Topics

Power MOSFET failure due to overheating?

Alan G3XAQ
 

I have an ongoing 500W class-E RF amplifier project. I just let the magic smoke out of one of the pair of paralleled IXYS IXZ210N50L2 FETs. I don't wish to bore everyone with all the gory details save to say it worked fine at 100V DC supply but a FET expired at 114V as soon as I applied drive.

There are several (well, lots of) possible failures but I would like to know if anyone here is familiar with how a power FET looks after it has failed through high temperature caused by inadequate heatsinking? I'll attach some photo if groups.io will allow it. I lifted one of the FETs off the PCB and as it happens it is the one with the drain-source short. The other FET has survived, at least for now.

There are photos of the 12.5x12.5mm metal pad on the underside of the FET and the matching heatsink footprint covered in thermal paste. I hope you can see a fairly small blackened area on both. It's not easy to see but each FETs has one of these small blackened areas. Is this what a thermally stressed power FET looks like? For completeness I've added a photo of the top of the failed FET which shows no signs of having been overheated.

So, is the heatsink inadequate or was the primary cause due to something else such as voltage breakdown, oscillation, body diode forward biasing, or a failure in the external circuit (LPF flashover or suchlike)?

73, Alan G3XAQ

PS does anyone have any of these FETs in a drawer somewhere that they don't want? IXYS-RF has been taken over by Littelfuse and this and many other FETs have been dropped

Alan Ralph
 
Edited

Hi,
I wonder if there is too much paste ?

It might be a good idea to cut a hole in the PCB above the FET so you can apply some downward pressure to hold the fet againts the heatsink rather than the PC board.

On my Icom rig , they cut a hole in the PCB which has the outline of the PA device  body and then a sprung clamp to press down on the device. The leadouts are soldered on the top of the board rather than the underside.



Alan

Dave GW4GTE
 

Having experimented with Class E some years ago, I established 'comprehensively' that getting the power-down sequence wrong can be very destructive! Not saying that's the problem here, but if you remove the load before removing the drive - and possibly also removing the HT before removing the drive - it's an early bath for the FETs. (nasty back-emfs as is obvious but easily done).

Dave

On 02/09/2019 20:29, Alan G3XAQ wrote:
I have an ongoing 500W class-E RF amplifier project. I just let the magic smoke out of one of the pair of paralleled IXYS IXZ210N50L2  FETs. I don't wish to bore everyone with all the gory details save to say it worked fine at 100V DC supply but a FET expired at 114V as soon as I applied drive.

There are several (well, lots of) possible failures but I would like to know if anyone here is familiar with how a power FET looks after it has failed through high temperature caused by inadequate heatsinking? I'll attach some photo if groups.io will allow it. I lifted one of the FETs off the PCB and as it happens it is the one with the drain-source short. The other FET has survived, at least for now.

There are photos of the 12.5x12.5mm metal pad on the underside of the FET and the matching heatsink footprint covered in thermal paste. I hope you can see a fairly small blackened area on both. It's not easy to see but each FETs has one of these small blackened areas. Is this what a thermally stressed power FET looks like? For completeness I've added a photo of the top of the failed FET which shows no signs of having been overheated.

So, is the heatsink inadequate or was the primary cause due to something else such as voltage breakdown, oscillation, body diode forward biasing, or a failure in the external circuit (LPF flashover or suchlike)?

73, Alan G3XAQ

PS does anyone have any of these FETs in a drawer somewhere that they don't want? IXYS-RF has been taken over by Littelfuse and this and many other FETs have been dropped



Alan G3XAQ
 

Hello Alan,

I'm no expert but I thought there might be a bit too much paste. I've followed the IXYS recommended mounting method as exactly as I can, with an 1/8" FR4 load stretcher and Belville washers and all. The details are attached, in the section on "High Average Power - With Heat Sink".

73, Alan G3XAQ

Alan G3XAQ
 

Hello Dave,

You make some good points but at this stage of testing I had a permanent dummy load attached and had followed the power up/down sequence you recommend. The failure occurred almost the instant I applied drive.

73, Alan G3XAQ

Andy G4JNT
 

The biggest killer of MOSFETs is overvoltage
If it worked OK at Vs 100V, but died at 114V that suggests it could be the case here.

Look for spikes and unexpected waveform glitches.   Once the gate channel breaks down that's it, instant fully on and full current flow though the channel - meltdown.

Look at the saga of my 700W 137kHz PA -  http://g4jnt.com/137tx.pdf  most of the failures there were related to overvoltage glitches or brief overloads




Virus-free. www.avg.com


On Mon, 2 Sep 2019 at 21:55, Dave GW4GTE <dave@...> wrote:

Having experimented with Class E some years ago, I established 'comprehensively' that getting the power-down sequence wrong can be very destructive! Not saying that's the problem here, but if you remove the load before removing the drive - and possibly also removing the HT before removing the drive - it's an early bath for the FETs. (nasty back-emfs as is obvious but easily done).

Dave

On 02/09/2019 20:29, Alan G3XAQ wrote:
I have an ongoing 500W class-E RF amplifier project. I just let the magic smoke out of one of the pair of paralleled IXYS IXZ210N50L2  FETs. I don't wish to bore everyone with all the gory details save to say it worked fine at 100V DC supply but a FET expired at 114V as soon as I applied drive.

There are several (well, lots of) possible failures but I would like to know if anyone here is familiar with how a power FET looks after it has failed through high temperature caused by inadequate heatsinking? I'll attach some photo if groups.io will allow it. I lifted one of the FETs off the PCB and as it happens it is the one with the drain-source short. The other FET has survived, at least for now.

There are photos of the 12.5x12.5mm metal pad on the underside of the FET and the matching heatsink footprint covered in thermal paste. I hope you can see a fairly small blackened area on both. It's not easy to see but each FETs has one of these small blackened areas. Is this what a thermally stressed power FET looks like? For completeness I've added a photo of the top of the failed FET which shows no signs of having been overheated.

So, is the heatsink inadequate or was the primary cause due to something else such as voltage breakdown, oscillation, body diode forward biasing, or a failure in the external circuit (LPF flashover or suchlike)?

73, Alan G3XAQ

PS does anyone have any of these FETs in a drawer somewhere that they don't want? IXYS-RF has been taken over by Littelfuse and this and many other FETs have been dropped



Dave GW4GTE
 

Maybe there's a cap in there that suffered a breakdown creating a spike? I'd agree with Andy spikes are instant killers. The other point is depending on how the output is (class E) tuned the drain RF volts can be many times the DC voltage. and it doesn't always follow that you can determine the ratio at say 12V then scale up as the 'varicap' nature of the FETs will alter the tuning unless you have big swamping caps.

On 02/09/2019 22:10, Andy TALBOT wrote:
The biggest killer of MOSFETs is overvoltage
If it worked OK at Vs 100V, but died at 114V that suggests it could be the case here.

Look for spikes and unexpected waveform glitches.   Once the gate channel breaks down that's it, instant fully on and full current flow though the channel - meltdown.

Look at the saga of my 700W 137kHz PA -  http://g4jnt.com/137tx.pdf  most of the failures there were related to overvoltage glitches or brief overloads




Virus-free. www.avg.com

On Mon, 2 Sep 2019 at 21:55, Dave GW4GTE <dave@...> wrote:

Having experimented with Class E some years ago, I established 'comprehensively' that getting the power-down sequence wrong can be very destructive! Not saying that's the problem here, but if you remove the load before removing the drive - and possibly also removing the HT before removing the drive - it's an early bath for the FETs. (nasty back-emfs as is obvious but easily done).

Dave

On 02/09/2019 20:29, Alan G3XAQ wrote:
I have an ongoing 500W class-E RF amplifier project. I just let the magic smoke out of one of the pair of paralleled IXYS IXZ210N50L2  FETs. I don't wish to bore everyone with all the gory details save to say it worked fine at 100V DC supply but a FET expired at 114V as soon as I applied drive.

There are several (well, lots of) possible failures but I would like to know if anyone here is familiar with how a power FET looks after it has failed through high temperature caused by inadequate heatsinking? I'll attach some photo if groups.io will allow it. I lifted one of the FETs off the PCB and as it happens it is the one with the drain-source short. The other FET has survived, at least for now.

There are photos of the 12.5x12.5mm metal pad on the underside of the FET and the matching heatsink footprint covered in thermal paste. I hope you can see a fairly small blackened area on both. It's not easy to see but each FETs has one of these small blackened areas. Is this what a thermally stressed power FET looks like? For completeness I've added a photo of the top of the failed FET which shows no signs of having been overheated.

So, is the heatsink inadequate or was the primary cause due to something else such as voltage breakdown, oscillation, body diode forward biasing, or a failure in the external circuit (LPF flashover or suchlike)?

73, Alan G3XAQ

PS does anyone have any of these FETs in a drawer somewhere that they don't want? IXYS-RF has been taken over by Littelfuse and this and many other FETs have been dropped



Alan G3XAQ
 

Hi Andy. If a one-time spike can kill a FET how can I look for spikes and unexpected waveform glitches? The scope traces all looked like the simulations at lower voltage.

So what's the practical procedure I should follow here?

73, Alan

On Mon, Sep 2, 2019 at 10:10 PM, Andy TALBOT wrote:


The biggest killer of MOSFETs is overvoltage
If it worked OK at Vs 100V, but died at 114V that suggests it could be the
case here.

Look for spikes and unexpected waveform glitches. Once the gate channel
breaks down that's it, instant fully on and full current flow though the
channel - meltdown.

Look at the saga of my 700W 137kHz PA - http://g4jnt.com/137tx.pdf most
of the failures there were related to overvoltage glitches or brief
overloads


Andy
www.g4jnt.com



<http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
Virus-free.
www.avg.com
<http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
<#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>

On Mon, 2 Sep 2019 at 21:55, Dave GW4GTE <dave@...> wrote:

Having experimented with Class E some years ago, I established
'comprehensively' that getting the power-down sequence wrong can be very
destructive! Not saying that's the problem here, but if you remove the load
before removing the drive - and possibly also removing the HT before
removing the drive - it's an early bath for the FETs. (nasty back-emfs as
is obvious but easily done).

Dave
On 02/09/2019 20:29, Alan G3XAQ wrote:

I have an ongoing 500W class-E RF amplifier project. I just let the magic
smoke out of one of the pair of paralleled IXYS IXZ210N50L2 FETs. I don't
wish to bore everyone with all the gory details save to say it worked fine at
100V DC supply but a FET expired at 114V as soon as I applied drive.

There are several (well, lots of) possible failures but I would like to know
if anyone here is familiar with how a power FET looks after it has failed
through high temperature caused by inadequate heatsinking? I'll attach some
photo if groups.io will allow it. I lifted one of the FETs off the PCB and as
it happens it is the one with the drain-source short. The other FET has
survived, at least for now.

There are photos of the 12.5x12.5mm metal pad on the underside of the FET
and the matching heatsink footprint covered in thermal paste. I hope you can
see a fairly small blackened area on both. It's not easy to see but each FETs
has one of these small blackened areas. Is this what a thermally stressed
power FET looks like? For completeness I've added a photo of the top of the
failed FET which shows no signs of having been overheated.

So, is the heatsink inadequate or was the primary cause due to something
else such as voltage breakdown, oscillation, body diode forward biasing, or a
failure in the external circuit (LPF flashover or suchlike)?

73, Alan G3XAQ

PS does anyone have any of these FETs in a drawer somewhere that they don't
want? IXYS-RF has been taken over by Littelfuse and this and many other FETs
have been dropped






Alan G3XAQ
 

Dave,

After the event I can't find any failed components except the dead FET. The decoupling and low pass filter caps are all 1500V/20A ATC 100B porcelain types.

[enter gory detail mode]

The circuit isn't real class-E. It's what Hans Summers calls "ghetto class E", in that the drive is sine wave but driving the gate quite hard and the drain just has an inductor feeding DC and being tuned by the FET output capacitance, although the LPF input cap probably does swamp the waveform a bit. I used LTspice and a VDMOStool model of the FETs. The observed scope waveforms (at lower voltage before destruction) were very similar to the simulations.

Someone else off this list has had a bit of a rant at me to say the clearances around the high voltage lands on the PCB are way too small. He also said I have used too much thermal paste!

73, Alan G3XAQ

Ken Eastty
 

Alan

I  doubt that this the cause of you problem however in the process of doing another repair I've blown the HF driver MOSFET in my IC 706 Mk 2g by applying volts to the drain whilst the source was floating. There's series R-C between Drain & Gate I suspect that this put a spike on the Gate. (unfortunately the HF & V/UHF drivers share the same heat spreader which makes the job of changing the device more difficult than it should be).

73...

Ken

G3LVP


---
This email has been checked for viruses by AVG.
https://www.avg.com

Dave GW4GTE
 

I wonder if the drive was a bit high - there's a limit of course. The theory of applying overdriven sine wave to steepen the relevant part of the rising and falling edge is ok as long as you don't bust the gate with overvolts.

I think I read somewhere that the optimum drive shape is sawtooth.

Maybe there's some parasitics or odd resonances. Try a few ohms in series with the gates next time - I found that cleaned things up quite a bit.

Dave

On 02/09/2019 22:38, Alan G3XAQ wrote:
Dave,

After the event I can't find any failed components except the dead FET. The decoupling and low pass filter caps are all 1500V/20A ATC 100B porcelain types.

[enter gory detail mode]

The circuit isn't real class-E. It's what Hans Summers calls "ghetto class E", in that the drive is sine wave but driving the gate quite hard and the drain just has an inductor feeding DC and being tuned by the FET output capacitance, although the LPF input cap probably does swamp the waveform a bit. I used LTspice and a VDMOStool model of the FETs. The observed scope waveforms (at lower voltage before destruction) were very similar to the simulations.

Someone else off this list has had a bit of a rant at me to say the clearances around the high voltage lands on the PCB are way too small. He also said I have used too much thermal paste!

73, Alan G3XAQ




Hans Summers
 

Hi Alan

> The circuit isn't real class-E. It's what Hans Summers calls "ghetto class E", in that the
> drive is sine wave but driving the gate quite hard and the drain just has an inductor
> feeding DC and being tuned by the FET output capacitance, although the LPF input
> cap probably does swamp the waveform a bit. I used LTspice and a VDMOStool
> model of the FETs. The observed scope waveforms (at lower voltage before
> destruction) were very similar to the simulations.

Hey hey hey your ghetto ain't the same as my ghetto, man... are you saying my QCX isn't really Class-E? http://qrp-labs.com/qcx - The "ghetto" method was my humorous way of describing how I arrive at the circuit values without needing to go through a lot of complex mathematics. In QCX the gate is driven by a squarewave. There is a resonant circuit at the drain. The waveforms do match what is expected for Class-E and so does the efficiency. 

My condolences on the demise of your MOSFETs. I have been playing with amps a bit lately though not as big as yours, nor Class E. It appears to me, based on recent experiences, that there are many ways for MOSFETs to meet their end. Over-voltage (drain-source, and let's not forget gate-source), over-current, over-heat. Spontaneous enormous oscillation self-destruction. Noisy with a huge big bang, or quietly without a whimper. 

They're probably not wrong about the high voltage lands and the thermal paste. But whatever happens, however right you do everything... I think you're still going to be sending some MOSFETs to meet their maker. It's par for the course, when playing QRO with MOSFETs. That's why I like playing with cheap types like IRF510  :-D

73 Hans G0UPL
http://qrp-labs.com

John Rabson
 

Another cause of demise of the FET in a Class E amplifier is not having a sufficiently square wave to drive it. I discovered this one day when I inadvertently set the signal generator for sine wave output.

John F5VLF

On 3 Sep 2019, at 08:18, Hans Summers <hans.summers@...> wrote:

Hi Alan

> The circuit isn't real class-E. It's what Hans Summers calls "ghetto class E", in that the
> drive is sine wave but driving the gate quite hard and the drain just has an inductor
> feeding DC and being tuned by the FET output capacitance, although the LPF input
> cap probably does swamp the waveform a bit. I used LTspice and a VDMOStool
> model of the FETs. The observed scope waveforms (at lower voltage before
> destruction) were very similar to the simulations.

Hey hey hey your ghetto ain't the same as my ghetto, man... are you saying my QCX isn't really Class-E? http://qrp-labs.com/qcx - The "ghetto" method was my humorous way of describing how I arrive at the circuit values without needing to go through a lot of complex mathematics. In QCX the gate is driven by a squarewave. There is a resonant circuit at the drain. The waveforms do match what is expected for Class-E and so does the efficiency. 

My condolences on the demise of your MOSFETs. I have been playing with amps a bit lately though not as big as yours, nor Class E. It appears to me, based on recent experiences, that there are many ways for MOSFETs to meet their end. Over-voltage (drain-source, and let's not forget gate-source), over-current, over-heat. Spontaneous enormous oscillation self-destruction. Noisy with a huge big bang, or quietly without a whimper. 

They're probably not wrong about the high voltage lands and the thermal paste. But whatever happens, however right you do everything... I think you're still going to be sending some MOSFETs to meet their maker. It's par for the course, when playing QRO with MOSFETs. That's why I like playing with cheap types like IRF510  :-D

73 Hans G0UPL
http://qrp-labs.com

Alan G3XAQ
 

Hi Ken. No, in my case I have no negative feedback RC between gate and drain. There is just the unavoidable Crss. My FET sources are firmly soldered to the ground plane.

73, Alan G3XAQ

On Mon, Sep 2, 2019 at 10:57 PM, Ken Eastty wrote:


Alan

I  doubt that this the cause of you problem however in the process of
doing another repair I've blown the HF driver MOSFET in my IC 706 Mk 2g
by applying volts to the drain whilst the source was floating. There's
series R-C between Drain & Gate I suspect that this put a spike on the
Gate. (unfortunately the HF & V/UHF drivers share the same heat spreader
which makes the job of changing the device more difficult than it should
be).

73...

Ken

G3LVP


---
This email has been checked for viruses by AVG.
https://www.avg.com

Alan G3XAQ
 

Dave,

I'm applying less than 13V peak drive to the gates. I already have extensive swamping. A sawtooth may be optimum but my transceiver only delivers sine waves.

73, Alan G3XAQ

On Tue, Sep 3, 2019 at 12:14 AM, Dave GW4GTE wrote:


I wonder if the drive was a bit high - there's a limit of course. The theory
of applying overdriven sine wave to steepen the relevant part of the rising
and falling edge is ok as long as you don't bust the gate with overvolts.

I think I read somewhere that the optimum drive shape is sawtooth.

Maybe there's some parasitics or odd resonances. Try a few ohms in series with
the gates next time - I found that cleaned things up quite a bit.

Dave

On 02/09/2019 22:38, Alan G3XAQ wrote:


Dave,

After the event I can't find any failed components except the dead
FET. The decoupling and low pass filter caps are all 1500V/20A ATC 100B
porcelain types.

[enter gory detail mode]

The circuit isn't real
class-E. It's what Hans Summers calls "ghetto class E", in that the drive
is sine wave but driving the gate quite hard and the drain just has an
inductor feeding DC and being tuned by the FET output capacitance,
although the LPF input cap probably does swamp the waveform a bit. I used
LTspice and a VDMOStool model of the FETs. The observed scope waveforms
(at lower voltage before destruction) were very similar to the
simulations.

Someone else off this list has had a bit of a rant at me to
say the clearances around the high voltage lands on the PCB are way too
small. He also said I have used too much thermal paste!

73, Alan G3XAQ



Alan G3XAQ
 

OK Hans. Sorry for misrepresenting you. I was thinking more about my drain circuit, which copies your elegant simplicity. Yes, I only have a sine wave drive source so I can only drive it as hard as is safe and accept lower efficiency. I am also losing efficiency in the 1.75 ohm Rds(on) of my FETs . At the 4A or more peak drain current (per FET) in my circuit there is the best part of 20W loss just in the drain "on" resistance.

This thread has generated more traffic than I expected so I'll raise the veil and give you some of the gory details of my circuit. I'll attach a screenshot of the circuit diagram and also an LTspiceXVII model file so you can all pick apart my "empirical" approach to circuit design. Just remember: my first degree is in Biochemistry. Oh, and if you remove the 25 ohm transmission line that represents the 2xRG316 coax interconnect between the PA and LPF boards the simulation waveforms become much "cleaner" and perhaps easier to fathom.

On Tue, Sep 3, 2019 at 07:18 AM, Hans Summers wrote:


Hi Alan

The circuit isn't real class-E. It's what Hans Summers calls "ghetto class
E", in that the
drive is sine wave but driving the gate quite hard and the drain just has an
inductor
feeding DC and being tuned by the FET output capacitance, although the LPF
input
cap probably does swamp the waveform a bit. I used LTspice and a VDMOStool
model of the FETs. The observed scope waveforms (at lower voltage before
destruction) were very similar to the simulations.
Hey hey hey your ghetto ain't the same as my ghetto, man... are you saying my
QCX isn't really Class-E? http://qrp-labs.com/qcx - The "ghetto" method was my
humorous way of describing how I arrive at the circuit values without needing
to go through a lot of complex mathematics. In QCX the gate is driven by a
squarewave. There is a resonant circuit at the drain. The waveforms do match
what is expected for Class-E and so does the efficiency.

My condolences on the demise of your MOSFETs. I have been playing with amps a
bit lately though not as big as yours, nor Class E. It appears to me, based on
recent experiences, that there are many ways for MOSFETs to meet their end.
Over-voltage (drain-source, and let's not forget gate-source), over-current,
over-heat. Spontaneous enormous oscillation self-destruction. Noisy with a
huge big bang, or quietly without a whimper.

They're probably not wrong about the high voltage lands and the thermal paste.
But whatever happens, however right you do everything... I think you're still
going to be sending some MOSFETs to meet their maker. It's par for the course,
when playing QRO with MOSFETs. That's why I like playing with cheap types like
IRF510  :-D

73 Hans G0UPL
http://qrp-labs.com

Alan G3XAQ
 

John, I only have sine wave drive available. My only option is to use lots of drive, but not so much that I damage the FET gates.

73, Alan G3XAQ

Brian Howie GM4DIJ
 

Plot vd*id on your LTSPice model and see if there is an instantaneous power spike. You may have done it already. I've had semiconductors blowing because of that, not Fets but same issue.

Brian

On 03/09/2019 14:41, Alan G3XAQ wrote:
John, I only have sine wave drive available. My only option is to use lots of drive, but not so much that I damage the FET gates.
73, Alan G3XAQ
--
Brian

Peter Chadwick
 

Alan.

Why not use a CMOS HEX invertor with say 4 devices in parallel for the output and two devices to drive them, and feed in a low level sine wave - about half the Vdd pk-pk? That could get you an up to 15 volts p-p reasonably square wave drive depending on the Vdd. How square would depend on the input capacity of course.

73

Peter G3RZP

------ Original Message ------
From: "Alan G3XAQ" <alan@...>
To: RSGB-Workshop@groups.io
Sent: Tuesday, 3 Sep, 19 At 14:41
Subject: Re: [RSGB-Workshop] Power MOSFET failure due to overheating?

John, I only have sine wave drive available. My only option is to use lots of drive, but not so much that I damage the FET gates.

73, Alan G3XAQ

Andy G4JNT
 

Or better, a proper MOSFET driver like the MCP14E4



Virus-free. www.avg.com


On Tue, 3 Sep 2019 at 16:05, Peter Chadwick via Groups.Io <g8on=btinternet.com@groups.io> wrote:
Alan.

Why not use a CMOS HEX invertor with say 4 devices in parallel for the
output and two devices to drive them, and feed in a low level sine wave
-  about half the Vdd pk-pk? That could get you an up to 15 volts p-p
reasonably square wave drive depending on the Vdd. How square would
depend on the input capacity of course.

73

Peter G3RZP

------ Original Message ------
From: "Alan G3XAQ" <alan@...>
To: RSGB-Workshop@groups.io
Sent: Tuesday, 3 Sep, 19 At 14:41
Subject: Re: [RSGB-Workshop] Power MOSFET failure due to overheating?

John, I only have sine wave drive available. My only option is to use
lots of drive, but not so much that I damage the FET gates.

73, Alan G3XAQ