Apparently, there is a problem to find small digital temperature sensors on the market that are resembling state-of-the-art technology. Thereby we are not talking about the sensor as such, but something like that:
This is a kind of temperature sensor that can be screwed into some device where it can be used for measuring the temperature of a fluid, for instance.
Requirements
What does one want from such a sensor? Of course, this does heavily depend on the application. A few important general points to think about are the following:
- Size
- Precision
- Mechanical ruggedness
- Temperature range
- A simple but effective interface
As for the size, usually there can be said: The smaller, the better – which is particularly important for a temperature sensor. This is because the more bulky and heavy it gets, the less responsive it is getting. I.e. it takes more time until the material resp. the actual temperature sensor adopts to a changing temperature of the fluid, the more material is there. This is usually not wanted.
Precision is two-fold and can be divided into an absolute precision as well as the resolution of individual measurements. A resolution of say one-hundredth of a Kelvin does not mean that the sensor is working at a precision of 0.01K. Usually the absolute precision is less – for instance 0.1K. A resolution which is higher than the absolute precision makes well sense though. The point is that it allows for a more precise differentiation. That is, while the absolute precision can be roughly considered as some sort of offset that applies to a more coarse grain, a higher resolution lets you know what is happening on a finer grain. That is, assuming an absolute precision of +/-0.1K and a resolution of 0.01K, two readings of say 50.00°C at time a) and 50.05°C at time b) correspond with actual temperatures of 49.90°C to 51.10°C for a) and 49.95°C to 51.15°C for b). However, although there is a certain amount of uncertainty about the actual temperature, one would know that the measurement at time b) is 0.05K above the measurement at time a). Of course, this only applies for a single exemplar of a sensor.
The temperature range is another important aspect es well. It certainly requires different technologies when one needs to measure a common range from say a few degrees Celsius minus to the boiling point of water, or the exhaust temperature of some jet turbine.
The interface is important as well.
Why a Digital Interface?
A traditional electronic temperature sensor usually is rather simple and is working on an analog base. A good example are Platin-based temperature sensors that are based on measuring the resistance of a wire primarily consisting of the precious metal Platin. Platin is showing a very defined and almost linear behavior of its resistance depending on temperature. So the final temperature measurement boils down to what the typical multimeter is doing when one is measuring the resistance of some wire, for instance. In practice, one could send some defined current through the actual sensor and use an analog-to-digital converter to derive the temperature from it. There are a few pitfalls:
- The cabling is affecting the measurement as well.
- External noise can be coupled into the cabling, hence compromising the measurement.
The longer the cabling, the larger the problems.
Another sensing technology is the use of so-called thermocouples that produce a voltage on their own depending on the temperature. These devices are usually not that precise, but usually they can be used for a rather large range and rather high temperatures. In general, they are facing the cabling issues as well as the sensor signal is transported in an analog fashion.
A digital temperature sensor that is limited to a rather small space does remove those cabling issues from the design. A digital temperature sensor does include all the analog circuitry. In theory, a digital temperature sensor could also employ technologies like Platin- or Thermocouple-based sensing. However, as of 2025 this seems not feasible to be manufactured on a chip-level. What is working well here is the use of silicon diodes, which do also show an excellent dependency from temperature with their characteristics. Once the temperature sample is present within such a digital sensor, it can be digitally transported through some interface and is less prone for corruption. Though, there are a few drawbacks here as well:
- The temperature range is limited to what the whole circuitry can withstand. This usually limits their application for -50°C to 150°C.
- There is more active analog and digital circuitry close to where the temperature is to be measured. As this circuitry requires power to operate and hence dissipates heat resp. causes self-heating of the sensor, this is is compromising the measurement as well – particularly in low-temperature measurements.
The chosen Temperature Sensor Device
The device that has been chosen here is the Texas Instruments TMP119. As of 2025 this is a state-of-the-art digital temperature sensor featuring an absolute precision of +/-0.2°C over the complete temperature range of -55°C to +150°C. For certain sub-ranges this precision is even higher. It does also feature a rather simple I2C interface. It comes in a very tiny BGA package with a footprint size of less than 1mm x 1.5mm.
A Sensor-carrying PCB
In order to make use of the sensor device, it needs to be soldered to somewhere. There has been made a miniature PCB that is just taking up the sensor and two decoupling capacitors for the supply voltage of the device. Since the actual PCB is very small (6mm long and 1.8mm wide) and can hardly be handled that way during the soldering process, there has been made a larger PCB with 5 small PCBs:
The PCB is made of so-called TG170 base material, hence allowing the whole sensor to be operated at the maximum the TMP119 can achieve (150°C). However, there might apply other limits later, such as the maximum temperature the cabling or the potting compound can withstand.
The carrier PCB has got four pads where cables for ground/supply as well as the two I2C lines have to be soldered to:
The cable shown here is a very slim 3-wire plus shielding one. The diameter of the cable that has been used is even slightly larger than the sensor PCB itself.
Housing the Sensor PCB
Such a sensor PCB can then be placed into it’s desired mechanical housing – e.g. some sort of a screw. The following 3D-view is showing the actual relations with a M5-based screw-like housing.
It can then be potted by using some suitable epoxy resin. The actual result is looking like that:
Limitations of the I2C Interface
I2C is intended for low-speed data transmissions across a few decimeters. Though, low-speed is relative. Considering a typical process temperature measurement, a sampling rate of a few Hertz is usually sufficient. There has been made some analysis, and the total transmission time to fetch a sample from the TMP119 at an I2C frequency of approx. 111kHz is around 880µs. At 390kHz it is around 240µs. So this would allow sample rates of 1kHz and above, which is far more than needed (leaving aside internal limitations of the sensor). However, it allows to operate tens of sensors through a single I2C host at reasonable sample rates.
As for the cable length, this does virtually boil down to a trade off between I2C operating frequency and length. The important point hereby is the capacitance of the cable. For the cable that has been used here there has been measured a capacitance of around 180pF per meter from a wire to the shield (which has been used as ground). There has been made some extensive analysis that is covering cable length, I2C frequency, operating voltage, and also the PullUps required on the I2C lines. Assuming there is just a single device on the particular I2C bus (resp. two devices – the TMP119 and the host), the following statements can be made:
- For 2m cable length the operation is safe at an I2C frequency of around 400kHz. Pullups should be around 1kOhm-1.5kOhm.
- For 10m cable length the I2C frequency should be limited to around 100kHz and the Pullups should be around 1kOhm.
- Even longer cable lengths seem feasible, although they call for a further reduction of the frequency.
- There is almost no effect of the actual supply voltage on stability.
Considering the pictures of the electronics shown above, one might ask where the PullUps are. Well, they have been banned from the sensor-end and are to be located at the host. There are two primary reason for that:
- The resistors require space – something that is precious close to the sensor.
- The resistors create heat. Although not much, but you don’t want a heater besides a temperature sensor… This is also part of a guideline set up by Texas Instruments with regard to a high-precision temperature measurement.
Not dedicated to temperature sensing, but there is some rumor about where to place the I2C Pullups – at the host or the device end? Mostly these are placed at the device end. In fact, it does not matter where they are located. Those are Pullups – not transmission line termination resistors.
Yet another point to note is that there should be only a single sensor device per I2C host. Actually, I2C is a multi-drop bus that allows the connection of many devices. However, when there are involved long cables this is going to become a mess regarding signal integrity. Furthermore, although the TMP119 allows a configuration of four different I2C addresses hence allowing to operate up to four of them at one I2C bus, the design used here does fix the address to a single value. So it is not possible anyway to attach more than one temperature sensor to an I2C bus. In case multiple temperature sensors are to be attached to some sort of micro controller, an I2C multiplexer has to be used. For a particular design, a Texas Instruments TCA9548A has been used which allows the attachment of up to 8 of these temperature sensors to some host computer.
As for the supply voltage, the TMP119 supports a rather broad range of 1.8V-5.5V. However, it is advised to set the voltage to the lowest possible value in order to minimize self-heating effects of the sensor. In a particular application the voltage has been set to approx. 1.9V. 1.9V has been chosen in order to ensure that the lower limit is met – especially in wake of long an thin cables that might incorporate a slight drop in voltage. Furthermore, these 1.9V have been created through a linear voltage regulator, hence delivering superior quality of the supply voltage with regard to ripple and noise in order to achieve the best possible precision for the measurement.
More rugged Alternatives to I2C
Apart from the fact that practical I2C implementations with cable lengths of several meters are well possible and are working stable, it should be noted that this certainly is not the way to go for a really rugged system. Here, one would place some conversion circuitry very close to the sensor – so that the I2C wires just span a few centimeters. This circuitry would then convert to some other communication system such as RS422/RS485 or similar or some sort of glas fiber or possibly even radio frequency transmitters (Bluetooth, WiFi, etc.), and might also include additional protection by means of error correction etc. However, for “normal” environments (whatever that means…) such an I2C connection across a few meters and high-quality cabling has been proven to work absolutely flawless.
Actual Performance
The M5-Screw-like sensor as shown in the pictures above has shown some weakness in that sense, that it does conduct a little bit too much heat away from the point of interest (or vice versa). Depending on the difference between the ambient and medium temperature, this is introducing an offset of a few K. That is, the sensor should be a bit longer, hence immerse further into the liquid that is being measured. This is a matter of the mechanical housing, however. In the pictures below there is also shown another mechanical incarnation virtually consisting just of a thin stainless steel pipe with 3mm outer diameter and 0.1mm wall thickness. The situation here is much much better. What might also contribute a bit to heat extraction/insertion is the fact that the copper cables – although rather thin already – act as a heat sink as well. So for ultra-high precision it might make sense to use extraordinary thin wires for the last millimeters. However, that has not been analyzed in detail yet.
As for the stability of the temperature readings via the TMP119 it can be stated that it is remarkable. In a particular application the built-in averaging of the TMP119 has been set to 32 (that is, for each reported sample there are done 32 measurements internally). This is providing a stable reading at a resolution of 0.01K.
A few Application Examples
Here is an example of the M5-type sensor shown above screwed into a PTFE (aka Teflon) block including hose-nipples. This setup has been proven to be somewhat suboptimal because there is too much of heat exchange between the ambient environment and the liquid being sensed.
Here is an improved system based on essentially the same PTFE block, but with a much more slim sensor housed into a simple pipe with an outer diameter of 3mm and 0.1mm wall thickness. It is fitting tight into the block in order to guarantee a leak-free operation up to a certain pressure.
Finally, this is a typical rod thermometer where a sensor housing is made from a lathed piece of stainless steel and tightly pressed into a longer rod made from PTFE.
In all those examples, the PTFE serves as sort of thermal insulator. It might not be the best material to do so, but it is better than any sort of metal.