A reconfigurable 'ANN' architecture
T.H. Madraswala, Bassam Jamil Mohd, Muhammad Masroor Ali, R. Premi, Magdy Bayoumi · 2003
Proposes a design of a digital artificial neural network (ANN). The architecture is based on a single-instruction multiple-data (SIMD) processing configuration. Communication is done through broadcasting and also by systolic methods. With the help of a microprogrammed control unit, the design is mainly capable of implementing the following three models: (1) Hamming, (2) Hopfield, and (3) Carpenter/Grossberg algorithms. The architecture was also designed to achieve parallelism, modularity, adaptability, flexibility, speed, low cost, smaller silicon area, and expandability.>