On Mechanosynthesis
Posted by Jeriaska on January 21st, 2008
Ralph Merkle is a pioneer in public key cryptography and an expert on the emerging technological applications of molecular nanotechnology and cryonics. He is a key member of the Institute for Molecular Manufacturing, part of the Nanofactory Collaboration. In the popular science fiction novel The Diamond Age by Neal Stephenson, Dr. Merkle is portrayed as one of the heroes of a future civilization where nanotechnology is ubiquitous. At the CRN conference on the Future of Nano and Bio he spoke on the subject of “A Minimal Toolset for Diamond Mechanosynthesis,” a paper with Robert Freitas on molecular nanotechnology scheduled to be published in JCTN (Journal of Computational and Theoretical Nanoscience).
The following transcript of Ralph Merkle’s CRN Conference presentation “On Mechanosynthesis” has not been approved by the author.
On Mechanosynthesis
So, I will be talking about mechanosynthesis. The reason is fairly simple. Many years ago Rob Freitas and I were at Zyvex and we were looking at what is involved in building an assembler. We sat down and designed an assembler and then we presented this design at Zyvex. We sorted through the various criticisms and found that at Zyvex, as at many places, the major concern was what do these molecular tools look like? What is the chemistry and what are the reactions? As a consequence we started looking at mechanosynthesis, what are the chemical reactions involved in actually building diamond?
A link that provides information to further information: www.molecularassembler.com. That is a discussion of a lot of these things related to how you build nanofactories. We are interested in the long term design and eventual construction of these molecular machine systems.
You’ve already seen this. There’s coal, there’s diamond, and they are really made out of the same elements. There’s a computer chip on top a pile of sand. That person on the bottom is unhealthy. The person who is bicycling along and smiling is healthy. Again, the difference between being unhealthy and healthy is the arrangement of atoms. As a consequence, we find a lot of people who have an interest in medicine are interested in nanotechnology. One of Rob Freitas’s books is Nanomedicine. In fact, there are other books in that series that are coming out, which will be discussing the various medical applications of nanotechnology.
We have seen basically three trends in manufacturing over the last many decades, even over the last many centuries. Flexibility in the things that we manufacture. We are able to build a wider variety of things today than we could in the past. Greater precision and lower manufacturing costs.
As we extend those trends into the future, we find that we should be able to achieve some rather remarkable things, which were pointed out by Feynman in the talk There’s Plenty of Room at the Bottom, in which he said we ought to be able to arrange atoms. We ought to be able to have tiny factories that can build and arrange. It does not seem to evade any physical law. It looks like something that is feasible, but has not yet been done.
Let’s suppose that we can build pretty much what we want. What do we want? The standard answer is that we want diamond. Why do we want diamond? There are the various reasons. It’s stronger, lighter, stiffer, it has better electrical properties, a better band gap, and a better refractive index. Pretty much, diamond is either the best or close to the best in most of the categories you can describe for various materials properties. So, it’s not surprising that attention turns to diamond when we say we can build what we want to build.
Just to give you a more concrete illustration of things made out of hydrogen and carbon atoms, here we have a bearing. It’s a theoretical bearing; it hasn’t been built. But if our understanding of basic chemistry is correct, it looks like that ought to work if we could build it.
Here we have a universal joint. Again, it hasn’t been built, but it looks like if we could build that kind of structure, it should work pretty much the way we think it should.
We can look at other elements in the periodic table. If we have a broader capability of synthesized structures with a lot of elements, you can start to have more colorful structures. These are more commonly seen in the press because they’re colorful. The oxygen is red, the sulfur is yellow, the hydrogen is white, the nitrogen is blue, and the carbon is black. Like I say, the hydrocarbon stuff gets short shrift. It only shows up occasionally in newspaper articles where they have black and white processing. This is something that we don’t know how to build today. If you built it, it looks like it ought to work. There have been some simulations of this down at Caltech. Indeed, it looks like it does work if you could build it.
This is a neon pump. There is the core of the pump on the right. You have the housing for the pump on the left. If you slide this core into the housing and rotate it, then it will pump neon atoms. Again, that was simulated down at Caltech and it looks like it ought to work reasonably well. It looks like it will probably pump a couple other atoms of similar size, but that’s not a critical point. Those were some examples of things we could make out of diamond: a rather different kind of design style than you see in standard organic synthesis.
So, we want to build diamond. How do you build diamond? Today, we can build diamond by various techniques. One of the techniques is to use diamond CVD (chemical vapor deposition). You take hydrogen and some source of carbon like methane. You stir in energy. You heat it or microwave it, and you get highly reactive molecular fragments that bang into a surface, and you grow diamond films. There are various chemical reactions that take place in that growth of diamond film.
We need to have something more than a gaseous vapor banging into a surface. In other words, if you have a gaseous vapor banging into a surface, you don’t have a lot of control over what you are building. What we would like to do is have a synthetic strategy for the synthesis of these diamonds involving positional control.
In the chemical vapor deposition growth of diamond you have a gas with particles banging into the surface at random. What we would like to do instead is have molecular tools selectively designed to interact with the surface at a particular spot and are moved away. The tools modify the surface in some controlled fashion so you have positional control of the tools. You would use highly reactive compounds. It looks as though that is a useful strategy for building diamond. In fact, it looks like a useful strategy for building up a wide array of structures. You also need an inert environment. If you are going to have radicals, you had better not have a reactive environment. So you need an inert environment like vacuum. That’s the general strategy.
Here is one proposal for a molecular positional device. Positional devices can be small. One of these days we should be able to build very small positional devices. This positional device has the property that even at room temperature it has quite good stability. I would like to run you through the basic equations for stability in the face of thermal noise.
On the left you see sigma is mean positional error. In other words, if I have a positional device and it’s at some temperature, it’s going to wiggle because of thermal noise. Boltzmann’s constant in this case is that fudge factor required to make the equation work. You don’t have to worry about what it is; it’s constant. T is the temperature in Kelvins. What does this equation say? The warmer it gets, the more positional uncertainty you have from thermal noise. Not surprising.
In the denominator we have k, the stiffness of the system. If I hold out my arm, it’s reasonably stiff. If I brace it, it’s stiffer. So you get better control over the position and get greater stiffness. You get better positional accuracy at given temperatures. So there are two parameters we can play with. One is the temperature, and the other is the stiffness of the system. If you play with those parameters, you can get a wide range of positional accuracies. In particular, if you plug in some plausible looking parameters, you can get positional accuracy at a fraction of an atomic diameter, two tenths of an angstrom. An angstrom is about the size of a hydrogen atom. k, this stiffness constant. You have scanning probe systems today with a k of 10 Newtons per meter. That’s a fairly typical stiffness.
At room temperature, if you have a molecular robotic arm with a stiffness of 10 Newtons per meter, you get an accuracy of two tenths of an angstrom. I should mention that there have been various people who have said thermal noise poses a fundamental problem in the manufacture of molecular machines. They ignore existing biological molecular machines, and they ignore this analysis, and indeed they seem to ignore everything. If you see people making doubtful statements about the feasibility of molecular machines just keep in mind that they’re wrong. I’ve seen lots of them and they’re just amazingly wrong.
On a more technical note, if you want to find out more, I’m going to give you a teaser about mechanosynthesis. There are a bunch of papers on the subject. There is an annotated bibliography up at molecularassember.com. It’s got over fifty entries in it. This is on a wide range of molecular tools. Tools that involve hydrogen, carbon, various other thing.
Now what I would like to do is show you the tool that has been analyzed the most. This tool was proposed well over a decade ago. The idea here is you have an ethynyl radical. The chemical structure is shown on the left. This ethynyl radical has a higher affinity for hydrogen than almost any other structure. This simulation was done a number of years ago, and it shows the abstraction of a hydrogen atom from a diamond (111) surface. On top we have the tool, the tool is brought down slowly to the surface, it pops off a hydrogen atom from the surface, and illustrates the basics of mechanosynthesis.
That’s been studied for a long time, and there are a lot of papers on it. This is one of those slides that is flashed up on the screen to impress you. In looking at various pieces of advice on how to present a talk, the usual statement is that one slide in your talk should be incomprehensible so that everyone will be properly impressed with how scientific it is. So, here it is. There are references to published papers.
So that was the hydrogen abstraction tool. Then when we were at Zyvex we started looking at tools that would allow you to deposit carbon on a surface. In fact, because of the resources that Jim Van Ehr made available at Zyvex, a lot of computational power as well as computational chemists and a lot of ability to purchase software, we got an analysis of a dimer placement tool. This is a tool which has two carbon atoms next to each other on the tip of the tool. The tool comes up to the surface and the dimer is deposited on the surface. By the way, one of the nice things about these computational models, if you put in more computational time, you can get better results. Notice I said you can get better results. You don’t have to get better results. So, more computational time, more accuracy. And computational power, as we all know, is increasing exponentially. As the years go by, it gets easier and easier to produce this kind of data and to have a large number of people who can publish a paper describing these computational tools.
So that’s a second tool that has been analyzed. The original idea of molecular manufacturing was to build with a full palette. There are over a hundred elements in the periodic table. Let’s have over a hundred tools. Each tool would have a different element in it, and just start building away. It turns out if you do that, the analysis is complicated. You get a combinatorial explosion in all those possibilities. The good news is though, you can build a huge range of structures and you also get good flexibility in the synthetic process because you have a wide range of tools.
Rob and I have a new paper in preparation. We’ve just sent it off. It is on a minimal toolset for positional diamond mechanosynthesis. We picked three elements: hydrogen, carbon, and germanium. To build diamond, you have to have hydrogen and carbon. You can also build Lonsdaleite, which is a hexagonal version of diamond. You can build graphite and buckytubes, fullerenes, organic compounds. So there are a lot of things you can build if you have hydrogen and carbon.
It turns out that when you start looking at the synthesis of hydrocarbons, it’s a little bit tricky if you want to have tools made out of hydrogen and carbon while building hydrogen and carbon. We tried out various elements, but finally we said it looks like we can get away with germanium as providing for synthetic flexibility. It does not add too much complexity to the toolset itself. Remember, every time you add an element, you also have to add all the tools required to synthesize structures with that element in it, too. You at least have to be able to synthesize the tools that use that element to build other things.
We picked those three elements, for better or for worse. Then we went ahead and started analyzing what would go on. It turns out, like I said, computing power is getting cheaper, so we could actually look over a fair number of reactions. One of the things we wanted to do was actually have the full set of reactions. In other words, we didn’t want to say, “Well, here’s a few, and you can imagine the rest.” We wanted to say, “Here it is, folks! Here is the set of reactions.” And we actually wound up looking at a large number of structures, over 1,500. There are 65 reaction sequences, 328 reaction steps, a lot of pathological side reactions. If you bring a tool up to a surface, sometimes it does what you want, and sometimes it doesn’t do what you want. So you have to look at the various ways it can do what you don’t want it to do, and make sure there are barriers to those bad reactions. Mind you, this actually means we tried out a lot of stuff and threw it away. We have now reached that glorious state of affairs where we too can fuss around and get absolutely nowhere. It’s not just an experimental model. We blew on the order of 100,000 hours of computer time. This is just on gigahertz processors.
Computational methods. We used a standard off-the-shelf computational chemistry program called Gaussian. We used Gaussian 98 because we could get a copy of that cheap. There are actually more recent versions, but we didn’t use those. When you ask Gaussian to analyze a structure, you have to tell it what the spins of the electrons are. You also have to tell it what sort of approximations to use. In other words, none of these programs solve Schrödinger’s wave equations exactly. They all use a variety of approximations. You have to tell it exactly which kinds of approximations you are using. So, we are using B3LYP with a 3-21G* basis set. Yeah, science!
But there’s more. It turns out when we are optimizing the geometry, we used this 3-21G* basis set. It turns out, if you’re a computational chemist, it’s a pretty goofy basis set. When we actually started doing the energy calculations, we used a 6-311+G(2d,p) basis set. You can tell that’s better because it has more punctuation marks and bigger numbers. I’m serious. A fast way, if someone gives you two basis sets, just look at the size of the numbers and the number of punctuation marks. In fact, in the computational chemistry community, no one knows whether these approximations are any good. The way you figure it out is you say, okay, we have an approximation. Let’s take a whole bunch of structures where we know what is going on and analyze them. You analyze them using your computational method and you see if your analysis matches the experimental data.
If you do that, you find that the mean absolute disparity–or MAD, as it’s called in the literature– is pretty small. It’s a little over a tenth of an electron volt, which is good enough for the work we’re doing, and should let us describe all of the reactions and get them qualitatively correct, and even get pretty close on a quantitative analysis. Now, in most of the cases, we have barriers of four-tenths of an electron volt against side reactions. Undesirable side reactions are selected against by about four-tenths EV. That’s the computational background.
What are out molecular tools?
Here they are. Nine of them. This is a hydrogen abstraction tool. It has been around, like I say, for over a decade. It has an ethynyl radical sticking out on the edge.
Here we have a hydrogen donation tool. Remember I said germanium was convenient? One of the reasons is that it forms weak bonds to hydrogen. So the hydrogen donation tool holds onto a hydrogen weakly, so you can donate a hydrogen to other structures when you want to.
We have the GM tool. What is that? That’s a germanium atom, that’s a carbon, and those are two hydrogens. This carbon here has three bonds, and carbon wants four. So that’s an unhappy carbon, otherwise known as a radical. As a consequence, that carbon atom is pretty active. If you bring that carbon atom up to a reactive spot on a surface, it will react. If you pull away, the germanium-carbon bond is weaker than the carbon-carbon bond. So you can bring it up to a surface, have it react with the carbon on the surface, pull the tool away, and cleavage will appear between the germanium and the carbon.
We have this germanium, so we have to have a tool that you can use to build germanium. Here is the tool for building the germanium. Again, it’s got a germanium, it’s got two hydrogens, so it’s a highly reactive form of germanium. You can bring it up to a surface, and you have to be careful when you are pulling it away.
Here we have a methylene tool. This is again a highly reactive carbon atom. Instead of having a germanium we have a carbon atom with an adamantane bridgehead. So we now have another tool which can bring up a carbon atom. It’s a little bit harder to pull away. When you look at this tool being used, you have to bring it up, you have to do some fiddling, to have it let go of the carbon atom.
Here we have a hydrogen transaction tool. This is almost like the hydrogen abstraction tool, except it’s got a hydrogen on the tip. We approach the carbon atom here with germanium. That means this is a chemically unhappy structure. It’s so unhappy it really would like to get rid of that hydrogen. In fact, the hydrogen is very weakly bonded in this structure. You can use this to donate a hydrogen. It’s a useful hydrogen donation tool. In fact, it’s so good that you can actually donate to the germanium radical and make that work. The other thing is, once you have donated the hydrogen you can pull away the germanium radical and suddenly you have a hydrogen abstraction tool that has a very high affinity for hydrogen. You can modulate the affinity for hydrogen of this tool by moving away the germanium radical or moving it up. So that’s a useful tool.
There’s the adamantane radical. You take the adamantane cag, pop off a hydrogen from the top, and voilà, you have a radical. It’s useful for various things involving a radical. Here we have the dimer placement tool. That’s the one that has been analyzed going back to Zyvex days. Finally, we have the germanium radical. This is simply a germanium on adamantane. It’s a radical, so it’s n unhappy germanium. It wants to bond with something.
Those are the nine tools. That’s it. We’re assuming we have high accuracy positional control. Remember, there were two ways of getting positional control. One was to have a stiff tool, so that you can position despite thermal noise. The other is to lower the temperature. We can get positional accuracy one way or another. We can also get a control device or a mechanical device to position them. But this is the core set of tools. When people ask, “What do the tools look like?” The answer is “There they are.” That’s the proposal.
What are the reactions? Let’s start out with hydrogen donation. Suppose I want to donate a hydrogen from our hydrogen donation tool to a radical site on a diamond surface. We have a donation onto a C (111) surface. That is energetically favored by .61 eV. In other words, the hydrogen wants to be on the surface. It prefers that to being on the hydrogen donation tool. It prefers it by .61 eV. Take the donation tool, bring it up, you push a bit, and the hydrogen is on the surface.
Or again, we can do hydrogen donation on the C (110) surface. If you bring up the hydrogen donation tool on the (110) surface, then you get .73 eV. It prefers it slightly more. So, we are looking at different diamond surfaces. You want to donate a hydrogen to the surface, pick your surface and your tool, decide what sort of energy preference you are going to get based on these quantum chemistry calculations. Here again we have our hydrogen transaction tool and a radical on the C (110) diamond surface. You bring up this tool. Remember, this hydrogen is bonded more weakly to this carbon than even it is to the germanium. It turns out when you look, you find that the energy is even greater. The hydrogen really prefers much more being on the carbon surface, even in the case of germanium. Again, this tool has a preference for donating a hydrogen to the surface. That’s a mechanism for doing hydrogen donation.
So you’re beginning to get the flavor of this. I can also recharge the hydrogen abstraction tool. You’ve got germanium radicals and you’ve got a hydrogen abstraction tool there. There’s our hydrogen abstraction tool, and it’s got a hydrogen stuck onto it, remember? You have a hydrogen abstraction tool, it abstracts a hydrogen, now it’s got a hydrogen stuck to its tip. It’s a really high affinity bond, so how do you break it? Well, you come up with a radical and you bond to the carbon atom at the tip of the abstraction tool. You come up with another germanium radical and bring it up to the hydrogen, and the hydrogen pops right off. In fact, it turns out that there are a lot of ways of approaching that tip carbon atom. If you’ve got a germanium radical, you can bring it up and rotate it around in a lot of ways. There are some guys in Russia who are friendly with us–this is one of our collaborative efforts–who went off and published a paper where they analyzed the energy profiles of the germanium radical approaching the tip of the hydrogen abstraction tool. They had a particular place that they liked, where it would just slide right in. So we used that and we get .43 eV. So the germanium radical likes to stick to the carbon atom at the tip of the hydrogen abstraction tool by .43eV. Then you bring up the second germanium radical and it slurps off the hydrogen at .83 eV. The hydrogen likes to be on the germanium radical by .83 eV. As the final step, you pull the germanium radical away from the tip and you have recharged the hydrogen abstraction tool.
This is one of the common questions. I used to give talks where I would say “Here’s the hydrogen abstraction too, and “Here is how you abstract the hydrogen from the surface.” And people would ask, “How do you recharge the tool?” And I left that as an exercise for the reader. It worked–everyone had a good laugh and figured there was a way of doing it. But now here we have it. It’s laid out, it’s analyzed.
Now we’ll start looking at ways of depositing a carbon atom on the surface. This adamantane is going to represent a really small piece of a surface. So we’ll pop off a hydrogen on the surface, and at that spot the surface becomes reactive. Now we have a radical site on the surface. The GM tool comes up and has this carbon atom on the tip. A radical-radical reaction is energetically favored. In fact, it is favored by 3.17 eV. After you have bonded that on, you pull it away, and bring up a hydrogen donation tool. You deposit the hydrogen on the carbon radical and you have donated to the surface.
One other little detail that I thought I would mention. When you pull this germanium radical away, you are actually having to put in 2.76 eV. That’s a fair amount of energy. It turns out that if you look at the potential wells that it actually is the case that it wants to break the germanium-carbon bond, but it turns out to be even stronger than that. Chemists always talk about reactions that are activated by thermal energy. They don’t talk about reactions that are activated by pulling. The rules that apply are slightly different. When you pull, the question is not what is the potential well depths, but what is the strength of the bond. You can actually have bonds that are weaker in terms of their force but stronger in terms of their well depth because they are a longer bond. In this case, the germanium-carbon bond is weaker and has a peak tensile strength which is smaller than a carbon-carbon bond. In other words, if you pull on something and apply force, it will break if you are ignoring thermal energy. If you pull at zero Kelvins, then what dominates is the tensile strength of the bond. The germanium-carbon bond is weaker in that sense as well as being weaker in the chemist’s sense of having lower energy.
That sequence will deposit a carbon atom on a surface. Again, we’re taking a surface, we’re putting a carbon atom on it, we’re putting a carbon atom adjacent to it, and we’re seeing if that works. We’re checking out a lot of reaction pathways to make sure they all work.
This is building a hydrogen abstraction tool from a dimer placement tool. You have a dimer placement tool, you have two carbon atoms, you bring up an adamantane radical and it pulls off the two carbon atoms. You have a hydrogen abstraction tool.
This is making the GM tool. The GM tool is a pretty major tool for depositing carbon on a surface, so having a synthetic reaction that will actually generate a desired GM tool is very important. You can also make other things like polyyne chains. That’s an illustration of what sort of things we can build.
This set of reactions has 100% process closure. In other words, we have all of the reactions laid out. You have the set of tools, you can build another set of tools, and build a bunch of other stuff. There are nine tools. For feedstock you are using methane, Ge2H6, and hydrogen. In this particular proposal we wanted to pick up the feedstock molecules from flat surfaces because that’s more consistent with existing technology. Hopefully these are experimentally accessible. The idea is you have flat diamond germanium surfaces and you display the feedstock molecules on the surfaces.
What is required in the future? If you want to have higher confidence in reaction you can have more than one group doing it with other computational chemistry tools. You can do molecular dynamics to see what happens when you’ve got thermal noise. Also, when you look at the literature, remember that the hydrogen abstraction tool had a bunch of papers, so that’s one reaction. We have a whole bunch of reactions, so presumably a thorough analysis of all the pathways is going to take some work. You also want to start looking at the set of reactions, saying “What are experimentally accessible pathways into this set of reactions?” Here is a set of reactions, can we do them in the lab? for that you have to draw in some experimentalists and there is going to be divergence of opinions.
We have not seen funding on long-term system design by conventional sources of funding. We have seen funding by individuals and occasionally select groups. The work you’ve been seeing, there has been funding from the Alcor Foundation. Alcor, the cryonics organization, is interesting because it has as part of its core mission the development of long-term technologies to revive people who have been cryopreserved. As a consequence, a committee of people involved at Alcor has as part of their core values to develop these kinds of advanced technologies. You also have the Institute for Molecular Manufacturing, Kurzweil‘s foundation, and the Life Extension Foundation, Nanorex. Jim Van Ehr at Zyvex has been kicking in money not only in this work and some of the work on mechanosynthesis.
The useful funding has been coming from individuals. If you look just at today’s experimental work, computational work and theoretical work needs to be funded because if we understand where we are going we are more likely to get there. And yet, it’s not being funded by the standard sources. Maybe this changes in the future, but who knows? If we don’t get this research that provides the context in which experiment can thrive we will wander in the desert for a long time. If you do not know where you are going, it takes longer to get there. That’s the talk.
