400 V Square wave oscillator

[sdouble]

hi guys,
I need to build a very fast switcher. I do need to switch a capacitor  (actually parallel plates used to deflect ions).
Only 400 V, so not that much but I do need  to switch from 0 to 400 V in 20 ns.
I was thinking about using a GAN FET in cascode configuration. I have the impression that the Transphorm product could fit my needs.
Does someone have some experience with those products ?
thanks in advance.

[sdouble]

forgot to mention that the oscillator needs to operate at 2.5 MHz

[NiHaoMike]

Maybe use a RF amplifier tube?

[calexanian]

What kind of current and what kind of load are you driving. Gonna be really hard to get a sharp square wave at those frequencies at any appreciable power level.

[daqq]

Behlke has some modules that might help you in this.

[ahbushnell]

Wolfspeed makes 900 V SiC MOSFET's that should be fast enough.  You need to think about ringing. 

[sdouble]

Behlke has some modules that might help you in this.

thanks, i also saw the Behlke modules. They look pretty interesting. They are more "high voltage" oriented (ie 2+ kV) as far as I could see

[sdouble]

What kind of current and what kind of load are you driving. Gonna be really hard to get a sharp square wave at those frequencies at any appreciable power level.

I have essentially a capacitive load (35pF, 25 nH). I agree that they are issues like ringing, EMC, current handling. The good thing is that I don't have to  deal with significantly high voltage.
The neat thing is that I can cope with moderate ringing.

[daqq]

That's true about the Behlke modules, though there are some low-ish voltage devices. Down to 500V if I remember correctly.

[kridri]

You can het inspired by looking to some older crt schematics. Especially the high speeds scopes have a broadband 'high voltage' (around 300-400V) x and y deflection drivers.

[T3sl4co1l]

Nothing special.  Any modern MOSFET will do, no need for GaN, SiC or exotica.

If it's a capacitive load (just an electrode deflecting ions?), you don't need DC gate drive.  The transistors can be driven from a transformer.  Very easy at 2MHz!

If you need adjustable duty cycle, then you need a DC restorer circuit as well, no big deal.

Tim

[sdouble]

any modern MOSFET will switch (on and off) 400 V in 20 ns ?
not really sure 

[ganzuul]

The on/off time here is 2000V/us. A higher-voltage transistor should be faster when it operates closer to its linear range, signified by its gain -3db point.* However, that means you need a HV PSU and snubber diodes for when something goes wrong.

Is 2.5MHz cycle repetition rate or the highest harmonic of the square wave? If you need very little ringing at 2.5 MHz you might want those snubbers, but depending on the resistance that could mean your 170MHz load oscillation frequency is breathing down your neck due to the high voltage, necessitating a filter.

Is this a chopper for an ion beam, a cyclic device, or a single-shot kind of deal?

* This might not be clear. This is what I mean:

[sdouble]

The on/off time here is 2000V/us. A higher-voltage transistor should be faster when it operates closer to its linear range, signified by its gain -3db point.* However, that means you need a HV PSU and snubber diodes for when something goes wrong.

Of course, I have both devices  in stock. 

Is 2.5MHz cycle repetition rate or the highest harmonic of the square wave? If you need very little ringing at 2.5 MHz you might want those snubbers, but depending on the resistance that could mean your 170MHz load oscillation frequency is breathing down your neck due to the high voltage, necessitating a filter.

Is this a chopper for an ion beam, a cyclic device, or a single-shot kind of deal?

Sorry , it seems that I was not very clear. The 2.5 MHz is the repetition rate. It's actually a chopper for an ion accelerator.

[capt bullshot]

any modern MOSFET will switch (on and off) 400 V in 20 ns ?
not really sure 

Yes, they do, even faster:
http://cmosedu.com/jbaker/papers/1992/MST31992.pdf

Just for fun, I had build a version for 900V / 6.5ns fall time:
http://wunderkis.de/pulser/pulser.pdf
http://wunderkis.de/pulser/

Somewhen later I tried this circuit with a modern SIC MOSFET instead of the IRFBG30, yielding a much slower fall time (don't remember the number right now, maybe 20...30ns)

[T3sl4co1l]

I did an industrial inverter, 5kW 400kHz, with middle-aged MOSFETs (VDMOS, 900-1200V, not SuperJunction -- so their performance sucked!).  650V edges in 50ns.  Good thing I didn't go with optoisolators, that's more dV/dt than any I've seen can handle.

Win Hill (of AoE fame; he posts on sci.electronics.design) has been working on a higher voltage pulser, using SiC MOSFETs.  I think that was 10ns for the whole edge (a few kV).  SiC aren't really much faster than regular Si, they're just higher voltage (so, higher gate and drain voltages).

Tim

[sdouble]

did you estimate the power efficiency of your circuit ?
I need mine to be quite efficient. It will operate in vacuum "slightly" hampering the natural convection. I foresaw a water cooling for the power transistors, but still, I'd prefer the circuit to be as efficient as possible.

any modern MOSFET will switch (on and off) 400 V in 20 ns ?
not really sure 

Yes, they do, even faster:
http://cmosedu.com/jbaker/papers/1992/MST31992.pdf

Just for fun, I had build a version for 900V / 6.5ns fall time:
http://wunderkis.de/pulser/pulser.pdf
http://wunderkis.de/pulser/

Somewhen later I tried this circuit with a modern SIC MOSFET instead of the IRFBG30, yielding a much slower fall time (don't remember the number right now, maybe 20...30ns)

[sdouble]

Win Hill (of AoE fame; he posts on sci.electronics.design) has been working on a higher voltage pulser, using SiC MOSFETs.  I think that was 10ns for the whole edge (a few kV).  SiC aren't really much faster than regular Si, they're just higher voltage (so, higher gate and drain voltages).

Tim

I was more thinking about a GAN FET . Those are actually faster with very low RDSon.
Another point that I didn't raise is that the board will be in a harsh environment (SF6 gaz around and some radiation). THerefore, I'm after a solution which would be as robust s possible. That's one of the reason for which I looked at integrated modules, with, possibly, integrated cooling.
There will be no chance to open the tank to change easily a burned MOSFET. Will take days before being allowed to have access to the board.

[BrianHG]

This one is a little pricey, but it has the low and high side driver built in with protection & it slightly exceeded your requirements:
http://www.ti.com/product/LMG3410?keyMatch=lmg3410&tisearch=Search-EN-Everything

If price is an issue, don't bother as it is a fancy slick component.

[T3sl4co1l]

Hmm, radiation might be more of a concern for modules?  SiC or GaN might be better than Si, too, with suitable arrangements to have the driver at a distance, or shielded properly.

SF6 is rather inert stuff IIRC, unless you mean it's particularly nasty from the radiation (radicals), which I can see being a possibility.

At worst, use mil spec hermetically sealed parts (SiC FETs are available in gold-lid TO-220-lookalike packages for about $100 each), or seal up the circuit (with something nice like N2 or Ar) inside a tin.

Tim

[sdouble]

This one is a little pricey, but it has the low and high side driver built in with protection & it slightly exceeded your requirements:
http://www.ti.com/product/LMG3410?keyMatch=lmg3410&tisearch=Search-EN-Everything

If price is an issue, don't bother as it is a fancy slick component.

Reliability is muuuch more important than price. Shuting down the accelerator cost ten grants.... opening the tanks ten more....
thanks for the datasheet. THis is a part that I did not know.

[sdouble]

No idea whether the module would be more prone to failure than individual components. I would guess that this depends more on the transistor technology than the integration of the circuit.
a Moderate shiedling is foreseen. It won't be 100% efficient though.
I was again a bit fast (bad habit, my wife said it too last night). The board will operate in vacuum. The SF6 will be close by. To have access to the board, we have to evacuate the SF6 into dedicated reservoirs fill the tank with air and only then open the tank. Strangly, Gold is not so good for me. Gold loves neutrons... inducing gamma ray emission.

Hmm, radiation might be more of a concern for modules?  SiC or GaN might be better than Si, too, with suitable arrangements to have the driver at a distance, or shielded properly.

SF6 is rather inert stuff IIRC, unless you mean it's particularly nasty from the radiation (radicals), which I can see being a possibility.

At worst, use mil spec hermetically sealed parts (SiC FETs are available in gold-lid TO-220-lookalike packages for about $100 each), or seal up the circuit (with something nice like N2 or Ar) inside a tin.

Tim

[ahbushnell]

This one is 8 ns switching speed. 
http://www.wolfspeed.com/media/downloads/949/C3M0120100J.pdf

Note you can control oscillations by adjusting the gate resistor.  It does increase switching losses.  The energy stored in the output capacitance of the switch will be dissipated in the device.  At 400 V and 2.5 MHz I get 8 watts. 

There is a 900 V device that's a  little faster in similar package.

Andy

[ahbushnell]

No idea whether the module would be more prone to failure than individual components. I would guess that this depends more on the transistor technology than the integration of the circuit.
a Moderate shiedling is foreseen. It won't be 100% efficient though.
I was again a bit fast (bad habit, my wife said it too last night). The board will operate in vacuum. The SF6 will be close by. To have access to the board, we have to evacuate the SF6 into dedicated reservoirs fill the tank with air and only then open the tank. Strangly, Gold is not so good for me. Gold loves neutrons... inducing gamma ray emission.

Hmm, radiation might be more of a concern for modules?  SiC or GaN might be better than Si, too, with suitable arrangements to have the driver at a distance, or shielded properly.

SF6 is rather inert stuff IIRC, unless you mean it's particularly nasty from the radiation (radicals), which I can see being a possibility.

At worst, use mil spec hermetically sealed parts (SiC FETs are available in gold-lid TO-220-lookalike packages for about $100 each), or seal up the circuit (with something nice like N2 or Ar) inside a tin.

Tim

Vacuum could be a problem at 400 V.  This will depend on how hard the vacuum is.  This is driven by Paschen curve.  It's a function of the gap x pressure.  The minimum is around 1 to 5 torr-cm. The minimum is around 300 volts.  If you can get down to under 0.1 torr-cm then it looks like you would be ok.  I'm not sure what low pressure would do to electronic packaging.   

https://en.wikipedia.org/wiki/Paschen%27s_law

[capt bullshot]

did you estimate the power efficiency of your circuit ?
I need mine to be quite efficient. It will operate in vacuum "slightly" hampering the natural convection. I foresaw a water cooling for the power transistors, but still, I'd prefer the circuit to be as efficient as possible.

Yes, they do, even faster:
http://cmosedu.com/jbaker/papers/1992/MST31992.pdf

No, this circuit isn't suitable for you purpose, since it switches the whole 900 Volts into a 50 Ohm resistor. It's usable for short pulses and low duty cycle. My circuit operates with 1us wide pulses at 100Hz repetition rate.
To generate a square wave, I'd suggest a half or full brigde architecture. Should work with standard MOSFETs like the IRF840 used in the original paper. As Teslacoil said, gate driving should work with a simple transformer driven by a beefy driver like the IX4427 I've used.
Power dissipation will then be rather low, mostly dissipating your loads charge as abushnell pointed out. Plus the MOSFETs parasitic capacitance, which would be higher.
You might also consider building a full bridge running at low voltage and stepping up the output to your desired voltage using a custom build transformer. 2.5MHz shouldn't be too difficult for a pulse transformer. This saves you a lot of parasitic capacitance switching at 400V.

BTW, @Teslacoil: 650V in 50ns is just 13kV/us - nothing a modern gate driver optocoupler can't handle, they're usually spec'd for at least 25kV/us