Calculation of Junction Temperature

Question

I have a PCB and trying to measure the maximum junction temperatures of the major components in the PCB which in full working mode. For example like Input Diodes, Buck Regulator ICs, Aluminium Electrolytic capacitors and other Discretes.

I have placed my PCB inside the thermal chamber at 95degC and made the board full operational and measured the temperatures of the components by placing the thermocouples (K-Type) on top of the ICs, diodes and other components as mentioned above.

My questions :

1. I measured the temperatures of the components on their surface and found to be in the range of 110degC to 120degC. But this is the Surface temperature of those components, right? How can I practically measuring the junction temperature of those components so as to find what margin I have ? Is it something like I add 20% to the measured surface temperature value to arrive at the maximum junction temperature of the components? Or is there another way to measure the junction temperature of the components?

2. What are the other important things that I need to calculate or note during this thermal stress testing?

Answer 1

You really have to know something about the thermal resistance between the case and the junction to make an estimate. And you need to know the dissipation in order to multiply that by the thermal resistance to get the rise. If the dissipation is not fairly constant in a time span of milliseconds you may also have to know something about the dynamic behavior.

Unless your measurements were made very, very carefully with an exceptionally tiny thermocouple they are likely be well on the optimistic side because of heat loss down the wires which will tend to make the reading closer to your ambient in the oven. Other types of sensors tend to be worse again.

IR measurement may help, especially if you can get an accurate reading of the PCB itself with thermocouple(s) and use the IR for the deltas.

If there are particular components you are especially concerned about, it may be worthwhile to measure them individually in mocked up situations, for which various techniques are available such as measuring Vf of any available junction (which can be calibrated) or measuring the resistance of a copper coil.

Answer 2

Try to check the Thermal Resistance Junction to Ambient (K/W)

for istance for 1N4007 is 100 K/W, considering a voltage drop of 1V and a current of 0.1A we will have a 0.1W therefore the junction temperature will be 10K higher than ambient.

the formula is Temperaturejunction=ThResistance*Power+Ambient temperature.

During thermal stress you could see if any offset or gain of the circuit is changing. Also if you have some hysteresis effect.

Answer 3

An approximately way to measure the temperature would be to glue an NTC sensor with really thin wires on top of the component (for minimum thermal transportation) and then you know the surface temp of the component and the calculate backwards according to datasheet, however this will not be accurate due to the number in the datasheet is not for the same condition as in your case, however it will give an indication.

The same can be done with the use of a thermal camera and by knowing what settings you should have according to the emission of the surface. You could also paint on a known emission, but that will add some extra thermal way, hence giving you a conservative answer, but it will still not be 100% accurate because the Junction to Ambient in datasheet is not the same as in your application.

Answer 4

The gold-standard method is to measure it inside the device itself, with a pulsed measurement.

This is a rather special case though.

Typical case for a MOSFET, is to first of all quantify the body diode Vf(T) characteristic. Apply your forward-bias test, then momentarily cut power, apply reverse current, wait some 10 to 100s µs for it to settle (but mind that the die is cooling down to package temp, and so on, during this time!), then resume the test.

You can also special-order test components from the manufacturer. How do they know their stuff works? Of course they know. Examples are plastic-packaged devices either with the package demolded above or near the die, or drilled in nearby (say for a nearby inner-tab temperature reading); or hermetically sealed devices with the cap off; or modules/packs without the silicone gorilla snot filling; etc.

Don't forget to use thermal paste on the temp sensor (or thermal epoxy if you don't need to reuse it). The sensor body/leads are not perfect insulators by a long shot! Also, thin sensor leads are preferable.

See also: https://www.nxp.com/files-static/rf_if/doc/app_note/AN1040.pdf Fig. B1 (p. 18) is particularly relevant.