Nanomanipulation, or positional control at the nanometer scale, is the first step towards molecular nanotechnology. Reactions must happen precisely where we plan for them to. We must provide highly accurate and repeatable positioning of a tool and workpiece with as many degrees of freedom as necessary. Three degrees of freedom (DOF) for the system is probably not enough, while six DOF for each of the tool and the workpiece is probably overkill.
It might be possible to self-assemble nanotechnology in a vat, the way life seems to have developed, but it seems more likely we will get where we want to be with positional control. Self-assembly will be important for some of the building blocks, but it seems exceedingly unlikely that we could get interesting structures, such as bearings, to self-assemble. Flagellar motors are an interesting existence proof that even this is possible, but we desire small, stiff, precise structures, and would like to design them, rather than evolve them.
Today's Atomic Force Microscope (AFM) is capable of suitably small motions (tenths of an angstrom), but cannot repeatably position to, and hold, sub-angstrom accuracy due to hysteresis and drift. It is also currently limited to three degrees of freedom, with no rotational control. As a scanning probe, it works quite well; as a construction device for nanotechnology, it is a bit like using a pencil to construct a car.
A useful positional control device must be able to grip, position to high accuracy, exert force, and release, all under external control, and with several degrees of freedom. It can be attached to a substrate or free-floating in solution; our current guess is something attached to a substrate that provides power and control.
Given such a device, along with a basic mechanosynthetic repertoire, system design to get materials to and from the synthesis site, and some kind of control and power distribution scheme, the pieces are in place to build a practical assembler.
------------------------------------------------------------------------
There is a considerable interest in building micromachines today; these are manufactured in bulk silicon by using techniques borrowed from semiconductor manufacturing. Such micromachines suffer from being at a scale where friction is very high, due to the relative roughness of sliding or rotating surfaces and inability to lubricate such surfaces. Above this scale, surfaces can be made relatively smooth, and lubricated. Below this scale, with precise placement of atoms relative to one another, friction has been shown to be almost nonexistent in well-designed devices (see Nanosystems).
The challenge of intermediate-scale positional control devices is to either minimize frictional contact (for example, in Texas Instruments' Digital Micromirror Display, movement is by electrostatically applied torsion, not sliding interfaces), or to build such systems out of atomically precise building blocks. The former approach can be utilized today, the latter appears to be a fruitful area for development.