Rapid upside

Added: Jun 07, 2013 9:25 am The Unconventional Processing of Signals for Intelligent Data Exploitation (UPSIDE) program will create a new generation of computing structures that will enable revolutionary advances in Intelligence, Surveillance and Reconnaissance (ISR) processing. The objective of the proposed research effort is to demonstrate a clear path toward the implementation of non-Boolean computational models in the form of an Inference Module (IM), designed around the unique capabilities of emerging non-CMOS devices. In the system developed here, inference will be the fundamental computational building block in the same way that the Boolean logic operation is the fundamental computational block in traditional digital systems. A mixed signal CMOS implementation of the basic Inference Modules shall be developed as an intermediate step in this program. The final result of this research will be a new UPSIDE computing infrastructure, incorporating emerging non-CMOS devices, that will demonstrate three orders of magnitude improvement in processing speed and four orders of magnitude improvement in power consumption for a variety of state-of-the-art object recognition and tracking applications.

HRL Laboratories, LLC (HRL), proposed the Revolutionary Analog-based Probabilistic Inference Devices for Unconventional Processing of Signals for Intelligent Data Exploitation (RAPID-UPSIDE) to develop a revolutionary non-Boolean, non-CMOS nanodevice technology for power-efficient, high-speed processing of sensor data and image analysis applications that identifies and tracks a large number of targets in real-time. The RAPID-UPSIDE team has laid a strong foundation to realize this technology. HRL proposed a novel concept to perform fast distance computation between any two highdimensional vectors based on spontaneous synchronization among weakly coupled oscillators. HRL's IM module design will leverage this concept by combining banks of weakly coupled oscillators to provide an unprecedented speed advantage over Boolean approaches for probabilistic inference-making because it can perform both distance computation and search for closest matching vectors in O(1) time. The other innovative aspect of this IM design is that it also offers the most energy efficient means to make probabilistic inferences since synchronization of oscillators corresponds to a stable minimum energy state. HRL's IM based IPP (or Gold IPP) will be designed using this novel non-Boolean computational paradigm to rapidly detect and track multiple objects in a power-efficient manner.