difficult to procure and use in practical applications. With less than infinite dissipation capability (or the presence of an insulating washer of some sort between package and heatsink) you'd be in trouble.
From your comments about vaccum vs. on-the-bench it sounds as if the heatsink simply isn't capable of getting rid of 44 watts of heat when there's no atmosphere to convect some of the heat away. ("In space, nobody can hear you frying eggs.")
At that current level, the FET will "look like" a 45-ohm resistor. It's a long way from being turned on all the way, and its Rds(on) isn't really relevant.
Using TO-247 packages might help *if* the problem is in the package-to-heatsink interface (i.e. excessive thermal resistance). However, if the problem is that your heater simply can't radiate away (or conduct away) 44 watts of heat under vacuum conditions, you'll still end up overheating.
Using several FETs in parallel (with a few ohms of degeneration between source and ground, to compensate for threshold-voltage differences) might help somewhat.
The other concern could be hot-spotting. A lot of the low-Rds-on FETs (and IGBTs) are *not* rated for linear service - they're intended for hard-on, hard-off switching. Reportedly, some of them can suffer hot spots on the die when pushed into high-dissipation linear service (a bit like BJT second breakdown) and go FOOF.
If noise isn't a problem (we should all be so lucky!) you could use a bang-bang controller of some sort (PWM). Dissipation in the FET would be very close to zero, since it would be either off (no current flow) or switched fully on (low Rds(on), little voltage drop across it) except during the switch-on/switch-off transitions. The resistors would have to handle all of the heat stress.
Right, I guess my picture, is that I'm changing the resistance of the fet with Vgs. (Ids vs Vgs plots) like from an ohm to 100's of ohms. For low Ron fets (well I've only looked a few) that change is over a smaller gate voltage. (I'm going to have to go read AoE on fets again.)
... then the FET junction could be very hot indeed. You'd better look at the data sheet, use the "junction to case" and "case to heatsink" thermal impedances to calculate the junction temperature at your worst-case power dissipation, and see if the part's going to survive it.
Let's see... STP100N8F6 (a 100-amp, 80-volt part) has an absolute-maximum junction temperature of 175C, and a junction-to-case thermal resistance of .85C/W. At 22 watts dissipation, that's almost
19C of delta, so the junction would be at around 139C. Nonconductive thermal pads for TO-220 seems to start at around .33C/W, so that's another 7C or so, so you're up to around 145C.
That's not a whole lot of safety margin. I'd expect that the lifetime of the FET might be seriously compromised if you run it that hot for very much time at all.
Possibly a silicon-carbide FET would be a better choice than silicon.
You need thick metal (to minimize thermal resistance, and carry the heat away from the package), with a *lot* of radiating surface facing outwards (to radiate the heat away as IR). Black-anodized would probably be the best.
None at all (>>grin At that current level, the FET will "look like" a 45-ohm resistor.
Not necessarily. Transconductance (delta I / delta Vgs), and the final on-resistance aren't the same thing.
Yup... and that means that you need more efficient heat-sinking (along with perhaps other things).
Maybe yes, maybe no... I don't think there's necessarily a direct relationship there. It's going to depend on the internal structure of the FET. A high-Rds(on) FET might simply have a smaller die, or poorer conductivity in the metal layers or etc. This _might_ result in it ballasting itself well enough to avoid hot spots, or it might have no beneficial effect at all.
You need to actually look at the characteristics of the specific part you're interested in.
To "correctly" use a FET as a heater, it does not matter what its on resistance is, within reason. This is because pretty much all fets can be set to blow themselves up with enough volts and current. All that really matters is that the fet can dissipate the amount of power you want from it. Its like, what gives out more heat, a 1KW oil field radiator, or a 1KW fan blower.
The idea is to use a temperature sensor and have the FET in a thermal loop. The loop is used to force the current X voltage to be whatever it needs to be.
For example, this is the sort of technique that is used in OCXO (oven controlled xtal oscillator) ASICS :-)