I see many references to input and output caps helping with "stability" and some outright references to preventing oscillation in both answers on this site, and even in some data sheets. I also have seen references saying that the 7805 is "inherently stable" in both answers on this site and even in some data sheets and app notes. Are the datasheets/people using the term "stability" confused or are there versions of the 7805 that actually can become unstable? If so, I would like to see an example of a circuit in which this would happen.
It's quite challenging to make them misbehave to the point of oscillation in practice, especially with light loads. Try heavy loading, no input capacitance at all and some source inductance, but I offer no guarantees. You can tell as oscillation approaches you'll see reduced phase margin, meaning overshoot/undershoot on load or line changes.
Based on simulation, it looks like something like 500uH with a 0.5A load will be close to oscillation typically. This is a pretty pathological arrangement. The plot below is with 400uH and a 0.55A load which decreases* to 0.5A at t=100usec
* the decrease is done with a tanh function over a period of about 1usec, so not quite an ideal switch opening.
This comment does not extend to other types of regulators (especially LDOs) which are easy to make oscillate. And, of course, in real circuits we prefer to live on the side of "guaranteed to be stable" rather than "guaranteed to oscillate", at least for voltage regulators. The opposite would be true of things that are intended to oscillate- as the old saw says "amplifiers oscillate, oscillators don't".
Edit: I tried a couple quick tests- with a relatively large input capacitor at the regulator (1uF) it shows low level (2.5mV p-p) oscillation at about 8kHz. Much below that it lowers the frequency of oscillation but magnitude remains high. Adding only a capacitor to the output with the input inductor present reduces the stability- 5-10nF is enough to make it oscillate with 400uH inductance on the input and a 0.5A load.
I had a 7805 in an application oscillate. However is was not quite in the way you are thinking here. I had a 7805 that was loaded in such way that the part was getting rather warm. It had a rather measly little heat sink but it was not enough to keep the parts temperature below the critical thermal cutout level of the regulator. As such the part would get really hot and cutout causing the output to drop to zero volts. As soon as the load was removed the part would start to cool off and eventually come back on again. It would oscillate like this at a couple of seconds rate. I also found I could change the oscillation frequency by placing a tool on the small heatsink!!
While a Vreg can be "inherently stable," it's still beneficial to add 'stabilizing' components externally, to prevent the possibility of load instabilities outweighing the internal stability & 'forcing' the regulator into unsta le operation.
To use a more obvious analogy, imagine you're feeding rope to someone as they rappel down a cliff.
- As long as the climber maintains a fairly predictable/steady pace, you (being inherently helpful) can maintain an even, stable tension on the rope.
- Now, imagine the same climber stops for a few minutes, trying to find a new route; then they suddenly slip off the face & drop; then catch themselves on the face again; climb up a foot or two; then fall yet again.
In this instance; while you may have a stable footing, be very good at maintaining stable tension on the rope, etc. Wouldn't it be incredibly useful to have a rope-stabilizing pulley/drag system to keep the rope from getting into unstable movements?