Exploiting the Surface Properties of Graphene for Polymorph Selectivity

Producing crystals of the desired form (polymorph) is currently a challenge as nucleation is yet to be fully understood. Templated crystallization is an efficient approach to achieve polymorph selectivity; however, it is still unclear how to design the template to achieve selective crystallization of specific polymorphs. More insights into the nanoscale interactions happening during nucleation are needed. In this work, we investigate crystallization of glycine using graphene, with different surface chemistry, as a template. We show that graphene induces the preferential crystallization of the metastable α-polymorph compared to the unstable β-form at the contact region of an evaporating droplet. Computer modeling indicates the presence of a small amount of oxidized moieties on graphene to be responsible for the increased stabilization of the α-form. In conclusion, our work shows that graphene could become an attractive material for polymorph selectivity and screening by exploiting its tunable surface chemistry.


S1. Crystallisation set-up
shows the experimental set-up used for the crystallisation of glycine. Crystallisations were carried out in the following way: a substrate was placed in the centre of a plastic box, a 2 L droplet of glycine solution was deposited on the substrate and then it was covered by a clean glass slide, without any contact between the droplet and the cover. The glass slide allowed a 1 mm gap to be present at the front of the sample box. All crystallisations were carried out in an incubator kept at 21 °C.

Figure S1
Image of the crystallisation set-up.

S2. Raman spectrum of glycine
Glycine crystallises in three distinct polymorphic forms at ambient conditions, denoted as , , and . The relative stabilities of these polymorphs are:  >  > . 1,2 The most commonly obtained metastable -form crystallises from aqueous solution; 3 the stable -form can be obtained from acidic or basic solutions; 4,5 while the unstable -form crystallises from mixtures of ethanol or methanol with aqueous glycine solutions and it readily transforms to the -form upon contact with humid air. 6 It has been shown that all three polymorphs of glycine crystallise simultaneously upon evaporation of aqueous solution microdroplets, which makes polymorphs of glycine classified as concomitant polymorphs. 7 Figure S2.1 shows the distinct Raman spectra of the CH region (2900-3050 cm -1 ) of the three polymorphs of glycine. These peaks represent the symmetric (lower shift) and asymmetric (higher shift) stretches of the C-H bonds. The positions of these modes are distinct for each polymorph, which were found to be at 2972 cm -1 and 3007 cm -1 for the -form, at 2953 cm -1 and 3008 cm -1 for the -form and at 2962 cm -1 and 3000 cm -1 for the -form. Only and forms were considered in our study as -form was not detected. A Raman map was obtained by scanning the whole area of a crystallised glycine droplet to ascertain the overall distribution of the polymorphs within the product crystal, as shown in Figure S2.2. This map shows the intensity ratio for the symmetric C-H stretch of the -form (2953 cm -1 ) to -form (2972 cm -1 ): a low ratio corresponds to -form, while a high ratio corresponds to -form. Raman spectroscopy confirms that the bulk of the droplet exclusively consists of the metastable -form, while the unstable -form preferentially nucleates at the edge of the droplet, as expected from literature. 8 To be more quantitative, ~10% of the crystals grown at the contact region are and ~90% are the -form. This outcome has been attributed to the higher supersaturation rate generated at the droplet contact region, which should allow formation of a less stable polymorph. 8 Since Raman mapping on such a large scale is extremely time consuming, the analysis of the polymorphic outcome in this work has been performed by taking individual measurements of each crystal at the droplet contact region.

S3. Graphene dispersion characterisation
After the electrochemical exfoliation process (see Methods), the graphene dispersions were characterised with several techniques. A detailed Raman characterisation of graphene produced by ECE has been recently provided by our group, 10 hence we only provide a brief overview of the results in this section. Figure   S3.1b compares the Raman spectra taken from isolated flakes of ECE(NH4) and ECE(K) samples. The Raman spectra show the characteristic peaks of solution processed graphene: 11 the G , D, D' and 2D peaks, which are observed at ~1580 cm -1 , ~1350 cm -1 , ~1620 cm -1 and ~2690 cm -1 , respectively, can be seen for both samples. The D peak, a defect activated feature, 12,13 is also observed in graphene produced by liquid phase exfoliation 11 as these flakes have a size smaller than the laser spot, so the edges act as defects. 14 However in our case, the average size of the flakes is a few micrometers, 15 so the D peak is likely to be activated by structural defects (e.g. formation of C-O bonds), which are known to form during the intercalation process, resulting in the partial oxidation of ECE graphene. 15,16 The Raman spectrum of defective graphene can be described with a phenomenological threestage amorphization trajectory. 12 In stage 1, starting from pristine graphene, the Raman spectrum evolves as follows: the D peak appears and the intensity ratio between the D and G peaks (ID/IG) increases; the D' appears; all the peaks broaden and G and D' begin to overlap.
In this stage, ID/IG can be used to estimate the amount of defects, 12,17 while ID/ID' can be used to distinguish between different type of defects. 13 Table S1 shows the results of the fits. From the full width at half maximum (FWHM) of the peaks, we can conclude that the samples are defective and belong to Stage 2, so defects quantification is not possible. However, all the Raman peaks are broader for the ECE(NH4) sample compared to the ECE(K) sample: in particular the FWHM of the G peak is ~35 cm -1 for ECE(K) and ~65 cm -1 for the ECE(NH4) (Table S1). Furthermore, the G and D' peaks are still distinguishable for ECE(K), whereas they overlap considerably for the ECE(NH4) sample. These observations indicate that both samples are highly defective, but ECE(NH4) graphene contains, on average, a higher concentration of defects than the ECE(K) sample. 12 The defects are likely to be oxygen-containing functional groups which are formed during the ECE process. 16   Figure S5 shows an example of a spherical cap. The surface areas of glycine droplets on different substrates were calculated by applying the obtained contact angles (see Section 5.2) as cap angles () to the spherical cap model. The fixed volume (V) of 2 L and the cap angles were applied to Equation 1 in order to obtain the radius (r) values: 18

S4. Surface area and volume ratio calculations
Following, the surface areas (SA) were obtained using Equation 2: 18 Finally, the surface area values were divided by a fixed volume of 2 L in order to obtain Surface area and Volume ratios.

S5.3 Morphology
The observation noted in the main text regarding the distinct morphologies of the crystals can be rationalised by considering the solubility curve of glycine and relative growth and nucleation rates, kG and J, respectively. Both kG and J are dependent on the driving force for crystallization, which is typically expressed as the ratio of the actual liquid concentration to the equilibrium/solubility concentration. The solution begins undersaturated and via evaporation of the solvent crosses the supersaturation threshold into glycine's metastable zone (MSZ). 19 In this region, crystal growth dominates over nucleation (a low J/kG ratio) so if the system were to crystallise, larger crystals would form. If the solvent continues to evaporate, the system will move past the MSZ into high supersaturation, where nucleation events dominate over crystal growth (a high J/kG ratio). 20 It must be noted here that the crystals obtained from a pure water system (example shown in Figure S2.2) were always compact, thus implying that the addition of IPA, an antisolvent, causes a widening of the MSZ width, as already observed. 21,22 The effect of graphene on the crystal morphology can be seen in Figure 2f  The dark blue spots that can be seen are the graphene deposits. Scale bars ≈ 10 m.  The broad peak at ~900 cm -1 is from the silicon substrate.

S5.5 Use of GO
The concentration of GO used for this set of experiments were made comparable to those of the ECE graphene samples, ranging from 5 x 10 -2 mg mL -1 to 5 x 10 -4 mg mL -1 . All other experimental parameters were kept identical to the ECE graphene additive-templated crystallisations. The resulting crystals are vastly different, in size and morphology, to those obtained from the experiments with ECE graphene. This is attributed to the different geometries of the droplets between ECE graphene, which is hydrophobic, and GO, which is hydrophilic. Polymorph analysis was attempted by Raman spectroscopy, but the signal from GO made the determination of the glycine's polymorph impossible. The coverages were calculated by subtracting the number of the pixels with no G peak signal from the total number of pixels and then the sum was divided by the total number of pixels and multiplied by one hundred. Table S2 shows the graphene coverage of the Si/SiO2 substrates for both the ECE(K) and ECE(NH4) samples.        Figure S6.6 shows the kinetic impact of graphene substrates on the crystallisation of glycine.

S6.1.4 Kinetic Effects
As stated in the main text, the induction times do not follow the expected behaviour if the times were dictated by the geometry of the droplets. The implication from the geometry results of Figure 3a is that droplets should evaporate slower on the graphene substrates than on the bare silicon substrates, thus a longer induction time should be seen. 29 However, the graphene substrates had a promoting effect on the induction times, which furthers the hypothesis of Hbonding between graphene and glycine. It is unclear at this time why there is a reduction in the induction times for these substrate-templated experiments whilst there was an apparent increase in the times for the additive-templated ones, but it is likely due to the intricacies of the different solvent systems used. It is also important to remember that the induction times are all qualitative.

S6.2.1 Coverage Analysis
To investigate the coverage of CVD graphene, we performed Raman mapping of a large area  A representative Raman spectrum of CVD Gr on Cu can be seen in Figure S6.8, showing the characteristic G and 2D peaks of CVD Gr. 30 The high crystalline quality of the film is demonstrated by the absence of the D peak and the sharp G and 2D peaks.

S7. Computer modelling
We computed the interaction of surfaces of and -glycine crystals modelled by slabs of increasing thickness, namely with 1, 2, 3 and 4 molecular layers in vacuo and interacting with pristine graphene and oxidised graphene. The oxidised graphene model had 12.5 % OH groups and was built by adding 30 OH groups in a 240 C atoms graphene surface of 2.6 nm x 2.5 nm in a highly regular fashion, Figure S7 Table S4.