Mechano-optical switching of a single molecule with doublet emission

Abstract

The ability to control the emission from single-molecule quantum emitters is an important step towards their implementation in optoelectronic technology. Phthalocyanine and derived metal complexes (MPc) on thin insulating layers studied by scanning tunneling microscope-induced luminescence (STML) offers an excellent playground for tuning their excitonic and electronic states by Coulomb interaction and to showcase their high environmental sensitivity. Copper phthalocyanine (CuPc) has an open-shell electronic structure and its lowest-energy exciton is a doublet which brings interesting prospects in its application for optospintronic devices. Here, we demonstrate that the excitonic state of a single CuPc molecule can be reproducibly switched by atomic scale manipulations permitting precise positioning of the molecule on the NaCl ionic crystal lattice. Using a combination of STML, AFM and ab-initio calculations, we show the modulation of electronic and optical bandgaps, and the exciton binding energy in CuPc by tens of meV. We explain this effect by spatially-dependent Coulomb interaction occurring at the molecule-insulator interface, which tunes the local dielectric environment of the emitter. Our results provide an initial step towards the creation of mechanically activated single molecule optical switches.

When a single molecule emitter (SMEs) is placed in a solid or onto a surface, it is subjected to hybridization or screening of its electronic orbital structure with the local environment due to Coulomb interaction. One of the typical consequences is a change in the energies of the molecular electronic excitations, sometimes called spectral diffusion. ¹ Photoluminescence and related time-resolved techniques have been widely used to measure exciton environmental effects in SMEs, ² but achieving high spatial resolution which makes it possible to resolve individual emitters has remained a challenge. Efforts to study the local environment of excitons in 3D molecular crystals, ³ 2D materials ^4,5 or 1D carbon nanotubes ^6,7 have been made, an atomic spatial resolution is, however, still lacking. Recently, scanning tunneling microscopy induced luminescence (STML) has emerged as a powerful tool to probe optoelectronic properties on the atomic scale, ⁸ such as exciton charge annihilation, ⁹ coupled modes on molecular dimers, ¹⁰ superradiance in artificially constructed chains ¹¹ or the charge state.¹² Spectral switching has been investigated on STML by using resonant energy transfer between different excitons in Pc heterodimers, ¹³ and by tautomerization of H2Pc.¹⁴ Nevertheless, the Coulombic effects of the local environment on the excitonic states of SMEs have not been extensively addressed. ^15,16

For closed-shell single molecule emitters excited in a tunnel junction, injection is not spin-selective and the formation probability is 75% for dark triplet and only 25% for bright singlet excitons.8 To obtain higher efficiencies for light-emitting purposes, triplet-to-singlet intersystem crossing (ISC) ¹⁷ and bright triplet states¹⁸ have been proposed to effectively increase the quantum yields. Recently, the discovery of electroluminescence from a doublet, rather than from a singlet or a triplet, has opened the possibility of reaching emission quantum yields up to 100%.¹⁹ However, doublet emission has, so far, been only demonstrated for ensembles of radicals species easily undergoing chemical degradation and doublet SMEs have not been reported.

Here, we show using a combined STML/AFM measurements and ab-initio calculations that the doublet excitonic state of SMEs can be controlled in a deterministic way by atomically precise manipulations. The use of scanning probe technique at cryogenic temperature allows to manipulate the adsorption configuration of individual copper phthalocyanine (CuPc) molecules on ultrathin NaCl films on Ag(111) with sub-Å precision and consequently control their excitonic state. By injection of charge carriers we create electron-hole pairs in the resulting SME´s and detect the photon emission enhanced by the plasmonic picocavity formed by the tunnel junction. We observe a change in the wavelength of the excitonic emission between both configurations. Combination of atomically resolved STM images with AFM frequency shift maps, local tunneling spectroscopy, and ab-initio calculations, allows to determine the modification of the electronic structure of the molecules and link it to the observed shift of its doublet excitonic line.

Fig.1 shows a sequence of STM images, AFM image and electroluminescence spectra obtained during a manipulation experiment of SMEs. In Fig. 1a-c the same two CuPc molecules are resolved in the two possible adsorption positions, namely dynamic and steady states (see below), on top of 2 ML - NaCl/Ag(111) lattice (for details of sample preparation see Methods section). In Fig. 1a both CuPc molecules manifest a 16-lobe pattern characteristic of molecules in a dynamic configuration arising from a bistable adsorption geometry. The distinctive STM appearance can be rationalized by the overlapping electronic structures of two chirally rotated adsorption configurations oscillating at a frequency well beyond the STM bandwidth. ²⁰ STML spectrum (1) in Fig. 1, typical for dynamic CuPc, taken at -2.5 V above a lobe consists of a sharp line at 1.9 eV and a red-shifted vibrational sideband. The spectrum (2) taken on the dynamic molecule on the right hand side of the panel in Fig.1a however obtains a different spectral fingerprint. It shows the sharp line at 1.9 eV accompanied by an intense shoulder at lower energies, which is an indication of a sudden modification of the molecular exciton state. Indeed, a subsequent image (Fig.1b) confirms the molecule changed its appearance to a 8-lobe shape typical for a steady stabilized molecule. STML spectrum (3) remeasured on the now steady molecule shows a broad peak at the shoulder position at 1.86 eV and disappearance of the 1.9 eV feature. Spectra (4) and (5) reproduce the same behavior on the second molecule upon switching from dynamic to the steady configuration, visualised by the STM in the Figs.1b,c. Therefore, spectrum (2) can be rationalized as a linear combination of the spectra from the dynamic and steady states.

Figure 1. An experiment probing the manipulation of the adsorption geometry and its influence on the STML spectra.

(a-c) sequence of STM constant-current images of two CuPc molecules at 2ML-NaCl/Ag(111), subsequently switched from dynamic to steady state. Image parameters: 6 x 6 nm², -2.5 V, 1 pA (d) submolecular-resolution AFM image of the area in (c), taken in constant height mode with a CO-tip, image parameters 4.2 x 3.2 nm², 10 mV. (e) STML spectra of the molecules in (a-c), numbered correspondingly (1-5) and reference spectrum taken on a bare NaCl (6). Acquisition parameters for the spectra were -2.5 V, 50 pA, 60 s.

In our investigation of the dynamic to steady state conversion of molecules on 2 ML and 3 ML of NaCl on Ag(111) we found that molecular switching occurs reproducibly after applying higher current and bias in excess of 150 pA and -3 V for few seconds on any part of the molecule. The converse from steady to a dynamic state process can be achieved by applying lateral forces with the tip, however with a significantly lower efficiency, strongly indicating that the steady state is energetically favoured over the dynamic state (for details see Supp. Info.). Switching CuPc to steady state produces systematic redshift and broadening of the STML lines of the spectrum. We have measured redshifts of (20.6 ± 3.4) meV for molecules at 3 ML and (34.0 ± 3.5) meV at 2 ML, plus broadenings of (63 ± 13) % and (103 ± 29) % (Fig. 2b, Fig. S2 and Tab. S1), respectively. Both effects are more pronounced on 2 ML of NaCl than on the 3 ML. The environmentally-induced exciton modification presented here has not been observed neither in photoluminescence nor in electroluminescence of CuPc single crystals²¹ or thin films. ^22,23 The exciton switching of the CuPc SME´s must therefore originate from the interaction of the electronic structure of the molecule with its nearest local environment.

Figure 2. Analysis of the adsorption geometry in relation to the STML fingerprint.

(a) STM constant-current image of CuPc molecules in the steady and dynamic configurations and their registration with the NaCl lattice obtained with a CO-functionalized tip. Parameters 7.5 x 3.2 nm², -2.3 V, 1 pA. (b) Representative STML fingerprints of the dynamic (green) and steady (violet) configuration on 3 ML NaCl. (c-f) computationally optimized theoretical models of the steady (c,e) and dynamic geometries (d,f). (g) AFM constant-height frequency shift map with two different tip heights enabling submolecular resolution on the CuPc and atomic resolution on the NaCl substrate for registration. Parameters 1.7 x 2.8 nm², 25 mV.

An atomic-scale characterization of the steady and dynamic adsorption configurations of CuPc on NaCl provides the link between their adsorption configuration and the spectral switching. High-resolution STM image obtained with a CO-functionalized tip in Fig. 2a shows two molecules, each in one of the configurations, together with the underlying NaCl lattice. For CO adsorbed on a metal tip it has been shown that under defined tunneling feedback parameters (-2 V, const. current feedback 1 pA), the protrusions in the STM image correspond to Na atom positions. ²⁴ Accordingly it can be determined that the steady molecule is adsorbed with its metal core located on top of a Na site while the dynamic state latches on top of a Cl site. This is further confirmed by the frequency shift channel, measured on the steady configuration in Fig.2g. The image is composed of two sections measured in constant tip-sample height at different offsets facilitating atomic resolution on both the molecule and the substrate. It permits a straightforward identification of the molecular backbone (in the upper section of the panel) and enables a very precise registration with the substrate (lower section). As the image is taken in a regime where repulsive forces dominate, the lower frequency shift regions can be unambiguously attributed to Na atoms. ²⁵ By extrapolation of the substrate lattice it is then easily demonstrated that the steady CuPc is centered above a Na site (see AFM simulations in Fig. S1).

These findings agree with first-principles simulations, in which a model CuPc molecule was positioned with different angles of in-plane rotation with respect to the 2 ML NaCl lattice, over Na or Cl ions. After a self-consistent minimization of the total energy of the system, the most stable configuration is found to be 45°-rotated molecule above the Na site, and the second most stable at an ±9° angle sitting on the Cl site (see Figs.2c,e and Figs.2d,f, respectively). These two configurations perfectly match the experimentally observed steady and dynamic configurations, respectively. The calculated total energy difference between them is 90 meV in favour of the steady geometry. The simulated relaxations of the molecular backbone in both configurations are comparable and only marginally different, as in both cases the peripheral parts are slightly bent towards the substrate.²⁶

The modification of the exciton energy described above can be reproduced by excited-state DFT calculations performed for isolated CuPc and the dynamic and steady configurations on the NaCl substrate, placed either on the Na or Cl sites. NaCl is modelled by point-charges and the equilibrium orientations are derived from the previous DFT simulations. The excitation is represented by a transfer of a spin-up or spin-down electron from the highest occupied molecular orbital (HOMO) of CuPc in its ground state D0 into the lowest unoccupied molecular orbital (LUMO), producing an excited state D1. In the ground state of the isolated molecule, the singly occupied (SOMO) orbital is energetically below the doubly occupied HOMO and HOMO-1 orbitals (Fig.3a). Corresponding unoccupied orbital (SUMO) is above the degenerate LUMO orbitals. Upon the HOMO to LUMO electron transfer, the HOMO orbital is destabilized and the degeneracy in LUMO orbitals is removed. The orbital ordering below HOMO is also a subject of rearrangement, mainly reversing the relative energy ordering of HOMO-1 and SOMO levels. Despite being able to obtain the emission energy solely for the isolated molecule, all three scenarios produce consistent values for the absorption in both spin branches and manifest the same orbital energy reordering upon excitation (see Tab.S3). Only subtle energy changes ensue as a result of the electrostatic field action, particular for each adsorption configuration on NaCl and spin asymmetry. Therefore, we estimate by extrapolation that the optical gap of the steady state is by 22 meV lower than of the dynamic state, in a good agreement with the experiment.

Figure 3. Theoretical analyses of the excited and ground states and the impact of the substrate.

(a) Scheme of the simulated orbital energy level reordering upon a transfer of an electron (marked by red color) from HOMO to LUMO within an isolated CuPc molecule, showing the occupied and virtual levels near the Fermi level (E_F). The ground state is denoted as D₀, excited state as D₁. Corresponding orbital geometries are depicted in the same order for clarity. (b-g) Calculated charge redistribution isosurface plots and profiles, showing the depletion (blue) and accumulation (red) (at ±0.003 e^-/ų, respectively) of electrons in the CuPc / 2ML-NaCl systems in the dynamic (b,c) and steady (e,f) configuration. (d,g) Corresponding profiles of the electron density, obtained by integration in the directions parallel to the NaCl surface. Plots are superimposed onto the atomistic model for orientation.

Knowing that the distinct spectral fingerprints are linked to their respective adsorption geometries, we now focus on a detailed electronic structure characterization of CuPc on 3 ML NaCl using differential conductance tunnelling spectroscopy (dI/dV). Figure 4a shows the dI/dV spectra in a bias range encompassing the peaks originating from LUMO and HOMO of the steady and dynamic molecules, measured before and after a controlled stabilization of the CuPc at their lobes, for various Z-setpoints. Using normalization we determine that the energy gap of the steady CuPc is about 100 mV narrower than of the dynamic CuPc (Fig. 4b,c). The apparent LUMO and HOMO positions of the stabilized molecule are shifted towards the Fermi level by 30 and 70 meV, respectively. Additional local contact potential differences measured with various tips over both configurations using Kelvin probe parabolas show a small shift up to 45 meV (see Fig. S5), consistent with a decrease of the overall electrostatic potential on the steady molecules with respect to the dynamic ones. ²⁷

Figure 4. Determination of the transport gaps by differential conductance spectroscopy.

(a) dI/dV spectra of one CuPc / 3ML-NaCl molecule in a dynamic state (green) and after stabilization (violet), taken at the positions above the molecule lobes, marked by arrows in the STM constant-current images in the insets. The spectra have been taken on the molecular lobes at various current setpoints (20 - 90 pA at -2.5 V). Normalized dI/dV plotted in logarithmic scale, corresponding to the HOMO (b) and LUMO (c).

One can anticipate that a lower HOMO-LUMO gap will be reflected in a lower exciton energy upon D₀ → D₁ transition, following an orbital level renormalization. However, the 100 meV value measured by the dI/dV exceeds by far the difference of the corresponding exciton energies obtained from the optical spectroscopy (21 meV). The apparent HOMO-LUMO energy difference measured from dI/dV represents the transport gap of the system, which may generally differ from the real HOMO-LUMO gap due to transient renormalization of the frontier orbital energies upon electron/hole injection.²⁸ The electronic gap of solid CuPc thin films on Au(111) has been reported to be 3.1 eV by direct and inverse photoemission.²⁹ These results compare well with the 3.0 and 2.9 eV gaps measured here by dI/dV for dynamic and steady molecules, respectively, and imply that neither the high electric fields of the tunnel junction nor the presence of the tip and insulating substrate alter the measured electronic bandgap of the molecules significantly.

The difference between the HOMO-LUMO gap derived from dI/dV and the optical gap measured from STML is a strong indication of a modification of the exciton binding energy (E_BE).³⁰ The difference (□E_BE) between its values on the Na and Cl sites (E_BE^Cl, E_BE^Na) can be estimated using all the measured transport and optical gaps. 5 In particular, taking the measured transport (E_GAP^Cl, E_GAP^Na) and optical gaps (E_OPT^Cl, E_OPT^Na) for the dynamic and steady CuPc molecules, □EBE follows:

$$\Delta E_{BE}=\left(E_{GAP}^{CI}-E_{OPT}^{CI}\right)-\left(E_{GAP}^{NA}-E_{OPT}^{NA}\right)=\left(E_{GAP}^{CI}-E_{GAP}^{NA}\right)-\left(E_{OPT}^{CI}-E_{OPT}^{NA}\right)=\left(100meV\right)-\left(21meV\right)-\left(21meV\right)=79meV,$$

meaning that the exciton of the steady molecule is less stable than of the dynamic molecule. We note that this derivation provides the relative differences among the configurations rather than absolute values of the exciton binding energies in CuPc (reported to be 0.6 eV 29). Since our measurements of the dI/dV gap and STML have been performed for the two very similar systems and with identical tip most of uncertainties are removed which allows us to conclude that the 79 meV can only be related to the change of the CuPc doublet exciton binding energy.

A detailed analysis of the ab-initio calculations of the ground states in Fig.3b-g allows to evaluate the total electron density redistribution on the two adsorption configurations and points out the origin of the exciton energy difference between the two states. Data shows the density difference (Δ⍴) of the fully perturbed CuPc and NaCl substrate relative to their unperturbed states. Isosurface plots (at Δ⍴ = ±0.003 e^-/ų) in Fig. 3b,c,e,f corresponding to the electron density increase (blue) and depletion (red) reveal the characteristic interaction of individual atoms within the molecule with the Na and Cl sites in their vicinity. A systematic electron depletion is visible under the atoms atop of the Na in the interface region; conversely for atoms residing above the Cl ions an electron density increase occurs. Most striking difference is on the metal core, where the charge redistribution is of opposite sign for the two configurations centered above Na and Cl. It also leads to the distinct spatial redistributions at the ligand. This can be understood in terms of simple electrostatic action of the substrate on the electron envelope of the molecule (see the Hartree potential in Fig. S6).

The net effect of the surface potential on the molecule is an electron accumulation in the interface region and depletion at the molecule, visible in the plots of Δ⍴ integrated across the axes parallel to the surface, plotted as a function of Z (Figs.3d,g). The electron density in the interface region is slightly larger in the case of the dynamic configuration, hinting at a higher overall positive field. Indeed, the total numbers of Na⁺ and Cl^- ions directly acting on the molecule vary between 9 Na, 12 Cl on the steady vs. 12 Na, 13 Cl at the dynamic configurations, respectively.

The altered charge redistribution demonstrates the extent to which the exact NaCl lattice orientation and registry with the molecular backbone affects charges within the system. An implication of the differences in the Coulomb interaction will impose an inevitable correction on the electron-hole bound state (exciton) and the transport gap. Apparently, as seen from the experiment, the stabilization of the molecule leads to lowering of the electronic gap, exciton and binding energies, and widening of the main spectral line. While the exciton and gap energetics can be explained using the interaction with the substrate potential and screening by the substrate, the spectral linewidth, which is notoriously difficult to split into individual contributions due to a number of both radiative and non-radiative recombination processes, ^11,31 remains an open question. We can merely state, based on the experiments, that the excitons on the steady CuPc appear to have a higher probability of non-radiative decay into the bulk (by e.g. coupling to the phonon bath), causing the spectral line broadening

In conclusion, we have shown an exciton mechano-optical switch with doublet emission consisting of a single CuPc molecule on NaCl on Ag(111). A change in energetics of its doublet excitonic state could be triggered through atomic-scale manipulations that allow to precisely define the adsorption configuration - with the CuPc molecule centered either above a Na or Cl atom. From the measured values of the optical transitions and changes in the system transport gaps, we estimate a difference of the exciton binding energies: exciton on the dynamic molecule is 79 meV more stable than on the steady one. Theoretical calculations confirm the experimentally determined geometries as the most favored, reproduce the changes in the exciton energy and provide details on the Coulomb-mediated charge redistribution within these systems. This allows to relate the unique local effect of the NaCl lattice potential on CuPc orbital levels with modifications in the exciton energy and electronic gaps. Our results represent a step towards a detailed understanding of the effect of the local environment in the many-body excitations in single-molecule emitters.

Methods

Preparation of sample and tip

The experiments were performed in an ultrahigh vacuum (UHV) system with a base pressure of 5x10^-11 mbar. NaCl was evaporated on Ag(111) kept at 400K to produce 2 ML and 3 ML NaCl islands. The CuPc molecules were evaporated onto the sample kept at T<10 K from a tantalum crucible. CO dosing has been achieved by exposition of the cold sample to CO partial pressure of 10^-7mbar for 1 min.

STM, STS, AFM and Kelvin probe

The combined STM and ncAFM sensor attached to a scanner of 4K UHV SPM instrument (Createc Gmbh) was based on a tuning-fork sensor 29 operating at 30 kHz, equipped with an Ag or Pt/Ir tip, sharpened by focused Xe-ion beam (Tescan FERA3). The resulting quality factor of the sensor at 7 K was >22000. The setpoint amplitude of oscillation for the frequency modulation mode was set to 50 pm.

STML

The optical path consists of two lenses, first fixed to the coarse scanner and second on the air side, coupling the photons into a fiber of a spectrograph. The spectrograph is a Shamrock 163i with iDus 401 BV or Newton 920 BEX2-DD CCD camera. The overall spectral resolution was 3.5 nm FWHM, the spectral range was 400 - 840 nm. All spectra and maps were collected in an accumulating regime.

Density Functional Theory calculations of the geometries and charge redistribution

The calculations were carried out using the FHI-aims code³² to describe the electronic behavior of the CuPc molecule on the bilayer NaCl(100) surface. The calculations were performed using the GGA-PBE approximation of the exchange-correlation potential including the Tkatchenko-Scheffler approach of the Van der Waals interactions.³³ The relativistic effects were taken into account by applying the scaled zeroth-order regular approximation.³⁴ The NaCl(100) surface was modeled by a 7x7 supercell made of two layers on which a single CuPc molecule was placed in on-top position at the Na or Cl ion respectively. The molecule was rotated by 15°, 30° and 45° with respect to the high-symmetry axes. Structural relaxations of the slab were performed for all the atoms, except the bottom NaCl layer. The calculations were considered converged when the remaining atomic forces and the total energy were found below 10^-2 eV/Å and 10^-5 eV respectively. A single gamma point was used for the integration in the Brillouin zone. The total energy calculations were conducted to find the best adsorption site. Furthermore the total density and the Hartree potential were used to determine the electronic interactions between the surface and the molecule.

Excited-state DFT calculations

Self-consistent solutions of the electron wavefunctions and geometrical optimization were obtained using the wB97XD ³⁵ and LC-wPBE ^36,37,38 functionals in the spin-unrestricted mode, employing the SVP basis set.^39,40 Visible absorption and emission spectra were derived using the comparison of the total energies of the D₀→ D₁ transiently excited systems with their corresponding ground states, independently for both spin branches. The electrostatic field of the NaCl substrate was simulated by a fixed bilayer of charges placed beneath the CuPc, calculated using B3LYP ^41,42,43 functional, geometrically equivalent to the Na and Cl ion coordinates in the steady and dynamic configurations of the system (for more details Supporting Information). The Gaussian program package (Gaussian 16, Revision C.01, M. J. Frisch et al. Gaussian, Inc., Wallingford CT, 2016) was used for the calculations.