Define the goals of this project:
- Formalize the SiC layer initiation and lattice extension.
- Create a method to compile matter into a sequence of tips.
- As a result, I can quantify the slowness from tip randomization.
End goal is to exactly know the building speed of SiC vs. amorphous carbon.
Silicon feedstocks are difficult to prepare. Metal hydrides and hydrosilanes are known to react with oxygen-containing compounds. On first glance, one would assume that any oxygen-containing compound reacts with silicon. To be more specific, the reaction of concern is ROH/RNH2 + RSiH3.
The first reagent is ROH or RNH2. Hydrosilanes are most stable under weakly acidic conditions. A strong base, like hydroxide, forms a pentacoordinated transition state with RSiH3. This intermediate decomposes into a compound where a Si-H bond has been replaced with Si-O. Amines are inherently basic, even trialkylamines. Therefore, one should avoid linkers containing nitrogen.
The second reagent is RSiH3. Pure hydrosilicon compounds, such as silane (R=H) and disilane (R=Si), are notoriously reactive. They ignite on contact with air. Methylsilane (R=C) is more stable. The stability of monoalkylsilanes slightly improves with chain length. In the feedstocks relevant to SiC methanosynthesis, R is the germanium of a germanium-substituted adamantane cage. The compound's reactivity likely falls between methylsilane and disilane.
It is not definitively known whether C3Ge-SiH3 with OH linkers can be synthesized. It seems plausible, given an aprotic solvent and no contamination from ions (e.g. hydroxides). It also seems plausible in purified tripods, where the solvent is other tripod molecules.
A more pressing concern is the addition of SiH3 to a bridgehead radical during mechanosynthesis. Si3C, C3Si, and C3Ge all have roughly the same binding energy (2.96–3.10 eV). This is not an issue when forming the very first cage of a SiC layer. All carbons are sidewall carbons (CH2). The binding energy of HSi2C is 3.51 eV. The stronger binding energy is likely due to less steric hindrance. There are no neighboring SiC unit cells; one of the C-Si bonds is replaced with a C-H.
There are four candidate solutions for the issue with binding energy:
- Si3Si feedstock holder (2.43 eV). This cannot be prepared as a tripod. The MinToolset paper makes liberal use of GeH3 bonded to a Ge(111) surface. However, we need the feedstock in a form with a high aspect ratio. One could plausibly create silicon shards or unpassivated Si nanoparticles, then flow Si2H6 gas over at a cryogenic temperature.
- N(CH2)3Sn or N(NMe)3Sn feedstock holder (2.34 eV).
- Organotin compounds are highly toxic. Synthesis of stannatrane compounds will require multiple steps where a chemist works with the compound, cleans glassware of it, and could be exposed to its vapors.
- Organotins cannot coexist with an alcohol linker. One could imagine a gold surface with Sn tripods linked through thiol linkers. Reactions on this surface involve SiH3 feedstocks, with small pulling forces. A separate Si surface holds the other tripods, for carbon feedstocks that involve larger forces. Switching between these surfaces will introduce delay into build sequences, slowing down atom placement. However, such tripod segregation methods may already be necessary to counteract tip randomization.
- While the C3Ge-SiH3 group has been synthesized in the literature, C3Sn-SiH3 has not. Preparation of the feedstock will likely involve STM lithography. C3Sn-H tripods are first deposited onto a surface. An STM selectively (if in a mixture with other species) depassivates the tin tripods. The process employs localized heating from inelastically scattered electrons, which is proven to bake off the H from Si3Si-H. Next, Si2H6 gas flows over the surface at a cryogenic temperature. Although the gas cracks when flowed over unpassivated Si, there is not confirmation that it cracks over isolated Sn bridgehead radicals.
- NS3Ge feedstock holder (2.55 eV). When I designed the first ever SiC lattice extension primitives in summer 2024, this feedstock was chosen. It was an attractive region of the energetic landscape. However, compounds with the S-Si or S-Ge bond are relatively unstable. Such a tripod likely cannot be synthesized.
- Create a SiH: diradical nearby. Hope that it inserts into the C-H bond of the Si3C-H moiety. This is probably the first thing people will attempt in experiment. It can be done with only C3Ge-SiH3 feedstocks. However, there are multiple pathologies. A hydrogen might migrate from the C bonded to the Si diradical, forming a pi bond and SiH2. Or nothing happens; it fails to insert into the C-H bond. I did not study the pathologies in detail last summer.
There is uncertainty over which solution (if any) works. If none of them works, it might be impossible to build any crystalline unit cell with 3DOF nanopositioners. One would have to resort to multiple tooltips resting on unrealistically sharp SPM probes.
The last concern is the SiH2· feedstock. It must be formed through mechanosynthesis on the SiH3 feedstock, and with extreme positional control. If the hydrogen abstraction tool is even one angstrom off course, the carbon radical will bond to the silicon in the feedstock.
It is also a great practical concern, as it involves formation of a temporarily conventional-mode tooltip on the SiC workpiece. The tooltip (HAbst-H) is then ripped off of a bridgehead C3Si through the pulling force of adamantane(C)-CH2·. Injecting the feedstock preparation steps into a build sequence adds significant complexity in a practical experimental setting. However, I see no reasons why it should fail in theory.
The uncertain solution to Si3C binding energy is informative for compiler design. The code should accept an arbitrary set of feedstocks, with their energies annotated, as an input argument. The compiler itself is agnostic to the exact choice of feedstock. The compiler will not experience software rot, when future experimental evidence reveals a different tripod was necessary.
The fourth solution to the energetic problem has another implication. Not only is the feedstock uncertain, but the exact sequence of reactions could vary. The compiler should generate a different sequence of reactions, based on the chosen solution. If every option can be compiled, this maximizes the chance that the compiler remains relevant after experimental confirmation.
Note that these measures do not guarantee the compiler's immunity to software rot. If none of them work, or a new solution outside the known set ends up working, the compiler will be useless as-is. At best, it is reference code that inspires a freshly written compiler, with the new reactions. At worst, it has no relevance, as the software stack required to operate the SPM has drastically more details.
The objective of this project is not to create something with practical use towards experimental realization of SiC mechanosynthesis. Nor is it to prove the technical feasibility of crystalline molecular machines. Rather, it is a fun project, giving the author something to enjoy. It will be years until SiC is confirmed in experiment, and the author doesn't want to wait that long to have fun. The other objective is to show superior technical prowess in computation and theory, compared to other direct-path competitors (CBN Nano Technologies and Machine Phase Systems).
Finally (not to forget), the compiled build sequences will aid in estimating the slowdown from tip randomization. And extrapolating near-term experimental results with amorphous C, to far-term results with silicon carbide.
During compilation, atoms should be represented through their spatial positions and atomic numbers. The bonding topology may or may not be required as an input. The compiler should automatically detect regions of crystal surface or bulk, then connect them together with a path chosen algorithmically.
Another constraint on the compiler, is to only compile one layer of unit cells in the Z direction. Assume that the build plate has the same cross-section as the layer being built. The compiler still has immense relevance. One could call it successively on the different layers of a 3D structure. In fact, the simplest 3D compiler probably decomposes a product into smaller 2D planes along the (111) direction.
If rigorous simulations reveal complete mechanosynthetic routes to extension of SiC(100) or other surfaces, one would probably design a fresh compiler for that specific surface. Its architecture may benefit greatly from decomposition into 2D planes, and restricting the current layer to remain inside the last layer's cross-section.
The very edges of a 2D cross-section may pose a challenge. Instead of a bulk C3Si-H bridgehead site, one is dealing with a HC2Si-H site. After depassivating that to HC2Si·, the remaining hydrogen is free to flip orientation. Even if the temperature is low enough, quantum tunneling may induce that flip.
If the remaining hydrogen does flip to upward, it can now pollute a CH2· moiety. The end product of the hydrogen stealing is R-CH3 + Si: diradical. The diradical state of the silicon group makes it unusually energetically favorable.
The best solution seems to be depositing carbon in CH3 form. I vaguely remember testing this route in simulation, and having it work correctly. It is difficult because the carbon cannot form an expanded octet. However, it could just be an SN2 reaction with an energetic barrier.
From here on out, it is legal to use CH3 to deposit carbons onto a C3Si radical. This choice makes the process of methylation simpler (compared to CH2· + HDon). The estimate is less conservative; biased toward overestimating the ease of executing a reaction. However, there are much bigger elephants in the room regarding whether SiC can be built at all. This concession is more reasonable than assuming it is possible to silylate a Si3C moiety.
There is another complication about forming SiC unit cells besides the very first cell of a layer. There are two groups relevant to the reaction. First, a CH3 connected to the SiC(111) surface. Second, a SiH3 partially floating in the air. The silicon group connects to what was once the sidewall carbon (CH2) of an adamantane-like cage. The two groups must be made to deterministically bond to each other.
As I only invested substantial effort in checking the "layer initiation" reactions, I did not simulate lattice extension with much rigor. There are two plausible pathways to form the Si-C bond with high success rate (≥99.9% at ≥4 K). The compiler should generate build sequences with both pathways, the specific one chosen on demand.
- Transform the SiH3 into SiH: floating in the air. The first abstraction occurs with HAbst (Ge-CC· or C-CC·). The second must be done more carefully, as the SiH2· has much more propensity to bond to a C radical. It should occur with a weakly bonding tooltip (e.g. C3Si·, C3Sn·) that forms a reversible covalent bond with the silicon. After forming the diradical, depassivate the CH3 to CH2·. It forms a covalent bond with the Si diradical, with almost no chance of stealing its remaining hydrogen.
- Depassivate the CH3 first. This may require an HAbst tool approaching from a non-vertical angle, to prevent unwanted side reactions with the SiH3. In experiment, one would need a novel tripod with R-CC· groups pointing sideways. Furthermore, one would need to search for various randomly oriented species. Find a tool with the R-CC· pointing in the exactly right quadrant of the XY plane. It would help to perform more simulations to resolve this uncertainty regarding steric closeness and distance to the SiH3. Next, depassivate the SiH3 in a concerted mechanism that simultaneously forms the Si-C bond. This way, there is no chance of hydrogen stealing.
An especially troubling combination comes from closing the final SiC cell in a crowded area. As more SiC cells are added to the first cell of the layer, the identity of the CH3 changes. More specifically:
- It starts out as C3Si-CH3.
- Upon adding the second cell, it becomes C3Si-SiCH2. One could rearrange the neighbors of the carbon atom, to find that this is the same HSi2C· with a 3.51 eV binding energy for SiH3.
- Upon adding the third cell, it becomes C3Si-Si2CH. There is only a single hydrogen to abstract. During this abstraction, there is no way to approach from a side opposite to the SiH3.
I have a strong aversion to reactions that form irreversible C-C bonds in 5-membered rings. Imagine a pair of C radicals, or a Si radical and C radical, would form a 4-membered ring upon joining. This reaction is highly unfavorable because of the steric strain from forming the ring. In summer 2024, I permitted reactions with such a radical pair. The same cannot be said for two C radicals formed on nearby bridgehead sites of an adamantane cage.
However, the particular carbon radical in this case has immense mechanical force keeping it still. The steric hindrance from the three neighboring Si atoms is enough to make it a poor acceptor of SiH3. Both carbon radicals involved have this hindrance, because it's the last SiC cell on the interior of a crystal layer. Such a situation may be permissible, but I still want rigorous simulations proving it.
This compiler must use a very conservative subset of all plausible reactions. Under no situation, can it be topologically valid to form a 5-membered ring with a C-C bond. Instead, form crystal unit cells by expanding outward from a seed cell. This approach was used for the "2nd generation tooltip" during summer 2024. The restriction likely limits the number of products, molecular shapes and geometries. It is either this, or you can use multiple tooltips with unrealistically sharp SPM probes. The restriction to 3DOF has difficult implications. If not this restriction, it could plausibly create a different restriction on geometry with similar severity.
Except...this restriction might invalidate the build sequence for the "2nd generation tooltip". I might have to perform a new theoretical analysis with my existing software stack (and not the upgraded renderer supporting Windows).
If you have a physically realistic (meaning can be done in the real world, not just a cheap imitation of Minecraft) reaction sequence, you can build a compiler. It would be "too good to be true" to be able to do this. It unlocks the ability to design molecules with VLSI-like decoupling of functionally separate parts.
The following must be studied in more rigor, whether through literature reviews or molecular simulations.
- Understanding how Si2H6 decomposes on an unpassivated surface. The papers suggest it splits into H + Si2H5 when cracked over a Si(100) dimer at a cryogenic temperature. Does Ge2H6 behave differently when cracked over its respective Ge surfaces? The splitting into GeH3 + GeH3 was what motivated the MinToolset paper. Perhaps, in the presence of a single bridgehead radical, a disilane molecule splits differently than with a Si(100) dimer. Or we need to use silane or elevated temperatures to prepare this feedstock. The last option is cracking disilane into SiH3· radicals, just like H2 is cracked into H· radicals over a metallic surface. These radicals (if possible to create) would likely create an equal population of Sn-H + SiH2: and Sn-SiH3. Since they're flowing over a gold surface, we don't have to worry about them performing homoepitaxy and extending a silicon lattice.
- Need a synthetic pathway to atrane(tin) tripod with thiol linkers. Need proof that both the N(CH2)3Sn and N(NMe)3Sn variants can be synthesized. Even if extremely hazardous, what matters is that it's physically possible. Perhaps the synthetic chemistry lab could be operated by a telepresence robot, with humans going in with HAZMAT suits when it messes up. And a high-field NMR machine is installed right next to the lab room, for the robot to place samples into. Look at how people handled radioactive substances when engineering nuclear energy.
- Need more thorough simulations of the process of depositing carbon in CH3 form. The MinToolset paper utilized CH3 feedstocks by first converting them to CH2·, then ripping them off the Ge(111) surface. Perhaps Nanofactory/CBN did more rigorous simulations of the CH3 pathway, but they've kept the results secret.
- Need an explanation of the pathology of forming irreversible 5-rings when completing certain SiC unit cells.
However, the author has a finite amount of patience. To save time, I'll focus on the two most important issues: Sn tripod synthesis and the 5-ring pathology.
Assume that Sn radicals can be formed through localized heating of inelastically scattered tunneling electrons. Without destroying the tripod in the process. If the hydrogen bakes off due to temperature, perhaps investigate the thermal stability of tributyltin. Tripods on gold surfaces can be deposited through solution techniques (with contamination and clustering issues that we'll ignore for brevity). They don't need to be elevated to their boiling point and deposited through MBE. Thus, I'll spend time investigating a third important issue, the thermal mechanism of Sn-H bond cleavage.