This week is the 20^th anniversary of FMS, the Future of Memory and Storage conference, previously known as the Flash Memory Summit. Multiple world conflicts and tariffs have messed with supply chains even as AI data centers do much of the same. Still, technology continues to advance, delivering higher bandwidth and capacities on all fronts. So, for me, this week will be spent in Santa Clara covering FMS 2026.

I expect to see more ways to take advantage of DDR5 in systems like Compute eXpress Link (CXL) as well as multiplexed rank DIMMs (MRDIMMs). The latter allows servers to pack in more and faster memory modules that use multiplexed ping-ponging and buffer registers to hide DRAM die on a memory module.

CXL can be used for more than just cache-coherent memory. Xcena’s computational CXL board (Fig. 1) provides RISC-V cores as well as memory compression support. It also features SSD RAID support.

Even as new technologies become readily available, older platforms like eMMC, DDR2, and DDR3 remain in the embedded space. Interfaces like quad SPI and octal SPI with matching memory chips are changing the way embedded systems are implemented as microcontrollers and SoCs support these interfaces.

Solid-state drives (SSDs) have moved to the Enterprise and Data Center Standard Form Factor (EDSFF) for servers with rising interest in liquid cooling of SSDs for higher performance (Fig. 2). Of course, NVMe has become the standard for PCIe-based non-volatile storage. Embedded developers may still contend with SATA, but hard drives and SATA/SAS interfaces are no longer the mainstream.

The challenge for developers is long-term availability of memory and storage components for long-lived systems. Medical, avionics, and automotive applications double the chore since any new alternatives need to be certified.