Korea

Samsung and SK Hynix Split on HBM4, Forcing a Design Choice for AI Chipmakers

At Hot Chips 2026, Samsung Electronics emphasized 'active base die' technology, transforming the HBM base die from a passive connector into an active logic chip using 4nm logic processes.

AsiaAI Publisher · August 24, 2026 · 3 min read · Source: 전자신문 ETNews · Issue #75

What this means

Why it matters: The specific technical details shared by Samsung and SK Hynix reveal their contrasting long-term visions for HBM. Samsung’s move to a more intelligent, logic-enabled base die fundamentally changes how memory and logic interact, shifting HBM from a memory module to a co-processor. SK Hynix is focused on mastering the physical challenges of stacking more DRAM layers, which is an incremental but equally vital engineering problem for production at scale.

For Western readers: Western AI chip designers and data center operators should recognize that future HBM will offer more than just increased bandwidth; Samsung’s approach suggests HBM could take on more localized processing tasks, altering how workloads are partitioned between the accelerator and memory stack. Conversely, SK Hynix’s focus on packaging implies continued improvements in density and thermal management, crucial for high-power AI systems.

What to watch: Monitor which of these distinct HBM architecture approaches—active base die logic vs. advanced thermal packaging for higher stacks—gains traction in adoption by major AI accelerator customers in the next 12-18 months.

This story ran in Issue #75, alongside three other stories.

Samsung and SK Hynix are taking different paths in foundational HBM development. Samsung wants to put logic directly into the HBM die. SK Hynix is focusing instead on packaging and heat management. This split means the AI chip market will likely see two different HBM designs for HBM4 and beyond. Chip designers must choose where they want processing power and heat control to live.

The Korean press, including *ETNews*, sees this as a big gamble for global AI leadership. It is more than a simple technical decision. For Samsung, using an “active base die” is a way to keep control of the design. This method turns a memory part into a smarter subsystem. Samsung wants to combine logic and memory by using its advanced chip factories. This is a bold move to change how memory connects to chips.

SK Hynix is focusing on how to make chips easily and in high volumes. They are working on MR-MUF and hybrid bonding to solve heat and warping issues. This shows they want to push current HBM limits with proven packaging methods. Their pragmatic plan focuses on engineering to keep HBM reliable for dense AI systems. This method uses ideas from Japanese materials science and precision manufacturing for new memory stacks.

Both companies face big risks with how customers adopt their technology. Samsung’s active base die requires big design changes from chipmakers. This change could slow sales if the extra speed does not justify the complex design. SK Hynix has a more gradual plan but still faces a hard task. They must get high yields on 16-layer stacks, where heat control is very difficult. The market might not choose one best HBM type, but could use both designs for different AI tasks.

To see how this battle ends, look for specific design wins in 2027. We must see which design Nvidia, AMD, or custom chipmakers choose. We should also watch the yield rates for 16-layer HBM stacks from both firms. Output and reliability will decide the winner, rather than just raw speed.

📊 Featured Chart

SK Hynix HBM4 Improvement Targets vs. HBM3E

HBM4 targets vs. HBM3E

For the wider picture, see Korea Semiconductor Ecosystem.

Original source (Korean)

삼성·SK, ‘핫칩스 2026’서 차세대 HBM 병목 다른 해법 제시

전자신문 ETNews

This story appeared in AsiaAI.FYI Issue #75.

Leave a Reply Cancel reply