Story perspectives

Intel and AMD Propose 10x Faster AI Memory Solutions

10/12/2025

Summary

Story summary

Researchers from Intel and Advanced Micro Devices (AMD) published a paper proposing enhancements to Universal Chiplet Interconnect Express (UCIe) for on-package memory.
The proposal targets AI memory bottlenecks by boosting power efficiency, bandwidth, and cost.
It suggests up to 10x higher bandwidth density and about 3x lower latency and power.
Amazon Web Services' (AWS) Trainium reduces Nvidia reliance for AI workloads, and Micron Technology and SK Hynix enable CXL-based disaggregation.

History

Historical Context

The Universal Chiplet Interconnect Express (UCIe), introduced in 2022, aims to standardize chiplet designs, enhancing semiconductor efficiency. This initiative follows the rise of 2.5D and 3D integration technologies, which improve performance in high-performance computing and AI. Collaborations between Intel and AMD could yield up to 10x higher bandwidth density, reshaping the semiconductor landscape amid increasing demand for advanced AI capabilities.

1 of 2

Story summary

Researchers from Intel and Advanced Micro Devices (AMD) published a paper proposing enhancements to Universal Chiplet Interconnect Express (UCIe) for on-package memory.
The proposal targets AI memory bottlenecks by boosting power efficiency, bandwidth, and cost.
It suggests up to 10x higher bandwidth density and about 3x lower latency and power.
Amazon Web Services' (AWS) Trainium reduces Nvidia reliance for AI workloads, and Micron Technology and SK Hynix enable CXL-based disaggregation.

1 / 2