Unconventional Processing of Signals for Intelligent Data Exploitation (UPSIDE)*
Abstract
*Formerly Unconventional Computation The Unconventional Processing of Signals for Intelligent Data Exploitation (UPSIDE) program will address the open problems facing real-time ISR systems and other power-constrained data-intensive applications. The objective of the UPSIDE program is to create a high-level, non-Boolean computational model and map it directly to the unique functional properties of new emerging devices to achieve significant increases in power efficiency and performance. The UPSIDE program will create a new generation of computing structures that will, in turn, enable revolutionary advances in ISR processing, particularly for DoD applications of embedded, real-time sensor data analysis. Because Boolean data representations are inherently power-inefficient for many datasets, particularly those produced by noisy analog real-time sensors, the UPSIDE program will establish an unconventional, non-Boolean, computing paradigm to enable new and needed capabilities in the area of sensor data analysis. UPSIDE intends to implement this new computing paradigm in the form of a specialized hardware component termed the inference module (IM). The inference module will be first developed through simulation, and then implemented using mixed-signal complementary metal-oxide semiconductor (CMOS), as well as using state of the art emerging (non-CMOS) devices. Throughout the program, the inference module will be benchmarked using a DoD relevant image processing pipeline, to verify gains in both computing throughput and power efficiency. The result will be a computing infrastructure and functional implementations that demonstrate three orders of magnitude improvement in processing speed and four orders of magnitude improvement in power efficiency. These gains will constitute a disruptive new level of embedded computational efficiency for future real-time sensor systems.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2014
- Source ID
- 483855fdf8feb77fb0fb550ecacfd961