White Paper

SPACE TECHNOLOGY SPECIAL REPORT 12 SEPTEMBER 2026 dry storage and moisture control, periodic solderability verification and re-bake, and attention to long-term data retention behavior. Die and wafer banking through authorized aftermarket channels can shift storage and continuing manufacturing to specialists that hold original die under franchise agreements with the manufacturers, a channel that has become more relevant now that legacy DRAM and SLC NAND products are moving through authorized aftermarket distribution. In defense programs, customer-funded lifetime-buy structures shift forecast risk to the customer, which is often the only way the economics close on low- volume, long-sustainment programs. One warning belongs in every buy plan: scarcity breeds counterfeits. As legacy memory prices spike, gray-market supply becomes more dangerous, not more attractive. Authorized channels, counterfeit-avoidance systems compliant with DFARS requirements, authentication testing per SAE AS6171, and GIDEP reporting are the minimum standard for any part carrying a radiation pedigree. Discipline 4: Mission-Class- Appropriate COTS The fourth discipline widens the trade space by matching assurance to mission class rather than defaulting to the highest qualification level. For risk- tolerant missions and proliferated low Earth orbit constellations, commercial and automotive-grade memory with lot-based radiation acceptance testing can be viable. Automotive parts qualified to AEC-Q100 offer a useful middle path: wider temperature range, tighter lot control, and longer retention specifications than consumer-grade parts, without any radiation guarantee. The engineering that makes commercial memory flyable happens at the system level. Error detection and correction, background scrubbing against protected golden copies, watchdog supervision, redundancy with voting, and latchup protection circuits substitute system-level mitigation for part-level pedigree. The governing question is not whether a part is radiation hardened but whether the mission, environment, application, and lifetime can tolerate the characterized failure rates after mitigation. The approach is now well proven in low Earth orbit, where data- center-class processors and COTS-heavy payloads have operated successfully and where constellation-level redundancy absorbs part-level risk that would be unacceptable on a flagship mission. Geostationary and deep-space missions, with long lives and no replenishment, still justify full QML-V pedigree and part- level radiation hardness assurance. One caveat deserves plain statement: the COTS escape hatch is narrowing too. The same demand distortion driving DDR4 and SLC NAND out of production makes commercial legacy memory harder to procure, with tight legacy SLC supply projected through at least 2026 and no significant new capacity expected before 2027. COTS strategies now require the same supply chain discipline as qualified-part strategies. The Demand Side is Coming to Orbit The final twist is that the workloads causing the shortage are migrating to space. Orbital data center ventures have demonstrated data-center-class GPUs operating in low Earth orbit, major silicon vendors have announced space computing modules and partner ecosystems, and European feasibility studies have validated large-scale orbital compute concepts. Whatever fraction of those ambitions materializes on schedule, the direction is clear: on-orbit AI inference and edge processing mean spacecraft memory demand is rising, in both capacity and bandwidth, at exactly the moment the qualified supply base is narrowing. Market forecasts for radiation-hardened electronics and space computing diverge widely in magnitude, but not in direction. Obsolescence as a Discipline The common thread across all four strategies is posture. Programs that treat memory obsolescence as a periodic crisis will keep paying crisis prices: emergency buys at spot pricing, reactive qualification at the back of a beam-time queue, and redesigns measured in years. Programs that treat it as a continuous engineering discipline will experience the next end-of-life notice as a scheduled event rather than an emergency. That discipline means monitoring lifecycle alerts as an operational function, architecting for die change from the first design review, qualifying successors ahead of need, sizing buys deliberately with their true carrying costs, and right- sizing assurance to mission class. The AI buildout is not going to hand back the wafer starts. The memory market that the space industry grew up with, in which legacy die stayed available for decades because nobody else needed the capacity, is gone. The programs that internalize that reality first will be the ones that fly on schedule. This article was written by Vincent Pribble, Space Product Manager at Mercury Systems. For more information, please visit mrcy.com. Mercury's SCFE6933 is a rad-tolerant AMD Versal™ AI Core based board that rapidly collects and processes on orbit sensor data with high density, real-time AI/DSP performance for beamforming, radar, comms, and radio applications. Spacecraft Memory

• Cover