A relatively informal visit. We'll have a wonderful seminar here. Afterwards, folks should feel free to hang around and talk with Paul, and at one point he will head on over to MBB I think so he can look around and talk some more, and go eat and drink that's up to him. Grant Wilson. Grant's been pestering me forever to get Paul Rothemund here. Paul has been out with the semitech people this morning. Greg. Paul Pastella. Semi-independent. So, he does DNA crystal stuff.
Beyond Watson and Crick,
I came here for the semiconductor industry but then wanted to visit everyone here. It's a mixed bag of left over other talks, so bare with me. I was talking to semiconductor industry about things towards integrated DNA self-assembly, artificial DNA self-assembly with nanoelectronics using this idea of programming DNA self-assembling. I come loosely from this field of DNA nanotechnology. I just want to give a real brief background of the elements of that field and a little bit about what I am interested in. I also want to entertain any discussion about anything here. Let's just have an open questions and discussion, and at the end, more towards biology and biochemistry type applications that you might see in the talk. Today might actually help single molecule biophysics. I might be able to hand wave about that to stimulate discussions.
There are is programming versus Programming. To (p)rogram means merely to "specify", this is what a chemist means by programming binding interactions. To a computer scientist, Programming means embedding a real computer algorithm, to embed something that has arbitrary computation in a molecular situation. Programs from molecules. Specify what the molecules do. There might be no computation in a formal sense. The reason I am bringing this up is that lately people in DNA computation have re-branded themselves as molecular programmers. Their related friends in syunthetic biology. Synthetic biologists nicely let us hang out with them. This is just what the effort is. We're trying to take molecules, principles from biology, and turn them towards our own purposes that might not be biological purposes. Synthetic biologists start with cells. Molecular programmers start with molecules. Some of them are (p)rogrammers, others are (P)rogrammers. The far most proponent of that has been Erik Winfree. Ron Weiss has interests in big piece stuff. Tom Knight is a big P kind of person. Drew to a certain extent. A lot of people are little (p) type. That has to do with philosophy, not any sort of value.
The field of DNA nanotech has three strains. The little (p) of programming structure. Ned Seeman started the field roughly thirty years ago. Over the course of thirty years, they used DNA as lego, and build any structure we might want. They laid out the structural motifs, making junctions, the things that were thermodynamically and kinetically stable. Ned had origins in crystallography, he tried and succeeded to make rationally designed 3D crystals for DNA. One is this programming functions in Dynamics. Bernie Yurke and Andrew Turberfield. Non-equilibrium DNA structures. DNA hybridization is not two-state. We can make little DNA tweezers, little DNA walkers, none of these dynamic things are as good as your protein motors. We're making them more autonomous. In 10, 20 years, we might catch up with the structural guys and have the dynamics and functions we expect. You're familiar with other functions of nucleic acids that are not just structural. This is very active.
Finally, Len Adleman started experimental DNA computation in 1994. Putting actual computation in molecules. I won't go into this today. The idea here is to put computation in molecules, but not to compete with electronic computers. Thermodynamics, kinetic error rates, cost, scaling would not allow us to compete with electronic computers. These can control other molecular systems. They can build shapes, structures. A little computation to determine what structures. Or computation to figure out GRNs, to turn genes on and off. Do a set of computation on biochemical inputs. Does this cell have a particular disease state? These are small and compact. A few dozen gates. Never the less, they are computations. If you put a little computation into molecules there may be many many things you can do. There are strains of structure, function and computation. This is the kind of place the field is going. We want to recapitulate what the living systems are doing. Along the way, we'll use the things we come up with, and put them into applications.
What was the original impact of Len? I view them as two different intellectual efforts. For construction tasks, Erik visualized this thing as not computation, but.. this is an aperiodic DNA crystal. These are little DNA monomers, grey and white bodies are 12 nanometers by 4 nanometers, that has the beginning of this fractal pattern. It's binary addition modulo two. It was about figuring out a complex pattern. The tie between computation and pattern formation was immediate, not just "explicitly doing computation". The argument is if Len had never existed, I was doing DNA computing before Len, if Len hadn't ... Len's implementation, which is being pursued in eastern Europe, the reason I bring it up, is that we don't have lots of things. What leads you to believe, "we're going to program with a big P, Len's big P isn't going to go anywhere, if we bring on these fast time cells, what you're learning now will help later?". One answer to that question is that independent of what we're learning now, I can give a whole talk on it, if you pick this one subfield, this has all the richness of computer programs, all the Kalgoromov complexity rolled into it, in non-trivial ways. It's the only implementation of computation in DNA, as far as I'm concerned, captures all of the interesting stuff in computer science, in a way that makes sense, all plays out in whether or not you can build something. If you just carve off one thing, whether or not it's practical is a different issue. There are particular subsets of this particular domain, that could capture all of the interesting things in this domain.
But, making structures towards things we might do with nanoelectronics. The first real successes with DNA nanostuff was periodic structures. Cubes. And tetrahedrons. This is a DNA tube. It's made of similar DNA motifs. They are made by 4 by 12 nm. They make this crystal. They have a curvature when they bind, so they make a tube, it's not a model for a microtubule. It's a crystalline tube. It's not too interesting of a periodic pattern. Purdue made these where they have these little 5-connected objects. 17 plane symmetry. Periodic pattern when symmetry breaks. We can make any periodic pattern you want. We can make any periodic pattern in the plane with any symmetry. That's great if you want to make a regular structure like computer memory, but what if you have the greatest amount of arbitrary and unique things in a finite structure. Arbitrary color pixels and make whatever pattern. A grad student posed this question to me, so I came up with DNA origami. This is a process where we take a long strand of DNA, 7k nt single stranded DNA, take a very long ssDNA and fold it into a desired shape or pattern. The basic way this is done is via DNA staples. In the real implementation they are 20 to 30 nt. Each staple conceptually has a left half, and that left half binds that long black scaffold strand. The right half doesn't have a continuous complementary. The right and left sides are complementary in other areas. You have to design this. It folds this into some shape or pattern. You have to respect the geometry of DNA. The idea is that these colored staple strands fold this long strand into this path where it makes a rectangle. The black hash marks make the hydrogen bonds. The blue ones bind here, 8 bases here, here by 16, these things are more complicated than what I described before, each one binds the scaffold together in different places. It's shocking and somewhat confusing that this works at all. This was a video by Shawn Douglas from William Shih's group at Harvard. We can't make decent movies of stuff that happens with molecules. It gives you some feeling of what must happen for these things to come together. These staples are binding together, falling off, being invaded, etc. etc. The molecuiles are wigling. You can heat it up to almost boiling. A double helical character emerges. The whole thing is composed of things composed of double helices. This is not a simulation. There are no kinetics. This is a movie. Nothing is moving. This was made in a time-reversed way.
Have you found issues with knot formation? That's been most rigorously explored by Shih because things in 2D are easier than things in 3D. You have these double helical axises held together by these crosslinks, where a staple strand winds around one hnelices and then reverses directions. One of these crosslinks enforces the other crosslink. It's kind of flike in homologous recombination, but not quite. There are no unbaired bases. These are just drawn to show you. There is a gap between helices, but not between helixes jump.
In each of these linkers, you have to worry about if a sequence in the linker chain is an exact match in the seven thousand .. by design, each staple has an exact match, it has domains that match exactly. Many things are wrong. The long strand itself can have self-complementary to itself. The short strands can have complementary to each other, complementarity to places they are not supposed to bind. In DNA nanotech, we typically use computers to design these sequences. We design the hell out of these sequences to avoid these problems. Here I used a natural scaffold from the M13 virus. You're stuck with the sequence of the virus. You're effectively stuck with the sequence of the staples because they are somewhat induced by the virus. You have no choice. I just close my eyes and get lucky. There are reasons you get lucky in this setting. In DNA self-assembly, Burnie Yurkie on non-equilibrium DNA stuff, exact matches have a mechanism- biochemists knew this for years- mechanisms by which they can displace incorrect matches. If they mismatched by a length of 10 here, if there's a guy who matches by 16, the other one can bind by four on the end, and through a random rock can get there. The reason why this doesn't get trapped in a billion different wrong ways, well, DNA hybridization is not two-state. The 10-mer binding domain, if you had to wait off, you don't have to wait centuries to drop off, well, there's this milliseconds mechanism before step. Without this mechanism, this would be dead probably. It's not just a difference in equilibrium constant. It has to be able to "walk off" bad things that occur. If you have something that matches inappropriately, and nobody will compete him off, you will have some garbage stuck there. Now since you can synthesize that, is it worthwhile to design a perfect scaffold for folding origamis? There have been no pathologies so far. Shih and others are going up to 50 kb. The address space gets more cramped. It might get worthwhile to design very carefully. The answer is that I don't know when it becomes necessary. Did you ever actually write some program that would ensure some tearing between the stitching in the molecules? I didn't have choices about the sequences. The virus induces those sequences. I had to close my eyes, disregard my best practices, and do it on a small scale. But I did do the following: I checked this thing for self-binding, it has a biologically important 20 nt hairpin. That hairpin is really thermodynamically stable, so I only folded the molecule, except that really important molecular hairpin. That's the pin that hangs off in everything. Shih will fold through that section sometimes. Oour ability to test if something is screwed up is not that great. Depending on what strands are bridging that hairpin, and whether or not they are overlapping enough, that's where you figure it out. Secondary structure can screw you up. Ned wrote algorithms very early. Absolutely. Oversequences of this length, you can totally get rid of it.
I made a variety of structures early on. Once I had basic geometry under control, I just wanted to fold in any way I wanted. So this was a scaffold path. Start up, make a turn around, fold over, fold back down, and be back here, and leave the little pin down here and out of the way. I wanted to do this to show that I could make structures with mechanically stable holes. I designed staples to do this. To give you a sense of scale, I had this 100 nm in both directions, within 6 nm in one direction, 3 nm in the other. To compare its size to a red blood cell, there's the smiley face guy. 70 of these can fit across a red blood cell. It's slightly smaller than lithography. The line widths are 5 to 10 times smaller than modern photolithography. Is there some reason why it needs to be continguous? You think it would cause pins. You could leave these structures as gaps, and the DNA helices stack across gaps. There are staples across the gaps. In this design, there are DNA strands that go across this gap and suture things with stability. It has problems with stability elsewhere, but not there. Can you make a multi-layered structure if you wanted to fold it? William Shih has been doing that.
Atomic force microscopy is where you have a micromachined sharp tip, you measure the deflection with a laser beam bounced off of it and build up an image one scanline at a time. You're not taking a picture, you're physically dragging a tip, or tapping a tip, interacting strongly with a sample. This is how we're going to visualize this structure. You design one, you send an email to IDT, you get 250 strands back, you do it in a magnesium buffer so that they form, then you're stuck with a hundred billions of these things in 100 micro liters, you pour them over a surface. The DNAs are negatively charged. The surface is negatively charged. You have magnesium in the buffer that serves to stick the DNA to the surface. They stick strongly enough so that you can take the AFM image and see a bunch of DNA origami like this. You can see that among other things, they aggregate. Some number of them are broken, some of them are fragmented. In this particular prep, 70% of them had the recognizable shape. You can zoom in and see the fine structure. here's the fine structure. Here's the fine structure of this thing. They take 4 to 8 minutes to image. Getting a tip that is sufficiently sharp. In the past, you could sit there, it could take two or three days, throwing away needles, changing the needle on your record player until it sounds good. If you make a structure much more rigid, the apparent yield rises. These triangles, which have, these are the schematics, they have DNA helices in all directions. It doesn't have just a few helices.. it has greater than 95% yield. The yield of these, the gross morphology of each staple, the gross morphology down to the 5 or 10 nm length scale, is dictated merely by the purity of the scaffold. The better the purity of the scaffold, when you break the structures, it's not hydrolyzed, you get the yielded structures. But it's not staple-dependent? You can pick multiple sets of staples to make the same structure. So there are some rectangles with no apparent yield.. it's not measured in a fine way. We don't try to optimize the yield. That 90% robustness is from different set of staples. It's shockingly robust. The fine detail works out.
A year after I did this, I got this picture in an email, Lulu Qian. Lulu is now a postdoc at Winfree's lab. What a job application. So she sends me this thing in an email. I was amazed that she did this on her own. She just wrote her own code to do it, screw this guy, she designed this really great fold. She has taiwan sitting right there. She has taiwan on the world's shortest leash. So this goes to show that this can be reconpitulated by reading supplemental materials. It's not rocket science. Layered structures. Shih has got 3D structures of several times. Nature, 2009. Douglas et al. He has used one where he uses DNA helices on a honeycomb lattice. On like a rectangular grid. I would want to do a rectangular grid if I went to 3D. You can do hexagonal grid. These are negative stained TM images. He has girders, interpenetrating structures, all kind of stuff. These are three origami at once. Chains of origami. All of these things have reasons. Everyone in the lab has different reasons for these structures. The whole idea behind this stuff for biology is to make custom instruments for biology. William Shih has a background in motor proteins. You can ask hypothesis like, they take these steps, they walk along a protein filament, you can ask, how does it know to release the backleg? The hypothesis might be that the back leg is released when under tension. I want to test that by varying the spaces of my track, maybe if I increase the step length, maybe it walks faster. The probability of a walk might be less with shorter steps. You can't engineer protein filaments. You can't change the step size on a microtubule. This will be a substrate for motor proteins to work along. You can do the DNA part and make it great, but the whole is meant to accept an active tetramer. Getting that right will be the hardest part. At least we can think about building custom instruments for biology. For little expense, you can design a custom instrument. The interpenetrating barbel. Imagine machining that thing. Andy's question about folding difficulties. To get these things to fold. The 2D structures that I make, and 3D shells, fold quickly in 2 hours and you're fine. To get these interpentrating bars, you have to let them fold for a week. If you are clever you might do long staples on the inside, so they fold first at higher temperatures, but nobody has been that detailed yet. Annealing schedule. Staple length. You can get these things to fold. Several groups are trying to get dense 3D structures. They made this structure or something. They tried to make this structure, but they didn't let it fold in a week. The yield of the interpenetrating bars after a week is 50%, for what they consider to be "well formed". What's the definition? Well, it's at least better than 1% yield. This is the point that Ned harps on, are these molecular and really all the same? So that's fine. So you can assemble multiple pieces. They have little extra DNA strands. They purify, they mix them together. The same for the polymers on the top? There are many ways to string these together. They are exploring that, a different one, if I have time for that.
Hendrick Dietz, William Shih, 2009.
Are they flat? I don't know. If you make a 3D structure or a DNA tube, with a total curvature, the mica might catalyze them open. I don't know if the smiley faces are flat. I designed them with symmetries that would tend to take out, would take out bowing in x or y, so according to the symmetries they might be taken out.. so that same structure is completely compatible with trust. Shih said that my structures are twisted. I went back two days later and realized I screwed up. You have to nail the inner spacings between the cross-overs. It's 10 bp/turn. If you don't compensate for it, your twist will have this super twist, not gentle at all. If you make a rectangle, it might be twisted from 90 to 180 degrees. They are twisted like pretzels in solutions. Everything I made was really twisted. He actually, this just came out in nature, he had these rectangles either left-hand twisted, right-hand twisted, or straight. We reconpitulated this with our stuff, it really helps. Things were twisted, if you know what to look for in the AFM. The long chains can also fragment. Hendrick Dietz was the one who thought about this. Bending and twisting based on this can lead to curved structures. Boomarang structures. They figure out the model, they can nail it. Gears, they can make curved 3D stuff. This was made before anyone was making this. This thing, if you look at the reconstruction, it suffers from the same twist and deconstruction (the boxes). Ebbe Andersen 3D box from 2D surfaces, Nature 2009. Rectilinear style. You can make everything.
caDNano open source software
For the purposes of patterning, if you want to arrange stuff, every staple goes to a unique position. The orange strand lands exactly there. In principle, you have to put chemical functionalizations. For the paper on this, I didn't put interesting functionalizations, just hairpins and bumps, which has been done in DNA forever. With 6 nm resolution you can do North America, snowflakes, two words of DNA with a continuous helix from one flake to another. There's a piece missing from this D. It's the case that you're trying to figure out how many functional groups you have. You can only visualize 94% of them. The AFM actually knocks them off, they disappear while you're sitting there. If you're trying to functionalize this thing. You need other techniques to try to nail this to yield the functionalization of this technique. But if it all works out, you want to put some active elements on some staples, for example, make a circuit, and once it formed, you wash the thing away, and you have some functioning nano circuit.
The design of a biochip of a self-assembling molecular scale memory device. Robinson & Seeman. Protein Engineering 1 (4) 295-300. 1987. Nealey Lab, Ruiz, have been using copolymers for 2D lithography and data storage. We get a little amount of money from the semiconductor industry through the semiconductor consortium. It has focused my thinking about what are the actual challenges to nanoelectronics integration and how to surpass them. This chart was made by people in this organiszation. Benchamrking Table for Nanopatterning Technologies. Here's DNA nanotechnology. The things that are in this column that might be a problem in yellow, and really definitely a problem in red. What kind of devices and how to make them compatible with DNA? Higher order structures? And foremost, registering them with underlying features on a surface. Here's this problem. Here's this problem of DNA origami on solution, and they go everywhere. We have FENA sponsored projects to address this problem. Yield, defect rates, scanning, compatibility. CNT cross junctions. And, in those projects, maybe some of these things, maybe the registration is more yellow than red, and anyway. So.
We have a collaboration at caltech, and everyone has graduated. This involved Siping Han, Robert Barish, Hareem Tariq, Marc Bockroth, Bill Goddard, Erik Winfree. These students did everything. There's a history of DNA plus CNTs to make devices. But there hasn't been any use of DNA self-assembly to assemble anything other than a one-dimensional device. You can combine DNA with CNTs, there's a nice paper by Keren et al. 2003. You might have a single CNT along its length that might be functionalized different, like for attaching electrodes, and the entire thing is gated by this carbon backing. It has a symmetry breaking in some sort of mild way. You don't have anything that approximates a two dimensional device. You have this same substrate-backing. It's this whole giant device. We wanted to organize this thing in 2D, and have 2D devices. There are many ways to functionalize DNA to CNTs, you can sonicate CNTs to DNA. Otherwise the CNTs won't go into aqueous solutions. Addressing the compatibility of CNTs and DNA origami. You can combine some components to the CNTs. As far as yield is concerned, the DNA is usually binding things later, but those things can also bind to the CNTs. So, they will bind and be partially accessible. So they stick DNA in hydrocellulose. Some stuff is bound, some stuff is not. What the students did is took a note from Bernie and put protecting groups on the strands. These little DNA sequences that are exposed are very short. We haven't measured this rigorously. The protecting groups have more available. This kind of tricks appears to have really helped us in organizing these things. We took two different sequences orthogonally to each other, made two kinds of CNTs with protecting groups on them. These protecting groups only release if and only if these things start to bind with this toehold with the appropriate DNA sequence. They wanted to make cross bars. If we used random mixtures with metallilc and one semiconducting one you in principle get transistors. One semiconducting, one metallic, they relied on random chance. Maybe they don't make.. they got crosses. You can deposit them on silicon, wire them up with electrodes. We got a few FETs, one of them was sort of stable.
Two dimensional organization of CNTs on DNA origami.
What drives the interactions of DNA and CNTs? It's not completely characterized. The best way to separate CNTs. Metallic CNTs are semiconducting. They have different chirality. You can close them and get chickenwire with different chirality. You can exquisitely separate the CNTs with different populations. It's fricking amazing. They had a paper in Nature in the last couple of months, where they can do this massive screen over short DNA sequences and selectively find 12 different types of CNTs. They proposed it to be a DNA "beta sheet". They see the pyrimidines specifically binding between. Once the tubes are put together, once you've directed the crossover, are there tubes to each other, will the tubes remain stuck? There is this origami in between them. They have no directions between them. This guy binds to this guy. The DNA origami works. You could imagine aligning things on the CNTs such that you have your toeholds and protecting groups. Breaking symmetry on CNTs is a bitch. You can functionalize the ends more than the internals. You broke symmetries, when you broke the CNTs, the whole tube is the whole thing. We're not talking about bands. The whole tube has a particular chicken wire structure. There are some CNTs that have heterogenerity between two structures. Nobody knows how to grow CNTs to grow one chirality and switch. What I was getting at was that, it's great to think towards this nanoelectronics era driven by DNA, and then thinking that this is the world's most expensive tech for this scale. But what if it was catalytic? As far as cost is concerned, if we could organize things, we could paint a Boeing for pennies. In the grand scheme in things, I haven't figured it out very precisely, you can paint large areas with this stuff relatively cheaply. I think I am much more worried about working with any yield whatsoever. That will stop it much longer than the cost. The yields that we want are appropriate. They organized these crosses. What's a metric for how well they actually did? We had CNTs that can bind either the horizontal or vertical stripe. How many bind that and how well? 50% of the tubes bind within 15 degrees. We didn't give them that long of a run to bind. If we give them a longer binding stripe, this is an objective measure of how well we actually did. There are some that are 90 degrees off from the desired orientation. They might be binding only by one DNA, they are still floppy, well, these are the distributions we get. Let's say we really make great crosses. Even if you make great crosses, you have to find them. You have to use ultramicroscopy to find them. You have deposition and aggregation. You try increasing the yield, say you have not many crosses and not many DNA molecules are binding to tubes, well, these CNTs are long. You get more aggregation, but at the end of the day after you find them, you have to use lithography to go down to these guys. A grad student is going blind trying to hook these things up but it's not fun. What can you do about this? Under the FENA research. Grant is gone? Looks like he ran off. Anyay, so this is a collaboration I have with IBM Research Almaden. Ryan Kershner, Luisa Bozano, CHristine Micheel, Albert Hung, Anne Fornoff, Charles Rettner, Marco Bersani, Jennifer Cha, Jane Frommer, Greg Wallraff.
The largest length scale is the smallest length scale we can do with electron beam lithography. The top-down length scale matches the bottom-up length scale. So we're going to make locks that are sticky-patches for shape-patched origami. Here's the idea. Ryan Kershner spearheaded one chemistry. Another two did another chemistry. Everyone was really important. Someone did all the lithography. The solution with the triangles, which are nice and rigid. Silicon dioxide or a silicon substrate. So pattern a template layer over it. Pattern a photoresist, use e-beam, transfer to the template layer using an oxygen plasma. This is a dry etch, you impinge it on the template layer, you oxidize it and change its chemical layer, you strip off the .. and in principle you get DNA binding sites. Magnesium is what binds these things to a MICA surface. The degree to which Mg will regulate this will depend on a few different factors. For the conditions of the materials we had, we had 10 nM Mg to make it, but you bind it in 100 nN Mg. If they haven't discoverred this, the whole thing wouldn't have happened. We have a couple differentchemistries. Syline on a straight dioxide surface. TMS/SiO2. We can do it on a diamond-like carbon surface (DLC/DLC on Si). We've done this with two systems. This is DNA triangles on diamond carbon. These have columns of up and down triangles. You can score this very tediously and show that the diamond like carbon, you have 2 populations, and guys are usually 60% oriented in the correct direction. They are oriented plus or minus 10 degrees. This is under buffer. Here, origami triangles are being ripped off by AFM tip. If you do this under the buffer, then you will necessarily rip them. If you dewett and get rid of the water, you have to dry in a way that doesn't get them off, and some people are doing that. This is the best data we could get. This data could be cleaned up. These are 20% off or something horrible. This is the random distribution, it would be + or - 30 degrees. We tried to bind them into shape features. This is not the route to getting th e organization of origami in features. Thsis is a movie that shows the size dependence of the binding. This is all under AFM. These things are binding and unbinding. For the small things, they have the highest residence time. They are about 127 nm. The best alignment and binding in the size-matched features. The important thing to know is that, if this binding were not reversible, we would not get the results we get. One of the good things about the noncovalent binding, the streptavidin would be stripped too fast.. The triangles have a chance to settle down. So that's a key feature if you're going to try this binding and alignment. So this is just some frames from that movie. This shows the salt dependence. In too high binding, the DNA itself actually overcharges, so it now, the DNA itself is nominally positively charged and won't bind to the surface. If you're a biochemist and think of DNA condensation, well, that's what this is, with a polycovalent cation. It doesn't mediate the binding. If it does overcharge, the DNA falls apart. Leave one strand on a triangle to be able to hybridize to another triangle. That's a different project. There's a whole, so here we want individual triangles that are not spaced, but then what if we have the triangles talk with substrands, let them have molecular precision with their interactions. We want asymmetric shapes. So that the origami goes down on the surface, all points north, a triangle can go down with 6 different orientations. Place the shapes then add the components. The components wouldn't cause the origami to aggregate in solution. The student wouldn't have to do this crazy ultramicroscopy to figure out where they are. Aggregation and finding the structures in one shot, using this registration technology. In the next couple of years, high orientation of the shapes, placing distinct shapes. I am seeking postdocs for this. IBM Albert Hung, Jennifer Cha, now at UCSD. Gold triplets of nanoparticles on these surfaces using DNA origami. They did however, they didn't want to do it on a surface, the yield of the coupling was too low. They got some aggregation, but they lucked out a lot, the triangle has this nice feature such that they dimerize or trimerize, they bind and are oriented, there's a monomer, they got lucky, but it's great stuff anyway. So anyway. There's lots of stuff to talk about, so I'll end there.
Before the origami, did everyone build a FET out of CNTs? With randomly spraying CNTs on a surface, you can find the crosses. The first one, where you explicitly use molecular self-assembly to do it. Otherwise we don't achieve a better yield than any of these other things. Are the behavior of these FETs, their capacitance is very tiny? We didn't explore how fast they switch. You'd be surprised. The number of devices that you can compare to the device that we can make, is like 2. People abandoned messing around with these FET CNTs. So everyone moved on because it was so hard. You can try to make 60 devices, we got one. But we didn't do any better than people who did it before. We did it in a different method, one in principle gives you 2D organization with self-organization. When we combine with other stuff, do you see ways in getting rid of building, getting some semiconductor straight out of it? Pulling a semiconductor into a sandwich. I am not a big fan of CNTs. A lot of the things we might go back and revisit, but organizing or functionalizing, pulling down semiconductor quantum dots or silver wires, will probably be the next steps. Even if you have 95% yield, how do you identify the things that work? How can you purify the devices for those that work? In each stage where there is less than 99.99% yield, what's the issue at that stage? The one point of intervention is purity of the scaffold. There is a funnier purification in things that are like impartial triangles don't bind to this site, so there's some purification that happens there. At every step of the way you have to figure out what you're going to do to maximize yield. I don't know how to say it. You'll work out some system and say, this is the device, this is the system, blah. When only exploring this, I don't care about yield. The way we approach lithography now is that you can imagine a self-healing device. You can build checksums like "hey look, I'm totally wrong", send a bunch of power, melt me, re-anneal, and try again. Self-healing crystals that have those kinds of properties, where the defects fall out and the thing can re-grow. The challenge is figuring out an end-to-end process that works with the devices. A lot of the concepts in this thing, when we actualyl build things, you're dry, the DNA is dead, imagining a way where the self-assembly with later stuff, it could be done, it's done in our bodies, but how you put that all together requires 50 years of cleverness. How do we adapt DNA nanotech to the wants of the electronic industry? Well. Maybe. That's the thing. Is there a fundamental incompatibility? We have error rates that are still too high. We have lots of problems that we have to optimize and debug. I don't think so, if you look, there are steps that are in the making of certain devices. There's this outfit called Alien Technologies and they make RFID chips and whatever and what they have are certain processes in the semicodncutor industry, RF subsets versus whatever, are just incompatible. They have this big machine that makes this surface, they make offline in another process, these little circuits. They pour over a billion of them, they slot down in their holes, they have a defect rate is low enough. It's a fluidic self-assembly process. That kind of existence proof shows me that we could get there. I don't feel like it's fundamentally incompatible.
Tethering components to study molecular interactions. Shih is going in that direction. The goal is 5 years from now to have 5 biophysical lab, and 5 physical sciences labs, to do some experiment they couldn't otherwise do without this tech. Single molecule guys now, take some protein, or some protein DNA complex, spray it, have a fluorphore or two, there is low signal, it looks like noise, you get these blips. If you look at those on a grid, you could forget everything except 1 or 2 microns on center. Orientation, position. Right now, if you want to put a single molecule on a spot, you could spot it. You get a Poisson number of molecules in each hole. That's great, you can assay him. The bioscience guys have a big array, a polymerase in each hole. They can do sequencing really fast. What if you want to study proteins in a group? By the time you take Poisson statistics to the fourth pattern, the number of guys with that, is very small. With DNA origami, you have 99.9% with those four guys, and 95% of the holes filled. Is anyone trying to do next gen sequencing. Tethering single molecules on sheets?