NETMAP: An integrated system for building analog neural networks

• Published in: Journal of systems architecture Vol. 44; no. 1; pp. 63 - 73
• Main Authors: Cancelo, Gustavo I.E., Hansen, Sten
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-10-1997; Elsevier Sequoia S.A
• Subjects: Algorithms; Analog signal processing; Computer architecture; Computer peripherals; Digital signal processors; Multi-layer perceptrons; Neural networks; Parallel implementation; Studies; Systems integration; Multi-layer perceptrons; Parallel implementation; Neural networks; Analog signal processing
• ISSN: 1383-7621; 1873-6165
• DOI: 10.1016/S1383-7621(96)00122-1

NETMAP hardware is described in detail along with our solutions to optimize inter-processor data throughput. Four processors of very different nature interact in the system: The PC central processor, a digital signal processor, a bit-slice microsequencer and a neural network processor called ETANN.
A description of the software modules for the processors is given. The weight setting and network training algorithms are also described. Finally, performance considerations and test results are analyzed.