Samsung First to Mass Produce HBM4 AI Memory – A Blow to SK Hynix?

Samsung Electronics has become the first company globally to begin mass production and shipment of its 6th generation High Bandwidth Memory (HBM4), a critical component for artificial intelligence (AI) semiconductors. This move marks a strategic response from Samsung, following a previous generation where SK Hynix took the lead in HBM technology.

Regaining Momentum in the AI Memory Market

The company announced on February 12th the shipment of the industry’s highest-performing HBM4, accelerating its release schedule by approximately one week following consultations with customers. Industry analysts interpret this as a clear signal of Samsung’s intent to reverse the momentum established by SK Hynix with the HBM3E (5th generation) products.

Did You Know? Samsung’s HBM4 achieves a data transfer rate of 11.7Gbps, exceeding the JEDEC standard of 8Gbps by approximately 46%.

The new HBM4 boasts a data transfer speed of 11.7 gigabits per second (Gbps), surpassing the JEDEC standard of 8Gbps by roughly 46%. It also represents a 1.22x increase over the previous generation, HBM3E, which operated at 9.6Gbps, with potential to reach up to 13Gbps. The focus is on alleviating data bottlenecks as AI models continue to grow in complexity.

Enhanced Performance and Capacity

The HBM4 offers a maximum memory bandwidth of 3.3 terabytes per second (TB/s) per stack, a 2.7x improvement over the previous generation. This exceeds current customer requirements of 3.0TB/s. Configurations with 12 layers provide capacities ranging from 24 to 36 gigabytes (GB), and can be expanded to 48GB with 16 layers.

A key differentiator lies in Samsung’s manufacturing processes. The company proactively adopted 10-nanometer 6th generation (1c) DRAM and utilized its 4-nanometer foundry process for the base die—the underlying layer of the HBM stack. This leverages Samsung’s unique capability of manufacturing both memory and logic components.

Expert Insight: Samsung’s decision to prioritize speed and adopt cutting-edge manufacturing processes for HBM4 demonstrates a commitment to regaining leadership in the AI memory market, particularly after facing competition from SK Hynix in the previous generation.

As a result of these advancements, energy efficiency is improved by approximately 40% compared to the previous generation, while thermal resistance and heat dissipation are enhanced by roughly 10% and 30%, respectively. According to Samsung’s Vice President of Memory Development, Hwang Sang-joon, the company broke with tradition by implementing the most advanced 1c DRAM and 4-nanometer foundry processes.

Future Roadmap and Expansion Plans

Samsung has outlined its future product roadmap, planning to sample HBM4E in the second half of 2026 and begin sampling custom HBM solutions tailored to specific customer needs starting in 2027. This strategy involves designing capacity, speed, power characteristics, and interfaces to maximize performance for AI accelerators and graphics processing units (GPUs).

The company anticipates that HBM4 sales will increase by more than three times year-over-year. To support this growth, Samsung is accelerating the expansion of its production capacity, designating the 5th line of its Pyeongtaek Plant 2, scheduled to be operational in 2028, as a primary HBM production hub.

Samsung’s CTO of the Device Solutions (DS) division, Song Jae-hyuk, stated that HBM4 is “the best in technology” and that customer feedback has been “very satisfactory.” He also expressed confidence in maintaining industry leadership with the development of HBM4E and HBM5.

A Competitive Response

According to Kim Yong-seok, a distinguished professor at Gachon University’s Semiconductor College, Samsung’s emphasis on being “world’s first” is unusual and reflects a strong sense of urgency to regain market share. He believes this signals a determination to lead not only the HBM4 generation but also the broader AI memory market.

Frequently Asked Questions

What is HBM4?

HBM4 is the 6th generation High Bandwidth Memory, a critical component for artificial intelligence (AI) semiconductors, offering increased speed and capacity compared to previous generations.

How does Samsung’s HBM4 compare to the previous generation?

Samsung’s HBM4 achieves a data transfer speed of 11.7Gbps, a 1.22x increase over the HBM3E (9.6Gbps), and offers a 2.7x improvement in memory bandwidth.

What are Samsung’s plans for future HBM development?

Samsung plans to sample HBM4E in the second half of 2026 and begin sampling custom HBM solutions in 2027, tailored to specific customer needs.

As AI technology continues to evolve, how might the demand for advanced memory solutions like HBM4 shape the future of semiconductor innovation?