That's not unreasonable. In switching apps, and even most of linear apps, you don't see the max voltage and the max current at the same time.
DC SOA is just the thermal resistance and max junction temp stuff. It's all a matter of power. The only exceptions to a pure power curve is the max voltage (breakdown) and max current (wirebond) limits.
One of the IR datasheets has some dpak fets that the datasheet proclaims, in big type, are rated for 110 amps or something insane like that. Then there's a tiny footnote somewhere on page 19 that says they're actually "package limited" to 70 amps, meaning they would do the 110 except for the wire bonds.
I think the SOA curves on the 2907 are pretty straight forward.
The fact that it includes 1000A at 100V (100KW) is interesting only in that it indicates that real thermal characteristics are determining the performance, not the semiconductor process or inherent flaws.
It must have some interesting internal wire-bonds, to do this in TO247.
An IGBT of similar ratings is likely to have a clipped upper right-hand corner on a similar plot , a region in which, after entering, it cannot be expected turned off in a controlled manner, or even non-destructively.
The 'Best Practices' Mil Spec recommended derating on Mosfets for current, voltage and power are 0.6, 0.75, and 0.5 respectively, with no restrictions on SOA except not to exceed it or the previously mentioned current or voltage limits under transient conditions. Tj limit is recommended as 110degC.
Thanks, I was thinking that it all seemed to be the same but it's always nice to have at least 2 different curves/measurements/equations give you the same answer as a reality check. :-)
On a couple of our products, we have a microprocessor use an adc to digitize fet currents, d-s voltage drop, and heatsink temperature and then run a realtime simulation of junction temperature. We display that for fun, and base a shutdown on junction temp. That lets us push fets really hard without the danger of blowing them up. The thermal model that we use, based on a lot of destructive fet testing, models the thermal mass of the junction as a simple 50 millisecond 1st order lag, and we then assume a single thermal resistance from that mass to the heatsink temperature. We run the models at rates around 1000 to
That SOA note means the junction starts at 25C before the single pulse, etc., and finishes at 175C by the end of the pulse. In your case the MOSFET junctions are preheated from steady use and at some point the lamp is turned on, triggering the event, you theorized. In such a case you have to massively derated the SOA curve, compute a prediction using the Effective Transient Thermal Impedance curves, or make a thermal-mass thermal-conductance model and run that. I again suggest you add some parallel servo'd FETs, and get the temperature down.
Yea, you're right. :-) For my testing, I had originally placed 4 FETs on the sink (individually servo'd) and the heat sink temperature directly next to FET was about 11C lower. But, I was having one heck of a time fitting everything on the PCB so I was testing how far I could go with 2 FETs.
Seems that, to get slightly more reasonable junction temps, I can either go with four FETs at my existing 250W load max. or two FETs at a 200W load.
Still gotta' do the copper heat spreader test (with two FETs). But, I suspect that it won't be as effective as doubling the number of FETs. Interested in seeing the results though.
Time to fire up Eagle again and get to cramming in more FETs! It was a fun (and thanks to everyone here, very educational) bunch of tests though.
The last big amp we did had 32 fets in the output stage, at about $9 each. So ever time something went wrong in the prototype, we'd lose $280 in a few milliseconds. That rate calculates out to about 2.9 trillion dollars per year.
Results of the heat spreader test... On the 2.5" x 2.5" aluminum base of the heat sink, 0.5" thick, I added a 2.25" x 2.0" piece of Alloy 110 copper, 3/16" thick, buffed with 0 and 000 steel wool until a bit mirrorlike. Using Aavid's UltraStick compound, I screwed it to the heat sink in 4 places, raised the temp of the sink high enough to phase change the compound, and re-tightened the screws (they barely turned). I then attached the two FETs as I've always done (screw, fender washer, doubled belleville washers, UltraStick compound) for these tests and brought the load up to 250W.
Without the spreader the temp of the heat sink directly next to the FETs was 87C-88C.
With the copper heat spreader, the temps were 82C-84C. About 5C lower.
This is always a good thing but probably not worth the trouble and expense of the spreader for this situation. I think the well-spaced FETs, 0.5" thick base of the sink helped to keep the heat spread well enough that the copper spreader couldn't do too much more....maybe? Or, the joint between the heat sink and spreader wasn't as good as I thought it was. I spent a lot of time making those two surfaces as smooth/flat/co-planar as possible though.
I tried the same test using Chemtronics thermal grease, well mixed in the tube before spreading. I also spread out the grease very thinly on all surfaces and did a test assembly/disassembly first to make sure that all surfaces were flat and that the grease was being pressed down on everywhere....it was.
With the copper heat spreader and silicone grease, the temps were
85C-86C.
This is about 2C-3C higher than the temps I got using the same setup and Aavid's UltraStick compound, which mirrors early tests results when I first started using the stuff. On paper, its specs are better than silicone grease. Grease is a heck of a lot easier to apply and clean up though.
Aavid's UltraStick is surprising good stuff. It almost looks like a deodorant stick. Easy to apply, no mess, works as well as silicon thermal pucky. Digi-Key carries the stuff.
My manufacturing people tried it, but we found that it wasn't as good as filled silicone grease in our applications. It was hard to avoid air gaps, and it didn't squish as thin as grease.
If you put a fat glob of grease on the sink, then apply, squish, and rotate the transistor, you can completely squeeze out any air gaps. Then applied pressure (clamp or screw) will extrude the grease out, with a final gap below 100 microinches. If the gap filler doesn't flow out and make a mess at the edges of the transistor, then it's staying underneath, in place, and it's remaining as thick as when it was first applied, which could be several mils. None of these organic things conduct heat very well, filled or not, so the way to get low theta is to minimize thickness.
Of course, the heatsink has to be very flat and smooth to get the grease thickness down into the microinch range. Most as-extruded heat sinks are nowhere close to flat enough.
With proper controls in your iterations, changes to the mounting method or to the thermal impedance between the semiconductor bodies and the heatsink will not alter the temperature of the heatsink. Only the temperature of the semiconductor's mounting surface and the semiconductor itself will vary.
The thermal impedance from the heatsink mounting surface to free air is not being changed.
Measure the temperature of the semiconductor's tabs and the temperature of the heatsink. The deltaT, for the same power, will reflect improvements in the mounting method and interface.
Changes in heatsink temperature indicate a poor control of possible variables present.
I was thinking... Since I have a spreader-to-heat interface using two different materials (grease and UltraStick), the heat spreader temperature ("heat sink" temp in my earlier posts) could be lower without any having any control problems, right?
Ok, here are the test results: The notches I mentioned earlier are lousy places to measure. I used the drain lead where it exited the case instead. The spreader temperatures are measured directly next to the middle of the case (checked on both sides). The plastic case temperature is near the center of the top of the plastic case (as it lays flat) just as another data point.
With heat spreader and silicone grease between spreader/sink and FETs/spreader: Spreader = 85C and 86C Drain lead = 112C Plastic case = 76C Spreader-to-Drain-lead Delta-T = 27C
With hat spreader and UltraStick between spreader/sink and FETs/spreader: Spreader = 84C and 84C Drain lead = 105C Plastic case = 75C Spreader-to-Drain-lead Delta-T = 21C
I'm guess that the lower spreader temperature (for UltraStick) reflects the better performance of the UltraStick between the sink and the spreader? I guess that 1C-2C isn't much of a difference though and might be explained by differences in taking the measurement. Very consistent differences though.
I should have measured the heat sink temperature to better factor in the heat sink/spreader interface performance but the 7C lower drain lead temperature gives enough of an indication, I believe, that UltraStick is outperforming the silicone grease I'm using?
Or, if my controls are lousy, it's at least a tie. But, the edge has always gone to UltraStick for all my different test numbers.
oh yes. large (and rapid) dTj stresses the die attach, which can (and does) degrade the connection. this increases the thermal resistance, eventually leading to a runaway failure mechanism.
powerex have some curves detailing this - eg dip_gen_3_app_note.pdf
OK its for IGBTs, but the mechanism is mechanical - its dependent of the die functionality. I've also seen a paper discussing this with TO-247s.
yep it does.
its unlikely to be conducted into your circuit, its either magnetic or capacitive coupling. the problem will either be a big loop or a very high impedance node. 0V planes are your friend. In addition, keep impedances low, especially the gate driver.
This resulted in a 25A
you already have an extremely hot junction, small wonder it goes bang. Get a decent thermal model and see for yourself. because its mostly mechanical, just pinch an infineon model for a similar-sized FET and SPICE it.
I'd look very hard at the lamp-related problem, and solve that before doing any more thermal tests.
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.