Reconfigurable Parallel Hardware for Computing Local Linear Neuro-Fuzzy Model

Because of the computational-intensive nature of local linear neuro-fuzzy model (LLNFM), effective employment of LLNFM requires dedicated hardware implementation that can be reconfigured for different applications such as pattern recognition, system identification with ensured reusability. This paper introduces a new versatile and reusable parallel hardware engine for computations of LOLIMOT algorithm in local linear neuro-fuzzy model. Our LOLIMOT hardware engine exploits the inherent parallelism and reusability of data and redundancies of computation with its effective caching scheme and parallel processing engines. The designed hardware element for feed forward step can also perform training step with acceptable overhead which is implemented. This overcomes the on-chip learning complexity cost. Synthesis and implementation results on FPGA beds are presented to show its power of computations on reconfigurable platforms