Gate resistor for lateral power MOSFET?

Winfield Hill wrote...
>
>Phil Allison wrote...
>>
>> As you have realised, lateral mosfets have lots of power
>> gain up into the VHF range and a little stray L or C
>> will set easily them off.  
>
> Ordinary VMOS power MOSFETs have *much* higher gm in the
> linear region than lateral MOSFETs, and techniques that
> work for them should be conservative for the laterals.

 I retract this statement, and bow to Phil's comments.


-- 
 Thanks,
    - Win

cassiope wrote:
>
>
> >>
> >>
> > ** Gate resistors of about 200ohms are necessary, but not sufficient to
> > ensure freedom from parasitic oscillations. As you have realised,
> > lateral mosfets have lots of power gain up into the VHF range and a
> > little stray L or C will set easily them off. Wiring layout should be
> > done with this in mind, keeping all tracks short as possible. You can
> > avoid heatsink capacitance by isolating it from ground and connecting
> > all the mosfet sources directly so it floats at output voltage.
> 
>
> Unfortunately I have three channels of amplifiers, and the physical 
> design (this is a "feature improved" replacement of a legacy device)
> makes separate heat sinks kinda unlikely/difficult.  But now that I see
> this I should revisit this, though it would take a huge package redesign
> to accommodate separate heat-sinks for N and P devices and retain
> compatibility with the existing units.


** Lateral mosfets have all have the case connected to the source, so one hestsink can be used for all devices in a channel without any insulators.

I assume you are not using source ballast resistors as most designs leave them out. 

> So far the heat sinks have
> metal exposed to the outside world, which would have to change if not
> grounded.


** Another option is to ground the sources of all the mosfets, via a common heatsink that can serve more than one channel. This requires that the centre point of the PSU be available to serve as the output as in this Haffler schematic:

http://bmamps.com/Schematics/Hafler/Hafler_9300,_9500_Schematic.pdf


> 
> How did you determine the 200 ohms?
>  

 ** I didn't, but many other designers have settled on values of 220 ohms to 470 ohms per device for the job. Often N channel devices having a higher value than used for the P channel ones. 


> > An output stabilising network is essential, consisting of a inductor of
> > about 5uH with RC Zobels on both ends to supply ground.
> 
> Ouch.  5uH has ~12ohms reactance at 400kHz.  I may need some inductance 
> here, but hopefully less.  The usual Zobel impedances are also too low,
> consuming far too much of the output at my much-higher-than-audio 
> frequencies.
> 

** Yeah, that is a problem. 

 Any chance you could operate the mosfets outside the NFB loop with the gates linked via 220ohm resistors? Lateral mosfets start conducting at 0.4V so there would be a little crossover distortion and the output impedance would be around 1 ohm. 

....   Phil 

On Thu, 29 Sep 2016 18:52:53 -0700 (PDT), Phil Allison
<pallison49@gmail.com> wrote:

>cassiope wrote:
>>
>>
>> >>
>> >>
>> > ** Gate resistors of about 200ohms are necessary, but not sufficient to
>> > ensure freedom from parasitic oscillations. As you have realised,
>> > lateral mosfets have lots of power gain up into the VHF range and a
>> > little stray L or C will set easily them off. Wiring layout should be
>> > done with this in mind, keeping all tracks short as possible. You can
>> > avoid heatsink capacitance by isolating it from ground and connecting
>> > all the mosfet sources directly so it floats at output voltage.
>> 
>>
>> Unfortunately I have three channels of amplifiers, and the physical 
>> design (this is a "feature improved" replacement of a legacy device)
>> makes separate heat sinks kinda unlikely/difficult.  But now that I see
>> this I should revisit this, though it would take a huge package redesign
>> to accommodate separate heat-sinks for N and P devices and retain
>> compatibility with the existing units.
>
>
>** Lateral mosfets have all have the case connected to the source, so one hestsink can be used for all devices in a channel without any insulators.
>
>I assume you are not using source ballast resistors as most designs leave them out. 
>
>> So far the heat sinks have
>> metal exposed to the outside world, which would have to change if not
>> grounded.
>
>
>** Another option is to ground the sources of all the mosfets, via a common heatsink that can serve more than one channel. This requires that the centre point of the PSU be available to serve as the output as in this Haffler schematic:
>
>http://bmamps.com/Schematics/Hafler/Hafler_9300,_9500_Schematic.pdf
>
>
>> 
>> How did you determine the 200 ohms?
>>  
>
> ** I didn't, but many other designers have settled on values of 220 ohms to 470 ohms per device for the job. Often N channel devices having a higher value than used for the P channel ones. 
>
>
>> > An output stabilising network is essential, consisting of a inductor of
>> > about 5uH with RC Zobels on both ends to supply ground.
>> 
>> Ouch.  5uH has ~12ohms reactance at 400kHz.  I may need some inductance 
>> here, but hopefully less.  The usual Zobel impedances are also too low,
>> consuming far too much of the output at my much-higher-than-audio 
>> frequencies.
>> 
>
>** Yeah, that is a problem. 
>
> Any chance you could operate the mosfets outside the NFB loop with the gates linked via 220ohm resistors? Lateral mosfets start conducting at 0.4V so there would be a little crossover distortion and the output impedance would be around 1 ohm. 
>
>....   Phil 
>


I would imagine that the gate resistor could be very high in value, at
least until it gets in the way of the desired slew rate or high end
frequency response of the amplifier ?

boB

On Thu, 29 Sep 2016 18:52:53 -0700, Phil Allison wrote:

> cassiope wrote:
>>
>>
>> >>
>> >>
>> > ** Gate resistors of about 200ohms are necessary, but not sufficient to
>> > ensure freedom from parasitic oscillations. As you have realised,
>> > lateral mosfets have lots of power gain up into the VHF range and a
>> > little stray L or C will set easily them off. Wiring layout should be
>> > done with this in mind, keeping all tracks short as possible. You can
>> > avoid heatsink capacitance by isolating it from ground and connecting
>> > all the mosfet sources directly so it floats at output voltage.
>> 
>>
>> Unfortunately I have three channels of amplifiers, and the physical 
>> design (this is a "feature improved" replacement of a legacy device)
>> makes separate heat sinks kinda unlikely/difficult.  But now that I see
>> this I should revisit this, though it would take a huge package redesign
>> to accommodate separate heat-sinks for N and P devices and retain
>> compatibility with the existing units.
> 
> 
> ** Lateral mosfets have all have the case connected to the source, so one hestsink can be used for all devices in a channel without any insulators.
> 
> I assume you are not using source ballast resistors as most designs leave them out. 

Actually I'd put some in :( figuring that they'd be convenient to measure
the quiescent current.  Before I fixed my bias generator (using the
thermal sense diode attached to one of the MOSFETS), the current wasn't
as stable as I wanted so I bumped these to 0.22ohms.

>> So far the heat sinks have
>> metal exposed to the outside world, which would have to change if not
>> grounded.
> 
> 
> ** Another option is to ground the sources of all the mosfets, via a common heatsink that can serve more than one channel. This requires that the centre point of the PSU be available to serve as the output as in this Haffler schematic:
> 
> http://bmamps.com/Schematics/Hafler/Hafler_9300,_9500_Schematic.pdf

I'd seen this design earlier - cute!  IIRC it requires (one? two?) separate 
floating power supplies for each channel, unfortunately, which makes the cure
probably worse than the disease in my situation.

>> How did you determine the 200 ohms?
>>  
> 
>  ** I didn't, but many other designers have settled on values of 220 ohms to 470 ohms per device for the job. Often N channel devices having a higher value than used for the P channel ones. 

Certainly the N channel device has less capacitance so the disparity makes
sense.  I'd sure like to understand the need for these resistors better than
the WAG about HF oscillations.

>> > An output stabilising network is essential, consisting of a inductor of
>> > about 5uH with RC Zobels on both ends to supply ground.
>> 
>> Ouch.  5uH has ~12ohms reactance at 400kHz.  I may need some inductance 
>> here, but hopefully less.  The usual Zobel impedances are also too low,
>> consuming far too much of the output at my much-higher-than-audio 
>> frequencies.
>> 
> 
> ** Yeah, that is a problem. 
> 
>  Any chance you could operate the mosfets outside the NFB loop with the gates linked via 220ohm resistors? Lateral mosfets start conducting at 0.4V so there would be a little crossover distortion and the output impedance would be around 1 ohm. 
> 
> ....   Phil

That's something I'm hoping to explore today.  The feedback network is
a standard pair of parallel RC (identical t.c.s); so I'm going to try
connecting the output-side resistor to the "real" output, and the 
capacitor to the output of the voltage gain stage.  The breakpoint is
around 1MHz, so distortion may be tolerable.

On Fri, 30 Sep 2016 00:54:45 -0700, boB wrote:

[snip]
> 
> I would imagine that the gate resistor could be very high in value, at
> least until it gets in the way of the desired slew rate or high end
> frequency response of the amplifier ?
> 
> boB

The input capacitances are so high that I'm observing ringing step
responses with "higher" (300ohm) gate resistors with no load; and more
problems with reactive loads.  Thus my effort to better understand
what's going on...

   -F

On Friday, September 30, 2016 at 11:53:51 AM UTC-4, cassiope wrote:
> On Thu, 29 Sep 2016 18:52:53 -0700, Phil Allison wrote:
> 
> > cassiope wrote:
> >>
> >>
> >> >>
> >> >>
> >> > ** Gate resistors of about 200ohms are necessary, but not sufficient to
> >> > ensure freedom from parasitic oscillations. As you have realised,
> >> > lateral mosfets have lots of power gain up into the VHF range and a
> >> > little stray L or C will set easily them off. Wiring layout should be
> >> > done with this in mind, keeping all tracks short as possible. You can
> >> > avoid heatsink capacitance by isolating it from ground and connecting
> >> > all the mosfet sources directly so it floats at output voltage.
> >> 
> >>
> >> Unfortunately I have three channels of amplifiers, and the physical 
> >> design (this is a "feature improved" replacement of a legacy device)
> >> makes separate heat sinks kinda unlikely/difficult.  But now that I see
> >> this I should revisit this, though it would take a huge package redesign
> >> to accommodate separate heat-sinks for N and P devices and retain
> >> compatibility with the existing units.
> > 
> > 
> > ** Lateral mosfets have all have the case connected to the source, so one hestsink can be used for all devices in a channel without any insulators.
> > 
> > I assume you are not using source ballast resistors as most designs leave them out. 
> 
> Actually I'd put some in :( figuring that they'd be convenient to measure
> the quiescent current.  Before I fixed my bias generator (using the
> thermal sense diode attached to one of the MOSFETS), the current wasn't
> as stable as I wanted so I bumped these to 0.22ohms.
> 
> >> So far the heat sinks have
> >> metal exposed to the outside world, which would have to change if not
> >> grounded.
> > 
> > 
> > ** Another option is to ground the sources of all the mosfets, via a common heatsink that can serve more than one channel. This requires that the centre point of the PSU be available to serve as the output as in this Haffler schematic:
> > 
> > http://bmamps.com/Schematics/Hafler/Hafler_9300,_9500_Schematic.pdf
> 
> I'd seen this design earlier - cute!  IIRC it requires (one? two?) separate 
> floating power supplies for each channel, unfortunately, which makes the cure
> probably worse than the disease in my situation.
> 
> >> How did you determine the 200 ohms?
> >>  
> > 
> >  ** I didn't, but many other designers have settled on values of 220 ohms to 470 ohms per device for the job. Often N channel devices having a higher value than used for the P channel ones. 
> 
> Certainly the N channel device has less capacitance so the disparity makes
> sense.  I'd sure like to understand the need for these resistors better than
> the WAG about HF oscillations.

I know little about these issues, but would it make any sense to try a big 
ferrite bead?  Maybe with a little R too.    

George H. 
> 
> >> > An output stabilising network is essential, consisting of a inductor of
> >> > about 5uH with RC Zobels on both ends to supply ground.
> >> 
> >> Ouch.  5uH has ~12ohms reactance at 400kHz.  I may need some inductance 
> >> here, but hopefully less.  The usual Zobel impedances are also too low,
> >> consuming far too much of the output at my much-higher-than-audio 
> >> frequencies.
> >> 
> > 
> > ** Yeah, that is a problem. 
> > 
> >  Any chance you could operate the mosfets outside the NFB loop with the gates linked via 220ohm resistors? Lateral mosfets start conducting at 0.4V so there would be a little crossover distortion and the output impedance would be around 1 ohm. 
> > 
> > ....   Phil
> 
> That's something I'm hoping to explore today.  The feedback network is
> a standard pair of parallel RC (identical t.c.s); so I'm going to try
> connecting the output-side resistor to the "real" output, and the 
> capacitor to the output of the voltage gain stage.  The breakpoint is
> around 1MHz, so distortion may be tolerable.

On Fri, 30 Sep 2016 15:53:36 +0000, Frank Miles wrote:

[snip]

> The feedback network is
> a standard pair of parallel RC (identical t.c.s); so I'm going to try
> connecting the output-side resistor to the "real" output, and the 
> capacitor to the output of the voltage gain stage.  The breakpoint is
> around 1MHz, so distortion may be tolerable.

Early testing looks very good!  Stability seems fine with a few different
loads, distortion (at least as viewed on the 'scope) is negligible, and
BW and slew rates are at least twice as fast as I need.  This is with gate
resistors around 240 ohms (no beads, at least not yet).

It would still be nice to have a better understanding of why these
resistors are needed.  For now I'll content myself with a fuller testing
with more reactive loads.  Hopefully that will shake out any weaknesses.

Thanks Phil and everyone else that contributed to the discussion!
   -F

Frank Miles wrote...
> Frank Miles wrote:
>>   [ snip ]
>> The breakpoint is around 1MHz, so distortion may be tolerable.
>
>Early testing looks very good!  Stability seems fine with a few different
>loads, distortion (at least as viewed on the 'scope) is negligible, and
>BW and slew rates are at least twice as fast as I need.  This is with gate
>resistors around 240 ohms (no beads, at least not yet).
>
>It would still be nice to have a better understanding of why these
>resistors are needed.  For now I'll content myself with a fuller testing
>with more reactive loads.  Hopefully that will shake out any weaknesses.
>
>Thanks Phil and everyone else that contributed to the discussion!
>   -F

 I wasn't in the discussion, but would like to point
 out / remind anyone interested, of my 200 W amplifier
 project of last fall.  This has a 1000 V/us slew rate
 and a DC to 10 MHz response (-3dB rolloff frequency).
 But ahem, it doesn't use lateral MOSFETs.


-- 
 Thanks,
    - Win

>Early testing looks very good!  Stability seems fine with a few different
>loads, distortion (at least as viewed on the 'scope) is negligible, and
>BW and slew rates are at least twice as fast as I need.  This is with gate
>resistors around 240 ohms (no beads, at least not yet).
>
>It would still be nice to have a better understanding of why these
>resistors are needed.  For now I'll content myself with a fuller testing
>with more reactive loads.  Hopefully that will shake out any weaknesses.

My recollection/belief is that you're dealing with the equivalent of
the "grid stopper" resistors that were used with many vacuum-tube
amplifier circuits.

MOSFETs, like tubes, have a significant amount of capacitance at their
gates/grids (both gate-to-source, and the Miller capacitance).  The
leads/traces connected to the gates have a non-zero inductance... so
you've got a series-resonant circuit, with the gate right in the
middle of it (maximum-voltage-excursion point).  This can be a recipe
for instability, up to and including parasitic oscillation at the
resonant frequency.  A term I recall from the vacuum-tube days was
"snivets" (these were thin vertical lines on a TV screen, caused by
parasitic oscillation of this general sort in the output tube).

Putting a stopper resistor in series with the gate or grid (ideally,
close to the device) both rolls off the bandwidth (resistor R
interacting with gate C) and kills the Q of the L/C resonant circuit.
It quiets the shrieking and screaming no end and can help keep the
Magic Blue Smoke where it belongs :-)

Anyhow, that's my (possibly-faulty) recollection and analysis.

I had to deal with a somewhat-related problem some years ago,
trouble-shooting a simple twin-audio-tone oscillator designed for
doing IM analysis of single-sideband ham transceivers.  Very simple
circuit, which a guy built based on an article in QST - two twin-T
audio oscillators using 2N2222 transistors.  It did oscillate, but the
frequencies and amplitudes were unstable and it'd misbehave if you
brought your fingers near the circuit.  The owner couldn't figure
out the cause, and gave me the box as a "use this for parts if you
like" gift.

A bit of poking around with a spectrum analyzer showed that this
"audio" oscillator was breaking into RF parasitic oscillation at
upwards of 100 MHz!

Sticking a ferrite bead around the base and collector leads of each
2N2222 killed the Q of the resonances and fixed the problem.  A
resistor of a few tens of ohms in the base would probably have done
just as well.

cassiope wrote:

>
> > I assume you are not using source ballast resistors as most designs
> > leave them out. 
> 
> Actually I'd put some in :( figuring that they'd be convenient to measure
> the quiescent current.  Before I fixed my bias generator (using the
> thermal sense diode attached to one of the MOSFETS), the current wasn't
> as stable as I wanted so I bumped these to 0.22ohms.
>

** Using a diode like that could result in over compensation. 

   Lose the 0.22ohms, especially if they are WW types. 

> > 
> > ** Another option is to ground the sources of all the mosfets, 
> > via a common heatsink that can serve more than one channel.
> > This requires that the centre point of the PSU be available to
> > serve as the output as in this Haffler schematic:
> > 
> > http://bmamps.com/Schematics/Hafler/Hafler_9300,_9500_Schematic.pdf
> 
> I'd seen this design earlier - cute!  IIRC it requires (one? two?) separate 
> floating power supplies for each channel, unfortunately, which makes the cure
> probably worse than the disease in my situation.
>

** The main DC supply needs to be floating but drive can be from an op-amp running on +/-15V rails with centre grounded. Note how there are no load isolating components - not even an RC zobel. 



....   Phil