hasu wrote: How about adding thin film between membranes as an insulator and sensing capacitance?
I don't think the pads are large enough. Capsense is tricky business, if I've learned anything reading wcass's and idollar's design threads. Things work best with large pads, small traces, and minimal overlap. The Model M membranes fail on all three counts. It's a neat idea though.
A diode needs to be in series with each switch for true NKRO. A thin film of diodes is certainly the right approach. The secret ingredient is niobium silicide.
This was demonstrated in 2014:

From:
http://www.pnas.org/content/111/33/1194 ... nsion.html
See also:
http://www.pnas.org/content/111/33/11943.full.pdf
Good luck making one. It's a long shot just for an NKRO retrofit for a Model M but it could enable arbitrary rubber dome matrix boards to be NKRO. Why in large quantities it could even be cost saving.