UCIe (Universal Chiplet Interconnect Express)

UCIe (Universal Chiplet Interconnect Express) is an open die-to-die interconnect standard designed for chiplet-based architectures. Its primary goal is to enable chiplets from different vendors, manufactured on different process nodes and even by different foundries, to communicate seamlessly within the same package.

Often described as the equivalent of PCI Express for chiplet systems, UCIe aims to establish a common ecosystem for interoperable chiplets and heterogeneous integration.

History

The UCIe Consortium was launched in March 2022 by a group of leading semiconductor and technology companies, including:

  • Intel
  • AMD
  • Arm
  • Qualcomm
  • Samsung Electronics
  • TSMC
  • Google
  • Microsoft
  • Meta

The consortium has since expanded to include hundreds of members across the semiconductor ecosystem, including IP providers, EDA vendors, packaging companies, foundries, and system developers.

Architecture

UCIe consists of three primary layers:

Physical Layer (PHY)

The PHY handles the electrical transmission of data between chiplets and includes:

  • High-speed signaling
  • Link initialization and training
  • Clocking and synchronization
  • Error detection
  • Power management

Because communication occurs over very short distances inside a package, UCIe PHYs achieve significantly higher energy efficiency than traditional board-level interfaces.

Die-to-Die Adapter Layer

The Adapter Layer provides a reliable transport mechanism and includes:

  • Packetization
  • Flow control
  • Error handling
  • Retry mechanisms
  • Link management
  • Protocol multiplexing

This layer abstracts protocol-specific details from the underlying physical interface.

Protocol Layer

UCIe supports multiple upper-layer protocols, including:

  • PCI Express (PCIe)
  • Compute Express Link (CXL)
    • CXL.io
    • CXL.cache
    • CXL.mem
  • Streaming protocols
  • Vendor-specific protocols

This flexibility allows UCIe to support a wide range of applications, from AI accelerators to memory expansion devices.

Related Articles

Related Blogs

Related News