Supplementary Information for Evolution of Cooperativity in the Spin Transition of an Iron ( II ) Complex on a Graphite Surface

Cooperative effects determine the spin-state bistability of spin-crossover molecules (SCMs). Herein, the ultimate scale limit at which cooperative spin switching becomes effective is investigated in a complex [Fe(H2B(pz)2)2(bipy)] deposited on a highly oriented pyrolytic graphite surface, using x-ray absorption spectroscopy. This system exhibits a complete thermal- and light-induced spin transition at thicknesses ranging from submonolayers to multilayers. On increasing the coverage from 0.35(4) to 10(1) monolayers, the width of the temperature-induced spin transition curve narrows significantly, evidencing the buildup of cooperative effects. While the molecules at the submonolayers exhibit an apparent anticooperative behavior, the multilayers starting from a double-layer exhibit a distinctly cooperative spin switching, with a free-molecule-like behavior indicated at around a monolayer. These observations will serve as useful guidelines in designing SCM-based devices.

(a) Change in γ HS during the heating (black dots) and cooling (orange dots) cycles between 5 and 96 K of 0.69(8) ML of Fe(bpz)-bipy on HOPG (solid lines are a guide to the eyes). The sample has been first heated from 5 to 96 K and then cooled back to 5 K at the rate of 4 K·min −1 under continuous illumination by a green LED (light emitting diode, λ = 520 nm) of the same intensity as used for the experiment shown in Figure 3b of the main text. The Fe L 3 spectra recorded during the heating and cooling cycles corresponding to the black and orange dots in (a) are shown in (b) and (c), respectively. Below 40 K, the spin-state is dominated by the lightinduced metastable HS state; between 40 and 60 K, there is a rapid interconversion from the HS to the LS state during the heating cycle and from the LS to the HS states during the cooling cycle. For temperatures > 60 K, the LS state becomes dominant due to the thermally activated back-switching from the HS state to the LS ground state. There is a residual HS fraction of about 5% in the thermally-activated regime, as can be directly inferred from the presence of a "bump" at the photon energy of 708.1 eV in (b) and (c) (deep-magenta-coloured spectra). It is due to the soft-x-ray-induced excited spin state trapping (SOXIESST).
m e a n s q u a r e d d e v .

Supplementary Notes 1: Thickness estimation of the molecular layer
The frequency shift of a quartz micro balance integrated into the evaporator was used as a relative measure of the thickness of the molecular layer. The absolute thickness was estimated by comparing the integrated Fe L 3 XAS intensity at low molecular coverage to that of an Fe octaethyl-porphyrin (Cl)/Cu(001) reference sample which had been measured both with STM and XAS [5]. The total electron yield for an adsorbate layer (A) of thickness ∆x on a semi-infinite substrate (C) and a gold grid reference (G) is given by [6,7]: assuming that the attenuation length of the secondary electrons λ e is material-independent and much smaller than that of the x rays (λ e 1/µ). After normalizing to the gold grid, the signal is normalized to the pre-edge absorption coefficient of the adsorbate. Assuming constant absorption of the substrate and grid in the relevant energy range (µ C = µ pre C , µ G = µ pre G ) yields: In the low-coverage regime this signal is proportional to the ratio of the resonance intensity of the adsorbate and the absorption coefficient of the substrate after subtracting unity. To compare the adsorbate signal on different substrates, the ratio of their absorption coefficients must be known. On metals the efficiency of electron extraction from the substrate changes with molecular coverage due to the change in work function. The XA intensity ratio between HOPG and Cu(001) substrates in the pre-edge region of the Fe L 3 edge has been determined by XA measurements under identical conditions with normalization to the gold grid. For clean HOPG and Cu(001) substrates the intensity ratio is 0.84(8) and decreases to 0.65(7) for a coverage of one monolayer. The reference sample has an areal density of 0.14 Fe ions/nm 2 . Following Ref.
3 and 8, an areal density of 0.82 Fe ions/nm 2 is assumed for 1 ML. Comparing the integrated Fe L 3 intensity of the two samples with lowest coverage, we obtain a factor of 205(25) Hz/ML that relates the frequency shift to the number of monolayers.
To analyze the thickness dependence of the absorption signal, the pre-edge signal can be subtracted after normalizing to the gold grid. Using the same assumptions as above yields: The best fit of this relationship to the experimental data is shown in Supplementary Figure 8. A comparably high value of λ e = 1780(230) Hz = 8.7(1.6) ML for the attenuation length of the electron yield is obtained. This may be attributed to a less dense packing of the atoms in the molecular layer and a lower degree of electronic conjugation compared to metals or graphite.