Using atomic manipulation, one can dissociate, form and rearrange bonds, as well as alter the conformation or charge state of molecules. The molecular structures of reactants, intermediates and products are revealed at unprecedented resolution by using atomic force microscopy (AFM) and a suitably functionalized tip. Our present capabilities of manipulation and imaging of molecules by AFM approach the level of control predicted by Richard P. Feynman in his famous lecture ‘There's plenty of room at the bottom’, in which he described how molecules and materials might be formed by attaching and detaching individual atoms at will. In this Review, we discuss recent progress and the future prospects of molecule generation by atom manipulation and molecular characterization by AFM.
Considerable recent progress has been made with respect to the generation, manipulation and characterization of molecules on surfaces using scanning probe microscopy (SPM). Such microscopy involves scanning an atomically sharp tip across a surface to obtain an image. The tip serves another important purpose in that it can also be used for atomic manipulation. Various SPM instruments exist today. The year 1982 saw the invention of the scanning tunnelling microscope1, the development of which led G. Binnig and H. Rohrer to be awarded the Nobel Prize in Physics in 1986. The principles behind scanning tunnelling microscopy (STM) imaging involve electron tunnelling, a quantum mechanical effect. A complementary method, and one that this Review will focus on more, is atomic force microscopy (AFM), which was developed in 1986 by G. Binnig, C. F. Quate and Ch. Gerber2 and led them to receive the Kavli Prize in Nanoscience in 2016. In AFM, the force acting between tip and sample is exploited to obtain images.
In 1990, D. Eigler and E. K. Schweizer pioneered atomic manipulation, demonstrating the placement of atoms with atomic precision3 using STM. Later milestones include reports by Sugimoto and co-workers describing atomic manipulation at room temperature4 and demonstrating chemical sensitivity on atom inlays with AFM5. However, only few examples of molecular reactions triggered by atomic manipulation have been reported until recently. Important works include the dissociation of dioxygen by Stipe and co-workers6, an Ullmann reaction induced by atomic manipulation by Hla and co-workers7 and polymerization by Y. Okawa and M. Aono8, all performed and characterized using STM. At the time, a great challenge facing these scientists was the identification of the reaction products. The direct assignment of molecular structure by atomic-resolution imaging was yet to be enabled, and this problem led to the development of inventive methods to characterize reaction products. For example, the generation of biphenyl from iodobenzene in the Ullmann reaction was confirmed by lateral manipulation: that is, pulling of the product7. Molecular orbital imaging by STM9 increased our insight into molecular reactions and, for example, allowed observation of metal complexation10 and tautomerization reactions11 triggered by atom manipulation.
It was in 2009 that AFM tip functionalization by atom manipulation allowed for molecular imaging at atomic resolution12. Subsequently, it became possible to identify even complex molecules from their images in real space13 and to study reaction products by AFM. Concurrently, on-surface synthesis of covalently bonded molecules became an important topic of research14, and molecular wires and graphene nanoribbons15 in particular are now grown by on-surface synthesis, which involves thermal annealing of custom-made precursor molecules. Naturally, atomic-resolution AFM became an important tool to characterize the products and intermediates of thermally induced on-surface reactions16,
AFM has proven particularly valuable in the study of reactions triggered by atom manipulation, allowing for the observation of radicals formed upon molecular dissociation23,
The microscope. Having just celebrated its 30th anniversary, the atomic force microscope is nowadays a technology widely used in many applications. In general, most AFM tools work under ambient conditions and in so-called contact or tapping mode, in which the sensor makes physical contact with the sample. To maximize resolution when imaging single molecules, it is advantageous to operate under very stable and clean conditions achieved by using cryogenic systems under ultrahigh vacuum. Imaging of the surface can be non-destructive if mechanical contact between tip and sample is avoided. Atomic imaging of molecules is performed in frequency modulated non-contact mode30,31, in which a cantilever holding the tip is actively oscillated at constant amplitude at its resonance frequency by a feedback circuit. Interaction of the tip with the sample surface causes the resonance frequency to shift by Δf, the key observable in non-contact AFM. In the limit of small amplitudes, the frequency shift, Δf, is proportional to the gradient of the interaction force, Fts, between tip and sample (Fig. 1). Using 3D force spectroscopy32, quantitative force maps can be extracted33, and the interaction forces and energies can be measured.
In general, and for all experiments described in this Review, achieving atomic resolution on molecules involves the use of combined low-temperature STM–AFM systems and qPlus sensors. Developed by Franz Giessibl, the qPlus sensor is based on quartz tuning forks analogous to those used as time keeping elements in wrist watches34,35. Such sensors operate by piezoelectric detection and are comparably stiff, allowing AFM operation at oscillation amplitudes down to fractions of an ångström (1 Å = 0.1 nm) as well as simultaneous STM experiments with uncompromised performance. Working at small oscillation amplitudes is crucial to optimizing sensitivity to short-range forces, a prerequisite for obtaining atomic resolution35,36. The AFM images presented in this Review were typically acquired using peak-to-peak oscillation amplitudes of about 1 Å to obtain atomic resolution. However, atomic resolution on molecules has also been demonstrated with silicon cantilevers37 using much larger oscillation amplitudes — even at room temperature38.
Before we discuss molecular reactions performed by atomic manipulation, it is instructive to review the capabilities of SPM for molecular imaging and analysis and the importance of atomic tip functionalization. As described in the following section, the complementary scanning probe techniques — AFM, STM and Kelvin probe force microscopy (KPFM) — uncover different properties about the molecules of interest. As an illustrative example, data obtained on naphthalocyanine using different SPM modes and tip functionalizations are compiled in Fig. 2.
Atomic force microscopy. Non-contact AFM is the most important technique for structure elucidation, because it enables atomic resolution on molecules12 and its contrast mechanisms are well understood. AFM images of molecules are usually obtained in constant-height mode, in which the tip is scanned in a fixed plane above the molecule while Δf is recorded. Typically, atomic resolution is obtained by scanning the tip at a height z lower than the minimum in the Δf(z) curve (Fig. 1b). The atomic contrast mechanism is based on Pauli repulsion — bright features in the constant-height AFM images indicate repulsive force contributions and are often observed above atoms and bonds. This repulsive atomic contrast arises from overlap of the electron distributions of the molecule imaged with that of the tip apex resulting in Pauli repulsion39. The van der Waals and electrostatic forces give rise to an overall attractive background with negligible spatial fluctuations on the atomic scale; such a background is evident in the dark halos surrounding molecules in AFM images. In the simplest description, the bright areas in AFM images can be interpreted as regions of high electron density in the molecule imaged. Figure 2a shows a structural model of naphthalocyanine and Fig. 2b shows its AFM image acquired using a tip functionalized with carbon monoxide (CO) molecules40. The faint contrast differences at the centre of the molecule can even reveal positions of the hydrogen atoms, which are difficult to image by many techniques, including transmission electron microscopy.
The nature of the tip functionalization is of utmost importance for AFM imaging12. First, it is desirable for the atoms or molecules decorating the tip to be small, because the lateral resolution limit generally scales with the tip radius. Second, the tip should be rather inert to prevent the imaged molecule from being picked up or displaced by the tip. It is for the latter reason that bare metal tips are often unsuitable, with their high reactivity resulting in the analyte molecule being picked up before atomic contrast is obtained12. Other properties of the tip functionalization that are important include relaxations of the tip and its charge distribution, and the resulting electrostatic field.
Tips functionalized with CO have proven to be of high utility in obtaining high-resolution molecular structures. CO is picked up from a surface with a metallic tip (often copper) to which it readily binds, with the carbon bonded to the metallic tip and the oxygen protruding outwards41. The protruding moiety has the requisite small radius and even a high aspect ratio on the atomic scale. Additionally, tip relaxations involving the tilting of CO lead to apparent distortions resulting in the sharpened appearance of bonds42,43 and facilitate bond-order analysis42. Such relaxation does have a side effect in that it can lead to artefacts that might appear as bonds44,45. However, these tip relaxation effects are well understood42,43,45,
Scanning tunnelling microscopy. STM characteristically allows imaging of molecular electronic structure in the form of orbital densities9. The lowest unoccupied molecular orbital (LUMO) of naphthalocyanine possesses two-fold symmetry such that the positions of pyrrolic hydrogen atoms can be located. These atoms move in a tautomerization reaction induced by electron attachment11 (Fig. 2f). In general, only the frontier molecular orbitals — that is, the LUMO and the highest occupied molecular orbital (HOMO) — can be resolved using STM, because the accessible bias window for tunnelling is limited to a few volts around the Fermi level. For a neutral molecule, the LUMO (HOMO) is imaged by applying a positive (negative) sample voltage to tunnel resonantly into the negative (positive) ion resonance. For orbital imaging it is beneficial to electronically decouple the molecule from a metallic substrate using a thin insulating film, such as bilayer NaCl (Ref. 9) or monolayer xenon50. In the case of naphthalocyanine, the calculated HOMO and the HOMO density measured using a metal (s-wave) tip are shown in Fig. 2c and Fig. 2d, respectively9,11,51. Figure 2e shows the same orbital imaged using a tip functionalized with CO molecules, the π orbitals of which contribute to the tip's strong p-wave character that allows for increased resolution in imaging the lateral gradient of the orbital density51,50.
Submolecular resolution on molecules can also be achieved when STM imaging is performed using functionalized tips at voltages below resonance, a technique referred to as scanning tunnelling hydrogen microscopy52. In this case, the forces, which vary over very short distances, lead to corresponding tip relaxations that modulate the measured tunnelling currents, thus providing information regarding molecular structure. In a sense, tip functionalization acts like a force-to-current converter53.
An important technique for chemical identification is inelastic electron tunnelling spectroscopy (IETS), a form of vibrational spectroscopy pioneered by the group of W. Ho54. This technique yields energies of molecular vibrational modes that can serve as molecular fingerprints and even allow isotope discrimination54,
Kelvin probe force microscopy. KPFM is a variant of non-contact AFM in which the frequency shift is measured as a function of applied voltage58,
The manipulation toolkit. Different methods can be used to induce chemical reactions by atomic manipulation. First, the force between the tip and an adsorbed species can be used. This tip–adsorbate force can be tuned by varying certain parameters, which include: the tip–adsorbate distance; the applied voltage (which changes the electric field and, consequently, the electrostatic forces at the junction); or the charge state and/or conformation of an adsorbate. Second, the tunnelling current can trigger reactions by means of electronic excitations (electron or hole attachment), inelastic energy transfer from tunnelling electrons to the adsorbate (IET) or via hot carriers in surface states. In addition, temperature is an important parameter because processes can be enabled or accelerated at elevated temperatures.
Atoms and molecules can be moved laterally and vertically, manipulations that literally involvi them being picked up and dropped from the tip3. Lateral manipulation can take the form of pulling, pushing and sliding modes66, which afford detailed information about the manipulation process66,67. AFM allows quantification of the forces needed to pull an atom on a surface68, and sufficiently large molecules can have their internal degrees of freedom manipulated69. When generation of a molecule by atomic manipulation is desired, lateral manipulation can be important for arranging the products on the surface and might also be used to verify successful bond formation by pulling the fused product7. Furthermore, manipulation allows access to different configurations and the ability to study single-molecule switches, with tunnelling electrons inducing conformational switching70,
The most important mechanism for bond dissociation and formation is thought to be IET. Energy, typically up to several electronvolts, can be transferred from the tunnelling electrons (or holes) to the molecule on the surface. Depending on the system of interest, different processes can occur. One possibility involves part of the energy of the tunnelling electrons being directly transferred to vibrational modes of the molecule6,75. Alternatively, electrons might be temporarily injected into a vibronic state of the molecule with part of their energy then being converted to vibrational energy76. In the former case, the threshold voltage corresponds to the energy of a vibrational mode, whereas in the latter case the threshold corresponds to the energy of an electronic resonance (molecular orbital). Some situations allow for these mechanisms to be distinguished, such that energy transfer can be unambiguously attributed to, for example, an electronic excitation of an antibonding orbital or a vibrational excitation77,78. As mentioned before, other processes are also possible, including energy transfer via hot carriers in surface states79,80.
Whenever the energy transferred exceeds the barrier to a certain reaction, the precursor molecules can, of course, undergo the transformation. Depending on the energy transferred per electron, one or several electrons might be required to trigger the reaction (vibrational heating75). Early on, pioneering STM studies provided insight into the triggering of reactions, allowing the measurement of yields as a function of voltage and current, and the collection of information about the energy and the number of electrons needed for the reaction6,75. By using IET, bond dissociation6,7,10,75,81 and formation7,10,55 can be performed in a controlled manner. Also, tautomerization reactions within individual molecules can be induced using inelastic tunnelling11,81,82 (Fig. 2f). However, because STM experiments only probe electronic properties, it is not always straightforward to infer the chemical structures of complex reaction products. Owing to the recent progress in submolecular resolution, it is now possible to go further and tackle the generation of more complex molecules by atomic manipulation.
Sample preparation. The preparation of samples for manipulation and analysis is typically performed under ultrahigh vacuum conditions. Several components of the sample are important and have to be considered: a suitably clean substrate is prepared on which the analyte molecules (or their precursors) are adsorbed, these being in addition to atoms or molecules required for tip functionalizations (Fig. 3).
The sample surface is important for several reasons, particularly the strength and geometry with which it anchors the molecules during imaging. It is often desirable for the surface–adsorbate coupling to be weak to preserve the reactivity, geometry and molecular structure of a species, as well as to allow charge-state manipulation. These considerations have led to the use of ultrathin insulating films such as bilayer (100)-oriented NaCl (Ref. 83). The surface is also important to prepare the tip, which is often indented in a metal substrate to form a stable, yet atomically sharp, conductive point to which the atoms or molecules for tip functionalization are attached. Moreover, the substrate might have an important catalytic role in on-surface reactions, with surface structures or defects possibly serving as reaction sites or guiding structures for manipulation experiments7.
The analyte or precursor molecules are typically deposited on the sample surface by thermal sublimation. Small molecules or atoms to be used for tip functionalization are introduced into the chamber and adsorbed onto the surface. Alternatively, atoms for tip functionalization can be picked out of the substrate (for example, chlorine from NaCl)12,49 or dissociated from precursor molecules (for example, bromine from 9,10-dibromoanthracene (DBA))24.
Molecule identification. The most fundamental application of atomic-resolution AFM is the identification of molecules. Structure elucidation by AFM has already been used to determine the connectivity of atoms in natural13,84 and synthetic products85, as well as to characterize complex molecular mixtures86. Figure 4a shows breitfussin A, the structure of which could only be determined by combining NMR, mass spectrometry and density functional theory calculations with AFM. These studies afforded not only an unambiguous structure of this molecule, but further provide clues about the nature of a novel class of natural products. In this context, AFM revealed the connectivity of the cyclic systems and the methoxy, bromo and iodo substituents84. Furthermore, AFM was used to identify products of on-surface synthesis using heating16–18, as well as confirming the atomic structure of zig-zag graphene nanoribbons (GNRs)21, fused GNRs87,88 and boron-doped GNRs19. In a recent application, AFM has proven its utility in revealing how molecules covalently couple to graphene edges22.
AFM analytes do not need to be pre-formed molecules, and the technique has been applied to the study of novel molecules generated by on-surface synthesis. For example, the thermally induced cyclodehydrogenation of 6,6′-bipentacene on Au(111) was found to yield peripentacene18 (Fig. 4b). Different reaction intermediates could be imaged, shown in Fig. 4c, and their abundance as a function of annealing temperature was quantified to provide unprecedented insight into the reaction pathway of such surface-mediated reactions20.
Metal–ligand complexes generated by atomic manipulation. Metal–ligand complexes can be formed by tip-induced manipulation, with subsequent AFM characterization revealing the structures of the products. For example, reversible bond formation between a gold centre and a PTCDA (perylene-3,4,9,10-tetracarboxylic acid dianhydride) ligand is triggered by electron or hole attachment on bilayer NaCl on Cu(111). The complex [Au(PTCDA)] can be doubly reduced by undergoing electron attachment, after which the interaction between the metal atom and PTCDA is broken to afford Au− and [PTCDA]− — fragments that experience Coulombic repulsion (Fig. 5a,b). If a hole is attached, the gold will again bond to PTCDA to afford the monoanion [Au(PTCDA)]−, as confirmed by AFM, STM and theoretical analyses (Fig. 5c,d). The bond formation and cleavage in the Au–PTCDA system is reversible, repeatable, reliable and directed (because electrons and holes are attached at opposite voltages), such that this species is potentially of interest for applications including single-molecule switching28.
A related complex has been generated by fusing two phenazine molecules with Au− on bilayer NaCl on Cu(111). The resulting linear complex forms upon electron tunnelling at a sample bias of −2.5 V. AFM made it possible to determine the geometry of the resulting complex (Fig. 5e,f), while STM was used to characterize the electronic structure, which was consistent with the covalent nature of bonding within the complex29.
Radicals generated by atomic manipulation. Bond dissociation is one of the most studied chemical reactions induced by atomic manipulation. Typically, bond dissociation is achieved by IET, a process in which energy is transferred to a molecule and typically causes the weakest bond(s) to break, provided that the electron energy is sufficiently high. This method often leads to the dissociation of halogen or hydrogen atoms7,75. When the hydrogen atoms are removed from aromatic groups, the resulting aryl radicals are reactive and become chemisorbed such that they form covalent bonds to a metallic substrate23,89,90. It is important to distinguish between σ and π radicals. Although π radicals typically maintain the planarity of their parent species23, σ radicals, similar to the aryls just described, often deviate strongly from planarity, in particular on metallic surfaces where they form carbon−metal bonds with the substrate25,26,89,90. The radical character can often be preserved by using inert surfaces, such as ultrathin insulating layers, to curb bond formation25,26.
As expected, bonds between carbon and iodine or bromine are weaker than those between carbon and hydrogen; in the case of an aromatic (for example, phenyl) system, the bond enthalpies of Ph−H, Ph−Br and Ph−I bonds are 4.9 eV, 3.6 eV and 2.9 eV, respectively91. In this regard, the strategic placement of halogen atoms in a precursor molecule can predefine the positions at which atoms will be dissociated and radical centres are formed. Dissociation often occurs by temporary electron attachment, a process that is efficient because the minimal electron energy (applied voltage multiplied by electron charge) needed to dissociate the bond is typically within a few hundred millielectronvolts of the respective bond enthalpy. Given a judiciously chosen precursor, this method can thus selectively afford the radical of choice, which might then be stabilized and characterized at low temperature on an inert substrate.
Arynes are prominent reactive intermediates for which characterization is typically precluded by the shortness of their lifetimes, particularly under ambient conditions. Aiming to generate and study an aryne using atomic manipulation25, diiodonaphthoperylene (DINP; Fig. 6a) was deposited on bilayer NaCl on Cu(111), a substrate on which the DINP is well resolved by AFM (Fig. 6b). Removal of the two ortho-iodo groups from the outermost ring affords the corresponding unsaturated species (drawn as a cumulene in Fig. 6c). This process involves electron attachment, whereby the LUMO of DINP is populated at sample voltages above 1.6 V, with bond cleavage induced by electronic excitation25,77,78. The product (Fig. 6d) features an extended planar aromatic hydrocarbon core (the perylene backbone) suitable for bond-order analysis by AFM42. Indeed, the cumulene (rather than the alkyne or diradical) was found to be the dominant resonance structure for the unsaturated product. It was further shown that the product could be converted back to the DINP starting material by applying a voltage pulse. In this case, two nearby iodine atoms reattach to the cumulene, with such reactivity being characteristic of an aryne25. The latter result is promising for the use of highly unsaturated organics for the generation of larger, more complex molecules using bond formation by atomic manipulation.
When a bare Cu(111) substrate is used instead, the aryne that is generated immediately bonds to the surface. The iodine atoms show up as faint depressions in Fig. 6e, but the complete left-hand-side ring of the organic product can no longer be seen in the constant-height AFM image in Fig. 6e. In Fig. 6f, acquired with the tip lowered in the region of this ring, it is revealed that the adsorption height of the ring is lowered owing to the bonds formed to the Cu(111) substrate25. A similar reaction had previously been reported for dehydrogenated benzene on Cu(110)78.
Analogous methodology to that described above for cumulene generation allowed for the formation of another diradical, this time by removing bromine atoms from DBA (Fig. 7a). Again, given the anticipated reactivity of the radical product, bilayer NaCl on Cu(111) was chosen as the substrate. In contrast to the ortho-aryne discussed above, the dissociation of bromine atoms in para positions occurred in two steps92. Although the voltage threshold for dissociation of the first bromine atom corresponded to resonant tunnelling into the molecule at above 1.6 V, appreciably higher electron potentials (>3.3 V) were required to remove the second bromine atom. The diradical product can undergo a further reaction induced by atomic manipulation, with electron attachment to the diradical inducing ring opening to afford a diyne featuring a 10-membered ring. Moreover, this process is reversible: the 10-membered ring can be reconverted to the diradical triggering a Bergman reaction. Both Bergman and retro-Bergman reactions proceed upon electron attachment, which occurs above the 1.6-V threshold required for electrons to resonantly tunnel into the molecule.
Tip-induced reactions are often accompanied by lateral displacement of the diradical or diyne, which can be suppressed by anchoring the molecule to a defect, such as a step edge. Exposure of an anchored molecule to electrons with energies above the switching threshold resulted in continuous switching between three different tunnel-current plateaus (Fig. 7b). The three plateaus correspond to the diradical and the two diyne molecules with 10-membered rings on either side of the molecule. The reaction can be stopped at any of these three forms by lowering the voltage below its threshold value. Under such conditions the molecule remains stable and its structure can be determined using AFM (Fig. 7c). This reversible switching is interesting in that it involves switching spin multiplicity: the diyne features a singlet ground state, whereas the diradical has a triplet ground state with unpaired electrons localized at the 9- and 10-positions26. Aside from the optical and magnetic properties, the reactivity of the diradical is also switched in this reaction, a potentially important aspect for future molecular generation by atomic manipulation.
We foresee some important goals — feasibly achievable in the next decade — for molecular identification by AFM and molecule generation by atomic manipulation. In terms of identification, improvements can be made in chemical sensitivity and the applicability to more complex molecules of a larger and/or non-planar nature. For molecular generation by manipulation, the most important goal is the controlled fusion of molecular segments and the quest for a deeper understanding of the mechanisms underlying molecular generation by atomic manipulation.
Telling different elements apart is extremely challenging within molecules. This is mainly owing to the huge effect of the bonding environment. For example, a carbon atom will appear different depending on whether it is sp-, sp2- or sp3-hybridized, and even minute differences in bond order can significantly affect contrast. Furthermore, the non-planarity of molecules (their topography) is not easily separated from the chemical contrast, and the relatively small interatomic distances within molecules makes distinguishing elements extremely difficult. We foresee that such elemental sensitivity could be established by fingerprinting molecular subgroups by AFM. For this, a database of AFM images and the contrasts of such measured (and/or calculated) molecular moieties could be established using different well-defined tip functionalizations.
There are two main obstacles to extending molecular identification by AFM to more complex and larger molecules. The first is the preparation of such molecules in necessary clean environments, a problem that is, in part, addressed by electrospray deposition, already feasible for large molecules (∼10,000 Da) on atomically clean surfaces93,
In terms of manipulation, the fusion of molecules by atomic manipulation, although achieved in some cases7, still remains a great challenge. Bond formation by atomic manipulation seems much more challenging than bond dissociation. To trigger the latter process, it usually suffices to simply provide the requisite energy, typically by IET. By contrast, bond formation requires not only the provision of energy, but also the fulfilment of other conditions. Importantly, the products to be fused have to be placed at a certain distance and orientation with respect to each other7. The nature of the tip functionalization could also have a role in bond formation, and such catalytic tip properties have not been studied in detail.
What can be expected from the study of reactions by atomic manipulation in the future? As demonstrated in several examples, this method provides a route to molecules that are unstable or cannot be generated, stabilized and/or characterized by other techniques. A particularly important class of these molecules are organic radicals, which have already been generated and studied by tip-induced dissociation and AFM imaging. Their study will improve our understanding of radical formation, adsorption and reactivity (both in an inter- and intramolecular sense). Manipulation experiments can also yield insights into reaction pathways and mechanisms of on-surface synthesis and catalysis. Thus, novel reaction schemes might be explored and catalysts discovered. Finally, the prospect of building custom-designed molecular or metal–ligand networks and molecular machines undoubtedly fuels a fundamental and long-term interest in the rapidly developing area of atomic manipulation.
The authors thank B. Schuler, N. Moll, Z. Majzik, S. Fatayer, G. Meyer, D. Peña and R. Allenspach for discussions. They acknowledge financial support from the European Research Council Consolidator Grant AMSEL (agreement no. 682144), the European Research Council Advanced Grant CEMAS (agreement no. 291194), the European Union project PAMS (agreement no. 610446) and the Initial Training Network QTea (agreement no. 317485).