Canl 2N3904 handle 1200V?

No great surprise there. The individual collecter-to-base leakage currents tend to be all over the place, so the device in the stack with the lowest leakage current sees most of the 1200V.

Stacking devices always seems to involve wasting a lot of current on making sure that each device sees much the same voltage drop.

I suppose you could set up an active scheme for keeping the voltage drops across each device more or less equal, but I've never seen it done.

Cascoded with a 6HV5 in grounded grid and I see no problem at all. ;)

I never stacked more than two and that worked fine, that is, if you could tolerate the large random pulse-to-pulse output voltage variations. I had no voltage balancing in the stack, so it's anyone's guess which transistor fires first.

Jeroen Belleman

Did you measure the current as you brought up the voltage?

The 3904 leaks 10uA at 60V. You are applying 120V/device (ideally). Please educate me on avalanche voltage.

The consistency seems pretty good, actually. (I spoke too soon, most or all the transistors are actually okay.) I've got 12pF at the top of the stack, which switches the cap into the output (50 ohm BNC). 500V pulse, 3.6ns rise time, 6ns width. A pretty solid 10A, not bad at all for the humble 2N3904. :)

Tim

On Fri, 8 Jul 2016 04:13:15 -0500, "Tim Williams" Gave us:

Is it not a simple task to perform a search for and find the spec sheet for it?

One must string the string back together for it to work.

Ah, OK. So much the better. It's a bit slow, no? Mine looks like it has a risetime in the 500ps ballpark. I think it's the scope (Tek 7104 with

I use a piece of UT141 coax as the top capacitor, so I get square pulses.

Jeroen Belleman

Vceo is just a recommendation -- a guarantee that won't get you into trouble.

Vces (if listed) is the actual "zener" voltage. Obviously, it's quite a bit more than Vceo, in some cases!

Avalanche really is just the mechanism of breakdown. In a critically strong electric field, free charges accelerate fast enough to impact atoms (most often impurities, so it depends on doping level), which frees additional charges, resulting in a cascade just like a gaseous spark discharge.

Avalanche has a positive tempco, characteristic of "zeners" over 8V or so (true Zener breakdown has lots of leakage, a very soft knee, a negative tempco, and occurs under 5V; the best zeners are inbetween, around 6.2V).

BJT avalanche is just the same, though the leakage current flows into the base, and therefore gets multiplied by hFE if the base is unconnected (which is why Vcbo >= Vceo). There is the additional quirk, however: probably due to the base layer, there's a poorly compensated excess of free charges, which means huge excess noise when you use a BJT as a zener.

If you've ever operated a zener "starved" (i.e., at low current), you'll have noticed its voltage resembles a relaxation oscillator with a random threshold.

In fact, under the right conditions, the transistor can turn on so hard, that it fully discharges the voltage across the junction, plus anything external.

So, you can use a suitable transistor, in avalanche mode, much as you can use a spark gap, albeit one with rather large leakage (fractional mA).

Apparently not all transistors avalanche. 2N2369 is the classic (see LT's AN47, IIRC), and avalanches very easily. Most RF transistors do. 2N3904 isn't often appreciated as a candidate, but apparently it works. Adjusting the B-E resistor, and providing enough supply current, is the trick. 4.7k and 1mA seems to be typical for 2N3904.

Diodes/Zetex makes a selected variety, FMMT415TD, which has been tested and burned-in for avalanche mode. It's a bit higher voltage and current, and slower (couple ns), and stupendously expensive (~$8/ea?), but that can be better than paying for design spins, let alone selecting and aging standard components!

Tim

Part of it is because the stack is "lit" from the bottom. So, the first transistor (the one touching the output) fires, dumping a little voltage down the line, while tugging on the one above it. Which quickly breaks down, tripping the next and so on. As each one fires, it delivers charge in the output direction, but from further and further away, so the wavefront is naturally spread out, approximately by the physical length of the stack (in my case, about 10cm) plus the total rise times of all the transistors.

If you ignite it from the far end, the wavefronts all pile up -- the stack looks more like a shock line, and you get a rise time somewhat under 1ns.

If you need more voltage, you can even build a Marx generator style topology, so you don't need stupendously high supply voltages.

I'm kind of tempted to make an EFT generator, so, the rise time /should/ be a little on the slow side -- saves me the trouble of filtering it! (Spec is

I'd like 2kV output for it, and it looks like I'd need four times the voltage, which is 16 times the peak power. And five times more energy (pulse duration), on top of that!

I have a box with HV supply in the bottom, and a pair of BNCs on top. One is the pulse line, so choose-your-own. :) It goes about as fast as my scope, I think.

Tim

Wow! Thanks for that, Tim. Good stuff! I have a lot to learn.

What's the circuit? Toasting the current limit resistor sounds normal to me, if there's enough current to keep the stack latched.

Series transistor avalanche stacks are common. Zetex makes SOT23 avalanche transistors and has appnotes on series stacks.

Those Zetex transistors are specifically grown to avalanche. I was told that they were made on an old Russian diffusion line. They can avalanche with very low charging currents and a small base trigger pulse to the bottom transistor in the stack. The Zetex's essentially saturate when they avalanche, clamp to close to zero volts. More modern epitaxial transistors don't avalanche well.

We've found some rectifier diodes, and the c-b junctions of some HV power transistors, that accidentally have the right diffusion profile to make great drift step-recovery diodes. There are probably devices that accidentally avalanche well, too.

The Tek 7S14 dual-channel sampling plugin used an avalanche transistor to make the sampling pulse, no step-recovery diode. It was spec'd for

It also used mercury cells to back-bias the sampling diodes, so don't expect old ones to work.

On Fri, 08 Jul 2016 09:04:16 -0700, John Larkin Gave us:

Zetex makes very good transistors.

I think they are Diodes Inc now.

They have some parts with astounding betas. AoE3, page 501, lists a bunch of Zetex parts in the low-noise-BJT table, including the winner, ZTX851 at 0.18 nV/rthz.

On Fri, 08 Jul 2016 12:06:46 -0700, John Larkin Gave us:

Used it on a 15kV HV supply that you could fit 6 of into a pack of cigarettes, was human contact medically approved, had overcurrent sense shut downs in it for that purpose and was for Batelle inc. for use on a medical device where all the bastards wanted was for us to develop it so they could filch it and have someone in China make millions of them for them. The product never did make market, but was astounding physical and medical technology.

Tim, is that what you are doing? Except for the bottom all the bases tied to the emitters? Are you only using the C-B junction? Does the b-e junction do anything? (any idea how much current flows out of the base and how much out the emitter?

George H.

They have been for some time.

They also had one of the only monolithic dual transistors (no more).