Biological Information Technology Systems - BITS: Architectures for Neuromimetic Information Systems

University Of Southern California, Los Angeles CA

Investigators

Abstract

EIA-130898-John Granacki-University of Southern California- Architectures for Neuromimetic Information System Biologically inspired computing modules performing spatio-temporal pattern recognition will be a key aspect of future computing systems. Current prototypes of biomimetic circuit models have demonstrated real-time performance at very low power levels and high physical density. Further, these simple structures are capable of recognizing signal patterns masked by significant noise levels. These properties are difficult to mimic in digital signal processing algorithms, which require substantial modification when adapting to new applications. We propose to implement second-order dynamic synapse neural models with digital equivalents that use features extracted from software simulations. Arrays of these digital equivalents can be readily implemented in dense re-useable commercial field-programmable gate arrays (FPGAs), enabling very high-speed operation. Additional features, such as writeable registers for user tuning of dynamics, will accelerate the evaluation of new kernels as well as enable investigation of high-speed "super-time" learning.