Description
Neuromorphic architectures like IBM TrueNorth promise to deliver massively data parallel computing combined with unprecedented energy efficiency, potentially offering an avenue for next generation high-performance systems in the post-Moore’s law era. To better understand the range of problems to which neuromorphic computing is advantageous, we have assembled a diverse portfolio of real-time, data-intensive scientific applications.
Here, we explore neuromorphic computing as a low-power computing platform for both on-sensor processing and its potential role in a hybrid exascale computing platform. Specifically, our goals are: (1) understand the challenges of using spiking neural networks, and (2) evaluate neuromorphic hardware relative to von Neumann architectures. This will provide under ‘problem space’ for which neuromorphic computing is applicable.
Author
Dr
Kristofer Bouchard
(LBL)
