Sorry for the news....but your thermistor design may be more costly than you think.  Have you ever wondered if you should use a digital temperature sensor instead of a thermistor? Probably not - most people design with thermistors for one of two reasons: 1) they are affordable and 2.) they are familiar. These are good reasons for using a thermistor, and historically I would say that replacing a thermistor with a digital temp sensor sounds expensive, until I happened upon page 16-18 of the TMP112 datasheet. This section is titled "Calibrating for Improved Accuracy" and it has a lengthy detail of how to achieve better than 0.5°C accuracy over specific temperature ranges through something called average slope error.

What  does this have to do with your friendly old thermistor? Well, I asked one of our apps guys how to achieve best case accuracy using thermistor based design - and he shared a circuit that looks something like the below (figure 1). He said it is pretty common for cost-sensitive designers to automatically choose a thermistor because it seems affordable, but for designs that need a high level of accuracy (better than 1°C for example) there is usually a multi-point calibration involved to compensate for its inherent nonlinearity, and a simplified version of the process goes something like this: select highest accuracy thermistor for the lowest possible price, drift over temperature in operating range of interest, calibrate multiple points for errors to get an accuracy over temperature curve, using this curve create a look-up table for the micro-controller, repeat for every thermistor in the design and...go!   A cheap and easy way to accurately measure temperature. Okay, that sounded to me like it might cost some serious time, and maybe would require some additional cost for calibration circuitry - and it didn't sound all that easy either.

.

Thermistor Cal Sample Circuit (Figure 1)...

So I asked him about the TMP112 and the average slope error. It turns out that the TMP112 has specifications for user-calibrated systems, which allows accuracy better than 0.5°C to be achieved over limited temperature ranges - using calculation, and a single point of calibration at 25°C. So the process now goes something like this: power up your TMP112 at 25°C, measure and zero-out the offset error, use provided slopes to predict worst case error over temperature, get digital output from TMP112 with better that 0.5°C accuracy.

Whether or not the TMP112 would be the lowest cost for your design needs, it definitely sounds like it might be the easiest.  Depending on what you are trying to accomplish you might want to think about leaving your trusty dusty thermistor on the shelf for your next high-accuracy design.

Check out more about the TMP112 and TI temperature sensor design activity at the TI Temperature Sensor Forum: http://e2e.ti.com/support/other_analog/temperature_sensors/f/243.aspx