By Norman Quesnel
Senior Member of Marketing Staff
Advanced Thermal Solutions, Inc.
Rapid advancements in fiber optic technology have increased transfer rates from 10GbE to 40/100GbE within data centers. With the emergence of 100GbE technologies, the creation of data center network architectures free from bandwidth constraints has been made possible. The major enabler of this performance increase is the QSFP optical transceiver.
QSFP is the Quad (4-channel) Small Form-Factor Pluggable optical transceiver standard. A QSFP transceiver interfaces a network device, e.g. switch, router, media converter, to a fiber optic or copper cable connection as part of a Fast Ethernet LAN.
The QSFP design became an industry standard via the Small Form Factor Committee in 2009. Since then, the format has steadily evolved to enable higher data rates. Today, the QSFP MSA (multi-source agreement) specification supports Ethernet, Fibre Channel (FC), InfiniBand and SONET/SDH standards with different data rate options.
The small QSFP form factor has significantly increased the number of ports per package. The increased density of transceivers can lead to heat issues. The optical modules can get hot due to their use of lasers to transmit data. Even though the popular QSFP28 provides lower power dissipation than earlier transceivers – abut 3.5W, the QSFP28 factor has also allowed a significant increase in port density.
Newer microQSFPs can dissipate even more heat. microQSFP interconnects fit more ports (up to 72) on a standard line card, saving significant design space.
The performance and longevity of the transceiver lasers depend on the ambient temperature they operate in and the thermal characteristics of the packaging of these devices. The typical thermal management approach combines heat dissipating fins, e.g. heat sinks, and directed airflow.
Recently, Advanced Thermal Solutions, Inc. (ATS) tested a variety of pin and fin-style heat sinks for their comparative cooling performance on a standard QSFP connector cage. For this setup, an even amount of heat was provided to each connector site via a heater block. Individual thermocouples measured the heat flux resulting with the different heat sink types.
A main goal of this test was how each of four heat sinks would perform while relying on airflow incoming from just one side. By the time it reached the fourth heat sink would the airflow provide enough conduction for adequate cooling? An image from this series of tests is below in Figure 4.
The tests results showed that the denser the heat sink pins or fins on the sink closest to the incoming air, the hotter the farthest away QSFP will be. Thus, the best solution used heat sinks whose pin/fin layouts were optimized to work in the actual airflow reaching them.
This meant more open layouts closer to the air source, allowing more air to reach denser pin/fin sinks farther from the air. The non-homogeneous heat sinks allowed for a low, uniform temperature across the QSFP for the most effective function of the QSFPs’ lasers.
Cooling solutions are different between QSFP28 designs and microQSFP installations. QSFP28 transceiver cooling is typically provided at multiple connector sites. microQSFP modules, e.g. from TE Connectivity, have an integrated heat sink in the individual optical module. Used with connection cages that are optimized for airflow, their heat is controlled in high density applications.
Finally, another factor affecting cooling performance is surface finish and flatness. Designers can reduce thermal spreading losses by keeping the heat sources close to the thermal interface area and by increasing the thermal conductivity of the case materials.
For QSFP, the size of the cage hole for heat sink contact given in the multi-source agreement (MSA) can be increased giving a reduction in the thermal interface resistance and therefore module temperature.
1. FMAD IO, http://fmad.io/images/blog/20160612-100g-connectors.png
3. microQSFP, http://www.microqsfp.com/
4. TE Connectivity, https://www.youtube.com/watch?v=k_qNj-yAKz4