WireByte

Home / AI & Machine Learning

Photo of network, robotics warehouse, drone
Image: via cdn.mos.cms.futurecdn.net
AI & Machine Learning

Intel Proposes New Memory Architecture to Ease AI Bottleneck

WireByte Staff · July 7, 2026

Intel has filed a patent for a new high-bandwidth memory architecture, dubbed cross-batch memory (XBM), which aims to reduce the packaging and cost bottleneck of traditional interposer-based HBM. The design uses backend transistors and UCIe links to match HBM4's footprint while shrinking the package and building in defect repair.

Key points

  • Intel has filed a patent for a new memory architecture called cross-batch memory (XBM), which aims to ease the packaging and cost bottleneck of traditional interposer-based HBM.
  • The XBM design uses backend transistors and UCIe links to match HBM4's footprint while shrinking the package and building in defect repair.
  • Each memory die in the XBM stack holds 1.5 GB of 1T1C DRAM, arranged in a 32-by-24 grid and grouped into eight channels of eight sub-channels each.
  • The XBM stack is designed to be scalable, with the potential to stack up to 16 dies high.
  • The proposed design is intended to address the high costs associated with traditional HBM, which are driven by the need for a silicon interposer.
  • The patent filing has sparked interest in the tech industry, with analysts saying it could be a significant development for AI applications.

Intel has filed a patent for a new memory architecture called cross-batch memory (XBM), which aims to ease the packaging and cost bottleneck of traditional interposer-based HBM. The design uses backend transistors and UCIe links to match HBM4's footprint while shrinking the package and building in defect repair.

The XBM stack is designed to be highly scalable, with the potential to stack up to 16 dies high. Each memory die holds 1.5 GB of 1T1C DRAM, arranged in a 32-by-24 grid and grouped into eight channels of eight sub-channels each.

The proposed design is intended to address the high costs associated with traditional HBM, which are driven by the need for a silicon interposer. By eliminating the interposer and shrinking the package, Intel hopes to make XBM a more cost-effective option for AI applications.

The patent filing has sparked interest in the tech industry, with analysts saying it could be a significant development for AI applications. However, it's worth noting that the patent is still in the early stages, and it will likely be several years before we see XBM in commercial products.

Sources

WireByte Staff — Editorial Team

The WireByte editorial team synthesises technology news from multiple primary sources, verifies the facts, and links every source. Articles are produced with AI assistance and reviewed under our editorial policy.