Neural network system for massively parallel signal processing

Future development of neural networks and their applications will be strongly affected by the availability of specialized hardware that meets the need for massively parallel communication and allow building of scalable networks. We present a neural network system which consists of special neural network chips, each including a neural processor and a specialized communication hardware to solve this problem. The neural processor enables the emulation of different types of neurons including biologically inspired models based on activity pulses, learnable synaptic weights and pulse delay times. The communication processor realizes full global communication despite locally interconnected chips.