[REVISION HISTORY]

Samsung 2nm and advanced semiconductor development

Updated 6 times since CLSTR started tracking revisions of this situation.

What changed

2026-10-21 00:56 UTC → 2026-10-21 13:07 UTC · added ~~removed~~

Samsung has introduced its SF2P 2nm Gate-All-Around (GAA) process, a third-generation platform utilizing a “hypercell” structure to improve performance-to-power ratios. Technical data indicates SF2P offers a performance-to-power ratio up to 2.6 times better than the 4nm (SF4) process. While mass production for SF2P is expected within the year, development for the 1.4nm node has been delayed until 2029. Commercial negotiations between Samsung and Qualcomm regarding 2nm processor manufacturing have experienced volatility. While the companies are in advanced discussions to provide Qualcomm with a critical alternative to TSMC, negotiations recently stalled due to pricing disputes. Qualcomm is seeking lower costs, while Samsung has maintained a firm stance, citing concerns that low production volumes may not offset lower margins. Despite these hurdles, the firms continue to collaborate on a 2.5D chip packaging system using organic bridges to connect multiple chips for AI and data center applications. Samsung is also transitioning its next-generation M2800 series 5G modems to a 2nm process, with development expected to conclude by 2027. This shift is intended to free up 4nm capacity for High Bandwidth Memory (HBM) base dies. Additionally, Samsung is expanding its mature process segment with the 8LPU+ platform, which utilizes Deep Ultraviolet (DUV) production to target edge AI, robotics, and automotive markets with improved efficiency over previous 8nm generations. Samsung Foundry is also reportedly nearing a manufacturing agreement with Anthropic to produce custom AI accelerators using the 2nm SF2 process, managed by Broadcom. In response to surging AI-driven data center demand, Samsung is upgrading its manufacturing facilities in China. The company is reportedly working to convert its Xi’an facility to produce 286-layer NAND memory, with plans to scale production from 5,000 wafers per month by the end of 2026 to 40,000 wafers per month by late 2027.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.