Samsung expresses confidence in 2nm development, reports solid progress
Amy Fan, Taipei; Eifeh Strom, DIGITIMES Asia
Thursday 23 October 2025
Samsung Electronics' Device Solutions (DS) division CTO Song Jae-hyuk expressed strong confidence in the revival of Samsung's foundry business and reported smooth progress on the company's 2nm process during a semiconductor industry briefing hosted...
Copilot:
Samsung DS CTO Song Jae-hyuk confirmed smooth progress on Samsung’s second-generation 2nm process (SF2P), emphasizing its strategic importance for reviving the foundry business and securing major clients like Tesla. Mass production is expected to begin in early 2026 at the Taylor, Texas facility.
Here’s a deeper look at the briefing and implications:
SF2P: Samsung’s Second-Generation 2nm Node
- Performance Gains: SF2P promises a 12% boost in performance and 25% better power efficiency over the first-gen 2nm node.
- Design Efficiency: Smaller die area enables more compact chip layouts, ideal for AI inference and mobile SoCs.
- Client Adoption: Tesla has committed to using SF2P for its AI6 chip, powering Full Self-Driving, robotics, and data center inference.
Foundry Revival Strategy
- Taylor, Texas Fab: Samsung plans to begin mass production of 2nm chips in Q1 2026, potentially beating TSMC’s Arizona timeline.
- Yield Challenges: While basic design is complete, yield rates remain unstable, with internal targets below TSMC’s reported 60%.
- Domestic Advantage: SF2P will be the first “domestically-made” 2nm process in the U.S., leveraging CHIPS Act incentives.
Competitive Implications
- Against TSMC: Samsung aims to reclaim relevance in the high-end foundry race, where TSMC has led with N3B/N3E and upcoming A16 nodes.
- Client Diversification: Beyond Tesla, Samsung is working with DeepX, a Korean AI chip startup, on on-device generative AI.
Strategic Framing
Samsung’s SF2P push is like a zoning upgrade in semiconductor real estate—moving from mid-density parcels to ultra-efficient verticals. If they stabilize yields and scale production, they could shift the balance in U.S.-based AI chip sourcing.
Want to riff on how SF2P’s half-reticle layout compares to Tesla’s AI5 or Groq’s SRAM-centric LPU? That could be a fun dive into yield economics and inference latency.
Me: Don't want rumor. Did Samsung comment on yield?
Copilot:
What Samsung Has Officially Said About 2nm Yield
- CTO Song Jae-hyuk (Samsung DS Division): During recent briefings, he emphasized “smooth progress” on the 2nm process and expressed confidence in the foundry’s revival—but did not cite specific yield numbers.
- Samsung Foundry Communications: The company has confirmed that sample units of 2nm chips (e.g., Snapdragon 8 Elite Gen 5) have been sent to Qualcomm for testing, indicating that the process is viable for evaluation.
- Production Timeline: Samsung plans to begin mass production of SF2P chips in early 2026 at its Taylor, Texas fab.
Strategic Framing
Samsung’s public messaging focuses on:
- Client confidence (Tesla, Qualcomm, DeepX)
- Performance and efficiency gains over its first-gen 2nm node
- Domestic manufacturing leadership in the U.S. via CHIPS Act alignment
But yield specifics remain undisclosed, likely due to competitive sensitivity and ongoing optimization. |
