Old Mosfet question-new problem

manie

Senior Member
I need to pass between 10 - 150 Amps at 12V through a semi resistive/capacitive load. The 28x1 works quite well, BUT ! (always but...) My mosfets keep frying !:mad:

description:
I use from 4 to 8 IRF3205's(55V/75A/TO-220) in parallel to carry the load. I drive a PN2222A npn transistor (with PWM at 4kHz, duty between 10% and 100%), when "on", the collector grounds ( 0V ) the bases of 4 to 8 BC327's(pnp) each which in turn switches regulated 12V onto the Gates of the Mosfets. Gate power is via 47R resistors to the gates. The mosfets are well heatsinked with air cooling via a 12V 92mm fan. I clamp the mosfets down with a piece of flat bar across the face, thus pinching the fets between heatsink at back and 5mm metal at front. I use heat transfer compound on both sides. My dual channel oscilloscope shows VERY FAST "on" switching ( < 2uSec) and quite good "off" switch times (about 5 uSec on very slight slope, with curved gradient the last few mV's at about another 5 uSec). I do not use the pnp's to short the gates to 0V but to actually drive the gates positive (at +- 7.5V to +- 11V Gv). This is the best driver I have had to date.

Problem:
At 100% duty (gates fully "on", low RDS +- 8 mOhm) the mosfets run beautifully....:D: at close to ambient temp. of +- 38 deg C.
At anything less than 93% duty, the temp starts rising dramatically ! At 85% today, at 55A via 5x mosfets (only 11A each), temp. went up to 88 - 96 deg C on the "gate side" on the face of the fet. Thats when the "snap/crackle/pop" starts....:confused:

To date I have "popped", "smoked", "cracked" and sublimated a round total of +- 100 fets !!

What am I doing wrong ? I am not a professional electronics expert, just a very serious hobbyist. Am I driving the gates too hard ? The spec sheets all show gates driven at 10V ??:confused:

Any help out there ?
Manie
 

BeanieBots

Moderator
Old Mosfet question-new problem

I think it's still Old Mosfet question-same old problem.

Any chance of a circuit of your driver, I couldn't quite picture it.
Sounded like it was all PNP's. Needs a few NPN's in there as well for a proper push-pull driver.

'Scope pictures would be really useful too if at all possible.
One of gate drive voltage and one of voltage across FET.

Failing that, integrate all the 'bad' component parts and determine which one is contibuting most to the heating problem. Bottom line, they're getting hot. That can only be caused by:-
a) sloppy switching times (I'd need the pics to do the sums)
b) insufficient gate voltage causing poor Ron (can't be that if it's OK @ 100%)
c) excessive gate drive, preventing full turn-off. (what's it look like?)
 
Last edited:

manie

Senior Member
Beanie, thanks for your reply. Will try get schematic up for you but (but again...) try and picture this:

PWM from pin1 on 28x1 to base of PN2222A which is NPN. Pulse high switches npn ON, in this case I have tied the emitter to 0V, so the Collector then goes to 0V as well. This grounded collector now pulls the base of the pnp BC327 down to 0V (normally held high by pull-up 2k7, keeping it "off"). When the base goes to 0V the pnp switches "on". The pnp emitter is tied directly to 12V, thus passing +12V to the collector and via 47R to the gate. Hope this helps in the meantime.
Regards
 

BeanieBots

Moderator
OK, got it.
What draws the charge out of the gate during the other phase?
Presumably there's some resistance between NPN emmitter and PNP base???
 

kevrus

New Member
Manie,

This link may be helpful...http://www.richieburnett.co.uk/mosfail.html

It gives some good info about driving mosfets.

As a minor thought, you say your load is capacitive (and resistive), have you considered the possible inrush curent when charging a 'discharged' capacitor?
they can present an almost short circuit momentarily. I believe that for some high current DC speed controllers, a pre-charge resistor is used prior to energising the controller proper to limit this effect. Commonly found on DIY electric vehicles
 

Andrew Cowan

Senior Member
What is the load you are driving? Knowing that may help find the problem. If it has inductance, you will need a free-wheeling diode.

A
 

BeanieBots

Moderator
Kevrus,
When I read the first paragraph:-
"After emptying many pounds worth of dead MOSFETs from the trash can in my workshop, I decided that it might be worth actually compiling a list of reasons why MOSFET devices might fail in solid state tesla coil applications.
"
I thought, Oh dear, here we go,
but that is actually a very good article. Thanks for posting.

Andrew, read post #1. The load is resistive/capacitive.
(but I'm intrigued to know what it actually is)
 
Last edited:

kevrus

New Member
As BeanieBots and Andrew says, I too would be interested in the actual load, one could speculate for hours...come on Manie, please enlighten us...
 

jglenn

Senior Member
There are several problems with his setup. First of all, don't parallel FETs if you don't have to. You can get one that will do the job. Of course, probably easier to use 4 ea. 50A N-channel devices. When paralleling, take care to provide the same length of heavy, bare wire, connection. Daisy chaining with loops of too thin wire, and tapping off the end is no good, use a star config, each fet "sees" the same length of wire, and keep them as short as possible, like 1 or 2". There should be some cap to decouple the supply, short leads, keep the whole power loop small.

A ground plane perf board is good. Your gate driver circuit is dangerous, try a real FET driver like the TC4427, made by Teledyne, I have many of them, if you want a couple let me know, will mail for free (just the one guy). Driving FETs is a science, study the waveform, you are getting switching loss that heats during pwm. Turn on and turn off are both important. I learned to sometimes tailor the gate waveform by tuning the gate resistor, sometimes adding a ferrite bead, to shape it. Turning on too fast is bad, possible with bipolar drive like you have. And do a search on snubbers, another useful device, for the output side. Using a normal diode for the freewheeling rect is bad, will heat and cause bad waveforms, use a FAST RECOVERY type.
Before you blow another 100 fets, send me 10 and I'll build your driver.
:p

johnf@apk.net
 

jglenn

Senior Member
You don't want the fet to turn on instantly, but to have a turn on time that does not create large transients. There are parasitics all over the place, L, R, C. The gate itself has capacitance, thus the need for a good fet driver, you need to push in a bunch of electrons and pull them out quickly.

http://www.linear.com/pc/productDetail.jsp?navId=H0,C1,C1003,C1142,C1041,P8291

I'm not saying NPN/PNP can't be used to drive fets, I just have a recollection of someone who said it might not be the best. Regardless of the driver, the series R controls the gate rise time, along with the gate capacitance.

I do vaguely remember a ref on a way to blow a transistor up if you turn it on too fast, but do not have any links at this time. Not sure if true.
 

jglenn

Senior Member
A couple things I forgot to say. When switching a current on and off to an inductive load, you have a voltage generated, L di/dt (calculus, just means inductance times the rate of change of current. If you switch it fast, you get large transients). So sometimes, tailoring the turn on slew rate of a fet can minimize blips on the output. Turn on too slow, and you get excessive switching loss, heating in the fet.

This square wave switching also generates EMI. Companies spend a lot on shielding, filtering, etc to minimize it.
 

MikeGyver

Senior Member
Paralleling that may mosfets, you'll end up with a HUGE gate capacitance. Scope the gate under load and look at the rise/fall times, they're probably going to hell. In which case you'll need a gate driver than can sink/source schloads of current instantaneously.

Is your load inductive at all? if so you need a (FAST) reverse protection diode across the load.
 

manie

Senior Member
Thanks for all the input guys... appreciated. Wish I have a way to download cell phone image to my laptop. Then I can show actual oscilloscope shots, I.ve been told they looked VERY good by one who should know, rise time is straight up, REALLY, fall time is +- 10 uSec, mostly straight down, well almost. Have a look at the schematic above, it seems to work on the scope. Tested a single Mosfet today pushing 12V 54A, at 54 deg C, ambient was 34 deg C, so not bad, BUT at 54A, the Mosfet did what ??? YEAH, it went short circuit, but it was working hard though for about 1 hour...
manie
 

BeanieBots

Moderator
That's no good.
It can't switch the FET OFF fast enough.
The LED has a volt or 2 drop and won't pull all the way down.
Don't know the FET spec without reading up on it but would suspect that it won't turn fully off.

If you don't want get a 'proper' FET driver, at least build a totempole driver.
Details here:-
http://www.planetanalog.com/showArticle.jhtml?articleID=192200748

Shorting out the LED and halving the 1k resistor will help, but you really should build/buy a proper driver.
 
Last edited:

Dippy

Moderator
Completely agree BB. Far too slow.

I don't think it'll turn it fully ON either.
The rating for that MOSFET of 8milliOhms is at Vgs = 10V for the I.R. 3205 device.
Driving it at Vgs = (under) 5V with a low ohm load is asking for trouble.
A nice little resistive heater you have there.

Proper driving would have much lower gate resistance with proper push-pull.
And home-brew push-pull without care or 'scope can either be too slow or suffer the shoot-through. You need sub microsecond Gate transition time, unless you are designing a room heater? ;)

Is there a problem with getting (or even designing) a proper driver??
 

BeanieBots

Moderator
A "cheapskate" method I've used with good results is a PNP/NPN totempole driven by an op-amp. The op-amp power rail is taken from the higher voltage supply and configured as a 2.5v comparitor.
The article I linked to in the earlier gives tips on how to avoid shoot-through with totempoles.
 

Andrew Cowan

Senior Member
You could (don't know if you still can), get free 4427 drivers from Microchip.

Its a dual driver, 6 pins. V+, 0V, in A, in B, out A, out B. What could be simpler?

You can connect the picaxe straight to the input, and then run the FET gate at any voltage (up to 18V).

A
 

Dippy

Moderator
Maybe you can.

But I received an email from Microchip last year that they were going to remove freebies due to frivolous requests.
 

jglenn

Senior Member
I have 12 freebee 4427, if most you guys are in the UK, I'm willing to mail them over in one box, spread them around. I'll never use the amount I have.

The LED in the gate pulldown circuit is not good. A resistor would have been better. There should be symmetry in the pull up and pull down driver circuit, run a few amps peak current.
 

Dippy

Moderator
Alternatively, if Manie is in UK, there are at least 2 big suppliers who have drivers.
Order today (Saturday) and they could be in your sweaty puddies on Monday.
Obviously you'd buy more than one, maybe of a couple of different types too to try out.

Postage costs?
Whats a fiver compared to 100 FETs.....

But as the load was never described in detail then thats a variable you'll have to deal with on your own maybe ;)

Manie, a 10 microsecond transition time for the Gate is not impressive. And also you don't say the voltage swing of the Gate on the 'scope with FET under load.

I'm sure you'll get it sorted. Good luck.
 

BeanieBots

Moderator
All the advice given so far has been "basic general FET driving" advice.
This is very well and good, if adhered to, should produce the desired results.
However, I can fully appreciate any frustrations you might have with a circuit that you have built and where it might be possible to make it "OK" with a very simple change (possibly just a single resistor).

The only way we can very specific advice, is with a few pictures.
Possibly the most important picture would be of the voltage across the FET with a very clear indication of both the 0v reference and the supply voltage.
Failing that, the only sensible advice is "do it properly" with a FET driver.
 

manie

Senior Member
Thanks

Beannie, Dippy, Jglen, thanks all for input. Now I'm thouroughly discouraged ( LOL LOL), incidentally, I work and live in South Africa, and I think I'm the only Picaxe user in the country !!!!

The load is the age old attempt at producing Brown's Gas with hydrolysis... works well too but that is a totally DIFFERENT story as to how I got this far with that effort. Suffice to say, the gas production and subsequent application is a big success, just that I can't afford the cost in Mosfets....

I've shown the very latest effort driving the gate at 5V to prevent "over driving". All previous attempts fired the Gate with 11V on the DVM, scope showed 7.5V at low PWM duty and up to 10V at higher duties. I've tried some 6pin DIP IC drivers, I've had a 1K across the gate-source on the to-220 legs, totem pole driver as per Zetex sheet, you name it, I've tried (I think...) all ! The totem poles did not get the mosfet going AT ALL ! Only showed about 1.8V to 2.1V across the Gate/Source...

Also, although the mosfets are used in parrallel to distribute the load, I drive each mosfet individually from its own driver. I use one single NPN between the Picaxe and the drivers to ensure sufficient signal to each driver.

I'll ask another question in a new thread with regard to power pulsing a load at high currents.

Thanks all and Regards
Manie
 

BeanieBots

Moderator
Well Manie, not only are you not alone being in south africa using a PICAXE, you're not alone having problems with MOSFETs driven by PICAXE in south africa. Must be a cultural thing;)

Driving the FETs direct from the PICAXE would probably be better than what you have. (and still not advised).
Have you tried what I suggested re the LED and resistor?
The totempole values you stated are about right if you had them driven from 5v. The 2 tranny totempole is just a current amplifier. It won't give more volts than whatever drives it. Try one driven as described for the "cheapskate" version.
 

manie

Senior Member
Beannie, no, haven't tried it yet, the totems i triggered from the axe with a level shift (expected) to 12V, ie. the totem is connected to 12V feed, separate from the 5V supply. Sounds funny now ?
 

Jaguarjoe

Senior Member
If the problem is slow turn off time why can't you use push pull drivers referenced to a -Ve supply instead of ground? Wouldn't that shut them off faster?
 

jglenn

Senior Member
That is an old trick with normal transistors, called "sweeping the base", using a bit of negative base drive, say -6V. I have used it with a triple darlington that controlled 400A DC at 60V. Decreases shutoff time by about 4 or 5 times.

Don't think you need that with FETs. An N channel you just need 0 to 5V, or whatever the gate needs for full turn on. It is the job of the gate driver to get that current out quickly. Note the use of the chip in this link, the 4427 or others may be even better. The dwg is of a step down converter, to just drive the FET you do not need L1, D1, R8, R9, or R4. Switching times are on the order of 40nS, way faster than even 1 uS. Again, you can influence the turn on and turn off times by the value of the gate resistor, here they don't even have one. May be in the chip.

http://www.linear.com/pc/productDetail.jsp?navId=H0,C1,C1003,C1142,C1041,P8291

I've got some TO-247 giant N channel FETs lying around. Could whip up a small driver board. Fire it up. Not sure why you are having such problems. That LED gate pulldown circuit is totally not good, however.
 

BeanieBots

Moderator
We are now going round in circles so I'll point out post #2 again.
Ironically, it's turned out to be all three issues and "the same old problem".
 

manie

Senior Member
Will use driver IC

I have decided to go the TC4420cpa driver IC way (or similar chip). Which driver IC would you suggest ? Or are they all much the same ? The TC4420 has good voltage spread so level shifting from 5V to 12V is easy at least.

Next I would like to slow the pwm freq. down to as low as 31kHz on 28x1. This chip will only really be used to drive PWM at low frequency. Yes, I hear you ! What a waste of 28x1 chips ! But if it gives the results and reliability in Mosfet performance, its worth it ! I will use two 28x1's, one at 8MHz doing the calc's, data stores etc, Readadc's etc. and then serial comms to the other slower chip to change duty% there. Will probably have to use interrupt request on slower chip to tell it data is comming. What is the minimum baud rate I would be able to use ? Will I have to slow down the master chip from 8MHz while comms is on, then reset to 8MHz ?

How will I calculate the period and duty on the chip running @ 31KHz ? AAAAARRRGGGHHHHHH !

The idea behind this is to reduce the number of high stress periods on the Mosfets per second. I'm sure this will help them run cooler as the number of switches from off/on/off etc is dramatically reduced no....?

Totally new idea, so PLEASE HELP AGAIN...
Manie
 

Dippy

Moderator
jglenn's post.
I think it had been said almost word for word several pages previously.
... and confirmed by BB's post about going round in circles.
It just made me titter ;)

4420 looks OK. There are other non-inverting options but that looks nice'n'beefy.
 

BeanieBots

Moderator
If you are going to use a 'proper' driver, then you shouldn't need to drop the PICAXE clock.
Use the FULL resolution which also gives the lowest frequency PWMout.
Namely PWMout pin,255,N where N is between 0 & 1023 for the duty value.

Clock frequency does NOT effect the duty value.
If you do underclock and you REALLY want to know what the output frequency is, then simply use the wizard at 'normal' frequency and then divide by the ratio of the frequencies. ie, if you halved the clock speed, then the PWM frequency will be halved.
 

manie

Senior Member
De ja vy Dippy, hhhhmmmmmmmm......

BB, thought that. I've been playing with uC's for 5 months now, hell there is a lot to get into..... so 31KHz clock speed will mean 128 times division from 4MHz ? I will give ONE more mosfet a try at "normal" clock, then I'll try slower speed. I have a reasonably good digital thermometer (temp. sensor), one of those with the twin red laser dots to indicate area being measured... it shows .5 to 2 deg C difference between Gate "side" and Source "side" of TO-220 mosfet. So I will be able to see which freq. runs cooler... Now just to get hold of those IC's. When you live in the country 30km from Kruger National Park gate (blessing...) you have to get everything by Speed Services (2-3 days) or by Royal Mail (10-14 days). What the heck, life goes on...
 
Last edited:

jglenn

Senior Member
I had read all the posts, was just offering a viewpoint which overlapped I guess.

Why do you need such a high pwm freq? It won't work slower? It's switching loss causing heating, if you run slower it will be a lot less. Then can refine the waveforms before speeding up if you have to.
 

Dippy

Moderator
Eh?
"...down to as low as 31kHz on 28x1".
31,000 Hz

In post #1 you say 4kHz.

I realise my eyesight is failing but is this a typo.
Saying "down to" implies "from 4kHz ... down to 31kHz". ???.

Besides that, 31kHz isn't very low.

With a 4MHz xtal and having a poke the lowest PWM you can get is just under 250Hz.
Or, just under 0.25kHz.

The clock value can have a small effect on Duty resolution but its not enough to worry about, so don't; I have briefly donned the Anorak.

I think its time for a beer.

Wait til you get your driver wired up before making any decisions.
All the best.
 
Last edited:

manie

Senior Member
Its ALIVE !!

Thanks to all your comments, suggestions and plain talking straight "I think she's got it....". If that does not reveal my age nothing will.... anyway, thought I'd report back.

Landed some IRFP064N's in TO-247 livery. Also some *-oin DIP TC4420CPA's. Some Veroboard and good faith, and so far it works well. Here is what I've learnt thus far:
1 Spec'ns like "70A @ 60V" is not true for hard CONTINUOUS work.
2 Apply Pi, even though its DC, like 70A becomes 21A, like semi "RMS'ing"
3 GET YOUR PULSE SWITCHING THE BEST YOU CAN !!!!!!!!
4 Pulse freq. makes NO DIFFERENCE, well below a certain level anyway
5 Ensure a large enough heatsink for ADEQUATE cooling
6 Use a 12V/24V 90mm fan for good airflow and heat removal
7 Don't just bolt the mosfet down, clamp it across the body face as well
and use heat transfer compound both front and back
8 When testing, be REALISTIC about current passed and mosfet temp.

So, after all that, I tested with only one IRFP064N @ 20-22A max, 12V, up to 79-80 deg C temp on front heatsink clamp (measured on Gate side, it's hotter there and cooler on the Source side by about 5-10 deg C). So far after THREE DAYS of various tests, no "snap/cracckle/pop's", no fire and sparks, no sublimation of the entire TO-247 package !!! WOW !!! Hope this thread will help someone else to get their own problems sorted out.

Thanks for the input guys.
Manie
 

BeanieBots

Moderator
I have to comment on your points 3 & 4.

Point 3, that's we were telling you (and so many others) all along.
Point 4. Think about this for a moment.

In your initial posting, you stated
At 100% duty (gates fully "on", low RDS +- 8 mOhm) the mosfets run beautifully....: at close to ambient temp. of +- 38 deg C.
At 0% there is NO current, so NO heating.
At different duties, you were having heating problems.
Hence, switching speed IS IMPORTANT.

Even with proper good drivers it is still an important part of the equation.
 
Top