Okay, I had some 12v 5W zeners lying around and it seems to work. I will try to implement 1.5KE68A because it's easily available for me in radio shop. Do I need resistor in series with diode, because it gets hot quick? Will resistor affect the suppressing of voltage spikes?
Well, you'll need quite a stack of those 12V zeners, but if you don't mind putting them together, no problem.
Resistor, nah. Mainly just pick a transistor (and thus zener/stack) rating high enough to avoid most of the energy that's supposed to go to the secondary anyway. Like magic said, you want fairly high voltages here.
What about avalanche rated MOSFETs?
Some of them really seem to suggest that they can take anything as long as Tjmax is not exceed (and you know it from transient thermal impedance plots). Not sure if it means they can be used as high power zeners at DC too. But they surely promise to withstand operation with unclamped inductive loads.
The non-rated types just have inferior handling (in terms of peak current or energy), I think a hotspot issue?, which happens to trade off directly with body diode t_rr. Rated types usually handle full energy, but not at "full" current: usually the avalanche rating is close to the DC current rating, but it's also nowhere near Id(max,pk). Due to the lower current, the inductance is also usually quite large and so the pulse relatively slow (but that's fine; find the spot on the transient thermal resistance plot and you should see the energy checks out, and so it stands to reason, the energy would be smaller if the current could be higher*).
*Or, they compromise and don't tell you the ultimate (limiting) current, but inflate the parameter somewhat as a result. No idea.
I don't recall if it's strictly due to deposited charge (so, the current is irrelevant, even to currents higher than given in the datasheet, as long as pulse width is short enough), or if there's some curve to it? Or of course, if hotspotting can't be avoided completely, so there's still a limiting current.
Anyways, the mechanism is hot carriers depositing in / disrupting / damaging the gate oxide. So, I guess, a similar mechanism to EPROM/Flash wear? (Not sure if it causes Vgs(th) shift in the same way. But, there's no floating electrode or ion migration for that to happen with either, I guess.) It's cumulative, so, at low enough currents and duty cycles, maybe it lasts long enough to make a viable product; most likely for anything more frequent than electromechanical switching, it's not going to survive.
Hence why the repetitive figure is so small; I'm not sure what count they test/estimate with, a million pulses maybe? Whereas the single figure is maybe good for a couple total or something. Not a thermal figure where it can do it forever at low enough duty (long enough time between pulses) (or, give or take thermal stress anyway).
A maybe giveaway is they never use avalanche in protected MOSFETs; instead they use a zener strapped from G to D. Fair, that might have to do with the integrated environment -- avalanche in the power section probably sends free charge carriers throughout the die? (Or at least neighboring sections of the control circuit.) Plus the channel is more than strong enough to do the job (i.e., Id(max,pk) at least), so it's not like it's a poor solution.
Tim