Interesting idea. Never really thought about it. If you create efficient circuitry and correctly implement low power wireless a replaceable battery can last months (or only need charged at those intervals).
Mechanical is really no different than membrane from an electrical standpoint. In both, the controllers scan for connections when the switch/contacts get closed and electricity can pass though them and that's what registers the key press.
You are correct that power consumption will be the number one concern next to functionality. I remember asking hasu about current draw on some of the hardware he has created so I sifted though some previous threads where i knew there were some real numbers. Here's what they were (for reference):
- Hasu's Alps64 pcb draws 15mA at 5V with current firmware configuration. (standard USB mechanical keyboard pcb, no wireless, no LEDs)
- Hasu's HHKB Alt Controller (has bluetooth for wireless) draws 50mA when not trying to charge a lipo battery. When trying to charge a battery it can jump up to 500mA (which is pretty standard USB/Lipo numbers)
Doing a quick search turned up some 5V 30mA 53X30mm solar cells
So, by going by these numbers it's certainly doable (at least in an operational state, to get to a charging state you might end up needing more solar cells than you'd like)
There are currently a bunch of wireless options for mechanical keyboards. Some are integrated into the pcb itself, some are external modules and some that can be implemented DIY. The biggest variable here will be efficiency, being, the most integrated is probably the most efficient (longest battery life).
Here's a nifty video (although a little long) about a solar R/C plane. He explains a bunch about solar cells, current, etc. It's not about keyboards, but all the concepts are the same.