Is die temperature calibrated after the chip manufacture?

[scytulip]

I tried to read out the temperature sensor of NRF51822. In light CPU load (turn on ADC and send data through BLE every 1s), the measured value is around 45 degC, compared to my temperature gun reading, 32 degC on the chip package surface. I know there should be some difference between die and surface, however, 45 degC seems not reasonable for this LOW POWER chip set.

I wonder whether the temp. sensor was calibrated after its fabrication? Is the reading reliable?

[Stefan Birnir Sverrisson]

Hi

The temperature sensor should give out a 10 bit 2’s compliment value but if you use the temperature_example in the SDK the output will be 32 bit 2’s compliment value. To find the actual deg C temperature value you should devide the TEMP output value with 4, as done in the temperature_example in the SDK.

The nRF51 temperature sensor might need offset calibration. If you are not seeing a meaningful value, you should calibrate the temperature value in software to get a meaningful value, i.e. if you know the temperature in your room, just add or subtract from the measured value to get the correct value.

The temperature meter doesn't reflect the room temperature perfectly, as it is on the die. I have seen the temperature measurement on the nRF51 increase by 2 deg C after one minute of operation, compared with a cold start, and that was with medium CPU and radio load. If your application has high activity and the chip is drawing a lot of current, you might see even more temperature offset from the room temperature. However, if your chip is drawing relatively low average current (perhaps <0.5mA), the nrf51 temperature meter should give you a fairly good estimate of the room temperature after you have calibrated it with the method mentioned above.

[Bouke]

I think the short answer is "no". Those 13 degrees between ambient and die temperature are too much, and should be corrected for, also see my question and data.

There is an undocumented temperature offset register (mentioned in a PAN) which might originally have been intended as a calibration mechanism, but it is always set to 0 (and if not, you're told to do so yourself). It seems the +/- 4 deg accuracy spec is not completely made good upon.