A model for Intelligent Random Access Memory architecture (IRAM): cellular automata algorithms on the Associative String Processing machine (ASTRA)

In the near future, the computer performance will be completely determined by how long it takes to access memory. There are bottle-necks in memory latency and memory-to processor interface bandwidth. The IRAM initiative could be the answer by putting Processor-In-Memory (PIM). Starting from the massively parallel processing concept, one reached a similar conclusion. The MPPC (Massively Parallel Processing Collaboration) project and the 8K processor ASTRA machine (Associative String Test bench for Research \& Applications) developed at CERN \cite{kuala} can be regarded as a forerunner of the IRAM concept. The computing power of the ASTRA machine, regarded as an IRAM with 64 one-bit processors on a 64$\times$64 bit-matrix memory chip machine, has been demonstrated by running statistical physics algorithms: one-dimensional stochastic cellular automata, as a simple model for dynamical phase transitions. As a relevant result for physics, the damage spreading of this model has been investigated.