High-speed analog-to-digital conversion is very difficult, so a common way to make high-speed oscilloscopes is to store the signal after the trigger in an analog form for long enough to analyze it at leisure. This is of course a description of how entirely-analog oscilloscopes work, but for example a multi-gigahertz digital oscilloscope vendor in the early 1990s told me his scope stored the data in an internal CRT — a sort of analog version of the Williams tube — until it had time to analyze it.
Oscilloscopes are a particularly tricky kind of thing to build out of random electronic crap you find in the junk pile because that crap typically doesn't include any ADCs over 10Msps (some scanners contain 6Msps ADCs), and you really need at least 40Msps or 60Msps for an entry-level 20MHz oscilloscope. (Keep in mind that an analog 20MHz oscilloscope isn’t incapable of viewing signals above 20MHz; that’s just its 3dB attenuation frequency. Sub-nanosecond signals will probably be phase-shifted and badly attenuated but they’ll still be there.)
So it occurred to me that maybe a discarded obsolete hard disk could bridge this gap. Suppose we’re talking about a current 15krpm Seagate Cheetah with its 204MB/s data transfer rate, which (if it’s on one head) implies that the waveform at the disk surface includes significant, reliably recoverable components at up to 800MHz. The disk is rotating at 250Hz. Once a waveform is recorded, it is then repeated at the read head over and over again, every 4 milliseconds, until either the head is moved to another track or a new waveform is recorded. We have 4 milliseconds of waveform recorded, which would amount to 3.3 million cycles of the highest frequencies recorded and could thus be fully digitized in about a second using the 6Msps scanner ADCs I mentioned earlier; but in a much more typical case, you only care about a few hundred or thousand sample points after the trigger event. And you can digitize a few of them on every revolution until you have them all digitized.
Considering digitizing 1000 points at 40 million samples per second, well, that’s 25 microseconds, which is 150 samples at 6 megasamples per second. You can digitize points #0, #7, #14, #21, and so on on the first revolution of the disk; #1, #8, #15, #22, etc., on the second; and in this way after 7 revolutions of the disk (28 ms) you have digitized the whole event. Digitizing at higher effective sample rates, or using a slower ADC, would require proportionally more revolutions.
Even ordinary disks (5400 rpm, 50 MB/s) should still be capable of functioning effectively in this role.
A problem with this pretty picture is that disks are not really designed for analog signal integrity, and so the signal may be corrupted with noise and subject to hard-to-characterize nonlinearities. And of course you need to degauss the track before recording small analog signals on it.
(See also files TV oscilloscope, VCR oscilloscope, Laser printer oscilloscope, and CCD oscilloscope.)