SK hynix Begins Mass Production of 192GB SOCAMM2 on 1c DRAM for NVIDIA Vera Rubin

By NineScrolls Team · 2026-04-22 · 4 min read · Industry

The Headline

SK hynix announced on April 20, 2026 that it has begun mass production of the industry's first 192GB SOCAMM2 (Small Outline Compression Attached Memory Module 2), built on its 1c-nanometer (sixth-generation 10nm-class) LPDDR5X DRAM process. The product is purpose-built for NVIDIA's next-generation Vera Rubin AI server platform.

SK hynix says the module delivers more than twice the bandwidth and over 75% better power efficiency than conventional RDIMM solutions, positioning SOCAMM2 as the bridge between high-bandwidth HBM stacks and capacity-focused DDR5 RDIMMs in the AI server memory hierarchy.

Inside the 192GB SOCAMM2 Module

SOCAMM2 adapts low-power memory — historically used in mobile devices — for server environments. The 192GB capacity per module is achieved by stacking and connecting 1c LPDDR5X dies in a compact form factor designed to be installed and serviced as a discrete unit in next-generation AI compute nodes.

According to SK hynix, the design targets the memory bottleneck that constrains training and inference of large language models with hundreds of billions of parameters. Justin Kim, SK hynix's President and Head of AI Infra, said the module is intended to "fundamentally resolve the memory bottlenecks encountered during the training and inference of large language models."

The 1c DRAM Process Node

The 1c node is SK hynix's sixth-generation 10nm-class DRAM process. Pushing DRAM cell pitch into the 1c regime requires the most advanced patterning, etch, and deposition technologies the industry has ever deployed at high volume — including extreme ultraviolet (EUV) lithography for critical layers and atomic layer deposition (ALD) for high-k dielectrics in storage capacitors.

Mass production of 192GB SOCAMM2 confirms that SK hynix is now ramping 1c DRAM in commercial volume — a milestone that pulls every layer of the deposition and etch toolchain forward with it.

The Three-Way Race: SK hynix, Samsung, Micron

SK hynix is first to mass production at 192GB. Micron shipped 256GB SOCAMM2 samples in early March, roughly 33% higher capacity per module. Samsung recently resolved manufacturing warpage issues by deploying a proprietary low-temperature solder process that drops soldering temperatures from above 260°C to below 150°C, a change designed to protect dense LPDDR5X stacks during module assembly.

All three suppliers are competing for design-in slots on NVIDIA's Vera Rubin platform and on hyperscale cloud customer roadmaps. The competitive pressure to scale module capacity, bandwidth, and yield is translating directly into incremental orders for advanced front-end and back-end equipment.

The Cleanroom Build-Out Behind It

The 1c DRAM ramp is happening inside SK hynix's accelerated cleanroom expansion. The company has pulled in Phase 4 completion of its M15X fab in Cheongju from April to March, after previously fast-tracking Phase 3 startup to February. M15X spans roughly 60,000 square meters of cleanroom — comparable to the combined footprint of the existing M11 and M12 plants — and is the primary site supporting the HBM4 and 1c DRAM ramp.

Samsung is running a parallel sprint at its Pyeongtaek P5 line, where cleanroom completion has moved from early 2028 to Q3 2026, with operational start pulled in to late 2027. Both projects translate into substantial near-term orders for etch, deposition, lithography, and metrology tools.

What This Means for Plasma Processing and Thin Film Deposition

A 1c-class LPDDR5X die packed into a 192GB module is a thin-film and plasma-processing problem at every step. Each shrink to a smaller DRAM node multiplies the number of ALD, PECVD, and etch steps required per wafer. For 1c specifically, capacitor high-k dielectric deposition (HfO₂, ZrO₂ and related stacks) demands ALD with atomic-level uniformity across 300mm wafers, and storage node electrode layers rely on PVD and CVD of titanium nitride and tungsten with tightening defect budgets.

Plasma etch tools take on more critical layers as well. Capacitor mold etch, contact etch, and word-line/bit-line patterning at 1c pitch require selective, low-damage plasma chemistries, fueling demand for cryogenic and pulsed-plasma etch reactors and the gas delivery systems that feed them. The dense LPDDR5X stacks inside SOCAMM2 also push back-end-of-line copper and barrier layers further toward ALD- and PVD-deposited films, with TaN/Ta liners and Cu seed layers becoming more critical to yield as metal pitches shrink.

For the equipment supply chain — plasma sources, RF generators, electrostatic chucks, vacuum components, gas panels, ALD precursor delivery and process monitoring — SK hynix's mass-production milestone is a leading indicator. M15X's accelerated phases and Samsung's Pyeongtaek P5 fast-track translate into a multi-quarter pull on every layer of the deposition and etch toolchain, from chamber consumables to entire process modules. SOCAMM2 is a memory module headline, but underneath it is another step-up in the global plasma-processing and thin-film-deposition load.