Presentation 2010-03-09
SHIN : A Hardware Architecture for over 100M Neurons Real-time Neural-net
Shinjiro TOYODA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) SHIN: Synchronous Hierarchical Information Network is a data transfer architecture which solves the enormous wiring issue among the over 10^8 neurons with 10^<12> synapses. At first, a virtual plane is divided into several hexagon cells. Then, the neurons in a neural-net are sprinkled into cells so that its layer formation should become vertical to the plane. Neurons in a cell can get information from neurons in the cell and neighbor cells. This architecture makes it possible to transfer information among the neurons by several local low speed signal lines. Only a signal line should be added when a new cell is applied to a system. The data transfer rate and the calculation speed of the neurons are not affected. Therefore, it is possible to make a neural-net hardware as big as we need.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) big scale neural-net / real-time operation / self-organization / spatial-divide / hardware / SHIN
Paper # NC2009-92
Date of Issue

Conference Information
Committee NC
Conference Date 2010/3/2(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) SHIN : A Hardware Architecture for over 100M Neurons Real-time Neural-net
Sub Title (in English)
Keyword(1) big scale neural-net
Keyword(2) real-time operation
Keyword(3) self-organization
Keyword(4) spatial-divide
Keyword(5) hardware
Keyword(6) SHIN
1st Author's Name Shinjiro TOYODA
1st Author's Affiliation Honda Research Institute Japan Co., Ltd.()
Date 2010-03-09
Paper # NC2009-92
Volume (vol) vol.109
Number (no) 461
Page pp.pp.-
#Pages 6
Date of Issue