Yes, to make it clear: a photoresistor is a monode -- no junction, just contacts on a blob of semiconductor. Doping would provide free carriers and thus dark current, so the semiconductor must be intrinsic (high purity, low defects).
With the common CdS LDRs there is another mechanism to increase the sensitivity: There are some relatively long lived trapping centers that catch the created extra carriers of one polarity. I don't know if this are the electrons or holes that are caught, but the principle is the same. As long as the hole is captured in a long lived state the corresponding electron has no easy way to recombine and can freely move and contribute to an increased conductivity. The longer the holes are trapped the more (the longer) the electrons can contribute to conductivity. Though half the carriers are caught and thus immobile, this mechanism can increase the sensitivity quite a bit, as the partner stays mobile and gets a much longer effective life time (e.g. ms instead of ns range). So slow reponse and good sensitivity are kind of linked.
Ah, this must be related to the effect I measured back in college...
The experiment was just measuring a simple impulse response, illuminating a CdS photocell with a camera flash and watching the current flow through a shunt resistor. Evidently, I had chosen the resistor too large, resulting in a large signal level (though not clipping, as I recall). I got a result different from the rest of class, which I had no explanation for. But it was a clear effect, having a sqrt(t) dependency. Or at least, a better fit than for the linear curve.
Ah, here's the relevant plot:
The lumps are quantization noise from the shitty scope used to acquire it (the right side corresponds to a tail of some 10s mV); it's pretty fair to average through them.
If the material had a simple time constant behind its conductivity (as you'd more or less expect from recombination), R should increase as a single power of time; but it actually goes at the half power. At least, so it seems, here. If that was actually the finding that others made (hm, I don't recall if I looked at anyone else's data), it could be that extra charges get trapped when the dwell time is longer, versus being cleared fairly quickly from the junction under bias; hence the time dependency that other students apparently didn't have.
sqrt(t) suggests either a diffusion mechanism, or an ensemble effect where -- if it's trapping centers, it might be there are a wide variety of them present, some short, some long, and they happen to be distributed in this manner.
Hmm, photoflash isn't exactly brief, by itself (~fractional ms usually), and this whole event is only a ms long. That should have a linear, or truncated exponential, light curve though, and wouldn't match the data here. I didn't happen to save the whole waveform (including rising edge and peak; only the decay). Any difference due to terminal voltage, can only amount to a few microseconds of difference, as the voltage drops to normal values fairly quickly...
(The prof found my answer of "Idunno" rather unsatisfying...
but also didn't have any insights himself. Heh, I wonder if he expected/wanted me to BS it. Wouldn't have known what to research for, at the time. Anyway, if it's trapping centers, how would I even be able to verify that hypothesis without a hell of a lot more work (likely including fab work)?... ah well.)
Tim