Are there companies working on such a project? Would this allow us to gain orders of magnitude of value in terms of performance.
Note: My question concerns reusing existing base manufacturing technologies, I'm not talking about something like Quantum Computing.
It is not out of the question that some silicon blueprint could or does result in a component which can safely distinguish between more than two states.
We could, but I don't think there's much of a point. 1s and 0s are easy to work with because it's simple to represent their absolute state with complete precision. Storing, retrieving, analyzing and operating-on non-binary data is computationally expensive for minimal gain in performance.
The end user hardware isn't perfect, but you'll be hard-pressed to match the efficiency of a classical computer.