The potential impact of neuromorphic computing in society is significant and manifold. As big data are overwhelming our world, we need tools to collect, analyze and act on these data in real time. Example applications include smarter security cameras and smart-city infrastructure, autonomous vehicles, etc.

Traditional Von Neumann architectures can be programmed to execute many billions of arithmetic operations on data per second, but they cannot keep up with the current data deluge. One challenging problem that these architectures face today is referred to as “hitting the memory wall” – the root of the problem is that every elementary instruction needs to bring both data and control from memory, thus increasing power consumption and turn-around latency over a shared resource. Neuromorphic computing, following the architecture of brains, operates on data in a more distributed (“data flow”) manner, thus overcoming the serialization and the power consumption of bringing data from and to memory. But even more importantly, the elementary functions for which current Von Neumann architectures are optimized for, although demonstrating one aspect of intelligence, are not directly linked with the higher–levels of cognition that we are after to at this stage of computing in modern societies.

Neuromorphic computing envisions new systems optimized for such higher levels of cognition, but in order to deliver these, it needs more data from neuroscience, which the neuronsXnets will offer.