My story starts with my reading a science fact book by Isaac Asimov which introduced me to the first Law of Thermodynamics (1LoT) that energy cannot be created or destroyed and 2LoT that a fraction of heat must become unavailable. I have not found the book again though I have found parts of the theme in several of his other books. I followed up on this by reading Radar Engineering by Donald G. Fink McGraw-Hill 1947 one of my father's college textbook where Johnson noise was presented. Johnson Noise was shown to be consistent for any value of resistance to the extent that electrical resistance wasn't part of the equation for power. High values of resistance express Johnson Noise in high voltage low current form. Low values of resistance express Johnson Noise in high current low voltage form. It was learned empirically when the equipment got good enough during WWII that Johnson Noise was subjected to impedance matching [impedance is resistance in resistors and reactance in capacitors and inductors] where maximum power transfer occurs when the electrical impedance is matched and that maximum power in a circuit was one quarter that of the theoretical. This concept applies to diodes which have a varying impedance. The low resistance high current low voltage form predominates in a massively parallel network. The matching impedance of the electrical load is very low when matching the highly predominant low forward impedance of the diodes. The low voltage is shared by all the elements in parallel. A diode with completely asymetrical conductivity would have one half of the one quarter of the theoretical power defined earlier. Power is Voltage squared divided by resistance so the way to increase the power is to increase the number of diodes.
One theorum states that the power of waves of many frequencies is the sum of the individual waves. Johnson Noise is white Noise out to a limit where the photon energy istoo much,~1THz. The units work out that a Hertz of bandwidth is one degree of thermodynamic freedom [realm of expression] The power of the diodes is then a chain of multipliers including bandwidth [Hertz], the absolute temperature [degrees kelvin], Boltzmann's constant [1.38 x 10-23], 1/2 because the diode passes predominately the forward going part of the power and suppressed the backward part. Next the output is a multiple of the number of diodes. Finally, there is device efficiency; there are losses in anything. I assign 50% here. The product is ~1 nanowatt per diode at room temperature. This used to be insignificant; it is highly practical with nanometer sized diodes.
A micrometer scale prototype was built and tested in 1993. 5,600 Au/GaAs diodes produced ~50 nanowatts in lab conditions. This is enough to demonstrate that it works but without being practical. Anyone can repeat this experiment.
C60 anodes are ~4 million times smaller by area than the Au anode diode used above. The individual diodes are expected to work better too. Prototypes using C60 anodes requires nanofabrication. The C60 anodes are spaced ~ 30 nm apart to provide room for the depletion region to grow under reverse bias. At this spacing there are 100 billion diodes / cm2.