Connect with us

Step recovery diodes and SPICE

Discussion in 'Electronic Design' started by Darwin, Mar 2, 2007.

Scroll to continue with content
  1. Darwin

    Darwin Guest

    I am trying to build a fast pulse generator for an UWB transmitter.
    My goal is to obtain a few V peak pulses in a 50Ohm load with a
    10%-90% rise time < 100ps. The repetition rate should be at least a
    few tenths of MHz. I have seen that step recovery diodes can be very
    useful for this application and I read the interesting HP application
    note 918 and a few articles about this and I am now playing a little
    with SPICE.
    Among all producers, I found Aeroflex/Metelics which provides SPICE
    models for their diodes:

    This sound very interesting, but SPICE does not include the step
    recovery diode's transition time.
    I am trying to run a few simulations now but I would like to know how
    to expect from SPICE results. SPICE calculates a transient response,
    but how should I expect this would be degraded with a certain recovery

    Part of my SPICE file follows; since I plan to use a SOT-23 package, I
    included a 1.5nH series inductance and a 0.2pF shunt capacitance to
    the diode to have a rough idea of the parasitics' influence.

    Thanks to everybody

    * A simple pulse sharpener

    ..TRAN 1ps 20ns

    VPLS 1 0 PULSE(0 10 5ns 1ns 1ns 20ns 50ns)
    Rg 1 2 50Ohm
    C1 2 3 1nF
    Lp 3 5 1.5nH
    Cp 0 5 0.2pF
    D1 0 5 D_840
    C2 3 4 1pF
    Rl 4 0 50Ohm
    If 3 0 10mA
  2. Darwin

    Darwin Guest

    .... I forgot the model... sorry!

    * A simple pulse sharpener

    ..TRAN 1ps 20ns

    VPLS 1 0 PULSE(0 10 5ns 1ns 1ns 20ns 50ns)
    Rg 1 2 50Ohm
    C1 2 3 1nF
    Lp 3 5 1.5nH
    Cp 0 5 0.2pF
    D1 0 5 D_840
    C2 3 4 1pF
    Rl 4 0 50Ohm
    If 3 0 10mA

    ..MODEL D_840 D (IS=500.0E-15 N=1.3 RS=0.22 CJ0=0.545p VJ=0.5 M=0.235
    XTI=3.0 EG=1.12 BV=60 IBV=10E-6 TT=10E-9)

  3. John Larkin

    John Larkin Guest

    I'd doubt that spice modeling will be worth the effort. You'd do
    better to experiment and measure. Getting below 100 ps Tr is hard,
    because package and pcb parasitics start to really matter, and because
    the faster srd's are hard to drive.

    Faster srd's don't store much charge, so the problem is to reverse
    bias them fast enough, to get enough reverse current through them,
    before they snap. The driver becomes non-trivial, and it's sometimes
    better to buy a "slower" diode just because you can slam so much more
    reverse current into it.

    The other way to get edges like this are to start with a fast ecl or
    cml edge - 40 ps is easy these days - and run it through a cheap 5 or
    10 GHz mmic.

  4. Joel Kolstad

    Joel Kolstad Guest

    Say John,

    When you're designing this stuff, what's your technique for prototyping?
    "Real" circuit boards (full layout and etched)? Copper clad (blank) PCBs that
    you Dremel away the bits you don't need from? Some of the fancier "universal"
    PCBs specifically meant for prototyping?

  5. John Larkin

    John Larkin Guest

    I just solder stuff to a piece of double-side copperclad, midair or
    onto crude traces cut with an xacto knife, all impedance matched of
    course. I usually use flanged sma connectors soldered to the board
    edge-mount style, or solder hardline coax straight onto the parts.
    There's tons of sma hardline cables on ebay.

    For more complex stuff, we lay out a board and have AP Circuits or
    somebody make us a few boards.

    The critical thing is to have a good, 12 GHz or so, sampling scope. An
    11801+SD22 can be had on ebay for $2K or less. Older 7000 series stuff
    isn't as good, but is a lot cheaper.

  6. Darwin

    Darwin Guest

    I agree with you; I am waiting for a few samples to play with and I
    wanted to simulate a little them while I am waiting...
    I was thinking to drive the SRD with a logic gate such as a 74AC14,
    paralleling a few gates in order to obtain enough current. I hope it
    is fast enough...
    I have had a look at some very nice SiGe ultrafast comparators. I will
    try them if I do not have good results with SRDs.
    I was thinking to prepare a few boards with matched 50Ohm striplines
    to prototyping a little. I have access to a nice 40 GHz scope, but it
    is 100km away from where I am working, and we have a 20GHz network
    analyzer in house.
  7. John Larkin

    John Larkin Guest

    That's one place where simulation or calculation is useful. But you
    have to know the approximate equivalent stored charge of the srd, to
    calculate what the reverse current will be at the instant of snap. If
    you drive the srd through a small inductor, fairly common, the
    risetime of the driver pumping the inductor is the thing that has to
    be integrated. You'll need roughly 60-70 mA at snap time.

    An NC7NZ34, all sections in parallel, powered by +6, will put 5 volts
    into 50 ohms in about 600 ps.
    You probably refer to the Analog Devices parts. One of them makes a
    nice 35 ps edge or some such. You can differentiate that with
    0603-size r/l/c parts, into anybody's fast mmic, and get about a 100
    ps blip. We're doing that to get a 90 ps, 6-volt fiducial pulse into a
    laser modulator, but we're using $180 Hittite distributed amps to get
    that much swing.

    But for a few volts at 100 ps, the srd is probably easier.

    Oh, try M-pulse too. Thay have some nice parts.

  8. Darwin

    Darwin Guest

    Thank you John, your posts are very informative; I have seen that that
    SRD stuff has been debated also a few months ago and I read the old
    If I have understood correctly, the transition time of the diode
    depends on the stored charge. Should I minimize that charge? The the
    current risetime is mainly fixed by the inductor, if the driver has a
    sufficiently small risetime. How should I choose the inductor then? I
    guess that the diode should snap when the current equals the supply
    voltage divided by the 50 Ohm load resistance. In this case, I should
    get a clean output pulse. The inductor should be sufficiently small to
    have a fast dI/dt, but big enough to isolate the generator when the
    diode snaps. I can then choose the forward current in order to make
    the SRD snap at the right time. Is that correct?
    Thanks. I will have a look at that part. These tiny logic devices are
    truly amazing.
    Yes, I do. Something like the ADCMP580.

    Thanks a lot!
  9. John  Larkin

    John Larkin Guest

    No, or at least only indirectly. Diodes that store a lot of charge
    tend to have slower transition times.
    Yes. The output voltage will be a bit less than (inductor current *
    load resistance) at the instant of snap. The problem is to get that
    much current into the srd before you use up all the stored charge. If
    you lose that race, it snaps when the current is too low, and you get
    a wimpy step. It's easier to pump a lot of current into an srd that
    stores a lot if charge, but those are the same ones that tend to have
    slow transition times.
    Email me your address and I'll send you some parts.

  10. Guest

    Hi everybody,
    I'm doing my study in Belgium and for my final work I have to do a
    pulse generator too.

    First I've planned to use a transistor (2N3904) in avalanche area. But
    with my circuit, I need a very fast signal generator because the width
    of my pulse is proportionnal to the rise time of the signal generator
    and I want a 1 ns width pulse, so I need a 1 ns rise time for the

    So I would like to use a SRD instead or inside my circuit. Could you
    please tell more about your work? Do you have a particular circuit for
    the pulse generator? Please email me and we can share

  11. Darwin

    Darwin Guest

    If you put a small transistor in the avalanche region, you should
    easily get a sub-nanosecond rise time.
    You can find a good example in Jim Williams' AN94, from Linear
    I found very interesting the old AN918 from HP:

    If you need a 1ns pulse rise time at a few volts, keep in mind that
    this can be obtained with a reasonably fast logic gate, such as the
    NC7NZ34 John suggested.
  12. John Larkin

    John Larkin Guest

    If the transistor avalanches, the output rise time is unrelated to the
    input trigger speed. It fires or it doesn't.

    Zetex sells, as far as I know, the only transistors specified for
    avalanche use. They are an old diffused process, fairly low Ft, made
    in Russia. Newer epitaxials don't seem to avalanche.

  13. Darwin

    Darwin Guest

    First I've planned to use a transistor (2N3904) in avalanche area. But
    @ Johan: I played a little with the transistors I have in my junkbox.
    I did not see avalanching on a 2n2222, but I tried a few 2n2907 (they
    are PNP transistors, so one needs to change the schematics
    accordingly), from several brands and they all worked nicely. Of
    course, they are not specified for avalanche use, but a lot of cheap
    transistors can be made avalanching well. I did not run them for a
    very long time, but they did not vanish in a puff of smoke. You have
    to try. Of course, you need at least a 250V max DC adjustable power
    supply. I used a flyback inverter I made a few years ago. Be careful,
    of course.

    @ John: I built several SRD pulse stretcher, but unfortunately I have
    to wait until I have access to a fast scope to test them. Thank you
  14. Guest

    @ Darwin: That sounds great!! Which schematics did you use to try the
    2n2222 and the 2n2907? What sort of input trigger did you use (and
    specially the input trigger rise time)?? Because in a book that I
    have, they say that it could work with a 2n918, 2n2222 and 2n3904.
    Here is the schematics I would like to use for the transistor in
    avalanche area: and
    this one is the one I would like to use with a SRD: . For you, which
    ones could give me the best result? I mean the smaller width impulse?

  15. Guest


    if I understand, I can put any rectangular trigger on my schematics'
    input, it will not influence the width of the output impulse (Even if
    the rise time is more than 5 ns)? In fact, I would like to have an
    impluse of maximum 1ns of duration... if it's smaller, it better!!

    What do you mean by the fire or not? Does it mean that the transistor
    died? Here is the schematics I would like to use: But I don't know
    what to put for Vbb ... Do you have any idea??

  16. John  Larkin

    John Larkin Guest

    That looks OK, but the avalanche will kick a lot of pulse back into
    the trigger input. I like to use a small trigger transformer, with the
    secondary connected directly b-e on the transistor. A little RF
    ferrite core with a few turns on each side, bifalar maybe, works fine
    for a ttl-level drive.

    You might also add a bit of inductance in series with the output 50
    ohm resistor, or even reduce its value, to speed things up a tad. Keep
    all lead lengths zero, of course, and build it on a solid ground

    I've never damaged a transistor by avalanching it. Many generations of
    Tek sampling heads used avalanche transistors, to either drive the
    sampling gate directly or to drive an SRD. There must be billions of
    device-hours there, and as far as I nkow, it's not a particular
    failure mode. The dual-channel 7S14 uses one avalanche transistor to
    drive both sampling bridges. It's spec'd at 350 ps risetime but is
    typically about twice that fast.

    Picosecond Pulse Labs and Kentec sell commercial avalanche-based
    pulsers. We may do one soon too, for a custom application.

  17. Guest

    Thank you for all these advices!!
    I've try to simulate the circuit with the avalanche transistor with
    PSpice but I think the simulation device don't know the avalanche
    Do you think there is a way to simulate that? May be with ADS? I've
    got it but I've never use it.

    I will use a 2N3904 because that's what we have in the laboratory. But
    to fixe -Vbb and Rb, I don't really know what to put there. Will it
    depend on my transistor breakdown voltage?if I put Rb = 1Mohm and -Vbb
    = -60V will it be ok?

  18. John  Larkin

    John Larkin Guest

    Simulation is useless here. Use solder.

    To test transistors for avalanche behavior, try...

    +---------+--- <-- 10M scope probe
    | |
    | |
    | 47 pF
    | |
    | |
    | gnd
    +----b NPN
    | e
    10k |
    | |
    +-----+-----------> } 50r coax to attanuator
    | gnd--> } and fast scope

    This will self-oscillate and tell you the avalanche voltage and
    risetime. Invert supply for PNP of course. Most transistors,
    especially modern ones, won't avalanche at all.

    I've noticed that some transistors will dump essentially all of the
    cap voltage into the load, and some just a fraction. The Zetex parts
    seem to turn on hard.

  19. Guest

    Just to be sure...

    50r coax to attanuator
    gnd--> } and fast scope

    are connected to b of the NPN? So I have b ---10k --- gnd and before
    gnd I connect fast scope??
    Is that correct??

  20. Fred Bartoli

    Fred Bartoli Guest

    a écrit :
    Nope. The 10k base resistor, the 50R resistor and the 50R coax/scope are
    all connected to the transistor emitter.
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day