Weekly recap cover, week ending July 31, 2026A 3D NAND word line staircase with channel holes through it, the lead topic of the week ending July 31, 2026; below it, one drawing for each topic the week covered: NAND flash (1), DRAM and HBM (2), SSDs and storage systems (2), Controllers and interfaces (1), Standards (2), Market and earnings (4), Research papers (3).THE WEEK IN MEMORY AND STORAGEWEEK ENDINGJuly 312026LEAD · NAND FLASH1221243

Week endingFriday, July 31, 2026Covers July 25 to July 31, 2026 · 15 items

Record Memory Results from Samsung, SK hynix and Kioxia, and NVMe Ships a Spec Wave

Second quarter earnings dominated the week. Samsung and SK hynix both reported record memory results on AI server demand, Seagate closed its fiscal year, and Kioxia reported a record first quarter led by data center SSDs. Samsung also signed a memorandum with Broadcom, and SK Group and NVIDIA announced a partnership that includes HBM4 supply.

On the technology side, Kioxia began sampling a ninth generation BiCS FLASH part, introduced its first PCIe 6.0 enterprise SSD on its tenth generation flash, and announced UFS 5.0 devices. NVM Express ratified a large set of specification revisions, headed by Base Specification 2.4.

Top stories

The rest of the week

DRAM and HBM

SSDs and storage systems

Controllers and interfaces

Standards

Market and earnings

Research papers

  1. A paper tests whether device models explain RowHammer and RowPress bitflips (opens in a new tab)

    The authors compare published device level mechanisms with experimental DRAM characterization, then use TCAD simulations to update the explanation of why read disturbance bitflips occur.

    arXiv ·

  2. A paper proposes a photonic CXL memory appliance for LLM KV cache (opens in a new tab)

    Marvell authors describe a switch-free design that uses a passive fiber shuffle to share a large CXL memory pool across many hosts, aiming to avoid cache eviction cliffs in multi-turn inference.

    arXiv ·

  3. A MICRO 2026 paper pairs GPUs with processing-near-memory over CXL for mixed length LLM serving (opens in a new tab)

    NELSSA routes short requests to GPUs and long ones to near-memory devices on a CXL system, and migrates requests as context grows. The authors report better decode throughput and tail latency than GPU-only serving.

    arXiv ·