MM nanofactory architecture
From Wise Nano
Contents |
Advanced Nanofactory Architecture and Operation
Although any design for an advanced nanofactory must be tentative, the rough outlines of a reference design can be sketched out. The design presented here is derived from the Burch/Drexler planar assembly architecture23 and the Phoenix primitive nanofactory. This section discusses transport of blocks within the factory and placement in the product, control of the factory including a possible data file organization, block fabrication machinery, physical layout, power and cooling, maintenance of internal cleanliness, and reliability in the face of random damage.
The planar assembly structure proposed for the primitive molecular manufacturing system is also an effective structure for high-performance nanofactories. The blocks that would be attached to the growing product would be somewhat larger than in the primitive design, on the order of 100 nm to 1 micron. (A 200-nm block contains about a billion atoms.)
Blocks would be manufactured from small feedstock molecules, and then transported to the assembly face. Maximum product deposition speed would be on the order of 1 mm/second (block placement frequency increases as block size decreases, so deposition rate is independent of block size), though the rate of deposition would in many cases be limited by the rate at which blocks could be fabricated unless the blocks could be prefabricated. Feedstock and cooling fluid would be distributed to the input side of the nanofactory via pipes. Design and analysis indicate that a 1-kilogram factory might manufacture and deposit 1 gram of product per second. In a kilogram of nanomachines, errors are inevitable, so each section of the factory must be designed for error detection and redundancy.
Block delivery
Product deposition speed will be limited by three factors: block delivery speed, block placement speed, and block fabrication speed (unless blocks are prefabricated). A kilogram of blocks distributed over a square meter would make a 1-mm thick layer. However, the use of high-strength covalent solids would allow the product to have very low density; product structural members may be inflated volumes with walls just a few microns thick, or composed of lightweight truss or foam. Accordingly, it is reasonable to provide a large product deposition area even for a kilogram-per-hour factory.
For efficient operation, the fabrication of blocks can continue during times when the mass deposition rate is lower than the mass fabrication rate; and when high concentrations of mass are to be deposited, blocks can be delivered from the entire fabrication volume. This means they will have to be transported from the area where they are manufactured to the area where they are deposited.
Blocks will be large enough to be handled mechanically, and can be transported by any convenient mechanism. A layer of small “cilia†that pass the block overhead is mechanically simple and allows redundant design. Grippers are unnecessary—contact points for surface forces are sufficient.
Depositing a thin, solid column of blocks at maximum speed requires blocks to be delivered from all over the factory to a small area. This determines the thickness of the routing structure, which is roughly the same as the desired thickness of the deposited part divided by the ratio of delivery speed to deposition speed. For example, to allow deposition of a 1-cm solid block at 1 mm/s, if the blocks can be moved internally at 1 cm/s, then the routing structure must be 1 mm thick.
Factory control and data architecture
There are enough different ways to implement digital logic that it seems safe to assume that any system capable of building a nanofactory can also include large numbers of digital computers. General-purpose (microprocessor style) computation is energetically expensive because bits must be erased at an irreducible cost of ln(2)kT per bit, so it will be infeasible to use many CPU cycles per atom in a high-throughput kilogram-scale nanofactory. But special-purpose logic (state machines) can be used to good effect for repetitive feedstock-placement tasks, and need not erase bits. To plan the handling of larger blocks (many millions of atoms), general-purpose computation will add minimal cost.
The control system will be deterministic and detailed, making it possible to specify any nanoblock at any position. The blueprint can still be small, since one nanoblock design can be re-used at many locations. Arrangements of nanoblocks can also be copied (tiled) and used to fill solid geometry volumes.24
There are only about 1015 100-nm blocks per gram, so with the computing resources that can be included in a kilogram-scale nanofactory, it will be quite possible to plan the path of each individual block from where it is fabricated to where it will be placed.
The blueprint/control file will be broadcast equally to all parts of the factory; this allows a large number of local computers to be accessed with a simple network design. The control file will be sent in two parts. First will be hierarchical solid geometry descriptions of the product, which describe the block patterns in each volume of the product. This information will instruct each fabricator as to what block pattern it needs to make. The number and type of blocks required for each layer can be calculated in advance. These can then be distributed over the capacity of the fabrication volume to keep all the fabricators busy.
The distribution planning is complicated by the fact that at different times, concentrations of mass in different parts of the product may draw blocks from all over the factory. Pathological cases can be designed in which it is impossible to deposit blocks at full speed. However, preprocessing (generating broad plans for each product) along with tolerance for mild inefficiency will allow block production and delivery to be planned with simple deterministic algorithms. Each fabrication region will be able to compute exactly what point in the product volume its block is destined for, and when the hierarchical geometry/pattern description of the product is delivered, each fabricator will be able to identify exactly which block type to build. (A processor's eye view: "I'm 54,925.2 microns from the edge of the product... that means I'm 2,142.6 microns from the edge of sub-pattern K, whatever that is... Here comes the spec for K... I'm in sub-sub-pattern XB, 152.4 microns in... That means I'm building block type KXB4W.")
Fabrication instructions for individual blocks will be delivered next. Each fabricator will know which part of the instruction stream to pay attention to. In practice, a local computer will likely control multiple fabricators, parsing the instruction stream and sending appropriate instructions to each fabricator it controls. A few redundant fabricators for each computer will allow broken fabricators to be left idle. Blocks will be built sequentially, so local processors will not need to remember the entire instruction stream. They will receive a string of instructions, and place the next ten thousand atoms while instructions for other block types are delivered. Again, it is possible to design pathological cases where this doesn't work well, such as all block types requiring complex instruction at the same point in each construction sequence, but this approach typically should be reasonably efficient.
This plan assumes that all fabricators will be working in parallel. If the fabrication mechanism allows blocks not to be built in lockstep (see next section), then blocks with simple blueprints (which can be remembered by local computers) can be built out of step. Alternatively, the blueprint may be sent with several different timings. This can be accomplished by having broader communication channels to each local computer. Or different fabricators can be hardwired to different communication channels to get blocks ready at different times. Blocks will probably take minutes to hours to build, which is enough time to transmit many gigabytes of data.
Block fabrication
The most flexible way to build large (million- to billion-atom) blocks from molecules is to use a general-purpose manipulator system to add molecular fragments one at a time to the block. In this scheme, the blocks would all be built in lockstep, and deposition would start after fabrication was finished. This is fast enough to provide high performance: scaling laws indicate that it might take an hour for a single 100-nm manipulator to build a billion-atom (200-nm) block one atom at a time. But the delay before the first block is finished could be several times longer than the time needed to deposit the product.
One way to speed up the process is by building several block components in parallel, either at general-purpose workstations or on special-purpose fixed-path mills, and then combining them to form the block. The sub-parts could be either general-purpose parts, such as lonsdaleite cylinders to be added to a diamond crystal, or special-purpose parts like computational elements. Either way, this could speed up block construction severalfold, allowing deposition to begin sooner.
Of course, the fastest way is not to manufacture the blocks in the factory, but to prefabricate them as described in the "Primitive" section, using any convenient combination of synthetic chemistry, self-assembly, mechanosynthesis, and simple mechanical assembly. A variety of design options is available as to what size the input blocks should be, how they should be delivered (solvated or clean and packed), how they will be fastened, and whether the nanofactory will include an intermediate block-assembly stage before depositing the blocks onto the product.
Physical architecture
The product is deposited from a block-placement plane which is studded with manipulators to push the blocks onto the product surface. The manipulators may hold the growing product, making it possible to build several disconnected parts that would be fastened together after further deposition. Alternatively, the product may be held externally; this would require disconnected parts to be connected by temporary scaffolding.
Between the manipulators are holes through which blocks are supplied from the routing structure. The routing structure is composed of braided or interwoven block delivery mechanisms (probably cilia). This allows blocks to be shipped crosswise, routed around damaged areas, and so on. Power and control for the delivery mechanisms, as well as for the placement plane mechanisms, run inside the delivery structure.
Below the routing structure is the fabrication volume. This contains most of the mass of the nanofactory. It will be arranged in long, thin fins with 1-micron gaps between them for cooling/feedstock fluid to circulate; cables or tension struts will bridge the coolant gaps to resist coolant pressure. The interior of each fin will be hollow, providing workspace and space for transporting parts and blocks. The walls will be lined on both sides with fabrication systems. The fin may be 4-6 times the width of a completed block--on the order of a micron--and about 2/3 empty space; added to the fluid channel volume, this means that the density of the fabrication volume will be about 0.1 g/cm3. For a kilogram-scale factory with a square-meter deposition area, the fabrication volume will be about 1 cm thick. This provides about 10,000 square meters of surface area for feedstock intake and cooling; diffusion and heat transfer will not limit the nanofactory speed.
Below the fins are fluid supply and return ducts. Low-viscosity fluid can flow a distance of 3 cm at 1 cm/s through a 1-micron wide channel with a pressure drop of 6 atm, so the fluid can be injected from the ducts, flow along the fins to the top of the fabrication volume, and return, without excessive supply pressure. No fractal plumbing is needed for a factory that manufactures 1 kilogram per hour per square meter.
Because the physical architecture of the nanofactory is planar, with feedstock intake and processing located adjacent to product deposition, there is no need to change any dimension of the factory's nanoscale components in order to increase the manufacturing capacity and deposition area. In effect, multiple square-meter designs can be abutted to make as large a factory as desired.
Power and cooling
Most of the energy used by the factory will be used in the block-fabrication area, since handling the smallest components (feedstock molecules) will require the majority of operations. Fortunately, this is the area that is closest to the cooling/feedstock fluid. A cooling fluid made of a low-viscosity carrier of small encapsulated ice particles can provide very efficient heat removal; a flow of 1 liter/second can cool 100 kW—more than an advanced nanofactory will need. Cooling by phase change also has the advantage of keeping the whole factory at an almost uniform and constant temperature. Because only about 1 gram per second of feedstock is needed, feedstock molecules can be dissolved in the cooling fluid at about 1000 PPM.
Within the nanofactory, power can be distributed very efficiently at all scales by rotating shafts. Electrostatic motor/generators can be used to interface with an external electrical power system.
A nanofactory manufacturing a kilogram per hour and drawing 1.4 kW, probably the upper limit for an air-cooled “desktop†model, would have an energy budget of about 100 zJ (10-19 J ) per atom. Achieving this would require recovering the chemical energy of bonding operations, since a single bond contains several hundred zJ of energy. Recovering energy from chemistry requires controlling the reactions with machinery that is stiffer than the bond strength, so that the reaction can pull the machinery along smoothly. This is thought to be possible with advanced design. A less advanced design would require an external cooling system.
Clean internal environment
The internal environment of the factory must be kept clean of contaminant molecules that could cause undesired reactions or jam the moving parts. The factory interfaces with the environment at feedstock delivery and at product deposition. The feedstock molecule intake mechanism will deterministically manipulate the desired molecules, which provides an opportunity to exclude other molecules. This will be relatively easy if the molecule is small, like acetylene, or at least compact so that a shaped shell closed around it can exclude all solvent molecules. Small feedstock is preferable for several other reasons as well, including feedstock cost and flexibility of manufacture.
Environmental contaminants can be kept out of the product deposition mechanism by extruding a thin shell or sealing sheet to cover the product and any unused area of the deposition surface. Before the product is removed, a second covering must be deposited to seal the deposition surface.
Reliability
Most of the mass of the nanofactory will be used for manufacturing blocks from feedstock. This implies that many fabricators per second will be damaged, and a percentage of blocks under construction will not be completed. This can be dealt with by building duplicates of all block types. Since the number of block types must be far smaller than the total number of blocks, this is not too onerous. Excess good blocks can be retained for later use or added to a reserved "dump" volume of the product. Damaged or incomplete blocks in a damaged fabrication area can be retained there permanently, since that area will not be used again. (Self-repair seems likely to add more mass and complexity than redundancy. If damaged parts need to be moved and stored, the required dump volume would be small.)
Blocks damaged after fabrication while in transit to the product assembly surface need not be replaced. The product's design will need to cope with radiation damage immediately after manufacture, and its lifetime will be far longer than the manufacture time; therefore, its design must be able to deal with a small fraction of damaged blocks. Missing blocks could pose a larger problem, but radiation damage will not significantly change a finished block's physical shape, so the block should still be able to be added to the product.
The transport mechanism will consist of many redundant arms / struts / cilia that work together to move the block along surfaces. The random failure of a small percentage of these will not compromise the ability of the rest to transport the blocks. If a patch of them fails, blocks can be diverted around the area.
There is limited room at the planar assembly surface, and the robotics may be more complex than for block transport. However, the volume of radiation-sensitive machinery is correspondingly small. A square-meter area, 100 nm thick, and with machinery occupying 5% of its volume, contains about 5 milligrams of machinery. The entire volume may be hit about 4 times per millisecond. If repair requires a few microseconds--which may be plausible given the small scale and correspondingly high operation frequency--then the entire placement operation could stop during each repair without greatly reducing performance. A simpler solution is to make the placement machinery flexible and large enough that machines can reach to do the work of a disabled neighbor. This would slow down the deposition slightly, but if most machines in an area were undamaged, then many machines could participate in taking up the slack by each doing some of their neighbor's work to distribute the load of the missing machine.
As discussed in sections 6.1 and 8.5 of the “primitive nanofactory†paper, computers can be made as redundant as necessary.
