Monitor AI Datacenter Temperatures with NTC Thermistors

作者:Kenton Williston

資料提供者:DigiKey 北美編輯群

Datacenters face unprecedented thermal management challenges as artificial intelligence (AI) demands intensify and power densities increase. Precise, real-time temperature monitoring is needed to optimize performance and efficiency while preventing overheating. These sensing solutions must be accurate, responsive, and robust to handle rapidly varying thermal loads on highly sensitive equipment.

This article examines the thermal management challenges facing designers of modern AI datacenters, exploring various cooling systems, including air conditioning, immersion cooling, and heat pipe solutions. It then introduces negative temperature coefficient (NTC) thermistor solutions from EPCOS (TDK) and shows how they can be used to meet these challenges.

Why AI datacenters present new thermal management challenges

AI hardware such as graphics processing units (GPUs) and tensor processing units (TPUs) typically draws significantly more power than traditional central processing units (CPUs). As a result, AI-focused datacenters tend to have considerably higher power densities with concentrated hotspots that are difficult to manage with conventional cooling approaches.

To make matters worse, AI workloads tend to be highly variable, with the potential for rapid escalation in thermal loads during intensive training or inferencing operations. Without proper thermal management, these conditions can lead to performance throttling, unplanned outages, and accelerated hardware degradation.

In response to these emerging demands, datacenters are adopting more advanced cooling methods. Direct-to-chip cooling is one popular choice. This technique positions cooling pipes, cold plates, or heat exchangers directly against high-power components such as CPUs, GPUs, and memory. Immersion cooling is another option; here, entire servers are submerged in a non-conductive liquid.

Air conditioning is also getting a variety of upgrades. For example, in-row and in-rack units supplement facility-wide air conditioning with localized cooling that can respond to hot spots in real time.

Although the specifics of these cooling systems vary, they are all driving demand for more distributed and responsive temperature monitoring. Take direct-to-chip systems as an example. Heatsink-mounted sensors are needed at each targeted chip to ensure that temperature standards are maintained. Pipe-mounted sensors are required to monitor the incoming flow of coolant, and additional sensors are necessary for coolant distribution units and heat exchangers to verify that the system is operating efficiently.

Benefits of NTC thermistor sensors for datacenter applications

NTC thermistors are a good fit for all these requirements. As the name implies, an NTC sensor exhibits a predictable decrease in electrical resistance as temperature increases. In the case of NTC thermistors, this is accomplished with a small, heat-sensitive oxidized ceramic element enclosed in a protective metal or epoxy housing.

Figure 1 shows typical temperature vs. resistance curves for thermistors with ratings of 2 to 5 kilohms (kΩ) at 25°C. As this chart illustrates, thermistors with higher resistances are better for high-temperature applications because the change in resistance is easier to measure.

Image of typical temperature vs. resistance curvesFigure 1: Shown are typical temperature vs. resistance curves for thermistors with ratings of 2 kΩ to 5 kΩ at 25°C. (Image source: EPCOS (TDK))

The benefits of NTC thermistors for AI datacenters include:

  • High accuracy and fast response time: NTC thermistors are extremely sensitive to slight temperature changes and can respond quickly due to their small thermal mass. These factors make NTC thermistors well-matched to the rapidly fluctuating thermal demands of AI datacenters.
  • Durability and stability: NTC thermistors are constructed from robust materials and offer excellent long-term reliability with minimal resistance drift over time. This stability minimizes maintenance requirements and reduces the risk of unexpected downtime.
  • Compact size and flexible mounting: Their small size enables easy integration into crowded datacenter environments where space is at a premium. They are available in a wide variety of form factors, allowing them to address the full range of AI datacenter cooling systems.

The EPCOS family of NTC thermistors illustrates these advantages. The lineup includes solutions for monitoring heatsinks and pipes, immersion cooling systems, and air handling units.

Monitoring high-power components with heatsink-mounted NTC thermistors

High-power processors such as GPUs and TPUs require close thermal monitoring to maintain performance and prevent overheating. The B57703M0103G040 (Figure 2) is designed for direct attachment to heatsinks, making it well-suited for this task. This screw-on sensor encapsulates an NTC thermistor in a metal-tag case with a protruding ring lug.

Image of EPCOS B57703M0103G040 ring lug thermistorFigure 2: The B57703M0103G040 ring lug thermistor enables accurate temperature monitoring on heatsinks for high-power processors. (Image source: EPCOS (TDK))

The screw-on sensor design is both convenient and important for ensuring good thermal coupling and consistent contact pressure with the heatsink surface, reducing thermal resistance and improving measurement accuracy during rapid load changes.

The sensor has been tested for long-term stability over 10,000 hours at +70°C, supporting its use in the elevated temperature conditions common in AI datacenter workloads. Its 10 kΩ rating at +25°C provides a reliable basis for measuring higher operating temperatures, enabling accurate feedback for temperature control systems.

Monitoring liquid cooling pipes with NTC thermistors

Liquid cooling systems depend on a consistent supply of coolant at the proper temperature. The B58100A0506A000 (Figure 3) is a 10 kΩ NTC thermistor designed for quick installation on pipes, making it a good choice for monitoring coolant supply lines. This molded assembly clips directly onto pipes with diameters of 18 mm to 19 mm, with other sizes available for different installations. Built-in contact tabs provide a straightforward connection to monitoring equipment.

Image of EPCOS B58100A0506A000 clip-on thermistorFigure 3: The B58100A0506A000 clip-on thermistor measures coolant temperature in liquid cooling systems. (Image source: EPCOS (TDK))

Monitoring the external pipe temperature provides an accurate and maintenance-friendly way to track coolant performance without immersing the sensor directly in liquid. This approach minimizes installation complexity, reduces leak risks, and enables rapid sensor replacement if needed.

A key performance parameter for temperature sensors is the thermal time constant, also known as thermal response time, which reflects how quickly the resistance of an NTC thermistor changes in response to external temperature fluctuations. This parameter is influenced by the sensor’s design, mounting configuration, and the surrounding environment.

For example, the B58100A0506A000 uses a copper housing to thermally couple the sensor with the pipe to achieve a thermal time constant of less than 5 seconds (s), measured on the pipe. This quick response time helps ensure a reliable coolant supply.

Monitoring coolant distribution systems with NTC thermistors

In addition to monitoring heat sources and coolant supply lines, liquid cooling systems require temperature sensing at the coolant distribution unit, heat exchanger, and other central components. The B57800K0103A001 (Figure 4) is well-suited for this role. Its cylindrical copper case provides excellent thermal conductivity, enabling accurate measurement of fluid temperatures at critical points of the system.

Image of EPCOS B57800K0103A001 cylindrical probe thermistorFigure 4: The B57800K0103A001 cylindrical probe thermistor monitors coolant temperatures in distribution systems. (Image source: EPCOS (TDK))

With an operating temperature capability of +150°C, this 10 kΩ sensor can be placed in hot-side locations without risk of overheating. It has a thermal time constant of about 8 s in water, allowing it to track changes in coolant temperature with sufficient speed for system-level control and protection.

Deploying sensors at both the inlet and outlet of these components allows operators to track temperature differentials across heat exchangers or distribution loops. Significant deviations can indicate issues such as reduced coolant flow, partial blockages, or fouled heat exchanger surfaces, prompting preventative maintenance before system performance is impacted.

Monitoring immersion cooling systems with NTC thermistors

Immersion cooling systems, which submerge servers in a non-conductive liquid such as dielectric oil, require sensors that can withstand potential corrosion. The B57504K0103A009 (Figure 5) is a 10 kΩ sensor purpose-built for such environments. Its stainless-steel case provides durability in the presence of mild corrosives while ensuring effective thermal coupling to the sensing element.

Image of EPCOS B57504K0103A009 stainless-steel probe thermistorFigure 5: The B57504K0103A009 stainless-steel probe thermistor monitors coolant temperature in immersion cooling systems. (Image source: EPCOS (TDK))

This sensor has a thermal time constant of less than 2 s in water, enabling precise tracking of temperature changes within the immersion bath.

Monitoring air conditioning with NTC thermistors

Finally, consider the B57500M0103A005 (Figure 6). This 10 kΩ device uses a simple epoxy encapsulation to achieve a compact form factor that is coupled with 470 mm wire leads for flexible routing options. For example, the small size and long leads allow installation close to evaporator coils or within airflow channels, where rapid detection of temperature changes helps maintain stable climate control.

Image of EPCOS B57500M0103A005 epoxy-encapsulated thermistorFigure 6: The B57500M0103A005 epoxy-encapsulated thermistor can monitor air conditioning systems. (Image source: EPCOS (TDK))

Among its other benefits, this sensor has been validated to withstand vibration, rapid temperature cycling, and other hazards common in AC systems with no physical damage or significant impact on measurement accuracy.

Conclusion

AI workloads are creating an urgent need for distributed temperature monitoring within the datacenter. With options for heatsinks, coolant pipes, immersion baths, air handling units, and more, the EPCOS (TDK) family of NTC thermistors supports reliable, responsive operation in demanding environments.

DigiKey logo

聲明:各作者及/或論壇參與者於本網站所發表之意見、理念和觀點,概不反映 DigiKey 的意見、理念和觀點,亦非 DigiKey 的正式原則。

關於作者

Image of Kenton Williston

Kenton Williston

Kenton Williston 於 2000 年獲得電氣工程學士學位,並開始處理器基準分析師的職業生涯。從那時起,他在 EE Times 集團擔任編輯,協助推出和引導電子產業的多項刊物與會議。

關於出版者

DigiKey 北美編輯群