I'm trying to design an RTC that delivers a timing error of better than 1ppm. So far the closest I've been able to get is the Maxim DS3232, but this is +/-3.5ppm (across full temp range) + around 1ppm / yr (ageing).
Seems virtually impossible since an OCXO is out of the question. We need to know a bit more about your design, what conditions it operates under etc.
For what it's worth many years ago we bought a load of Dallas 2PPM RTCs but found it was very hit and miss if they met the spec.
It sounds like you need to periodically re-sync your clock, or find some way of mitigating drift as a problem. For example a product I have been working on needs extremely accurate timing (a few 10s of uS variation between two completely separate devices maximum) but it wasn't possible to do in a low power and low cost way. We eventually found a way of syncing the two devices even though there was no comms between them :-) I'd tell you how but then I'd have to kill you, but the point is there may be a way around your rather extreme requirement.
Yep, I was kind of figuring a periodic sync may be the best option here. We do have a wireless interface which could be used....intrigued by the no comms / sync technique though....
The power question is still a bit open in that there will be a single battery supplying power to the rest of the system as well (and this is still being worked on). What is defined is that we need 5 year battery life and I was kind of hoping I could just throw a backup battery in there with the RTC so the time only has to be set once and is retained even if batteries are changed. In short, I've kind of assumed I need uA consumption for time keeping rather than mA (which is of course a problem as soon as I look at any decent precision oscillators).
Separately I've also wondered about the distribution of timekeeping error across multiple devices and whether you could essentially correct for error by averaging the time from multiple devices. I suspect though that much of the error will be temperature dependent and similar between devices. This doesn't help with power consumption either...
That is a very hard problem - you should maybe look again at why you need that level of accuracy and see if you can do it a different way.
Depending where you are geographically, a MSF/DCF radio-controlled clock may me an option to periodically adjust the internal timebase.
The requirement is actually a regulatory one which kind of makes it hard to avoid.
I'll have to look into whether there are clock sync signals available we could use (in Australia).