It was shown more recently in our Laboratory [1,2,3] that having a substrate/diffusant/thin-film/cap-layer structure (the thin film was typically several 10 nm thick, with the same order of magnitude of grain size; the refractory metal cap layer was used just to avoid the oxidation), first the diffusant atoms migrated very fast across the thin film and segregated at the film/cap-layer interface. The accumulated atoms at the film/cap layer interface form a secondary diffusion reservoir and atoms diffuse back to the layer. Later on, the thin film was gradually filled up with the diffusing atoms and composition depth profiles, determined by Secondary Neutral Mass Spectroscopy (SNMS), showed a maximum at the cap layer-thin film interface. The accumulated atoms at this interface formed a secondary diffusion reservoir and atoms diffused back to the layer. These observations can be interpreted supposing a bimodal grain boundary structure with different (fast and low) diffusivities. The observed grain boundary diffusion phenomena can be classified as C-type diffusion. The appearance of the peak observed at the cap layer interface can be used as a tool to determine the grain boundary diffusivity along the fast boundaries. Because the fast boundaries were saturated in the first stage of the process, this back-diffusion took place along the low-diffusivity boundaries only. Thus the SNMS depth-profiling is a good method to determine grain boundary diffusivities in a bimodal structure. In addition, from the overall impurity content inside the film the segregation can also be estimated, if the bulk solubility is low and the GB density is known. Numerical simulations of C-type GB diffusion in thin films with a bimodal structure confirmed that the interpretation of the result depicted above is reasonable . In order to estimate roughly the GB diffusion data we determined the fast diffusivity using the first appearance method. The lower diffusivity was determined from the time evolution of the broadening of the diffusant/thin film interface. In addition both (slow and fast) diffusivities were also estimated from fitting numerical solutions obtained in  too.