Bottom-up formation of robust gold carbide

A new phenomenon of structural reorganization is discovered and characterized for a gold-carbon system by in-situ atomic-resolution imaging at temperatures up to 1300 K. Here, a graphene sheet serves in three ways, as a quasi transparent substrate for aberration-corrected high-resolution transmission electron microscopy, as an in-situ heater, and as carbon supplier. The sheet has been decorated with gold nanoislands beforehand. During electron irradiation at 80 kV and at elevated temperatures, the accumulation of gold atoms has been observed on defective graphene sites or edges as well as at the facets of gold nanocrystals. Both resulted in clustering, forming unusual crystalline structures. Their lattice parameters and surface termination differ significantly from standard gold nanocrystals. The experimental data, supported by electron energy loss spectroscopy and density-functional theory calculations, suggests that isolated gold and carbon atoms form – under conditions of heat and electron irradiation – a novel type of compound crystal, Au-C in zincblende structure. The novel material is metastable, but surprisingly robust, even under annealing condition.


Growth on facets, on edges and crystallization:
The growth of AuC crystals has been observed for every possible sort of nucleation site, i.e. on the facets of gold particles, on the edges of defective or holey graphene and even on the boundaries of fullerene agglomerations (Figures S1 to S5). Defective two-- and three--dimensional carbon structures formed by heat and electron irradiation provide active sources of atomic carbon under electron irradiation. Some sequences reveal, that the growing AuC structures do not only collect diffusing gold atoms but seemingly even `eat' fullerenes (sequence M1b with the corresponding snapshots in

Thermal stability of the AuC particles:
The AuC structures have been observed to exhibit a very large thermal stability as demonstrated by Figure S6 and the sequence M4 (snapshot in Figure S7).

Figure S6.
A gold particle larger than 10 nm (indicated by an arrow) starts to evaporate. This indicates a temperature of at least 1300 K. Interestingly, the somewhat smaller AuC cubes still move around but do not change their size and even preserve their crystalline structure. This image series was taken from a sequence lasting 60 seconds. Figure S7. Snapshot of sequence M4: The temperature has further increased via increasing the electrical current through the multilayer graphene sheet. Finally, also the AuC starts to decompose. The frame rate is 10 fps.
We performed local electron energy loss spectroscopy (EELS) to clarify the presence or absence of a possible second elements in our AuC structures. Figure S8 shows the AuC crystal that has been analyzed. Figure S8. Particle that has been analyzed via EELS under a beam diameter of approximately 5nm.

Image simulations
We simulated TEM images for the AuC crystals in order to compare the most reasonable crystal structures (ZnS versus NaCl). Considering noise and the underlying graphene substrate, light elements like carbon embedded in the gold fcc lattice are practically invisible for both structures.
Only in case of heavier elements like silicon, we would expect to reveal the lack of a fourfold symmetry in the simulated (100) projection of ZnS ( Figure S9). The images are obtained via the multi--slice program MUSLI [1].) Figure  S9. Simulated TEM images in order to compare the most reasonable crystal structures. Image simulation conditions: 80 kV, C S =0.02 mm, df=--10 nm.

Additional data obtained by the DFT calculations:
The last discussion in the main article deals with the comparison of possibly formed gold compounds.
In one regard we compared different surface terminations for an Au--C compound with sodium chloride (NaCl) and zinc blende (ZnS) structure. Here, we exemplify shortly the data obtained by our density functional theory (DFT) calculations: Figure  S10 shows the formation energy of gold carbide in dependence of the number of atomic planes. Here, we have to differentiate between two situations: • In the first case, the atomic planes which belong to a certain crystallographic orientation are occupied exclusively by gold or carbon atoms (gold and carbon atomic layers are alternately stacked). This is valid for ZnS(100), ZnS(111) and NaCl(111). In Figure S10 all odd numbers correspond to the layers which are occupied by gold atoms. If we consider this, we can conclude that a surface terminated with gold atoms is energetically preferred.
• In the second case, the atomic planes are always equally occupied by gold and carbon atoms. This is valid for NaCl(100), NaCl(110) and ZnS(110).  that the formation of the (100) plane is the most favorable one in case of ZnS but the most unfavorable one in case of NaCl.
The comparison between the density of states of the isolated cluster and the one of the interacting system (model presented in Figure S11a) shows that hardly any AuC--related peaks are shifted ( Figure   S11b). The total density of states can rather be regarded as a superposition of the densities of the AuC cluster and the graphene substrate. This indicates that there is a weak chemical interaction between AuC and graphene.

Figure S11. (a) Structure of the AuC cluster on graphene used to determine the interaction between
AuC and graphene; b) Total density of states of the AuC cluster on graphene and of the pure graphene sheet and the isolated cluster.