Designing a High-Voltage Nanosecond Pulse Generator

[Brandon Curtis]

Hey all,

I am attempting to build a high-voltage repetitive pulse generator with the following characteristics:
• Output square-wave pulse at 5kV (can reduce to 2kV if impossible or prohibitively expensive)
• Pulse peak width of 10ns (non-negotiable)
• Rise time of <5ns (non-negotiable)
• Pulse repetition frequency of 30kHz (can reduce to 10kHz if impossible or prohibitively expensive)

I have done quite a bit of research already and I have narrowed down the design choices:
• the high-voltage square-wave requirement suggests a Transmission Line Pulser (TLP)
• the high pulse repetition frequency suggests an avalanche transistor

I've also looked at these options and they don't appear to meet the specifications:
• Mercury-Wetted Reed Relays
• Silicon-Controlled Rectifiers (SCRs) and other thyristors

In the meantime I have just begun researching other possibilities:
• MOSFET cascode topologies (looks promising; not sure how to scale to 5kV... possible with a voltage ladder?)

Figure 4 in this paper shows a schematic for an avalanche transistor TLP, so I'm really looking to adapt that to much higher voltages. **broken link removed** (warning: paywall) says:

Power MOSFETs driving a 50 R load can be switched on in less than 3 ns when driven with an avalanche transistor.”

Their example circuit is limited to ~1kHz, and my understanding of what they're doing and how they're able to apply such high voltages to the MOSFET is insufficient for me to tell if this design is scalable to 10-30kHz.

While researching, I've become aware that there are MOSFETs out there with perversely high ratings: IXZR08N120B, many IGBTs. Wikipedia's description of IGBTs in general sounds pretty promising:

The extremely high pulse ratings of second- and third-generation devices also make them useful for generating large power pulses in areas like particle and plasma physics, where they are starting to supersede older devices like thyratrons and triggered spark gaps. Their high pulse ratings, and low prices on the surplus market, also make them attractive to the high-voltage hobbyist for controlling large amounts of power to drive devices like solid-state Tesla coils and coilguns.

Buying a $20,000 HV pulse generator is not an option; I am a graduate student with limited means. Ideally I can DIY this for less than $250.

I have variable high-voltage supplies available from 5-25kV.

Thanks in advance!

 

[chuckey]

What load are you going to use? Why do you want a PRF so low? I would consider synthesizing it from a series of RF sine wave generators. If the pulse width is 10 nS, and the rise time is < 5nS then its hardly a square wave. So build a 100 MHZ RF transmitter, use tuned circuits to step to output up to your 5 KV. Sniff of a bit of 100 MHZ, triple it to 300 MHZ and build another transmitter of 1/9 the power of the first, used tuned circuits to get your 5KV/3 which you add to the 100 MHZ output. Use a phasing circuit in the "sniff" feed to get the 300 MHZ bit in the correct phase (reverse transformer connections if required). Repeat as required with 5 X 100 MHZ...
Frank

 

[Brandon Curtis]

What load are you going to use? Why do you want a PRF so low?

The output will be applied to electrodes to yield a glow discharge used to study high-energy plasma-catalyzed chemistry. The PRF is purposefully kept very low to reduce the total energy dissipation in the spark gap and keep the gas temperature as close to ambient as possible.

The waveform I'm looking for looks something like this:


Rise and fall times should be as low as possible; I only specified <5ns because that is what a collaborator was capable of doing with a $25,000 industrial high-voltage function generator.

A transmission line pulser with a mercury-wetted reed relay would be perfect, if only the relay could support a higher PRF!

 

[dick_freebird]

The rep rate does not seem high to me, although it could
be high for high voltage switching devices.

Are you against a transformer based scheme? Since the
duty cycle is quite low, a pulse transformer ought to be
pretty well behaved and you could use more capable (for
high speed devices) provided your transformed current
is adequate.

If I were playing around, I'd get a car ignition coil or a
TV flyback transformer, either is well capable of the
voltage and the rep rate is also consistent with the
TV flyback horizontal rate (within an octave anyway).
Winding your own 1000:1 transformer is probably a
drag.

But there are outfits selling magnetics for this use and
at low current, pretty cheap.

Whether you'd get pulse shape fidelity out the back, I
couldn't say. But banging a car coil with a high speed
logic gate, ot a ton of 'em in parallel, would be cheap
enough to try.

Now hanging "whatever" onto the transformer, or for
that matter any delivery system, is liable to really change
things, and that goes for your 'scope probe as well as
the driven load. You might need to consider the whole
assembly and its resonance; your pulse as described
does not seem that different from a half-sine. Maybe
you want a tuned secondary and an impulse primary
drive wave form. Or something.

 

[erikl]

Figure 4 in this paper shows a schematic for an avalanche transistor TLP, so I'm really looking to adapt that to much higher voltages.

In principle this is a well appropriate configuration (Transmission Line (TL), avalanche transistor chain, possibly with the Marx bank configuration). With a 5kV supply, however, the pulse current from a 50Ω TL will be 100A, so you have to find avalanche transistors which can stand this current. HV MOSFETs or IGBTs would be an option, probably.

**broken link removed** (warning: paywall) says:

Their example circuit is limited to ~1kHz, and my understanding of what they're doing and how they're able to apply such high voltages to the MOSFET is insufficient for me to tell if this design is scalable to 10-30kHz.

I think the loading of the TL can be done fast enough: 100kΩ*50pF TL capacitance would be enough for a 30kHz operation; the actual limitation, however, is power: a 5kV*100A*10ns pulse has just 5mWs, but @ 30kHz repetition rate this means a power transfer of 150W.

So you'd probably have to use TLs with a much higher impedance.

 

[FvM]

Required output impedance is one of the unspecified points in the question.

Commercially available kV&ns pulsers are mostly using MOSFET cascades. They should work for the intended risetime range. For fixed width squarewave generation, a TL pulser topology is in fact preferable.

A transmission line pulser with a mercury-wetted reed relay would be perfect, if only the relay could support a higher PRF!

Usual mercury-wetted relays have in fact two limitations:
- the switching time of about 1 ms, which is essentially set by the spring constant and mass of the contact. You might consider faster reed relays with respective stronger magnet drive, but I never heard about any.
- above 1 kV, current risetime detoriates due to preceeding gas discharge and/or partial vaporization of the mercury film after first contact closure. I remember a breakdown voltage of about 2.5 kV

In addition, reed relay action is affected by jitter in a µs range, which may be inacceptable for some applications.

 

[Brandon Curtis]

My temporary solution has been to use a CRT monitor flyback transformer. It doesn't give me anywhere close to a square wave, but I have been able to tune things to get very short pulses on a very wide range of Pulse Repetition Frequencies. Though this works for the most part, I want to learn more and try my hand at a TL pulser or cascaded MOSFET design.

the actual limitation, however, is power: a 5kV*100A*10ns pulse has just 5mWs, but @ 30kHz repetition rate this means a power transfer of 150W.

The impedance of a spark gap is quite high, so it is my understanding that the pulse current is not generally a limiting factor. To impedance match this on a TL pulser, I would definitely require a transmission line with a much higher impedance.

Commercially available kV&ns pulsers are mostly using MOSFET cascades. They should work for the intended risetime range.

Most power MOSFETs have rise times on the order of 20-100ns, for instance the 43ns quoted for the **broken link removed**. For a rise time of only a few nanoseconds, the MOSFET will need to be driven by an avalanche transistor. To get into the voltage ranges that I'm interested in, I will need to **broken link removed** (warning: paywall). Is it relatively straightforward to extend this design concept to stack, say, six IRF840's rated at 500V each to switch 2.5kV?

 

[dick_freebird]

You might check out the EPC eGaN FETs, if you can live
with a 200V rating; much snappier risetimes provided you
keep the driving gate impedance low. Whether you can
tune a cascode stack to survive robustly, is another
matter.

Perhaps a higher turns ratio and a faster FET would
give you a better tradeoff than a tall stack of slow
FETs and a junkbox transformer-of-opportunity. But
I'm not much good for magnetics.

 

[shahbaz.ele]

can anyone suggest me from where i can get a pulser
having following features
400v peak voltage
upto 20KHz PRF
rise time < 10ns
I need this for ultra sonic flaw detection

 

[kamranravian]

shahbaz.ele
yo can design your own.
main parts are
1. low voltage pulse generator
2. amplifier
3. transformer
4. controlling unit (uicrocontroller etc)

 

[shahbaz.ele]

thanks dear kamranravian
Actually I want to save my design time.
If anyone knows about the programmable pulser's avaliability please share.