Reading would actually be the easy part- drop a phototransistor in (optical)
line with the NE - pricy but doesn't affect the stored data.
I believe somebody made a neon matrix ROM -- have an electically
rectangular array of neons with some present, others not. By applying
suitable voltages to the X and Y wires, a particular location is
addressed and the neon there fires (if there is a neon at that location),
but none of the others do.
There was a photomultiplier aimed at the neons to detect the light flash.
I seem to rememebr there was the well-known problem that neons in the
dark don't fire reliably, which was got round by firing another neon just
before addressing a location (this neon could have a high enough voltage
applied to it to make sure it always fired reliably) and then ignoring
the extra pulse from the PM tube.
Of course this did not use the memory property of the neons themselves --
the data was stored by which neons were fitted.
I wish I could remember which machine used this. Maybe something like
EDSAC 2?
-tony