Could you construct universal sequential digital logic with just LEDs?
It’s straightforward to use LEDs for diode logic, which can give you sum-of-products logic, up to monotonicity — you can get any monotonic logical function that way, but diode logic alone doesn’t give you inversion.
LEDs can function as photodiodes, although not very good photodiodes. So you could imagine using the light from one LED to switch another LED. But it would seem that you can’t get any current gain that way: each charge-carrier pair that gets annihilated in the transmitting LED produces at most one photon, and then produces at most one charge-carrier pair in the receiving LED. And there are losses at every stage of this process, thanks to non-ideal quantum efficiencies and the like, so you can’t even get to unity gain.
I think there are at least four ways to solve this problem, which sort of blur into each other.
You can get voltage gain, because the voltage in the emitting LED will be close to its usual forward voltage, say 1.6V, while I think the voltage in the receiving LED can be much higher if it’s back-biased, say 5V. But it’s easy to trade that off for current gain by putting LEDs in series. For example, you can put three 1.6V LEDs in series and thus generate three photons per charge carrier.
You may be able to increase the number of photons with fluorescence, though at the cost of speed.
You can use a “regenerative” design using positive feedback, in which the back-biased receiving LED is in series with one or more forward-biased LEDs which also illuminate it. This way, most of the electrons produce one or more photons on their way to wherever they’re going, thus allowing another charge carrier pair to spawn in the receiving LED.
(One problem that occurs to me with the above techniques is that it’s going to be hard to get more irradiance at the photodetector junction than, like, in the rest of these diodes that are glued together.)
You can initiate an avalanche discharge in the receiving LED and directly get current amplification after all, similar to how SPADs work. Like, if you’re close to the diode’s reverse avalanche voltage, maybe you can reduce that voltage threshold by varying irradiation, and thus get both voltage gain and current gain.
You can get amplification through a bridge configuration. As long as you don’t exceed the reverse breakdown voltage, an LED works as a (not very sensitive) differential voltage detector. However, this still suffers from a lack of current gain.
You can use LEDs as if they were PIN diodes to switch RF signals by changing their capacitance with a DC bias, providing enormous current gain (like a JFET, leakage current down in the femtoamp range controlling an RF current up in the milliamp range) despite below-unity voltage gain. But then how do you rectify the RF signal? A faster LED, I suppose.
Apparently a 5-mm red LED can generate over 20 μA as a photovoltaic diode in full sunlight, while 1N4148 diodes only generate about 10 nA. Assuming a 19.6 mm² area and 1000 W/m², the total solar power incident on the diode is 19.6 mW, which would be 12.3 mA at 1.6 V. So that’s an efficiency of about 0.16%, compared with 16% for common low-cost photovoltaic cells.
Typically LEDs work better to detect slightly shorter-wavelength light, which is a major reason the red LED has such poor efficiency in sunlight. So that 0.16% might really be a quantum efficiency on the order of 2% or so in the right wavelength band.
These data make me think that getting even a current “gain” of 0.1 is going to be quite difficult with the first three approaches, much less getting a current gain above 1.0. The situation can be improved somewhat by heating up the LEDs you want to have lower bandgaps and cooling down the ones you want to have larger bandgaps, but maybe not to the point where those approaches are feasible.