nRF52840: At low temperatures I can't achieve the specified accuracy for the internal temperature sensor

[Frank Viganske]

While testing our custom Hardware with the nRF52840 in a climate chamber in the -40°C to +85°C range i noticed that the value read from the internal temperature sensor was much too low at -40°C. According to the specification the deviation should be within the -5°C to +5°C range over the whole operating temperature range. I am aware of the erratum 66 and have implemented the initialization code for this.

At high temperatures the accuracy is OK, i only see this large deviation at very low temperatures. I tested 12 PCBs simultaneously. Of 12 boards only 3 gave a temperature measurement result of -45°C or higher at -40°C climate chamber temperature. All remaining boards showed values below -45°C The lowest value i have seen on one board was -55,25°C.

I compared the internal CPU temperatures to a TMP1075 I²C temperature sensor mounted on the same board.. The TMP1075 sensors saw a temperature within the range of -39°C to -37°C during the same measurements..

Below is the code used to initialize and read the temperature sensor. I'm not 100% sure about this, as i didn't have any example code available, but the initialization should be ok, i just copied the suggested workaround from the errata sheet..

void init_cpu_temp_sensor(void)
{
  NRF_TEMP->A0 = NRF_FICR->TEMP.A0;
  NRF_TEMP->A1 = NRF_FICR->TEMP.A1;
  NRF_TEMP->A2 = NRF_FICR->TEMP.A2;
  NRF_TEMP->A3 = NRF_FICR->TEMP.A3;
  NRF_TEMP->A4 = NRF_FICR->TEMP.A4;
  NRF_TEMP->A5 = NRF_FICR->TEMP.A5;
  NRF_TEMP->B0 = NRF_FICR->TEMP.B0;
  NRF_TEMP->B1 = NRF_FICR->TEMP.B1;
  NRF_TEMP->B2 = NRF_FICR->TEMP.B2;
  NRF_TEMP->B3 = NRF_FICR->TEMP.B3;
  NRF_TEMP->B4 = NRF_FICR->TEMP.B4;
  NRF_TEMP->B5 = NRF_FICR->TEMP.B5;
  NRF_TEMP->T0 = NRF_FICR->TEMP.T0;
  NRF_TEMP->T1 = NRF_FICR->TEMP.T1;
  NRF_TEMP->T2 = NRF_FICR->TEMP.T2;
  NRF_TEMP->T3 = NRF_FICR->TEMP.T3;
  NRF_TEMP->T4 = NRF_FICR->TEMP.T4;

}

float read_cpu_temp_sensor(void)
{
  int result;
  NRF_TEMP->TASKS_START = 1;
  NRF_TEMP->EVENTS_DATARDY = 0;
  // Read it back to ensure it is written (needed on cortex m4)
  volatile uint32_t dummy = NRF_TEMP->EVENTS_DATARDY;
  (void) dummy;
  while (0 == NRF_TEMP->EVENTS_DATARDY);
  result = NRF_TEMP->TEMP;
  NRF_TEMP->TASKS_STOP = 1;
  return (float)result / 4;
}

Assuming that the code is ok (i hope so), what else could cause this high deviation of the temperature reading?

Best regards

Frank

[Kenneth]

Hi,

Make sure to check out:
https://infocenter.nordicsemi.com/topic/ps_nrf52840/temp.html

"To achieve the measurement accuracy stated in the electrical specification, the crystal oscillator must be selected as the HFCLK source, see CLOCK — Clock control for more information."

Best regards,
Kenneth

[Frank Viganske]

Hi Kenneth,

this has been taken care of. We have a 32MHz crystal connected to XC1 and XC2, and this was also selected as the HFCLK source during the test.

Best regards,

Frank

[Kenneth]

Can you share the code snippet that startup the HFCLK before you execute temperature measurement? If you go to sleep, then you must startup the HFCLK every time you wakeup before you can do a temperature measurement.

Kenneth

[Frank Viganske]

Hi Kenneth,

Sorry, i don't have the source code for starting the HF clock. We are using a third party radio protocol stack, and i only have that as a precompiled binary. The clock is controlled by this module. The radio was active and working fine during the test, so i'm sure that the HFCLK was running from the external crystal.

i can do a debug print of some of the status registers for the HFCLK if that helps..

Best regards

Frank

[Kenneth]

Frank Viganske said:

i can do a debug print of some of the status registers for the HFCLK if that helps..

That is a good idea. It may be that the radio protocol stack will start and stop the external HFCLK before and after the radio usage (this is the case even for Nordic softdevices, though there is an api that typically can be used to keep it ON all the time, which is what you want for this test).

Kenneth

[Frank Viganske]

Hi kenneth,

I inserted a debug print of the HFCLKSTAT at the beginning of the temperature measurement, and the value was 0x10000 (HFCLK running, but on the internal RC oscillator). So it seems that you're right,  the inaccuracy might be caused by the incorrect clock source.

I'll try to change my firmware to keep the crystal oscillator running during the temperature measurement and see if that improves my accuracy.

Best regards

Frank

[Frank Viganske]

Hi Kenneth,

I changed the Firmware to switch over to the HFXO during the temperature measurement. Now at -40°C i get the result -41.5°C from a board that measured -53.75°C in my previous tests.

Problem solved. Thanks for your help.

Best regards

Frank