High Performance Computing with the NorthPole Inferencing Chip

The Air Force Research Laboratory (AFRL) continues to mature advanced Multi-Function Processor capabilities, techniques and methods using IBM’s NorthPole chip. NorthPole is a newly developed artificial intelligence inferencing engine that supports diverse sensors and deep networks. The system is optimized for 8-, 4-, and 2-bit low-precision. During this years’ Government sponsored experiment, called Trident Spectre 2023, our team extended upon the state-of-the-art by immediately using one of the very first NorthPole chips to rapidly process newly available Gorgon Stare 2 data. Gorgon Stare 2 data is challenging because it is a large data set consisting of targets that can be identified by machine learning-based automatic target recognition techniques and methods.

Our research consists of designing high-performance embedded computing (HPEC) capabilities that are integral to enabling autonomy and upstream information exploitation – at the edge, where the data is collected. The extensibility of powerful systems and the associated technologies deliver capabilities that continue to provide and inform processing and exploitation technologies for various domains, i.e., ground, air, at high altitudes, and in space.

Specifically, we successfully processed the data and assessed our performance – these processing metrics will be discussed.

Discussion will also cover advanced technologies that promise to impact future computing solutions. For example, future NorthPole architectures can be fabricated, using a 3-nanometer or even 2-nanomater process and achieve even more efficiency and decreased size, weight, and power (SWaP) characteristics. Other innovative science includes ultra-low power computing, e.g., Intel’s neuromorphic chip, called Loihi2 and Agile Condor®. AFRL and SRC’s Agile Condor, US Patent (10,915,152), HPEC system is an ideal candidate for an edge computing solution on various platforms – including attritables. These systems afford engineers flexible compute options for future missions (ground, air, high altitude and space).

PRESENTER

Wise, David
david.wise.13@us.af.mil
315-330-4264

AFRL/RI

CO-AUTHOR

Barnell, Mark
mark.barnell.1@us.af.mil

CATEGORY

Artificial Intelligence / Machine Learning usage for HPC Applications

SYSTEMS USED

AFRL/RI HPC-ARC

SECRET

No