An M-SIMD hardware architecture for neural and digital hybrid applications

A modular, reconfigurable, parallel and linearly scalable hardware architecture for realization of large-scale neural networks has been developed. Called the modular neural ring, the architecture has been prototyped and shown to be highly effective in hardware implementation of large-scale neural computing models. The authors extend the application of this neural ring architecture to neural and digital processing. The proposed hybrid computing architecture has been tested and has been found to offer a uniform hardware platform for highly parallel, modular, and reconfigurable implementations of both digital and neural processing tasks. Performance evaluation of neural model implementations and examples of application to matrix and vector digital computing are presented