Successive independent batches of data are written into a memory and then read out. The behavior of a memory cell in each read/write cycle is dependent (this is the “aftereffect”) on its previous history. Each cell can be modelled as an automaton. Using a large number of identical cells and coding/decoding across the memory, what is the maximum throughput that can be achieved in N cycles with negligible error probability?
This problem is equivalent (by a time/space interchange) to finding the total capacity of a certain multiuser interference channel.
Exact answers are obtained for the cell-model in which the output of a cell is the exclusive OR of the two most recent inputs. For the (more realistic) inclusive OR lower and upper bounds are determined.
The increase in throughput obtainable by delaying some of the read cycles is also determined or bounded.
