There's been worrying in this NG about the need to use artificial evolution techniques to build systems containing a number of parts (atoms?) that approaches Avagadro's Number (N_A), ie, about 6.2 x 10^23.
This is the number of particles in a mole of any substance. A mole of carbon-12 has N_A atoms, and masses (by definition) 12.000 grams.
That's a really, _really_ big nanomachine.
Drexler's plans for a jointed robot arm, about 100 nm long, contains about four million atoms. His conservative estimate for a complete mobile assembler robot is about a billion.
This is a lot of atoms, but remember that they're all built into (fairly) rigid diamondoid structures -- precisely so to minimize thermal and quantum noise in the system. His plans are very careful in defining structures so that atoms from, say, a gear, don't suddenly pop off and bond to the surrounding crankcase.
Now, say you're building a self-deploying emergency tent. This certainly contains many moles of atoms, but they're all built into small subcomponents, not much larger than a single assembler. You don't _have_ to worry about the interaction of all O(N_A) parts here; just the interaction of each part with its immediate neighbors (more or less). The problem is a lot simpler.
If you're trying to construct a rocket engine made of a single crystal of diamond and sapphire, I suppose you might have encounter some "emergent properties" that haven't been seen with the Hope Diamond.
And if you're trying to build a meter-cube super-duper nano-computer, with no modularity whatsoever, _then_ I'll admit you'll have a design problem.
But, IMHO, I don't see too many applications requiring anything so drastic.
-- =========================================================================
= PHILLIP THORNE, <thornp2@rpi.edu> URL: http://www.rpi.edu/~thornp2 =
= RENSSELAER POLYTECHNIC INSTITUTE (Always under construction!) =
= TROY, NEW YORK, USA "It's the boundary conditions that get you." =
=========================================================================