L-Neuro 2.3: a VLSI for image processing by neural networks

Real-time and embedded applications of image processing like pattern recognition, shape analysis etc. (using classical or less classical methods such as neural networks) are computer intensive tasks that lead to complex systems. Furthermore, the skyrocketting demand for those techniques has led to a flurry of algorithms that must be rapidly implemented, evaluated and finally tuned to real-world cases. This is why LEP has developed the fully programmable vectorial processor L-Neuro 2.3, which is a parallel chip composed of an array of twelve DSPs (Digital Signal Processors). It can be used for neurocomputing, fuzzy logics applications, real-time image processing, digital signal processing and all applications that can take advantage of cooperating DSPs. The now available chip is able to perform up to 2 Giga arithmetic operations per second, and has a peak throughput of 1.5 Gigabytes per second.