Applied Materials Partners with Kioxia to Advance AI Memory and 3D Stacking R&D

nashnova research
今天发布阅读约 12 分钟

Applied Materials announced that Kioxia will join its $5 billion EPIC R&D center to co-develop next-generation memory for AI — this means the pipeline from memory materials research to mass production is tightening.

01

What is the EPIC center, and why is Kioxia joining?

EPIC — the Equipment and Process Innovation and Commercialization center — is a $5 billion semiconductor R&D facility Applied Materials is building in Silicon Valley, set to open in 2026.
In plain terms = it is a bridge that moves lab breakthroughs into factory production lines faster, cutting the time from invention to shipment.
Kioxia, one of the world's leading flash-memory makers, joins as an "innovation partner" alongside chipmakers, chip designers, equipment firms, and universities already on site.
02

What four technology areas will the partnership tackle?

Higher-density memory cells — developing advanced storage structures with greater performance and density.
Multi-chip stacking — piling more memory chips in a limited space while boosting performance.
Advanced packaging and interconnects — improving how stacked chips connect and communicate with each other.
Materials engineering for manufacturing — solving the gap between "works in the lab" and "works in the factory." This means → the partnership covers the full chain from a single chip to a finished package.
03

What did management say — and is there a timeline?

Applied Materials CEO Gary Dickerson: "AI infrastructure requires memory solutions that deliver dramatically higher density, greater bandwidth, and better energy efficiency."
Kioxia CEO Hiroo Ota said memory progress depends on pushing the boundaries of materials and process innovation.
Neither company disclosed financial terms or a specific timeline for commercial products. This means → the partnership is still at the early-framework stage, with production-ready products some distance away.
04

How hot is memory-equipment demand right now?

Applied Materials' fiscal Q3 2026 data: DRAM revenue rose to 26% of its semiconductor-systems business, up from 22% a year earlier.
This reflects that HBM — high-bandwidth memory, the high-speed memory built for AI chips — and AI-server-driven storage investment have become a key growth driver for equipment demand.
The company is optimistic about DRAM, advanced foundry, logic, and advanced-packaging demand in the second half of 2026, and expects strong growth to continue into 2027.
05

What do the six new equipment systems announced this quarter solve?

Centura Prime Epi epitaxy system: selectively grows specialized materials at critical transistor junctions to boost DRAM and HBM speed and energy efficiency.
Opta Quad CMP — chemical-mechanical planarization, essentially polishing a chip surface flat — improves within-wafer uniformity for hybrid bonding (joining two chips directly, face-to-face).
Nokota Vmax 2 electrochemical deposition system: covers the full 3D-stacking flow from TSV (through-silicon via — a vertical channel linking chip layers) filling to fine-pitch micro-bumps. Producer Avila 2 PECVD system strengthens ultra-thin DRAM dies mechanically, enabling 12-layer, 16-layer, and higher HBM stacks.
06

Who else has joined EPIC, and what does the bigger picture look like?

Three new partnerships this quarter: Broadcom on next-gen AI chip packaging; UC Berkeley on materials and process research for AI computing; Screen Semiconductor Solutions on advanced chip-process development combining wafer cleaning and materials engineering.
In plain terms = EPIC is becoming a semiconductor industry consortium — from design (Broadcom) to manufacturing (Kioxia) to equipment (Screen) to fundamental research (Berkeley), every link in the chain now has a seat.
The key proof point after Kioxia's entry: whether materials-engineering breakthroughs can translate into mass-producible HBM and 3D NAND products — that is the ultimate test of this partnership's commercial value.

市场有风险,内容仅供研究参考,不构成投资建议。