Molex targets AI data centres with co-packaged copper solutions

As hyperscale data centres continue to expand to support artificial intelligence (AI) and high-performance computing, signal integrity and power efficiency are becoming increasingly crucial design constraints. Molex has introduced innovative copper solutions as a connectivity platform aimed at addressing these challenges in next-generation server architectures.

For eeNews Europe readers involved in designing high-speed interconnects or data-centre hardware, this development underscores how copper-based solutions may continue to evolve alongside optical technologies in AI infrastructure.

Co-packaged Architecture for High-Speed Copper Solutions

Molex has unveiled its Impress platform, a cutting-edge connector and cable system specifically designed to facilitate ultra-high-speed communication between application-specific integrated circuits (ASICs) and external connectivity. This approach involves placing the interconnect directly on the ASIC package substrate, thereby shortening signal paths across the printed circuit board and potentially reducing signal loss and crosstalk.

The system utilizes a compression-attached socket in conjunction with a mating cable assembly to enable high-speed data transfer for data-centre equipment. According to Molex, the architecture supports transmission speeds of up to 224Gbps PAM-4 and is engineered with scalability in mind for future network generations.

“As AI workloads continue to push data centres to their physical limits, our focus is on maximizing efficiency without compromising signal integrity,” stated Jairo Guerrero, VP & GM, Copper Solutions, Molex. “Impress represents our latest innovation aimed at facilitating the scaling of infrastructures without exponential increases in power consumption or cost. By enabling high performance at the rack level, Molex is making next-generation compute more technically and economically viable.”

Designed for Emerging AI and 224G Ecosystems

The Impress co-packaged copper platform is part of Molex’s comprehensive portfolio of high-speed interconnect technologies that target the transition to 224G networking and beyond. The demand for higher data rates is escalating as AI training, large language models, and cloud services propel rapid growth in data-centre bandwidth.

This connector system is engineered to streamline system maintenance and upgrades by allowing the socket to attach to the substrate without causing damage. Molex asserts that this approach can also facilitate rework and maintenance operations more efficiently, which could be particularly valuable in complex server platforms.

Impress builds upon the company’s earlier NearStack On-the-Substrate connector technology, which aimed to bring high-speed paths closer to the processor to reduce latency and enhance space efficiency in dense server designs. Molex reports that over one million NearStack units have already been deployed in data-centre environments.

The current implementation supports 224Gbps PAM-4 applications, with ongoing development efforts focused on validating the architecture for future 336G and 448G signalling speeds.