JPL unveils radiation-hardened AI chip

NASA’s Jet Propulsion Laboratory has moved into testing on a next-generation radiation-hardened space processor that the agency says is designed to give future spacecraft far more onboard computing power. NASA and JPL said in releases published last week that the processor is part of the High-Performance Spaceflight Computing project, a public-private effort to build a new space-qualified system-on-chip. Early testing at JPL has shown the processor working as designed and operating at 500 times the performance of radiation-hardened chips currently in use, according to the agency. The project is aimed at missions that need to process more data onboard and make more decisions without waiting for instructions from Earth. ### What exactly did JPL announce? JPL said the current milestone is a test campaign, not a flight-ready deployment. The lab said engineers began testing in February in Southern California and are putting the processor through radiation, thermal and shock tests meant to simulate the space environment. Jim Butler, High Performance Space Computing project manager at JPL, said the team is evaluating both survivability and performance during the campaign. (jpl.nasa.gov) NASA said the processor is the centerpiece of its High-Performance Spaceflight Computing project. The agency described it as a next-generation system-on-chip that combines computing and networking in one device, with a target of delivering more than 100 times the computing capability of current space processors while lowering system cost and power use. (jpl.nasa.gov) ### Why does radiation hardening matter for a space AI chip? NASA said radiation-hardened processors are built to keep working in environments where electronics can be degraded by electromagnetic radiation and extreme temperature swings. JPL’s space-radiation materials describe the lab’s role in modeling and studying those environments for flight projects. The constraint for NASA has been that rad-hard processors are resilient but far behind commercial chips in raw performance. (nasa.gov) A 2024 NASA technical briefing said current rad-hard processors are so performance-limited that deploying state-of-the-art AI frameworks on them is generally infeasible. That gap is why NASA and JPL have been pursuing both higher-performance flight computing and separate low-power AI acceleration work. (jpl.nasa.gov) ### Is this the same thing as an AI accelerator? NASA records show two related but distinct tracks. The High-Performance Spaceflight Computing project is a general-purpose next-generation processor for flight computing, while a separate NASA TechPort project describes a radiation-hardened AI inference accelerator that can expand existing systems’ AI capabilities by orders of magnitude at less than 0.4 watts. (ntrs.nasa.gov) The social-media framing around an “AI chip” appears to draw on the broader autonomy use case rather than a single newly published technical paper. JPL’s public AI materials and conference papers describe future deep-space missions using onboard planning, scheduling, target identification and data summarization so spacecraft can respond to changing conditions and reduce dependence on continuous ground contact. That is an inference from NASA’s published project descriptions and papers, not a separate new JPL statement. (techport.nasa.gov) ### What would spacecraft do differently with more onboard computing? JPL researchers wrote in a 2023 SpaceOps paper that future deep-space robotic explorers will use advanced onboard autonomy to observe fast-changing phenomena and adapt to dynamic environments. Another JPL paper said future missions will need stronger onboard autonomy both to respond quickly to science events and to reduce the amount of instrument data that must be sent back to Earth. (ai.jpl.nasa.gov) NASA’s framing of the processor is consistent with that goal. More computing onboard would let spacecraft run planning, science selection and data-processing tasks locally instead of routing every decision through Earth-based operators, especially where communication delays are long. That description is based on NASA and JPL research documents about autonomy, rather than a new operational claim about a specific mission. (ai.jpl.nasa.gov) ### Who is building it, and what comes next? NASA awarded a $50 million contract in 2022 to Microchip Technology to develop the next-generation space-qualified compute processor platform under the High-Performance Spaceflight Computing effort, with Microchip also contributing research and development spending, JPL said at the time. The agency said the project uses a public-private partnership model. (ai.jpl.nasa.gov) Testing that began in February will continue for several months, according to NASA and JPL. The next public milestones are likely to come through additional NASA or JPL updates on the High-Performance Spaceflight Computing project, which both organizations are now featuring in recent technology coverage. (jpl.nasa.gov 1) (jpl.nasa.gov 2)