Technology Integration Impact: Major Conf: 95%

SK Hynix Jumps to TSMC 3nm for HBM4E Logic Die to Counter Samsung's 4nm Lead

Summary

SK Hynix plans to use TSMC's 3nm process for the logic die in its 7th-gen HBM4E, a leap from the 12nm used in HBM4. This aims to reverse the performance gap with Samsung (which used 4nm logic in HBM4) and deliver higher bandwidth and power efficiency for next-gen AI chips like NVIDIA's Vera Rubin Ultra.

Key Takeaways

SK Hynix is reportedly prioritizing TSMC's 3nm process for the logic die in its 7th-gen HBM4E, a significant upgrade from the 12nm used in HBM4. The core DRAM die will use 10nm-class 1c process. In contrast, Samsung used its own 4nm logic and 1c DRAM in HBM4, claiming performance leadership and first mass production. SK Hynix aims to reverse this by adopting finer 3nm logic, reducing electron travel distance and operating voltage, targeting NVIDIA's next-gen flagship 'Vera Rubin Ultra'. The custom HBM market is expected to expand from HBM4E, allowing customer-specific logic die designs, but SK Hynix plans to primarily use 3nm for most market shipments. AMD and Google have also announced HBM4E adoption in their next-gen AI chips.

Why It Matters

This move is a direct siege on Samsung's HBM4 performance lead. By skipping 4nm and jumping to TSMC 3nm, SK Hynix aims to dominate tail latency and power efficiency. However, two hidden traps exist: 1) Supply chain lock-in: Full reliance on TSMC 3nm makes SK Hynix+NVIDIA vulnerable to capacity and yield issues, while Samsung retains internal foundry flexibility. 2) Physical limitations downplayed: 3nm logic die's higher power density exacerbates thermal constraints in HBM stacks, potentially throttling real-world performance. The article omits thermal management costs and 3nm yield ramp challenges, which will inflate TCO. Additionally, custom HBM shifts logic design control to customers, eroding SK Hynix's architectural autonomy.

PRO Decision

[Vendors (Samsung, Micron)]: Accelerate own 3nm logic development and pitch internal foundry integration (e.g., Samsung 3nm GAA) as a supply chain stability advantage. Highlight thermal co-design capabilities to expose SK Hynix's 3nm thermal bottleneck in stacked HBM. [Enterprises]: Demand power-thermal curves and 3nm yield data from HBM4E suppliers. Insert multi-sourcing clauses to avoid SK Hynix+TSMC lock-in. For custom HBM, ensure logic die design is process-agnostic to preserve architectural flexibility. [Investors]: Watch SK Hynix's over-reliance on TSMC 3nm capacity. Samsung's vertical integration offers resilience during yield disruptions. Custom HBM trend may commoditize SK Hynix's logic role, pressuring margins.

Source: TrendForce
View Original →

Get 3-5 key AI infrastructure signals weekly →

💬 Comments (0)