1
\$\begingroup\$

Question

Is there a temperature sensor (or else) which allows for the measurement of a copper pour on a PCB (and therefor the approximate temperature of a component)?

I want to read the temperature via an ADC (due to firmware complexity) to provide telemetry and shut the system down when the temperature reaches a certain threshold. It needs to be consistent and quick enough to be useful, and I realize accuracy will be limited.

Background

I would like to know how hot my MOSFETs are getting. They use SMD packages, and im cooling them through the PCB with a heatsink and a fan on the other side.

My main issue is that it seems to me that most SMD temperature sensors are geared for ambient temperature measurement. I cannot connect any of their pins electrically (and thus thermally via the copper pour) to the hot part, and it would be too expensive for my application to add thermal jumpers (not that I know for sure that a strong thermal connection to the pads would yield a good response time)

It feels like my best bet could be to just place the temperature sensor (example) really close to each MOSFET's tab, and surround it in a copper pour. Perhaps doing something fancy like a comb pattern between the MOSFET's pour and the pour on the sensor's pins to increase surface area

The fact that this seems to not have a lot of questions about it (that I could find), or products that do this (that I could find) makes me think my approach may be wrong, or that im searching in the wrong place. Perhaps with a heatsink, the temperature differential would be slow enough to not need a strong thermal link?

\$\endgroup\$
2
  • \$\begingroup\$ What response time are you after? If you are doing a pulsed power application, you may be better served by live calculating or assuming SOA, rather than trying to measure it directly. (Which can still use a sensor, but inferring only a later part of the thermal system; the rest needs to be characterized i.e. RC equivalent driven by a power calculation.) \$\endgroup\$ May 14, 2023 at 20:29
  • 1
    \$\begingroup\$ Im already using a current shunt to do cycle by cycle current limiting. With that, I am safe from something like an extreme shot circuit or a very powerful load. But the cycle by cycle current protection kicks in at 20% higher than the maximum design current, so the temperature sensors are to protect from such a case where the user stays in that region. I think something around 200-500 ms response time is plenty. I can always set the trip temperature lower to account for a slow response time, which is why I dont have a strict response time requirement, but I still need to know it. \$\endgroup\$
    – Anas Malas
    May 18, 2023 at 14:17

2 Answers 2

0
\$\begingroup\$

An SMD temperature sensor is going to measure the PCB temperature primarily, so I think that would server your purpose with reasonable layout. Keep in mind the way the heat flows and possible resulting thermal gradients. You could use an easy semiconductor sensor or a thermistor or RTD, for example. It won't be instant, of course, there will be a lag from the die to the case and the case to the PCB where the sensor is. Temperature sensors are cheap and you can pepper the board with them if it helps.

Another consideration is to keep noise from the switching MOSFETs from affecting the temperature reading.

\$\endgroup\$
3
  • \$\begingroup\$ Before I accept your answer, is there a way to estimate the sensor's response time? The datasheets include "air to hot stirred oil" times, but that needs the PCB itself to heat up whereas in my application the PCB will be "hot already" \$\endgroup\$
    – Anas Malas
    May 18, 2023 at 14:20
  • \$\begingroup\$ @AnasMalas I think it will track changes in the PCB fairly closely, in other words the time constant will be dominated by the heat capacity of the PCB and components and the time constant of the PCB to element will be much much less. To figure the latter out would require modeling the heat losses as thermal time constant equation. Frankly, I think it will be more accurate to just try it with a very small (polyimide stick-on) thermocouple to compare. But the calculation will help tell you if it's worth measuring. \$\endgroup\$ May 18, 2023 at 16:49
  • \$\begingroup\$ To give an example, it would take many minutes for a PCB to stabilize (maybe 30) to within 9x% of the change in one product I designed, so maybe a time constant of 10 minutes. The time constant of the sensor on a PCB will be more like a fraction of a second. \$\endgroup\$ May 18, 2023 at 16:51
0
\$\begingroup\$

Yes, that is doable. Beware that the time constant is somewhere between the thermal diffusivity to that location, and the thermal time constant of the components and board (or part of it).

As example, I made this transient limiter,

Thermistor surrounded by copper pour, EDA view

Thermistor surrounded by copper pour, photo

TVS diodes are used to dissipate power during an overload condition, semi-continuously. The dissipation and time constant are such that, at maximum ratings, the surface temperature of the diodes may exceed ratings (over 150°C), but only briefly; within a few seconds at most, the thermistor reads the increase and shuts off the circuit.

Thermal diffusivity is relevant in this case. As a thermal pulse, the diodes heat the local area, and heat spreads out as distance \$\propto \sqrt{t}\$. FR-4 is in the 0.1 to 1 mm2/s range, up to 10 or so when heavy copper is present on multiple layers. The distance here is about 3mm from SMC pad to 0805 thermistor chip, so it takes on the order of a second to begin heating up significantly, and tens of seconds to "soak" in.

During that time, heat is flowing into the surrounding material, making a thermal resistance divider, but not just resistance but the effective capacitance (heat capacity) of the materials too, hence the time dependence.

Eventually (well, not in this case, but in yours perhaps), the rest of the board reaches equilibrium, and thermistor reads a temperature given by the thermal resistances. Which means you'll always have higher TJ because the transistors are generating heat in the first place, and there is some drop from there to the surroundings (and thermistor). And so on, as temperature drops across the board (due to removal by convection or conduction).

A similar argument gives a steady state thermal spreading distance of an inch or two for typical (say 2 layer 2oz copper) PCB, meaning something like a D(2)PAK is good for a watt or two in typical conditions. With thermal pad to a heat spreader/sink, more like 10W is feasible. Obviously in those conditions, with more heat sunk across board materials, the error (between TJ and Tthermistor) will increase.

In a switching application, be careful of switching noise (heavily filter the thermistor terminal(s) -- don't neglect the common mode!), and also be careful of breakdown voltage if at higher voltages.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.