narrow pulse generator

EEVblog #306 has an episode of that, 520 ps rise time pulse. It cites a design of LTC, now AD, Application Note 47 by Jim Williams. You need to operate a BJT on the avalanche operating region to get very sharp pulses.

Those risetimes are probably within 65nm CMOS capability.

Not clear to me, whether this is supposed to be a transistor
implementation, or a simulation "widget". Nor is, just what
"acceptable tolerance" means.

It could be as simple as an N-inverter / NAND "race pulse" (giving
N gate delays' worth of width, rise/fall are intrinsic / parasitics
driven).

If you want control / variability you might current-starve one
or more of those inverters (I'd want the back of the delay
string to be crisp, still, which CSIs don't really give you
especially when "starved real good").

just for reference reading ....

The programmable unijunction transistor seems naturally inclined to generate brief pulses.
I don't know whether it's easier to build as BJT, or CMOS.
All components must be adjusted carefully, to achieve sustained oscillations.

The capacitor charges slowly through R1, then discharges quickly through R2. The transistor arrangement turns on until current drops to zero (just like an SCR). The 220 ohm resistor serves as load. Pulses are referenced to 1.2 V supply.

Do you wish to have positive pulses which are referenced to ground? Then flip the circuit (putting N-device for P-device and vice-versa).

I've seen the UJT used for relaxation oscillators (which have
the abrupt-ish return stroke) but I don't expect the UJT to be
especially fast. I've only used them in audio play-projects.

As I understand its operation the UJT really operates by
conductivity modulation and the body is a lightly doped
resistive region (which minority carrier injection modulates
downward).

But that means there's a minority carrier lifetime involved,
and if you want a high modulation you are probably looking
at a high lifetime, and a not-that-short pulse. Maybe by
audio standards, but surely not photonic / high speed clock
standards.

I looked at one old-timey UJT datasheet found here:


and while it shows an "impulse generator" app circuit, notable
are the rounded pulse and long tail shown in the attending figure,
and a complete lack of time- and frequency-domain attributes
specified.

The long tail is the minority carrier lifetime / recombination at
work.

State-of-the-art pulse generators like Keysight 81133A outperform your requirements.

If you want to design something on your own, you can either use classical bipolar techniques with avalanche transistors and transmission line pulse formers used since the 60th or e.g. recent microwave FET.

Hi,

To add to the avalanche transistor part of the conversation, probably below level of unix_amr, but a good starting point could be the following three links for anyone interested:

Avalanche Transistor (Wikipedia - has formulas to calculate circuit, and useful links)

Avalanche Pulse Generator (it's one of the links in the Wikipedia article, to a person's homebrew PCB version - nice, helpful example with photosotos)

Avalanche Pulse Generator - An Introduction (random web search, seems a good introduction to the concept for novices)

There are multiple related forum threads on this topic here, too.

Poster could define current output/input required for mentioned NMOS gate beyond 1.2V amplitude.

Don't envy you the task, unix_amr - 2000 MHz... Good to know it can be done without expensive speciality devices, e.g. lowly 2N3904 and its relatives show up a lot in high-speed pulse circuit descriptions.

As an approximation to the speed involved, just mentally calculating a capacitor and resistor combination whose 5Tau equals ~100 pS or even entire ~500 pS was enlightening (100 pS = e.g. 1pF or less was my inexperienced conclusion because of the RC resistor (for 1k), otherwise 10pF to 100pF, unless 1.2V/1R = 1.2A or 1.2V/10R = 120mA is needed to drive NMOS gate, 12mA would be 100R); makes you wonder if using PCB trace and lead capacitances, etc. might be a simpler way of getting a tiny capacitor, if needed.

Good luck!

A transmission line pulser does make a good square pulse, provided that you want Zo=Zline (and have a switch handy that's well faster than desired edge rate). But at 1ns/ft the line will not be integrated on chip.

There are fast small SCRs that allow nS pulse generators with relatively simple ckts

please see: https://www.mouser.com/datasheet/2/268/WT8_35-1592491.pdf

also: