Thermocouple Nodules, Cold Junctions & Integration Opportunities

When measurement instruments reach their specified performance limits, one comes to a gray zone separating reality from fantasy. Thanks to sensors, this zone is widening.

This region is approached in digital multimeters (DMMs) as the quantity being measured approaches zero. One major DMM company forces a zero reading on the display when the measurement is within a certain small band around zero. A similar issue exists in data acquisition (DAQ) systems, which are essentially multichannel DMMs.

The encroaching fantasy is being exacerbated by sensors, particularly thermocouples (TCs), which are conceptually simple sensors. When two dissimilar metals are joined, a voltage appears across their junction that is approximately proportional to temperature. Two wires — one chromel and one alumel, for instance — can be spot welded together at the sensing end by process chemists (who like to use TCs) and connected to the DAQ screwdown terminals, shown in orange below.

However, the DAQ-end connections also form thermocouples and generate voltages in series with the intended sensor junction. These additional undesired sources are called cold junctions. If the two wire-to-terminal connections are at the same temperature, the cold-junction voltages cancel each other. Therefore, the goal is to have the same temperature for both connections on the terminal strip. This is not an electronic design problem as such, but is one of the many opportunities for engineers to exercise versatility.

If the cold-junction connections are near each other and somewhat enclosed, they will be within 0.5°C of each other and typically within 0.3°C. Consequently, one would think this would set the limit on the accuracy specification for DAQ products. In reality, it does, but process chemists want to be able to resolve to 0.1°C or better. With filtering, the DAQ electronics can be designed to provide such resolution.

The trouble arises when the chemist wants to see a steady reading to within 0.1°C. Cold-junction temperatures vary with air movement and cannot be maintained to this resolution, yet some DAQ suppliers offer it. How is it achieved? A similar kind of zero-scale sophistry is applied to DMMs. It is not so illogical to consider a measurement so close to zero that no better can be achieved as a measured zero value. However, a constant non-zero reading can give the illusion of a solid measured value.

The zero-scale problem with cold junctions presents a semiconductor product opportunity. More significantly, an excellent opportunity for integrated analog on silicon exists to minimize the cold-junction problem by moving it to the silicon chip. One of the usually undesirable characteristics of monolithic silicon for circuits is its high thermal conductivity. In the case of cold-junction compensation, it might be possible to maintain to within 0.1°C the on-chip connections of two wires. One method for doing this is to take fixed lengths of TC wire — long enough to be out of the heat — and affix a processing chip with a digital or noise-insensitive processed analog output. TC wires are thin enough to be attached (but not die bonded) to large pads on a chip.

Alternatively, a tiny leadless chip package can be given an unusual pin-out to accommodate the TC wires. Two power pins and one (digital) or two (differential analog) output pins complete the chip connections, and the small processing granule is enclosed in a way suitable for the general environment — perhaps in a blob of epoxy. Such nodular TCs would reduce TC wire length and cost and could sell by the millions. True 0.1°C resolution (and maybe accuracy) could be achieved by TC nodules. The process chemists would benefit. DAQ electronics would be simplified. Sensor costs would drop, since TC wire is not cheap, and the market would reward the IC company offering this.

The manufacturing process for packaging TC nodules is the central design issue. For a greater challenge, the IC design goal could include elimination of the power pins by deriving chip micropower from the sensor itself, making use of the cold junctions to supply power. This would increase the analog integration needed on the IC.

Creative IC developers and vendors should be able to find the optimal combination of features for such a product (including TC linearization) so that they appeal broadly to process chemists, chemical engineers, and electronics engineers.

Related posts:

16 comments on “Thermocouple Nodules, Cold Junctions & Integration Opportunities

  1. eafpres
    May 17, 2013

    Hi Dennis–nice idea, and suitable for a lot of applications.  On the other hand, TCs are pretty rugged and people stick them into all sorts of things like kilns and furnaces etc.  Generally the IC you are talking about might live up to 60C, or maybe 85C in some cases, or even 100C for some really rugged stuff.

    Any thoughts on what to do beyond that?  

    A challenge with temperature measurements in particular is the heat flux along the TC wires or along the signal wires in your design.  Since they exit the “box” we draw around the process, under some conditions a thermal gradient exists and can distort the precise measurement you are going for.  I'm thinking that in biologicial systems the temperature limits are modest, and you probably could bring out signal wires that are much finer than TC wires, and so reduce any distortion in the measurement.

    I would guess this has great applications in DNA labs and automated sequencers, for example?

  2. Davidled
    May 17, 2013

    In my experience, thermalcouples are used in the automotive industry.  Wire length is much longer than 3 feet to connect the wire mounted near oxygen sensor all the way to measurement device located in outside chamber. I thought that if wire length could be minimized, measurement distortion can be reduced, but, there is no much difference for performance viewpoint in the automotive application level.

    Similarly, I am wondering how much finer wire provides the better performance in above application.

  3. D Feucht
    May 18, 2013

    The problem was addressed, albeit briefly, in the article:

    “One method for doing this is to take fixed lengths of TC wire — long enough to be out of the heat …”

    If the TC wire is extended beyond the region of measured temperature to a somewhat ambient place, then ordinary temperature bounds for IC design should apply. The wire would not be as long as in the typical case nowadays where it has to extend all the way to the DAS. Instead, it need only be removed far enough from the measured temperature to accommodate the IC nodule. From it can be output cheap wire back to the (cheaper) DAS.

    I wouldn't know about DNA lab applications, but for process chemists – and there are many of them – this would, I suspect, be an improvement.


  4. D Feucht
    May 18, 2013

    “I am wondering how much finer wire provides the better performance in above application.”

    The TC wire need not be special, or differ from what is presently used, though this might be a design detail to be investigated in regard to attachment to the IC.

    I might add that TCs are used because they are cheap and can be made with a spot welder and a roll of TC wire. However, they are one of the worst temperature sensors from a measurement electronics standpoint – hence an opportunity for improvement by a semiconductor company or a big distributor such as Omega Engineering.


  5. Davidled
    May 19, 2013

    Auto get recall all vehicle on the earth, since testing related to temperature is incorrect by TC. Unfortunately, TCS is one of popular temperature sensors in the Auto Chamber Lab. Auto company might call manger who orders TCs and measurement tool to get all refund from vendor. I think that there is measurement standard spec for each sensor before they sell it on the market. I am curious for electronic measurement standpoint that you comments.

  6. eafpres
    May 20, 2013

    @DaeJ–I think you are overstating the “problem” with thermocouples.  For qualification testing in Automotive, if good/no good depends on better than 1 degree resolution then you are putting a product on the market at high risk.  

    Most automotive parts outside the engine compartment are supposed to be good to at least 85°C.  Are you suggesting that bad TCs result in parts going into vehicles that fail and cause recalls?  From my direct expereince, that seems like a stretch.

  7. bjcoppa
    May 20, 2013

    Having a good fundamental understanding of thermocouples is critical for work in clean rooms for the fabrication of microchips. Nearly all process tools have some sort of Tc for temp measurement and their accuracy often dictates the performance of a particular process. Occassionally, the wrong specified type is inserted into a system such as a non-oxygen compatible type which leads to quick degradation and failure. 

    May 21, 2013

    One other improvement thought is to think of the heat-sink capability of a thermocouple. Though in most cases the thermocouple is very small compared to the object being measured, but if the masses are similar, then the thermocouple can act like a heatsink with the leads being fins. Furthermore, if any air flow exists, the leads can further act as heat sinks.

  9. Brad Albing
    May 21, 2013

    I would hope that the engineers and technicians working in and around the clean room would have a complete and thorough understanding of thermocouples – else the company fabricating the microchips will be in big trouble in a hurry.

  10. TheMeasurementBlues
    May 21, 2013

    Brad, you'd be surprised how few people even get the basic concept of a thermocouple. It's a case of how wrong information, repeated enough, becomes accepted. Simple, but Misunderstood.

  11. Dirceu
    May 22, 2013

       I worry if developers involved with temperature related instruments, especially the low cost ones, actually do all that end to end error propagation analysis (sensitivity, mean square, derivatives,…) – from the sensor, conditioning circuits, ADC (o C/bit) and the impact on the final equation (fixed point Q format or floating). Or a more direct way: Post-design statistical measures in order to attain the desired specifications.

  12. Brad Albing
    May 23, 2013

    DK – that's related to Heisenberg, isn't it? Or maybe it's the observer effect. You can't measure something without changing what you thought you just measured.

  13. Brad Albing
    May 23, 2013

    Yep – I knew that – but you're right – lots of folks get that one wrong. That'd be a good question to give on a job interview.

  14. Brad Albing
    May 28, 2013

    Dirceu – did you mean that you are concerned that many engineers designing products don't go thru this sort of analysis?

  15. Dirceu
    May 28, 2013

    Good question,

        when you launch a measuring device on the market, regulatory agencies require that Certificate of Calibration includes the measurement uncertainty or other statistical values. But it does not provide much information on how the product was engineered from the beginning, in order to reach the final specifications. I'm referring to the mathematical tools (Total Differential, Taylor Series, Mean Square Error, etc. ..) used to evaluate the error propagation, select components (eg tolerance, ADC resolution,…) and write the software (int32, int64, …). I'm not talking here about significant figures – which is something that physicists like, but engineers seem do not to pay much attention. For example, some engineering students often ask me why the LCD is showing 1.22, 1.24, 1.26, … I wonder if the classic courses are teaching this kind of thing to our future professionals. I also think there is little literature specifically for the filed of electronics.

  16. Brad Albing
    May 28, 2013

    I hadn't thought about the implications of selling a piece of test equipment with regard to the possibility of needing to get it certified by a testing agency. Good point. But beyond that, good observations on techniques to evaluate error sources as part of the design process.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.