Open source EFT generator (and other transients)

[T3sl4co1l]

Gauging interest on this topic:

I think it's feasible, potentially even with junkbox parts (albeit needing a fair bit of repetition).  Off-the-shelf is doable with stocked, reasonably priced parts.

What range of waveforms and ratings would be desirable?

I think IEC 61000-4-4, 2kV (measured into 50 ohms) would be pretty reasonable.  More is always possible just by combining more generators, it's just a pain.  I think even getting to 2kV will require some combination already, and going higher requires quadratically more (double voltage = quadruple power, etc.) so it doesn't scale real well.

I'm not sure about ESD.  The voltages are much higher, and a spark gap would be most attractive.  Good luck calibrating that, though.  I suspect commercial units use vacuum relays (but I haven't taken one apart to see, heh).  Which are available, but they're not exactly weekend-project prices.

Also, at some point ITAR comes into play; I don't remember exactly what rate or amplitude of pulse is currently restricted.  Note this applies to sales, not publication -- it's not a direct limitation on an open-source project.

Tim

[NiHaoMike]

Joe Smith made one that goes up to 15kV or so, but I don't recall there being any documentation of how it was done.

[capt bullshot]

I'm not sure about ESD.  The voltages are much higher, and a spark gap would be most attractive.  Good luck calibrating that, though.  I suspect commercial units use vacuum relays (but I haven't taken one apart to see, heh).  Which are available, but they're not exactly weekend-project prices.

Yes, they use HV relays (reed type, maybe mercury wetted, maybe vacuum don't know exactly). One can hear them operating, it's the typical reed relay "pling".

[Marco]

How do they do 100 kHz with a relay?

With a 50 Ohm output impedance and 4 kV you need "only" max 80 A. What's wrong with abusing some SiC mosfets a little beyond their datasheet specs?

[T3sl4co1l]

How do they do 100 kHz with a relay?

ESD gun, not EFT.

With a 50 Ohm output impedance and 4 kV you need "only" max 80 A. What's wrong with abusing some SiC mosfets a little beyond their datasheet specs?

Exactly. It's an eminently doable project; filling in the pulse generator, drivers, and output networks is the bulk of it.

And, what I mean by salvaged parts is, I've got a FQA9N90C here doing 500V in 10ns (into 22 ohms); similar ratings are a common sight in flyback supplies.

The LISN projects on here seem to have been a modest success, I wonder if other EMC equipment might prove interesting.


Other devices that could fit the bill:
- Hydrogen thyratron.  Quite specialized, though can get lucky finding them in surplus radar units.  Radar is essentially the same peak power level as EFT, just replace the PFN (shorter pulse width, less flat pulse) and add a matching transformer (since the thyratron will be at full ratings at a somewhat higher load resistance than 50 ohms).
- Spark gaps.  100% All Natural(R).  Good luck getting any consistency with it, though?  Amplitude can't be controlled, needs attenuators for control (rated for the peak power!).
- Hard vacuum tubes: maybe.  Definitely won't be cheap, or widely available.  I haven't been able to get more than 3kW peak from a TV tube; you're basically looking at transmitter types here.  Or modulator tubes -- again, radar hardware FTW.  Biggest downside: the load impedance needs to be rather high, maybe 500 ohms or more, so the bandwidth drops very quickly, and the magic 5ns rise time may quickly disappear.  Ceramic (planar and radial beam) types have been used in physics before (klystron drivers, beam diverters, etc.).
- Avalanche transistors (BJTs).  These don't do all that much power by themselves, so don't scale well.  A MMBT3904 might do a few hundred watts; thousands would be needed, synchronized perfectly.  FMMT417 and friends are good for more power, but hundreds might still be needed.  They are rather expensive too (avalanche ratings are very niche), maybe the regular commercial equivalents can be used as-is, or with qualification testing, who knows.  Risetime may suffer due to timing errors; I'm not sure how reliable avalanche triggering is.  Like a spark gap, these can't be throttled; needs relays (to change how many pulsers get connected to the output) or attenuators.

Speaking of bandwidth, peaking networks can be used to extend it somewhat, on the amplifying types (vacuum tubes, MOSFETs) -- but I am not aware of any means to extend the bandwidth of a capacitive source, into a resistive load, more than about 40% (namely, using a series peaking network -- tee coils only apply for capacitive sources and loads!)

Tim

[Mazo]

Definitely interested in the topic and possibly building such device.

[TimNJ]

Practicality, cost, and design work aside, my dream EFT generator would be 4KV @ 100KHz repetition rate. IIRC, 4KV is now the standard (?) for medical products, or is at least expected for modern power supplies/products.

[T3sl4co1l]

Hype

[coppice]

Practicality, cost, and design work aside, my dream EFT generator would be 4KV @ 100KHz repetition rate. IIRC, 4KV is now the standard (?) for medical products, or is at least expected for modern power supplies/products.

4kV is now the minimum for most serious applications. If you want to supply to extremely dry places, people often demand 8kV testing for some products. However, 4kV (preferably 4.5kV so you can test just a little higher than 4kV, and ensure your results aren't marginal) is what you need for most applications.

[temperance]

Perhaps you might find this interesting:

https://www.fordemc.com/

And this:
http://www.highenergydevices.com/technical_resources/application_notes/index.html

[chris_leyson]

Nice project Tim, I wish you the best of luck. At work I usually test L, N, PE, L+N, L+PE, N+PE and L+N+PE at 7kV or 6.5kV for L+N+PE. It's a customer requirement to test at 8kV but they let us test at 7kV. The Emtest EFT generator will only manage 7kV. Most testing is done with 5kHz burst but sometimes 100kHz. I would like to take the covers off to get a sneak peek but it's not allowed  . I've often wondered how they generate fast pulses with around 1kV/ns risetimes and how do you measure that anyway ? Not with a scope probe for sure, 1kV/ns will give you 1A peak into a 1pF load. As far as I know the fast transients are capacitively coupled into L,N or PE via 10n capacitors, Emtest drew a simple block diagram on the front panel. EDIT: As the EFT generator is quite flexible as far as voltage,  pulse frequency and burst length is concerned I'm guessing they use avalanche devices or maybe thyratrons and probably not relays.

[T3sl4co1l]

Like I said, could be a hydrogen thyratron. *shrug*

Or the same kind of thing again (MOSFETs), just enough wired together to get the amplitude.

Probing is easy.  Just get an attenuator rated for the peak power, voltage, whatever.

For a crude (qualitative) measurement, use a capacitive divider -- put the pulse on an exposed wire and hold a scope probe in its general proximity.  Can use a known pulse generator to calibrate it, making it quantitative.

I was thinking 2kV, as that's a common commercial test level -- and I don't intend to compete with, y'know, full test lab capability -- if you need many kV, you probably need something more reliable than a weekend project!

Tim

[chris_leyson]

Not even sure where you would buy thyratrons these days. I think they must use a string of HV transistors or more likely mosfets with transformer coupled drives. Need to have a stab at a 100:1 probe with very low input capacitance. EDIT: 2kV would be a good place to start, a few of the EMC test guys think that 1kV is probably good enough but I have had some flyback controllers pass at 1kV or 2kV but let the magic smoke out at 3kV or 4kV.

[Marco]

Unless you use the G2R120MT33J you'd probably already use a stack at 2kV with MOSFETs ... and once you start using a stack you might as well design for a higher stack. That said, just using a relatively expensive transistor to ease the design is of course an option.

PS. actually 2 kV into 50 Ohm means switching 4 kV, so you'd need a stack either way.

[coppice]

I was thinking 2kV, as that's a common commercial test level

Do you find many customers still only asking for 2kV EFT testing?

[penfold]

It sounds like a pretty good case for a MOSFET stack and all can be pretty well isolated with a pulse transformer, that's all companies like Behlke do in their switches

[mrjoda]

Hi,

I have some experiences with burst/surge testing and I have passed training from Langer EMV-Technik in Dresden in good old times, when we were able to travel

Related to bursts. In short, there are two ways how to do it. First - charge capacitor and discharge it via relay contacts. Another way is to use repetitive avalanche breakdown of semiconductors. Burst are not so critical. Basically, any HV generator with ANY frequency (5kHz or 10kHz is more then enough for testing) with standard distribution will do the job. Another topic are probes...

For surges, charge capacitor (standard specify 12R 9uF or 2R 18uF ) and discharge via thyristor or big IGBT. Surges are quite "slow" (50us +/-) and therefore easier to generate (dwell time specified in standard is around 20s!).
 

[T3sl4co1l]

Yeah, surges are a no brainer, just use enough SCRs.  I even have such a generator on my shelf... it has a big old stack of them inside.

One tricky bit is the coupling network.  IEC 61000-4-5 isn't quite a resistive source, its shape depends on load: 8/20us into a short, 1.5/50us into an open.  The ratio of those peaks gives the 2 ohm impedance -- a peculiar definition of impedance for sure, but it's roughly the right idea.  Anyway, one might wonder what kind of network is needed to do that.  Well, that's not too hard either:



Tim

[mrjoda]

T3sl4co1l

Is there any synchronization of those SCRs necessary ? bcs they will not switch ON in one exact time. Is not there any issue with di/dt or some sort of repetitive avalanche breakdown ?

If you take 20x SCR in series, they will react "randomly" every time when you switch them (of course random means ns delays) but it can be enough for breakdown or local overheating/cracking of SCR die.


or it is not an issue at all and my brain is overthinking.     

Thanks !

[T3sl4co1l]

The surge network limits dI/dt, according to the 8us risetime of course.

Note that synchronization is an issue not just between SCRs, but over the area of a single SCR as well.  Adequate dI/dt and Ipk ratings are required, and strong gate drive is preferred.

This issue is much more critical for EFT and other fast pulses like that.

Tim

[temperance]

I think a series connection of SCR's can be used to generate very fast pulses. Trigger one or more SCR's and let some of them go into avalanche breakdown.

This one looks like a good candidate: S4070WTP

https://www.littelfuse.com/products/power-semiconductors/discrete-thyristors/scr/sxx65x-sxx70x/s4070wtp.aspx

[T3sl4co1l]

I've tested avalanche on power transistors.  The problem is formation of a current filament.  In short, between avalanche breakdown, and localized heating, the carrier density rises to the point where the base layer becomes intrinsic, and therefore a BJT effectively experiences punch-through in the local area.  It ceases to be NPN, and becomes NIN or something like that.  This explains the extreme current capacity, and the long saturation time, of avalanche operation.

The current filament is only able to spread as wide as carrier diffusion takes it in a given time frame, and I suppose thermal diffusion as well.  The point of failure (Ipk at given dI/dt?) is essentially constant across junction size, probably being higher for more Vceo (wider junction = wider current filament).  Or if anything, a bigger junction fails with less external capacitance, because it has to discharge its own junction capacitance first.

The failure mode I measured, on line output transistors (Vceo = 1500V), is the device goes resistive.  Transistor action works essentially as normal, but there is a high resistance (10s kohm) short C-E, or C-B or something like that, I forget which way exactly.  So after failure, it can't cut off properly.  Which is as you would expect for a point failure, probably something like microscopically melted or vaporized die.

I have no reason to expect an SCR to behave any better; in particular, due to the multiplication effect of its multiple layers, and the low transient voltage drop while turning on, it takes precious time for conduction to spread out across the die.  This sets the dI/dt rating of the device.

If you consider ~1us to be fast, absolutely, SCRs are great -- they are indeed perfectly suitable for surge generators like that.  As far as I know, they are unsuitable for fast pulses as in the OP.

Tim

[temperance]

Just toying around with the idea. Unfortunately I can not comment on your observations only that current spreading in the device will be a problem.

Searching for SCR's working in avalanche breakdown I found this:
https://patents.google.com/patent/US3513328A/en

Micro semi has an ultra fast SCR:
https://www.microsemi.com/product-directory/silicon-controlled-rectifier-scr/5230-upga301a

I have no idea about what's inside a modern EFT generator.

[Marco]

The rise time of that SCR is still in the range of 50 ns, fast enough for EFT ... not fast enough for ESD. But MOSFETs are already fast enough for EFT and easier to work with.

A mercury relay is the easiest way to do switched ESD, avalanche transistors and various forms of triggered sparkgaps are used in other applications because they have lower jitter ... but jitter is hardly relevant here.

PS. or maybe just do a mechanically driven sparkgap using a plastic rod from the trigger to bring two rounded bolts together, bit noisier though.

[T3sl4co1l]

The rise time of that SCR is still in the range of 50 ns, fast enough for EFT ... not fast enough for ESD. But MOSFETs are already fast enough for EFT and easier to work with.

EFT is 5ns rise though??

Impressed that it has no dI/dt limit.  Downside, quite low voltage rating so you have to stack a lot.  Let's see, 100A into 50 ohms is 5kV, you'd need more than 50 in series to handle that.  Not sure what kinda-fast surge would be on that order.  Would be adequate to drive a sharpening (NLTL) network; but then, a few IGBTs would do just as well, and are readily available.

Tim