An article by Jamais Cascio makes a good start at describing three levels of nanotechnology: nanoparticles, nanomachines, and nanofactories. Astute readers will note that the title of this blog entry replaces "nanomachines" with "nanodevices." I'll explain why later on.
There are a lot of "nano" concepts to be sorted out. They have to be plotted in two dimensions, ranging from primitive to advanced, and from mundane through fantastic to impossible. Jamais's article, though quite useful, does not draw enough distinctions to clear up all the confusions.
By "mundane," I mean things that are completely consistent with physics and engineering as they are understood today; things that may require a lot of work to develop, but don't require new scientific breakthroughs; things that we could start designing and developing, and in some cases already have.
A primitive example of a mundane nanotechnology is nanoparticles. A more advanced example is nanodevices that are starting to be developed today: active things like transistors and molecular actuators. Nanomanufacturing is a term for early attempts to take a more active role in building more intricate nanostructures -- though not necessarily using molecular bonds or molecular precision. This is gradually being developed in order to build better nanodevices, reflecting the realization that it's not enough to make small stuff -- to be really useful, the stuff has to be intricate (information-rich), and this requires more than bulk processes (including pure self-assembly).
Perhaps the most advanced mundane nanotechnology that has been proposed is the nanofactory outlined in 1992 by Eric Drexler (Chapter 14 of Nanosystems) and fleshed out over the last few years by Burch and Drexler, and by CRN projects. Calculating the capabilities of nanofactories is the best way I know to understand what molecular manufacturing will be capable of -- at least at first.
A lot of less mundane (fantastic) nanotechnologies have been proposed: things that would be very powerful or useful, that are probably doable, but that we don't know enough to really start designing yet. A lot of nanomedical devices fall into this category. We can speculate about them, sometimes in more or less detail, but we can't really finish the design until we learn more about how the body works. Some of the more advanced aerospace proposals fall into this category.
Nanobots fall somewhere between fantastic and impossible. Science fiction hardly ever manages to get it right; even when the authors try to be physically plausible, they usually mess up some constraint like heat or raw materials or operation speed.
Before Drexler developed the nanofactory concept, he had an earlier idea of small cooperating fabrication robots, which he called "assemblers." The operation of assemblers was never described in detail, and making them work together would require major advances if not breakthroughs in robotics. In short, I would have to classify them as non-mundane -- fantastic -- though probably not impossible. Unfortunately, the assemblers became almost immediately mythologized as "universal" nanobots that could do nearly-impossible tasks like efficiently pulling apart unsorted raw materials to rearrange the atoms into new products.
This is why I suggest that "nanodevices" is a better word than "nanomachines" to describe the evolution of practical nanotechnology. To me, "nanomachine" sounds potentially a lot more complicated and functional than "nanodevice." This is reflected in Jamais's article, where he describes two things that he classifies as nanomachines. One is a simple molecular actuator that's being developed today. The other is Drexler's superseded semi-fantastic assembler concept. It's better to use a word like "nanodevice" that helps to make it clear that the simple devices being developed today have no relation to assemblers or other nanobots.
In just a few years, with only the general guiding principle that small novel stuff is likely to be worth studying, the National Nanotechnology Initiative has moved from nanoparticles to nanodevices and nanomanufacturing. (By the way, I don't like the term "nanomanufacturing" because it is confusing: it sounds like it includes nanofactory-level technology, although it is actually a far more modest and special-purpose set of technologies. But the NNI has adopted it, so we're stuck with it.)
The natural course of nanotech development is to fasten together smaller particles, with more precision, with stronger bonds, with more programmability, to make more intricate devices. In perhaps 20 years, work funded by the NNI could lead to nanofactory-level technologies without much planning or direction. However, it's worth talking about nanofactories and their consequences today, for two reasons. First, they will be easier to develop than most nanotechnologists realize -- and because a few do realize it, nanofactories might be developed early and take a lot of people even in the nanotech community by surprise. Second, they are an extremely powerful tool for nanoscale R&D, and at the same time an extremely powerful tool for making advanced products.
As nanotech develops, it will become clearer that there is a continuum from nanoparticles, through nanodevices, to nanofactories. There are times when it's important to make the distinction clear: molecular manufacturing is fundamentally different from nanoscale technologies in many ways, both in approach and implications. But it is becoming increasingly possible and useful to chart the evolution of goals and the development pathways.