Particle.news

Samsung Ships World’s First 12‑Layer HBM4E Samples

The sample rollout signals Samsung’s push to win early AI‑memory contracts by moving to mass production as customers complete qualification.

Overview

  • Samsung said Friday, May 29, 2026, that it has begun shipping global samples of its 12‑layer HBM4E to customers as the next step before qualification and mass production.
  • The 12‑layer stack ships in a 48GB configuration with per‑pin speeds of 14–16 Gbps and about 3.6 terabytes per second of bandwidth per stack, aiming to ease memory bottlenecks in large language model training and inference.
  • Samsung built the HBM4E on its sixth‑generation 1c 10‑nanometer‑class DRAM process paired with a 4‑nanometer logic base die from Samsung Foundry to raise yields, reduce power use, and improve thermal resistance.
  • The announcement lifted Samsung shares roughly 6% as customers such as AMD, Nvidia and Google prepare to test the samples and Samsung ties mass production timing to their qualification schedules.
  • Rivals SK Hynix and Micron are reported to be preparing their own HBM4E sample programs, making the market competitive and putting pressure on Samsung to scale Pyeongtaek capacity and convert early tests into large orders.