"... the crystal runs with 32.940Hz"

If that's really true, something is very wrong! This would be a deviation of 0.5% from 32768Hz, while a typical watch-crystal is accurate to well within ±100ppm (=0.01%). Have you read Dallas/Maxim Application Note 58?

As an alternative to the DS1302, you might want to consider STM's M41T56, which has a built-in calibration mechanism.

Note that your requirement of 1 minute per year (at least, that's what I think you mean by "1min per ano") equates to ±2ppm. You need a really stable clock source to achieve that, because eventhough a watch crystal has a fairly good accuracy, it has a very high temperature coefficient; typically around 4ppm/°C


Best regards,
Rob Klein.