A hardware supported system for a special online application of self-organizing map

In this paper we present a hardware supported system which enables the parallel computing of Euclidean distance with Kohonen´s self-organizing map (SOM) for a special online application. The hardware is implemented as neural coprocessors connected to a Sun workstation via a parallel interface. For the aim of the computing time requirement, a four step pipeline and parallelism are used. The algorithmic description of the hardware and a simulation environment have been written in VHDL. Field programmable gate arrays are used to implement these neural coprocessors. The system can be used in various applications of neural networks