For many decades, dynamic random access memory, or DRAM, has been
the technology of choice for use
as core processor memory storage. Although the functionality and
general access characteristics of DRAM
have not changed dramatically since its inception, the technology
has evolved by continually improving in
overall bit density. However, the success of DRAM has also enabled
the emergence of the processor-memory
latency gap. Much effort has been spent improving processor
functionality and redesigning
memory hierarchies to limit the effects of this growing gap in
performance. In recent years, there has been
a rapid onset of different designs proposed to attack the problem
at the memory itself.
A summary of current DRAM and SRAM technologies focused on this
problem is given, enumerating the
specific design characteristics that differ among the proposals.
Next, we present two technologies,
intelligent RAM and magnetic RAM, which provide a fundamentally
different perspective from
conventional DRAM architectures for bridging the processor-memory
gap. Our study is concluded by a
discussion on the future of memory architectures.
Paper available as: Postscript
or PDF