Complex products overview
From Wise Nano
This page explores the steps necessary for building large-scale nanosystem-based products. The analysis covers all physical stages from building molecules to products.
At each level, several options are presented. Ideally, it will be possible to mix-and-match, providing increased design flexibility and probability of success.
The manufacture of large, heterogeneous, high-performance, useful, engineered products will be very difficult without at least one of the low-level steps (topics 1 or 2) being programmable--that is, carried out under direct computer control of sequential operations with lots of choice points.
Nanofabrication options that involve large machines and low throughput are not listed here, because they cannot be used in producing large products. Arrays of MEMS-built scanning probe microscopes are an attempt to bypass this problem, but it'll take some convincing before I'll consider them even potentially viable. (Why not? - we need to discuss this)
Contents |
1) Building molecules (machine parts)
Carbon lattice Mechanosynthesis
Pros:
- Parts can be
- Large
- Complex shapes
- Stiff and strong (high covalent bond density)
- Mechanosynthetic manipulator can also position/assemble parts
- Process is relatively fast (billion-atom nanoblock in hours, even without mills)
- Programmable
- Diverse material properties/functions possible (carbon or silicon lattice with a wide variety of dopants)
Cons:
- Not demonstrated yet
- Requires vacuum
- Requires complex machinery (bootstrapping problem)
Buckytubes
Pros:
- Stiff
- Simple to build in bulk
Cons:
- Requires additional chemistry to join
- Requires unknown techniques to sort
Nucleic acid
Pros:
- Can be built in bulk
- Programmable shapes
- Lots of organic chemistry options/derivatives
Cons:
- Expensive
- Relatively low strength
- Relatively high feature size (a loop requires several nucleotides)
Protein
Pros:
- Stronger than nucleic acid
Cons:
- More expensive than nucleic acid
- Hard to engineer/program shapes (though this may be changing)
- Structure is not stiff/precise
Schafmeister's stiff polymers
Abstract of talk at 2004 Foresight conference
The idea is simple: synthesize special amino acids, string them together, then double-link them so the chain stiffens, implementing a definite structure: a twig, rather than a string.
Pros:
- Feature size is the same as that of proteins.
- Shape is programmable.
Cons:
- A little more expensive than putting together proteins.
- Slow synthesis; 1 monomer per hour IIRC (at least for now).
Note (Added by Christian Schafmeister): 1 hour coupling speed really isn't a problem.
We plan to use our method to synthesize macromolecules between 1,000 and 10,000 Daltons. That is the size of small proteins. These macromolecules can serve as building blocks for mechano-synthesis and for self-assembly of much more complicated devices.
Mechanosynthesis of polymers
To date, with one exciting but toy exception (Ned Seeman's programmable DNA-joiner), new polymer sequences have been built by bulk chemistry. What if it were possible to protect or deprotect the growing ends of sequences, or catalyze only certain reactions depending on mechanical positioning or actuation, so that a general-purpose chemical bath could build individually programmed molecules?
If a promoter catalyst were used, rather than a blocking approach (sterically hindering a reaction that otherwise would've happened), this could substantially increase synthesis rates.
2) Building nanoblocks (nanomachines)
Nanoblock Contents
First question: How big is the nanoblock?
What machines and structures can be built into a nanoblock? How general purpose? Parameterized or limited parts count? (BTW, I figured out a cool way to solve the smooth curved surfaces problem from limited number of parts - how do I include jpgs?)
- Structural (Static and Dynamic - e.g. skeleton vs. controlled pore-size membranes)
- Solid State: Thermal (conductive or insulating) and Electronic (conductive, insulating, and semiconducting)
- Actuators
- Sensing devices
- Logic
- Fastening systems to join to other nanoblocks/parts
What environment does the machinery need to work?
- Water
- Solvent
- Inert gas or vacuum
- air
What performance can be expected from the machinery?
- Actuation and motion speed (linear and rotational)
- Precision
- Efficiency
- Strength (compressive and tensile)
Robotic assembly
Pros:
- Good fit for diamond mechanosynthesis
- Programmable
Cons:
- Requires complicated robots, not yet demonstrated
Self-assembly
Pros:
- Simple (Production, by putting into a box and shaking, yes; design however is very difficult - a large number of 3D Wang cubes require an exponentially increasing number of smart glue oligomers.) (Maybe I should study this a bit more formally to show how difficult it is. - ttf))
- Fast
Cons:
- Requires recognition sites (e.g. half-DNA strings)
- May be limited to building smaller blocks (diffusion time, etc)
Top-down (ie lithography)
Pros:
- Established technology
- May be good for proof-of-concept demos
Cons:
- Low performance
- Imprecise features
- Cost
- Requires machines that can't be built with litho; thus, not self-contained
3) Combining nanoblocks into products...
Self-assembly
(mentioned only for completeness: I don't think this can work with large blocks, and it certainly can't make heterogeneous products.) (I suspect it can, but with great difficulty -ttf)
3D printing
Load nanoblocks into an inkjet or something similar; deposit them.
Pros:
- Established technology
Cons:
- Poor placement
- Very poor alignment
- Poor to very poor inter-block connections
Fluidic assembly
Blocks are washed along by fluid, which can align them, until they "fall" into their appropriate position. Mentioned for completeness; more or less a cross between self-assembly and 3D printing.
Cellular product
(Note: the words "cell" and "cellular" here have nothing to do with biological cells.)
The most famous cellular product in the molecular manufacturing community is utility fog. At MIT, it would probably be amorphous computing. The idea of a cellular product is that the factory doesn't have to produce the structure. The micro- or nano-scale components can form their own structure, either mechanically or programmatically.
Cellular products may have advantages, such as the ability to flow through cracks, but they will also have disadvantages, such as not being very strong for their weight. (Utility fog is comparable to plastic.)
Control of a mechanical cellular product is likely to be difficult. And if it's possible to pack enough robotics into a nano-built cell to let it move around, surely it's possible to pack enough into a factory's assembly face to move nanoblocks into position. So a cellular approach will not be necessary, though it may be useful in the very early stages before a complete factory is designed. A robotic assembly factory (see next section) could easily produce a cellular product.
Robotic assembly
This has several sub-variants, including convergent assembly (small cubes make bigger cubes, repeatedly), working face aggregation (small parts added directly to full-scale product, conceptually similar to but more precise than 2D->3D inkjet rapid prototyping tools), and general-purpose robotics (handling large heterogeneous parts and a variety of assembly operations).
Working face aggregation appears to be better than convergent assembly in every way except speed, and the speed appears adequate as long as the blocks are on the same scale as the handling robots--that is, if a single block can contain a robot. At that point, scaling laws say size doesn't matter. If robots are made of many blocks, then divide (I'm guessing) 1 cm/sec by the number of blocks per robot for a rough estimate of product assembly speed.
The advantage of general-purpose robotics is that you can use large, odd-shaped parts. But that only matters if you can build large, odd-shaped parts. There's no known way to do this from the bottom up that doesn't involve attaching smaller blocks to each other; and if you can do that, then you can probably build your odd-shaped part in pieces and attach the pieces one at a time using working-face aggregation (or convergent assembly).
4) ...including nanofactories that can do 1, 2, and 3
A nanofactory is a large (human-scale) machine that takes in simple physical inputs and complicated blueprint files (and power, unless it burns some of the chemicals), and outputs human-scale nano-structured product.
Ideally, the nanofactory would take in single molecules and use it to build a complete nanofactory. But earlier versions may (for example) use complex molecules such as short strings of nucleic acid, fastening them together in programmable ways.
Thus early nanofactories may or may not do any of the three steps:
- Chemical processing
- Nanoblock fabrication
- Product assembly
However, at least one of these steps will need to be programmable, and thus require a nanofactory or robot. Otherwise the product will be either small or unstructured. Non-programmable steps don't need a nanofactory; they can just happen in a test tube.
In theory, a kinematic cellular product (or modular robot) could be built by bulk chemistry followed by self-assembly. This would imply that no nanofactory would be needed at all. But it's hard to imagine that a cell built using no programmable steps could be complex enough to be a robot capable of forming useful products.
In order to do programmable nanoscale operations, the nanofactory will have to have large numbers of nanoscale features. In other words, it will have to be built by a nanofactory. (This recurses back to the smallest nanofactory, which can be built by non-scalable processes.)
So this means that, whatever the nanofactory does, it will have to output product that can do that. <huh? what is "that"? -ttf>
- For chemical processing, the product must be either water/solvent-tight or vacuum-tight.
- For mechanosynthetic operation, the nanofactory must contain nanoscale actuators with chemically useful tips.
- For chemical-sequence operations (e.g. to build biopolymers), the nanofactory must control the flow of chemicals through the workspace. (I expect this to be too slow and error-prone to actually be useful.) <this sounds like directed self-assembly -ttf>
- For programmable nanoblock fabrication, again nanoscale actuators (capable of handling molecular parts) are needed, though with less precision and perhaps more range of motion than mechanosynthetic work.
- For self-assembled nanoblocks, the factory doesn't have to do anything.
- To assemble product (assuming the product isn't kinematically cellular), the factory would have to pick-and-place blocks so that they were joined. This appears quite simple, assuming the joint mechanism doesn't require twiddling. It's conceivable that a few simple blocks, self-assembled, could make a structure that would grab blocks from solution and place them in programmed patterns (similar to sorting rotors that take the random solute parts and control their position, orientation, and timing for the next step).
5) How the products work
Assume nanoblocks are placed and fastened in programmable 3D array. What can you do with that?
(This section is unfinished.)
Right away we get into further assumptions that will feed back and constrain earlier choices. First, it'd be really nice to be able to send electricity through wires--though nanoscale wiring doesn't scale very well: resistance increases, so that a 1 nm cube of copper-equivalent has a resistance of 17 ohms. But for power, larger conductors would be used; a 1-mm square-micron wire would again have a 17-ohm resistance. Surrounding that wire with 50 nm of diamond insulation (increasing its area by 21%) would allow it to carry 100 volts, thus ten milliamps per watt of power transmitted, thus losing about 0.2% of power per millimeter. <What about conductive buckytubes and superconductors? - ttf>
In carbon lattice, electrical actuation is pretty easy; electrostatics work great at the nanoscale. In organic chemistry, electrical actuation might invoke redox reactions.
Power could also be transported mechanically; a quickly rotating diamond rod transmits power on the order of watts per square micron. But high speed would cause significant frictional losses, which still need to be calculated.
There are lots of ways of implementing digital logic and thus computers. Efficiency would not be a problem for computers on the order of today's desktops. Theory says that rod-logic switching at 100 pHz should enable Earth Simulator-class massively(!)-parallel machines to be run for two watts.
