GMIF2026 展覽回顧：人工智能記憶體與儲存鏈的創新展示

The GMIF2026 exhibition highlighted a push towards specialized storage solutions designed to meet the increasing demands of AI workloads for speed, capacity, and durability. Sandisk, for instance, presented its SN861 NVMe drive, which offers read throughput up to 13700 MB/s, and its BiCS8 enterprise-grade QLC NAND, providing up to 8TB per package for AI data storage and data lakes. These products suggest an industry focus on accelerating data access and increasing storage density for AI-specific data.

Data integrity and efficient processing are also key areas of development. Silicon Motion's SM8466 controller, built on TSMC's 4nm process, is reported to achieve 28GB/s sequential read speeds and 7M random read IOPS for enterprise platforms. This controller integrates an ARM Cortex-R82 core and CoCo 16K LDPC ECC, features designed to enhance processing capability and data endurance in demanding AI environments. Its support for post-quantum cryptography also indicates a focus on future-proofing data security.

New approaches to high-capacity storage are emerging, such as Numbers Law's Parallel Magnetic Drive (PMD). The company claims its PMD delivers significantly higher bandwidth and random IOPS compared to conventional HDDs, aiming to provide a high price-performance option for AI data centers and big data clusters in China. The reported domestic supply chain and national cryptographic compliance for PMD suggest an effort to localize critical components and meet specific regulatory requirements within the region.