Back to Industries & Applications
In 2009, Facebook set out to optimize its data centers to better handle consumer demand as its platform grew in popularity. The company assembled a small team to create the world’s most efficient data center, and that team was able to increase energy efficiency by 38% while cutting operating costs by 24%.
Facebook made these designs public in 2011, then, in collaboration with companies such as Intel and Goldman Sachs, helped launch the Open Compute Project (OCP). Over the past decade, OCP has worked to replicate the collaborative open-source software environment for computer hardware, an environment in which cross-functional engineering teams create reference designs that accelerate the pace of innovation for the data center.
Today, the transformative pace of data center growth is accelerating demand for system optimization, especially in sectors that leverage high-performance computing (HPC) and artificial intelligence (AI). More than 200 companies, including Molex, have joined the OCP community to solve difficult and unique challenges. By applying a broader perspective, they aim to rapidly innovate and make hyperscale infrastructure more efficient, flexible, and scalable.
A New Artificial Intelligence Accelerator Solution
The world is experiencing an explosion of new AI applications that improve our everyday lives—from multiplayer gaming to cybersecurity protection, disease detection to supply chain optimization. Today’s cloud computing solutions just aren’t optimized for tomorrow’s AI workloads.
The Open Accelerator Infrastructure (OAI) group, a subgroup of OCP’s Server group, took up the task of defining an open standard for accelerator infrastructure systems, comprised of accelerator modules, a baseboard, tray, and chassis to handle these intense workloads. In 2019, the group released their Open Accelerator Module (OAM) v1.0 standard incorporating the Molex Mirror Mezz mezzanine connector.
The design specification called for a high-density board-to-board connection—from the GPU module to the baseboard—capable of handling 56Gbps. The connection had to be small enough to leave room for other key components, enable a high differential pair count, and maintain extremely low crosstalk levels. Molex Mirror Mezz connectors have been specifically designed for such tight constraints. With an exceptionally high density of 107 to 115 differential pairs (DP)/in² and 56Gbps PAM-4 signal integrity requirements, these connectors were designed to handle the data demands of intense AI applications.
Mirror Mezz Innovation
The Molex Mirror Mezz Connector is a hermaphroditic interface, meaning the connector mates to itself, reducing part numbers, simplifying bills-of-material, and consolidating supply chain efforts. It is available in a variety of mating heights, from 11.00mm to 5.00mm, and it can be used with flex cable links to increase board-to-board distance without sacrificing performance.
Although small in stature, Mirror Mezz Connectors are mighty, packed with innovative design features that enhance reliability and maintain high-speed signal integrity.
- The shrouded housing helps protect the pins and gives the user blind mating guidance, virtually eliminating the possibility of incorrect connector mating.
- The pins are shaped to improve connection reliability by providing constant two-point contact. These contact points provide a force on one another that resists vibrations and thermal drift.
- The connector is attached to the printed circuit board (PCB) using a proven, reliable, and easy to assemble ball grid array (BGA) matrix.
- Unique geometry features in the terminal design and solder ball maintain tighter impedance control.
- Wide ground pins provide signal isolation from neighboring differential pairs.
- The bottom paddles and the centers of each contact point are at a slightly different pitch and offset from one another to further reduce crosstalk.
Beyond 56 Gbps: Overcoming New Challenges with Mirror Mezz Pro
The innovative characteristics of the Mirror Mezz connector helped advance the first-generation OAM design, but OCP is not stopping there. The OCP OAI group is committed to advancing the accelerator capabilities, aiming for a new specification that supports speeds of 112Gbps, 224Gbps or more.
Molex has created an improved OCP-standard version of the Mirror Mezz connector to address the need for 112Gbps PAM-4 connectivity—Mirror Mezz Pro. This design delivers improved performance and signal integrity at twice the maximum data rate, while fitting the original compact PCB footprint. The size of the new connector’s BGA has been reduced to minimize return loss at higher frequencies, and the terminal and housings have been tuned to minimize impedance variation.
The latest and most innovative variant of the Mirror Mezz connector family is Mirror Mezz Enhanced. This product improves maximum data speeds to 224Gbps NRZ, enabling next-generation capacity and signal integrity for high-performance applications. The innovative design accommodates the same number of differential pairs in the same footprint as earlier Mirror Mezz versions, but with improved electrical performance that supports 224G data rates. The result: 224Gbps performance with the same mechanical and electrical robustness customers have already come to know.
Looking to the Future
Molex is honored that Mirror Mezz Pro has been selected by the Open Compute Project’s OAI group as the board-to-board mezzanine connector standard for OAM v1.5. We are thrilled to support the many collaborators that comprise OCP and welcome the opportunity to push the limits of next- generation hyperscale infrastructure with products like Mirror Mezz Enhanced. The Mirror Mezz Enhanced release is the latest in a long line of innovations that Molex looks to unveil through its collaboration with OCP.
Additional Content
Share