I think the best way to do this is to research out completely how biological cells work, then reverse engineer them. Cells clearly self-replicate and can form a wide variety of systems (e.g. multi-celular life-forms). Biology, being based on cells, is modular. Any kind of industrial nanotech would have to have similar modularity to be scalable from the nano-level to making macroscale products. Biology, being modular, has the needed redundancy built-in, where every cell does not have to be perfectly functional for the whole system to work. Within each cell, there is built-in redundancy and modularity in the celullar processes that build and power an individual cell.
It seems to me that biologically-derived nanotech is the way to go. Needless to say, I am very sceptical of any kind of “dry” nanotech concept. I have worked with AFMs in the past and find them a very useful analytical instrument, but use for nanofabrication involves scaling issues that are very difficult.