The math is going to be back-of-the-envelope OOM analysis anyway, so you could work it out per bit or per unit area and multiply it out from there, given your particular configuration.

The statistics get more complicated when multiple bits are flipped (either by a single event, or over time) before an error correcting pass can look at the bits in question. In those cases, the particular flipped bits and where they are spatially will matter a lot more.