61265Re: [beam] Memristor
- Jan 13, 2014That is really turning out nicely. I like the silver paint tweak! I also want to know more about crossbar latches, as they sound very useful and intriguing. They remind me vaguely of PAL devices. While the concept is aimed at nanoscale digital computing, I would think that it could make for a practical neural computer as well, seeing as memristors have more than two states.
Taking things a step further, I've always been interested in the idea of two PLDs arranged symmetrically, able to communicate between each other, but each one also capable of programming the other. The reason for my interest being that one PLD could act as the processor, and the other act as the memory. Since PLDs are capable of compressing information down to sums, and reproducing it quickly, the PLD dubbed as memory would be polynomially more efficient at storing information than any memory array. And the PLD dubbed the processor could potentially perform very complex operations immediately and with minimal gate delay, without growing in size or complexity. And the stored program has the potential to alter the very architecture of the computer!
The only problem is this: When sets of terms are reduced to sums of products and products of sums, the individual terms themselves cannot be distinguished, making it impossible to determine if a particular bit belongs to just one, or many terms. An attempt to remove a term could inadvertently introduce an undesired term, or remove another desired term. Therefore, such a computer cannot operate in a deterministic fashion. The machine would either have to be nondeterministic, in which it would be able to correctly guess which bits to modify every time, or quantum, in which the original terms are still existent in the form of a probability distribution across each qubit. In a nondeterministic scenario, the only applicable principle is to perform every possible alteration of the target information in parallel at each step, and converge on the most satisfactory sequence.
Why am I saying this? Because a memristor based crossbar latch could be very capable of operating in a nondeterministic way. While the machine would still hold only Boolean values, each bit can have any analog value above or below logic threshold. Each analog value and its inverse would indicate the probability of a bit belonging to any particular sequence. Therefore choosing a satisfactory sequence boils down to selecting a probability, and matching the most probable bits, a simple operation for an analog circuit.
Hence, it may be possible to build a universal computer out of only memristors, which you are working on an easily producible array of right now. Not only a full computer, but a fundamentally very powerful one at that.
- << Previous post in topic Next post in topic >>