It occurred to me that, sometimes, I just need one hell of a fast pulse.
One reasonably available method is using transistors in their avalanche region. This phenomena occurs when a transistor is biased well in excess of its Vceo rating. Leakage current flows through the collector-base depletion region, caused by spontaneous electron-hole pair production (a result of impurities modifying energy levels and thermodynamics kicking things around, the same phenomenon responsible for the conductivity of a P-N junction). In the middle of the depletion region, there is a strong electric field, so electrons are drawn towards the collector and holes towards the base and emitter (in an NPN structure). Due to the strong electric field, they are accelerated to enough velocity that, in collisions with ions in the crystal, more electrons and holes can get knocked out. For moderate currents, the characteristic looks like a zener diode (which isn't coincidental, as high voltage zeners use avalanche breakdown rather than the actual zener effect).
At the breakdown voltage, the junction acts like a plasma, where charges move around and ionize more charges to maintain the current. More current causes more charges to be knocked about, so a relatively large current flows freely. Resistance drops, so it is a region (on the V-I graph) of negative dynamic resistance, the classic condition for a simple amplifier. If sufficient capacitance is placed across the device in this region, a relaxation oscillator can be made. The large amount of displaced charge in the transistor gradually dissipates, over perhaps a few microseconds. I don't know if an avalanche condition can be maintained statically, but I wouldn't want to try; peak currents are typically up to an ampere.
The remarkable characteristic of this whole phenomenon is that, once the tipping point is reached, the voltage on the depletion region is quickly discharged. Typical observed discharge times are as short as 300 picoseconds, and that's with an additional capacitor and load, necessary to observe it. With rather common parts, it's possible to get single digit nanoseconds or less.
In practical terms, an avalanche pulse generator consists, of course, of the avalanche transistor itself. The 2N2369 is typically used, getting sub-ns pulses quite easily. I have found the common 2N3904 works, at somewhat higher voltage and probably a bit slower. It's still near my oscilloscope's maximum risetime, which is fast enough for my purposes. Some hardware must be provided to channel the discharge, which (as I've hinted) means a storage capacitor, a transmission line (for output), and a high voltage power supply to charge the capacitance. Two designs are given by Jim Williams of Linear Technology: one in AN47, High Speed Techniques, pp.93-95, and more of a "luxury model" in AN94, The Taming Of The Slew. Both of these are available for download from Linear Technology.
This is the pulse generator I built. (The complete circuit is here, and is a little wide to put inline.) The unit consists of a power supply PCB on the bottom and the high-falootin' stuff wired tightly point-to-point. The power supply consists of a resonant inverter (the toroid and associated parts on the left), voltage doubler (furnishing 150-300VDC, depending on the 9-18V supply) and a variable linear regulator (100-150V or so) with current limiting (2mA). The high voltage feeds through an RFC and filter capacitor to the charging resistor. The charge feeds into the capacitor and any transmission line I may happen to have connected. When the voltage reaches the breakdown voltage of the 2N3904, it switches on hard and the charge side is quickly connected to the output. The pulse is reigned in by a little attenuator and sent out. By storing charge in an unterminated transmission line, I can get a pulse with a sharp rising edge, flat top and well defined width. For my sampler tests, I used a 3' cable, which gives about 10ns width. Observed rise and fall times are on the order of my scope's rating (1 ns), so it's pretty fast.
Hmm, I don't actually have any pictures of this thing's waveform! I'd better go get pictures some day.