75..125C temp sensing with <0.5C accuracy (T-type thermocouple)

Document created by analog-archivist Employee on Feb 23, 2016
Version 1Show Document
  • View in full screen mode

Q1. Our research team need to measure the temperature profile on the surface of
a chip using multiple thermocouples (4x at least, at different positions). For
our experiments, we assume having T-type thermocouples with a fast response
time and accuracy better than 0.5°C. We then need to amplify the signal coming
from all the thermocouples and send a structured packet to the PC for
temperature trace analysis, via USB.

For this reason, we need a thermocouple amplifier that is able to provide
accurate amplification of the signal. We came across the AD594/5 IC, but from
the specs we realize that it guarantees only 1°C of accuracy, and this is too
low for us.

Do you have an IC in your portfolio that guarantees 0.5°C of accuracy or even
better (e.g., 0.1°C)?

Q2. How could we properly calibrate the AD8495 to deal with T-type


Q1. Something close to that is likely possible with the AD8495 and some
calibration. The "Initial Error" spec is due to amplifier offset voltage and
can be calibrated out with a 1 point temperature calibration, and then AD8495C
specifies < 0.1% gain error. But also Type T thermocouples themselves are quite
non-linear. You would definitely need linearity compensation to get anywhere
near 0.5°C. We can provide a linearity compensation polynomial or table if

The best solution is more likely something like the ADT7320 and the AD7793 in
this article :


Q2. In the narrow range, you still see some nonlinearity. For very high
accuracy, I still don't think linearity is a very good assumption. If you use
AD8495 with a type T thermocouple and you assume the output is linear between
75°C and 125°C, you end up with a maximum of about 2.8mV output error vs the
straight line. That's a little bit more than 0.5°C.

Here is the graph and the table.

The table:

The polynomial calculated by Excel's best-fit line over this range was :

The 23°C ± 5°C ambient temperature rejection error was negligible, calculated
to change the output voltage by ±0.1mV. Also, the stated Polynomial Error takes
that ambient temp error into account.

Chances are the look-up table is computationally cheaper than the polynomial in
this case.

I would stress once again the importance of an accurate one-point temperature
calibration to remove the Initial Accuracy error. Any error in that calibration
measurement will persist as residual offset.