Using BS270 MOSFET as Temperature Sensor

Question

I have previously asked a question about using thermocouples to maintain a temperature range of 36-39° inside an insulated and dry environment.

The more enlightened engineers advised I look into using either a semiconductor device or an RTD instead.

Parts availability is limited where I'm currently located in the third world and mail order is unreliable at best.

I have a lot of bipolar transistors as well as MOSFETs at my disposal.

I've seen some examples of using 2N3904/3906 as thermal sensors.

The BS270 looks a bit more appealing to me because of the positive temperature coefficient and lower currents required.

My one concern is the inherent D-S resistance warming the device. But if the current is low enough I think the resistance curve with respect to junction temperature looks easier to calibrate than something based on bipolar transistors.

Am I travelling down the wrong path here? Bipolars seem to be more popular for this sort of hack job sensor.

The alternative is for me to manufacture an RTD which is indeed a possibility if that's a better option.

Answer 1

Vbe of a diode-connected transistor is a fairly predictable and stable characteristic of a BJT. It is quite linear and yields a large signal (-2mV/K, approximately). It would probably require at least single point calibration in your application, since variability is of the order of +/-10°C without calibration. I expect you will use this because it is very, very simple (resistor + transistor -> ADC) and "good enough" for your application. Now, for general interest--

Better is difference in Vbe of two matched BJT devices at exactly the same temperature (i.e. on the same die) at two currents differing by a large factor such as 10:1. Or Vbe of a single device at two currents, switched fast enough that the temperature does not vary between measurements. For a device with low base spreading resistance such as an BC547 it is almost independent of individual device characteristics. Diode-connected transistors have an ideality factor close to one and varying little from that (perhaps 1.008 or so) This yields a smaller signal (order of 200uV/K) but more than 10x more accurate without calibration (proportional to absolute temperature) - typically better than +/-1°C.

Almost every characteristic of every component has some temperature dependence so one can't say that any given idea is unsuitable, however some are definitely better than others, and I would say the Vbe of a diode-connected transistor or delta-Vbe is very good. The two-current method is the same principle used to measure the die temperature in your computer CPU, for example, and the PTAT voltage is used in band-gap voltage references.

If r is the ratio between currents, the voltage difference is

\$\Delta V_{BE} = \frac{nkT}{q} \ln(r)\$

where n is the ideality factor, k is Boltzman's constant, T is temperature (Kelvin) and q is the charge on an electron.

Note that, except for n (which is very close to 1.00 for a diode-connected transistor, and does not vary much with a given type of transistor), there are only fundamental constants in the equation plus the current ratio (which you control).

(there are a few other sources of error but that's a very good first approximation)

The key thing is that Is (saturation current) in the Ebers-Moll equation is eliminated. Is varies widely and is not constant with temperature.