Transistor temperature accuracy when used as a sensor

Hello,

I am using a BC547 transistor and a BC548 transistor as temperature sensors.

I see that there is some deviation compared to air temperature, between 2-4 oC and that the temperature sometimes increases for ~4oC.

Do you know the temperature accuracy of these transistors? Is there a way to figure this out from their datasheet?

Also, would you recommend better an LM35 temperature sensor, like this one?

https://www.ti.com/lit/ds/symlink/lm35.pdf

Note that these transistors are connected to an AMC6821 IC.


Thank you,

blade88 said:

would you recommend better an LM35 temperature sensor, like this one?

Click to expand...

Why can't you the same LM35?
Or else try LM75AD

Hi,

BJTs can be used as approximate temperature sensors, the same as diodes, but they can't really be compared to a temperature sensor IC, for reasonably accurate measurments the LM35 or LM335 are far better, I like the LM35 - it's accurate and simple to implement for analog, which doesn't mean to say it isn't worth looking at other analog temp. sensors to see which is most appropriate for your design.

...Looking at the AMC6821 datasheet, it recommends a BJT, search for temp. sensor BJT or similar, there is at least one which is considered a good choice for this function, but the 547 and 548 are unlikely to be comparable. Also, from the datasheet, it looks improbable that the LM35 can be used there instead of a BJT, the operation is somewhat different for both. I think it's the (glass) 1N4148 diode used in hospitals as a throwaway thermometer.

It looks like it's measuring the deltaVBE of the transistor, you could try putting a resistor between ground and the emitter of the 547/8 and seeing if that linearises the temp. rise/fall a little better - I think - as I haven't read the AMC6821 datasheet.

Here's an ap note on using a 2N3906 as a temp sensor.

The old A/C system in my house used plain glass
diodes at the end of long wire runs, for per-room
temp sensing. "Good enough" for that, I guess.

Self heating may impose a few degrees worth of
sensed-temp (as Vf / Vbe voltage) error.

Every integrated circuit temperature sensor
fundamentally uses the transistor as the sense
element. But they have the advantage of close-
in gain and linearization done with matched
(make, supply, die temp) devices, which a bare
element remote from analog signal conditioning
can't really come close to.

So how good is good enough? And for the trivial
cost difference and near-same-form-factor, why
not the IC?

Something that has not been mentioned in this thread, is a comment from an old National Semi app note:
Plastic packaged temp sensors (and that includes transistors) sense the surrounding temperature thru their leads.

It is a very common pitfall that heat is conducted from another heat-producing device thru the PWB's copper traces into the sensor.
Copper is after all, an excellent heat conductor, and the techniques that are used for low electrical impedance also promote low thermal impedance.

This is not a theoretical speculation, it has happened to me.

Hello,

Thank you all for your replies.

I have been testing an LM35 sensor and a common npn transistor.
The difference in normal conditions is that the pnp transistor shows approximately ~3-4oC degrees more than LM35.
When I heat them with a thermal source, I see that the common transistor shows a larger increase in temperature than the LM35. This difference can go about ~10oC.

So, while 3oC degrees difference is not that much, the question is how the common transistor will behave, when it is close to a thermal source (for example an active unit, generating heat when operating).
I guess this needs to be tested in actual conditions, to see the actual behavior.

- - - Updated - - -

schmitt trigger said:

Something that has not been mentioned in this thread, is a comment from an old National Semi app note:
Plastic packaged temp sensors (and that includes transistors) sense the surrounding temperature thru their leads.

It is a very common pitfall that heat is conducted from another heat-producing device thru the PWB's copper traces into the sensor.
Copper is after all, an excellent heat conductor, and the techniques that are used for low electrical impedance also promote low thermal impedance.

This is not a theoretical speculation, it has happened to me.

Click to expand...

I found the respective National Semiconductor Application Note here **broken link removed**

Please also note that the common transistor is inside a heatshrinkable tube.
The LM35 is inside a metal case, like in the photo: http://i.stack.imgur.com/IlKKD.jpg

blade88 said:

I have been testing an LM35 sensor and a common npn transistor.
The difference in normal conditions is that the pnp transistor shows approximately ~3-4oC degrees more than LM35.
When I heat them with a thermal source, I see that the common transistor shows a larger increase in temperature than the LM35. This difference can go about ~10oC.

So, while 3oC degrees difference is not that much, the question is how the common transistor will behave, when it is close to a thermal source (for example an active unit, generating heat when operating).
I guess this needs to be tested in actual conditions, to see the actual behavior.

Click to expand...

Hi,

I feel you have seen with the above experimentation pretty much what is to be expected. On-chip temperature sensors are - from what I've read - still not a perfect method as even single packages have hot-spots, cooler parts, etc, and a pcb will be more noticable in this respect.

The LM35 is very accurate, however, there is a section in the datasheet offering advice on layout, as it also senses temp. through pins, and if heat-shrinked, you can get it to go up quite a few degrees by holding it by the pins, not just the plastic IC housing.

Related, somewhat, playing around with discrete bandgaps some time ago, either BJT or Brokaw op amp one - not worth the time for a serious application: one BJT had suffciently different temp. to one half a centimeter away - with no heat creating devices nearby... and blowing on the circuit is funny to see how hopeless such things are, adding a heat source deliberately makes it hysterical. All that said, potting might offer some solution to that issue.

Again, I don't believe the LM35 can be connected to the AMC6821, those pins for the temp. sensing device are a current source, the LM35 only uses a supply voltage and outputs a voltage, not a current, so a BJT would be needed.

I think, and I might be wrong, the best approach is to consider what temperature you wish to monitor, and place the BJT as close to that spot on the PCB as is physically possible, assuming you aren't measuring ambient air temperature.