OverUnity Research

While looking for a way to switch 3.5kV DC (via a MOSFET or IGBT) for another project, i found this youtube video from Jeg:

https://www.youtube.com/watch?v=manMnNOFdNI

He shows a board with 3 switching circuits each equipped with 2 MOSFET's in series able to switch some kV's, see picture.
But looking at the video alone i can not quite make up how his circuit is working, so by opening this thread, i hope Jeg (a member here) will be available to explain how his circuit is laid out.

I see / hear that there are 2 MOSFETs (1kV each) in series (total 6 MOSFET's) and there are 3 TC4420 MOSFET drivers (each driving 2 MOSFET's).
The gates of the MOSFET's are isolated by using some toroidal coils, but the setup is not clear to me.

First i think i need a diagram on how the circuit is laid out, especially how the toroidal coils are build / connected.

Secondly, do i understand correctly that each of the 2 series MOSFET's is capable of switching 2kV and that several of these 2 MOSFET's circuits can be put in series to handle 2, 4 or even 6kV?

Regards Itsu 

The circuit for which i need this HV switcher is looking like this:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47537;image)


Itsu

Hi Itsu,

I think Jeg used two + two MOSFETs  instead of the usual one + one  to build a normal push pull output circuit.
I think he connected the two MOSFETs as a bidirectional switch you surely heard about such, see the principle attached. I copied it from the bottom of this page: https://electronics.stackexchange.com/questions/175917/substrate-connection-when-driving-a-bi-directional-switch (https://electronics.stackexchange.com/questions/175917/substrate-connection-when-driving-a-bi-directional-switch)

In Jeg's circuit I think any one of the toroidal transformers drives two MOSFETs connected in bidirectional fashion.  Isolation is needed for driving bidirectional switches because the common source pins can be at the AC (or DC) potential that is switched by the drain pins.  Each transformer drives the common gate and common source pins directly I suppose but there is this possibility shown here:
https://www.analog.com/en/technical-articles/transformerdriver-ic-controls-bidirectional-switch.html (https://www.analog.com/en/technical-articles/transformerdriver-ic-controls-bidirectional-switch.html)  but this latter circuit may not give a fast switch-off time.

Later tomorrow I return to your HV switcher with a suggestion.

Gyula

It will be better probably.
It has taken from apnote IR
But this was working bad to me.

Thanks guys,

so Gyula, you say that he uses 2 MOSFET pairs (each with 2 MOSFETs in series) to build this Push Pull configuration.

But what about this 3th MOSFET pair?


I understood that he can use that too to increase the voltage handling capacity of his circuit.


I will try to alert Jeg to take a look here.

Regards itsu

Hi Itsu,

I cannot see the 3rd MOSFET pair on the left side of Jeg's circuit board, it is not applied. 
And the IC socket (for the MOSFET driver IC) under the 3rd toroidal tarnsformer on the left side is also empty.

Gyula

Gyula, 

i agree, but in an internal conversation with Jeg, he mentioned:

"If all 3 switches are connected in series at the output then it can handle up to 6kV.    (1kV MOSFET's)
If 1.2kV MOSFET's are used then it can reach 7.8kV"    (I guess he meant 7.2kV.)

Waiting for Jeg to confirm.

Itsu

Hi Itsu,


Okay, it is an 'IF' possibility for all the 3 switches in series.  But not in the circuit shown in the video because he spoke about push pull circuit in the video and shown 4 MOSFETs in use, I attached a schema as I think he used them those 2 pairs in push pull as he may have showed in his video.  His 24 V battery input may include a choke coil seen at the top left side of his circuit board (looks like a tv yoke core too)  Of course the drive pulses should be 180 degree out of phase going to the gate - source inputs of the push pull stage, probably the phase invertion can be done by the drive transformers too.

I attached a 2nd schema I suggest to use for your 3.5 kV switching circuit, utilizing 3 pairs of MOSFETs connected all in series as per bidirectional fashion within each pair.  I took the idea from this paper
http://bromine.cchem.berkeley.edu/grppub/frbm2.pdf (http://bromine.cchem.berkeley.edu/grppub/frbm2.pdf)  The paper includes how the drive transformer is made.

I see one problem : if the input DC supply is indeed as high as 3.5 kV, then the flyback pulse after the switch-off moment may be much higher than 3.5 kV, so protection is needed. The MOV devices may be one solution. But it may be better to utilize the flyback pulses by redirecting to the output transformer primary with HV diodes (like a mechanical relay coil is clamped by a parallel diode). 

Gyula

Gyula,

so you think Jeg is switching 24V which then is being transformed up to HV?     (i think you are right about that).

This is opposite to what i want which is HV being switched so it can be transformed down to low voltage.

It makes sense to use 1kV MOSFET's as, like you mentioned lateron, the flyback pulse will be high.


Thanks for your 2nd diagram, thats about how i envisioned it, but the flyback pulse will be massive so some countermeasures will need to be taken to protect the components like you suggested.


Itsu

Hi Itsu, Gyula, all
This switching board is a generic board which i use in almost all of my experiments between about 3 Khz- 1.8 Mhz. The idea was to build a tough switch able to handle many KV. The board contains three independent switches. Each switch utilize two mosfets in series so to increase the breakdown voltage. Recently i modified it and i used lm7815 to drive tc4420 instead of 12V i had been using. This gave me more frequency range without changing turns at the gate transformer. About the isolation tx coils, i need to count their turns because i cant find any notes. The 1 uf caps especially those at the primary side need to be of a low esr or few caps in parallel with the total of 1uf.
I drive this board from my frequency generator.

If i drive all three switches with the same signal i can connect all six mosfets is series.

Hi Jeg,

Thanks for chiming in.  O0    So you do not use the bidirectional MOSFET pair as I figured but you use them in normal series connection, thanks for clarifying.  In the meantime I found an old thread here
https://overunity.com/13995/mosfet-stack-for-higher-breakdown-voltage/ (https://overunity.com/13995/mosfet-stack-for-higher-breakdown-voltage/)   you surely recall to have done the pioneering work on stacking.

EDIT:  How you meant the push pull output?  one such seriesly connected pair on one side and another seriesly connected pair on the other side, driving them out of phase?

Gyula

Jeg wanted to publish some pictures, but he says he cannot upload them here, so i will put them up here for him:


Itsu

Quote from: gyula on 2023.03.20, 17:32:27

EDIT:  How you meant the push pull output?  one such seriesly connected pair on one side and another seriesly connected pair on the other side, driving them out of phase?

Gyula

Hi Gyula. Exactly that.

If you go a little more back in time, before some nine years, you were the first who had taught me how to make a decent mosfet driver. :-)

Thank you itsu. It was more for you to see the layout as an Idea. There is a smaller board as you see at the input which is a filter for power supply decoupling.

Jeg,

OK, i see.

These blue gate isolation toroids, are these coming from PC PS's, or did you buy them separately?

Itsu 

Separately. I had chosen them according the frequency range they work with. I hope i ll find any relevant notes on this.

Ps. Thankfully i found their type. Check this link
https://product.tdk.com/en/search/ferrite/ferrite/ferrite-core/info?part_no=B64290L0618X087

Thanks Jeg,

so they are: "Ferrite (N87) core materials are an oxide made from Fe (iron), Mn (manganese), and Zn (zinc), which are commonly referred to as manganese zinc ferrites" effective up to 3MHz:
https://www.netl.doe.gov/sites/default/files/netl-file/Core-Loss-Datasheet---MnZn-Ferrite---N87%5B1%5D.pdf

Itsu

Itsu i checked the transformer and it is a 27 turns  trifiliar. So we are talking about a one to one ratio. I ll try to spot a pdf i had which provides a table with inductance vs turns vs frequency range for gate isolation transformers.

https://www.google.com/url?sa=t&source=web&rct=j&url=https://ferroxcube.home.pl/news/gate%2520drive%2520trafo.pdf&ved=2ahUKEwj5yqCm-ez9AhU9VvEDHWX2BagQFnoECCkQAQ&usg=AOvVaw2ofOU16NR9USdC4L7YLEAS

Here are some useful info

Guys,

i think i will settle for a less complex setup that Jeg has and use a mix of Jeg his setup and Gyula like this:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47563;image)

I need to find the correct flyback diodes across the transformer coil as there will be serious voltage to be diverted.

Itsu 

Hi Itsu,  the schema looks good.  Yes, fast HV diodes are needed to clamp flyback spikes.
The gate driver transformer's insulation needs attention from material quality point of view.

EDIT: I wonder if the primary of the drive transformer needs a flyback clamp diode too?   C.C  (to protect the driver IC when it is just off)  Maybe an overkill...

EDIT 2: well, no need for a diode across the input of the drive transformer because the IXDI614 shunts it right after the switch off moment (shunting resistance is around 1 Ohm), 

Gyula

OK Gyula,

I have used an isolated gate transformer before when building a SSTC 2:   https://www.loneoceans.com/labs/sstc2/
It uses the UCC27425 gate driver and gate isolation transformer like shown in the attached diagram.

I used magnet wire for the transformer wire which is rated 3000V for its isolation, see the picture below.

By the way, i changed the gate driver in my above drawing from an IXDI614PI (inverting) to an IXDD614PI (non-inverting) as that is what i have and "inverting" is not needed.


Thanks,   Itsu

Quote from: Itsu on 2023.03.21, 21:35:18
Guys,

i think i will settle for a less complex setup that Jeg has and use a mix of Jeg his setup and Gyula like this:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47563;image)

I need to find the correct flyback diodes across the transformer coil as there will be serious voltage to be diverted.

Itsu

Itsu,

IMO, the schematic as shown will not successfully switch 3000v.  If you consider a positive pulse on the secondary of the gate drive transformer, both gates will be "turned on" so to speak, but the bottom mosfet will conduct through the substrate diode leaving only the top mosfet rated at 1700 Vds to hold off the 3000 volts applied to the HV transformer primary.  Avalanche of the top mosfet will certainly occur.  This type of mosfet topology will work for switching bipolar waveforms but only to a 1700v drain to source maximum.

The mosfets shown must be connected in series as previously shown by Jeg.

Regards,
Pm

Partzman,

thanks, i tend to agree with you about the bottom MOSFET "conduct through the substrate diode".

So i have redrawn the diagram as i understand you about "The mosfets shown must be connected in series as previously shown by Jeg."

Itsu

Hi Partzman,

I would like to understand the process you mention.  I assume you mean the body diode on the substrate diode, or is it a different one? I find the term instrictic diode also for body diode. Is the substrate diode not the body or instrictic diode?

I see the operation of such bidirectional switch as follows: When both mosfets are turned on,  both will conduct via their own drain - source pins. Both of their body diodes would actually be shorted by the conducting drain source channels during the switched-on moments.  The control pulse appearing across the common gate and common source pins are ground independent pulses. 
In solid state relays, power mosfets are also used in this bidirectional connection,  and AC switching is also involved for them. 

Thanks,
Gyula

Quote from: gyula on 2023.03.22, 16:37:06
Hi Partzman,

I would like to understand the process you mention.  I assume you mean the body diode on the substrate diode, or is it a different one? I find the term instrictic diode also for body diode. Is the substrate diode not the body or instrictic diode?

Basically, it is the diode that is normally created by the various fabrication processes used in the enhancement mode mosfets and has been called many names in the past.  In an n-channel mosfet as you know, the cathode is connected to the drain and the anode is connected to the source.

Quote
I see the operation of such bidirectional switch as follows: When both mosfets are turned on,  both will conduct via their own drain - source pins. Both of their body diodes would actually be shorted by the conducting drain source channels during the switched-on moments.  The control pulse appearing across the common gate and common source pins are ground independent pulses. 
In solid state relays, power mosfets are also used in this bidirectional connection,  and AC switching is also involved for them.

To clarify the bidirectional mode, only one of the mosfets is normally turned on and that would be the one with the drain more positive than the source.  The other will have the opposite polarity.  What is not well known is that the reversed biased mosfet is capable of reverse conduction. IOW, with the gate having a positive voltage, the mosfet will conduct even with the drain more negative than the source.  In this mode, the voltage across the drain-source will be less than the normal substrate diode voltage drop due to reverse conduction.  Not all mosfets will work in this mode but most seem to.

Regards,
Pm

[/quote]
Thanks,
Gyula
[/quote]

Quote from: Itsu on 2023.03.22, 15:57:52
Partzman,

thanks, i tend to agree with you about the bottom MOSFET "conduct through the substrate diode".

So i have redrawn the diagram as i understand you about "The mosfets shown must be connected in series as previously shown by Jeg."

Itsu

Itsu,

Yes, that's better!

Pm

Hi Partzman,

Okay, thanks. So the diode you meant is the body diode found between the drain and source pins but with reversed polarity.

I will look for more pieces of information  on the bidirectional configuration. 
Once switched on, the drain source channel of a insulated gate power mosfet is able to conduct current in both directions, you mentioned  this as reverse conduction and I agree.
This is known and this feature is what the bidirectional configuration is based on. 

Anyway, thanks for the comments.   O0


Gyula

Hi,

If you're using SiC MOSFETs then you might want to consider using a negative voltage (e.g. -5V) to turn off the gate.

https://eepower.com/technical-articles/how-sic-mosfets-are-made-and-how-they-work-best
https://www.wolfspeed.com/knowledge-center/article/gate-drives-and-gate-driving-with-sic-mosfets

Best regards,

Lee

Hi Lee,

thanks, indeed, my ordered MOSFET specs say:  VGSop Gate - Source Voltage -5/+20 V, so it could be used to switch it off via a negative voltage.
The 2nd link you provide shows how to do that  O0

Itsu

Hi Itsu,

The Cree SiC MOSFETs are a nice piece of equipment. Did you get the C2M0045170D (45 mΩ RDS ON) or C2M1000170D (1000 mΩ)?

I'm using the Microchip MSC035SMA170B4 (35 mΩ, 7ns rise, 17ns fall) and Infineon 1ED3124MU12H (14 A, 5.7 kV single-channel isolated gate driver) gate driver in my MOSFET switching boards.

I chose the Microchip MOSFET because the price was very good (around $35 USD (https://octopart.com/search?q=MSC035SMA170B4&currency=USD&specs=0)), and the specs. are really nice too.

I'm nearly finished soldering the latest revision of my MOSFET driver PCB. I've done 7 boards so far.

(https://i.ibb.co/DtGLMQT/IMG-4531.jpg) (https://ibb.co/DtGLMQT)

(https://i.ibb.co/6vVNV1c/IMG-4532.jpg) (https://ibb.co/6vVNV1c)

Best regards,

Lee

Thanks Lee,

looking very nice, i am impressed, are you producing these boards to sell, or are you working on a project?

I ordered the C2M0045170D (45 mΩ RDS ON) because of that low RDSon, but the MSC035SMA170B4 looks very nice too.

So you use DC2DC separation and isolated gate driver to isolate in from out.

Itsu

Yeah, it uses an isolated DC/DC converter powering an isolated gate driver. I use the Recom R12P22005D (https://recom-power.com/en/products/dc-dc-converters/dc-dc-unregulated/rec-p-R12P22005D.html?9) DC/DC converter.

I'm just using them for my own project(s). I figured I needed a reliable & robust high voltage solid state switch. I researched it and put together the PCB design and used JLCPCB to manufacture a batch. Finding the best parts to use and how they fit together was the most time consuming part. I've lost count of the number of hours I've trawled through datasheets and application notes. I hadn't done any PCB work before this, so it was quite a steep learning curve!

Here's the schematic:

(https://i.ibb.co/Mpw0ctY/Schematic-Magnetic-Energy-Generator-2023-03-23.png) (https://ibb.co/Mpw0ctY)

OK Lee,

I know how important it is to have a reliable switch for several projects, i myself also use some, based on the Quadratron for which Chris Sykes made some PCBs available here:
https://www.aboveunity.com/thread/reliable-and-flexible-switching-system/

It's somewhat outdated now (the components), but works fine, but yours look more "state of the art".

I worked with KiCad for designing PCB's, so i know how steep the learning curve can be, but the results are awesome.

Any chance you would publish the Gerber files of your PCB, so people can order them at JLCPCB or other PCB manufacturer?
It would have to include a BOM (Bill of Material) so people can optimally benefit from all your hard work  :P

Regards Itsu

Hi Itsu,

I saw Chris's post describing his robust switching solution. It looks like he put a lot of effort into designing and building that, and I'm sure many people are very greatful that he provided such detailed instructions on how to build it themselves. I think it is a good switching solution, but as you say it is no longer state of the art and I wanted to use SiC MOSFETs so I took the (deep) plunge and designed one myself.

I've attached all of the files you need to manufacture one of these for yourself (or anyone else for that matter!). I have included the BOM & Gerber files, along with the EasyEDA (https://easyeda.com) JSON file that you can import so that you can customize it if wanted. Feel free to do whatever you want with it.

I hope that others can benefit from this. I'm not in this for the money, so I'm happy to share the fruits of my labour or detailed findings from my research. I simply want to help move the world forward in whatever way I can. Sharing is caring!

Corrections: I just had a glance at the BOM and noticed that I included the Murata MGJ2D122005SC as the DC/DC converter, but I'm actually using the Recom R12P22005D. They are pin compatible and interchangeable. I don't think there is much difference between them. I opted for the Recom because the Murata was out of stock.

Best regards,

Lee

One more small change. The heatsink footprint for the C40-058-VE is incorrect. It was a community contributed footprint, so I guess they got the measurements wrong. The Ohmite C247-050-2VE (https://www.digikey.com/en/products/detail/ohmite/C247-050-2VE/2627381) fits nicely, although you need to knock the feet pins in slightly.

Thanks a million Lee.

My KiCad is not able to import the Json file, so perhaps i need to install EasyEDA myself.
I am sure many will be grateful for this switching unit as it can serve many purposes,


I have uploaded the Gerber.zip file to JLCPCB, and it accepted it as OK, so i will order the minimum 5 PCB's.

It will take some time to order the parts, but with your BOM and corrections it will be no problem.

Itsu

Awesome job Lee thanks for sharing. What is the frequency range please that you can achieve with that mosfet driver?

Unfortunately there is no easy way to import from EasyEDA into KiCad. There is a 3rd party conversion tool here (https://wokwi.com/tools/easyeda2kicad), but your mileage may vary.

I want to move away from using EasyEDA and start using KiCad because it seems like the better option. EasyEDA was easy (!) to get started with, but now that I know what to do I think it should be straightforward to move over to KiCad for future work. I'll probably migrate my MOSFET switching PCB to KiCad as my first project. Just need to find some time (that is not better spent doing other things)!

I'm using a PWM module (https://www.aliexpress.com/item/1005003560974429.html?spm=a2g0o.productlist.main.1.23d25970PylwOR&algo_pvid=7f8885d4-7521-4b62-b9b5-16fc3dc1e765&algo_exp_id=7f8885d4-7521-4b62-b9b5-16fc3dc1e765-0&pdp_ext_f=%7B%22sku_id%22%3A%2212000026289800876%22%7D&pdp_npi=3%40dis%21GBP%211.6%211.5%21%21%21%21%21%40211be3cd16796681775971342d06e6%2112000026289800876%21sea%21UK%211989806021&curPageLogUid=RbWsRLYD5Shd) to test the boards.

(https://i.ibb.co/F0nH0LT/R.jpg) (https://ibb.co/F0nH0LT)

I've then got an incandescent bulb wired up to a 12V supply, and by adjusting the PWM frequency I can see the bulb flashing accordingly.

(https://i.ibb.co/VSqdYQV/IMG-4539.jpg) (https://ibb.co/VSqdYQV)

Quote from: Jeg on 2023.03.24, 14:42:30
Awesome job Lee thanks for sharing. What is the frequency range please that you can achieve with that mosfet driver?

No worries, glad to help in any way I can :)

The datasheet for the driver (https://www.infineon.com/dgdl/Infineon-1ED31xxMU12H-DataSheet-v02_01-EN.pdf?fileId=5546d46274cf54d50174d97c37be1f67) says 1MHz max switching speed, but I've been able to go up to 5MHz. I've put small heatsinks on the driver chips and I've added some fans (there are 3x 12V fan headers on the board) to help with cooling, because they get quite hot at that speed.

Soldering the SOIC driver chip is a little fiddly if you're not used to soldering such small components. My advice is to apply plenty of flux to the pads and then use a small tipped soldering iron, apply a small amount of solder directly to the soldering iron tip and then touch each one of the pins in turn. The flux will encourage the solder to stick to the right places.

You might also want to apply some more solder to the main traces for the drain & source outputs if you are going to switch large currents.

I do it this way: arrange package in place then solder two opposite legs then change soldering iron tip to flat one , take a nice amount of solder on it and smear on all soic legs on one side then another side.

Quote from: forest on 2023.03.24, 16:31:56
I do it this way: arrange package in place then solder two opposite legs then change soldering iron tip to flat one , take a nice amount of solder on it and smear on all soic legs on one side then another side.

This is the way. It's the technique that I haven't yet perfected :) When I try I end up shorting the pins, then I've got to desolder it and start again.

I guess I'll have to watch some more YouTube tutorials and practice.

Quote from: lfarrand

but I've been able to go up to 5MHz..
/quote]

Amazing range Lee really very impressive!!

Someone might find this useful when looking for a SiC MOSFET. Here's a table to compare the various SiC MOSFETs available from both Microchip and Wolfspeed (Cree), sorted by price ascending.

(https://i.ibb.co/m65FSnd/Screenshot-2023-03-24-at-19-56-54.png) (https://ibb.co/HnTpXcR)

Quote from: Itsu on 2023.03.19, 21:25:45
The circuit for which i need this HV switcher is looking like this:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47537;image)


Itsu

Direct selection of high-voltage MOSFETs,
Simpler and more reliable than using in series!
Even if it is connected in parallel due to the increased current,
It is also simpler and more reliable than tandem.

I wish you every success.

pan

Hi Pan,

i agree, but these very HV MOSFETs have some drawbacks like the price and the high RDSon (20 Ohm), not sure if that last one is a problem in my setup.

Itsu

While waiting for parts, i decided to throw together a quick breadboard setup of the both discussed setups meaning Jeg his setup see this:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47583;image)

and the Bidirectional setup proposed by Gyula here:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47585;image)

The breadboard layout looks like this (here Jeg his setup is shown):

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47593;image)

So i use some low voltage IRF540 MOSFETs for now and a 12V bulb as load, just to see if things work.

Both setups seem to work equally well in this low voltage layout.

There are some differences and agreements in the drain - source signals, see these screenshots (1st is from Jeg his setup, 2nd is from the Bidirectional setup):

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47589;image)

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47591;image)

Both seems to show that the upper (white trace) MOSFET is doing more work looking at the mean and rms voltages (current through both MOSFETs should be the same)
and indicated by the temperature meter showing some 5 degrees C more on the top MOSFET.

But in the Bidirectional setup it shows the top MOSFET having more voltage across the drain - source compared to Jeg his setup.

Itsu

From the above experiment (and after increasing the voltage across the MOSFETs / load to 40V), it seems that the voltage across the both MOSFETs is not equally divided across them.
The top MOSFET is taking the most (80 to 90% roughly) of the voltage across it, so when we need to work with 3000V it seems that the top MOSFET will get some 2400V across it.

Itsu

Hi Itsu,

Thanks for doing these tests and sharing the results.   

Interesting outcome and very likely the use of the MOVs (or other components) to help the HV distribute more or less equally between the series connected fets is a must.  Even if not equally but at least not let HV exceed the rated break down specs for any of the fets.

Gyula

Quote from: gyula on 2023.03.26, 15:47:43

very likely the use of the MOVs (or other components) to help the HV distribute more or less equally between the series connected fets

Yes indeed. I use a 10 Mohm resistance between source and drain for each of the two mosfets.

Hi Jeg,

I forgot you did include the 10 MOhm resistors in your circuit schema here https://www.overunityresearch.com/index.php?topic=4459.msg104395#msg104395 (https://www.overunityresearch.com/index.php?topic=4459.msg104395#msg104395)   

Regarding the bidirectional circuit topology I had suggested to Itsu earlier,  we should not use it here.   

Gyula

Quote from: Jeg on 2023.03.27, 12:55:00
Yes indeed. I use a 10 Mohm resistance between source and drain for each of the two mosfets.

Jeg, did you measure the drain source voltage across your MOSFETs?

If i do that i get instable readings, depending on what MOSFET i use in what position (see the 1st screenshot here: https://www.overunityresearch.com/index.php?topic=4459.msg104456#msg104456)

I have this breadboard setup now, so i have the 10M resistors across the MOSFETs:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47596;image)

But the 40V across the load (bulb) is either on the top MOSFET or, if i change over the both MOSFETs, on the bottom MOSFET.
It seems there is some preference for a certain MOSFET to switch, and then it takes it all (almost like 80 / 20 % across the MOSFETs).


Itsu

Hi Itsu
How do you scope your two drains? Try to connect your ground clip to the source of the lower switch and probe both drains with that same reference. The upper's drain normally would be the double of the voltage than that measured on the lower drain.

Also check your resistances if are indeed the same.

Jeg,

If i follow that, i have 40V on the upper MOSFET drain and 40V on the lower MOSFET drain (reference is the lower MOSFET source).

The way i measured it earlier, is scoping each MOSFET drain - source separate.

Make sure you have no ground connected to the circuit, that way you can use ONE (1) probe of your scope to scope any component, like the both MOSFET drain - source.

Itsu

If i make a LTspice simulation of the Jeg setup, then it works as expected (each MOSFET takes half the voltage), see the below simulation output.  (.asc file attached).

I use IRF840 MOSFETs (500V) at 300V load supply.

Itsu

What i did to find the problem is to swap some connections / components.

The problem moves with the MOSFET, so if the bottom MOSFET takes all the voltage, i swap it with the top MOSFET and now the top MOSFET takes all the voltage.

When swapping the gate toroid coil connections, the problem stays at the same  MOSFET.
Also swapping the 1uF cap, 1.8 Ohm and diodes between top and bottom secondary circuits keeps the problem at the same MOSFET.

So it seems to me that the problem is within the MOSFET, some parameters are different and causing this inbalans.
I tried different MOSFET pairs (IRF540, IRF840), but both type have the same problem.

Itsu

Hi Itsu,

Yes, there are no MOSFETs with identical parameters.  If you use say small value source resistors in series with the source pins, and the difference between them is only 0.01 Ohm, the drain-source voltages will already differ.
See the simulation attached (asc file is below it).
If you set the two resistors equal, or simply short circuit them, the drain source voltages will nicely overlay. 

The difference in the MOSFETs even within the same type or batch can cause easily what you have found in practice. 
Or If you use say IRF840LC type for the bottom fet in the simulator for instance its drain voltage will drastically get reduced to a few volts.

Gyula

Gyula,

thanks, i thought so, so that is why i asked Jeg to measure his drain - source voltages, because he could have the same problem.

I changed the "Ron" parameter on one IRF840 in the .Model file in the "LTspiceXVII/lib/cmp/M standard.mos" file from 850m to 500m:


.model IRF840 NMOS(Level=3 Gamma=0 Delta=0 Eta=0 Theta=0 Kappa=0.2 Vmax=0 Xj=0 Tox=100n Uo=600 Phi=.6 Rs=6.382m Kp=20.85u W=.68 L=2u Vto=3.879 Rd=.6703 Rds=2.222MEG Cbd=1.415n Pb=.8 Mj=.5 Fc=.5 Cgso=1.625n Cgdo=133.4p Rg=.6038 Is=56.03p N=1 Tt=710n mfg=International_Rectifier Vds=500 Ron=850m Qg=63n)

but that did not change the LTspice simulation outcome,  i guess that "Ron" parameter is just an indicator, not a parameter for the RDSon (Drain-source ON resistance).

Anyway, so i have to "match" 2 IRF840 MOSFETs (or any other MOSFET pair) to get it right.

I did try to use a DMM in resistance mode to measure the Ron resistance, but they both measure 0.6 Ohm which is not the specified 850mOhm RDSon.


For info: I measured the "coupling coefficient" (K) of my coils on the gate isolation toroid to be 0.99999, so very close to 1, so to have the LTspice simulation correct.


Itsu

There could be a few reasons why one of the two MOSFETs is taking more voltage than the other.

    1. Mismatched MOSFETs: The MOSFETs may not be perfectly matched in terms of their threshold voltage or on-resistance, which can cause one of the MOSFETs to conduct more than  the other.

    2. Unequal gate drive: The gate drive for one MOSFET may be stronger than the other, causing it to conduct more.

    3. Parasitic elements: Parasitic capacitance or inductance in the circuit can cause one MOSFET to turn on faster than the other, leading to an imbalance in voltage across the two MOSFETs.

Most likely is that there is a difference at ON-resistance or unequal threshold voltage.

Try replacing the MOSFET that is exhibiting the higher voltage with a new one and see if the issue is resolved. If the issue persists, then there may be an issue with the gate drive circuit or parasitic elements in the circuit that are causing the voltage imbalance.

I may have the same issue but didnt notice. It is because i use them to share the voltage of the kickback spike and so i was interesting more on what happens when switch goes OFF and not when switch goes ON as you do. I will test it though cause i am curious.

Thanks Jeg,

reason 2 and 3 are not in play here i think because it swapped out the gate drive parts, and it did not change the problem, also the turn on time was measured to be the same.

So indeed, probably reason 1 is the cause.

The gate voltages are above the MOSFET gate threshold level, so they should not be in the linear region, but fully on, so that leaves indeed the RDSon values.

I changed several MOSFETs, but they all show similar behavior up till now.

Itsu

As suggested by Gyula, here i use 2 different MOSFETs, bottom the normal IRF840, and top the IRF840LC which has some different parameters.
The effect is the same as i see on my bench circuit, 1 MOSFETs takes 80% of the voltage, the other 20% or so when measuring across each MOSFET drain - source.

Itsu

Hi Itsu,

Here are the MOSFET model parameters listed which LTSpice and in general Spice based simulators use:
http://www.simonbramble.co.uk/lt_spice/ltspice_lt_spice_tutorial_6.htm (http://www.simonbramble.co.uk/lt_spice/ltspice_lt_spice_tutorial_6.htm) 

It turns out that LTSpice ignores the Ron parameter...   >:(

At the bottom of the link there is an Rdson test jig asc circuit file to play with gate bias and drain current and display ON resistance for a MOSFET model. It mentions the Kp parameter which is the transconductance (in Siemens), this shows how the drain current changes in the function of the gate-source control voltage.

But the important thing would be to match the MOSFETs in practice of course. Data sheets specify Rdson at given bias test conditions. And even so, the individual values can differ from it as we find.

You wrote: "I did try to use a DMM in resistance mode to measure the Ron resistance, but they both measure 0.6 Ohm which is not the specified 850mOhm RDSon."

Try to bias the gate source by a PS for say 15 VDC first and then remove it. The Cgs gate source capacitance will keep this voltage for a certain time and during this short time use the Ohm meter to see the Rds resistance. (Connect the positive lead of the Ohm meter to the drain and the negative to the source.)  Sorry if you already did it so.
And then do the same with the other MOSFET(s) within the same type.   The gate-source charge will slowly discharge as the Ohmmeter imposes its own measuring current (it will be the drain-source current).

Gyula

Thanks Gyula,

so the "Ron" parameter is indeed ignored, just for aiding the user to compare MOSFETs.

The RDSon test jig runs OK, 700.05mOhm  for the IRF840, and 680.75mOhm for the IRF840LC (10V on gate / 4.8A  through drain), so this 20mOhm difference causes the 80/20% drain voltage difference seen.


Concerning measuring the RDSon, yes i used that method to measure the RDSon, but it seems inaccurate this way (both measure 0.6 Ohm).
I guess one has to be lucky to have 2 MOSFETs that match.

Itsu

Addition:

I find that the transconductance parameter Kp is the main factor in simulation. If we choose for instance MOSFET type IRF840LC for the top fet (Kp=6.626) and the IRFBC40LC for the bottom fet (Kp=4.96), these are relatively close to each other.
It is known that the use of a source resistor reduces transconductance (it gives negative feedback) and indeed adding 0.05 Ohm resistor in series with the source of the IRF840LC, we can get the desired response, see attachment. Without the resistor, the blue trace amplitude goes up to 210 V peak voltage from the desired 150 V.

Note that such "solution" is valid for a given average drain current the R3=100 Ohm load in this simulated circuit establishes from the 300 VDC supply. Changing the supply voltage and / or the load will change the common drain current and the drain source voltages will also change.

So the use of a series source resistor for the MOSFET having the higher transconductance to match it to the other MOSFET is not the best solution, may serve as a secondary solution.

Gyula

Well, the main problem is that RDSon does not really change linearly, it depends mainly on the actual average drain current, and this involves the actual supply voltage and load resistance.  So it is not directly the 20 mOhm difference that causes the problem.   The RDSon resistance of the two series MOSFETs make a voltage divider and these resistors change their actual value as the drain current changes by the duty cycle, by the supply voltage and by the load. But as the simulations show the transconductance is the main parameter, the IRF840 has a Kp of 11.192.

Gyula

Quote from: Itsu on 2023.03.29, 15:24:42
Thanks Gyula,

so the "Ron" parameter is indeed ignored, just for aiding the user to compare MOSFETs.

The RDSon test jig runs OK, 700.05mOhm  for the IRF840, and 680.75mOhm for the IRF840LC (10V on gate / 4.8A  through drain), so this 20mOhm difference causes the 80/20% drain voltage difference seen.


Concerning measuring the RDSon, yes i used that method to measure the RDSon, but it seems inaccurate this way (both measure 0.6 Ohm).
I guess one has to be lucky to have 2 MOSFETs that match.

Itsu

Itsu,

Perhaps you might want to look at techniques for balancing series connected mosfets via snubber capacitors.  See attached pdf for one example .

Pm

Hi Pm,

Thanks for this paper, it offers a clever idea and simple solution. 
In the hardware configuration Figure 5, diodes are shown between the fets at each source pin and the text does not mention about them. Do you know why they are used?
(I assume they block possible switching spikes.)

Gyula

Quote from: gyula on 2023.03.29, 19:30:52
Hi Pm,

Thanks for this paper, it offers a clever idea and simple solution. 
In the hardware configuration Figure 5, diodes are shown between the fets at each source pin and the text does not mention about them. Do you know why they are used?
(I assume they block possible switching spikes.)

Gyula

Hi Gyula,

After looking again at the paper, I'm not sure about their function!  There are a number of papers out there using the same technique and some on IEEE but my membership has lapsed.

Pm

Quote from: Itsu on 2023.03.29, 15:24:42

I guess one has to be lucky to have 2 MOSFETs that match.

Itsu

Hey Itsu,

I'm unsure of the power demands of your circuit but seeing that you're trying to operate a 3KV switch, you might be able to get away with a lower current FET with a higher blocking voltage. 

Here is one that might work:
IXTF1R4N450 --- MOSFET N-CH 4500V 1.4A
https://www.digikey.com/en/products/detail/ixys/IXTF1R4N450/6109391


This is the one that I personally used to make a gate isolated, HV full bridge circuit for various HV experiments:
IXTF02N450 --- MOSFET N-CH 4500V 200MA
https://www.digikey.com/en/products/detail/ixys/ixtf02n450/3737625

Take Care,

Dave

Quote from: partzman on 2023.03.29, 18:36:30
Itsu,

Perhaps you might want to look at techniques for balancing series connected mosfets via snubber capacitors.  See attached pdf for one example .

Pm

Partzman,

thanks for the paper, i looked at it, and could be worth to do something with those caps, but the paper is not so detailed i think.

Itsu

Quote from: web000x on 2023.03.30, 06:24:53
Hey Itsu,

I'm unsure of the power demands of your circuit but seeing that you're trying to operate a 3KV switch, you might be able to get away with a lower current FET with a higher blocking voltage. 

Here is one that might work:
IXTF1R4N450 --- MOSFET N-CH 4500V 1.4A
https://www.digikey.com/en/products/detail/ixys/IXTF1R4N450/6109391


This is the one that I personally used to make a gate isolated, HV full bridge circuit for various HV experiments:
IXTF02N450 --- MOSFET N-CH 4500V 200MA
https://www.digikey.com/en/products/detail/ixys/ixtf02n450/3737625

Take Care,

Dave

Hi Dave,

thanks, it looks like that would be a better solution instead of going "in series" and it was suggested before.

But i expected these HV MOSFETs to be expensive, but it turns out they are reasonably prized.

The "current" demand is not that high, so these MOSFETs you suggested could do the job i think.

Itsu 

Hi Pm,

Okay, thanks for the hints.  I have found some papers and among the freely available ones these two sound informative and practical:

https://ntnuopen.ntnu.no/ntnu-xmlui/bitstream/handle/11250/2472861/Peftitsis.pdf (https://ntnuopen.ntnu.no/ntnu-xmlui/bitstream/handle/11250/2472861/Peftitsis.pdf)   

https://iopscience.iop.org/article/10.1088/1742-6596/2401/1/012052/pdf (https://iopscience.iop.org/article/10.1088/1742-6596/2401/1/012052/pdf)

Side note OF course the simplest solution would be to use a single HV switching device capable of working  directly at the needed high voltage levels.

Gyula

Quote from: partzman on 2023.03.29, 22:34:45
Hi Gyula,

After looking again at the paper, I'm not sure about their function!  There are a number of papers out there using the same technique and some on IEEE but my membership has lapsed.

Pm

https://item.taobao.com/item.htm?spm=a1z10.3-c.w4002-11819735349.9.3dee2ac7f4hMFg&id=573936200334

IXGF30N400

I often use disassembly elements

Thanks Panyuming,

a good example of a usable HV IGBT.

Itsu

To be sure that the breadboard does not add some parasitic capacitance on one side of the gate / MOSFET, i put this Jeg circuit on a PCB and made it as symmetrical as possible see the picture:


I now use some HV (1700V) MOSFETs G3R450MT17D's, but there still is asymmetry in the voltages across the Drain / Sources of the both MOSFETs, see screenshot:



Itsu

Hi Itsu,

From Figure 11, Page 6 in this data sheet https://genesicsemi.com/sic-mosfet/G3R450MT17D/G3R450MT17D.pdf (https://genesicsemi.com/sic-mosfet/G3R450MT17D/G3R450MT17D.pdf)  the output capacitance C_OSS varies roughly between 30 pF and 120 pF when the drain source voltage, V_DS changes between zero and about 70 V.  So if you feel like using a snubber capacitor across the drain source of the MOSFET which now has the higher V_DS  (white wave form) it may help achieve a better balance.  For the capacitor, perhaps an air variable capacitor would be the best to use,  the 40 V should not cause any arcing between the capacitor plates. Do this only if you agree, of course.  OF course, even a fix ceramic or any good quality capacitor (of 56pF or 72pF etc) would do instead of a variable cap.

Thanks for sharing.

Gyula 

Gyula,

Without any snubber caps, i have 41.8Vpp across the left (highest voltage) MOSFET and 9.1Vpp across the other MOSFET (40V input).

When using a 6-100pF trimmer cap across the left (highest voltage) MOSFET Drain - Source, while scoping the voltage across its DS i did not see any noticeable difference, but when moving the probe to the other MOSFET DS, i could increase the voltage from 9.1 to 14.9V when turning the trimmer cap. to max.

I added then some ceramic caps to this trimmer cap and saw the change in voltages, see table below.

Finally i used a fixed 2nF ceramic cap across the left MOSFET instead of the trimmer cap, and its voltage dropped to 35.4V, while the voltage across the other MOSFET was increased to 34.2V.

Not what i expected as what i would expect is to have 20V across each MOSFET when balanced.

I do notice some rounded off corners on the left MOSFET with the 2nF cap, see screenshot (yellow trace, White is the original trace without any caps).

So it seems that the snubber caps are doing something, but not what we want, here is a table of the results:

Cap value |Left MOSFET|  Right MOSFET
----------|------------|---------------
   0pF      |    41.8V     |   9.1V
100pF      |    41.2V     |  14.9V
200pF      |    39.5V     |  18.0V
570pF      |    38.5V     |  25.6V
   2nF      |    35.4V     |  34.2V   with rounded off corners visible.


Itsu

Hi Itsu,

Thanks for the tests.

I find quasi similar behaviour in simulation.  When the voltage across the top fet's DS was 91V,  the voltage across the bottom fet's DS was 207V.  By adding a 150 pF snubber cap across the bottom fet, the voltages changed to 99V and 201V respectively. And when I used 1.3 nF snubber, both voltages changed to 150 V, the waveform nicely covered each other.  Supply voltage was 300V.

When I reduced the supply voltage to 40V for the same circxuit, the voltage across the top fet's DS was 22.5V and 17.5V across the bottom fet's DS. A 700 pF snubber cap (instead of the 1.3 nF) equlized the two voltages to 20V.  See the picture attached on the latter simulation.

Gyula

Gyula,

i do not call that a quasi similar behavior as you can balance the voltages to have half of the supply voltage over each MOSFET (20V at 40V input), but my bench does show equalization, but not to half the voltages, as i have 35V / 34V across the both MOSFETs, but with 40V input.

Itsu

Sorry, you are right.  More study is needed. 

Gyula

Itsu, i did some quick tests in this direction.
V=48 V dc
15Khz
Load: 42V 40W incandescent bulb.

At 50 % duty cycle the upper mosfet is at 46V, and the lower mosfet 2V!!!
As long as i decrease the duty cycle the difference starts to decrease. At 39% we have equality in voltage. As i go lower than 39 % the voltage at the lower mosfet starts to increase its voltage level at higher values than the upper. Even if i didn't try to change frequency it is obvious that the analogy will also be affected.

Thanks for pointing out this issue.

Ps. There is a technique which introduces a small delay to one of the two gate pulses and equalise voltages. It is something like a small capacitor between gate and source of one of the mosfets. Normally the one that takes the higher voltage across it.

Jeg,

thanks for doing these tests, so it confirms what i see here.

But i do not have the outcome you describe, when changing the duty cycle from 50% to 10%, there is hardly any difference in the 2 voltages, in my case it stays around 41Vpp / 9Vpp with 40V input.

So any type of MOSFET could have its own behavior.

Itsu

I did a test using my DMM in DC voltage mode, I had the FG set to 10kHz @ 50% duty cycle, input 41.5V.

With no snubber caps installed:
Left MOSFET   15V   (Scope shows 42Vpp)
Right MOSFET  4.8V (scope shows 10.9Vpp)

With a 2nF snubber cap installed on the left MOSFET:
Left MOSFET   3.6V??!!   (Scope shows 35Vpp)
Right MOSFET  17V         (scope shows 35Vpp)


So also the DMM shows a difference in the no snubber situation, like my infrared temperature meter, Left MOSFET 22 degrees C,  right MOSFET 17 degrees C.   (17 degrees is room temp.).
Why the DMM shows 3.6V with the 2nF cap across it I don't know, but the infrared meter still shows same temperatures across both MOSFETs 22 / 17 degrees C).

Itsu

Itsu normally it can also be done without capacitor but by varying the resistances between gate and source.
This will affect again the charching time of the gate input capacitor.

Jeg,

lets forget the solution for a moment, i made some further tests and have some questions.

When you say: "At 50 % duty cycle the upper mosfet is at 46V, and the lower mosfet 2V!!!" how did you measure that? (DMM or scope and if with the scope what was it p2p, mean or rms?)

You have a 42V 40W bulb as load, so at 48V input, what is the voltage drop across this bulb?

Thanks,  Itsu

Itsu i am talking about waveforms like yours above. 46v is the peak level. I used an oscilloscope, but by using the same reference for both of my probes. And this is the source of the lower mosfet. So when both voltages are equal then my upper probe shows exautly the double the voltage of the lower mosfet.

Thanks Jeg.

I see that with every change in supply voltage, frequency and duty cycle the voltages across the both MOSFETs vary, so i will be not easy to come to a stable situation across these variables.

Itsu

Quote from: Itsu on 2023.04.03, 20:20:21
Thanks Jeg.

I see that with every change in supply voltage, frequency and duty cycle the voltages across the both MOSFETs vary, so i will be not easy to come to a stable situation across these variables.

Itsu

Yes Itsu it seems that it needs to be calibrated on its final position/placement with fixed values of frequency duty and voltage. Not that bad for such a cheap mosfet switch. 

Jeg,

that's true, but in an experimental environment like we have where we need to vary those parameters on the fly to search for optimal results in our circuits, its not so good.

Itsu

Gyula pointed me to this article: http://www.power-mag.com/pdf/feature_pdf/1260811353_CT-Concept_PEE_0809.pdf in where is described some active clamping methods for stabilizing IGBT/MOSFETs.

It turns out that the gate clamping method shown in Fig. 4 also works to stabilize / balance the drain - source voltages of our series MOSFETs.

I build the below circuit, and it works fine in balancing the 40V input across the load and both MOSFETs:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47654;image)

It stays balanced across a wide range of frequency, duty cycle and input voltage settings.

Care should be taken to not let the zener current increase too much leading to overheating etc. and to not let the Gate - source voltage to increase above the MOSFET specifications.

The 10K resistor in series with the zener can be adjusted accordingly.

As shown in the diagram, the zener / diode combination can be changed out for a bidirectional TVS diode (Transient Voltage Suppression) like the shown 1.5KE18CA.

I am now building a HV edition of this circuit using 2 C2M0045170D MOSFETs (1700V) and some 1.5KE440CA (440V) TVS diodes to see if it can switch the 3kV from my source.

Regards Itsu

I build Lee his switching board as presented here:  https://www.overunityresearch.com/index.php?topic=4459.msg104430#msg104430

I compared this switching unit with another older one which i was using up till now, see picture:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47729;image)

But when i saw the signal when the MOSFET closes i was kind of disappointed.

here is a screenshot of a 10kHz 10% duty cycle signal taken from both circuits:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47731;image)

Yellow signal is Lee his circuit (using the 1700V MOSFET) and white signal the older used switching board (also a 1700V MOSFET).
Load is a 470 Ohm resistor at 41V.

I use 3.9 Ohm resistors as source / sink resistors in Lee his circuit.

I will try to use a short for the sink resistor to see if this improves the turn-off time.


Itsu

Changing the sink resistor R3 for a short does not change the turn-off signal.

Itsu

I made some further tests, especially to test the gate driver signals, but these turn out to be OK.

So finally i changed the used 1700V MSC035SMA170B4 (TO-247-4 layout) MOSFET for a CREE 1700V C2M0045170D (TO-247-3 layout) MOSFET.

I had to modify the board / MOSFET connections due to the layout differences, but then the turn-off time was much better, see screenshot:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47748;image)

Yellow is the CREE drain source signal and blue the gate signal.


I cannot imagine i had a damaged MOSFET as i hardly used it, so i ask Lee if he can do the same test as i did by switching a 470 Ohm resistor load with 40 or so volt at a frequency of 10kHz and at 10% duty cycle and look at the turn-off time.
Does it have the same 1us turn-off time i had as shown in this (yellow) trace:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47750;image)

Thanks,   regards Itsu

Hi Itsu,

Sorry for the late reply to your original post about slow turn off time for the MOSFET. I only just saw it and your subsequent post. For some reason this thread wasn't highlighted as having new content on the main page.
It's certainly very peculiar that it takes so long to turn off, especially when you have shown that the Cree turns off much quicker. The MSC035SMA170B4 has a gate charge of 178nC and the C2M0045170D has a gate charge of 200nC, so if anything I'd expect the C2M0045170D to turn off slower because there is more charge to discharge.

In my experience, a MOSFET turning off slowly is due to the charge on the gate not being discharged quick enough. This would usually be caused by a high resistance between the gate and the gate driver. Reducing this resistance should improve the gate discharge time, at the expense of some ringing. Having said that, you are using 3.9ohm resistors which is what I'm also using.

What modification did you make to mount the 3 pin Cree MOSFET? Can you please post a photo?

I have some boards built with the MSC035SMA170B4, so I'll run a test tomorrow. I don't have a 470ohm load resistor but I do have a 500ohm load resistor. Is that close enough?
Best regards,
Lee

I used this hardly used 1700V MSC035SMA170B4 MOSFET in my old switching board and also here i have this long (1us) turn-off time:
Yellow DS signal, blue gate signal.

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47752;image)

So either my 1700V MSC035SMA170B4 is damaged or this MOSFET is working as normal this way.

Itsu

Quote from: lfarrand on 2023.04.28, 19:39:25
Hi Itsu,

Sorry for the late reply to your original post about slow turn off time for the MOSFET. I only just saw it and your subsequent post. For some reason this thread wasn't highlighted as having new content on the main page.
It's certainly very peculiar that it takes so long to turn off, especially when you have shown that the Cree turns off much quicker. The MSC035SMA170B4 has a gate charge of 178nC and the C2M0045170D has a gate charge of 200nC, so if anything I'd expect the C2M0045170D to turn off slower because there is more charge to discharge.

In my experience, a MOSFET turning off slowly is due to the charge on the gate not being discharged quick enough. This would usually be caused by a high resistance between the gate and the gate driver. Reducing this resistance should improve the gate discharge time, at the expense of some ringing. Having said that, you are using 3.9ohm resistors which is what I'm also using.

What modification did you make to mount the 3 pin Cree MOSFET? Can you please post a photo?

I have some boards built with the MSC035SMA170B4, so I'll run a test tomorrow. I don't have a 470ohm load resistor but I do have a 500ohm load resistor. Is that close enough?
Best regards,
Lee

Hi lee,

the 500 Ohm will be fine.

Here a picture of my mod, it is a simple cross-over between the pins and board.

By the way, for R3 i used a 0 Ohm (short, see picture) resistor, but that did not make any difference with the earlier used 3.9Ohm resistor.

Itsu

Well at least the MOSFET is consistent in it's inability to turn off quickly! :)

If it is behaving the same way in a completely different board with different gate driver etc. then it is either the MOSFET that is damaged in some way, or this is the inherent behaviour of this MOSFET. With that said, the spec. sheet lists a 17ns turn off time, so 1 microsecond is considerably longer.

I'll run some tests in the morning and let you know my findings.

Hi Itsu,

I ran the test. I placed a 500ohm load resistor across the switch and powered it with 40V from my bench power supply.

The results were..interesting! They seem to confirm your finding about slowness, although I see it as being slow to turn off rather than slow to turn on.

The pink trace is from my signal gen, and the yellow trace is a probe placed across the load resistor.

(https://i.ibb.co/wSf3qdw/SDS00001.png) (https://ibb.co/D9ZsF18)

(https://i.ibb.co/gJL6YSB/SDS00002.png) (https://ibb.co/4pnJk85)

Hi Lee,

well, that confirms that this type of MOSFET has a slow turn-off time (as that was slow with me too all the time) only it is inverted by you because you measure across the resistor while i measured across the drain - source.

Perhaps it's only slow at this relative low drain voltage as this type of MOSFET is designed to handle HV (1700V).

I will see if i can make a measurement with those kinds of voltages.

Thanks for your test.

Regards Itsu

Yes, it would be interesting to see the same test done with a much higher voltage. If it has the same behaviour then some questions have to be asked about the Microchip spec sheet. I'll reach out to them to see if they have any explanation.

I guess I need to take another look at the Cree/Wolfspeed MOSFETs. They are super expensive though...

I filed a support case with Microchip to report the problem. Let's see what they say.

(https://i.ibb.co/hBtLsKT/IMG-1567.jpg) (https://ibb.co/x1RXhMr)

Thanks Lee,

yes, let's see what they have to say.

Itsu

I tried to measure the turn-off time at a higher voltage (1700V), but therefor i had to use a higher resistance resistor of 1Mohm to stay within the resistor wattage limits.
But it seems that due to this higher resistance, the RC time (R load resistor, C output capacitance of the MOSFET) becomes so long that it won't switch anymore.

So basically i can not measure the turn-off time at 1700V using a 470 Ohm resistor.


I then measured the capacitance of the both used MOSFETs, and it shows a much higher capacitance of the MSC035SMA170B4 between its ports then the C2M0045170D:
 
Tested with my LCR meter at 10kHz:

MSC035SMA170B4 (Lee MOSFET)   
GS          DS         GD
4.9nF      4nF        3.4nF

C2M0045170D (CREE)
GS           DS         GD
331pF      291pF      324pF

This could be the reason that the MSC035SMA170B4 behaves worse than the C2M0045170D.

But this is NOT how the data sheets show they measure the input/output capacitance!!

Itsu

Hi Itsu,

The strange thing is that your capacitance tests show pretty close values for the MSC035SMA170B4 as per the data sheet includes. 

And the capacitance tests on the C2M0045170D type show very low pF values by the same C meter.  Of course , the capacitances depend very much on the voltage levels between the MOSFET pins and the data sheets show the curves in the function of the drain-source DC voltage, they used 25 mV AC at 1 MHz frequency.  (I do not think your 10 kHz test frequency is a problem here. )
Basically these Silicon Carbid MOSFETs have very similar behaviour capacitance wise in the function of the drain source voltage to that of "normal" power MOSFETs. i.e. as the drain source voltage increases, all the capacitances decrease with respect to their zero bias capacitance values.   
   
Probably the capacitance measurement can be done by applying a static DC bias across the drain source pins via the 470 k or even 1 MegaOhm drain resistor and use the C meter between the drain and source via a coupling capacitor of say 1 nF in series with the meter.   The gate and source pins should be connected together (Vgs=0).

I attach below the relevant Figures from both data sheets. 

Perhaps the answer from the manufacturer Lee turned to may reveal some explanation on the long turn off time issue.

Gyula 

Edit: I wrote applying static DC bias across the drain source pins but of course a variable DC is to be used.  Normally at zero voltages all the capacitances are at their highest values (some nF) and these decrease as the voltage levels increase, the higher the voltage levels (up to the max rated ones) the lower the capacitances go down.  Varicap diodes behave like this too.

Itsu,

Microsemi recommends a gate drive that switches between +20 and -5 volts.  The negative V on the gate assists turn off.

As an experiment, and without needing to modify your driver, you could insert a floating 5 volt supply between the source lead and ground so that the gate goes negative relative to source when turned off.  Bypass the 5 volt supply with a cap very close to the device.

Also, keep lead lengths as short as possible...

PW   

Gyula,

thanks, i will give that a try later today.

Itsu

Quote from: picowatt on 2023.05.01, 13:37:53
Itsu,

Microsemi recommends a gate drive that switches between +20 and -5 volts.  The negative V on the gate assists turn off.

As an experiment, and without needing to modify your driver, you could insert a floating 5 volt supply between the source lead and ground so that the gate goes negative relative to source when turned off.  Bypass the 5 volt supply with a cap very close to the device.

Also, keep lead lengths as short as possible...

PW

PW,

in my original test of this Microsemi MSC035SMA170B4 MOSFET which can be found in post #91 of this thread here:  https://www.overunityresearch.com/index.php?topic=4459.msg104637#msg104637
i used the PCB and circuit designed by Lee which contains a R12p22005d ( https://recom-power.com/pdf/Econoline/RxxP2xxyy.pdf ) which does switches the gate between +20V and -5V.

But still it had this long turn-off time compared to the CREE C2M0045170D MOSFET as also can be seen in that post.

Regards Itsu

Quote from: gyula on 2023.05.01, 10:52:32
Hi Itsu,

The strange thing is that your capacitance tests show pretty close values for the MSC035SMA170B4 as per the data sheet includes. 

And the capacitance tests on the C2M0045170D type show very low pF values by the same C meter.  Of course , the capacitances depend very much on the voltage levels between the MOSFET pins and the data sheets show the curves in the function of the drain-source DC voltage, they used 25 mV AC at 1 MHz frequency.  (I do not think your 10 kHz test frequency is a problem here. )
Basically these Silicon Carbid MOSFETs have very similar behaviour capacitance wise in the function of the drain source voltage to that of "normal" power MOSFETs. i.e. as the drain source voltage increases, all the capacitances decrease with respect to their zero bias capacitance values.   
   
Probably the capacitance measurement can be done by applying a static DC bias across the drain source pins via the 470 k or even 1 MegaOhm drain resistor and use the C meter between the drain and source via a coupling capacitor of say 1 nF in series with the meter.   The gate and source pins should be connected together (Vgs=0).

I attach below the relevant Figures from both data sheets. 

Perhaps the answer from the manufacturer Lee turned to may reveal some explanation on the long turn off time issue.

Gyula 

Edit: I wrote applying static DC bias across the drain source pins but of course a variable DC is to be used.  Normally at zero voltages all the capacitances are at their highest values (some nF) and these decrease as the voltage levels increase, the higher the voltage levels (up to the max rated ones) the lower the capacitances go down.  Varicap diodes behave like this too.

Gyula,

i followed your suggestion and used a 470K resistor in the drain lead to feed in 0 to 12V.
The Gate was shorted to the Source (Vgs=0) and the LCR meter was connected via a series capacitor of 47nF to the drain and source.

The results was:

MSC035SMA170B4   4.5nF at 0V decreasing to 1.6nF at 12V
C2M0045170D        390pF at 0V decreasing to  70pF at 12V

So they both follow the graphs you showed meaning a decreasing capacitance with increasing voltage across DS.

The MSC035SMA170B4 follows its graph reasonably accurate, but the C2M0045170D shows a much lower Coss then in the graph for some reason.

Itsu

Hi Itsu,

Thanks for doing the drain-source capacitance check for the two MOSFET types (both are Silicon Carbide technology).  Indeed it is strange that the C2M0045170D type has such a low capacitance (around 390 pF instead of the around 1200 pF at 10V as it can be estimated from its data sheet https://assets.wolfspeed.com/uploads/2020/12/C2M0045170D.pdf (https://assets.wolfspeed.com/uploads/2020/12/C2M0045170D.pdf) .

Just out of curiosity, I looked up Wolfspeed's new 650V SiC MOSFET family (3rd generation SiC technology) and this type for instance C3M0045065L 650V, 49A 45mOhm  ( https://assets.wolfspeed.com/uploads/2022/10/C3M0045065L.pdf (https://assets.wolfspeed.com/uploads/2022/10/C3M0045065L.pdf) ) has around 600 pF Coss output capacitance at 10V from its data sheet, Figure 17, Page 6.

I do think that the data sheets include correct output capacitances and I also think your output capacitance measurements gave correct, very close to reality results on those particular devices. Yet there is an order of magnitude difference between the two types contrary to their data sheets. 

Hopefully, the manufacturer (Lee turned to) can give explanation on the longer turn off time you and also Lee experienced in practice.

Gyula

Microchip replied to me and asked me to provide the gate waveform. I provided them the signal gen trace as well as the load resistor trace, but I didn't include the gate trace. I'll reply to them shortly and see what they say.

Thanks lee,

in my post # 95  (https://www.overunityresearch.com/index.php?topic=4459.msg104658#msg104658) i show a screenshot of the Drain - Source signal (yellow) and the Gate - Source signal (Blue).
Its not in your circuit but my old board, so without the +20V / -5V gate drive signal.

Itsu

I provided the following scope shots to Microchip.

Blue trace: load resistor
Purple trace: MOSFET gate
Yellow trace: signal gen

MOSFET gate (Vgs) voltage showing +20/-5V

(https://i.ibb.co/tBD5FJL/MOSFET-gate-Vgs-voltage.png) (https://ibb.co/p2dp5Lh)

No load rise & fall showing <10ns rise & fall

(https://i.ibb.co/s1gWjGF/No-load-rise-fall.png) (https://ibb.co/xg2Fh0M)

500-ohm dummy load (40V) rise & fall showing ~10ns rise and ~1.25us fall

(https://i.ibb.co/TmrVhMp/500-ohm-dummy-load-rise-fall.png) (https://ibb.co/ZWNnTzt)

Dear Lee,

Thanks for showing the scopeshots.  Would you mind providing a simple schematic with the measuring points indicated that you used in these tests?

Thanks,
Gyula

I, too, was wondering how the used circuit looks like and where the probe points (tip and ground) are as it seems strange you can measure the "signal gen" signal together with the gate signal and the signal
across the load resistor.

They all have different grounding points IMO.

Here i used Lee his circuit including the MSC035SMA170B4 MOSFET, so with the +20V / -5V gate signal and measure only the DS (CH1 yellow) and GS (CH2 blue) signals as they have a common ground:
These are the signals that are most useful IMO.

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47768;image)

The circuit and measurement points i used are:

(https://www.overunityresearch.com/index.php?action=dlattach;topic=4459.0;attach=47772;image)

Kees

Quote from: lfarrand on 2023.05.04, 15:14:18
I provided the following scope shots to Microchip.

Blue trace: load resistor
Purple trace: MOSFET gate
Yellow trace: signal gen


No load rise & fall showing <10ns rise & fall

(https://i.ibb.co/s1gWjGF/No-load-rise-fall.png) (https://ibb.co/xg2Fh0M)

Hi Lee,

Hope you are doing fine.  Have you received any new feedback from Microchip?

The reason I asked you for a simple schematic with the probe points was that I do not understand the no load rise and fall time : with no load where the probe was placed?  On load I mean the 500 Ohm of course.
The other scopeshots are ok for me.

Thanks,
Gyula

Hi folks. Sorry for the delay in replying. I've been busy trying to make some bifilar pancake coils, which turned out to be a bit more complicated than I'd anticipated. They're quite tricky to wind, so I ended up building a jig to wind them on. This went through a few revisions as I tweaked it to my needs, but in the end it turned out alright.

I did get a follow up response from Microchip today.

This is what they said:

QuoteI'm assuming this is a source follower configuration (Drain to 40V, source to 500ohm resistor).  If that is true then the waveform looks correct.  The blue trace (input signal to the isolated driver) goes low, then 50ns later, the gate drive output (VGS) quickly goes low and the FET turns off.  Once the FET is off, the source node (across the resistor) is discharged with a time constant of Coss x 500ohms. The effective capacitance between 0V and 40V is around 2nf and RC = 1us.

I'm not sure why the Wolfspeed part doesn't also do this since it's Coss is approximately the same as the Microchip FET.

I'll put together a schematic showing the probe points. I'm using isolated probes to avoid grounding issues.

I just saw that Infineon have released a new line of SiC 2000V MOSFETs. Here's the link (https://www.mouser.co.uk/c/?q=IMYH200R) to view the range at Mouser.

The cheapest one has a rise time of 3ns and a fall time of 5ns with a RDS(on) of 131 mOhms.

I've ordered a selection of them to test them out.

The Infineon MOSFETs have arrived, and I was surprised to find out that they are a different form factor to the other 4 pin TO247 MOSFETs. The distance between pin 1 & 2 is 7.62mm (2.54*3), whereas it is 5.08mm (2.54*2) for the Microchip & Wolfspeed 4 pin MOSFETs. The other pins are spaced 2.54mm apart.

I guess it's because the Infineon MOSFETs are rated to 2000V, so there needs to be a greater distance between drain and source to prevent arc over.

I wish they would use a more distinct naming scheme for the package types. It's not super obvious looking at the package name - TO-247-4-PLUS (Infineon) vs TO-247-4L (Microchip). Call it something different to TO-247 if there is a significant difference! Sigh..  :(

I guess I'll have to bend pin 1 and hope for the best. I'll try it out with one of them to see how it works. I'll probably have to revise the PCB layout and get some more manufactured to do it properly. I might as well try to shrink the size of the PCB down as well and maybe replace some through hole components with SMD. That's a project for another day, I don't want that to distract me from experimenting.

Here's an interesting PDF from Infineon (https://www.infineon.com/dgdl/Infineon+Friedrichs+CoolSiC+2+kV.pdf?fileId=8ac78c8b808544e20180b78d8f03005c) explaining all about 2kV SiC technology. Page 13 is of particular interest when it explains 'Higher voltage in combination with higher frequency requires much higher clearance'.

Thanks, Lee, for your findings, interesting, and good to know.

Itsu

Hi Itsu,

Is the HV pushpull working well?
I assume 2 good matching Mosfets was the issue?

Greetings,
Ape

Hi Ape,

the test was if 2 MOSFETs in series could be set up to be switching together and dividing the Voltage across them equally, so not a push pull setup.

This worked at low voltage levels / MOSFETs using some active clamping as shown here:  https://www.overunityresearch.com/index.php?topic=4459.msg104570#msg104570

But trying to do this with HV and ditto MOSFETs dit not work out that way.

So indeed, i guess the problem still is the difference in MOSFET parameters, but looking for a matched pair is not realistic as we don't know which parameter(s) are involved, and they are expensive.

Itsu

If you look through the IVO thread on youtube he shows how to do a 3000 volt circuit using MosFets

andf more when i can find it in a large chain.

Sil

Quote from: AlienGrey on 2023.06.02, 09:28:42
If you look through the IVO thread on youtube he shows how to do a 3000 volt circuit using MosFets

andf more when i can find it in a large chain.

Sil

this is one method i'm sure there are others similar.
once you start to produce sparks your defeating energy.

Sil

I had to make some changes to my MOSFET driver PCB to support the TO-247-4-Plus footprint, so I decided to move away from EasyEDA and redesign it using Kicad 7. I felt more secure not having to rely on a cloud service to store my designs.

I've attached the Gerber files, as well as the Kicad 7 design files, should you wish to use it or modify it to suit your purposes.

I've just sent it off to JLCPCB for manufacturing. I'll keep you posted on how it goes when I receive it.

Significant changes over the last version:

• Support for TO-247-4-Plus as well as TO-247-4 footprints
• Much larger exposed copper section for the MOSFET & diode connections to support larger currents and provide some heatsinking for the connected pins
• Optional series TO-247-2 diodes to protect MOSFET from inductive kickback
• Optional series TO-247-2 diodes to protect MOSFET from reverse polarity
• 6 layer design which uses two separate ground planes connected using vias, and dedicated layers for power & signals
• Addition of a 12V regulator for 12V section
• Addition of bulk capacitors for input, 5V & 12V power planes
• Added optional gate driver inverted input BNC. Can be tied to ground using onboard jumper.
• Added NTC (negative temperature coefficient) fuse to limit inrush current
• Changed gate resistors from through hole to SMD

3D View
(https://i.ibb.co/kX60RD4/PCB-3D.png) (https://ibb.co/vkvm7d3)

All Layers Composite
(https://i.ibb.co/wrS00NK/PCB-all-layers.png) (https://ibb.co/n87ggsD)

Signal & Power
(https://i.ibb.co/M7qr6GN/PCB-signal-and-power.png) (https://ibb.co/3ct2zMh)

Schematic
(https://i.ibb.co/SVTV8SP/PCB-schematic.png) (https://ibb.co/KVZVknX)

Looking great Lee  O0

Thanks for the files, i like to see the PCB when finished.

Itsu

The PCBs are due to be delivered either tomorrow or Wednesday, so I'll post some photos then.

I had a hard time trying to work out which MOSFET was the best for my particular requirements. I found myself switching between endless datasheets trying to determine the meaningful differences. In the end I decided to put together a spreadsheet, and I've attached a PDF export of the comparison sheet. There's a heat map on each of the attributes, and they are organised into groups according to the voltage class. It's a lot easier to quickly glance and compare now.

The spreadsheet only lists TO-247-4 footprint MOSFETs since that's what my driver board uses, but most models have TO-247-3 equivalents and the attributes should be broadly the same.

Lee,

that must have been a tedious work, but it will come in very handy for a lot of people, so thanks a lot  O0

Itsu

I'm not going to lie, it was painfully boring at times :) I think it will pay off in the mid to long term though.

I've also attached a comparison of TO-247-2 SiC Schottky diodes.