NASA JPL radiation-hardened AI processor

NASA’s Jet Propulsion Laboratory is testing a next-generation radiation-hardened processor designed to give spacecraft far more onboard computing power in deep space, according to NASA releases published on May 12. The chip is part of NASA’s High Performance Spaceflight Computing project, or HPSC, which the agency says is intended to support autonomy, AI, image processing and other workloads on future missions. (jpl.nasa.gov) NASA said early indications show the processor operating at 500 times the performance of the radiation-hardened chips currently in use. (nasa.gov) The May 20 social-media framing around a “radiation-hardened AI processor” appears to draw on two related NASA efforts rather than a single new May 20 unveiling. (nasa.gov) NASA’s published JPL and NASA.gov accounts describe the HPSC flight processor and its ongoing testing, while NASA TechPort separately describes an active radiation-hardened AI inference accelerator project aimed at low-power neural-network workloads. ### Was there actually a new NASA JPL unveiling on May 20? NASA’s public releases located for this story were dated May 12, not May 20. The JPL and NASA.gov articles said the HPSC processor was undergoing testing at JPL in Southern California and described the work as part of a broader effort to modernize space-qualified computing. (jpl.nasa.gov) The May 20 date in circulation appears to come from social posts summarizing NASA technology work. Based on NASA’s own pages, the verified milestone is that HPSC was already in testing by May 12, with testing having begun in February. ### What is NASA building? (jpl.nasa.gov) NASA describes HPSC as a radiation-hardened, high-performance system-on-a-chip meant to address spacecraft needs through 2040 and beyond. The agency says it is designed to deliver more than 100 times the computing capability of current space processors while surviving radiation and other hazards that can disrupt electronics in space. (jpl.nasa.gov) Eugene Schwanbeck, program element manager in NASA’s Game Changing Development program, said the new multicore system is “fault-tolerant, flexible, and extremely high-performing.” Jim Butler, HPSC project manager at JPL, said engineers are carrying out radiation, thermal and shock tests while running a functional test campaign. (nasa.gov) ### Where does the AI claim come in? NASA’s HPSC page says future missions will need onboard computing for “advanced autonomy, AI (artificial intelligence) and machine learning, image and signal processing, data flow management, and object detection and classification.” The agency says communication delays beyond Earth orbit make real-time onboard decision-making more important. (nasa.gov) A separate NASA TechPort project describes a “radiation-hardened Artificial Intelligence (AI) inference accelerator” that can expand existing systems’ AI capabilities “by orders of magnitudes” while consuming less than 0.4 watts of power. TechPort says that project has the potential to reach 50 TOPS/W efficiency and is targeted at onboard data analysis, sensor enhancement and system autonomy. (jpl.nasa.gov) ### Does NASA verify the “10 trillion operations per second” figure? NASA TechPort’s AI accelerator page supports a figure of 50 TOPS/W at 0.4 watts, which implies about 20 trillion operations per second in that project description. The same page does not use the exact social-media phrasing about “full neural networks in real time,” but it does identify the technology as an AI inference accelerator for space use. (nasa.gov) NASA’s HPSC releases, by contrast, frame performance against existing space processors rather than in TOPS. Those releases say the processor is designed for up to 100 times current computational capacity, and that early testing indicates roughly 500 times the performance of radiation-hardened chips now in use. (techport.nasa.gov) ### Why does radiation hardening matter for spacecraft? NASA says radiation in space can cause long-term damage to components and errors that disrupt computing, including faults that can push a spacecraft into safe mode. The agency says those constraints have left many missions using older but proven chips, even as mission demands for autonomy and onboard analysis have increased. (techport.nasa.gov) The next public milestone is continued testing at JPL over the coming months. NASA says HPSC testing began in February and will continue for several months, while the separate TechPort AI accelerator project lists an end date of Oct. 31, 2026. (nasa.gov) (jpl.nasa.gov)