An energy-efficient computing approach by filling the connectome gap

Abstract

This paper presents an energy-efficient neuromorphic computing approach by filling the connectome gap between algorithm, brain, and VLSI. The gap exists in structural features such as the average number of synaptic connections per neural node as well as in dimensional features. We argue that the energy dissipation in complex computing tasks is more predominantly bounded by the control processes that synchronize and redirect both computing processes and data rather than the computing processes themselves. Therefore, it is crucial to fill the connectome gap and to avoid excessive interactions of the computing process and data with the control processes when achieving energy-efficient computing for large-scale cognitive computing tasks. The use of freespace optics is proposed as a means to efficiently handle sparse but still heavily entangled connections. © 2014 Springer International Publishing Switzerland.