Space Micro Inc. — Department of Defense SBIR Phase II: AFX234-DCSO2

Description

This SBIR addresses the United States Space Force (USSF) strategic capabilities of a) Space Superiority and b) Space Domain Awareness. The USSF is tasked with protecting and defending US interests in space. USSF now has an even greater surveillance task for space domain awareness and control in that region, but its current capabilities and architecture are limited by technologies and an architecture designed for the legacy GEO mission. Existing ground and near-earth computing systems are not only stressed by the increased range and volume, but also by limited Earth to satellite communication options in that space and navigation challenges due to the chaotic nature of cislunar orbits. To address the challenges posed to the current architecture, the USSF is exploring trusted autonomous space-based computing systems with artificial intelligence and machine learning capabilities operating in cislunar space. Space Micro Inc. proposes the development of a radiation-tolerant, high-performance, reconfigurable single board computer (SBC) based on the AMD/Xilinx Versal® AI Edge Series Adaptive Compute Acceleration Platform (ACAP) with artificial intelligence (AI)/machine learning (ML) engines. This new SBC will be available in the 3U SpaceVPX form-factor and Space Micro has christened this new SBC the Proton High-Performance Reconfigurable (PHiRe™ - pronounced “fire”) SBC. The goal of this proposal, and its subsequent technologies, is to begin the transition from sensor design to mission capabilities through the development of highly integrated, intelligent computational systems. This technology will allow for the implementation of onboard, high performance data processing which allows for more advanced machine learning, data analysis, and communication enhancement opportunities. The PhiRe™ SBC produced under this direct-to-Phase-II (D2P2) effort uses true AI/ML engines, not traditional central processing units (CPUs), programmable logic (PL), and/or graphical processing units (GPUs) to execute AI algorithms, which are not compute or power efficient. The PHiRe™ SBC attacks this problem with 34 AI/ML tiles generating 45 tera-operations per second performance dissipating only 25 watts of power.