There was a paper in the last few years proposing to make DNA a practical mass-storage technology, as the last and slowest level of the hierarchy. It'd need lots of error correction, and to be chunked up into small-enough strings to synthesize, and each string would need to carry address bits to match on (unless you want a content-addressable store). But IIRC the storage density and the cost trends made it at least kind of plausible.

And of course there's http://e-drexler.com/d/06/00/Nanosystems/toc.html chapter 12.