VFF - The signal in the noise
NewsTrending

IBM's Vertical Chip Design Extends Moore's Law Another Decade

Read original
Share
IBM's Vertical Chip Design Extends Moore's Law Another Decade

IBM has demonstrated a prototype chip with 100 billion transistors stacked vertically in two layers, doubling the density of its 2021 design. The nanostack architecture uses complementary field-effect transistors (CFETs) with staggered alignment to pack more transistors without shrinking them further, potentially extending Moore's Law by another decade. The company reports 50% performance gains and 70% energy efficiency improvements over its previous state-of-the-art design.

  • IBM built a prototype chip with 100 billion transistors on a fingernail-sized area, double the density of its 2021 technology
  • The nanostack architecture stacks transistors vertically in two layers using staggered complementary field-effect transistors (CFETs)
  • Chips using this approach could deliver 50% more performance and 70% better energy efficiency than IBM's previous generation
  • IBM expects nanostacking to be widely deployed in data centers within a decade, with partnerships planned with semiconductor manufacturers

Transistors have approached the physical limits of miniaturization at just a few dozen nanometers, making traditional Moore's Law scaling impossible. IBM's vertical stacking approach offers a path forward by building upward rather than shrinking further, potentially sustaining chip performance improvements for another 10 to 15 years. This matters because computing infrastructure, particularly data centers, depends on continued efficiency gains to manage power consumption and costs.

Semiconductor manufacturers including Intel, Samsung, and TSMC are all pursuing similar vertical stacking strategies, making this a critical competitive battleground. IBM's claimed advantages in alignment precision and simplified wiring could influence which companies win design partnerships and market share. Data center operators face mounting energy costs, making 70% efficiency gains a material business consideration for infrastructure investment decisions.

  • IBM will license the nanostack architecture to semiconductor manufacturers rather than producing chips directly, positioning it as an industry standard rather than a proprietary advantage
  • The technology is expected to be deployed across multiple chip types including GPUs and CPUs, broadening its impact beyond specialized applications
  • Competing approaches from Intel, Samsung, TSMC, and Imec suggest the industry is converging on vertical stacking as the solution to Moore's Law limits, but IBM's staggered alignment method may offer performance or manufacturing advantages

Monitor announcements from semiconductor manufacturers about adopting IBM's nanostack architecture and timelines for commercial deployment. Track competing CFET designs from Intel, Samsung, TSMC, and Imec to assess whether IBM's staggered alignment approach delivers claimed performance and efficiency advantages in production. Watch for data center adoption rates and energy consumption metrics once nanostacking chips become available, as these will validate the technology's real-world impact.

OneUpAI
OneUp Your Business. Get More Done. OneUp Your Business. Get More Done. OneUp Your Business. Get More Done.
Learn More
Share

Subscribe to the newsletter

The latest stories and analysis, delivered to your inbox.

Free. No spam. Unsubscribe any time.

Related stories

Meta Launches Wearable Device for Muse AI Agent
TrendingNews

Meta Launches Wearable Device for Muse AI Agent

Meta has created a small wearable device that serves as a hardware platform for its Muse AI agent, expanding the ways users can interact with the AI beyond traditional mobile interfaces. The device functions similarly to a Tamagotchi, suggesting a portable, always-on form factor for the AI assistant. This move represents Meta's strategy to embed its AI agent across multiple hardware platforms and form factors.

by Lucas Ropek· TechCrunch AI
Anthropic Seeks 1GW Compute Lease from Apollo-Backed Stream Data Centers

Anthropic Seeks 1GW Compute Lease from Apollo-Backed Stream Data Centers

Anthropic is in early-stage negotiations to lease up to 1 gigawatt of compute capacity from Stream Data Centers, which is majority-owned by Apollo Global Management. The facility would house tensor processing units designed by Broadcom and Google. The deal reflects ongoing competition among AI labs to secure dedicated infrastructure for large-scale model training and deployment.

by Laura Mandaro· The Information
Rabbit Launches Cloud AI Agent That Works Without R1

Rabbit Launches Cloud AI Agent That Works Without R1

Rabbit has launched OS3, a cloud-based AI agent that runs on Windows, Mac, and Linux devices without requiring its R1 hardware. The system allows users to connect up to five devices to a single account, select preferred AI models, and access the agent through desktop, messaging apps like Telegram or iMessage, or the R1 device itself. OS3 automatically determines which devices, files, apps, and models it needs to complete tasks.

by Emma Roth· The Verge AI
NVIDIA Isaac ROS 5.0 Brings AI Agents to Robotics Development

NVIDIA Isaac ROS 5.0 Brings AI Agents to Robotics Development

NVIDIA released Isaac ROS 5.0 at ROSCon in Toronto, introducing agentic workflows and new platform support for robotics development. The update adds GPU-accelerated tools, AI agent-ready documentation, and new skills like FoundationStereo fine-tuning and FoundationPose inference that enable faster perception and object tracking. The release targets the 1.3 million ROS users and includes support for ROS Lyrical and Ubuntu 24.04, positioning NVIDIA's accelerated computing as a standard for high-performance robotics applications.

by Katie Washabaugh· NVIDIA Blog (AI)