If you connect the transistor as a diode and bias it with a reasonable current, maybe a couple hundred uA, you can read the voltage. It will require calibration (say at room temperature and 0°C in an ice-water slurry.

Sensitivity will be about -2mV/K so with a 5mV resolution ADC you’ll have 2.5 degrees C resolution, not great.

If you connect it as a Vbe multiplier, say with 5:1, and average many measurements you might be able to get a usable resolution of about 0.5°C. Or just use an op-amp.

Edit: Here is a simulation result used as a simple Vbe multiplier.

0°C 3.591V

25°C 3.328V

50°C 3.057V

75°C 2.781V

100°C 2.500V

The current drops as the voltage increases (due to the use of a simple resistor as a current source) so the linearity isn't great= 11.2mV/K average at 0..25°C and 10.5mV/K average at 75..100°C, but for narrow excursions around room temperature it should be fine, or it can be corrected digitally. With a 10-bit ADC and 5V Vref the resolution is about 0.5°C, which is adequate for many purposes.

There are much better ways to measure temperature with a transistor, however the complexity is increased. Using matched transistors or multiple (2 or 3) currents with a single transistor allows cancellation of many of the transistor parameters that vary from unit to unit, as well as connection resistance. Unfortunately, the sensitivity is reduced by at least an order of magnitude so better analog circuitry is required.

If you connect the transistor as a diode and bias it with a reasonable current, maybe a couple hundred uA, you can read the voltage. It will require calibration (say at room temperature and 0°C in an ice-water slurry.

Sensitivity will be about -2mV/K so with a 5mV resolution ADC you’ll have 2.5 degrees C resolution, not great.

If you connect it as a Vbe multiplier, say with 5:1, and average many measurements you might be able to get a usable resolution of about 0.5°C. Or just use an op-amp.

Edit: Here is a simulation result used as a simple Vbe multiplier.

0°C 3.591V

25°C 3.328V

50°C 3.057V

75°C 2.781V

100°C 2.500V

The current drops as the voltage increases (due to the use of a simple resistor as a current source) so the linearity isn't great= 11.2mV/K average at 0..25°C and 10.5mV/K average at 75..100°C, but for narrow excursions around room temperature it should be fine, or it can be corrected digitally. With a 10-bit ADC and 5V Vref the resolution is about 0.5°C, which is adequate for many purposes.

There are much better ways to measure temperature with a transistor, however the complexity is increased. Using matched transistors or multiple (2 or 3) currents with a single transistor allows cancellation of many of the transistor parameters that vary from unit to unit, as well as connection resistance. Unfortunately, the sensitivity is reduced by at least an order of magnitude so better analog circuitry is required.

Edit/Epilog:-

I tried this with a single diode-connected BJT (base connected to collector) with a 10K resistor to the regulated (and otherwise unused) 3.3V rail on an Arduino Nano. Reference set to nominal 1.1V as suggested by @EdgarBonet, summed 100 sequential readings.

1. Calibrated the voltage by adding a multiplier so the reading was accurate in mV compared to a handheld 3.5 digit DMM.

2. Wrote the equation for temperature based on the current room temperature and the mV, with an estimate of -2.0mV/K for the Vbe temperature coefficient.

3. Tested it at 0°C and 45°C against a type K bead thermocouple.

4. Adjusted the 2.0 to 2.2 to reduce the error at the temperature extremes.

adc *= 1.0532319391 * 0.01;  // calibrated voltage in millivolts, 100 summed
temp = -(adc - 556)/2.2 + 24.0;

Works quite well with only one unexpected thing- the INTERNAL1V1 constant was not defined even though the Arduino environment knows it is an ATMega328p (using an Arduino Nano). I added one line to code the definition:

#define INTERNAL1V1 2
analogReference(INTERNAL1V1);

If you connect the transistor as a diode and bias it with a reasonable current, maybe a couple hundred uA, you can read the voltage. It will require calibration (say at room temperature and 0°C in an ice-water slurry.

Sensitivity will be about -2mV/K so with a 5mV resolution ADC you’ll have 2.5 degrees C resolution, not great.

If you connect it as a Vbe multiplier, say with 5:1, and average many measurements you might be able to get a usable resolution of about 0.5°C. Or just use an op-amp.

If you connect the transistor as a diode and bias it with a reasonable current, maybe a couple hundred uA, you can read the voltage. It will require calibration (say at room temperature and 0°C in an ice-water slurry.

Sensitivity will be about -2mV/K so with a 5mV resolution ADC you’ll have 2.5 degrees C resolution, not great.

If you connect it as a Vbe multiplier, say with 5:1, and average many measurements you might be able to get a usable resolution of about 0.5°C. Or just use an op-amp.

Edit: Here is a simulation result used as a simple Vbe multiplier.

0°C 3.591V

25°C 3.328V

50°C 3.057V

75°C 2.781V

100°C 2.500V

The current drops as the voltage increases (due to the use of a simple resistor as a current source) so the linearity isn't great= 11.2mV/K average at 0..25°C and 10.5mV/K average at 75..100°C, but for narrow excursions around room temperature it should be fine, or it can be corrected digitally. With a 10-bit ADC and 5V Vref the resolution is about 0.5°C, which is adequate for many purposes.

There are much better ways to measure temperature with a transistor, however the complexity is increased. Using matched transistors or multiple (2 or 3) currents with a single transistor allows cancellation of many of the transistor parameters that vary from unit to unit, as well as connection resistance. Unfortunately, the sensitivity is reduced by at least an order of magnitude so better analog circuitry is required.