Hot MosFet

[K9ACT]

The attached schematic is the driver I recently  built for my 8000 rig. 

I have destroyed at least 6 transistors in the process but only on the bench.  I have run it on the air for over a month and it works most of the time but never blows transistors in actual use.

The strange thing is that it is always the same transistor which I think of as the upper one on the schematic.

What is even stranger is that the heat sink temp of the naughty one runs substantially hotter (50%) than the well behaved one.  I have measured it with infrared and thermocouples and this difference is always there.  I see numbers like 100F and 160F.

One might suspect that a single bias supply for both might be the problem but this would not explain how 6 transistors from two different sources would all be off in the same direction.

Any ideas?  On this or the design in general. Any problems?

js

[KD6VXI]

Jack,

I have a similiar problem....  And can't figure it out either, so I'm going to be watching your thread.

What voltage are you using for bias....  I've done simple voltage dividers on mine, using a silicon diode, but it's a bit too late for math at this time in the evening.


Thanks!

--Shane

[KF1Z]

Check the Drain of each of the 510s and see that the waveform looks the same on both.
And that the voltage isn't going too high.

I'd suspect that it might be rising to about the voltage limit of the devices (100v)
May need to add a capacitor or 2 either from each Drain to ground, or from Drain to Drain. To keep the voltage down.


Also check the gates as well.
You maybe able to tell easily what's happening once you get a scope on it.

[W3RSW]

Hmmm, an imbalance somewhere. Yes, I think split secondary windings to each device with separately adjustable bias would more than help. Perhaps the windings are not accurately center tapped.  Even if accurately center tapped if the primary is offset by a few turns then an imbalance may occur.  

Also, generally when two devices are in parallel, one will hog the current (yes I know they're in push pull) so perhaps a, say 0.5 ohm, resistor in the source leads might help stabilize things.  What are the parameters of the circuit? - power, current to each 'final,' measured voltages at the gates, etc.?  Have you taken voltages under operating conditions? Adding temporary, small value source resistors will allow you to see a voltage difference between the two devices at the souces even if you decide to take them out later.

[KM1H]

Id lean to parasitics. Try a few Ohms of carbon in each source for starters and also in the drains as a next step.

Be sure the gates are at an absolute minimum inductance ground. The .1 caps need to be low inductance also. The usual el cheapo square block style are poor choices.

Carl
KM1H

[WD5JKO]

Jack,

   If the drive from T1 to T2 is not symmetrical, then each power FET will see a different gate drive waveform. This will alter the conduction period between the two FET's from the excitation. There will also be a DC offset on the gate bias, but that will be common to both FET's. So why the 9.1K between your TL431 Zener and T2 centertap?

   Since this driver is for a class C tube amplifier, I'd use something like what the Class E folks use, and not worry about making my driver a linear amplifier.

Good Luck!
Jim
WD5JKO

[WA1GFZ]

You may have a layout problem. Assuming you have a perfect layout with real short leads I would increase the bypass on the CT of the output transformer. Add a couple more .1s and a .01 in parallel. Also consider adding a 1 ohm noninductive resistor in series with each source to help balance the bias.
One fet will hog current if the FETs don't thermal track each other.
Check it with a scope to see if one FET is oscillating. An IRF510 will easily oscillate at 80 MHz under the right conditions. a long gate connection will make them take off. A small series resistor on the gate will help 4.7 or 10 ohms.
You may have the bias set too high. Since this is a CW amplifier you can get away with the bias set to just into conduction. Class AB. 50 ma should be fine for two FETs. This would put the resting current dissipation at .3 watts each. 
I'm working on a FQA11N90 linear and will use a pair of 1 ohm 2 watt surface mount resistor in series with each source. This with RF feedback helps linearity.
You could add RF feedback with say 100 ohms in series with .01 uf between each drain and gate.

[KF1Z]

You said it hasn't blown a transistor "in actual use".

When you lost fets on the bench, how did you terminate the output for testing?


Just curious.

[WA1GFZ]

The class a driver is look into a non resistive load you might consider a couple 100 ohm resistors across the final gate transformer secondary ends to ct. You have 1:2 so 50 to 200 ohms. The driver gain might be pretty high with that load on it. It wants to see a broadband 50 ohm load. I have seen bipolar class A amplifiers like that oscillate at VHF with a high Z load on it. 2N3553 is a good device. You could try a 2N3866 or 2N5109 for higher frequency operation

[Ian VK3KRI]

Gate thresholds on MOSFETS tend to be all over the place, more so if theyre from a different batch and manafacturer.  Maybe six dead mosfets arn't off in the same direction, maybe the one that hasn't died has a higher than normal gate threshold. ie off in the other direction.

Also the gate threshold voltage decreases as temperature increases. So a mosfet in linear mode with fixed bias can warm up, dropping its threshold, increasing the standing current , making it hotter and round it goes.
 
                         Ian VK3KRI

[K9ACT]

You said it hasn't blown a transistor "in actual use".

When you lost fets on the bench, how did you terminate the output for testing?

Good question but first let me thank everyone for all the good ideas and questions.

This one is of particular interest because of what I learned today.

I made a few changes and it seemed very clean and stable, putting out more power with less drain current so I put it back with the 8000 rig and used it on the net today.

Just before un-keying after each of my buzzardly transmissions, I checked the temp of the two Mosfets  and they were always within a few degrees of each other and never over 90F.

This is looking into the neutralized grid circuit of the 8000's as discussed in a previous thread.

All bench testing is done with an MFJ tuner/dummy and always shows gross differences in temp.

After the net, I took it back to the bench and ran it into the dummy load and again 80F and 140F.

Put it back on the 8000's tonite and hooked up with Brent W1IA for about 40 min, checking the temp regularly and again.... the same low temp on both.

So, I am not sure that answering all the other questions would lead anywhere now.

Besides, why fix what aint broke?

Cause it's fun and I hate unsolved mysteries.

js

[K9ACT]

> So why the 9.1K between your TL431 Zener and T2 centertap?

It's part of the divider that puts 3.6v on the CT.  The voltage is 6.3 on the other side of the resistor.

>   Since this driver is for a class C tube amplifier, I'd use something like what the Class E folks use, and not worry about making my driver a linear amplifier.

My thoughts too but I can't seem to find any info on doing this.  I looked through the Class E files and found nothing.  All I recall reading is that they are less efficient in Class C which seems a bit strange.

I would think I could get the power I need (12W) from a single transistor and eliminate all the balance problems.

My first attempt was such but I blew up a lot of transistors and assumed I was pushing it too had and switched to the PP circuit.

js

 

[KF1Z]

Seems like you've answered it then....


The dummy load is a pure resistance at 50ohms.

I'd have to assume the "8000's" are presenting a different impedance...

Obviously one that your driver circuit enjoys!

Rock-on!

 

[WA1GFZ]

Good chance the portion of the cycle you draw grid current is when the load z will go down. below 100f is good. That is where my class e final runs unless it is very warm in the shack. I glued a thermalcouple on one of the fet cases so I can monitor the case temperature.
Another thing you can do is add temperature compensation to the bias circuit. 4 or 5 diodes in series attached to the heatsink connect one end to the CT and cathode to ground. As the diodes heat up the junction voltage will drop pulling the bias down.
10 watts on a pair of IRF510s sounds fine. I run 1500 PEP on 160 with 14 IRF840s in push pull parallel and it has been many years since I took out a FET. The last time was when a tree pulled my antenna down in the middle of a qso. I kept keying the rig to see why it wasn't happy. The last time I noticed the reflected power was pegged as the FET blew. 

[Opcom]

Ok. So there is something right about the real load, and something wrong about the bench test. If I am on the right track. the words "MFJ tuner" are thrown in with the dummy load... so what about the tuner? maybe it's fine generally but something, of a scientific nature, about the setup is causing the FET to get hot.

Can you substiture a different dummy sans tuner?

Could there be some kind of ground loop feedback happening on the bench that does not happen in the rig? A test would be to reverse the polarity on the output transformer and see of the other fet then suffers?

I'm scratching my head -the diagram looks fine.

[WA1GFZ]

I didn't catch you were using a tuner between the amp and dummy load. I would eliminate the tuner and go straight into the dummy load. You could be reflecting a reactive load back at the amplifier and make it act weird. You have a low pass filter on the output of the amo so don't see why you need the tuner. Life is easier when stages are running at a fixed Z like 50 ohms.

[KF1Z]

I assumed he meant he was using the 300 watt dummy load built into the tuner...
Bypassing the tuner portion.... (flip of a switch)

[K9ACT]

I assumed he meant he was using the 300 watt dummy load built into the tuner...
Bypassing the tuner portion.... (flip of a switch)

Roger.  I think the tuner is completely bypassed when switched to dummy. I also hang a 15W light bulb on one of the tuned channels as an alternate but I get the same results either way.

It also draws more current on the dummy than in the rig.  Two vs about 1.5A.

js

[WA1GFZ]

so the rig is class C with a smaller conduction angle than a load resistor. The avreage load resistance must be higher than 50 ohms.