DocumentCode
596833
Title
Dedicated FPGA communication architecture and design for a large-scale neuromorphic system
Author
Thanasoulis, Vasilis ; Partzsch, Johannes ; Hartmann, Steve ; Mayr, Christian ; Schuffny, Rene
Author_Institution
Parallel VLSI Syst. & Neuromorphic Circuits, Tech. Univ. Dresden, Dresden, Germany
fYear
2012
fDate
9-12 Dec. 2012
Firstpage
877
Lastpage
880
Abstract
The communication infrastructure is a crucial part of neuromorphic VLSI systems. A sophisticated network is required for the transmission of pulse events among the individual neuron-and-synapse circuits and the configuration data for the components of the system. We present a packet-based communication network implemented on FPGA that provides pulse stimulation, inter-node connectivity and allows the complete control and configuration of the neuromorphic system by a host unit. The design deploys a specific architecture for optimized throughput and implements a reliable and ordered data delivery. It employs also a large Playback/Trace memory that offers pulse event replay and recording of the produced neuronal activity. Altogether, our implementation is more versatile than recent FPGA-based neuromorphic infrastructures, allowing for fast system configuration, stimulation and monitoring.
Keywords
VLSI; field programmable gate arrays; neural nets; communication infrastructure; dedicated FPGA communication architecture; internode connectivity; large-scale neuromorphic system; neuromorphic VLSI system; neuron-and-synapse circuit; packet-based communication network; playback memory; pulse event recording; pulse event replay; pulse event transmission; pulse stimulation; trace memory; Computer architecture; Field programmable gate arrays; IP networks; Neuromorphics; Neurons; Reliability; Throughput;
fLanguage
English
Publisher
ieee
Conference_Titel
Electronics, Circuits and Systems (ICECS), 2012 19th IEEE International Conference on
Conference_Location
Seville
Print_ISBN
978-1-4673-1261-5
Electronic_ISBN
978-1-4673-1259-2
Type
conf
DOI
10.1109/ICECS.2012.6463548
Filename
6463548
Link To Document