Characterisation of PTCDA/Si(111) and molecular switchingIndividual PTCDA molecules adsorb at the corner hole site of the Si(111)-7 × 7 surfaces, forming a bridging structure with four O–Si bonds between the acyl O atoms and Si adatoms adjacent to the corner hole27,28. With STM at positive sample bias (Vbias > 0), the single molecule is imaged as a four-petal-shaped protrusion (Fig. 1d), where two darker (brighter) petals correspond to the long (short) side of the rectangular molecule27. Our density functional theory (DFT) calculations revealed that the molecule remains nearly flat (Fig. 1e, f), unlike a convex structure predicted previously28 (see Supplementary Note 1 and Supplementary Fig. 1).Plasmon-induced switching occurs when a plasmonic Ag tip under laser irradiation approaches the PTCDA molecule. The switching is manifested as a two-state telegraph noise of the STM current (jSTM) (Fig. 1b), indicating a reversible reaction of the molecule. Single-molecule junctions can be further characterised by vibrational spectroscopy recorded simultaneously with conductance measurements38,39,40,41. As shown in Fig. 1c (see Supplementary Note 2 and Supplementary Fig. 2 for measurement details), the intense Raman peaks originating from PTCDA are observed at the high conductance state (namely, ON state), while no molecular peaks appear at the low conductance (OFF) state. Note that the peak at 520 cm−1 corresponds to the optical phonon mode of the bulk Si42,43. It has been previously shown that the TERS intensity is largely enhanced by the point-contact formation between a plasmonic tip and a single adatom or molecule on surfaces42,44,45. Therefore, we assign the high conductance state to a point-contact configuration, as depicted in Fig. 1a, where the O–Si bonds of one of the two anhydride groups of the molecule are broken and the dissociated half is lifted. The tip-height dependence of the junction conductance supports the assignment of the ON state geometry, as described in the next subsection.Tip-height dependence of the switching activity: 0.1-Å order controlThe switching is activated in a particular range of tip–molecule gap distance (d) and we categorise d ranges into Zones I (tip far) to III (tip close) (Fig. 2a), based on the characteristic behaviour of the jSTM (Fig. 2b) and TERS spectra (Fig. 2d–h) recorded simultaneously. This measurement was conducted at 10 K to eliminate thermal drift effects, but the same Zone behaviours were observed at 78 K (Supplementary Fig. 3). The forward (tip approach) and backward (tip retraction) traces show the same tip-height dependence, indicating that the Ag–PCTDA–Si junction acts reversibly and the process was not destructive. Note that the switching event is not coincidental but well reproducible; once we found the proper tip conditions, we could repeatedly obtain switching features until the junction deformed (see also Supplementary Note 3 for details).Fig. 2: Activation and control of the molecular switch by tip-height tuning.a Schematics of the Ag–PTCDA–Si(111) junction at four different gap distances: from left (tip far) to right (tip close), Zone I, Zone II with low ON-state occupation, Zone II with high ON-state occupation and Zone III. b Time trace of jSTM during the approach-and-retraction stroke of an Ag tip over PTCDA/Si(111) (20-ms time resolution, Vbias = −400 mV, T = 10 K, λext = 532 nm, Pext = 5.6 mW). The first (last) half of the record time, i.e. t = 0–33 s (33–66 s), is the data during the tip approach (retraction) process, where d was varied by −0.2 (+0.2) Å every 3 s. Red (blue) ribbons indicate guides for the high (low) current level corresponding to the ON (OFF) states. c TERS intensity of a 1375-cm−1 peak measured simultaneously with jSTM in (b). The purple dotted curve shows the ON-state occupation ratio calculated from the jSTM trace at each d. d–h TERS spectra in individual Zones acquired during the trace in (b). The spectra in d–f (g, h) were recorded during the tip approach (retraction) process (from d to h, d = 3.2 Å, 2.1 Å, 1.0 Å, 2.1 Å and 3.2 Å).In Zones I and III, jSTM increases exponentially as the tip approaches the molecule (blue ribbon in Fig. 2b), indicating that no configurational changes or chemical reactions are involved. In addition, the absence of the TERS peaks in these Zones (Fig. 2d, f, and h) suggests that the molecule remains in the OFF state. In contrast, in Zone II, jSTM shows the telegraph noise (Fig. 2b) and the TERS peaks appear (Fig. 2e, g). In the high-conductance state (ON state), the measured current does not vary with the tip-sample distance (red ribbon in Fig. 2b). The invariant conductance across different tip heights indicates the formation of point contact between the tip and the molecule44, consistent with the assignment by TERS described above.The ON-state occupation derived from the telegraph noise of jSTM shows sensitive d dependence (purple dotted curve in Fig. 2c), which indicates that a change in tip height affects the relative stability between the OFF and ON states. In other words, the ratio between the forward and backward reaction rates can be adjusted by changes with 0.1-Å-order precision in d (see also Supplementary Fig. 4). The d dependence of the ON-state occupation is in good agreement with that of the TERS intensity (black solid curve in Fig. 2c). This is consistent with the ON- and OFF-state TERS spectra shown in Fig. 1c because only the ON state of the switching molecule during the signal accumulation time (3 s) contributes to the TERS intensity. We also confirm that when the tip approaches further than Zone III and contacts the OFF-state molecule (d = 0; namely, Zone IV), the TERS signals of the OFF state are detected (Supplementary Fig. 3). The spectral features differ from those of the ON state, indicating the different molecular configurations between the two states (Supplementary Note 4).The switching behaviour of PTCDA/Si(111) is independent of the lateral tip position (Supplementary Fig. 5), and it occurs both over the molecular centre (perylene part) and over the edge (anhydride part). We conclude, therefore, that in the ON state, the perylene part attractively interacts with the tip apex, analogous to the point-contact formation between a metal tip and a benzene ring reported previously34 (see Supplementary Note 5 for a detailed discussion). Assuming that PTCDA has a rigid plane without intramolecular deformation, the tilting angle of the ON-state molecule (in Zone II) is estimated to be about 10–15∘ from the flat OFF configuration.Chemical dependence: active anhydride, silent imidePTCDA has various derivatives with a similar molecular frame but different chemical properties46,47. They can be used to tune the switching property without a significant configurational change in the tip–molecule–substrate junction. In this work, we examined two derivatives, perylene-3,4,9,10-tetracarboxylic monoimide monoanhydride (PMI) and perylene-3,4,9,10-tetracarboxylic diimide (PDI). In the STM images at positive Vbias, PTCDA, PMI and PDI appear as a symmetric protrusion (Fig. 1d), a protrusion-and-depression pair (Fig. 3a) and a symmetric depression (Fig. 3i), respectively (see Supplementary Fig. 6 for more images at various Vbias). These appearances suggest that the brighter (darker) half of the image of PMI corresponds to an anhydride (imide) group (Fig. 3a, b).Fig. 3: Chemical tailoring of the switch.a STM image of a PMI molecule on Si(111)-7 × 7 (Vbias = 800 mV). b Structure of PMI. c, d Side-view schematics of the Zone-II junctions for PMI with different lateral tip positions: over the imide side (c) vs the anhydride side (d). e–h jSTM traces recorded at different lateral tip positions over PMI/Si(111) (Vbias = −300 mV, λext = 532 nm, Pext = 8 μW, d = 1.5 Å). The four panels have identical scale axes. The coloured markers in a and b correspond to the tip location for the current traces with the same colours. i STM image of a PDI molecule on Si(111)-7 × 7 (Vbias = 500 mV). j Structure of PDI. k Side-view schematic of the PDI junction. l jSTM trace recorded over PDI at the position marked with the grey circles in (i, j) (Vbias = 500 mV, λext = 532 nm, Pext = 5.6 mW, d = 1.5 Å). In the figure, all measurements were conducted at T = 78 K.To evaluate the switching behaviour, we recorded jSTM at d comparable to the active range of the PTCDA switching (Zone II). In contrast to PTCDA, switching of PMI shows a clear dependence on the lateral tip position (indicated by the markers in Fig. 3a, b). When the tip is placed over the anhydride side, the telegraph noise appears in jSTM (Fig. 3e, f), whereas the tip over the imide side keeps jSTM in a low conductance state (Fig. 3g, h). For PDI, no switching was observed at any tip locations (Fig. 3l). These results indicate that the anhydride side of PMI can be lifted (Fig. 3d) in the same manner as PTCDA, whereas PDI and the imide side of PMI do not react (Fig. 3c, k). The inertness of the imide groups strongly supports that the switching behaviour originates from the reactivity of the molecule–substrate system (Fig. 1a), ruling out the possibility of the switching due to the atomic-scale deformation of the plasmonic electrode (tip apex) in the picocavity reported previously48. For PMI, the separation between the switchable and non-switchable positions is 4 Å (orange and cyan in Fig. 3a, b), which highlights the importance of the sub-nanoscale tip positioning for the reaction control.The different switching behaviours between the three molecules are explained by their adsorption energies Eads. Our DFT calculations identify that PMI and PDI adsorb at the corner hole site of the surface via the four acyl O atoms in the same manner as PTCDA (Supplementary Fig. 1). Eads is calculated to be −4.21 eV, −4.41 eV, and −4.85 eV for PTCDA, PMI and PDI, respectively, which indicates the stronger interaction of the imide group with the Si surface than the anhydride group.In terms of junction sustainability, PMI has an advantage over PTCDA. Although non-destructive switching is feasible for both molecules (Fig. 2), lateral diffusion or pick-up of PTCDA by the tip was occasionally observed in Zone II or at smaller tip heights (cf. d ≈ 6 Å for STM imaging vs d ≈ 2 Å for switching). In contrast, such irreversible events were rarely observed for PMI because the non-reactive imide side acts as a stable anchor bound to the surface. This provides an important insight into designing stable single-molecule switching in nanojunctions.Reaction mechanismThe results above suggest that switching of the PTCDA and PMI molecules on the Si surface (referred to as the anhydride/Si switch) is mediated by LSP in the Ag–Si nanojunction. However, other stimuli, such as Vbias-derived tunnelling electrons, electrostatic field or tip–sample attractive/repulsive forces, inevitably coexist in the STM junction, all of which could also contribute to the molecular reaction21,49. We conducted the following control experiments to identify the driving force of the O–Si dissociation to activate the anhydride/Si switch.First, we confirm that light irradiation to the STM junction is necessary for the switch activation (Fig. 4a and see Supplementary Fig. 7 for the laser-power dependence); the telegraph noise in jSTM disappeared in the absence of laser irradiation independent of Vbias. Second, by modifying the plasmonic resonance of the junction through the tip-apex shaping (see “Method”), we verify the near-field contribution to switching. We prepared Tips 1 and 2 and confirmed their different plasmon-resonance energy profiles (Fig. 4b) by STM-induced luminescence (STML)50. The resonance at around 500 (750) nm for Tip 1 (2) suggests that the LSP is resonantly excited by incident light with a wavelength λext of 532 (780) nm (see the markers c and e in Fig. 4b). Under the on-resonance conditions, the switching was detected (Fig. 4c, e). In contrast, when 780-nm laser irradiates Tip 1 (i.e. off-resonance), no switching was observed (Fig. 4d). We also confirmed that an LSP-resonant Ag tip with λext = 633 nm induced the switching (Supplementary Fig. 8). We conclude, therefore, that the anhydride/Si switch is initiated by plasmon excitation.Fig. 4: Characterising plasmons as the driving force of the reaction.a jSTM traces for PTCDA/Si(111) with (upper panel) and without (lower) laser irradiation (laser on: λext = 532 nm, Pext = 56 μW, d = 2.0 Å; laser off: Pext = 0, d = 1.9 Å and 2.0 Å for traces at +3 V and +1 V, respectively). The voltage value above each trace denotes Vbias used. The upper trace in each panel (−1 V and +3 V) is displayed in arbitrary units and offset vertically for clarity. b Normalised STML spectra recorded on a bare Si(111) surface with two Ag tips, namely, Tip 1 and Tip 2 (jSTM = 10 nA and 40 nA, respectively, Vbias = 3 V). The circles in the spectra indicate the STML intensities at the same wavelength as the irradiating laser used for the jSTM trace measurements in (c–e). c jSTM trace measured over PTCDA using Tip 1 at λext = 532 nm (Vbias = −300 mV, Pex = 5.6 mW, d = 2.0 Å). d, e jSTM traces recorded over PTCDA using Tips 1 and 2, respectively, at λext = 780 nm (Vbias = −300 mV, Pext = 35.5 mW, d = 2.0 Å). The junction schematics are displayed on the left side of each trace in (c–e). In the figure all measurements were conducted at T = 78 K. All the horizontal scale bars in the traces denote 1 s. f Schematic potential energy diagram of the anhydride/Si switch under the plasmon-driven HC transfer. Red arrows indicate sequential steps: the transition from the ground state (GS) to a transient charged state (CS) ①, gain of kinetic energy due to the different equilibrium nuclear configurations between GS and CS ②, relaxation to a vibrational excited state in GS ③ and transfer across the barrier ④.Plasmon-induced reactions can be triggered not only by direct transitions due to optical absorption of the reactant but also by hot carriers (HCs; high-energy nonequilibrium electrons and holes) and heat generated in the relaxation process of LSP8,51,52,53. In our case, the linear laser-power dependence of the switching rate (Supplementary Fig. 7) rules out the plasmon-mediated heat- or electric field-driven mechanism where nonlinear dependence is expected (see Supplementary Note 6 for a detailed discussion). The fact that the switching is observable with several λext also excludes the contribution of a mechanism based on direct optical excitation10,15. The described features resemble the HC-mediated dissociation of oxygen molecules strongly bound to an Ag surface in a plasmonic metal junction12,13. Therefore, we suggest that the anhydride/Si switch follows the HC transfer mechanism54 (Supplementary Note 6 and Supplementary Fig. 7). In this mechanism, the reaction is triggered by HCs excited in the plasmonic electrode and transferred to the molecule, analogous to the Antoniewicz model established for electron-stimulated desorption55. The system gains sufficient kinetic energy to overcome the activation barrier by the excitation from the ground state (GS) to a transient charged state (CS) and its relaxation due to the different equilibrium internuclear distances between the two states (Fig. 4f).The HC-driven mechanism has widely been proposed for plasmon-induced chemistry in metal–molecule–metal nanojunctions9,12,13,14. However, we emphasise that the Ag–molecule–Si nanojunction is distinct from the conventional systems. Our finite element method (FEM) simulations confirmed that, unlike an Ag surface, a Si substrate does not contribute effectively to plasmonic field enhancement in the STM junction (Supplementary Fig. 9). Consequently, HCs are dominantly provided from the plasmonic Ag tip, which probably contributes to the high controllability and target-selectivity of the switching by tip positioning (Figs. 2 and 3e–h). This contrasts with the metal–metal junctions, where the strong plasmonic electric field spreads spatially (>10 nm); HCs from the metal substrate can contribute to unspecific reactions of multiple molecules distributed over the surface area9,10.Vbias independence of the switching behaviour (Fig. 4a) excludes the contribution of Vbias-derived tunnelling electrons33,56 to the reaction. Theoretically, high-energy tunnelling electrons injected into the molecule by high Vbias potentially induce the same reactions as plasmon-mediated HCs9,12,13,14. Nevertheless, we have not observed Vbias-induced switching of anhydride/Si even at a few volts (Fig. 4a; 3 V is the highest Vbias that could be used for the jSTM trace in Zone II non-destructively). Instead, the tip-apex structure tends to be irreparably destroyed at such high voltages probably because of an excessive electric field in the junction57. Based on the FEM simulations, the plasmon-derived electric field is estimated to be less than 2.3 × 108 V/m, while Vbias of a few hundred mV will lead to an even higher direct-current field in the junction (Supplementary Note 7). This strongly suggests that the LSP-resonant tip acts as a supplier of high-energy electrons/holes (HCs) to invoke single-molecule reactions efficiently and non-destructively.The short duration of the ON state restricts the obtainable information compared to the OFF state and impedes in-depth analysis of the ON-to-OFF reaction mechanism; nevertheless, there is considerable promise for addressing this through further chemical tailoring of the target molecule. Since the forward and backward reactions proceed on a comparable time scale (Fig. 1b and Supplementary Fig. 4), we predict that the HC-transfer mechanism is also responsible for the ON-to-OFF reaction. The ON-state occupation (Fig. 2c) is affected by the double-well potential shape of the GS (Fig. 4f). The 0.1-Å scale displacement of the tip height varies the ON-state potential energy and deforms the reaction barrier, thereby modifying the reaction rate.We characterised single-molecule nanojunctions between an Ag tip and a Si surface using PTCDA and its derivatives, PMI and PDI, and controlled their plasmon-induced photoreactivity. Approaching the LSP-resonant tip under illumination to PTCDA/Si(111) induces the O–Si bond breaking to form a point-contact junction between the molecule and the tip. The Si surface leads to strong chemisorption of the anhydride molecule via O–Si bonds and reduces electric field enhancement, which hinders undesired reactions or junction destruction. The LSP also induces the backward reaction, i.e. the dissociation of the contact point between the tip and molecule, leading to the switching behaviour. The switch operation requires both an effective plasmon resonance and an optimum tip–molecule gap distance, allowing for unprecedented control of the reaction rate by changing the gap distance with 0.1-Å precision. Moreover, by comparing the three target molecules, we reveal that an imide moiety on the surface is inert against the plasmon in contrast to the reactive anhydride, although they have little difference in adsorption geometries. Such pinpoint chemical tailoring promises advanced design of single-molecule devices. The design and control of metal–molecule–semiconductor junctions should pave the way for the development of plasmon chemistry with high selectivity and controllability beyond conventional nano-optoelectronics, towards pico-optoelectronics.
