I have a pile of bluish white LEDs salvaged from Christmas lights. They look like ordinary LEDs, with one electrode having a little cup with the light emitter, and the other with a tiny wire going to the top of the emitter. When I apply a milliamp or so from a 9v supply, they light up just fine. If I apply 10 microamps or so, they barely light when power is applied, then get brighter over the next few seconds. It is clearly visible to the eye, and makes them useless for making optisolators with. Obviously the LEDs used in commercial optoisolators don't have this problem. Why do these LEDs do this, and what kinds of LEDs do it?
Bare LEDs are generally quite fast. The ones with phosphor to re-emit a different color are slower. How much slower depends on the phosphor. White LEDs, for example, have phosphors. Those would be silly in a opto-isolator.
Commercial opto-isolators use infrared LEDs, usually in the 9xx nm wavelength.
Look at the current spec for commercial opto-isolators. Note that they require significantly more than 10 µA of input current to work. What a LED does with only 10 µA is irrelevant to opto-coupler use since they aren't being run at such low currents. Usually a few mA is expected.
The most plausible reason for this behavior is that you only think the current though the LED is constant where in fact it isn't. Plot the current with a scope for the few seconds while the brightness changes: I'm pretty sure you won't see the flat line you'd expect.
They are UV LEDs with multiple phosphors, designed for efficiency at medium current even more than high brightness (i.e. they're not illumination but decoration). The UV is detectable if I look for it. They do, in fact, take observably long times for the phosphors to both activate and deactivate.
re LED illumination, I found this in another forum, and a densely mathematical article that backs it up down to the quantum level, where individual photons are emitted.
"There is no absolute minimum current required to make an LED emit light; using a sensitive photodetector, I've detected light output from LEDs (green "ultrabright" indicator lamps) driven by currents as low as one nanoampere. They probably emit at even lower currents, but that's where my photodetector bottomed out."
A datasheet for something similar to what I build only shows the LED current/opto resistance graph out to 100k, also states that the off resistance is 50MEG, and suggests contacting the factory for specially selected units if you want more accuracy at low current levels.
And a commercial use of such an optoisolator shows a maximum current through the LED of less than 1ma, linear use rather than on-off switching, and an active resistance range for the optoresistor up to 680k or so, successfully running it at microamp levels.
So it appears that both the LEDs and photoresistors work at those levels, just with less precision.