Circuit Protection – Server Thermal Management

What’s Old is New Again

The nature of computing has changed a great deal over the time it’s been around. Initially, computers with any decent processing power were expensive and bulky; thus, multiple users would use terminals to interact with and share a single computer. As computer technology improved, user machines (such as home PCs) became powerful enough to run the majority of applications, including tasks like modelling, simulation, and development. As a result, large mainframe computers became less important in everyday applications.

The introduction of the internet saw a surge of websites and users, which put a demand on specialized computers that could handle a massive amount of traffic and connections. At this point, the server farm was born, but even then these farms were dedicated to web hosting. The rise of the Internet of Things and machine learning has expanded the server farm’s purpose and believe it or not, has led to large mainframe machines becoming important once again.

A Hot Topic

Small devices that can take sensory readings and perform basic tasks typically do not have the processing capability to run artificial intelligence and machine learning algorithms, and so in these scenarios the data can be sent to the cloud for processing. However, cloud computing is not limited to just IoT devices — many applications that once ran on local machines are starting to be executed on the server side (such as Google Docs). The use of server-side, browser-based applications allows for a wider range of devices to use the same services but at the cost of higher server requirements.

This higher demand on servers requires more processing power which, in turn, typically results in more heat generation. Since uptime with servers translates to income, it is critical that servers do not experience any downtime. As such, circuit protection via thermal monitoring is a big concern in the industry.

One method for cooling down servers involves the use of cold countries like Iceland, where the outside air is cold enough to keep servers cool without spending money on air-conditioning. While this is practical for simple applications (such as website and file hosting), it is not practical for other applications; especially those involving low-latency data processing. For example, financial companies may have servers that require low-latency between the server and the stock market, but a server farm in Iceland will be far slower than a sever farm down the road from the stock market (in the stock market, profit is found in milliseconds).

Therefore, servers that need to be located near their customers or information resource need to employ complex thermal management solutions to not only protect themselves, but to also reduce operational costs.

Understanding the Importance of Thermal Management

Thermal management is a multi-dimensional problem and so can be problematic from a number of different angles. The first, and most obvious, is that heat can destroy and / or deteriorate electronics. Integrated circuits, components, and even circuit boards have a specific temperature range for which they are rated at and it is critical that designers ensure this range is not exceeded. Poor thermal management can also lead to performance loss with components that perform thermal throttling (many modern CPUs do this). When a device detects that it is getting too hot, some will reduce its clock speed (and therefore process less data) to reduce the heat generated, while others will simply shut down abruptly.

Thermal management is not just about temperature ranges — thermal cycles are also very important to consider. A PCB that is kept within its temperature range can still be damaged by thermal activity if the PCB experiences too many wide thermal cycles (where, say, it goes from very cold to very hot back to very cold). As the temperature of a PCB increases, the board can warp due to thermal expansion. This effect is hardly noticeable on old fashioned through-hole boards, however, boards that use BGA ICs can be particularly vulnerable to this. Therefore, it is better to keep PCBs at a consistent temperature regardless of the processing load.

Where Thermal Management Solutions Stand Today

Typically, circuit protection in the thermal realm will be concerned with component placement and product installation as opposed to the actual circuitry (although it is worth stating the obvious here – that some components such as thermal fuses do exist for thermal protection). When designing server PCBs and peripherals it is important to determine how the unit will be cooled and how that cooling will affect the overall board.

For example, boards that are to be air cooled will need to consider the fact that as air flows across the board, it will heat up; in turn, this will reduce the air’s capability to cool components further down the line. The use of air in this manner can also result in a thermal gradient that can cause warping of the PCB and therefore damage to component connections. If the PCB in question is too large for air cooling (the thermal gradient is too great), then liquid cooling can be considered; however, this comes with a major increase in cost and complexity. Another disadvantage of liquid cooling is that a failure in pipe connections can result in damaged circuitry (assuming that a conductive liquid such as building water is used).

AI Solutions in Thermal Management

While actively removing heat from a server is important for its operation, modern electronics allows for far more complex solutions to thermal management. The Industrial Internet of Things is still in its infancy, but it is already demonstrating how machinery and equipment can be monitored to not only improve performance but also protect from potential damage. A server farm that is fitted with small edge-computing IoT sensors that monitor temperature can be used to not only monitor which server racks are at risk of damage from overheating, but also predict server racks that will experience thermal issues. The data from the network of IoT sensors could also help load the server more efficiently, such that workloads from one server rack can be transmitted to other server racks that may be idle (and therefore cool).

Exploring Energy Harvesting Solutions

With the need for more energy efficient devices, the future of thermal management may be in the form of energy harvesting modules. Servers are large sources of heat, which can be a potential energy source if paired with a large enough thermal gradient. One example of how waste heat can be used to generate electricity is with the use of Peltier coolers — when an appropriate temperature difference is applied to them, they generate electricity. If cool air is applied across the cool side of a Peltier cooler, and the hot side is directly mounted to heat generating components, then the generated electricity can be fed back into the system to improve the overall efficiency of the operation to which it’s been applied.

Another example of energy harvesting would be the use of Stirling engines, which rely on a temperature gradient to move pistons which, in turn, can be used for things like generating electricity or operating the fluid pumps that remove the heat from hot components. While the energy generated from such a system can never be more than what is put into a server, it can help reduce overall energy consumption.

Conclusion

Thermal management in servers has never been more important than it is right now. With many services now turning to cloud-based computing and the expectancy that there will soon be more than 22 billion IoT devices the world over, servers will continue to see increased demand. Ensuring that thermal properties are carefully monitored will help to reduce downtime and integrating modern solutions like artificial intelligence into thermal management systems looks to be the likely next step forward for this particular field.