My following prediction is extremely speculative and probably would not work, but a similar design may work. Current diamond research does not have the precision for it, and they create their diamonds by smashing layers of carbon together.

I predict that if the 3 degrees Kelvin background radiation can be blocked or very much reduced, and you can connect many atoms with single atom precision, then the following can be built:

A group of connected atoms (mostly carbon), contained in a large maze-like shield with at least 1 hole thats 1 atom wide that has a straight tunnel of carbon.
That tunnel allows only 1 type of laser light to enter.
That tunnel will lead to a battery-like group of atoms where the light bounces around while it waits for enough photons to be there (I dont know exactly how).
When enough photons are there, they would somehow be let out in the same laser pattern, pointing at a specific atom in a smaller structure floating inside the shell.
Position of the floating structure would be maintained by throwing photons in the opposite direction to move, and out a hole in the shell (Which would require 1-way tunnels).
That smaller structure has turing-machine parts and quantum-computer parts, and directs the light to detach atoms from specific known locations and attach them into other locations which are calculated, using the extremely precise single photon laser.
It can self replicate within the large shell, limited to the number of atoms inside the large shell.
At least 2 of the battery-like things are needed, paired on opposite sides of the floating structures, so constant position can be maintained.

A more practical design is for it not to build things and only use it as a quantum computer that takes laser inputs and gives laser outputs.

Actually, no. At that size (and consequently required energy level), with tools that can self-replicate, and multitudes of them can work on an assembly at once, it becomes an unbelievably cheap manufacturing technique.

Was any part of that research done with government grants? Doesn’t the public have a right to see what they are paying for?

Drexler and Smalley make the case for and against ‘molecular assemblers’
http://pubs.acs.org/cen/coverstory/8148/8148counterpoint.html

Debate About Assemblers — Smalley Rebuttal
http://www.imm.org/publications/sciamdebate2/smalley/

“How is it possible that an otherwise respectable publication would publish these attacks? None of their technical criticisms of molecular assemblers has withstood scrutiny — all have fallen by the wayside when it became obvious that they were incorrect.”

- http://www.imm.org/publications/sciamdebate2/

Another problem is that, more than twenty years after it was theorized, and despite millions of dollars of R&D funding trying to accomplish it, diamondoid mechanosynthesis has not been accomplished. It has only been theorized under some dubious Newtonian simplifications of quantum mechanics. Many chemists, such Nobel Prize winner Richard Smalley, think it won’t work.

I would so go far as to speculate that we’ll see true artificial (biochemical) life long before we’ll have large, self-replicating factories. The need to impose constraints is less at the scale of cells.

This may be a key component in the future of computer manufacturing.