The Mathematical Theory of Communication, by Claude Shannon and Warren Weaver, is a good place to start.
A programmed EPROM has a signal (in the conductivity of its memory
cell transistors) that can be read out, and has a large signal-to-noise
ratio. Erasure reduces the signal, until the logarithm of the signal-plus-noise
to noise ratio becomes zero.
There's a way to identi fy, from any given erasure efficacy, the amount
of information in the EPROM after erasure (Shannon gives the formula,
and Weaver explains it).
The usual circuitry in the EPROM will call all the erased bits '1' (all the words 'FF')
because that defines the erasure level that is complete (in the
hardware data sheet), so you can't expect the EPROM to simply plug into
a socket and read out the erased parts. In that sense, there's no
practical way to read the erased data, because the EPROM internal
readout amplifiers have a threshold below which all inputs are interpreted
as '1', which is the erased state as well as the unprogrammed state.
Even with signal-to-noise ratios in the zero range, however, one might
extraordinary readout procedures (like electron microprobing with
power applied to the opened-up package of the EPROM).
In theory, that allows one to
identify all the erased bits, and get more than half of those
identifications right. Guessing, one expects exactly half
right, of course.