His plan is to take a silicon surface, carefully terminated with one layer of hydrogen; use a scanning probe microscope to remove the hydrogen in certain spots; hit it with a chemical that will deposit a single additional silicon layer in the "depassivated" areas; and repeat to build up multiple layers. As long as the scanning probe can remove single, selected hydrogens -- and this capability has existed for a while, at least in the lab -- then this should be capable of building 3D structures (or at least, 2.5D) with atomic precision.
This is interesting for several reasons. First, the approach has a lot of possible variants. You could use multi-tip arrays for higher throughput. You don't need atomic precision to make the basic approach work. You could use materials other than silicon -- silicon carbide was mentioned (this might be a way to avoid the silicon atoms wandering around the surface). It might be possible to use light to depassivate the surface, allowing higher throughput (though of course with less precision).
So this approach appears to fit into a family of technologies that form a continuum grounded in what's available today. The atomic-precision version seems a reasonable goal for a single smallish company to develop in a decade or less.
So what does this mean for molecular manufacturing and the tabletop nanofactory revolution?
APM is not a nanofactory. It would not be able to build large products, or even small products, or even small machines -- it would be able to build nanoscale structures, and that's it. Now, you can do a lot with nanoscale structures, so this is a strong enabling technology for nanofactories. With additional R&D, it would probably be possible to develop structures that would function as bearings. With enough work, you could probably build up a toolkit of APM-built structures that could be combined into computer logic, perhaps sensors, maybe even motors... Of course, you'd also have to integrate manipulators to take the structures and assemble them into the machines.
One interesting possibility is that, if you can build silicon shapes, you can build MEMS -- perhaps even the same type of scanning probe array that would form the heart of a parallel APM system. So the system might be able to build one of its more intricate components. Of course, there are many components that it would not be able to build. So it would not be exponential manufacturing; but it could represent an interesting step in that direction.
Tomorrow, I'll write about how -- paradoxically -- this might actually slow down the development of molecular manufacturing.