Abstract

Scientific computation typically involves processing large amounts of data in which operations using main memory and mass storage are frequent. A performance bottleneck to and from main memory arises because only a small number of input and output pins are provided on electronic memory integrated circuits, which means that large portions of the memory cannot be accessed in parallel. This bottleneck is compounded by optical memories, in which entire pages can be brought into a system in parallel. Here, we describe a concept architecture that improves the access time to a parallel memory through the use of an optical interface.

© 1995 Optical Society of America