A quiet forward converter likes only one MOSFET -- why?

Hi,

I have been trying to build a low-power multiple-output converter that 
has no regulation at the secondary side. Its purpose is to power many 
floating MOSFET drivers, so a reasonably too low voltage is not a 
problem, but to high certainly is. Particularly, the experiments 
involved powering a mock-up SiC MOSFET driver with +15/-3.3V rails.
Therefore, the smoothness of the transition from no load to 10mA load
is considered critical.

My experiments involved multiple topologies and transformer designs, but
they all have failed miserably: if the converter was loaded, then the 
regulation was good to very good, but the no-load overshot was a killer:
in the case of a GaN half-bridge the 15V power rail merilly went to 25V 
due to leakage inductance spikes on the secondary.

It turned out that I had been trying to build the converter from way too 
good parts and the reasons behind the spike were not excessively long
MOSFET switching/dead times, but quite the opposite: the secondary 
started ringing due to the high slew rate. So I decided to confirm this
theory by building the simplest possible forward converter: a single 
switch, core reset based on a reset winding + a Schottky diode, fixed 
50% duty cycle @176kHz, no inductor at the output and a half-wave 
rectifier (peak detection is fine), fixed 12V input voltage. Plus the 
magic at the primary side: the MOSFET gate is driven by a current source 
to slow down the switching process. Since Vth is around 3V and the
driving waveform is ~9V, the approximation of a current source by
a 1k variable resistor turned out to be sufficient. Below are the 
results taken from the secondary winding -- I show the edge first,
then I zoom out to show the entire cycle.

With no gate resistance and no load we start with this regular overshot:

https://i.postimg.cc/rp8Qt2py/DS1-Z-Quick-Print52.png

Then the gate current is throttled down:

https://i.postimg.cc/9f8KmQBH/DS1-Z-Quick-Print53.png

And we approach critical damping:

https://i.postimg.cc/jjF3Vkss/DS1-Z-Quick-Print54.png

Zoom out, still no load:

https://i.postimg.cc/Fs6YLysY/DS1-Z-Quick-Print56.png

Heavy overdamping with its nice round edges:

https://i.postimg.cc/q7xBjy0X/DS1-Z-Quick-Print47.png

In the critically damped case at no load the output voltage is 15.4V, 
with 15mA output current (1k resistor) it is 14.84V and with a 
significant 76mA overload (47 Ohm) it still bravely keeps 13.9V.
Of curse no overshots whatsoever. Efficiency is around 84%. This
is a huge success, given the simplicity of the POC converter,
but there are two interesting findings:

1. The converter consumes 15mA@12V when idle. Not bad, but the
LT3439 chip, which has ben designed to address exactly this kind
of tasks, consumes 45mA. In spite of its push-pull nature the
output voltage is not nearly as stiff as the forward's: with the very 
same transformer and a full-bridge rectifier on BAS4002 it outputs
16.3V at no load and 14V when loaded with a 1k resistor. I can see
significant voltage drop at the secondary winding, so it's not the 
rectifier's fault. My circuit is much simpler, but performs much
better, most interesting.

2. This slew-rate limited forward works great, but... only with IRFR825.
I tried a number of other transistors: the high-voltage ones behave more
or less as expected, but as the RDS_ON goes down, the regulation region
gets narrower. E.g. the FCP20N60 with 150mOhm is barely usable. Even 
better transistors are not usable at all. I suspect the Qg gets too big
to fully charge the gate and switching losses dominate. It is possible 
to get decently round edges with them, but at the expense of ~200mA idle
current. Good low-voltage MOSFETs, e.g. SQJA62EP simply don't work at 
all. Interestingly, their Qg is comparable to that of the worse HV 
transistors, but R_DS_ON is *much* smaller. So my theory breaks down here.

Typically for me, a solved practical problem has been replaced by a 
theoretical one. What could possibly make this 500V/1 Ohm IRFR825 MOSFET 
SO special? It is the only part that works PERFECTLY in this applicaton.

	Best regards, Piotr

On 31/12/2019 11:15 am, Piotr Wyderski wrote:
> Hi,
> 
> I have been trying to build a low-power multiple-output converter that 
> has no regulation at the secondary side. Its purpose is to power many 
> floating MOSFET drivers, so a reasonably too low voltage is not a 
> problem, but to high certainly is. Particularly, the experiments 
> involved powering a mock-up SiC MOSFET driver with +15/-3.3V rails.
> Therefore, the smoothness of the transition from no load to 10mA load
> is considered critical.
> 
> My experiments involved multiple topologies and transformer designs, but
> they all have failed miserably: if the converter was loaded, then the 
> regulation was good to very good, but the no-load overshot was a killer:
> in the case of a GaN half-bridge the 15V power rail merilly went to 25V 
> due to leakage inductance spikes on the secondary.
> 
> It turned out that I had been trying to build the converter from way too 
> good parts and the reasons behind the spike were not excessively long
> MOSFET switching/dead times, but quite the opposite: the secondary 
> started ringing due to the high slew rate. So I decided to confirm this
> theory by building the simplest possible forward converter: a single 
> switch, core reset based on a reset winding + a Schottky diode, fixed 
> 50% duty cycle @176kHz, no inductor at the output and a half-wave 
> rectifier (peak detection is fine), fixed 12V input voltage. Plus the 
> magic at the primary side: the MOSFET gate is driven by a current source 
> to slow down the switching process. Since Vth is around 3V and the
> driving waveform is ~9V, the approximation of a current source by
> a 1k variable resistor turned out to be sufficient. Below are the 
> results taken from the secondary winding -- I show the edge first,
> then I zoom out to show the entire cycle.
> 
> With no gate resistance and no load we start with this regular overshot:
> 
> https://i.postimg.cc/rp8Qt2py/DS1-Z-Quick-Print52.png
> 
> Then the gate current is throttled down:
> 
> https://i.postimg.cc/9f8KmQBH/DS1-Z-Quick-Print53.png
> 
> And we approach critical damping:
> 
> https://i.postimg.cc/jjF3Vkss/DS1-Z-Quick-Print54.png
> 
> Zoom out, still no load:
> 
> https://i.postimg.cc/Fs6YLysY/DS1-Z-Quick-Print56.png
> 
> Heavy overdamping with its nice round edges:
> 
> https://i.postimg.cc/q7xBjy0X/DS1-Z-Quick-Print47.png
> 
> In the critically damped case at no load the output voltage is 15.4V, 
> with 15mA output current (1k resistor) it is 14.84V and with a 
> significant 76mA overload (47 Ohm) it still bravely keeps 13.9V.
> Of curse no overshots whatsoever. Efficiency is around 84%. This
> is a huge success, given the simplicity of the POC converter,
> but there are two interesting findings:
> 
> 1. The converter consumes 15mA@12V when idle. Not bad, but the
> LT3439 chip, which has ben designed to address exactly this kind
> of tasks, consumes 45mA. In spite of its push-pull nature the
> output voltage is not nearly as stiff as the forward's: with the very 
> same transformer and a full-bridge rectifier on BAS4002 it outputs
> 16.3V at no load and 14V when loaded with a 1k resistor. I can see
> significant voltage drop at the secondary winding, so it's not the 
> rectifier's fault. My circuit is much simpler, but performs much
> better, most interesting.
> 
> 2. This slew-rate limited forward works great, but... only with IRFR825.
> I tried a number of other transistors: the high-voltage ones behave more
> or less as expected, but as the RDS_ON goes down, the regulation region
> gets narrower. E.g. the FCP20N60 with 150mOhm is barely usable. Even 
> better transistors are not usable at all. I suspect the Qg gets too big
> to fully charge the gate and switching losses dominate. It is possible 
> to get decently round edges with them, but at the expense of ~200mA idle
> current. Good low-voltage MOSFETs, e.g. SQJA62EP simply don't work at 
> all. Interestingly, their Qg is comparable to that of the worse HV 
> transistors, but R_DS_ON is *much* smaller. So my theory breaks down here.
> 
> Typically for me, a solved practical problem has been replaced by a 
> theoretical one. What could possibly make this 500V/1 Ohm IRFR825 MOSFET 
> SO special? It is the only part that works PERFECTLY in this applicaton.
> 
>      Best regards, Piotr

Thanks for sharing this Piotr.

May I be heretical and suggest considering a bipolar? One of the Zetex 
low Vcesat offerings (designed as mosfet rivals) like ZXT849 or FZT853 
might also perform well. When quietness is more important than 
efficiency they can be helpful.

piglet

On Tue, 31 Dec 2019 12:15:26 +0100, Piotr Wyderski
<peter.pan@neverland.mil> wrote:

>Hi,
>
>I have been trying to build a low-power multiple-output converter that 
>has no regulation at the secondary side. Its purpose is to power many 
>floating MOSFET drivers, so a reasonably too low voltage is not a 
>problem, but to high certainly is. Particularly, the experiments 
>involved powering a mock-up SiC MOSFET driver with +15/-3.3V rails.
>Therefore, the smoothness of the transition from no load to 10mA load
>is considered critical.
>
>My experiments involved multiple topologies and transformer designs, but
>they all have failed miserably: if the converter was loaded, then the 
>regulation was good to very good, but the no-load overshot was a killer:
>in the case of a GaN half-bridge the 15V power rail merilly went to 25V 
>due to leakage inductance spikes on the secondary.
>
>It turned out that I had been trying to build the converter from way too 
>good parts and the reasons behind the spike were not excessively long
>MOSFET switching/dead times, but quite the opposite: the secondary 
>started ringing due to the high slew rate. So I decided to confirm this
>theory by building the simplest possible forward converter: a single 
>switch, core reset based on a reset winding + a Schottky diode, fixed 
>50% duty cycle @176kHz, no inductor at the output and a half-wave 
>rectifier (peak detection is fine), fixed 12V input voltage. Plus the 
>magic at the primary side: the MOSFET gate is driven by a current source 
>to slow down the switching process. Since Vth is around 3V and the
>driving waveform is ~9V, the approximation of a current source by
>a 1k variable resistor turned out to be sufficient. Below are the 
>results taken from the secondary winding -- I show the edge first,
>then I zoom out to show the entire cycle.
>
>With no gate resistance and no load we start with this regular overshot:
>
>https://i.postimg.cc/rp8Qt2py/DS1-Z-Quick-Print52.png
>
>Then the gate current is throttled down:
>
>https://i.postimg.cc/9f8KmQBH/DS1-Z-Quick-Print53.png
>
>And we approach critical damping:
>
>https://i.postimg.cc/jjF3Vkss/DS1-Z-Quick-Print54.png
>
>Zoom out, still no load:
>
>https://i.postimg.cc/Fs6YLysY/DS1-Z-Quick-Print56.png
>
>Heavy overdamping with its nice round edges:
>
>https://i.postimg.cc/q7xBjy0X/DS1-Z-Quick-Print47.png
>
>In the critically damped case at no load the output voltage is 15.4V, 
>with 15mA output current (1k resistor) it is 14.84V and with a 
>significant 76mA overload (47 Ohm) it still bravely keeps 13.9V.
>Of curse no overshots whatsoever. Efficiency is around 84%. This
>is a huge success, given the simplicity of the POC converter,
>but there are two interesting findings:
>
>1. The converter consumes 15mA@12V when idle. Not bad, but the
>LT3439 chip, which has ben designed to address exactly this kind
>of tasks, consumes 45mA. In spite of its push-pull nature the
>output voltage is not nearly as stiff as the forward's: with the very 
>same transformer and a full-bridge rectifier on BAS4002 it outputs
>16.3V at no load and 14V when loaded with a 1k resistor. I can see
>significant voltage drop at the secondary winding, so it's not the 
>rectifier's fault. My circuit is much simpler, but performs much
>better, most interesting.
>
>2. This slew-rate limited forward works great, but... only with IRFR825.
>I tried a number of other transistors: the high-voltage ones behave more
>or less as expected, but as the RDS_ON goes down, the regulation region
>gets narrower. E.g. the FCP20N60 with 150mOhm is barely usable. Even 
>better transistors are not usable at all. I suspect the Qg gets too big
>to fully charge the gate and switching losses dominate. It is possible 
>to get decently round edges with them, but at the expense of ~200mA idle
>current. Good low-voltage MOSFETs, e.g. SQJA62EP simply don't work at 
>all. Interestingly, their Qg is comparable to that of the worse HV 
>transistors, but R_DS_ON is *much* smaller. So my theory breaks down here.
>
>Typically for me, a solved practical problem has been replaced by a 
>theoretical one. What could possibly make this 500V/1 Ohm IRFR825 MOSFET 
>SO special? It is the only part that works PERFECTLY in this applicaton.
>
>	Best regards, Piotr

Can you post the schematic? What transformer are you using?

I've blown out Cree SiC fets going just a bit over abs max gate
voltage. They are serious about that spec.

The low-load increase is probably from leakage inductance ringing in
the transformer. That shouldn't be much energy if you keep the
frequency low. Personally, I'd put a 15 volt zener on the isolated
side, rather than damping the primary. 



-- 

John Larkin         Highland Technology, Inc

lunatic fringe electronics 

jlarkin@highlandsniptechnology.com wrote...
>
> I've blown out Cree SiC fets going just a bit over
> abs max gate voltage. They are serious about that spec.

 I have found some of their parts work well at
 4V below the suggested gate voltage.


-- 
 Thanks,
    - Win

jlarkin@highlandsniptechnology.com wrote:

> Can you post the schematic?

In a bit of a hurry to celebrate, so here's the CAD file:

https://i.postimg.cc/htPWH8v2/quiet-forward.png

> What transformer are you using?

A hand-wound 20mm toroid made of F938 with one instance of the secondary 
rails. Wound with some random wires of different colour for easy 
prototyping, hence no isolation. The production unit will use a proper 
TIW wire.

> I've blown out Cree SiC fets going just a bit over abs max gate
> voltage. They are serious about that spec.

Some more measurements: the negative rail is -3.54V when unloaded and
-3.46V when the 15V rail is loaded with 1k. No ringing at all. This is 
an outstanding achievment, given the simplicity of the circuit.

> The low-load increase is probably from leakage inductance ringing in
> the transformer.

Sure, but even that is enough to kill your precious SiC parts.

> Personally, I'd put a 15 volt zener on the isolated
> side, rather than damping the primary.

That would work too, but since it is the secondary that is ringing,
every clone of it would need a zener. That makes 20 more parts, while
on the primary side it is just a single gate resistor.

Some more discoveries:

1. The inability to use any other MOSFET that IRFT825 is not related to 
much lower Cgd of the low voltage parts: an SQJA62 with 680pF between 
its drain and gate is similarly usless, but in a different setting of
the trimpot.

2. Ditto for the series channel resistance. The SQJA62 with 1 Ohm in the 
source (good for current sensing, anyway) behaves exactly the same way.

No idea what physics is behind that and why the IRFR825 is just 
perfectly operational and predictable over the entire trimpot range. The 
resulting unit just rocks, but I would like to understand why -- this is 
a sci group after all. :-)

Some MOSFET god has smiled upon me, because it was the first suitable 
MOSFET that I found. If I started with any other part, the entire idea 
would get canned.

And finally, Happy New Year to all of you!

	Best regards, Piotr

On 31/12/2019 5:13 pm, Piotr Wyderski wrote:
> jlarkin@highlandsniptechnology.com wrote:
> 
>> Can you post the schematic?
> 
> In a bit of a hurry to celebrate, so here's the CAD file:
> 
> https://i.postimg.cc/htPWH8v2/quiet-forward.png
> 
>> What transformer are you using?
> 
> A hand-wound 20mm toroid made of F938 with one instance of the secondary 
> rails. Wound with some random wires of different colour for easy 
> prototyping, hence no isolation. The production unit will use a proper 
> TIW wire.
> 
>> I've blown out Cree SiC fets going just a bit over abs max gate
>> voltage. They are serious about that spec.
> 
> Some more measurements: the negative rail is -3.54V when unloaded and
> -3.46V when the 15V rail is loaded with 1k. No ringing at all. This is 
> an outstanding achievment, given the simplicity of the circuit.
> 
>> The low-load increase is probably from leakage inductance ringing in
>> the transformer.
> 
> Sure, but even that is enough to kill your precious SiC parts.
> 
>> Personally, I'd put a 15 volt zener on the isolated
>> side, rather than damping the primary.
> 
> That would work too, but since it is the secondary that is ringing,
> every clone of it would need a zener. That makes 20 more parts, while
> on the primary side it is just a single gate resistor.
> 
> Some more discoveries:
> 
> 1. The inability to use any other MOSFET that IRFT825 is not related to 
> much lower Cgd of the low voltage parts: an SQJA62 with 680pF between 
> its drain and gate is similarly usless, but in a different setting of
> the trimpot.
> 
> 2. Ditto for the series channel resistance. The SQJA62 with 1 Ohm in the 
> source (good for current sensing, anyway) behaves exactly the same way.
> 
> No idea what physics is behind that and why the IRFR825 is just 
> perfectly operational and predictable over the entire trimpot range. The 
> resulting unit just rocks, but I would like to understand why -- this is 
> a sci group after all. :-)
> 
> Some MOSFET god has smiled upon me, because it was the first suitable 
> MOSFET that I found. If I started with any other part, the entire idea 
> would get canned.
> 
> And finally, Happy New Year to all of you!
> 
>      Best regards, Piotr

Just wondering if moving the catch diode to the non-dot end of the reset 
winding changes anything - ringing due to interwinding stray capacity etc?

Could you save the dual secondary and instead of making +15V and -3.5V 
just make 18.5V and syntheise zero with a resistor and 15V zener (or 
3.6V zener) then you'd also get a preload on the converter?

Happy New Year!

piglet

piglet wrote:

> Thanks for sharing this Piotr.

The pleasure is on my side. :-)

> May I be heretical and suggest considering a bipolar?

This circuit is heretical enough to make more heresies go unnoticed. :-)

I have only ZTX851. No luck, the part is too slow, total disaster.

	Best regards, Piotr

Why not a push pull 50% duty?

That would make you drive all the transformers in parallel and is dead simple 

Happy new year 

On Tue, 31 Dec 2019 18:13:06 +0100, Piotr Wyderski
<peter.pan@neverland.mil> wrote:

>jlarkin@highlandsniptechnology.com wrote:
>
>> Can you post the schematic?
>
>In a bit of a hurry to celebrate, so here's the CAD file:
>
>https://i.postimg.cc/htPWH8v2/quiet-forward.png
>
>> What transformer are you using?
>
>A hand-wound 20mm toroid made of F938 with one instance of the secondary 
>rails. Wound with some random wires of different colour for easy 
>prototyping, hence no isolation. The production unit will use a proper 
>TIW wire.
>
>> I've blown out Cree SiC fets going just a bit over abs max gate
>> voltage. They are serious about that spec.
>
>Some more measurements: the negative rail is -3.54V when unloaded and
>-3.46V when the 15V rail is loaded with 1k. No ringing at all. This is 
>an outstanding achievment, given the simplicity of the circuit.
>
>> The low-load increase is probably from leakage inductance ringing in
>> the transformer.
>
>Sure, but even that is enough to kill your precious SiC parts.
>
>> Personally, I'd put a 15 volt zener on the isolated
>> side, rather than damping the primary.
>
>That would work too, but since it is the secondary that is ringing,
>every clone of it would need a zener. That makes 20 more parts, while
>on the primary side it is just a single gate resistor.
>
>Some more discoveries:
>
>1. The inability to use any other MOSFET that IRFT825 is not related to 
>much lower Cgd of the low voltage parts: an SQJA62 with 680pF between 
>its drain and gate is similarly usless, but in a different setting of
>the trimpot.
>
>2. Ditto for the series channel resistance. The SQJA62 with 1 Ohm in the 
>source (good for current sensing, anyway) behaves exactly the same way.
>
>No idea what physics is behind that and why the IRFR825 is just 
>perfectly operational and predictable over the entire trimpot range. The 
>resulting unit just rocks, but I would like to understand why -- this is 
>a sci group after all. :-)
>
>Some MOSFET god has smiled upon me, because it was the first suitable 
>MOSFET that I found. If I started with any other part, the entire idea 
>would get canned.
>
>And finally, Happy New Year to all of you!
>
>	Best regards, Piotr

Post the schematic. 

-- 

John Larkin         Highland Technology, Inc
picosecond timing   precision measurement 

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

On 2019-12-31 12:13, Piotr Wyderski wrote:
> jlarkin@highlandsniptechnology.com wrote:
> 
>> Can you post the schematic?
> 
> In a bit of a hurry to celebrate, so here's the CAD file:
> 
> https://i.postimg.cc/htPWH8v2/quiet-forward.png
> 
>> What transformer are you using?
> 
> A hand-wound 20mm toroid made of F938 with one instance of the secondary 
> rails. Wound with some random wires of different colour for easy 
> prototyping, hence no isolation. The production unit will use a proper 
> TIW wire.
> 
>> I've blown out Cree SiC fets going just a bit over abs max gate
>> voltage. They are serious about that spec.
> 
> Some more measurements: the negative rail is -3.54V when unloaded and
> -3.46V when the 15V rail is loaded with 1k. No ringing at all. This is 
> an outstanding achievment, given the simplicity of the circuit.

Provided you can do it twice!

Happy New Year

Phil Hobbs


-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com