A common thermal debug and characterization technique is the use of one or more thermocouples or thermistors placed in the system to get real-time temperature measurements. This can give much more insight to the thermal performance and limitations in the system. There are some precautions for this to ensure the results gathered are accurate and properly interpret thermal operation.
Xilinx does not recommend placing a thermocouple between the heatsink and the device to perform case measurements. This can often provide erroneous readings, compromise the performance of the system, and in some cases cause damage to the device. It is suggested instead to first measure at multiple points on the external base and/or fins of the heatsink to gauge thermal conduction to the heatsink. If enough temperature variance is seen between the SYSMON reading and heatsink measurement, investigate the heatsink contact to the device and/or heatsink construction as a general first course of action. If no issues are found there and a device case measurement is still desired, the best method to do so is to drill a hole through the heatsink to the area of the device that is intended to be probed.
Ensure there is very good contact between the thermocouple and the device through the given hole. This can be difficult, and poor contact to the device often leads to erroneous readings. Due to this reason as well as the fact that the thermal system must be modified/damaged to do this measurement, it is suggested to only be done as a last resort as in most cases this is not a necessary step to understand thermal issues in the system. When adhering the thermocouple to the measurement point, as repeatably mentioned, good contact to the measuring surface must be maintained. Improper application of thermal epoxy, such as when the thermocouple does not have direct contact to the case or excessive epoxy is applied, is shown to cause several degrees of difference to actual surface measurement.
Using thermal tape is a more common method to fix the thermocouple to the measurement surface because it is less permanent. However, the common issue here again is contact. If the tape is not taut enough and/or has a good surface to affix to, the thermocouple can lift, causing reduced readings. Coupling these common issues with the inherent uncertainties of the thermocouple measurement itself can often lead to misinterpretation of results. Thus, these and other possible sources of error should be eliminated before making conclusions from the data.