Kioxia All Set to Raise the NAND Game in AI SSDs

2026-07-06 · Source: Big Data & AI News - EE Times · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Advanced, short

Summary

Kioxia has launched a 332-layer NAND flash with a 4.8 Gb/s interface speed, utilizing its 10th-generation BiCS FLASH 3D memory technology. This development positions the Japanese chipmaker ahead of competitors like Micron, Samsung, and SK Hynix in the NAND market, specifically targeting enterprise and data center SSDs crucial for AI inference. Kioxia's technological lead is attributed to its focus on NAND over HBM and innovations such as CMOS directly bonded to array (CBA) and on-pitch select gate drain (OPS) technology, which enhance performance and reduce power consumption by 10% at input and 34% at output. The 10th-generation chips offer a 59% improvement in data storage density and 30% better power efficiency compared to its 8th-generation. This strategic focus has driven Kioxia's market capitalization to over \$250 billion, making it Japan's largest company by market value, with mass production slated for next year at its Kitakami fab.

Key takeaway

For AI Architects designing data center infrastructure, Kioxia's 10th-generation 332-layer BiCS FLASH NAND offers a compelling solution for high-capacity, high-speed AI inference SSDs. Your designs should consider this technology's 59% higher density and 30% better power efficiency to optimize performance and reduce operational costs. Evaluate sample shipments for functional testing to validate its suitability for your specific AI workloads, but remain aware of potential competition from other memory giants.

Key insights

Kioxia's 332-layer BiCS FLASH 3D NAND, with CBA and OPS technologies, establishes a significant lead in high-performance, power-efficient AI data center SSDs.

Principles

• Focus on a specific market segment can yield competitive advantage.
• Wafer bonding technology (CBA) enhances NAND performance and power.
• Vertical 3D flash memory (BiCS FLASH) surpasses 2D capacity limits.

Method

Kioxia's CMOS directly bonded to array (CBA) technology manufactures memory cells and control circuits separately on wafers, then bonds them to boost performance and reduce power consumption.

In practice

• Evaluate 10th-gen BiCS FLASH for AI inference workloads.
• Consider CBA-enabled NAND for high-density, low-power storage.
• Prioritize NAND suppliers focused on data center SSDs.

Topics

• NAND Flash
• 3D BiCS FLASH
• AI SSDs
• Data Center Storage
• Wafer Bonding
• CMOS Directly Bonded to Array

Best for: MLOps Engineer, Investor, CTO, AI Hardware Engineer, AI Architect, Director of AI/ML

Related on AIssential

• Massive AI Storage Demand Creates a New Memory Wall — AI's massive memory demands necessitate new architectures like high-bandwidth flash, optimizing capacity and sequential bandwidth over traditional DRAM.
• Imec reimagines CCD technology as high-density AI memory — Imec's 3D CCD memory architecture merges DRAM and NAND benefits to address AI's memory wall.
• OpenAI unveils its first custom chip, built by Broadcom — OpenAI's custom chip, Jalapeño, signifies a strategic vertical integration to optimize AI inference costs and scale.
• Micron Readies Client Storage for AI — Local AI on client devices is driving demand for high-performance, power-efficient storage solutions like PCIe Gen5 QLC SSDs.
• SoftBank’s Saimemory joins forces with Intel on next-gen memory — Intel and SoftBank's Saimemory are commercializing Z-Angle Memory to address AI and HPC memory bottlenecks.
• Further Accelerating Kimi-K2.5 on AMD Instinct™ MI325X: W4A8 & W8A8 Quantization with AMD Quark — Quantization to W4A8 or W8A8 significantly accelerates Kimi-K2.5 inference on AMD MI325X GPUs with minimal accuracy impact.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.