Applications of a connection code computer for neural network calculations

A model of a connection code computer, which we proposed as a reconfigurable hardware, is developed to examine application to neural network models. This hardware is able to speed up the parallel and pipelined calculations by realizing the optimal composition (hardware structure) for various problems. As the example, one of typical neural network models, that is, back-propagation model, are simulated Each component is improved to reduce a chip-size. All components are designed and are implemented to check a size and their performance.