Researchers at the University of California have designed a hierarchical, fully-scalable machine, composed of multiple processing nodes and a global communications network, that can support multiple grains of parallelism. Based on an advanced associative processing model that emphasizes the access of compound data objects by content rather than location, this new architecture extends the fundamental operations of content-based retrieval and processing to complex, dynamic objects and should significantly increase the processing speed for numerous applications. The model and architecture design incorporate recent algorithmic techniques to reduce expected-case asymptotic times, so hardware speedups can significantly increase tractable problem size. The researchers estimate that such a machine with 16K processing elements will be about 1000 times as fast (well over 100,000 thirty-two bit MIPS) as recent workstations for key algorithms.
APPLICATIONS: This massively parallel, associative processing architecture will provide improved symbolic computation capabilities that will benefit work in AI, machine learning, information retrieval, programming languages, cognitive sciences, psychology, and other areas. Consequently, this new architecture design would lead to the generation of a large body of cost-effective, end-user applications in a wide variety of fields such as management information systems, enterprise modeling, knowledge-based systems, computational chemistry and biology, medical informatics, speech and character recognition, natural language understanding, robotics, and computer-aided design.
ADVANTAGES:
REFERENCE: UC Case 94-082