Cellular automata are a well-known model for computation and
simulation of phenomena in physics, biology, or social sciences.  From
a computational point of view, the prospect of actual implementation
as a basic computing device has always been considered.  

For scaling an implementation we assume a division of the cellular
automaton into identical `chips', which have to be interconnected. To
reduce the cost of the interconnection it is assumed that the state
information is compressed on one site and decompressed on the other,
exploiting redundancies in space and time of the cellular automata.
Several questions regarding the issues of achievable compression,
complexity of the compression and decompression circuits are evaluated
with simulations and formal means.  

One insight gained is that a
computationally powerful cellular automaton does not exhibit the
maximum bandwidth requirements, or in other words, a cellular
automaton with given resources can only do one, either communicate
with full bandwidth or perform sophisticated computations.