Recent developments of photonic computers and memristor memories brings us closer to being able to implement the reverse Von Neumon CPU I have talked about previously. By eliminating the bottleneck of copper between the CPU and memory and taking 1024's of CPU to the memory, the potential dead time of CPU use is eliminated.
Memristors: The lastest in molecular memory. A single atom state represents the bit; ie: the atom is in one of two states of excitation. However there is a big problem, no copper, gold, silicon wire is going to be able to connect to a single atom on the grid scale required for a desnse memory structure. But there is one one way, Optical.
Suppose we set up an optical Bidi, ie: one of those black/white squares) where each "pixel" block represents the state of a single atom. Redundancy could be incorporated by allowing each "pixel" to cover squares 2x2, 4x4 etc of atoms and we read the mean.
So we create a bidi with the CPU, store in the memristor. To read, the reverse, we flood the section of memory with a optical span and note the disturbance to reconstruct the bidi. Since we are dealing with one of these new photonic CPU, we can flood in all the data and compute the result at the same time. Now if we use the Harvard approach both the program and data are read at the same time, the result is produced almost at the same time, so by finding a means to modulate the flood to bidi intensity we store the result.
Obviously 1024's of CPU can be paralleled using then same synchronised flood-bidi sequence. If this can be done using bi-colour states that would make it easier, the separation of events.
We still have the problem entering the program, initial data and retrieving the results as this will be to copper world, so whta is needed is a copper based memory with optical i/o. Flood the copper side i/o and focus the result on the memristor. Likewise to get the result, flood the memristor and and focus on the copper side.