[time-nuts] A look inside the DS3231

Attila Kinali attila at kinali.ch
Sun Jul 30 08:13:27 EDT 2017

On Sun, 30 Jul 2017 12:23:17 +0200
Pete Stephenson <pete at heypete.com> wrote:

> > > - I find it remarkable that this circuit can operate on less than a
> > > microamp during normal usage, including temperature conversion.
> > 
> > That's not so remarkable. If you make the transistors long, then
> > you get very low leakage. Couple that with small clock frequency
> > and you use very little current. Modern ICs only use so much current
> > because they have so many transistors, which are also optimized
> > for being fast, rather then low leakage. 
> Good point! I admit the details of optimizing transistors for different
> purposes is beyond my ken, and I appreciate the insight.

There are multiple optimization points. One is to select a prodcution
process that is optimized for low leakage. I.e. thick gate oxide
and high threshold voltage. Both of these parameters imply higher
suplly voltage.
Then, in the design, you make your transistors long and large.

The problem here is, that power consumption scales proportional
to the square of supply voltage, the gate capacitance and the
switching frequency. This means, if you choose a low leakage
process, and thus high supply voltage, your power consumtion
will go up. The same goes for choosing large transistors. 
Hence it becomes a trade-off between static (leakage) and 
dynamic (through gate capacitance) power consumption.

> > > The DS3231 has on-board temperature monitoring to correct the crystal
> > > frequency: is this something where they would have bothered putting a
> > > separate sensor next to the crystal itself, or are the die and the
> > > crystal are close enough and in the same package that they could use an
> > > on-die sensor like a diode and call that "good enough"?
> > 
> > My guess would be that it's a PN-junction or a bandgap temperature
> > sensor somewhere on the chip. Adding another part increases the cost
> > of production quite considerably.
> Indeed. At first glance, I was surprised not to see tiny discrete
> capacitors within the chip package itself, as I assumed (incorrectly)
> that getting sufficient capacitance to steer a crystal a little would
> require larger capacitors than could be easily put on a die, but then I
> remembered that each LSB in the aging register only changes the
> frequency by 0.1ppm at 25C, so that wouldn't need a large amount of
> capacitance. 

As a rule of thumb, you can assume that in an "old" (aka large node size)
process the gate capacitance is approximately 1nF per mm^2. So, you can
build quite easily 10-100pF of capacitors on-chip.

			Attila Kinali
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor

More information about the time-nuts mailing list