New! 

Nanotech Scenario Series

 Join the  conversation at CRNtalk!

CRN Science & Technology Essays - 2006

"Four stages of acceptance: 1) this is worthless nonsense; 2) this is an interesting, but perverse, point of view; 3) this is true, but quite unimportant; 4) I always said so." — Geneticist J.B.S. Haldane, on the stages scientific theory goes through

Each issue of the C-R-Newsletter features a brief article explaining technical aspects of advanced nanotechnology. They are gathered in these archives for your review. If you have comments or questions, please email Research Director Chris Phoenix.

  1. Powering Civilization Sustainably (January 2006)
  2. Who remembers analog computers? (February 2006)
  3. Trends in Medicine (March 2006)
  4. Bottom-up Design (April 2006)
  5. Types of Nanotechnology (May 2006)
  6. History of the Nanofactory Concept (June 2006)
  7. Inapplicable Intuitions (July 2006)
  8. Military Implications of Molecular Manufacturing (August 2006)
  9. New Opportunities for DNA Design (September 2006)
10. Recycling Nano-Products (October 2006)
11. Preventing Errors in Molecular Manufacturing (November 2006)
 

     2004 Essays Archive
     2005 Essays Archive
     2007 Essays Archive
     2008 Essays Archive
 

Preventing Errors in Molecular Manufacturing
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

What kind of system can perform a billion manufacturing operations without an error?

Many people familiar with today's manufacturing technologies will assume that such a thing is near-impossible. Today's manufacturing operations are doing very well to get one error in a million products. To reduce the error to one in a billion--to say nothing of one in a million billion, which Drexler talks about in Nanosystems--seems ridiculously difficult. None of today's technologies could do it, despite many decades of improvement. So how can molecular manufacturing theorists reasonably expect to develop systems with such low error rates--much less, to develop them on a schedule measured in years rather than decades?

There are physical systems today that have error rates even lower than molecular manufacturing systems would require. A desktop computer executes more than a billion instructions each second, and can operate for years without a single error. Each instruction involves tens of millions of transistors, flipping on or off with near-perfect reliability. If each transistor operation were an atom, the computer would process about a gram of atoms each day--and they would all be flawlessly handled.

The existence of computers demonstrates that an engineered, real-world system, containing millions of interacting components, can handle simple operations in vast numbers with an insignificant error rate. The computer must continue working flawlessly despite changes in temperature, line voltage, and electromagnetic noise, and regardless of what program it is asked to run.

The computer can do this because it is digital. Digital values are discrete--each signal in the computer is either on or off, never in-between. The signals are generated by transistor circuits which have a non-linear response to input signals; an input that is anywhere near the ideal "on" or "off" level will produce an output that is closer to the ideal. Deviations from perfection, rather than building up, are reduced as the signal propagates from one transistor to another. Even without active error detection, correction, or feedback, the non-linear behavior of transistors means that error rates can be kept as low as desired: for the purposes of computer designers, the error rate of any signal is effectively zero.

An error rate of zero means that the signals inside a computer are perfectly characterized: each signal and computation is exactly predictable. This allows a very powerful design technique to be used, called "levels of abstraction." Error-free operations can be combined in intricate patterns and in large numbers, with perfectly predictable results. The result of any sequence of operations can be calculated with certainty and precision. Thousands of transistors can be combined into number-processing circuits that do arithmetic and other calculations. Thousands of those circuits can be combined in general-purpose processor chips that execute simple instructions. Thousands of those instructions can be combined into data-processing functions. And those functions can be executed thousands or even billions of times, in any desired sequence, to perform any calculation that humans can invent... performing billions of billions of operations with reliably zero errors.

Modern manufacturing operations, for all their precision, are not digital. There is no discrete line between a good and a bad part--just as it's impossible to say exactly when someone who loses one hair at a time becomes bald. Worse, there is no mechanism in manufacturing that naturally restores precision. Difficult and complicated processes are required to construct a machine more precise than the machines used to build it. To build a modern machine such as a computer-controlled lathe requires so many different techniques--polymer chemistry, semiconductor lithography, metallurgy and metal working, and thousands of others--that the "real world" will inevitably create errors that must be detected and corrected. And to top it off, machines suffer from wear--their dimensions change as they are used.

Given the problems inherent in today's manufacturing methods and machine designs, the idea of building a fully automated general-purpose manufacturing system that could build a complete duplicate of itself... is ridiculous.

The ability to form covalent solids by placing individual molecules changes all that. Fundamentally, covalent bonds are digital: two atoms are either bonded, or they are held some distance apart by a repelling force. (Not all bond types are fully covalent, but many useful bonds including carbon-carbon bonds are.) If a covalent bond is stretched out of shape, it will return to its ideal configuration all by itself, without any need for external error detection, correction, or feedback.

If a covalent-bonding manufacturing system performs an operation with less than one atom out of place, then the resulting product will have exactly zero atoms out of place. Just like transistor signals in a digital computer, imperfections fade away all by themselves. (In both cases, a bit of energy is used up in making the imperfections disappear.) In digital systems, there is no physical law that requires imperfections to accumulate into errors--not in digital computer logic, and not in atomic fabrication systems.

Atomic fabrication operations, like transistor operations, can be characterized with great reliability. Only a few transistor operations are a sufficient toolkit with which to design a computer. A general-purpose molecular manufacturing system may use a dozen or so different kinds of atoms, and perhaps 100 reactions between the atoms. That is a small enough number to study each reaction in detail, and know how it works with as much precision as necessary. Each reaction can proceed in a predictable way each and every time it is attempted.

A sequence of completely predictable operations will itself have a completely predictable outcome, regardless of the length of the sequence. If each one of a sequence of a billion reactions is carried out as expected, then the final product can be produced reliably.

Chemists who read this may be objecting that there's no such thing as a reaction with 100% yield. Answering that objection in detail would require a separate essay--but briefly, mechanical manipulation and control of reactants can in many cases prevent unwanted reaction pathways as well as shifting the energetics so far (hundreds of zJ/bond or kJ/mole) that the missed reaction rate is reduced by many orders of magnitude.

At this point, it is necessary to consider the "real world." What factors, in practice, will reduce the predictability of mechanically-guided molecular reaction machinery?

One factor that doesn't have to be considered in a well-designed system of this type is wear. Again, it would take a separate essay to discuss wear in detail, but wear in a covalent solid requires breaking strong inter-atomic bonds, and a well-designed system will never, in normal operation, exert enough force on any single atom to cause its bonds to break. Likewise, machines built with the same sequence of reliable operations will themselves be identical. Once a machine is characterized, all of its siblings will be just as fully understood.

Mechanical vibration from outside the system is unlikely to be a problem. It is a problem in today's nanotech tools because the tools are far bigger than the manipulations or measurements those tools perform--big enough to have slow vibrational periods and high momentum. Nanoscale tools, such as would be used in a molecular manufacturing system, would have vibrational frequencies in the gigahertz or higher, and momentum vanishingly small compared to restoring forces.

It is possible that vibrations generated within the system, from previous operations of the system or of neighboring systems, could be a problem. In computers, transistor operations can cause voltage ripples that cause headaches for designers, and are probably analogous. But these problems are practical, not fundamental.

Contaminant molecules should not be a problem in a well-designed system. The ability to build covalent solids without error implies the ability to build hermetically sealed enclosures. Feedstock molecules would have to be taken in through the enclosures, but sorting mechanisms have been planned that should reject any contaminants in the feedstock stream with extremely low error rates. There are ways for a manufacturing system inside a sealed enclosure to build another system of the same size or larger without breaking the seal. It would take a third essay to discuss these topics in detail, but they have been considered and none of the problems appears unlikely to be addressable in practice.

Despite everything written above, there will be some fraction of molecular manufacturing systems that suffer from errors--if nothing else, background levels of ionizing radiation will cause at least some bond breakage. In theory, an imperfect machine could fabricate more imperfect machines, perpetuating and perhaps exacerbating the error. However, in practice, this seems unlikely. Whereas a perfect manufacturing machine could do a billion operations without error, an imperfect machine would probably make at least one atom-placement error fairly early in the fabrication sequence. That first error would leave an atom out of its expected position on the surface of the workpiece. A flawed workpiece surface would usually cause a cascade of further fabrication errors in the same product, and long before a product could be completed, the process would be hopelessly jammed. Thus, imperfect machines would quickly become inert, before producing even one imperfect product.

The biggest source of unpredictability probably will be thermal noise, sometimes referred to as Brownian motion. (Quantum uncertainty and Heisenberg uncertainty are similar but smaller sources of unpredictability.) Thermal noise means that the exact dimensions of a mechanical system will change unpredictably, too rapidly to permit active compensation. In other words, the exact position of the operating machinery cannot be known. The degree of variance depends on the temperature of the system, as well as the stiffness of the mechanical design. If the position varies too far, then a molecule-bonding operation may result in one of several unpredictable outcomes. This is a practical problem, not a fundamental limitation; in any given system, the variance is limited, and there are a number of ways to reduce it. More research on this point is needed, but so far, high-resolution computational chemistry experiments by Robert Freitas seem to show that even without using some of the available tricks, difficult molecule-placement operations can be carried out with high reliability at liquid nitrogen temperatures and possibly at room temperature. If positional variance can be reduced to the point where the molecule is placed in approximately the right position, the digital nature of covalent bonding will do the rest.

This is a key point: The mechanical unpredictability in the system does not have to be reduced to zero, or even extremely close to zero, in order to achieve extremely high levels of reliability in the product. As long as each reaction trajectory leads closer to the right outcome than to competing outcomes, the atoms will naturally be pulled into their proper configuration each time--and by the time the next atoms are deposited, any positional error will have dissipated into heat, leaving the physical bond structure perfectly predictable for the next operation.

Molecular manufacturing requires an error rate that is extremely low by most standards, but is quite permissive compared to the error rates necessary for digital computers. Error rate is an extremely important topic, and unfortunately, understanding of errors in mechanically guided chemistry is susceptible to incorrect intuitions from chemistry, manufacturing, and even physics (many physicists assume that entropy must increase without considering that the system is not closed). It appears that the nature of covalent bonds provides an automatic error-reducing mechanism that will make molecular manufacturing closer in significant ways to computer logic than to today's manufacturing or chemistry.

Three previous science essays have touched on related topics:

Who remembers analog computers? (February 2006)
Coping with Nanoscale Errors (September 2004)
The Bugbear of Entropy (May 2004)

Subsequent CRN science essays will cover topics that this essay raised but did not have space to cover in detail.
 

Recycling Nano-Products
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

We are often asked, "How will nanofactory-built products be recycled?"

One of the advantages of molecular manufacturing is that it will use very strong and high-performance materials. Most of them probably will not be biodegradable. So what will save us from being buried in nano-litter?

The first piece of good news is that nano-built products will use materials more efficiently. Mechanical parts can be built mostly without defects, making them a lot stronger than today's materials. Active components can be even more compact, because scaling laws are advantageous to small machines: motors may have a million times the power density, and computers may be a million times as compact. So for equivalent functionality, nano-built products will use perhaps 100-1000 times less material. In fact, some products may be so light that they have to be ballasted with water. (This would also make carbon-based products fireproof.)

The second good news is that carbon-based products will burn once any water ballast is removed. Traditionally, incineration has been a dirty way to dispose of trash; heavy metals, chlorine compounds, and other nasty stuff goes up the smokestack and pollutes wherever the wind blows. Fortunately, one of the first products of molecular manufacturing will be efficient molecular sorting systems. It will be possible to remove the harmless and useful gases from the combustion products--perhaps using them to build the next products--and send the rest back to be re-burned.

The third good news is that fewer exotic materials and elements should be needed. Today's products use a lot of different substances for different jobs. Molecular manufacturing, by contrast, will be able to implement different functions by building different molecular shapes out of a much smaller set of materials. For example, carbon can be either an insulator or a conductor, shapes built of carbon can be both flexible or rigid, and carbon molecules can be transparent (diamond) or opaque (graphite).

Finally, it may be possible to build many full-size products out of modular building blocks: microscopic nanoblocks that might contain a billion atoms and provide flexible functionality. In theory, rather than discarding and recycling a product, it could be pulled apart into its constituent blocks, which could then be reassembled into a new product. However, this may be impractical, since the nanoblocks would have to be carefully cleaned in order to fit together precisely enough to make a reliable product. But re-using rather than scrapping products is certainly a possibility that's worth investigating further.

Not surprisingly, there is also some bad news. The first bad news is that carbon is not the only possible material for molecular manufacturing. It is probably the most flexible, but others have been proposed. For example, sapphire (corundum, alumina) is a very strong crystal of aluminum oxide. It will not burn, and alumina products probably will have to be scrapped into landfills if their nanoblocks cannot be re-used. Of course, if we are still using industrial abrasives, old nano-built products might simply be crushed and used for that purpose.

The second bad news is that nano-built products will come in a range of sizes, and some will be small enough that they will be easy to lose. Let me stress that a nano-built product is not a grey goo robot, any more than a toaster is. Tiny products may be sensors, computer nodes, or medical devices, but they will have specialized functionality--not general-purpose manufacturing capability. A lost product will likely be totally inert. But enough tiny lost products could add up to an irritating dust.

The third bad news is that widespread use of personal nanofactories will make it very easy and inexpensive to build stuff. Although each product will use far less material than today's versions, we may be using far more products.

Some readers may be wondering about "disassemblers" and whether they could be used for recycling. Unfortunately, the "disassembler" described in Eric Drexler's Engines of Creation was a slow and energy-intensive research tool, not an efficient way of taking apart large amounts of matter. It might be possible to take apart old nano-products molecule by molecule, but it would probably be less efficient than incinerating them.

Collecting products for disposal of is an interesting problem. Large products can be handled one at a time. Small and medium-sized products might be enough of a hassle to keep track of that people will be tempted to use them and forget them. For example, networked sensors with one-year batteries might be scattered around, used for two months, and then forgotten--better models would have been developed long before the battery would wear out. In such cases, the products might need to be collected robotically. Any product big enough to have an RFID antenna would be able to be interrogated as to its age and when it was last used. Ideally, it would also tell who its owner had been, so the owner could be billed, fined, or warned as appropriate.

This essay has described what could be. Environmentally friendly cleanup and disposal schemes will not be difficult to implement in most cases. However, as with so much else about molecular manufacturing, the availability of good choices does not mean that the best options necessarily will be chosen. It is likely that profligate manufacturing and bad design will lead to some amount of nano-litter. But the world will be very fortunate if nano-litter turns out to be the biggest problem created by molecular manufacturing.

New Opportunities for DNA Design
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

DNA is a very versatile molecule, if you know how to use it. Of course, the genetic material for all organisms (except some viruses) is made of DNA. But it is also useful for building shapes and structures, and it is this use that is most interesting to a nanotechnologist.

Readers familiar with DNA complementarity should skip this paragraph. Non-technical readers should read my earlier science essay on DNA folding. Briefly, DNA is made of four molecules, abbreviated A, C, G, and T, in a long string (polymer). G and C attract each other, as do A and T. A string with the sequence AACGC will tend to attach itself to another string with the sequence GCGTT (the strings match head-to-tail). Longer sequences attach more permanently. Heating up a mixture of DNA makes the matched strings vibrate apart; slowly cooling a mixture makes the strings reattach in (hopefully) their closest-matched configuration.

Until recently, designing a shape out of DNA was a painstaking process of planning sequences that would match in just the right way – and none of the wrong ways. Over the years, a number of useful design patterns were developed, including ways to attach four strands of DNA side by side for extra stiffness; ways to make structures that would contract or twist when a third strand was added to bridge two strands in the structure; and three-way junctions between strands, useful for building geometric shapes. A new structure or technique would make the news every year or so. In addition to design difficulties, it was hard to make sufficiently long error-free strands to form useful shapes.

A few months ago, a new technique was invented by Dr. Paul Rothemund. Instead of building all the DNA artificially for his shapes, he realized that half of it could be derived from a high-quality natural source with a fixed but random sequence, and the other half could be divided into short, easily synthesized pieces – “staples” – with sequences chosen to match whatever sequence the natural strand happens to have at the place the staple needs to attach. Although the random strand will tend to fold up on itself randomly to some extent, the use of large numbers of precisely-matching staples will pull it into the desired configuration.

If a bit of extra DNA is appended to the end of a staple piece, the extra bit will stick out from the shape. This extra DNA can be used to attach multiple shapes together, or to grab on to a DNA-tagged molecule or particle. This implies that DNA-shape structures can be built that include other molecules for increased strength or stiffness, or useful features such as actuation.

Although the first shapes designed were planar, because planar shapes are easier to scan with atomic force microscopes so as to verify what’s been built, the stapling technique can also be used to pull the DNA strand into a three-dimensional shape. So this implies that with a rather small design effort (at least by the standards of a year ago), 3D structures built of DNA can be constructed, with “Velcro” hanging off of them to attach them to other DNA structures, and with other molecules either attached to the surface or embedded in the interior.

The staple strands are short and easy to synthesize (and don’t need to be purified), so the cost is quite low. According to page 81 of Rothemund’s notes [PDF], a single staple costs about $7.00 – for 80 nmol, or 50 quintillion molecules. Enough different staples to make a DNA shape cost about $1,500 to synthesize. The backbone strand costs about $12.50 per trillion molecules. Now, those trillion molecules only add up to 4 micrograms. Building a human-scale product out of that material would be far too costly. But a computer chip with only 100 million transistors costs a lot more than $12.50.

The goal that’s the focus of this essay is combining a lot of these molecular “bricks” to build engineered heterogeneous structures with huge numbers of atoms. In other words, rather than creating simple tilings of a few bricks, stick them together in arbitrary computer-controlled patterns, constructs in which every brick can be different and independently designed.

I was hoping that nano-manipulation robotics had advanced to the point where the molecular shapes could be attached to large handles that would be grabbed and positioned by a robot, making the brick go exactly where it’s wanted relative to the growing construct, but I’m told that the state of the art probably isn’t there yet. Just one of the many problems is that if you can’t sense the molecule as you’re positioning it, you don’t know if temperature shifts have caused the handle to expand or contract. However, there may be another way to do it.

An atomic force microscope (AFM) uses a small tip. With focused ion beam (FIB) nano-machining, the tip can be hollowed out so as to form a pocket suitable for a brick to nestle in. By depositing DNA Velcro with different sequences at different places in the pocket (which could probably be done by coating the whole tip, burning away a patch with the FIB, then depositing a different sequence), it should be possible to orient the brick relative to the tip. (If the brick has two kinds of Velcro on each face, and the tip only has one kind deposited, the brick will stick less strongly to the tip than to its target position.)

Now, the tip can be used for ordinary microscopy, except that instead of having a silicon point, it will have a corner of the DNA brick. It should still be usable to scan the construct, hopefully with enough resolution to tell where the tip is relative to the construct. This would solve the nano-positioning problem.

I said above that the brick would have DNA Velcro sticking out all over. For convenience, it may be desirable to have a lot of bricks of identical design, floating around the construct – as long as they would not get stuck in places they’re not wanted. This would allow the microscope tip to pick up a brick from solution, then deposit it, then pick up another right away, without having to move away to a separate “inkwell.” But why don’t the bricks stick to the construct and each other, and if they don’t, then how can the tip deposit them, and why do they stay where they’re put?

To make the bricks attach only when and where they’re put requires three conditions. First, the Velcro should be designed to be sticky, so that the bricks will stay firmly in place once attached. Second, the Velcro should be capped with other DNA strands so that the bricks will not attach by accident. Third, the capping strands should be designed so that physically pushing the brick against a surface will weaken the bond between Velcro and cap, allowing the Velcro to get free and bind to the target surface. For example, if the cap strands stick stiffly out away from the block (perhaps by being bound to two Velcro strands at once), then mechanical pressure will weaken the connection.

Mechanical pressure, of course, can be applied by an AFM. Scan with low force, and when the brick is in the right place, press down with the microscope. Wait for the cap strands to float away and the Velcro to pair up, and the brick is deposited. With multiple Velcro strands between each brick, the chance of them all coming loose at once and allowing the brick to be re-capped can be made miniscule, especially since the effective concentration of Velcro strands would be far higher than the concentration of cap strands. But before the brick was pushed into place, the chance of all the cap strands coming loose at once also would have been miniscule. (For any physicists reading this, thermodynamic equilibrium between bound and free bricks still applies, but the transition rate can be made even slower than the above concentration argument implies, since the use of mechanical force allows an extremely high energy barrier. If the mechanical force applied is 100 pN over 5 nm, that is 500 zJ, approximately the dissociation energy of a C-C bond.)

So, it seems that with lots of R&D (but without a whole lot of DNA design), it might be possible to stick DNA bricks (plus attached molecules) together in arbitrary patterns, using an AFM. But an AFM is going to be pretty slow. It would be nice to make the work go faster by doing it in parallel. My NIAC project suggests a way to do that.

The plan is to build an array of “towers” or “needles” out of DNA bricks. In the tip of each one, put a brick-binding cavity. Use an AFM to build the first one in the middle of a flat surface. Then use that to build a second and third needle on an opposing surface. (One of the surfaces would be attached to a nano-positioner.) Use those two towers to build a fourth, fifth, sixth, and seventh on the first surface. The number of towers could grow exponentially.

By the time this is working, there may be molecules available that can act as fast, independently addressable, nanoscale actuators. Build a mechanism into each tower that lets it extend or retract – just a nanometer or so. Now, when the towers are used to build something, the user can select which bricks to place and which ones to hold back. This means that different towers, all attached to the same surface and moved by the same nano-positioner, can be doing different things. Now, instead of an exponential number of identical designs, it has become possible to build an exponential number of different designs, or to work on many areas of a large heterogeneous design in parallel.

A cubic micron is not large by human standards, but it is bigger than most bacteria. There would be about 35,000 DNA bricks in a cubic micron. If a brick could be placed every fifteen seconds, then it would take a week to build a cubic micron out of bricks. This seems a little fast for a single AFM that has to bind bricks from solution, find a position, and then push the brick into place, even if all steps were fully automated – but it might be doable with a parallel array (either an array of DNA needles, or a multi-tip AFM). If every brick were different, it would cost about $50 million for the staples, but of course not every brick will be different. For 1,000 different bricks, it would cost only about $1.5 million.

We will want the system to deposit any of a number of brick types in any location. How to select one of numerous types? The simplest way is to make all bricks bind to the same tip, then flush them through one type at a time. This is slow and wasteful. Better to include several tips in one machine, and then flush through a mixture of bricks that will each bind to only one tip. The best answer, once really high-function bricks are available and you’re using DNA-built tips instead of hollowed-out AFM tips, is to make the tips reconfigurable by using fast internal actuators to present various combinations of DNA strands for binding of various differently-tagged bricks.

I started by suggesting that a scanning probe microscope be used to build the first construct. Self-assembly also could be used to build small constructs, if you can generate enough distinct blocks. But you may not have to build hundreds of different bricks to make them join in arbitrary patterns. Instead, build identical bricks, and cap the Velcro strands with a second-level “Velcro staple.” Start with a generic brick coated with Velcro – optionally, put a different Velcro sequence on each side. Stir that together with strands that are complementary to the Velcro at one end and contain a recognition pattern on the other end. Now, with one generic brick and six custom-made Velcro staples, you have a brick with a completely unique recognition pattern on each side. Do that for a number of bricks, and you can make them bind together any way you want. One possible problem with this is that DNA binding operations usually need to be “annealed” – heated to a temperature where the DNA falls apart, then cooled slowly. This implies that the Velcro-staple approach would need three different temperature ranges: one to form the shapes, one to attach the staples, and one to let the shapes join together.

The Velcro-staple idea might even be tested today, using only the basic DNA-shape technology, with one low-cost shape and a few dozen very-low-cost staples. Plus, of course, whatever analysis tools you’ll need to convince you that you’re making what you think you’re making.

There is a major issue involved here that I have not yet touched on. Although the DNA staple technique makes a high percentage of good shapes, it also makes a lot of broken or incomplete shapes. How can the usable shapes be sorted from the broken shapes? Incomplete shapes may be sorted out by chromatography. Broken shapes might possibly be rejected by adding a fluorescence pair and using a cell sorter to reject shapes that did not fluoresce properly. Another possibility, if using a scanning probe microscope (as opposed to the “blind” multi-needle approach) is to detect the overall shape of the brick by deconvolving it against a known surface feature, and if an unwanted shape is found, heat the tip to make it dissociate.

This is just a sketch of some preliminary ideas. But it does go to show that the new DNA staple technology makes things seem plausible that would not have been thinkable before it was developed.

Military Implications of Molecular Manufacturing
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

(Originally published in the July 2006 issue of NanoNews-Now -- reprinted by permission)

This essay will survey the technology of molecular manufacturing, the basic capabilities of its products, some possible weapon systems, some tactical and strategic considerations, and some possible effects of molecular manufacturing on the broader context of societies and nations. However, all of this discussion must take place in the context of the underlying fact that the effects and outcome of molecular manufacturing will be almost inconceivable, and certainly not susceptible to shallow or linear analysis.

Take a minute and try to imagine a modern battlefield without electricity. No radar or radios; no satellites; no computers; no night vision, or even flashlights; no airplanes, and few ground vehicles of any kind. Imagination is not sufficient to generate this picture—it simply doesn't make sense to talk of a modern military without electricity.

Molecular manufacturing will have a similarly profound effect on near-future military affairs.

Electricity is a general-purpose energy technology, useful for applications from motors to data processing. A few inventions, ramified and combined—the storage battery, transistor, electromagnet, and a few others—are powerful enough to be necessary components of almost all modern military equipment and activities.

If it is impossible to conceive of a modern military without electricity—a technology that exists, and the use of which we can study—it will be even less feasible to try to imagine a military with molecular manufacturing.

Molecular manufacturing will be the world's first general-purpose manufacturing technology. Its products will be many times more plentiful, more intricate, and higher performance than any existing product. They will be built faster and less expensively, speeding research and development. They will cover a far greater range of size, energy, and distance than today's weapons systems. As increasingly powerful weapons make the battlefield untenable for human soldiers, computers vastly more powerful and compact than today's will enable far higher degrees of automation and remote operation. Kilogram-scale manufacturing systems, building directly from the latest blueprints in minutes, will utterly transform supply, logistics, and deployment.

Radium and X-rays were discovered within months of each other. Within a few years, X-rays had inspired stories about military uses of “death rays.” Decades later, Madame Curie gave speeches on the wonderful anti-cancer properties of radium. It would have been difficult or impossible to predict that a few decades after that, X-rays would be a ubiquitous medical technology, and nuclear radiation would be the basis of the world's most horrific weapons. While reading the rest of this article, keep in mind that the implications of various molecular manufacturing products and capabilities will be at least as unpredictable and counterintuitive.

Technical Basis of Molecular Manufacturing

At its foundation, molecular manufacturing works by doing a few precise fabrication operations, very rapidly, at the molecular level, under computer control. It can thus be viewed as a combination of mechanical engineering and chemistry, with some additional help from rapid prototyping, automated assembly, and related fields of research.

Atoms and inter-atomic bonds are completely precise: every atom of a type is identical to every other, and there are only a few types. Barring an identifiable error in fabrication, two molecules manufactured according to the same blueprint will be identical in structure and shape (with transient variations of predictable scale due to thermal noise and other known physical effects). This consistency will allow fully automated fabrication. Computer controlled addition of molecular fragments, creating a few well-characterized bond types in a multitude of selected locations, will enable a vast range of components to be built with extremely high reliability. Building with reliable components, higher levels of structure can retain the same predictability and engineerability.

A fundamental “scaling law” of physics is that small systems operate faster than large systems. Moving at moderate speed over tiny distances, a nanoscale fabrication system could perform many millions of operations per second, creating products of its own mass and complexity in hours or even minutes. Along with faster operation comes higher power density, again proportional to the shrinkage: nanoscale machines might be a million times more compact than today's technology. Computers would shrink even more profoundly, and non-electronic technologies already analyzed could dissipate enough less power to make the shrinkage feasible. Although individual nanoscale machines would have small capacity, massive arrays could work together; it appears that gram-scale computer and motor systems, and ton-scale manufacturing systems, preserving nanoscale performance levels, can be built without running afoul of scaling laws or other architectural constraints including cooling. Thus, products will be buildable in a wide range of sizes.

A complete list of advantages and capabilities of molecularly manufactured products, much less an analysis of the physical basis of the advantages, would be beyond the scope of this paper. But several additional advantages should be noted. Precisely fabricated covalent materials will be much stronger than materials formed by today's imprecise manufacturing processes. Precise, well-designed, covalently structured bearings should suffer neither from wear nor from static friction (stiction). Carbon can be an excellent conductor, an excellent insulator, or a semiconductor, allowing a wide range of electrical and electronic devices to be built in-place by a molecular manufacturing system.

Development of Molecular Manufacturing

Although its capabilities will be far-reaching, the development of molecular manufacturing may require a surprisingly small effort. A finite, and possibly small, number of deposition reactions may suffice to build molecular structures with programmable shape—and therefore, diverse and engineerable function. High-level architectures for integrated kilogram-scale arrays of nanoscale manufacturing systems have already been worked out in some detail. Current-day tools are already able to remove and deposit atoms from selected locations in covalent solids. Engineering of protein and other biopolymers is another pathway to molecularly precise fabrication of engineered nanosystem components. Analysis tools, both physical and theoretical, are developing rapidly.

As a general rule, nanoscale research and development capabilities are advancing in proportion to Moore's Law—even faster in some cases. Conceptual barriers to developing molecular manufacturing systems are also falling rapidly. It seems likely that within a few years, a program to develop a nanofactory will be launched; some observers believe that one or more covert programs may already have been launched. It also seems likely that, within a few years of the first success, the cost of developing an independent capability will have dropped to the point where relatively small groups can tackle the project. Without stringent and widespread restrictions on technology, it most likely will not be possible to prevent the development of multiple molecular manufacturing systems with diverse owners.

Products of Molecular Manufacturing

All exploratory engineering in the field to date has pointed to the same set of conclusions about molecular manufacturing-built products:

1. Manufacturing systems can build more manufacturing systems.

2. Small products can be extremely compact.

3. Human-scale products can be extremely inexpensive and lightweight.

4. Large products can be astonishingly powerful.

If a self-contained manufacturing system can be its own product, then manufacturing systems can be inexpensive—even non-scarce. Product cost can approach the cost of the feedstock and energy required to make it (plus licensing and regulatory overhead). Although molecular manufacturing systems will be extremely portable, most products will not include one—it will be more efficient to manufacture at a dedicated facility with installed feedstock, energy, and cooling supplies.

The feature size of nanosystems will probably be about 1 nanometer (nm), implying a million features in a bacteria-sized object, a billion features per cubic micron, or a trillion features in the volume of a ten-micron human cell. A million features is enough to implement a simple CPU, along with sensors, actuators, power supply, and supporting structure. Thus, the smallest robots may be bacteria-sized, with all the scaling law advantages that implies, and a medical system (or weapon system based thereon) could be able to interact with cells and even sub-cellular structures on an equal footing. (See Nanomedicine Vol. I: Basic Capabilities for further exploration.)

As a general rule of thumb, human-scale products may be expected to be 100-1000 times lighter than today's versions. Covalent carbon-based materials such as buckytubes should be at least 100 times stronger than steel, and materials could be used more efficiently with more elegant construction techniques. Active components will shrink even more. (Of course, inconveniently light products could be ballasted with water.)

Large nanofactories could build very large products, from spacecraft to particle accelerators. Large products, like smaller ones, could benefit from stronger materials and from active systems that are quite compact. Nanofactories should scale to at least ton-per-hour production rates for integrated products, though this might require massive cooling capacity depending on the sophistication of the nanofactory design.

Possible Weapons Systems

The smallest systems may not be actual weapons, but computer platforms for sensing and surveillance. Such platforms could be micron-scale. The power requirement of a 1-MIPS computer might be on the order of 10-100 pW; at that rate, a cubic micron of fuel might last for 100-1000 seconds. The computer itself would occupy approximately one cubic micron.

Very small devices could deliver fatal quantities of toxins to unprotected humans.

Even the smallest ballistic projectiles (bullets) could contain supercomputers, sensors, and avionics sufficient to guide them to targets with great accuracy. Flying devices could be quite small. It should be noted that small devices could benefit from a process of automated design tuning; milligram-scale devices could be built by the millions, with slight variations in each design, and the best designs used as the basis for the next “generation” of improvements; this could enable, for example, UAV's in the laminar regime to be developed without a full understanding of the relevant physics. The possibility of rapid design is far more general than this, and will be discussed below.

The line between bullets, missiles, aircraft, and spacecraft would blur. With lightweight motors and inexpensive manufacturing, a vehicle could contain a number of different disposable propulsion systems for different flight regimes. A “briefcase to orbit” system appears feasible, though such a small device might have to fly slowly to conserve fuel until it reached the upper atmosphere. It might be feasible to use 1 kg of airframe (largely discarded) and 20 kg of fuel (not counting oxidizer) to place 1 kg into orbit; some of the fuel would be used to gather and liquify oxygen in the upper atmosphere for the rocket portion of its flight. (Engineering studies have not yet been done for such a device, and it might require somewhat more fuel than stated here.)

Advanced construction could produce novel energy-absorbing materials involving high-friction mechanical slippage under high stress via micro- or nano-scale mechanical components. In effect, every molecule would be a shock absorber, and the material could probably absorb mechanical energy until it was destroyed by heat.

New kinds of weapons might be developed more quickly with rapid inexpensive fabrication. Many classes of device will be buildable monolithically. For example, a new type of aircraft or even spacecraft might be tested an order of magnitude more rapidly and inexpensively, reducing the cost of failure and allowing further acceleration in schedule and more aggressive experimentation. Although materials and molecular structures would not encompass today's full range of manufactured substances, they could encompass many of the properties of those substances, especially mechanical and electrical properties. This may enable construction of weapons such as battlefield lasers, rail guns, and even more exotic technologies.

Passive armor certainly could not stop attacks from a rapid series of impacts by precisely targeted projectiles. However, armor could get a lot smarter, detecting incoming attacks and rapidly shifting to interpose material at the right point. There may be a continuum from self-reconfiguring armor, to armor that detaches parts of itself to hurl in the path of incoming attacks, to armor that consists of a detached cloud of semi-independent counterweapons.

A new class of weapon for wide-area destruction is kinetic impact from space. Small impactors would be slowed by the atmosphere, but medium-to-large asteroids, redirected onto a collision course, could destroy many square miles. The attack would be detectable far in advance, but asteroid deflection and destruction technology is not sufficiently advanced at this time to say whether a defender with comparable space capacity could avoid being struck, especially if the asteroid was defended by the attacker. Another class of space impactor is lightweight solar sails accelerated to a respectable fraction of light speed by passage near the sun. These could require massive amounts of inert shielding to stop; it is not clear whether or not the atmosphere would perform this function adequately.

A hypothetical device often associated with molecular manufacturing is a small, uncontrolled, exponentially self-replicating system. However, a self-replicating system would not make a very good weapon. In popular conception, such a system could be built to use a wide range of feedstocks, deriving energy from oxidizing part of the material (or from ambient light), and converting the rest into duplicate systems. In practice, such flexibility would be quite difficult to achieve; however, a system using a few readily available chemicals and bypassing the rest might be able to replicate itself—though even the simplest such system would be extremely difficult to design. Although unrestrained replication of inorganic systems poses a theoretical risk of widespread biosphere destruction through competition for resources—the so-called “grey goo” threat—it seems unlikely that anyone would bother to develop grey goo as a weapon, even a doomsday deterrent. It would be more difficult to guide than a biological weapon. It would be slower than a device designed simply to disrupt the physical structure of its target. And it would be susceptible to detection and cleanup by the defenders.

Tactics

A detailed analysis of attack and defense is impossible at this point. It is not known whether sensor systems will be able to effectively detect and repel an encroachment by small, stealthy robotic systems; it should be noted that the smallest such systems might be smaller than a wavelength of visible light, making detection at a distance problematic. It is unknown whether armor will be able to stop the variety of penetrating objects and forces that could be directed at it. Semi-automated R&D may or may not produce new designs so quickly that the side with the better software will have an overwhelming advantage. The energy cost of construction has only been roughly estimated, and is uncertain within at least two orders of magnitude; active systems, including airframes for nano-built weapons, will probably be cost-effective in any case, but passive or static systems including armor may or may not be worth deploying.

Some things appear relatively certain. Unprotected humans, whether civilian or soldier, will be utterly vulnerable to nano-built weapons. In a scenario of interpenetrating forces, where a widespread physical perimeter cannot be established, humans on both sides can be killed at will unless protected at great expense and inconvenience. Even relatively primitive weapons such as hummingbird-sized flying guns with human target recognition and poisoned bullets could make an area unsurvivable without countermeasures; the weight of each gun platform would be well under one gram. Given the potential for both remote and semi-autonomous operation of advanced robotics and weapons, a force with a developed molecular manufacturing capability should have no need to field soldiers; this implies that battlefield death rates will be low to zero for such forces.

A concern commonly raised in discussions of nanotech weapons is the creation of new diseases. Molecular manufacturing seems likely to reduce the danger of this. Diseases act slowly and spread slowly. A sufficiently capable bio-sensor and diagnostic infrastructure should allow a very effective and responsive quarantine to be implemented. Rapid testing of newly manufactured treatment methods, perhaps combined with metabolism-slowing techniques to allow additional R&D time, could minimize disease even in infected persons 

Despite the amazing power and flexibility of molecular manufactured devices, a lesson from World War I should not be forgotten: Dirt makes a surprisingly effective shield. It is possible that a worthwhile defensive tactic would be to embed an item to be protected deeply in earth or water. Without active defenses, which would also be hampered by the embedding material, this would be at best a delaying tactic.

Information is likely to be a key determiner of military success. If, as seems likely, unexpected offense with unexpected weapons can overwhelm defense, then rapid detection and analysis of an attacker's weapons will be very important. Information-gathering systems are likely to survive more by stealth than by force, leading to a “spy vs. spy” game. To the extent that this involves destruction of opposing spy-bots, it is similar to the problem of defending against small-scale weapons. Note that except for the very smallest systems, the high functional density of molecularly constructed devices will frequently allow both weapon and sensor technology to be piggybacked on platforms primarily intended for other purposes.

It seems likely that, with the possible exception of a few small, fiercely defended volumes, a robotic battleground would consist of interpenetrated forces rather than defensive lines (or defensive walls). This implies that any non-active matter could be destroyed with little difficulty unless shielded heavily enough to outlast the battle.

Strategy

As implied above, a major strategy is to avoid putting soldiers on the battlefield via the use of autonomous or remotely operated weapons. Unfortunately, this implies that an enemy wanting to damage human resources will have to attack either civilian populations or people in leadership positions. To further darken the picture, civilian populations will be almost impossible to protect from a determined attack without maintaining a near-hermetic seal around their physical location. Since the resource cost of such a shield increases as the shield grows (and the vulnerability and unreliability probably also increase), this implies that civilians under threat will face severe physical restrictions from their own defenders.

A substantial variety of attack mechanisms will be available, including kinetic impact, cutting, sonic shock and pressure, plasma, electromagnetic beam, electromagnetic jamming and EMP, heat, toxic or destructive chemicals, and perhaps more exotic technologies such as particle beam and relativistic projectile. A variety of defensive techniques will be available, including camouflage, small size, physical avoidance of attack, interposing of sacrificial mass, jamming or hacking of enemy sensors and computers, and preemptive strike. Many of these offensive and defensive techniques will be available to devices across a wide range of sizes. As explored above, development of new weapon systems may be quite rapid, especially if automated or semi-automated design is employed.

In addition to the variety of physical modes of attack and defense, the cyber sphere will become an increasingly important and complex battleground, as weapon systems increasingly depend on networking and computer control. It remains to be seen whether a major electronic security breach might destroy one side's military capacity, but with increasing system complexity, such an occurrence cannot be ruled out.

Depending on what is being defended, it may or may not be possible to prepare an efficient defense for all possible modes of attack. If defense is not possible, then the available choices would seem to be either preemptive strike or avoidance of conflict. Defense of civilians, as stated above, is likely to be difficult to impossible. Conflict may be avoided by deterrence only in certain cases—where the opponent has a comparable amount to lose. In asymmetric situations, where opponents may have very different resources and may value them very differently, deterrence may not work at all. Conflict may also be avoided by reducing the sources of tension 

Broader Context

Military activity does not take place in isolation. It is frequently motivated by non-military politics, though warlords can fight simply to improve their military position. Molecular manufacturing will be able to revolutionize economic infrastructures, creating material abundance and security that may reduce the desire for war—if it is distributed wisely.

It appears that an all-out war between molecular manufacturing powers would be highly destructive of humans and of natural resources; the objects of protection would be destroyed long before the war-fighting ability of the enemy. In contrast, a war between molecular manufacturing and a conventionally armed power would probably be rapid and decisive. The same is true against a nuclear power that was prevented from using its nuclear weapons, either by politics or by anti-missile technologies. Even if nuclear weapons were used, the decentralization allowed by self-contained exponentially manufacturing nanofactories would allow survival, continued prosecution of the war, and rapid post-war rebuilding.

The line between policing and military action is increasingly blurred. Civilians are becoming very effective at attacking soldiers. Meanwhile, soldiers are increasingly expected to treat civilians under occupation as citizens (albeit second-class citizens) rather than enemy. At least in the US, paramilitary organizations (both governmental and commercial) are being deployed in internal civilian settings, such as the use of SWAT teams in some crime situations, and Blackwater in post-Katrina New Orleans.

Many molecular manufactured weapon systems will be useable for policing. Several factors will make the systems desirable for police activity: a wide range of weapon effects and intensities to choose from; less risk to police as telepresence is employed; maintaining parity with increasingly armed criminals; and increased deterrence due to increased information-gathering and surveillance. This means that even without military conflict, a variety of military-type systems will be not only developed, but also deployed and used.

It is tempting to think that the absence of nuclear war after six decades of nuclear weapons implies that we know how to handle insanely destructive weapons. However, a number of factors will make molecular manufacturing arms races less stable than the nuclear arms race—and it should be remembered that on several different occasions, a single fortuitous person or event has prevented a nuclear attack. Nuclear weapons are hard to design, hard to build, require easily monitored testing, do indiscriminate and lasting damage, do not rapidly become obsolete, have almost no peaceful use, and are universally abhorred. Molecular manufactured weapons will be easy to build, will in many cases allow easily concealable testing, will be relatively easy to control and deactivate, and would become obsolete very rapidly; almost every design is dual-use, and peaceful and non-lethal (police) use will be common. Nuclear weapons are easier to stockpile than to use; molecular manufactured weapons will be the opposite.

Interpenetrating arrays of multi-scale complex weapons cannot be stable for long. Sooner or later, and probably sooner, a perceived attack will be answered by an actual attack. Whether this mushrooms out of control into a full-scale conflict will depend on the programming of the weapon systems. As long as it is only inanimate hardware at stake, probing attacks and small-scale accidental attacks may be tolerated.

Given the amount of damage that a hostile power armed with molecular manufacturing products could do to the civilian sector, it seems likely that hostile actors will be tolerated only as a last resort, and even apparently non-hostile but untrustworthy actors will be highly undesirable. As mentioned above, an asymmetry in values may prevent deterrence from working. An asymmetry in force, such as between a molecular manufacturing and a pre-MM power, may tempt a preemptive strike to prevent molecular manufacturing proliferation. Likewise, a substantial but decreasing lead in military capability may lead to a preemptive strike. It is unclear whether in general a well-planned surprise attack would lead to rapid and/or inexpensive victory; this may not become clear until offensive and defensive systems are actually developed.

One stable situation appears to be that in which a single power deploys sufficient sensors and weapons to prevent any other power from developing molecular manufacturing. This would probably require substantial oppression of civilians and crippling of industrial and scientific capacity. The government in power would have near-absolute control, being threatened only by internal factors; near-absolute power, combined with an ongoing need for oppression, would likely lead to disastrous corruption.

Widespread recognition of the dangers of arms race, preemptive strike, and war might inspire widespread desire to avoid such an outcome. This would require an unprecedented degree of trust and accountability, worldwide. Current government paradigms are probably not compatible with allowing foreign powers such intimate access to their secrets; however, in the absence of this degree of openness, spying and hostile inspections will only raise tension and reduce trust. One possible solution is for governments to allow their own citizens to observe them, and then allow the information gained by such distributed and non-combative (and thus presumably more trustworthy) observation to be made available to foreign powers.

Conclusion

Molecular manufacturing will introduce a wide diversity of new weapon systems and modes of warfighting. In the absence of actual systems to test, it is difficult if not impossible to know key facts about offensive and defensive capability, and how the balance between offense and defense may change over time. Incentives for devastating war are unknown, but potentially large—the current geopolitical context may favor a strategy of preemptive strike.

Full information about molecular manufacturing's capabilities will probably be lacking until a nanofactory is developed. At that point, once an exponential manufacturing capacity exists that can make virtually unlimited quantities of high-performance products, sudden development of unfamiliar and powerful weapon systems appears likely. It is impossible, from today's knowledge, to predict what a molecular manufacturing-enabled war will be like—but it is possible to predict that it would be most destructive to our most precious resources.

Given these facts and observations, an immediate and urgent search for alternatives to arms races and armed conflict is imperative.

Inapplicable Intuitions
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

Experts in a field develop intuitions about the way things work. For example, a biochemist will develop intuitions about the complexity of interactions between biomolecules. When faced with a new idea, a scientist will first evaluate it in light of existing intuitions.

Interest in molecular manufacturing is rapidly growing, and many scientists may be encountering the ideas for the first time. Because molecular manufacturing cuts across a number of fields -- physics, chemistry, mechanical engineering, software, and more -- and because it uses a rather novel approach to building stuff, almost any scientist will find something in the proposal that violates one or more intuitions. It is worth examining some of these intuitions. Notice that each intuition is true, although in a limited context, and molecular manufacturing avoids that context.

In addition to personally developed intuitions, scientists new to molecular manufacturing may run across objections formerly raised by others in different fields. In general, these objections were the result of similarly misplaced intuitions. The intent here is not to re-fight old battles, but simply to explain what the battles were about.

Here in a nutshell is the molecular manufacturing plan: Build a system that does a billion chemical reactions, one after the other, on the same molecule, with very high reliability, to make perfect molecular products. The system does chemical reactions by holding molecules and moving them into place through a vacuum, to transfer atoms to the product, adding a few atoms at a time to build molecular shapes. Use that system to build nanoscale machine components, and assemble the components into nanoscale machines. Control a bunch of these machines to build more machine components, one deposition at a time; then combine those machine components into large products. This will need huge numbers of machines, arrayed in a factory. Use an initial small factory to make another bigger factory, repeating enough times to grow to kilogram scale. Use the resulting big factory to make products from downloaded blueprints.

As we will see, nearly every phrase in this description may evoke skepticism from someone; however, all of these objections, and many others, have been addressed. The technical foundation for the modern approach to molecular manufacturing was laid with the 1992 publication of Nanosystems. After so many years, any objection that comes readily to mind has probably been thought of before. We encourage those who are just encountering the ideas of MM to work through the initial skepticism and misunderstanding that comes from unfamiliarity, recognizing that a large number of scientists have been unable to identify any showstoppers. Although the theory has not yet reached the point of being proved by the existence of a nanofactory, it has reached the point where a conversation that assumes most of it is correct will be more productive than a conversation that assumes it's fatally flawed.

The following is an imagined conversation between an MM researcher (MMer) and a room full of scientists who are new to the ideas.

MMer: OK, we're going to build a system that does a billion chemical reactions, one after the other, on the same molecule, with very high reliability.

Chemist: Wait a minute. 99% is an excellent yield, but 99% times 99% times 99%... a billion times is a big fat ZERO. You would reliably get zero molecules of desired product.

MMer: A chemist is used to reactions between molecules that bump into each other randomly. In molecular manufacturing, the molecules would be held in place, and only allowed to react at chosen locations. Yield could be many "nines" better than 99%.
 

MMer: So we take a system that does chemical reactions by holding molecules and moving them into place through a vacuum...

Chemist: Wait. You're going to hold the molecules in a vacuum and make them react as you want? Chemistry's more complex than that; you need more control, and you may even need water to help out with really complex reactions.

MMer: Yes, chemistry is complex when you have lots of potentially reactive molecules bumping around. But if the motion of the molecules is constrained, then the set of potential reaction products is also constrained. Also, there are new kinds of freedom that traditional chemistry doesn't have, including freedom to select from nearly identical reaction sites, and freedom to keep very reactive molecules from touching anything until you're ready. And by the way, even enzymes evolved for water don't necessarily need water -- this has been known since the mid-80's.
 

MMer: So we move the molecules into place to transfer atoms...

Chemist: Atoms are more reactive than that.

MMer: MM wouldn't be grabbing individual unbound atoms -- it would transfer molecular fragments from a "tool" molecule to a "workpiece" molecule, in reactions that work according to standard chemistry laws.
 

MMer: We add a few atoms at a time to build molecular shapes...

Biochemist: Proteins make molecular shapes, and they are very, very hard to design.

MMer: Natural proteins are indeed hard to understand. They have to fold into shape under the influence of a large number of weak forces. But even with proteins, desired shapes have been engineered. DNA, another biomolecule, is a lot easier to design shapes with. And MM plans to build three-dimensional shapes directly, not build long stringy molecules that have to fold up to make shapes.
 

MMer: Then we're going to use that system to build nanoscale machine components...

Micro-mechanical system researcher: Wait a minute! We've tried building machine components, and friction kills them. The smaller you make them, the worse it gets.

MMer: The micromachines were built with a fabrication technique that left the surfaces rough. Friction and wear between rough surfaces are in fact worse as machines get smaller. But if the surfaces are atomically precise and smooth, and the atoms are spaced differently on the two surfaces, they can have extremely low friction and wear. This has been verified experimentally with nested carbon nanotubes and with graphite sheets; it's called "superlubricity."
 

MMer: Assemble the components into nanoscale machines...

Molecular biologist: Why not use machines inspired by nature? Biology does a great job and has lots of designs we could adapt.

MMer: This isn't an argument against the feasibility of MM. If biology-based designs work even better than mechanical designs and are more convenient to develop, then MM could use them. The main advantage of biology is that a technical toolkit to work with biomolecules has already been developed. However, there are several fundamental reasons why biomachines, as good as they are, aren't nearly as good as what MM expects to build. (For example, any machine immersed in water must move slowly to avoid excessive drag.) And mechanical designs will almost certainly be easier to understand and engineer than biological designs.
 

MMer: So we take a bunch of these machines and control them...

Nanotechnologist: How can you hope to control them? It's very, very hard to get information to the nanoscale.

MMer: MM intends to build nanoscale data-processing systems as well as machines. And MM also proposes to build large and multi-scale systems that can get info to the nanoscale without requiring external nanoscale equipment to do so.
 

MMer: We control the machines to build more machine components, one deposition at a time...

Skeptic: That'll take forever to build anything!

MMer: It would indeed take almost forever for a large scanning probe microscope to build its own mass of product. But as the size of the tool decreases, the time required to build its own mass shrinks as the fourth power of the size. Shrink by 10X, decrease the time by 10,000X. By the time you get down to a 100-nanometer scanning probe microscope, the scaling laws of volume and operation frequency suggest it should be able to build its own mass in about 100 seconds.
 

MMer: Then we'll combine those machine components into large products...

Skeptic: You plan to build large products with nanoscale systems? It'll take billions of years!

MMer: MM won't be using just a few nanoscale systems; it'll be using huge numbers of them, working together under the control of nanocomputers. Each workstation will build one tiny sub-part.
 

MMer: So we take huge numbers of machines, arrayed in a factory...

Self-assembly expert: Whoa, how do you plan to put together this factory? Self-assembly isn't nearly there yet.

MMer: Use a factory, with robotic component-handling etc., to make a factory. Use a small factory to make a bigger factory. (The first tiny sub-micron factory would be made painstakingly in the lab.)
 

MMer: So we take this factory and make another bigger factory...

Skeptic: Wait, how can you have a dumb machine making something more complex than itself? Only life can do things like that.

MMer: The complexity of the manufacturing system is the physical system plus the software that drives it. The physical manufacturing system need not be more physically complex than the thing it makes, as long as the software makes up the difference. And the software can be as complex as human brains can design.
 

MMer: We take this big factory and make a product...

Mechanical engineer: How are you going to design a product with zillions of parts?

MMer: The product will not have zillions of different parts. It will have to be engineered in a hierarchical approach, with well-characterized re-usable structures at all levels. Software engineers design computer programs along these lines; the technique is called "levels of abstraction."
 

MMer: Download a blueprint to the factory to make a product...

Programmer: The factory would need amazingly advanced software to run zillions of operations to build zillions of parts.

MMer: Just as the product would contain zillions of parts, but only relatively few distinct parts, so the nanofactory would contain relatively few different types of machines to be controlled. The blueprint file format could be designed to be divided into hierarchical patterns and sub-patterns. Distributing the file fragments to the correct processors, and processing the instructions to drive the workstations, would be straightforward operations.

And so on. As you can see, each objection brought by intuition from within a specific field has an answer that comes from the interdisciplinary approach of molecular manufacturing theory. We are not, of course, asking anyone to take it on faith that molecular manufacturing will work as planned. We are only asking newcomers to the ideas to refrain from snap judgments that it can't work for some apparently obvious reason.

History of the Nanofactory Concept
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

When CRN talks about molecular manufacturing, we usually focus on one particular implementation: a nanofactory. A nanofactory is basically a box with a whole lot of molecular manufacturing machines inside; feedstock and energy go in, and products come out. But why do we focus on nanofactories? Where did the idea come from? I'll tackle the second question first.

Richard Feynman is often credited as a founder of nanotechnology, though the word would not exist until decades after his now famous talk, “There's Plenty of Room at the Bottom,” in 1959. In that talk, Feynman proposed that machines could build smaller machines until the smallest of them was working with atomic precision, and indeed “maneuvering things atom by atom.” Materials could be built under direct control: “Put the atoms down where the chemist says, and so you make the substance.” Along the way to this goal, he said, “I want to build a billion tiny factories, models of each other, which are manufacturing simultaneously...” However, these factories would have been on the border between microtech and nanotech, with individual machines larger than 100 nanometers. Atom manipulation would come “ultimately---in the great future.”

In the 1980's, Eric Drexler introduced most of the ideas of molecular manufacturing (then called simply “nanotechnology”). However, instead of using machines to make smaller machines, Drexler's plan started directly with molecules engineered to have mechanical functionality. Build a number of intricate molecules, he said, join them together into a programmable robotic system, and that system could be used to perform more molecule-building and joining operations.

Both Feynman and Drexler recognized that small machines can't do much individually. Feynman planned to have his manufacturing process make multiple copies of each tiny machine in parallel, growing the number exponentially with each stage of shrinkage. Drexler, starting from nanoscale machines, planned to design his machine so that it could build a complete duplicate. The first machine would build two, then they would build four, then eight, and so on. This is actually an easier problem in many ways than designing a factory to build smaller machines than those in the factory.

Drexler was working from a biological model, in which cells build more cells. Rather than designing a factory, Drexler pictured vast numbers of self-contained, independent robotic fabrication systems. The systems, “assemblers,” were intended to cooperate to build large products. In his 1986 book Engines of Creation, Drexler described a vat of assemblers, floating in fluid, building a rocket engine.

By 1992, when he published Nanosystems, Drexler's plans had evolved somewhat. Instead of vast quantities of free-floating assemblers, each with its own manufacturing system, control system, power system, shell, and chemical input system, he planned to fasten down vast numbers of manufacturing devices into a framework. Instead of cooperating to attach molecules to an external product, each manufacturing workstation would build a tiny fragment of the product. These fragments would then be combined into larger and larger components, using a system much like a tree of assembly lines feeding larger assembly lines.

Drexler's nanofactory proposal in Nanosystems was to be refined several times. In Drexler's proposal, the assembly lines occupied a three-dimensional branching structure. This structure is more complex than it looks, because some of the smaller lines must be bent aside in order to avoid the larger ones. In Merkle's 1997 refinement, the assembly lines occupied a simpler stacked configuration. The price of this is constraining the allowable dimensions of sub-parts. Essentially, Merkle's system works best if the product is easily divisible into cubes and sub-cubes.

In my 2003 paper “Design of a Primitive Nanofactory”, I continued to use a convergent assembly approach, accepting the limitations of dividing a product into sub-cubes. Another limitation that should be noted with convergent assembly is that the product must be small enough to fit in the assembly line: significantly smaller than the factory. The paper includes an entire chapter on product design, much of which is guided by the problems inherent in building diverse products out of small dense rigid multi-scale cubes. Basically, the plan was to build the product folded up, and then unfold it after completion and removal from the nanofactory. My design, as well as Drexler's and Merkle's, required large internal factory volumes for handling the product in various stages of completion.

A few months after my Primitive Nanofactory paper was published, John Burch and Eric Drexler unveiled their newest nanofactory concept. Instead of many levels of converging assembly lines, the Burch/Drexler factory design deposits tiny blocks directly onto a planar surface of a product under construction. Although this requires many thousands of deposition operations at each position to build each centimeter of product, the process is not actually slow, because the smaller the blocks are, the faster each one can be placed. (Although the physical layout of my nanofactory is now obsolete, most of the calculations in my paper are still useful.)

Instead of requiring the product to be divisible into sub-cubes at numerous size scales, the Burch/Drexler architecture requires only that the product be made of aggregated tiny components—which would be necessary in any case for anything constructed by molecular manufacturing workstations. Instead of requiring a large internal volume for product handling, the factory only needs enough internal volume to handle the tiny components; the growing product can be attached to an external surface of the factory.

Focus on the Factory

So, that is how the nanofactory concept has evolved. Why does CRN use it as the basis for talking about molecular manufacturing? The answer is that a nanofactory will be a general-purpose manufacturing technology. Although it could not build every product that could possibly be built by molecular manufacturing, it will be able to build a very wide range of very powerful products. At the same time, a personal nanofactory would be perhaps the most user-friendly way to package molecular manufacturing. Technologies that are user-friendly, assuming they are adequate, tend to be used more widely than more powerful but less convenient alternatives. Although there may come a time when computer-aided design processes run into the limits of the nanofactory approach, it seems unlikely that humans using current design techniques would be able even to fully map, let alone explore, the range of possible designs.

A nanofactory is easy to conceptualize. At the highest level, it's a computer-controlled box that makes stuff, sort of like a 3D inkjet printer. Add in a couple of key facts, and its importance becomes clear:

1. It can make more nanofactories.
2. Its products will be extremely powerful.
3. Rapid programmable manufacture implies rapid prototyping and rapid design.

It is difficult to see how “diamondoid mechanosynthesis of multi-scale nanosystem-based products” can revolutionize the world. It is much easier to imagine a nanofactory being flown in to a disaster area, used to produce more nanofactories and feedstock factories, and then all of them producing water filters, tents, and whatever else is needed, in any quantity desired—within just a few days.

Nanotechnology today is largely the province of the laboratory, where most people cannot participate. But a personal nanofactory could be made easy enough for untrained people to use, even to the point of making new product designs. This advantage comes with a cost: the simpler the design software, the more limited the range of products. But molecularly constructed products will be so intricate and high-performance that a certain amount of tradeoff will be quite acceptable for most applications. If a design has an array of a thousand tiny motors where one hundred would suffice, that probably would not even be noticeable.

A final advantage of conceptualizing the means of production as a human-scale box is that it helps to separate the production system from the product. In the pre-nanofactory days of molecular manufacturing discussion, when tiny assemblers were the presumed manufacturing system, a lot of people came to assume that every product would include assemblers—and thus be prone to a variety of risks, such as making more of itself without limit. The nanofactory concept makes it much clearer that products of molecular manufacturing will not have any spooky self-replicating attributes, and the manufacturing apparatus itself—the nanofactory—may be about as dangerous as a printer.

Types of Nanotechnology
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

Now that nanotechnology has been in the public eye for twenty years, and well-funded for half a decade, it's worth a quick look at just what it is—and how it got that way.

When the word “nanotechnology” was introduced to the public by Eric Drexler's 1986 book Engines of Creation, it meant something very specific: small precise machines built out of molecules, which could build more molecular machines and products—large, high-performance products. This goal or aspect of nanotechnology now goes by several names, including molecular nanotechnology, molecular manufacturing, and productive nanosystems. The reason for this renaming is that “nanotechnology” has become a broad and inclusive term, but it's still important to distinguish molecular manufacturing from all the other types. I'll talk about molecular manufacturing, and why it is unique and important, after surveying some of the other types of nanotechnology.

With the funding of the U.S. National Nanotechnology Initiative (NNI), there has been a strong financial incentive to define nanotechnology so that one's own research counts—but not so broadly that everyone's research counts. There has been a less focused, but still real, incentive to define the goals of nanotechnology aggressively, to justify major funding, but not too aggressively, lest it sound scary or implausible.

With all the different research fields applying the above rules to a wide variety of research, it is not surprising that there's no single hard-edged definition of nanotechnology that everyone can agree on. Perhaps the most commonly quoted definition of nanotechnology is the one used by the NNI: “Nanotechnology is the understanding and control of matter at dimensions of roughly 1 to 100 nanometers, where unique phenomena enable novel applications.” I don't know how they decided on the size scale; thinking cynically, it might have had something to do with the fact that computer chips were just about to gain features smaller than 100 nanometers, so they were guaranteed at least one early success.

Nanotechnology can be even broader than that. A rough rule of thumb is: if it's too small to see with an ordinary light microscope, it's likely to be considered nanotechnology. Without using special physics tricks, light can't be used to see anything smaller than half a wavelength of light, which is a few hundred nanometers (I can't be more precise because light comes in different colors with different wavelengths). Because some optics technology uses structures smaller than light (such as photonic crystals) to manipulate light, you will sometimes see optics researchers describe their work as nanotechnology. However, because these structures tend to be larger than the official 100-nm cutoff, many nanotechnologists will reject this usage.

Another point of contention is how unique the “unique phenomena enabl[ing] novel applications” have to be. For example, some nanotechnology simply uses ordinary materials like clay, in smaller chunks, in fairly ordinary ways. They can get new material properties; they are using nanoscale materials; they are studying them with new techniques; but is it really nanotechnology, or is it just materials science? It might as well be called nanotech, seems to be the consensus. It's providing early successes for the field and it’s putting “nano” into consumers' hands in a beneficial, non-threatening way.

Another kind of nanotechnology involves building increasingly large and intricate molecules. Some of these molecules can be very useful: for example, it appears possible to combine a cancer-cell-recognizer, a toxic drug, and a component that shows up in MRI scans, into a single molecule that kills cancer cells while showing you where they were and leaving the rest of the body untouched. This is a little bit different from traditional chemistry in that the chemist isn't trying to create a new molecule with a single function, but rather to join together several different functions into one connected package.

Some new nanomaterials have genuinely new properties. For example, small mineral particles can be transparent to visible light, which makes them useful in sunscreen. Even smaller particles can glow in useful colors, forming more-stable markers for biomedical research. For related reasons, small particles can be useful additions to computer circuits, lending their quantum effects to make smaller and better transistors.

We should talk about semiconductors (computer chips), a major application of nanotechnology. Feature sizes on mainstream silicon chips are well below 100 nanometers now. This obviously is a great success for nanotechnology (as defined by the NNI). From one point of view, semiconductor makers are continuing to do what they have always done: make chips smaller and faster using silicon-based transistors. From another point of view, as sizes shrink, their task is rapidly getting harder, and they are inventing new technology every day just to keep up with expectations. There are more unusual computer-chip designs underway as well, most of which use nanotechnology of one form or another, from quantum-dot transistors to sub-wavelength optics (plasmonics) to holographic storage to buckytube-based mechanical switches.

Which brings us to buckytubes. Buckytubes are remarkable molecules that were discovered not long ago. They are tiny strips of graphite, rolled up with the sides fastened together to form a seamless tube. They are very strong, very stiff, and can be quite long in proportion to their width; four-centimeter long buckytubes have been reported, which is more than ten million times the width of the tube. Some buckytubes are world-class conductors and electron emitters. They may be useful in a wide variety of applications.

And what about those quantum effects? According to the NNI, “At the nanoscale, the physical, chemical, and biological properties of materials differ in fundamental and valuable ways from the properties of individual atoms and molecules or bulk matter.” Materials are of course made up of atoms, which contain electrons, and it is the interaction of electrons that gives materials most of their properties. In very small chunks of material, the electrons interact differently, which can create new material properties. Nanoparticles can be more chemically active; as mentioned above, they can fluoresce; they can even participate in weird physics such as quantum computers. But, as the above overview should make clear, a lot of “nanotechnology” does not make use of these quantum effects.

Molecular manufacturing (MM) is a fairly mundane branch of nanotech, or it would be if not for the political controversy that has swirled around it. The idea is simple: Use nanoscale machines as construction tools, joining molecular fragments into more machines. Every biological cell contains molecular machines that do exactly that. There are, however, a few reasons why molecular manufacturing has been highly controversial.

Much of the controversy stems from the fact that MM proposes to use engineered devices to build duplicate devices. Although biology can do this, intuition suggests that such self-duplication requires some special spark of complexity or something even more numinous: surely a simple engineered machine can't be so lifelike! This ultimate spark of vitalism is fading as we learn how machinelike cellular molecules actually are, and as increasingly detailed plans make it clear that hardware does not have to be very complex in order to make duplicate hardware. (Even the software doesn't have to be very complex, just intricate and well-designed. This has been known by computer scientists for many decades, but the paradigm has taken a while to shift in the wider world.)

There is another problem with self-replication: in some forms, it may be dangerous. In 1986, Eric Drexler warned that tiny engineered self-replicators could outcompete natural life, turning the biosphere into boring copies of themselves: “grey goo.” This formed a cornerstone of Bill Joy's essay “Why The Future Doesn't Need Us,” which hit just as the NNI was ramping up. No nanoscientist wanted to be associated with a poorly-understood technology that might destroy the world, and the easiest thing was to assert that MM was simply impossible. (Modern MM designs do not use small self-replicators; in fact, they have been obsolete since Drexler's 1992 technical book Nanosystems.)

A third source of controversy is that MM plans to use diamond as its major building material, not bio-based polymers like protein and DNA. (Some pathways to this capability, including the pathway favored by Drexler, go through a biopolymer stage.) Although there is a wide variety of reactions that can form diamond and graphite, living organisms do not build with these materials, so there is no existence proof that such structures can be built using point-by-point computer-controlled molecular deposition.

If diamond-like structures can be built by molecular manufacturing techniques, they should have astonishingly high performance characteristics. To those who study MM, its projected high performance indicates that researchers should work toward this goal with a focused intensity not seen since the Manhattan Project. To those who have not studied MM, talk of motors a million times more powerful than today's merely seems fanciful, a reason (or an excuse) to discount the entire field.

At least as problematic as the extreme technical claims are the concerns about the extreme implications of molecular manufacturing. It is rare that a technology comes along which revolutionizes society in a decade or so, and even more rare that such things are correctly predicted in advance. It is very tempting to dismiss claims of unstable arms races, wholesale destruction of existing jobs, and widespread personal capacity for mass destruction, as improbable.

However, all the skepticism in the world won't change the laws of physics. In more than two decades (almost five, if you count from Richard Feynman's visionary speech), no one has found a reason why MM, even diamond-based MM, shouldn't work. In fact, the more work that's done, the less complex it appears. Predicting social responses to technology is even more difficult than predicting technology itself, but it seems beyond plausibility that such a powerful capability won't have at least some disruptive effects—perhaps fatally disruptive, unless we can understand the potential and find ways to bypass the worst pitfalls.

In the near future, nanotechnology in the broad sense will continue to develop dozens of interesting technologies and capabilities, leading to hundreds of improved capabilities and applications. Meanwhile, molecular manufacturing will continue to move closer, despite the (rapidly fading) opposition to the idea. Sometime in the next few years, someone will have the vision to fund a targeted study of molecular manufacturing's potential; less than a decade after that, general-purpose nanoscale manufacturing will be a reality that the world will have to deal with. Molecular manufacturing will build virtually unlimited quantities of new products as rapidly as the software can be designed—and it should be noted that most of today's physical products are far less complex than today's software. Molecular manufacturing will both enable and eclipse large areas of nanotechnology, further accelerating the achievements of the field. We are in for some interesting times.

Bottom-up Design
Chris Phoenix, Director of Research, Center for Responsible Nanotechnology

At first encounter, the idea of designing products with 100,000,000,000,000,000,000,000 atoms, each in an engineered position, and each one placed without error, may seem ridiculous. But the goal is not as implausible as it sounds. Today's personal computers do that number of transistor operations every few weeks. The operations are done without error, and each one was engineered—though not directly. There are two reasons why computers can do this: digital operations and levels of abstraction. I've talked about both of these in previous essays, but it bears repeating: at the lowest level of operations, personal computers do a mole of engineered, reliable transistor operations every few weeks, and the techniques used to accomplish this can be applied to molecular manufacturing.

Computers can be so precise and reliable because they are based on digital operations. A digital operation uses discrete values: either a 1 or a 0. A value of 0.95 will be corrected to 1, and a value of 0.05 will be corrected to 0. This correction happens naturally with every transistor operation. Transistors can do this correction because they are nonlinear: there is a large region of input where the output is very close to 1, and another large region of input where the output is very close to 0. A little bit of energy is used to overcome entropy at each step. Rather than letting inaccuracies accumulate into errors, they are fixed immediately. Thermal noise and quantum effects are corrected before they compound into errors.

Forces between atoms are nonlinear. As atoms approach each other, they feel a weak attractive force. Then, at a certain distance, the force becomes repulsive. If they are pushed together even more closely, the force becomes more strongly attractive than before; finally, it becomes sharply repulsive. Chemists and physicists know the region of weak distant attraction as “surface forces”; the closer, stronger attraction is “covalent bonds”; and the intervening zone of repulsion is responsible for the “activation energy” that is required to make reactions happen. Of course, this picture is over-simplified; covalent bonds are not the only type of bond. But for many types of atoms, especially carbon, this is a pretty good description.

Several types of errors must be considered in fabricating and using a mechanical component. A fabrication operation may fail, causing the component to be damaged during manufacture. The operations may be correct but imprecise, causing small variances in the manufactured part. During use, the part may wear, causing further variance. As we will see, the nonlinear nature of molecular bonds can be used (with good design) to virtually eliminate all three classes of error.

Nonlinear forces between atoms can be used to correct inaccuracies in fabrication operations before they turn into errors. If the atom is placed in slightly the wrong location, it will be pulled to the correct location by inter-atomic forces. The correction happens naturally. If the placement tool is inaccurate, then energy will be lost as the atom moves into place; as with transistors, entropy isn't overcome for free. But, as with transistors, reliability can be maintained over virtually unlimited numbers of operations by spending a little bit of energy at each step.

In practice, there are several different kinds of errors that must be considered when a moiety—an atom or a molecular fragment—is added to a part under construction. It may fail to transfer from the “tool” molecule to the “workpiece” molecule. This kind of error can be detected and the operation can be retried. The moiety may bond to the wrong atom on the workpiece. Or it may exert