On May 29, Samsung Electronics announced it has begun shipping engineering samples of the industry's first 12-layer stacked HBM4E to major global customers, further consolidating its leading position in the next-generation high-bandwidth memory (HBM) market. The move addresses escalating AI computing demands, with Samsung stating the 12-layer HBM4E delivers dual improvements in energy efficiency and thermal performance to meet next-generation AI high-load computing requirements. Samsung's customer base includes AI chip manufacturers such as NVIDIA and AMD, as well as cloud service providers like Google and Microsoft—all core enterprises supporting AI computing needs.
Product Specifications and Performance Metrics
Samsung HBM4E, the company's fourth-generation enhanced high-bandwidth memory, achieves a stable pin rate of 14 gigabits per second (Gbps) with scalability to 16Gbps to address surging data processing demands. Compared to Samsung HBM4, performance increases by over 20%, with maximum single-stack memory bandwidth reaching 3.6 terabytes per second (TB/s) to maximize computing performance for large language models (LLM) and next-generation AI systems.
In storage capacity, Samsung's 12-layer HBM4E offers 48 gigabytes (GB) per stack, representing over 30% capacity increase versus the previous generation. Samsung Electronics stated it plans to expand the product line based on customer needs, adding 32GB (8-layer stack) and 64GB (16-layer stack) configuration versions.
Manufacturing Technology and Process Integration
Samsung Electronics stated HBM4E's core advantage lies in fully integrating Samsung's full-chain semiconductor technology, reusing and optimizing cutting-edge processes validated during HBM4 mass production. The product employs the industry's most advanced sixth-generation 10-nanometer-class DRAM process (1c process), paired with Samsung Foundry's 4-nanometer logic base wafer, giving HBM4E higher process stability and manufacturability.
Energy Efficiency and Thermal Performance
Design and process optimizations in HBM4E's storage and logic architecture have improved performance, energy efficiency, and yield. Compared to HBM4, HBM4E delivers approximately 16% energy efficiency improvement and over 14% thermal resistance reduction. Samsung stated these improvements enable more efficient heat dissipation, enhancing operational reliability of high-load next-generation data centers while reducing energy consumption.
Production Timeline and Market Position
Samsung stated it will initiate HBM4E mass production according to customer schedules after completing sample delivery and customer adaptation optimization. HBM4, launched in February of this year, has received high recognition from global customers, with performance and energy efficiency particularly well-received.
Financial Performance
On April 30, Samsung Electronics disclosed Q1 results, with final verified operating profit calculated on a consolidated financial statement basis increasing 756.1% year-over-year to 57.2328 trillion Korean won (approximately 263.9 billion RMB), marking the second consecutive quarter of record-high single-quarter results. The Device Solutions (DS) division responsible for semiconductor business drove overall company performance growth, achieving Q1 sales of 81.7 trillion won and operating profit of 53.7 trillion won—both historical single-quarter highs.
FAQ
What is the capacity of Samsung's 12-layer HBM4E?
Samsung's 12-layer HBM4E offers 48GB per stack, representing over 30% capacity increase versus the previous generation. Samsung plans to add 32GB (8-layer) and 64GB (16-layer) configurations based on customer needs.
How much faster is HBM4E compared to HBM4?
HBM4E delivers over 20% performance improvement compared to HBM4, with stable pin rate of 14Gbps scalable to 16Gbps, and maximum single-stack memory bandwidth reaching 3.6TB/s.
What manufacturing process does Samsung use for HBM4E?
HBM4E employs Samsung's sixth-generation 10-nanometer-class DRAM process (1c process) paired with Samsung Foundry's 4-nanometer logic base wafer.
