In-situ nanospectroscopic imaging of plasmon-induced two-dimensional [4+4]-cycloaddition polymerization on Au(111)

Plasmon-induced chemical reactions (PICRs) have recently become promising approaches for highly efficient light-chemical energy conversion. However, an in-depth understanding of their mechanisms at the nanoscale still remains challenging. Here, we present an in-situ investigation by tip-enhanced Raman spectroscopy (TERS) imaging of the plasmon-induced [4+4]-cycloaddition polymerization within anthracene-based monomer monolayers physisorbed on Au(111), and complement the experimental results with density functional theory (DFT) calculations. This two-dimensional (2D) polymerization can be flexibly triggered and manipulated by the hot carriers, and be monitored simultaneously by TERS in real time and space. TERS imaging provides direct evidence for covalent bond formation with ca. 3.7 nm spatial resolution under ambient conditions. Combined with DFT calculations, the TERS results demonstrate that the lateral polymerization on Au(111) occurs by a hot electron tunneling mechanism, and crosslinks form via a self-stimulating growth mechanism. We show that TERS is promising to be plasmon-induced nanolithography for organic 2D materials.


S4
System compatibility. We have to choose a proper molecular model (e.g., specific LUMO level, Fig. 5) to match the first incident wavelength (e.g., the shorter one), while choosing a fitting gap-mode configuration (the tip/substrate material and geometry) to match the second incident wavelength (e.g., the longer one). (ii) Thermal drift. For TERS under ambient conditions, the thermal drift of the setups is unavoidable, estimated to be on the order of a nanometer per minute.
Consequently, we might not scan the exact same location during these two processes (see Supplementary Note 5).

Plasmon-induced photodamage during the TERS imaging
LSPs in confined volumes (hotspots) lead to highly enhanced electromagnetic fields, charge transfer, local heat generation, as well as hot carrier excitation. 3,4,5,6 These effects can open novel photoelectrocatalytic reaction pathways, which are not accessible thermally. For example, plasmon-induced photodamage is a general but undesired effect in SERS and TERS experiments. These chemical transformations in hotspots are perceived as sample degradation and dissociation due to the high energy of hot carriers, 9 of which energies are in excess of a certain threshold. These induced products (e.g., amorphous carbonaceous) and side reactions (e.g., desorption of molecular fragments) are similar to that known from X-ray or e-beammediated surface chemistry, despite more than 100-fold energy difference between visible and X-ray photons. 9 Supplementary Note 4 Self-impeding and self-stimulating mechanisms S5 In a previous review, 10 Schlüter and co-workers analyzed Bragg and diffuse X-ray scattering of monomer/2Dpolymer single crystals for various conversion numbers, ranging from about 15-100%. Irrespective of what happens on the local scale, effective global strain distribution by self-impeding (or random) propagation mechanisms is ideal for 2D polymerization to happen smoothly. This means that the distance of unreacted anthracene pairs increases slightly in the direct vicinity of a site where dimerization has occurred, thus lowering the probability of a second reaction next to where the previous one took place. Other 2D polymerizations in single crystals were also analyzed but not to this level of scrutiny. Except for one case with peculiar packing likely renders polymerization propagation to be self-stimulating, in all other cases the self-impeding mechanism was likely folloed. 28 In the exceptional case, the layer had to undergo enormous rearrangements to make polymerization possible at all. These rearrangements, which involve all three dimensions, bring unreacted anthracene pairs located next to a reaction site in a concerted action closer together than normal, which in turn increases the probability of dimerization. This is the reason why this particular monomer follows the self-stimulating mechanism.
It is difficult to make a clear statement in our current 2D polymerization on Au(111) case, as the interplay between monomer adlayer and Au(111) substrate is not known. Given the above insights, although solely referring to single crystals, it is reasonable to assume to first approximation that the interactions between adlayer and substrate are responsible for bringing unreacted anthracene pairs near a reaction site somewhat closer together. However, whether the two lattices (monomer and Au(111)) are commensurate to one another, which would likely be a condition for the validity of such an assumption, is not known. Compared to the above mentioned exception, which solely refers to arguments forming a rationalization chain, the current work is the first experimental proof for self-stimulating growth in a polymerization leading to a 2D polymer. Such growth results in many polymer phases growing inside a S6 'dispersing' monomer phase and forming growth fronts encircling each polymer phase, separating it from the surrounding monomer. These growth fronts are expected to be seen from HR-AFM or HR-STM imaging. However, it is still a massive challenge to obtain STM/AFM imaging of the 2DP sheets with molecular resolution under ambient conditions (see Supplementary Note 5).

Supplementary Note 5 STM imaging of the 2DP monolayers under ambient conditions
Structural analysis of 2DP monolayers is exceedingly challenging because of the very small thickness and the tiny amount of material. 11 This has the consequence that most analytical methods are not sensitive enough and causes problems with sample preparation and handlings, such as the folding/distortion of the 2D sheets and the irregular undulation between the sheets and substrates. Several groups have intensely tried to reveal the periodic structures of 2DPs by AFM or STM imaging at the molecular level, yet only two successful cases on HOPG have been reported by Murray et al. 12 and Müller et al. 11 For our TERS setup under ambient conditions, although it allows for obtaining chemical information of 2DPs at the nanoscale, the Ag tip used for TERS is not optimized to obtain atomic-resolution STM images. To avoid scratching the covalent sheet, the STM parameters are set to keep a relatively large distance between the tip and sample. Furthermore, thermal drift plays a role during our STM and TERS imaging. Consequently, it is difficult to obtain STM or TERS images of the 2DP monolayers at the molecular level. Regarding the way forward to achieve a higher resolution and to associate microscopic images with TERS results, AFM/STM and TERS imaging should be carried out in a well-controlled environment in the future, e.g., under ultra-high vacuum and at cryogenic temperature.

Plasmon-induced nanolithography of molecular monolayers
Nanolithography can create nanoscale patterns on different media, e.g., on silicon wafers and molecular monolayers, used in various fields of technology from electronic to biomedical devices. 13,14 Our current plasmon-induced nanolithography is potential to write new 2D patterns onto a molecular monolayer. In order to obtain more custom-designed 2D organic patterns, we need to combine photochemistry and surface monolayers with TERS techniques subtly. For the plasmon-induced [4+4]-cycloaddition polymerization system, more practical parameter optimization, such as the molecular structure, the incident wavelength, the tip/substrate material and geometry, and the scanning sequence, should be taken into consideration before realizing custom-designed nanolithography.
The key for the successful fabrication of 2D organic patterns is usually bound with a suitable molecule design that can restrict its rotation in the monolayer and allow the occurrence of photopolymerization in the plane. The self-stimulating nature however only means that anthracene dimers next to a reaction site are moved a bit closer together. This makes them more amenable for reaction but does not result in reaction. Bond formation still needs to be triggered.
Moreover, the spatial selectivity with which one can trigger dictates the spatial selectivity of the chemical response. Besides the proposed [4+4]-cycloaddition, several possible reaction candidates may be used in plasmon-induced nanolithography, e.g., [2+2]-cycloaddition. 15 Our new findings will be reported in due course.       Fig. 5c in main text). The average spectrum of UV-driven 2DP 1 was averaged from 32×32 pixels (1024 spectra, 100 × 100 nm 2 ). The average spectrum of monomer 1 was averaged from 32×32 pixels (1024 spectra, 100 × 100 nm 2 , see Fig. 5a in main text). Spline baseline corrections were performed in Origin 9.1 software by the 2nd derivative method with 30 anchor points. The detailed calculation of conversion numbers and the standard errors are shown below in Table S1. a.u., arbitrary units.

Supplementary Table 1. Estimation of polymerization conversions of UV-driven and
plasmon-induced 2DP 1. All peak intensity/ratio data are collected from TERS maps with 32×32 (the UV-driven sample, see Supplementary Figure 29) and 10×10 (the plasmon-induced sample, see Supplementary Figure 29) pixels, and the values are given in the format of mean(M) ± standard error (SE, δ   respectively. Distinct phase segregation between the monomer and polymer becomes visible at the nanoscale when sufficient peculiar hot carriers are generated. Such pattern formation is probably due to a different packing array of the monomer MLs or a different facet orientation on Au(111), and thus leading to the discrepancy in the plasmon-assisted catalytic activity. 18,19 Note that the packing configuration of monomer arrays, the distance between two anthracene blades, the angle of π-π stacking modes, as well as the contact between the MLs and Au(111) could impact and tune the plasmon-induced [4+4]-polymerization, and thus affect the fabrication of localized nanopatterns. 20, 21 a.u., arbitrary units.

Langmuir-Blodgett (LB) Trough and Monolayer (ML) Preparation
Amphiphilic monomers 1 and 2 were synthesized as described in our previous work. Prior to use, they were rinsed with EtOH and then millipore water.

Langmuir-Blodgett (LB) Film Transfer and Spanning
Film transfer was achieved by raising the stage carrying the substrate at a lifting speed of 0.5 mm/min. The stages were either horizontal or had a tilt angle of 45°. In the case of film transfer onto TEM grids, the copper grids with a mesh size of 1000 (PLANO, G2780C) were gently placed from the top onto the film with tweezers. A piece of writing paper was then placed onto the grids and, upon adsorption of the grids to the paper, peeled off the water surface and dried.

Scanning Electron Microscopy (SEM)
The TEM grids were placed on a holder (PLANO, G3662) and imaged with FEG-SEM (Zeiss LEO Gemini 1530, Germany) microscope with an in-lens detector.

Atomic Force Microscopy (AFM)
Tapping mode AFM height analysis and imaging were carried out on a Bruker Dimension Icon (Bruker Corporation, USA) using OMCL-AC160TS silicon tips (Asylum Research Cypher S, Oxford Instruments, UK) with a resonance frequency between 200 and 400 kHz and a spring constant of about 42 N/m.

UV-Vis Spectroscopy
UV-Vis spectra of monomer 1 in acetonitrile and the corresponding LB monolayer on quartz were recorded by a UV-Vis spectrophotometer (UV-1800, Shimadzu, Japan). The scanning speed was 400 nm/min with a spectral resolution of 1 nm. The substrates used were Suprasil quartz (Suprasil 1, 0.5 mm thick, Heraeus, Hanau, Germany).

Raman and TERS measurements
All TERS and confocal Raman spectra were acquired on a combined STM/Raman microscope (Ntegra Spectra, NT-MDT, Zelenograd, Russia) enclosed by a home-made acoustic isolation box and operated under ambient conditions. The instrument is equipped with an air objective (100 ×, NA = 0.7, Mitutoyo, Japan) and an electron-multiplying charge-coupled device (EMCCD, Newton 971 UVB, Andor, Belfast, UK). The spectrometer was calibrated by a standard neon lamp (Renishaw, UK). TERS probes were prepared by electrochemical etching of a silver wire (diameter 0.25 mm, 99.9985% purity, Alfa Aesar).

FDTD Calculations
The spatial distribution and absorption spectra of a local electric field in the nanogap between the Ag tip and Au substrate were calculated using the FDTD method (Lumerical Solution). The simulation model comprised an electrochemically etched Ag tip (curvature radius r=50 nm, cone angle θ=40°, Supplementary Figure 15), an Au metal slab, and a vacuum layer. The tipsurface distance is set at 1 nm, and the incident wavelength set at 633 nm. The dielectric function of Au and Ag was taken from the experimental data reported by Johnson and Christy. 24 A plane wave, which was polarized along the x-direction and propagated along the z-direction, impinged on the nanogap with different incident angles . In order to obtain the absorption spectra, the incident angles fixed while changing the incident wavelengths ( Supplementary Figs. 16-19). 25, 26

DFT and TERS Calculations
Theoretical Raman spectra were calculated using the Gaussian 09 software (Gaussian, Wallingford, USA) by means of DFT. All calculations, including full geometry optimizations and frequency predictions, were performed using B3LYP/6-31+G(d) basis. The keyword Integral = (Grid=UltraFine, Acc2E=11) was used to increase the two-electron integral accuracy when SCF calculations failed to converge using default run parameters. All calculated frequencies were scaled with proper factors compared to confocal Raman spectra. The optimized geometry and the calculated vibrational modes were visualized using the Gaussview 5 package.
TERS calculations based on the dipole approximation were carried out to study molecular orientation dependence as described previously. 27,28 The Raman responses are related to the magnitude of Raman tensor α and the intensity of the local electric field. When a molecule is S42 adsorbed to a flat metallic surface, the Raman signals change with the orientation of molecules because the Raman tensor α will be modified after taking surface selection rules into consideration. The intensity of gap-mode TERS signals can be approximately expressed as follows: 27 The Raman tensor can be obtained from DFT calculations, expressed as X, Y, and Z components. Moreover, the excitation of gap-mode surface plasmon resonance is based on the interaction between the incident laser and the tip-substrate gap junction by lightning rod and antenna effects. 29 Under our STM-TERS configuration, the major local field enhancement is contributed by the vertical components (ZZ axis), which are much larger than the horizontal components (XY axis). 25 where is the adsorption energy, / is the total energy for a molecule on the surface, and are the energy of isolated molecule and surface.

Conversion Ratio Calculations
In order to estimate polymerization conversion, an internal standard method was applied for analysis, as described in our previous work. 1 For example, the intensive 1425 cm -1 band of monomer 1, which majorly corresponds to ring breathing vibrations of the 9-(hydroxymethyl)anthracene blades (see Movie S1), was chosen as an indicator for this compound. After polymerization, the weak 1447 cm -1 band in the polymer suggests some residual monomer. The bands at 462, 684, and 808 cm -1 , which mostly correspond to vibrations of the center triptycene core, are basically unaffected by the polymerization and are therefore selected as internal references. After normalizing these signals for both monomer 1 (m) and 2DP 1 (p), conversions were obtained by determining the peak intensity ratios of the reference S44 vibrations with the signal at 1425/1447 cm -1 . The calculation details of conversion ratio for monomer 1 and 2DP 1 are shown below: where is the mean value of the normalized peak ratio (for residual monomer signal) from polymer samples, and is the mean value of normalized peak ratio (for original monomer signal) from monomer samples. In our measurements and calculations, and are independent variables.

Spectral Processing
For all experimental TERS imaging and spectra shown, the data were fed into MatLab Peak heights were determined by means of the automatic peak finder (maximums in the selected area) based on MatLab, which was used for intensity analysis. For a selected peak, the noise level was calculated by Matlab using a root mean square (RMS) function in the nearby area, and the individual threshold (signal-to-noise ratio over three) was calculated for each spectrum (after aforementioned preprocessing) at every pixel of the maps. In the peak ratio maps, the intensity of individual peaks would be set as zero when the value of their signal-to-noise ratio was less than three, otherwise set as the maximum. For all individual TER spectra extracted from imaging and DFT calculated spectra, the data were processed in Origin (9.0, OriginLab, USA) to normalize and fit with a Gaussian function.

S45
Supplementary Figure 34 | Data batch processing of a typical TERS spectrum in the TERS maps for intensity and peak ratio analysis. 37,38 In order to evaluate the LSP contribution in the (semi-)quantitative spectral analysis, we also followed a reported method (by Ren and co-workers) to correct the influence of the plasmon resonance dispersion on the relative intensity of the TERS peaks. 39 Interestingly, the LSP shape (background (Bg)) is almost negligible for the given data set/analysis with and without the background division (Supplementary Figure P2). However, some variations in relative band intensities (e.g., 1566, 1600, and 1670 cm -1 ) can be found when applying the two processing methods ( Supplementary Figs. P1, 2). That is mainly due to the difference in defining and fitting the background baselines (MSBACKADJ vs. ALSS).