It occurred to me that a back-biased red or green LED or base-emitter junction might make an interesting substitute for a spark gap in a tiny Marx generator. These often break down in avalanche mode with minimal or no damage to the semiconductor, often at voltages under 10 volts. So you could easily produce a pulse train at 100 volts or more, perhaps from a 5-volt source using only tiny surface-mount components. This led me to dig a bit into the existing literature on similar devices.
The average achievable power of the device probably will not be large, since especially base-emitter junctions are not optimized for power dissipation, and, according to folklore, high-power LEDs generally do not tolerate avalanche breakdown well. Moreover, the properties the device relies on here (avalanche energy tolerance, reverse breakdown voltage) are not among the properties the manufacturer normally specifies, except I guess on JFETs. But JFETs are tiny high-precision devices, so JFET gate junctions are probably even worse for this application.
I suspect that, at least with silicon, you won’t get a good avalanche effect until 7 volts or more; below that, the less abrupt Zener effect will dominate. Purpose-built “zener” avalanche diodes might also work, but since they’re sold as voltage references, they might be designed to spike as little as possible — the spikes we’re after here are troublesome noise in voltage-reference applications. At least you’d have a well-characterized and optimized power rating. And of course diacs would definitely work and have been used in production Marx generators in the past. (Can you get MOS latchup out of power MOSFETs? Probably not.)
Looking at Can you get JLCPCB to fabricate a CPU for you affordably from “basic” parts?, I see that 5% 1k resistors occupying half an 0402-equivalent cost 0.385¢ soldered (and a few other values are available); discrete 1% 0402s are 0.305¢ in a wide variety of sizes; 0402 MLCCs are 0.4¢, while 1206 MLCCs are 1.3¢; NPN S9013 transistors are 1.58¢ and MMBT3904 is 1.32¢; red 0805 LEDs are 1.54¢; zeners only come in 5.6V and 3.3V and cost 1.35¢; 5.8V TVSs cost 2.91¢. Each Marx stage requires an avalanche element, a cap, and two resistors, so maybe an MMBT3904 (3 mm × 3 mm?), an 0402 cap (1 mm × ½ mm), and two 0402 resistors, for a total of 2.33¢ and 10½ mm², probably 25 mm² in practice.
A particularly interesting question is how fast the device completes its avalanche discharge and recovers (through recombination of the freed-up charge carriers). It wouldn’t surprise me to find a transistor was capable of producing higher frequencies in this avalanche-diode mode than when being used as a real transistor. But are we talking about potential pulse repetition rates of 10 kHz, 100 kHz, 1 MHz, 10 MHz, 100 MHz, 1 GHz, 10 GHz? How high do the harmonics go? I’m guessing that they’re much faster than a corona-stabilized spark gap (≈20 kHz repetition rate) but I don’t know how much. Even high-voltage SCRs typically manage 10 kHz.
One particularly amusing hackish thought: the avalanche elements can be back-biased diodes, the capacitors can be beefier back-biased diodes, and the charging speed limiters could also maybe be tiny diodes instead of resistors, though that’s a trickier proposition; this would enable you to do digital logic entirely out of diodes! If it’s possible, and fast, this seems like it would have been a killer advantage in the 01950s and early 01960s, when transistors were expensive and hard to get, and vacuum tubes more so; some Russian electronics were “ferrite/diode” systems in which the diodes took care of the combinational logic and (square-loop?) ferrite transformers handled memory and inversion.
When I was 9, I proposed solid-state switching elements for computation that worked on the neon-lamp-like principle of avalanche breakdown in ionic solids. Those are not practical, because the device would require very high voltages, and its crystal structure would rapidly lose integrity (almost certainly, anyway). But avalanche discharge in solid-state semiconductors is commonplace; it’s the way diacs, triacs, and other thyristors and SCRs work.
Kerry Wong built a 2N3904-based pulse generator, running it off 120V (!!) and back-biasing the collector—base junction, which he says consistently avalanches around 100V, saying:
Avalanche transistors can be used to generate fast rise time pulses. Their usage in the hobby world was made popular following an application note ([Linear] AN72 2 [now at Analog[9]]) by Jim Williams and was further publicized via this EEVBlog video. ...
R2, C1 along with the NPN transistor form a relaxation oscillator. The capacitor gets charged via R2 and then rapidly discharges when the collector-emitter voltage reaches the avalanche voltage. The discharge current flows through R1 during the avalanche and forms a fast-rise pulse between ground and the emitter. The choice of R2 and C1 is pretty liberal. In general, C1 can range from a few pF’s to tens of pF’s and R2 can range from 100K to 1M. The larger the value of C1, the wider the avalanche pulses due to increased discharging RC (R1C1) constant. But C1 cannot be too large as the energy released during the short avalanche period could cause the PN junction to fail. The RC constant (R2C1) determines the operation frequency. For the values given [220kΩ and 22pF], the pulsing frequency is at roughly 30 kHz. R1 is chosen to match the characteristic impedance of the load. ...
During my build process, I sampled a large batch of 2N3904’s, and found that most can avalanche pretty consistently at around 100V. ...
...The following picture shows the same pulse observed on a Tektronix 2445 (150MHz bandwidth) with matching input impedance. The measured rise time is around 1.5 ns which corresponds to a bandwidth of approximately 230 Mhz [0.35/Tᵣ].
His pulses look like only 50 volts, though, suggesting that they might actually be much faster than 1.5 ns, and being limited by the oscilloscope’s 150MHz input bandwidth. Oddly, log(120V/100V)/(220kΩ 22pF) works out to about 16 kHz, not 30 kHz.
Using the huge collector-base junction instead of the teensy emitter-base junction probably means you can handle a lot more power, and it is at least a somewhat controlled process parameter, since people actually do often require that their transistors resist a back bias on the base-collector junction; ST’s 2N3904 datasheet specifies a minimum of 60 V for the base-collector reverse breakdown and, surprisingly, provides a minimum value for the base-emitter reverse breakdown voltage as well: 6 V. Interestingly, its delay time and rise time (for normal transistor operation) are specced as 35 ns, with 200 ns for storage time and 50 ns for fall time, and a 270 MHz transition frequency.
This means that the pulse’s rise time is more than 20 times faster than the pulse you’d get using the transistor as a transistor switch, but maybe no faster or even a bit slower than if you were using it in its linear region. But probably the transistor is not the limiting feature here.
He also cites a 01997 paper by Kilpelä and Kostamovaara and a pulse generator project by Andrew Holme.
Holme used a 2N3904 and an open coax transmission line rather than a 22-pF cap to get a rectangular pulse with about a 400-ps rise time, which he says is limited by his oscilloscope. Astonishingly, he did this with through-hole components.
The coax transmission line suggests how to get arbitrarily high gain from such a circuit, considered as an amplifier: an arbitrarily short input pulse can produce an arbitrarily long output pulse, as long as the current is high enough to maintain the avalanche but not high enough to overheat the transistor. (I think you can do this with a capacitor too — it’s just messier.)
Holme mentions that you can trigger the circuit by applying short pulses to the base, which is a thing I hadn’t thought of; both Wong and Holme are taking their main signal from the emitter and just tying the base to ground through a big resistor. I suppose that you’d pull the base negative to trigger it in that case, thus increasing |VCB| enough to cause an avalanche — just treating the transistor as a pair of back-to-back diodes? (This is wrong; see below.)
Holme also cites Jim Williams’s Linear AN72 and AN94. I guess when Analog bought Linear they broke the link, but I found it anyway.
Jim Williams’s AN94 is about measuring an amplifier slew rate at 2.8 GV/s, for which he had to build a 360ps-rise-time 15–20V pulse generator for this purpose, because his fancy 1-ns rise-time pulse generator was too slow for the amplifier he was measuring, but subnanosecond-rise-time pulse generators cost US$10k–30k. So he used a 2N2501 (or maybe a 2N2369) as an avalanche transistor, biasing its collector to 70 volts above ground. Interestingly, my understanding of triggering with a base pulse is incorrect, at least for this circuit: he uses a positive-going trigger pulse into the base of the avalanche transistor to trigger the avalanche, which I’d’ve thought would be counterproductive. He AC-couples the trigger pulse with a 5pF cap and protects the avalanche transistor’s base with a Schottky up from ground.
The 2N2501 looks like a perfectly ordinary (but old) small-signal NPN transistor: 350 MHz, β≈50 (or >3.5 for small signals), 40V minimum V(BR)CBO, 1.2 W, 100 mA; the 2N2369 is pretty similar, but maybe 500 MHz and 200 mA. The datasheets show them in a 01960s-style TO-18 metal can rather than a modern TO-220 or similar epoxy package; an advertisement for the 2N2501 appeared in the May 4, 01964 issue of Electronics, though with only 20V “BVCBO”, and both transistors appear in the 1965 Motorola Semiconductor Data Manual, with ratings more like the 40V I mentioned earlier. Neither is billed as an avalanche transistor or has a datasheet with avalanche characteristics, and there’s nothing to suggest that they can in any way be used to generate 400-picosecond edges.
I wonder if Williams used them even in 02003 instead of more modern parts because the modern parts were “much improved” — in the sense of having an inconveniently higher base-collector breakdown voltage. (Does that also imply a larger junction capacitance?) Williams comments that not every transistor of this model was suitable:
Q5 requires selection for optimal avalanche behavior. Such behavior, while characteristic of the device specified, is not guaranteed by the manufacturer. A sample of 30 2N2501s, spread over a 17-year date code span, yielded 90%. All "good" devices switched in less than 475ps with some below 300ps.⁶ In practice, Q5 should be selected for “in-circuit” rise time under 400 picoseconds.
Note 6: 2N2501s are available from Semelab plc. Sales@semelab.co.uk; Tel. 44-0-1455-556565 A more common transistor, the 2N2369, may also be used but switching times are rarely less than 450ps. See also Footnotes 10 and 11.
Previously Williams wrote AN72, which also covers the technique, but at less length; most of AN72 is a primer on the basics of working with VHF and faster circuits, or explanations of what you might want to use Linear’s new high-speed LT1394 comparator for. But in pp. 32–34 and in appendix B, he gives a couple of simplified versions of the AN94 design using a 2N2369, with a 2pF capacitor instead of a transmission line. Here he also explains the measures necessary to prevent the 250-ps-rise-time avalanche pulse from overwhelming the output of the comparator providing the trigger pulse: 100Ω and six ferrite beads! There must be a reason a diode wasn’t enough but I don’t know what it is.
This suggests that after writing AN72 he found that the 2N2501 gave better results than the 2N2369, which implies that probably most transistors will be significantly worse.
He describes the transistor a couple of times as “a 40V breakdown device”.
The avalanche pulse measures 8V [4.8V where text is duplicated in appendix for slightly different circuit] high with a 1.2ns base. Rise time is 250ps [216ps in appendix], with fall time indicating 200ps [232ps in appendix]. The times are probably slightly faster, as the oscilloscope’s 90ps rise time influences the measurement.
Q5 may require selection to get avalanche behavior. Such behavior, while characteristic of the device specified, is not guaranteed by the manufacturer. A sample of 50 Motorola 2N2369s, spread over a 12-year date code span, yielded 82%. All “good” devices switched in less than 600ps [650ps in appendix]. C1 is selected for a 10V amplitude output. Value spread is typically 2pF to 4pF. Ground plane type construction with high speed layout, connection and termination techniques is essential for good results from this circuit.
Note that 8V is a lot less than the 70V or 90V to which the 2pF capacitor is charged via a 1MΩ resistor; he says that 90V gives about a 200kHz free-running frequency.
In AN94 he reported that he was having a hard time getting down to 300 ps, so maybe that 250-ps transistor was just a super good one, or maybe he decided it wasn’t really 250 ps.
You might think that it would be hard to get less than 2pF parasitic capacitance between PCB traces and stuff, but in the photo it seems he just constructed that part of the circuit soldered to the backside of a BNC connector.
This circuit also reconfirms that, contrary to my previous expectations, he was triggering the avalanche by forward-biasing the base-emitter junction, just like you normally would to operate a transistor.
For more information about avalanche transistors for pulse generation Williams refers us to these references:
17: Williams, J. “High Speed Amplifier Techniques,” Linear Technology Corporation, Application Note 47 (August 1991)
20: Tektronix, Inc., Type 111 Pretrigger Pulse Generator Operating and Service Manual, Tektronix, Inc. (1960) (Williams says his method “borrows heavily from” this device.)
22: Williams, J., “Practical Circuitry for Measurement and Control Problems,” Linear Technology Corporation, Application Note 61 (August 1994)
27: Haas, Isy, “Millimicrosecond Avalanche Switching Circuits Utilizing Double-Diffused Silicon Transistors,” Fairchild Semiconductor, Application Note 8/2 (December 1961)
28: Beeson, R. H., Haas, I., Grinich, V. H., “Thermal Response of Transistors in the Avalanche Mode,” Fairchild Semiconductor, Technical Paper 6 (October 1959)
This could be used with the Tektronix Type N Sampling Plug-In Unit, back when Tektronix was an oscilloscope company; in 01960 the Type N claimed an 0.6-ns rise time on its front panel; it was used to trigger an oscilloscope to repeatedly sample an otherwise-too-fast signal in the analog domain:
The sampling system thus formed permits the display of repetitive signals with fractional nanosecond (10⁻⁹ second or nsec) risetimes. By taking successive samples at a slightly later time at each recurrence of the pulse under observation, the Type N reconstructs the pulse on a relatively long time base. ... The sampling system formed by the combination of the N Unit and a conventional oscilloscope is quite different in operation from normal oscilloscope systems. A conventional oscilloscope system traces out a virtually continuous picture of waveforms applied to the oscilloscope input; a complete display is formed for each input waveform. The sampling system, however, samples the input waveform at successively later points in relative time on a large number of input pulses. From this sampling process, a series of signal samples is obtained. The amplitude of each signal sample is proportional to the amplitude of the input signal during the short time the sample is made. Input waveforms are then reconstructed on the screen of the oscilloscope, as a series of dots, from these signal samples. The oscilloscope bandpass required to pass the “time stretched” signal samples is much less than the bandpass which would be required to pass the original input signal.
The Type 111 pulse generator was used to transmit the timing information to the type-N sampler:
As described previously, to trigger the Type N Unit you must first connect a triggering signal to either the TRIGGER INPUT or REGENERATED TRIGGER INPUT connector. When triggering signals are applied to the TRIGGER INPUT connector of the N Unit you must adjust the TRIGGER SENSITIVITY control for stable triggered operation.
...
When the Type 111 Pretrigger Pulse Generator is used, no triggering adjustments are necessary except to turn the TRIGGER SENSITIVITY control of the N Unit fully counterclockwise. The N Unit is started automatically each time a pulse from the 111 is applied to the REGENERATED TRIGGER INPUT connector of the N Unit.
I haven’t been able to find the “Type 111 Operating and Service Manual”, just the “Instruction Manual” from 01965 (59 pp.) This explains that it runs at up to 100 kHz (“kc”) and has a risetime of 500 ps at at least 10 volts, which is astounding for 01960: “Determined from observed system risetime of 615 psec using a Tektronix sampling oscilloscope with a risetime of 350 psec. See Calibration section.”
It evidently used an external coax “charge line” to produce a rectangular pulse, so you could hook up different lengths of cable there to produce pulses of different lengths, but only up to 142 ns for serial numbers below 800: “Exceeding these limits may damage the avalanche transistor, Q84.” Higher serial numbers could use pulse widths up to 1500 ns, so I guess they beefed up Q84.
You could also couple the pulse generator’s output pulse into the device under test, I guess so that what you were viewing on the oscilloscope was its pulse response.
The circuit is explained (p. 3–2, 22/59):
Output Pulse Generator (S/N 800-Up)
The positive output pulse from the Comparator blocking oscillator is applied to the Output Pulse Generator (avalanche stage) through C75 and D80. Since the collector voltage of Q84 is set just short of the point where the transistor will avalanche, when the voltage pulse from T60 turns on D80, a fast current pulse is applied to the base of Q84, causing the transistor to avalanche. This allows the internal charge line (and the external charge line, if any) to begin to discharge. The resulting positive voltage step at the emitter of Q84 produces the start of the output pulse.
...
Output Pulse Generator (S/N 101-799 only)
The positive output pulse from the Comparator blocking oscillator is applied to the Output Pulse Generator (avalanche stage) through two paths.
One path is through C75 and R75 to the collector of Q84. The pulse which takes this path is a current pulse and is most effective when short time duration charge lines are used. The collector voltage of Q84 is set just short of the point where the transistor will avalanche. Consequently, when the positive pulse from Q60 is applied to the collector of Q84, the signal is sufficient to cause Q84 to avalanche.
The second path, from T60 through C76 to the outer conductor of the internal charge line and to R77 and R78, couples a positive voltage pulse to the collector of Q84. This pulse is more effective than the current pulse at getting Q84 to avalanche when long charge lines are used. The internal charge line is passed through a ferrite toroid core (T78) to prevent the voltage pulse from being shorted to ground. The toroid core effectively isolates one end of the internal charge line.
So, fascinatingly, they redesigned the circuit to trigger through the base instead of by adding more voltage to the collector, starting with serial number 800! I guess they didn’t realize they could do that in 01960 and only figured it out around 01965.
There’s a parts list in the manual starting on p. 49 and absolutely beautiful schematics on pp.54–55/59 (initialed TR 964 and TR 366), annotated with expected oscilloscope traces in callouts and dc voltage levels as well. There are only three transistors in the whole instrument!
The all-important Q84 avalanche transistor was originally “Selected from 2N636”, but switched at serial number 800 to “Silicon Avalanche, checked”, with Tektronix part numbers. In the pre-800 schematic I think its VBC is given as 37 volts, and its base is pulled down to a (germanium) diode drop below ground.
C75 is a 47 pF ceramic up to S/N 799, 10 pF in 800 and up, 500 V. The resistors are [carbon?] composition; R75 is 1kΩ, ±10% ½W up to 799, ±5% 1W in 800 and up. R77 and R78 are ½-W 10-Ω jobbies, deleted in 800 and up. T60, cleverly arranged so that the trigger-pulse-generating transistor Q60 that triggers Q84 turns itself off, is a TD20 toroidal transformer up to 799, a 4T bifilar transformer in 800 and up (actually trifilar on the schematic).
D80 is exotic: for serial numbers X241–799 it’s a Tektronix germanium diode, and for 800 and up it’s a Tektronix gallium arsenide diode. (To be fair, most of the 16 diodes were germanium; only 2–4 were silicon, plus five more in a typewritten erratum stuck in the back of the manual.)
This probably explains why Williams didn’t use a diode to block the current pulse surging back through the base of his avalanche transistor: his diodes were too slow! He probably didn’t have a superfast GaAs diode handy, so he opted for ferrites.
The General Electric 2N636 was a 15MHz germanium NPN transistor specified for 20 volts of “BVcb”, 200 mA, and β=35, according to one 01962 compendium, or 300 mA and β=70, according to one from 01973. It appears in GE’s 01958 Transistor Manual, categorized as “computer” rather than “audio”, “amplifier & computer”, “unijunction”, “tetrode”, or “IF”, and rated for 300 mA, β=35, and only a 15-volt “punch through voltage” (p. 145, 143/167).
Fullwood says:
The pnp transistor types 2N501, 2N502, 2N504, 2N588, as well as the npn types 2N635, 2N636, 2N697, 2N706, and 2N1168 have all been found to avalanche with the same order of rise time [which he states in the abstract and later to be about 1 ns]. However, the decay time that is observed varies greatly with type, being related to the transistor’s normal performance as a switch. ... One hundred and twenty 2N504 were tested as to whether or not they would avalanche at all in the circuit of Fig. 1. About 80% were found to operate satisfactorily with the zero bias arrangement as shown and without oscillating at this steady current.
An interesting thing about this is that he was triggering the avalanches with a pulse on the base, unlike the 01960 version of the Tektronix device. Because the 2N504 was pnp, it was a negative-going pulse, and the circuit was driven from a -300 V power supply. Trigger pulses were supplied from a “mercury pulser”.
DOI 10.1063/1.1716847, “On the Use of 2N504 Transistors in the Avalanche Mode for Nuclear Instrumentation", by Ralph Fullwood (under Walter Selove) at U Penn (later at RPI), Review of Scientific Instruments, Volume 31, Number 11, November, 01960, interestingly the same journal that published Kilpelä and Kostamovaara 37 years later (see below).
He cites:
This paper is interesting because it has a number of very simple circuits that do interesting things, like amplify the tiny pulses from a photomultiplier tube.
After Holme’s project, Williams and David Beebe revisited pulse generators in Linear AN122 in 02009. In Appendix B, “Subnanosecond Rise Time Pulse Generators for the Rich and Poor”, on p. 11/20, they explain:
The Tektronix type 111 has edge times of 500ps, with fully variable repetition rate and external trigger capabilities. Pulse width is set by external charge line length. Price is usually about [US]$25. ... Residents of Silicon Valley tend towards inbred techno-provincialism. Citizens of other locales cannot simply go to a flea market, junk store or garage sale and buy a sub-nanosecond pulse generator.
Then they again present the circuit from AN94, unmodified as far as I can tell, but this time its performance has been derated again, to a 400ps rise time. And in Appendix F, they explain, “The Tektronix type 109 mercury wetted reed relay based pulse generator will put a 50V pulse into 50Ω (1A) in 250ps.” Perhaps this is the “mercury pulser” Fullwood was talking about.
Kerry Wong revisited the theme in 02014 using the lower-voltage emitter-base junction as I suggested above, producing the following table of emitter reverse breakdown voltages with a 1000μF (!!!) cap:
| 2N4401 | ~12.5V |
| SS9014 | ~12.5V |
| 2N4124 | ~12V |
| 2N3904 | ~12V |
| BD137 | ~11V |
| BD139 | ~11V |
| BC337 | ~9V |
| SS9018 | ~8.2V |
He found some important limitations:
Also, while I could get most NPN transistors to oscillate in their reverse breakdown regions I could only get a couple of BD138 PNP transistors to oscillate using the same circuit above (power polarity is reversed). And the oscillation only occurred at a very tight voltage interval (e.g. ±0.05V).
One of the useful features of a standard avalanche pulser (like this one [linking to his other project]) is its extremely fast rise time (subnanosecond), so can we use negistors to build similar pulsers?
Well, the short answer is no. After some experiments it appeared that the rise time of a negistor pulser is magnitudes higher (e.g. ~100ns) than a typical avalanche pulser.
...the capacitance cannot be arbitrarily small. In my case, 100nF seems to be near the lower limit.
Importantly, he says in the comments:
Just the e-b junction won’t work, it would just act like a Zener diode.
analogspiceman posted the following SPICE model in the comments:
* UpsideDown.asc – a single transistor relaxation oscillator model for LTspice
V1 1 0 10
R1 1 2 1k5
C1 2 0 1µ Rser=8m
XQ1 0 NC_01 2 2N2222r
*
.subckt 2N2222r e b c ; this subckt just turns the NPN upside down
Q1 c b e 2N2222r
.model 2N2222r npn Is=10f Xtb=1.5 Rb=10 ; nondirectional parameters
+ Br=200 Ikr=0.3 Var=100 tr=400p ; reverse (forward) parameters
+ Bf=7 Ikf=0.5 Vaf=10 tf=100n Itf=1 Vtf=2 Xtf=3 Ptf=180 ; fwd (rev) params
+ Re=.3 Cje=8p Ise=5p ; emitter (collector) parameters
+ Rc=.2 Cjc=25p Isc=1p BVcbo=7 ; collector (emitter) parameters
.ends 2N2222r
*
.opt plotwinsize=0
.tran 0 10m 0 1u uic
Wong mentions the term “negistor” Richard Phares used in Popular Electronics in 01975 for this configuration (an avalanche discharge has negative differential resistance, so “negative resistance transistor”). Phares notes that germanium transistors and pnp transistors will not work, recommending the MPS-5172, the 2N2218 (7.7V), the 2N2222, or the 2N697. Unfortunately, the term “negistor” seems to have been largely co-opted by Keelynet crackpots lacking even the most basic knowledge of physics and electronics. However, Alan Yates, for example, built some oscillators using the term. Prolific electronics hacker sv3ora reports, “The 2N4124 gave the lowest oscillation voltage, around 6.8V,” and confirms that grounding the base kills the oscillation. Jean-Louis Naudin reports oscillation at 16.4 volts on a 2N2222A and also characterizes its available stable avalanche currents ranging from 5.47 V at 10 mA up to 6.54 V at 2 mA.
These folks wanted to make 5–10-ns semiconductor laser pulses for LiDAR, but at tens of amps. They said a transistor in avalanche mode is faster than a thyristor or MOSFET, though GaAs thyristors were reported to have reached the 500-ps-level most of the above discussion has centered on. They tried an MJE200, a 2N5190, two 2N5192s, and a Zetex ZTX415 SOT-23 avalanche transistor; the MJE200 started breaking down below 100V but was consistently only about 15A no matter how high the voltage, while the others all required 250V to break down, reaching 70 A at 400 V. These all got rise times in the 2.5–4 ns range; this extreme slowness (ha!) is probably because parasitic inductances matter more at 70 amps than at 1 amp.
Their circuit is very different from all the others I’ve seen, full of inductors, and I don’t understand it yet. The paper has lots of good explanation about how avalanche transistors work, though.
70 A at 400 V for 10 ns is 28 kW, but only about 0.28 mJ.
Alex McCown (onebiozz) built a pulse generator to test his oscilloscope around a 2N3904, getting a 1.56-ns rise time (which he thinks is the scope’s limit, not the circuit’s) but wished he’d used a BFR505:
I have to say this was a fun $0 project, but if i were to spend some cash what would i have done differently knowing what i do now? Well for one i would not use an 2N3904, the BFR505 appears to be a better solution at a simple 30v avalanche of ~200-300pS.
Michel I. Gallant put 20ns pulses through an infrared LED using a 2N2369a avalanche transistor to measure the speed of light to within about 1% in his living room, but the 25 MHz Vishay TSFF5210 LED he chose slowed their rise time to 10ns. Very simple perfboard circuit. As a detector he used a 200 MHz Vishay BPV10 PIN photodiode amplified by an AD8001 configured for 35× gain and 50MHz, but they also built the circuit on a solderless breadboard, so it might have suffered some signal integrity problems from that and from the long leads on their components too.
Also interesting for fast-circuit purposes, he measured the response of different common LEDs up to 10MHz: the TSFF5210 had drooped less than 1dB at 10MHz, a red 08LCHR5 AlInGaP drooped 3dB, and a white 08LCHW3 InGaN drooped 3dB at 2MHz and 6dB at 3MHz. Presumably that’s a composite of fast blue and slow yellow, but the pulse response he shows doesn’t show much fast blue.
Covington notes that the avalanche effect of the emitter-base junction makes a good white noise source, and also a good low-leakage low-capacitance “zener diode”, citing EEVBlog #1157.