Multiscale system engineering
From Wise Nano
This explores the engineering approaches to building large products from nanoscale components. See Nanomachine foundations for discussion of the starting assumptions: strong materials, nanoscale features, kW/mm^3 motors, nanometer-scale digital logic.
Contents |
Reliability through simple redundancy at multiple levels
Any macro-scale product with atomically precise components is going to have a lot of errors--from background radiation if nothing else. (An error is defined as an atomically precise component with an atom out of place. To be conservative, we assume that each such structural error causes the machine it's part of to stop working correctly.) A gram product might have a trillion errors. Obviously, the product must be fault tolerant.
We know how to build fault-tolerant systems with a few components. Just put in multiple redundant components, and some way to detect failure and switch between them. But does this work in systems with quintillions of components?
Yes, it does. For example, if the failure rate of a component at level N is less than 3.2%, then supplying one spare for every eight will be enough to make the failure rate at level N+1 less than the failure rate at level N. Of course, this also increases the system mass by 12% for each level where it is used; but if you start small enough, it need only be used at a few levels. See the Nanofactory paper for an application of this, including a program to calculate error rates at various levels.
Strength through anisotropy
In many materials such as glass fibers, tensile strength is limited by defects. Strain concentrates at the tip of a crack, propagating the crack until the material falls apart. This is a general problem with stiff homogeneous materials.
Metal is not brittle because, although it has lots of defects, they can migrate. Local migration keeps strain from concentrating. But the presence of migrating defects significantly reduces the ultimate strength of the metal.
From a defect/crack point of view, covalent solids appear to be more like glass than like metal. This would be bad news for a solid chunk of diamond. But we don't have to build solid chunks. In the same way that mostly-reliable machines can be combined with just a little redundancy in fault-tolerant systems, mostly-reliable fibers can be combined in anisotropic sheaves. A billion-atom diamond rod (10x10 nm by 50 microns) should be defect-free with very high probability (??%). Combine 100 of these in a sheaf, and the result should support ??% of the ideal (theoretical covalent bond) load with ??% reliability. ?? of these sheaves can be attached end-to-end for ??% load with ??% reliability, and then ?? of these constructions put in parallel for additional redundancy, resulting in a ?? mm cord holding ??% of ideal load with ?? reliability. Similar numbers hold for buckytubes.
Design flexibility through functional density
Any human-scale product will be mostly empty space (possibly including tankage). The reason is simple: with kilowatts per cubic millimeter of motor power, and petaflops per cubic millimeter (or more) of computer power, there will rarely be any reason to fill even a fraction of a percent of the structure with active components. Similarly, with materials 100 times as strong as steel, most of the volume of a human-scale product will not be structural.
Power and data transmission are similarly dense. A rotating diamond rod can transport 14 W/micron^2. Data transmission using logic rods can transport on the order of a Gb/s/nm^2, though with some delay for frequent (e.g. 5 micron) repeater stations. Micron-scale coax, or even nanometer-scale coax made of buckytubes, might be quite a bit better.
This implies that the functional equipment of a product can be basically pasted on the walls, which will themselves be very thin sheets either guyed and pressurized or backed by extremely lightweight trusses.
A few applications, such as medical nanorobots, distributed sensors, and unmanned spacecraft, may be sensitive to size. By today's standards, such things could be built extremely compactly simply by treating them as smaller versions of familiar manufactured machinery.
Macro-scale bearings
Although in theory a zero-static-friction bearing interface could be wrapped completely around a macro-scale shaft, radiation damage could be a problem. A square centimeter of graphite has 3.8E15 atoms, making it very likely to have an error. And one error in a sliding surface could propagate, releasing reactive debris that tear up the interface elsewhere. So this doesn't look like a very reliable solution. A 1-micron by 1-cm sliding band has 7.6E11 atoms in its two surfaces, so may be reasonably reliable. Separating the surface into bands should limit the propagation of errors. Layering each band into a concentric ring of bands would allow it to keep working even if one band siezed up. This deserves a closer look to see how long it'll last. Note this also looks like a good environmental seal, at least as long as the environment doesn't have lots of long sticky chemicals that could bridge the gap and be ripped apart.
Another way to make a large bearing is to suspend a surface on many tiny bearings. This allows occasional bumps (perhaps left by block-assembly manufacturing techniques) to be passed without much resistance. This does not completely resolve the concern about how (chemical) failures might propagate between roller and surface.
Taking advantage of scaling laws
In an electrostatic motor, power density is inversely proportional to size. This implies that whatever the final power desired, it will be best to produce it by ganging together a sufficient number of motors built as small as possible. (This has the advantage that once one motor is designed, it can be used for everything; only the gears need to be built bigger as the application grows.) In an application requiring gross motion with relatively little torque, the motors can
