Bio-inspired System Architecture for Energy Efficient, BIGDATA Computing With Application to Wide Area Motion Imagery

TOMAS FIGLIOLIA; GASPAR TOGNETTI; MARTIN VILLEMUR; RALPH ETIENNE-CUMMINGS; ANDREAS G. ANDREOU; THOMAS S. MURRAY; JAMAL L. MOLIN; PEDRO JULIAN; KAYODE SANNI; DANIEL R. MENDAT; PHILIPPE O. POULIQUEN; ISIDOROS DOXAS

Florianopolis

Congreso; LASCAS; 2016

Universidad Federal de Santa Catarina

In this paper we discuss a brain-inspired systemarchitecture for real-time big velocity BIGDATA processing thatoriginates in large format tiled imaging arrays used in wide areamotion imagery ubiquitous surveillance. High performance andhigh throughput is achieved through approximate computing andfixed point arithmetic in a variable precision (6 bits to 18 bits)architecture. The architecture implements a variety of processingalgorithms classes ranging from convolutional networks (ConvNets)to linear and non-linear morphological processing, probabilisticinference using exact and approximate Bayesian methodsand ConvNet based classification. The processing pipelineis implemented entirely using event based neuromorphic andstochastic computational primitives. The system is capable ofprocessing in real-time 160 x 120 raw pixel data running on areconfigurable computing platform (5 Xilinx Kintex-7 FPGAs).The reconfigurable computing implementation was developedto emulate the computational structures for a 3D System onChip (3D-SOC) that will be fabricated in the 55nm CMOStechnology and it has a dual goal: (i) algorithm exploration and(ii) architecture exploration.