AN-1058: Using CBC348xx EnerChip RTC in High Accuracy Applications

©2014 Cymbet Corporation • Tel: +1-763-633-1780 • www.cymbet.com

Doc AN-72-1058 Rev B

Page 2 of 7

Temperature Compensation

Tuning fork crystals typically exhibit a temperature versus parts per million (ppm) frequency characteristic that

is centered on 25°C. Figure 1 below shows a typical characteristic.

Figure 1: Parts per Million Frequency Error vs. Temperature.

The most common frequency versus temperature coefficient is -0.035 ppm, but that operates on the square of

the temperature delta from 25°C (thus the parabola). Notice that if the temperature is either above or below

25°C the crystal will slow down. It will never speed up. Also notice that if the temperature is always near 25°C,

say 25°C ± 6.5°C, there is only a -1.5ppm error which is quite minor. 1.5ppm equates to about 47 seconds per

year. The next three sections present techniques for compensating for this parabolic curve at the system level.

The techniques are: a static compensation based on use-case information, an active compensation using the

host microcontroller, and an adaptive compensation using the host microcontroller that requires less energy.

Use-Case Compensation

This technique requires estimation of the average temperature the system will be exposed to and

compensation at final test time for the difference between the estimated average temperature and 25°C.

There is no active temperature compensation in this technique after the system is initially calibrated. Because

the crystal temperature profile shown in Figure 1 is a parabola that plots ppm errors versus temperature, the

crystal will slow down if the temperature is either above or below 25°C. This means that the oscillator needs

to be sped up at assembly time by calibrating the oscillator to a faster clock. To determine the speed of this

reference clock, sum the products of the hours per year the system is expected to be at a given temperature

times the ppm error associated with each temperature and divide the sum by the total hours in a year. This

weighted average ppm figure will be how much to speed up the clock used for calibrating the system at test

time.