When I was an undergrad I took a 6 week 6 hour per day winter session class about DNA manipulation (cloning, sequencing, restriction enzymes, etc). Sometime during those six weeks I decided there had to be a better/faster way to sequence DNA, and over the next couple of years I fleshed out some concepts.

The basic idea is that a myosin 'engine' pulls a strand of DNA along a actin-lined channel and sensors in that channel would read each nucleotide as it passes by. The DNA is single stranded and has been cut with a particular restriction enzyme and a matching piece of DNA is bonded to the actin engine. ATP would be in the solution for motive power. Physically, I picture two small wells separated by a membrane, or thin wall, with actin channels connecting the two. Larger areas surrounding the actin channels may be needed to insure a good supply of ATP, perhaps capped at one end to prevent/slow diffusion of DNA.

The sensors could be one or both of:
A protein that has one of the four nucleotides bonded to it, or in some other way can recognize a particular nucleotide. Confirmational changes when it matches a nucleotide would have to be picked up by some larger structure (more proteins, or maybe a short cascade of enzymes which produces a small current across an artificial membrane).
Or, something like an atomic force microscope tip could be mounted in the channel and used to recognize nucleotides.
The sensors may not have to be 100% accurate. Given the small scale, a large number of sensors could be placed in each channel and statisical methods could be used to arrive at a sequence. And each strand of DNA could be read multiple times, until a desired degree of accuracy is reached.

Questions/Data needed