I'm currently trying to integrate a type K thermocouple into an electronics project. I intend on interfacing this thermocouple with the MAX31855 cold-junction compensated thermocouple-to-digital converter IC.

I am not an expert on thermocouples and the Seebeck effect, so I'm somewhat curious as to how I should actually land the thermocouple onto the PCB. Assume I'm using loose thermocouple wires.


  • Can I use any type of terminal (say a simple through hole screw terminal) to land the wires onto the PCB? Or will these new junctions introduce errors?

  • Is the termination method dependent on the thermocouple type? E.g. use this terminal for type K, use this other terminal for type J, etc.

  • \$\begingroup\$ You should read up a bit about how these work, in short, they rely on being different materials. If you change materials along the path, things change. If they are too much for your plans or not depends on what you want. Do calculations and then decide if its fine. It likely is. \$\endgroup\$
    – PlasmaHH
    Dec 2, 2016 at 15:58

6 Answers 6


Using a terminal block made with ordinary materials is quite sufficient for a relatively modest accuracy system as you're aiming for.

The cold junction compensation depends on the cold junction sensor (in this case, the chip itself) being at the same temperature as the two junctions where the thermocouple wire transitions to copper. In other words, all three should be isothermal- so you want to minimize gradients caused by dissipation on the PCB and by gradients caused by heat flowing down the wires. You can help this along greatly with ground planes or at least pours and by keeping anything that dissipates a lot of heat well away from the T/C block. Keep air currents away from the terminal block too. Of course you will put the chip as close as practical to the terminal block, physically as well as thermally.

There is no great difference between most sensors as far as this goes as most thermocouples are fairly linear (a couple percent) so 1°C error at the cold junction is around 1°C error in the temperature reading.

If the connection is out hanging in the breeze or is at an elevated (for example) temperature, it's better to use connectors that are made of thermocouple materials, and this is typically done for panel-mount connectors and inline connectors. They are usually color coded. In North America we use the ISA color codes, and type K (Chromel-Alumel) is yellow, type J (Iron-Constantan) is black. You could, for example, have a bulkhead K connector and connect that inside an enclosure to the PCB. You MUST use the proper thermocouple extension wire on the inside as well as outside in this example, and it MUST be connected the correct way (if you swap the polarity the error is actually doubled). Keep in mind that red = negative in North America T/C color codes.


Any change in material will introduce an error on the signal. However if the changes are the same (in terms of both material and temperature) on both terminals then those errors will cancel out. So you can use just about any method you like as long as you keep things symmetrical.

Two things to watch out for - Firstly the temperature that is measured is the difference between the thermocouple junction and the temperature where the two wires become the same material, normally the copper trace on your PCB or the terminal block. This means that if you use screw terminals then your cold junction temperature sensor needs to be at the same temperature as those screw terminals. Any difference will be reflected as an error in your measurement.

Secondly be careful of temperature gradients on your PCB e.g. if one terminal is closer to the power supply or further away from some source of airflow it could be a tiny bit warmer which will impact the end result. If the terminals are on the outside of the box and the cold junction reference is inside getting heated by a CPU then you're going to get a big error.

  • 1
    \$\begingroup\$ So this is what I was curious about. Referring to figure 4 here (ohio.edu/people/bayless/seniorlab/thermocouple.pdf), you could see that the added junctions would be symmetric and cancel out. However, are these junction voltages dependent on what materials are be connected? In the case of figure 4, the fact that the thermocouple wires are of different material essentially makes one junction (copper to copper) disappear, thus making it no longer symmetric. Thoughts on this? \$\endgroup\$
    – Izzo
    Dec 2, 2016 at 16:21
  • \$\begingroup\$ In figure 4 the the Cu-Cu junction has no material change and so has no impact, that leaves you with two Cu - C junctions in opposite directions, J1 and J2 with the associated voltages V1 and V2. The final output voltage will depend on the temperature difference between J1 and J2. So you are still measuring the difference between the end of the wire and the junction where they first become the same material. \$\endgroup\$
    – Andrew
    Dec 2, 2016 at 16:36

You must change metals at some point, to connect to the 31855.

I order to do cold junction compensation, the 31855 must know the temperature of this point. Any difference between the temperature of this point, and the point that the 31855 senses, will appear as an error in the temperature reading.

The 31855 measures its die temperature as the 'cold junction'. This means that as long as the 31855 and your 'thermocouple wires to copper' junction are at the same temperature, all will be good. This generally means make the junction as close to the IC as possible, and don't have hot components producing thermal gradients anywhere near the IC.


The quick answer: Yes.

The long answer: thermocouples use the difference between two metals at temperatures to generate a voltage that is representative of the temperature. This means that all of the cable, from the thermocouple to the sensor, needs to be this special combination of metals to give you the temperature measurement accurately. You can buy thermal couple wire for this purpose.

That being said, I have used systems where we had a K-type thermocouple, going into a standard header on a PCB, through standard tracks before going into the sensor, and that could give us a value. The issue with using something other than a proper header is the accuracy of the reading. We knew roughly what the offset was (appeared to be an offset of around +4C at the temperatures we were interested in) and just adjusted accordingly. But that was the advantage of being in our own test cells, where we had other sensors to compare the value to.

So, while you should use proper connectors, you can get away without doing so.

  • \$\begingroup\$ What do you mean by: the K-type thermocouple went into a standard header on the PCB, then through standard tracks, then into the sensor. The thermocouple is surely the sensor. Did you mean to say amplifier or ADC where you wrote "sensor"? And the "offset" you discuss was an offset between what and what? And what do you mean by "being in a test cell"? \$\endgroup\$
    – gwideman
    Sep 28, 2021 at 13:33

Each time you change to a different metal thermocouple wire->connector metal->solder->copper, you are adding another thermocouple junction. Fortunately, you always add a pair of junctions at a time- one in the wire going to the thermocouple and one in the wire coming back- and as long as both of the junctions are at the same temperature, the two will cancel out.

The overall effect will be the same as having just two junctions: the one at the end of the thermocouple (hot) and the one at the junction between the two thermocouple wires and the connector (cold).

Ideally you should do your cold junction compensation by measuring the temperature of the cold junction (ie the connector). If the chip that that you are using is doing the CJC, you should place it as close as possible to the connector, and as far away as possible from any heat sources.


Lets put some numbers on the PCB design. Standard copper foil is 70 degree Cent per watt per square. Thus 1cm or 1mm or 100micron or 10cm squares, with one watt injected uniformly along one edge and that heat flowing ONLY to the opposite edge, will have 70 degree Cent temperature gradient. Should you have a 100 millWatt MCU (its usually busy handling USB data for a bunch of other ICs) needed to dump that 100mW to a metal bolt 3 squares away, the temperature gradient will be 0.1W * 3 * 70 degC/watt = 21 degrees. C.

Get a quadrille pad, and start drawing heat sources and heat exits, and slits in the PCB to guide the heat flows. And consider using VDD and GND planes to thermally move the heat around.

If you perform a finite-element model, using a grid of resistors in SPICE, with low value Rs to model the copper, and Rs 100X higher to model the FR-4, you'll get an idea how much overlap of planes is needed to dump half the heat in the hotter plane into the cooler plane.

  • \$\begingroup\$ What does this have to do with the poster's question? \$\endgroup\$
    – gwideman
    Sep 28, 2021 at 13:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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