These transistors sure do scream...

"Winfield Hill" <hill@rowland.harvard.edu> wrote in message 
news:o8i5h60lhj@drn.newsguy.com...
> You should have two sets of drivers and gate
> transformers, one each for high and low sides.
> Then you can create an adjustable deadtime.
>
> Another attractive idea: pos/neg pulsing to
> turn a set of MOSFETs on or off, with the gate
> capacitance storing the state.  Combine fast
> on/off times with low switching frequencies.

What you can't see about the driver is, I've already done that. :)  It's 
actually an astable driving a pair of monostables driving a pair of gate 
driver ICs (TC4420).  So, one side pulses up, other side stays down; other 
side pulses up, one side stays down; etc.  Pulse width is ~2.5us, repeat 
frequency low ~kHz.

Output is square-ish as long as it's dominant capacitive.  At low 
frequencies, the 10M probe bleeds it down, of course, leaving a rounded 
wave.

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com 

On Tue, 21 Feb 2017 21:51:06 -0600, "Tim Williams"
<tiwill@seventransistorlabs.com> wrote:

>"John Larkin" <jjlarkin@highlandtechnology.com> wrote in message 
>news:3tnoacd1vfrgrlgb6l5gcf7tiqr7eoejma@4ax.com...
>> Another way to get insane edge speeds is avalanche transistors. A
>> modest stack of the Zetex SOT23 parts could switch a couple of KV in a
>> couple of ns.
>
>Yup, though not too much current before life expectancy plummets.  I can't 
>imagine it's too easy connecting them in parallel either (yes, they can be 
>triggered, with reasonable jitter, but that's a lot more bother).
>

They seem to have good lifetimes. You can't fire an avalanche stack at
a very high rate!

There is a Zetex appnote about using them in series and parallel. The
cool thing about a series stack is that a small pulse, TTL class, can
trigger the bottom one, and the whole stack zippers.



>I did a minor study of transistors I have laying around; few did not exhibit 
>latching avalanche discharge, but many were finicky (the region of base 
>resistance vs. collector current for pulsing behavior was small, and 
>inconsistent between samples).  None really seemed to be more powerful and 
>faster than the usual tiny suspects (like 2N3904).
>
>Avalanche also spreads out poorly -- so a very large transistor (like a 15A 
>1500V HOT) only 'ignites' in some random spot location, and becomes damaged 
>at basically the same surge current as a 2N3904.  (Afterwards, instead of 
>collector leakage, there's C-E resistance, usually on the order of 40kohms. 
>Characteristic of a microscopic burn hole.)

The Zetex parts are designed to avalanche. They are made in Russia,
probably on an ancient diffusion line. They have really low Fts, like
40 MHz, which good avalanche transistors seem to.


-- 

John Larkin         Highland Technology, Inc

lunatic fringe electronics 

Aaaand...it's dead.  Oh well, fun while it lasted. :^)

https://www.seventransistorlabs.com/Images/HighVoltageBridge2.jpg

No visible damage, but the three high side are ~shorted.

Just a show of the last version: 10pF Miller caps, which kicked dV/dt up to 
17k/us or thereabouts (that's 100ns at 1.7kV).

Tried 220 ohm damper resistors in series with the 10p's, but that wasn't 
actually helpful/useful (actually, may've been harmful? no idea).  (The 
ringing was something else.)

Ferrite beads (type #31) on the gates, and then bigger ones on the gate 
drive pairs (emphasis on the middle ones, because, guess why!), fixed the 
100-400MHz screaming pretty nicely, as confirmed by near-field loop 
"sniffer".

I realized early on, my conspicuous lack of high voltage, wideband probes... 
well, that won't do.

So....I know!
https://www.seventransistorlabs.com/Images/HighVoltageProbe.jpg
10 x 1M 0805, in parallel with (20p + 487), and then 110k || (varicap + 
49.9) at the bottom.  Requires 1 meter coax and 15pF || 1M scope.

Probe circuits are traditionally drawn with "speed up caps" across the 
divider resistors, and nothing else, but that's dumb.  It's an impedance 
equalizer, going from ~megs at DC, down to whatever the transmission line 
ratio needs to be at HF.  It looks like a capacitor at middle frequencies, 
and a resistor (of one value or the other) at the asymptotes.  In this case, 
100:1 into 50 ohms is 5k, and 5k/10 is 500, so the distributed resistors are 
500 ish.

Failure to observe the correct design process results in heavily ringing 
waveforms (the coax is shorted by caps at either end, so resonates like an 
inductor), and in mistakes, like having to solder your resistors in 
after-the-fact in a layout that wasn't made to fit them. :-)

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com


"Tim Williams" <tiwill@seventransistorlabs.com> wrote in message 
news:o8egdp$65c$1@dont-email.me...
> Playing with a switching circuit, yet I seem to have made the observation 
> that these things are fantastic for linear.
>
> Infineon SPA07N60C3, but everyone has their line of SuperJunction MOSFETs.
>
> Circuit, for posterity:
> https://www.seventransistorlabs.com/Images/High%20Voltage%20Bridge.pdf
>
> Intended load is high impedance, capacitive.  It can easily source 5A peak 
> though.
>
> Since the load is capacitive, it operates in hard switching.  The 
> transistors start singing as soon as they get into the Miller plateau. 
> Which in this circuit, I've intentionally exaggerated (27pF D-G), to help 
> keep the transistor voltages matched.
>
> A word about SuperJunction transistors: Coss tanks by two decades, over 
> the 5 to 20V range.  Very nonlinear, brutal.  This is fantastic for 
> switching converters, because it "cushions" the switching edge, doing a 
> better job of snubbing than an external network ever could.  By pushing 
> all the Miller effect to the low voltage end, switching loss can be very 
> low.
>
> With stacked transistors, that works against me, because they'll probably 
> be mismatched in the low-capacitance region.  So the switching times, and 
> voltages, probably won't be matched, forcing much more voltage across just 
> one over-performing transistor.
>
> So I increase Miller capacitance, so the rise is slower, and more linear.
>
> And to protect against accidental turn-on or damage, due to opposite side 
> hard-switching or output sparks, I put zener diodes on G-S.  (Back-to-back 
> pairs, since the drive is transformer coupled.)
>
> I think between the zeners and the Miller cap, I've got a particularly 
> nasty loop that makes a wonderful oscillator.  In the 200 to 400MHz 
> region, depending on which transistor you ask.
>
> (Ferrite beads on the gate leads solves the oscillation, more or less.)
>
> I'm definitely going to try an RF amplifier with these, soon.  I can't do 
> very much power, because of thermal limitations, and bandwidth won't be 
> fantastic because of the high load resistance versus Coss (note that the 
> load resistance has to be high, i.e. the supply voltage high, and because 
> of power limits, the current relatively low, to stay in the low-Coss 
> range). The useful frequency range seemingly should be worthy of vacuum 
> tubes, though!  Assuming lead parasitics don't trash it first, which is 
> likely. :)
>
> Tim
>
> -- 
> Seven Transistor Labs, LLC
> Electrical Engineering Consultation and Contract Design
> Website: http://seventransistorlabs.com 

On Tue, 21 Feb 2017 22:05:09 -0600, "Tim Williams"
<tiwill@seventransistorlabs.com> wrote:

>"John Larkin" <jjlarkinxyxy@highlandtechnology.com> wrote in message 
>news:ic7pacphaoie5kcv79k4e5vbj7pp1d692e@4ax.com...
>> The old RadLab radar pulsers were cool. They used the stored energy in
>> a transmission line to make rectangular pulses, with a single switch.
>> I've made some very pretty pulses with a coax line and avalanche
>> transistors: very small and simple circuit. I'm thinking one could
>> merge the pulse storage line with the step-up-transformer function.
>
>A friend of mine has been taking apart some modest-vintage radar equipment 
>(civil, ATC I think).  He's got a handful of hydrogen thyratrons and hard 
>glass modulators (transmitter tetrodes, with fucking huge cathodes) from it.
>
>We've been thinking about pulse generator applications, like making a 
>retro-high-tech Tesla coil.
>
>I suspect they use hydrogen thyratrons in EFT generators, but I've not taken 
>one apart.
>
>Some day, I'd like to design and build an EFT tester, just because.  But 
>getting a hydrogen thyratron would be too easy, and not really all that 
>helpful.  I'd rather do it with an IGBT and shock line: could be very cheap 
>and highly available, only needing custom wound inductors (saturable 
>reactors, really).  Don't know how easy it will be to adjust to make a clean 
>pulse, though...
>
>Tim

Spark gaps?

Krytron!

https://dl.dropboxusercontent.com/u/53724080/Tubes/Kry_Danger.jpg


-- 

John Larkin         Highland Technology, Inc

lunatic fringe electronics 

"John Larkin" <jjlarkin@highlandtechnology.com> wrote in message 
news:md5qac1s0ekk8j5visjol7qndiolh9slv1@4ax.com...
> They seem to have good lifetimes. You can't fire an avalanche stack at
> a very high rate!

You maybe.  I've blasted a 2N3904 at 100kHz. :)

And I said with current.  This is in the spec sheet too: you don't get too 
many peak kilowatts, before the life curve goes from "she'll outlive me" to 
"you have one second of run time, better make it important".


> There is a Zetex appnote about using them in series and parallel. The
> cool thing about a series stack is that a small pulse, TTL class, can
> trigger the bottom one, and the whole stack zippers.

Yup.  Might also be neat to build a Marx generator with them, though again, 
you have the problem of current.  Any transmission line you hang on the 
output is going to be hundreds of ohms, and after a few kV, you get into 
transistor-popping currents, no problem.

And then the problem spirals geometrically out of control.

To make something like IEC 61000-4-4 EFT, you need about a hundred of the 
poor things, in series-parallel.  Such an innocent noise signal -- but it's 
brutally spiky.


> The Zetex parts are designed to avalanche. They are made in Russia,
> probably on an ancient diffusion line. They have really low Fts, like
> 40 MHz, which good avalanche transistors seem to.

It's too bad HOTs are no good.  They have really low fT and were made on 
ancient diffusion lines (or maybe some epitaxy too).

That at least leaves some HOTs useful for SRD, as you've found.

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com 

"John Larkin" <jjlarkin@highlandtechnology.com> wrote in message 
news:2v5qactcbg6qhmcqfpjch982tu50np4scq@4ax.com...
> Spark gaps?
>
> Krytron!
>
> https://dl.dropboxusercontent.com/u/53724080/Tubes/Kry_Danger.jpg

Much shorter lifetime -- though, in their intended use, a slightly bigger 
bang. ;D

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com 

On 22/02/17 16:05, Tim Williams wrote:
> "John Larkin" <jjlarkin@highlandtechnology.com> wrote in message
> news:md5qac1s0ekk8j5visjol7qndiolh9slv1@4ax.com...
>> They seem to have good lifetimes. You can't fire an avalanche stack at
>> a very high rate!
>
> You maybe.  I've blasted a 2N3904 at 100kHz. :)

Find that appnote Tim. They'll avalanche 60A/20ns for >10E11 cycles.
See e.g.
<http://datasheet.octopart.com/FMMT415TD-Zetex-datasheet-8374939.pdf>

Clifford Heath.

On Tuesday, February 21, 2017 at 7:53:43 PM UTC-8, Tim Williams wrote:
> <mrdarrett@gmail.com> wrote in message 
> news:21c519f7-9b06-4b72-b950-94a3a4cea273@googlegroups.com...
> > Nice!  Did you draw this with Autodesk's Inventor?
> >
> 
> GAH!  Choke!
> 
> No, Altium.  The real deal. ;-)
> 
> (You might recognize some shapes and default colors from Win's drawings as 
> well; he uses an old copy of Protel, IIRC.  The ground symbols haven't 
> changed a pixel!)
> 
> Tim
> 


Oh, nice!  I'll have to check Altium out.  :)

Your schematic looked a bit Inventorish.  I took an AutoCAD class at the junior college last year, and the prof also made us use Inventor a few times.  I found it a bit cumbersome, and I just preferred AutoCAD.  

Thanks,

Michael

"Clifford Heath" <no.spam@please.net> wrote in message 
news:58ad2e7b$0$32564$b1db1813$19ace300@news.astraweb.com...
> Find that appnote Tim. They'll avalanche 60A/20ns for >10E11 cycles.
> See e.g.
> <http://datasheet.octopart.com/FMMT415TD-Zetex-datasheet-8374939.pdf>
>

2N3904 likely isn't as robust, but also, notice how much it drops with pulse 
width -- returning to the 61000-4-4 example, the nominal pulse width is 
50ns, but that's only to the 50% level of an exponential decay.  Diodes Inc. 
defines pulse width as half-cycle sinusoid,
https://www.diodes.com/diodes-part-files/DJ/FMMT417/Application%20Notes/201342.pdf
so the equivalent duration would probably be even longer (100ns?).

That simple hyperbola seems to suggest a deeper truth: that the conductive 
channel has constant voltage drop, and the delivered energy needs to be 
limited to a constant.

Let's see.  If the curve is an exponential decay, with a half-life of 50ns, 
then the time constant is 72ns, and the total area under the curve, divided 
by the amplitude, is simply the time constant.  So it's more than 50ns, but 
less than 100 at least, not quite as bad as I had been thinking.

There is quite a large disparity between the "no failure" and modest-life 
curves.  For this curve, it's about 16A vs. 60A.

Which would be switching impedances of 19 ohms, and 5 ohms respectively, 
which is pretty damn low!  And, a stack of ten gets you 3kV into 50 ohms --  
supplying up to 10 hours of continuous EFT duty!

Not as bad as I remember calculating before.

Still, if you need long life, the point about parallel strings stands --  
that's a ~4x reduction in load current, so you need a four strings of ten to 
switch it.

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com 

Tim Williams wrote...
>
> (You might recognize some shapes and default
> colors from Win's drawings as well; he uses
> an old copy of Protel, IIRC.  The ground
> symbols haven't changed a pixel!)

 No, I use the latest rev of Altium Designer.
 However my PCB engineer, Chuck Fisk, who
 retired a few years ago, stayed with Protel.
 Re: colors, I'm constantly honing my choice.
 I dislike Altium's default ground symbol,
 but haven't figured out how to change it.


-- 
 Thanks,
    - Win