Demand for computer memory grows with each data center. The ever-increasing needs are driving memory prices higher and impacting diverse industries, including programs in space.

To address the changing memory landscape, an article from Mercury Systems, “Spacecraft Memory in the Age of AI Hyperscaling: Turning Obsolescence into an Engineering Discipline,” written by Vincent Pribble, Space Product Manager, suggests that device obsolescence be planned for rather than shunned. It can be the heart of an innovative approach to use aging memory chips in space-based electronic systems.

The whitepaper addresses the changing performance and applications of dynamic random-access memory (DRAM) and NOT AND (NAND) memory devices, including in low-Earth-orbit (LEO) satellite systems (see image above).

Using payloads reliant on commercial-off-the-shelf (COTS) electronics, these systems relied on DRAM and NAND devices in which legacy memory die have been available for decades. But with the rapid expansion of artificial intelligence (AI), and the need for higher-bandwidth memory chips in AI accelerators, data-center-class processors are being used in LEO satellite systems.

On-orbit AI inference and edge processing drive spacecraft memory demands even higher. The legacy memory market, which once supported memory devices for decades, is now changing rapidly and memory design requirements are resulting in faster product turnarounds. 

The whitepaper explains that even COTS memory solutions for space are more difficult to procure with these growing demands. However, planning for memory device obsolescence as part of a continuous engineering discipline by regularly monitoring memory product lifecycle alerts as part of the development of space-based systems can circumvent the loss of legacy memory products.