Above: TIMe-Ge molecules on Si(100): 77 K STM scanning. (a) Experimental constant-current filled-states 77 K STM image of four TIMe-Ge molecules on Si(100)-(2×1) (I = 50 pA, V −2.0 V). Molecule deposition was performed onto a room temperature Si(100) substrate. (b) The four symmetry-equivalent on-dimer configurations of TIMe-Ge. An experimental STM tri-lobe is overlaid on the central triply-iodinated proxy. (c,d) Representative 77 K STM images of TIMe-Ge molecules deposited on Si(100) substrates held at <240 K and >400 K, respectively, prior to imaging. The dashed white ellipse in (c) marks a pair of dissociated, surface-bound iodine atoms. A minor low-pass inverse fast Fourier transform filter has been applied to highlight TIMe-Ge positions on the Si lattice. z = 0 for STM heights is referenced to the highest point of the Si dimer-row surface. Figure 2 in the article.
Those keeping track of diamondoid mechanosynthesis research of the mid-to-late 2000's and the very small number of us engaged in this very specific activity may note that the last of these papers was published in 2011 before effectively disappearing completely from academic (publication) activities. Those taking extra steps to keep track of those activities will note that the journal article submissions were replaced by patent applications. Lots of and lots of patent applications.
Saving the history of the 2010's and some of the early 2020's for a far future post, I am very pleased to report that the first academic and experimental foray into the applications of molecular tools for interesting future applications has been posted to arXiv recently (journal article acceptance to hopefully follow) thanks to the efforts of a very large and diverse group of researchers at CBN Nano Technologies (of which I am proud to call myself one).
Here, synthesized and activate-able molecules are deposited on Si100 and present, upon activation, nearly the smallest feedstock one could come up with for the carbon-based mechanosynthesis of anything – CH2.
Figure 1. DFT predictions of chemisorbed TIMe-Ge surface configurations. (a) The Si(100)-(2×1) surface and gas-phase TIMe-Ge (Ge(CH2I)4) molecule with relevant computed dimensions for lattice matching. (b,c) DFT-optimized geometries for the on-dimer and inter-row three-legged surface bindings, respectively. (d) Representative three rotational isomers (rotamers) of the pendent CH2I group. (e-h) Simulated on-dimer STM images for (e) a CI3 variant used as a time-averaged proxy for STM temperatures where the CH2I facilely rotates between rotamers, (f) an intact CH2I in a single “frozen” rotamer configuration, (g) the deiodinated, chemisorbed TIMe-Ge alkyl radical (R3GeCH2●; R = surface-bound CH2), and (h) the deiodomethylated, chemisorbed TIMe-Ge germyl radical (R3Ge●). “AH” refers to the STM “apparent height” of the tallest part of the molecule with respect to the local maxima of the Si dimer surface iso-current contour (analogous to the experimental STM height measurements presented in this work). A legend of atom colors is provided.
I point out as a particular highlight the consideration of design criteria for molecules we and others might want to use in the future for varied applications, which are listed below in a highly less-verbose form than the original draft included:
High molecular symmetry (makes many considerations easier)
Covalent bond formation (these are not weakly-bonding fragments)
“Loose legs, rigid body” (for surface sampling)
Lattice-matching (Legs "… should be at least long enough to reach from his body to the ground." – A. Lincoln)
Confidence in surface-bound orientations (nice to know they're there when you look)
An accessible, stable radical on the surface-bound molecule (the "business-end")
This work begins to bridge the gap between Molecular Toolsets (as proposed by Rob and Ralph) and demonstrations of molecules that contain functional groups that can be used for similar applications. As such, this work is partly "on the path" and, in light of all that was learned to get this specific work to the point of publication, partly identifying where one should start walking in the first place.
Still thinking small…
Authors: Taleana Huff, Brandon Blue, Terry McCallum, Mathieu Morin, Damian G. Allis, Rafik Addou, Jeremy Barton, Adam Bottomley, Doreen Cheng, Nina M. Ćulum, Michael Drew, Tyler Enright, Alan T.K. Godfrey, Ryan Groome, Aru J. Hill, Alex Inayeh, Matthew R. Kennedy, Robert J. Kirby, Mykhaylo Krykunov, Sam Lilak, Hadiya Ma, Cameron J. Mackie, Oliver MacLean, Jonathan Myall, Ryan Plumadore, Adam Powell, Henry Rodriguez, Luis Sandoval, Marc Savoie, Benjamin Scheffel, Marco Taucer, Denis A.B. Therien, Dušan Vobornik
Abstract: Scanning probe microscopy (SPM) investigations of on-surface chemistry on passivated silicon have only shown in-plane chemical reactions, and studies on bare silicon are limited in facilitating additional reactions post-molecular-attachment. Here, we enable subsequent reactions on Si(100) through selectively adsorbing 3D, silicon-specific "molecular tools". Following an activation step, the molecules present an out-of-plane radical that can function both to donate or accept molecular fragments, thereby enabling applications across multiple scales, e.g., macroscale customizable silicon-carbon coatings or nanoscale tip-mediated mechanosynthesis. Creation of many such molecular tools is enabled by broad molecular design criteria that facilitate reproducibility, surface specificity, and experimental verifiability. These criteria are demonstrated using a model molecular tool tetrakis(iodomethyl)germane (Ge(CH2I)4; TIMe-Ge), with experimental validation by SPM and X-ray photoelectron spectroscopy (XPS), and theoretical support by density functional theory (DFT) investigations. With this framework, a broad and diverse range of new molecular engineering capabilities are enabled on silicon.
Above: Multiple views of the proposed mFET (7694 atoms), with colors at left used to distinguish all components in the design. The total volume of the mFET is 46.52 nm3 based on van der Waals radii.
Ralph Merkle, Robert A. Freitas Jr., and I had worked on a design for a molecular field effect transistor a few years ago, then all sorts of distractions got in the way of its completion (IYKYK). A recent return to it led to quick turnarounds of several significant iterations and improvements and, now, recent addition to the Reports section of the Institute for Molecular Manufacturing (IMM) website (www.imm.org) Like all such reports, it's free and open to anyone interested – I direct you to www.imm.org/Reports/rep056.pdf for the latest version.
Above: Give it a spin (and zoom) courtesy of 3DMol.
Below, I'm including some additional notes on the overall process for anyone wanting to take a stab at redesigns, considering similar design efforts, or just have questions about aspects of the analysis. There is a lot someone could do in any future optimization and assessment process with readily available theoretical tools. This first pass is just to, as such designs have always been for, get the conversation started.
DFTB
And I can't stress this enough. I'm now approaching 25 years of "thinking small" about nanotechnology (N), molecular nanotechnology (MNT), molecular manufacturing (MM), advanced molecular manufacturing (AMM), and now atomically-precise manufacturing (APM). Don't look at me, I was just doing theory.
One of the great challenges throughout the entirety of this effort has been trying to get more done with a sub-par coverage of theoretical tools for the size range of structures and behaviors being considered. Interfacing theoretical approaches to bridge the gap between molecular mechanics (MM) and density functional theory (DFT) – the two workhorses for all of this effort to date – have historically been too approximate or too slow (sometimes both). The MM-to-DFT divide is a gigantic (no pun intended) bridge to span, and it left the entirety of this field in the 200N's divided into (1) quantum chemical analyses of mechanosynthetic operations (dimers, ex: 1, 2, 3) and (2) molecular dynamics simulations of diamondoid behavior (bearings, gears, sorting rotors – see what Bryan Bishop has kept spinning at github.com/kanzure/nanoengineer).
From the paper: Figure 7. The model channel-insulator structure produced for assessing the performanceof the DFTB PBC-0-3 parameter set against full DFT calculations (see text). The overallstructure and components were optimized under C2v symmetry constraints. Click for a larger view.
I made a tutorial about bearings and nanotubes many years ago (A Low-Friction Molecular Bearing Assembly Tutorial, v1) and lamented it suffering mightily from being a molecular mechanics-heavy assessment. At the time, however, theoretical options were limited. The calculations were all done in CAChe (I can't even find links!) on a G4 Mac graciously donated to me by an Apple Rep (after my driving two hours to a seminar he was giving on Apple and Scientific Computing. I think everyone else in attendance was from within the same building). GAMESS-US was the only free program one could get their hands on to do DFT on a Mac, but virtually nothing about that tutorial or those structures was quantum-friendly.
Density functional tight binding (DFTB) methods are fast becoming that bridge, with optimizations of large components in this mFET structure taking only a "reasonable" bit longer than molecular mechanics would have taken 25 years ago. That's a tremendous improvement in the state of theoretical methods when one can work through an entire design and analysis process with calculations that, thanks to considerable parameterization efforts, are in very good agreement with much more involved theoretical calculations (in the mFET study, some time is spent comparing DFTB and DFT just to show the reasonable agreements).
From the paper: Figure 13. Modifications that can be made to the proposed insulator design to approachmore “vacuum-like” behavior by simple modification to the Lonsdaleite framework. Click for a larger view.
The spirit of this paper is the proposal of a molecular FET, not an exhaustive comparison of theoretical methods on model systems to show where good agreement could be obtained for selections of calculable properties. That comparative study is something we are looking at in the mid-term.
That said, and this has come up in conversation on X in the not-too-distant past, the universe described by DFTB and a given parameter set is not our reality. The same is true for DFT. The same is still true for CCSD(T) (despite it being damn-close if you're a molecule!). In the Venn diagram of DFT, DFTB, and reality, there may be designs and devices that are equally well-described by DFT and DFTB, and that description might be very close to the behavior observed for an actual device. If you believe that to be the case, and you continue to accept that we're a looooong way away from making something like this, then a plausible path to production is to design exhaustively in the DFTB space and use that assessment as a "first gate" for feeding into DFT-level work. You, of course, risk missing a much better design that would have come about had you started with DFT calculations, but you, at least, improve your chances through DFTB (for instance) of narrowing the total design space to a clunky, slow, sub-par, smoke-billowing device THAT DOES WHAT YOU WANT. After that, you improve the design – and the basic science gets handed off to the engineers.
From the paper: Figure 8. Two views of the (20,20) CNT gate (1774 atoms) from the optimized mFET. Atright, the connectivity of the (4,4) CNT to the (20,20) CNT. Click for a larger view.
There are also some sharp divides between the quantum chemistry of the components and the formulaic assessments of performance and behavior in the mFET paper. The bridging of that gap is ever-ongoing by the entire community and we await determination of where the gaps in both reside at the nanoscale.
Programs and Parameters
Gaussian 16 was the DFT workhorse for this work. DFTB+ has become the structure optimization go-to for virtually every part of design processes I find myself doing because it is fast and, for the properties of interest, accurate. PBC-0-3 might not be the best of the available parameter sets for C/H-only systems, but it has shown itself here to be a completely reasonable set for optimizations and some property predictions (again, the goal of this report was not a method survey).
From the paper: Figure 4. Three views of the proposed Lonsdaleite channel, including identification of thesingle repeat unit and representative six-unit supercell. Blue dot at lower-right is thechannel axis. Click for a larger view.
The one code added late in the game was XTB (which also sees extensive use in active work for, primarily, configuration sampling with Crest) for the specific prediction of electronic coupling between the channel and Lonsdaleite insulator through the calculation of orbital overlap integrals. The issue it was used to provide some first-pass address of – the potential for insulator electrons tunneling into channel orbitals due to the uncomfortably-ordered frontier orbital energies of both – is a very nontrivial one to deal with at the design level, one that has definite methodological dependencies, and is even worse when you account for thermal behavior. The brute-force solution is in Figure 13 (above), where the insulator can be carved out to reduce close contacts (the breakdown issues associated with that are discussed in the paper).
From the paper: Figure 14. Assessment of orbital overlap between the channel (C) and insulator (I) in ourmodel test system, employing the DIPRO method within XTB to obtain JAB,eff for twovariants of the C/I model and a benzene-nitrobenzene complex (shown enlarged and toscale with the C/I models) [27]. Click for a larger view.
The fact that you can get an estimate (at all) in 15 seconds and cite a paper from which the XTB-generated values are not too dissimilar from wB97X-D4/TZ2P calculations across the HAB79 dataset is another testament to the quality of DFTB methods.
Bonus pics from left: The highest-lying channel-localized orbital, one of the few frontier orbitals with any contribution from the channel, and the insulator-localized HOMO (from CAM-B3LYP-GD3(BJ)/6-31G(d,p) calculations).
GUI
Ralph did the first pass in HyperChem. I, since 2007, continue to use NanoEngineer-1 almost daily. I am encouraged by the possibilities within MSEP for doing this work and find some of the build approaches to be, like NE-1, less chemical and more CAD-like – which is great for this type of design work (all the better that it's new and supported on modern OSs with no issues). SAMSON remains gorgeous, opens and assigns all of the atoms in this structure on first try from an .xyz file, and is very fast on the optimizations.
From the paper: Figure 3. Top: Londsaleite channel with attached (4,4) CNT leads (502 atoms). Bottom:close up of the channel in van der Waals (left) and ball-and-stick (right) representations. Click for a larger view.
Ignoring the capabilities of a sufficiently advanced technology, the above statement is quite correct. After a little bit of math, the full-speed assessment of the proposed mFET is summarized as follows:
This hypothetical sugar-cube-sized computer switching at 10 THz would generate the energy of the Hiroshima blast every six seconds. Cooling would be challenging.
That's the more immediate concern to the proposed design, with solutions proposed that bring the exothermal nature of the operation down to reasonable levels. As an homage to some previous work addressing much improved overall efficiency and thermal behavior by way of mechanical computing, a brief discussion that ends with the following snippet follows the above:
This switching energy per operation is below the Landauer limit, meaning that a conventional (irreversible) logic device would be thermodynamically impossible to operate at this power level. This highlights the need to use reversible computing to achieve such densities.
LLMs – Use Them Before They Get Too Garbage-ified By Lousy Science In Predatory Open Access Journals
Ralph threw draft V2.6 into o3-mini-high to see what it thought of our work, then asked for a technical assessment. I'm 54 peer-reviewed papers into this whole "science thing" and I can count on one hand the number of times I've gotten reviewer comments back that were as thorough and informative as what we got back from o3-mini-high. We were all very pleasantly surprised.
It also helps that LLMs are so good at making you feel like a knowledgeable authority who knows what you're talking about. Luxuriating in the constant praise of o3-mini-high for one's "thorough consideration" and "astute insight" into some topic while it tells you that you missed an exponent or spelled "FET" wrong is certainly better than Reviewer #2 telling you that the paper should never have been submitted in the first place.
From the paper: Figure 5. Three views of the mFET Lonsdaleite insulator (5418 atoms). Click for a larger view.
A long-short LLM take ca. March, 2025 – when it's good, it's terrific. Just like wikipedia, within which tabloid-esque updates of famous people can be fought over all day long, but no respectable contributor wants to be the dummy who wrote "F = ma" wrong in a physics page. Cited references from o3-mini-high were, however, barely in the ballpark. That said, there's A LOT to train on concerning field effect transistors, basic physics, and computational chemistry. For this particular topic, the review was excellent and I'm sold on including a research-grade LLM assessment into any future technical writings (having been pleased with the physics, we'll see how good the chemistry is in a future paper).
Design Of A Molecular Field Effect Transistor (mFET)
Field Effect Transistors (FETs) are ubiquitous in electronics. As we scale FETs to ever smaller sizes, it becomes natural to ask how small a practical FET might be. We propose and analyze an atomically precise molecular FET (herein referred to as an “mFET”) with 7,694 atoms made only of hydrogen and carbon atoms. It uses metallic (4,4) carbon nanotubes as the conductive leads, a linear segment of Lonsdaleite (hexagonal diamond) as the channel, Lonsdaleite as the insulating layer between the channel and the gate, and a (20,20) metallic carbon nanotube as the surrounding gate. The (4,4) nanotube leads are bonded to the channel using a mix of 5- and 6-membered rings, and to the gate using 5-,6- and 7-membered rings. Issues of component design assessment and optimization using quantum chemical methods are discussed throughout. A 10 watt sugar-cube-sized computer made with 1018 such mFETs could deliver ~1025 switching operations per second.
The bionano site is a dramatic improvement from the NAMOT + sed'ing I worked up many years ago while getting something DNA-related stood up.
As a still-not-irregular user myself, I can appreciate the small hurdles needed to keep the program upright in modern OSs (or the need to just run VirtualBox and be done with it) – therefore noting both Bryan Bishop's dev page (worth the visit to github for the preserved gallery alone!) and Bruce Allen's Molecular Dynamics Studio effort and sourceforge-available download.
