HSI Accuracy on the Value line STM32L

elil · ‎2014-02-21

Posted on February 21, 2014 at 15:26

Hello,

The following question I have applied to the support team,on yet another forum and still awaiting an answer. Once I get an answer,I will share it here.

The Value Line series STM32L datasheet claims the HSI accuracy is +-10%, which is without calibration. Since my product is defined as outdoor and HSI frequency is affected by temperarture, the accuracy issue is important for me. To simulate the temperature changes, I've connected a signal generator to LSE and have increased/decreased the frequency by 10%. I have implemented HSI calibration vs LSE in the software. I have expected to fix 10% drift , but, in fact, I've managed to to fix only 5%,which is absolutely unacceptable. Apparently ,the reason for that was the trimming step of 0.4%. There is a 5-bit calibration register that defines the multiplier to the trimming step.

From among these bits the MSB is a sign. Thus, the maximal value I should have reached was 0.4 x 16= 6.4%, which does not cover the mentioned 10% . Inthedatasheet you can find that there 2 optional trimming step: 0.4 or 1.5%. With the latter, I think, it should have worked properly, but it is not clear for me who decides, according to what criteria, which trimming step to put into device in fab ?

Thank you in advance for your opinions,

Eli

waclawek.jan · ‎2014-02-21

Posted on February 21, 2014 at 16:39

> In the datasheet you can find that there 2 optional trimming step: 0.4 or 1.5%.

You misunderstood the meaning of this. It means, that the calibration curve (frequency versus CALIB+TRIM) is not linear, and at every 16th step, there is a 1.5% ''jumpback''. Have a look at AN3300 ''How to calibrate an STM32L1xx internal RC oscillator''.

> T

he Value Line series STM32L datasheet claims the HSI accuracy is +-10%

The reason probably is, that the calibration step is not performed during the manufacturing to achieve the promised lower price. I don't have a first-hand experience, but I believe you'll find all STM32L100 chips' calibration value to be the same, I'd expect at the half of its range. So you might find out that you are unable to trim it to required precision. I believe there's only one way out of this - pay the extra pennies and use the L151 instead.

JW