Diode vs BJT temperature sensors

[philpem]

Hi there,

In one of my fits of insanity, I decided it'd be fun to try and build a laser power meter. What I have is a thin chunk of brass, painted black on one side, with a 100R resistor and some form of temperature sensor with sub-degree-C resolution. A second sensor is held at room temperature and used to eliminate ambient temperature from the equation (zero the measurement). An instrumentation amplifier, precision current source (the good old REF200) and "balance" pot (used to compensate for Vf/Vbe difference between the two sensors by adjusting the forward current) completes the circuit.

The theory is, the black paint absorbs the energy from the laser beam and its temperature rises as a result. Measuring the increase in temperature allows you to work backwards to calculate optical power. The 100R resistor can have a known power (say, 10mW DC) fed through it to "calibrate" the sensor head.

I've noticed that there are two ways to implement a "silicon bandgap" temperature sensor -- a silicon diode (e.g. 1N4148) or a diode-wired bipolar transistor (e.g. 2N3904, BC547). The BJT seems to be more prevalent. What I'm struggling to find is any rationale behind this; a lot of silicon temperature sensor chipmakers say "always use a BJT", but give no reason for the statement.

Does anyone know why you'd use a diode-connected BJT? Are they more accurate or repeatable than silicon diode sensors?

Cheers,
Phil.

[edavid]

You wouldn't want to use a 1N4148, since it's gold-doped.  I don't know if that explains the transistor preference or not.

[Andreas]

http://cds.linear.com/docs/en/application-note/an137f.pdf

see page 6

usually you would want to use a PNP transistor.

With best regards

Andreas

[wraper]

Better use PT100 or PT1000 sensor if you want reliable "sub-degree-C resolution".

[muvideo]

Better use PT100 or PT1000 sensor if you want reliable "sub-degree-C resolution".

I would consider also using a simple NTC, a good one it's repeteable enough,
coupled to a good resistor I think it's the simplest and cheapest way to have
an high resolution temperature sensor.

[wraper]

I would consider also using a simple NTC, a good one it's repeteable enough,
coupled to a good resistor I think it's the simplest and cheapest way to have
an high resolution temperature sensor.

Why to bother, when for $6-7 you can get PT100 class b, completely linear, with 0.4 degrees C initial resistance accuracy at room temperature. 0.05% stability. You even don't need to do calibration in many cases.       

http://www.micropik.com/PDF/pt100.pdf
http://www.farnell.com/datasheets/1702835.pdf

[Eviltech]

What i use for a sub-degree temperature sensor is a NTC from a digital thermometer for medical pourposes. If you use a verry stable power supply you might even get to 0.01C accuracy. Downside is that the particular NTC is only usable up to about 80 deg. Celsius.

[wraper]

I have calorimeters at work which uses PT100, resolution is 0.0001 C (on display, actually uses precision 24bit ADC) C and at least 0.0003 C is usable.

[Eviltech]

I have calorimeters at work which uses PT100, resolution is 0.0001 C (on display, actually uses precision 24bit ADC) C and at least 0.0003 C is usable.

I do belive this will be perfectly adiquate, presuming the rest of the circuit is up to the task...