Prototyping a Resilient, Versatile, and Future-Proof Spaceflight Coprocessor Platform

State-of-the-art NASA coprocessing for Digital Signal Processing (DSP) and Artificial Intelligence (AI) applications lacks the versatility, performance, and energy efficiency needed for future space missions which also require radiation resilience not found in commercial devices. This research and development accelerates high-bandwidth, real-time sensor DSP and AI data processing for autonomous perception, planning, and control applications. Our proposed work integrates radiation fault tolerance, health monitoring, and power reduction techniques into open-source General Purpose GPU (GPGPU) soft cores on latest generation radiation-tolerant, reprogrammable Field Programmable Gate Array (FPGA) devices. Additionally, GPGPU software programming toolchains are leveraged to enable flexible, parallel coprocessing needed for future spaceflight missions. Initial estimations show we outperform the baseline FPGA coprocessing technology found on the Mars Perseverance Rover by 42x for similar SWaP. A low-power, radiation-tolerant Application Specific Integrated Circuit (ASIC) translation boosts this gain to over 100x while retaining the GPGPU open compute flexibility and inflight reprogrammability. Phase I funded work includes a proof-of-concept development for the FPGA-based GPGPU design. FPGA simulations validate our design choices by showing increase in versatility, performance, and energy efficiency on a radiation-tolerant, space-ready platform. NASA, DoD, and private space companies benefit from our future-proofed application coprocessing capabilities and resilience to natural and emerging, adversarial radiation space threats.