Investigating Architectural Issues in Neuromorphic Computing
Abstract
This effort has explored the issues associated with the efficient mapping of neuromorphic computing strategies onto advanced computational architectures. This multidisciplinary effort combined concepts and research from diverse fields including computer architecture, neuroscience, cognitive psychology, cognitive modeling, dynamical systems, software and computer engineering. It explored multiple columnar cortical models reported in the literature, and produced new models by combining ideas with insights developed by the research team. These models range in scale of abstraction from cell assemblies of individual minicolumns to models that represent abstractions of hundreds of thousands of synapses and neurons. Selected models were also emulated. Columnar model software produced by this effort includes C code and FPGA VHDL code. The VHDL code consists of "accelerations" of Bayesian tree recall algorithms, and Brain State in the Box point attractors. The C code consists of these algorithms, models of minicolumns and functional columns, spiky neuron models, and Confabulation models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2009
- Accession Number
- ADA501732
Entities
People
- Daniel Burns
- Michael J Moore
- Qing Wu
- Qinru Qiu
- Richard W. Linderman
- Tarek Taha
Organizations
- Air Force Research Laboratory