I had a few minutes to play with a super capacitor and a Domino. Here are my results:
I used a .47F 5.5V super capacitor.
I had general concern about inrush current. This is the initial current draw that happens when the uncharged capacitor is connected to a voltage. To limit this intial current I used a 100 ohm resistor.
At the low current draw of the RTC, under battery conditions, the voltage drop across the resistor is very small. During intial charging, with higher currents, the resistor will ease the load on the 3.3V supply.
For my testing, I was able to use an analog input to monitor the voltage at the capacitor.
With nothing connected, using the analog input, I measured a voltage on the VBat pin of the Domino. it was ~3.3V.
I then connected the minus of the capacitor to ground and the plus to a 100 ohm resistor and the other end of the resistor to VBat.
This did not work… The volltage a VBat was dragged to under a 100 mv and was increasing very very slowly. Putting in a large resistor increased the charge rate, but it was still very slow. I was concerned about the effect of dragging down the voltage source that was power VBat for a long time.
I then tried a modified version of schematic that Gus suggested above. I modified it by adding an inrush protection resistor in addition to a diode.
I connected the capacitor minus to ground. I connected the plus side of the capacitor to VBat. I then connected a diode in series with a 100 ohm resistor between the 3.3V pin and the plus side of the capacitor. (Anode of diode to 3.3V)
Within a minute the voltage at the plus side of the capacitor is about 2.7 volts. The diode I used has a ~.7V drop across it, so 2.7V makes sense. To reduce this drop, a Shockley diode could be used.
The NXP data sheet says that the RTC backup fails at about 1.6V, so 2.7V is fine.
I have not made tests to see how long the capacitor keeps the RTC going. But at 2 microamps, it should be a while. If the Panda has a VBat pin, I will buy one and do a long term test.
It turns out the super capacitor is a lot bigger than I thought. it is about the size of eight cr2032 stacked coin batteries. A coin battery in a holder is smaller and neater. Less parts.
I will be repeating these tests on a Cobra. The Cobra might not need the diode.