new paper: A massively parallel neurocomputer

[henrik@robots.ox.ac.uk]

Here is another one ...
I just have placed this preprint (the paper is submitted to MicroNeuro 93)
in the neuroprose archive, file klagges.massively-parallel.ps.Z. 

Cheers, Henrik (henrik at robots.ox.ac.uk)

Abstract((

We have developed a SIMD massively parallel digital neural network 
simulator --- called GeNet for Generic Network --- which can evaluate
large networks with a variety of learning algorithms at high speed.
A medium-size installation with 256 physical nodes and 1 Gbyte of 
memory can sustain e.g. 1.7 giga 16bit-connection crossings/sec at 
network sizes of 2 layers with 64K neurons each, a fan-in of 1K and a 
random wired topology. The neural network core operations are supported
by optimized and balanced computation and communication hardware that
sustains heavily pipelined processing. In addition to an array of 
processing units with one global scalar (16 bit) bus, the system is 
equipped with a ring-shifter (32 bit) and a parallel (256\times16 bit) 
vector bus that feeds a tree-shaped global vector accumulator. This eases 
backward communication and the calculation of scalar products of 
distributed vectors. The VLIW-architecture is highly scalable. A prototype
has been cost-effectively implemented without custom VLSI chips. 

))

FTP instructions:
$ ftp archive.cis.ohio-state.edu

ftp> user ftp
ftp> password <your email-address>
ftp> binary
ftp> cd pub/neuroprose
ftp> get Getps
ftp> bye

$ chmod +x Getps
$ Getps klagges.massively-parallel.ps.Z
$ uncompress kl*.ps.Z
$ lpr -Plp kl*.ps      (or whatever cmd you use for your postscript printer)
=========