What research have you done; have you referenced IEC 61000-4-4 itself?
For example, this figure comes from Indian open standard IS 14700, available here:
Unfortunately, they don't provide component values; at these speeds, device selection and circuit layout are critical anyway, so it's kind of not a big deal, in that you'll have to figure out much more than just this cartoon of a diagram. (This is unfortunate, but a reality of EMC standards; ESD also gives a cartoon (a simple RC circuit), yet it is shown to produce a double-peaked waveform!)
These are kind of maybe your first hints that, while you could go the effort to make one... you're probably better off just renting the machine for the amount of time it'll take you to make one, let alone calibrate it to the standard. (Not that calibration is needed for precertification testing, but a rental brings that confidence as part of the package.)
This is the waveform to target. (A note that your screenshots show only a couple pixels of this aspect of your waveform, if indeed they do at all -- hopelessly zoomed out.)
We can determine a few things about this already:
- It must have half amplitude (±10%) into a 50 Ω load (relative to open circuit peak voltage) and the waveform shown.
- It must have full amplitude (±20%) into a 1 kΩ load, same tr, td -15 to +100 ns.
- Cd shall be 10 nF (±20%).
- Energy is sourced from Cc, so if we let Rm = 0 and Rs = 50 Ω, we get a 25/50 Ω load for Cc and switch (which seems agreeable, td in the 1 kΩ case would merely be +45 ns). Then, U ≥ Vp (enough to cover switch losses) and Cc = 2.9 nF.
- Cc and Switch can be swapped, depending on desired output polarity, and at minor expense to the idle level between pulses (because charge current then flows through the load; the idle level is not specified so I don't think this is important to the standard). Having a common-ground switch is likely of great value here, as making a truly floating switch, at these voltages and speeds, is not simple. (Output polarity can be swapped with a transmission line transformer, so a fixed configuration is fine.)
- Vp up to 4 kV is suggested, so we need to handle on the order of 23 mJ, and 160 A peak. Switch will not be trivial, especially at these speeds.
- tr limits stray inductance to a maximum of 3 nH or so (inversely proportional to Cc). (Note the asymptotic value can be higher than this, if the inductance is a property of transmission line sections; but we still need the waves on those sections synchronized so that the pulse isn't smeared out, for the same effect in the end.)
We can relax the energy and current a bit by increasing Rs, but also keep in mind Rm will have to trade off with switch losses/speed to get the required output impedance (basically they want a well behaved ~50 Ω source).
We might consider additional transformers, for impedance matching, and as power combiners, to use multiple channels in parallel. (The alternative is a large cascode, which is likely to be troublesome both in operation (5 ns isn't much time to distribute drive voltages along it), and even a single component is likely to exceed the stray inductance, let alone a stack of them (about four will be needed, considering commercially available ~1200 V MOSFETs).
At this point, we can clearly see device characteristics will be a limiting factor. We might now ask: what can we get from a given MOSFET?
Quick survey of what's cataloged at Digi-Key:
- MOSFETs ≥ 4000 V: maximum 2 A rating. Likely too slow besides.
- 1000 < VDS < 4000 V: ooh, there's a GaN part, GPIHV30DFN, a likely pick. Not much among Si; IXFH32N100X is about as good as they get. SiC is the most promising, with some voltage range, and lots of parts available (the tech is pretty mature these days); G3R75MT12J, UF3C120040K3S, G3R160MT17J and others are likely suitable.
- Under 1000 V: I'm not going to sort through these at the moment, but although the voltage is lower, the performance can be better so it's not an automatic exclusion. SiC are readily available in 1200+ V so this would only be relevant for GaN (more mature in the ~600 V class I think) and Si. I would not consider anything less than 600 V as low enough inductance simply won't be possible to meet a low load impedance, matching transformers get more difficult to build with higher ratios, and too many parts will be needed.
Incidentally, another device is capable of producing such pulses, but at much lower power levels: BJT avalanche breakdown. When avalanched in a certain way, effectively the base layer is punched through, momentarily shorting C-E; a 2N3904 can go from blocking about a hundred volts, to sinking a couple amperes peak, within a nanosecond. But the amount of power we need here (~40 kW) means hundreds would be required, and you'll never synchronize that many plus wind all the power combiners needed to build such a beast.
This is only the beginning of an overall system design, but these two angles of attack (the waveform; potential devices) can can be filled in further, meeting in the middle with a promising final design.
Note that you don't gain much from the pulsed nature, as any transistor still needs to handle the peak current, with low enough voltage drop to deliver most of the supply voltage to the output. That is, you're basically making an amplifier here (maybe neither exactly linear nor switching, but some manner of large-signal operation to be sure), and full operating capacity is required during the peak; the only thing you really skimp on is the average power dissipation (heatsinking).
Only two* other devices are capable of this function: a sparking contact itself (what's ultimately being modeled here; but calibration of an air gap is impossible), and a hydrogen thyratron. Some vacuum tubes may also be suitable but they will all be more trouble to use than a thyratron. Note that general purpose thyratrons (usually Hg or Xe fill), and thyristors, are not suitable: they operate in the microsecond range at best. They are suitable for IEC 61000-4-5 style surges however, and are indeed commonly used in such generators.
*Fine, there are some arcane devices out there, like pulse x-ray commutated silicon switches, or vacuum spark devices; but these are not generally available, or not without significant budget (or licensing!). Hydrogen thyratrons are probably on the obscure side as it is, but are occasionally available surplus as far as I know.