Unconventional Processing of Signals for Intelligent Data Exploitation (UPSIDE)

Added: May 17, 2013 9:03 am

SIDEARM demonstrates that Bayesian Probabilistic Inference, implemented on emerging analog devices, affords the computational expressivity approaching purpose machinery and is capable of handling the high demands of complex image processing pipelines, such as ARGUS-IS, with 10,000x less energy and 1,000x more processing capacity.

SIDEARM (Sensor Inference and Decision Engine via Adaptive Regulation of Memory arrays) provides a comprehensive solution through co-design of the inference module (IM). It uses the strengths and weaknesses of achievable Mixed Signal (MS) CMOS and memristor device technology combined with the requirements of state-of-the-art processing pipelines, such as the Wide Area Motion Imagery (WAMI) pipeline of BAE Systems‟ Autonomous Realtime Ground Ubiquitous Surveillance Imaging System (ARGUS-IS) In particular, SIDEARM includes eight innovations to produce a generally applicable compute model that meets or exceeds 1,000x speed increase and 10,000x power reduction targets.