New Cooling Approaches Needed for Servers: Report

With artificial intelligence and quantum computing driving the need for more powerful servers, data center engineers need to juggle increasing data handling needs with higher power density and, subsequently, more heat dissipation. Some solutions might be found in new cooling methods, according to a report from interconnections supplier Molex designated In-Depth Report of Thermal Management Solutions for I/O Modules.

The report addresses the limitations of legacy approaches for thermal characterization and management and explores new innovations in server and optical module cooling to better support 112G and 224G connectivity.

 PAM-4 Driving Data Needs and Cooling

According to the report, the move to 224 Gbps PAM-4 interconnects between servers and network infrastructure represents a doubling of the per-lane data rate. Power consumption is also surging, with optical modules alone reaching as high as 40 W over long-range coherent links, up from 12 W just a few years ago. These trends are prompting companies to look at various cooling solutions.

The report noted that optical I/O modules inside servers and rack-mounted network infrastructure systems are cooled directly through active cooling systems. While modules for 56G and 112G can get by with air cooling, the move toward 224G optical modules may mean that exhaustion of forced air will be needed for the higher-speed parts.

Related:Engineering Solutions by Design News Talks Corvette Z06 Thermal Management

In the report, Molex explores the latest in air cooling, along with integrating creative liquid cooling solutions within existing form factors to address increased power and thermal demands on I/O modules. Direct-to-chip liquid cooling, immersion cooling, and the role of passive components to enhance active cooling are addressed. The report also delineates cooling methods that may be most effective to accommodate power demands in chips and I/O modules that scale to high levels.

 Molex cites one example of a liquid cooling solution, called the integrated floating pedestal. In this scenario, each pedestal that contacts the module is spring-taught and moves independently, allowing implementation of a single cold plate to different 1xN and 2xN single row and stacked cage configurations. This solution for a 1x6 QSFP-DD module utilizes six independently moving pedestals which can compensate for varying port stack heights while ensuring seamless thermal contact. As a result, heat flows directly from the module generating heat to the pedestal over the shortest possible conduction path to minimize thermal resistance and maximize heat transfer efficiency.

The report also looks at the inherent costs and risks associated with immersion cooling, which offers highly effective thermal cooling that exceeds roughly 50 kW per rack but requires a complete overhaul of a data center’s architecture.

Beyond liquid cooling, the report details advanced approaches to module design and thermal characterization poised to transform the performance of high-speed network interconnects. For I/O specifically, new solutions can be integrated into servers and switches for greater levels of heat sinking without compromising reliability. The report describes a Drop Down Heat Sink (DDHS) solution, developed by Molex, that maximizes the heat transfer capability of a traditional riding heat sink while minimizing metal-to-metal contact, which can create wear-and-tear on components.

As an active participant in the Open Compute Project (OCP), Molex is teaming with other industry leaders to develop next-gen cooling technologies that meet the evolving thermal management needs of today’s most demanding data center environments.