Silicon Brains

Abstract

Beyond the digital neural networks of Chap. 16, the more radical mapping of brain-like structures and processes into VLSI substrates has been pioneered by Carver Mead more than 30 years ago [1]. The basic idea was to exploit the massive parallelism of such circuits and to create low-power and fault-tolerant information-processing systems. Neuromorphic engineering has recently seen a revival with the availability of deep-submicron CMOS technology, which allows for the construction of very-large-scale mixed-signal systems combining local analog processing in neuronal cells with binary signalling via action potentials. Modern implementations are able to reach the complexity-scale of large functional units of the human brain, and they feature the ability to learn by plasticity mechanisms found in neuroscience. Combined with high-performance programmable logic and elaborate software tools, such systems are currently evolving into user-configurable non-von-Neumann computing systems, which can be used to implement and test novel computational paradigms. The chapter introduces basic properties of biological brains with up to 200 Billion neurons and their 10¹⁴ synapses, where action on a synapse takes ∼10 ms and involves an energy of ∼10 fJ. We outline 10x programs on neuromorphic electronic systems in Europe and the USA, which are intended to integrate 10⁸ neurons and 10¹² synapses, the level of a cat’s brain, in a volume of 1 L and with a power dissipation <1 kW. For a balanced view on intelligence, we references Hawkins’ view to first perceive the task and then design an intelligent technical response.