Building up neuromimetic machines with LNeuro 1.0

The state of experiments on neural networks simulations on a parallel architecture is presented. The computing device, LNeuro 1.0, is based on an existing coarse-grain parallel framework (INMOS Transputers), improved with finer grain parallel abilities through VLSI modules. A digital architecture, scalable and flexible enough to be useful for simulating various kinds of networks and paradigms, was retained. A small-scale machine has been realized using 16 LNeuros arranged in clusters composed of four circuits and a controller, to experimentally study the behavior of neuromimetic processes (communication, control, primitives required, etc.). Results are presented on an integer version of Kohonen feature maps, the speedup factor increasing regularly with the number of clusters involved (up to a factor of 80). Some ways to improve this family of neural networks simulation machines are also investigated