discrete Gate driver

[Nikan]

Hello,

I am working on a discrete gate driver with a disable function.
I've already tried some IC's but my mosfets (8 Mosfets are beeing switched at the same time with 4 of them inverted) were not that satisfied with those IC's...

Anyways, I want to use Q2 and Q3 to amplify the signal. Q4 and Q5 are there to allow a higher curent flow. Q1 is the disable function, when the base of Q1 is pulled high, the gates of Q4 and Q5 should be low thus the output of Q4 / Q5 should be high.

The same should be happening with Q7/Q6 , Q9/Q11 and Q8 except that Q10 is there to invert the input signal and thus invert the output.

Is Q10 even able to invert the signal?

Will this work and drive the gate-drive-transformer theoretically?!?

*** The GDT I am going to use has only 1 primary and 8 secondaries.

I appreciate any help/suggestion.

Thanks in advance!!!

[Yansi]

Start with specifying your mosfets, required switching frequency/switching times first.

[Nikan]

I am using 8x IRFP22N50A.
The mosfets need to switch with a frequency of around 700KHz but I'm likely going to change some things...
minimum should be at around 300Khz and maximum frequency is not going to exceed 1Mhz.

[Marco]

Your MOSFET totem poles will cross conduct and your "disable" will turn on Q5&Q11 and cause a short.

[Nikan]

That is what I wanted to achieve if Q5 and Q11 both turn on, the potentials on the primary of the gate drive transformer should be almost equal and almost no current should flow through the GDT. Or am I missing something out here?

[Marco]

Yeah, you're right I misread that.

Still, your MOSFET totem pole will cross conduct, the diode resistor combos don't provide dead time because the totem pole's gates are shared. If you wanted to do deadtime with diode+resistor, you'd have to give each MOSFET its own set (with the diodes reversed). You don't need the diode+resistor on the BJT totem pole at all, that inherently won't cross conduct so just a resistor will do fine.

You probably want the disable before the BJT totem poles too, less current to sink.

Are you sure you need the MOSFET stage at all? 8x IRFP22N50A with 5 Ohm gate resistors and 1:1 GDTs would take less than 20A to switch, couldn't you find BJTs to do that directly without the MOSFET stage and its annoyances?

[Nikan]

So if I understood you right...

I need for each Mosfet a resistor and a didoe to achieve the deadtime.

And a single resistor is enough for the BJT.

Well I dont have BJT's with a high current capability...

Speaking of the GDT... Can I use the ga3550 from coilcraft? SRF is at 5MHz is it possible to run those at a frequency of around 1MHz or will I run into problems with the driving frequency beeing so high?

[Nikan]

This is how far I got.

[tautech]

Google this for further study:
mosfet shoot through

[Nikan]

I've already changed the schematic for a dead-time on the Mosfets as Marco already pointed out that my mosfets will cross conduct...
Or did I do something wrong again?

[tautech]

I've already changed the schematic for a dead-time on the Mosfets as Marco already pointed out that my mosfets will cross conduct...
Or did I do something wrong again?

Probably not.

Others here on the forum have pulled me up on 'shoot through' when I've used IRF5851 as a gate driver but it's worked and still works perfectly fine in applications where I've used it. However it's been at slower speeds than you're planning. Really you need be aware of it and verify it's not affecting your design.

The only way to be perfectly sure is to monitor the PSU for unexpected current peaks at the gate drive changeover from P to N channel. You'll need a scope current probe for that.

[Damianos]

On the last schematic:
- The V_BE of Q11 prevents the upper part to operate. It keeps the Q3 always low.
- The parallel B-E connections of Q10 and Q9 will "antagonize", which one will be more conductive.

Also there are different delays between the upper and lower part of the circuit, mostly because of the added inverting stage...

[Nikan]

Right and putting a resistor on  the base of Q11 will have a positive effect on Q3 etc. but it will also result in a bigger delay between the upper and lower part.

Any suggestion for fixing this problem Is more than welcome.

Does it matter if Q11 and Q9 are parallel?
For example the base of Q9 is low thus the resistance of Q9 e-c goes into infinity. Wouldn’t we be able to just think for example of a 1M Ohm resistor in parallel with Q11 e-c?

By turning Q9 on, the Base of Q1 and Q6 will always be low doesn’t matter which signal goes into Q11.
I hope that is right...

Thanks for all the help I really appreciate that! 😁

[T3sl4co1l]

What's wrong with an off the shelf dual driver IC with disable?  They're out there...

Tim

[Damianos]

I think that the best solution is this:

What's wrong with an off the shelf dual driver IC with disable?  They're out there...

Tim

[David Hess]

On the original schematic:

Cross conduction of Q4 and Q5 during switching is going to create a massive current spike from the supply to ground.

Q1 is going to fight Q3 through R2.  Why not us it to pull the base of Q2 and Q3 negative or do the disable in logic?

Q10 will invert the signal but its base-emitter voltage will limit the input to Q2 and Q3 to basically zero.

How are you planning on keeping the gate drive transformer out of saturation?

Common source output stages are tricky because of high and poorly controlled gate threshold voltage.  I would use bipolar transistors for the output for this reason.

What kind of output currents and edge speeds are you looking for?

[Mechatrommer]

What's wrong with an off the shelf dual driver IC with disable?  They're out there...
Tim

sometime a few transistors, resistors and diodes are alot cheaper and easier to get than the IC. although i will agree on the IC if small footprint is on the highest priority...

[ArdWar]

I'm not so sure with discrete circuit, switching MOSFET with 120nC Qg at 700kHz is quite a challenge even for integrated solution. Might be easier if you can tolerate long turn on/off time (the MOSFET might not, though).

[T3sl4co1l]

I'm not so sure with discrete circuit, switching MOSFET with 120nC Qg at 700kHz is quite a challenge even for integrated solution. Might be easier if you can tolerate long turn on/off time (the MOSFET might not, though).

Oh, is that how much it is?  Yeeeeeaahhh.... you can't do that with GDTs.

I mean, you can, with great effort refining the design of the GDT itself, and some beefy drivers.  (That's several watts of gate drive, remember.)

That doesn't mean you want to.

That's an isolated gate driver application right there.

Tim

[Nikan]

I've tried using these:
http://www.ixysic.com/home/pdfs.nsf/www/IXD_630.pdf/$file/IXD_630.pdf

I think these should be beefy enough... But I had a pretty bad gate drive signal on the mosfets.
Would a picture of the the gate signal and my setup help to find a good solution for my problem?

[rstofer]

I've tried using these:
http://www.ixysic.com/home/pdfs.nsf/www/IXD_630.pdf/$file/IXD_630.pdf

I think these should be beefy enough... But I had a pretty bad gate drive signal on the mosfets.
Would a picture of the the gate signal and my setup help to find a good solution for my problem?

Absolutely!  Scope traces are always a help.

You know the size of the capacitor you are trying to charge and you know how much voltage swing you need so you can calculate coulombs of charge as Q = C*v (in farads and volts).

Now you know how much charge (in coulombs) you need to transfer to turn the MOSFET on and you know about how fast you need to make the transition (in seconds) so you can calculate Amps as Coulombs/second.

Thirty amps seems like a lot but it isn't.  You want the transition to be as short as possible to reduce heating in the MOSFET.  Ideally, you want the transition to take zero time.  The driver you linked has a 20 ns rise and fall time (under the best possible condition) and it looks like they intend to have it operate at around 15 kHz.  There are graphs that show the rise and fall times to be MUCH longer under real world operating conditions.

You need to reflect on every graph and table in the datasheet and digest what they are trying to tell you.  Every word printed in the description comes with a caveat that is shown in the graphs.  They don't lie but they always paint the prettiest picture.  It takes time to figure out if the device is appropriate.  And a little math...

You really don't want any unnecessary resistance in the gate circuit in an attempt to slow down switching.  This leads to heating in the MOSFET because it doesn't fully transition fast enough.  If you need to deal with punch-through, do it ahead of the driver/MOSFET circuitry.  Once you gate the driver, it's time to go!  No messing around!

[T3sl4co1l]

Thirty amps seems like a lot but it isn't.  You want the transition to be as short as possible to reduce heating in the MOSFET.  Ideally, you want the transition to take zero time.

Well... not really.  Gamma ray bursts from a gate driver would be rather extreme.

Even ignoring the absurd case (literally zero ), there are very good reasons not to have transitions of, say, single nanoseconds.  Even besides device characteristics (you'll be somewhat hard pressed to find a gate driver, in stock, with less than 20ns risetime, or a transistor that can be pushed so fast besides) and EMI concerns, the physical layout of the circuit itself may prohibit such short time scales.

This is simply because, as current flow rises, part size rises, the number and size of bypass capacitors rises, and trace widths and lengths rise.  The dimension scale rises, and stray inductance rises proportionally.

The solution is, dividing the circuit into smaller pieces, which can handle it.  This is part of the reason why PC motherboards have so damned many phases supplying Vcore.

This is also why FM radio transmitters use many small modules with power combiners, rather than a single monolithic device.  The only thing that can handle that much power, in a single unit, at that frequency, is a vacuum tube!  But those are inefficient and less reliable, and therefore less cost-effective today.

And why large industrial modules are only available in IGBT flavor -- why bother with MOSFETs at all, when the part is so large that you can't possibly harness the speed advantage?  (Efficiency is a bonus, and the historical priority -- but no longer true with newer developments in SJ Si, SiC and GaN FETs.  Hmm, it'll be interesting to see if anyone tries introducing a power switching module that's GaN based, with integrated bypass caps and gate drivers.)

You really don't want any unnecessary resistance in the gate circuit in an attempt to slow down switching.  This leads to heating in the MOSFET because it doesn't fully transition fast enough.  If you need to deal with punch-through, do it ahead of the driver/MOSFET circuitry.  Once you gate the driver, it's time to go!  No messing around!

It is easy to find situations where increasing gate resistance reduces switching loss.

Incidentally, it's also easy to find situations where increasing stray gate drive inductance reduces rise time.  (That's just good old fashioned series peaking.)

Tim

[rstofer]

Thirty amps seems like a lot but it isn't.  You want the transition to be as short as possible to reduce heating in the MOSFET.  Ideally, you want the transition to take zero time.

Well... not really.  Gamma ray bursts from a gate driver would be rather extreme.

Even ignoring the absurd case (literally zero ), there are very good reasons not to have transitions of, say, single nanoseconds.  Even besides device characteristics (you'll be somewhat hard pressed to find a gate driver, in stock, with less than 20ns risetime, or a transistor that can be pushed so fast besides) and EMI concerns, the physical layout of the circuit itself may prohibit such short time scales.

I realize it isn't possible to get anywhere near 0 ns.  It isn't even possible to get anywhere near 20 ns with the proposed driver and a real MOSFET with any amount of gate capacitance.

The point is, don't take that splendid 20 ns rise/fall time in the description as a real capability.  Look up the gate capacitance and then look at the graphs.

Yes, fast rise times produce EMC but slow rise times produce heat.  Somewhere in the middle seems good...

[Mechatrommer]

instead of using 500V part on 12V circuit, maybe he can use lower V part with lower gate capacitance, hence larger current capability such as IRFZ44N and its PMOSFET counterpart... anyway, what kind of (high current?) transformer that needs sub MHz switching frequency? i'm imagining the transformer must be very small...

I am using 8x IRFP22N50A.

[Nikan]

The Fullbridge is going to be used on rectified mains (220V) so 500V Mosfets are the minimum I wanted to use...
The coil is not that small but the inductance is pretty low... It is an air-coil.