‘The socket is becoming the motherboard’ — how a Samsung-backed startup wants to change computing forever by integrating everything in an elegant packaging to tackle the AI beast once and for all
Many system designers are exploring chiplet-based SiPs that move beyond the limitations and costs of huge single-die implementations, but which depend heavily on silicon interposers as substrates for mounting and interconnecting the dies.
Silicon interposers deliver a higher data rate than organic substrates, but are not without drawbacks. They are costly, proprietary and limit the number of chiplets that can be placed on one substrate due to size restrictions, while increasing TCO.
Many CPUs are multi-die assemblies which make use of organic substrates (implemented using chiplet interconnect standards like UCIe and the Open Compute Project’s Bunch of Wires (BoW)), but they can’t compete performance-wise with silicon interposers.
Build a better ‘Blackwell’
Now, Eliyan has come up with what it believes is a viable solution that offers the performance advantages of silicon interposers but without their limitations. The solution to this dilemma, the firm says, is to look at the dies themselves – “at the tiny electronic circuits that drive the interconnect lines”. Most of today’s chiplet interconnect standards use a physical-layer (PHY) IP block, but Eliyan’s NuLink PHY reportedly reaches the same maximum performance levels on organic substrates that alternative PHYs can only achieve on silicon interposers.
The company says NuLink PHY, “enables systems with higher performance (more memory) and lower TCO (no interposer),” with results that deliver up to 4x the bandwidth, 4x the power efficiency, up to 4x the SiP size and up to 10x the AI performance.
Eliyan’s Co-founder and chief executive officer Ramin Farjadrad recently talked to The Next Platform about how NuLink PHY can be used to “build better, cheaper, and more powerful compute engines than can be done with current packaging techniques based on silicon interposers.”
The article, titled “How To Build A Better ‘Blackwell’ GPU Than Nvidia Did”, explains how its technology could be used to significantly improve on Nvidia’s superchip design.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
It’s a fascinating read, and hints at what could be possible in the future. As The Next Platform’s Timothy Prickett Morgan sums up, “Any memory, any co-packaged optics, any PCI-Express or other controller, can be linked using NuLink to any XPU. At this point, the socket really has become the motherboard.”
There’s certainly a lot of interest in Eliyan’s NuLink PHY. The company recently closed a $60 Million Series B funding round, co-led by Samsung Catalyst Fund and Tiger Global Management. “This investment reflects the confidence in our approach to integrating multi-chip architectures that address the critical challenges of high costs, low yield, power consumption, manufacturing complexity, and size limitations,” Farjadrad said, following the funding announcement.
More from TechRadar Pro
Many system designers are exploring chiplet-based SiPs that move beyond the limitations and costs of huge single-die implementations, but which depend heavily on silicon interposers as substrates for mounting and interconnecting the dies. Silicon interposers deliver a higher data rate than organic substrates, but are not without drawbacks. They are…
Recent Posts
- Amazon develops a warehouse robot workers can speak to
- This App Makes Google TV Actually Usable
- Google Wallet ID passes will be available in select EU states this summer
- Shokz upgraded its open earbuds with better sound and a lighter design
- Shokz says its clip-on OpenDots 2 earbuds focus on improved volume and bass
Archives
- June 2026
- May 2026
- April 2026
- March 2026
- February 2026
- January 2026
- December 2025
- November 2025
- October 2025
- September 2025
- August 2025
- July 2025
- June 2025
- May 2025
- April 2025
- March 2025
- February 2025
- January 2025
- December 2024
- November 2024
- October 2024
- September 2024
- August 2024
- July 2024
- June 2024
- May 2024
- April 2024
- March 2024
- February 2024
- January 2024
- December 2023
- November 2023
- October 2023
- September 2023
- August 2023
- July 2023
- June 2023