MOSFET damage

Hello,

I've noticed a very peculiar problem with a MOSFET in one of my designs. For reference, I am using an IRL3715, and I should be well within the rated voltage, current, and switching times. I have the source connected to GND, the gate to my uC (with a large value pull-down), and the drain goes to a connector so I can sink current through a bunch of LEDs.

Now onto the weird part. When the gate is at 5V, I show 0V difference between GND and the connector, and 12V difference between 12V and the connector. This tells me that the MOSFET is switched on like it should be.

However, when the gate is grounded, I show 0V difference between GND and the connector, but a 4.5V difference between 12V and the connector. While the 0V between GND and the connector can be indicative of an open (which should be the case), where exactly is this 4.5V comming from? Changing the polarization of my multimeter reads -4.5V. Placing an LED+resistor from 12V to the connector does not light up, so the 4.5V is extremeley weak.

Additionally, my 5V rail is nowhere near this area of the PCB, and I have 3 other channels that opperate as I would expect (0V between the connector independent of which power supply I measure against).

Obviously, this MOSFET is damaged, and I will be replacing it, but I'm curious as to why this happened. All assembly took place on an anti-static mat, with me carefully grounded to it. Admitantly though, I know I touched the drain several times while not grounded, but I would have touched all the channel's drains checking for heat. And, from my understanding, it is only supopsed to be the gate that is extremely sensitive to ESD.

And, on a side note, why did component MFGs attach the heat tabs of TO220s to the drain? I know you can get full-packs, and have isolation, but why was this done in the first place? Drain is always going to be hot with respect to GND, and I can't think of any reason to have your heatsinks at a different potential than your conducting chassis...

Looking forward to an explanation, Nathan

I read in sci.electronics.design that Nathan Hunsperger wrote (in ) about 'MOSFET damage', on Sat, 29 Jan 2005:

No, everything is very probably OK. Your voltmeter is simply responding to a **very** small leakage current through the FET. If you connect a few LEDs to the connector you will find everything works. Almost certainly.

--
Regards, John Woodgate, OOO - Own Opinions Only. 
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

The die is bonded to the heatsink. The die substrate is usually the drain ( or collector for bipolars ).

As for your other question, I suggest you look at some currents as well as voltage. A voltage reading can be meaningless ( leakage currents etc ).

Graham

I read in sci.electronics.design that Terry Given wrote (in ) about 'MOSFET damage', on Sun, 30 Jan 2005:

I'm not sure what you mean by that, but silicon isn't glass. Nor is silica (silicon dioxide); almost all glass is metallic silicate and/or borate.

--
Regards, John Woodgate, OOO - Own Opinions Only. 
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

I read in sci.electronics.design that Terry Given wrote (in ) about 'MOSFET damage', on Sun, 30 Jan 2005:

This probably explains why two successive staple guns I bought failed after about 50 staples. I had a full refund, otherwise I would have changed the triac and, of course, bolted the new one to the heat sink.

I agree that spring clamps are even safer than bolts, BUT you need very strong springs to ensure good thermal contact.

--
Regards, John Woodgate, OOO - Own Opinions Only. 
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

Mechanically, it's pretty glass-like.

Hi Nathan,

Others have covered the MOSFET readings, and Graham mentioned the die attachment. But your thought process *is* correct - connecting the wobbly bit to a large lump of copper is less than desirable from an emi standpoint, and is often inconvenient electrically too. Some v*cor power supply modules use single-ended or push-pull converters with FETs to the positive, rather than negative, DC rail. Although the gate drive needs to be level shifted, it allows the FET to be directly mounted on a DCB substrate (ie heatsink) without injecting a huge amount of noise through the drain-substrate capacitance, as would occur with the conventional implementation. I gutted one once, and it was a marvel of production engineering. The Faraday shields around the cores were almost works of art.

But wait, there's more. Copper and Silicon have different CTEs, and so change size at different rates as the temperature changes, just like a bimetallic strip. This then stresses the die-tab interface, which is usually a solder joint [I assume this is true for TO220]. These stresses tend to make the solder joint deteriorate - and the greater the temperature change, the worse the degradation. Luckily the degaraded connection increases the temperature rise, thereby exacerbating the problem, eventually leading to thermal-fatigue induced runaway failure - this is a leading cause of failure among high-power IGBT modules, and I just read an interesting paper on its occurrence in TO-247 packages too.

As if thats not bad enough, the copper tab does a great job of transmitting forces to the silicon (ie glass). Over-tightening of a bolt can deform the copper tab, placing a direct mechanical strain on the die, degrading lifetime. Another interesting thing that can happen with tight bolts is the copper tab bends at the bolt interface, actually lifting the die off the heatsink. Pop-riveting applies a fairly uncontrolled force with a very rapid risetime, and can easily break the die. motorola AN1040 covers this subject in detail, is on-line and a must to read.

[anecdotal evidence] A product I "inherited" once had a 30% failure rate (from 600pcs) on the motorola/on-semi 7805 regulator - but nobody in production thought that was at all unusual - the official explanation was "motorola makes lousy regulators." Yeah right - one look at the pcb showed the device was pop-riveted on. We changed to a nut, bolt and belleville spring washer, tightened with a preset torque wrench, and the failure rate dropped to zero, over 17,000 units. We also removed the "nylok" self-locking nut from the smps transistor, because the inset nylon washer melted at about 95C - in some units it ended up as a puddle on the pcb, and the nut would spin freely. We used a washer, a belleville spring washer and a nut. Since then I have found really neat all-metal self-locking nuts - basically a spring washer is crimped into the end of the nut. Spring clips are even better, as they provide a much more controlled force over a wider area, and are a lot easier to assemble.

Cheers Terry

Which was entirely the point - thump it and it will break.

Cheers Terry