In the previous article, we have covered how to monitor the temperature of the circuit board. However, power management in high-performance processors such as central processing units (CPUs), graphics processing units (GPUs), application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs) is often more complex. Through temperature monitoring, these systems can not only initiate safe system shutdown procedures, but also use temperature data to dynamically adjust performance.

Monitoring process temperature increases system reliability and maximizes performance. As shown in the figure below, high-performance processors typically use heat sinks to absorb excess heat from the die. Higher temperatures may activate cooling fans, modify the system clock, or quickly shut down the system when the processor exceeds its temperature threshold.

Motherboards with high-performance processors usually require a heatsink

Design Considerations for Die Temperature Monitoring

To achieve efficient temperature monitoring, high-performance processors have two design considerations: temperature accuracy and sensor placement. The temperature accuracy of the processor is directly related to the sensor position.

Improve system performance with high-accuracy temperature monitoring

As shown in the figure above, processor performance can be maximized with high-accuracy temperature monitoring, pushing the system to its temperature design limits. While most integrated circuits have built-in temperature sensors, the accuracy of these sensors is not uniform due to wafer-to-wafer and other batch-to-batch variability. In addition, the processor must be conditioned against the benchmark, adjusting the coefficient relative to die temperature. High-performance processors inherently have complex circuits and cause self-heating, and thus generate temperature errors that increase with temperature. If a system is designed with lower accuracy and temperature error, the performance of the system will not be maximized within its temperature design limits.

Sensor placement and accuracy

An integrated temperature sensor or temperature diode or an external temperature sensor can monitor the thermal performance of the processor. In some cases, using both internal and external sensors can maximize system performance and improve reliability.

Bipolar Junction Transistor Integrated Temperature Sensor

Some high-performance processors contain bipolar junction transistors (BJTs) for temperature sensing. The BJT has a very predictable transfer function that is temperature dependent. Remote temperature sensors use this principle to measure die temperature. The most common BJT in CMOS process is P-channel N-channel P-channel (PNP). The figure below shows a remote temperature monitoring circuit used to measure the PNP transistor connection configuration.

Base-emitter voltage change (ΔVBE) measured with two currents

The process of designing a remote temperature monitoring system can be challenging due to noise and errors caused by wafer and lot-to-lot variability. Temperature diode errors can be caused by:

• Ideality factor change. The characteristics of BJT temperature diodes depend on process geometry factors and other process variables. If the ideality factor n is known, the n-factor error can be corrected using the n-factor register. Alternatively, software calibration methods can be used to correct for ideality factor changes over the desired temperature range.

• Series resistance. Any resistance in the signal path will cause a voltage offset due to the current source. Modern remote temperature sensors use a series resistance algorithm that eliminates temperature errors caused by resistances up to 1-2kΩ. The algorithm achieves robust and accurate measurements even when combined with a resistor-capacitor filter.

• Noise injection. Electromagnetic interference or inductance coupling into remote printed circuit board traces can cause errors when diode traces are run in parallel with high-frequency signal traces that carry high currents. This is one of the most important board design considerations for remote temperature sensors.

• Beta Compensation. A temperature transistor integrated into an FPGA or processor may have a beta value of less than 1. Remote temperature sensors with beta compensation are specifically designed to work with these transistors and correct for temperature measurement errors associated with them. The beta compensation feature provides no benefit when used with discrete transistors.

Device Recommendations

The TMP421 provides a single channel to monitor the BJT; there is also a multi-channel remote temperature sensor supporting up to eight channels to measure temperature locally and remotely.

The TMP451 provides high accuracy (0.0625°C) temperature measurement both locally and remotely. Server, laptop and automotive sensor fusion applications can benefit from multi-channel remote sensors.

External temperature sensor

While the built-in temperature sensor is optimally positioned, its accuracy is as low as ±5°C. Adding an external local temperature sensor can improve die temperature accuracy and improve system performance. Local temperature sensors can also be used when an integrated die temperature sensor is not available. However, for local temperature sensors, sensor location is an important design consideration. The image below shows some options for placing local temperature sensors: positions a, b, and c.

High-performance processor temperature monitoring by placing sensors

• Location a. The sensor is located in the center hole of the microprocessor heatsink very close to the die. The heatsink can be clipped onto the processor or attached to the top of the processor with epoxy. Temperature sensors in this location typically require longer leads, and as thermal conductivity from the heat sink to the microprocessor gradually degrades, the sensor data will become incorrect.

• Position b. Another potential location for sensors is in the cavity below the processor socket, where assembly is very straightforward. Since the sensor is isolated from the airflow, ambient temperature has minimal effect on the sensor reading. Also, if the heatsink is detached from the processor, the sensor will show an increase in processor temperature. Nonetheless, with this sensor placement, the temperature difference between the sensor and the processor can be between 5°C and 10°C.

• Location c. The sensor can be mounted on a circuit board next to the microprocessor unit (MPU). While this type of installation is easy to implement, the correlation between sensor temperature and MPU temperature is much weaker.

Device Recommendations

Footprint size is a factor to consider when choosing a local temperature sensor. The TMP112 is packaged in a 1.6mm x 1.6mm package and can be used close to the processor. The 0.5°C accuracy of the TMP112 device maximizes performance compared to the typical 5°C to 20°C accuracy of temperature sensors integrated inside the processor.