Part #/ Keyword
All Products

Alphawave Unveils 64 Gbit/s UCIe with 3nm Interconnect

2024-12-24 16:03:52Mr.Ming
twitter photos
twitter photos
twitter photos
Alphawave Unveils 64 Gbit/s UCIe with 3nm Interconnect

Alphawave Semi, a leading semiconductor IP company based in Canada, has recently showcased the world's first 64 Gbit/s Die-to-Die (D2D) IP subsystem designed for Universal Chiplet Interconnect Express (UCIe) technology.

This third-generation chiplet interconnect IP is set to be built using TSMC's cutting-edge 3nm process technology, leveraging TSMC's advanced packaging techniques and industry standards. It is based on the latest Gen2 36 Gbit/s and Gen1 24 Gbit/s implementations, also utilizing TSMC's 3nm process. The high bandwidth interconnect provides over 20 Tbit/s/mm bandwidth density, offering significant advantages in low power consumption and low latency.

The IP is highly configurable and supports multiple protocols, including AXI-4, AXI-S, CXS, CHI, and CHI-C2C, addressing the growing demand for high-performance connectivity in applications such as High-Performance Computing (HPC), data centers, and artificial intelligence (AI).

Compliant with the latest UCIe specifications, the IP features a scalable architecture with advanced testability functions, including real-time channel health monitoring. This D2D interconnect solution is designed to facilitate a range of both established and emerging chiplet connection scenarios. Common use cases include linking compute chiplets for low-latency, consistent connectivity via UCIe's streaming functionality, and connecting compute to I/O chiplets through UCIe interfaces with PCIe, CXL, or Ethernet.

To enhance system connectivity, optical retimers are employed to create reliable, low-latency optical I/O links via optical engines. This supports the development of low-power, high-speed solutions in data centers and AI/ML systems.

For high-performance applications, the creation of custom HBM-based chips using the latest UCIe standards represents a groundbreaking approach. It involves tightly integrating memory chips with compute chips to achieve ultra-high bandwidth and low-latency communication between components. This method optimizes memory transactions, especially in AI applications, by reusing the die-to-die interfaces already present on the main chip for kernel-to-kernel or kernel-to-I/O connectivity.

Brian Rea, Chair of the UCIe Alliance Marketing Working Group, stated, “The UCIe Alliance is thrilled to see our members reaching key milestones, such as successful tape-outs, reflecting the growing adoption of UCIe standards. UCIe is the cornerstone of the chiplet industry, providing robust solutions for high-speed, low-latency die-to-die interconnects. By embracing open standards, we enable the industry to accelerate innovation, shorten time-to-market, and deliver breakthrough technologies.”

Mohit Gupta, Senior Vice President and General Manager of Alphawave Semi, added, “We've successfully taped out Gen2 UCIe IP at 36 Gbit/s using 3nm technology, building upon our pioneering silicon-validated 3nm UCIe IP with CoWoS packaging. This sets the stage for our 64 Gbit/s Gen3 UCIe IP, which aims to deliver high-performance, 20 Tbit/s/mm throughput capabilities to meet the critical AI bandwidth demands of our customers by 2025.”

* Solemnly declare: The copyright of this article belongs to the original author. The reprinted article is only for the purpose of disseminating more information. If the author's information is marked incorrectly, please contact us to modify or delete it as soon as possible. Thank you for your attention!