When i was an apprentice so long ago, I remember the company i worked for using one junction of a transistor for doing 0.1degC. Accuracy was actually a lot better, but thats all they could claim becasue of the equipment used for calibration.
I remember that everytime they got a new batch of transistors, i would have to spend days with an old 4wire thermistor and meter to make readings every 0.05degC. Sitting with an ice bath waiting for the ice to melt and trying to zero that meter was a right pain in the arse. Calibrating the standard device was the same, and that had to be done every three months.