I'm popping MOSFETS....linear derating factor involved?

I'm running at just above 27V and 4.6A per TO-247 IRFP2907. Pretty good safety margin for the current limit. :-)

John

-- remove SPAMMENOT for e-mail responses --

Hmm...I have a 1K resistor from the op-amp output to the gate and

John

-- remove SPAMMENOT for e-mail responses --

Darn good point. I can easily raise the source to 9V and test again. The gate voltage was about 3.6V at full power and the two gates were being driven to within 0.1V of each other (IIRC). Checking the Vds drop across the FETs and the sense resistors showed a pretty good match (but I've forgotten the numbers). I'll definitely be looking closer at those numbers for the next round of testing though, thanks.

I'm beginning to agree with that, at least for my situation. I got a

John

-- remove SPAMMENOT for e-mail responses --

Another question (slightly related)....

How can you properly calculate how much power these MOSFETs can handle without popping as the heat sink warms up? That is, if measurements show me that they can handle 100W continuous (operated linearly as part of an active load) once the heat sink has heated up and all temperatures are stable, what's the max. power it can handle while the heat sink is coming to temp?

When discharging batteries under high load, I'd love to be able to take into account that their under-load voltage takes a nose dive in the first minute. That is, I'd love to start the constant-current discharge at 150W per MOSFET knowing that the power level would drop to 100W (or less) before the heat sink heated up.

Can this be calculated?

Is this where the derating factor can be effectively be used? Just set a max junction temperature rise, and using the derating factor, calculate the power left available to be dissipated as the load?

John

-- remove SPAMMENOT for e-mail responses --

the MOSFET can oscillate, all by its self. There is a nice Siliconix app note (in the mospower apps manual) that does a Routh-Hurwitz stability analysis on a MOSFET, showing why it oscillates, and how to stop it.

Cheers Terry

I use decent bolts (not cheese) and apply rated torque.

Just use gruntier spring clips :)

unless you have enough compliance in your spring, those reflow-type thermal interfaces require you to re-torque the fastener. I've done that with 600V 1200A IGBTs - heat to 80C for a while, let cool then re-torque.

whats the surface finish like on your heatsink? have you checked the failed untis, to ensure an even spread of goop (a lack thereof indicates a bent heatsink)

Cheers Terry

nah, just get a decent thermal model of the fet. look at, say, the Infineon SPP04N60C3 datasheet to get an idea of what a good thermal model looks like.

or use the transient thermal impedance curves (same thing really)

Cheers Terry

For pulsed power, the immediate capacity of the sink material alone to absorb power dominates.

Take the heatsink material's specific heat, multiply by watt-seconds and you have a delta-T on the heatsink for the impulse. The rise is not actually linear, but variations due to convection and radiation give more modest deltas.

RL

RL

I looked for that manual but couldn't find it even using searches for "stability", "Routh-Hurwitz", "analysis", "MOSPOWER", "applications" and "manual". May not be available anymore?

I'll check the web though, thanks!

John

-- remove SPAMMENOT for e-mail responses --

You use a belleville washer (with a flat washer), "bottom it out" and continue to tighten until the rated torque is reached?

Hmm...maybe I need to make the investment and get a decent torque wrench for this. More toys is always a good thing! :-)

LOL, for half a second I was saying to myself "Hmm...I've never heard of a company named Gruntier selling spring clips".

Any recommendations for these guys? The ones I have are rated for

I'm thinking that one of my tests needs to be with plain ol' silicone grease, just to get the case temperature as a benchmark for the other tests.

That's what I've done for screw-mounting in my prototype (since no belleville washers were used) but you shouldn't need to do it when using springs, should you? The springs I had were "pulled up" the thickness of the entire TO-247 body and would surely push back down the couple of mils needed to squeeze out any excess compound (when it reflowed)?

It's the "ready-to-use" milled finish straight from the manufacturer which I buff with 000 steel wool and rinse thoroughly. It was starting to become mirror-like.

I did take a look at the mounting surface of the FETs and sink after they blew and so far all of them have had a good even layer of goop. I'm guessing, but it sure looked as thin or thinner than a sheet of paper.

John

-- remove SPAMMENOT for e-mail responses --

Thanks guys. More math, oh joy. :-)

I did do an experiment at 300W with an earlier set of FETs and plotted the case temp vs. time. It took over 45 seconds to reach 90C and about 2 minutes to reach 100C.

But, the FETs blew in a few seconds when doing the tests again. Lotsa of damage I guess.

John

-- remove SPAMMENOT for e-mail responses --

Siliconix MOSPOWER Applications, Ed. R.P. Severns, Siliconix, ISBN

one of the reasons I download and archive everything I see....

besides, web pages have a 6-month half-life.

Cheers Terry

no, spring washers should never be flattened out. pick a belleville that gives 50-75% compression at the rated torque of the bolt.

cant argue with that one.

antipodean slang.

pass. I never picked the nice ones we stuck in our 2500W dc-dc converters, but they were pretty stiff.

I rather like John Larkins approach of a well-anodised heatsink/insulator. sometimes less is more :)

AIUI this is especially true from split-ring type spring washers, but they tend to relax after being flattened out, so the original spring force is not restored.

thats a good sign then. I only ask because of a nightmarish 2 weeks a decade or so ago - crux of story is mech eng threw out off-tool heatsink sample, a tech fished it out of the bin to sell for scrap, another tech used it to build a prototype, prototype mysteriously failed, quite repeatably, for 2 weeks.

nowadays I always check surface finish and coplanarity - usually with a combination of eyeometry and a straight edge.

Cheers Terry

that pretty much says its an oscillation problem, assuming you didnt fiddle with any of the mounting hardware between tests - that it got to a suitable steady-state the first time says the heatsink interface must be OK.

with a 1k gate resistor, you may be getting noise up the gate (perhaps due to Cmiller). You could try dropping it to, say 10R

your connection to low-level-gnd might be causing the problem (eg picking up some H). ditto for construction technique - an intermittent open-circuit might be just the ticket to blow your FETs up.

another thing that can cause problems for this circuit is the current feedback picks up the gate charging/discharging current. if you use a complementary emitter-follower and low Rg to drive the gate, sit the pnp collector at the source, place a 100nF from pnp-c to npn-c, and connect npn-c to +5V with a 10R resistor. Now the gate charge/discharge currents do not flow thru the current sense resistor - although the npn/pnp base currents will, they are 1 beta'th the gate current. now you can really push the gate around without it buggering up your feedback.

Cheers Terry

Thanks! Founds lots of sources, several for only a couple of dollars.

John

-- remove SPAMMENOT for e-mail responses --

Ahhh...OK, that makes sense.

You read my mind. I'm pretty sure I'll be going the hard anodized route for this one (instead of using insulating pads) but I'll still need a grease of some type between the FETs and the anodized sink.

Ouch, I can just picture you guys absolutely pulling your hair out for those 2 weeks.

I just tried mounting two of the blown FETs via screw and flat washer, heating the sink up to phase-change the compound and then popping off the FETs to check how well the compound spread. It is definitely a much thinner layer of compound when screw mounted (almost non-existent) and looks to be pretty even along the length of the TO-247 body. Unless I can find clips with 100lb. of pressure, I'm sticking with screws for this application (or a screw-mounted bar pressing down on the FETs).

John

-- remove SPAMMENOT for e-mail responses --

Nope, didn't touch the setup.

But, the 300W test is wayyyyyy above what the linear derating says the FETs can deal with. I can easily see the junction temperature going too high in a few seconds, before I could see the case temperature rising too high.

I was thinking that this damaged one or more of the FETs during the first 300W test and it finally blew during the second one. That's not a probable option?

At 200W and 250W, I can get dozens of thermal cycles (25C to 125C Tj and back again in 4 minutes or so) and had the FETs going for 28 hours straight at 200W. The circuit seemed stable. But "seemed" is a tricky word sometimes.

I guess I just don't want to deal with the possibility of the thing oscillating. :-)

I guess it's time to dig deeper into this.

Thanks, I'll give it a try.

I'll verify that connections are tight and leads routed carefully and as short as possible.

Hmm, interesting idea. For this practically-DC application, something like this would help prevent possible problems even for only the one burst of gate charging current and the gate discharge when the FETs were turned off a few minutes later?

John

-- remove SPAMMENOT for e-mail responses --

oh yes. it was just after we fixed a subtle gatedrive problem, so we were pretty paranoid about blowups. Eventually the mech guy wandered past as another tech was dismantling it, picked up the heatsink and said "what the f*ck is this doing here". Half an hour later, we had the drive up and running, and it stayed that way.

the latter is a nice approach, a good compromise wrt parts count.

Cheers Terry

I ran some numbers and I'm clearly doing something wrong. :-)

Let's say I run at 125W for 60 seconds, that's 7500W/S or 7500 Joules (I think). If the specific heat of aluminum is 900 J/(kg*K) then shouldn't the temperature rise be the amount of heat divided by the specific heat?

(7500 J) / (900 J/kg*K) = 8.33 degrees-K rise.

But, the heat sink gets much hotter than an 8.33K rise (after 1 minute with a 125W load). And where the heck does the "kg" end up going in the calculation?

Or does this number assuming I have 1kg of material and if I don't, I should divide the temp rise by the amount of material?

(8.33K rise) / (0.268kg) = 31.1 degrees-K rise for the heat sink

Running the specific heat numbers (gathered up while reading about this stuff):

(heat change) / (mass * specific heat capacity) = temperature change

(7500 J) / (.268Kg) * (900 J/kg*K) = 31 degrees-K temperature rise.

This seems to confirm the earlier calculation. But, I'm sure this makes all sorts of assumptions about temperature gradients across the heat sink (or lack of them), etc. If the result is a 31K rise, could that be a 50K rise at sink next to the MOSFET and a 10K rise at the other end of the heat sink (averaging out to 31K)? Or does the temp rise number only apply to directly next to the heat source?

Call me confused, about a lot of things. :-)

John

-- remove SPAMMENOT for e-mail responses --