Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network

  Advanced Search  

FP 14.1 Intelligent RAM (IRAM): Chips that remember and compute

Summary: FP 14.1 Intelligent RAM (IRAM): Chips that
remember and compute
David Patterson, Thomas Anderson, Neal Cardwell, Rich-
ard Fromm, Kimberley Keeton, Christoforos Kozyrakis,
Randi Thomas, and Kathy Yelick
Computer Science/EECS University of California, Berkeley CA
The division of the semiconductor industry into microprocessor
and memory camps provides many advantages: fabrication lines
can be tailored to a device, packages are tailored to the pinout
and power of a device, and the number of memory chips in a
computer is independent of the number of processors.
The split has disadvantages as well. While microprocessors have
been improving performance by 60% per year, DRAM access
time has been improving by 7% per year. This processor-memory
performance gap limits many applications. For example, one
microprocessor spends 75% of its time in the memory hierarchy
for data base and matrix computations[1]. These delays occur
despite tremendous resources being spent trying the bridge this
gap. Table 1 shows that up to 60% of the area and 90% of the
transistors of recent microprocessors are dedicated to the grow-


Source: Anderson, Tom - Department of Computer Science and Engineering, University of Washington at Seattle
Kozyrakis, Christos - Departments of Electrical Engineering & Computer Science, Stanford University
Zakhor, Avideh - Department of Electrical Engineering and Computer Sciences, University of California at Berkeley


Collections: Computer Technologies and Information Sciences; Engineering