• DocumentCode
    3496548
  • Title

    A MULTINEURO computer system by using high performance digital neuro processor

  • Author

    Saito, Takashi ; Baba, Yasushi ; Gawa, Koichi Naka ; Fuwa, Yutaka ; Shimokawa, Yoshiyuki

  • Author_Institution
    Dept. of Microelectron. Syst. Instrum., Toshiba Corp., Tokyo, Japan
  • fYear
    1995
  • fDate
    26-28 Jul 1995
  • Firstpage
    1419
  • Lastpage
    1424
  • Abstract
    We have developed an accelerator device “MULTINEURO-A” which incorporates multiple processors (one master node and plural slave nodes) dedicated to neural network high-speed computing realized on an engineering workstation. Nodes are connected by two data paths, broadcast bus and ring bus. Processing speed is in proportion to the number of processors. It is built in two boards (master and slave) having 64 VLSI processors that offers to 1.5 billion connections per seconds. It processes forward/backward calculation of multilayer perceptron type neural networks, feedback type neural networks such as Hopfield model, and any other types by programming for parallel processing
  • Keywords
    Hopfield neural nets; VLSI; engineering workstations; multilayer perceptrons; multiprocessing systems; neural nets; special purpose computers; system buses; Hopfield neural networks; MULTINEURO-A; VLSI processors; accelerator device; broadcast bus; engineering workstation; feedback neural networks; forward/backward calculation; high performance digital neuro processor; multilayer perceptron; multiple processors; neural network high-speed computing; parallel processing; ring bus; Computer networks; Feedforward neural networks; High performance computing; Hopfield neural networks; Master-slave; Multi-layer neural network; Multilayer perceptrons; Neural networks; Very large scale integration; Workstations;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    SICE '95. Proceedings of the 34th SICE Annual Conference. International Session Papers
  • Conference_Location
    Hokkaido
  • Print_ISBN
    0-7803-2781-0
  • Type

    conf

  • DOI
    10.1109/SICE.1995.526722
  • Filename
    526722