Light-enhanced liquid-phase exfoliation and current photoswitching in graphene–azobenzene composites

Multifunctional materials can be engineered by combining multiple chemical components, each conferring a well-defined function to the ensemble. Graphene is at the centre of an ever-growing research effort due to its combination of unique properties. Here we show that the large conformational change associated with the trans–cis photochemical isomerization of alkyl-substituted azobenzenes can be used to improve the efficiency of liquid-phase exfoliation of graphite, with the photochromic molecules acting as dispersion-stabilizing agents. We also demonstrate reversible photo-modulated current in two-terminal devices based on graphene–azobenzene composites. We assign this tuneable electrical characteristics to the intercalation of the azobenzene between adjacent graphene layers and the resulting increase in the interlayer distance on (photo)switching from the linear trans-form to the bulky cis-form of the photochromes. These findings pave the way to the development of new optically controlled memories for light-assisted programming and high-sensitive photosensors.

G raphene is a one-atom-thick two-dimensional material with unique mechanical, optical, thermal and electrical properties 1 , promising future emerging technologies, including flexible and wearable electronics 2 . Two main approaches are being followed for graphene production: bottom-up and top-down 3 . The former relies on the assembly of suitably designed molecular building blocks, undergoing chemical reaction to form covalently linked networks 4 . The latter occurs via exfoliation of graphite into graphene 3,5 . Bottomup techniques, in particular those based on organic syntheses starting from small molecular modules 6,7 , when performed in liquid media, are both size limited, because macromolecules become less soluble with increasing size 6 , and suffer from the occurrence of side reactions with increasing molecular weight 6 . The growth on solid (ideally catalytically active) surfaces allows one to circumvent these issues. Substrate-based growth can also be achieved by chemical vapour deposition (CVD) 8,9 or via silicon evaporation from silicon carbide 10 , which rely on the ability to follow a narrow thermodynamic path. Top-down approaches can be accomplished under different environmental conditions. Among them, liquid-phase exfoliation (LPE) 2,3,5,[11][12][13][14][15] is an attractive strategy, being extremely versatile and applicable to a variety of environments and on different substrates. While bottom-up methods, in particular CVD, can yield large size, LPE gives limited sheet sizes 11,[16][17][18] . Nevertheless, LPE has several advantages. It is a viable low-cost process, which can be easily upscaled to mass-produced dispersions processable by well-established techniques such as spin-coating, drop-casting, screen-printing and ink-jet printing 17 . Increasing research efforts are being devoted to the production of graphene via LPE to improve the material's physicochemical and electrical properties.
Small organic molecules, such as surfactants or dispersionstabilizing agents, can promote the exfoliation of graphite into graphene in organic solvents 5,11 , in particular when such molecules have higher adsorption energies on the basal plane of graphene than those involved in solvent-graphene interaction 11 . We recently found that long alkanes are particularly suitable for enhancing the yield of exfoliation 19,20 .
Photochromic molecules, in particular azobenzene-based molecules, covalently linked to reduced graphene oxide 21 , physisorbed directly on graphene 22,23 or through pyrene anchoring groups 24 on the graphene surface, or even when graphene is adsorbed on an azobenzene self-assembled monolayer 25 , can be used to reversibly modulate the graphene's electronic properties. Moreover, it was shown that covalent functionalization of gold surface 26 or carbon nanotubes 27 with azobenzene molecules can be used as a route for fabrication of light-driven electronic switches. Nonetheless, the use of additional responsive functions provided by a suitably designed molecule to assist the LPE of graphite into functional inks is still unexplored. This approach could lead to the production of functional hybrid materials and nanocomposites in a one-pot process.
Here we combine photochromic systems and LPE graphene exploiting the properties of these materials. We focus on a commercial alkoxy-substituted azobenzene, i.e., 4-(decyloxy) azobenzene. The presence of a long alkoxy side chain is expected to enhance the molecular interaction with graphene 28 , and promote the exfoliation of graphite towards graphene. Here we demonstrate that the use of 4-(decyloxy)azobenzene has two major advantages: first, the amount of graphene dispersed in N-methyl-2-pyrrolidone (NMP) can be enhanced by exploiting the photoisomerization of 4-(decyloxy)azobenzene molecules during exfoliation. Second, the isomerization of 4-(decyloxy)azobenzene from trans to cis and vice versa, when physisorbed between adjacent graphene layers leads to a reversible modulation of the inter-flake distance on the sub-Ångström scale. This is reflected in a light response of the electrical properties of the hybrid material.

Results
Liquid-phase exfoliation. To test the ability of alkoxy-substituted azobenzenes to increase Y W (%), which is the yield of graphene exfoliation, defined as the ratio between the weight of dispersed graphitic material and that of the starting graphite flakes, as well as to exploit the photochromic nature of such molecules when interacting with graphene, vials containing NMP, graphite powder and azobenzene are exposed to tip sonication for 3 h (see Methods and Supplementary Note 1 for details). This experiment is carried out either in dark or under ultraviolet irradiation, using a portable ultraviolet lamp, at two temperatures: 20 and 40°C (see experimental set-up in Supplementary Fig. 1). Subsequently, the dispersions are allowed to settle for 15 min, then decanted and centrifuged for 1 h at 10,000 r.p.m. Control samples are also prepared, consisting of dispersions prepared in the absence of 4-(decyloxy)azobenzene both in dark and irradiated at 365 nm at either 20 or 40°C. To quantify the concentration of graphene after centrifugation, a mixture of graphene dispersion and 2-propanol (IPA) is first heated to 50°C for 30 min and then passed through polytetrafluoroethylene membrane filters. The remaining solvent and weakly interacting 4-(decyloxy) azobenzene molecules are washed out several times with diethyl ether and IPA. Measurements of the filtered mass are performed on a microbalance to infer the concentration of graphitic material in dispersion, needed to quantify Y W (%) (see Fig. 1b). The presence of adsorbed molecules may affect the mass measurements and Y W (%). Thus, the heating step is necessary to completely desorb the 4-(decyloxy)azobenzene molecules, as proven with X-ray photoelectron spectroscopy (Supplementary Fig. 2 and Supplementary Note 2). By analysing 20 independent experiments, we find that when the exfoliation is performed in the absence of 4-(decyloxy)azobenzene (control experiments), comparable Y W are observed when dispersions are kept at 20°C (0.64%) or heated at 40°C (0.62%). The irradiation at 365 nm does not affect Y W , neither at 20 nor at 40°C (0.63% in both cases).
LPE in the presence of alkoxy-substituted azobenzene acting as dispersion-stabilizing agent can undergo different mechanisms depending on the isomeric form of the photochromic molecule (Fig. 1c). In our case, the presence of 4-(decyloxy)azobenzene has no major influence on Y W of dispersions prepared in dark (0.71% and 0.72% at 20 and 40°C, respectively). As previously reported 11 , the use of small organic molecules, in particular those based on aromatic cores 11 and alkyl functionalization 19,20 , can promote graphite exfoliation because of their high affinity for the basal plane of graphene hindering interflake stacking. The relation between interaction energy and exfoliation yield is not straightforward. In some cases, molecules with lower adsorption energy on graphene give higher exfoliation yield, as demonstrated for graphite exfoliation 29 , as well as for exfoliation of a wide range of other layered materials 30 . This is because a critical step of the process is the adsorption of the molecule on the graphene surface and the displacement of the adsorbed solvent layer, overcoming an activation energy barrier that depends significantly on molecular structure and conformation. Therefore, even a minor increase of Y W (0.06-0.09%) on addition of a small amount of 4-(decyloxy)azobenzene (5 wt%) highlights the importance of these molecules during LPE.
In contrast to the experiments performed in dark, irradiation with ultraviolet light at 365 nm increases the concentration of exfoliated material. At 20°C we obtain B81 ± 5 mg ml À 1 , corresponding to B30% increase in Y W , when compared with reference samples (63 mg ml À 1 ). Yet, the largest increase in concentration is obtained by irradiating the dispersions heated at 40°C, reaching 110 ± 8 mg ml À 1 , which corresponds to a 75% increase in Y W . These results demonstrate that the addition of 4-(decyloxy)azobenzene during LPE leads to a significant increase in Y W only in samples containing 4-(decyloxy)azobenzene irradiated at 365 nm. Molecules rapidly switching conformation, such as azobenzenes on ultraviolet (UV) illumination, heating and sonication, can overcome the activation barriers to adsorption, acting as 'nano-spinners' that disrupt the adsorbed solvent layer more effectively than molecules having a static conformation. Therefore, the increase in concentration can be attributed to the motion resulting by the trans-to-cis photoisomerization. The Y W temperature dependence may be due to differences in molecule-graphene interactions at 20 and 40°C, as well as different kinetics of the trans-to-cis photoisomerization.
Trans-to-cis photoisomerization and cis-to-trans relaxation. The transition from the thermodynamically stable trans to the metastable cis form can be induced by irradiation with UV light, and takes place on the picosecond timescale 31,32 , whereas the cis-to-trans isomerization is typically triggered by irradiation with visible light or heat 33 . The latter can also be activated by the use of ultrasound-induced mechanical force 33 . Ref. 34 claimed that mechanoisomerization of azobenzene typically results in the cleavage of the N ¼ N azo bond. However, more recent theoretical calculations 35 and experiments 36 have challenged this view.
To gain a quantitative understanding of the photoisomerization of 4-(decyloxy)azobenzene, we perform a kinetic analysis of both trans-to-cis and cis-to-trans isomerization. The UV-visible spectrum of 4-(decyloxy)azobenzene dissolved in NMP (Fig. 2a, black line) shows an absorption maximum at 353 nm, arising from the p-p* transition 33 , and a peak at 442 nm, due to n-p* transitions 33 . The UV-visible spectrum of the above solution when irradiated at 365 nm shows a reduction of the p-p* transition band at 353 nm, and an increase in absorbance of the 442 nm (n-p* transition) peak, which is a characteristic feature of the trans-to-cis photoisomerization (Fig. 2a, red line; and Supplementary Fig. 3). The photostationary state can be reversibly converted back to the trans isomer on irradiation at 450 nm (see Methods for details). The yield of the conversion back to the thermodynamically stable trans-4-(decyloxy)azobenzene is higher than 95% (Fig. 2a, black dashed line).
To probe the cis-to-trans isomerization of the azobenzene molecules under experimental conditions corresponding to those used during the LPE process, i.e., at 20 and 40°C, in the presence of ultrasound-induced mechanical forces, a stock solution of 4-(decyloxy)azobenzene in NMP (16.2 mM) is prepared, and then exposed to 365 nm light and sonicated for 3 h (see Supplementary Note 1 for details). Every 30 min Trans 4-(decyloxy)azobenzene (except for a first measurement performed after 5 min) 0.2 ml of UV-irradiated solution is transferred (in the dark) into a quartz cuvette, diluted 10 times with NMP. The optical absorption spectrum is then recorded (shown in Supplementary Fig. 4). Finally, to ensure the stability of the azo bond during sonication, a solution sonicated for 3 h is exposed to white light and an absorption spectrum is recorded. A similar procedure is applied for blank experiments, i.e., in the absence of ultrasound-induced mechanical forces. Analysis of the absorption and proton nuclear magnetic resonance spectra ( Supplementary Fig. 5) acquired for different experimental conditions allows us to quantify the percentage of conversion from trans-to cis-4-(decyloxy)azobenzene, as well as the rate constants for trans-to-cis isomerization. Both nuclear magnetic resonance and high-performance liquid chromatography data (Supplementary Figs 5 and 6, and Supplementary Note 3) show that the decyloxy chains are not being chopped from the azobenzene core. The slow kinetics of the trans-to-cis isomerization can be explained by the fact that both spectroscopic experiments and LPE process are carried out under unusual isomerization conditions, including a low UV light power density (0.34 mW cm À 2 ). A different time (30 min) is needed for reaching a photostationary state, in the reference and sonicated samples, as displayed in Fig. 2b. Previous studies 36,37 on the trans-to-cis isomerization of azobenzene in the presence of UV light showed that the reaction follows first-order kinetics and the rate constant k can be written as follows: where A 0 , A t and A N are the absorbances before irradiation, at irradiation time t and after irradiation for a prolonged time (ca. 6 h). Applying equation (1), we find that the isomerization constant k trans-cis at 20°C for the 4-(decyloxy)azobenzene linked Trans-to-cis isomerization Cis-to-trans relaxation Trans-to-cis isomerization at Cis-to-trans relaxation at: Cis-to-trans mechano-relaxation at: Sonication assisted trans-to-cis isomerization at 20   to graphene is similar to k trans-cis at 40°C: 2.96 and 2.89 h À 1 , respectively. Significantly different k trans-cis kinetics are observed for the samples exposed to ultrasound-induced mechanical forces, i.e., 2.28 and 0.79 h À 1 for dispersions prepared at 20 and 40°C, respectively. To ensure the stability of the azo bond during sonication, the two samples are exposed to white light, and the absorption spectra recorded. The absorption at 365 nm is consistent with the trans-4-(decyloxy)azobenzene spectra in both cases ( Supplementary Figs 3 and 4), indicating that the molecules, in particular the azo bond, are stable during sonication at various temperatures.
Since the trans-to-cis isomerization and the LPE processes are performed under UV light, the possible cis-to-trans backconversion can be considered promoted by thermal or mechano-thermal relaxation, for experiments performed in the absence or presence of ultrasound-induced forces, respectively.
To probe both thermal and mechano-thermal isomerization, four solutions (16.2 mM) are prepared in NMP and the UV-visible spectra recorded. The solutions are then irradiated with UV light (365 nm) for 3 h to photoisomerize the 4-(decyloxy)azobenzene from its trans to cis configuration ( Supplementary Fig. 3). The solutions are then kept for 3 h at 20 and/or 40°C, in the presence and/or absence of ultrasoundinduced mechanical forces. Every 30 min, 0.2 ml of solution is transferred (in dark) onto a quartz cuvette and diluted 10 times with NMP. Absorption spectra are then recorded.
Owing to the lack of white light during the 3 h relaxation, 4-(decyloxy)azobenzene molecules do not convert fully from their cis-to trans-isomers. As shown in Fig. 2c, the percentage of cis-to-trans conversion in the blank samples (without sonication) at 20 and 40°C is B8.6% and 27.6%, respectively. In contrast, the ultrasound-induced conversion is much higher, i.e., 29.8% and 51.2% for samples kept at 20 and 40°C, respectively. Similar to the trans-to-cis isomerization, the cis-to-trans relaxation of 4-(decyloxy)azobenzene follows first-order kinetics, and k can be obtained from equation (1). k cis-trans for 4-(decyloxy)azobenzene kept at 20°C is lower than k cis-trans for 4-(decyloxy)azobenzene relaxed at 40°C, i.e., 0.03 and 0.11 h À 1 , respectively (Fig. 2d). Notably different k cis-trans are also observed for samples exposed to ultrasound-induced mechanical forces, i.e., 0.10 and 0.24 h À 1 for dispersions kept at 20 and 40°C, respectively.
These results indicate that the trans-to-cis isomerization of 4-(decyloxy)azobenzene strongly depends on the experimental conditions. In particular, it is hindered by ultrasound-induced mechanical forces, which most likely cause the occurrence of the competing cis-to-trans isomerization. Furthermore, comparison of k trans-cis and k cis-trans at 20 and 40°C reveals that, at higher temperature, the process is more dynamic. As a result, the higher concentration of flakes (110 mg ml À 1 ) obtained at 40°C can be attributed to the dynamic conformational trans-to-cis and cis-totrans changes of 4-(decyloxy)azobenzene molecules triggered simultaneously by the competing effect activated by UV light, temperature and ultrasounds.
Analysis of dispersions. To fully characterize the exfoliated flakes, both qualitative and quantitative information are required. While quantitative insights can be assessed by providing the Y W , a qualitative analysis must give more relevant details, such as the percentage of single-layer graphene (SLG) and multi-layer graphene (MLG) flakes, the lateral size of the flakes and the presence/absence of defects. The number of graphene layers (N) can be quantified using high resolution-transmission electron microscopy (HR-TEM) 3,5 and Raman spectroscopy 17,18 . Together with the information coming from electron diffraction patterns, in HR-TEM N can be directly counted by analysing the folded edges 38 . We focus on the samples prepared by LPE at 40°C in the presence of azobenzene, while irradiating with UV light (highest increase in Y W ), and compare them with those exfoliated in NMP at 40°C under UV irradiation (in the absence of azobenzene). The statistical analyses of N and the lateral flake size are reported in Supplementary Fig. 7a,b, respectively (Supplementary Note 4). In terms of N, minor differences are observed between the reference sample and flakes exfoliated in the presence of azobenzene molecules, where the percentage of SLG amounts to 20% and 14%, respectively. The lateral size of the flakes is not affected by the presence of azobenzene molecules during LPE (Supplementary Fig. 7b). Representative HR-TEM micrographs are reported in Supplementary Figs 8, 9 and 10.
The Raman spectrum of graphite and MLG consists of two fundamentally different sets of peaks. Those, such as D, G and 2D, present also in SLG, due to in-plane vibrations [39][40][41] , and others such as the shear (C) modes 42 and the layer-breathing modes (LBM) 40,43 , due to the relative motions of the planes themselves, are either perpendicular or parallel to their normal. The G peak corresponds to the high-frequency E 2g phonon at C. The D peak is due to the breathing modes of six-atom rings and requires a defect for its activation 41,44,45 . It comes from transverse optical phonons around the Brillouin Zone edge K (refs 41,45), is active by double resonance 46 and is strongly dispersive with excitation energy due to a Kohn anomaly at K (ref. 47). Double resonance can also happen as intra-valley process, i.e., connecting two points belonging to the same cone around K or K 0 . This gives the so-called D 0 peak. The 2D peak is the D peak overtone while the 2D 0 peak is the D 0 overtone. Since 2D and 2D 0 originate from a process where momentum conservation is satisfied by two phonons with opposite wave vectors, no defects are required for their activation, and are thus always present 38 . The 2D peak is composed of a single Lorentzian in SLG, whereas it splits into several components as N increases, reflecting the evolution of the electronic band structure 38 .
In disordered carbons, the G peak position, Pos(G), increases as the excitation wavelength decreases from infrared to ultraviolet 44 . Therefore, the dispersion of the G peak, Disp(G), i.e. the rate of change of Pos(G) with the laser excitation wavelength, increases with disorder 44 . Similar to Disp(G), also the full width at half maximum of the G peak, FWHM(G), always increases with disorder 44 . The analysis of the intensity ratio of the D and G peaks, I(D)/I(G), combined with that of FWHM(G) and Disp(G) allows us to discriminate between disorder localized at the edges and disorder in the bulk of the samples. In the latter case, a higher I(D)/I(G) would correspond to higher FWHM(G) and Disp(G).
Raman measurements are done on the same set of samples to characterize the quality of the graphitic material. Figure 3a,b show a small correlation between these parameters. This implies that the D peak is mostly due to edges, as well as the presence of some defects in the samples. FWHM(2D) (Fig. 3d) is larger with respect to that of graphene flakes produced by micromechanical cleavage, but still has a single Lorentzian line shape. This is consistent with the presence of MLGs composed of folded or juxtaposed SLGs, but still electronically decoupled 17 . Figure 4a plots the Raman spectra of a graphene-azobenzene film after visible and UV light illumination, acquired at 514.5 nm with power o250 mW to avoid any Raman laser-induced isomerization of azobenzene. None of the graphene Raman parameters show significant change on UV or visible light irradiation indicating negligible change in the defects and doping.
The trans-to-cis isomerization can also be monitored by Raman spectroscopy 24,48 . Figure 4b plots the Raman spectrum of azobenzene after both visible and UV light illumination. The spectrum is composed of five main peaks, labelled C1-C5 (ref. 48). These can also be seen on the spectra of the grapheneazobenzene hybrids (Fig. 4c). The intensity ratio of peaks C3 and C4 can be used to monitor the isomerization of azobenzene as described in ref. 48. On illumination with visible light, the photochromic molecules fall into the trans configuration and I(C3)/I(C4) approaches B1. When the sample is illuminated with UV light, azobenzene molecules undergo trans-to-cis isomerization and I(C3)/I(C4) decreases markedly. The effect is fully reversible and the trans form is recovered by a further visible light exposure as shown in Fig. 4d.
X-ray analysis. Powder X-ray diffraction is further used to characterize the structure of the graphene-azobenzene hybrid powder and to compare it with the control samples (Supplementary Note 5). Similarly to graphite, the powder prepared from LPE has a sharp peak at 26.7°, corresponding to an interlayer spacing of B0.33 nm (Supplementary Fig. 11a). The X-ray diffraction patterns of powders prepared from graphene/ trans-azobenzene and graphene/cis-azobenzene show new peaks at 2y ¼ 12.3°(d spacing B0.72 nm) and 9.9°(d spacing B0.89 nm), respectively ( Supplementary Fig. 11b,c). This confirms that trans and/or cis-azobenzenes can sandwich between graphene layers, thus increasing the overall spacing.
Electrical characterization. To probe the electrical properties of the graphene-azobenzene hybrid and, in particular, to explore the potential light-responsive nature of the material, we drop-cast a B100-nm-thick graphene-azobenzene film on a n þ þ Si/SiO 2 substrates exposing pre-patterned interdigitated gold electrodes (channel length ¼ 10 mm; Fig. 6a,b).
Typical I-V curves exhibit a nonlinear resistive behaviour, as shown in Fig. 5c. The conductivity of the graphene-azobenzene hybrid is lower than that measured in control devices, i.e.,  comprising as active layer graphene exfoliated in NMP in the absence of azobenzene ( Supplementary Fig. 12). Figure 6d shows the current modulation in a two-terminal device with grapheneazobenzene on cycles of UV and visible light irradiation, for 1 V applied between the two electrodes. One cycle consists of 10-s UV light (365 nm), 1-min rest in dark followed by 40-s visible light (450 nm) irradiation. Under UV exposure (magenta background) a current decrease is detected, which reversibly increases to about the initial value under visible light irradiation (violet background) over various cycles. No fatigue is observed during six cycles of photoisomerization. Conversely, the control devices show no response to light ( Supplementary Fig. 12). Noteworthy, in some cases the increase of current in the hybrid films can be the result of p-electron interactions between the molecules and graphene 49 ; however, such phenomenon, known as increased photoconductivity, is irreversible and can only be modulated by varying the ratio between graphene and ad hoc molecules.
In the azobenzene-graphene film ( Supplementary Figs 13  and 14), the molecules are physisorbed on top and in between the graphene sheets. Under UV irradiation, the azobenzene molecules undergo a conformational change from the less-bulky trans isomer to the bulkier cis isomer. Such a process is therefore accompanied by an increase in the few-layer graphene (FLG) intersheet distance, thereby hindering the charge transport via hopping between the sheets, resulting in lower conductivity of the hybrid film. The conductivity of the hybrid material can be fully restored by irradiating with visible light, causing the cis-to-trans photoisomerization.
To gain greater insight into the conduction mechanism and the effect of isomerization, we then perform Kelvin probe measurements (Supplementary Note 6). We find that the work function is reversibly photo-modulated over various cycles ( Supplementary  Fig. 15), providing further evidence for the isomerization of the azobenzene molecules. This difference in work function has a direct effect on charge injection from the electrodes to the graphene/azobenzene conductive material, which is less favourable in the presence of the cis isomer. Light modulation can be applied multiple times to the systems, with no evidence of decrease in performance. We assign this to graphene's shielding of the azobenzene molecules from the outer environment. An ideal case consists of stacks of alternating graphene and azobenzenes, the latter having a thickness, which can be photo-modulated (Fig. 5e). In reality, the flakes (with average lateral sizes of 200 nm) form aggregates possessing a poor degree of order. Within these aggregates the azobenzenes are intercalated in between adjacent layers of graphene. The control samples show no change in work function on UV and visible light cycling (Supplementary Fig. 15).

Molecular modelling simulations.
To gain a better understanding of the interactions of the azobenzene molecules with graphene, we perform molecular dynamics simulations (see Supplementary Note 7 for details). All calculations are done using the Groningen Machine for Chemical Simulations (GROMACS) 50 package and a modified version of the all-atom Optimized Potentials for Liquid Simulations (OPLSAA) force field developed for azobenzenes 51,52 . To assess the relative adsorption affinity of the trans versus cis conformations, we first compute the interaction energy for a single azobenzene molecule by running a 1-ns simulation in vacuum. The graphene layer is taken as an infinite rigid body in these calculations. The computed interaction energies of À 130.1 ± 7.5 and À 115.7±8.0 kJ mol À 1 for trans-and cis-azobenzene, respectively, are consistent with the expected stronger adsorption of trans-azobenzene, resulting from its planar geometry and the concomitant maximized p-p interactions between the phenyl  groups and graphene. The average adsorption distance for the trans molecule is B0.07 nm shorter than cis (0.34 versus 0.41 nm). Figure 6a,b plots top and side views of the adsorbed cis and trans molecules from the last snapshots along the molecular dynamics trajectories.
It should be noted that the conformation giving the higher exfoliation yield is the cis one, with lower interaction energy with graphene (14.4 kJ mol À 1 difference). This counterintuitive finding highlights the importance of kinetic factors on exfoliation yield rather than thermodynamic factors, in good agreement with previous results obtained with rigid molecules 29,30 . The 70% increase in exfoliation yield could be due to the rapid, dynamic change of conformation between cis and trans that the molecules undergo during the exfoliation, which is more rapid at high temperature and under sonication (Fig. 2).
We then study the supramolecular organization of trans-and cis-azobenzenes on SLG in vacuum. In the absence of experimental structural data, we consider highly regular self-assembled monolayers (Supplementary Fig. 16) and test their stability by performing simulations of the assemblies at room temperature. The regular assemblies are unstable at 300 K and disassemble after few tens of picoseconds, suggesting that the azobenzene molecules form disordered layers at the SLG surface. To investigate the formation and structure of such disordered layers, we use the following protocol. Azobenzene molecules are introduced one by one in the simulation box at a distance of 1.8 nm above the plane of the infinite SLG with a random orientation. The initial atomic velocities are assigned so that the resulting vectors point towards the SLG, thereby prompting physical adsorption on the surface. Room-temperature molecular dynamics simulations of a few hundreds of picoseconds are then performed to explore the configurational space after each molecular deposition, and the last saved configuration is used as input for the landing of the subsequent molecule. The scheme is iterated multiple times until the desired surface coverage is reached. A 5-ns simulation is then performed on the final structure to equilibrate the monolayer at room temperature. A more detailed description of the methodology is provided in Supplementary Note 8. Full coverage of the graphene periodic layer of 115.8 nm 2 is reached when 64 (72) molecules of trans (cis)-azobenzene are adsorbed, which translates into a surface contact area on graphene of B1.8 nm 2 for the trans and B1.6 nm 2 for the cis conformation. Figure 6c,d plots a top view of the azobenzene monolayers obtained for the two conformers in the case of full coverage. Supplementary Figs 17 and 18 show the formation of those monolayers as a function of the number of molecules adsorbed on SLG.
To validate the hypothesis that the insertion of a trans-/cis-monolayer sandwiched between two SLGs can result in different interlayer spacing according to the ideal model, a second SLG is placed on top of the azobenzene monolayers prepared using the deposition protocol described above. The 1-ns simulations are then done at room temperature. The results show that cis-azobenzenes maintain the SLG at a distance B0.85 nm, while the layer separation decreases to B0.73 nm for trans (Fig. 6e,f), in excellent agreement with the X-ray data. This supports the view that photoisomerization of the molecules from trans to cis can favour the exfoliation process. To investigate the effect of this switchable separation on charge transport, we perform quantum chemical calculations (at the intermediate neglect of differential overlap level 53 ) on two square-shaped nanographenes (B20 nm 2 ) interacting either through-space or in the presence of the sandwiched (trans or cis) azobenzenes ( Supplementary Fig. 19). In the first case (no sandwiched layer), the electronic couplings mediating positive and negative charge carrier tunnelling from one nanographene to the other, as obtained from the diabatization scheme reported in ref. 54, decrease by a factor B100 when the interlayer distance is increased from 0.73 (corresponding to the azobenzenes in trans conformation) to 0.85 nm (cis conformation). In both cases, these interactions are very small due to the fast fall-off decay of the nanographene wavefunction overlap. Interestingly, the corresponding couplings calculated in the presence of the sandwiched azobenzenes are several orders of magnitude larger and show a less pronounced change (by a factor B10) on photoisomerization, suggesting that the loss in electrical transmission associated with the larger inter-graphene distances is at least partly compensated by conformational-dependent superexchange effects involving the molecular orbitals of the azobenzenes. Our experimental current modulation on photoisomerization of azobenzenes (Fig. 5d) is much smaller than the calculated one (factor B100). This can be understood considering that the simulations are run for two SLG intercalated by a tightly packed, physisorbed, self-assembled layer of azobenzene, whereas the experiments are done for hybrid films with a distribution of layer thicknesses ( Supplementary Fig. 7a) separated by a small quantity of azobenzene (below 5% of the area of graphene).
As aforementioned the ideal model does not fully describe our real system that consists of poorly ordered aggregates comprising flakes intercalated with azobenzenes, and charges hopping between adjacent graphenes. The distance between the adjacent graphene sheets can be varied on photoisomerization of the azobenzenes.

Discussion
We demonstrated that alkoxy-substituted photochromic molecules can act as photo-addressable surfactant and as dispersion-stabilizing agents to enhance the yield of exfoliation in an upscalable molecule-assisted LPE-based method. The simultaneous use of UV light, promoting the trans-to-cis isomerization, as well as thermal annealing at 40°C and mechanical forces generated by sonication, both favouring cis-to-trans isomerization, promotes the exfoliation in liquid media. The most effective exfoliation is obtained with azobenzene molecules irradiated with UV light in NMP at 40°C, with a concentration of exfoliated graphene of 110 mg ml À 1 . This corresponds to an B80% increase in exfoliation yield when compared with pure NMP (63 mg ml À 1 ). By depositing the hybrid film onto Au pre-patterned SiO 2 substrates, lightresponsive thin hybrid films, formed in a one-step co-deposition process, can be realized, whose conductivity can be reversibly modulated by the trans-cis photoisomerization of the azobenzenes. By combining this approach with cost-effective techniques, such as ink-jet printing, more complex responsive device designs and architectures may be realized. This paves the way to future applications such as optically controllable memory switch elements for light-assisted programming and photosensors. Liquid-phase exfoliation. Graphene dispersions are prepared by adding graphite powder (100 mg) to NMP (10 ml), and tip-sonication (Labsonic M, Platinum tip, diameter 2 mm, sound rating density 300 W cm À 2 ) for 3 h either in dark or under UV light irradiation, using a portable laboratory ultraviolet lamp (8 W, 365 nm, 0.34 mW cm À 2 , Herolab GmbH), in glass vials (Pyrex), in the presence of azobenzene molecules (5.5 mg). The dispersions are then allowed to settle for 15 min, then decanted and centrifuged (Eppendorf, centrifuge 5804) for 1 hour at 10,000 r.p.m. From the centrifuged dispersions 70 vol% are pipetted off the top for characterization and film deposition. To quantify the concentration of flakes after centrifugation, the dispersions are passed through polytetrafluoroethylene membrane filters (pore size 100 nm). Measurements of the filtered mass are performed with a microbalance (Sartorius MSA2.75) Device fabrication. n þ þ Si substrates with a thermally grown SiO 2 layer (230 nm) and pre-patterned interdigitated gold source and drain electrodes (IPMS Fraunhofer) with different channel length (L ¼ 2.5, 5, 10 and 20 mm) and constant channel width (W ¼ 10 mm) are used. The substrates are cleaned before device fabrication in an ultrasonic bath (FB 15047, Fisher Scientific) of acetone and isopropanol, 15 min in each solvent, and treated 5 min ( þ 30 min incubation) with an UV surface decontamination system (PSD-UV, Novascan) to improve wetting of the solvent. The dispersions (50 ml) are drop-cast on clean substrates and dried for 48 h in vacuum at 30°C.

Methods
Characterization. The electrical characterization is carried out in inert atmosphere (glovebox) with an electrometer (Keithley 2636A) interfaced with LabTracer software. For the light-induced switch, a Polychrome V monochromator (Till Photonics) is used as ultraviolet light (350 nm, 5.64 mW cm À 2 ) and visible light source (450 nm, 4.83 mW cm À 2 ). For the Raman experiments, graphene dispersions are drop-cast onto pre-cleaned n þ þ Si substrates with a thermally grown SiO 2 layer (300 nm) and dried for 48 h. Raman spectra are collected with a Renishaw InVia spectrometer at 457, 514.5 and 633 nm. The excitation power is kept below 1 mW to avoid the effects of local heating. The scattered light is collected with a Â 100 objective. HR-TEM is done in a FEI Tecnai F20 equipped with a Schottky emitter and operated at 120 keV primary beam energy. Scanning electron microscopy images are recorded with a Quanta FEG 250 from FEI.
The thickness of the hybrid films is determined with an Alpha-Step IQ Surface Profiler from KLA Tencor. X-ray photoelectron spectroscopy analyses are carried out on a Thermo Scientific K-Alpha X-ray photoelectron spectrometer, with a basic chamber pressure B10 À 8 mbar and an Al anode as the X-ray source (X-ray radiation of 1,486 eV). Spot sizes of 400 mm, pass energies of 200 eV for survey scans and 50 eV for high-resolution scans are used. A volume of 150 ml of the dispersions are spin-coated on Au substrate for 1 min at 1,000 r.p.m. and substrates are annealed for 1 day at 100°C in a oven under vacuum. Samples for X-ray diffraction are prepared by precipitating graphene and/or graphene/azobenzene by adding water:ethanol (1:1, vol:vol) into NMP. The collected precipitate is dried under vacuum for 24 h. The powder X-ray diffraction patterns are obtained using a Bruker AXS D2 Phaser (LYNXEYE detector) with Ni-filtrated Cu-Ka radiation (l ¼ 1.5406 Å) with a 1-mm air-scattering slit and 0.1-mm equatorial slit. Samples are deposited on the surface of a single crystal Si wafer (cut of (911)). X-ray diffraction patterns are collected with 0.016°steps and 10 s per step increments from 8°to 80°.