Samsung Begins Shipping Samples of 12‑Layer HBM4E Memory to Major Customers

Samsung Electronics said in a May 29 press release in Korea that it has begun shipping samples of its 12-layer HBM4E memory to “major global customers,” marking an early step toward the next generation of memory used in AI accelerators.

HBM, or high-bandwidth memory, is stacked DRAM placed close to AI chips and data-center GPUs to deliver far more bandwidth and lower latency than conventional memory, making it critical for training and running large AI models and other memory-intensive workloads. As demand for AI hardware rises, memory suppliers are competing on bandwidth, capacity, power efficiency and manufacturability.

Samsung described the new product as the “industry’s first” 12-layer HBM4E, a characterization that was not independently verified in the announcement. The company said the sample is a 48-gigabyte, 12-layer stack aimed at next-generation AI accelerators. It also said it plans additional versions based on customer requirements, including 32GB 8-layer and 64GB 16-layer configurations.

According to Samsung, the HBM4E sample supports a stable pin speed of 14 gigabits per second, with scalability to 16Gbps, and can deliver up to 3.6 terabytes per second of memory bandwidth per stack. Samsung also said HBM4E improves energy efficiency by about 16% and thermal resistance by more than 14% compared with HBM4.

Samsung said the product uses its sixth-generation 10-nanometer-class DRAM process, known as 1c, along with a 4-nanometer logic base die manufactured by Samsung Foundry. Those process choices are intended to improve both performance and power characteristics as AI accelerators demand more memory capacity and throughput.

The company did not identify the customers receiving the samples. It also did not disclose a timetable for mass production, expected shipment volumes or pricing. Samsung said mass production would begin after customer evaluation and optimization, aligned with customer schedules.

The announcement follows Samsung’s Feb. 12 statement that it had begun mass production and commercial shipment of HBM4. That gives Samsung a public marker in the transition from HBM4 to HBM4E, though rivals are also moving on similar road maps. Public reporting has indicated that SK hynix is targeting HBM4E sample supply in the second half of 2026 and mass production in 2027, while Micron’s publicly discussed HBM4E plans generally point to 2027. Those timelines provide context for the competitive race, but they do not independently establish Samsung’s “industry-first” claim.

“Following the successful mass production of HBM4, Samsung has once again demonstrated its distinct technological edge with HBM4E,” Sang Joon Hwang, executive vice president and head of memory development at Samsung Electronics, said in the company’s release.