Positionally-Controlled Mechanosynthetic Addition

• Positionally-controlled mechanosynthetic addition is the synthesis of specific atomic structures by mechanically directing C2 precursors to predetermined silicon sites.
• Experiments using inverted-mode STM on both clean and H-passivated Si(100) demonstrate controlled covalent bond formation with selectivities up to 93%.
• Sequential assembly through stepwise C–C bond formation highlights the importance of bond strength matching and fine trajectory control in achieving deterministic build sequences.

Positionally-controlled mechanosynthetic addition is the synthesis of specific atomic structures by mechanically manipulating reactive precursors in three dimensions so that a chosen moiety forms covalent bonds at a preselected surface address. In current experimental realizations, the transferable fragment is a carbon dimer, $\mathrm{C_2}$, donated from a functionalized molecular tool in inverted-mode scanning tunneling microscopy (IM-STM). Two complementary demonstrations define the topic on silicon build sites: localized carbon donation to atomically clean crystalline Si(100), yielding carbon-added motifs such as IR-C2, OD-C2, and rarely ID-C2, and higher-selectivity donation to pre-patterned dangling-bond targets on hydrogen-passivated Si(100), including single-site IR-C2 placement, multi-site patterning, and stepwise conversion of IR-C2 into IR-C4 through successive C–C bond formation (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026).

1. Definition and distinguishing criteria

In this context, mechanosynthesis is explicitly defined as synthesis of specific atomic structures by the mechanical manipulation of precursors in 3D space with precise, positional control over covalent bond formation (Blue et al., 11 Jun 2026). The defining feature is not merely that STM is present, nor merely that a reaction occurs at a surface, but that a pre-functionalized reagent is oriented relative to a selected build site and then driven through a bond-forming event by controlled motion.

Several distinctions are central. The reactive moiety is preloaded on a molecular tool rather than supplied by gas-phase precursor exposure during transfer. The critical donation step is performed with no applied bias, so the decisive chemistry is not conventional electron-induced surface chemistry. The operation is in situ, uses atomically structured tool and build-site geometries, and is intended to localize both where addition occurs and what covalent transformation is attempted (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026).

A common misconception is that any STM-created surface product constitutes positional mechanosynthesis. The clean-Si(100) donation experiments show local positional control over where transfer is attempted and where products appear, but not complete deterministic control over which nearby carbon bonding motif results under simple $z$-only trajectories. By contrast, the hydrogen-passivated Si(100) experiments couple patterned dangling bonds to force-mediated transfer and then to sequential C–C bond formation, providing stronger evidence for predetermined placement and chemically specific product formation at lattice-defined addresses (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026).

2. Inverted-mode STM and molecular tool architecture

The experimental platform is IM-STM. Here the usual STM geometry is inverted: sparsely deposited upright molecules on the sample act as imaging probes, while the nominal “tip” is a silicon probe chip (SPC) whose flat crystalline apex serves as the build site. Properly bound molecules generate reflected probe images (RPIs), so the same junction reports both the state of the molecular tool and the local structure of the SPC apex mesa (Blue et al., 11 Jun 2026).

The principal donation tool is EAOGe-C2I, an adamantane-tripod derivative with flexible legs and chemisorbing feet, an adamantane core that constrains orientation, a Ge bridgehead, and a terminal $-\mathrm{C_2I}$ feedstock (Blue et al., 11 Jun 2026). In the hydrogen-passivated implementation, the tool is described as a Ge-substituted adamantane derivative bearing a C$_2$ functional group, an iodine cap, and three OH-terminated legs; a subset stand upright so that the Ge–C$_2$–I axis is approximately perpendicular to the surface (Cowie et al., 26 May 2026). These upright molecules are simultaneously reagents and imaging probes.

The iodine cap protects the reactive feedstock until deliberate deiodination. After activation, the reactive species is the de-iodinated radical donor EAOGe-C$_2^\bullet$ or EAOGe-C$_2\!\bullet$, not the intact precursor molecule (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026). On clean Si(100), deiodination is identified by a discontinuous increase in baseline tunneling position, $\Delta z \approx 300\ \mathrm{pm}$, together with a change in RPI appearance (Blue et al., 11 Jun 2026). On H:Si(100), post-reaction changes in imaging modality and apparent height indicate that the tool termination has changed because the C$_2$ fragment has been lost (Cowie et al., 26 May 2026).

3. Build sites and the operational meaning of positional control

Two build-site regimes are now established. One uses atomically clean crystalline Si(100) on the SPC apex mesa, with a build site larger than $>5 \times 5\ \mathrm{nm^2}$. The Si(100) dimer rows and troughs define the positional vocabulary of the donation products: on-dimer, inter-dimer, and inter-row or trough-centered (Blue et al., 11 Jun 2026). In this regime, positional control means selecting an individual molecular tool, selecting an individual target region on the clean Si(100) build site, laterally aligning the radical feedstock with sub-ångström precision, and then executing a defined approach trajectory, initially emphasized as $z$0-only (Blue et al., 11 Jun 2026).

The second regime uses an H-terminated, $z$1-reconstructed Si(100) surface. Hydrogen passivation suppresses general reactivity, so chemistry is enabled only where dangling bonds (DBs) are deliberately written by bias pulsing through electron-stimulated desorption of H (Cowie et al., 26 May 2026). The fundamental target is an inter-row DB pair, IR-DB, spanning a trough between adjacent dimer rows. Because the DB pair is created before transfer and the donor is aligned beneath the center of that pair before donation is attempted, the reaction is localized to a predetermined lattice-defined address (Cowie et al., 26 May 2026).

This difference in substrate preparation matters. On clean Si(100), the selected region is controlled, but the exact donated geometry can relax into IR-C2, OD-C2, or rarely ID-C2. On H:Si(100), the inert background and pre-patterned reactive sites permit a stronger form of addressability, and after one C$z$2 is placed, either newly written DB patterns or the previously transferred carbon motif can serve as the next programmed target (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026). This suggests that passivation plus explicit reactive-site writing is a key route from local placement toward deterministic build sequences.

4. Donation chemistry and mechanistic pathways

The fundamental activation step is deiodination: $z$3 After activation, the bond-forming event is carried out with the bias set to $z$4, eliminating tunneling current during the actual transfer attempt. The STM $z$5-controller is disengaged, and the SPC is moved toward and away from the molecular tool in iterative approach–retraction cycles (Cowie et al., 26 May 2026). In the H:Si(100) experiments, the maximum approach depth is increased in $z$6 increments, less than half the $z$7 C$z$8-dimer length, and a constant-current IM-STM image is acquired after each increment (Cowie et al., 26 May 2026).

For donation to a dangling-bond pair, the reaction is summarized as

$z$9

and for extension of an existing carbon motif,

$-\mathrm{C_2I}$0

The second transformation is the crucial stepwise C–C bond-forming event that generates a four-carbon chain assigned as IR-C$-\mathrm{C_2I}$1, described as a surface-bound polyyne-like diyne structure (Cowie et al., 26 May 2026).

The mechanistic model for single-site donation on H:Si(100) uses QM/MM with xTB(GFN0)/DFT $-\mathrm{C_2I}$2. As the SPC–molecule separation $-\mathrm{C_2I}$3 decreases, a critical distance $-\mathrm{C_2I}$4 is reached at which the first C–Si bond forms, accompanied by a sharp drop in potential energy. During retraction, the Ge–C bond cleaves, leaving a transient upright pendent C$-\mathrm{C_2I}$5 intermediate that is coordinatively stabilized by the remaining EAOGe$-\mathrm{C_2I}$6 fragment; continued retraction weakens that interaction until relaxation into IR-C$-\mathrm{C_2I}$7 becomes barrierless (Cowie et al., 26 May 2026). At larger approach depths, an alternative pathway can also yield IR-C$-\mathrm{C_2I}$8, now with the second Si–C bond forming while carbon is still attached to Ge.

The IR-C$-\mathrm{C_2I}$9 pathway is more elaborate. Approach of EAOGe-C$_2$0 beneath IR-C$_2$1 produces radical addition and a vinyl-radical intermediate; during retraction, that intermediate partially detaches from the underlying Si and rearranges into a four-carbon diyne intermediate bound to both the surface and the tool; Ge–C cleavage then generates a transient pendent C$_2$2 that relaxes to IR-C$_2$3 (Cowie et al., 26 May 2026). This provides direct evidence for successive C–C bond formation during controlled carbon-chain assembly.

On clean Si(100), the mechanistic description is qualitative but chemically parallel. During bias-free approach, a new Si–C bond forms between the ethynyl radical and the selected site. On withdrawal, one of three outcomes follows: desired donation, in which the Ge–C bond breaks and $_2$4 remains on the surface; a null pathway, in which the newly formed Si–C bond breaks instead; or abstraction, in which a Si atom remains attached to the C$_2$5 moiety and is removed from the build site as EAOGe-C$_2$6Si$_2$7, leaving a silicon vacancy (SiV) (Blue et al., 11 Jun 2026). The energetic competition is framed by nearly degenerate bond dissociation enthalpies,

$_2$8

$_2$9

with substrate Si–Si bonding much weaker,

$_2$0

(Blue et al., 11 Jun 2026). On H:Si(100), the authors explicitly state that “all barriers in the reaction pathway are overcome by the mechanical work performed by varying $_2$1” (Cowie et al., 26 May 2026).

5. Products, verification, and quantitative performance

The clean-Si(100) donation products are IR-C2, OD-C2, and more rarely ID-C2. Their assignments rest on before/after IM-STM images, comparison with simulated STM images, and correlated changes in tool imaging modality. Donation products on the build site correlate with the Ge$_2$2-terminated tool modality, whereas abstraction products correlate with the C$_2$3Si$_2$4-terminated tool modality (Blue et al., 11 Jun 2026). Under one standardized condition—EAOGe-C2I, $_2$5-only, on-dimer trajectory—donation occurred in $_2$6 interactions, corresponding to an estimated true donation rate of $_2$7 with 95% confidence by the Wilson method; more broadly, simple $_2$8-only trajectories gave $_2$9 selectivity with strong interaction-site dependence (Blue et al., 11 Jun 2026).

The hydrogen-passivated Si(100) work provides a higher-selectivity and more explicitly programmable version of the same basic operation. Single-site donation places IR-C$_2^\bullet$0 exactly at the former IR-DB location, with a nearby unchanged single DB serving as an internal positional reference. Multi-site patterning includes two adjacent IR-C$_2^\bullet$1 units, three adjacent IR-C$_2^\bullet$2 units, and a nine-unit “X” pattern containing 18 carbon atoms, all built by repeated site-specific transfers (Cowie et al., 26 May 2026). Sequential growth revisits the same local arrangement through

$_2^\bullet$3

demonstrating controlled chemical advancement of selected substructures (Cowie et al., 26 May 2026).

The corresponding targeted yields in that multistep build sequence are:

• IR-C$_2^\bullet$4: $_2^\bullet$5, 95% CI 89–96%
• 2IR-C$_2^\bullet$6: $_2^\bullet$7, 95% CI 91–99%
• IR-C$_2^\bullet$8/C$_2^\bullet$9: $_2\!\bullet$0, 95% CI 81–97%
• 2IR-C$_2\!\bullet$1: $_2\!\bullet$2, 95% CI 68–93%

The outcome plots use Wilson-score 95% confidence intervals, and each step in the build sequence was replicated from tens to more than 100 times (Cowie et al., 26 May 2026).

Verification of site-selective addition on H:Si(100) combines reproducible imaging-modality changes, comparison with simulated filled-state STM images, exclusion of alternative H:Si defect assignments, and mechanistic calculations. For IR-C$_2\!\bullet$3, the simulated image reproduces the experimental trough-centered hexagonal feature with twofold mirror symmetry and little height contrast relative to surrounding H:Si. For IR-C$_2\!\bullet$4, theory predicts a symmetric high-contrast feature and identifies it as lower in energy than alternative geometries by more than $_2\!\bullet$5 (Cowie et al., 26 May 2026).

The off-target outcomes are also quantified. The main competing pathway is H abstraction from nearby H-passivated Si by EAOGe-C$_2\!\bullet$6, observed in $_2\!\bullet$7 of $_2\!\bullet$8 interactions, or $_2\!\bullet$9 with a 95% confidence interval of $\Delta z \approx 300\ \mathrm{pm}$0. A likely Si atom abstraction event occurred in $\Delta z \approx 300\ \mathrm{pm}$1 of $\Delta z \approx 300\ \mathrm{pm}$2 interactions, or $\Delta z \approx 300\ \mathrm{pm}$3 with 95% CI $\Delta z \approx 300\ \mathrm{pm}$4. In addition, $\Delta z \approx 300\ \mathrm{pm}$5 of $\Delta z \approx 300\ \mathrm{pm}$6 interactions, about $\Delta z \approx 300\ \mathrm{pm}$7 with 95% CI $\Delta z \approx 300\ \mathrm{pm}$8, gave isolated unrepeated outcomes (Cowie et al., 26 May 2026). The explicit categorization of these events is important because it shows that controlled donation is being measured against a defined off-path manifold rather than inferred from anecdotal successes.

6. Design principles, constraints, and significance

Several general principles emerge from the present demonstrations. Bond-strength matching controls transfer tendency: for donation, the bridgehead–feedstock bond must be strong enough to preserve the tool but not so strong that retraction favors rupture of the newly formed substrate bond instead. The EAOGe-C2I donation tool sits near that boundary because $\Delta z \approx 300\ \mathrm{pm}$9 (Blue et al., 11 Jun 2026). By contrast, replacing the Ge bridgehead with C in MAOC-C2I raises the bridgehead–feedstock bond to

$_2$0

which suppresses donation and strongly favors abstraction. In the cited abstraction experiments, MAOC-C2I gave $_2$1 selectivity at $_2$2 and $_2$3 at $_2$4, with no observed C$_2$5 donation (Blue et al., 11 Jun 2026). The comparison shows that donation and abstraction are competing outcomes of a common interaction manifold and that tool engineering is a problem of bond energetics, orientation, radical localization, and trajectory.

Geometric constraints are equally important. The tripod architecture is intended to anchor the tool reproducibly, expose the feedstock, and constrain orientation relative to the target. The papers also emphasize that trajectory matters: the present addition work on clean Si(100) used primarily simple $_2$6-only trajectories, whereas the hydrogen-passivated study used a more tightly controlled protocol of deiodination, lattice-addressed target preparation, zero-bias approach/retraction, and post-transfer verification (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026). This suggests that mechanosynthetic functionality is not purely a molecular property; it is a coupled property of molecule, substrate, target-site programming, and motion protocol.

The limitations are explicit. Current operation requires UHV, cryogenic temperature—typically $_2$7 for donation studies—atomically prepared silicon surfaces, and serial manipulation (Blue et al., 11 Jun 2026, Cowie et al., 26 May 2026). On H:Si(100), donor depletion is inherent because once a molecule has donated its C$_2$8, the SPC must be moved over a fresh upright molecule for subsequent steps. Scalability is therefore limited by sparse upright molecular tools, repeated registration to fresh donors, 4 K operation, and fragment-by-fragment serial build sequences (Cowie et al., 26 May 2026). On clean Si(100), exact donated geometry remains incompletely deterministic under simple $_2$9-only trajectories (Blue et al., 11 Jun 2026).

Within those constraints, the significance is precise. The clean-Si(100) work establishes the first experimental demonstrations of atom addition by mechanically controlled covalent transfer, specifically carbon donation to a crystalline surface build site, with the decisive step performed without applied bias (Blue et al., 11 Jun 2026). The hydrogen-passivated Si(100) work then demonstrates simultaneous spatial and chemical control: a passive surface background, deliberately written reactive sites, a specific transferable fragment, a force-mediated vertical transfer protocol, patterned multi-site placement, and iterative growth through successive C–C bond formation (Cowie et al., 26 May 2026). Taken together, these results define positionally-controlled mechanosynthetic addition as an experimentally realized mode of atomically precise fabrication in which covalent placement is addressable, mechanically induced, and in at least one platform programmable over multiple build steps.