Back to power, from the analysis above (or below, given the order how log entries are rendered) there is only one instantly viable combination for the low-powered-wireless class of gadgets: photovoltaic (solar) primary source and capacitor as backup storage. (The mechanical solution for primary source and backup/storage sounds very attractive, but is something which needs more thinking, investigation and work).
Using photovoltaics as primary source somewhat limits the placement of the gadgets, but we knew it won't be all that easy. On the other hand, this is something which could be considered at an early stage, designing appropriate locations for gadgets with respect to windows and their orientation, when designing the house itself.
Part of the survival strategies is also RECONSIDERING the SCHEDULING and NEEDS of the particular gadgets so that it matches the varying nature of energy supply. The simplest thing is to schedule the power-demanding tasks, if this is possible. However, this may involve certain "intelligence", so instead of some simplistic logic a microcontroller and developing algorithms may be involved.
Let's have a look at some examples. Say, we want to log the temperature every half a hour even during the night, so that we can optimize heating in variously used rooms; but we need the data only the next day, so that the thermometer does not need to transmit them during the night, it's enough to store them locally and transmit them later.
In case of clock, the clock may stop displaying or move the indicators when dark - nobody looks at it anyway. It's enough to keep the time. If a radio receiver is employed to adjust the clock periodically, it may be run only when there's enough power. The longwave receivers such as the european DCF77 are usually lost in the "digital/switching" noise in areas remote from the transmitter (such as my location around Bratislava, Slovakia) so nighttime reception is a must; but common clock crystals are precise enough to be able to run autonomously with acceptable precision for days, so it's enough to try to get synchronized say once a night, and only if there was enough sunlight the previous day so that a bigger capacitor could be charged fully.
Let's now go for the nasty details.
PHOTOVOLTAIC panels come in various sizes, and the common variety could produce around . Sure, inside a house the light level will be lower, but that could be compensated by using a larger panel. Let's have an example - this is a relatively small, 6x4 cm panel, delivering cca 3.9V*15mA=50mW at 50klux illumination. The catalogue has a nice table detailing available light at various environments - turns out that moderate electric lighting delivers two orders of magnitude less power (~500lux) than non-direct sunlight outdoor (~50klux); so even taking into account some degradation during the years, it should be able to power a clock (see analysis in the first blog entry - it consumes a few hundreds of uW in average). As there's no sunlight in the night, we need to have roughly twice that much power and store half of it for the night; but it's easy to add another panel or buy a twice-as-big one.
Solar panels are typically constructed (by connecting the individual cells into suitable serial/parallel connections) so that output voltage of solar panels is around 3-4V. This may simplify the voltage conditioning electronics, possibly reducing it to a simple protection device (Zener diode or similar). On the other hand, using sophisticated energy harvesting ICs (e.g. the ADP5090) can help to squeeze out the maximum available power using MPPT and boost conversion; and may run at as low as 80mV of input voltage, ie. delivering power even at low light level.
CAPACITORS are unfortunately not very good as long-term energy storage. To be able to power the clock at hundreds of uW during night, we need to store say 300 uW * 10 h = cca 10J (yes, that is that daily windup of the pendulum clock!). The energy content of a charged capacitor is (CV^2)/2; considering discharge from 3V to 1V this results in capacitance of around 2.5F. That's way out of what's possible, especially since we ruled out the electrolythics (and supercaps too). But maybe the clock's consumption decreases with the voltage, so let's put it in other way, 300uW at 3V is some 100uA of current, so the clock behaves roughly as a 30 kOhm resistor. If it should take 10 hours to discharge a capacitor from 3V to roughly one third i.e. 1V, that means RC constant is 36000 seconds, resulting in 1F... still no go.
So this is the moment where we need to reconsider the real need for power. We may for example drop the idea of mechanical indication, and go for LCDs; they should consume somewhat less, although the large area ones (and we want these things to be nice, once they will be around us for a long time) may still go as high as tens of a hundred of uA. So let's consider something else. While the clock may in average draw around 100uA, if during night only the timekeeping runs, that can go quite well below one uA, into hundreds of nA. There are RTCs available, running off as little as 40nA. Calculating 300nA instead of 300uA brings down the requirement for capacity down to a few mF, which may be at the brim of possible. The mechanism then may readjust for the current time when the sun comes up, or when somebody switches on the light (maybe for having a look at the clock).
Unfortunately, it's still not done. Capacitors leak, i.e. self-discharge. Even if the considered ceramic and foil types leak much less than electrolytics, it still may be limiting.
Leakage can be modeled as a resistance parallel to the capacitor, through which current from charged capacitor flows. This current may flow around the dielectric, or across it. The first mechanism results in a parasitic resistance, given by the capacitor's construction, but at a given construction/size independent of capacity. The second mechanism directly scales with the dielectric's area i.e. with capacity. As with higher capacities the second mechanism is prevalent, no matter how big the capacity is or if we stack capacitances in parallel, they will discharge roughly in the same time. The manufacturers often give the leakage for capacitors of higher values in seconds (which equals to Ohm * Farad, or MegaOhm * microFarad). The typical value for high-permitivity ceramics (X5R, Y7R/V) is in hundreds of seconds. We need 10 hours, this is some 36000 seconds.
So the solution is to use foil capacitors. The best dielectric material in this regard is polypropylene. Typical high-quality PP film capacitors, such as these, have rated leakage of 30000s - as this is given at 100V, and as the leakage will scale down with voltage (given many leakage mechanisms depend on electric field intensity), it will suffice for our purpose. They also have a guaranteed lifetime of >30000 hrs which is 35 years, at 40 degrees; so it's almost sure they will last 100 years at slightly lower temperature. Unfortunately, the largest model in this line is 10uF and they are also quite costly - the needed 100pcs would cost around 700 Euros... Not even the enlarged budget would hold that.
We need to investigate this further. So, maybe THIS is going to be the real challenge!