Memristive and memcapacitive event-based neuromorphic computing
Academy of Finland
This research is motivated by the fact that the computing efficiency of biological systems exceeds overwhelmingly that achieved by the state-of-the-art integrated circuit technology. For example, it can be estimated that the human brain processes information at least 8 orders of magnitude more efficiently than is feasible in digital integrated circuits due to the energy efficiency wall, and at least 5 orders of magnitude more efficiently than is currently possible in analog domain. Partly this efficiency gap can be explained by the highly efficient use of physical and chemical computing primitives in biological systems. Another source of extremely high computing efficiency is that biological systems are driven by events taking place in the outside world, asynchronously and in continuous time. Such event-based computing scheme yields high compression in data communication as events — in the case of biological neurons, spikes — are generated only when the input of a neuron changes by a sufficient amount. In this project, the above described biological information processing principles are applied to computing systems which take advantage of the physical properties of memristors and memcapacitors, in order to develop extremely efficient hardware solutions for sensory information processing.