Tracking temp changes with MCU

[niconiconi]

Another place to find silicon temperature sensors is some references which include a PTAT (proportional to absolute temperature) output, but an LM35 is probably less expensive at $13.30/10, and there are board mount temperature sensors much less expensive than that.

Interesting. I didn't know some voltage references made the internal PTAT circuit available.

[cv007]

Some of my implementation has DS18B20 and it works through UART (STM32 doesn't have native implementation of OneWire for it). Recently, some guy suggested me to use NTC instead of 18B20, I google it, and looks like it's pretty common way.

So you are currently using digital, although the 'interface' is unknown- bare ic to object, ic on a pcb, encapsulated, etc. I am no expert in this field, but if the temp range allows my first inclination would be to use some digital temp ic where you can let the ic handle all the little details and you can get a value you know will be within the datasheet specifications. Like you are doing currently.

There are many options for digital temp ic's- from just Texas Instruments as an example you have tmp117 (i2c), tmp1826 (1wire), LMT01 (pulse), and a lot more. They also have a lot of useful app notes.

There may be good reasons for looking at other options (large quantities/cost), and you will have a need to 'interface' the sensor to the object in some way for any solution, but the biggest advantage to the digital temp ic is you know you will be getting a temperature value that is accurate within the specification. No need to calibrate, no analog to get wrong, minimal parts count. I would also use something more accurate than needed, mainly for a little more confidence you are within the accuracy wanted. Also make sure you check the datasheet for accuracy over the range you need- for example the ds18b20 lists as +/-0.5C but at 100C goes to +/-2C (which is still ok according to your stated accuracy requirement).