PCB transmission line transformer

one layer will be better, with 6 interleaved layers you have the L2-L3 and the L4-L5 parasitic capacitance working against your goal,

with the coax transformer how much coax does it take before you don't need a core?

In wide layers, the HF currents will be forced to the outer part of the winding, still resulting in high impedance. I'd suggest to etch smaller parallel lines on the 'wide' winding to counter this skin effect.

joe

If a secondary trace is sandwiched between two primary traces, it becomes a stripline of sorts. The impedance of the inner, referenced to the outers, gets lower as the traces get wider or the dielectric gets thinner. That's analogous to the impedance of the coax used to wind a transmission line transformer. Which, I guess, doesn't matter too much is the winding is short compared to the risetime of the signal.

Something like that.

I guess I could float the gate driver electronics on the mosfet source (and power that somehow, kilovolts off ground) and run the transformer at signal levels. That's less interesting (aka less risky.)

Not to change the subject, but I've been exploring mosfets for fast kilovolt switching. Most kilovolt-level power fets kill you from gate charge requirements and/or source lead inductance. SiC looks like a winner until you see the internal gate resistances, 5 ohms or so, too much to allow you to drive the gate hard and fast.

GaN is nice, but too low voltage. Maybe a GaN gate driver into one of those planar Ixys mosfets is the best one can do. My customer wants

If the mosfet equivalent of beta is peak Id/Ig, at these speeds the number is approaching 1.

The coax would have to be really long to sustain wide pulses, and the the prop delay becomes a problem. I don't entirely understand this, but it seems to me that the output impedance becomes the Zo of the coax when the coax is long relative to the pulse rise time.

Here's a version we did, but only for 200 ns 5-volt pulses.

We've also done some 2:1 step-up versions, to put 100 volts into 50 ohms, worked fine.

This is fun, a picosecond speed pulse inverter.

The step response is flat for about a ns, until the generator wakes up and realizes that it's shorted. If you slip a ferrite over the coax, it handles much longer pulses. PSPL sells this in a box, for some kilobucks.

You may want to think that over. A piece of coax is a near- perfectly coupled transformer, because of its symmetry. In a sense, it's a pair of perfectly interleaved conductors. They have the same centroid. Both centre and shield conductors follow the same path and therefore see the same outer flux, at least for frequencies high enough for the shield to act as a shield.

One of the limits on transformer bandwidth is the ratio of coupled to leakage inductance. Common ratios are of the order of 10^3, making for transformers with 3 decades of bandwidth. I've made transformers having 6 decades of bandwidth, implying a coupling factor of the order of 0.999999, using coax. The absolute value of the leakage inductance was in the fractional nH ballpark. No way you're ever going to do that with wire, let alone striplines and similar. There are other parasitics that also limit transformer bandwidth, of course, so this number is probably somewhat conservative.

Jeroen Belleman

What did you think I meant with my question: "Inside a transformer?"

Anyway, the path to making transformers with good coupling factors is symmetry. Both conductors have to cut the same common-mode flux. A design with a narrow trace running over or along a wide trace is doomed to be poor.

The trick to limiting dispersion is designing the transmission line such that all e-fields are confined to a (low-loss) dielectric with the same epsilon everywhere.

Jeroen Belleman

ve wound them from micro-coax on ferrite toroids, with the shield being the primary and the inner conductor the secondary. Sub-ns speed and low leakag e inductance. But it's labor intensive.

. Layers 1/3/5 could be one to three layers of spiral trace, primary, and 2 /4/6 ditto, secondary. I'm not sure how to think about the impedances, but it ought to have wide traces and thin dielectrics, I guess.

indeed that: Low leakage inductance. Using PCB transmission lines instead would yield higher leakage inductance, maybe similar to what you'd expect of twisted-pair windings, maybe a little worse.

plane above and below a buried trace - isn't dispersive. The field spreads sideways to some extent, but it dies away very rapidly with distance, and s hould be quite a lot better than twisted pair. If the odd-numbered layers i n John's multiplayer PCB were all ground planes, strip-line traces on all t he even numbered inner layers (not the last) would be pretty innocuous.

One doesn't have to know much about magnetics to know more than John. In th at sense we have any number of magnetics experts around here.

The late Tony Williams made a lot of his money out of making special purpos e transformers, and I was enormously gratified by once being able to give h im advice - on a ratio transformer - that he found useful, but that's as cl ose as I've ever come to being a magnetics expert. I do know enough about m agnetics to look like an expert until a real one shows up.

That sounds a bit odd. Bifilar wound transformers are famously closely coup led, and the bifilar winding is just a single length of twisted pair.

Leakage inductance is flux that threads one coil without threading the othe r, and the thing about ferrite cores and shields is that they offer a much lower reluctance path to the flux for the path that goes through the middle of the whole coil and returns around the outside the whole coil.

That's the point about transmission-line transformers - the capacitance for ms part of the transmission line.

At high frequencies very little of the flux gets outside the transmission l ine.

They do couple, but the sign of the coupling reverses with every twist, so the couplings cancel pretty exactly.

I did a transmission line transformer with miniature coax in about 1989, an d it worked fine but was going to be a swine to manufacture, so - to avoid the obvious question - I built a version with twisted pair, and was surpris ed to found that it worked just as well. Better actually - the twisted pai r was a bit thinner than the coax, so I could put on more turns and get a l ower low frequency limit. This did mean that the twisted pair winding was a lso longer, so the high frequency limit dropped from 500MHz to about 150MHz , but both were much higher than I needed

urn

I'm wondering about more elaborate transmission line structures. Two parall el tracks are a transmission line, but one with a relatively large external field.

If you stacked two such lines, one above the other, and tied the diagonal p airs together, the external field would fall off faster.

A hexagon arrangement, spread over three layers of a multilayer board, is e ven sillier. Of course, if you put in a central track, and drive the six ou tside tracks together, you've just built a crude - but possibly useful - ap proximation to a coax cable.

John wants 4kV isolation, so maybe a square "screen" on - say - layers 1 an d 3 with an "inner" on layer 2 might work.

Didn't have a clue. I guess it should have got me thinking - but it took "shorted turn" to get my attention.

Then twisted-pair transmission line transformers wouldn't work. Mind you, that's probably because the asymmetries cancel out.

So make the layers out of one of the Rogers low-loss printed circuit substrates, intended for microwave work. They tend to be a lot more uniform than FR4, and less lossy.

And you might be able to make more symmetrical transmission line structures if you coupled together more than two or three traces.

The coupling of course is in regards to a conventional (two section "shell", or let alone bank wound) construction, which doesn't couple nearly so well.

Yes, and there's plenty of space between the wires of a transmission line -- if there were none, it wouldn't transmit (or rather, it would violate causality -- no inductance means infinite speed of light). Analogously, no material has an infinite speed of sound, or stiffness (though people often say "water is incompressible", which has the same error or approximation).

There are many cases where a (single element) transmission line transformer won't suffice; high frequency power conversion is an easy example. Put simply, the circuit impedance is lower than most transmission line structures or winding constructions, hence LL dominates.

Correct, particularly literal in the case of coax. But it's also not mutual inductance: in the case of coax, the flux that loops around the inner conductor necessarily does not also enclose the outer conductor. Thus, fully describing leakage inductance.

The external flux (corresponding to current flowing on the shield) is what's doing the transformering.

I may've overstated the impact. Last I recall measuring one, the leakage was consistent with the length of twisted pair used. Capacitance should be higher, given the proximity, and since the proximity is to subsequent turns, it will be dispersive, at least in the common mode.

Indeed, you go from dipole to quadrapole and so on. It would be a big transformer to have the edge-to-edge distance in the far field though (say,

Each facing pair of traces crudely looks like a parallel plate transmission line in isolation, so that a stack of N layers of alternating traces (P, S, P, S, ..) looks like N-1 transmission lines in parallel and thus N-1 times lower impedance: same as using a very wide transmission line, except not. (And also saving on edge effect, which was mentioned earlier.)

The layers will inevitably couple between each other, which I think should increase the impedance slightly, but not by nearly as much as the construction reduces it to begin with.

Pretty neat:

Tim

What, I thought MOSFETs were "the harder you drive them, the faster they go"? ;-)

I doubt they make SiC much slower than regular types. Rg and Vgs are higher, and Qg is smaller. It all scales.

Superjunction FETs are pretty amazing, at least above 20V or so. I imagine you wouldn't have a very good pulse generator using those, if you ran into the saturation region. Perhaps some sort of level-shifted pre-Cds-catastrophe active clamp thingy to make a sharp pulse without fully saturating. And plenty of opportunity for "interesting/risky" in that driver circuit thing.

I take it your customer is a bit more picky than the level of, say, avalanche chains? And the voltage is higher than, say, step recovery can provide? But that's a silly limitation, because 4kV single junctions are available. And they're slow and probably have really s***ty recovery. Maybe another research project. Interesting. Risky.

Or if you can find SiC diodes in that range, too. The junction (not schottky) types are just as awful as silicon, for instance, the performance of SiC BJTs, or MOSFET body diodes. AFAIK, they are constructed exactly the same as their Si counterparts, just in different material.

Tim

Until you blow out the gate, yes.

The small SiC chips make the gate resistances high, according to the appnotes. A 40 amp 1200 volt silicon mosfet may have a gate resistance of a couple tenths of an ohm, and a corresponding SiC may be 6 ohms. And the Qg advantage might be around 2:1. At nanosecond switching speeds, you can't pump the gate fast enough, through that 6 ohms.

A Zetex avalanche chain would be great at 100 Hz. It would melt at

And the voltage is higher than, say, step recovery can

Lots of diodes are unintentional drift-step-recovery (Grehkov) diodes, and if there isn't a 4KV part somewhere, one could series some 1400 volt parts. There are some high-voltage power transistors whose c-b junctions make great DSRDs.

Water-cooled 2KV 400KHz DSRD pulser:

Schottky diodes don't do the step-recovery thing; diffused silicon PNs seem to have the best doping profiles.

What I may wind up doing is talking the customer down to something more reasonable, 2-4 ns. The sine-squared transfer function of a Pockels cell speeds up the optical risetime a bit.

I guess we may as well get some SiC transistors and bang the gates until we blow them up. Then we'll know. The Cree, Rohm, and Ixys parts seem to be remarkably similar.

Cool, big airline coax.

ok. you said how fast in the original post, but not how slow.

as for delay:

what if you do this

.------> gate | drive >-------------(|) | | | | \/\ coax longer than shown. \/\ | | | | --------(|) __|_ | //// `-----> souce

What if you put a diac in series with the gate - would that make a latching mosfet?

Only works at low frequencies. A large diameter open wire "ladder" transmission line would probably work equally well, but would require very big baluns at both ends.

I'm also wondering about the losses. At low RF frequencies, skin depth and resistive losses predominate. I don't think there's enough copper in the wires to be really low loss. I've built broadband wire cage dipoles where such losses almost became a problem. At megawatt power levels, any transmission line losses can be a serious problem.

I wouldn't worry (much) about the wind. The wires should all swing in unison. For wide gaps, spreaders can be added. However, I suspect the real danger is some bird landing inside the cage, resulting in an RF arc barbequed bird. Flying branches might also be a problem, but it looks like they've cleared all the trees for miles. Rain would not be a problem, unless the insulators get both dirty and wet.

l",

l.

-- if there were none, it wouldn't transmit (or rather, it would violate caus ality -- no inductance means infinite speed of light). Analogously, no ma terial has an infinite speed of sound, or stiffness (though people often say "water is incompressible", which has the same error or approximation).

er

You can build transmission lines from printed circuit traces with a rather wider range of impedances than you can get with coaxial cable and twisted p air.

Since the transmission line is the transformer at high frequencies, the lea kage inductance simply doesn't the kind of problem that you seem to be imag ining.

You do have to be careful about the lead dress as you enter and leave the t ransformer structure - that was the area where I managed to tell Tony Willi ams something that he didn't already know about transformers, much to my su rprise and gratification.

ual

t's

Wrong. It's the internal flux - in the region between the inner and the out er - that's doing most of the work. Induced voltage is L.di/dt and at high frequencies, di/dt is very high, so you don't need much L.

be

ns,

Again, the capacitative currents alternate and largely cancel.

y,

for

With transmission lines there shouldn't be a lot. Threading the structure w ith occasional - spiral - ground lines might help a lot. Low frequency tran sformers make use of screens to minimise inter-winding capacitances.

One standard technique - albeit expensive - for minimising cross-talk in pa rallel data cables is just to ground every second wire at each end of the c able. It lowers the characteristic impedance and makes the cables harder to drive, but what comes out at the receiver end is a lot cleaner.

on

S,

This coupling will be strictly second order - current flowing in screen cre ating resistive voltage drop in screen which then induces current in adjace nt screen. It's a real - but not usually detectable - effect in transmissio n line transformers.

Not a problem if you are putting all the traces on a multilayer printed cir cuit board stack.

wound them from micro-coax on ferrite toroids, with the shield being the p rimary and the inner conductor the secondary. Sub-ns speed and low leakage inductance. But it's labor intensive.

Layers 1/3/5 could be one to three layers of spiral trace, primary, and 2/4 /6 ditto, secondary. I'm not sure how to think about the impedances, but it ought to have wide traces and thin dielectrics, I guess.

ill be better, with 6 interleaved layers you have the L2-L3 and the L4-L5 p arasitic capacitance working against your goal.

B inductors?

prop delay becomes a problem. I don't entirely understand this, but it see ms to me that the output impedance becomes the Zo of the coax when the coax is long relative to the pulse rise time.

The coax doesn't have to be long to sustain wide pulses. The joy of the tra nsmission line transformer is that the ferrite cores make it a conventional transformer for the low frequency content.

In fact the transmission line transformer stops being a flat, frequency-ind ependent coupling when the pulse width gets down to close to equalling the length of the transmission line. Winfield Hill rubbed my nose in that here many years ago - much to my delight. I'd been puzzled as to why my twisted pair transmission line transformer had been falling over at 150MHz when the coax version was good to 500MHz, and the difference was just the length of the winding.

worked fine.

I published the same idea - in this case for level-shifting - back in 1979

Ghiggino, K.P., Phillips, D., and Sloman, A.W. "Nanosecond pulse stretcher" ,Journal of Physics E: Scientific Instruments, 12, 686-687 (1979).

I didn't think the circuit was worth publishing, but Ken Ghiggino wanted th e publication, and wrote the first version of the paper. I had to revise it extensively to get it published. Ken was a chemist - now professor of Phys ical Chemistry at Melbourne, where I did my Ph.D. - and he hadn't touched a ll the bases.

GaN cascoded into a MOSFET = fast, HV MOSFET?

Cheers, James

Maybe. At the point that the peak gate current approaches the peak drain current, somewhere south of 2 ns maybe, the GaN will have to sink twice the drain current. It might be better to just ground the mosfet source and use the GaN as the gate driver. A 7 AM, not having had any coffee yet, I could well be wrong.