The analysis of parallel information-processing systems, independent of their realization in localized or distributed form, reveals that the necessary data exchange between data-storage and data-processing parts of the system represents a major limiting factor for the system performance. Therefore, in addition to the number-crunching power of the processing parts, innovations which substantially improve the bandwidth of the data exchange are essential for advances in the overall system capabilities. To be practically useful, such innovations must in particular carefully balance the bandwidth of the exchanged data amount against the required power dissipation for this purpose. We review the main methods for achieving the data exchange improvements, namely (a) an increased memory-access bandwidth by multi-porting of the memory and (b) a unification of memory and processing parts of the information-processing system. We further present the advances in VLSI architectures for realizing the higher data-exchange bandwidth by applying advanced nano-technologies and discuss practical implementation examples for parallel processors as well as for pattern-matching and pattern-recognition systems.
展开▼
