Coding in neuromorphic VLSI networks

Abstract

As recent advances in neuroscience are revealing the principles of neural coding used in the mammalian brain (Douglas and Martin 2004, Dayan and Abbott 2001), modeling studies are beginning to show how neural architectures composed of diversified and unreliable computing elements (such as neurons and synapses) can support robust and reliable computation, making use of computational primitives that are of both analog and digital nature (Hahnloser et al. 2000, Rutishauser and Douglas 2009). These studies arise from the observation that the principles of computation used by nervous systems are radically different from those generally used in current computers. Biological neural networks process information using energy-efficient asynchronous, event-based, methods. Biology uses self-construction, self-repair, self-programming, and has learned how to flexibly compose complex behaviors from simpler slow and inhomogeneous elements. Hardware implementations of these biological computational principles offer an attractive alternative to conventional computing technology, and could have enormous consequences for future generations of artificial information processing and behaving systems.