micro-chip

I've been pushing (ie, blowing up) come Cree SiC power fets, the 1200 volt 280 mohm part in TO247. It sure looks like thermal runaway.

Just for fun I ground one down.

formatting link

No wonder it blows up. The actual chip is tiny, so all the heat is in one spot. Why bother to put it in that giant TO247 package?

A healthy mosfet die would use most of the available area.

formatting link

I've ordered some ST SiC parts. They have similar capacitance and twice the Rds-on, but Ron barely changes with temperature. The Cree Rds-on takes off radically with temperature, so the ST wins above

150C.

I'll gring down an ST part too.

We are considering getting an x-ray machine. It will inspect BGA solder joints and will count parts, loose in trays or on reels. Engineering could use it to snoop inside parts like this, maybe.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin
Loading thread data ...

No surprise, SiC is very conductive.

These numbers look bad but note they're per cm, not per m as usual. Copper is ~4 W/cm/K.

formatting link

Interesting that it peaks around room temperature, then drops so sharply at higher temperatures. Not that you'll be able to operate so high, with a plastic packaged device.

What is it about plastic packaging that dies so suddenly at 150C+, anyway? Local expansion ripping the bondwires out? You can solder the buggers at nearly 300C, and they don't always fail. (Not that they necessarily use the same compound in TO-247s and SOT-23s.) Soldering likely has less gradient than operation, particularly under high heat fluxes --> high thermal gradient between die surface and backing plate. Maybe that's it.

formatting link
Silicon also peaks in conductivity, but lower (LN2 temps), and is only ~1.3 W/cm/K at RT. So you need bigger dies for the power alone, but more importantly, for the Rds(on) at Vds(max). Consequence: fuckoff massive Qg.

SiC just performs that much better, but also has yield problems (persistent defects -- screw defects IIRC, from the bulk substrate, even after epitaxy and annealing), making large area devices impractical.

I wonder if SiC transistors are any good as RF amps. Not going to threaten GaN of course, but Superjunction Si is already quite good, and damn cheap compared to the alternatives.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Design 
 Click to see the full signature
Reply to
Tim Williams

That's interesting. SiC thermal conductivity seems to peak at around

100K, and then drops off. Combine that with an upward curve of Rds-on with temperature, and things get bad.

SiC tends to have high internal series gate resistance, not so good for RF, but there are 1200 and 1700 volt parts, good for power switching.

The Cree part that I'm (ab)using, C2M0280120D, ia rated for 150C max Tj and 62 watts. The ST parts with similar electrical specs (but higher Rds-on at room temp) SCT10N120, is spec'd for 200C and 150 watts. I'm thinking that the ST will be a bigger physical chip, easier to cool.

I think people have worked out the defect issues in SiC. But they are dead serious about the abs max gate voltages.

I have discovered that this plastic container

formatting link

is the ideal size to fill with water, freeze, and drop into my water-cooling bucket. I need to buy lots more of them.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

I regularly run plastic power transistors at 200'C, eg TIP3055 and several newer ones. I don't have much choice, and I don't suppose they'll last forever, but that doesn't matter. I can't recall any failures, but the equipment doesn't generally have a long service life.

Cheers

--
Clive
Reply to
Clive Arthur

John, You know, ...I wouldn't mind seeing the inside of an IRFP4668 FET. ...next time you find some spare time on your hands. :)

The datasheet claims this part will not fail even in avalanche, provided you don't overheat it.

The RdsON = 9.7 mOhm @ 81A, 10V

Of course, this is only a 200V Vdss part, so nothing like your's.

Hope you figure it out.

Reply to
mpm

Just put one sideways into a good vise and squeeze it. The epoxy will shatter and you can see the chip.

The Cree SiC parts are avalanche rated, so don't die immediately if the voltage spikes. Some GaN parts do.

There are 1200 volt mosfets, but the capacitances are huge compared to SiC.

It doesn't loook like better heat sinking will much help the Cree part; the chip is tiny, 0.004 sq inches, and the copper slab inside the package is big, so most of the thermal resistance is right around the chip itself. The ST parts may be a bit better.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

I've tested a number of transistors that popped right at the RthJC in the datasheet, plus a modest guard band (~20%). So I wonder if it's stress. Which implies Tj(max) actually rises at low power, which I haven't tested.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Design 
 Click to see the full signature
Reply to
Tim Williams

so pop the covers off & repot with copper filled epoxy :) Or not.

NT

Reply to
tabbypurr

The Cree has a bunch of tiny wire bonds that loop from the source and gate pins to the silly little chip. I could see them as I was sanding the epoxy down.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

00

ded you don't overheat it.

I was joking of course, exposing the die to the world then repotting with a n epoxy of unknown suitability does not produce something I'd put in testge ar, nor is it in any way labour efficient. If it were the 1960s someone mig ht try repotting them, maybe adding a secondary heatsink tab to get more po wer through it. Now you got what you got. I'm sure your budget stretches to another one.

NT

Reply to
tabbypurr

Yes, I've been using the same part, here's my image of the die.

formatting link

Luckily I haven't blown any up yet, even though I'm switching them at a 10MHz repetition rate. I'd like to have the part in a TO-220 package, but I suppose the thermal resistance is a little better with the larger package.

I think you'll find they all have rather small dies.

--
 Thanks, 
    - Win
Reply to
Winfield Hill

That pic would have saved me a bunch of sanding! That filled epoxy is a pain to grind down.

I blew up mine with a too-agressive gate driver, swinging to abs max in both directions in a couple ns, probably with some overshoot. Backing down a bit makes it reliable again, and doesn't affect dissipation measurably.

I'm using a TO220 size AlN insulator to keep the capacitance down. The insulated mounting hole on the TO247 is nice mechanically. It increases the creepage distance too.

The ST has a much lower, near zero, Rds-on tempco; I think I'm seeing thermal runaway on the Cree when I push my pulser.

The ST is rated for 150 watts and 200C Tj, too.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

You might try the SCT10N120. It has about half the max theta-jc of the Cree, so is probably a bigger die. Initial Rds-on is higher, but it stays low as temperature rises. Gate resistance is lower too.

It Spices better than the Cree in my circuit at 100C Tj. I'll actually try one on Tuesday.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

What I'm hearing is to expect even smaller dies with lower RdsOn in future SiC generations.

Reply to
boB

It might help a little to epoxy a copper disk to the top to spread out the hot spot. Probably not much.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

This maybe off topic but I saw a pdf where a manufacturer said that their high voltage SiC fets can take only 50V when used in linear circuit. (I think it was SiC fet.)

Reply to
LM

The Cree and ST data sheets do have SOAR graphs, and they look a lot like mosfets. Like a lot of mosfets, they weren't optimized for linear operation.

Both vendors' Spice models have 5 pins, two of which are thermal. They violate all available physical conservation laws and do bizarre things to simulation. You can add external parts to include thermal dynamics in a sim, but I haven't done that so far.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

nd

h an epoxy of unknown suitability does not produce something I'd put in tes tgear, nor is it in any way labour efficient. If it were the 1960s someone might try repotting them, maybe adding a secondary heatsink tab to get more power through it. Now you got what you got. I'm sure your budget stretches to another one.

Metal tab heat path versus through a plastic case - you wouldn't win much.

It would cost little to sandwich the die between 2 metal tabs, but using a

2nd heatsink is more money. Make that 2nd tab a piece of stamped metal with a minimum sized well for the semiconductor & you wouldn't need a 2nd heats ink.

NT

Reply to
tabbypurr

John Larkin wrote: ...

Most Power MOSFET usable for linear operation have gone end-of-life.

--
Uwe Bonnes                bon@elektron.ikp.physik.tu-darmstadt.de 

Institut fuer Kernphysik  Schlossgartenstrasse 9  64289 Darmstadt 
 Click to see the full signature
Reply to
Uwe Bonnes

We tested a bunch of fets to destruction for use in a big linear amp. Envision shrapnel. Some of the switchmode parts were pretty good.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

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.