AM623: On-die temperature sensor accuracy

 Hello Daniel Laffoon,

Daniel Laffoon said:

The latest information we received from TI was that these temperature readings were accurate to +/- 8degC, pending full part qualification.

From a software perspective, we can perform tests at different temperatures and note down the ADC count values.

Based on these count values, we can implement an algorithm such as linear interpolation or another method to match your ADC count values to a reference temperature.

However, I'm unsure if these same offset values can be used for all variants of the AM62X SOC.

Also, can you tell me what you are doing with this temperature information?

could you let me know what your specific requirements are?

Regards,

S.Anil.

Hi Swargam,

I am not performing any averaging at present.... simply reading the ADC counts from both temperature sensors, converting each to Degrees Celcius and then printing the values out.

Example:

// Temp 0
//--------------------------------------------------
uint32_t SOC_CONT_MASK = 0x30;
uint32_t SOC_MASK = 0x20;
uint32_t CtrlReg = *(uint32_t *)(Address + VTM_REG_VTM_TMPSENS_CTRL_0);
CtrlReg &= ~SOC_CONT_MASK; //Clear SOC and CONT
CtrlReg |= SOC_MASK; // Set SOC
*(uint32_t *)(Address + VTM_REG_VTM_TMPSENS_CTRL_0) = CtrlReg;

while (1)
{
    CtrlReg = *(uint32_t *)(Address +VTM_REG_VTM_TMPSENS_CTRL_0);

    if ((CtrlReg & SOC_MASK) == SOC_MASK)
        nanospin_ns(5000);
    else
        break;
}

uint32_t Temp = (*(uint32_t *)(Address + VTM_REG_VTM_COMMON_0_STAT)) & 0x3FF;
float TempInDegC = ConvertToDegC(Temp);
printf("Temp_Sensor_Main_0 (DegC): %f\n", TempInDegC);

float ConvertToDegC(uint32_t RawTenBitTemp)
{
// From Table 6-3136 in spruiv7.pdf
const double a4 = -9.2627e-12;
const double a3 = 6.0373e-8;
const double a2 = -1.7058e-4;
const double a1 = 3.2512e-1;
const double a0 = -4.9002e1;

double x = RawTenBitTemp;

float y = (float)((a4 * x*x*x*x) + (a3 * x*x*x) + (a2 * x*x) + (a1 * x) + a0);

return y;
}

We are simply interested in ascertaining the junction temperature while running various stressors while at the upper operating limit of our product and trying to verify junction temperature stays below the rating.  Right now, the accuracy stated is 16degC spread.  I don't have a specific amount, but if we could get that down to say 5degC that would be great.  if you could state a tighter accuracy just in this upper range (say 85degC+), that would suffice as well.

We are currently using XAM6254 (with GPU disabled, and single-core); we are targeting AM6231 - we will have production AM6231 on our next revision here in ~2 months.

Thanks,

Dan

Hello Dan, 

Thank you for the inputs.

Let me check internally if we have any additional performance data and update you.

Quick question - have you done some measurements or characterization and is there some data on your observations.

Regards,

Sreenivasa