Of paramount importance is that in the limit,
locality of action will dominate operational behavior
of nanoscale computing systems and that bandwidth
of storage access will determine the ultimate
sustained performance of systems of fixed size.
While locality even today is of importance and drives
much of the attention on cache based hierarchies,
these effects are trivial when compared to the degree
and tightness of locality that will be imposed on
nanoscale computing structures and methods. The
metric W introduced in this paper suggests at least a
three order of magnitude difference between current
technologies at or just below 0.1 micron and the
asymptotic technologies at the end of the next decade
that will exhibit basic feature sizes almost a hundred
times smaller. New architectures will be required that
enable the exploitation of unprecedented amounts of
fine grain parallelism and that are latency tolerant for
remote interactions. In addition, the overhead
mechanisms for managing the parallel physical
structures and concurrent logical activities must be
very efficient to achieve real scalability through the
massively parallel execution of fine grain tasks.