Diode recovery pulse generator

reverse

carriers

everything

I've

(1.5kV,

Its easily swallowed up by any capacitance, ive used sub nanosecond samplers wich use a step recovery diode, puting a scope probe on the step recovery diode kills the sampling output, so I know its there but ive never been able to view it on a scope.

Colin =^.^=

Yup. SRDs are designed to do this. The doping profile has to be hyperbolic or something like that so that all the reverse charge gets used up at once. Plus, it needs to have low junction capacitance.

Getting below 100 ps isn't hard with a decent SRD. Below 40 is hard.

There is a dilemma: fast-switching SRDs store a small amount of charge, and for a healthy snap we want a lot of reverse current flowing at the instant it snaps. If we ramp up the reverse current too slowly, we'll deplete all the available charge while the current is still low, so the snap will be wimpy. The faster the SRD, the less stored charge, so the faster your fet must turn on and the smaller the series inductor must be. To get a 100 ps edge, the drive current risetime would need to be well below 10 ns maybe. Your circuit might not be fast enough to snap the speedier SRDs.

A couple of tricks:

Some varicap diodes apparently make decent SRDs.

Larger junctions, like rectifier diodes, don't snap well, but get better if, instead of DC forward bias, they are forward biased for a short time, 100 ns maybe, before being reverse driven. This is the Grehkov "drift step recovery" effect. In one of our pulse generators, we forward bias the diode +48 volts (no typo!) for about 80 ns before we slam it with a -400 reverse drive.

Some transistor b-c junctions will snap.

Ma/Com makes probably the only SRDs that are stocked by distributors, although they are fairly slow...

Most other people seem to make them to order, and don't usually stock packaged parts.

Of course, you need a fast scope, preferably a sampler, to resolve 100 ps edges, and a pure 50 ohm system or else a resistive divider probe, to avoid the capacitive loading of a scope probe. And a very tight layout, of course.

Email me your address and I'll send you a few of the Ma/Com parts to play with.

John

Eh, I'm not convinced about capacitance. Precise readings I'll do my best to make a good connection, but my probe isn't *this* bad- I can watch the

1-2ns pulse output of an avalanche generator with it, albeit not too pretty with all the wire I use to hook it up.

To John: I don't know what dI/dt my FET is getting, but it is at the very least

1A/10ns, given the 15 ohm load resistor I tested with. I tried 1 ohm (i.e., 15A), but things got too hot and loaded at the duty cycle (5 or 10%) to get anything good off of it. I'll rebuild that one-shot circuit I made to get a better duty cycle.

I don't have any varactors, that I know of.

I might try the short forward bias trick, but it seems to me that'll need two FETs, and careful timing to prevent them kicking each other in the groin. Might be a happier solution to that, I'll have to work it out.

Tim

--
Deep Fryer: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

"Tim Williams" a écrit dans le message de news:FIQeg.515$ snipped-for-privacy@fe04.lga...

recovery

pretty

(i.e.,

get

a

How can you achieve "at the very least" 100A/us with a 15V supply and a

200nH inductor? That's _at best_ 75A/us.

--
Thanks,
Fred.

When I worked at Cambridge Instruments we used step-recovery diodes to generate 100psec snap-off steps to test our 0.5nsec long electron beam pulses in the stroboscopic electron beam testers.

We bought the step-recovery diodes from Hewlett-Packard via U.K. distributors, and drove them with an H-P pulse generator. I had a plan to make a pulse generator board with a couple of BFR93 (NPN) and BFT93 (PNP) wideband transistors driving a BFR96 output transistor, but never got around to it. The BFR93 and the BFT93 are still available from Farnell for about a $1.50 each. The BFR96 seems to have been superseded by things like the BFQ19 ($2.50)and the BFG97 ($1.0) and Farnell now carries the 8 GHz BFG135 ($4.5) which might be interesting. I always had to put a "base-stopper" resistor in series with every base - something between 22R and 33R to stop the parts oscillating at a couple of GHz.

I'd been using 5GHz bndwidth transistors for some time - see

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

which owes a lot to good advice from colleagues at EMI Central Research.

For sources of step-recovery diodes see

--
Bill Sloman, Nijmegen

recovery

pretty

(i.e.,

get

a

I tried a similar thing as you some time ago, I put the circuit right next to the scope input and fed it through a 1pf capacitor, however I had the inductor in the grounded side of the diode, this gives a large negative pulse when the current is switched and a small positive pulse when the srd snaps, also you can put an ultra low capacitance schotky diode to rectify the pulse, with a bit of added bias and a very high impedance meter gives a very good indication. varying the current has a significant effect, I was using the output of an ultra fast comparator so was seeing 100ma compared to your 1 amp. I tried with many diodes but only one or two showed any significant pulse wich was way too small for my needs, I gave up after I found a Metalics 2ghz sampler wich is just a tiny module with a SRD and 2 very low capacitance diodes conected by 2 capacitors so small they dont measure on my 1 pf resolution cap meter. It made part of a nice 2ghz sampling unit for my scope.

I was going to hook the sampling unit up to the SRD circuit but never got round to it.

I tried simulating a SRD ciruit but spice only has storage recovery time not the transition time, however a normal spice diode exhibits the step recovery effect, in fact to model soft recovery diodes is harder, the effect of the junction capacitance seems to dominate the pulse height.

If you weant to create a short sharp pulse you can turn on a transistor in avalanche mode wich turns on extremly abruptly, an open circuit transmision line in the collector stores the charge for the pulse and determines its length, in fact you can easily generate nice clean 1ns pulses this way, I was using a bfy90 in a previous sampling unit from the ciruit wich apeared in wireless world.

heres a nice app note wich gives a circuit for this with some detail ...

Colin =^.^=

HP/Agilent/Avago quit selling SRDs about a decade ago, because their name doesn't start with "M".

John

The reference is to using cheap PIN diodes - HSMP-3820 - in a frequency multiplier, which is very much a step-recovery diode application.

Farnell still stock the diodes - order code 549-812 - and sell them for less than a dollar each, aolng with a range of packaged dual versions of the same diode.

I don't know how useful they'd be to the OP, but it looked interesting.

--
Bill Sloman, Nijmegen

The 1N4007 is often referred to as a poor man's PIN diode. If that's what it takes to make a SRD, you couldn't find one much cheaper:) Regards,

Mike Monett

"Fred Bartoli" wrote in message news:447be417$0$10255$ snipped-for-privacy@news.free.fr...

I'm estimating from the load resistor and the voltage slope, which is around

15V or I = V/R = 15V/15ohm = 1A in about 10ns. That was without the inductor; with, I would measure the same or better, *at the drain*, so I'll go check what it says at the resistor...

About 32V supply (the full supply, so current will be about 2A). It rises for a moment to 33V (probably G-D coupling), then it falls to about 14V in

10ns, then it drops another 10V in 15ns. The saturation region is pretty slopey too. I would have to guess parasitic ringing is breaking the smooth LR decay into what looks like a piecewise curve.

This is more like 2A in 25ns ~= 80A/us, about as you expected.

Tim

--
Deep Fryer: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Decided how to accomplish this. So I may build this:

The one-shots will probably be: Although too slow for the short forward bias delay you speak of. Plain old propagation delay through this circuit may fill that in, perhaps with some RC slowdownness to adjust it.

Theory: double pulse generated and sent to FET. Big L charges for a few microseconds, reaching some number of amps, then the FET lets off briefly and smacks the diode, which keeps the voltage relatively low (Vf) so current doesn't decay too bad (relatively constant If, if that makes any difference) before the FET clamps it again, this time which the reverse recovery fires, pulsing the smaller drain inductance and sending some dV/dt to the output line. After the excitement, inductor current decays through its DCR and the diode until the next cycle arrives. Maximum rate depends on the ratio of current rise to current decay in the inductor between ON and OFF states, and of course, heating of the circuit.

I tried some last night but didn't get anything better for the same setup I posted earlier.

Tim

--
Deep Fryer: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Hello John,

Avalanching the old 2N2369 is pretty neat:

But not for 100psec stuff. Then he probably needs your MA/COM devices. And some not so cheap gear to see those slopes.

--
Regards, Joerg

http://www.analogconsultants.com

Ya, I built one with a PH2369:

Seems to max out my Tek 475, showing a 1-2ns pulse: (10ns/div)

I'd be happy enough with 1ns edges. I'm not even getting 10ns quite yet.

Best results I've had are from FMU22S and FMPG2F pulls I have on hand, not so much by speed as by amplitude. I'll take some pics.

Tim

--
Deep Fryer: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Hello Tim,

1nsec is all? You can get 500psec from avalanching a 2N2369. The PMBT2369 (SMT version) is just $0.13 in single qties at Digikey, and it goes below $0.03 for production qties. A true bargain for such a hot transistor. A while ago I was surprised it was still around.

You can also get a TO92 from Fairchild there but for fast stuff I'd go SOT23 and stripline. A Sauerkraut wiring scheme isn't going to work up there. This is in essence GHz stuff you are doing so everything needs to be neat and properly impedance calculated.

--
Regards, Joerg

http://www.analogconsultants.com

Yep. The 2N2369 dates to MY youth ;-)

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
I love to cook with wine.      Sometimes I even put it in the food.

There are lots of cheap cmos gates that will get you well below 1 ns edges for a 5 or 6 volt swing. Max ratings? We don't need no stinkin' max ratings! I like the NL37WZ16US with all three sections in parallel. Only problem is that you can barely see the damned thing.

For more fun, use it to drive one of those 1 or 2-amp Idss PHEMTs.

Some of the LVDS-to-TTL converters get down in the 700 ps edge rate turf.

John

2N107 dates to mine!

John

Also CTL22S.

Two pics:

The little folded loop of red wire, between the TO-220 cases, is the inductor. At this speed it seems to work better with less L. There's some extra bypass caps on board not shown on the schematic, too.

5V/div vertical, 20ns/div horizontal (the ringing corresponds to about 35MHz, possibly reflection on the 3' cable, even though termination doesn't seem to change the reflection..). This is with a 50 ohm cable fed from a 56pF capacitor connected to the "scope point" on the schematic. Rep rate is um... 500kHz or so.

Tim

--
Deep Fryer: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Got Ringing, Eh?

Tim, I'm sure most people here would suggest that fast circuits often work a lot better on a copperclad ground plane. With very short leads, and good rf bypassing.

Many people like to mount the ic's upside-down, dead bug style. This makes it difficult to read the markings so you can identify the circuit six months later. I simply bend the leads up so they won't touch the copper, then mount the parts face up. This allows you to lay the circuit out in a cad program, route it, then simply hand wire the traces. Heat-strippable magnet wire is great for making connections.

Regards,

Mike Monett