The modular neural ring (MNR) architecture for hardware implementation of neural networks has been extended to highly parallel vector processing of digital, as well as neural, applications. The distinguishing advantages of modularity and dynamic reconfigurability are maintained in the modular parallel ring (MPR) architecture. Along with the high degree of parallelism achieved, these properties make the MPR architecture well suited to intelligent vehicle applications, where hybrid coprocessing (neural and digital) is essential for high efficiency. The authors present the MPR architecture as an attached hybrid coprocessor to a host computer; the MPR is closely coupled to, and is activated by, the host computer as needed. The MPR coprocessor is easily expandable with modular additions, so that it may equip the host computer with the needed high efficiency neural and vector digital processing facilities. The hardware MPR architecture is programmed through the host computer, and it is dynamically reconfigurable to suit the requirements of the vector digital and neural processing task. The required system and utility software, which will make the MPR coprocessor user friendly, are discussed.<>
A high performance modular architecture for hardware implementations of neural and digital applications
1992-01-01
636183 byte
Conference paper
Electronic Resource
English
Neural networks: implementations and applications
IEEE | 1996
|Hardware Implementations of Data Compression Algorithms for Aerospace Type Applications
British Library Conference Proceedings | 1993
|