Some ideas in various stages of bakedness:
Silica-nucleating proteins (e.g. silicatein) might be used to make silica structures. Especially interesting would be if a proximal-probe-bound protein could be used to “grow” a silica shape, with deposition only taking place where the protein was brought into contact with the existing shape. In principle, this could be used to fabricate mechanically functional shapes, blurring the line between covalent solid machines and biomimetic systems. Eventually, it could be useful that silica could maintain its structure in a non-solvated regime.
Nano-manipulation (cf Zyvex) and electron-beam buckytube welding/cutting (cf Zettl and Banhart) may have reached the point where a 3D structure of buckytubes could be built up by feeding in a single buckytube, bonding it to the structure, then cutting to fit–somewhat analogous to lamp-worked glass beads. Similarly, what if the nano-manipulated tube was welded but not cut–could it serve to transmit force from the external actuator directly to the nanoscale? Of course the stiffness would be quite low, but that could be improved by pulling part of the structure against another part functioning as a stop.
I hypothesize that efficient biological protein machines balance their energy by means of entropic springs–easily tunable by evolution, but limiting the speed of operation. This hypothesis is testable, and has implications for protein study and for nanomachine design. See http://tinyurl.com/wnleq for more discussion.
With regard to Rob Freitas’s post: Positionally controlled fabrication has lots of dimensions of possibility: size of molecular building blocks or moieties; strength of bonds (weaker bonds can be compensated by parallel bonds); materials constructed; medium (polar or nonpolar solvent, supercritical CO2 or xenon, or vacuum) and of course flexibility of the deposition mechanism. Freitas’s proposal is near an extreme; very strong bonds, very small moieties, vacuum, relatively small set of operations. This does not make it bad; diamondoid appears to be a “sweet spot” in the space of machines. My point is that mechanosynthesis of structures is much broader than diamondoid or Drexler, and blends into approaches that don’t even require covalent chemistry.