IBM Unveils 0.7nm NanoStack Node With 666 Million Transistors/mm² as CFET Pushes Beyond 2nm
Updated
Updated · morethanmoore.substack.com · Jun 25
IBM Unveils 0.7nm NanoStack Node With 666 Million Transistors/mm² as CFET Pushes Beyond 2nm
3 articles · Updated · morethanmoore.substack.com · Jun 25
Summary
IBM said its 0.7nm NanoStack research node uses a staggered sequential CFET design to deliver 50% logic area scaling versus its 2nm process, with roughly 100 billion transistors on a fingernail-sized chip.
The architecture stacks nFET and pFET devices with different silicon orientations and a proprietary front-end bonding method, which IBM says improves routing and enables up to 65% more effective gate width than aligned sCFET designs.
IBM projects 50% higher performance at the same power or 70% lower power at the same performance, plus 40% SRAM scaling; the company estimates about five years before commercial adoption.
High-NA EUV at Albany NanoTech is expected to help cut patterning steps rather than make NanoStack possible, while EDA, thermal management, bonding yield and inspection remain key hurdles to production.
The announcement extends IBM's post-2nm roadmap and positions NanoStack as a candidate path for early smartphone processors or small AI chiplets if CFET manufacturing matures.
Will IBM's efficient chips curb AI's massive energy use, or simply accelerate the demand for more power-hungry data centers?
With TSMC launching 2nm chips in 2025, can IBM’s superior technology overcome a two-year market delay with its partner Rapidus?
As IBM leads in research but not mass production, is America's chip strategy shifting from manufacturing dominance to innovation control?
IBM NorthPole: Brain-Inspired AI Chip Delivers 25x Energy Efficiency for Edge Applications
Overview
IBM has unveiled NorthPole, a groundbreaking AI hardware chip inspired by the human brain’s efficiency. Designed to overcome the limitations of traditional computing, NorthPole delivers exceptional speed and energy efficiency, especially for demanding edge applications. By fundamentally enhancing neural inference, it sets a new standard for AI hardware. NorthPole’s architecture merges memory and processing, enabling faster and more efficient data handling. This innovation not only boosts performance but also reduces power consumption, making advanced AI more accessible in environments with limited resources. NorthPole represents IBM’s bold step toward the future of artificial intelligence.